Christoph Hellwig | 3dcf60bc | 2019-04-30 14:42:43 -0400 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 2 | /* |
| 3 | * Common Block IO controller cgroup interface |
| 4 | * |
| 5 | * Based on ideas and code from CFQ, CFS and BFQ: |
| 6 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
| 7 | * |
| 8 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> |
| 9 | * Paolo Valente <paolo.valente@unimore.it> |
| 10 | * |
| 11 | * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> |
| 12 | * Nauman Rafique <nauman@google.com> |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 13 | * |
| 14 | * For policy-specific per-blkcg data: |
| 15 | * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it> |
| 16 | * Arianna Avanzini <avanzini.arianna@gmail.com> |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 17 | */ |
| 18 | #include <linux/ioprio.h> |
Vivek Goyal | 2208419 | 2009-12-03 12:59:49 -0500 | [diff] [blame] | 19 | #include <linux/kdev_t.h> |
Vivek Goyal | 9d6a986 | 2009-12-04 10:36:41 -0500 | [diff] [blame] | 20 | #include <linux/module.h> |
Ingo Molnar | 174cd4b | 2017-02-02 19:15:33 +0100 | [diff] [blame] | 21 | #include <linux/sched/signal.h> |
Stephen Rothwell | accee78 | 2009-12-07 19:29:39 +1100 | [diff] [blame] | 22 | #include <linux/err.h> |
Divyesh Shah | 9195291 | 2010-04-01 15:01:41 -0700 | [diff] [blame] | 23 | #include <linux/blkdev.h> |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 24 | #include <linux/backing-dev.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 25 | #include <linux/slab.h> |
Gui Jianfeng | 34d0f17 | 2010-04-13 16:05:49 +0800 | [diff] [blame] | 26 | #include <linux/genhd.h> |
Tejun Heo | 72e06c2 | 2012-03-05 13:15:00 -0800 | [diff] [blame] | 27 | #include <linux/delay.h> |
Tejun Heo | 9a9e8a2 | 2012-03-19 15:10:56 -0700 | [diff] [blame] | 28 | #include <linux/atomic.h> |
Tejun Heo | 36aa9e5 | 2015-08-18 14:55:31 -0700 | [diff] [blame] | 29 | #include <linux/ctype.h> |
Tejun Heo | eea8f41 | 2015-05-22 17:13:17 -0400 | [diff] [blame] | 30 | #include <linux/blk-cgroup.h> |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 31 | #include <linux/tracehook.h> |
Josef Bacik | fd112c7 | 2019-07-09 14:41:29 -0700 | [diff] [blame] | 32 | #include <linux/psi.h> |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 33 | #include "blk.h" |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 34 | |
Tejun Heo | 838f13b | 2015-07-09 16:39:47 -0400 | [diff] [blame] | 35 | /* |
| 36 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. |
| 37 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire |
| 38 | * policy [un]register operations including cgroup file additions / |
| 39 | * removals. Putting cgroup file registration outside blkcg_pol_mutex |
| 40 | * allows grabbing it from cgroup callbacks. |
| 41 | */ |
| 42 | static DEFINE_MUTEX(blkcg_pol_register_mutex); |
Tejun Heo | bc0d650 | 2012-04-13 13:11:26 -0700 | [diff] [blame] | 43 | static DEFINE_MUTEX(blkcg_pol_mutex); |
Tejun Heo | 923adde | 2012-03-05 13:15:13 -0800 | [diff] [blame] | 44 | |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 45 | struct blkcg blkcg_root; |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 46 | EXPORT_SYMBOL_GPL(blkcg_root); |
Vivek Goyal | 9d6a986 | 2009-12-04 10:36:41 -0500 | [diff] [blame] | 47 | |
Tejun Heo | 496d5e7 | 2015-05-22 17:13:21 -0400 | [diff] [blame] | 48 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
Tejun Heo | 9b0eb69 | 2019-06-27 13:39:48 -0700 | [diff] [blame] | 49 | EXPORT_SYMBOL_GPL(blkcg_root_css); |
Tejun Heo | 496d5e7 | 2015-05-22 17:13:21 -0400 | [diff] [blame] | 50 | |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 51 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
Tejun Heo | 035d10b | 2012-03-05 13:15:04 -0800 | [diff] [blame] | 52 | |
Tejun Heo | 7876f93 | 2015-07-09 16:39:49 -0400 | [diff] [blame] | 53 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
| 54 | |
Tejun Heo | 07b0fde | 2019-07-16 07:58:31 -0700 | [diff] [blame] | 55 | bool blkcg_debug_stats = false; |
Tejun Heo | d3f77df | 2019-06-27 13:39:52 -0700 | [diff] [blame] | 56 | static struct workqueue_struct *blkcg_punt_bio_wq; |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 57 | |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 58 | static bool blkcg_policy_enabled(struct request_queue *q, |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 59 | const struct blkcg_policy *pol) |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 60 | { |
| 61 | return pol && test_bit(pol->plid, q->blkcg_pols); |
| 62 | } |
| 63 | |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 64 | /** |
| 65 | * blkg_free - free a blkg |
| 66 | * @blkg: blkg to free |
| 67 | * |
| 68 | * Free @blkg which may be partially allocated. |
| 69 | */ |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 70 | static void blkg_free(struct blkcg_gq *blkg) |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 71 | { |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 72 | int i; |
Tejun Heo | 549d3aa | 2012-03-05 13:15:16 -0800 | [diff] [blame] | 73 | |
| 74 | if (!blkg) |
| 75 | return; |
| 76 | |
Tejun Heo | db61367 | 2013-05-14 13:52:31 -0700 | [diff] [blame] | 77 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
Tejun Heo | 001bea7 | 2015-08-18 14:55:11 -0700 | [diff] [blame] | 78 | if (blkg->pd[i]) |
| 79 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 80 | |
Tejun Heo | f733164 | 2019-11-07 11:18:03 -0800 | [diff] [blame] | 81 | free_percpu(blkg->iostat_cpu); |
Tejun Heo | ef069b9 | 2019-06-13 15:30:39 -0700 | [diff] [blame] | 82 | percpu_ref_exit(&blkg->refcnt); |
Tejun Heo | 549d3aa | 2012-03-05 13:15:16 -0800 | [diff] [blame] | 83 | kfree(blkg); |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 84 | } |
| 85 | |
Dennis Zhou | 7fcf2b0 | 2018-12-05 12:10:38 -0500 | [diff] [blame] | 86 | static void __blkg_release(struct rcu_head *rcu) |
| 87 | { |
| 88 | struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head); |
| 89 | |
Tejun Heo | d3f77df | 2019-06-27 13:39:52 -0700 | [diff] [blame] | 90 | WARN_ON(!bio_list_empty(&blkg->async_bios)); |
| 91 | |
Dennis Zhou | 7fcf2b0 | 2018-12-05 12:10:38 -0500 | [diff] [blame] | 92 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
| 93 | css_put(&blkg->blkcg->css); |
| 94 | if (blkg->parent) |
| 95 | blkg_put(blkg->parent); |
Dennis Zhou | 7fcf2b0 | 2018-12-05 12:10:38 -0500 | [diff] [blame] | 96 | blkg_free(blkg); |
| 97 | } |
| 98 | |
| 99 | /* |
| 100 | * A group is RCU protected, but having an rcu lock does not mean that one |
| 101 | * can access all the fields of blkg and assume these are valid. For |
| 102 | * example, don't try to follow throtl_data and request queue links. |
| 103 | * |
| 104 | * Having a reference to blkg under an rcu allows accesses to only values |
| 105 | * local to groups like group stats and group rate limits. |
| 106 | */ |
| 107 | static void blkg_release(struct percpu_ref *ref) |
| 108 | { |
| 109 | struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt); |
| 110 | |
| 111 | call_rcu(&blkg->rcu_head, __blkg_release); |
| 112 | } |
| 113 | |
Tejun Heo | d3f77df | 2019-06-27 13:39:52 -0700 | [diff] [blame] | 114 | static void blkg_async_bio_workfn(struct work_struct *work) |
| 115 | { |
| 116 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, |
| 117 | async_bio_work); |
| 118 | struct bio_list bios = BIO_EMPTY_LIST; |
| 119 | struct bio *bio; |
Xianting Tian | 192f1c6 | 2020-09-10 14:15:06 +0800 | [diff] [blame] | 120 | struct blk_plug plug; |
| 121 | bool need_plug = false; |
Tejun Heo | d3f77df | 2019-06-27 13:39:52 -0700 | [diff] [blame] | 122 | |
| 123 | /* as long as there are pending bios, @blkg can't go away */ |
| 124 | spin_lock_bh(&blkg->async_bio_lock); |
| 125 | bio_list_merge(&bios, &blkg->async_bios); |
| 126 | bio_list_init(&blkg->async_bios); |
| 127 | spin_unlock_bh(&blkg->async_bio_lock); |
| 128 | |
Xianting Tian | 192f1c6 | 2020-09-10 14:15:06 +0800 | [diff] [blame] | 129 | /* start plug only when bio_list contains at least 2 bios */ |
| 130 | if (bios.head && bios.head->bi_next) { |
| 131 | need_plug = true; |
| 132 | blk_start_plug(&plug); |
| 133 | } |
Tejun Heo | d3f77df | 2019-06-27 13:39:52 -0700 | [diff] [blame] | 134 | while ((bio = bio_list_pop(&bios))) |
| 135 | submit_bio(bio); |
Xianting Tian | 192f1c6 | 2020-09-10 14:15:06 +0800 | [diff] [blame] | 136 | if (need_plug) |
| 137 | blk_finish_plug(&plug); |
Tejun Heo | d3f77df | 2019-06-27 13:39:52 -0700 | [diff] [blame] | 138 | } |
| 139 | |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 140 | /** |
| 141 | * blkg_alloc - allocate a blkg |
| 142 | * @blkcg: block cgroup the new blkg is associated with |
| 143 | * @q: request_queue the new blkg is associated with |
Tejun Heo | 1597499 | 2012-06-04 20:40:52 -0700 | [diff] [blame] | 144 | * @gfp_mask: allocation mask to use |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 145 | * |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 146 | * Allocate a new blkg assocating @blkcg and @q. |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 147 | */ |
Tejun Heo | 1597499 | 2012-06-04 20:40:52 -0700 | [diff] [blame] | 148 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
| 149 | gfp_t gfp_mask) |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 150 | { |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 151 | struct blkcg_gq *blkg; |
Tejun Heo | f733164 | 2019-11-07 11:18:03 -0800 | [diff] [blame] | 152 | int i, cpu; |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 153 | |
| 154 | /* alloc and init base part */ |
Tejun Heo | 1597499 | 2012-06-04 20:40:52 -0700 | [diff] [blame] | 155 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 156 | if (!blkg) |
| 157 | return NULL; |
| 158 | |
Tejun Heo | ef069b9 | 2019-06-13 15:30:39 -0700 | [diff] [blame] | 159 | if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask)) |
| 160 | goto err_free; |
| 161 | |
Tejun Heo | f733164 | 2019-11-07 11:18:03 -0800 | [diff] [blame] | 162 | blkg->iostat_cpu = alloc_percpu_gfp(struct blkg_iostat_set, gfp_mask); |
| 163 | if (!blkg->iostat_cpu) |
Tejun Heo | 77ea733 | 2015-08-18 14:55:24 -0700 | [diff] [blame] | 164 | goto err_free; |
| 165 | |
Tejun Heo | c875f4d | 2012-03-05 13:15:22 -0800 | [diff] [blame] | 166 | blkg->q = q; |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 167 | INIT_LIST_HEAD(&blkg->q_node); |
Tejun Heo | d3f77df | 2019-06-27 13:39:52 -0700 | [diff] [blame] | 168 | spin_lock_init(&blkg->async_bio_lock); |
| 169 | bio_list_init(&blkg->async_bios); |
| 170 | INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn); |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 171 | blkg->blkcg = blkcg; |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 172 | |
Tejun Heo | f733164 | 2019-11-07 11:18:03 -0800 | [diff] [blame] | 173 | u64_stats_init(&blkg->iostat.sync); |
| 174 | for_each_possible_cpu(cpu) |
| 175 | u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync); |
| 176 | |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 177 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 178 | struct blkcg_policy *pol = blkcg_policy[i]; |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 179 | struct blkg_policy_data *pd; |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 180 | |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 181 | if (!blkcg_policy_enabled(q, pol)) |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 182 | continue; |
Tejun Heo | 549d3aa | 2012-03-05 13:15:16 -0800 | [diff] [blame] | 183 | |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 184 | /* alloc per-policy data and attach it to blkg */ |
Tejun Heo | cf09a8e | 2019-08-28 15:05:51 -0700 | [diff] [blame] | 185 | pd = pol->pd_alloc_fn(gfp_mask, q, blkcg); |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 186 | if (!pd) |
| 187 | goto err_free; |
Tejun Heo | 549d3aa | 2012-03-05 13:15:16 -0800 | [diff] [blame] | 188 | |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 189 | blkg->pd[i] = pd; |
| 190 | pd->blkg = blkg; |
Tejun Heo | b276a87 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 191 | pd->plid = i; |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 192 | } |
| 193 | |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 194 | return blkg; |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 195 | |
| 196 | err_free: |
| 197 | blkg_free(blkg); |
| 198 | return NULL; |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 199 | } |
| 200 | |
Tejun Heo | 24f2904 | 2015-08-18 14:55:17 -0700 | [diff] [blame] | 201 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
| 202 | struct request_queue *q, bool update_hint) |
Tejun Heo | 80fd997 | 2012-04-13 14:50:53 -0700 | [diff] [blame] | 203 | { |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 204 | struct blkcg_gq *blkg; |
Tejun Heo | 80fd997 | 2012-04-13 14:50:53 -0700 | [diff] [blame] | 205 | |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 206 | /* |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 207 | * Hint didn't match. Look up from the radix tree. Note that the |
| 208 | * hint can only be updated under queue_lock as otherwise @blkg |
| 209 | * could have already been removed from blkg_tree. The caller is |
| 210 | * responsible for grabbing queue_lock if @update_hint. |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 211 | */ |
| 212 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 213 | if (blkg && blkg->q == q) { |
| 214 | if (update_hint) { |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 215 | lockdep_assert_held(&q->queue_lock); |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 216 | rcu_assign_pointer(blkcg->blkg_hint, blkg); |
| 217 | } |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 218 | return blkg; |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 219 | } |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 220 | |
Tejun Heo | 80fd997 | 2012-04-13 14:50:53 -0700 | [diff] [blame] | 221 | return NULL; |
| 222 | } |
Tejun Heo | ae11889 | 2015-08-18 14:55:20 -0700 | [diff] [blame] | 223 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); |
Tejun Heo | 80fd997 | 2012-04-13 14:50:53 -0700 | [diff] [blame] | 224 | |
Tejun Heo | 1597499 | 2012-06-04 20:40:52 -0700 | [diff] [blame] | 225 | /* |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 226 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
| 227 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. |
Tejun Heo | 1597499 | 2012-06-04 20:40:52 -0700 | [diff] [blame] | 228 | */ |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 229 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 230 | struct request_queue *q, |
| 231 | struct blkcg_gq *new_blkg) |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 232 | { |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 233 | struct blkcg_gq *blkg; |
Tejun Heo | f427d90 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 234 | int i, ret; |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 235 | |
Tejun Heo | cd1604f | 2012-03-05 13:15:06 -0800 | [diff] [blame] | 236 | WARN_ON_ONCE(!rcu_read_lock_held()); |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 237 | lockdep_assert_held(&q->queue_lock); |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 238 | |
Dennis Zhou | 0273ac3 | 2018-12-11 18:03:08 -0500 | [diff] [blame] | 239 | /* request_queue is dying, do not create/recreate a blkg */ |
| 240 | if (blk_queue_dying(q)) { |
| 241 | ret = -ENODEV; |
| 242 | goto err_free_blkg; |
| 243 | } |
| 244 | |
Tejun Heo | 7ee9c56 | 2012-03-05 13:15:11 -0800 | [diff] [blame] | 245 | /* blkg holds a reference to blkcg */ |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 246 | if (!css_tryget_online(&blkcg->css)) { |
Tejun Heo | 20386ce | 2015-08-18 14:55:28 -0700 | [diff] [blame] | 247 | ret = -ENODEV; |
Tejun Heo | 93e6d5d | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 248 | goto err_free_blkg; |
Tejun Heo | 1597499 | 2012-06-04 20:40:52 -0700 | [diff] [blame] | 249 | } |
Tejun Heo | cd1604f | 2012-03-05 13:15:06 -0800 | [diff] [blame] | 250 | |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 251 | /* allocate */ |
| 252 | if (!new_blkg) { |
| 253 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN); |
| 254 | if (unlikely(!new_blkg)) { |
| 255 | ret = -ENOMEM; |
Christoph Hellwig | 8c911f3 | 2020-07-01 11:06:21 +0200 | [diff] [blame] | 256 | goto err_put_css; |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 257 | } |
| 258 | } |
| 259 | blkg = new_blkg; |
Tahsin Erdogan | 7fc6b87 | 2017-03-09 00:05:31 -0800 | [diff] [blame] | 260 | |
Tejun Heo | db61367 | 2013-05-14 13:52:31 -0700 | [diff] [blame] | 261 | /* link parent */ |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 262 | if (blkcg_parent(blkcg)) { |
| 263 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); |
| 264 | if (WARN_ON_ONCE(!blkg->parent)) { |
Tejun Heo | 20386ce | 2015-08-18 14:55:28 -0700 | [diff] [blame] | 265 | ret = -ENODEV; |
Christoph Hellwig | 8c911f3 | 2020-07-01 11:06:21 +0200 | [diff] [blame] | 266 | goto err_put_css; |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 267 | } |
| 268 | blkg_get(blkg->parent); |
| 269 | } |
| 270 | |
Tejun Heo | db61367 | 2013-05-14 13:52:31 -0700 | [diff] [blame] | 271 | /* invoke per-policy init */ |
| 272 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
| 273 | struct blkcg_policy *pol = blkcg_policy[i]; |
| 274 | |
| 275 | if (blkg->pd[i] && pol->pd_init_fn) |
Tejun Heo | a9520cd | 2015-08-18 14:55:14 -0700 | [diff] [blame] | 276 | pol->pd_init_fn(blkg->pd[i]); |
Tejun Heo | db61367 | 2013-05-14 13:52:31 -0700 | [diff] [blame] | 277 | } |
| 278 | |
| 279 | /* insert */ |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 280 | spin_lock(&blkcg->lock); |
| 281 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
| 282 | if (likely(!ret)) { |
| 283 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
| 284 | list_add(&blkg->q_node, &q->blkg_list); |
Tejun Heo | f427d90 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 285 | |
| 286 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
| 287 | struct blkcg_policy *pol = blkcg_policy[i]; |
| 288 | |
| 289 | if (blkg->pd[i] && pol->pd_online_fn) |
Tejun Heo | a9520cd | 2015-08-18 14:55:14 -0700 | [diff] [blame] | 290 | pol->pd_online_fn(blkg->pd[i]); |
Tejun Heo | f427d90 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 291 | } |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 292 | } |
Tejun Heo | f427d90 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 293 | blkg->online = true; |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 294 | spin_unlock(&blkcg->lock); |
| 295 | |
Tejun Heo | ec13b1d | 2015-05-22 17:13:19 -0400 | [diff] [blame] | 296 | if (!ret) |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 297 | return blkg; |
Tejun Heo | 1597499 | 2012-06-04 20:40:52 -0700 | [diff] [blame] | 298 | |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 299 | /* @blkg failed fully initialized, use the usual release path */ |
| 300 | blkg_put(blkg); |
| 301 | return ERR_PTR(ret); |
| 302 | |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 303 | err_put_css: |
Tejun Heo | 496fb78 | 2012-04-19 16:29:23 -0700 | [diff] [blame] | 304 | css_put(&blkcg->css); |
Tejun Heo | 93e6d5d | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 305 | err_free_blkg: |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 306 | blkg_free(new_blkg); |
Tejun Heo | 93e6d5d | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 307 | return ERR_PTR(ret); |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 308 | } |
Tejun Heo | 3c96cb3 | 2012-04-13 13:11:34 -0700 | [diff] [blame] | 309 | |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 310 | /** |
Christoph Hellwig | 8c54628 | 2020-06-27 09:31:52 +0200 | [diff] [blame] | 311 | * blkg_lookup_create - lookup blkg, try to create one if not there |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 312 | * @blkcg: blkcg of interest |
| 313 | * @q: request_queue of interest |
| 314 | * |
| 315 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 316 | * create one. blkg creation is performed recursively from blkcg_root such |
| 317 | * that all non-root blkg's have access to the parent blkg. This function |
Christoph Hellwig | 8c54628 | 2020-06-27 09:31:52 +0200 | [diff] [blame] | 318 | * should be called under RCU read lock and takes @q->queue_lock. |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 319 | * |
Dennis Zhou | beea9da | 2018-12-05 12:10:28 -0500 | [diff] [blame] | 320 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
| 321 | * down from root. |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 322 | */ |
Christoph Hellwig | 8c54628 | 2020-06-27 09:31:52 +0200 | [diff] [blame] | 323 | static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
| 324 | struct request_queue *q) |
Tejun Heo | 3c96cb3 | 2012-04-13 13:11:34 -0700 | [diff] [blame] | 325 | { |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 326 | struct blkcg_gq *blkg; |
Christoph Hellwig | 8c54628 | 2020-06-27 09:31:52 +0200 | [diff] [blame] | 327 | unsigned long flags; |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 328 | |
| 329 | WARN_ON_ONCE(!rcu_read_lock_held()); |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 330 | |
Christoph Hellwig | 8c54628 | 2020-06-27 09:31:52 +0200 | [diff] [blame] | 331 | blkg = blkg_lookup(blkcg, q); |
Tejun Heo | 86cde6b | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 332 | if (blkg) |
| 333 | return blkg; |
| 334 | |
Christoph Hellwig | 8c54628 | 2020-06-27 09:31:52 +0200 | [diff] [blame] | 335 | spin_lock_irqsave(&q->queue_lock, flags); |
| 336 | blkg = __blkg_lookup(blkcg, q, true); |
| 337 | if (blkg) |
| 338 | goto found; |
| 339 | |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 340 | /* |
| 341 | * Create blkgs walking down from blkcg_root to @blkcg, so that all |
Dennis Zhou | beea9da | 2018-12-05 12:10:28 -0500 | [diff] [blame] | 342 | * non-root blkgs have access to their parents. Returns the closest |
| 343 | * blkg to the intended blkg should blkg_create() fail. |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 344 | */ |
| 345 | while (true) { |
| 346 | struct blkcg *pos = blkcg; |
| 347 | struct blkcg *parent = blkcg_parent(blkcg); |
Dennis Zhou | beea9da | 2018-12-05 12:10:28 -0500 | [diff] [blame] | 348 | struct blkcg_gq *ret_blkg = q->root_blkg; |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 349 | |
Dennis Zhou | beea9da | 2018-12-05 12:10:28 -0500 | [diff] [blame] | 350 | while (parent) { |
| 351 | blkg = __blkg_lookup(parent, q, false); |
| 352 | if (blkg) { |
| 353 | /* remember closest blkg */ |
| 354 | ret_blkg = blkg; |
| 355 | break; |
| 356 | } |
Tejun Heo | 3c54786 | 2013-01-09 08:05:10 -0800 | [diff] [blame] | 357 | pos = parent; |
| 358 | parent = blkcg_parent(parent); |
| 359 | } |
| 360 | |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 361 | blkg = blkg_create(pos, q, NULL); |
Christoph Hellwig | 8c54628 | 2020-06-27 09:31:52 +0200 | [diff] [blame] | 362 | if (IS_ERR(blkg)) { |
| 363 | blkg = ret_blkg; |
| 364 | break; |
| 365 | } |
Dennis Zhou | beea9da | 2018-12-05 12:10:28 -0500 | [diff] [blame] | 366 | if (pos == blkcg) |
Christoph Hellwig | 8c54628 | 2020-06-27 09:31:52 +0200 | [diff] [blame] | 367 | break; |
Dennis Zhou | b978962 | 2018-12-05 12:10:27 -0500 | [diff] [blame] | 368 | } |
| 369 | |
Christoph Hellwig | 8c54628 | 2020-06-27 09:31:52 +0200 | [diff] [blame] | 370 | found: |
| 371 | spin_unlock_irqrestore(&q->queue_lock, flags); |
Dennis Zhou | b978962 | 2018-12-05 12:10:27 -0500 | [diff] [blame] | 372 | return blkg; |
| 373 | } |
| 374 | |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 375 | static void blkg_destroy(struct blkcg_gq *blkg) |
Tejun Heo | 72e06c2 | 2012-03-05 13:15:00 -0800 | [diff] [blame] | 376 | { |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 377 | struct blkcg *blkcg = blkg->blkcg; |
Dennis Zhou (Facebook) | 6b06546 | 2018-08-31 16:22:42 -0400 | [diff] [blame] | 378 | int i; |
Tejun Heo | 03aa264 | 2012-03-05 13:15:19 -0800 | [diff] [blame] | 379 | |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 380 | lockdep_assert_held(&blkg->q->queue_lock); |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 381 | lockdep_assert_held(&blkcg->lock); |
Tejun Heo | 03aa264 | 2012-03-05 13:15:19 -0800 | [diff] [blame] | 382 | |
| 383 | /* Something wrong if we are trying to remove same group twice */ |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 384 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 385 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 386 | |
Dennis Zhou (Facebook) | 6b06546 | 2018-08-31 16:22:42 -0400 | [diff] [blame] | 387 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
| 388 | struct blkcg_policy *pol = blkcg_policy[i]; |
| 389 | |
| 390 | if (blkg->pd[i] && pol->pd_offline_fn) |
| 391 | pol->pd_offline_fn(blkg->pd[i]); |
| 392 | } |
| 393 | |
Tejun Heo | f427d90 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 394 | blkg->online = false; |
| 395 | |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 396 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 397 | list_del_init(&blkg->q_node); |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 398 | hlist_del_init_rcu(&blkg->blkcg_node); |
Tejun Heo | 03aa264 | 2012-03-05 13:15:19 -0800 | [diff] [blame] | 399 | |
Tejun Heo | 03aa264 | 2012-03-05 13:15:19 -0800 | [diff] [blame] | 400 | /* |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 401 | * Both setting lookup hint to and clearing it from @blkg are done |
| 402 | * under queue_lock. If it's not pointing to @blkg now, it never |
| 403 | * will. Hint assignment itself can race safely. |
| 404 | */ |
Paul E. McKenney | ec6c676a | 2014-02-17 13:35:57 -0800 | [diff] [blame] | 405 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
Tejun Heo | a637120 | 2012-04-19 16:29:24 -0700 | [diff] [blame] | 406 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
| 407 | |
| 408 | /* |
Tejun Heo | 03aa264 | 2012-03-05 13:15:19 -0800 | [diff] [blame] | 409 | * Put the reference taken at the time of creation so that when all |
| 410 | * queues are gone, group can be destroyed. |
| 411 | */ |
Dennis Zhou | 7fcf2b0 | 2018-12-05 12:10:38 -0500 | [diff] [blame] | 412 | percpu_ref_kill(&blkg->refcnt); |
Tejun Heo | 03aa264 | 2012-03-05 13:15:19 -0800 | [diff] [blame] | 413 | } |
| 414 | |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 415 | /** |
| 416 | * blkg_destroy_all - destroy all blkgs associated with a request_queue |
| 417 | * @q: request_queue of interest |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 418 | * |
Tejun Heo | 3c96cb3 | 2012-04-13 13:11:34 -0700 | [diff] [blame] | 419 | * Destroy all blkgs associated with @q. |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 420 | */ |
Tejun Heo | 3c96cb3 | 2012-04-13 13:11:34 -0700 | [diff] [blame] | 421 | static void blkg_destroy_all(struct request_queue *q) |
Tejun Heo | 03aa264 | 2012-03-05 13:15:19 -0800 | [diff] [blame] | 422 | { |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 423 | struct blkcg_gq *blkg, *n; |
Tejun Heo | 72e06c2 | 2012-03-05 13:15:00 -0800 | [diff] [blame] | 424 | |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 425 | spin_lock_irq(&q->queue_lock); |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 426 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 427 | struct blkcg *blkcg = blkg->blkcg; |
Tejun Heo | 72e06c2 | 2012-03-05 13:15:00 -0800 | [diff] [blame] | 428 | |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 429 | spin_lock(&blkcg->lock); |
| 430 | blkg_destroy(blkg); |
| 431 | spin_unlock(&blkcg->lock); |
Tejun Heo | 72e06c2 | 2012-03-05 13:15:00 -0800 | [diff] [blame] | 432 | } |
Tejun Heo | 6fe810b | 2015-09-05 15:47:36 -0400 | [diff] [blame] | 433 | |
| 434 | q->root_blkg = NULL; |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 435 | spin_unlock_irq(&q->queue_lock); |
Tejun Heo | 72e06c2 | 2012-03-05 13:15:00 -0800 | [diff] [blame] | 436 | } |
| 437 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 438 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
| 439 | struct cftype *cftype, u64 val) |
Divyesh Shah | 303a3ac | 2010-04-01 15:01:24 -0700 | [diff] [blame] | 440 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 441 | struct blkcg *blkcg = css_to_blkcg(css); |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 442 | struct blkcg_gq *blkg; |
Tejun Heo | f733164 | 2019-11-07 11:18:03 -0800 | [diff] [blame] | 443 | int i, cpu; |
Divyesh Shah | 303a3ac | 2010-04-01 15:01:24 -0700 | [diff] [blame] | 444 | |
Tejun Heo | 838f13b | 2015-07-09 16:39:47 -0400 | [diff] [blame] | 445 | mutex_lock(&blkcg_pol_mutex); |
Divyesh Shah | 303a3ac | 2010-04-01 15:01:24 -0700 | [diff] [blame] | 446 | spin_lock_irq(&blkcg->lock); |
Tejun Heo | 997a026 | 2012-03-08 10:53:58 -0800 | [diff] [blame] | 447 | |
| 448 | /* |
| 449 | * Note that stat reset is racy - it doesn't synchronize against |
| 450 | * stat updates. This is a debug feature which shouldn't exist |
| 451 | * anyway. If you get hit by a race, retry. |
| 452 | */ |
Sasha Levin | b67bfe0 | 2013-02-27 17:06:00 -0800 | [diff] [blame] | 453 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
Tejun Heo | f733164 | 2019-11-07 11:18:03 -0800 | [diff] [blame] | 454 | for_each_possible_cpu(cpu) { |
| 455 | struct blkg_iostat_set *bis = |
| 456 | per_cpu_ptr(blkg->iostat_cpu, cpu); |
| 457 | memset(bis, 0, sizeof(*bis)); |
| 458 | } |
| 459 | memset(&blkg->iostat, 0, sizeof(blkg->iostat)); |
Tejun Heo | 77ea733 | 2015-08-18 14:55:24 -0700 | [diff] [blame] | 460 | |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 461 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 462 | struct blkcg_policy *pol = blkcg_policy[i]; |
Tejun Heo | 549d3aa | 2012-03-05 13:15:16 -0800 | [diff] [blame] | 463 | |
Tejun Heo | a9520cd | 2015-08-18 14:55:14 -0700 | [diff] [blame] | 464 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
| 465 | pol->pd_reset_stats_fn(blkg->pd[i]); |
Tejun Heo | bc0d650 | 2012-04-13 13:11:26 -0700 | [diff] [blame] | 466 | } |
Divyesh Shah | 303a3ac | 2010-04-01 15:01:24 -0700 | [diff] [blame] | 467 | } |
Vivek Goyal | f0bdc8c | 2011-05-19 15:38:30 -0400 | [diff] [blame] | 468 | |
Divyesh Shah | 303a3ac | 2010-04-01 15:01:24 -0700 | [diff] [blame] | 469 | spin_unlock_irq(&blkcg->lock); |
Tejun Heo | bc0d650 | 2012-04-13 13:11:26 -0700 | [diff] [blame] | 470 | mutex_unlock(&blkcg_pol_mutex); |
Divyesh Shah | 303a3ac | 2010-04-01 15:01:24 -0700 | [diff] [blame] | 471 | return 0; |
| 472 | } |
| 473 | |
Tejun Heo | dd165eb | 2015-08-18 14:55:33 -0700 | [diff] [blame] | 474 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
Divyesh Shah | 303a3ac | 2010-04-01 15:01:24 -0700 | [diff] [blame] | 475 | { |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 476 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
Jan Kara | dc3b17c | 2017-02-02 15:56:50 +0100 | [diff] [blame] | 477 | if (blkg->q->backing_dev_info->dev) |
Yufen Yu | d51cfc5 | 2020-05-04 14:47:55 +0200 | [diff] [blame] | 478 | return bdi_dev_name(blkg->q->backing_dev_info); |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 479 | return NULL; |
Divyesh Shah | 303a3ac | 2010-04-01 15:01:24 -0700 | [diff] [blame] | 480 | } |
| 481 | |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 482 | /** |
| 483 | * blkcg_print_blkgs - helper for printing per-blkg data |
| 484 | * @sf: seq_file to print to |
| 485 | * @blkcg: blkcg of interest |
| 486 | * @prfill: fill function to print out a blkg |
| 487 | * @pol: policy in question |
| 488 | * @data: data to be passed to @prfill |
| 489 | * @show_total: to print out sum of prfill return values or not |
| 490 | * |
| 491 | * This function invokes @prfill on each blkg of @blkcg if pd for the |
| 492 | * policy specified by @pol exists. @prfill is invoked with @sf, the |
Tejun Heo | 810ecfa | 2013-01-09 08:05:13 -0800 | [diff] [blame] | 493 | * policy data and @data and the matching queue lock held. If @show_total |
| 494 | * is %true, the sum of the return values from @prfill is printed with |
| 495 | * "Total" label at the end. |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 496 | * |
| 497 | * This is to be used to construct print functions for |
| 498 | * cftype->read_seq_string method. |
| 499 | */ |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 500 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 501 | u64 (*prfill)(struct seq_file *, |
| 502 | struct blkg_policy_data *, int), |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 503 | const struct blkcg_policy *pol, int data, |
Tejun Heo | ec39934 | 2012-04-13 13:11:27 -0700 | [diff] [blame] | 504 | bool show_total) |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 505 | { |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 506 | struct blkcg_gq *blkg; |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 507 | u64 total = 0; |
| 508 | |
Tejun Heo | 810ecfa | 2013-01-09 08:05:13 -0800 | [diff] [blame] | 509 | rcu_read_lock(); |
Linus Torvalds | ee89f81 | 2013-02-28 12:52:24 -0800 | [diff] [blame] | 510 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 511 | spin_lock_irq(&blkg->q->queue_lock); |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 512 | if (blkcg_policy_enabled(blkg->q, pol)) |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 513 | total += prfill(sf, blkg->pd[pol->plid], data); |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 514 | spin_unlock_irq(&blkg->q->queue_lock); |
Tejun Heo | 810ecfa | 2013-01-09 08:05:13 -0800 | [diff] [blame] | 515 | } |
| 516 | rcu_read_unlock(); |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 517 | |
| 518 | if (show_total) |
| 519 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); |
| 520 | } |
Tejun Heo | 829fdb5 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 521 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 522 | |
| 523 | /** |
| 524 | * __blkg_prfill_u64 - prfill helper for a single u64 value |
| 525 | * @sf: seq_file to print to |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 526 | * @pd: policy private data of interest |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 527 | * @v: value to print |
| 528 | * |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 529 | * Print @v to @sf for the device assocaited with @pd. |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 530 | */ |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 531 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 532 | { |
Tejun Heo | f95a04a | 2012-04-16 13:57:26 -0700 | [diff] [blame] | 533 | const char *dname = blkg_dev_name(pd->blkg); |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 534 | |
| 535 | if (!dname) |
| 536 | return 0; |
| 537 | |
| 538 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); |
| 539 | return v; |
| 540 | } |
Tejun Heo | 829fdb5 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 541 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
Tejun Heo | d3d32e6 | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 542 | |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 543 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ |
| 544 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, |
| 545 | const struct blkcg_policy *pol, |
| 546 | struct request_queue *q) |
| 547 | { |
| 548 | WARN_ON_ONCE(!rcu_read_lock_held()); |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 549 | lockdep_assert_held(&q->queue_lock); |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 550 | |
| 551 | if (!blkcg_policy_enabled(q, pol)) |
| 552 | return ERR_PTR(-EOPNOTSUPP); |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 553 | return __blkg_lookup(blkcg, q, true /* update_hint */); |
| 554 | } |
| 555 | |
Tejun Heo | 16b3de6 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 556 | /** |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 557 | * blkcg_conf_open_bdev - parse and open bdev for per-blkg config update |
Tejun Heo | 015d254 | 2019-08-28 15:05:53 -0700 | [diff] [blame] | 558 | * @inputp: input string pointer |
| 559 | * |
| 560 | * Parse the device node prefix part, MAJ:MIN, of per-blkg config update |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 561 | * from @input and get and return the matching bdev. *@inputp is |
Tejun Heo | 015d254 | 2019-08-28 15:05:53 -0700 | [diff] [blame] | 562 | * updated to point past the device node prefix. Returns an ERR_PTR() |
| 563 | * value on error. |
| 564 | * |
| 565 | * Use this function iff blkg_conf_prep() can't be used for some reason. |
| 566 | */ |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 567 | struct block_device *blkcg_conf_open_bdev(char **inputp) |
Tejun Heo | 015d254 | 2019-08-28 15:05:53 -0700 | [diff] [blame] | 568 | { |
| 569 | char *input = *inputp; |
| 570 | unsigned int major, minor; |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 571 | struct block_device *bdev; |
| 572 | int key_len; |
Tejun Heo | 015d254 | 2019-08-28 15:05:53 -0700 | [diff] [blame] | 573 | |
| 574 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) |
| 575 | return ERR_PTR(-EINVAL); |
| 576 | |
| 577 | input += key_len; |
| 578 | if (!isspace(*input)) |
| 579 | return ERR_PTR(-EINVAL); |
| 580 | input = skip_spaces(input); |
| 581 | |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 582 | bdev = blkdev_get_no_open(MKDEV(major, minor)); |
| 583 | if (!bdev) |
Tejun Heo | 015d254 | 2019-08-28 15:05:53 -0700 | [diff] [blame] | 584 | return ERR_PTR(-ENODEV); |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 585 | if (bdev_is_partition(bdev)) { |
| 586 | blkdev_put_no_open(bdev); |
Tejun Heo | 015d254 | 2019-08-28 15:05:53 -0700 | [diff] [blame] | 587 | return ERR_PTR(-ENODEV); |
| 588 | } |
| 589 | |
| 590 | *inputp = input; |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 591 | return bdev; |
Tejun Heo | 015d254 | 2019-08-28 15:05:53 -0700 | [diff] [blame] | 592 | } |
| 593 | |
| 594 | /** |
| 595 | * blkg_conf_prep - parse and prepare for per-blkg config update |
Tejun Heo | 3a8b31d | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 596 | * @blkcg: target block cgroup |
Tejun Heo | da8b066 | 2012-04-13 13:11:29 -0700 | [diff] [blame] | 597 | * @pol: target policy |
Tejun Heo | 3a8b31d | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 598 | * @input: input string |
| 599 | * @ctx: blkg_conf_ctx to be filled |
| 600 | * |
| 601 | * Parse per-blkg config update from @input and initialize @ctx with the |
Tejun Heo | 36aa9e5 | 2015-08-18 14:55:31 -0700 | [diff] [blame] | 602 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
| 603 | * part of @input following MAJ:MIN. This function returns with RCU read |
| 604 | * lock and queue lock held and must be paired with blkg_conf_finish(). |
Tejun Heo | 3a8b31d | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 605 | */ |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 606 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
Tejun Heo | 36aa9e5 | 2015-08-18 14:55:31 -0700 | [diff] [blame] | 607 | char *input, struct blkg_conf_ctx *ctx) |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 608 | __acquires(rcu) __acquires(&bdev->bd_disk->queue->queue_lock) |
Gui Jianfeng | 34d0f17 | 2010-04-13 16:05:49 +0800 | [diff] [blame] | 609 | { |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 610 | struct block_device *bdev; |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 611 | struct request_queue *q; |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 612 | struct blkcg_gq *blkg; |
Tejun Heo | 015d254 | 2019-08-28 15:05:53 -0700 | [diff] [blame] | 613 | int ret; |
Gui Jianfeng | 34d0f17 | 2010-04-13 16:05:49 +0800 | [diff] [blame] | 614 | |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 615 | bdev = blkcg_conf_open_bdev(&input); |
| 616 | if (IS_ERR(bdev)) |
| 617 | return PTR_ERR(bdev); |
Tejun Heo | e56da7e | 2012-03-05 13:15:07 -0800 | [diff] [blame] | 618 | |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 619 | q = bdev->bd_disk->queue; |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 620 | |
Tejun Heo | e56da7e | 2012-03-05 13:15:07 -0800 | [diff] [blame] | 621 | rcu_read_lock(); |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 622 | spin_lock_irq(&q->queue_lock); |
Tejun Heo | da8b066 | 2012-04-13 13:11:29 -0700 | [diff] [blame] | 623 | |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 624 | blkg = blkg_lookup_check(blkcg, pol, q); |
Tejun Heo | 4bfd482 | 2012-03-05 13:15:08 -0800 | [diff] [blame] | 625 | if (IS_ERR(blkg)) { |
| 626 | ret = PTR_ERR(blkg); |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 627 | goto fail_unlock; |
Vivek Goyal | 062a644 | 2010-09-15 17:06:33 -0400 | [diff] [blame] | 628 | } |
Tejun Heo | e56da7e | 2012-03-05 13:15:07 -0800 | [diff] [blame] | 629 | |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 630 | if (blkg) |
| 631 | goto success; |
| 632 | |
| 633 | /* |
| 634 | * Create blkgs walking down from blkcg_root to @blkcg, so that all |
| 635 | * non-root blkgs have access to their parents. |
| 636 | */ |
| 637 | while (true) { |
| 638 | struct blkcg *pos = blkcg; |
| 639 | struct blkcg *parent; |
| 640 | struct blkcg_gq *new_blkg; |
| 641 | |
| 642 | parent = blkcg_parent(blkcg); |
| 643 | while (parent && !__blkg_lookup(parent, q, false)) { |
| 644 | pos = parent; |
| 645 | parent = blkcg_parent(parent); |
| 646 | } |
| 647 | |
| 648 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 649 | spin_unlock_irq(&q->queue_lock); |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 650 | rcu_read_unlock(); |
| 651 | |
| 652 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); |
| 653 | if (unlikely(!new_blkg)) { |
| 654 | ret = -ENOMEM; |
| 655 | goto fail; |
| 656 | } |
| 657 | |
Gabriel Krisman Bertazi | f255c19 | 2020-10-22 16:58:42 -0400 | [diff] [blame] | 658 | if (radix_tree_preload(GFP_KERNEL)) { |
| 659 | blkg_free(new_blkg); |
| 660 | ret = -ENOMEM; |
| 661 | goto fail; |
| 662 | } |
| 663 | |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 664 | rcu_read_lock(); |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 665 | spin_lock_irq(&q->queue_lock); |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 666 | |
| 667 | blkg = blkg_lookup_check(pos, pol, q); |
| 668 | if (IS_ERR(blkg)) { |
| 669 | ret = PTR_ERR(blkg); |
Gabriel Krisman Bertazi | 52abfcb | 2020-10-22 16:58:41 -0400 | [diff] [blame] | 670 | blkg_free(new_blkg); |
Gabriel Krisman Bertazi | f255c19 | 2020-10-22 16:58:42 -0400 | [diff] [blame] | 671 | goto fail_preloaded; |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 672 | } |
| 673 | |
| 674 | if (blkg) { |
| 675 | blkg_free(new_blkg); |
| 676 | } else { |
| 677 | blkg = blkg_create(pos, q, new_blkg); |
Kefeng Wang | 98d669b | 2019-06-05 22:24:27 +0800 | [diff] [blame] | 678 | if (IS_ERR(blkg)) { |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 679 | ret = PTR_ERR(blkg); |
Gabriel Krisman Bertazi | f255c19 | 2020-10-22 16:58:42 -0400 | [diff] [blame] | 680 | goto fail_preloaded; |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 681 | } |
| 682 | } |
| 683 | |
Gabriel Krisman Bertazi | f255c19 | 2020-10-22 16:58:42 -0400 | [diff] [blame] | 684 | radix_tree_preload_end(); |
| 685 | |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 686 | if (pos == blkcg) |
| 687 | goto success; |
| 688 | } |
| 689 | success: |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 690 | ctx->bdev = bdev; |
Tejun Heo | 3a8b31d | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 691 | ctx->blkg = blkg; |
Tejun Heo | 015d254 | 2019-08-28 15:05:53 -0700 | [diff] [blame] | 692 | ctx->body = input; |
Tejun Heo | 726fa694 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 693 | return 0; |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 694 | |
Gabriel Krisman Bertazi | f255c19 | 2020-10-22 16:58:42 -0400 | [diff] [blame] | 695 | fail_preloaded: |
| 696 | radix_tree_preload_end(); |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 697 | fail_unlock: |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 698 | spin_unlock_irq(&q->queue_lock); |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 699 | rcu_read_unlock(); |
| 700 | fail: |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 701 | blkdev_put_no_open(bdev); |
Tahsin Erdogan | 457e490f | 2017-03-29 11:27:19 -0600 | [diff] [blame] | 702 | /* |
| 703 | * If queue was bypassing, we should retry. Do so after a |
| 704 | * short msleep(). It isn't strictly necessary but queue |
| 705 | * can be bypassing for some time and it's always nice to |
| 706 | * avoid busy looping. |
| 707 | */ |
| 708 | if (ret == -EBUSY) { |
| 709 | msleep(10); |
| 710 | ret = restart_syscall(); |
| 711 | } |
| 712 | return ret; |
Gui Jianfeng | 34d0f17 | 2010-04-13 16:05:49 +0800 | [diff] [blame] | 713 | } |
Pavel Begunkov | 89f3b6d | 2019-09-14 20:31:50 +0300 | [diff] [blame] | 714 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
Gui Jianfeng | 34d0f17 | 2010-04-13 16:05:49 +0800 | [diff] [blame] | 715 | |
Tejun Heo | 3a8b31d | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 716 | /** |
| 717 | * blkg_conf_finish - finish up per-blkg config update |
| 718 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() |
| 719 | * |
| 720 | * Finish up after per-blkg config update. This function must be paired |
| 721 | * with blkg_conf_prep(). |
| 722 | */ |
Tejun Heo | 829fdb5 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 723 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 724 | __releases(&ctx->bdev->bd_disk->queue->queue_lock) __releases(rcu) |
Gui Jianfeng | 34d0f17 | 2010-04-13 16:05:49 +0800 | [diff] [blame] | 725 | { |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 726 | spin_unlock_irq(&ctx->bdev->bd_disk->queue->queue_lock); |
Tejun Heo | 3a8b31d | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 727 | rcu_read_unlock(); |
Christoph Hellwig | 22ae8ce | 2020-11-26 09:23:26 +0100 | [diff] [blame] | 728 | blkdev_put_no_open(ctx->bdev); |
Gui Jianfeng | 34d0f17 | 2010-04-13 16:05:49 +0800 | [diff] [blame] | 729 | } |
Pavel Begunkov | 89f3b6d | 2019-09-14 20:31:50 +0300 | [diff] [blame] | 730 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
Gui Jianfeng | 34d0f17 | 2010-04-13 16:05:49 +0800 | [diff] [blame] | 731 | |
Boris Burkov | cd1fc4b | 2020-06-01 13:11:43 -0700 | [diff] [blame] | 732 | static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src) |
| 733 | { |
| 734 | int i; |
| 735 | |
| 736 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { |
| 737 | dst->bytes[i] = src->bytes[i]; |
| 738 | dst->ios[i] = src->ios[i]; |
| 739 | } |
| 740 | } |
| 741 | |
| 742 | static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src) |
| 743 | { |
| 744 | int i; |
| 745 | |
| 746 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { |
| 747 | dst->bytes[i] += src->bytes[i]; |
| 748 | dst->ios[i] += src->ios[i]; |
| 749 | } |
| 750 | } |
| 751 | |
| 752 | static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src) |
| 753 | { |
| 754 | int i; |
| 755 | |
| 756 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { |
| 757 | dst->bytes[i] -= src->bytes[i]; |
| 758 | dst->ios[i] -= src->ios[i]; |
| 759 | } |
| 760 | } |
| 761 | |
| 762 | static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
| 763 | { |
| 764 | struct blkcg *blkcg = css_to_blkcg(css); |
| 765 | struct blkcg_gq *blkg; |
| 766 | |
Johannes Weiner | dc26532 | 2021-04-29 22:56:23 -0700 | [diff] [blame] | 767 | /* Root-level stats are sourced from system-wide IO stats */ |
| 768 | if (!cgroup_parent(css->cgroup)) |
| 769 | return; |
| 770 | |
Boris Burkov | cd1fc4b | 2020-06-01 13:11:43 -0700 | [diff] [blame] | 771 | rcu_read_lock(); |
| 772 | |
| 773 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
| 774 | struct blkcg_gq *parent = blkg->parent; |
| 775 | struct blkg_iostat_set *bisc = per_cpu_ptr(blkg->iostat_cpu, cpu); |
| 776 | struct blkg_iostat cur, delta; |
| 777 | unsigned int seq; |
| 778 | |
| 779 | /* fetch the current per-cpu values */ |
| 780 | do { |
| 781 | seq = u64_stats_fetch_begin(&bisc->sync); |
| 782 | blkg_iostat_set(&cur, &bisc->cur); |
| 783 | } while (u64_stats_fetch_retry(&bisc->sync, seq)); |
| 784 | |
| 785 | /* propagate percpu delta to global */ |
| 786 | u64_stats_update_begin(&blkg->iostat.sync); |
| 787 | blkg_iostat_set(&delta, &cur); |
| 788 | blkg_iostat_sub(&delta, &bisc->last); |
| 789 | blkg_iostat_add(&blkg->iostat.cur, &delta); |
| 790 | blkg_iostat_add(&bisc->last, &delta); |
| 791 | u64_stats_update_end(&blkg->iostat.sync); |
| 792 | |
Johannes Weiner | dc26532 | 2021-04-29 22:56:23 -0700 | [diff] [blame] | 793 | /* propagate global delta to parent (unless that's root) */ |
| 794 | if (parent && parent->parent) { |
Boris Burkov | cd1fc4b | 2020-06-01 13:11:43 -0700 | [diff] [blame] | 795 | u64_stats_update_begin(&parent->iostat.sync); |
| 796 | blkg_iostat_set(&delta, &blkg->iostat.cur); |
| 797 | blkg_iostat_sub(&delta, &blkg->iostat.last); |
| 798 | blkg_iostat_add(&parent->iostat.cur, &delta); |
| 799 | blkg_iostat_add(&blkg->iostat.last, &delta); |
| 800 | u64_stats_update_end(&parent->iostat.sync); |
| 801 | } |
| 802 | } |
| 803 | |
| 804 | rcu_read_unlock(); |
| 805 | } |
| 806 | |
Boris Burkov | ef45fe4 | 2020-06-01 13:12:05 -0700 | [diff] [blame] | 807 | /* |
Johannes Weiner | dc26532 | 2021-04-29 22:56:23 -0700 | [diff] [blame] | 808 | * We source root cgroup stats from the system-wide stats to avoid |
| 809 | * tracking the same information twice and incurring overhead when no |
| 810 | * cgroups are defined. For that reason, cgroup_rstat_flush in |
| 811 | * blkcg_print_stat does not actually fill out the iostat in the root |
| 812 | * cgroup's blkcg_gq. |
Boris Burkov | ef45fe4 | 2020-06-01 13:12:05 -0700 | [diff] [blame] | 813 | * |
| 814 | * However, we would like to re-use the printing code between the root and |
| 815 | * non-root cgroups to the extent possible. For that reason, we simulate |
| 816 | * flushing the root cgroup's stats by explicitly filling in the iostat |
| 817 | * with disk level statistics. |
| 818 | */ |
| 819 | static void blkcg_fill_root_iostats(void) |
| 820 | { |
| 821 | struct class_dev_iter iter; |
| 822 | struct device *dev; |
| 823 | |
| 824 | class_dev_iter_init(&iter, &block_class, NULL, &disk_type); |
| 825 | while ((dev = class_dev_iter_next(&iter))) { |
Christoph Hellwig | 0d02129 | 2020-11-27 16:43:51 +0100 | [diff] [blame] | 826 | struct block_device *bdev = dev_to_bdev(dev); |
| 827 | struct blkcg_gq *blkg = |
| 828 | blk_queue_root_blkg(bdev->bd_disk->queue); |
Boris Burkov | ef45fe4 | 2020-06-01 13:12:05 -0700 | [diff] [blame] | 829 | struct blkg_iostat tmp; |
| 830 | int cpu; |
| 831 | |
| 832 | memset(&tmp, 0, sizeof(tmp)); |
| 833 | for_each_possible_cpu(cpu) { |
| 834 | struct disk_stats *cpu_dkstats; |
| 835 | |
Christoph Hellwig | 0d02129 | 2020-11-27 16:43:51 +0100 | [diff] [blame] | 836 | cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu); |
Boris Burkov | ef45fe4 | 2020-06-01 13:12:05 -0700 | [diff] [blame] | 837 | tmp.ios[BLKG_IOSTAT_READ] += |
| 838 | cpu_dkstats->ios[STAT_READ]; |
| 839 | tmp.ios[BLKG_IOSTAT_WRITE] += |
| 840 | cpu_dkstats->ios[STAT_WRITE]; |
| 841 | tmp.ios[BLKG_IOSTAT_DISCARD] += |
| 842 | cpu_dkstats->ios[STAT_DISCARD]; |
| 843 | // convert sectors to bytes |
| 844 | tmp.bytes[BLKG_IOSTAT_READ] += |
| 845 | cpu_dkstats->sectors[STAT_READ] << 9; |
| 846 | tmp.bytes[BLKG_IOSTAT_WRITE] += |
| 847 | cpu_dkstats->sectors[STAT_WRITE] << 9; |
| 848 | tmp.bytes[BLKG_IOSTAT_DISCARD] += |
| 849 | cpu_dkstats->sectors[STAT_DISCARD] << 9; |
| 850 | |
| 851 | u64_stats_update_begin(&blkg->iostat.sync); |
| 852 | blkg_iostat_set(&blkg->iostat.cur, &tmp); |
| 853 | u64_stats_update_end(&blkg->iostat.sync); |
| 854 | } |
| 855 | } |
| 856 | } |
| 857 | |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 858 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
| 859 | { |
| 860 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); |
| 861 | struct blkcg_gq *blkg; |
| 862 | |
Boris Burkov | ef45fe4 | 2020-06-01 13:12:05 -0700 | [diff] [blame] | 863 | if (!seq_css(sf)->parent) |
| 864 | blkcg_fill_root_iostats(); |
| 865 | else |
| 866 | cgroup_rstat_flush(blkcg->css.cgroup); |
| 867 | |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 868 | rcu_read_lock(); |
| 869 | |
| 870 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
Tejun Heo | f733164 | 2019-11-07 11:18:03 -0800 | [diff] [blame] | 871 | struct blkg_iostat_set *bis = &blkg->iostat; |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 872 | const char *dname; |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 873 | char *buf; |
Tejun Heo | 636620b | 2018-07-18 04:47:41 -0700 | [diff] [blame] | 874 | u64 rbytes, wbytes, rios, wios, dbytes, dios; |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 875 | size_t size = seq_get_buf(sf, &buf), off = 0; |
| 876 | int i; |
| 877 | bool has_stats = false; |
Tejun Heo | f733164 | 2019-11-07 11:18:03 -0800 | [diff] [blame] | 878 | unsigned seq; |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 879 | |
Tejun Heo | b081436 | 2019-11-05 08:09:51 -0800 | [diff] [blame] | 880 | spin_lock_irq(&blkg->q->queue_lock); |
| 881 | |
| 882 | if (!blkg->online) |
| 883 | goto skip; |
| 884 | |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 885 | dname = blkg_dev_name(blkg); |
| 886 | if (!dname) |
Tejun Heo | b081436 | 2019-11-05 08:09:51 -0800 | [diff] [blame] | 887 | goto skip; |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 888 | |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 889 | /* |
| 890 | * Hooray string manipulation, count is the size written NOT |
| 891 | * INCLUDING THE \0, so size is now count+1 less than what we |
| 892 | * had before, but we want to start writing the next bit from |
| 893 | * the \0 so we only add count to buf. |
| 894 | */ |
| 895 | off += scnprintf(buf+off, size-off, "%s ", dname); |
| 896 | |
Tejun Heo | f733164 | 2019-11-07 11:18:03 -0800 | [diff] [blame] | 897 | do { |
| 898 | seq = u64_stats_fetch_begin(&bis->sync); |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 899 | |
Tejun Heo | f733164 | 2019-11-07 11:18:03 -0800 | [diff] [blame] | 900 | rbytes = bis->cur.bytes[BLKG_IOSTAT_READ]; |
| 901 | wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE]; |
| 902 | dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD]; |
| 903 | rios = bis->cur.ios[BLKG_IOSTAT_READ]; |
| 904 | wios = bis->cur.ios[BLKG_IOSTAT_WRITE]; |
| 905 | dios = bis->cur.ios[BLKG_IOSTAT_DISCARD]; |
| 906 | } while (u64_stats_fetch_retry(&bis->sync, seq)); |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 907 | |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 908 | if (rbytes || wbytes || rios || wios) { |
| 909 | has_stats = true; |
| 910 | off += scnprintf(buf+off, size-off, |
Tejun Heo | 636620b | 2018-07-18 04:47:41 -0700 | [diff] [blame] | 911 | "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
| 912 | rbytes, wbytes, rios, wios, |
| 913 | dbytes, dios); |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 914 | } |
| 915 | |
Tejun Heo | 07b0fde | 2019-07-16 07:58:31 -0700 | [diff] [blame] | 916 | if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) { |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 917 | has_stats = true; |
| 918 | off += scnprintf(buf+off, size-off, |
| 919 | " use_delay=%d delay_nsec=%llu", |
| 920 | atomic_read(&blkg->use_delay), |
| 921 | (unsigned long long)atomic64_read(&blkg->delay_nsec)); |
| 922 | } |
| 923 | |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 924 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
| 925 | struct blkcg_policy *pol = blkcg_policy[i]; |
| 926 | size_t written; |
| 927 | |
| 928 | if (!blkg->pd[i] || !pol->pd_stat_fn) |
| 929 | continue; |
| 930 | |
| 931 | written = pol->pd_stat_fn(blkg->pd[i], buf+off, size-off); |
| 932 | if (written) |
| 933 | has_stats = true; |
| 934 | off += written; |
| 935 | } |
Tejun Heo | 07b0fde | 2019-07-16 07:58:31 -0700 | [diff] [blame] | 936 | |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 937 | if (has_stats) { |
Tejun Heo | f539da8 | 2019-06-13 15:30:38 -0700 | [diff] [blame] | 938 | if (off < size - 1) { |
| 939 | off += scnprintf(buf+off, size-off, "\n"); |
| 940 | seq_commit(sf, off); |
| 941 | } else { |
| 942 | seq_commit(sf, -1); |
| 943 | } |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 944 | } |
Tejun Heo | b081436 | 2019-11-05 08:09:51 -0800 | [diff] [blame] | 945 | skip: |
| 946 | spin_unlock_irq(&blkg->q->queue_lock); |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 947 | } |
| 948 | |
| 949 | rcu_read_unlock(); |
| 950 | return 0; |
| 951 | } |
| 952 | |
Bart Van Assche | e1f3b94 | 2016-06-14 17:04:32 +0200 | [diff] [blame] | 953 | static struct cftype blkcg_files[] = { |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 954 | { |
| 955 | .name = "stat", |
| 956 | .seq_show = blkcg_print_stat, |
| 957 | }, |
| 958 | { } /* terminate */ |
| 959 | }; |
| 960 | |
Bart Van Assche | e1f3b94 | 2016-06-14 17:04:32 +0200 | [diff] [blame] | 961 | static struct cftype blkcg_legacy_files[] = { |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 962 | { |
Divyesh Shah | 84c124d | 2010-04-09 08:31:19 +0200 | [diff] [blame] | 963 | .name = "reset_stats", |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 964 | .write_u64 = blkcg_reset_stats, |
Vivek Goyal | 2208419 | 2009-12-03 12:59:49 -0500 | [diff] [blame] | 965 | }, |
Tejun Heo | 4baf6e3 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 966 | { } /* terminate */ |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 967 | }; |
| 968 | |
Dennis Zhou (Facebook) | 59b5771 | 2018-08-31 16:22:43 -0400 | [diff] [blame] | 969 | /* |
| 970 | * blkcg destruction is a three-stage process. |
| 971 | * |
| 972 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked |
| 973 | * which offlines writeback. Here we tie the next stage of blkg destruction |
| 974 | * to the completion of writeback associated with the blkcg. This lets us |
| 975 | * avoid punting potentially large amounts of outstanding writeback to root |
| 976 | * while maintaining any ongoing policies. The next stage is triggered when |
| 977 | * the nr_cgwbs count goes to zero. |
| 978 | * |
| 979 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called |
| 980 | * and handles the destruction of blkgs. Here the css reference held by |
| 981 | * the blkg is put back eventually allowing blkcg_css_free() to be called. |
| 982 | * This work may occur in cgwb_release_workfn() on the cgwb_release |
| 983 | * workqueue. Any submitted ios that fail to get the blkg ref will be |
| 984 | * punted to the root_blkg. |
| 985 | * |
| 986 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. |
| 987 | * This finally frees the blkcg. |
| 988 | */ |
| 989 | |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 990 | /** |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 991 | * blkcg_css_offline - cgroup css_offline callback |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 992 | * @css: css of interest |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 993 | * |
Dennis Zhou (Facebook) | 59b5771 | 2018-08-31 16:22:43 -0400 | [diff] [blame] | 994 | * This function is called when @css is about to go away. Here the cgwbs are |
| 995 | * offlined first and only once writeback associated with the blkcg has |
| 996 | * finished do we start step 2 (see above). |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 997 | */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 998 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 999 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 1000 | struct blkcg *blkcg = css_to_blkcg(css); |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1001 | |
Dennis Zhou (Facebook) | 59b5771 | 2018-08-31 16:22:43 -0400 | [diff] [blame] | 1002 | /* this prevents anyone from attaching or migrating to this blkcg */ |
| 1003 | wb_blkcg_offline(blkcg); |
| 1004 | |
Tejun Heo | d866dbf | 2019-07-24 10:37:22 -0700 | [diff] [blame] | 1005 | /* put the base online pin allowing step 2 to be triggered */ |
| 1006 | blkcg_unpin_online(blkcg); |
Dennis Zhou (Facebook) | 59b5771 | 2018-08-31 16:22:43 -0400 | [diff] [blame] | 1007 | } |
| 1008 | |
| 1009 | /** |
| 1010 | * blkcg_destroy_blkgs - responsible for shooting down blkgs |
| 1011 | * @blkcg: blkcg of interest |
| 1012 | * |
| 1013 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock |
| 1014 | * is nested inside q lock, this function performs reverse double lock dancing. |
| 1015 | * Destroying the blkgs releases the reference held on the blkcg's css allowing |
| 1016 | * blkcg_css_free to eventually be called. |
| 1017 | * |
| 1018 | * This is the blkcg counterpart of ioc_release_fn(). |
| 1019 | */ |
| 1020 | void blkcg_destroy_blkgs(struct blkcg *blkcg) |
| 1021 | { |
Baolin Wang | 6c635ca | 2021-01-28 13:58:15 +0800 | [diff] [blame] | 1022 | might_sleep(); |
| 1023 | |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 1024 | spin_lock_irq(&blkcg->lock); |
Tejun Heo | 7ee9c56 | 2012-03-05 13:15:11 -0800 | [diff] [blame] | 1025 | |
Dennis Zhou (Facebook) | 6b06546 | 2018-08-31 16:22:42 -0400 | [diff] [blame] | 1026 | while (!hlist_empty(&blkcg->blkg_list)) { |
| 1027 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, |
| 1028 | struct blkcg_gq, blkcg_node); |
Tejun Heo | c875f4d | 2012-03-05 13:15:22 -0800 | [diff] [blame] | 1029 | struct request_queue *q = blkg->q; |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1030 | |
Baolin Wang | 6c635ca | 2021-01-28 13:58:15 +0800 | [diff] [blame] | 1031 | if (need_resched() || !spin_trylock(&q->queue_lock)) { |
| 1032 | /* |
| 1033 | * Given that the system can accumulate a huge number |
| 1034 | * of blkgs in pathological cases, check to see if we |
| 1035 | * need to rescheduling to avoid softlockup. |
| 1036 | */ |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 1037 | spin_unlock_irq(&blkcg->lock); |
Baolin Wang | 6c635ca | 2021-01-28 13:58:15 +0800 | [diff] [blame] | 1038 | cond_resched(); |
Dan Carpenter | a556793 | 2012-03-29 20:57:08 +0200 | [diff] [blame] | 1039 | spin_lock_irq(&blkcg->lock); |
Baolin Wang | 6c635ca | 2021-01-28 13:58:15 +0800 | [diff] [blame] | 1040 | continue; |
Jens Axboe | 0f3942a | 2010-05-03 14:28:55 +0200 | [diff] [blame] | 1041 | } |
Baolin Wang | 6c635ca | 2021-01-28 13:58:15 +0800 | [diff] [blame] | 1042 | |
| 1043 | blkg_destroy(blkg); |
| 1044 | spin_unlock(&q->queue_lock); |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 1045 | } |
Jens Axboe | 0f3942a | 2010-05-03 14:28:55 +0200 | [diff] [blame] | 1046 | |
Tejun Heo | 9f13ef6 | 2012-03-05 13:15:21 -0800 | [diff] [blame] | 1047 | spin_unlock_irq(&blkcg->lock); |
Tejun Heo | 7ee9c56 | 2012-03-05 13:15:11 -0800 | [diff] [blame] | 1048 | } |
| 1049 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 1050 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
Tejun Heo | 7ee9c56 | 2012-03-05 13:15:11 -0800 | [diff] [blame] | 1051 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 1052 | struct blkcg *blkcg = css_to_blkcg(css); |
Tejun Heo | bc915e6 | 2015-08-18 14:55:08 -0700 | [diff] [blame] | 1053 | int i; |
Tejun Heo | 7ee9c56 | 2012-03-05 13:15:11 -0800 | [diff] [blame] | 1054 | |
Tejun Heo | 7876f93 | 2015-07-09 16:39:49 -0400 | [diff] [blame] | 1055 | mutex_lock(&blkcg_pol_mutex); |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 1056 | |
Tejun Heo | 7876f93 | 2015-07-09 16:39:49 -0400 | [diff] [blame] | 1057 | list_del(&blkcg->all_blkcgs_node); |
Tejun Heo | 7876f93 | 2015-07-09 16:39:49 -0400 | [diff] [blame] | 1058 | |
Tejun Heo | bc915e6 | 2015-08-18 14:55:08 -0700 | [diff] [blame] | 1059 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 1060 | if (blkcg->cpd[i]) |
| 1061 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); |
| 1062 | |
| 1063 | mutex_unlock(&blkcg_pol_mutex); |
| 1064 | |
Tejun Heo | bc915e6 | 2015-08-18 14:55:08 -0700 | [diff] [blame] | 1065 | kfree(blkcg); |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1066 | } |
| 1067 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 1068 | static struct cgroup_subsys_state * |
| 1069 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1070 | { |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1071 | struct blkcg *blkcg; |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 1072 | struct cgroup_subsys_state *ret; |
| 1073 | int i; |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1074 | |
Tejun Heo | 7876f93 | 2015-07-09 16:39:49 -0400 | [diff] [blame] | 1075 | mutex_lock(&blkcg_pol_mutex); |
| 1076 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 1077 | if (!parent_css) { |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1078 | blkcg = &blkcg_root; |
Tejun Heo | bc915e6 | 2015-08-18 14:55:08 -0700 | [diff] [blame] | 1079 | } else { |
| 1080 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
| 1081 | if (!blkcg) { |
| 1082 | ret = ERR_PTR(-ENOMEM); |
weiping zhang | 4c18c9e | 2017-08-25 23:49:32 +0800 | [diff] [blame] | 1083 | goto unlock; |
Tejun Heo | bc915e6 | 2015-08-18 14:55:08 -0700 | [diff] [blame] | 1084 | } |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 1085 | } |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1086 | |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 1087 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { |
| 1088 | struct blkcg_policy *pol = blkcg_policy[i]; |
| 1089 | struct blkcg_policy_data *cpd; |
| 1090 | |
| 1091 | /* |
| 1092 | * If the policy hasn't been attached yet, wait for it |
| 1093 | * to be attached before doing anything else. Otherwise, |
| 1094 | * check if the policy requires any specific per-cgroup |
| 1095 | * data: if it does, allocate and initialize it. |
| 1096 | */ |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 1097 | if (!pol || !pol->cpd_alloc_fn) |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 1098 | continue; |
| 1099 | |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 1100 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 1101 | if (!cpd) { |
| 1102 | ret = ERR_PTR(-ENOMEM); |
| 1103 | goto free_pd_blkcg; |
| 1104 | } |
Tejun Heo | 8143764 | 2015-08-18 14:55:15 -0700 | [diff] [blame] | 1105 | blkcg->cpd[i] = cpd; |
| 1106 | cpd->blkcg = blkcg; |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 1107 | cpd->plid = i; |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 1108 | if (pol->cpd_init_fn) |
| 1109 | pol->cpd_init_fn(cpd); |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 1110 | } |
| 1111 | |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1112 | spin_lock_init(&blkcg->lock); |
Tejun Heo | d866dbf | 2019-07-24 10:37:22 -0700 | [diff] [blame] | 1113 | refcount_set(&blkcg->online_pin, 1); |
Tejun Heo | e00f4f4 | 2016-11-21 18:03:32 -0500 | [diff] [blame] | 1114 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1115 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 1116 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 1117 | INIT_LIST_HEAD(&blkcg->cgwb_list); |
| 1118 | #endif |
Tejun Heo | 7876f93 | 2015-07-09 16:39:49 -0400 | [diff] [blame] | 1119 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
| 1120 | |
| 1121 | mutex_unlock(&blkcg_pol_mutex); |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1122 | return &blkcg->css; |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 1123 | |
| 1124 | free_pd_blkcg: |
| 1125 | for (i--; i >= 0; i--) |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 1126 | if (blkcg->cpd[i]) |
| 1127 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); |
weiping zhang | 4c18c9e | 2017-08-25 23:49:32 +0800 | [diff] [blame] | 1128 | |
| 1129 | if (blkcg != &blkcg_root) |
| 1130 | kfree(blkcg); |
| 1131 | unlock: |
Tejun Heo | 7876f93 | 2015-07-09 16:39:49 -0400 | [diff] [blame] | 1132 | mutex_unlock(&blkcg_pol_mutex); |
Arianna Avanzini | e48453c | 2015-06-05 23:38:42 +0200 | [diff] [blame] | 1133 | return ret; |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1134 | } |
| 1135 | |
Tejun Heo | 4308a43 | 2019-07-24 10:37:55 -0700 | [diff] [blame] | 1136 | static int blkcg_css_online(struct cgroup_subsys_state *css) |
| 1137 | { |
| 1138 | struct blkcg *blkcg = css_to_blkcg(css); |
| 1139 | struct blkcg *parent = blkcg_parent(blkcg); |
| 1140 | |
| 1141 | /* |
| 1142 | * blkcg_pin_online() is used to delay blkcg offline so that blkgs |
| 1143 | * don't go offline while cgwbs are still active on them. Pin the |
| 1144 | * parent so that offline always happens towards the root. |
| 1145 | */ |
| 1146 | if (parent) |
| 1147 | blkcg_pin_online(parent); |
| 1148 | return 0; |
| 1149 | } |
| 1150 | |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 1151 | /** |
| 1152 | * blkcg_init_queue - initialize blkcg part of request queue |
| 1153 | * @q: request_queue to initialize |
| 1154 | * |
Christoph Hellwig | c62b37d | 2020-07-01 10:59:43 +0200 | [diff] [blame] | 1155 | * Called from blk_alloc_queue(). Responsible for initializing blkcg |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 1156 | * part of new request_queue @q. |
| 1157 | * |
| 1158 | * RETURNS: |
| 1159 | * 0 on success, -errno on failure. |
| 1160 | */ |
| 1161 | int blkcg_init_queue(struct request_queue *q) |
| 1162 | { |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 1163 | struct blkcg_gq *new_blkg, *blkg; |
| 1164 | bool preloaded; |
Tejun Heo | ec13b1d | 2015-05-22 17:13:19 -0400 | [diff] [blame] | 1165 | int ret; |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 1166 | |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 1167 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
| 1168 | if (!new_blkg) |
| 1169 | return -ENOMEM; |
| 1170 | |
| 1171 | preloaded = !radix_tree_preload(GFP_KERNEL); |
| 1172 | |
Jiang Biao | bea5488 | 2018-04-19 12:04:26 +0800 | [diff] [blame] | 1173 | /* Make sure the root blkg exists. */ |
Tejun Heo | ec13b1d | 2015-05-22 17:13:19 -0400 | [diff] [blame] | 1174 | rcu_read_lock(); |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 1175 | spin_lock_irq(&q->queue_lock); |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 1176 | blkg = blkg_create(&blkcg_root, q, new_blkg); |
Jiang Biao | 901932a | 2018-04-19 12:06:09 +0800 | [diff] [blame] | 1177 | if (IS_ERR(blkg)) |
| 1178 | goto err_unlock; |
| 1179 | q->root_blkg = blkg; |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 1180 | spin_unlock_irq(&q->queue_lock); |
Tejun Heo | ec13b1d | 2015-05-22 17:13:19 -0400 | [diff] [blame] | 1181 | rcu_read_unlock(); |
| 1182 | |
Jens Axboe | d708f0d | 2017-03-29 11:25:48 -0600 | [diff] [blame] | 1183 | if (preloaded) |
| 1184 | radix_tree_preload_end(); |
| 1185 | |
Tejun Heo | ec13b1d | 2015-05-22 17:13:19 -0400 | [diff] [blame] | 1186 | ret = blk_throtl_init(q); |
Christoph Hellwig | 04be60b | 2018-11-14 17:02:12 +0100 | [diff] [blame] | 1187 | if (ret) |
| 1188 | goto err_destroy_all; |
Yufen Yu | 27029b4 | 2020-08-10 22:21:16 -0400 | [diff] [blame] | 1189 | |
| 1190 | ret = blk_iolatency_init(q); |
| 1191 | if (ret) { |
| 1192 | blk_throtl_exit(q); |
| 1193 | goto err_destroy_all; |
| 1194 | } |
Christoph Hellwig | 04be60b | 2018-11-14 17:02:12 +0100 | [diff] [blame] | 1195 | return 0; |
Jiang Biao | 901932a | 2018-04-19 12:06:09 +0800 | [diff] [blame] | 1196 | |
Christoph Hellwig | 04be60b | 2018-11-14 17:02:12 +0100 | [diff] [blame] | 1197 | err_destroy_all: |
Christoph Hellwig | 04be60b | 2018-11-14 17:02:12 +0100 | [diff] [blame] | 1198 | blkg_destroy_all(q); |
Christoph Hellwig | 04be60b | 2018-11-14 17:02:12 +0100 | [diff] [blame] | 1199 | return ret; |
Jiang Biao | 901932a | 2018-04-19 12:06:09 +0800 | [diff] [blame] | 1200 | err_unlock: |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 1201 | spin_unlock_irq(&q->queue_lock); |
Jiang Biao | 901932a | 2018-04-19 12:06:09 +0800 | [diff] [blame] | 1202 | rcu_read_unlock(); |
| 1203 | if (preloaded) |
| 1204 | radix_tree_preload_end(); |
| 1205 | return PTR_ERR(blkg); |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 1206 | } |
| 1207 | |
| 1208 | /** |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 1209 | * blkcg_exit_queue - exit and release blkcg part of request_queue |
| 1210 | * @q: request_queue being released |
| 1211 | * |
Marcos Paulo de Souza | 7585d50 | 2019-01-25 00:01:42 -0200 | [diff] [blame] | 1212 | * Called from blk_exit_queue(). Responsible for exiting blkcg part. |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 1213 | */ |
| 1214 | void blkcg_exit_queue(struct request_queue *q) |
| 1215 | { |
Tejun Heo | 3c96cb3 | 2012-04-13 13:11:34 -0700 | [diff] [blame] | 1216 | blkg_destroy_all(q); |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 1217 | blk_throtl_exit(q); |
| 1218 | } |
| 1219 | |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1220 | /* |
| 1221 | * We cannot support shared io contexts, as we have no mean to support |
| 1222 | * two tasks with the same ioc in two different groups without major rework |
| 1223 | * of the main cic data structures. For now we allow a task to change |
| 1224 | * its cgroup only if it's the only owner of its ioc. |
| 1225 | */ |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 1226 | static int blkcg_can_attach(struct cgroup_taskset *tset) |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1227 | { |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 1228 | struct task_struct *task; |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 1229 | struct cgroup_subsys_state *dst_css; |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1230 | struct io_context *ioc; |
| 1231 | int ret = 0; |
| 1232 | |
| 1233 | /* task_lock() is needed to avoid races with exit_io_context() */ |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 1234 | cgroup_taskset_for_each(task, dst_css, tset) { |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 1235 | task_lock(task); |
| 1236 | ioc = task->io_context; |
| 1237 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) |
| 1238 | ret = -EINVAL; |
| 1239 | task_unlock(task); |
| 1240 | if (ret) |
| 1241 | break; |
| 1242 | } |
Vivek Goyal | 31e4c28 | 2009-12-03 12:59:42 -0500 | [diff] [blame] | 1243 | return ret; |
| 1244 | } |
| 1245 | |
Tejun Heo | 69d7fde | 2015-08-18 14:55:36 -0700 | [diff] [blame] | 1246 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
| 1247 | { |
| 1248 | int i; |
| 1249 | |
| 1250 | mutex_lock(&blkcg_pol_mutex); |
| 1251 | |
| 1252 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
| 1253 | struct blkcg_policy *pol = blkcg_policy[i]; |
| 1254 | struct blkcg *blkcg; |
| 1255 | |
| 1256 | if (!pol || !pol->cpd_bind_fn) |
| 1257 | continue; |
| 1258 | |
| 1259 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) |
| 1260 | if (blkcg->cpd[pol->plid]) |
| 1261 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); |
| 1262 | } |
| 1263 | mutex_unlock(&blkcg_pol_mutex); |
| 1264 | } |
| 1265 | |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1266 | static void blkcg_exit(struct task_struct *tsk) |
| 1267 | { |
| 1268 | if (tsk->throttle_queue) |
| 1269 | blk_put_queue(tsk->throttle_queue); |
| 1270 | tsk->throttle_queue = NULL; |
| 1271 | } |
| 1272 | |
Tejun Heo | c165b3e | 2015-08-18 14:55:29 -0700 | [diff] [blame] | 1273 | struct cgroup_subsys io_cgrp_subsys = { |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 1274 | .css_alloc = blkcg_css_alloc, |
Tejun Heo | 4308a43 | 2019-07-24 10:37:55 -0700 | [diff] [blame] | 1275 | .css_online = blkcg_css_online, |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 1276 | .css_offline = blkcg_css_offline, |
| 1277 | .css_free = blkcg_css_free, |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1278 | .can_attach = blkcg_can_attach, |
Tejun Heo | f733164 | 2019-11-07 11:18:03 -0800 | [diff] [blame] | 1279 | .css_rstat_flush = blkcg_rstat_flush, |
Tejun Heo | 69d7fde | 2015-08-18 14:55:36 -0700 | [diff] [blame] | 1280 | .bind = blkcg_bind, |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 1281 | .dfl_cftypes = blkcg_files, |
Tejun Heo | 880f50e2 | 2015-08-18 14:55:30 -0700 | [diff] [blame] | 1282 | .legacy_cftypes = blkcg_legacy_files, |
Tejun Heo | c165b3e | 2015-08-18 14:55:29 -0700 | [diff] [blame] | 1283 | .legacy_name = "blkio", |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1284 | .exit = blkcg_exit, |
Tejun Heo | 1ced953b | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 1285 | #ifdef CONFIG_MEMCG |
| 1286 | /* |
| 1287 | * This ensures that, if available, memcg is automatically enabled |
| 1288 | * together on the default hierarchy so that the owner cgroup can |
| 1289 | * be retrieved from writeback pages. |
| 1290 | */ |
| 1291 | .depends_on = 1 << memory_cgrp_id, |
| 1292 | #endif |
Tejun Heo | 676f7c8 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 1293 | }; |
Tejun Heo | c165b3e | 2015-08-18 14:55:29 -0700 | [diff] [blame] | 1294 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
Tejun Heo | 676f7c8 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 1295 | |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1296 | /** |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1297 | * blkcg_activate_policy - activate a blkcg policy on a request_queue |
| 1298 | * @q: request_queue of interest |
| 1299 | * @pol: blkcg policy to activate |
| 1300 | * |
| 1301 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through |
| 1302 | * bypass mode to populate its blkgs with policy_data for @pol. |
| 1303 | * |
| 1304 | * Activation happens with @q bypassed, so nobody would be accessing blkgs |
| 1305 | * from IO path. Update of each blkg is protected by both queue and blkcg |
| 1306 | * locks so that holding either lock and testing blkcg_policy_enabled() is |
| 1307 | * always enough for dereferencing policy data. |
| 1308 | * |
| 1309 | * The caller is responsible for synchronizing [de]activations and policy |
| 1310 | * [un]registerations. Returns 0 on success, -errno on failure. |
| 1311 | */ |
| 1312 | int blkcg_activate_policy(struct request_queue *q, |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1313 | const struct blkcg_policy *pol) |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1314 | { |
Tejun Heo | 4c55f4f | 2015-08-18 14:55:09 -0700 | [diff] [blame] | 1315 | struct blkg_policy_data *pd_prealloc = NULL; |
Tejun Heo | 9d179b8 | 2019-10-15 09:03:47 -0700 | [diff] [blame] | 1316 | struct blkcg_gq *blkg, *pinned_blkg = NULL; |
Tejun Heo | 4c55f4f | 2015-08-18 14:55:09 -0700 | [diff] [blame] | 1317 | int ret; |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1318 | |
| 1319 | if (blkcg_policy_enabled(q, pol)) |
| 1320 | return 0; |
| 1321 | |
Jens Axboe | 344e9ff | 2018-11-15 12:22:51 -0700 | [diff] [blame] | 1322 | if (queue_is_mq(q)) |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 1323 | blk_mq_freeze_queue(q); |
Tejun Heo | 9d179b8 | 2019-10-15 09:03:47 -0700 | [diff] [blame] | 1324 | retry: |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 1325 | spin_lock_irq(&q->queue_lock); |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1326 | |
Tejun Heo | 9d179b8 | 2019-10-15 09:03:47 -0700 | [diff] [blame] | 1327 | /* blkg_list is pushed at the head, reverse walk to allocate parents first */ |
Tejun Heo | 71c8140 | 2019-06-13 15:30:40 -0700 | [diff] [blame] | 1328 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
Tejun Heo | 4c55f4f | 2015-08-18 14:55:09 -0700 | [diff] [blame] | 1329 | struct blkg_policy_data *pd; |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1330 | |
Tejun Heo | 4c55f4f | 2015-08-18 14:55:09 -0700 | [diff] [blame] | 1331 | if (blkg->pd[pol->plid]) |
| 1332 | continue; |
| 1333 | |
Tejun Heo | 9d179b8 | 2019-10-15 09:03:47 -0700 | [diff] [blame] | 1334 | /* If prealloc matches, use it; otherwise try GFP_NOWAIT */ |
| 1335 | if (blkg == pinned_blkg) { |
| 1336 | pd = pd_prealloc; |
| 1337 | pd_prealloc = NULL; |
| 1338 | } else { |
| 1339 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q, |
| 1340 | blkg->blkcg); |
| 1341 | } |
| 1342 | |
Tejun Heo | 4c55f4f | 2015-08-18 14:55:09 -0700 | [diff] [blame] | 1343 | if (!pd) { |
Tejun Heo | 9d179b8 | 2019-10-15 09:03:47 -0700 | [diff] [blame] | 1344 | /* |
| 1345 | * GFP_NOWAIT failed. Free the existing one and |
| 1346 | * prealloc for @blkg w/ GFP_KERNEL. |
| 1347 | */ |
| 1348 | if (pinned_blkg) |
| 1349 | blkg_put(pinned_blkg); |
| 1350 | blkg_get(blkg); |
| 1351 | pinned_blkg = blkg; |
| 1352 | |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 1353 | spin_unlock_irq(&q->queue_lock); |
Tejun Heo | 9d179b8 | 2019-10-15 09:03:47 -0700 | [diff] [blame] | 1354 | |
| 1355 | if (pd_prealloc) |
| 1356 | pol->pd_free_fn(pd_prealloc); |
| 1357 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q, |
| 1358 | blkg->blkcg); |
| 1359 | if (pd_prealloc) |
| 1360 | goto retry; |
| 1361 | else |
| 1362 | goto enomem; |
Tejun Heo | 4c55f4f | 2015-08-18 14:55:09 -0700 | [diff] [blame] | 1363 | } |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1364 | |
| 1365 | blkg->pd[pol->plid] = pd; |
| 1366 | pd->blkg = blkg; |
Tejun Heo | b276a87 | 2013-01-09 08:05:12 -0800 | [diff] [blame] | 1367 | pd->plid = pol->plid; |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1368 | } |
| 1369 | |
Tejun Heo | 9d179b8 | 2019-10-15 09:03:47 -0700 | [diff] [blame] | 1370 | /* all allocated, init in the same order */ |
| 1371 | if (pol->pd_init_fn) |
| 1372 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) |
| 1373 | pol->pd_init_fn(blkg->pd[pol->plid]); |
| 1374 | |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1375 | __set_bit(pol->plid, q->blkcg_pols); |
| 1376 | ret = 0; |
Tejun Heo | 4c55f4f | 2015-08-18 14:55:09 -0700 | [diff] [blame] | 1377 | |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 1378 | spin_unlock_irq(&q->queue_lock); |
Tejun Heo | 9d179b8 | 2019-10-15 09:03:47 -0700 | [diff] [blame] | 1379 | out: |
Jens Axboe | 344e9ff | 2018-11-15 12:22:51 -0700 | [diff] [blame] | 1380 | if (queue_is_mq(q)) |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 1381 | blk_mq_unfreeze_queue(q); |
Tejun Heo | 9d179b8 | 2019-10-15 09:03:47 -0700 | [diff] [blame] | 1382 | if (pinned_blkg) |
| 1383 | blkg_put(pinned_blkg); |
Tejun Heo | 001bea7 | 2015-08-18 14:55:11 -0700 | [diff] [blame] | 1384 | if (pd_prealloc) |
| 1385 | pol->pd_free_fn(pd_prealloc); |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1386 | return ret; |
Tejun Heo | 9d179b8 | 2019-10-15 09:03:47 -0700 | [diff] [blame] | 1387 | |
| 1388 | enomem: |
| 1389 | /* alloc failed, nothing's initialized yet, free everything */ |
| 1390 | spin_lock_irq(&q->queue_lock); |
| 1391 | list_for_each_entry(blkg, &q->blkg_list, q_node) { |
| 1392 | if (blkg->pd[pol->plid]) { |
| 1393 | pol->pd_free_fn(blkg->pd[pol->plid]); |
| 1394 | blkg->pd[pol->plid] = NULL; |
| 1395 | } |
| 1396 | } |
| 1397 | spin_unlock_irq(&q->queue_lock); |
| 1398 | ret = -ENOMEM; |
| 1399 | goto out; |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1400 | } |
| 1401 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); |
| 1402 | |
| 1403 | /** |
| 1404 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue |
| 1405 | * @q: request_queue of interest |
| 1406 | * @pol: blkcg policy to deactivate |
| 1407 | * |
| 1408 | * Deactivate @pol on @q. Follows the same synchronization rules as |
| 1409 | * blkcg_activate_policy(). |
| 1410 | */ |
| 1411 | void blkcg_deactivate_policy(struct request_queue *q, |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1412 | const struct blkcg_policy *pol) |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1413 | { |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1414 | struct blkcg_gq *blkg; |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1415 | |
| 1416 | if (!blkcg_policy_enabled(q, pol)) |
| 1417 | return; |
| 1418 | |
Jens Axboe | 344e9ff | 2018-11-15 12:22:51 -0700 | [diff] [blame] | 1419 | if (queue_is_mq(q)) |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 1420 | blk_mq_freeze_queue(q); |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 1421 | |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 1422 | spin_lock_irq(&q->queue_lock); |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1423 | |
| 1424 | __clear_bit(pol->plid, q->blkcg_pols); |
| 1425 | |
| 1426 | list_for_each_entry(blkg, &q->blkg_list, q_node) { |
Tejun Heo | 001bea7 | 2015-08-18 14:55:11 -0700 | [diff] [blame] | 1427 | if (blkg->pd[pol->plid]) { |
Dennis Zhou (Facebook) | 6b06546 | 2018-08-31 16:22:42 -0400 | [diff] [blame] | 1428 | if (pol->pd_offline_fn) |
Tejun Heo | a9520cd | 2015-08-18 14:55:14 -0700 | [diff] [blame] | 1429 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
Tejun Heo | 001bea7 | 2015-08-18 14:55:11 -0700 | [diff] [blame] | 1430 | pol->pd_free_fn(blkg->pd[pol->plid]); |
| 1431 | blkg->pd[pol->plid] = NULL; |
| 1432 | } |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1433 | } |
| 1434 | |
Christoph Hellwig | 0d945c1 | 2018-11-15 12:17:28 -0700 | [diff] [blame] | 1435 | spin_unlock_irq(&q->queue_lock); |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 1436 | |
Jens Axboe | 344e9ff | 2018-11-15 12:22:51 -0700 | [diff] [blame] | 1437 | if (queue_is_mq(q)) |
Jens Axboe | bd166ef | 2017-01-17 06:03:22 -0700 | [diff] [blame] | 1438 | blk_mq_unfreeze_queue(q); |
Tejun Heo | a2b1693 | 2012-04-13 13:11:33 -0700 | [diff] [blame] | 1439 | } |
| 1440 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); |
| 1441 | |
| 1442 | /** |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1443 | * blkcg_policy_register - register a blkcg policy |
| 1444 | * @pol: blkcg policy to register |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1445 | * |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1446 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
| 1447 | * successful registration. Returns 0 on success and -errno on failure. |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1448 | */ |
Jens Axboe | d5bf029 | 2014-06-22 16:31:56 -0600 | [diff] [blame] | 1449 | int blkcg_policy_register(struct blkcg_policy *pol) |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 1450 | { |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1451 | struct blkcg *blkcg; |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1452 | int i, ret; |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 1453 | |
Tejun Heo | 838f13b | 2015-07-09 16:39:47 -0400 | [diff] [blame] | 1454 | mutex_lock(&blkcg_pol_register_mutex); |
Tejun Heo | bc0d650 | 2012-04-13 13:11:26 -0700 | [diff] [blame] | 1455 | mutex_lock(&blkcg_pol_mutex); |
| 1456 | |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1457 | /* find an empty slot */ |
| 1458 | ret = -ENOSPC; |
| 1459 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1460 | if (!blkcg_policy[i]) |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1461 | break; |
Jens Axboe | 01c5f85 | 2018-09-11 10:59:53 -0600 | [diff] [blame] | 1462 | if (i >= BLKCG_MAX_POLS) { |
| 1463 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); |
Tejun Heo | 838f13b | 2015-07-09 16:39:47 -0400 | [diff] [blame] | 1464 | goto err_unlock; |
Jens Axboe | 01c5f85 | 2018-09-11 10:59:53 -0600 | [diff] [blame] | 1465 | } |
Tejun Heo | 035d10b | 2012-03-05 13:15:04 -0800 | [diff] [blame] | 1466 | |
weiping zhang | e840107 | 2017-10-17 23:56:21 +0800 | [diff] [blame] | 1467 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
| 1468 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || |
| 1469 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) |
| 1470 | goto err_unlock; |
| 1471 | |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1472 | /* register @pol */ |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1473 | pol->plid = i; |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1474 | blkcg_policy[pol->plid] = pol; |
| 1475 | |
| 1476 | /* allocate and install cpd's */ |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 1477 | if (pol->cpd_alloc_fn) { |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1478 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
| 1479 | struct blkcg_policy_data *cpd; |
| 1480 | |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 1481 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
Bart Van Assche | bbb427e | 2016-09-29 08:33:30 -0700 | [diff] [blame] | 1482 | if (!cpd) |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1483 | goto err_free_cpds; |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1484 | |
Tejun Heo | 8143764 | 2015-08-18 14:55:15 -0700 | [diff] [blame] | 1485 | blkcg->cpd[pol->plid] = cpd; |
| 1486 | cpd->blkcg = blkcg; |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1487 | cpd->plid = pol->plid; |
Tejun Heo | 86a5bba | 2019-08-28 15:05:52 -0700 | [diff] [blame] | 1488 | if (pol->cpd_init_fn) |
| 1489 | pol->cpd_init_fn(cpd); |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1490 | } |
| 1491 | } |
| 1492 | |
Tejun Heo | 838f13b | 2015-07-09 16:39:47 -0400 | [diff] [blame] | 1493 | mutex_unlock(&blkcg_pol_mutex); |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1494 | |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1495 | /* everything is in place, add intf files for the new policy */ |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 1496 | if (pol->dfl_cftypes) |
| 1497 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, |
| 1498 | pol->dfl_cftypes)); |
Tejun Heo | 880f50e2 | 2015-08-18 14:55:30 -0700 | [diff] [blame] | 1499 | if (pol->legacy_cftypes) |
Tejun Heo | c165b3e | 2015-08-18 14:55:29 -0700 | [diff] [blame] | 1500 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
Tejun Heo | 880f50e2 | 2015-08-18 14:55:30 -0700 | [diff] [blame] | 1501 | pol->legacy_cftypes)); |
Tejun Heo | 838f13b | 2015-07-09 16:39:47 -0400 | [diff] [blame] | 1502 | mutex_unlock(&blkcg_pol_register_mutex); |
| 1503 | return 0; |
| 1504 | |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1505 | err_free_cpds: |
weiping zhang | 58a9edc | 2017-10-10 22:53:46 +0800 | [diff] [blame] | 1506 | if (pol->cpd_free_fn) { |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1507 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 1508 | if (blkcg->cpd[pol->plid]) { |
| 1509 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); |
| 1510 | blkcg->cpd[pol->plid] = NULL; |
| 1511 | } |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1512 | } |
| 1513 | } |
| 1514 | blkcg_policy[pol->plid] = NULL; |
Tejun Heo | 838f13b | 2015-07-09 16:39:47 -0400 | [diff] [blame] | 1515 | err_unlock: |
Tejun Heo | bc0d650 | 2012-04-13 13:11:26 -0700 | [diff] [blame] | 1516 | mutex_unlock(&blkcg_pol_mutex); |
Tejun Heo | 838f13b | 2015-07-09 16:39:47 -0400 | [diff] [blame] | 1517 | mutex_unlock(&blkcg_pol_register_mutex); |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1518 | return ret; |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 1519 | } |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1520 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 1521 | |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1522 | /** |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1523 | * blkcg_policy_unregister - unregister a blkcg policy |
| 1524 | * @pol: blkcg policy to unregister |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1525 | * |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1526 | * Undo blkcg_policy_register(@pol). Might sleep. |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1527 | */ |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1528 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 1529 | { |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1530 | struct blkcg *blkcg; |
| 1531 | |
Tejun Heo | 838f13b | 2015-07-09 16:39:47 -0400 | [diff] [blame] | 1532 | mutex_lock(&blkcg_pol_register_mutex); |
Tejun Heo | bc0d650 | 2012-04-13 13:11:26 -0700 | [diff] [blame] | 1533 | |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1534 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
Tejun Heo | 8bd435b | 2012-04-13 13:11:28 -0700 | [diff] [blame] | 1535 | goto out_unlock; |
| 1536 | |
| 1537 | /* kill the intf files first */ |
Tejun Heo | 2ee867dc | 2015-08-18 14:55:34 -0700 | [diff] [blame] | 1538 | if (pol->dfl_cftypes) |
| 1539 | cgroup_rm_cftypes(pol->dfl_cftypes); |
Tejun Heo | 880f50e2 | 2015-08-18 14:55:30 -0700 | [diff] [blame] | 1540 | if (pol->legacy_cftypes) |
| 1541 | cgroup_rm_cftypes(pol->legacy_cftypes); |
Tejun Heo | 44ea53d | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 1542 | |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1543 | /* remove cpds and unregister */ |
Tejun Heo | 838f13b | 2015-07-09 16:39:47 -0400 | [diff] [blame] | 1544 | mutex_lock(&blkcg_pol_mutex); |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1545 | |
weiping zhang | 58a9edc | 2017-10-10 22:53:46 +0800 | [diff] [blame] | 1546 | if (pol->cpd_free_fn) { |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1547 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
Tejun Heo | e4a9bde | 2015-08-18 14:55:16 -0700 | [diff] [blame] | 1548 | if (blkcg->cpd[pol->plid]) { |
| 1549 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); |
| 1550 | blkcg->cpd[pol->plid] = NULL; |
| 1551 | } |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1552 | } |
| 1553 | } |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1554 | blkcg_policy[pol->plid] = NULL; |
Tejun Heo | 06b285b | 2015-07-09 16:39:50 -0400 | [diff] [blame] | 1555 | |
Tejun Heo | bc0d650 | 2012-04-13 13:11:26 -0700 | [diff] [blame] | 1556 | mutex_unlock(&blkcg_pol_mutex); |
Tejun Heo | 838f13b | 2015-07-09 16:39:47 -0400 | [diff] [blame] | 1557 | out_unlock: |
| 1558 | mutex_unlock(&blkcg_pol_register_mutex); |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 1559 | } |
Tejun Heo | 3c79839 | 2012-04-16 13:57:25 -0700 | [diff] [blame] | 1560 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 1561 | |
Tejun Heo | d3f77df | 2019-06-27 13:39:52 -0700 | [diff] [blame] | 1562 | bool __blkcg_punt_bio_submit(struct bio *bio) |
| 1563 | { |
| 1564 | struct blkcg_gq *blkg = bio->bi_blkg; |
| 1565 | |
| 1566 | /* consume the flag first */ |
| 1567 | bio->bi_opf &= ~REQ_CGROUP_PUNT; |
| 1568 | |
| 1569 | /* never bounce for the root cgroup */ |
| 1570 | if (!blkg->parent) |
| 1571 | return false; |
| 1572 | |
| 1573 | spin_lock_bh(&blkg->async_bio_lock); |
| 1574 | bio_list_add(&blkg->async_bios, bio); |
| 1575 | spin_unlock_bh(&blkg->async_bio_lock); |
| 1576 | |
| 1577 | queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work); |
| 1578 | return true; |
| 1579 | } |
| 1580 | |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1581 | /* |
| 1582 | * Scale the accumulated delay based on how long it has been since we updated |
| 1583 | * the delay. We only call this when we are adding delay, in case it's been a |
| 1584 | * while since we added delay, and when we are checking to see if we need to |
| 1585 | * delay a task, to account for any delays that may have occurred. |
| 1586 | */ |
| 1587 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) |
| 1588 | { |
| 1589 | u64 old = atomic64_read(&blkg->delay_start); |
| 1590 | |
Tejun Heo | 54c52e1 | 2020-04-13 12:27:55 -0400 | [diff] [blame] | 1591 | /* negative use_delay means no scaling, see blkcg_set_delay() */ |
| 1592 | if (atomic_read(&blkg->use_delay) < 0) |
| 1593 | return; |
| 1594 | |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1595 | /* |
| 1596 | * We only want to scale down every second. The idea here is that we |
| 1597 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain |
| 1598 | * time window. We only want to throttle tasks for recent delay that |
| 1599 | * has occurred, in 1 second time windows since that's the maximum |
| 1600 | * things can be throttled. We save the current delay window in |
| 1601 | * blkg->last_delay so we know what amount is still left to be charged |
| 1602 | * to the blkg from this point onward. blkg->last_use keeps track of |
| 1603 | * the use_delay counter. The idea is if we're unthrottling the blkg we |
| 1604 | * are ok with whatever is happening now, and we can take away more of |
| 1605 | * the accumulated delay as we've already throttled enough that |
| 1606 | * everybody is happy with their IO latencies. |
| 1607 | */ |
| 1608 | if (time_before64(old + NSEC_PER_SEC, now) && |
| 1609 | atomic64_cmpxchg(&blkg->delay_start, old, now) == old) { |
| 1610 | u64 cur = atomic64_read(&blkg->delay_nsec); |
| 1611 | u64 sub = min_t(u64, blkg->last_delay, now - old); |
| 1612 | int cur_use = atomic_read(&blkg->use_delay); |
| 1613 | |
| 1614 | /* |
| 1615 | * We've been unthrottled, subtract a larger chunk of our |
| 1616 | * accumulated delay. |
| 1617 | */ |
| 1618 | if (cur_use < blkg->last_use) |
| 1619 | sub = max_t(u64, sub, blkg->last_delay >> 1); |
| 1620 | |
| 1621 | /* |
| 1622 | * This shouldn't happen, but handle it anyway. Our delay_nsec |
| 1623 | * should only ever be growing except here where we subtract out |
| 1624 | * min(last_delay, 1 second), but lord knows bugs happen and I'd |
| 1625 | * rather not end up with negative numbers. |
| 1626 | */ |
| 1627 | if (unlikely(cur < sub)) { |
| 1628 | atomic64_set(&blkg->delay_nsec, 0); |
| 1629 | blkg->last_delay = 0; |
| 1630 | } else { |
| 1631 | atomic64_sub(sub, &blkg->delay_nsec); |
| 1632 | blkg->last_delay = cur - sub; |
| 1633 | } |
| 1634 | blkg->last_use = cur_use; |
| 1635 | } |
| 1636 | } |
| 1637 | |
| 1638 | /* |
| 1639 | * This is called when we want to actually walk up the hierarchy and check to |
| 1640 | * see if we need to throttle, and then actually throttle if there is some |
| 1641 | * accumulated delay. This should only be called upon return to user space so |
| 1642 | * we're not holding some lock that would induce a priority inversion. |
| 1643 | */ |
| 1644 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) |
| 1645 | { |
Josef Bacik | fd112c7 | 2019-07-09 14:41:29 -0700 | [diff] [blame] | 1646 | unsigned long pflags; |
Tejun Heo | 5160a5a | 2020-09-01 14:52:52 -0400 | [diff] [blame] | 1647 | bool clamp; |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1648 | u64 now = ktime_to_ns(ktime_get()); |
| 1649 | u64 exp; |
| 1650 | u64 delay_nsec = 0; |
| 1651 | int tok; |
| 1652 | |
| 1653 | while (blkg->parent) { |
Tejun Heo | 5160a5a | 2020-09-01 14:52:52 -0400 | [diff] [blame] | 1654 | int use_delay = atomic_read(&blkg->use_delay); |
| 1655 | |
| 1656 | if (use_delay) { |
| 1657 | u64 this_delay; |
| 1658 | |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1659 | blkcg_scale_delay(blkg, now); |
Tejun Heo | 5160a5a | 2020-09-01 14:52:52 -0400 | [diff] [blame] | 1660 | this_delay = atomic64_read(&blkg->delay_nsec); |
| 1661 | if (this_delay > delay_nsec) { |
| 1662 | delay_nsec = this_delay; |
| 1663 | clamp = use_delay > 0; |
| 1664 | } |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1665 | } |
| 1666 | blkg = blkg->parent; |
| 1667 | } |
| 1668 | |
| 1669 | if (!delay_nsec) |
| 1670 | return; |
| 1671 | |
| 1672 | /* |
| 1673 | * Let's not sleep for all eternity if we've amassed a huge delay. |
| 1674 | * Swapping or metadata IO can accumulate 10's of seconds worth of |
| 1675 | * delay, and we want userspace to be able to do _something_ so cap the |
Tejun Heo | 5160a5a | 2020-09-01 14:52:52 -0400 | [diff] [blame] | 1676 | * delays at 0.25s. If there's 10's of seconds worth of delay then the |
| 1677 | * tasks will be delayed for 0.25 second for every syscall. If |
| 1678 | * blkcg_set_delay() was used as indicated by negative use_delay, the |
| 1679 | * caller is responsible for regulating the range. |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1680 | */ |
Tejun Heo | 5160a5a | 2020-09-01 14:52:52 -0400 | [diff] [blame] | 1681 | if (clamp) |
| 1682 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1683 | |
Josef Bacik | fd112c7 | 2019-07-09 14:41:29 -0700 | [diff] [blame] | 1684 | if (use_memdelay) |
| 1685 | psi_memstall_enter(&pflags); |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1686 | |
| 1687 | exp = ktime_add_ns(now, delay_nsec); |
| 1688 | tok = io_schedule_prepare(); |
| 1689 | do { |
| 1690 | __set_current_state(TASK_KILLABLE); |
| 1691 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) |
| 1692 | break; |
| 1693 | } while (!fatal_signal_pending(current)); |
| 1694 | io_schedule_finish(tok); |
Josef Bacik | fd112c7 | 2019-07-09 14:41:29 -0700 | [diff] [blame] | 1695 | |
| 1696 | if (use_memdelay) |
| 1697 | psi_memstall_leave(&pflags); |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1698 | } |
| 1699 | |
| 1700 | /** |
| 1701 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked |
| 1702 | * |
| 1703 | * This is only called if we've been marked with set_notify_resume(). Obviously |
| 1704 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we |
| 1705 | * check to see if current->throttle_queue is set and if not this doesn't do |
| 1706 | * anything. This should only ever be called by the resume code, it's not meant |
| 1707 | * to be called by people willy-nilly as it will actually do the work to |
| 1708 | * throttle the task if it is setup for throttling. |
| 1709 | */ |
| 1710 | void blkcg_maybe_throttle_current(void) |
| 1711 | { |
| 1712 | struct request_queue *q = current->throttle_queue; |
| 1713 | struct cgroup_subsys_state *css; |
| 1714 | struct blkcg *blkcg; |
| 1715 | struct blkcg_gq *blkg; |
| 1716 | bool use_memdelay = current->use_memdelay; |
| 1717 | |
| 1718 | if (!q) |
| 1719 | return; |
| 1720 | |
| 1721 | current->throttle_queue = NULL; |
| 1722 | current->use_memdelay = false; |
| 1723 | |
| 1724 | rcu_read_lock(); |
| 1725 | css = kthread_blkcg(); |
| 1726 | if (css) |
| 1727 | blkcg = css_to_blkcg(css); |
| 1728 | else |
| 1729 | blkcg = css_to_blkcg(task_css(current, io_cgrp_id)); |
| 1730 | |
| 1731 | if (!blkcg) |
| 1732 | goto out; |
| 1733 | blkg = blkg_lookup(blkcg, q); |
| 1734 | if (!blkg) |
| 1735 | goto out; |
Dennis Zhou | 7754f66 | 2018-12-05 12:10:39 -0500 | [diff] [blame] | 1736 | if (!blkg_tryget(blkg)) |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1737 | goto out; |
| 1738 | rcu_read_unlock(); |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1739 | |
| 1740 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); |
| 1741 | blkg_put(blkg); |
Josef Bacik | cc7ecc25 | 2018-07-31 12:39:03 -0400 | [diff] [blame] | 1742 | blk_put_queue(q); |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1743 | return; |
| 1744 | out: |
| 1745 | rcu_read_unlock(); |
| 1746 | blk_put_queue(q); |
| 1747 | } |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1748 | |
| 1749 | /** |
| 1750 | * blkcg_schedule_throttle - this task needs to check for throttling |
Bart Van Assche | 537d71b | 2019-03-20 13:18:45 -0700 | [diff] [blame] | 1751 | * @q: the request queue IO was submitted on |
| 1752 | * @use_memdelay: do we charge this to memory delay for PSI |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1753 | * |
| 1754 | * This is called by the IO controller when we know there's delay accumulated |
| 1755 | * for the blkg for this task. We do not pass the blkg because there are places |
| 1756 | * we call this that may not have that information, the swapping code for |
| 1757 | * instance will only have a request_queue at that point. This set's the |
| 1758 | * notify_resume for the task to check and see if it requires throttling before |
| 1759 | * returning to user space. |
| 1760 | * |
| 1761 | * We will only schedule once per syscall. You can call this over and over |
| 1762 | * again and it will only do the check once upon return to user space, and only |
| 1763 | * throttle once. If the task needs to be throttled again it'll need to be |
| 1764 | * re-set at the next time we see the task. |
| 1765 | */ |
| 1766 | void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) |
| 1767 | { |
| 1768 | if (unlikely(current->flags & PF_KTHREAD)) |
| 1769 | return; |
| 1770 | |
Chunguang Xu | 49d1822 | 2021-01-25 13:05:28 +0800 | [diff] [blame] | 1771 | if (current->throttle_queue != q) { |
| 1772 | if (!blk_get_queue(q)) |
| 1773 | return; |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1774 | |
Chunguang Xu | 49d1822 | 2021-01-25 13:05:28 +0800 | [diff] [blame] | 1775 | if (current->throttle_queue) |
| 1776 | blk_put_queue(current->throttle_queue); |
| 1777 | current->throttle_queue = q; |
| 1778 | } |
| 1779 | |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1780 | if (use_memdelay) |
| 1781 | current->use_memdelay = use_memdelay; |
| 1782 | set_notify_resume(current); |
| 1783 | } |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1784 | |
| 1785 | /** |
| 1786 | * blkcg_add_delay - add delay to this blkg |
Bart Van Assche | 537d71b | 2019-03-20 13:18:45 -0700 | [diff] [blame] | 1787 | * @blkg: blkg of interest |
| 1788 | * @now: the current time in nanoseconds |
| 1789 | * @delta: how many nanoseconds of delay to add |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1790 | * |
| 1791 | * Charge @delta to the blkg's current delay accumulation. This is used to |
| 1792 | * throttle tasks if an IO controller thinks we need more throttling. |
| 1793 | */ |
| 1794 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) |
| 1795 | { |
Tejun Heo | 54c52e1 | 2020-04-13 12:27:55 -0400 | [diff] [blame] | 1796 | if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) |
| 1797 | return; |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1798 | blkcg_scale_delay(blkg, now); |
| 1799 | atomic64_add(delta, &blkg->delay_nsec); |
| 1800 | } |
Josef Bacik | d09d8df | 2018-07-03 11:14:55 -0400 | [diff] [blame] | 1801 | |
Christoph Hellwig | 28fc591 | 2020-06-27 09:31:51 +0200 | [diff] [blame] | 1802 | /** |
| 1803 | * blkg_tryget_closest - try and get a blkg ref on the closet blkg |
Christoph Hellwig | 13c7863 | 2020-06-27 09:31:54 +0200 | [diff] [blame] | 1804 | * @bio: target bio |
| 1805 | * @css: target css |
Christoph Hellwig | 28fc591 | 2020-06-27 09:31:51 +0200 | [diff] [blame] | 1806 | * |
Christoph Hellwig | 13c7863 | 2020-06-27 09:31:54 +0200 | [diff] [blame] | 1807 | * As the failure mode here is to walk up the blkg tree, this ensure that the |
| 1808 | * blkg->parent pointers are always valid. This returns the blkg that it ended |
| 1809 | * up taking a reference on or %NULL if no reference was taken. |
Christoph Hellwig | 28fc591 | 2020-06-27 09:31:51 +0200 | [diff] [blame] | 1810 | */ |
Christoph Hellwig | 13c7863 | 2020-06-27 09:31:54 +0200 | [diff] [blame] | 1811 | static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio, |
| 1812 | struct cgroup_subsys_state *css) |
Christoph Hellwig | 28fc591 | 2020-06-27 09:31:51 +0200 | [diff] [blame] | 1813 | { |
Christoph Hellwig | 13c7863 | 2020-06-27 09:31:54 +0200 | [diff] [blame] | 1814 | struct blkcg_gq *blkg, *ret_blkg = NULL; |
Christoph Hellwig | 28fc591 | 2020-06-27 09:31:51 +0200 | [diff] [blame] | 1815 | |
Christoph Hellwig | 13c7863 | 2020-06-27 09:31:54 +0200 | [diff] [blame] | 1816 | rcu_read_lock(); |
Christoph Hellwig | 309dca30 | 2021-01-24 11:02:34 +0100 | [diff] [blame] | 1817 | blkg = blkg_lookup_create(css_to_blkcg(css), |
| 1818 | bio->bi_bdev->bd_disk->queue); |
Christoph Hellwig | 28fc591 | 2020-06-27 09:31:51 +0200 | [diff] [blame] | 1819 | while (blkg) { |
| 1820 | if (blkg_tryget(blkg)) { |
| 1821 | ret_blkg = blkg; |
| 1822 | break; |
| 1823 | } |
| 1824 | blkg = blkg->parent; |
| 1825 | } |
Christoph Hellwig | 13c7863 | 2020-06-27 09:31:54 +0200 | [diff] [blame] | 1826 | rcu_read_unlock(); |
Christoph Hellwig | 28fc591 | 2020-06-27 09:31:51 +0200 | [diff] [blame] | 1827 | |
| 1828 | return ret_blkg; |
| 1829 | } |
| 1830 | |
| 1831 | /** |
| 1832 | * bio_associate_blkg_from_css - associate a bio with a specified css |
| 1833 | * @bio: target bio |
| 1834 | * @css: target css |
| 1835 | * |
| 1836 | * Associate @bio with the blkg found by combining the css's blkg and the |
| 1837 | * request_queue of the @bio. An association failure is handled by walking up |
| 1838 | * the blkg tree. Therefore, the blkg associated can be anything between @blkg |
| 1839 | * and q->root_blkg. This situation only happens when a cgroup is dying and |
| 1840 | * then the remaining bios will spill to the closest alive blkg. |
| 1841 | * |
| 1842 | * A reference will be taken on the blkg and will be released when @bio is |
| 1843 | * freed. |
| 1844 | */ |
| 1845 | void bio_associate_blkg_from_css(struct bio *bio, |
| 1846 | struct cgroup_subsys_state *css) |
| 1847 | { |
Christoph Hellwig | 28fc591 | 2020-06-27 09:31:51 +0200 | [diff] [blame] | 1848 | if (bio->bi_blkg) |
| 1849 | blkg_put(bio->bi_blkg); |
| 1850 | |
Christoph Hellwig | a5b9752 | 2020-06-27 09:31:53 +0200 | [diff] [blame] | 1851 | if (css && css->parent) { |
Christoph Hellwig | 13c7863 | 2020-06-27 09:31:54 +0200 | [diff] [blame] | 1852 | bio->bi_blkg = blkg_tryget_closest(bio, css); |
Christoph Hellwig | a5b9752 | 2020-06-27 09:31:53 +0200 | [diff] [blame] | 1853 | } else { |
Christoph Hellwig | 309dca30 | 2021-01-24 11:02:34 +0100 | [diff] [blame] | 1854 | blkg_get(bio->bi_bdev->bd_disk->queue->root_blkg); |
| 1855 | bio->bi_blkg = bio->bi_bdev->bd_disk->queue->root_blkg; |
Christoph Hellwig | a5b9752 | 2020-06-27 09:31:53 +0200 | [diff] [blame] | 1856 | } |
Christoph Hellwig | 28fc591 | 2020-06-27 09:31:51 +0200 | [diff] [blame] | 1857 | } |
| 1858 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); |
| 1859 | |
| 1860 | /** |
| 1861 | * bio_associate_blkg - associate a bio with a blkg |
| 1862 | * @bio: target bio |
| 1863 | * |
| 1864 | * Associate @bio with the blkg found from the bio's css and request_queue. |
| 1865 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is |
| 1866 | * already associated, the css is reused and association redone as the |
| 1867 | * request_queue may have changed. |
| 1868 | */ |
| 1869 | void bio_associate_blkg(struct bio *bio) |
| 1870 | { |
| 1871 | struct cgroup_subsys_state *css; |
| 1872 | |
| 1873 | rcu_read_lock(); |
| 1874 | |
| 1875 | if (bio->bi_blkg) |
| 1876 | css = &bio_blkcg(bio)->css; |
| 1877 | else |
| 1878 | css = blkcg_css(); |
| 1879 | |
| 1880 | bio_associate_blkg_from_css(bio, css); |
| 1881 | |
| 1882 | rcu_read_unlock(); |
| 1883 | } |
| 1884 | EXPORT_SYMBOL_GPL(bio_associate_blkg); |
| 1885 | |
| 1886 | /** |
| 1887 | * bio_clone_blkg_association - clone blkg association from src to dst bio |
| 1888 | * @dst: destination bio |
| 1889 | * @src: source bio |
| 1890 | */ |
| 1891 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) |
| 1892 | { |
| 1893 | if (src->bi_blkg) { |
| 1894 | if (dst->bi_blkg) |
| 1895 | blkg_put(dst->bi_blkg); |
| 1896 | blkg_get(src->bi_blkg); |
| 1897 | dst->bi_blkg = src->bi_blkg; |
| 1898 | } |
| 1899 | } |
| 1900 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); |
| 1901 | |
Christoph Hellwig | db18a53 | 2020-06-27 09:31:58 +0200 | [diff] [blame] | 1902 | static int blk_cgroup_io_type(struct bio *bio) |
| 1903 | { |
| 1904 | if (op_is_discard(bio->bi_opf)) |
| 1905 | return BLKG_IOSTAT_DISCARD; |
| 1906 | if (op_is_write(bio->bi_opf)) |
| 1907 | return BLKG_IOSTAT_WRITE; |
| 1908 | return BLKG_IOSTAT_READ; |
| 1909 | } |
| 1910 | |
| 1911 | void blk_cgroup_bio_start(struct bio *bio) |
| 1912 | { |
| 1913 | int rwd = blk_cgroup_io_type(bio), cpu; |
| 1914 | struct blkg_iostat_set *bis; |
| 1915 | |
| 1916 | cpu = get_cpu(); |
| 1917 | bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu); |
| 1918 | u64_stats_update_begin(&bis->sync); |
| 1919 | |
| 1920 | /* |
| 1921 | * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split |
| 1922 | * bio and we would have already accounted for the size of the bio. |
| 1923 | */ |
| 1924 | if (!bio_flagged(bio, BIO_CGROUP_ACCT)) { |
| 1925 | bio_set_flag(bio, BIO_CGROUP_ACCT); |
Colin Ian King | 0b8cc25 | 2020-06-30 16:54:41 +0100 | [diff] [blame] | 1926 | bis->cur.bytes[rwd] += bio->bi_iter.bi_size; |
Christoph Hellwig | db18a53 | 2020-06-27 09:31:58 +0200 | [diff] [blame] | 1927 | } |
| 1928 | bis->cur.ios[rwd]++; |
| 1929 | |
| 1930 | u64_stats_update_end(&bis->sync); |
| 1931 | if (cgroup_subsys_on_dfl(io_cgrp_subsys)) |
| 1932 | cgroup_rstat_updated(bio->bi_blkg->blkcg->css.cgroup, cpu); |
| 1933 | put_cpu(); |
| 1934 | } |
| 1935 | |
Tejun Heo | d3f77df | 2019-06-27 13:39:52 -0700 | [diff] [blame] | 1936 | static int __init blkcg_init(void) |
| 1937 | { |
| 1938 | blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio", |
| 1939 | WQ_MEM_RECLAIM | WQ_FREEZABLE | |
| 1940 | WQ_UNBOUND | WQ_SYSFS, 0); |
| 1941 | if (!blkcg_punt_bio_wq) |
| 1942 | return -ENOMEM; |
| 1943 | return 0; |
| 1944 | } |
| 1945 | subsys_initcall(blkcg_init); |
| 1946 | |
Josef Bacik | 903d23f | 2018-07-03 11:14:52 -0400 | [diff] [blame] | 1947 | module_param(blkcg_debug_stats, bool, 0644); |
| 1948 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |