blob: d6283b3f5db50674d18ae485970f1e24ed44569d [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 Heoeea8f412015-05-22 17:13:17 -040027#include <linux/blk-cgroup.h>
Tejun Heo5efd6112012-03-05 13:15:12 -080028#include "blk.h"
Vivek Goyal3e252062009-12-04 10:36:42 -050029
Divyesh Shah84c124d2010-04-09 08:31:19 +020030#define MAX_KEY_LEN 100
31
Tejun Heo838f13b2015-07-09 16:39:47 -040032/*
33 * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation.
34 * blkcg_pol_register_mutex nests outside of it and synchronizes entire
35 * policy [un]register operations including cgroup file additions /
36 * removals. Putting cgroup file registration outside blkcg_pol_mutex
37 * allows grabbing it from cgroup callbacks.
38 */
39static DEFINE_MUTEX(blkcg_pol_register_mutex);
Tejun Heobc0d6502012-04-13 13:11:26 -070040static DEFINE_MUTEX(blkcg_pol_mutex);
Tejun Heo923adde2012-03-05 13:15:13 -080041
Arianna Avanzinie48453c2015-06-05 23:38:42 +020042struct blkcg blkcg_root;
Tejun Heo3c798392012-04-16 13:57:25 -070043EXPORT_SYMBOL_GPL(blkcg_root);
Vivek Goyal9d6a9862009-12-04 10:36:41 -050044
Tejun Heo496d5e72015-05-22 17:13:21 -040045struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css;
46
Tejun Heo3c798392012-04-16 13:57:25 -070047static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
Tejun Heo035d10b2012-03-05 13:15:04 -080048
Tejun Heo7876f932015-07-09 16:39:49 -040049static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */
50
Tejun Heoa2b16932012-04-13 13:11:33 -070051static bool blkcg_policy_enabled(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -070052 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -070053{
54 return pol && test_bit(pol->plid, q->blkcg_pols);
55}
56
Tejun Heo03814112012-03-05 13:15:14 -080057/**
58 * blkg_free - free a blkg
59 * @blkg: blkg to free
60 *
61 * Free @blkg which may be partially allocated.
62 */
Tejun Heo3c798392012-04-16 13:57:25 -070063static void blkg_free(struct blkcg_gq *blkg)
Tejun Heo03814112012-03-05 13:15:14 -080064{
Tejun Heoe8989fa2012-03-05 13:15:20 -080065 int i;
Tejun Heo549d3aa2012-03-05 13:15:16 -080066
67 if (!blkg)
68 return;
69
Tejun Heodb613672013-05-14 13:52:31 -070070 for (i = 0; i < BLKCG_MAX_POLS; i++)
71 kfree(blkg->pd[i]);
Tejun Heoe8989fa2012-03-05 13:15:20 -080072
Tejun Heoa0516612012-06-26 15:05:44 -070073 blk_exit_rl(&blkg->rl);
Tejun Heo549d3aa2012-03-05 13:15:16 -080074 kfree(blkg);
Tejun Heo03814112012-03-05 13:15:14 -080075}
76
77/**
78 * blkg_alloc - allocate a blkg
79 * @blkcg: block cgroup the new blkg is associated with
80 * @q: request_queue the new blkg is associated with
Tejun Heo15974992012-06-04 20:40:52 -070081 * @gfp_mask: allocation mask to use
Tejun Heo03814112012-03-05 13:15:14 -080082 *
Tejun Heoe8989fa2012-03-05 13:15:20 -080083 * Allocate a new blkg assocating @blkcg and @q.
Tejun Heo03814112012-03-05 13:15:14 -080084 */
Tejun Heo15974992012-06-04 20:40:52 -070085static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
86 gfp_t gfp_mask)
Tejun Heo03814112012-03-05 13:15:14 -080087{
Tejun Heo3c798392012-04-16 13:57:25 -070088 struct blkcg_gq *blkg;
Tejun Heoe8989fa2012-03-05 13:15:20 -080089 int i;
Tejun Heo03814112012-03-05 13:15:14 -080090
91 /* alloc and init base part */
Tejun Heo15974992012-06-04 20:40:52 -070092 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
Tejun Heo03814112012-03-05 13:15:14 -080093 if (!blkg)
94 return NULL;
95
Tejun Heoc875f4d2012-03-05 13:15:22 -080096 blkg->q = q;
Tejun Heoe8989fa2012-03-05 13:15:20 -080097 INIT_LIST_HEAD(&blkg->q_node);
Tejun Heo03814112012-03-05 13:15:14 -080098 blkg->blkcg = blkcg;
Tejun Heoa5049a82014-06-19 17:42:57 -040099 atomic_set(&blkg->refcnt, 1);
Tejun Heo03814112012-03-05 13:15:14 -0800100
Tejun Heoa0516612012-06-26 15:05:44 -0700101 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
102 if (blkcg != &blkcg_root) {
103 if (blk_init_rl(&blkg->rl, q, gfp_mask))
104 goto err_free;
105 blkg->rl.blkg = blkg;
106 }
107
Tejun Heo8bd435b2012-04-13 13:11:28 -0700108 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -0700109 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heoe8989fa2012-03-05 13:15:20 -0800110 struct blkg_policy_data *pd;
Tejun Heo03814112012-03-05 13:15:14 -0800111
Tejun Heoa2b16932012-04-13 13:11:33 -0700112 if (!blkcg_policy_enabled(q, pol))
Tejun Heoe8989fa2012-03-05 13:15:20 -0800113 continue;
Tejun Heo549d3aa2012-03-05 13:15:16 -0800114
Tejun Heoe8989fa2012-03-05 13:15:20 -0800115 /* alloc per-policy data and attach it to blkg */
Tejun Heo15974992012-06-04 20:40:52 -0700116 pd = kzalloc_node(pol->pd_size, gfp_mask, q->node);
Tejun Heoa0516612012-06-26 15:05:44 -0700117 if (!pd)
118 goto err_free;
Tejun Heo549d3aa2012-03-05 13:15:16 -0800119
Tejun Heoe8989fa2012-03-05 13:15:20 -0800120 blkg->pd[i] = pd;
121 pd->blkg = blkg;
Tejun Heob276a872013-01-09 08:05:12 -0800122 pd->plid = i;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800123 }
124
Tejun Heo03814112012-03-05 13:15:14 -0800125 return blkg;
Tejun Heoa0516612012-06-26 15:05:44 -0700126
127err_free:
128 blkg_free(blkg);
129 return NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800130}
131
Tejun Heo16b3de62013-01-09 08:05:12 -0800132/**
133 * __blkg_lookup - internal version of blkg_lookup()
134 * @blkcg: blkcg of interest
135 * @q: request_queue of interest
136 * @update_hint: whether to update lookup hint with the result or not
137 *
138 * This is internal version and shouldn't be used by policy
139 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
140 * @q's bypass state. If @update_hint is %true, the caller should be
141 * holding @q->queue_lock and lookup hint is updated on success.
142 */
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700143struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
144 bool update_hint)
Tejun Heo80fd9972012-04-13 14:50:53 -0700145{
Tejun Heo3c798392012-04-16 13:57:25 -0700146 struct blkcg_gq *blkg;
Tejun Heo80fd9972012-04-13 14:50:53 -0700147
Tejun Heoa6371202012-04-19 16:29:24 -0700148 blkg = rcu_dereference(blkcg->blkg_hint);
149 if (blkg && blkg->q == q)
150 return blkg;
151
152 /*
Tejun Heo86cde6b2013-01-09 08:05:10 -0800153 * Hint didn't match. Look up from the radix tree. Note that the
154 * hint can only be updated under queue_lock as otherwise @blkg
155 * could have already been removed from blkg_tree. The caller is
156 * responsible for grabbing queue_lock if @update_hint.
Tejun Heoa6371202012-04-19 16:29:24 -0700157 */
158 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800159 if (blkg && blkg->q == q) {
160 if (update_hint) {
161 lockdep_assert_held(q->queue_lock);
162 rcu_assign_pointer(blkcg->blkg_hint, blkg);
163 }
Tejun Heoa6371202012-04-19 16:29:24 -0700164 return blkg;
Tejun Heo86cde6b2013-01-09 08:05:10 -0800165 }
Tejun Heoa6371202012-04-19 16:29:24 -0700166
Tejun Heo80fd9972012-04-13 14:50:53 -0700167 return NULL;
168}
169
170/**
171 * blkg_lookup - lookup blkg for the specified blkcg - q pair
172 * @blkcg: blkcg of interest
173 * @q: request_queue of interest
174 *
175 * Lookup blkg for the @blkcg - @q pair. This function should be called
176 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
177 * - see blk_queue_bypass_start() for details.
178 */
Tejun Heo3c798392012-04-16 13:57:25 -0700179struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
Tejun Heo80fd9972012-04-13 14:50:53 -0700180{
181 WARN_ON_ONCE(!rcu_read_lock_held());
182
183 if (unlikely(blk_queue_bypass(q)))
184 return NULL;
Tejun Heo86cde6b2013-01-09 08:05:10 -0800185 return __blkg_lookup(blkcg, q, false);
Tejun Heo80fd9972012-04-13 14:50:53 -0700186}
187EXPORT_SYMBOL_GPL(blkg_lookup);
188
Tejun Heo15974992012-06-04 20:40:52 -0700189/*
190 * If @new_blkg is %NULL, this function tries to allocate a new one as
191 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
192 */
Tejun Heo86cde6b2013-01-09 08:05:10 -0800193static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
194 struct request_queue *q,
195 struct blkcg_gq *new_blkg)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400196{
Tejun Heo3c798392012-04-16 13:57:25 -0700197 struct blkcg_gq *blkg;
Tejun Heoce7acfe2015-05-22 17:13:38 -0400198 struct bdi_writeback_congested *wb_congested;
Tejun Heof427d902013-01-09 08:05:12 -0800199 int i, ret;
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400200
Tejun Heocd1604f2012-03-05 13:15:06 -0800201 WARN_ON_ONCE(!rcu_read_lock_held());
202 lockdep_assert_held(q->queue_lock);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500203
Tejun Heo7ee9c562012-03-05 13:15:11 -0800204 /* blkg holds a reference to blkcg */
Tejun Heoec903c02014-05-13 12:11:01 -0400205 if (!css_tryget_online(&blkcg->css)) {
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800206 ret = -EINVAL;
207 goto err_free_blkg;
Tejun Heo15974992012-06-04 20:40:52 -0700208 }
Tejun Heocd1604f2012-03-05 13:15:06 -0800209
Tejun Heoce7acfe2015-05-22 17:13:38 -0400210 wb_congested = wb_congested_get_create(&q->backing_dev_info,
211 blkcg->css.id, GFP_ATOMIC);
212 if (!wb_congested) {
213 ret = -ENOMEM;
214 goto err_put_css;
215 }
216
Tejun Heo496fb782012-04-19 16:29:23 -0700217 /* allocate */
Tejun Heo15974992012-06-04 20:40:52 -0700218 if (!new_blkg) {
219 new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
220 if (unlikely(!new_blkg)) {
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800221 ret = -ENOMEM;
Tejun Heoce7acfe2015-05-22 17:13:38 -0400222 goto err_put_congested;
Tejun Heo15974992012-06-04 20:40:52 -0700223 }
224 }
225 blkg = new_blkg;
Tejun Heoce7acfe2015-05-22 17:13:38 -0400226 blkg->wb_congested = wb_congested;
Tejun Heocd1604f2012-03-05 13:15:06 -0800227
Tejun Heodb613672013-05-14 13:52:31 -0700228 /* link parent */
Tejun Heo3c547862013-01-09 08:05:10 -0800229 if (blkcg_parent(blkcg)) {
230 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
231 if (WARN_ON_ONCE(!blkg->parent)) {
Tejun Heo2423c9c2013-05-14 13:52:30 -0700232 ret = -EINVAL;
Tejun Heoce7acfe2015-05-22 17:13:38 -0400233 goto err_put_congested;
Tejun Heo3c547862013-01-09 08:05:10 -0800234 }
235 blkg_get(blkg->parent);
236 }
237
Tejun Heodb613672013-05-14 13:52:31 -0700238 /* invoke per-policy init */
239 for (i = 0; i < BLKCG_MAX_POLS; i++) {
240 struct blkcg_policy *pol = blkcg_policy[i];
241
242 if (blkg->pd[i] && pol->pd_init_fn)
243 pol->pd_init_fn(blkg);
244 }
245
246 /* insert */
Tejun Heoa6371202012-04-19 16:29:24 -0700247 spin_lock(&blkcg->lock);
248 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
249 if (likely(!ret)) {
250 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
251 list_add(&blkg->q_node, &q->blkg_list);
Tejun Heof427d902013-01-09 08:05:12 -0800252
253 for (i = 0; i < BLKCG_MAX_POLS; i++) {
254 struct blkcg_policy *pol = blkcg_policy[i];
255
256 if (blkg->pd[i] && pol->pd_online_fn)
257 pol->pd_online_fn(blkg);
258 }
Tejun Heoa6371202012-04-19 16:29:24 -0700259 }
Tejun Heof427d902013-01-09 08:05:12 -0800260 blkg->online = true;
Tejun Heoa6371202012-04-19 16:29:24 -0700261 spin_unlock(&blkcg->lock);
262
Tejun Heoec13b1d2015-05-22 17:13:19 -0400263 if (!ret)
Tejun Heoa6371202012-04-19 16:29:24 -0700264 return blkg;
Tejun Heo15974992012-06-04 20:40:52 -0700265
Tejun Heo3c547862013-01-09 08:05:10 -0800266 /* @blkg failed fully initialized, use the usual release path */
267 blkg_put(blkg);
268 return ERR_PTR(ret);
269
Tejun Heoce7acfe2015-05-22 17:13:38 -0400270err_put_congested:
271 wb_congested_put(wb_congested);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800272err_put_css:
Tejun Heo496fb782012-04-19 16:29:23 -0700273 css_put(&blkcg->css);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800274err_free_blkg:
Tejun Heo15974992012-06-04 20:40:52 -0700275 blkg_free(new_blkg);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800276 return ERR_PTR(ret);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500277}
Tejun Heo3c96cb32012-04-13 13:11:34 -0700278
Tejun Heo86cde6b2013-01-09 08:05:10 -0800279/**
280 * blkg_lookup_create - lookup blkg, try to create one if not there
281 * @blkcg: blkcg of interest
282 * @q: request_queue of interest
283 *
284 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
Tejun Heo3c547862013-01-09 08:05:10 -0800285 * create one. blkg creation is performed recursively from blkcg_root such
286 * that all non-root blkg's have access to the parent blkg. This function
287 * should be called under RCU read lock and @q->queue_lock.
Tejun Heo86cde6b2013-01-09 08:05:10 -0800288 *
289 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
290 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
291 * dead and bypassing, returns ERR_PTR(-EBUSY).
292 */
Tejun Heo3c798392012-04-16 13:57:25 -0700293struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
294 struct request_queue *q)
Tejun Heo3c96cb32012-04-13 13:11:34 -0700295{
Tejun Heo86cde6b2013-01-09 08:05:10 -0800296 struct blkcg_gq *blkg;
297
298 WARN_ON_ONCE(!rcu_read_lock_held());
299 lockdep_assert_held(q->queue_lock);
300
Tejun Heo3c96cb32012-04-13 13:11:34 -0700301 /*
302 * This could be the first entry point of blkcg implementation and
303 * we shouldn't allow anything to go through for a bypassing queue.
304 */
305 if (unlikely(blk_queue_bypass(q)))
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100306 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800307
308 blkg = __blkg_lookup(blkcg, q, true);
309 if (blkg)
310 return blkg;
311
Tejun Heo3c547862013-01-09 08:05:10 -0800312 /*
313 * Create blkgs walking down from blkcg_root to @blkcg, so that all
314 * non-root blkgs have access to their parents.
315 */
316 while (true) {
317 struct blkcg *pos = blkcg;
318 struct blkcg *parent = blkcg_parent(blkcg);
319
320 while (parent && !__blkg_lookup(parent, q, false)) {
321 pos = parent;
322 parent = blkcg_parent(parent);
323 }
324
325 blkg = blkg_create(pos, q, NULL);
326 if (pos == blkcg || IS_ERR(blkg))
327 return blkg;
328 }
Tejun Heo3c96cb32012-04-13 13:11:34 -0700329}
Tejun Heocd1604f2012-03-05 13:15:06 -0800330EXPORT_SYMBOL_GPL(blkg_lookup_create);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500331
Tejun Heo3c798392012-04-16 13:57:25 -0700332static void blkg_destroy(struct blkcg_gq *blkg)
Tejun Heo72e06c22012-03-05 13:15:00 -0800333{
Tejun Heo3c798392012-04-16 13:57:25 -0700334 struct blkcg *blkcg = blkg->blkcg;
Tejun Heof427d902013-01-09 08:05:12 -0800335 int i;
Tejun Heo03aa2642012-03-05 13:15:19 -0800336
Tejun Heo27e1f9d2012-06-05 13:36:44 +0200337 lockdep_assert_held(blkg->q->queue_lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800338 lockdep_assert_held(&blkcg->lock);
Tejun Heo03aa2642012-03-05 13:15:19 -0800339
340 /* Something wrong if we are trying to remove same group twice */
Tejun Heoe8989fa2012-03-05 13:15:20 -0800341 WARN_ON_ONCE(list_empty(&blkg->q_node));
Tejun Heo9f13ef62012-03-05 13:15:21 -0800342 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
Tejun Heoa6371202012-04-19 16:29:24 -0700343
Tejun Heof427d902013-01-09 08:05:12 -0800344 for (i = 0; i < BLKCG_MAX_POLS; i++) {
345 struct blkcg_policy *pol = blkcg_policy[i];
346
347 if (blkg->pd[i] && pol->pd_offline_fn)
348 pol->pd_offline_fn(blkg);
349 }
350 blkg->online = false;
351
Tejun Heoa6371202012-04-19 16:29:24 -0700352 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
Tejun Heoe8989fa2012-03-05 13:15:20 -0800353 list_del_init(&blkg->q_node);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800354 hlist_del_init_rcu(&blkg->blkcg_node);
Tejun Heo03aa2642012-03-05 13:15:19 -0800355
Tejun Heo03aa2642012-03-05 13:15:19 -0800356 /*
Tejun Heoa6371202012-04-19 16:29:24 -0700357 * Both setting lookup hint to and clearing it from @blkg are done
358 * under queue_lock. If it's not pointing to @blkg now, it never
359 * will. Hint assignment itself can race safely.
360 */
Paul E. McKenneyec6c6762014-02-17 13:35:57 -0800361 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
Tejun Heoa6371202012-04-19 16:29:24 -0700362 rcu_assign_pointer(blkcg->blkg_hint, NULL);
363
364 /*
Tejun Heo03aa2642012-03-05 13:15:19 -0800365 * Put the reference taken at the time of creation so that when all
366 * queues are gone, group can be destroyed.
367 */
368 blkg_put(blkg);
369}
370
Tejun Heo9f13ef62012-03-05 13:15:21 -0800371/**
372 * blkg_destroy_all - destroy all blkgs associated with a request_queue
373 * @q: request_queue of interest
Tejun Heo9f13ef62012-03-05 13:15:21 -0800374 *
Tejun Heo3c96cb32012-04-13 13:11:34 -0700375 * Destroy all blkgs associated with @q.
Tejun Heo9f13ef62012-03-05 13:15:21 -0800376 */
Tejun Heo3c96cb32012-04-13 13:11:34 -0700377static void blkg_destroy_all(struct request_queue *q)
Tejun Heo03aa2642012-03-05 13:15:19 -0800378{
Tejun Heo3c798392012-04-16 13:57:25 -0700379 struct blkcg_gq *blkg, *n;
Tejun Heo72e06c22012-03-05 13:15:00 -0800380
Tejun Heo6d18b002012-04-13 13:11:35 -0700381 lockdep_assert_held(q->queue_lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800382
Tejun Heo9f13ef62012-03-05 13:15:21 -0800383 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
Tejun Heo3c798392012-04-16 13:57:25 -0700384 struct blkcg *blkcg = blkg->blkcg;
Tejun Heo72e06c22012-03-05 13:15:00 -0800385
Tejun Heo9f13ef62012-03-05 13:15:21 -0800386 spin_lock(&blkcg->lock);
387 blkg_destroy(blkg);
388 spin_unlock(&blkcg->lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800389 }
390}
391
Tejun Heo2a4fd072013-05-14 13:52:31 -0700392/*
393 * A group is RCU protected, but having an rcu lock does not mean that one
394 * can access all the fields of blkg and assume these are valid. For
395 * example, don't try to follow throtl_data and request queue links.
396 *
397 * Having a reference to blkg under an rcu allows accesses to only values
398 * local to groups like group stats and group rate limits.
399 */
400void __blkg_release_rcu(struct rcu_head *rcu_head)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800401{
Tejun Heo2a4fd072013-05-14 13:52:31 -0700402 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
Tejun Heodb613672013-05-14 13:52:31 -0700403 int i;
404
405 /* tell policies that this one is being freed */
406 for (i = 0; i < BLKCG_MAX_POLS; i++) {
407 struct blkcg_policy *pol = blkcg_policy[i];
408
409 if (blkg->pd[i] && pol->pd_exit_fn)
410 pol->pd_exit_fn(blkg);
411 }
412
Tejun Heo3c547862013-01-09 08:05:10 -0800413 /* release the blkcg and parent blkg refs this blkg has been holding */
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800414 css_put(&blkg->blkcg->css);
Tejun Heoa5049a82014-06-19 17:42:57 -0400415 if (blkg->parent)
Tejun Heo3c547862013-01-09 08:05:10 -0800416 blkg_put(blkg->parent);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800417
Tejun Heoce7acfe2015-05-22 17:13:38 -0400418 wb_congested_put(blkg->wb_congested);
419
Tejun Heo2a4fd072013-05-14 13:52:31 -0700420 blkg_free(blkg);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800421}
Tejun Heo2a4fd072013-05-14 13:52:31 -0700422EXPORT_SYMBOL_GPL(__blkg_release_rcu);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800423
Tejun Heoa0516612012-06-26 15:05:44 -0700424/*
425 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
426 * because the root blkg uses @q->root_rl instead of its own rl.
427 */
428struct request_list *__blk_queue_next_rl(struct request_list *rl,
429 struct request_queue *q)
430{
431 struct list_head *ent;
432 struct blkcg_gq *blkg;
433
434 /*
435 * Determine the current blkg list_head. The first entry is
436 * root_rl which is off @q->blkg_list and mapped to the head.
437 */
438 if (rl == &q->root_rl) {
439 ent = &q->blkg_list;
Jun'ichi Nomura65c77fd2012-10-22 10:15:37 +0900440 /* There are no more block groups, hence no request lists */
441 if (list_empty(ent))
442 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -0700443 } else {
444 blkg = container_of(rl, struct blkcg_gq, rl);
445 ent = &blkg->q_node;
446 }
447
448 /* walk to the next list_head, skip root blkcg */
449 ent = ent->next;
450 if (ent == &q->root_blkg->q_node)
451 ent = ent->next;
452 if (ent == &q->blkg_list)
453 return NULL;
454
455 blkg = container_of(ent, struct blkcg_gq, q_node);
456 return &blkg->rl;
457}
458
Tejun Heo182446d2013-08-08 20:11:24 -0400459static int blkcg_reset_stats(struct cgroup_subsys_state *css,
460 struct cftype *cftype, u64 val)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700461{
Tejun Heo182446d2013-08-08 20:11:24 -0400462 struct blkcg *blkcg = css_to_blkcg(css);
Tejun Heo3c798392012-04-16 13:57:25 -0700463 struct blkcg_gq *blkg;
Tejun Heobc0d6502012-04-13 13:11:26 -0700464 int i;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700465
Tejun Heo838f13b2015-07-09 16:39:47 -0400466 mutex_lock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700467 spin_lock_irq(&blkcg->lock);
Tejun Heo997a0262012-03-08 10:53:58 -0800468
469 /*
470 * Note that stat reset is racy - it doesn't synchronize against
471 * stat updates. This is a debug feature which shouldn't exist
472 * anyway. If you get hit by a race, retry.
473 */
Sasha Levinb67bfe02013-02-27 17:06:00 -0800474 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo8bd435b2012-04-13 13:11:28 -0700475 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -0700476 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heo549d3aa2012-03-05 13:15:16 -0800477
Tejun Heoa2b16932012-04-13 13:11:33 -0700478 if (blkcg_policy_enabled(blkg->q, pol) &&
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700479 pol->pd_reset_stats_fn)
480 pol->pd_reset_stats_fn(blkg);
Tejun Heobc0d6502012-04-13 13:11:26 -0700481 }
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700482 }
Vivek Goyalf0bdc8c2011-05-19 15:38:30 -0400483
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700484 spin_unlock_irq(&blkcg->lock);
Tejun Heobc0d6502012-04-13 13:11:26 -0700485 mutex_unlock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700486 return 0;
487}
488
Tejun Heo3c798392012-04-16 13:57:25 -0700489static const char *blkg_dev_name(struct blkcg_gq *blkg)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700490{
Tejun Heod3d32e62012-04-01 14:38:42 -0700491 /* some drivers (floppy) instantiate a queue w/o disk registered */
492 if (blkg->q->backing_dev_info.dev)
493 return dev_name(blkg->q->backing_dev_info.dev);
494 return NULL;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700495}
496
Tejun Heod3d32e62012-04-01 14:38:42 -0700497/**
498 * blkcg_print_blkgs - helper for printing per-blkg data
499 * @sf: seq_file to print to
500 * @blkcg: blkcg of interest
501 * @prfill: fill function to print out a blkg
502 * @pol: policy in question
503 * @data: data to be passed to @prfill
504 * @show_total: to print out sum of prfill return values or not
505 *
506 * This function invokes @prfill on each blkg of @blkcg if pd for the
507 * policy specified by @pol exists. @prfill is invoked with @sf, the
Tejun Heo810ecfa2013-01-09 08:05:13 -0800508 * policy data and @data and the matching queue lock held. If @show_total
509 * is %true, the sum of the return values from @prfill is printed with
510 * "Total" label at the end.
Tejun Heod3d32e62012-04-01 14:38:42 -0700511 *
512 * This is to be used to construct print functions for
513 * cftype->read_seq_string method.
514 */
Tejun Heo3c798392012-04-16 13:57:25 -0700515void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
Tejun Heof95a04a2012-04-16 13:57:26 -0700516 u64 (*prfill)(struct seq_file *,
517 struct blkg_policy_data *, int),
Tejun Heo3c798392012-04-16 13:57:25 -0700518 const struct blkcg_policy *pol, int data,
Tejun Heoec399342012-04-13 13:11:27 -0700519 bool show_total)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400520{
Tejun Heo3c798392012-04-16 13:57:25 -0700521 struct blkcg_gq *blkg;
Tejun Heod3d32e62012-04-01 14:38:42 -0700522 u64 total = 0;
523
Tejun Heo810ecfa2013-01-09 08:05:13 -0800524 rcu_read_lock();
Linus Torvaldsee89f812013-02-28 12:52:24 -0800525 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo810ecfa2013-01-09 08:05:13 -0800526 spin_lock_irq(blkg->q->queue_lock);
Tejun Heoa2b16932012-04-13 13:11:33 -0700527 if (blkcg_policy_enabled(blkg->q, pol))
Tejun Heof95a04a2012-04-16 13:57:26 -0700528 total += prfill(sf, blkg->pd[pol->plid], data);
Tejun Heo810ecfa2013-01-09 08:05:13 -0800529 spin_unlock_irq(blkg->q->queue_lock);
530 }
531 rcu_read_unlock();
Tejun Heod3d32e62012-04-01 14:38:42 -0700532
533 if (show_total)
534 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
535}
Tejun Heo829fdb52012-04-01 14:38:43 -0700536EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
Tejun Heod3d32e62012-04-01 14:38:42 -0700537
538/**
539 * __blkg_prfill_u64 - prfill helper for a single u64 value
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 * @v: value to print
543 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700544 * Print @v 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_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
Tejun Heod3d32e62012-04-01 14:38:42 -0700547{
Tejun Heof95a04a2012-04-16 13:57:26 -0700548 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700549
550 if (!dname)
551 return 0;
552
553 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
554 return v;
555}
Tejun Heo829fdb52012-04-01 14:38:43 -0700556EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
Tejun Heod3d32e62012-04-01 14:38:42 -0700557
558/**
559 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
560 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700561 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700562 * @rwstat: rwstat to print
563 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700564 * Print @rwstat to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700565 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700566u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
Tejun Heo829fdb52012-04-01 14:38:43 -0700567 const struct blkg_rwstat *rwstat)
Tejun Heod3d32e62012-04-01 14:38:42 -0700568{
569 static const char *rwstr[] = {
570 [BLKG_RWSTAT_READ] = "Read",
571 [BLKG_RWSTAT_WRITE] = "Write",
572 [BLKG_RWSTAT_SYNC] = "Sync",
573 [BLKG_RWSTAT_ASYNC] = "Async",
574 };
Tejun Heof95a04a2012-04-16 13:57:26 -0700575 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700576 u64 v;
577 int i;
578
579 if (!dname)
580 return 0;
581
582 for (i = 0; i < BLKG_RWSTAT_NR; i++)
583 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
584 (unsigned long long)rwstat->cnt[i]);
585
586 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
587 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
588 return v;
589}
Tejun Heob50da392013-01-09 08:05:12 -0800590EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700591
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700592/**
593 * blkg_prfill_stat - prfill callback for blkg_stat
594 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700595 * @pd: policy private data of interest
596 * @off: offset to the blkg_stat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700597 *
598 * prfill callback for printing a blkg_stat.
599 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700600u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700601{
Tejun Heof95a04a2012-04-16 13:57:26 -0700602 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
Tejun Heod3d32e62012-04-01 14:38:42 -0700603}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700604EXPORT_SYMBOL_GPL(blkg_prfill_stat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700605
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700606/**
607 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
608 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700609 * @pd: policy private data of interest
610 * @off: offset to the blkg_rwstat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700611 *
612 * prfill callback for printing a blkg_rwstat.
613 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700614u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
615 int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700616{
Tejun Heof95a04a2012-04-16 13:57:26 -0700617 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
Tejun Heod3d32e62012-04-01 14:38:42 -0700618
Tejun Heof95a04a2012-04-16 13:57:26 -0700619 return __blkg_prfill_rwstat(sf, pd, &rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700620}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700621EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700622
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700623/**
Tejun Heo16b3de62013-01-09 08:05:12 -0800624 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
625 * @pd: policy private data of interest
626 * @off: offset to the blkg_stat in @pd
627 *
628 * Collect the blkg_stat specified by @off from @pd and all its online
629 * descendants and return the sum. The caller must be holding the queue
630 * lock for online tests.
631 */
632u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off)
633{
634 struct blkcg_policy *pol = blkcg_policy[pd->plid];
635 struct blkcg_gq *pos_blkg;
Tejun Heo492eb212013-08-08 20:11:25 -0400636 struct cgroup_subsys_state *pos_css;
Tejun Heobd8815a2013-08-08 20:11:27 -0400637 u64 sum = 0;
Tejun Heo16b3de62013-01-09 08:05:12 -0800638
639 lockdep_assert_held(pd->blkg->q->queue_lock);
640
Tejun Heo16b3de62013-01-09 08:05:12 -0800641 rcu_read_lock();
Tejun Heo492eb212013-08-08 20:11:25 -0400642 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
Tejun Heo16b3de62013-01-09 08:05:12 -0800643 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
644 struct blkg_stat *stat = (void *)pos_pd + off;
645
646 if (pos_blkg->online)
647 sum += blkg_stat_read(stat);
648 }
649 rcu_read_unlock();
650
651 return sum;
652}
653EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
654
655/**
656 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
657 * @pd: policy private data of interest
658 * @off: offset to the blkg_stat in @pd
659 *
660 * Collect the blkg_rwstat specified by @off from @pd and all its online
661 * descendants and return the sum. The caller must be holding the queue
662 * lock for online tests.
663 */
664struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
665 int off)
666{
667 struct blkcg_policy *pol = blkcg_policy[pd->plid];
668 struct blkcg_gq *pos_blkg;
Tejun Heo492eb212013-08-08 20:11:25 -0400669 struct cgroup_subsys_state *pos_css;
Tejun Heobd8815a2013-08-08 20:11:27 -0400670 struct blkg_rwstat sum = { };
Tejun Heo16b3de62013-01-09 08:05:12 -0800671 int i;
672
673 lockdep_assert_held(pd->blkg->q->queue_lock);
674
Tejun Heo16b3de62013-01-09 08:05:12 -0800675 rcu_read_lock();
Tejun Heo492eb212013-08-08 20:11:25 -0400676 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
Tejun Heo16b3de62013-01-09 08:05:12 -0800677 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
678 struct blkg_rwstat *rwstat = (void *)pos_pd + off;
679 struct blkg_rwstat tmp;
680
681 if (!pos_blkg->online)
682 continue;
683
684 tmp = blkg_rwstat_read(rwstat);
685
686 for (i = 0; i < BLKG_RWSTAT_NR; i++)
687 sum.cnt[i] += tmp.cnt[i];
688 }
689 rcu_read_unlock();
690
691 return sum;
692}
693EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
694
695/**
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700696 * blkg_conf_prep - parse and prepare for per-blkg config update
697 * @blkcg: target block cgroup
Tejun Heoda8b0662012-04-13 13:11:29 -0700698 * @pol: target policy
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700699 * @input: input string
700 * @ctx: blkg_conf_ctx to be filled
701 *
702 * Parse per-blkg config update from @input and initialize @ctx with the
703 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
Tejun Heoda8b0662012-04-13 13:11:29 -0700704 * value. This function returns with RCU read lock and queue lock held and
705 * must be paired with blkg_conf_finish().
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700706 */
Tejun Heo3c798392012-04-16 13:57:25 -0700707int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
708 const char *input, struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700709 __acquires(rcu) __acquires(disk->queue->queue_lock)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800710{
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700711 struct gendisk *disk;
Tejun Heo3c798392012-04-16 13:57:25 -0700712 struct blkcg_gq *blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700713 unsigned int major, minor;
714 unsigned long long v;
715 int part, ret;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800716
Tejun Heo726fa692012-04-01 14:38:43 -0700717 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
718 return -EINVAL;
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700719
Tejun Heo726fa692012-04-01 14:38:43 -0700720 disk = get_gendisk(MKDEV(major, minor), &part);
Tejun Heo5f6c2d22015-07-22 18:05:53 -0400721 if (!disk)
Tejun Heo726fa692012-04-01 14:38:43 -0700722 return -EINVAL;
Tejun Heo5f6c2d22015-07-22 18:05:53 -0400723 if (part) {
724 put_disk(disk);
725 return -EINVAL;
726 }
Tejun Heoe56da7e2012-03-05 13:15:07 -0800727
728 rcu_read_lock();
Tejun Heo4bfd4822012-03-05 13:15:08 -0800729 spin_lock_irq(disk->queue->queue_lock);
Tejun Heoda8b0662012-04-13 13:11:29 -0700730
Tejun Heoa2b16932012-04-13 13:11:33 -0700731 if (blkcg_policy_enabled(disk->queue, pol))
Tejun Heo3c96cb32012-04-13 13:11:34 -0700732 blkg = blkg_lookup_create(blkcg, disk->queue);
Tejun Heoa2b16932012-04-13 13:11:33 -0700733 else
734 blkg = ERR_PTR(-EINVAL);
Tejun Heoe56da7e2012-03-05 13:15:07 -0800735
Tejun Heo4bfd4822012-03-05 13:15:08 -0800736 if (IS_ERR(blkg)) {
737 ret = PTR_ERR(blkg);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700738 rcu_read_unlock();
Tejun Heoda8b0662012-04-13 13:11:29 -0700739 spin_unlock_irq(disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700740 put_disk(disk);
741 /*
742 * If queue was bypassing, we should retry. Do so after a
743 * short msleep(). It isn't strictly necessary but queue
744 * can be bypassing for some time and it's always nice to
745 * avoid busy looping.
746 */
747 if (ret == -EBUSY) {
748 msleep(10);
749 ret = restart_syscall();
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400750 }
Tejun Heo726fa692012-04-01 14:38:43 -0700751 return ret;
Vivek Goyal062a6442010-09-15 17:06:33 -0400752 }
Tejun Heoe56da7e2012-03-05 13:15:07 -0800753
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700754 ctx->disk = disk;
755 ctx->blkg = blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700756 ctx->v = v;
757 return 0;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800758}
Tejun Heo829fdb52012-04-01 14:38:43 -0700759EXPORT_SYMBOL_GPL(blkg_conf_prep);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800760
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700761/**
762 * blkg_conf_finish - finish up per-blkg config update
763 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
764 *
765 * Finish up after per-blkg config update. This function must be paired
766 * with blkg_conf_prep().
767 */
Tejun Heo829fdb52012-04-01 14:38:43 -0700768void blkg_conf_finish(struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700769 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800770{
Tejun Heoda8b0662012-04-13 13:11:29 -0700771 spin_unlock_irq(ctx->disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700772 rcu_read_unlock();
773 put_disk(ctx->disk);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800774}
Tejun Heo829fdb52012-04-01 14:38:43 -0700775EXPORT_SYMBOL_GPL(blkg_conf_finish);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800776
Tejun Heo3c798392012-04-16 13:57:25 -0700777struct cftype blkcg_files[] = {
Vivek Goyal31e4c282009-12-03 12:59:42 -0500778 {
Divyesh Shah84c124d2010-04-09 08:31:19 +0200779 .name = "reset_stats",
Tejun Heo3c798392012-04-16 13:57:25 -0700780 .write_u64 = blkcg_reset_stats,
Vivek Goyal22084192009-12-03 12:59:49 -0500781 },
Tejun Heo4baf6e32012-04-01 12:09:55 -0700782 { } /* terminate */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500783};
784
Tejun Heo9f13ef62012-03-05 13:15:21 -0800785/**
Tejun Heo92fb9742012-11-19 08:13:38 -0800786 * blkcg_css_offline - cgroup css_offline callback
Tejun Heoeb954192013-08-08 20:11:23 -0400787 * @css: css of interest
Tejun Heo9f13ef62012-03-05 13:15:21 -0800788 *
Tejun Heoeb954192013-08-08 20:11:23 -0400789 * This function is called when @css is about to go away and responsible
790 * for shooting down all blkgs associated with @css. blkgs should be
Tejun Heo9f13ef62012-03-05 13:15:21 -0800791 * removed while holding both q and blkcg locks. As blkcg lock is nested
792 * inside q lock, this function performs reverse double lock dancing.
793 *
794 * This is the blkcg counterpart of ioc_release_fn().
795 */
Tejun Heoeb954192013-08-08 20:11:23 -0400796static void blkcg_css_offline(struct cgroup_subsys_state *css)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500797{
Tejun Heoeb954192013-08-08 20:11:23 -0400798 struct blkcg *blkcg = css_to_blkcg(css);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500799
Tejun Heo9f13ef62012-03-05 13:15:21 -0800800 spin_lock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800801
Tejun Heo9f13ef62012-03-05 13:15:21 -0800802 while (!hlist_empty(&blkcg->blkg_list)) {
Tejun Heo3c798392012-04-16 13:57:25 -0700803 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
804 struct blkcg_gq, blkcg_node);
Tejun Heoc875f4d2012-03-05 13:15:22 -0800805 struct request_queue *q = blkg->q;
Vivek Goyalb1c35762009-12-03 12:59:47 -0500806
Tejun Heo9f13ef62012-03-05 13:15:21 -0800807 if (spin_trylock(q->queue_lock)) {
808 blkg_destroy(blkg);
809 spin_unlock(q->queue_lock);
810 } else {
811 spin_unlock_irq(&blkcg->lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800812 cpu_relax();
Dan Carpentera5567932012-03-29 20:57:08 +0200813 spin_lock_irq(&blkcg->lock);
Jens Axboe0f3942a2010-05-03 14:28:55 +0200814 }
Tejun Heo9f13ef62012-03-05 13:15:21 -0800815 }
Jens Axboe0f3942a2010-05-03 14:28:55 +0200816
Tejun Heo9f13ef62012-03-05 13:15:21 -0800817 spin_unlock_irq(&blkcg->lock);
Tejun Heo52ebea72015-05-22 17:13:37 -0400818
819 wb_blkcg_offline(blkcg);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800820}
821
Tejun Heoeb954192013-08-08 20:11:23 -0400822static void blkcg_css_free(struct cgroup_subsys_state *css)
Tejun Heo7ee9c562012-03-05 13:15:11 -0800823{
Tejun Heoeb954192013-08-08 20:11:23 -0400824 struct blkcg *blkcg = css_to_blkcg(css);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800825
Tejun Heo7876f932015-07-09 16:39:49 -0400826 mutex_lock(&blkcg_pol_mutex);
827 list_del(&blkcg->all_blkcgs_node);
828 mutex_unlock(&blkcg_pol_mutex);
829
Arianna Avanzinia322baa2015-07-07 03:08:15 +0200830 if (blkcg != &blkcg_root) {
831 int i;
832
833 for (i = 0; i < BLKCG_MAX_POLS; i++)
834 kfree(blkcg->pd[i]);
Ben Blum67523c42010-03-10 15:22:11 -0800835 kfree(blkcg);
Arianna Avanzinia322baa2015-07-07 03:08:15 +0200836 }
Vivek Goyal31e4c282009-12-03 12:59:42 -0500837}
838
Tejun Heoeb954192013-08-08 20:11:23 -0400839static struct cgroup_subsys_state *
840blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500841{
Tejun Heo3c798392012-04-16 13:57:25 -0700842 struct blkcg *blkcg;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200843 struct cgroup_subsys_state *ret;
844 int i;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500845
Tejun Heo7876f932015-07-09 16:39:49 -0400846 mutex_lock(&blkcg_pol_mutex);
847
Tejun Heoeb954192013-08-08 20:11:23 -0400848 if (!parent_css) {
Tejun Heo3c798392012-04-16 13:57:25 -0700849 blkcg = &blkcg_root;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500850 goto done;
851 }
852
Vivek Goyal31e4c282009-12-03 12:59:42 -0500853 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200854 if (!blkcg) {
855 ret = ERR_PTR(-ENOMEM);
856 goto free_blkcg;
857 }
Vivek Goyal31e4c282009-12-03 12:59:42 -0500858
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200859 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
860 struct blkcg_policy *pol = blkcg_policy[i];
861 struct blkcg_policy_data *cpd;
862
863 /*
864 * If the policy hasn't been attached yet, wait for it
865 * to be attached before doing anything else. Otherwise,
866 * check if the policy requires any specific per-cgroup
867 * data: if it does, allocate and initialize it.
868 */
869 if (!pol || !pol->cpd_size)
870 continue;
871
872 BUG_ON(blkcg->pd[i]);
873 cpd = kzalloc(pol->cpd_size, GFP_KERNEL);
874 if (!cpd) {
875 ret = ERR_PTR(-ENOMEM);
876 goto free_pd_blkcg;
877 }
878 blkcg->pd[i] = cpd;
879 cpd->plid = i;
880 pol->cpd_init_fn(blkcg);
881 }
882
Vivek Goyal31e4c282009-12-03 12:59:42 -0500883done:
884 spin_lock_init(&blkcg->lock);
Tejun Heoa6371202012-04-19 16:29:24 -0700885 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500886 INIT_HLIST_HEAD(&blkcg->blkg_list);
Tejun Heo52ebea72015-05-22 17:13:37 -0400887#ifdef CONFIG_CGROUP_WRITEBACK
888 INIT_LIST_HEAD(&blkcg->cgwb_list);
889#endif
Tejun Heo7876f932015-07-09 16:39:49 -0400890 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
891
892 mutex_unlock(&blkcg_pol_mutex);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500893 return &blkcg->css;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200894
895free_pd_blkcg:
896 for (i--; i >= 0; i--)
897 kfree(blkcg->pd[i]);
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200898free_blkcg:
899 kfree(blkcg);
Tejun Heo7876f932015-07-09 16:39:49 -0400900 mutex_unlock(&blkcg_pol_mutex);
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200901 return ret;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500902}
903
Tejun Heo5efd6112012-03-05 13:15:12 -0800904/**
905 * blkcg_init_queue - initialize blkcg part of request queue
906 * @q: request_queue to initialize
907 *
908 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
909 * part of new request_queue @q.
910 *
911 * RETURNS:
912 * 0 on success, -errno on failure.
913 */
914int blkcg_init_queue(struct request_queue *q)
915{
Tejun Heoec13b1d2015-05-22 17:13:19 -0400916 struct blkcg_gq *new_blkg, *blkg;
917 bool preloaded;
918 int ret;
Tejun Heo5efd6112012-03-05 13:15:12 -0800919
Tejun Heoec13b1d2015-05-22 17:13:19 -0400920 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
921 if (!new_blkg)
922 return -ENOMEM;
923
924 preloaded = !radix_tree_preload(GFP_KERNEL);
925
926 /*
927 * Make sure the root blkg exists and count the existing blkgs. As
928 * @q is bypassing at this point, blkg_lookup_create() can't be
929 * used. Open code insertion.
930 */
931 rcu_read_lock();
932 spin_lock_irq(q->queue_lock);
933 blkg = blkg_create(&blkcg_root, q, new_blkg);
934 spin_unlock_irq(q->queue_lock);
935 rcu_read_unlock();
936
937 if (preloaded)
938 radix_tree_preload_end();
939
940 if (IS_ERR(blkg)) {
941 kfree(new_blkg);
942 return PTR_ERR(blkg);
943 }
944
945 q->root_blkg = blkg;
946 q->root_rl.blkg = blkg;
947
948 ret = blk_throtl_init(q);
949 if (ret) {
950 spin_lock_irq(q->queue_lock);
951 blkg_destroy_all(q);
952 spin_unlock_irq(q->queue_lock);
953 }
954 return ret;
Tejun Heo5efd6112012-03-05 13:15:12 -0800955}
956
957/**
958 * blkcg_drain_queue - drain blkcg part of request_queue
959 * @q: request_queue to drain
960 *
961 * Called from blk_drain_queue(). Responsible for draining blkcg part.
962 */
963void blkcg_drain_queue(struct request_queue *q)
964{
965 lockdep_assert_held(q->queue_lock);
966
Tejun Heo0b462c82014-07-05 18:43:21 -0400967 /*
968 * @q could be exiting and already have destroyed all blkgs as
969 * indicated by NULL root_blkg. If so, don't confuse policies.
970 */
971 if (!q->root_blkg)
972 return;
973
Tejun Heo5efd6112012-03-05 13:15:12 -0800974 blk_throtl_drain(q);
975}
976
977/**
978 * blkcg_exit_queue - exit and release blkcg part of request_queue
979 * @q: request_queue being released
980 *
981 * Called from blk_release_queue(). Responsible for exiting blkcg part.
982 */
983void blkcg_exit_queue(struct request_queue *q)
984{
Tejun Heo6d18b002012-04-13 13:11:35 -0700985 spin_lock_irq(q->queue_lock);
Tejun Heo3c96cb32012-04-13 13:11:34 -0700986 blkg_destroy_all(q);
Tejun Heo6d18b002012-04-13 13:11:35 -0700987 spin_unlock_irq(q->queue_lock);
988
Tejun Heo5efd6112012-03-05 13:15:12 -0800989 blk_throtl_exit(q);
990}
991
Vivek Goyal31e4c282009-12-03 12:59:42 -0500992/*
993 * We cannot support shared io contexts, as we have no mean to support
994 * two tasks with the same ioc in two different groups without major rework
995 * of the main cic data structures. For now we allow a task to change
996 * its cgroup only if it's the only owner of its ioc.
997 */
Tejun Heoeb954192013-08-08 20:11:23 -0400998static int blkcg_can_attach(struct cgroup_subsys_state *css,
999 struct cgroup_taskset *tset)
Vivek Goyal31e4c282009-12-03 12:59:42 -05001000{
Tejun Heobb9d97b2011-12-12 18:12:21 -08001001 struct task_struct *task;
Vivek Goyal31e4c282009-12-03 12:59:42 -05001002 struct io_context *ioc;
1003 int ret = 0;
1004
1005 /* task_lock() is needed to avoid races with exit_io_context() */
Tejun Heo924f0d92014-02-13 06:58:41 -05001006 cgroup_taskset_for_each(task, tset) {
Tejun Heobb9d97b2011-12-12 18:12:21 -08001007 task_lock(task);
1008 ioc = task->io_context;
1009 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1010 ret = -EINVAL;
1011 task_unlock(task);
1012 if (ret)
1013 break;
1014 }
Vivek Goyal31e4c282009-12-03 12:59:42 -05001015 return ret;
1016}
1017
Tejun Heo073219e2014-02-08 10:36:58 -05001018struct cgroup_subsys blkio_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08001019 .css_alloc = blkcg_css_alloc,
1020 .css_offline = blkcg_css_offline,
1021 .css_free = blkcg_css_free,
Tejun Heo3c798392012-04-16 13:57:25 -07001022 .can_attach = blkcg_can_attach,
Tejun Heo55779642014-07-15 11:05:09 -04001023 .legacy_cftypes = blkcg_files,
Tejun Heo1ced9532014-07-08 18:02:57 -04001024#ifdef CONFIG_MEMCG
1025 /*
1026 * This ensures that, if available, memcg is automatically enabled
1027 * together on the default hierarchy so that the owner cgroup can
1028 * be retrieved from writeback pages.
1029 */
1030 .depends_on = 1 << memory_cgrp_id,
1031#endif
Tejun Heo676f7c82012-04-01 12:09:55 -07001032};
Tejun Heo073219e2014-02-08 10:36:58 -05001033EXPORT_SYMBOL_GPL(blkio_cgrp_subsys);
Tejun Heo676f7c82012-04-01 12:09:55 -07001034
Tejun Heo8bd435b2012-04-13 13:11:28 -07001035/**
Tejun Heoa2b16932012-04-13 13:11:33 -07001036 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1037 * @q: request_queue of interest
1038 * @pol: blkcg policy to activate
1039 *
1040 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1041 * bypass mode to populate its blkgs with policy_data for @pol.
1042 *
1043 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1044 * from IO path. Update of each blkg is protected by both queue and blkcg
1045 * locks so that holding either lock and testing blkcg_policy_enabled() is
1046 * always enough for dereferencing policy data.
1047 *
1048 * The caller is responsible for synchronizing [de]activations and policy
1049 * [un]registerations. Returns 0 on success, -errno on failure.
1050 */
1051int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -07001052 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -07001053{
1054 LIST_HEAD(pds);
Tejun Heoec13b1d2015-05-22 17:13:19 -04001055 struct blkcg_gq *blkg;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001056 struct blkg_policy_data *pd, *nd;
Tejun Heoa2b16932012-04-13 13:11:33 -07001057 int cnt = 0, ret;
1058
1059 if (blkcg_policy_enabled(q, pol))
1060 return 0;
1061
Tejun Heoec13b1d2015-05-22 17:13:19 -04001062 /* count and allocate policy_data for all existing blkgs */
Tejun Heoa2b16932012-04-13 13:11:33 -07001063 blk_queue_bypass_start(q);
Tejun Heoa2b16932012-04-13 13:11:33 -07001064 spin_lock_irq(q->queue_lock);
Tejun Heoa2b16932012-04-13 13:11:33 -07001065 list_for_each_entry(blkg, &q->blkg_list, q_node)
1066 cnt++;
Tejun Heoa2b16932012-04-13 13:11:33 -07001067 spin_unlock_irq(q->queue_lock);
1068
Tejun Heo06b285b2015-07-09 16:39:50 -04001069 /* allocate per-blkg policy data for all existing blkgs */
Tejun Heoa2b16932012-04-13 13:11:33 -07001070 while (cnt--) {
Tejun Heof95a04a2012-04-16 13:57:26 -07001071 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
Tejun Heoa2b16932012-04-13 13:11:33 -07001072 if (!pd) {
1073 ret = -ENOMEM;
1074 goto out_free;
1075 }
1076 list_add_tail(&pd->alloc_node, &pds);
1077 }
1078
1079 /*
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001080 * Install the allocated pds and cpds. With @q bypassing, no new blkg
Tejun Heoa2b16932012-04-13 13:11:33 -07001081 * should have been created while the queue lock was dropped.
1082 */
1083 spin_lock_irq(q->queue_lock);
1084
1085 list_for_each_entry(blkg, &q->blkg_list, q_node) {
Tejun Heo06b285b2015-07-09 16:39:50 -04001086 if (WARN_ON(list_empty(&pds))) {
Tejun Heoa2b16932012-04-13 13:11:33 -07001087 /* umm... this shouldn't happen, just abort */
1088 ret = -ENOMEM;
1089 goto out_unlock;
1090 }
1091 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
1092 list_del_init(&pd->alloc_node);
1093
1094 /* grab blkcg lock too while installing @pd on @blkg */
1095 spin_lock(&blkg->blkcg->lock);
1096
1097 blkg->pd[pol->plid] = pd;
1098 pd->blkg = blkg;
Tejun Heob276a872013-01-09 08:05:12 -08001099 pd->plid = pol->plid;
Tejun Heof9fcc2d2012-04-16 13:57:27 -07001100 pol->pd_init_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -07001101
1102 spin_unlock(&blkg->blkcg->lock);
1103 }
1104
1105 __set_bit(pol->plid, q->blkcg_pols);
1106 ret = 0;
1107out_unlock:
1108 spin_unlock_irq(q->queue_lock);
1109out_free:
1110 blk_queue_bypass_end(q);
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001111 list_for_each_entry_safe(pd, nd, &pds, alloc_node)
Tejun Heoa2b16932012-04-13 13:11:33 -07001112 kfree(pd);
1113 return ret;
1114}
1115EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1116
1117/**
1118 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1119 * @q: request_queue of interest
1120 * @pol: blkcg policy to deactivate
1121 *
1122 * Deactivate @pol on @q. Follows the same synchronization rules as
1123 * blkcg_activate_policy().
1124 */
1125void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -07001126 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -07001127{
Tejun Heo3c798392012-04-16 13:57:25 -07001128 struct blkcg_gq *blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -07001129
1130 if (!blkcg_policy_enabled(q, pol))
1131 return;
1132
1133 blk_queue_bypass_start(q);
1134 spin_lock_irq(q->queue_lock);
1135
1136 __clear_bit(pol->plid, q->blkcg_pols);
1137
1138 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1139 /* grab blkcg lock too while removing @pd from @blkg */
1140 spin_lock(&blkg->blkcg->lock);
1141
Tejun Heof427d902013-01-09 08:05:12 -08001142 if (pol->pd_offline_fn)
1143 pol->pd_offline_fn(blkg);
Tejun Heof9fcc2d2012-04-16 13:57:27 -07001144 if (pol->pd_exit_fn)
1145 pol->pd_exit_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -07001146
1147 kfree(blkg->pd[pol->plid]);
1148 blkg->pd[pol->plid] = NULL;
1149
1150 spin_unlock(&blkg->blkcg->lock);
1151 }
1152
1153 spin_unlock_irq(q->queue_lock);
1154 blk_queue_bypass_end(q);
1155}
1156EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1157
1158/**
Tejun Heo3c798392012-04-16 13:57:25 -07001159 * blkcg_policy_register - register a blkcg policy
1160 * @pol: blkcg policy to register
Tejun Heo8bd435b2012-04-13 13:11:28 -07001161 *
Tejun Heo3c798392012-04-16 13:57:25 -07001162 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1163 * successful registration. Returns 0 on success and -errno on failure.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001164 */
Jens Axboed5bf0292014-06-22 16:31:56 -06001165int blkcg_policy_register(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001166{
Tejun Heo06b285b2015-07-09 16:39:50 -04001167 struct blkcg *blkcg;
Tejun Heo8bd435b2012-04-13 13:11:28 -07001168 int i, ret;
Tejun Heoe8989fa2012-03-05 13:15:20 -08001169
Tejun Heof95a04a2012-04-16 13:57:26 -07001170 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
1171 return -EINVAL;
1172
Tejun Heo838f13b2015-07-09 16:39:47 -04001173 mutex_lock(&blkcg_pol_register_mutex);
Tejun Heobc0d6502012-04-13 13:11:26 -07001174 mutex_lock(&blkcg_pol_mutex);
1175
Tejun Heo8bd435b2012-04-13 13:11:28 -07001176 /* find an empty slot */
1177 ret = -ENOSPC;
1178 for (i = 0; i < BLKCG_MAX_POLS; i++)
Tejun Heo3c798392012-04-16 13:57:25 -07001179 if (!blkcg_policy[i])
Tejun Heo8bd435b2012-04-13 13:11:28 -07001180 break;
1181 if (i >= BLKCG_MAX_POLS)
Tejun Heo838f13b2015-07-09 16:39:47 -04001182 goto err_unlock;
Tejun Heo035d10b2012-03-05 13:15:04 -08001183
Tejun Heo06b285b2015-07-09 16:39:50 -04001184 /* register @pol */
Tejun Heo3c798392012-04-16 13:57:25 -07001185 pol->plid = i;
Tejun Heo06b285b2015-07-09 16:39:50 -04001186 blkcg_policy[pol->plid] = pol;
1187
1188 /* allocate and install cpd's */
1189 if (pol->cpd_size) {
1190 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1191 struct blkcg_policy_data *cpd;
1192
1193 cpd = kzalloc(pol->cpd_size, GFP_KERNEL);
1194 if (!cpd) {
1195 mutex_unlock(&blkcg_pol_mutex);
1196 goto err_free_cpds;
1197 }
1198
1199 blkcg->pd[pol->plid] = cpd;
1200 cpd->plid = pol->plid;
1201 pol->cpd_init_fn(blkcg);
1202 }
1203 }
1204
Tejun Heo838f13b2015-07-09 16:39:47 -04001205 mutex_unlock(&blkcg_pol_mutex);
Tejun Heo8bd435b2012-04-13 13:11:28 -07001206
Tejun Heo8bd435b2012-04-13 13:11:28 -07001207 /* everything is in place, add intf files for the new policy */
Tejun Heo3c798392012-04-16 13:57:25 -07001208 if (pol->cftypes)
Tejun Heo2cf669a2014-07-15 11:05:09 -04001209 WARN_ON(cgroup_add_legacy_cftypes(&blkio_cgrp_subsys,
1210 pol->cftypes));
Tejun Heo838f13b2015-07-09 16:39:47 -04001211 mutex_unlock(&blkcg_pol_register_mutex);
1212 return 0;
1213
Tejun Heo06b285b2015-07-09 16:39:50 -04001214err_free_cpds:
1215 if (pol->cpd_size) {
1216 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1217 kfree(blkcg->pd[pol->plid]);
1218 blkcg->pd[pol->plid] = NULL;
1219 }
1220 }
1221 blkcg_policy[pol->plid] = NULL;
Tejun Heo838f13b2015-07-09 16:39:47 -04001222err_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -07001223 mutex_unlock(&blkcg_pol_mutex);
Tejun Heo838f13b2015-07-09 16:39:47 -04001224 mutex_unlock(&blkcg_pol_register_mutex);
Tejun Heo8bd435b2012-04-13 13:11:28 -07001225 return ret;
Vivek Goyal3e252062009-12-04 10:36:42 -05001226}
Tejun Heo3c798392012-04-16 13:57:25 -07001227EXPORT_SYMBOL_GPL(blkcg_policy_register);
Vivek Goyal3e252062009-12-04 10:36:42 -05001228
Tejun Heo8bd435b2012-04-13 13:11:28 -07001229/**
Tejun Heo3c798392012-04-16 13:57:25 -07001230 * blkcg_policy_unregister - unregister a blkcg policy
1231 * @pol: blkcg policy to unregister
Tejun Heo8bd435b2012-04-13 13:11:28 -07001232 *
Tejun Heo3c798392012-04-16 13:57:25 -07001233 * Undo blkcg_policy_register(@pol). Might sleep.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001234 */
Tejun Heo3c798392012-04-16 13:57:25 -07001235void blkcg_policy_unregister(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001236{
Tejun Heo06b285b2015-07-09 16:39:50 -04001237 struct blkcg *blkcg;
1238
Tejun Heo838f13b2015-07-09 16:39:47 -04001239 mutex_lock(&blkcg_pol_register_mutex);
Tejun Heobc0d6502012-04-13 13:11:26 -07001240
Tejun Heo3c798392012-04-16 13:57:25 -07001241 if (WARN_ON(blkcg_policy[pol->plid] != pol))
Tejun Heo8bd435b2012-04-13 13:11:28 -07001242 goto out_unlock;
1243
1244 /* kill the intf files first */
Tejun Heo3c798392012-04-16 13:57:25 -07001245 if (pol->cftypes)
Tejun Heo2bb566c2013-08-08 20:11:23 -04001246 cgroup_rm_cftypes(pol->cftypes);
Tejun Heo44ea53d2012-04-01 14:38:43 -07001247
Tejun Heo06b285b2015-07-09 16:39:50 -04001248 /* remove cpds and unregister */
Tejun Heo838f13b2015-07-09 16:39:47 -04001249 mutex_lock(&blkcg_pol_mutex);
Tejun Heo06b285b2015-07-09 16:39:50 -04001250
1251 if (pol->cpd_size) {
1252 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1253 kfree(blkcg->pd[pol->plid]);
1254 blkcg->pd[pol->plid] = NULL;
1255 }
1256 }
Tejun Heo3c798392012-04-16 13:57:25 -07001257 blkcg_policy[pol->plid] = NULL;
Tejun Heo06b285b2015-07-09 16:39:50 -04001258
Tejun Heobc0d6502012-04-13 13:11:26 -07001259 mutex_unlock(&blkcg_pol_mutex);
Tejun Heo838f13b2015-07-09 16:39:47 -04001260out_unlock:
1261 mutex_unlock(&blkcg_pol_register_mutex);
Vivek Goyal3e252062009-12-04 10:36:42 -05001262}
Tejun Heo3c798392012-04-16 13:57:25 -07001263EXPORT_SYMBOL_GPL(blkcg_policy_unregister);