blob: af665fe7f4f2db2ed2241325790eada804ec2520 [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>
12 */
13#include <linux/ioprio.h>
Vivek Goyal22084192009-12-03 12:59:49 -050014#include <linux/kdev_t.h>
Vivek Goyal9d6a9862009-12-04 10:36:41 -050015#include <linux/module.h>
Stephen Rothwellaccee782009-12-07 19:29:39 +110016#include <linux/err.h>
Divyesh Shah91952912010-04-01 15:01:41 -070017#include <linux/blkdev.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090018#include <linux/slab.h>
Gui Jianfeng34d0f172010-04-13 16:05:49 +080019#include <linux/genhd.h>
Tejun Heo72e06c22012-03-05 13:15:00 -080020#include <linux/delay.h>
Tejun Heo9a9e8a22012-03-19 15:10:56 -070021#include <linux/atomic.h>
Tejun Heo72e06c22012-03-05 13:15:00 -080022#include "blk-cgroup.h"
Tejun Heo5efd6112012-03-05 13:15:12 -080023#include "blk.h"
Vivek Goyal3e252062009-12-04 10:36:42 -050024
Divyesh Shah84c124d2010-04-09 08:31:19 +020025#define MAX_KEY_LEN 100
26
Tejun Heobc0d6502012-04-13 13:11:26 -070027static DEFINE_MUTEX(blkcg_pol_mutex);
Tejun Heo923adde2012-03-05 13:15:13 -080028static DEFINE_MUTEX(all_q_mutex);
29static LIST_HEAD(all_q_list);
30
Tejun Heo3381cb82012-04-01 14:38:44 -070031struct blkio_cgroup blkio_root_cgroup = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT };
Vivek Goyal9d6a9862009-12-04 10:36:41 -050032EXPORT_SYMBOL_GPL(blkio_root_cgroup);
33
Tejun Heo035d10b2012-03-05 13:15:04 -080034static struct blkio_policy_type *blkio_policy[BLKIO_NR_POLICIES];
35
Vivek Goyal31e4c282009-12-03 12:59:42 -050036struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
37{
38 return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
39 struct blkio_cgroup, css);
40}
Vivek Goyal9d6a9862009-12-04 10:36:41 -050041EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
Vivek Goyal31e4c282009-12-03 12:59:42 -050042
Tejun Heo4f85cb92012-03-05 13:15:28 -080043static struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk)
Vivek Goyal70087dc2011-05-16 15:24:08 +020044{
45 return container_of(task_subsys_state(tsk, blkio_subsys_id),
46 struct blkio_cgroup, css);
47}
Tejun Heo4f85cb92012-03-05 13:15:28 -080048
49struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio)
50{
51 if (bio && bio->bi_css)
52 return container_of(bio->bi_css, struct blkio_cgroup, css);
53 return task_blkio_cgroup(current);
54}
55EXPORT_SYMBOL_GPL(bio_blkio_cgroup);
Vivek Goyal70087dc2011-05-16 15:24:08 +020056
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 */
63static void blkg_free(struct blkio_group *blkg)
64{
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 Heoe8989fa2012-03-05 13:15:20 -080070 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
Tejun Heo9ade5ea2012-04-01 14:38:44 -070071 struct blkio_policy_type *pol = blkio_policy[i];
Tejun Heoe8989fa2012-03-05 13:15:20 -080072 struct blkg_policy_data *pd = blkg->pd[i];
73
Tejun Heo9ade5ea2012-04-01 14:38:44 -070074 if (!pd)
75 continue;
76
77 if (pol && pol->ops.blkio_exit_group_fn)
78 pol->ops.blkio_exit_group_fn(blkg);
79
Tejun Heo9ade5ea2012-04-01 14:38:44 -070080 kfree(pd);
Tejun Heo03814112012-03-05 13:15:14 -080081 }
Tejun Heoe8989fa2012-03-05 13:15:20 -080082
Tejun Heo549d3aa2012-03-05 13:15:16 -080083 kfree(blkg);
Tejun Heo03814112012-03-05 13:15:14 -080084}
85
86/**
87 * blkg_alloc - allocate a blkg
88 * @blkcg: block cgroup the new blkg is associated with
89 * @q: request_queue the new blkg is associated with
Tejun Heo03814112012-03-05 13:15:14 -080090 *
Tejun Heoe8989fa2012-03-05 13:15:20 -080091 * Allocate a new blkg assocating @blkcg and @q.
Tejun Heo03814112012-03-05 13:15:14 -080092 */
93static struct blkio_group *blkg_alloc(struct blkio_cgroup *blkcg,
Tejun Heoe8989fa2012-03-05 13:15:20 -080094 struct request_queue *q)
Tejun Heo03814112012-03-05 13:15:14 -080095{
96 struct blkio_group *blkg;
Tejun Heoe8989fa2012-03-05 13:15:20 -080097 int i;
Tejun Heo03814112012-03-05 13:15:14 -080098
99 /* alloc and init base part */
100 blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node);
101 if (!blkg)
102 return NULL;
103
Tejun Heoc875f4d2012-03-05 13:15:22 -0800104 blkg->q = q;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800105 INIT_LIST_HEAD(&blkg->q_node);
Tejun Heo03814112012-03-05 13:15:14 -0800106 blkg->blkcg = blkcg;
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800107 blkg->refcnt = 1;
Tejun Heo03814112012-03-05 13:15:14 -0800108 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
109
Tejun Heoe8989fa2012-03-05 13:15:20 -0800110 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
111 struct blkio_policy_type *pol = blkio_policy[i];
112 struct blkg_policy_data *pd;
Tejun Heo03814112012-03-05 13:15:14 -0800113
Tejun Heoe8989fa2012-03-05 13:15:20 -0800114 if (!pol)
115 continue;
Tejun Heo549d3aa2012-03-05 13:15:16 -0800116
Tejun Heoe8989fa2012-03-05 13:15:20 -0800117 /* alloc per-policy data and attach it to blkg */
118 pd = kzalloc_node(sizeof(*pd) + pol->pdata_size, GFP_ATOMIC,
119 q->node);
120 if (!pd) {
121 blkg_free(blkg);
122 return NULL;
123 }
Tejun Heo549d3aa2012-03-05 13:15:16 -0800124
Tejun Heoe8989fa2012-03-05 13:15:20 -0800125 blkg->pd[i] = pd;
126 pd->blkg = blkg;
Tejun Heo03814112012-03-05 13:15:14 -0800127 }
128
Tejun Heo549d3aa2012-03-05 13:15:16 -0800129 /* invoke per-policy init */
Tejun Heoe8989fa2012-03-05 13:15:20 -0800130 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
131 struct blkio_policy_type *pol = blkio_policy[i];
132
133 if (pol)
134 pol->ops.blkio_init_group_fn(blkg);
135 }
136
Tejun Heo03814112012-03-05 13:15:14 -0800137 return blkg;
138}
139
Tejun Heocd1604f2012-03-05 13:15:06 -0800140struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg,
141 struct request_queue *q,
Tejun Heocd1604f2012-03-05 13:15:06 -0800142 bool for_root)
143 __releases(q->queue_lock) __acquires(q->queue_lock)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400144{
Vivek Goyal1cd9e032012-03-08 10:53:56 -0800145 struct blkio_group *blkg;
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400146
Tejun Heocd1604f2012-03-05 13:15:06 -0800147 WARN_ON_ONCE(!rcu_read_lock_held());
148 lockdep_assert_held(q->queue_lock);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500149
Tejun Heocd1604f2012-03-05 13:15:06 -0800150 /*
151 * This could be the first entry point of blkcg implementation and
152 * we shouldn't allow anything to go through for a bypassing queue.
153 * The following can be removed if blkg lookup is guaranteed to
154 * fail on a bypassing queue.
155 */
156 if (unlikely(blk_queue_bypass(q)) && !for_root)
157 return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY);
158
Tejun Heoe8989fa2012-03-05 13:15:20 -0800159 blkg = blkg_lookup(blkcg, q);
Tejun Heocd1604f2012-03-05 13:15:06 -0800160 if (blkg)
161 return blkg;
162
Tejun Heo7ee9c562012-03-05 13:15:11 -0800163 /* blkg holds a reference to blkcg */
Tejun Heocd1604f2012-03-05 13:15:06 -0800164 if (!css_tryget(&blkcg->css))
165 return ERR_PTR(-EINVAL);
166
167 /*
168 * Allocate and initialize.
Tejun Heocd1604f2012-03-05 13:15:06 -0800169 */
Vivek Goyal1cd9e032012-03-08 10:53:56 -0800170 blkg = blkg_alloc(blkcg, q);
Tejun Heocd1604f2012-03-05 13:15:06 -0800171
172 /* did alloc fail? */
Vivek Goyal1cd9e032012-03-08 10:53:56 -0800173 if (unlikely(!blkg)) {
Tejun Heocd1604f2012-03-05 13:15:06 -0800174 blkg = ERR_PTR(-ENOMEM);
175 goto out;
176 }
177
178 /* insert */
179 spin_lock(&blkcg->lock);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500180 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
Tejun Heoe8989fa2012-03-05 13:15:20 -0800181 list_add(&blkg->q_node, &q->blkg_list);
Tejun Heocd1604f2012-03-05 13:15:06 -0800182 spin_unlock(&blkcg->lock);
183out:
Tejun Heocd1604f2012-03-05 13:15:06 -0800184 return blkg;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500185}
Tejun Heocd1604f2012-03-05 13:15:06 -0800186EXPORT_SYMBOL_GPL(blkg_lookup_create);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500187
Vivek Goyal31e4c282009-12-03 12:59:42 -0500188/* called under rcu_read_lock(). */
Tejun Heocd1604f2012-03-05 13:15:06 -0800189struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
Tejun Heoe8989fa2012-03-05 13:15:20 -0800190 struct request_queue *q)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500191{
192 struct blkio_group *blkg;
193 struct hlist_node *n;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500194
Tejun Heoca32aef2012-03-05 13:15:03 -0800195 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node)
Tejun Heoe8989fa2012-03-05 13:15:20 -0800196 if (blkg->q == q)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500197 return blkg;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500198 return NULL;
199}
Tejun Heocd1604f2012-03-05 13:15:06 -0800200EXPORT_SYMBOL_GPL(blkg_lookup);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500201
Tejun Heoe8989fa2012-03-05 13:15:20 -0800202static void blkg_destroy(struct blkio_group *blkg)
Tejun Heo72e06c22012-03-05 13:15:00 -0800203{
Tejun Heo03aa2642012-03-05 13:15:19 -0800204 struct request_queue *q = blkg->q;
Tejun Heo9f13ef62012-03-05 13:15:21 -0800205 struct blkio_cgroup *blkcg = blkg->blkcg;
Tejun Heo03aa2642012-03-05 13:15:19 -0800206
207 lockdep_assert_held(q->queue_lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800208 lockdep_assert_held(&blkcg->lock);
Tejun Heo03aa2642012-03-05 13:15:19 -0800209
210 /* Something wrong if we are trying to remove same group twice */
Tejun Heoe8989fa2012-03-05 13:15:20 -0800211 WARN_ON_ONCE(list_empty(&blkg->q_node));
Tejun Heo9f13ef62012-03-05 13:15:21 -0800212 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
Tejun Heoe8989fa2012-03-05 13:15:20 -0800213 list_del_init(&blkg->q_node);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800214 hlist_del_init_rcu(&blkg->blkcg_node);
Tejun Heo03aa2642012-03-05 13:15:19 -0800215
Tejun Heo03aa2642012-03-05 13:15:19 -0800216 /*
217 * Put the reference taken at the time of creation so that when all
218 * queues are gone, group can be destroyed.
219 */
220 blkg_put(blkg);
221}
222
Tejun Heoe8989fa2012-03-05 13:15:20 -0800223/*
224 * XXX: This updates blkg policy data in-place for root blkg, which is
225 * necessary across elevator switch and policy registration as root blkgs
226 * aren't shot down. This broken and racy implementation is temporary.
227 * Eventually, blkg shoot down will be replaced by proper in-place update.
228 */
229void update_root_blkg_pd(struct request_queue *q, enum blkio_policy_id plid)
230{
231 struct blkio_policy_type *pol = blkio_policy[plid];
232 struct blkio_group *blkg = blkg_lookup(&blkio_root_cgroup, q);
233 struct blkg_policy_data *pd;
234
235 if (!blkg)
236 return;
237
238 kfree(blkg->pd[plid]);
239 blkg->pd[plid] = NULL;
240
241 if (!pol)
242 return;
243
244 pd = kzalloc(sizeof(*pd) + pol->pdata_size, GFP_KERNEL);
245 WARN_ON_ONCE(!pd);
246
Tejun Heoe8989fa2012-03-05 13:15:20 -0800247 blkg->pd[plid] = pd;
248 pd->blkg = blkg;
249 pol->ops.blkio_init_group_fn(blkg);
250}
251EXPORT_SYMBOL_GPL(update_root_blkg_pd);
252
Tejun Heo9f13ef62012-03-05 13:15:21 -0800253/**
254 * blkg_destroy_all - destroy all blkgs associated with a request_queue
255 * @q: request_queue of interest
256 * @destroy_root: whether to destroy root blkg or not
257 *
258 * Destroy blkgs associated with @q. If @destroy_root is %true, all are
259 * destroyed; otherwise, root blkg is left alone.
260 */
Tejun Heoe8989fa2012-03-05 13:15:20 -0800261void blkg_destroy_all(struct request_queue *q, bool destroy_root)
Tejun Heo03aa2642012-03-05 13:15:19 -0800262{
263 struct blkio_group *blkg, *n;
Tejun Heo72e06c22012-03-05 13:15:00 -0800264
Tejun Heo9f13ef62012-03-05 13:15:21 -0800265 spin_lock_irq(q->queue_lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800266
Tejun Heo9f13ef62012-03-05 13:15:21 -0800267 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
268 struct blkio_cgroup *blkcg = blkg->blkcg;
Tejun Heo72e06c22012-03-05 13:15:00 -0800269
Tejun Heo9f13ef62012-03-05 13:15:21 -0800270 /* skip root? */
271 if (!destroy_root && blkg->blkcg == &blkio_root_cgroup)
272 continue;
Tejun Heo03aa2642012-03-05 13:15:19 -0800273
Tejun Heo9f13ef62012-03-05 13:15:21 -0800274 spin_lock(&blkcg->lock);
275 blkg_destroy(blkg);
276 spin_unlock(&blkcg->lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800277 }
Tejun Heo9f13ef62012-03-05 13:15:21 -0800278
279 spin_unlock_irq(q->queue_lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800280}
Tejun Heo03aa2642012-03-05 13:15:19 -0800281EXPORT_SYMBOL_GPL(blkg_destroy_all);
Tejun Heo72e06c22012-03-05 13:15:00 -0800282
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800283static void blkg_rcu_free(struct rcu_head *rcu_head)
284{
285 blkg_free(container_of(rcu_head, struct blkio_group, rcu_head));
286}
287
288void __blkg_release(struct blkio_group *blkg)
289{
290 /* release the extra blkcg reference this blkg has been holding */
291 css_put(&blkg->blkcg->css);
292
293 /*
294 * A group is freed in rcu manner. But having an rcu lock does not
295 * mean that one can access all the fields of blkg and assume these
296 * are valid. For example, don't try to follow throtl_data and
297 * request queue links.
298 *
299 * Having a reference to blkg under an rcu allows acess to only
300 * values local to groups like group stats and group rate limits
301 */
302 call_rcu(&blkg->rcu_head, blkg_rcu_free);
303}
304EXPORT_SYMBOL_GPL(__blkg_release);
305
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700306static int
Divyesh Shah84c124d2010-04-09 08:31:19 +0200307blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700308{
Tejun Heo997a0262012-03-08 10:53:58 -0800309 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700310 struct blkio_group *blkg;
311 struct hlist_node *n;
Tejun Heobc0d6502012-04-13 13:11:26 -0700312 int i;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700313
Tejun Heobc0d6502012-04-13 13:11:26 -0700314 mutex_lock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700315 spin_lock_irq(&blkcg->lock);
Tejun Heo997a0262012-03-08 10:53:58 -0800316
317 /*
318 * Note that stat reset is racy - it doesn't synchronize against
319 * stat updates. This is a debug feature which shouldn't exist
320 * anyway. If you get hit by a race, retry.
321 */
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700322 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
Tejun Heobc0d6502012-04-13 13:11:26 -0700323 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
324 struct blkio_policy_type *pol = blkio_policy[i];
Tejun Heo549d3aa2012-03-05 13:15:16 -0800325
Tejun Heobc0d6502012-04-13 13:11:26 -0700326 if (pol && pol->ops.blkio_reset_group_stats_fn)
Tejun Heo9ade5ea2012-04-01 14:38:44 -0700327 pol->ops.blkio_reset_group_stats_fn(blkg);
Tejun Heobc0d6502012-04-13 13:11:26 -0700328 }
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700329 }
Vivek Goyalf0bdc8c2011-05-19 15:38:30 -0400330
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700331 spin_unlock_irq(&blkcg->lock);
Tejun Heobc0d6502012-04-13 13:11:26 -0700332 mutex_unlock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700333 return 0;
334}
335
Tejun Heod3d32e62012-04-01 14:38:42 -0700336static const char *blkg_dev_name(struct blkio_group *blkg)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700337{
Tejun Heod3d32e62012-04-01 14:38:42 -0700338 /* some drivers (floppy) instantiate a queue w/o disk registered */
339 if (blkg->q->backing_dev_info.dev)
340 return dev_name(blkg->q->backing_dev_info.dev);
341 return NULL;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700342}
343
Tejun Heod3d32e62012-04-01 14:38:42 -0700344/**
345 * blkcg_print_blkgs - helper for printing per-blkg data
346 * @sf: seq_file to print to
347 * @blkcg: blkcg of interest
348 * @prfill: fill function to print out a blkg
349 * @pol: policy in question
350 * @data: data to be passed to @prfill
351 * @show_total: to print out sum of prfill return values or not
352 *
353 * This function invokes @prfill on each blkg of @blkcg if pd for the
354 * policy specified by @pol exists. @prfill is invoked with @sf, the
355 * policy data and @data. If @show_total is %true, the sum of the return
356 * values from @prfill is printed with "Total" label at the end.
357 *
358 * This is to be used to construct print functions for
359 * cftype->read_seq_string method.
360 */
Tejun Heo829fdb52012-04-01 14:38:43 -0700361void blkcg_print_blkgs(struct seq_file *sf, struct blkio_cgroup *blkcg,
Tejun Heod366e7e2012-04-01 14:38:44 -0700362 u64 (*prfill)(struct seq_file *, void *, int),
Tejun Heo829fdb52012-04-01 14:38:43 -0700363 int pol, int data, bool show_total)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400364{
Tejun Heod3d32e62012-04-01 14:38:42 -0700365 struct blkio_group *blkg;
366 struct hlist_node *n;
367 u64 total = 0;
368
369 spin_lock_irq(&blkcg->lock);
370 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node)
371 if (blkg->pd[pol])
Tejun Heod366e7e2012-04-01 14:38:44 -0700372 total += prfill(sf, blkg->pd[pol]->pdata, data);
Tejun Heod3d32e62012-04-01 14:38:42 -0700373 spin_unlock_irq(&blkcg->lock);
374
375 if (show_total)
376 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
377}
Tejun Heo829fdb52012-04-01 14:38:43 -0700378EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
Tejun Heod3d32e62012-04-01 14:38:42 -0700379
380/**
381 * __blkg_prfill_u64 - prfill helper for a single u64 value
382 * @sf: seq_file to print to
Tejun Heod366e7e2012-04-01 14:38:44 -0700383 * @pdata: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700384 * @v: value to print
385 *
Tejun Heod366e7e2012-04-01 14:38:44 -0700386 * Print @v to @sf for the device assocaited with @pdata.
Tejun Heod3d32e62012-04-01 14:38:42 -0700387 */
Tejun Heod366e7e2012-04-01 14:38:44 -0700388u64 __blkg_prfill_u64(struct seq_file *sf, void *pdata, u64 v)
Tejun Heod3d32e62012-04-01 14:38:42 -0700389{
Tejun Heod366e7e2012-04-01 14:38:44 -0700390 const char *dname = blkg_dev_name(pdata_to_blkg(pdata));
Tejun Heod3d32e62012-04-01 14:38:42 -0700391
392 if (!dname)
393 return 0;
394
395 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
396 return v;
397}
Tejun Heo829fdb52012-04-01 14:38:43 -0700398EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
Tejun Heod3d32e62012-04-01 14:38:42 -0700399
400/**
401 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
402 * @sf: seq_file to print to
Tejun Heod366e7e2012-04-01 14:38:44 -0700403 * @pdata: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700404 * @rwstat: rwstat to print
405 *
Tejun Heod366e7e2012-04-01 14:38:44 -0700406 * Print @rwstat to @sf for the device assocaited with @pdata.
Tejun Heod3d32e62012-04-01 14:38:42 -0700407 */
Tejun Heod366e7e2012-04-01 14:38:44 -0700408u64 __blkg_prfill_rwstat(struct seq_file *sf, void *pdata,
Tejun Heo829fdb52012-04-01 14:38:43 -0700409 const struct blkg_rwstat *rwstat)
Tejun Heod3d32e62012-04-01 14:38:42 -0700410{
411 static const char *rwstr[] = {
412 [BLKG_RWSTAT_READ] = "Read",
413 [BLKG_RWSTAT_WRITE] = "Write",
414 [BLKG_RWSTAT_SYNC] = "Sync",
415 [BLKG_RWSTAT_ASYNC] = "Async",
416 };
Tejun Heod366e7e2012-04-01 14:38:44 -0700417 const char *dname = blkg_dev_name(pdata_to_blkg(pdata));
Tejun Heod3d32e62012-04-01 14:38:42 -0700418 u64 v;
419 int i;
420
421 if (!dname)
422 return 0;
423
424 for (i = 0; i < BLKG_RWSTAT_NR; i++)
425 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
426 (unsigned long long)rwstat->cnt[i]);
427
428 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
429 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
430 return v;
431}
432
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700433/**
434 * blkg_prfill_stat - prfill callback for blkg_stat
435 * @sf: seq_file to print to
436 * @pdata: policy private data of interest
437 * @off: offset to the blkg_stat in @pdata
438 *
439 * prfill callback for printing a blkg_stat.
440 */
441u64 blkg_prfill_stat(struct seq_file *sf, void *pdata, int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700442{
Tejun Heod366e7e2012-04-01 14:38:44 -0700443 return __blkg_prfill_u64(sf, pdata, blkg_stat_read(pdata + off));
Tejun Heod3d32e62012-04-01 14:38:42 -0700444}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700445EXPORT_SYMBOL_GPL(blkg_prfill_stat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700446
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700447/**
448 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
449 * @sf: seq_file to print to
450 * @pdata: policy private data of interest
451 * @off: offset to the blkg_rwstat in @pdata
452 *
453 * prfill callback for printing a blkg_rwstat.
454 */
455u64 blkg_prfill_rwstat(struct seq_file *sf, void *pdata, int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700456{
Tejun Heod366e7e2012-04-01 14:38:44 -0700457 struct blkg_rwstat rwstat = blkg_rwstat_read(pdata + off);
Tejun Heod3d32e62012-04-01 14:38:42 -0700458
Tejun Heod366e7e2012-04-01 14:38:44 -0700459 return __blkg_prfill_rwstat(sf, pdata, &rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700460}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700461EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700462
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700463/**
464 * blkg_conf_prep - parse and prepare for per-blkg config update
465 * @blkcg: target block cgroup
466 * @input: input string
467 * @ctx: blkg_conf_ctx to be filled
468 *
469 * Parse per-blkg config update from @input and initialize @ctx with the
470 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
471 * value. This function returns with RCU read locked and must be paired
472 * with blkg_conf_finish().
473 */
Tejun Heo829fdb52012-04-01 14:38:43 -0700474int blkg_conf_prep(struct blkio_cgroup *blkcg, const char *input,
475 struct blkg_conf_ctx *ctx)
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700476 __acquires(rcu)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800477{
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700478 struct gendisk *disk;
479 struct blkio_group *blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700480 unsigned int major, minor;
481 unsigned long long v;
482 int part, ret;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800483
Tejun Heo726fa692012-04-01 14:38:43 -0700484 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
485 return -EINVAL;
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700486
Tejun Heo726fa692012-04-01 14:38:43 -0700487 disk = get_gendisk(MKDEV(major, minor), &part);
Tejun Heo4bfd4822012-03-05 13:15:08 -0800488 if (!disk || part)
Tejun Heo726fa692012-04-01 14:38:43 -0700489 return -EINVAL;
Tejun Heoe56da7e2012-03-05 13:15:07 -0800490
491 rcu_read_lock();
492
Tejun Heo4bfd4822012-03-05 13:15:08 -0800493 spin_lock_irq(disk->queue->queue_lock);
Tejun Heoaaec55a2012-04-01 14:38:42 -0700494 blkg = blkg_lookup_create(blkcg, disk->queue, false);
Tejun Heo4bfd4822012-03-05 13:15:08 -0800495 spin_unlock_irq(disk->queue->queue_lock);
Tejun Heoe56da7e2012-03-05 13:15:07 -0800496
Tejun Heo4bfd4822012-03-05 13:15:08 -0800497 if (IS_ERR(blkg)) {
498 ret = PTR_ERR(blkg);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700499 rcu_read_unlock();
500 put_disk(disk);
501 /*
502 * If queue was bypassing, we should retry. Do so after a
503 * short msleep(). It isn't strictly necessary but queue
504 * can be bypassing for some time and it's always nice to
505 * avoid busy looping.
506 */
507 if (ret == -EBUSY) {
508 msleep(10);
509 ret = restart_syscall();
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400510 }
Tejun Heo726fa692012-04-01 14:38:43 -0700511 return ret;
Vivek Goyal062a6442010-09-15 17:06:33 -0400512 }
Tejun Heoe56da7e2012-03-05 13:15:07 -0800513
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700514 ctx->disk = disk;
515 ctx->blkg = blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700516 ctx->v = v;
517 return 0;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800518}
Tejun Heo829fdb52012-04-01 14:38:43 -0700519EXPORT_SYMBOL_GPL(blkg_conf_prep);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800520
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700521/**
522 * blkg_conf_finish - finish up per-blkg config update
523 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
524 *
525 * Finish up after per-blkg config update. This function must be paired
526 * with blkg_conf_prep().
527 */
Tejun Heo829fdb52012-04-01 14:38:43 -0700528void blkg_conf_finish(struct blkg_conf_ctx *ctx)
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700529 __releases(rcu)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800530{
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700531 rcu_read_unlock();
532 put_disk(ctx->disk);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800533}
Tejun Heo829fdb52012-04-01 14:38:43 -0700534EXPORT_SYMBOL_GPL(blkg_conf_finish);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800535
Vivek Goyal31e4c282009-12-03 12:59:42 -0500536struct cftype blkio_files[] = {
537 {
Divyesh Shah84c124d2010-04-09 08:31:19 +0200538 .name = "reset_stats",
539 .write_u64 = blkiocg_reset_stats,
Vivek Goyal22084192009-12-03 12:59:49 -0500540 },
Tejun Heo4baf6e32012-04-01 12:09:55 -0700541 { } /* terminate */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500542};
543
Tejun Heo9f13ef62012-03-05 13:15:21 -0800544/**
545 * blkiocg_pre_destroy - cgroup pre_destroy callback
Tejun Heo9f13ef62012-03-05 13:15:21 -0800546 * @cgroup: cgroup of interest
547 *
548 * This function is called when @cgroup is about to go away and responsible
549 * for shooting down all blkgs associated with @cgroup. blkgs should be
550 * removed while holding both q and blkcg locks. As blkcg lock is nested
551 * inside q lock, this function performs reverse double lock dancing.
552 *
553 * This is the blkcg counterpart of ioc_release_fn().
554 */
Tejun Heo959d8512012-04-01 12:30:01 -0700555static int blkiocg_pre_destroy(struct cgroup *cgroup)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500556{
557 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
558
Tejun Heo9f13ef62012-03-05 13:15:21 -0800559 spin_lock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800560
Tejun Heo9f13ef62012-03-05 13:15:21 -0800561 while (!hlist_empty(&blkcg->blkg_list)) {
562 struct blkio_group *blkg = hlist_entry(blkcg->blkg_list.first,
563 struct blkio_group, blkcg_node);
Tejun Heoc875f4d2012-03-05 13:15:22 -0800564 struct request_queue *q = blkg->q;
Vivek Goyalb1c35762009-12-03 12:59:47 -0500565
Tejun Heo9f13ef62012-03-05 13:15:21 -0800566 if (spin_trylock(q->queue_lock)) {
567 blkg_destroy(blkg);
568 spin_unlock(q->queue_lock);
569 } else {
570 spin_unlock_irq(&blkcg->lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800571 cpu_relax();
Dan Carpentera5567932012-03-29 20:57:08 +0200572 spin_lock_irq(&blkcg->lock);
Jens Axboe0f3942a2010-05-03 14:28:55 +0200573 }
Tejun Heo9f13ef62012-03-05 13:15:21 -0800574 }
Jens Axboe0f3942a2010-05-03 14:28:55 +0200575
Tejun Heo9f13ef62012-03-05 13:15:21 -0800576 spin_unlock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800577 return 0;
578}
579
Li Zefan761b3ef2012-01-31 13:47:36 +0800580static void blkiocg_destroy(struct cgroup *cgroup)
Tejun Heo7ee9c562012-03-05 13:15:11 -0800581{
582 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
583
Ben Blum67523c42010-03-10 15:22:11 -0800584 if (blkcg != &blkio_root_cgroup)
585 kfree(blkcg);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500586}
587
Li Zefan761b3ef2012-01-31 13:47:36 +0800588static struct cgroup_subsys_state *blkiocg_create(struct cgroup *cgroup)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500589{
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700590 static atomic64_t id_seq = ATOMIC64_INIT(0);
Li Zefan03415092010-05-07 08:57:00 +0200591 struct blkio_cgroup *blkcg;
592 struct cgroup *parent = cgroup->parent;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500593
Li Zefan03415092010-05-07 08:57:00 +0200594 if (!parent) {
Vivek Goyal31e4c282009-12-03 12:59:42 -0500595 blkcg = &blkio_root_cgroup;
596 goto done;
597 }
598
Vivek Goyal31e4c282009-12-03 12:59:42 -0500599 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
600 if (!blkcg)
601 return ERR_PTR(-ENOMEM);
602
Tejun Heo3381cb82012-04-01 14:38:44 -0700603 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700604 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500605done:
606 spin_lock_init(&blkcg->lock);
607 INIT_HLIST_HEAD(&blkcg->blkg_list);
608
609 return &blkcg->css;
610}
611
Tejun Heo5efd6112012-03-05 13:15:12 -0800612/**
613 * blkcg_init_queue - initialize blkcg part of request queue
614 * @q: request_queue to initialize
615 *
616 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
617 * part of new request_queue @q.
618 *
619 * RETURNS:
620 * 0 on success, -errno on failure.
621 */
622int blkcg_init_queue(struct request_queue *q)
623{
Tejun Heo923adde2012-03-05 13:15:13 -0800624 int ret;
625
Tejun Heo5efd6112012-03-05 13:15:12 -0800626 might_sleep();
627
Tejun Heo923adde2012-03-05 13:15:13 -0800628 ret = blk_throtl_init(q);
629 if (ret)
630 return ret;
631
632 mutex_lock(&all_q_mutex);
633 INIT_LIST_HEAD(&q->all_q_node);
634 list_add_tail(&q->all_q_node, &all_q_list);
635 mutex_unlock(&all_q_mutex);
636
637 return 0;
Tejun Heo5efd6112012-03-05 13:15:12 -0800638}
639
640/**
641 * blkcg_drain_queue - drain blkcg part of request_queue
642 * @q: request_queue to drain
643 *
644 * Called from blk_drain_queue(). Responsible for draining blkcg part.
645 */
646void blkcg_drain_queue(struct request_queue *q)
647{
648 lockdep_assert_held(q->queue_lock);
649
650 blk_throtl_drain(q);
651}
652
653/**
654 * blkcg_exit_queue - exit and release blkcg part of request_queue
655 * @q: request_queue being released
656 *
657 * Called from blk_release_queue(). Responsible for exiting blkcg part.
658 */
659void blkcg_exit_queue(struct request_queue *q)
660{
Tejun Heo923adde2012-03-05 13:15:13 -0800661 mutex_lock(&all_q_mutex);
662 list_del_init(&q->all_q_node);
663 mutex_unlock(&all_q_mutex);
664
Tejun Heoe8989fa2012-03-05 13:15:20 -0800665 blkg_destroy_all(q, true);
666
Tejun Heo5efd6112012-03-05 13:15:12 -0800667 blk_throtl_exit(q);
668}
669
Vivek Goyal31e4c282009-12-03 12:59:42 -0500670/*
671 * We cannot support shared io contexts, as we have no mean to support
672 * two tasks with the same ioc in two different groups without major rework
673 * of the main cic data structures. For now we allow a task to change
674 * its cgroup only if it's the only owner of its ioc.
675 */
Li Zefan761b3ef2012-01-31 13:47:36 +0800676static int blkiocg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500677{
Tejun Heobb9d97b2011-12-12 18:12:21 -0800678 struct task_struct *task;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500679 struct io_context *ioc;
680 int ret = 0;
681
682 /* task_lock() is needed to avoid races with exit_io_context() */
Tejun Heobb9d97b2011-12-12 18:12:21 -0800683 cgroup_taskset_for_each(task, cgrp, tset) {
684 task_lock(task);
685 ioc = task->io_context;
686 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
687 ret = -EINVAL;
688 task_unlock(task);
689 if (ret)
690 break;
691 }
Vivek Goyal31e4c282009-12-03 12:59:42 -0500692 return ret;
693}
694
Tejun Heo923adde2012-03-05 13:15:13 -0800695static void blkcg_bypass_start(void)
696 __acquires(&all_q_mutex)
697{
698 struct request_queue *q;
699
700 mutex_lock(&all_q_mutex);
701
702 list_for_each_entry(q, &all_q_list, all_q_node) {
703 blk_queue_bypass_start(q);
Tejun Heoe8989fa2012-03-05 13:15:20 -0800704 blkg_destroy_all(q, false);
Tejun Heo923adde2012-03-05 13:15:13 -0800705 }
706}
707
708static void blkcg_bypass_end(void)
709 __releases(&all_q_mutex)
710{
711 struct request_queue *q;
712
713 list_for_each_entry(q, &all_q_list, all_q_node)
714 blk_queue_bypass_end(q);
715
716 mutex_unlock(&all_q_mutex);
717}
718
Tejun Heo676f7c82012-04-01 12:09:55 -0700719struct cgroup_subsys blkio_subsys = {
720 .name = "blkio",
721 .create = blkiocg_create,
722 .can_attach = blkiocg_can_attach,
Tejun Heo959d8512012-04-01 12:30:01 -0700723 .pre_destroy = blkiocg_pre_destroy,
Tejun Heo676f7c82012-04-01 12:09:55 -0700724 .destroy = blkiocg_destroy,
Tejun Heo676f7c82012-04-01 12:09:55 -0700725 .subsys_id = blkio_subsys_id,
Tejun Heo4baf6e32012-04-01 12:09:55 -0700726 .base_cftypes = blkio_files,
Tejun Heo676f7c82012-04-01 12:09:55 -0700727 .module = THIS_MODULE,
728};
729EXPORT_SYMBOL_GPL(blkio_subsys);
730
Vivek Goyal3e252062009-12-04 10:36:42 -0500731void blkio_policy_register(struct blkio_policy_type *blkiop)
732{
Tejun Heoe8989fa2012-03-05 13:15:20 -0800733 struct request_queue *q;
734
Tejun Heobc0d6502012-04-13 13:11:26 -0700735 mutex_lock(&blkcg_pol_mutex);
736
Tejun Heo923adde2012-03-05 13:15:13 -0800737 blkcg_bypass_start();
Tejun Heo035d10b2012-03-05 13:15:04 -0800738
739 BUG_ON(blkio_policy[blkiop->plid]);
740 blkio_policy[blkiop->plid] = blkiop;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800741 list_for_each_entry(q, &all_q_list, all_q_node)
742 update_root_blkg_pd(q, blkiop->plid);
Tejun Heobc0d6502012-04-13 13:11:26 -0700743
Tejun Heo923adde2012-03-05 13:15:13 -0800744 blkcg_bypass_end();
Tejun Heo44ea53d2012-04-01 14:38:43 -0700745
746 if (blkiop->cftypes)
747 WARN_ON(cgroup_add_cftypes(&blkio_subsys, blkiop->cftypes));
Tejun Heobc0d6502012-04-13 13:11:26 -0700748
749 mutex_unlock(&blkcg_pol_mutex);
Vivek Goyal3e252062009-12-04 10:36:42 -0500750}
751EXPORT_SYMBOL_GPL(blkio_policy_register);
752
753void blkio_policy_unregister(struct blkio_policy_type *blkiop)
754{
Tejun Heoe8989fa2012-03-05 13:15:20 -0800755 struct request_queue *q;
756
Tejun Heobc0d6502012-04-13 13:11:26 -0700757 mutex_lock(&blkcg_pol_mutex);
758
Tejun Heo44ea53d2012-04-01 14:38:43 -0700759 if (blkiop->cftypes)
760 cgroup_rm_cftypes(&blkio_subsys, blkiop->cftypes);
761
Tejun Heo923adde2012-03-05 13:15:13 -0800762 blkcg_bypass_start();
Tejun Heo035d10b2012-03-05 13:15:04 -0800763
764 BUG_ON(blkio_policy[blkiop->plid] != blkiop);
765 blkio_policy[blkiop->plid] = NULL;
Tejun Heo035d10b2012-03-05 13:15:04 -0800766
Tejun Heoe8989fa2012-03-05 13:15:20 -0800767 list_for_each_entry(q, &all_q_list, all_q_node)
768 update_root_blkg_pd(q, blkiop->plid);
Tejun Heo923adde2012-03-05 13:15:13 -0800769 blkcg_bypass_end();
Tejun Heobc0d6502012-04-13 13:11:26 -0700770
771 mutex_unlock(&blkcg_pol_mutex);
Vivek Goyal3e252062009-12-04 10:36:42 -0500772}
773EXPORT_SYMBOL_GPL(blkio_policy_unregister);