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