Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 2 | #ifndef _BLK_CGROUP_H |
| 3 | #define _BLK_CGROUP_H |
| 4 | /* |
| 5 | * Common Block IO controller cgroup interface |
| 6 | * |
| 7 | * Based on ideas and code from CFQ, CFS and BFQ: |
| 8 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
| 9 | * |
| 10 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> |
| 11 | * Paolo Valente <paolo.valente@unimore.it> |
| 12 | * |
| 13 | * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> |
| 14 | * Nauman Rafique <nauman@google.com> |
| 15 | */ |
| 16 | |
| 17 | #include <linux/cgroup.h> |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 18 | #include <linux/percpu_counter.h> |
Tejun Heo | 829fdb5 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 19 | #include <linux/seq_file.h> |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 20 | #include <linux/radix-tree.h> |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 21 | #include <linux/blkdev.h> |
Tejun Heo | a5049a8 | 2014-06-19 17:42:57 -0400 | [diff] [blame] | 22 | #include <linux/atomic.h> |
Shaohua Li | 902ec5b | 2017-09-14 14:02:06 -0700 | [diff] [blame] | 23 | #include <linux/kthread.h> |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 24 | |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 25 | /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */ |
| 26 | #define BLKG_STAT_CPU_BATCH (INT_MAX / 2) |
| 27 | |
Vivek Goyal | 9355aed | 2010-10-01 21:16:41 +0200 | [diff] [blame] | 28 | /* Max limits for throttle policy */ |
| 29 | #define THROTL_IOPS_MAX UINT_MAX |
| 30 | |
Tejun Heo | f48ec1d | 2012-04-13 13:11:25 -0700 | [diff] [blame] | 31 | #ifdef CONFIG_BLK_CGROUP |
| 32 | |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 33 | enum blkg_rwstat_type { |
| 34 | BLKG_RWSTAT_READ, |
| 35 | BLKG_RWSTAT_WRITE, |
| 36 | BLKG_RWSTAT_SYNC, |
| 37 | BLKG_RWSTAT_ASYNC, |
| 38 | |
| 39 | BLKG_RWSTAT_NR, |
| 40 | BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR, |
Divyesh Shah | 303a3ac | 2010-04-01 15:01:24 -0700 | [diff] [blame] | 41 | }; |
| 42 | |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 43 | struct blkcg_gq; |
| 44 | |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 45 | struct blkcg { |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 46 | struct cgroup_subsys_state css; |
| 47 | spinlock_t lock; |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 48 | |
| 49 | struct radix_tree_root blkg_tree; |
Bart Van Assche | 55679c8 | 2016-09-23 09:07:56 -0700 | [diff] [blame] | 50 | struct blkcg_gq __rcu *blkg_hint; |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 51 | struct hlist_head blkg_list; |
Tejun Heo | 9a9e8a2 | 2012-03-19 15:10:56 -0700 | [diff] [blame] | 52 | |
Tejun Heo | 8143764 | 2015-08-18 14:55:15 -0700 | [diff] [blame] | 53 | struct blkcg_policy_data *cpd[BLKCG_MAX_POLS]; |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 54 | |
Tejun Heo | 7876f93 | 2015-07-09 16:39:49 -0400 | [diff] [blame] | 55 | struct list_head all_blkcgs_node; |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 56 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 57 | struct list_head cgwb_list; |
| 58 | #endif |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 59 | }; |
| 60 | |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 61 | /* |
| 62 | * blkg_[rw]stat->aux_cnt is excluded for local stats but included for |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 63 | * recursive. Used to carry stats of dead children, and, for blkg_rwstat, |
| 64 | * to carry result values from read and sum operations. |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 65 | */ |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 66 | struct blkg_stat { |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 67 | struct percpu_counter cpu_cnt; |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 68 | atomic64_t aux_cnt; |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 69 | }; |
| 70 | |
| 71 | struct blkg_rwstat { |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 72 | struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR]; |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 73 | atomic64_t aux_cnt[BLKG_RWSTAT_NR]; |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 74 | }; |
| 75 | |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 76 | /* |
| 77 | * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a |
| 78 | * request_queue (q). This is used by blkcg policies which need to track |
| 79 | * information per blkcg - q pair. |
| 80 | * |
Tejun Heo | 001bea7 | 2015-08-18 14:55:11 -0700 | [diff] [blame] | 81 | * There can be multiple active blkcg policies and each blkg:policy pair is |
| 82 | * represented by a blkg_policy_data which is allocated and freed by each |
| 83 | * policy's pd_alloc/free_fn() methods. A policy can allocate private data |
| 84 | * area by allocating larger data structure which embeds blkg_policy_data |
| 85 | * at the beginning. |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 86 | */ |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 87 | struct blkg_policy_data { |
Tejun Heo | b276a87 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 88 | /* the blkg and policy id this per-policy data belongs to */ |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 89 | struct blkcg_gq *blkg; |
Tejun Heo | b276a87 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 90 | int plid; |
Joseph Qi | 4c69948 | 2018-03-16 14:51:27 +0800 | [diff] [blame] | 91 | bool offline; |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 92 | }; |
| 93 | |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 94 | /* |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 95 | * Policies that need to keep per-blkcg data which is independent from any |
| 96 | * request_queue associated to it should implement cpd_alloc/free_fn() |
| 97 | * methods. A policy can allocate private data area by allocating larger |
| 98 | * data structure which embeds blkcg_policy_data at the beginning. |
| 99 | * cpd_init() is invoked to let each policy handle per-blkcg data. |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 100 | */ |
| 101 | struct blkcg_policy_data { |
Tejun Heo | 8143764 | 2015-08-18 14:55:15 -0700 | [diff] [blame] | 102 | /* the blkcg and policy id this per-policy data belongs to */ |
| 103 | struct blkcg *blkcg; |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 104 | int plid; |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 105 | }; |
| 106 | |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 107 | /* association between a blk cgroup and a request queue */ |
| 108 | struct blkcg_gq { |
Tejun Heo | c875f4d | 2012-03-05 13:15:22 -0800 | [diff] [blame] | 109 | /* Pointer to the associated request_queue */ |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 110 | struct request_queue *q; |
| 111 | struct list_head q_node; |
| 112 | struct hlist_node blkcg_node; |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 113 | struct blkcg *blkcg; |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 114 | |
Tejun Heo | ce7acfe | 2015-05-22 17:13:38 -0400 | [diff] [blame] | 115 | /* |
| 116 | * Each blkg gets congested separately and the congestion state is |
| 117 | * propagated to the matching bdi_writeback_congested. |
| 118 | */ |
| 119 | struct bdi_writeback_congested *wb_congested; |
| 120 | |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 121 | /* all non-root blkcg_gq's are guaranteed to have access to parent */ |
| 122 | struct blkcg_gq *parent; |
| 123 | |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 124 | /* request allocation list for this blkcg-q pair */ |
| 125 | struct request_list rl; |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 126 | |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 127 | /* reference count */ |
Tejun Heo | a5049a8 | 2014-06-19 17:42:57 -0400 | [diff] [blame] | 128 | atomic_t refcnt; |
Vivek Goyal | 2208419 | 2009-12-03 12:59:49 -0500 | [diff] [blame] | 129 | |
Tejun Heo | f427d90 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 130 | /* is this blkg online? protected by both blkcg and q locks */ |
| 131 | bool online; |
| 132 | |
Tejun Heo | 77ea733 | 2015-08-18 14:55:24 -0700 | [diff] [blame] | 133 | struct blkg_rwstat stat_bytes; |
| 134 | struct blkg_rwstat stat_ios; |
| 135 | |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 136 | struct blkg_policy_data *pd[BLKCG_MAX_POLS]; |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 137 | |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 138 | struct rcu_head rcu_head; |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 139 | }; |
| 140 | |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 141 | typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp); |
Tejun Heo | 8143764 | 2015-08-18 14:55:15 -0700 | [diff] [blame] | 142 | typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd); |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 143 | typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd); |
Tejun Heo | 69d7fde | 2015-08-18 14:55:36 -0700 | [diff] [blame] | 144 | typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd); |
Tejun Heo | 001bea7 | 2015-08-18 14:55:11 -0700 | [diff] [blame] | 145 | typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node); |
Tejun Heo | a9520cd | 2015-08-18 14:55:14 -0700 | [diff] [blame] | 146 | typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd); |
| 147 | typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd); |
| 148 | typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd); |
Tejun Heo | 001bea7 | 2015-08-18 14:55:11 -0700 | [diff] [blame] | 149 | typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd); |
Tejun Heo | a9520cd | 2015-08-18 14:55:14 -0700 | [diff] [blame] | 150 | typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd); |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 151 | typedef size_t (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd, char *buf, |
| 152 | size_t size); |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 153 | |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 154 | struct blkcg_policy { |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 155 | int plid; |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 156 | /* cgroup files for the policy */ |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 157 | struct cftype *dfl_cftypes; |
Tejun Heo | 880f50e | 2015-08-18 14:55:30 -0700 | [diff] [blame] | 158 | struct cftype *legacy_cftypes; |
Tejun Heo | f9fcc2d | 2012-04-16 13:57:27 -0700 | [diff] [blame] | 159 | |
| 160 | /* operations */ |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 161 | blkcg_pol_alloc_cpd_fn *cpd_alloc_fn; |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 162 | blkcg_pol_init_cpd_fn *cpd_init_fn; |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 163 | blkcg_pol_free_cpd_fn *cpd_free_fn; |
Tejun Heo | 69d7fde | 2015-08-18 14:55:36 -0700 | [diff] [blame] | 164 | blkcg_pol_bind_cpd_fn *cpd_bind_fn; |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 165 | |
Tejun Heo | 001bea7 | 2015-08-18 14:55:11 -0700 | [diff] [blame] | 166 | blkcg_pol_alloc_pd_fn *pd_alloc_fn; |
Tejun Heo | f9fcc2d | 2012-04-16 13:57:27 -0700 | [diff] [blame] | 167 | blkcg_pol_init_pd_fn *pd_init_fn; |
Tejun Heo | f427d90 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 168 | blkcg_pol_online_pd_fn *pd_online_fn; |
| 169 | blkcg_pol_offline_pd_fn *pd_offline_fn; |
Tejun Heo | 001bea7 | 2015-08-18 14:55:11 -0700 | [diff] [blame] | 170 | blkcg_pol_free_pd_fn *pd_free_fn; |
Tejun Heo | f9fcc2d | 2012-04-16 13:57:27 -0700 | [diff] [blame] | 171 | blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn; |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 172 | blkcg_pol_stat_pd_fn *pd_stat_fn; |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 173 | }; |
| 174 | |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 175 | extern struct blkcg blkcg_root; |
Tejun Heo | 496d5e7 | 2015-05-22 17:13:21 -0400 | [diff] [blame] | 176 | extern struct cgroup_subsys_state * const blkcg_root_css; |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 177 | |
Tejun Heo | 24f2904 | 2015-08-18 14:55:17 -0700 | [diff] [blame] | 178 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
| 179 | struct request_queue *q, bool update_hint); |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 180 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
| 181 | struct request_queue *q); |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 182 | int blkcg_init_queue(struct request_queue *q); |
| 183 | void blkcg_drain_queue(struct request_queue *q); |
| 184 | void blkcg_exit_queue(struct request_queue *q); |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 185 | |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 186 | /* Blkio controller policy registration */ |
Jens Axboe | d5bf029 | 2014-06-22 16:31:56 -0600 | [diff] [blame] | 187 | int blkcg_policy_register(struct blkcg_policy *pol); |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 188 | void blkcg_policy_unregister(struct blkcg_policy *pol); |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 189 | int blkcg_activate_policy(struct request_queue *q, |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 190 | const struct blkcg_policy *pol); |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 191 | void blkcg_deactivate_policy(struct request_queue *q, |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 192 | const struct blkcg_policy *pol); |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 193 | |
Tejun Heo | dd165eb | 2015-08-18 14:55:33 -0700 | [diff] [blame] | 194 | const char *blkg_dev_name(struct blkcg_gq *blkg); |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 195 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 196 | u64 (*prfill)(struct seq_file *, |
| 197 | struct blkg_policy_data *, int), |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 198 | const struct blkcg_policy *pol, int data, |
Tejun Heo | ec39934 | 2012-04-13 13:11:27 -0700 | [diff] [blame] | 199 | bool show_total); |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 200 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v); |
| 201 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
Tejun Heo | 829fdb5 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 202 | const struct blkg_rwstat *rwstat); |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 203 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off); |
| 204 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
| 205 | int off); |
Tejun Heo | 77ea733 | 2015-08-18 14:55:24 -0700 | [diff] [blame] | 206 | int blkg_print_stat_bytes(struct seq_file *sf, void *v); |
| 207 | int blkg_print_stat_ios(struct seq_file *sf, void *v); |
| 208 | int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v); |
| 209 | int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v); |
Tejun Heo | 829fdb5 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 210 | |
Tejun Heo | f12c74c | 2015-08-18 14:55:23 -0700 | [diff] [blame] | 211 | u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg, |
| 212 | struct blkcg_policy *pol, int off); |
| 213 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, |
| 214 | struct blkcg_policy *pol, int off); |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 215 | |
Tejun Heo | 829fdb5 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 216 | struct blkg_conf_ctx { |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 217 | struct gendisk *disk; |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 218 | struct blkcg_gq *blkg; |
Tejun Heo | 36aa9e5 | 2015-08-18 14:55:31 -0700 | [diff] [blame] | 219 | char *body; |
Tejun Heo | 829fdb5 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 220 | }; |
| 221 | |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 222 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
Tejun Heo | 36aa9e5 | 2015-08-18 14:55:31 -0700 | [diff] [blame] | 223 | char *input, struct blkg_conf_ctx *ctx); |
Tejun Heo | 829fdb5 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 224 | void blkg_conf_finish(struct blkg_conf_ctx *ctx); |
| 225 | |
| 226 | |
Tejun Heo | a7c6d55 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 227 | static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css) |
| 228 | { |
| 229 | return css ? container_of(css, struct blkcg, css) : NULL; |
| 230 | } |
| 231 | |
Tejun Heo | b1208b5 | 2012-06-04 20:40:57 -0700 | [diff] [blame] | 232 | static inline struct blkcg *bio_blkcg(struct bio *bio) |
| 233 | { |
Shaohua Li | 902ec5b | 2017-09-14 14:02:06 -0700 | [diff] [blame] | 234 | struct cgroup_subsys_state *css; |
| 235 | |
Tejun Heo | b1208b5 | 2012-06-04 20:40:57 -0700 | [diff] [blame] | 236 | if (bio && bio->bi_css) |
Tejun Heo | a7c6d55 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 237 | return css_to_blkcg(bio->bi_css); |
Shaohua Li | 902ec5b | 2017-09-14 14:02:06 -0700 | [diff] [blame] | 238 | css = kthread_blkcg(); |
| 239 | if (css) |
| 240 | return css_to_blkcg(css); |
| 241 | return css_to_blkcg(task_css(current, io_cgrp_id)); |
Tejun Heo | fd383c2 | 2015-05-22 17:13:23 -0400 | [diff] [blame] | 242 | } |
| 243 | |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 244 | /** |
Josef Bacik | c7c98fd | 2018-07-03 11:14:51 -0400 | [diff] [blame] | 245 | * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg |
| 246 | * @return: true if this bio needs to be submitted with the root blkg context. |
| 247 | * |
| 248 | * In order to avoid priority inversions we sometimes need to issue a bio as if |
| 249 | * it were attached to the root blkg, and then backcharge to the actual owning |
| 250 | * blkg. The idea is we do bio_blkcg() to look up the actual context for the |
| 251 | * bio and attach the appropriate blkg to the bio. Then we call this helper and |
| 252 | * if it is true run with the root blkg for that queue and then do any |
| 253 | * backcharging to the originating cgroup once the io is complete. |
| 254 | */ |
| 255 | static inline bool bio_issue_as_root_blkg(struct bio *bio) |
| 256 | { |
Josef Bacik | 0d1e0c7 | 2018-07-03 11:14:53 -0400 | [diff] [blame^] | 257 | return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0; |
Josef Bacik | c7c98fd | 2018-07-03 11:14:51 -0400 | [diff] [blame] | 258 | } |
| 259 | |
| 260 | /** |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 261 | * blkcg_parent - get the parent of a blkcg |
| 262 | * @blkcg: blkcg of interest |
| 263 | * |
| 264 | * Return the parent blkcg of @blkcg. Can be called anytime. |
| 265 | */ |
| 266 | static inline struct blkcg *blkcg_parent(struct blkcg *blkcg) |
| 267 | { |
Tejun Heo | 5c9d535 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 268 | return css_to_blkcg(blkcg->css.parent); |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 269 | } |
| 270 | |
| 271 | /** |
Tejun Heo | 24f2904 | 2015-08-18 14:55:17 -0700 | [diff] [blame] | 272 | * __blkg_lookup - internal version of blkg_lookup() |
| 273 | * @blkcg: blkcg of interest |
| 274 | * @q: request_queue of interest |
| 275 | * @update_hint: whether to update lookup hint with the result or not |
| 276 | * |
| 277 | * This is internal version and shouldn't be used by policy |
| 278 | * implementations. Looks up blkgs for the @blkcg - @q pair regardless of |
| 279 | * @q's bypass state. If @update_hint is %true, the caller should be |
| 280 | * holding @q->queue_lock and lookup hint is updated on success. |
| 281 | */ |
| 282 | static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, |
| 283 | struct request_queue *q, |
| 284 | bool update_hint) |
| 285 | { |
| 286 | struct blkcg_gq *blkg; |
| 287 | |
Tejun Heo | 85b6bc9 | 2015-08-18 14:55:18 -0700 | [diff] [blame] | 288 | if (blkcg == &blkcg_root) |
| 289 | return q->root_blkg; |
| 290 | |
Tejun Heo | 24f2904 | 2015-08-18 14:55:17 -0700 | [diff] [blame] | 291 | blkg = rcu_dereference(blkcg->blkg_hint); |
| 292 | if (blkg && blkg->q == q) |
| 293 | return blkg; |
| 294 | |
| 295 | return blkg_lookup_slowpath(blkcg, q, update_hint); |
| 296 | } |
| 297 | |
| 298 | /** |
| 299 | * blkg_lookup - lookup blkg for the specified blkcg - q pair |
| 300 | * @blkcg: blkcg of interest |
| 301 | * @q: request_queue of interest |
| 302 | * |
| 303 | * Lookup blkg for the @blkcg - @q pair. This function should be called |
| 304 | * under RCU read lock and is guaranteed to return %NULL if @q is bypassing |
| 305 | * - see blk_queue_bypass_start() for details. |
| 306 | */ |
| 307 | static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, |
| 308 | struct request_queue *q) |
| 309 | { |
| 310 | WARN_ON_ONCE(!rcu_read_lock_held()); |
| 311 | |
| 312 | if (unlikely(blk_queue_bypass(q))) |
| 313 | return NULL; |
| 314 | return __blkg_lookup(blkcg, q, false); |
| 315 | } |
| 316 | |
| 317 | /** |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 318 | * blkg_to_pdata - get policy private data |
| 319 | * @blkg: blkg of interest |
| 320 | * @pol: policy of interest |
| 321 | * |
| 322 | * Return pointer to private data associated with the @blkg-@pol pair. |
| 323 | */ |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 324 | static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg, |
| 325 | struct blkcg_policy *pol) |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 326 | { |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 327 | return blkg ? blkg->pd[pol->plid] : NULL; |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 328 | } |
| 329 | |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 330 | static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg, |
| 331 | struct blkcg_policy *pol) |
| 332 | { |
Tejun Heo | 8143764 | 2015-08-18 14:55:15 -0700 | [diff] [blame] | 333 | return blkcg ? blkcg->cpd[pol->plid] : NULL; |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 334 | } |
| 335 | |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 336 | /** |
| 337 | * pdata_to_blkg - get blkg associated with policy private data |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 338 | * @pd: policy private data of interest |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 339 | * |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 340 | * @pd is policy private data. Determine the blkg it's associated with. |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 341 | */ |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 342 | static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 343 | { |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 344 | return pd ? pd->blkg : NULL; |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 345 | } |
| 346 | |
Tejun Heo | 8143764 | 2015-08-18 14:55:15 -0700 | [diff] [blame] | 347 | static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd) |
| 348 | { |
| 349 | return cpd ? cpd->blkcg : NULL; |
| 350 | } |
| 351 | |
Tejun Heo | 54e7ed1 | 2012-04-16 13:57:23 -0700 | [diff] [blame] | 352 | /** |
| 353 | * blkg_path - format cgroup path of blkg |
| 354 | * @blkg: blkg of interest |
| 355 | * @buf: target buffer |
| 356 | * @buflen: target buffer length |
| 357 | * |
| 358 | * Format the path of the cgroup of @blkg into @buf. |
| 359 | */ |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 360 | static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen) |
Vivek Goyal | afc24d4 | 2010-04-26 19:27:56 +0200 | [diff] [blame] | 361 | { |
Tejun Heo | 4c737b4 | 2016-08-10 11:23:44 -0400 | [diff] [blame] | 362 | return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen); |
Vivek Goyal | afc24d4 | 2010-04-26 19:27:56 +0200 | [diff] [blame] | 363 | } |
| 364 | |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 365 | /** |
| 366 | * blkg_get - get a blkg reference |
| 367 | * @blkg: blkg to get |
| 368 | * |
Tejun Heo | a5049a8 | 2014-06-19 17:42:57 -0400 | [diff] [blame] | 369 | * The caller should be holding an existing reference. |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 370 | */ |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 371 | static inline void blkg_get(struct blkcg_gq *blkg) |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 372 | { |
Tejun Heo | a5049a8 | 2014-06-19 17:42:57 -0400 | [diff] [blame] | 373 | WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0); |
| 374 | atomic_inc(&blkg->refcnt); |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 375 | } |
| 376 | |
Tejun Heo | 2a4fd07 | 2013-05-14 13:52:31 -0700 | [diff] [blame] | 377 | void __blkg_release_rcu(struct rcu_head *rcu); |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 378 | |
| 379 | /** |
| 380 | * blkg_put - put a blkg reference |
| 381 | * @blkg: blkg to put |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 382 | */ |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 383 | static inline void blkg_put(struct blkcg_gq *blkg) |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 384 | { |
Tejun Heo | a5049a8 | 2014-06-19 17:42:57 -0400 | [diff] [blame] | 385 | WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0); |
| 386 | if (atomic_dec_and_test(&blkg->refcnt)) |
Tejun Heo | 2a4fd07 | 2013-05-14 13:52:31 -0700 | [diff] [blame] | 387 | call_rcu(&blkg->rcu_head, __blkg_release_rcu); |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 388 | } |
| 389 | |
Tejun Heo | dd4a4ff | 2013-05-14 13:52:30 -0700 | [diff] [blame] | 390 | /** |
| 391 | * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants |
| 392 | * @d_blkg: loop cursor pointing to the current descendant |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 393 | * @pos_css: used for iteration |
Tejun Heo | dd4a4ff | 2013-05-14 13:52:30 -0700 | [diff] [blame] | 394 | * @p_blkg: target blkg to walk descendants of |
| 395 | * |
| 396 | * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU |
| 397 | * read locked. If called under either blkcg or queue lock, the iteration |
| 398 | * is guaranteed to include all and only online blkgs. The caller may |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 399 | * update @pos_css by calling css_rightmost_descendant() to skip subtree. |
Tejun Heo | bd8815a | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 400 | * @p_blkg is included in the iteration and the first node to be visited. |
Tejun Heo | dd4a4ff | 2013-05-14 13:52:30 -0700 | [diff] [blame] | 401 | */ |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 402 | #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \ |
| 403 | css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \ |
| 404 | if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \ |
Tejun Heo | dd4a4ff | 2013-05-14 13:52:30 -0700 | [diff] [blame] | 405 | (p_blkg)->q, false))) |
| 406 | |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 407 | /** |
Tejun Heo | aa539cb | 2013-05-14 13:52:31 -0700 | [diff] [blame] | 408 | * blkg_for_each_descendant_post - post-order walk of a blkg's descendants |
| 409 | * @d_blkg: loop cursor pointing to the current descendant |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 410 | * @pos_css: used for iteration |
Tejun Heo | aa539cb | 2013-05-14 13:52:31 -0700 | [diff] [blame] | 411 | * @p_blkg: target blkg to walk descendants of |
| 412 | * |
| 413 | * Similar to blkg_for_each_descendant_pre() but performs post-order |
Tejun Heo | bd8815a | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 414 | * traversal instead. Synchronization rules are the same. @p_blkg is |
| 415 | * included in the iteration and the last node to be visited. |
Tejun Heo | aa539cb | 2013-05-14 13:52:31 -0700 | [diff] [blame] | 416 | */ |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 417 | #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \ |
| 418 | css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \ |
| 419 | if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \ |
Tejun Heo | aa539cb | 2013-05-14 13:52:31 -0700 | [diff] [blame] | 420 | (p_blkg)->q, false))) |
| 421 | |
| 422 | /** |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 423 | * blk_get_rl - get request_list to use |
| 424 | * @q: request_queue of interest |
| 425 | * @bio: bio which will be attached to the allocated request (may be %NULL) |
| 426 | * |
| 427 | * The caller wants to allocate a request from @q to use for @bio. Find |
| 428 | * the request_list to use and obtain a reference on it. Should be called |
| 429 | * under queue_lock. This function is guaranteed to return non-%NULL |
| 430 | * request_list. |
| 431 | */ |
| 432 | static inline struct request_list *blk_get_rl(struct request_queue *q, |
| 433 | struct bio *bio) |
| 434 | { |
| 435 | struct blkcg *blkcg; |
| 436 | struct blkcg_gq *blkg; |
| 437 | |
| 438 | rcu_read_lock(); |
| 439 | |
| 440 | blkcg = bio_blkcg(bio); |
| 441 | |
| 442 | /* bypass blkg lookup and use @q->root_rl directly for root */ |
| 443 | if (blkcg == &blkcg_root) |
| 444 | goto root_rl; |
| 445 | |
| 446 | /* |
| 447 | * Try to use blkg->rl. blkg lookup may fail under memory pressure |
| 448 | * or if either the blkcg or queue is going away. Fall back to |
| 449 | * root_rl in such cases. |
| 450 | */ |
Tejun Heo | ae11889 | 2015-08-18 14:55:20 -0700 | [diff] [blame] | 451 | blkg = blkg_lookup(blkcg, q); |
| 452 | if (unlikely(!blkg)) |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 453 | goto root_rl; |
| 454 | |
| 455 | blkg_get(blkg); |
| 456 | rcu_read_unlock(); |
| 457 | return &blkg->rl; |
| 458 | root_rl: |
| 459 | rcu_read_unlock(); |
| 460 | return &q->root_rl; |
| 461 | } |
| 462 | |
| 463 | /** |
| 464 | * blk_put_rl - put request_list |
| 465 | * @rl: request_list to put |
| 466 | * |
| 467 | * Put the reference acquired by blk_get_rl(). Should be called under |
| 468 | * queue_lock. |
| 469 | */ |
| 470 | static inline void blk_put_rl(struct request_list *rl) |
| 471 | { |
Tejun Heo | 401efbf | 2015-08-18 14:55:06 -0700 | [diff] [blame] | 472 | if (rl->blkg->blkcg != &blkcg_root) |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 473 | blkg_put(rl->blkg); |
| 474 | } |
| 475 | |
| 476 | /** |
| 477 | * blk_rq_set_rl - associate a request with a request_list |
| 478 | * @rq: request of interest |
| 479 | * @rl: target request_list |
| 480 | * |
| 481 | * Associate @rq with @rl so that accounting and freeing can know the |
| 482 | * request_list @rq came from. |
| 483 | */ |
| 484 | static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) |
| 485 | { |
| 486 | rq->rl = rl; |
| 487 | } |
| 488 | |
| 489 | /** |
| 490 | * blk_rq_rl - return the request_list a request came from |
| 491 | * @rq: request of interest |
| 492 | * |
| 493 | * Return the request_list @rq is allocated from. |
| 494 | */ |
| 495 | static inline struct request_list *blk_rq_rl(struct request *rq) |
| 496 | { |
| 497 | return rq->rl; |
| 498 | } |
| 499 | |
| 500 | struct request_list *__blk_queue_next_rl(struct request_list *rl, |
| 501 | struct request_queue *q); |
| 502 | /** |
| 503 | * blk_queue_for_each_rl - iterate through all request_lists of a request_queue |
| 504 | * |
| 505 | * Should be used under queue_lock. |
| 506 | */ |
| 507 | #define blk_queue_for_each_rl(rl, q) \ |
| 508 | for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q))) |
| 509 | |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 510 | static inline int blkg_stat_init(struct blkg_stat *stat, gfp_t gfp) |
Peter Zijlstra | 90d3839 | 2013-11-12 19:42:14 -0800 | [diff] [blame] | 511 | { |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 512 | int ret; |
| 513 | |
| 514 | ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp); |
| 515 | if (ret) |
| 516 | return ret; |
| 517 | |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 518 | atomic64_set(&stat->aux_cnt, 0); |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 519 | return 0; |
| 520 | } |
| 521 | |
| 522 | static inline void blkg_stat_exit(struct blkg_stat *stat) |
| 523 | { |
| 524 | percpu_counter_destroy(&stat->cpu_cnt); |
Peter Zijlstra | 90d3839 | 2013-11-12 19:42:14 -0800 | [diff] [blame] | 525 | } |
| 526 | |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 527 | /** |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 528 | * blkg_stat_add - add a value to a blkg_stat |
| 529 | * @stat: target blkg_stat |
| 530 | * @val: value to add |
| 531 | * |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 532 | * Add @val to @stat. The caller must ensure that IRQ on the same CPU |
| 533 | * don't re-enter this function for the same counter. |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 534 | */ |
| 535 | static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val) |
| 536 | { |
Nikolay Borisov | 104b4e5 | 2017-06-20 21:01:20 +0300 | [diff] [blame] | 537 | percpu_counter_add_batch(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH); |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 538 | } |
| 539 | |
| 540 | /** |
| 541 | * blkg_stat_read - read the current value of a blkg_stat |
| 542 | * @stat: blkg_stat to read |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 543 | */ |
| 544 | static inline uint64_t blkg_stat_read(struct blkg_stat *stat) |
| 545 | { |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 546 | return percpu_counter_sum_positive(&stat->cpu_cnt); |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 547 | } |
| 548 | |
| 549 | /** |
| 550 | * blkg_stat_reset - reset a blkg_stat |
| 551 | * @stat: blkg_stat to reset |
| 552 | */ |
| 553 | static inline void blkg_stat_reset(struct blkg_stat *stat) |
| 554 | { |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 555 | percpu_counter_set(&stat->cpu_cnt, 0); |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 556 | atomic64_set(&stat->aux_cnt, 0); |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 557 | } |
| 558 | |
| 559 | /** |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 560 | * blkg_stat_add_aux - add a blkg_stat into another's aux count |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 561 | * @to: the destination blkg_stat |
| 562 | * @from: the source |
| 563 | * |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 564 | * Add @from's count including the aux one to @to's aux count. |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 565 | */ |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 566 | static inline void blkg_stat_add_aux(struct blkg_stat *to, |
| 567 | struct blkg_stat *from) |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 568 | { |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 569 | atomic64_add(blkg_stat_read(from) + atomic64_read(&from->aux_cnt), |
| 570 | &to->aux_cnt); |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 571 | } |
| 572 | |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 573 | static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp) |
| 574 | { |
| 575 | int i, ret; |
| 576 | |
| 577 | for (i = 0; i < BLKG_RWSTAT_NR; i++) { |
| 578 | ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp); |
| 579 | if (ret) { |
| 580 | while (--i >= 0) |
| 581 | percpu_counter_destroy(&rwstat->cpu_cnt[i]); |
| 582 | return ret; |
| 583 | } |
| 584 | atomic64_set(&rwstat->aux_cnt[i], 0); |
| 585 | } |
| 586 | return 0; |
| 587 | } |
| 588 | |
| 589 | static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat) |
Peter Zijlstra | 90d3839 | 2013-11-12 19:42:14 -0800 | [diff] [blame] | 590 | { |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 591 | int i; |
| 592 | |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 593 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 594 | percpu_counter_destroy(&rwstat->cpu_cnt[i]); |
Peter Zijlstra | 90d3839 | 2013-11-12 19:42:14 -0800 | [diff] [blame] | 595 | } |
| 596 | |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 597 | /** |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 598 | * blkg_rwstat_add - add a value to a blkg_rwstat |
| 599 | * @rwstat: target blkg_rwstat |
Christoph Hellwig | ef295ec | 2016-10-28 08:48:16 -0600 | [diff] [blame] | 600 | * @op: REQ_OP and flags |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 601 | * @val: value to add |
| 602 | * |
| 603 | * Add @val to @rwstat. The counters are chosen according to @rw. The |
| 604 | * caller is responsible for synchronizing calls to this function. |
| 605 | */ |
| 606 | static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat, |
Christoph Hellwig | ef295ec | 2016-10-28 08:48:16 -0600 | [diff] [blame] | 607 | unsigned int op, uint64_t val) |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 608 | { |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 609 | struct percpu_counter *cnt; |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 610 | |
Mike Christie | 63a4cc2 | 2016-06-05 14:32:14 -0500 | [diff] [blame] | 611 | if (op_is_write(op)) |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 612 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE]; |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 613 | else |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 614 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ]; |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 615 | |
Nikolay Borisov | 104b4e5 | 2017-06-20 21:01:20 +0300 | [diff] [blame] | 616 | percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH); |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 617 | |
Christoph Hellwig | d71d9ae | 2016-11-01 07:40:03 -0600 | [diff] [blame] | 618 | if (op_is_sync(op)) |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 619 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC]; |
| 620 | else |
| 621 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC]; |
| 622 | |
Nikolay Borisov | 104b4e5 | 2017-06-20 21:01:20 +0300 | [diff] [blame] | 623 | percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH); |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 624 | } |
| 625 | |
| 626 | /** |
| 627 | * blkg_rwstat_read - read the current values of a blkg_rwstat |
| 628 | * @rwstat: blkg_rwstat to read |
| 629 | * |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 630 | * Read the current snapshot of @rwstat and return it in the aux counts. |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 631 | */ |
Tejun Heo | c94bed89 | 2012-04-16 13:57:22 -0700 | [diff] [blame] | 632 | static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat) |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 633 | { |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 634 | struct blkg_rwstat result; |
| 635 | int i; |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 636 | |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 637 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
| 638 | atomic64_set(&result.aux_cnt[i], |
| 639 | percpu_counter_sum_positive(&rwstat->cpu_cnt[i])); |
| 640 | return result; |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 641 | } |
| 642 | |
| 643 | /** |
Tejun Heo | 4d5e80a | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 644 | * blkg_rwstat_total - read the total count of a blkg_rwstat |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 645 | * @rwstat: blkg_rwstat to read |
| 646 | * |
| 647 | * Return the total count of @rwstat regardless of the IO direction. This |
| 648 | * function can be called without synchronization and takes care of u64 |
| 649 | * atomicity. |
| 650 | */ |
Tejun Heo | 4d5e80a | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 651 | static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat) |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 652 | { |
| 653 | struct blkg_rwstat tmp = blkg_rwstat_read(rwstat); |
| 654 | |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 655 | return atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) + |
| 656 | atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]); |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 657 | } |
| 658 | |
| 659 | /** |
| 660 | * blkg_rwstat_reset - reset a blkg_rwstat |
| 661 | * @rwstat: blkg_rwstat to reset |
| 662 | */ |
| 663 | static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat) |
| 664 | { |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 665 | int i; |
| 666 | |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 667 | for (i = 0; i < BLKG_RWSTAT_NR; i++) { |
| 668 | percpu_counter_set(&rwstat->cpu_cnt[i], 0); |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 669 | atomic64_set(&rwstat->aux_cnt[i], 0); |
Tejun Heo | 24bdb8e | 2015-08-18 14:55:22 -0700 | [diff] [blame] | 670 | } |
Tejun Heo | edcb072 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 671 | } |
| 672 | |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 673 | /** |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 674 | * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 675 | * @to: the destination blkg_rwstat |
| 676 | * @from: the source |
| 677 | * |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 678 | * Add @from's count including the aux one to @to's aux count. |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 679 | */ |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 680 | static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to, |
| 681 | struct blkg_rwstat *from) |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 682 | { |
Arnd Bergmann | ddc2123 | 2018-01-16 16:01:36 +0100 | [diff] [blame] | 683 | u64 sum[BLKG_RWSTAT_NR]; |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 684 | int i; |
| 685 | |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 686 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
Arnd Bergmann | ddc2123 | 2018-01-16 16:01:36 +0100 | [diff] [blame] | 687 | sum[i] = percpu_counter_sum_positive(&from->cpu_cnt[i]); |
| 688 | |
| 689 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
| 690 | atomic64_add(sum[i] + atomic64_read(&from->aux_cnt[i]), |
Tejun Heo | e6269c4 | 2015-08-18 14:55:21 -0700 | [diff] [blame] | 691 | &to->aux_cnt[i]); |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 692 | } |
| 693 | |
Tejun Heo | ae11889 | 2015-08-18 14:55:20 -0700 | [diff] [blame] | 694 | #ifdef CONFIG_BLK_DEV_THROTTLING |
| 695 | extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg, |
| 696 | struct bio *bio); |
| 697 | #else |
| 698 | static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg, |
| 699 | struct bio *bio) { return false; } |
| 700 | #endif |
| 701 | |
| 702 | static inline bool blkcg_bio_issue_check(struct request_queue *q, |
| 703 | struct bio *bio) |
| 704 | { |
| 705 | struct blkcg *blkcg; |
| 706 | struct blkcg_gq *blkg; |
| 707 | bool throtl = false; |
| 708 | |
| 709 | rcu_read_lock(); |
| 710 | blkcg = bio_blkcg(bio); |
| 711 | |
Shaohua Li | 007cc56 | 2017-07-12 11:49:54 -0700 | [diff] [blame] | 712 | /* associate blkcg if bio hasn't attached one */ |
| 713 | bio_associate_blkcg(bio, &blkcg->css); |
| 714 | |
Tejun Heo | ae11889 | 2015-08-18 14:55:20 -0700 | [diff] [blame] | 715 | blkg = blkg_lookup(blkcg, q); |
| 716 | if (unlikely(!blkg)) { |
| 717 | spin_lock_irq(q->queue_lock); |
| 718 | blkg = blkg_lookup_create(blkcg, q); |
| 719 | if (IS_ERR(blkg)) |
| 720 | blkg = NULL; |
| 721 | spin_unlock_irq(q->queue_lock); |
| 722 | } |
| 723 | |
| 724 | throtl = blk_throtl_bio(q, blkg, bio); |
| 725 | |
Tejun Heo | 77ea733 | 2015-08-18 14:55:24 -0700 | [diff] [blame] | 726 | if (!throtl) { |
| 727 | blkg = blkg ?: q->root_blkg; |
Christoph Hellwig | ef295ec | 2016-10-28 08:48:16 -0600 | [diff] [blame] | 728 | blkg_rwstat_add(&blkg->stat_bytes, bio->bi_opf, |
Tejun Heo | 77ea733 | 2015-08-18 14:55:24 -0700 | [diff] [blame] | 729 | bio->bi_iter.bi_size); |
Christoph Hellwig | ef295ec | 2016-10-28 08:48:16 -0600 | [diff] [blame] | 730 | blkg_rwstat_add(&blkg->stat_ios, bio->bi_opf, 1); |
Tejun Heo | 77ea733 | 2015-08-18 14:55:24 -0700 | [diff] [blame] | 731 | } |
| 732 | |
Tejun Heo | ae11889 | 2015-08-18 14:55:20 -0700 | [diff] [blame] | 733 | rcu_read_unlock(); |
| 734 | return !throtl; |
| 735 | } |
| 736 | |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 737 | #else /* CONFIG_BLK_CGROUP */ |
| 738 | |
Tejun Heo | efa7d1c | 2015-05-22 17:13:18 -0400 | [diff] [blame] | 739 | struct blkcg { |
| 740 | }; |
Jens Axboe | 2f5ea47 | 2009-12-03 21:06:43 +0100 | [diff] [blame] | 741 | |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 742 | struct blkg_policy_data { |
| 743 | }; |
| 744 | |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 745 | struct blkcg_policy_data { |
| 746 | }; |
| 747 | |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 748 | struct blkcg_gq { |
Jens Axboe | 2f5ea47 | 2009-12-03 21:06:43 +0100 | [diff] [blame] | 749 | }; |
| 750 | |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 751 | struct blkcg_policy { |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 752 | }; |
| 753 | |
Tejun Heo | 496d5e7 | 2015-05-22 17:13:21 -0400 | [diff] [blame] | 754 | #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL)) |
| 755 | |
Tejun Heo | efa7d1c | 2015-05-22 17:13:18 -0400 | [diff] [blame] | 756 | #ifdef CONFIG_BLOCK |
| 757 | |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 758 | static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; } |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 759 | static inline int blkcg_init_queue(struct request_queue *q) { return 0; } |
| 760 | static inline void blkcg_drain_queue(struct request_queue *q) { } |
| 761 | static inline void blkcg_exit_queue(struct request_queue *q) { } |
Jens Axboe | d5bf029 | 2014-06-22 16:31:56 -0600 | [diff] [blame] | 762 | static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; } |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 763 | static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { } |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 764 | static inline int blkcg_activate_policy(struct request_queue *q, |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 765 | const struct blkcg_policy *pol) { return 0; } |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 766 | static inline void blkcg_deactivate_policy(struct request_queue *q, |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 767 | const struct blkcg_policy *pol) { } |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 768 | |
Tejun Heo | b1208b5 | 2012-06-04 20:40:57 -0700 | [diff] [blame] | 769 | static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; } |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 770 | |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 771 | static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg, |
| 772 | struct blkcg_policy *pol) { return NULL; } |
| 773 | static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; } |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 774 | static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; } |
| 775 | static inline void blkg_get(struct blkcg_gq *blkg) { } |
| 776 | static inline void blkg_put(struct blkcg_gq *blkg) { } |
Vivek Goyal | afc24d4 | 2010-04-26 19:27:56 +0200 | [diff] [blame] | 777 | |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 778 | static inline struct request_list *blk_get_rl(struct request_queue *q, |
| 779 | struct bio *bio) { return &q->root_rl; } |
| 780 | static inline void blk_put_rl(struct request_list *rl) { } |
| 781 | static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { } |
| 782 | static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; } |
| 783 | |
Tejun Heo | ae11889 | 2015-08-18 14:55:20 -0700 | [diff] [blame] | 784 | static inline bool blkcg_bio_issue_check(struct request_queue *q, |
| 785 | struct bio *bio) { return true; } |
| 786 | |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 787 | #define blk_queue_for_each_rl(rl, q) \ |
| 788 | for ((rl) = &(q)->root_rl; (rl); (rl) = NULL) |
| 789 | |
Tejun Heo | efa7d1c | 2015-05-22 17:13:18 -0400 | [diff] [blame] | 790 | #endif /* CONFIG_BLOCK */ |
Tejun Heo | 36558c8 | 2012-04-16 13:57:24 -0700 | [diff] [blame] | 791 | #endif /* CONFIG_BLK_CGROUP */ |
| 792 | #endif /* _BLK_CGROUP_H */ |