blob: 290792a13e3cc184f0ea67176589050e6323b675 [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 -080028
Tejun Heoe71357e2013-01-09 08:05:10 -080029struct blkcg blkcg_root = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT,
30 .cfq_leaf_weight = 2 * CFQ_WEIGHT_DEFAULT, };
Tejun Heo3c798392012-04-16 13:57:25 -070031EXPORT_SYMBOL_GPL(blkcg_root);
Vivek Goyal9d6a9862009-12-04 10:36:41 -050032
Tejun Heo3c798392012-04-16 13:57:25 -070033static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
Tejun Heo035d10b2012-03-05 13:15:04 -080034
Tejun Heoa2b16932012-04-13 13:11:33 -070035static bool blkcg_policy_enabled(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -070036 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -070037{
38 return pol && test_bit(pol->plid, q->blkcg_pols);
39}
40
Tejun Heo03814112012-03-05 13:15:14 -080041/**
42 * blkg_free - free a blkg
43 * @blkg: blkg to free
44 *
45 * Free @blkg which may be partially allocated.
46 */
Tejun Heo3c798392012-04-16 13:57:25 -070047static void blkg_free(struct blkcg_gq *blkg)
Tejun Heo03814112012-03-05 13:15:14 -080048{
Tejun Heoe8989fa2012-03-05 13:15:20 -080049 int i;
Tejun Heo549d3aa2012-03-05 13:15:16 -080050
51 if (!blkg)
52 return;
53
Tejun Heodb613672013-05-14 13:52:31 -070054 for (i = 0; i < BLKCG_MAX_POLS; i++)
55 kfree(blkg->pd[i]);
Tejun Heoe8989fa2012-03-05 13:15:20 -080056
Tejun Heoa0516612012-06-26 15:05:44 -070057 blk_exit_rl(&blkg->rl);
Tejun Heo549d3aa2012-03-05 13:15:16 -080058 kfree(blkg);
Tejun Heo03814112012-03-05 13:15:14 -080059}
60
61/**
62 * blkg_alloc - allocate a blkg
63 * @blkcg: block cgroup the new blkg is associated with
64 * @q: request_queue the new blkg is associated with
Tejun Heo15974992012-06-04 20:40:52 -070065 * @gfp_mask: allocation mask to use
Tejun Heo03814112012-03-05 13:15:14 -080066 *
Tejun Heoe8989fa2012-03-05 13:15:20 -080067 * Allocate a new blkg assocating @blkcg and @q.
Tejun Heo03814112012-03-05 13:15:14 -080068 */
Tejun Heo15974992012-06-04 20:40:52 -070069static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
70 gfp_t gfp_mask)
Tejun Heo03814112012-03-05 13:15:14 -080071{
Tejun Heo3c798392012-04-16 13:57:25 -070072 struct blkcg_gq *blkg;
Tejun Heoe8989fa2012-03-05 13:15:20 -080073 int i;
Tejun Heo03814112012-03-05 13:15:14 -080074
75 /* alloc and init base part */
Tejun Heo15974992012-06-04 20:40:52 -070076 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
Tejun Heo03814112012-03-05 13:15:14 -080077 if (!blkg)
78 return NULL;
79
Tejun Heoc875f4d2012-03-05 13:15:22 -080080 blkg->q = q;
Tejun Heoe8989fa2012-03-05 13:15:20 -080081 INIT_LIST_HEAD(&blkg->q_node);
Tejun Heo03814112012-03-05 13:15:14 -080082 blkg->blkcg = blkcg;
Tejun Heo1adaf3d2012-03-05 13:15:15 -080083 blkg->refcnt = 1;
Tejun Heo03814112012-03-05 13:15:14 -080084
Tejun Heoa0516612012-06-26 15:05:44 -070085 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
86 if (blkcg != &blkcg_root) {
87 if (blk_init_rl(&blkg->rl, q, gfp_mask))
88 goto err_free;
89 blkg->rl.blkg = blkg;
90 }
91
Tejun Heo8bd435b2012-04-13 13:11:28 -070092 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -070093 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heoe8989fa2012-03-05 13:15:20 -080094 struct blkg_policy_data *pd;
Tejun Heo03814112012-03-05 13:15:14 -080095
Tejun Heoa2b16932012-04-13 13:11:33 -070096 if (!blkcg_policy_enabled(q, pol))
Tejun Heoe8989fa2012-03-05 13:15:20 -080097 continue;
Tejun Heo549d3aa2012-03-05 13:15:16 -080098
Tejun Heoe8989fa2012-03-05 13:15:20 -080099 /* alloc per-policy data and attach it to blkg */
Tejun Heo15974992012-06-04 20:40:52 -0700100 pd = kzalloc_node(pol->pd_size, gfp_mask, q->node);
Tejun Heoa0516612012-06-26 15:05:44 -0700101 if (!pd)
102 goto err_free;
Tejun Heo549d3aa2012-03-05 13:15:16 -0800103
Tejun Heoe8989fa2012-03-05 13:15:20 -0800104 blkg->pd[i] = pd;
105 pd->blkg = blkg;
Tejun Heob276a872013-01-09 08:05:12 -0800106 pd->plid = i;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800107 }
108
Tejun Heo03814112012-03-05 13:15:14 -0800109 return blkg;
Tejun Heoa0516612012-06-26 15:05:44 -0700110
111err_free:
112 blkg_free(blkg);
113 return NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800114}
115
Tejun Heo16b3de62013-01-09 08:05:12 -0800116/**
117 * __blkg_lookup - internal version of blkg_lookup()
118 * @blkcg: blkcg of interest
119 * @q: request_queue of interest
120 * @update_hint: whether to update lookup hint with the result or not
121 *
122 * This is internal version and shouldn't be used by policy
123 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
124 * @q's bypass state. If @update_hint is %true, the caller should be
125 * holding @q->queue_lock and lookup hint is updated on success.
126 */
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700127struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
128 bool update_hint)
Tejun Heo80fd9972012-04-13 14:50:53 -0700129{
Tejun Heo3c798392012-04-16 13:57:25 -0700130 struct blkcg_gq *blkg;
Tejun Heo80fd9972012-04-13 14:50:53 -0700131
Tejun Heoa6371202012-04-19 16:29:24 -0700132 blkg = rcu_dereference(blkcg->blkg_hint);
133 if (blkg && blkg->q == q)
134 return blkg;
135
136 /*
Tejun Heo86cde6b2013-01-09 08:05:10 -0800137 * Hint didn't match. Look up from the radix tree. Note that the
138 * hint can only be updated under queue_lock as otherwise @blkg
139 * could have already been removed from blkg_tree. The caller is
140 * responsible for grabbing queue_lock if @update_hint.
Tejun Heoa6371202012-04-19 16:29:24 -0700141 */
142 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800143 if (blkg && blkg->q == q) {
144 if (update_hint) {
145 lockdep_assert_held(q->queue_lock);
146 rcu_assign_pointer(blkcg->blkg_hint, blkg);
147 }
Tejun Heoa6371202012-04-19 16:29:24 -0700148 return blkg;
Tejun Heo86cde6b2013-01-09 08:05:10 -0800149 }
Tejun Heoa6371202012-04-19 16:29:24 -0700150
Tejun Heo80fd9972012-04-13 14:50:53 -0700151 return NULL;
152}
153
154/**
155 * blkg_lookup - lookup blkg for the specified blkcg - q pair
156 * @blkcg: blkcg of interest
157 * @q: request_queue of interest
158 *
159 * Lookup blkg for the @blkcg - @q pair. This function should be called
160 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
161 * - see blk_queue_bypass_start() for details.
162 */
Tejun Heo3c798392012-04-16 13:57:25 -0700163struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
Tejun Heo80fd9972012-04-13 14:50:53 -0700164{
165 WARN_ON_ONCE(!rcu_read_lock_held());
166
167 if (unlikely(blk_queue_bypass(q)))
168 return NULL;
Tejun Heo86cde6b2013-01-09 08:05:10 -0800169 return __blkg_lookup(blkcg, q, false);
Tejun Heo80fd9972012-04-13 14:50:53 -0700170}
171EXPORT_SYMBOL_GPL(blkg_lookup);
172
Tejun Heo15974992012-06-04 20:40:52 -0700173/*
174 * If @new_blkg is %NULL, this function tries to allocate a new one as
175 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
176 */
Tejun Heo86cde6b2013-01-09 08:05:10 -0800177static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
178 struct request_queue *q,
179 struct blkcg_gq *new_blkg)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400180{
Tejun Heo3c798392012-04-16 13:57:25 -0700181 struct blkcg_gq *blkg;
Tejun Heof427d902013-01-09 08:05:12 -0800182 int i, ret;
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400183
Tejun Heocd1604f2012-03-05 13:15:06 -0800184 WARN_ON_ONCE(!rcu_read_lock_held());
185 lockdep_assert_held(q->queue_lock);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500186
Tejun Heo7ee9c562012-03-05 13:15:11 -0800187 /* blkg holds a reference to blkcg */
Tejun Heo15974992012-06-04 20:40:52 -0700188 if (!css_tryget(&blkcg->css)) {
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800189 ret = -EINVAL;
190 goto err_free_blkg;
Tejun Heo15974992012-06-04 20:40:52 -0700191 }
Tejun Heocd1604f2012-03-05 13:15:06 -0800192
Tejun Heo496fb782012-04-19 16:29:23 -0700193 /* allocate */
Tejun Heo15974992012-06-04 20:40:52 -0700194 if (!new_blkg) {
195 new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
196 if (unlikely(!new_blkg)) {
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800197 ret = -ENOMEM;
198 goto err_put_css;
Tejun Heo15974992012-06-04 20:40:52 -0700199 }
200 }
201 blkg = new_blkg;
Tejun Heocd1604f2012-03-05 13:15:06 -0800202
Tejun Heodb613672013-05-14 13:52:31 -0700203 /* link parent */
Tejun Heo3c547862013-01-09 08:05:10 -0800204 if (blkcg_parent(blkcg)) {
205 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
206 if (WARN_ON_ONCE(!blkg->parent)) {
Tejun Heo2423c9c2013-05-14 13:52:30 -0700207 ret = -EINVAL;
Tejun Heo3c547862013-01-09 08:05:10 -0800208 goto err_put_css;
209 }
210 blkg_get(blkg->parent);
211 }
212
Tejun Heodb613672013-05-14 13:52:31 -0700213 /* invoke per-policy init */
214 for (i = 0; i < BLKCG_MAX_POLS; i++) {
215 struct blkcg_policy *pol = blkcg_policy[i];
216
217 if (blkg->pd[i] && pol->pd_init_fn)
218 pol->pd_init_fn(blkg);
219 }
220
221 /* insert */
Tejun Heoa6371202012-04-19 16:29:24 -0700222 spin_lock(&blkcg->lock);
223 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
224 if (likely(!ret)) {
225 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
226 list_add(&blkg->q_node, &q->blkg_list);
Tejun Heof427d902013-01-09 08:05:12 -0800227
228 for (i = 0; i < BLKCG_MAX_POLS; i++) {
229 struct blkcg_policy *pol = blkcg_policy[i];
230
231 if (blkg->pd[i] && pol->pd_online_fn)
232 pol->pd_online_fn(blkg);
233 }
Tejun Heoa6371202012-04-19 16:29:24 -0700234 }
Tejun Heof427d902013-01-09 08:05:12 -0800235 blkg->online = true;
Tejun Heoa6371202012-04-19 16:29:24 -0700236 spin_unlock(&blkcg->lock);
237
Tejun Heoa6371202012-04-19 16:29:24 -0700238 if (!ret)
239 return blkg;
Tejun Heo15974992012-06-04 20:40:52 -0700240
Tejun Heo3c547862013-01-09 08:05:10 -0800241 /* @blkg failed fully initialized, use the usual release path */
242 blkg_put(blkg);
243 return ERR_PTR(ret);
244
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800245err_put_css:
Tejun Heo496fb782012-04-19 16:29:23 -0700246 css_put(&blkcg->css);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800247err_free_blkg:
Tejun Heo15974992012-06-04 20:40:52 -0700248 blkg_free(new_blkg);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800249 return ERR_PTR(ret);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500250}
Tejun Heo3c96cb32012-04-13 13:11:34 -0700251
Tejun Heo86cde6b2013-01-09 08:05:10 -0800252/**
253 * blkg_lookup_create - lookup blkg, try to create one if not there
254 * @blkcg: blkcg of interest
255 * @q: request_queue of interest
256 *
257 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
Tejun Heo3c547862013-01-09 08:05:10 -0800258 * create one. blkg creation is performed recursively from blkcg_root such
259 * that all non-root blkg's have access to the parent blkg. This function
260 * should be called under RCU read lock and @q->queue_lock.
Tejun Heo86cde6b2013-01-09 08:05:10 -0800261 *
262 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
263 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
264 * dead and bypassing, returns ERR_PTR(-EBUSY).
265 */
Tejun Heo3c798392012-04-16 13:57:25 -0700266struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
267 struct request_queue *q)
Tejun Heo3c96cb32012-04-13 13:11:34 -0700268{
Tejun Heo86cde6b2013-01-09 08:05:10 -0800269 struct blkcg_gq *blkg;
270
271 WARN_ON_ONCE(!rcu_read_lock_held());
272 lockdep_assert_held(q->queue_lock);
273
Tejun Heo3c96cb32012-04-13 13:11:34 -0700274 /*
275 * This could be the first entry point of blkcg implementation and
276 * we shouldn't allow anything to go through for a bypassing queue.
277 */
278 if (unlikely(blk_queue_bypass(q)))
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100279 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800280
281 blkg = __blkg_lookup(blkcg, q, true);
282 if (blkg)
283 return blkg;
284
Tejun Heo3c547862013-01-09 08:05:10 -0800285 /*
286 * Create blkgs walking down from blkcg_root to @blkcg, so that all
287 * non-root blkgs have access to their parents.
288 */
289 while (true) {
290 struct blkcg *pos = blkcg;
291 struct blkcg *parent = blkcg_parent(blkcg);
292
293 while (parent && !__blkg_lookup(parent, q, false)) {
294 pos = parent;
295 parent = blkcg_parent(parent);
296 }
297
298 blkg = blkg_create(pos, q, NULL);
299 if (pos == blkcg || IS_ERR(blkg))
300 return blkg;
301 }
Tejun Heo3c96cb32012-04-13 13:11:34 -0700302}
Tejun Heocd1604f2012-03-05 13:15:06 -0800303EXPORT_SYMBOL_GPL(blkg_lookup_create);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500304
Tejun Heo3c798392012-04-16 13:57:25 -0700305static void blkg_destroy(struct blkcg_gq *blkg)
Tejun Heo72e06c22012-03-05 13:15:00 -0800306{
Tejun Heo3c798392012-04-16 13:57:25 -0700307 struct blkcg *blkcg = blkg->blkcg;
Tejun Heof427d902013-01-09 08:05:12 -0800308 int i;
Tejun Heo03aa2642012-03-05 13:15:19 -0800309
Tejun Heo27e1f9d2012-06-05 13:36:44 +0200310 lockdep_assert_held(blkg->q->queue_lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800311 lockdep_assert_held(&blkcg->lock);
Tejun Heo03aa2642012-03-05 13:15:19 -0800312
313 /* Something wrong if we are trying to remove same group twice */
Tejun Heoe8989fa2012-03-05 13:15:20 -0800314 WARN_ON_ONCE(list_empty(&blkg->q_node));
Tejun Heo9f13ef62012-03-05 13:15:21 -0800315 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
Tejun Heoa6371202012-04-19 16:29:24 -0700316
Tejun Heof427d902013-01-09 08:05:12 -0800317 for (i = 0; i < BLKCG_MAX_POLS; i++) {
318 struct blkcg_policy *pol = blkcg_policy[i];
319
320 if (blkg->pd[i] && pol->pd_offline_fn)
321 pol->pd_offline_fn(blkg);
322 }
323 blkg->online = false;
324
Tejun Heoa6371202012-04-19 16:29:24 -0700325 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
Tejun Heoe8989fa2012-03-05 13:15:20 -0800326 list_del_init(&blkg->q_node);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800327 hlist_del_init_rcu(&blkg->blkcg_node);
Tejun Heo03aa2642012-03-05 13:15:19 -0800328
Tejun Heo03aa2642012-03-05 13:15:19 -0800329 /*
Tejun Heoa6371202012-04-19 16:29:24 -0700330 * Both setting lookup hint to and clearing it from @blkg are done
331 * under queue_lock. If it's not pointing to @blkg now, it never
332 * will. Hint assignment itself can race safely.
333 */
334 if (rcu_dereference_raw(blkcg->blkg_hint) == blkg)
335 rcu_assign_pointer(blkcg->blkg_hint, NULL);
336
337 /*
Tejun Heo03aa2642012-03-05 13:15:19 -0800338 * Put the reference taken at the time of creation so that when all
339 * queues are gone, group can be destroyed.
340 */
341 blkg_put(blkg);
342}
343
Tejun Heo9f13ef62012-03-05 13:15:21 -0800344/**
345 * blkg_destroy_all - destroy all blkgs associated with a request_queue
346 * @q: request_queue of interest
Tejun Heo9f13ef62012-03-05 13:15:21 -0800347 *
Tejun Heo3c96cb32012-04-13 13:11:34 -0700348 * Destroy all blkgs associated with @q.
Tejun Heo9f13ef62012-03-05 13:15:21 -0800349 */
Tejun Heo3c96cb32012-04-13 13:11:34 -0700350static void blkg_destroy_all(struct request_queue *q)
Tejun Heo03aa2642012-03-05 13:15:19 -0800351{
Tejun Heo3c798392012-04-16 13:57:25 -0700352 struct blkcg_gq *blkg, *n;
Tejun Heo72e06c22012-03-05 13:15:00 -0800353
Tejun Heo6d18b002012-04-13 13:11:35 -0700354 lockdep_assert_held(q->queue_lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800355
Tejun Heo9f13ef62012-03-05 13:15:21 -0800356 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
Tejun Heo3c798392012-04-16 13:57:25 -0700357 struct blkcg *blkcg = blkg->blkcg;
Tejun Heo72e06c22012-03-05 13:15:00 -0800358
Tejun Heo9f13ef62012-03-05 13:15:21 -0800359 spin_lock(&blkcg->lock);
360 blkg_destroy(blkg);
361 spin_unlock(&blkcg->lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800362 }
Jun'ichi Nomura65635cb2012-10-17 17:45:36 +0900363
364 /*
365 * root blkg is destroyed. Just clear the pointer since
366 * root_rl does not take reference on root blkg.
367 */
368 q->root_blkg = NULL;
369 q->root_rl.blkg = NULL;
Tejun Heo72e06c22012-03-05 13:15:00 -0800370}
371
Tejun Heo2a4fd072013-05-14 13:52:31 -0700372/*
373 * A group is RCU protected, but having an rcu lock does not mean that one
374 * can access all the fields of blkg and assume these are valid. For
375 * example, don't try to follow throtl_data and request queue links.
376 *
377 * Having a reference to blkg under an rcu allows accesses to only values
378 * local to groups like group stats and group rate limits.
379 */
380void __blkg_release_rcu(struct rcu_head *rcu_head)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800381{
Tejun Heo2a4fd072013-05-14 13:52:31 -0700382 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
Tejun Heodb613672013-05-14 13:52:31 -0700383 int i;
384
385 /* tell policies that this one is being freed */
386 for (i = 0; i < BLKCG_MAX_POLS; i++) {
387 struct blkcg_policy *pol = blkcg_policy[i];
388
389 if (blkg->pd[i] && pol->pd_exit_fn)
390 pol->pd_exit_fn(blkg);
391 }
392
Tejun Heo3c547862013-01-09 08:05:10 -0800393 /* release the blkcg and parent blkg refs this blkg has been holding */
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800394 css_put(&blkg->blkcg->css);
Tejun Heo2a4fd072013-05-14 13:52:31 -0700395 if (blkg->parent) {
396 spin_lock_irq(blkg->q->queue_lock);
Tejun Heo3c547862013-01-09 08:05:10 -0800397 blkg_put(blkg->parent);
Tejun Heo2a4fd072013-05-14 13:52:31 -0700398 spin_unlock_irq(blkg->q->queue_lock);
399 }
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800400
Tejun Heo2a4fd072013-05-14 13:52:31 -0700401 blkg_free(blkg);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800402}
Tejun Heo2a4fd072013-05-14 13:52:31 -0700403EXPORT_SYMBOL_GPL(__blkg_release_rcu);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800404
Tejun Heoa0516612012-06-26 15:05:44 -0700405/*
406 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
407 * because the root blkg uses @q->root_rl instead of its own rl.
408 */
409struct request_list *__blk_queue_next_rl(struct request_list *rl,
410 struct request_queue *q)
411{
412 struct list_head *ent;
413 struct blkcg_gq *blkg;
414
415 /*
416 * Determine the current blkg list_head. The first entry is
417 * root_rl which is off @q->blkg_list and mapped to the head.
418 */
419 if (rl == &q->root_rl) {
420 ent = &q->blkg_list;
Jun'ichi Nomura65c77fd2012-10-22 10:15:37 +0900421 /* There are no more block groups, hence no request lists */
422 if (list_empty(ent))
423 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -0700424 } else {
425 blkg = container_of(rl, struct blkcg_gq, rl);
426 ent = &blkg->q_node;
427 }
428
429 /* walk to the next list_head, skip root blkcg */
430 ent = ent->next;
431 if (ent == &q->root_blkg->q_node)
432 ent = ent->next;
433 if (ent == &q->blkg_list)
434 return NULL;
435
436 blkg = container_of(ent, struct blkcg_gq, q_node);
437 return &blkg->rl;
438}
439
Tejun Heo3c798392012-04-16 13:57:25 -0700440static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype,
441 u64 val)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700442{
Tejun Heo3c798392012-04-16 13:57:25 -0700443 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
444 struct blkcg_gq *blkg;
Tejun Heobc0d6502012-04-13 13:11:26 -0700445 int i;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700446
Tejun Heobc0d6502012-04-13 13:11:26 -0700447 mutex_lock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700448 spin_lock_irq(&blkcg->lock);
Tejun Heo997a0262012-03-08 10:53:58 -0800449
450 /*
451 * Note that stat reset is racy - it doesn't synchronize against
452 * stat updates. This is a debug feature which shouldn't exist
453 * anyway. If you get hit by a race, retry.
454 */
Sasha Levinb67bfe02013-02-27 17:06:00 -0800455 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo8bd435b2012-04-13 13:11:28 -0700456 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -0700457 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heo549d3aa2012-03-05 13:15:16 -0800458
Tejun Heoa2b16932012-04-13 13:11:33 -0700459 if (blkcg_policy_enabled(blkg->q, pol) &&
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700460 pol->pd_reset_stats_fn)
461 pol->pd_reset_stats_fn(blkg);
Tejun Heobc0d6502012-04-13 13:11:26 -0700462 }
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700463 }
Vivek Goyalf0bdc8c2011-05-19 15:38:30 -0400464
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700465 spin_unlock_irq(&blkcg->lock);
Tejun Heobc0d6502012-04-13 13:11:26 -0700466 mutex_unlock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700467 return 0;
468}
469
Tejun Heo3c798392012-04-16 13:57:25 -0700470static const char *blkg_dev_name(struct blkcg_gq *blkg)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700471{
Tejun Heod3d32e62012-04-01 14:38:42 -0700472 /* some drivers (floppy) instantiate a queue w/o disk registered */
473 if (blkg->q->backing_dev_info.dev)
474 return dev_name(blkg->q->backing_dev_info.dev);
475 return NULL;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700476}
477
Tejun Heod3d32e62012-04-01 14:38:42 -0700478/**
479 * blkcg_print_blkgs - helper for printing per-blkg data
480 * @sf: seq_file to print to
481 * @blkcg: blkcg of interest
482 * @prfill: fill function to print out a blkg
483 * @pol: policy in question
484 * @data: data to be passed to @prfill
485 * @show_total: to print out sum of prfill return values or not
486 *
487 * This function invokes @prfill on each blkg of @blkcg if pd for the
488 * policy specified by @pol exists. @prfill is invoked with @sf, the
Tejun Heo810ecfa2013-01-09 08:05:13 -0800489 * policy data and @data and the matching queue lock held. If @show_total
490 * is %true, the sum of the return values from @prfill is printed with
491 * "Total" label at the end.
Tejun Heod3d32e62012-04-01 14:38:42 -0700492 *
493 * This is to be used to construct print functions for
494 * cftype->read_seq_string method.
495 */
Tejun Heo3c798392012-04-16 13:57:25 -0700496void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
Tejun Heof95a04a2012-04-16 13:57:26 -0700497 u64 (*prfill)(struct seq_file *,
498 struct blkg_policy_data *, int),
Tejun Heo3c798392012-04-16 13:57:25 -0700499 const struct blkcg_policy *pol, int data,
Tejun Heoec399342012-04-13 13:11:27 -0700500 bool show_total)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400501{
Tejun Heo3c798392012-04-16 13:57:25 -0700502 struct blkcg_gq *blkg;
Tejun Heod3d32e62012-04-01 14:38:42 -0700503 u64 total = 0;
504
Tejun Heo810ecfa2013-01-09 08:05:13 -0800505 rcu_read_lock();
Linus Torvaldsee89f812013-02-28 12:52:24 -0800506 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo810ecfa2013-01-09 08:05:13 -0800507 spin_lock_irq(blkg->q->queue_lock);
Tejun Heoa2b16932012-04-13 13:11:33 -0700508 if (blkcg_policy_enabled(blkg->q, pol))
Tejun Heof95a04a2012-04-16 13:57:26 -0700509 total += prfill(sf, blkg->pd[pol->plid], data);
Tejun Heo810ecfa2013-01-09 08:05:13 -0800510 spin_unlock_irq(blkg->q->queue_lock);
511 }
512 rcu_read_unlock();
Tejun Heod3d32e62012-04-01 14:38:42 -0700513
514 if (show_total)
515 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
516}
Tejun Heo829fdb52012-04-01 14:38:43 -0700517EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
Tejun Heod3d32e62012-04-01 14:38:42 -0700518
519/**
520 * __blkg_prfill_u64 - prfill helper for a single u64 value
521 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700522 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700523 * @v: value to print
524 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700525 * Print @v to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700526 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700527u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
Tejun Heod3d32e62012-04-01 14:38:42 -0700528{
Tejun Heof95a04a2012-04-16 13:57:26 -0700529 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700530
531 if (!dname)
532 return 0;
533
534 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
535 return v;
536}
Tejun Heo829fdb52012-04-01 14:38:43 -0700537EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
Tejun Heod3d32e62012-04-01 14:38:42 -0700538
539/**
540 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
541 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700542 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700543 * @rwstat: rwstat to print
544 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700545 * Print @rwstat to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700546 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700547u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
Tejun Heo829fdb52012-04-01 14:38:43 -0700548 const struct blkg_rwstat *rwstat)
Tejun Heod3d32e62012-04-01 14:38:42 -0700549{
550 static const char *rwstr[] = {
551 [BLKG_RWSTAT_READ] = "Read",
552 [BLKG_RWSTAT_WRITE] = "Write",
553 [BLKG_RWSTAT_SYNC] = "Sync",
554 [BLKG_RWSTAT_ASYNC] = "Async",
555 };
Tejun Heof95a04a2012-04-16 13:57:26 -0700556 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700557 u64 v;
558 int i;
559
560 if (!dname)
561 return 0;
562
563 for (i = 0; i < BLKG_RWSTAT_NR; i++)
564 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
565 (unsigned long long)rwstat->cnt[i]);
566
567 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
568 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
569 return v;
570}
Tejun Heob50da392013-01-09 08:05:12 -0800571EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700572
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700573/**
574 * blkg_prfill_stat - prfill callback for blkg_stat
575 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700576 * @pd: policy private data of interest
577 * @off: offset to the blkg_stat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700578 *
579 * prfill callback for printing a blkg_stat.
580 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700581u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700582{
Tejun Heof95a04a2012-04-16 13:57:26 -0700583 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
Tejun Heod3d32e62012-04-01 14:38:42 -0700584}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700585EXPORT_SYMBOL_GPL(blkg_prfill_stat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700586
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700587/**
588 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
589 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700590 * @pd: policy private data of interest
591 * @off: offset to the blkg_rwstat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700592 *
593 * prfill callback for printing a blkg_rwstat.
594 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700595u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
596 int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700597{
Tejun Heof95a04a2012-04-16 13:57:26 -0700598 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
Tejun Heod3d32e62012-04-01 14:38:42 -0700599
Tejun Heof95a04a2012-04-16 13:57:26 -0700600 return __blkg_prfill_rwstat(sf, pd, &rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700601}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700602EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700603
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700604/**
Tejun Heo16b3de62013-01-09 08:05:12 -0800605 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
606 * @pd: policy private data of interest
607 * @off: offset to the blkg_stat in @pd
608 *
609 * Collect the blkg_stat specified by @off from @pd and all its online
610 * descendants and return the sum. The caller must be holding the queue
611 * lock for online tests.
612 */
613u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off)
614{
615 struct blkcg_policy *pol = blkcg_policy[pd->plid];
616 struct blkcg_gq *pos_blkg;
617 struct cgroup *pos_cgrp;
618 u64 sum;
619
620 lockdep_assert_held(pd->blkg->q->queue_lock);
621
622 sum = blkg_stat_read((void *)pd + off);
623
624 rcu_read_lock();
625 blkg_for_each_descendant_pre(pos_blkg, pos_cgrp, pd_to_blkg(pd)) {
626 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
627 struct blkg_stat *stat = (void *)pos_pd + off;
628
629 if (pos_blkg->online)
630 sum += blkg_stat_read(stat);
631 }
632 rcu_read_unlock();
633
634 return sum;
635}
636EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
637
638/**
639 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
640 * @pd: policy private data of interest
641 * @off: offset to the blkg_stat in @pd
642 *
643 * Collect the blkg_rwstat specified by @off from @pd and all its online
644 * descendants and return the sum. The caller must be holding the queue
645 * lock for online tests.
646 */
647struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
648 int off)
649{
650 struct blkcg_policy *pol = blkcg_policy[pd->plid];
651 struct blkcg_gq *pos_blkg;
652 struct cgroup *pos_cgrp;
653 struct blkg_rwstat sum;
654 int i;
655
656 lockdep_assert_held(pd->blkg->q->queue_lock);
657
658 sum = blkg_rwstat_read((void *)pd + off);
659
660 rcu_read_lock();
661 blkg_for_each_descendant_pre(pos_blkg, pos_cgrp, pd_to_blkg(pd)) {
662 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
663 struct blkg_rwstat *rwstat = (void *)pos_pd + off;
664 struct blkg_rwstat tmp;
665
666 if (!pos_blkg->online)
667 continue;
668
669 tmp = blkg_rwstat_read(rwstat);
670
671 for (i = 0; i < BLKG_RWSTAT_NR; i++)
672 sum.cnt[i] += tmp.cnt[i];
673 }
674 rcu_read_unlock();
675
676 return sum;
677}
678EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
679
680/**
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700681 * blkg_conf_prep - parse and prepare for per-blkg config update
682 * @blkcg: target block cgroup
Tejun Heoda8b0662012-04-13 13:11:29 -0700683 * @pol: target policy
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700684 * @input: input string
685 * @ctx: blkg_conf_ctx to be filled
686 *
687 * Parse per-blkg config update from @input and initialize @ctx with the
688 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
Tejun Heoda8b0662012-04-13 13:11:29 -0700689 * value. This function returns with RCU read lock and queue lock held and
690 * must be paired with blkg_conf_finish().
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700691 */
Tejun Heo3c798392012-04-16 13:57:25 -0700692int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
693 const char *input, struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700694 __acquires(rcu) __acquires(disk->queue->queue_lock)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800695{
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700696 struct gendisk *disk;
Tejun Heo3c798392012-04-16 13:57:25 -0700697 struct blkcg_gq *blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700698 unsigned int major, minor;
699 unsigned long long v;
700 int part, ret;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800701
Tejun Heo726fa692012-04-01 14:38:43 -0700702 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
703 return -EINVAL;
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700704
Tejun Heo726fa692012-04-01 14:38:43 -0700705 disk = get_gendisk(MKDEV(major, minor), &part);
Tejun Heo4bfd4822012-03-05 13:15:08 -0800706 if (!disk || part)
Tejun Heo726fa692012-04-01 14:38:43 -0700707 return -EINVAL;
Tejun Heoe56da7e2012-03-05 13:15:07 -0800708
709 rcu_read_lock();
Tejun Heo4bfd4822012-03-05 13:15:08 -0800710 spin_lock_irq(disk->queue->queue_lock);
Tejun Heoda8b0662012-04-13 13:11:29 -0700711
Tejun Heoa2b16932012-04-13 13:11:33 -0700712 if (blkcg_policy_enabled(disk->queue, pol))
Tejun Heo3c96cb32012-04-13 13:11:34 -0700713 blkg = blkg_lookup_create(blkcg, disk->queue);
Tejun Heoa2b16932012-04-13 13:11:33 -0700714 else
715 blkg = ERR_PTR(-EINVAL);
Tejun Heoe56da7e2012-03-05 13:15:07 -0800716
Tejun Heo4bfd4822012-03-05 13:15:08 -0800717 if (IS_ERR(blkg)) {
718 ret = PTR_ERR(blkg);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700719 rcu_read_unlock();
Tejun Heoda8b0662012-04-13 13:11:29 -0700720 spin_unlock_irq(disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700721 put_disk(disk);
722 /*
723 * If queue was bypassing, we should retry. Do so after a
724 * short msleep(). It isn't strictly necessary but queue
725 * can be bypassing for some time and it's always nice to
726 * avoid busy looping.
727 */
728 if (ret == -EBUSY) {
729 msleep(10);
730 ret = restart_syscall();
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400731 }
Tejun Heo726fa692012-04-01 14:38:43 -0700732 return ret;
Vivek Goyal062a6442010-09-15 17:06:33 -0400733 }
Tejun Heoe56da7e2012-03-05 13:15:07 -0800734
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700735 ctx->disk = disk;
736 ctx->blkg = blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700737 ctx->v = v;
738 return 0;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800739}
Tejun Heo829fdb52012-04-01 14:38:43 -0700740EXPORT_SYMBOL_GPL(blkg_conf_prep);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800741
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700742/**
743 * blkg_conf_finish - finish up per-blkg config update
744 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
745 *
746 * Finish up after per-blkg config update. This function must be paired
747 * with blkg_conf_prep().
748 */
Tejun Heo829fdb52012-04-01 14:38:43 -0700749void blkg_conf_finish(struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700750 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800751{
Tejun Heoda8b0662012-04-13 13:11:29 -0700752 spin_unlock_irq(ctx->disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700753 rcu_read_unlock();
754 put_disk(ctx->disk);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800755}
Tejun Heo829fdb52012-04-01 14:38:43 -0700756EXPORT_SYMBOL_GPL(blkg_conf_finish);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800757
Tejun Heo3c798392012-04-16 13:57:25 -0700758struct cftype blkcg_files[] = {
Vivek Goyal31e4c282009-12-03 12:59:42 -0500759 {
Divyesh Shah84c124d2010-04-09 08:31:19 +0200760 .name = "reset_stats",
Tejun Heo3c798392012-04-16 13:57:25 -0700761 .write_u64 = blkcg_reset_stats,
Vivek Goyal22084192009-12-03 12:59:49 -0500762 },
Tejun Heo4baf6e32012-04-01 12:09:55 -0700763 { } /* terminate */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500764};
765
Tejun Heo9f13ef62012-03-05 13:15:21 -0800766/**
Tejun Heo92fb9742012-11-19 08:13:38 -0800767 * blkcg_css_offline - cgroup css_offline callback
Tejun Heo9f13ef62012-03-05 13:15:21 -0800768 * @cgroup: cgroup of interest
769 *
770 * This function is called when @cgroup is about to go away and responsible
771 * for shooting down all blkgs associated with @cgroup. blkgs should be
772 * removed while holding both q and blkcg locks. As blkcg lock is nested
773 * inside q lock, this function performs reverse double lock dancing.
774 *
775 * This is the blkcg counterpart of ioc_release_fn().
776 */
Tejun Heo92fb9742012-11-19 08:13:38 -0800777static void blkcg_css_offline(struct cgroup *cgroup)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500778{
Tejun Heo3c798392012-04-16 13:57:25 -0700779 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500780
Tejun Heo9f13ef62012-03-05 13:15:21 -0800781 spin_lock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800782
Tejun Heo9f13ef62012-03-05 13:15:21 -0800783 while (!hlist_empty(&blkcg->blkg_list)) {
Tejun Heo3c798392012-04-16 13:57:25 -0700784 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
785 struct blkcg_gq, blkcg_node);
Tejun Heoc875f4d2012-03-05 13:15:22 -0800786 struct request_queue *q = blkg->q;
Vivek Goyalb1c35762009-12-03 12:59:47 -0500787
Tejun Heo9f13ef62012-03-05 13:15:21 -0800788 if (spin_trylock(q->queue_lock)) {
789 blkg_destroy(blkg);
790 spin_unlock(q->queue_lock);
791 } else {
792 spin_unlock_irq(&blkcg->lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800793 cpu_relax();
Dan Carpentera5567932012-03-29 20:57:08 +0200794 spin_lock_irq(&blkcg->lock);
Jens Axboe0f3942a2010-05-03 14:28:55 +0200795 }
Tejun Heo9f13ef62012-03-05 13:15:21 -0800796 }
Jens Axboe0f3942a2010-05-03 14:28:55 +0200797
Tejun Heo9f13ef62012-03-05 13:15:21 -0800798 spin_unlock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800799}
800
Tejun Heo92fb9742012-11-19 08:13:38 -0800801static void blkcg_css_free(struct cgroup *cgroup)
Tejun Heo7ee9c562012-03-05 13:15:11 -0800802{
Tejun Heo3c798392012-04-16 13:57:25 -0700803 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800804
Tejun Heo3c798392012-04-16 13:57:25 -0700805 if (blkcg != &blkcg_root)
Ben Blum67523c42010-03-10 15:22:11 -0800806 kfree(blkcg);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500807}
808
Tejun Heo92fb9742012-11-19 08:13:38 -0800809static struct cgroup_subsys_state *blkcg_css_alloc(struct cgroup *cgroup)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500810{
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700811 static atomic64_t id_seq = ATOMIC64_INIT(0);
Tejun Heo3c798392012-04-16 13:57:25 -0700812 struct blkcg *blkcg;
Li Zefan03415092010-05-07 08:57:00 +0200813 struct cgroup *parent = cgroup->parent;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500814
Li Zefan03415092010-05-07 08:57:00 +0200815 if (!parent) {
Tejun Heo3c798392012-04-16 13:57:25 -0700816 blkcg = &blkcg_root;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500817 goto done;
818 }
819
Vivek Goyal31e4c282009-12-03 12:59:42 -0500820 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
821 if (!blkcg)
822 return ERR_PTR(-ENOMEM);
823
Tejun Heo3381cb82012-04-01 14:38:44 -0700824 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
Tejun Heoe71357e2013-01-09 08:05:10 -0800825 blkcg->cfq_leaf_weight = CFQ_WEIGHT_DEFAULT;
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700826 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500827done:
828 spin_lock_init(&blkcg->lock);
Tejun Heoa6371202012-04-19 16:29:24 -0700829 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500830 INIT_HLIST_HEAD(&blkcg->blkg_list);
831
832 return &blkcg->css;
833}
834
Tejun Heo5efd6112012-03-05 13:15:12 -0800835/**
836 * blkcg_init_queue - initialize blkcg part of request queue
837 * @q: request_queue to initialize
838 *
839 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
840 * part of new request_queue @q.
841 *
842 * RETURNS:
843 * 0 on success, -errno on failure.
844 */
845int blkcg_init_queue(struct request_queue *q)
846{
847 might_sleep();
848
Tejun Heo3c96cb32012-04-13 13:11:34 -0700849 return blk_throtl_init(q);
Tejun Heo5efd6112012-03-05 13:15:12 -0800850}
851
852/**
853 * blkcg_drain_queue - drain blkcg part of request_queue
854 * @q: request_queue to drain
855 *
856 * Called from blk_drain_queue(). Responsible for draining blkcg part.
857 */
858void blkcg_drain_queue(struct request_queue *q)
859{
860 lockdep_assert_held(q->queue_lock);
861
862 blk_throtl_drain(q);
863}
864
865/**
866 * blkcg_exit_queue - exit and release blkcg part of request_queue
867 * @q: request_queue being released
868 *
869 * Called from blk_release_queue(). Responsible for exiting blkcg part.
870 */
871void blkcg_exit_queue(struct request_queue *q)
872{
Tejun Heo6d18b002012-04-13 13:11:35 -0700873 spin_lock_irq(q->queue_lock);
Tejun Heo3c96cb32012-04-13 13:11:34 -0700874 blkg_destroy_all(q);
Tejun Heo6d18b002012-04-13 13:11:35 -0700875 spin_unlock_irq(q->queue_lock);
876
Tejun Heo5efd6112012-03-05 13:15:12 -0800877 blk_throtl_exit(q);
878}
879
Vivek Goyal31e4c282009-12-03 12:59:42 -0500880/*
881 * We cannot support shared io contexts, as we have no mean to support
882 * two tasks with the same ioc in two different groups without major rework
883 * of the main cic data structures. For now we allow a task to change
884 * its cgroup only if it's the only owner of its ioc.
885 */
Tejun Heo3c798392012-04-16 13:57:25 -0700886static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500887{
Tejun Heobb9d97b2011-12-12 18:12:21 -0800888 struct task_struct *task;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500889 struct io_context *ioc;
890 int ret = 0;
891
892 /* task_lock() is needed to avoid races with exit_io_context() */
Tejun Heobb9d97b2011-12-12 18:12:21 -0800893 cgroup_taskset_for_each(task, cgrp, tset) {
894 task_lock(task);
895 ioc = task->io_context;
896 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
897 ret = -EINVAL;
898 task_unlock(task);
899 if (ret)
900 break;
901 }
Vivek Goyal31e4c282009-12-03 12:59:42 -0500902 return ret;
903}
904
Tejun Heo676f7c82012-04-01 12:09:55 -0700905struct cgroup_subsys blkio_subsys = {
906 .name = "blkio",
Tejun Heo92fb9742012-11-19 08:13:38 -0800907 .css_alloc = blkcg_css_alloc,
908 .css_offline = blkcg_css_offline,
909 .css_free = blkcg_css_free,
Tejun Heo3c798392012-04-16 13:57:25 -0700910 .can_attach = blkcg_can_attach,
Tejun Heo676f7c82012-04-01 12:09:55 -0700911 .subsys_id = blkio_subsys_id,
Tejun Heo3c798392012-04-16 13:57:25 -0700912 .base_cftypes = blkcg_files,
Tejun Heo676f7c82012-04-01 12:09:55 -0700913 .module = THIS_MODULE,
914};
915EXPORT_SYMBOL_GPL(blkio_subsys);
916
Tejun Heo8bd435b2012-04-13 13:11:28 -0700917/**
Tejun Heoa2b16932012-04-13 13:11:33 -0700918 * blkcg_activate_policy - activate a blkcg policy on a request_queue
919 * @q: request_queue of interest
920 * @pol: blkcg policy to activate
921 *
922 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
923 * bypass mode to populate its blkgs with policy_data for @pol.
924 *
925 * Activation happens with @q bypassed, so nobody would be accessing blkgs
926 * from IO path. Update of each blkg is protected by both queue and blkcg
927 * locks so that holding either lock and testing blkcg_policy_enabled() is
928 * always enough for dereferencing policy data.
929 *
930 * The caller is responsible for synchronizing [de]activations and policy
931 * [un]registerations. Returns 0 on success, -errno on failure.
932 */
933int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700934 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -0700935{
936 LIST_HEAD(pds);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800937 struct blkcg_gq *blkg, *new_blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -0700938 struct blkg_policy_data *pd, *n;
939 int cnt = 0, ret;
Tejun Heo15974992012-06-04 20:40:52 -0700940 bool preloaded;
Tejun Heoa2b16932012-04-13 13:11:33 -0700941
942 if (blkcg_policy_enabled(q, pol))
943 return 0;
944
Tejun Heo15974992012-06-04 20:40:52 -0700945 /* preallocations for root blkg */
Tejun Heo86cde6b2013-01-09 08:05:10 -0800946 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
947 if (!new_blkg)
Tejun Heo15974992012-06-04 20:40:52 -0700948 return -ENOMEM;
949
Tejun Heoa2b16932012-04-13 13:11:33 -0700950 blk_queue_bypass_start(q);
951
Jun'ichi Nomurae5072662013-04-09 15:01:21 +0200952 preloaded = !radix_tree_preload(GFP_KERNEL);
953
Tejun Heo86cde6b2013-01-09 08:05:10 -0800954 /*
955 * Make sure the root blkg exists and count the existing blkgs. As
956 * @q is bypassing at this point, blkg_lookup_create() can't be
957 * used. Open code it.
958 */
Tejun Heoa2b16932012-04-13 13:11:33 -0700959 spin_lock_irq(q->queue_lock);
960
961 rcu_read_lock();
Tejun Heo86cde6b2013-01-09 08:05:10 -0800962 blkg = __blkg_lookup(&blkcg_root, q, false);
963 if (blkg)
964 blkg_free(new_blkg);
965 else
966 blkg = blkg_create(&blkcg_root, q, new_blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -0700967 rcu_read_unlock();
968
Tejun Heo15974992012-06-04 20:40:52 -0700969 if (preloaded)
970 radix_tree_preload_end();
971
Tejun Heoa2b16932012-04-13 13:11:33 -0700972 if (IS_ERR(blkg)) {
973 ret = PTR_ERR(blkg);
974 goto out_unlock;
975 }
976 q->root_blkg = blkg;
Tejun Heoa0516612012-06-26 15:05:44 -0700977 q->root_rl.blkg = blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -0700978
979 list_for_each_entry(blkg, &q->blkg_list, q_node)
980 cnt++;
981
982 spin_unlock_irq(q->queue_lock);
983
984 /* allocate policy_data for all existing blkgs */
985 while (cnt--) {
Tejun Heof95a04a2012-04-16 13:57:26 -0700986 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
Tejun Heoa2b16932012-04-13 13:11:33 -0700987 if (!pd) {
988 ret = -ENOMEM;
989 goto out_free;
990 }
991 list_add_tail(&pd->alloc_node, &pds);
992 }
993
994 /*
995 * Install the allocated pds. With @q bypassing, no new blkg
996 * should have been created while the queue lock was dropped.
997 */
998 spin_lock_irq(q->queue_lock);
999
1000 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1001 if (WARN_ON(list_empty(&pds))) {
1002 /* umm... this shouldn't happen, just abort */
1003 ret = -ENOMEM;
1004 goto out_unlock;
1005 }
1006 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
1007 list_del_init(&pd->alloc_node);
1008
1009 /* grab blkcg lock too while installing @pd on @blkg */
1010 spin_lock(&blkg->blkcg->lock);
1011
1012 blkg->pd[pol->plid] = pd;
1013 pd->blkg = blkg;
Tejun Heob276a872013-01-09 08:05:12 -08001014 pd->plid = pol->plid;
Tejun Heof9fcc2d2012-04-16 13:57:27 -07001015 pol->pd_init_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -07001016
1017 spin_unlock(&blkg->blkcg->lock);
1018 }
1019
1020 __set_bit(pol->plid, q->blkcg_pols);
1021 ret = 0;
1022out_unlock:
1023 spin_unlock_irq(q->queue_lock);
1024out_free:
1025 blk_queue_bypass_end(q);
1026 list_for_each_entry_safe(pd, n, &pds, alloc_node)
1027 kfree(pd);
1028 return ret;
1029}
1030EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1031
1032/**
1033 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1034 * @q: request_queue of interest
1035 * @pol: blkcg policy to deactivate
1036 *
1037 * Deactivate @pol on @q. Follows the same synchronization rules as
1038 * blkcg_activate_policy().
1039 */
1040void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -07001041 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -07001042{
Tejun Heo3c798392012-04-16 13:57:25 -07001043 struct blkcg_gq *blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -07001044
1045 if (!blkcg_policy_enabled(q, pol))
1046 return;
1047
1048 blk_queue_bypass_start(q);
1049 spin_lock_irq(q->queue_lock);
1050
1051 __clear_bit(pol->plid, q->blkcg_pols);
1052
Tejun Heo6d18b002012-04-13 13:11:35 -07001053 /* if no policy is left, no need for blkgs - shoot them down */
1054 if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
1055 blkg_destroy_all(q);
1056
Tejun Heoa2b16932012-04-13 13:11:33 -07001057 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1058 /* grab blkcg lock too while removing @pd from @blkg */
1059 spin_lock(&blkg->blkcg->lock);
1060
Tejun Heof427d902013-01-09 08:05:12 -08001061 if (pol->pd_offline_fn)
1062 pol->pd_offline_fn(blkg);
Tejun Heof9fcc2d2012-04-16 13:57:27 -07001063 if (pol->pd_exit_fn)
1064 pol->pd_exit_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -07001065
1066 kfree(blkg->pd[pol->plid]);
1067 blkg->pd[pol->plid] = NULL;
1068
1069 spin_unlock(&blkg->blkcg->lock);
1070 }
1071
1072 spin_unlock_irq(q->queue_lock);
1073 blk_queue_bypass_end(q);
1074}
1075EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1076
1077/**
Tejun Heo3c798392012-04-16 13:57:25 -07001078 * blkcg_policy_register - register a blkcg policy
1079 * @pol: blkcg policy to register
Tejun Heo8bd435b2012-04-13 13:11:28 -07001080 *
Tejun Heo3c798392012-04-16 13:57:25 -07001081 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1082 * successful registration. Returns 0 on success and -errno on failure.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001083 */
Tejun Heo3c798392012-04-16 13:57:25 -07001084int blkcg_policy_register(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001085{
Tejun Heo8bd435b2012-04-13 13:11:28 -07001086 int i, ret;
Tejun Heoe8989fa2012-03-05 13:15:20 -08001087
Tejun Heof95a04a2012-04-16 13:57:26 -07001088 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
1089 return -EINVAL;
1090
Tejun Heobc0d6502012-04-13 13:11:26 -07001091 mutex_lock(&blkcg_pol_mutex);
1092
Tejun Heo8bd435b2012-04-13 13:11:28 -07001093 /* find an empty slot */
1094 ret = -ENOSPC;
1095 for (i = 0; i < BLKCG_MAX_POLS; i++)
Tejun Heo3c798392012-04-16 13:57:25 -07001096 if (!blkcg_policy[i])
Tejun Heo8bd435b2012-04-13 13:11:28 -07001097 break;
1098 if (i >= BLKCG_MAX_POLS)
1099 goto out_unlock;
Tejun Heo035d10b2012-03-05 13:15:04 -08001100
Tejun Heo8bd435b2012-04-13 13:11:28 -07001101 /* register and update blkgs */
Tejun Heo3c798392012-04-16 13:57:25 -07001102 pol->plid = i;
1103 blkcg_policy[i] = pol;
Tejun Heo8bd435b2012-04-13 13:11:28 -07001104
Tejun Heo8bd435b2012-04-13 13:11:28 -07001105 /* everything is in place, add intf files for the new policy */
Tejun Heo3c798392012-04-16 13:57:25 -07001106 if (pol->cftypes)
1107 WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes));
Tejun Heo8bd435b2012-04-13 13:11:28 -07001108 ret = 0;
1109out_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -07001110 mutex_unlock(&blkcg_pol_mutex);
Tejun Heo8bd435b2012-04-13 13:11:28 -07001111 return ret;
Vivek Goyal3e252062009-12-04 10:36:42 -05001112}
Tejun Heo3c798392012-04-16 13:57:25 -07001113EXPORT_SYMBOL_GPL(blkcg_policy_register);
Vivek Goyal3e252062009-12-04 10:36:42 -05001114
Tejun Heo8bd435b2012-04-13 13:11:28 -07001115/**
Tejun Heo3c798392012-04-16 13:57:25 -07001116 * blkcg_policy_unregister - unregister a blkcg policy
1117 * @pol: blkcg policy to unregister
Tejun Heo8bd435b2012-04-13 13:11:28 -07001118 *
Tejun Heo3c798392012-04-16 13:57:25 -07001119 * Undo blkcg_policy_register(@pol). Might sleep.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001120 */
Tejun Heo3c798392012-04-16 13:57:25 -07001121void blkcg_policy_unregister(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001122{
Tejun Heobc0d6502012-04-13 13:11:26 -07001123 mutex_lock(&blkcg_pol_mutex);
1124
Tejun Heo3c798392012-04-16 13:57:25 -07001125 if (WARN_ON(blkcg_policy[pol->plid] != pol))
Tejun Heo8bd435b2012-04-13 13:11:28 -07001126 goto out_unlock;
1127
1128 /* kill the intf files first */
Tejun Heo3c798392012-04-16 13:57:25 -07001129 if (pol->cftypes)
1130 cgroup_rm_cftypes(&blkio_subsys, pol->cftypes);
Tejun Heo44ea53d2012-04-01 14:38:43 -07001131
Tejun Heo8bd435b2012-04-13 13:11:28 -07001132 /* unregister and update blkgs */
Tejun Heo3c798392012-04-16 13:57:25 -07001133 blkcg_policy[pol->plid] = NULL;
Tejun Heo8bd435b2012-04-13 13:11:28 -07001134out_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -07001135 mutex_unlock(&blkcg_pol_mutex);
Vivek Goyal3e252062009-12-04 10:36:42 -05001136}
Tejun Heo3c798392012-04-16 13:57:25 -07001137EXPORT_SYMBOL_GPL(blkcg_policy_unregister);