blob: 55512dd626336eae49b758def08d601bc3515b74 [file] [log] [blame]
Vivek Goyal31e4c282009-12-03 12:59:42 -05001/*
2 * Common Block IO controller cgroup interface
3 *
4 * Based on ideas and code from CFQ, CFS and BFQ:
5 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
6 *
7 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
8 * Paolo Valente <paolo.valente@unimore.it>
9 *
10 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
11 * Nauman Rafique <nauman@google.com>
Arianna Avanzinie48453c2015-06-05 23:38:42 +020012 *
13 * For policy-specific per-blkcg data:
14 * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it>
15 * Arianna Avanzini <avanzini.arianna@gmail.com>
Vivek Goyal31e4c282009-12-03 12:59:42 -050016 */
17#include <linux/ioprio.h>
Vivek Goyal22084192009-12-03 12:59:49 -050018#include <linux/kdev_t.h>
Vivek Goyal9d6a9862009-12-04 10:36:41 -050019#include <linux/module.h>
Stephen Rothwellaccee782009-12-07 19:29:39 +110020#include <linux/err.h>
Divyesh Shah91952912010-04-01 15:01:41 -070021#include <linux/blkdev.h>
Tejun Heo52ebea72015-05-22 17:13:37 -040022#include <linux/backing-dev.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090023#include <linux/slab.h>
Gui Jianfeng34d0f172010-04-13 16:05:49 +080024#include <linux/genhd.h>
Tejun Heo72e06c22012-03-05 13:15:00 -080025#include <linux/delay.h>
Tejun Heo9a9e8a22012-03-19 15:10:56 -070026#include <linux/atomic.h>
Tejun Heo36aa9e52015-08-18 14:55:31 -070027#include <linux/ctype.h>
Tejun Heoeea8f412015-05-22 17:13:17 -040028#include <linux/blk-cgroup.h>
Tejun Heo5efd6112012-03-05 13:15:12 -080029#include "blk.h"
Vivek Goyal3e252062009-12-04 10:36:42 -050030
Divyesh Shah84c124d2010-04-09 08:31:19 +020031#define MAX_KEY_LEN 100
32
Tejun Heo838f13b2015-07-09 16:39:47 -040033/*
34 * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation.
35 * blkcg_pol_register_mutex nests outside of it and synchronizes entire
36 * policy [un]register operations including cgroup file additions /
37 * removals. Putting cgroup file registration outside blkcg_pol_mutex
38 * allows grabbing it from cgroup callbacks.
39 */
40static DEFINE_MUTEX(blkcg_pol_register_mutex);
Tejun Heobc0d6502012-04-13 13:11:26 -070041static DEFINE_MUTEX(blkcg_pol_mutex);
Tejun Heo923adde2012-03-05 13:15:13 -080042
Arianna Avanzinie48453c2015-06-05 23:38:42 +020043struct blkcg blkcg_root;
Tejun Heo3c798392012-04-16 13:57:25 -070044EXPORT_SYMBOL_GPL(blkcg_root);
Vivek Goyal9d6a9862009-12-04 10:36:41 -050045
Tejun Heo496d5e72015-05-22 17:13:21 -040046struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css;
47
Tejun Heo3c798392012-04-16 13:57:25 -070048static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
Tejun Heo035d10b2012-03-05 13:15:04 -080049
Tejun Heo7876f932015-07-09 16:39:49 -040050static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */
51
Tejun Heoa2b16932012-04-13 13:11:33 -070052static bool blkcg_policy_enabled(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -070053 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -070054{
55 return pol && test_bit(pol->plid, q->blkcg_pols);
56}
57
Tejun Heo03814112012-03-05 13:15:14 -080058/**
59 * blkg_free - free a blkg
60 * @blkg: blkg to free
61 *
62 * Free @blkg which may be partially allocated.
63 */
Tejun Heo3c798392012-04-16 13:57:25 -070064static void blkg_free(struct blkcg_gq *blkg)
Tejun Heo03814112012-03-05 13:15:14 -080065{
Tejun Heoe8989fa2012-03-05 13:15:20 -080066 int i;
Tejun Heo549d3aa2012-03-05 13:15:16 -080067
68 if (!blkg)
69 return;
70
Tejun Heodb613672013-05-14 13:52:31 -070071 for (i = 0; i < BLKCG_MAX_POLS; i++)
Tejun Heo001bea72015-08-18 14:55:11 -070072 if (blkg->pd[i])
73 blkcg_policy[i]->pd_free_fn(blkg->pd[i]);
Tejun Heoe8989fa2012-03-05 13:15:20 -080074
Tejun Heo994b7832015-08-18 14:55:07 -070075 if (blkg->blkcg != &blkcg_root)
76 blk_exit_rl(&blkg->rl);
Tejun Heo77ea7332015-08-18 14:55:24 -070077
78 blkg_rwstat_exit(&blkg->stat_ios);
79 blkg_rwstat_exit(&blkg->stat_bytes);
Tejun Heo549d3aa2012-03-05 13:15:16 -080080 kfree(blkg);
Tejun Heo03814112012-03-05 13:15:14 -080081}
82
83/**
84 * blkg_alloc - allocate a blkg
85 * @blkcg: block cgroup the new blkg is associated with
86 * @q: request_queue the new blkg is associated with
Tejun Heo15974992012-06-04 20:40:52 -070087 * @gfp_mask: allocation mask to use
Tejun Heo03814112012-03-05 13:15:14 -080088 *
Tejun Heoe8989fa2012-03-05 13:15:20 -080089 * Allocate a new blkg assocating @blkcg and @q.
Tejun Heo03814112012-03-05 13:15:14 -080090 */
Tejun Heo15974992012-06-04 20:40:52 -070091static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
92 gfp_t gfp_mask)
Tejun Heo03814112012-03-05 13:15:14 -080093{
Tejun Heo3c798392012-04-16 13:57:25 -070094 struct blkcg_gq *blkg;
Tejun Heoe8989fa2012-03-05 13:15:20 -080095 int i;
Tejun Heo03814112012-03-05 13:15:14 -080096
97 /* alloc and init base part */
Tejun Heo15974992012-06-04 20:40:52 -070098 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
Tejun Heo03814112012-03-05 13:15:14 -080099 if (!blkg)
100 return NULL;
101
Tejun Heo77ea7332015-08-18 14:55:24 -0700102 if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) ||
103 blkg_rwstat_init(&blkg->stat_ios, gfp_mask))
104 goto err_free;
105
Tejun Heoc875f4d2012-03-05 13:15:22 -0800106 blkg->q = q;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800107 INIT_LIST_HEAD(&blkg->q_node);
Tejun Heo03814112012-03-05 13:15:14 -0800108 blkg->blkcg = blkcg;
Tejun Heoa5049a82014-06-19 17:42:57 -0400109 atomic_set(&blkg->refcnt, 1);
Tejun Heo03814112012-03-05 13:15:14 -0800110
Tejun Heoa0516612012-06-26 15:05:44 -0700111 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
112 if (blkcg != &blkcg_root) {
113 if (blk_init_rl(&blkg->rl, q, gfp_mask))
114 goto err_free;
115 blkg->rl.blkg = blkg;
116 }
117
Tejun Heo8bd435b2012-04-13 13:11:28 -0700118 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -0700119 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heoe8989fa2012-03-05 13:15:20 -0800120 struct blkg_policy_data *pd;
Tejun Heo03814112012-03-05 13:15:14 -0800121
Tejun Heoa2b16932012-04-13 13:11:33 -0700122 if (!blkcg_policy_enabled(q, pol))
Tejun Heoe8989fa2012-03-05 13:15:20 -0800123 continue;
Tejun Heo549d3aa2012-03-05 13:15:16 -0800124
Tejun Heoe8989fa2012-03-05 13:15:20 -0800125 /* alloc per-policy data and attach it to blkg */
Tejun Heo001bea72015-08-18 14:55:11 -0700126 pd = pol->pd_alloc_fn(gfp_mask, q->node);
Tejun Heoa0516612012-06-26 15:05:44 -0700127 if (!pd)
128 goto err_free;
Tejun Heo549d3aa2012-03-05 13:15:16 -0800129
Tejun Heoe8989fa2012-03-05 13:15:20 -0800130 blkg->pd[i] = pd;
131 pd->blkg = blkg;
Tejun Heob276a872013-01-09 08:05:12 -0800132 pd->plid = i;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800133 }
134
Tejun Heo03814112012-03-05 13:15:14 -0800135 return blkg;
Tejun Heoa0516612012-06-26 15:05:44 -0700136
137err_free:
138 blkg_free(blkg);
139 return NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800140}
141
Tejun Heo24f29042015-08-18 14:55:17 -0700142struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
143 struct request_queue *q, bool update_hint)
Tejun Heo80fd9972012-04-13 14:50:53 -0700144{
Tejun Heo3c798392012-04-16 13:57:25 -0700145 struct blkcg_gq *blkg;
Tejun Heo80fd9972012-04-13 14:50:53 -0700146
Tejun Heoa6371202012-04-19 16:29:24 -0700147 /*
Tejun Heo86cde6b2013-01-09 08:05:10 -0800148 * Hint didn't match. Look up from the radix tree. Note that the
149 * hint can only be updated under queue_lock as otherwise @blkg
150 * could have already been removed from blkg_tree. The caller is
151 * responsible for grabbing queue_lock if @update_hint.
Tejun Heoa6371202012-04-19 16:29:24 -0700152 */
153 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800154 if (blkg && blkg->q == q) {
155 if (update_hint) {
156 lockdep_assert_held(q->queue_lock);
157 rcu_assign_pointer(blkcg->blkg_hint, blkg);
158 }
Tejun Heoa6371202012-04-19 16:29:24 -0700159 return blkg;
Tejun Heo86cde6b2013-01-09 08:05:10 -0800160 }
Tejun Heoa6371202012-04-19 16:29:24 -0700161
Tejun Heo80fd9972012-04-13 14:50:53 -0700162 return NULL;
163}
Tejun Heoae118892015-08-18 14:55:20 -0700164EXPORT_SYMBOL_GPL(blkg_lookup_slowpath);
Tejun Heo80fd9972012-04-13 14:50:53 -0700165
Tejun Heo15974992012-06-04 20:40:52 -0700166/*
167 * If @new_blkg is %NULL, this function tries to allocate a new one as
Tejun Heod93a11f2015-08-18 14:55:01 -0700168 * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return.
Tejun Heo15974992012-06-04 20:40:52 -0700169 */
Tejun Heo86cde6b2013-01-09 08:05:10 -0800170static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
171 struct request_queue *q,
172 struct blkcg_gq *new_blkg)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400173{
Tejun Heo3c798392012-04-16 13:57:25 -0700174 struct blkcg_gq *blkg;
Tejun Heoce7acfe2015-05-22 17:13:38 -0400175 struct bdi_writeback_congested *wb_congested;
Tejun Heof427d902013-01-09 08:05:12 -0800176 int i, ret;
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400177
Tejun Heocd1604f2012-03-05 13:15:06 -0800178 WARN_ON_ONCE(!rcu_read_lock_held());
179 lockdep_assert_held(q->queue_lock);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500180
Tejun Heo7ee9c562012-03-05 13:15:11 -0800181 /* blkg holds a reference to blkcg */
Tejun Heoec903c02014-05-13 12:11:01 -0400182 if (!css_tryget_online(&blkcg->css)) {
Tejun Heo20386ce2015-08-18 14:55:28 -0700183 ret = -ENODEV;
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800184 goto err_free_blkg;
Tejun Heo15974992012-06-04 20:40:52 -0700185 }
Tejun Heocd1604f2012-03-05 13:15:06 -0800186
Tejun Heoce7acfe2015-05-22 17:13:38 -0400187 wb_congested = wb_congested_get_create(&q->backing_dev_info,
Tejun Heod93a11f2015-08-18 14:55:01 -0700188 blkcg->css.id, GFP_NOWAIT);
Tejun Heoce7acfe2015-05-22 17:13:38 -0400189 if (!wb_congested) {
190 ret = -ENOMEM;
191 goto err_put_css;
192 }
193
Tejun Heo496fb782012-04-19 16:29:23 -0700194 /* allocate */
Tejun Heo15974992012-06-04 20:40:52 -0700195 if (!new_blkg) {
Tejun Heod93a11f2015-08-18 14:55:01 -0700196 new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT);
Tejun Heo15974992012-06-04 20:40:52 -0700197 if (unlikely(!new_blkg)) {
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800198 ret = -ENOMEM;
Tejun Heoce7acfe2015-05-22 17:13:38 -0400199 goto err_put_congested;
Tejun Heo15974992012-06-04 20:40:52 -0700200 }
201 }
202 blkg = new_blkg;
Tejun Heoce7acfe2015-05-22 17:13:38 -0400203 blkg->wb_congested = wb_congested;
Tejun Heocd1604f2012-03-05 13:15:06 -0800204
Tejun Heodb613672013-05-14 13:52:31 -0700205 /* link parent */
Tejun Heo3c547862013-01-09 08:05:10 -0800206 if (blkcg_parent(blkcg)) {
207 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
208 if (WARN_ON_ONCE(!blkg->parent)) {
Tejun Heo20386ce2015-08-18 14:55:28 -0700209 ret = -ENODEV;
Tejun Heoce7acfe2015-05-22 17:13:38 -0400210 goto err_put_congested;
Tejun Heo3c547862013-01-09 08:05:10 -0800211 }
212 blkg_get(blkg->parent);
213 }
214
Tejun Heodb613672013-05-14 13:52:31 -0700215 /* invoke per-policy init */
216 for (i = 0; i < BLKCG_MAX_POLS; i++) {
217 struct blkcg_policy *pol = blkcg_policy[i];
218
219 if (blkg->pd[i] && pol->pd_init_fn)
Tejun Heoa9520cd2015-08-18 14:55:14 -0700220 pol->pd_init_fn(blkg->pd[i]);
Tejun Heodb613672013-05-14 13:52:31 -0700221 }
222
223 /* insert */
Tejun Heoa6371202012-04-19 16:29:24 -0700224 spin_lock(&blkcg->lock);
225 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
226 if (likely(!ret)) {
227 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
228 list_add(&blkg->q_node, &q->blkg_list);
Tejun Heof427d902013-01-09 08:05:12 -0800229
230 for (i = 0; i < BLKCG_MAX_POLS; i++) {
231 struct blkcg_policy *pol = blkcg_policy[i];
232
233 if (blkg->pd[i] && pol->pd_online_fn)
Tejun Heoa9520cd2015-08-18 14:55:14 -0700234 pol->pd_online_fn(blkg->pd[i]);
Tejun Heof427d902013-01-09 08:05:12 -0800235 }
Tejun Heoa6371202012-04-19 16:29:24 -0700236 }
Tejun Heof427d902013-01-09 08:05:12 -0800237 blkg->online = true;
Tejun Heoa6371202012-04-19 16:29:24 -0700238 spin_unlock(&blkcg->lock);
239
Tejun Heoec13b1d2015-05-22 17:13:19 -0400240 if (!ret)
Tejun Heoa6371202012-04-19 16:29:24 -0700241 return blkg;
Tejun Heo15974992012-06-04 20:40:52 -0700242
Tejun Heo3c547862013-01-09 08:05:10 -0800243 /* @blkg failed fully initialized, use the usual release path */
244 blkg_put(blkg);
245 return ERR_PTR(ret);
246
Tejun Heoce7acfe2015-05-22 17:13:38 -0400247err_put_congested:
248 wb_congested_put(wb_congested);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800249err_put_css:
Tejun Heo496fb782012-04-19 16:29:23 -0700250 css_put(&blkcg->css);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800251err_free_blkg:
Tejun Heo15974992012-06-04 20:40:52 -0700252 blkg_free(new_blkg);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800253 return ERR_PTR(ret);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500254}
Tejun Heo3c96cb32012-04-13 13:11:34 -0700255
Tejun Heo86cde6b2013-01-09 08:05:10 -0800256/**
257 * blkg_lookup_create - lookup blkg, try to create one if not there
258 * @blkcg: blkcg of interest
259 * @q: request_queue of interest
260 *
261 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
Tejun Heo3c547862013-01-09 08:05:10 -0800262 * create one. blkg creation is performed recursively from blkcg_root such
263 * that all non-root blkg's have access to the parent blkg. This function
264 * should be called under RCU read lock and @q->queue_lock.
Tejun Heo86cde6b2013-01-09 08:05:10 -0800265 *
266 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
267 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
268 * dead and bypassing, returns ERR_PTR(-EBUSY).
269 */
Tejun Heo3c798392012-04-16 13:57:25 -0700270struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
271 struct request_queue *q)
Tejun Heo3c96cb32012-04-13 13:11:34 -0700272{
Tejun Heo86cde6b2013-01-09 08:05:10 -0800273 struct blkcg_gq *blkg;
274
275 WARN_ON_ONCE(!rcu_read_lock_held());
276 lockdep_assert_held(q->queue_lock);
277
Tejun Heo3c96cb32012-04-13 13:11:34 -0700278 /*
279 * This could be the first entry point of blkcg implementation and
280 * we shouldn't allow anything to go through for a bypassing queue.
281 */
282 if (unlikely(blk_queue_bypass(q)))
Tejun Heo20386ce2015-08-18 14:55:28 -0700283 return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800284
285 blkg = __blkg_lookup(blkcg, q, true);
286 if (blkg)
287 return blkg;
288
Tejun Heo3c547862013-01-09 08:05:10 -0800289 /*
290 * Create blkgs walking down from blkcg_root to @blkcg, so that all
291 * non-root blkgs have access to their parents.
292 */
293 while (true) {
294 struct blkcg *pos = blkcg;
295 struct blkcg *parent = blkcg_parent(blkcg);
296
297 while (parent && !__blkg_lookup(parent, q, false)) {
298 pos = parent;
299 parent = blkcg_parent(parent);
300 }
301
302 blkg = blkg_create(pos, q, NULL);
303 if (pos == blkcg || IS_ERR(blkg))
304 return blkg;
305 }
Tejun Heo3c96cb32012-04-13 13:11:34 -0700306}
Vivek Goyal31e4c282009-12-03 12:59:42 -0500307
Tejun Heo3c798392012-04-16 13:57:25 -0700308static void blkg_destroy(struct blkcg_gq *blkg)
Tejun Heo72e06c22012-03-05 13:15:00 -0800309{
Tejun Heo3c798392012-04-16 13:57:25 -0700310 struct blkcg *blkcg = blkg->blkcg;
Tejun Heo77ea7332015-08-18 14:55:24 -0700311 struct blkcg_gq *parent = blkg->parent;
Tejun Heof427d902013-01-09 08:05:12 -0800312 int i;
Tejun Heo03aa2642012-03-05 13:15:19 -0800313
Tejun Heo27e1f9d2012-06-05 13:36:44 +0200314 lockdep_assert_held(blkg->q->queue_lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800315 lockdep_assert_held(&blkcg->lock);
Tejun Heo03aa2642012-03-05 13:15:19 -0800316
317 /* Something wrong if we are trying to remove same group twice */
Tejun Heoe8989fa2012-03-05 13:15:20 -0800318 WARN_ON_ONCE(list_empty(&blkg->q_node));
Tejun Heo9f13ef62012-03-05 13:15:21 -0800319 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
Tejun Heoa6371202012-04-19 16:29:24 -0700320
Tejun Heof427d902013-01-09 08:05:12 -0800321 for (i = 0; i < BLKCG_MAX_POLS; i++) {
322 struct blkcg_policy *pol = blkcg_policy[i];
323
324 if (blkg->pd[i] && pol->pd_offline_fn)
Tejun Heoa9520cd2015-08-18 14:55:14 -0700325 pol->pd_offline_fn(blkg->pd[i]);
Tejun Heof427d902013-01-09 08:05:12 -0800326 }
Tejun Heo77ea7332015-08-18 14:55:24 -0700327
328 if (parent) {
329 blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes);
330 blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios);
331 }
332
Tejun Heof427d902013-01-09 08:05:12 -0800333 blkg->online = false;
334
Tejun Heoa6371202012-04-19 16:29:24 -0700335 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
Tejun Heoe8989fa2012-03-05 13:15:20 -0800336 list_del_init(&blkg->q_node);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800337 hlist_del_init_rcu(&blkg->blkcg_node);
Tejun Heo03aa2642012-03-05 13:15:19 -0800338
Tejun Heo03aa2642012-03-05 13:15:19 -0800339 /*
Tejun Heoa6371202012-04-19 16:29:24 -0700340 * Both setting lookup hint to and clearing it from @blkg are done
341 * under queue_lock. If it's not pointing to @blkg now, it never
342 * will. Hint assignment itself can race safely.
343 */
Paul E. McKenneyec6c6762014-02-17 13:35:57 -0800344 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
Tejun Heoa6371202012-04-19 16:29:24 -0700345 rcu_assign_pointer(blkcg->blkg_hint, NULL);
346
347 /*
Tejun Heo03aa2642012-03-05 13:15:19 -0800348 * Put the reference taken at the time of creation so that when all
349 * queues are gone, group can be destroyed.
350 */
351 blkg_put(blkg);
352}
353
Tejun Heo9f13ef62012-03-05 13:15:21 -0800354/**
355 * blkg_destroy_all - destroy all blkgs associated with a request_queue
356 * @q: request_queue of interest
Tejun Heo9f13ef62012-03-05 13:15:21 -0800357 *
Tejun Heo3c96cb32012-04-13 13:11:34 -0700358 * Destroy all blkgs associated with @q.
Tejun Heo9f13ef62012-03-05 13:15:21 -0800359 */
Tejun Heo3c96cb32012-04-13 13:11:34 -0700360static void blkg_destroy_all(struct request_queue *q)
Tejun Heo03aa2642012-03-05 13:15:19 -0800361{
Tejun Heo3c798392012-04-16 13:57:25 -0700362 struct blkcg_gq *blkg, *n;
Tejun Heo72e06c22012-03-05 13:15:00 -0800363
Tejun Heo6d18b002012-04-13 13:11:35 -0700364 lockdep_assert_held(q->queue_lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800365
Tejun Heo9f13ef62012-03-05 13:15:21 -0800366 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
Tejun Heo3c798392012-04-16 13:57:25 -0700367 struct blkcg *blkcg = blkg->blkcg;
Tejun Heo72e06c22012-03-05 13:15:00 -0800368
Tejun Heo9f13ef62012-03-05 13:15:21 -0800369 spin_lock(&blkcg->lock);
370 blkg_destroy(blkg);
371 spin_unlock(&blkcg->lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800372 }
Tejun Heo6fe810b2015-09-05 15:47:36 -0400373
374 q->root_blkg = NULL;
375 q->root_rl.blkg = NULL;
Tejun Heo72e06c22012-03-05 13:15:00 -0800376}
377
Tejun Heo2a4fd072013-05-14 13:52:31 -0700378/*
379 * A group is RCU protected, but having an rcu lock does not mean that one
380 * can access all the fields of blkg and assume these are valid. For
381 * example, don't try to follow throtl_data and request queue links.
382 *
383 * Having a reference to blkg under an rcu allows accesses to only values
384 * local to groups like group stats and group rate limits.
385 */
386void __blkg_release_rcu(struct rcu_head *rcu_head)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800387{
Tejun Heo2a4fd072013-05-14 13:52:31 -0700388 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
Tejun Heodb613672013-05-14 13:52:31 -0700389
Tejun Heo3c547862013-01-09 08:05:10 -0800390 /* release the blkcg and parent blkg refs this blkg has been holding */
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800391 css_put(&blkg->blkcg->css);
Tejun Heoa5049a82014-06-19 17:42:57 -0400392 if (blkg->parent)
Tejun Heo3c547862013-01-09 08:05:10 -0800393 blkg_put(blkg->parent);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800394
Tejun Heoce7acfe2015-05-22 17:13:38 -0400395 wb_congested_put(blkg->wb_congested);
396
Tejun Heo2a4fd072013-05-14 13:52:31 -0700397 blkg_free(blkg);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800398}
Tejun Heo2a4fd072013-05-14 13:52:31 -0700399EXPORT_SYMBOL_GPL(__blkg_release_rcu);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800400
Tejun Heoa0516612012-06-26 15:05:44 -0700401/*
402 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
403 * because the root blkg uses @q->root_rl instead of its own rl.
404 */
405struct request_list *__blk_queue_next_rl(struct request_list *rl,
406 struct request_queue *q)
407{
408 struct list_head *ent;
409 struct blkcg_gq *blkg;
410
411 /*
412 * Determine the current blkg list_head. The first entry is
413 * root_rl which is off @q->blkg_list and mapped to the head.
414 */
415 if (rl == &q->root_rl) {
416 ent = &q->blkg_list;
Jun'ichi Nomura65c77fd2012-10-22 10:15:37 +0900417 /* There are no more block groups, hence no request lists */
418 if (list_empty(ent))
419 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -0700420 } else {
421 blkg = container_of(rl, struct blkcg_gq, rl);
422 ent = &blkg->q_node;
423 }
424
425 /* walk to the next list_head, skip root blkcg */
426 ent = ent->next;
427 if (ent == &q->root_blkg->q_node)
428 ent = ent->next;
429 if (ent == &q->blkg_list)
430 return NULL;
431
432 blkg = container_of(ent, struct blkcg_gq, q_node);
433 return &blkg->rl;
434}
435
Tejun Heo182446d2013-08-08 20:11:24 -0400436static int blkcg_reset_stats(struct cgroup_subsys_state *css,
437 struct cftype *cftype, u64 val)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700438{
Tejun Heo182446d2013-08-08 20:11:24 -0400439 struct blkcg *blkcg = css_to_blkcg(css);
Tejun Heo3c798392012-04-16 13:57:25 -0700440 struct blkcg_gq *blkg;
Tejun Heobc0d6502012-04-13 13:11:26 -0700441 int i;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700442
Tejun Heo838f13b2015-07-09 16:39:47 -0400443 mutex_lock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700444 spin_lock_irq(&blkcg->lock);
Tejun Heo997a0262012-03-08 10:53:58 -0800445
446 /*
447 * Note that stat reset is racy - it doesn't synchronize against
448 * stat updates. This is a debug feature which shouldn't exist
449 * anyway. If you get hit by a race, retry.
450 */
Sasha Levinb67bfe02013-02-27 17:06:00 -0800451 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo77ea7332015-08-18 14:55:24 -0700452 blkg_rwstat_reset(&blkg->stat_bytes);
453 blkg_rwstat_reset(&blkg->stat_ios);
454
Tejun Heo8bd435b2012-04-13 13:11:28 -0700455 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -0700456 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heo549d3aa2012-03-05 13:15:16 -0800457
Tejun Heoa9520cd2015-08-18 14:55:14 -0700458 if (blkg->pd[i] && pol->pd_reset_stats_fn)
459 pol->pd_reset_stats_fn(blkg->pd[i]);
Tejun Heobc0d6502012-04-13 13:11:26 -0700460 }
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700461 }
Vivek Goyalf0bdc8c2011-05-19 15:38:30 -0400462
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700463 spin_unlock_irq(&blkcg->lock);
Tejun Heobc0d6502012-04-13 13:11:26 -0700464 mutex_unlock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700465 return 0;
466}
467
Tejun Heodd165eb2015-08-18 14:55:33 -0700468const char *blkg_dev_name(struct blkcg_gq *blkg)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700469{
Tejun Heod3d32e62012-04-01 14:38:42 -0700470 /* some drivers (floppy) instantiate a queue w/o disk registered */
471 if (blkg->q->backing_dev_info.dev)
472 return dev_name(blkg->q->backing_dev_info.dev);
473 return NULL;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700474}
Tejun Heodd165eb2015-08-18 14:55:33 -0700475EXPORT_SYMBOL_GPL(blkg_dev_name);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700476
Tejun Heod3d32e62012-04-01 14:38:42 -0700477/**
478 * blkcg_print_blkgs - helper for printing per-blkg data
479 * @sf: seq_file to print to
480 * @blkcg: blkcg of interest
481 * @prfill: fill function to print out a blkg
482 * @pol: policy in question
483 * @data: data to be passed to @prfill
484 * @show_total: to print out sum of prfill return values or not
485 *
486 * This function invokes @prfill on each blkg of @blkcg if pd for the
487 * policy specified by @pol exists. @prfill is invoked with @sf, the
Tejun Heo810ecfa2013-01-09 08:05:13 -0800488 * policy data and @data and the matching queue lock held. If @show_total
489 * is %true, the sum of the return values from @prfill is printed with
490 * "Total" label at the end.
Tejun Heod3d32e62012-04-01 14:38:42 -0700491 *
492 * This is to be used to construct print functions for
493 * cftype->read_seq_string method.
494 */
Tejun Heo3c798392012-04-16 13:57:25 -0700495void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
Tejun Heof95a04a2012-04-16 13:57:26 -0700496 u64 (*prfill)(struct seq_file *,
497 struct blkg_policy_data *, int),
Tejun Heo3c798392012-04-16 13:57:25 -0700498 const struct blkcg_policy *pol, int data,
Tejun Heoec399342012-04-13 13:11:27 -0700499 bool show_total)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400500{
Tejun Heo3c798392012-04-16 13:57:25 -0700501 struct blkcg_gq *blkg;
Tejun Heod3d32e62012-04-01 14:38:42 -0700502 u64 total = 0;
503
Tejun Heo810ecfa2013-01-09 08:05:13 -0800504 rcu_read_lock();
Linus Torvaldsee89f812013-02-28 12:52:24 -0800505 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo810ecfa2013-01-09 08:05:13 -0800506 spin_lock_irq(blkg->q->queue_lock);
Tejun Heoa2b16932012-04-13 13:11:33 -0700507 if (blkcg_policy_enabled(blkg->q, pol))
Tejun Heof95a04a2012-04-16 13:57:26 -0700508 total += prfill(sf, blkg->pd[pol->plid], data);
Tejun Heo810ecfa2013-01-09 08:05:13 -0800509 spin_unlock_irq(blkg->q->queue_lock);
510 }
511 rcu_read_unlock();
Tejun Heod3d32e62012-04-01 14:38:42 -0700512
513 if (show_total)
514 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
515}
Tejun Heo829fdb52012-04-01 14:38:43 -0700516EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
Tejun Heod3d32e62012-04-01 14:38:42 -0700517
518/**
519 * __blkg_prfill_u64 - prfill helper for a single u64 value
520 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700521 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700522 * @v: value to print
523 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700524 * Print @v to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700525 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700526u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
Tejun Heod3d32e62012-04-01 14:38:42 -0700527{
Tejun Heof95a04a2012-04-16 13:57:26 -0700528 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700529
530 if (!dname)
531 return 0;
532
533 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
534 return v;
535}
Tejun Heo829fdb52012-04-01 14:38:43 -0700536EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
Tejun Heod3d32e62012-04-01 14:38:42 -0700537
538/**
539 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
540 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700541 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700542 * @rwstat: rwstat to print
543 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700544 * Print @rwstat to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700545 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700546u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
Tejun Heo829fdb52012-04-01 14:38:43 -0700547 const struct blkg_rwstat *rwstat)
Tejun Heod3d32e62012-04-01 14:38:42 -0700548{
549 static const char *rwstr[] = {
550 [BLKG_RWSTAT_READ] = "Read",
551 [BLKG_RWSTAT_WRITE] = "Write",
552 [BLKG_RWSTAT_SYNC] = "Sync",
553 [BLKG_RWSTAT_ASYNC] = "Async",
554 };
Tejun Heof95a04a2012-04-16 13:57:26 -0700555 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700556 u64 v;
557 int i;
558
559 if (!dname)
560 return 0;
561
562 for (i = 0; i < BLKG_RWSTAT_NR; i++)
563 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700564 (unsigned long long)atomic64_read(&rwstat->aux_cnt[i]));
Tejun Heod3d32e62012-04-01 14:38:42 -0700565
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700566 v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) +
567 atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]);
Tejun Heod3d32e62012-04-01 14:38:42 -0700568 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
569 return v;
570}
Tejun Heob50da392013-01-09 08:05:12 -0800571EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700572
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700573/**
574 * blkg_prfill_stat - prfill callback for blkg_stat
575 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700576 * @pd: policy private data of interest
577 * @off: offset to the blkg_stat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700578 *
579 * prfill callback for printing a blkg_stat.
580 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700581u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700582{
Tejun Heof95a04a2012-04-16 13:57:26 -0700583 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
Tejun Heod3d32e62012-04-01 14:38:42 -0700584}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700585EXPORT_SYMBOL_GPL(blkg_prfill_stat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700586
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700587/**
588 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
589 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700590 * @pd: policy private data of interest
591 * @off: offset to the blkg_rwstat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700592 *
593 * prfill callback for printing a blkg_rwstat.
594 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700595u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
596 int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700597{
Tejun Heof95a04a2012-04-16 13:57:26 -0700598 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
Tejun Heod3d32e62012-04-01 14:38:42 -0700599
Tejun Heof95a04a2012-04-16 13:57:26 -0700600 return __blkg_prfill_rwstat(sf, pd, &rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700601}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700602EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700603
Tejun Heo77ea7332015-08-18 14:55:24 -0700604static u64 blkg_prfill_rwstat_field(struct seq_file *sf,
605 struct blkg_policy_data *pd, int off)
606{
607 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off);
608
609 return __blkg_prfill_rwstat(sf, pd, &rwstat);
610}
611
612/**
613 * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes
614 * @sf: seq_file to print to
615 * @v: unused
616 *
617 * To be used as cftype->seq_show to print blkg->stat_bytes.
618 * cftype->private must be set to the blkcg_policy.
619 */
620int blkg_print_stat_bytes(struct seq_file *sf, void *v)
621{
622 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
623 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
624 offsetof(struct blkcg_gq, stat_bytes), true);
625 return 0;
626}
627EXPORT_SYMBOL_GPL(blkg_print_stat_bytes);
628
629/**
630 * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios
631 * @sf: seq_file to print to
632 * @v: unused
633 *
634 * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private
635 * must be set to the blkcg_policy.
636 */
637int blkg_print_stat_ios(struct seq_file *sf, void *v)
638{
639 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
640 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
641 offsetof(struct blkcg_gq, stat_ios), true);
642 return 0;
643}
644EXPORT_SYMBOL_GPL(blkg_print_stat_ios);
645
646static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf,
647 struct blkg_policy_data *pd,
648 int off)
649{
650 struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg,
651 NULL, off);
652 return __blkg_prfill_rwstat(sf, pd, &rwstat);
653}
654
655/**
656 * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes
657 * @sf: seq_file to print to
658 * @v: unused
659 */
660int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v)
661{
662 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
663 blkg_prfill_rwstat_field_recursive,
664 (void *)seq_cft(sf)->private,
665 offsetof(struct blkcg_gq, stat_bytes), true);
666 return 0;
667}
668EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive);
669
670/**
671 * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios
672 * @sf: seq_file to print to
673 * @v: unused
674 */
675int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v)
676{
677 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
678 blkg_prfill_rwstat_field_recursive,
679 (void *)seq_cft(sf)->private,
680 offsetof(struct blkcg_gq, stat_ios), true);
681 return 0;
682}
683EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive);
684
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700685/**
Tejun Heo16b3de62013-01-09 08:05:12 -0800686 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
Tejun Heof12c74c2015-08-18 14:55:23 -0700687 * @blkg: blkg of interest
688 * @pol: blkcg_policy which contains the blkg_stat
689 * @off: offset to the blkg_stat in blkg_policy_data or @blkg
Tejun Heo16b3de62013-01-09 08:05:12 -0800690 *
Tejun Heof12c74c2015-08-18 14:55:23 -0700691 * Collect the blkg_stat specified by @blkg, @pol and @off and all its
692 * online descendants and their aux counts. The caller must be holding the
693 * queue lock for online tests.
694 *
695 * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is
696 * at @off bytes into @blkg's blkg_policy_data of the policy.
Tejun Heo16b3de62013-01-09 08:05:12 -0800697 */
Tejun Heof12c74c2015-08-18 14:55:23 -0700698u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
699 struct blkcg_policy *pol, int off)
Tejun Heo16b3de62013-01-09 08:05:12 -0800700{
Tejun Heo16b3de62013-01-09 08:05:12 -0800701 struct blkcg_gq *pos_blkg;
Tejun Heo492eb212013-08-08 20:11:25 -0400702 struct cgroup_subsys_state *pos_css;
Tejun Heobd8815a2013-08-08 20:11:27 -0400703 u64 sum = 0;
Tejun Heo16b3de62013-01-09 08:05:12 -0800704
Tejun Heof12c74c2015-08-18 14:55:23 -0700705 lockdep_assert_held(blkg->q->queue_lock);
Tejun Heo16b3de62013-01-09 08:05:12 -0800706
Tejun Heo16b3de62013-01-09 08:05:12 -0800707 rcu_read_lock();
Tejun Heof12c74c2015-08-18 14:55:23 -0700708 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
709 struct blkg_stat *stat;
Tejun Heo16b3de62013-01-09 08:05:12 -0800710
Tejun Heof12c74c2015-08-18 14:55:23 -0700711 if (!pos_blkg->online)
712 continue;
713
714 if (pol)
715 stat = (void *)blkg_to_pd(pos_blkg, pol) + off;
716 else
717 stat = (void *)blkg + off;
718
719 sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt);
Tejun Heo16b3de62013-01-09 08:05:12 -0800720 }
721 rcu_read_unlock();
722
723 return sum;
724}
725EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
726
727/**
728 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
Tejun Heof12c74c2015-08-18 14:55:23 -0700729 * @blkg: blkg of interest
730 * @pol: blkcg_policy which contains the blkg_rwstat
731 * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg
Tejun Heo16b3de62013-01-09 08:05:12 -0800732 *
Tejun Heof12c74c2015-08-18 14:55:23 -0700733 * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its
734 * online descendants and their aux counts. The caller must be holding the
735 * queue lock for online tests.
736 *
737 * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it
738 * is at @off bytes into @blkg's blkg_policy_data of the policy.
Tejun Heo16b3de62013-01-09 08:05:12 -0800739 */
Tejun Heof12c74c2015-08-18 14:55:23 -0700740struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
741 struct blkcg_policy *pol, int off)
Tejun Heo16b3de62013-01-09 08:05:12 -0800742{
Tejun Heo16b3de62013-01-09 08:05:12 -0800743 struct blkcg_gq *pos_blkg;
Tejun Heo492eb212013-08-08 20:11:25 -0400744 struct cgroup_subsys_state *pos_css;
Tejun Heobd8815a2013-08-08 20:11:27 -0400745 struct blkg_rwstat sum = { };
Tejun Heo16b3de62013-01-09 08:05:12 -0800746 int i;
747
Tejun Heof12c74c2015-08-18 14:55:23 -0700748 lockdep_assert_held(blkg->q->queue_lock);
Tejun Heo16b3de62013-01-09 08:05:12 -0800749
Tejun Heo16b3de62013-01-09 08:05:12 -0800750 rcu_read_lock();
Tejun Heof12c74c2015-08-18 14:55:23 -0700751 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
Tejun Heo3a7faea2015-08-18 14:55:26 -0700752 struct blkg_rwstat *rwstat;
Tejun Heo16b3de62013-01-09 08:05:12 -0800753
754 if (!pos_blkg->online)
755 continue;
756
Tejun Heof12c74c2015-08-18 14:55:23 -0700757 if (pol)
758 rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off;
759 else
760 rwstat = (void *)pos_blkg + off;
761
Tejun Heo16b3de62013-01-09 08:05:12 -0800762 for (i = 0; i < BLKG_RWSTAT_NR; i++)
Tejun Heo3a7faea2015-08-18 14:55:26 -0700763 atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) +
764 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]),
765 &sum.aux_cnt[i]);
Tejun Heo16b3de62013-01-09 08:05:12 -0800766 }
767 rcu_read_unlock();
768
769 return sum;
770}
771EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
772
773/**
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700774 * blkg_conf_prep - parse and prepare for per-blkg config update
775 * @blkcg: target block cgroup
Tejun Heoda8b0662012-04-13 13:11:29 -0700776 * @pol: target policy
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700777 * @input: input string
778 * @ctx: blkg_conf_ctx to be filled
779 *
780 * Parse per-blkg config update from @input and initialize @ctx with the
Tejun Heo36aa9e52015-08-18 14:55:31 -0700781 * result. @ctx->blkg points to the blkg to be updated and @ctx->body the
782 * part of @input following MAJ:MIN. This function returns with RCU read
783 * lock and queue lock held and must be paired with blkg_conf_finish().
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700784 */
Tejun Heo3c798392012-04-16 13:57:25 -0700785int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
Tejun Heo36aa9e52015-08-18 14:55:31 -0700786 char *input, struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700787 __acquires(rcu) __acquires(disk->queue->queue_lock)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800788{
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700789 struct gendisk *disk;
Tejun Heo3c798392012-04-16 13:57:25 -0700790 struct blkcg_gq *blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700791 unsigned int major, minor;
Tejun Heo36aa9e52015-08-18 14:55:31 -0700792 int key_len, part, ret;
793 char *body;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800794
Tejun Heo36aa9e52015-08-18 14:55:31 -0700795 if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
Tejun Heo726fa692012-04-01 14:38:43 -0700796 return -EINVAL;
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700797
Tejun Heo36aa9e52015-08-18 14:55:31 -0700798 body = input + key_len;
799 if (!isspace(*body))
800 return -EINVAL;
801 body = skip_spaces(body);
802
Tejun Heo726fa692012-04-01 14:38:43 -0700803 disk = get_gendisk(MKDEV(major, minor), &part);
Tejun Heo5f6c2d22015-07-22 18:05:53 -0400804 if (!disk)
Tejun Heo20386ce2015-08-18 14:55:28 -0700805 return -ENODEV;
Tejun Heo5f6c2d22015-07-22 18:05:53 -0400806 if (part) {
807 put_disk(disk);
Tejun Heo20386ce2015-08-18 14:55:28 -0700808 return -ENODEV;
Tejun Heo5f6c2d22015-07-22 18:05:53 -0400809 }
Tejun Heoe56da7e2012-03-05 13:15:07 -0800810
811 rcu_read_lock();
Tejun Heo4bfd4822012-03-05 13:15:08 -0800812 spin_lock_irq(disk->queue->queue_lock);
Tejun Heoda8b0662012-04-13 13:11:29 -0700813
Tejun Heoa2b16932012-04-13 13:11:33 -0700814 if (blkcg_policy_enabled(disk->queue, pol))
Tejun Heo3c96cb32012-04-13 13:11:34 -0700815 blkg = blkg_lookup_create(blkcg, disk->queue);
Tejun Heoa2b16932012-04-13 13:11:33 -0700816 else
Tejun Heo20386ce2015-08-18 14:55:28 -0700817 blkg = ERR_PTR(-EOPNOTSUPP);
Tejun Heoe56da7e2012-03-05 13:15:07 -0800818
Tejun Heo4bfd4822012-03-05 13:15:08 -0800819 if (IS_ERR(blkg)) {
820 ret = PTR_ERR(blkg);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700821 rcu_read_unlock();
Tejun Heoda8b0662012-04-13 13:11:29 -0700822 spin_unlock_irq(disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700823 put_disk(disk);
824 /*
825 * If queue was bypassing, we should retry. Do so after a
826 * short msleep(). It isn't strictly necessary but queue
827 * can be bypassing for some time and it's always nice to
828 * avoid busy looping.
829 */
830 if (ret == -EBUSY) {
831 msleep(10);
832 ret = restart_syscall();
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400833 }
Tejun Heo726fa692012-04-01 14:38:43 -0700834 return ret;
Vivek Goyal062a6442010-09-15 17:06:33 -0400835 }
Tejun Heoe56da7e2012-03-05 13:15:07 -0800836
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700837 ctx->disk = disk;
838 ctx->blkg = blkg;
Tejun Heo36aa9e52015-08-18 14:55:31 -0700839 ctx->body = body;
Tejun Heo726fa692012-04-01 14:38:43 -0700840 return 0;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800841}
Tejun Heo829fdb52012-04-01 14:38:43 -0700842EXPORT_SYMBOL_GPL(blkg_conf_prep);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800843
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700844/**
845 * blkg_conf_finish - finish up per-blkg config update
846 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
847 *
848 * Finish up after per-blkg config update. This function must be paired
849 * with blkg_conf_prep().
850 */
Tejun Heo829fdb52012-04-01 14:38:43 -0700851void blkg_conf_finish(struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700852 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800853{
Tejun Heoda8b0662012-04-13 13:11:29 -0700854 spin_unlock_irq(ctx->disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700855 rcu_read_unlock();
856 put_disk(ctx->disk);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800857}
Tejun Heo829fdb52012-04-01 14:38:43 -0700858EXPORT_SYMBOL_GPL(blkg_conf_finish);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800859
Tejun Heo2ee867dc2015-08-18 14:55:34 -0700860static int blkcg_print_stat(struct seq_file *sf, void *v)
861{
862 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
863 struct blkcg_gq *blkg;
864
865 rcu_read_lock();
866
867 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
868 const char *dname;
869 struct blkg_rwstat rwstat;
870 u64 rbytes, wbytes, rios, wios;
871
872 dname = blkg_dev_name(blkg);
873 if (!dname)
874 continue;
875
876 spin_lock_irq(blkg->q->queue_lock);
877
878 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
879 offsetof(struct blkcg_gq, stat_bytes));
880 rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
881 wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
882
883 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
884 offsetof(struct blkcg_gq, stat_ios));
885 rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
886 wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
887
888 spin_unlock_irq(blkg->q->queue_lock);
889
890 if (rbytes || wbytes || rios || wios)
891 seq_printf(sf, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n",
892 dname, rbytes, wbytes, rios, wios);
893 }
894
895 rcu_read_unlock();
896 return 0;
897}
898
899struct cftype blkcg_files[] = {
900 {
901 .name = "stat",
902 .seq_show = blkcg_print_stat,
903 },
904 { } /* terminate */
905};
906
Tejun Heo880f50e2015-08-18 14:55:30 -0700907struct cftype blkcg_legacy_files[] = {
Vivek Goyal31e4c282009-12-03 12:59:42 -0500908 {
Divyesh Shah84c124d2010-04-09 08:31:19 +0200909 .name = "reset_stats",
Tejun Heo3c798392012-04-16 13:57:25 -0700910 .write_u64 = blkcg_reset_stats,
Vivek Goyal22084192009-12-03 12:59:49 -0500911 },
Tejun Heo4baf6e32012-04-01 12:09:55 -0700912 { } /* terminate */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500913};
914
Tejun Heo9f13ef62012-03-05 13:15:21 -0800915/**
Tejun Heo92fb9742012-11-19 08:13:38 -0800916 * blkcg_css_offline - cgroup css_offline callback
Tejun Heoeb954192013-08-08 20:11:23 -0400917 * @css: css of interest
Tejun Heo9f13ef62012-03-05 13:15:21 -0800918 *
Tejun Heoeb954192013-08-08 20:11:23 -0400919 * This function is called when @css is about to go away and responsible
920 * for shooting down all blkgs associated with @css. blkgs should be
Tejun Heo9f13ef62012-03-05 13:15:21 -0800921 * removed while holding both q and blkcg locks. As blkcg lock is nested
922 * inside q lock, this function performs reverse double lock dancing.
923 *
924 * This is the blkcg counterpart of ioc_release_fn().
925 */
Tejun Heoeb954192013-08-08 20:11:23 -0400926static void blkcg_css_offline(struct cgroup_subsys_state *css)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500927{
Tejun Heoeb954192013-08-08 20:11:23 -0400928 struct blkcg *blkcg = css_to_blkcg(css);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500929
Tejun Heo9f13ef62012-03-05 13:15:21 -0800930 spin_lock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800931
Tejun Heo9f13ef62012-03-05 13:15:21 -0800932 while (!hlist_empty(&blkcg->blkg_list)) {
Tejun Heo3c798392012-04-16 13:57:25 -0700933 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
934 struct blkcg_gq, blkcg_node);
Tejun Heoc875f4d2012-03-05 13:15:22 -0800935 struct request_queue *q = blkg->q;
Vivek Goyalb1c35762009-12-03 12:59:47 -0500936
Tejun Heo9f13ef62012-03-05 13:15:21 -0800937 if (spin_trylock(q->queue_lock)) {
938 blkg_destroy(blkg);
939 spin_unlock(q->queue_lock);
940 } else {
941 spin_unlock_irq(&blkcg->lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800942 cpu_relax();
Dan Carpentera5567932012-03-29 20:57:08 +0200943 spin_lock_irq(&blkcg->lock);
Jens Axboe0f3942a2010-05-03 14:28:55 +0200944 }
Tejun Heo9f13ef62012-03-05 13:15:21 -0800945 }
Jens Axboe0f3942a2010-05-03 14:28:55 +0200946
Tejun Heo9f13ef62012-03-05 13:15:21 -0800947 spin_unlock_irq(&blkcg->lock);
Tejun Heo52ebea72015-05-22 17:13:37 -0400948
949 wb_blkcg_offline(blkcg);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800950}
951
Tejun Heoeb954192013-08-08 20:11:23 -0400952static void blkcg_css_free(struct cgroup_subsys_state *css)
Tejun Heo7ee9c562012-03-05 13:15:11 -0800953{
Tejun Heoeb954192013-08-08 20:11:23 -0400954 struct blkcg *blkcg = css_to_blkcg(css);
Tejun Heobc915e62015-08-18 14:55:08 -0700955 int i;
Tejun Heo7ee9c562012-03-05 13:15:11 -0800956
Tejun Heo7876f932015-07-09 16:39:49 -0400957 mutex_lock(&blkcg_pol_mutex);
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700958
Tejun Heo7876f932015-07-09 16:39:49 -0400959 list_del(&blkcg->all_blkcgs_node);
Tejun Heo7876f932015-07-09 16:39:49 -0400960
Tejun Heobc915e62015-08-18 14:55:08 -0700961 for (i = 0; i < BLKCG_MAX_POLS; i++)
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700962 if (blkcg->cpd[i])
963 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
964
965 mutex_unlock(&blkcg_pol_mutex);
966
Tejun Heobc915e62015-08-18 14:55:08 -0700967 kfree(blkcg);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500968}
969
Tejun Heoeb954192013-08-08 20:11:23 -0400970static struct cgroup_subsys_state *
971blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500972{
Tejun Heo3c798392012-04-16 13:57:25 -0700973 struct blkcg *blkcg;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200974 struct cgroup_subsys_state *ret;
975 int i;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500976
Tejun Heo7876f932015-07-09 16:39:49 -0400977 mutex_lock(&blkcg_pol_mutex);
978
Tejun Heoeb954192013-08-08 20:11:23 -0400979 if (!parent_css) {
Tejun Heo3c798392012-04-16 13:57:25 -0700980 blkcg = &blkcg_root;
Tejun Heobc915e62015-08-18 14:55:08 -0700981 } else {
982 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
983 if (!blkcg) {
984 ret = ERR_PTR(-ENOMEM);
985 goto free_blkcg;
986 }
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200987 }
Vivek Goyal31e4c282009-12-03 12:59:42 -0500988
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200989 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
990 struct blkcg_policy *pol = blkcg_policy[i];
991 struct blkcg_policy_data *cpd;
992
993 /*
994 * If the policy hasn't been attached yet, wait for it
995 * to be attached before doing anything else. Otherwise,
996 * check if the policy requires any specific per-cgroup
997 * data: if it does, allocate and initialize it.
998 */
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700999 if (!pol || !pol->cpd_alloc_fn)
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001000 continue;
1001
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001002 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001003 if (!cpd) {
1004 ret = ERR_PTR(-ENOMEM);
1005 goto free_pd_blkcg;
1006 }
Tejun Heo81437642015-08-18 14:55:15 -07001007 blkcg->cpd[i] = cpd;
1008 cpd->blkcg = blkcg;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001009 cpd->plid = i;
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001010 if (pol->cpd_init_fn)
1011 pol->cpd_init_fn(cpd);
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001012 }
1013
Vivek Goyal31e4c282009-12-03 12:59:42 -05001014 spin_lock_init(&blkcg->lock);
Tejun Heod93a11f2015-08-18 14:55:01 -07001015 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT);
Vivek Goyal31e4c282009-12-03 12:59:42 -05001016 INIT_HLIST_HEAD(&blkcg->blkg_list);
Tejun Heo52ebea72015-05-22 17:13:37 -04001017#ifdef CONFIG_CGROUP_WRITEBACK
1018 INIT_LIST_HEAD(&blkcg->cgwb_list);
1019#endif
Tejun Heo7876f932015-07-09 16:39:49 -04001020 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
1021
1022 mutex_unlock(&blkcg_pol_mutex);
Vivek Goyal31e4c282009-12-03 12:59:42 -05001023 return &blkcg->css;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001024
1025free_pd_blkcg:
1026 for (i--; i >= 0; i--)
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001027 if (blkcg->cpd[i])
1028 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001029free_blkcg:
1030 kfree(blkcg);
Tejun Heo7876f932015-07-09 16:39:49 -04001031 mutex_unlock(&blkcg_pol_mutex);
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001032 return ret;
Vivek Goyal31e4c282009-12-03 12:59:42 -05001033}
1034
Tejun Heo5efd6112012-03-05 13:15:12 -08001035/**
1036 * blkcg_init_queue - initialize blkcg part of request queue
1037 * @q: request_queue to initialize
1038 *
1039 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
1040 * part of new request_queue @q.
1041 *
1042 * RETURNS:
1043 * 0 on success, -errno on failure.
1044 */
1045int blkcg_init_queue(struct request_queue *q)
1046{
Tejun Heoec13b1d2015-05-22 17:13:19 -04001047 struct blkcg_gq *new_blkg, *blkg;
1048 bool preloaded;
1049 int ret;
Tejun Heo5efd6112012-03-05 13:15:12 -08001050
Tejun Heoec13b1d2015-05-22 17:13:19 -04001051 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
1052 if (!new_blkg)
1053 return -ENOMEM;
1054
1055 preloaded = !radix_tree_preload(GFP_KERNEL);
1056
1057 /*
1058 * Make sure the root blkg exists and count the existing blkgs. As
1059 * @q is bypassing at this point, blkg_lookup_create() can't be
1060 * used. Open code insertion.
1061 */
1062 rcu_read_lock();
1063 spin_lock_irq(q->queue_lock);
1064 blkg = blkg_create(&blkcg_root, q, new_blkg);
1065 spin_unlock_irq(q->queue_lock);
1066 rcu_read_unlock();
1067
1068 if (preloaded)
1069 radix_tree_preload_end();
1070
1071 if (IS_ERR(blkg)) {
Tejun Heo994b7832015-08-18 14:55:07 -07001072 blkg_free(new_blkg);
Tejun Heoec13b1d2015-05-22 17:13:19 -04001073 return PTR_ERR(blkg);
1074 }
1075
1076 q->root_blkg = blkg;
1077 q->root_rl.blkg = blkg;
1078
1079 ret = blk_throtl_init(q);
1080 if (ret) {
1081 spin_lock_irq(q->queue_lock);
1082 blkg_destroy_all(q);
1083 spin_unlock_irq(q->queue_lock);
1084 }
1085 return ret;
Tejun Heo5efd6112012-03-05 13:15:12 -08001086}
1087
1088/**
1089 * blkcg_drain_queue - drain blkcg part of request_queue
1090 * @q: request_queue to drain
1091 *
1092 * Called from blk_drain_queue(). Responsible for draining blkcg part.
1093 */
1094void blkcg_drain_queue(struct request_queue *q)
1095{
1096 lockdep_assert_held(q->queue_lock);
1097
Tejun Heo0b462c82014-07-05 18:43:21 -04001098 /*
1099 * @q could be exiting and already have destroyed all blkgs as
1100 * indicated by NULL root_blkg. If so, don't confuse policies.
1101 */
1102 if (!q->root_blkg)
1103 return;
1104
Tejun Heo5efd6112012-03-05 13:15:12 -08001105 blk_throtl_drain(q);
1106}
1107
1108/**
1109 * blkcg_exit_queue - exit and release blkcg part of request_queue
1110 * @q: request_queue being released
1111 *
1112 * Called from blk_release_queue(). Responsible for exiting blkcg part.
1113 */
1114void blkcg_exit_queue(struct request_queue *q)
1115{
Tejun Heo6d18b002012-04-13 13:11:35 -07001116 spin_lock_irq(q->queue_lock);
Tejun Heo3c96cb32012-04-13 13:11:34 -07001117 blkg_destroy_all(q);
Tejun Heo6d18b002012-04-13 13:11:35 -07001118 spin_unlock_irq(q->queue_lock);
1119
Tejun Heo5efd6112012-03-05 13:15:12 -08001120 blk_throtl_exit(q);
1121}
1122
Vivek Goyal31e4c282009-12-03 12:59:42 -05001123/*
1124 * We cannot support shared io contexts, as we have no mean to support
1125 * two tasks with the same ioc in two different groups without major rework
1126 * of the main cic data structures. For now we allow a task to change
1127 * its cgroup only if it's the only owner of its ioc.
1128 */
Tejun Heoeb954192013-08-08 20:11:23 -04001129static int blkcg_can_attach(struct cgroup_subsys_state *css,
1130 struct cgroup_taskset *tset)
Vivek Goyal31e4c282009-12-03 12:59:42 -05001131{
Tejun Heobb9d97b2011-12-12 18:12:21 -08001132 struct task_struct *task;
Vivek Goyal31e4c282009-12-03 12:59:42 -05001133 struct io_context *ioc;
1134 int ret = 0;
1135
1136 /* task_lock() is needed to avoid races with exit_io_context() */
Tejun Heo924f0d92014-02-13 06:58:41 -05001137 cgroup_taskset_for_each(task, tset) {
Tejun Heobb9d97b2011-12-12 18:12:21 -08001138 task_lock(task);
1139 ioc = task->io_context;
1140 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1141 ret = -EINVAL;
1142 task_unlock(task);
1143 if (ret)
1144 break;
1145 }
Vivek Goyal31e4c282009-12-03 12:59:42 -05001146 return ret;
1147}
1148
Tejun Heo69d7fde2015-08-18 14:55:36 -07001149static void blkcg_bind(struct cgroup_subsys_state *root_css)
1150{
1151 int i;
1152
1153 mutex_lock(&blkcg_pol_mutex);
1154
1155 for (i = 0; i < BLKCG_MAX_POLS; i++) {
1156 struct blkcg_policy *pol = blkcg_policy[i];
1157 struct blkcg *blkcg;
1158
1159 if (!pol || !pol->cpd_bind_fn)
1160 continue;
1161
1162 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node)
1163 if (blkcg->cpd[pol->plid])
1164 pol->cpd_bind_fn(blkcg->cpd[pol->plid]);
1165 }
1166 mutex_unlock(&blkcg_pol_mutex);
1167}
1168
Tejun Heoc165b3e2015-08-18 14:55:29 -07001169struct cgroup_subsys io_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08001170 .css_alloc = blkcg_css_alloc,
1171 .css_offline = blkcg_css_offline,
1172 .css_free = blkcg_css_free,
Tejun Heo3c798392012-04-16 13:57:25 -07001173 .can_attach = blkcg_can_attach,
Tejun Heo69d7fde2015-08-18 14:55:36 -07001174 .bind = blkcg_bind,
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001175 .dfl_cftypes = blkcg_files,
Tejun Heo880f50e2015-08-18 14:55:30 -07001176 .legacy_cftypes = blkcg_legacy_files,
Tejun Heoc165b3e2015-08-18 14:55:29 -07001177 .legacy_name = "blkio",
Tejun Heo1ced9532014-07-08 18:02:57 -04001178#ifdef CONFIG_MEMCG
1179 /*
1180 * This ensures that, if available, memcg is automatically enabled
1181 * together on the default hierarchy so that the owner cgroup can
1182 * be retrieved from writeback pages.
1183 */
1184 .depends_on = 1 << memory_cgrp_id,
1185#endif
Tejun Heo676f7c82012-04-01 12:09:55 -07001186};
Tejun Heoc165b3e2015-08-18 14:55:29 -07001187EXPORT_SYMBOL_GPL(io_cgrp_subsys);
Tejun Heo676f7c82012-04-01 12:09:55 -07001188
Tejun Heo8bd435b2012-04-13 13:11:28 -07001189/**
Tejun Heoa2b16932012-04-13 13:11:33 -07001190 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1191 * @q: request_queue of interest
1192 * @pol: blkcg policy to activate
1193 *
1194 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1195 * bypass mode to populate its blkgs with policy_data for @pol.
1196 *
1197 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1198 * from IO path. Update of each blkg is protected by both queue and blkcg
1199 * locks so that holding either lock and testing blkcg_policy_enabled() is
1200 * always enough for dereferencing policy data.
1201 *
1202 * The caller is responsible for synchronizing [de]activations and policy
1203 * [un]registerations. Returns 0 on success, -errno on failure.
1204 */
1205int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -07001206 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -07001207{
Tejun Heo4c55f4f2015-08-18 14:55:09 -07001208 struct blkg_policy_data *pd_prealloc = NULL;
Tejun Heoec13b1d2015-05-22 17:13:19 -04001209 struct blkcg_gq *blkg;
Tejun Heo4c55f4f2015-08-18 14:55:09 -07001210 int ret;
Tejun Heoa2b16932012-04-13 13:11:33 -07001211
1212 if (blkcg_policy_enabled(q, pol))
1213 return 0;
1214
1215 blk_queue_bypass_start(q);
Tejun Heo4c55f4f2015-08-18 14:55:09 -07001216pd_prealloc:
1217 if (!pd_prealloc) {
Tejun Heo001bea72015-08-18 14:55:11 -07001218 pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node);
Tejun Heo4c55f4f2015-08-18 14:55:09 -07001219 if (!pd_prealloc) {
Tejun Heoa2b16932012-04-13 13:11:33 -07001220 ret = -ENOMEM;
Tejun Heo4c55f4f2015-08-18 14:55:09 -07001221 goto out_bypass_end;
Tejun Heoa2b16932012-04-13 13:11:33 -07001222 }
Tejun Heoa2b16932012-04-13 13:11:33 -07001223 }
1224
Tejun Heoa2b16932012-04-13 13:11:33 -07001225 spin_lock_irq(q->queue_lock);
1226
1227 list_for_each_entry(blkg, &q->blkg_list, q_node) {
Tejun Heo4c55f4f2015-08-18 14:55:09 -07001228 struct blkg_policy_data *pd;
Tejun Heoa2b16932012-04-13 13:11:33 -07001229
Tejun Heo4c55f4f2015-08-18 14:55:09 -07001230 if (blkg->pd[pol->plid])
1231 continue;
1232
Tejun Heo001bea72015-08-18 14:55:11 -07001233 pd = pol->pd_alloc_fn(GFP_NOWAIT, q->node);
Tejun Heo4c55f4f2015-08-18 14:55:09 -07001234 if (!pd)
1235 swap(pd, pd_prealloc);
1236 if (!pd) {
1237 spin_unlock_irq(q->queue_lock);
1238 goto pd_prealloc;
1239 }
Tejun Heoa2b16932012-04-13 13:11:33 -07001240
1241 blkg->pd[pol->plid] = pd;
1242 pd->blkg = blkg;
Tejun Heob276a872013-01-09 08:05:12 -08001243 pd->plid = pol->plid;
Tejun Heo3e418712015-08-18 14:55:10 -07001244 if (pol->pd_init_fn)
Tejun Heoa9520cd2015-08-18 14:55:14 -07001245 pol->pd_init_fn(pd);
Tejun Heoa2b16932012-04-13 13:11:33 -07001246 }
1247
1248 __set_bit(pol->plid, q->blkcg_pols);
1249 ret = 0;
Tejun Heo4c55f4f2015-08-18 14:55:09 -07001250
Tejun Heoa2b16932012-04-13 13:11:33 -07001251 spin_unlock_irq(q->queue_lock);
Tejun Heo4c55f4f2015-08-18 14:55:09 -07001252out_bypass_end:
Tejun Heoa2b16932012-04-13 13:11:33 -07001253 blk_queue_bypass_end(q);
Tejun Heo001bea72015-08-18 14:55:11 -07001254 if (pd_prealloc)
1255 pol->pd_free_fn(pd_prealloc);
Tejun Heoa2b16932012-04-13 13:11:33 -07001256 return ret;
1257}
1258EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1259
1260/**
1261 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1262 * @q: request_queue of interest
1263 * @pol: blkcg policy to deactivate
1264 *
1265 * Deactivate @pol on @q. Follows the same synchronization rules as
1266 * blkcg_activate_policy().
1267 */
1268void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -07001269 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -07001270{
Tejun Heo3c798392012-04-16 13:57:25 -07001271 struct blkcg_gq *blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -07001272
1273 if (!blkcg_policy_enabled(q, pol))
1274 return;
1275
1276 blk_queue_bypass_start(q);
1277 spin_lock_irq(q->queue_lock);
1278
1279 __clear_bit(pol->plid, q->blkcg_pols);
1280
1281 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1282 /* grab blkcg lock too while removing @pd from @blkg */
1283 spin_lock(&blkg->blkcg->lock);
1284
Tejun Heo001bea72015-08-18 14:55:11 -07001285 if (blkg->pd[pol->plid]) {
Tejun Heoa9520cd2015-08-18 14:55:14 -07001286 if (pol->pd_offline_fn)
1287 pol->pd_offline_fn(blkg->pd[pol->plid]);
Tejun Heo001bea72015-08-18 14:55:11 -07001288 pol->pd_free_fn(blkg->pd[pol->plid]);
1289 blkg->pd[pol->plid] = NULL;
1290 }
Tejun Heoa2b16932012-04-13 13:11:33 -07001291
1292 spin_unlock(&blkg->blkcg->lock);
1293 }
1294
1295 spin_unlock_irq(q->queue_lock);
1296 blk_queue_bypass_end(q);
1297}
1298EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1299
1300/**
Tejun Heo3c798392012-04-16 13:57:25 -07001301 * blkcg_policy_register - register a blkcg policy
1302 * @pol: blkcg policy to register
Tejun Heo8bd435b2012-04-13 13:11:28 -07001303 *
Tejun Heo3c798392012-04-16 13:57:25 -07001304 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1305 * successful registration. Returns 0 on success and -errno on failure.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001306 */
Jens Axboed5bf0292014-06-22 16:31:56 -06001307int blkcg_policy_register(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001308{
Tejun Heo06b285b2015-07-09 16:39:50 -04001309 struct blkcg *blkcg;
Tejun Heo8bd435b2012-04-13 13:11:28 -07001310 int i, ret;
Tejun Heoe8989fa2012-03-05 13:15:20 -08001311
Tejun Heo838f13b2015-07-09 16:39:47 -04001312 mutex_lock(&blkcg_pol_register_mutex);
Tejun Heobc0d6502012-04-13 13:11:26 -07001313 mutex_lock(&blkcg_pol_mutex);
1314
Tejun Heo8bd435b2012-04-13 13:11:28 -07001315 /* find an empty slot */
1316 ret = -ENOSPC;
1317 for (i = 0; i < BLKCG_MAX_POLS; i++)
Tejun Heo3c798392012-04-16 13:57:25 -07001318 if (!blkcg_policy[i])
Tejun Heo8bd435b2012-04-13 13:11:28 -07001319 break;
1320 if (i >= BLKCG_MAX_POLS)
Tejun Heo838f13b2015-07-09 16:39:47 -04001321 goto err_unlock;
Tejun Heo035d10b2012-03-05 13:15:04 -08001322
Tejun Heo06b285b2015-07-09 16:39:50 -04001323 /* register @pol */
Tejun Heo3c798392012-04-16 13:57:25 -07001324 pol->plid = i;
Tejun Heo06b285b2015-07-09 16:39:50 -04001325 blkcg_policy[pol->plid] = pol;
1326
1327 /* allocate and install cpd's */
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001328 if (pol->cpd_alloc_fn) {
Tejun Heo06b285b2015-07-09 16:39:50 -04001329 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1330 struct blkcg_policy_data *cpd;
1331
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001332 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
Tejun Heo06b285b2015-07-09 16:39:50 -04001333 if (!cpd) {
1334 mutex_unlock(&blkcg_pol_mutex);
1335 goto err_free_cpds;
1336 }
1337
Tejun Heo81437642015-08-18 14:55:15 -07001338 blkcg->cpd[pol->plid] = cpd;
1339 cpd->blkcg = blkcg;
Tejun Heo06b285b2015-07-09 16:39:50 -04001340 cpd->plid = pol->plid;
Tejun Heo81437642015-08-18 14:55:15 -07001341 pol->cpd_init_fn(cpd);
Tejun Heo06b285b2015-07-09 16:39:50 -04001342 }
1343 }
1344
Tejun Heo838f13b2015-07-09 16:39:47 -04001345 mutex_unlock(&blkcg_pol_mutex);
Tejun Heo8bd435b2012-04-13 13:11:28 -07001346
Tejun Heo8bd435b2012-04-13 13:11:28 -07001347 /* everything is in place, add intf files for the new policy */
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001348 if (pol->dfl_cftypes)
1349 WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys,
1350 pol->dfl_cftypes));
Tejun Heo880f50e2015-08-18 14:55:30 -07001351 if (pol->legacy_cftypes)
Tejun Heoc165b3e2015-08-18 14:55:29 -07001352 WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys,
Tejun Heo880f50e2015-08-18 14:55:30 -07001353 pol->legacy_cftypes));
Tejun Heo838f13b2015-07-09 16:39:47 -04001354 mutex_unlock(&blkcg_pol_register_mutex);
1355 return 0;
1356
Tejun Heo06b285b2015-07-09 16:39:50 -04001357err_free_cpds:
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001358 if (pol->cpd_alloc_fn) {
Tejun Heo06b285b2015-07-09 16:39:50 -04001359 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001360 if (blkcg->cpd[pol->plid]) {
1361 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1362 blkcg->cpd[pol->plid] = NULL;
1363 }
Tejun Heo06b285b2015-07-09 16:39:50 -04001364 }
1365 }
1366 blkcg_policy[pol->plid] = NULL;
Tejun Heo838f13b2015-07-09 16:39:47 -04001367err_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -07001368 mutex_unlock(&blkcg_pol_mutex);
Tejun Heo838f13b2015-07-09 16:39:47 -04001369 mutex_unlock(&blkcg_pol_register_mutex);
Tejun Heo8bd435b2012-04-13 13:11:28 -07001370 return ret;
Vivek Goyal3e252062009-12-04 10:36:42 -05001371}
Tejun Heo3c798392012-04-16 13:57:25 -07001372EXPORT_SYMBOL_GPL(blkcg_policy_register);
Vivek Goyal3e252062009-12-04 10:36:42 -05001373
Tejun Heo8bd435b2012-04-13 13:11:28 -07001374/**
Tejun Heo3c798392012-04-16 13:57:25 -07001375 * blkcg_policy_unregister - unregister a blkcg policy
1376 * @pol: blkcg policy to unregister
Tejun Heo8bd435b2012-04-13 13:11:28 -07001377 *
Tejun Heo3c798392012-04-16 13:57:25 -07001378 * Undo blkcg_policy_register(@pol). Might sleep.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001379 */
Tejun Heo3c798392012-04-16 13:57:25 -07001380void blkcg_policy_unregister(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001381{
Tejun Heo06b285b2015-07-09 16:39:50 -04001382 struct blkcg *blkcg;
1383
Tejun Heo838f13b2015-07-09 16:39:47 -04001384 mutex_lock(&blkcg_pol_register_mutex);
Tejun Heobc0d6502012-04-13 13:11:26 -07001385
Tejun Heo3c798392012-04-16 13:57:25 -07001386 if (WARN_ON(blkcg_policy[pol->plid] != pol))
Tejun Heo8bd435b2012-04-13 13:11:28 -07001387 goto out_unlock;
1388
1389 /* kill the intf files first */
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001390 if (pol->dfl_cftypes)
1391 cgroup_rm_cftypes(pol->dfl_cftypes);
Tejun Heo880f50e2015-08-18 14:55:30 -07001392 if (pol->legacy_cftypes)
1393 cgroup_rm_cftypes(pol->legacy_cftypes);
Tejun Heo44ea53d2012-04-01 14:38:43 -07001394
Tejun Heo06b285b2015-07-09 16:39:50 -04001395 /* remove cpds and unregister */
Tejun Heo838f13b2015-07-09 16:39:47 -04001396 mutex_lock(&blkcg_pol_mutex);
Tejun Heo06b285b2015-07-09 16:39:50 -04001397
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001398 if (pol->cpd_alloc_fn) {
Tejun Heo06b285b2015-07-09 16:39:50 -04001399 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001400 if (blkcg->cpd[pol->plid]) {
1401 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1402 blkcg->cpd[pol->plid] = NULL;
1403 }
Tejun Heo06b285b2015-07-09 16:39:50 -04001404 }
1405 }
Tejun Heo3c798392012-04-16 13:57:25 -07001406 blkcg_policy[pol->plid] = NULL;
Tejun Heo06b285b2015-07-09 16:39:50 -04001407
Tejun Heobc0d6502012-04-13 13:11:26 -07001408 mutex_unlock(&blkcg_pol_mutex);
Tejun Heo838f13b2015-07-09 16:39:47 -04001409out_unlock:
1410 mutex_unlock(&blkcg_pol_register_mutex);
Vivek Goyal3e252062009-12-04 10:36:42 -05001411}
Tejun Heo3c798392012-04-16 13:57:25 -07001412EXPORT_SYMBOL_GPL(blkcg_policy_unregister);