blob: e8092276af58381ca2cb322c29f8b12928d5509d [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>
Vivek Goyal575969a2011-05-19 15:38:29 -040017#include <linux/u64_stats_sync.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
Vivek Goyal9355aed2010-10-01 21:16:41 +020023/* Max limits for throttle policy */
24#define THROTL_IOPS_MAX UINT_MAX
25
Tejun Heof48ec1d2012-04-13 13:11:25 -070026#ifdef CONFIG_BLK_CGROUP
27
Tejun Heoedcb0722012-04-01 14:38:42 -070028enum blkg_rwstat_type {
29 BLKG_RWSTAT_READ,
30 BLKG_RWSTAT_WRITE,
31 BLKG_RWSTAT_SYNC,
32 BLKG_RWSTAT_ASYNC,
33
34 BLKG_RWSTAT_NR,
35 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
Divyesh Shah303a3ac2010-04-01 15:01:24 -070036};
37
Tejun Heoa6371202012-04-19 16:29:24 -070038struct blkcg_gq;
39
Tejun Heo3c798392012-04-16 13:57:25 -070040struct blkcg {
Tejun Heo36558c82012-04-16 13:57:24 -070041 struct cgroup_subsys_state css;
42 spinlock_t lock;
Tejun Heoa6371202012-04-19 16:29:24 -070043
44 struct radix_tree_root blkg_tree;
45 struct blkcg_gq *blkg_hint;
Tejun Heo36558c82012-04-16 13:57:24 -070046 struct hlist_head blkg_list;
Tejun Heo9a9e8a22012-03-19 15:10:56 -070047
Tejun Heo81437642015-08-18 14:55:15 -070048 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
Tejun Heo52ebea72015-05-22 17:13:37 -040049
Tejun Heo7876f932015-07-09 16:39:49 -040050 struct list_head all_blkcgs_node;
Tejun Heo52ebea72015-05-22 17:13:37 -040051#ifdef CONFIG_CGROUP_WRITEBACK
52 struct list_head cgwb_list;
53#endif
Vivek Goyal31e4c282009-12-03 12:59:42 -050054};
55
Tejun Heoe6269c42015-08-18 14:55:21 -070056/*
57 * blkg_[rw]stat->aux_cnt is excluded for local stats but included for
58 * recursive. Used to carry stats of dead children.
59 */
Tejun Heoedcb0722012-04-01 14:38:42 -070060struct blkg_stat {
61 struct u64_stats_sync syncp;
62 uint64_t cnt;
Tejun Heoe6269c42015-08-18 14:55:21 -070063 atomic64_t aux_cnt;
Tejun Heoedcb0722012-04-01 14:38:42 -070064};
65
66struct blkg_rwstat {
67 struct u64_stats_sync syncp;
68 uint64_t cnt[BLKG_RWSTAT_NR];
Tejun Heoe6269c42015-08-18 14:55:21 -070069 atomic64_t aux_cnt[BLKG_RWSTAT_NR];
Tejun Heoedcb0722012-04-01 14:38:42 -070070};
71
Tejun Heof95a04a2012-04-16 13:57:26 -070072/*
73 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
74 * request_queue (q). This is used by blkcg policies which need to track
75 * information per blkcg - q pair.
76 *
Tejun Heo001bea72015-08-18 14:55:11 -070077 * There can be multiple active blkcg policies and each blkg:policy pair is
78 * represented by a blkg_policy_data which is allocated and freed by each
79 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
80 * area by allocating larger data structure which embeds blkg_policy_data
81 * at the beginning.
Tejun Heof95a04a2012-04-16 13:57:26 -070082 */
Tejun Heo03814112012-03-05 13:15:14 -080083struct blkg_policy_data {
Tejun Heob276a872013-01-09 08:05:12 -080084 /* the blkg and policy id this per-policy data belongs to */
Tejun Heo3c798392012-04-16 13:57:25 -070085 struct blkcg_gq *blkg;
Tejun Heob276a872013-01-09 08:05:12 -080086 int plid;
Tejun Heo03814112012-03-05 13:15:14 -080087};
88
Arianna Avanzinie48453c2015-06-05 23:38:42 +020089/*
Tejun Heoe4a9bde2015-08-18 14:55:16 -070090 * Policies that need to keep per-blkcg data which is independent from any
91 * request_queue associated to it should implement cpd_alloc/free_fn()
92 * methods. A policy can allocate private data area by allocating larger
93 * data structure which embeds blkcg_policy_data at the beginning.
94 * cpd_init() is invoked to let each policy handle per-blkcg data.
Arianna Avanzinie48453c2015-06-05 23:38:42 +020095 */
96struct blkcg_policy_data {
Tejun Heo81437642015-08-18 14:55:15 -070097 /* the blkcg and policy id this per-policy data belongs to */
98 struct blkcg *blkcg;
Arianna Avanzinie48453c2015-06-05 23:38:42 +020099 int plid;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200100};
101
Tejun Heo3c798392012-04-16 13:57:25 -0700102/* association between a blk cgroup and a request queue */
103struct blkcg_gq {
Tejun Heoc875f4d2012-03-05 13:15:22 -0800104 /* Pointer to the associated request_queue */
Tejun Heo36558c82012-04-16 13:57:24 -0700105 struct request_queue *q;
106 struct list_head q_node;
107 struct hlist_node blkcg_node;
Tejun Heo3c798392012-04-16 13:57:25 -0700108 struct blkcg *blkcg;
Tejun Heo3c547862013-01-09 08:05:10 -0800109
Tejun Heoce7acfe2015-05-22 17:13:38 -0400110 /*
111 * Each blkg gets congested separately and the congestion state is
112 * propagated to the matching bdi_writeback_congested.
113 */
114 struct bdi_writeback_congested *wb_congested;
115
Tejun Heo3c547862013-01-09 08:05:10 -0800116 /* all non-root blkcg_gq's are guaranteed to have access to parent */
117 struct blkcg_gq *parent;
118
Tejun Heoa0516612012-06-26 15:05:44 -0700119 /* request allocation list for this blkcg-q pair */
120 struct request_list rl;
Tejun Heo3c547862013-01-09 08:05:10 -0800121
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800122 /* reference count */
Tejun Heoa5049a82014-06-19 17:42:57 -0400123 atomic_t refcnt;
Vivek Goyal22084192009-12-03 12:59:49 -0500124
Tejun Heof427d902013-01-09 08:05:12 -0800125 /* is this blkg online? protected by both blkcg and q locks */
126 bool online;
127
Tejun Heo36558c82012-04-16 13:57:24 -0700128 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800129
Tejun Heo36558c82012-04-16 13:57:24 -0700130 struct rcu_head rcu_head;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500131};
132
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700133typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
Tejun Heo81437642015-08-18 14:55:15 -0700134typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700135typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
Tejun Heo001bea72015-08-18 14:55:11 -0700136typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node);
Tejun Heoa9520cd2015-08-18 14:55:14 -0700137typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
138typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
139typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
Tejun Heo001bea72015-08-18 14:55:11 -0700140typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
Tejun Heoa9520cd2015-08-18 14:55:14 -0700141typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
Vivek Goyal3e252062009-12-04 10:36:42 -0500142
Tejun Heo3c798392012-04-16 13:57:25 -0700143struct blkcg_policy {
Tejun Heo36558c82012-04-16 13:57:24 -0700144 int plid;
Tejun Heo36558c82012-04-16 13:57:24 -0700145 /* cgroup files for the policy */
146 struct cftype *cftypes;
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700147
148 /* operations */
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700149 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200150 blkcg_pol_init_cpd_fn *cpd_init_fn;
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700151 blkcg_pol_free_cpd_fn *cpd_free_fn;
152
Tejun Heo001bea72015-08-18 14:55:11 -0700153 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700154 blkcg_pol_init_pd_fn *pd_init_fn;
Tejun Heof427d902013-01-09 08:05:12 -0800155 blkcg_pol_online_pd_fn *pd_online_fn;
156 blkcg_pol_offline_pd_fn *pd_offline_fn;
Tejun Heo001bea72015-08-18 14:55:11 -0700157 blkcg_pol_free_pd_fn *pd_free_fn;
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700158 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
Vivek Goyal3e252062009-12-04 10:36:42 -0500159};
160
Tejun Heo3c798392012-04-16 13:57:25 -0700161extern struct blkcg blkcg_root;
Tejun Heo496d5e72015-05-22 17:13:21 -0400162extern struct cgroup_subsys_state * const blkcg_root_css;
Tejun Heo36558c82012-04-16 13:57:24 -0700163
Tejun Heo24f29042015-08-18 14:55:17 -0700164struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
165 struct request_queue *q, bool update_hint);
Tejun Heo3c798392012-04-16 13:57:25 -0700166struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
167 struct request_queue *q);
Tejun Heo36558c82012-04-16 13:57:24 -0700168int blkcg_init_queue(struct request_queue *q);
169void blkcg_drain_queue(struct request_queue *q);
170void blkcg_exit_queue(struct request_queue *q);
Tejun Heo5efd6112012-03-05 13:15:12 -0800171
Vivek Goyal3e252062009-12-04 10:36:42 -0500172/* Blkio controller policy registration */
Jens Axboed5bf0292014-06-22 16:31:56 -0600173int blkcg_policy_register(struct blkcg_policy *pol);
Tejun Heo3c798392012-04-16 13:57:25 -0700174void blkcg_policy_unregister(struct blkcg_policy *pol);
Tejun Heo36558c82012-04-16 13:57:24 -0700175int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700176 const struct blkcg_policy *pol);
Tejun Heo36558c82012-04-16 13:57:24 -0700177void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700178 const struct blkcg_policy *pol);
Vivek Goyal3e252062009-12-04 10:36:42 -0500179
Tejun Heo3c798392012-04-16 13:57:25 -0700180void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
Tejun Heof95a04a2012-04-16 13:57:26 -0700181 u64 (*prfill)(struct seq_file *,
182 struct blkg_policy_data *, int),
Tejun Heo3c798392012-04-16 13:57:25 -0700183 const struct blkcg_policy *pol, int data,
Tejun Heoec399342012-04-13 13:11:27 -0700184 bool show_total);
Tejun Heof95a04a2012-04-16 13:57:26 -0700185u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
186u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
Tejun Heo829fdb52012-04-01 14:38:43 -0700187 const struct blkg_rwstat *rwstat);
Tejun Heof95a04a2012-04-16 13:57:26 -0700188u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
189u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
190 int off);
Tejun Heo829fdb52012-04-01 14:38:43 -0700191
Tejun Heo16b3de62013-01-09 08:05:12 -0800192u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off);
193struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
194 int off);
195
Tejun Heo829fdb52012-04-01 14:38:43 -0700196struct blkg_conf_ctx {
Tejun Heo36558c82012-04-16 13:57:24 -0700197 struct gendisk *disk;
Tejun Heo3c798392012-04-16 13:57:25 -0700198 struct blkcg_gq *blkg;
Tejun Heo36558c82012-04-16 13:57:24 -0700199 u64 v;
Tejun Heo829fdb52012-04-01 14:38:43 -0700200};
201
Tejun Heo3c798392012-04-16 13:57:25 -0700202int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
203 const char *input, struct blkg_conf_ctx *ctx);
Tejun Heo829fdb52012-04-01 14:38:43 -0700204void blkg_conf_finish(struct blkg_conf_ctx *ctx);
205
206
Tejun Heoa7c6d552013-08-08 20:11:23 -0400207static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
208{
209 return css ? container_of(css, struct blkcg, css) : NULL;
210}
211
Tejun Heob1208b52012-06-04 20:40:57 -0700212static inline struct blkcg *task_blkcg(struct task_struct *tsk)
213{
Tejun Heo073219e2014-02-08 10:36:58 -0500214 return css_to_blkcg(task_css(tsk, blkio_cgrp_id));
Tejun Heob1208b52012-06-04 20:40:57 -0700215}
216
217static inline struct blkcg *bio_blkcg(struct bio *bio)
218{
219 if (bio && bio->bi_css)
Tejun Heoa7c6d552013-08-08 20:11:23 -0400220 return css_to_blkcg(bio->bi_css);
Tejun Heob1208b52012-06-04 20:40:57 -0700221 return task_blkcg(current);
222}
223
Tejun Heofd383c22015-05-22 17:13:23 -0400224static inline struct cgroup_subsys_state *
225task_get_blkcg_css(struct task_struct *task)
226{
227 return task_get_css(task, blkio_cgrp_id);
228}
229
Tejun Heo03814112012-03-05 13:15:14 -0800230/**
Tejun Heo3c547862013-01-09 08:05:10 -0800231 * blkcg_parent - get the parent of a blkcg
232 * @blkcg: blkcg of interest
233 *
234 * Return the parent blkcg of @blkcg. Can be called anytime.
235 */
236static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
237{
Tejun Heo5c9d5352014-05-16 13:22:48 -0400238 return css_to_blkcg(blkcg->css.parent);
Tejun Heo3c547862013-01-09 08:05:10 -0800239}
240
241/**
Tejun Heo24f29042015-08-18 14:55:17 -0700242 * __blkg_lookup - internal version of blkg_lookup()
243 * @blkcg: blkcg of interest
244 * @q: request_queue of interest
245 * @update_hint: whether to update lookup hint with the result or not
246 *
247 * This is internal version and shouldn't be used by policy
248 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
249 * @q's bypass state. If @update_hint is %true, the caller should be
250 * holding @q->queue_lock and lookup hint is updated on success.
251 */
252static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
253 struct request_queue *q,
254 bool update_hint)
255{
256 struct blkcg_gq *blkg;
257
Tejun Heo85b6bc92015-08-18 14:55:18 -0700258 if (blkcg == &blkcg_root)
259 return q->root_blkg;
260
Tejun Heo24f29042015-08-18 14:55:17 -0700261 blkg = rcu_dereference(blkcg->blkg_hint);
262 if (blkg && blkg->q == q)
263 return blkg;
264
265 return blkg_lookup_slowpath(blkcg, q, update_hint);
266}
267
268/**
269 * blkg_lookup - lookup blkg for the specified blkcg - q pair
270 * @blkcg: blkcg of interest
271 * @q: request_queue of interest
272 *
273 * Lookup blkg for the @blkcg - @q pair. This function should be called
274 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
275 * - see blk_queue_bypass_start() for details.
276 */
277static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
278 struct request_queue *q)
279{
280 WARN_ON_ONCE(!rcu_read_lock_held());
281
282 if (unlikely(blk_queue_bypass(q)))
283 return NULL;
284 return __blkg_lookup(blkcg, q, false);
285}
286
287/**
Tejun Heo03814112012-03-05 13:15:14 -0800288 * blkg_to_pdata - get policy private data
289 * @blkg: blkg of interest
290 * @pol: policy of interest
291 *
292 * Return pointer to private data associated with the @blkg-@pol pair.
293 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700294static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
295 struct blkcg_policy *pol)
Tejun Heo03814112012-03-05 13:15:14 -0800296{
Tejun Heof95a04a2012-04-16 13:57:26 -0700297 return blkg ? blkg->pd[pol->plid] : NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800298}
299
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200300static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
301 struct blkcg_policy *pol)
302{
Tejun Heo81437642015-08-18 14:55:15 -0700303 return blkcg ? blkcg->cpd[pol->plid] : NULL;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200304}
305
Tejun Heo03814112012-03-05 13:15:14 -0800306/**
307 * pdata_to_blkg - get blkg associated with policy private data
Tejun Heof95a04a2012-04-16 13:57:26 -0700308 * @pd: policy private data of interest
Tejun Heo03814112012-03-05 13:15:14 -0800309 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700310 * @pd is policy private data. Determine the blkg it's associated with.
Tejun Heo03814112012-03-05 13:15:14 -0800311 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700312static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
Tejun Heo03814112012-03-05 13:15:14 -0800313{
Tejun Heof95a04a2012-04-16 13:57:26 -0700314 return pd ? pd->blkg : NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800315}
316
Tejun Heo81437642015-08-18 14:55:15 -0700317static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
318{
319 return cpd ? cpd->blkcg : NULL;
320}
321
Tejun Heo54e7ed12012-04-16 13:57:23 -0700322/**
323 * blkg_path - format cgroup path of blkg
324 * @blkg: blkg of interest
325 * @buf: target buffer
326 * @buflen: target buffer length
327 *
328 * Format the path of the cgroup of @blkg into @buf.
329 */
Tejun Heo3c798392012-04-16 13:57:25 -0700330static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
Vivek Goyalafc24d42010-04-26 19:27:56 +0200331{
Tejun Heoe61734c2014-02-12 09:29:50 -0500332 char *p;
Tejun Heo54e7ed12012-04-16 13:57:23 -0700333
Tejun Heoe61734c2014-02-12 09:29:50 -0500334 p = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
335 if (!p) {
Tejun Heo54e7ed12012-04-16 13:57:23 -0700336 strncpy(buf, "<unavailable>", buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500337 return -ENAMETOOLONG;
338 }
339
340 memmove(buf, p, buf + buflen - p);
341 return 0;
Vivek Goyalafc24d42010-04-26 19:27:56 +0200342}
343
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800344/**
345 * blkg_get - get a blkg reference
346 * @blkg: blkg to get
347 *
Tejun Heoa5049a82014-06-19 17:42:57 -0400348 * The caller should be holding an existing reference.
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800349 */
Tejun Heo3c798392012-04-16 13:57:25 -0700350static inline void blkg_get(struct blkcg_gq *blkg)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800351{
Tejun Heoa5049a82014-06-19 17:42:57 -0400352 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
353 atomic_inc(&blkg->refcnt);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800354}
355
Tejun Heo2a4fd072013-05-14 13:52:31 -0700356void __blkg_release_rcu(struct rcu_head *rcu);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800357
358/**
359 * blkg_put - put a blkg reference
360 * @blkg: blkg to put
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800361 */
Tejun Heo3c798392012-04-16 13:57:25 -0700362static inline void blkg_put(struct blkcg_gq *blkg)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800363{
Tejun Heoa5049a82014-06-19 17:42:57 -0400364 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
365 if (atomic_dec_and_test(&blkg->refcnt))
Tejun Heo2a4fd072013-05-14 13:52:31 -0700366 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800367}
368
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700369/**
370 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
371 * @d_blkg: loop cursor pointing to the current descendant
Tejun Heo492eb212013-08-08 20:11:25 -0400372 * @pos_css: used for iteration
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700373 * @p_blkg: target blkg to walk descendants of
374 *
375 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
376 * read locked. If called under either blkcg or queue lock, the iteration
377 * is guaranteed to include all and only online blkgs. The caller may
Tejun Heo492eb212013-08-08 20:11:25 -0400378 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
Tejun Heobd8815a2013-08-08 20:11:27 -0400379 * @p_blkg is included in the iteration and the first node to be visited.
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700380 */
Tejun Heo492eb212013-08-08 20:11:25 -0400381#define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
382 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
383 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700384 (p_blkg)->q, false)))
385
Tejun Heoedcb0722012-04-01 14:38:42 -0700386/**
Tejun Heoaa539cb2013-05-14 13:52:31 -0700387 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
388 * @d_blkg: loop cursor pointing to the current descendant
Tejun Heo492eb212013-08-08 20:11:25 -0400389 * @pos_css: used for iteration
Tejun Heoaa539cb2013-05-14 13:52:31 -0700390 * @p_blkg: target blkg to walk descendants of
391 *
392 * Similar to blkg_for_each_descendant_pre() but performs post-order
Tejun Heobd8815a2013-08-08 20:11:27 -0400393 * traversal instead. Synchronization rules are the same. @p_blkg is
394 * included in the iteration and the last node to be visited.
Tejun Heoaa539cb2013-05-14 13:52:31 -0700395 */
Tejun Heo492eb212013-08-08 20:11:25 -0400396#define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
397 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
398 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
Tejun Heoaa539cb2013-05-14 13:52:31 -0700399 (p_blkg)->q, false)))
400
401/**
Tejun Heoa0516612012-06-26 15:05:44 -0700402 * blk_get_rl - get request_list to use
403 * @q: request_queue of interest
404 * @bio: bio which will be attached to the allocated request (may be %NULL)
405 *
406 * The caller wants to allocate a request from @q to use for @bio. Find
407 * the request_list to use and obtain a reference on it. Should be called
408 * under queue_lock. This function is guaranteed to return non-%NULL
409 * request_list.
410 */
411static inline struct request_list *blk_get_rl(struct request_queue *q,
412 struct bio *bio)
413{
414 struct blkcg *blkcg;
415 struct blkcg_gq *blkg;
416
417 rcu_read_lock();
418
419 blkcg = bio_blkcg(bio);
420
421 /* bypass blkg lookup and use @q->root_rl directly for root */
422 if (blkcg == &blkcg_root)
423 goto root_rl;
424
425 /*
426 * Try to use blkg->rl. blkg lookup may fail under memory pressure
427 * or if either the blkcg or queue is going away. Fall back to
428 * root_rl in such cases.
429 */
Tejun Heoae118892015-08-18 14:55:20 -0700430 blkg = blkg_lookup(blkcg, q);
431 if (unlikely(!blkg))
Tejun Heoa0516612012-06-26 15:05:44 -0700432 goto root_rl;
433
434 blkg_get(blkg);
435 rcu_read_unlock();
436 return &blkg->rl;
437root_rl:
438 rcu_read_unlock();
439 return &q->root_rl;
440}
441
442/**
443 * blk_put_rl - put request_list
444 * @rl: request_list to put
445 *
446 * Put the reference acquired by blk_get_rl(). Should be called under
447 * queue_lock.
448 */
449static inline void blk_put_rl(struct request_list *rl)
450{
Tejun Heo401efbf2015-08-18 14:55:06 -0700451 if (rl->blkg->blkcg != &blkcg_root)
Tejun Heoa0516612012-06-26 15:05:44 -0700452 blkg_put(rl->blkg);
453}
454
455/**
456 * blk_rq_set_rl - associate a request with a request_list
457 * @rq: request of interest
458 * @rl: target request_list
459 *
460 * Associate @rq with @rl so that accounting and freeing can know the
461 * request_list @rq came from.
462 */
463static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
464{
465 rq->rl = rl;
466}
467
468/**
469 * blk_rq_rl - return the request_list a request came from
470 * @rq: request of interest
471 *
472 * Return the request_list @rq is allocated from.
473 */
474static inline struct request_list *blk_rq_rl(struct request *rq)
475{
476 return rq->rl;
477}
478
479struct request_list *__blk_queue_next_rl(struct request_list *rl,
480 struct request_queue *q);
481/**
482 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
483 *
484 * Should be used under queue_lock.
485 */
486#define blk_queue_for_each_rl(rl, q) \
487 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
488
Peter Zijlstra90d38392013-11-12 19:42:14 -0800489static inline void blkg_stat_init(struct blkg_stat *stat)
490{
491 u64_stats_init(&stat->syncp);
Tejun Heoe6269c42015-08-18 14:55:21 -0700492 atomic64_set(&stat->aux_cnt, 0);
Peter Zijlstra90d38392013-11-12 19:42:14 -0800493}
494
Tejun Heoa0516612012-06-26 15:05:44 -0700495/**
Tejun Heoedcb0722012-04-01 14:38:42 -0700496 * blkg_stat_add - add a value to a blkg_stat
497 * @stat: target blkg_stat
498 * @val: value to add
499 *
500 * Add @val to @stat. The caller is responsible for synchronizing calls to
501 * this function.
502 */
503static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
504{
505 u64_stats_update_begin(&stat->syncp);
506 stat->cnt += val;
507 u64_stats_update_end(&stat->syncp);
508}
509
510/**
511 * blkg_stat_read - read the current value of a blkg_stat
512 * @stat: blkg_stat to read
513 *
Tejun Heoe6269c42015-08-18 14:55:21 -0700514 * Read the current value of @stat. The returned value doesn't include the
515 * aux count. This function can be called without synchroniztion and takes
516 * care of u64 atomicity.
Tejun Heoedcb0722012-04-01 14:38:42 -0700517 */
518static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
519{
520 unsigned int start;
521 uint64_t v;
522
523 do {
Eric W. Biederman57a77442014-03-13 21:26:42 -0700524 start = u64_stats_fetch_begin_irq(&stat->syncp);
Tejun Heoedcb0722012-04-01 14:38:42 -0700525 v = stat->cnt;
Eric W. Biederman57a77442014-03-13 21:26:42 -0700526 } while (u64_stats_fetch_retry_irq(&stat->syncp, start));
Tejun Heoedcb0722012-04-01 14:38:42 -0700527
528 return v;
529}
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{
537 stat->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
Peter Zijlstra90d38392013-11-12 19:42:14 -0800555static inline void blkg_rwstat_init(struct blkg_rwstat *rwstat)
556{
Tejun Heoe6269c42015-08-18 14:55:21 -0700557 int i;
558
Peter Zijlstra90d38392013-11-12 19:42:14 -0800559 u64_stats_init(&rwstat->syncp);
Tejun Heoe6269c42015-08-18 14:55:21 -0700560
561 for (i = 0; i < BLKG_RWSTAT_NR; i++)
562 atomic64_set(&rwstat->aux_cnt[i], 0);
Peter Zijlstra90d38392013-11-12 19:42:14 -0800563}
564
Tejun Heo16b3de62013-01-09 08:05:12 -0800565/**
Tejun Heoedcb0722012-04-01 14:38:42 -0700566 * blkg_rwstat_add - add a value to a blkg_rwstat
567 * @rwstat: target blkg_rwstat
568 * @rw: mask of REQ_{WRITE|SYNC}
569 * @val: value to add
570 *
571 * Add @val to @rwstat. The counters are chosen according to @rw. The
572 * caller is responsible for synchronizing calls to this function.
573 */
574static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
575 int rw, uint64_t val)
576{
577 u64_stats_update_begin(&rwstat->syncp);
578
579 if (rw & REQ_WRITE)
580 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
581 else
582 rwstat->cnt[BLKG_RWSTAT_READ] += val;
583 if (rw & REQ_SYNC)
584 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
585 else
586 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
587
588 u64_stats_update_end(&rwstat->syncp);
589}
590
591/**
592 * blkg_rwstat_read - read the current values of a blkg_rwstat
593 * @rwstat: blkg_rwstat to read
594 *
595 * Read the current snapshot of @rwstat and return it as the return value.
596 * This function can be called without synchronization and takes care of
597 * u64 atomicity.
598 */
Tejun Heoc94bed892012-04-16 13:57:22 -0700599static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
Tejun Heoedcb0722012-04-01 14:38:42 -0700600{
601 unsigned int start;
602 struct blkg_rwstat tmp;
603
604 do {
Eric W. Biederman57a77442014-03-13 21:26:42 -0700605 start = u64_stats_fetch_begin_irq(&rwstat->syncp);
Tejun Heoedcb0722012-04-01 14:38:42 -0700606 tmp = *rwstat;
Eric W. Biederman57a77442014-03-13 21:26:42 -0700607 } while (u64_stats_fetch_retry_irq(&rwstat->syncp, start));
Tejun Heoedcb0722012-04-01 14:38:42 -0700608
609 return tmp;
610}
611
612/**
Tejun Heo4d5e80a2013-01-09 08:05:12 -0800613 * blkg_rwstat_total - read the total count of a blkg_rwstat
Tejun Heoedcb0722012-04-01 14:38:42 -0700614 * @rwstat: blkg_rwstat to read
615 *
616 * Return the total count of @rwstat regardless of the IO direction. This
617 * function can be called without synchronization and takes care of u64
618 * atomicity.
619 */
Tejun Heo4d5e80a2013-01-09 08:05:12 -0800620static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
Tejun Heoedcb0722012-04-01 14:38:42 -0700621{
622 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
623
624 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
625}
626
627/**
628 * blkg_rwstat_reset - reset a blkg_rwstat
629 * @rwstat: blkg_rwstat to reset
630 */
631static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
632{
Tejun Heoe6269c42015-08-18 14:55:21 -0700633 int i;
634
Tejun Heoedcb0722012-04-01 14:38:42 -0700635 memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
Tejun Heoe6269c42015-08-18 14:55:21 -0700636
637 for (i = 0; i < BLKG_RWSTAT_NR; i++)
638 atomic64_set(&rwstat->aux_cnt[i], 0);
Tejun Heoedcb0722012-04-01 14:38:42 -0700639}
640
Tejun Heo16b3de62013-01-09 08:05:12 -0800641/**
Tejun Heoe6269c42015-08-18 14:55:21 -0700642 * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count
Tejun Heo16b3de62013-01-09 08:05:12 -0800643 * @to: the destination blkg_rwstat
644 * @from: the source
645 *
Tejun Heoe6269c42015-08-18 14:55:21 -0700646 * Add @from's count including the aux one to @to's aux count.
Tejun Heo16b3de62013-01-09 08:05:12 -0800647 */
Tejun Heoe6269c42015-08-18 14:55:21 -0700648static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to,
649 struct blkg_rwstat *from)
Tejun Heo16b3de62013-01-09 08:05:12 -0800650{
651 struct blkg_rwstat v = blkg_rwstat_read(from);
652 int i;
653
Tejun Heo16b3de62013-01-09 08:05:12 -0800654 for (i = 0; i < BLKG_RWSTAT_NR; i++)
Tejun Heoe6269c42015-08-18 14:55:21 -0700655 atomic64_add(v.cnt[i] + atomic64_read(&from->aux_cnt[i]),
656 &to->aux_cnt[i]);
Tejun Heo16b3de62013-01-09 08:05:12 -0800657}
658
Tejun Heoae118892015-08-18 14:55:20 -0700659#ifdef CONFIG_BLK_DEV_THROTTLING
660extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
661 struct bio *bio);
662#else
663static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
664 struct bio *bio) { return false; }
665#endif
666
667static inline bool blkcg_bio_issue_check(struct request_queue *q,
668 struct bio *bio)
669{
670 struct blkcg *blkcg;
671 struct blkcg_gq *blkg;
672 bool throtl = false;
673
674 rcu_read_lock();
675 blkcg = bio_blkcg(bio);
676
677 blkg = blkg_lookup(blkcg, q);
678 if (unlikely(!blkg)) {
679 spin_lock_irq(q->queue_lock);
680 blkg = blkg_lookup_create(blkcg, q);
681 if (IS_ERR(blkg))
682 blkg = NULL;
683 spin_unlock_irq(q->queue_lock);
684 }
685
686 throtl = blk_throtl_bio(q, blkg, bio);
687
688 rcu_read_unlock();
689 return !throtl;
690}
691
Tejun Heo36558c82012-04-16 13:57:24 -0700692#else /* CONFIG_BLK_CGROUP */
693
Tejun Heoefa7d1c2015-05-22 17:13:18 -0400694struct blkcg {
695};
Jens Axboe2f5ea472009-12-03 21:06:43 +0100696
Tejun Heof95a04a2012-04-16 13:57:26 -0700697struct blkg_policy_data {
698};
699
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200700struct blkcg_policy_data {
701};
702
Tejun Heo3c798392012-04-16 13:57:25 -0700703struct blkcg_gq {
Jens Axboe2f5ea472009-12-03 21:06:43 +0100704};
705
Tejun Heo3c798392012-04-16 13:57:25 -0700706struct blkcg_policy {
Vivek Goyal3e252062009-12-04 10:36:42 -0500707};
708
Tejun Heo496d5e72015-05-22 17:13:21 -0400709#define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
710
Tejun Heofd383c22015-05-22 17:13:23 -0400711static inline struct cgroup_subsys_state *
712task_get_blkcg_css(struct task_struct *task)
713{
714 return NULL;
715}
716
Tejun Heoefa7d1c2015-05-22 17:13:18 -0400717#ifdef CONFIG_BLOCK
718
Tejun Heo3c798392012-04-16 13:57:25 -0700719static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
Tejun Heo5efd6112012-03-05 13:15:12 -0800720static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
721static inline void blkcg_drain_queue(struct request_queue *q) { }
722static inline void blkcg_exit_queue(struct request_queue *q) { }
Jens Axboed5bf0292014-06-22 16:31:56 -0600723static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
Tejun Heo3c798392012-04-16 13:57:25 -0700724static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
Tejun Heoa2b16932012-04-13 13:11:33 -0700725static inline int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700726 const struct blkcg_policy *pol) { return 0; }
Tejun Heoa2b16932012-04-13 13:11:33 -0700727static inline void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700728 const struct blkcg_policy *pol) { }
Vivek Goyal3e252062009-12-04 10:36:42 -0500729
Tejun Heob1208b52012-06-04 20:40:57 -0700730static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
Tejun Heoa0516612012-06-26 15:05:44 -0700731
Tejun Heof95a04a2012-04-16 13:57:26 -0700732static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
733 struct blkcg_policy *pol) { return NULL; }
734static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
Tejun Heo3c798392012-04-16 13:57:25 -0700735static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
736static inline void blkg_get(struct blkcg_gq *blkg) { }
737static inline void blkg_put(struct blkcg_gq *blkg) { }
Vivek Goyalafc24d42010-04-26 19:27:56 +0200738
Tejun Heoa0516612012-06-26 15:05:44 -0700739static inline struct request_list *blk_get_rl(struct request_queue *q,
740 struct bio *bio) { return &q->root_rl; }
741static inline void blk_put_rl(struct request_list *rl) { }
742static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
743static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
744
Tejun Heoae118892015-08-18 14:55:20 -0700745static inline bool blkcg_bio_issue_check(struct request_queue *q,
746 struct bio *bio) { return true; }
747
Tejun Heoa0516612012-06-26 15:05:44 -0700748#define blk_queue_for_each_rl(rl, q) \
749 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
750
Tejun Heoefa7d1c2015-05-22 17:13:18 -0400751#endif /* CONFIG_BLOCK */
Tejun Heo36558c82012-04-16 13:57:24 -0700752#endif /* CONFIG_BLK_CGROUP */
753#endif /* _BLK_CGROUP_H */