blob: 4d1659c7f84bfc60fb014b8a90df3143f8f54676 [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 Heoedcb0722012-04-01 14:38:42 -070056struct blkg_stat {
57 struct u64_stats_sync syncp;
58 uint64_t cnt;
59};
60
61struct blkg_rwstat {
62 struct u64_stats_sync syncp;
63 uint64_t cnt[BLKG_RWSTAT_NR];
64};
65
Tejun Heof95a04a2012-04-16 13:57:26 -070066/*
67 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
68 * request_queue (q). This is used by blkcg policies which need to track
69 * information per blkcg - q pair.
70 *
Tejun Heo001bea72015-08-18 14:55:11 -070071 * There can be multiple active blkcg policies and each blkg:policy pair is
72 * represented by a blkg_policy_data which is allocated and freed by each
73 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
74 * area by allocating larger data structure which embeds blkg_policy_data
75 * at the beginning.
Tejun Heof95a04a2012-04-16 13:57:26 -070076 */
Tejun Heo03814112012-03-05 13:15:14 -080077struct blkg_policy_data {
Tejun Heob276a872013-01-09 08:05:12 -080078 /* the blkg and policy id this per-policy data belongs to */
Tejun Heo3c798392012-04-16 13:57:25 -070079 struct blkcg_gq *blkg;
Tejun Heob276a872013-01-09 08:05:12 -080080 int plid;
Tejun Heo03814112012-03-05 13:15:14 -080081};
82
Arianna Avanzinie48453c2015-06-05 23:38:42 +020083/*
Tejun Heoe4a9bde2015-08-18 14:55:16 -070084 * Policies that need to keep per-blkcg data which is independent from any
85 * request_queue associated to it should implement cpd_alloc/free_fn()
86 * methods. A policy can allocate private data area by allocating larger
87 * data structure which embeds blkcg_policy_data at the beginning.
88 * cpd_init() is invoked to let each policy handle per-blkcg data.
Arianna Avanzinie48453c2015-06-05 23:38:42 +020089 */
90struct blkcg_policy_data {
Tejun Heo81437642015-08-18 14:55:15 -070091 /* the blkcg and policy id this per-policy data belongs to */
92 struct blkcg *blkcg;
Arianna Avanzinie48453c2015-06-05 23:38:42 +020093 int plid;
Arianna Avanzinie48453c2015-06-05 23:38:42 +020094};
95
Tejun Heo3c798392012-04-16 13:57:25 -070096/* association between a blk cgroup and a request queue */
97struct blkcg_gq {
Tejun Heoc875f4d2012-03-05 13:15:22 -080098 /* Pointer to the associated request_queue */
Tejun Heo36558c82012-04-16 13:57:24 -070099 struct request_queue *q;
100 struct list_head q_node;
101 struct hlist_node blkcg_node;
Tejun Heo3c798392012-04-16 13:57:25 -0700102 struct blkcg *blkcg;
Tejun Heo3c547862013-01-09 08:05:10 -0800103
Tejun Heoce7acfe2015-05-22 17:13:38 -0400104 /*
105 * Each blkg gets congested separately and the congestion state is
106 * propagated to the matching bdi_writeback_congested.
107 */
108 struct bdi_writeback_congested *wb_congested;
109
Tejun Heo3c547862013-01-09 08:05:10 -0800110 /* all non-root blkcg_gq's are guaranteed to have access to parent */
111 struct blkcg_gq *parent;
112
Tejun Heoa0516612012-06-26 15:05:44 -0700113 /* request allocation list for this blkcg-q pair */
114 struct request_list rl;
Tejun Heo3c547862013-01-09 08:05:10 -0800115
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800116 /* reference count */
Tejun Heoa5049a82014-06-19 17:42:57 -0400117 atomic_t refcnt;
Vivek Goyal22084192009-12-03 12:59:49 -0500118
Tejun Heof427d902013-01-09 08:05:12 -0800119 /* is this blkg online? protected by both blkcg and q locks */
120 bool online;
121
Tejun Heo36558c82012-04-16 13:57:24 -0700122 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800123
Tejun Heo36558c82012-04-16 13:57:24 -0700124 struct rcu_head rcu_head;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500125};
126
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700127typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
Tejun Heo81437642015-08-18 14:55:15 -0700128typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700129typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
Tejun Heo001bea72015-08-18 14:55:11 -0700130typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node);
Tejun Heoa9520cd2015-08-18 14:55:14 -0700131typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
132typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
133typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
Tejun Heo001bea72015-08-18 14:55:11 -0700134typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
Tejun Heoa9520cd2015-08-18 14:55:14 -0700135typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
Vivek Goyal3e252062009-12-04 10:36:42 -0500136
Tejun Heo3c798392012-04-16 13:57:25 -0700137struct blkcg_policy {
Tejun Heo36558c82012-04-16 13:57:24 -0700138 int plid;
Tejun Heo36558c82012-04-16 13:57:24 -0700139 /* cgroup files for the policy */
140 struct cftype *cftypes;
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700141
142 /* operations */
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700143 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200144 blkcg_pol_init_cpd_fn *cpd_init_fn;
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700145 blkcg_pol_free_cpd_fn *cpd_free_fn;
146
Tejun Heo001bea72015-08-18 14:55:11 -0700147 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700148 blkcg_pol_init_pd_fn *pd_init_fn;
Tejun Heof427d902013-01-09 08:05:12 -0800149 blkcg_pol_online_pd_fn *pd_online_fn;
150 blkcg_pol_offline_pd_fn *pd_offline_fn;
Tejun Heo001bea72015-08-18 14:55:11 -0700151 blkcg_pol_free_pd_fn *pd_free_fn;
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700152 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
Vivek Goyal3e252062009-12-04 10:36:42 -0500153};
154
Tejun Heo3c798392012-04-16 13:57:25 -0700155extern struct blkcg blkcg_root;
Tejun Heo496d5e72015-05-22 17:13:21 -0400156extern struct cgroup_subsys_state * const blkcg_root_css;
Tejun Heo36558c82012-04-16 13:57:24 -0700157
Tejun Heo24f29042015-08-18 14:55:17 -0700158struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
159 struct request_queue *q, bool update_hint);
Tejun Heo3c798392012-04-16 13:57:25 -0700160struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
161 struct request_queue *q);
Tejun Heo36558c82012-04-16 13:57:24 -0700162int blkcg_init_queue(struct request_queue *q);
163void blkcg_drain_queue(struct request_queue *q);
164void blkcg_exit_queue(struct request_queue *q);
Tejun Heo5efd6112012-03-05 13:15:12 -0800165
Vivek Goyal3e252062009-12-04 10:36:42 -0500166/* Blkio controller policy registration */
Jens Axboed5bf0292014-06-22 16:31:56 -0600167int blkcg_policy_register(struct blkcg_policy *pol);
Tejun Heo3c798392012-04-16 13:57:25 -0700168void blkcg_policy_unregister(struct blkcg_policy *pol);
Tejun Heo36558c82012-04-16 13:57:24 -0700169int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700170 const struct blkcg_policy *pol);
Tejun Heo36558c82012-04-16 13:57:24 -0700171void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700172 const struct blkcg_policy *pol);
Vivek Goyal3e252062009-12-04 10:36:42 -0500173
Tejun Heo3c798392012-04-16 13:57:25 -0700174void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
Tejun Heof95a04a2012-04-16 13:57:26 -0700175 u64 (*prfill)(struct seq_file *,
176 struct blkg_policy_data *, int),
Tejun Heo3c798392012-04-16 13:57:25 -0700177 const struct blkcg_policy *pol, int data,
Tejun Heoec399342012-04-13 13:11:27 -0700178 bool show_total);
Tejun Heof95a04a2012-04-16 13:57:26 -0700179u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
180u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
Tejun Heo829fdb52012-04-01 14:38:43 -0700181 const struct blkg_rwstat *rwstat);
Tejun Heof95a04a2012-04-16 13:57:26 -0700182u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
183u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
184 int off);
Tejun Heo829fdb52012-04-01 14:38:43 -0700185
Tejun Heo16b3de62013-01-09 08:05:12 -0800186u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off);
187struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
188 int off);
189
Tejun Heo829fdb52012-04-01 14:38:43 -0700190struct blkg_conf_ctx {
Tejun Heo36558c82012-04-16 13:57:24 -0700191 struct gendisk *disk;
Tejun Heo3c798392012-04-16 13:57:25 -0700192 struct blkcg_gq *blkg;
Tejun Heo36558c82012-04-16 13:57:24 -0700193 u64 v;
Tejun Heo829fdb52012-04-01 14:38:43 -0700194};
195
Tejun Heo3c798392012-04-16 13:57:25 -0700196int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
197 const char *input, struct blkg_conf_ctx *ctx);
Tejun Heo829fdb52012-04-01 14:38:43 -0700198void blkg_conf_finish(struct blkg_conf_ctx *ctx);
199
200
Tejun Heoa7c6d552013-08-08 20:11:23 -0400201static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
202{
203 return css ? container_of(css, struct blkcg, css) : NULL;
204}
205
Tejun Heob1208b52012-06-04 20:40:57 -0700206static inline struct blkcg *task_blkcg(struct task_struct *tsk)
207{
Tejun Heo073219e2014-02-08 10:36:58 -0500208 return css_to_blkcg(task_css(tsk, blkio_cgrp_id));
Tejun Heob1208b52012-06-04 20:40:57 -0700209}
210
211static inline struct blkcg *bio_blkcg(struct bio *bio)
212{
213 if (bio && bio->bi_css)
Tejun Heoa7c6d552013-08-08 20:11:23 -0400214 return css_to_blkcg(bio->bi_css);
Tejun Heob1208b52012-06-04 20:40:57 -0700215 return task_blkcg(current);
216}
217
Tejun Heofd383c22015-05-22 17:13:23 -0400218static inline struct cgroup_subsys_state *
219task_get_blkcg_css(struct task_struct *task)
220{
221 return task_get_css(task, blkio_cgrp_id);
222}
223
Tejun Heo03814112012-03-05 13:15:14 -0800224/**
Tejun Heo3c547862013-01-09 08:05:10 -0800225 * blkcg_parent - get the parent of a blkcg
226 * @blkcg: blkcg of interest
227 *
228 * Return the parent blkcg of @blkcg. Can be called anytime.
229 */
230static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
231{
Tejun Heo5c9d5352014-05-16 13:22:48 -0400232 return css_to_blkcg(blkcg->css.parent);
Tejun Heo3c547862013-01-09 08:05:10 -0800233}
234
235/**
Tejun Heo24f29042015-08-18 14:55:17 -0700236 * __blkg_lookup - internal version of blkg_lookup()
237 * @blkcg: blkcg of interest
238 * @q: request_queue of interest
239 * @update_hint: whether to update lookup hint with the result or not
240 *
241 * This is internal version and shouldn't be used by policy
242 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
243 * @q's bypass state. If @update_hint is %true, the caller should be
244 * holding @q->queue_lock and lookup hint is updated on success.
245 */
246static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
247 struct request_queue *q,
248 bool update_hint)
249{
250 struct blkcg_gq *blkg;
251
Tejun Heo85b6bc92015-08-18 14:55:18 -0700252 if (blkcg == &blkcg_root)
253 return q->root_blkg;
254
Tejun Heo24f29042015-08-18 14:55:17 -0700255 blkg = rcu_dereference(blkcg->blkg_hint);
256 if (blkg && blkg->q == q)
257 return blkg;
258
259 return blkg_lookup_slowpath(blkcg, q, update_hint);
260}
261
262/**
263 * blkg_lookup - lookup blkg for the specified blkcg - q pair
264 * @blkcg: blkcg of interest
265 * @q: request_queue of interest
266 *
267 * Lookup blkg for the @blkcg - @q pair. This function should be called
268 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
269 * - see blk_queue_bypass_start() for details.
270 */
271static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
272 struct request_queue *q)
273{
274 WARN_ON_ONCE(!rcu_read_lock_held());
275
276 if (unlikely(blk_queue_bypass(q)))
277 return NULL;
278 return __blkg_lookup(blkcg, q, false);
279}
280
281/**
Tejun Heo03814112012-03-05 13:15:14 -0800282 * blkg_to_pdata - get policy private data
283 * @blkg: blkg of interest
284 * @pol: policy of interest
285 *
286 * Return pointer to private data associated with the @blkg-@pol pair.
287 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700288static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
289 struct blkcg_policy *pol)
Tejun Heo03814112012-03-05 13:15:14 -0800290{
Tejun Heof95a04a2012-04-16 13:57:26 -0700291 return blkg ? blkg->pd[pol->plid] : NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800292}
293
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200294static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
295 struct blkcg_policy *pol)
296{
Tejun Heo81437642015-08-18 14:55:15 -0700297 return blkcg ? blkcg->cpd[pol->plid] : NULL;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200298}
299
Tejun Heo03814112012-03-05 13:15:14 -0800300/**
301 * pdata_to_blkg - get blkg associated with policy private data
Tejun Heof95a04a2012-04-16 13:57:26 -0700302 * @pd: policy private data of interest
Tejun Heo03814112012-03-05 13:15:14 -0800303 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700304 * @pd is policy private data. Determine the blkg it's associated with.
Tejun Heo03814112012-03-05 13:15:14 -0800305 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700306static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
Tejun Heo03814112012-03-05 13:15:14 -0800307{
Tejun Heof95a04a2012-04-16 13:57:26 -0700308 return pd ? pd->blkg : NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800309}
310
Tejun Heo81437642015-08-18 14:55:15 -0700311static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
312{
313 return cpd ? cpd->blkcg : NULL;
314}
315
Tejun Heo54e7ed12012-04-16 13:57:23 -0700316/**
317 * blkg_path - format cgroup path of blkg
318 * @blkg: blkg of interest
319 * @buf: target buffer
320 * @buflen: target buffer length
321 *
322 * Format the path of the cgroup of @blkg into @buf.
323 */
Tejun Heo3c798392012-04-16 13:57:25 -0700324static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
Vivek Goyalafc24d42010-04-26 19:27:56 +0200325{
Tejun Heoe61734c2014-02-12 09:29:50 -0500326 char *p;
Tejun Heo54e7ed12012-04-16 13:57:23 -0700327
Tejun Heoe61734c2014-02-12 09:29:50 -0500328 p = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
329 if (!p) {
Tejun Heo54e7ed12012-04-16 13:57:23 -0700330 strncpy(buf, "<unavailable>", buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500331 return -ENAMETOOLONG;
332 }
333
334 memmove(buf, p, buf + buflen - p);
335 return 0;
Vivek Goyalafc24d42010-04-26 19:27:56 +0200336}
337
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800338/**
339 * blkg_get - get a blkg reference
340 * @blkg: blkg to get
341 *
Tejun Heoa5049a82014-06-19 17:42:57 -0400342 * The caller should be holding an existing reference.
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800343 */
Tejun Heo3c798392012-04-16 13:57:25 -0700344static inline void blkg_get(struct blkcg_gq *blkg)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800345{
Tejun Heoa5049a82014-06-19 17:42:57 -0400346 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
347 atomic_inc(&blkg->refcnt);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800348}
349
Tejun Heo2a4fd072013-05-14 13:52:31 -0700350void __blkg_release_rcu(struct rcu_head *rcu);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800351
352/**
353 * blkg_put - put a blkg reference
354 * @blkg: blkg to put
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800355 */
Tejun Heo3c798392012-04-16 13:57:25 -0700356static inline void blkg_put(struct blkcg_gq *blkg)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800357{
Tejun Heoa5049a82014-06-19 17:42:57 -0400358 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
359 if (atomic_dec_and_test(&blkg->refcnt))
Tejun Heo2a4fd072013-05-14 13:52:31 -0700360 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800361}
362
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700363/**
364 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
365 * @d_blkg: loop cursor pointing to the current descendant
Tejun Heo492eb212013-08-08 20:11:25 -0400366 * @pos_css: used for iteration
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700367 * @p_blkg: target blkg to walk descendants of
368 *
369 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
370 * read locked. If called under either blkcg or queue lock, the iteration
371 * is guaranteed to include all and only online blkgs. The caller may
Tejun Heo492eb212013-08-08 20:11:25 -0400372 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
Tejun Heobd8815a2013-08-08 20:11:27 -0400373 * @p_blkg is included in the iteration and the first node to be visited.
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700374 */
Tejun Heo492eb212013-08-08 20:11:25 -0400375#define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
376 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
377 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700378 (p_blkg)->q, false)))
379
Tejun Heoedcb0722012-04-01 14:38:42 -0700380/**
Tejun Heoaa539cb2013-05-14 13:52:31 -0700381 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
382 * @d_blkg: loop cursor pointing to the current descendant
Tejun Heo492eb212013-08-08 20:11:25 -0400383 * @pos_css: used for iteration
Tejun Heoaa539cb2013-05-14 13:52:31 -0700384 * @p_blkg: target blkg to walk descendants of
385 *
386 * Similar to blkg_for_each_descendant_pre() but performs post-order
Tejun Heobd8815a2013-08-08 20:11:27 -0400387 * traversal instead. Synchronization rules are the same. @p_blkg is
388 * included in the iteration and the last node to be visited.
Tejun Heoaa539cb2013-05-14 13:52:31 -0700389 */
Tejun Heo492eb212013-08-08 20:11:25 -0400390#define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
391 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
392 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
Tejun Heoaa539cb2013-05-14 13:52:31 -0700393 (p_blkg)->q, false)))
394
395/**
Tejun Heoa0516612012-06-26 15:05:44 -0700396 * blk_get_rl - get request_list to use
397 * @q: request_queue of interest
398 * @bio: bio which will be attached to the allocated request (may be %NULL)
399 *
400 * The caller wants to allocate a request from @q to use for @bio. Find
401 * the request_list to use and obtain a reference on it. Should be called
402 * under queue_lock. This function is guaranteed to return non-%NULL
403 * request_list.
404 */
405static inline struct request_list *blk_get_rl(struct request_queue *q,
406 struct bio *bio)
407{
408 struct blkcg *blkcg;
409 struct blkcg_gq *blkg;
410
411 rcu_read_lock();
412
413 blkcg = bio_blkcg(bio);
414
415 /* bypass blkg lookup and use @q->root_rl directly for root */
416 if (blkcg == &blkcg_root)
417 goto root_rl;
418
419 /*
420 * Try to use blkg->rl. blkg lookup may fail under memory pressure
421 * or if either the blkcg or queue is going away. Fall back to
422 * root_rl in such cases.
423 */
Tejun Heoae118892015-08-18 14:55:20 -0700424 blkg = blkg_lookup(blkcg, q);
425 if (unlikely(!blkg))
Tejun Heoa0516612012-06-26 15:05:44 -0700426 goto root_rl;
427
428 blkg_get(blkg);
429 rcu_read_unlock();
430 return &blkg->rl;
431root_rl:
432 rcu_read_unlock();
433 return &q->root_rl;
434}
435
436/**
437 * blk_put_rl - put request_list
438 * @rl: request_list to put
439 *
440 * Put the reference acquired by blk_get_rl(). Should be called under
441 * queue_lock.
442 */
443static inline void blk_put_rl(struct request_list *rl)
444{
Tejun Heo401efbf2015-08-18 14:55:06 -0700445 if (rl->blkg->blkcg != &blkcg_root)
Tejun Heoa0516612012-06-26 15:05:44 -0700446 blkg_put(rl->blkg);
447}
448
449/**
450 * blk_rq_set_rl - associate a request with a request_list
451 * @rq: request of interest
452 * @rl: target request_list
453 *
454 * Associate @rq with @rl so that accounting and freeing can know the
455 * request_list @rq came from.
456 */
457static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
458{
459 rq->rl = rl;
460}
461
462/**
463 * blk_rq_rl - return the request_list a request came from
464 * @rq: request of interest
465 *
466 * Return the request_list @rq is allocated from.
467 */
468static inline struct request_list *blk_rq_rl(struct request *rq)
469{
470 return rq->rl;
471}
472
473struct request_list *__blk_queue_next_rl(struct request_list *rl,
474 struct request_queue *q);
475/**
476 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
477 *
478 * Should be used under queue_lock.
479 */
480#define blk_queue_for_each_rl(rl, q) \
481 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
482
Peter Zijlstra90d38392013-11-12 19:42:14 -0800483static inline void blkg_stat_init(struct blkg_stat *stat)
484{
485 u64_stats_init(&stat->syncp);
486}
487
Tejun Heoa0516612012-06-26 15:05:44 -0700488/**
Tejun Heoedcb0722012-04-01 14:38:42 -0700489 * blkg_stat_add - add a value to a blkg_stat
490 * @stat: target blkg_stat
491 * @val: value to add
492 *
493 * Add @val to @stat. The caller is responsible for synchronizing calls to
494 * this function.
495 */
496static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
497{
498 u64_stats_update_begin(&stat->syncp);
499 stat->cnt += val;
500 u64_stats_update_end(&stat->syncp);
501}
502
503/**
504 * blkg_stat_read - read the current value of a blkg_stat
505 * @stat: blkg_stat to read
506 *
507 * Read the current value of @stat. This function can be called without
508 * synchroniztion and takes care of u64 atomicity.
509 */
510static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
511{
512 unsigned int start;
513 uint64_t v;
514
515 do {
Eric W. Biederman57a77442014-03-13 21:26:42 -0700516 start = u64_stats_fetch_begin_irq(&stat->syncp);
Tejun Heoedcb0722012-04-01 14:38:42 -0700517 v = stat->cnt;
Eric W. Biederman57a77442014-03-13 21:26:42 -0700518 } while (u64_stats_fetch_retry_irq(&stat->syncp, start));
Tejun Heoedcb0722012-04-01 14:38:42 -0700519
520 return v;
521}
522
523/**
524 * blkg_stat_reset - reset a blkg_stat
525 * @stat: blkg_stat to reset
526 */
527static inline void blkg_stat_reset(struct blkg_stat *stat)
528{
529 stat->cnt = 0;
530}
531
532/**
Tejun Heo16b3de62013-01-09 08:05:12 -0800533 * blkg_stat_merge - merge a blkg_stat into another
534 * @to: the destination blkg_stat
535 * @from: the source
536 *
537 * Add @from's count to @to.
538 */
539static inline void blkg_stat_merge(struct blkg_stat *to, struct blkg_stat *from)
540{
541 blkg_stat_add(to, blkg_stat_read(from));
542}
543
Peter Zijlstra90d38392013-11-12 19:42:14 -0800544static inline void blkg_rwstat_init(struct blkg_rwstat *rwstat)
545{
546 u64_stats_init(&rwstat->syncp);
547}
548
Tejun Heo16b3de62013-01-09 08:05:12 -0800549/**
Tejun Heoedcb0722012-04-01 14:38:42 -0700550 * blkg_rwstat_add - add a value to a blkg_rwstat
551 * @rwstat: target blkg_rwstat
552 * @rw: mask of REQ_{WRITE|SYNC}
553 * @val: value to add
554 *
555 * Add @val to @rwstat. The counters are chosen according to @rw. The
556 * caller is responsible for synchronizing calls to this function.
557 */
558static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
559 int rw, uint64_t val)
560{
561 u64_stats_update_begin(&rwstat->syncp);
562
563 if (rw & REQ_WRITE)
564 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
565 else
566 rwstat->cnt[BLKG_RWSTAT_READ] += val;
567 if (rw & REQ_SYNC)
568 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
569 else
570 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
571
572 u64_stats_update_end(&rwstat->syncp);
573}
574
575/**
576 * blkg_rwstat_read - read the current values of a blkg_rwstat
577 * @rwstat: blkg_rwstat to read
578 *
579 * Read the current snapshot of @rwstat and return it as the return value.
580 * This function can be called without synchronization and takes care of
581 * u64 atomicity.
582 */
Tejun Heoc94bed892012-04-16 13:57:22 -0700583static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
Tejun Heoedcb0722012-04-01 14:38:42 -0700584{
585 unsigned int start;
586 struct blkg_rwstat tmp;
587
588 do {
Eric W. Biederman57a77442014-03-13 21:26:42 -0700589 start = u64_stats_fetch_begin_irq(&rwstat->syncp);
Tejun Heoedcb0722012-04-01 14:38:42 -0700590 tmp = *rwstat;
Eric W. Biederman57a77442014-03-13 21:26:42 -0700591 } while (u64_stats_fetch_retry_irq(&rwstat->syncp, start));
Tejun Heoedcb0722012-04-01 14:38:42 -0700592
593 return tmp;
594}
595
596/**
Tejun Heo4d5e80a2013-01-09 08:05:12 -0800597 * blkg_rwstat_total - read the total count of a blkg_rwstat
Tejun Heoedcb0722012-04-01 14:38:42 -0700598 * @rwstat: blkg_rwstat to read
599 *
600 * Return the total count of @rwstat regardless of the IO direction. This
601 * function can be called without synchronization and takes care of u64
602 * atomicity.
603 */
Tejun Heo4d5e80a2013-01-09 08:05:12 -0800604static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
Tejun Heoedcb0722012-04-01 14:38:42 -0700605{
606 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
607
608 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
609}
610
611/**
612 * blkg_rwstat_reset - reset a blkg_rwstat
613 * @rwstat: blkg_rwstat to reset
614 */
615static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
616{
617 memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
618}
619
Tejun Heo16b3de62013-01-09 08:05:12 -0800620/**
621 * blkg_rwstat_merge - merge a blkg_rwstat into another
622 * @to: the destination blkg_rwstat
623 * @from: the source
624 *
625 * Add @from's counts to @to.
626 */
627static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
628 struct blkg_rwstat *from)
629{
630 struct blkg_rwstat v = blkg_rwstat_read(from);
631 int i;
632
633 u64_stats_update_begin(&to->syncp);
634 for (i = 0; i < BLKG_RWSTAT_NR; i++)
635 to->cnt[i] += v.cnt[i];
636 u64_stats_update_end(&to->syncp);
637}
638
Tejun Heoae118892015-08-18 14:55:20 -0700639#ifdef CONFIG_BLK_DEV_THROTTLING
640extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
641 struct bio *bio);
642#else
643static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
644 struct bio *bio) { return false; }
645#endif
646
647static inline bool blkcg_bio_issue_check(struct request_queue *q,
648 struct bio *bio)
649{
650 struct blkcg *blkcg;
651 struct blkcg_gq *blkg;
652 bool throtl = false;
653
654 rcu_read_lock();
655 blkcg = bio_blkcg(bio);
656
657 blkg = blkg_lookup(blkcg, q);
658 if (unlikely(!blkg)) {
659 spin_lock_irq(q->queue_lock);
660 blkg = blkg_lookup_create(blkcg, q);
661 if (IS_ERR(blkg))
662 blkg = NULL;
663 spin_unlock_irq(q->queue_lock);
664 }
665
666 throtl = blk_throtl_bio(q, blkg, bio);
667
668 rcu_read_unlock();
669 return !throtl;
670}
671
Tejun Heo36558c82012-04-16 13:57:24 -0700672#else /* CONFIG_BLK_CGROUP */
673
Tejun Heoefa7d1c2015-05-22 17:13:18 -0400674struct blkcg {
675};
Jens Axboe2f5ea472009-12-03 21:06:43 +0100676
Tejun Heof95a04a2012-04-16 13:57:26 -0700677struct blkg_policy_data {
678};
679
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200680struct blkcg_policy_data {
681};
682
Tejun Heo3c798392012-04-16 13:57:25 -0700683struct blkcg_gq {
Jens Axboe2f5ea472009-12-03 21:06:43 +0100684};
685
Tejun Heo3c798392012-04-16 13:57:25 -0700686struct blkcg_policy {
Vivek Goyal3e252062009-12-04 10:36:42 -0500687};
688
Tejun Heo496d5e72015-05-22 17:13:21 -0400689#define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
690
Tejun Heofd383c22015-05-22 17:13:23 -0400691static inline struct cgroup_subsys_state *
692task_get_blkcg_css(struct task_struct *task)
693{
694 return NULL;
695}
696
Tejun Heoefa7d1c2015-05-22 17:13:18 -0400697#ifdef CONFIG_BLOCK
698
Tejun Heo3c798392012-04-16 13:57:25 -0700699static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
Tejun Heo5efd6112012-03-05 13:15:12 -0800700static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
701static inline void blkcg_drain_queue(struct request_queue *q) { }
702static inline void blkcg_exit_queue(struct request_queue *q) { }
Jens Axboed5bf0292014-06-22 16:31:56 -0600703static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
Tejun Heo3c798392012-04-16 13:57:25 -0700704static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
Tejun Heoa2b16932012-04-13 13:11:33 -0700705static inline int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700706 const struct blkcg_policy *pol) { return 0; }
Tejun Heoa2b16932012-04-13 13:11:33 -0700707static inline void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700708 const struct blkcg_policy *pol) { }
Vivek Goyal3e252062009-12-04 10:36:42 -0500709
Tejun Heob1208b52012-06-04 20:40:57 -0700710static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
Tejun Heoa0516612012-06-26 15:05:44 -0700711
Tejun Heof95a04a2012-04-16 13:57:26 -0700712static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
713 struct blkcg_policy *pol) { return NULL; }
714static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
Tejun Heo3c798392012-04-16 13:57:25 -0700715static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
716static inline void blkg_get(struct blkcg_gq *blkg) { }
717static inline void blkg_put(struct blkcg_gq *blkg) { }
Vivek Goyalafc24d42010-04-26 19:27:56 +0200718
Tejun Heoa0516612012-06-26 15:05:44 -0700719static inline struct request_list *blk_get_rl(struct request_queue *q,
720 struct bio *bio) { return &q->root_rl; }
721static inline void blk_put_rl(struct request_list *rl) { }
722static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
723static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
724
Tejun Heoae118892015-08-18 14:55:20 -0700725static inline bool blkcg_bio_issue_check(struct request_queue *q,
726 struct bio *bio) { return true; }
727
Tejun Heoa0516612012-06-26 15:05:44 -0700728#define blk_queue_for_each_rl(rl, q) \
729 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
730
Tejun Heoefa7d1c2015-05-22 17:13:18 -0400731#endif /* CONFIG_BLOCK */
Tejun Heo36558c82012-04-16 13:57:24 -0700732#endif /* CONFIG_BLK_CGROUP */
733#endif /* _BLK_CGROUP_H */