blob: 4e491d9b529255476939b8608d91718c40140059 [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 Heo577cee12013-08-28 14:26:50 -0400238 if (!ret) {
239 if (blkcg == &blkcg_root) {
240 q->root_blkg = blkg;
241 q->root_rl.blkg = blkg;
242 }
Tejun Heoa6371202012-04-19 16:29:24 -0700243 return blkg;
Tejun Heo577cee12013-08-28 14:26:50 -0400244 }
Tejun Heo15974992012-06-04 20:40:52 -0700245
Tejun Heo3c547862013-01-09 08:05:10 -0800246 /* @blkg failed fully initialized, use the usual release path */
247 blkg_put(blkg);
248 return ERR_PTR(ret);
249
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800250err_put_css:
Tejun Heo496fb782012-04-19 16:29:23 -0700251 css_put(&blkcg->css);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800252err_free_blkg:
Tejun Heo15974992012-06-04 20:40:52 -0700253 blkg_free(new_blkg);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800254 return ERR_PTR(ret);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500255}
Tejun Heo3c96cb32012-04-13 13:11:34 -0700256
Tejun Heo86cde6b2013-01-09 08:05:10 -0800257/**
258 * blkg_lookup_create - lookup blkg, try to create one if not there
259 * @blkcg: blkcg of interest
260 * @q: request_queue of interest
261 *
262 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
Tejun Heo3c547862013-01-09 08:05:10 -0800263 * create one. blkg creation is performed recursively from blkcg_root such
264 * that all non-root blkg's have access to the parent blkg. This function
265 * should be called under RCU read lock and @q->queue_lock.
Tejun Heo86cde6b2013-01-09 08:05:10 -0800266 *
267 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
268 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
269 * dead and bypassing, returns ERR_PTR(-EBUSY).
270 */
Tejun Heo3c798392012-04-16 13:57:25 -0700271struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
272 struct request_queue *q)
Tejun Heo3c96cb32012-04-13 13:11:34 -0700273{
Tejun Heo86cde6b2013-01-09 08:05:10 -0800274 struct blkcg_gq *blkg;
275
276 WARN_ON_ONCE(!rcu_read_lock_held());
277 lockdep_assert_held(q->queue_lock);
278
Tejun Heo3c96cb32012-04-13 13:11:34 -0700279 /*
280 * This could be the first entry point of blkcg implementation and
281 * we shouldn't allow anything to go through for a bypassing queue.
282 */
283 if (unlikely(blk_queue_bypass(q)))
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100284 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800285
286 blkg = __blkg_lookup(blkcg, q, true);
287 if (blkg)
288 return blkg;
289
Tejun Heo3c547862013-01-09 08:05:10 -0800290 /*
291 * Create blkgs walking down from blkcg_root to @blkcg, so that all
292 * non-root blkgs have access to their parents.
293 */
294 while (true) {
295 struct blkcg *pos = blkcg;
296 struct blkcg *parent = blkcg_parent(blkcg);
297
298 while (parent && !__blkg_lookup(parent, q, false)) {
299 pos = parent;
300 parent = blkcg_parent(parent);
301 }
302
303 blkg = blkg_create(pos, q, NULL);
304 if (pos == blkcg || IS_ERR(blkg))
305 return blkg;
306 }
Tejun Heo3c96cb32012-04-13 13:11:34 -0700307}
Tejun Heocd1604f2012-03-05 13:15:06 -0800308EXPORT_SYMBOL_GPL(blkg_lookup_create);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500309
Tejun Heo3c798392012-04-16 13:57:25 -0700310static void blkg_destroy(struct blkcg_gq *blkg)
Tejun Heo72e06c22012-03-05 13:15:00 -0800311{
Tejun Heo3c798392012-04-16 13:57:25 -0700312 struct blkcg *blkcg = blkg->blkcg;
Tejun Heof427d902013-01-09 08:05:12 -0800313 int i;
Tejun Heo03aa2642012-03-05 13:15:19 -0800314
Tejun Heo27e1f9d2012-06-05 13:36:44 +0200315 lockdep_assert_held(blkg->q->queue_lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800316 lockdep_assert_held(&blkcg->lock);
Tejun Heo03aa2642012-03-05 13:15:19 -0800317
318 /* Something wrong if we are trying to remove same group twice */
Tejun Heoe8989fa2012-03-05 13:15:20 -0800319 WARN_ON_ONCE(list_empty(&blkg->q_node));
Tejun Heo9f13ef62012-03-05 13:15:21 -0800320 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
Tejun Heoa6371202012-04-19 16:29:24 -0700321
Tejun Heof427d902013-01-09 08:05:12 -0800322 for (i = 0; i < BLKCG_MAX_POLS; i++) {
323 struct blkcg_policy *pol = blkcg_policy[i];
324
325 if (blkg->pd[i] && pol->pd_offline_fn)
326 pol->pd_offline_fn(blkg);
327 }
328 blkg->online = false;
329
Tejun Heoa6371202012-04-19 16:29:24 -0700330 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
Tejun Heoe8989fa2012-03-05 13:15:20 -0800331 list_del_init(&blkg->q_node);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800332 hlist_del_init_rcu(&blkg->blkcg_node);
Tejun Heo03aa2642012-03-05 13:15:19 -0800333
Tejun Heo03aa2642012-03-05 13:15:19 -0800334 /*
Tejun Heoa6371202012-04-19 16:29:24 -0700335 * Both setting lookup hint to and clearing it from @blkg are done
336 * under queue_lock. If it's not pointing to @blkg now, it never
337 * will. Hint assignment itself can race safely.
338 */
339 if (rcu_dereference_raw(blkcg->blkg_hint) == blkg)
340 rcu_assign_pointer(blkcg->blkg_hint, NULL);
341
342 /*
Tejun Heo577cee12013-08-28 14:26:50 -0400343 * If root blkg is destroyed. Just clear the pointer since root_rl
344 * does not take reference on root blkg.
345 */
346 if (blkcg == &blkcg_root) {
347 blkg->q->root_blkg = NULL;
348 blkg->q->root_rl.blkg = NULL;
349 }
350
351 /*
Tejun Heo03aa2642012-03-05 13:15:19 -0800352 * Put the reference taken at the time of creation so that when all
353 * queues are gone, group can be destroyed.
354 */
355 blkg_put(blkg);
356}
357
Tejun Heo9f13ef62012-03-05 13:15:21 -0800358/**
359 * blkg_destroy_all - destroy all blkgs associated with a request_queue
360 * @q: request_queue of interest
Tejun Heo9f13ef62012-03-05 13:15:21 -0800361 *
Tejun Heo3c96cb32012-04-13 13:11:34 -0700362 * Destroy all blkgs associated with @q.
Tejun Heo9f13ef62012-03-05 13:15:21 -0800363 */
Tejun Heo3c96cb32012-04-13 13:11:34 -0700364static void blkg_destroy_all(struct request_queue *q)
Tejun Heo03aa2642012-03-05 13:15:19 -0800365{
Tejun Heo3c798392012-04-16 13:57:25 -0700366 struct blkcg_gq *blkg, *n;
Tejun Heo72e06c22012-03-05 13:15:00 -0800367
Tejun Heo6d18b002012-04-13 13:11:35 -0700368 lockdep_assert_held(q->queue_lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800369
Tejun Heo9f13ef62012-03-05 13:15:21 -0800370 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
Tejun Heo3c798392012-04-16 13:57:25 -0700371 struct blkcg *blkcg = blkg->blkcg;
Tejun Heo72e06c22012-03-05 13:15:00 -0800372
Tejun Heo9f13ef62012-03-05 13:15:21 -0800373 spin_lock(&blkcg->lock);
374 blkg_destroy(blkg);
375 spin_unlock(&blkcg->lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800376 }
377}
378
Tejun Heo2a4fd072013-05-14 13:52:31 -0700379/*
380 * A group is RCU protected, but having an rcu lock does not mean that one
381 * can access all the fields of blkg and assume these are valid. For
382 * example, don't try to follow throtl_data and request queue links.
383 *
384 * Having a reference to blkg under an rcu allows accesses to only values
385 * local to groups like group stats and group rate limits.
386 */
387void __blkg_release_rcu(struct rcu_head *rcu_head)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800388{
Tejun Heo2a4fd072013-05-14 13:52:31 -0700389 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
Tejun Heodb613672013-05-14 13:52:31 -0700390 int i;
391
392 /* tell policies that this one is being freed */
393 for (i = 0; i < BLKCG_MAX_POLS; i++) {
394 struct blkcg_policy *pol = blkcg_policy[i];
395
396 if (blkg->pd[i] && pol->pd_exit_fn)
397 pol->pd_exit_fn(blkg);
398 }
399
Tejun Heo3c547862013-01-09 08:05:10 -0800400 /* release the blkcg and parent blkg refs this blkg has been holding */
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800401 css_put(&blkg->blkcg->css);
Tejun Heo2a4fd072013-05-14 13:52:31 -0700402 if (blkg->parent) {
403 spin_lock_irq(blkg->q->queue_lock);
Tejun Heo3c547862013-01-09 08:05:10 -0800404 blkg_put(blkg->parent);
Tejun Heo2a4fd072013-05-14 13:52:31 -0700405 spin_unlock_irq(blkg->q->queue_lock);
406 }
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800407
Tejun Heo2a4fd072013-05-14 13:52:31 -0700408 blkg_free(blkg);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800409}
Tejun Heo2a4fd072013-05-14 13:52:31 -0700410EXPORT_SYMBOL_GPL(__blkg_release_rcu);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800411
Tejun Heoa0516612012-06-26 15:05:44 -0700412/*
413 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
414 * because the root blkg uses @q->root_rl instead of its own rl.
415 */
416struct request_list *__blk_queue_next_rl(struct request_list *rl,
417 struct request_queue *q)
418{
419 struct list_head *ent;
420 struct blkcg_gq *blkg;
421
422 /*
423 * Determine the current blkg list_head. The first entry is
424 * root_rl which is off @q->blkg_list and mapped to the head.
425 */
426 if (rl == &q->root_rl) {
427 ent = &q->blkg_list;
Jun'ichi Nomura65c77fd2012-10-22 10:15:37 +0900428 /* There are no more block groups, hence no request lists */
429 if (list_empty(ent))
430 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -0700431 } else {
432 blkg = container_of(rl, struct blkcg_gq, rl);
433 ent = &blkg->q_node;
434 }
435
436 /* walk to the next list_head, skip root blkcg */
437 ent = ent->next;
438 if (ent == &q->root_blkg->q_node)
439 ent = ent->next;
440 if (ent == &q->blkg_list)
441 return NULL;
442
443 blkg = container_of(ent, struct blkcg_gq, q_node);
444 return &blkg->rl;
445}
446
Tejun Heo182446d2013-08-08 20:11:24 -0400447static int blkcg_reset_stats(struct cgroup_subsys_state *css,
448 struct cftype *cftype, u64 val)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700449{
Tejun Heo182446d2013-08-08 20:11:24 -0400450 struct blkcg *blkcg = css_to_blkcg(css);
Tejun Heo3c798392012-04-16 13:57:25 -0700451 struct blkcg_gq *blkg;
Tejun Heobc0d6502012-04-13 13:11:26 -0700452 int i;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700453
Tejun Heobc0d6502012-04-13 13:11:26 -0700454 mutex_lock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700455 spin_lock_irq(&blkcg->lock);
Tejun Heo997a0262012-03-08 10:53:58 -0800456
457 /*
458 * Note that stat reset is racy - it doesn't synchronize against
459 * stat updates. This is a debug feature which shouldn't exist
460 * anyway. If you get hit by a race, retry.
461 */
Sasha Levinb67bfe02013-02-27 17:06:00 -0800462 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo8bd435b2012-04-13 13:11:28 -0700463 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -0700464 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heo549d3aa2012-03-05 13:15:16 -0800465
Tejun Heoa2b16932012-04-13 13:11:33 -0700466 if (blkcg_policy_enabled(blkg->q, pol) &&
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700467 pol->pd_reset_stats_fn)
468 pol->pd_reset_stats_fn(blkg);
Tejun Heobc0d6502012-04-13 13:11:26 -0700469 }
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700470 }
Vivek Goyalf0bdc8c2011-05-19 15:38:30 -0400471
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700472 spin_unlock_irq(&blkcg->lock);
Tejun Heobc0d6502012-04-13 13:11:26 -0700473 mutex_unlock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700474 return 0;
475}
476
Tejun Heo3c798392012-04-16 13:57:25 -0700477static const char *blkg_dev_name(struct blkcg_gq *blkg)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700478{
Tejun Heod3d32e62012-04-01 14:38:42 -0700479 /* some drivers (floppy) instantiate a queue w/o disk registered */
480 if (blkg->q->backing_dev_info.dev)
481 return dev_name(blkg->q->backing_dev_info.dev);
482 return NULL;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700483}
484
Tejun Heod3d32e62012-04-01 14:38:42 -0700485/**
486 * blkcg_print_blkgs - helper for printing per-blkg data
487 * @sf: seq_file to print to
488 * @blkcg: blkcg of interest
489 * @prfill: fill function to print out a blkg
490 * @pol: policy in question
491 * @data: data to be passed to @prfill
492 * @show_total: to print out sum of prfill return values or not
493 *
494 * This function invokes @prfill on each blkg of @blkcg if pd for the
495 * policy specified by @pol exists. @prfill is invoked with @sf, the
Tejun Heo810ecfa2013-01-09 08:05:13 -0800496 * policy data and @data and the matching queue lock held. If @show_total
497 * is %true, the sum of the return values from @prfill is printed with
498 * "Total" label at the end.
Tejun Heod3d32e62012-04-01 14:38:42 -0700499 *
500 * This is to be used to construct print functions for
501 * cftype->read_seq_string method.
502 */
Tejun Heo3c798392012-04-16 13:57:25 -0700503void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
Tejun Heof95a04a2012-04-16 13:57:26 -0700504 u64 (*prfill)(struct seq_file *,
505 struct blkg_policy_data *, int),
Tejun Heo3c798392012-04-16 13:57:25 -0700506 const struct blkcg_policy *pol, int data,
Tejun Heoec399342012-04-13 13:11:27 -0700507 bool show_total)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400508{
Tejun Heo3c798392012-04-16 13:57:25 -0700509 struct blkcg_gq *blkg;
Tejun Heod3d32e62012-04-01 14:38:42 -0700510 u64 total = 0;
511
Tejun Heo810ecfa2013-01-09 08:05:13 -0800512 rcu_read_lock();
Linus Torvaldsee89f812013-02-28 12:52:24 -0800513 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo810ecfa2013-01-09 08:05:13 -0800514 spin_lock_irq(blkg->q->queue_lock);
Tejun Heoa2b16932012-04-13 13:11:33 -0700515 if (blkcg_policy_enabled(blkg->q, pol))
Tejun Heof95a04a2012-04-16 13:57:26 -0700516 total += prfill(sf, blkg->pd[pol->plid], data);
Tejun Heo810ecfa2013-01-09 08:05:13 -0800517 spin_unlock_irq(blkg->q->queue_lock);
518 }
519 rcu_read_unlock();
Tejun Heod3d32e62012-04-01 14:38:42 -0700520
521 if (show_total)
522 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
523}
Tejun Heo829fdb52012-04-01 14:38:43 -0700524EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
Tejun Heod3d32e62012-04-01 14:38:42 -0700525
526/**
527 * __blkg_prfill_u64 - prfill helper for a single u64 value
528 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700529 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700530 * @v: value to print
531 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700532 * Print @v to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700533 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700534u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
Tejun Heod3d32e62012-04-01 14:38:42 -0700535{
Tejun Heof95a04a2012-04-16 13:57:26 -0700536 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700537
538 if (!dname)
539 return 0;
540
541 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
542 return v;
543}
Tejun Heo829fdb52012-04-01 14:38:43 -0700544EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
Tejun Heod3d32e62012-04-01 14:38:42 -0700545
546/**
547 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
548 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700549 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700550 * @rwstat: rwstat to print
551 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700552 * Print @rwstat to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700553 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700554u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
Tejun Heo829fdb52012-04-01 14:38:43 -0700555 const struct blkg_rwstat *rwstat)
Tejun Heod3d32e62012-04-01 14:38:42 -0700556{
557 static const char *rwstr[] = {
558 [BLKG_RWSTAT_READ] = "Read",
559 [BLKG_RWSTAT_WRITE] = "Write",
560 [BLKG_RWSTAT_SYNC] = "Sync",
561 [BLKG_RWSTAT_ASYNC] = "Async",
562 };
Tejun Heof95a04a2012-04-16 13:57:26 -0700563 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700564 u64 v;
565 int i;
566
567 if (!dname)
568 return 0;
569
570 for (i = 0; i < BLKG_RWSTAT_NR; i++)
571 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
572 (unsigned long long)rwstat->cnt[i]);
573
574 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
575 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
576 return v;
577}
Tejun Heob50da392013-01-09 08:05:12 -0800578EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700579
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700580/**
581 * blkg_prfill_stat - prfill callback for blkg_stat
582 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700583 * @pd: policy private data of interest
584 * @off: offset to the blkg_stat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700585 *
586 * prfill callback for printing a blkg_stat.
587 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700588u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700589{
Tejun Heof95a04a2012-04-16 13:57:26 -0700590 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
Tejun Heod3d32e62012-04-01 14:38:42 -0700591}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700592EXPORT_SYMBOL_GPL(blkg_prfill_stat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700593
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700594/**
595 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
596 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700597 * @pd: policy private data of interest
598 * @off: offset to the blkg_rwstat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700599 *
600 * prfill callback for printing a blkg_rwstat.
601 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700602u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
603 int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700604{
Tejun Heof95a04a2012-04-16 13:57:26 -0700605 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
Tejun Heod3d32e62012-04-01 14:38:42 -0700606
Tejun Heof95a04a2012-04-16 13:57:26 -0700607 return __blkg_prfill_rwstat(sf, pd, &rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700608}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700609EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700610
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700611/**
Tejun Heo16b3de62013-01-09 08:05:12 -0800612 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
613 * @pd: policy private data of interest
614 * @off: offset to the blkg_stat in @pd
615 *
616 * Collect the blkg_stat specified by @off from @pd and all its online
617 * descendants and return the sum. The caller must be holding the queue
618 * lock for online tests.
619 */
620u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off)
621{
622 struct blkcg_policy *pol = blkcg_policy[pd->plid];
623 struct blkcg_gq *pos_blkg;
Tejun Heo492eb212013-08-08 20:11:25 -0400624 struct cgroup_subsys_state *pos_css;
Tejun Heobd8815a2013-08-08 20:11:27 -0400625 u64 sum = 0;
Tejun Heo16b3de62013-01-09 08:05:12 -0800626
627 lockdep_assert_held(pd->blkg->q->queue_lock);
628
Tejun Heo16b3de62013-01-09 08:05:12 -0800629 rcu_read_lock();
Tejun Heo492eb212013-08-08 20:11:25 -0400630 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
Tejun Heo16b3de62013-01-09 08:05:12 -0800631 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
632 struct blkg_stat *stat = (void *)pos_pd + off;
633
634 if (pos_blkg->online)
635 sum += blkg_stat_read(stat);
636 }
637 rcu_read_unlock();
638
639 return sum;
640}
641EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
642
643/**
644 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
645 * @pd: policy private data of interest
646 * @off: offset to the blkg_stat in @pd
647 *
648 * Collect the blkg_rwstat specified by @off from @pd and all its online
649 * descendants and return the sum. The caller must be holding the queue
650 * lock for online tests.
651 */
652struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
653 int off)
654{
655 struct blkcg_policy *pol = blkcg_policy[pd->plid];
656 struct blkcg_gq *pos_blkg;
Tejun Heo492eb212013-08-08 20:11:25 -0400657 struct cgroup_subsys_state *pos_css;
Tejun Heobd8815a2013-08-08 20:11:27 -0400658 struct blkg_rwstat sum = { };
Tejun Heo16b3de62013-01-09 08:05:12 -0800659 int i;
660
661 lockdep_assert_held(pd->blkg->q->queue_lock);
662
Tejun Heo16b3de62013-01-09 08:05:12 -0800663 rcu_read_lock();
Tejun Heo492eb212013-08-08 20:11:25 -0400664 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
Tejun Heo16b3de62013-01-09 08:05:12 -0800665 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
666 struct blkg_rwstat *rwstat = (void *)pos_pd + off;
667 struct blkg_rwstat tmp;
668
669 if (!pos_blkg->online)
670 continue;
671
672 tmp = blkg_rwstat_read(rwstat);
673
674 for (i = 0; i < BLKG_RWSTAT_NR; i++)
675 sum.cnt[i] += tmp.cnt[i];
676 }
677 rcu_read_unlock();
678
679 return sum;
680}
681EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
682
683/**
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700684 * blkg_conf_prep - parse and prepare for per-blkg config update
685 * @blkcg: target block cgroup
Tejun Heoda8b0662012-04-13 13:11:29 -0700686 * @pol: target policy
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700687 * @input: input string
688 * @ctx: blkg_conf_ctx to be filled
689 *
690 * Parse per-blkg config update from @input and initialize @ctx with the
691 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
Tejun Heoda8b0662012-04-13 13:11:29 -0700692 * value. This function returns with RCU read lock and queue lock held and
693 * must be paired with blkg_conf_finish().
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700694 */
Tejun Heo3c798392012-04-16 13:57:25 -0700695int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
696 const char *input, struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700697 __acquires(rcu) __acquires(disk->queue->queue_lock)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800698{
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700699 struct gendisk *disk;
Tejun Heo3c798392012-04-16 13:57:25 -0700700 struct blkcg_gq *blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700701 unsigned int major, minor;
702 unsigned long long v;
703 int part, ret;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800704
Tejun Heo726fa692012-04-01 14:38:43 -0700705 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
706 return -EINVAL;
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700707
Tejun Heo726fa692012-04-01 14:38:43 -0700708 disk = get_gendisk(MKDEV(major, minor), &part);
Tejun Heo4bfd4822012-03-05 13:15:08 -0800709 if (!disk || part)
Tejun Heo726fa692012-04-01 14:38:43 -0700710 return -EINVAL;
Tejun Heoe56da7e2012-03-05 13:15:07 -0800711
712 rcu_read_lock();
Tejun Heo4bfd4822012-03-05 13:15:08 -0800713 spin_lock_irq(disk->queue->queue_lock);
Tejun Heoda8b0662012-04-13 13:11:29 -0700714
Tejun Heoa2b16932012-04-13 13:11:33 -0700715 if (blkcg_policy_enabled(disk->queue, pol))
Tejun Heo3c96cb32012-04-13 13:11:34 -0700716 blkg = blkg_lookup_create(blkcg, disk->queue);
Tejun Heoa2b16932012-04-13 13:11:33 -0700717 else
718 blkg = ERR_PTR(-EINVAL);
Tejun Heoe56da7e2012-03-05 13:15:07 -0800719
Tejun Heo4bfd4822012-03-05 13:15:08 -0800720 if (IS_ERR(blkg)) {
721 ret = PTR_ERR(blkg);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700722 rcu_read_unlock();
Tejun Heoda8b0662012-04-13 13:11:29 -0700723 spin_unlock_irq(disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700724 put_disk(disk);
725 /*
726 * If queue was bypassing, we should retry. Do so after a
727 * short msleep(). It isn't strictly necessary but queue
728 * can be bypassing for some time and it's always nice to
729 * avoid busy looping.
730 */
731 if (ret == -EBUSY) {
732 msleep(10);
733 ret = restart_syscall();
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400734 }
Tejun Heo726fa692012-04-01 14:38:43 -0700735 return ret;
Vivek Goyal062a6442010-09-15 17:06:33 -0400736 }
Tejun Heoe56da7e2012-03-05 13:15:07 -0800737
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700738 ctx->disk = disk;
739 ctx->blkg = blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700740 ctx->v = v;
741 return 0;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800742}
Tejun Heo829fdb52012-04-01 14:38:43 -0700743EXPORT_SYMBOL_GPL(blkg_conf_prep);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800744
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700745/**
746 * blkg_conf_finish - finish up per-blkg config update
747 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
748 *
749 * Finish up after per-blkg config update. This function must be paired
750 * with blkg_conf_prep().
751 */
Tejun Heo829fdb52012-04-01 14:38:43 -0700752void blkg_conf_finish(struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700753 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800754{
Tejun Heoda8b0662012-04-13 13:11:29 -0700755 spin_unlock_irq(ctx->disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700756 rcu_read_unlock();
757 put_disk(ctx->disk);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800758}
Tejun Heo829fdb52012-04-01 14:38:43 -0700759EXPORT_SYMBOL_GPL(blkg_conf_finish);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800760
Tejun Heo3c798392012-04-16 13:57:25 -0700761struct cftype blkcg_files[] = {
Vivek Goyal31e4c282009-12-03 12:59:42 -0500762 {
Divyesh Shah84c124d2010-04-09 08:31:19 +0200763 .name = "reset_stats",
Tejun Heo3c798392012-04-16 13:57:25 -0700764 .write_u64 = blkcg_reset_stats,
Vivek Goyal22084192009-12-03 12:59:49 -0500765 },
Tejun Heo4baf6e32012-04-01 12:09:55 -0700766 { } /* terminate */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500767};
768
Tejun Heo9f13ef62012-03-05 13:15:21 -0800769/**
Tejun Heo92fb9742012-11-19 08:13:38 -0800770 * blkcg_css_offline - cgroup css_offline callback
Tejun Heoeb954192013-08-08 20:11:23 -0400771 * @css: css of interest
Tejun Heo9f13ef62012-03-05 13:15:21 -0800772 *
Tejun Heoeb954192013-08-08 20:11:23 -0400773 * This function is called when @css is about to go away and responsible
774 * for shooting down all blkgs associated with @css. blkgs should be
Tejun Heo9f13ef62012-03-05 13:15:21 -0800775 * removed while holding both q and blkcg locks. As blkcg lock is nested
776 * inside q lock, this function performs reverse double lock dancing.
777 *
778 * This is the blkcg counterpart of ioc_release_fn().
779 */
Tejun Heoeb954192013-08-08 20:11:23 -0400780static void blkcg_css_offline(struct cgroup_subsys_state *css)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500781{
Tejun Heoeb954192013-08-08 20:11:23 -0400782 struct blkcg *blkcg = css_to_blkcg(css);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500783
Tejun Heo9f13ef62012-03-05 13:15:21 -0800784 spin_lock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800785
Tejun Heo9f13ef62012-03-05 13:15:21 -0800786 while (!hlist_empty(&blkcg->blkg_list)) {
Tejun Heo3c798392012-04-16 13:57:25 -0700787 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
788 struct blkcg_gq, blkcg_node);
Tejun Heoc875f4d2012-03-05 13:15:22 -0800789 struct request_queue *q = blkg->q;
Vivek Goyalb1c35762009-12-03 12:59:47 -0500790
Tejun Heo9f13ef62012-03-05 13:15:21 -0800791 if (spin_trylock(q->queue_lock)) {
792 blkg_destroy(blkg);
793 spin_unlock(q->queue_lock);
794 } else {
795 spin_unlock_irq(&blkcg->lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800796 cpu_relax();
Dan Carpentera5567932012-03-29 20:57:08 +0200797 spin_lock_irq(&blkcg->lock);
Jens Axboe0f3942a2010-05-03 14:28:55 +0200798 }
Tejun Heo9f13ef62012-03-05 13:15:21 -0800799 }
Jens Axboe0f3942a2010-05-03 14:28:55 +0200800
Tejun Heo9f13ef62012-03-05 13:15:21 -0800801 spin_unlock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800802}
803
Tejun Heoeb954192013-08-08 20:11:23 -0400804static void blkcg_css_free(struct cgroup_subsys_state *css)
Tejun Heo7ee9c562012-03-05 13:15:11 -0800805{
Tejun Heoeb954192013-08-08 20:11:23 -0400806 struct blkcg *blkcg = css_to_blkcg(css);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800807
Tejun Heo3c798392012-04-16 13:57:25 -0700808 if (blkcg != &blkcg_root)
Ben Blum67523c42010-03-10 15:22:11 -0800809 kfree(blkcg);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500810}
811
Tejun Heoeb954192013-08-08 20:11:23 -0400812static struct cgroup_subsys_state *
813blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500814{
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700815 static atomic64_t id_seq = ATOMIC64_INIT(0);
Tejun Heo3c798392012-04-16 13:57:25 -0700816 struct blkcg *blkcg;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500817
Tejun Heoeb954192013-08-08 20:11:23 -0400818 if (!parent_css) {
Tejun Heo3c798392012-04-16 13:57:25 -0700819 blkcg = &blkcg_root;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500820 goto done;
821 }
822
Vivek Goyal31e4c282009-12-03 12:59:42 -0500823 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
824 if (!blkcg)
825 return ERR_PTR(-ENOMEM);
826
Tejun Heo3381cb82012-04-01 14:38:44 -0700827 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
Tejun Heoe71357e2013-01-09 08:05:10 -0800828 blkcg->cfq_leaf_weight = CFQ_WEIGHT_DEFAULT;
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700829 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500830done:
831 spin_lock_init(&blkcg->lock);
Tejun Heoa6371202012-04-19 16:29:24 -0700832 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500833 INIT_HLIST_HEAD(&blkcg->blkg_list);
834
835 return &blkcg->css;
836}
837
Tejun Heo5efd6112012-03-05 13:15:12 -0800838/**
839 * blkcg_init_queue - initialize blkcg part of request queue
840 * @q: request_queue to initialize
841 *
842 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
843 * part of new request_queue @q.
844 *
845 * RETURNS:
846 * 0 on success, -errno on failure.
847 */
848int blkcg_init_queue(struct request_queue *q)
849{
850 might_sleep();
851
Tejun Heo3c96cb32012-04-13 13:11:34 -0700852 return blk_throtl_init(q);
Tejun Heo5efd6112012-03-05 13:15:12 -0800853}
854
855/**
856 * blkcg_drain_queue - drain blkcg part of request_queue
857 * @q: request_queue to drain
858 *
859 * Called from blk_drain_queue(). Responsible for draining blkcg part.
860 */
861void blkcg_drain_queue(struct request_queue *q)
862{
863 lockdep_assert_held(q->queue_lock);
864
865 blk_throtl_drain(q);
866}
867
868/**
869 * blkcg_exit_queue - exit and release blkcg part of request_queue
870 * @q: request_queue being released
871 *
872 * Called from blk_release_queue(). Responsible for exiting blkcg part.
873 */
874void blkcg_exit_queue(struct request_queue *q)
875{
Tejun Heo6d18b002012-04-13 13:11:35 -0700876 spin_lock_irq(q->queue_lock);
Tejun Heo3c96cb32012-04-13 13:11:34 -0700877 blkg_destroy_all(q);
Tejun Heo6d18b002012-04-13 13:11:35 -0700878 spin_unlock_irq(q->queue_lock);
879
Tejun Heo5efd6112012-03-05 13:15:12 -0800880 blk_throtl_exit(q);
881}
882
Vivek Goyal31e4c282009-12-03 12:59:42 -0500883/*
884 * We cannot support shared io contexts, as we have no mean to support
885 * two tasks with the same ioc in two different groups without major rework
886 * of the main cic data structures. For now we allow a task to change
887 * its cgroup only if it's the only owner of its ioc.
888 */
Tejun Heoeb954192013-08-08 20:11:23 -0400889static int blkcg_can_attach(struct cgroup_subsys_state *css,
890 struct cgroup_taskset *tset)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500891{
Tejun Heobb9d97b2011-12-12 18:12:21 -0800892 struct task_struct *task;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500893 struct io_context *ioc;
894 int ret = 0;
895
896 /* task_lock() is needed to avoid races with exit_io_context() */
Tejun Heod99c8722013-08-08 20:11:27 -0400897 cgroup_taskset_for_each(task, css, tset) {
Tejun Heobb9d97b2011-12-12 18:12:21 -0800898 task_lock(task);
899 ioc = task->io_context;
900 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
901 ret = -EINVAL;
902 task_unlock(task);
903 if (ret)
904 break;
905 }
Vivek Goyal31e4c282009-12-03 12:59:42 -0500906 return ret;
907}
908
Tejun Heo676f7c82012-04-01 12:09:55 -0700909struct cgroup_subsys blkio_subsys = {
910 .name = "blkio",
Tejun Heo92fb9742012-11-19 08:13:38 -0800911 .css_alloc = blkcg_css_alloc,
912 .css_offline = blkcg_css_offline,
913 .css_free = blkcg_css_free,
Tejun Heo3c798392012-04-16 13:57:25 -0700914 .can_attach = blkcg_can_attach,
Tejun Heo676f7c82012-04-01 12:09:55 -0700915 .subsys_id = blkio_subsys_id,
Tejun Heo3c798392012-04-16 13:57:25 -0700916 .base_cftypes = blkcg_files,
Tejun Heo676f7c82012-04-01 12:09:55 -0700917 .module = THIS_MODULE,
918};
919EXPORT_SYMBOL_GPL(blkio_subsys);
920
Tejun Heo8bd435b2012-04-13 13:11:28 -0700921/**
Tejun Heoa2b16932012-04-13 13:11:33 -0700922 * blkcg_activate_policy - activate a blkcg policy on a request_queue
923 * @q: request_queue of interest
924 * @pol: blkcg policy to activate
925 *
926 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
927 * bypass mode to populate its blkgs with policy_data for @pol.
928 *
929 * Activation happens with @q bypassed, so nobody would be accessing blkgs
930 * from IO path. Update of each blkg is protected by both queue and blkcg
931 * locks so that holding either lock and testing blkcg_policy_enabled() is
932 * always enough for dereferencing policy data.
933 *
934 * The caller is responsible for synchronizing [de]activations and policy
935 * [un]registerations. Returns 0 on success, -errno on failure.
936 */
937int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700938 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -0700939{
940 LIST_HEAD(pds);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800941 struct blkcg_gq *blkg, *new_blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -0700942 struct blkg_policy_data *pd, *n;
943 int cnt = 0, ret;
Tejun Heo15974992012-06-04 20:40:52 -0700944 bool preloaded;
Tejun Heoa2b16932012-04-13 13:11:33 -0700945
946 if (blkcg_policy_enabled(q, pol))
947 return 0;
948
Tejun Heo15974992012-06-04 20:40:52 -0700949 /* preallocations for root blkg */
Tejun Heo86cde6b2013-01-09 08:05:10 -0800950 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
951 if (!new_blkg)
Tejun Heo15974992012-06-04 20:40:52 -0700952 return -ENOMEM;
953
Tejun Heoa2b16932012-04-13 13:11:33 -0700954 blk_queue_bypass_start(q);
955
Jun'ichi Nomurae5072662013-04-09 15:01:21 +0200956 preloaded = !radix_tree_preload(GFP_KERNEL);
957
Tejun Heo86cde6b2013-01-09 08:05:10 -0800958 /*
959 * Make sure the root blkg exists and count the existing blkgs. As
960 * @q is bypassing at this point, blkg_lookup_create() can't be
961 * used. Open code it.
962 */
Tejun Heoa2b16932012-04-13 13:11:33 -0700963 spin_lock_irq(q->queue_lock);
964
965 rcu_read_lock();
Tejun Heo86cde6b2013-01-09 08:05:10 -0800966 blkg = __blkg_lookup(&blkcg_root, q, false);
967 if (blkg)
968 blkg_free(new_blkg);
969 else
970 blkg = blkg_create(&blkcg_root, q, new_blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -0700971 rcu_read_unlock();
972
Tejun Heo15974992012-06-04 20:40:52 -0700973 if (preloaded)
974 radix_tree_preload_end();
975
Tejun Heoa2b16932012-04-13 13:11:33 -0700976 if (IS_ERR(blkg)) {
977 ret = PTR_ERR(blkg);
978 goto out_unlock;
979 }
Tejun Heoa2b16932012-04-13 13:11:33 -0700980
981 list_for_each_entry(blkg, &q->blkg_list, q_node)
982 cnt++;
983
984 spin_unlock_irq(q->queue_lock);
985
986 /* allocate policy_data for all existing blkgs */
987 while (cnt--) {
Tejun Heof95a04a2012-04-16 13:57:26 -0700988 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
Tejun Heoa2b16932012-04-13 13:11:33 -0700989 if (!pd) {
990 ret = -ENOMEM;
991 goto out_free;
992 }
993 list_add_tail(&pd->alloc_node, &pds);
994 }
995
996 /*
997 * Install the allocated pds. With @q bypassing, no new blkg
998 * should have been created while the queue lock was dropped.
999 */
1000 spin_lock_irq(q->queue_lock);
1001
1002 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1003 if (WARN_ON(list_empty(&pds))) {
1004 /* umm... this shouldn't happen, just abort */
1005 ret = -ENOMEM;
1006 goto out_unlock;
1007 }
1008 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
1009 list_del_init(&pd->alloc_node);
1010
1011 /* grab blkcg lock too while installing @pd on @blkg */
1012 spin_lock(&blkg->blkcg->lock);
1013
1014 blkg->pd[pol->plid] = pd;
1015 pd->blkg = blkg;
Tejun Heob276a872013-01-09 08:05:12 -08001016 pd->plid = pol->plid;
Tejun Heof9fcc2d2012-04-16 13:57:27 -07001017 pol->pd_init_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -07001018
1019 spin_unlock(&blkg->blkcg->lock);
1020 }
1021
1022 __set_bit(pol->plid, q->blkcg_pols);
1023 ret = 0;
1024out_unlock:
1025 spin_unlock_irq(q->queue_lock);
1026out_free:
1027 blk_queue_bypass_end(q);
1028 list_for_each_entry_safe(pd, n, &pds, alloc_node)
1029 kfree(pd);
1030 return ret;
1031}
1032EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1033
1034/**
1035 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1036 * @q: request_queue of interest
1037 * @pol: blkcg policy to deactivate
1038 *
1039 * Deactivate @pol on @q. Follows the same synchronization rules as
1040 * blkcg_activate_policy().
1041 */
1042void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -07001043 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -07001044{
Tejun Heo3c798392012-04-16 13:57:25 -07001045 struct blkcg_gq *blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -07001046
1047 if (!blkcg_policy_enabled(q, pol))
1048 return;
1049
1050 blk_queue_bypass_start(q);
1051 spin_lock_irq(q->queue_lock);
1052
1053 __clear_bit(pol->plid, q->blkcg_pols);
1054
Tejun Heo6d18b002012-04-13 13:11:35 -07001055 /* if no policy is left, no need for blkgs - shoot them down */
1056 if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
1057 blkg_destroy_all(q);
1058
Tejun Heoa2b16932012-04-13 13:11:33 -07001059 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1060 /* grab blkcg lock too while removing @pd from @blkg */
1061 spin_lock(&blkg->blkcg->lock);
1062
Tejun Heof427d902013-01-09 08:05:12 -08001063 if (pol->pd_offline_fn)
1064 pol->pd_offline_fn(blkg);
Tejun Heof9fcc2d2012-04-16 13:57:27 -07001065 if (pol->pd_exit_fn)
1066 pol->pd_exit_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -07001067
1068 kfree(blkg->pd[pol->plid]);
1069 blkg->pd[pol->plid] = NULL;
1070
1071 spin_unlock(&blkg->blkcg->lock);
1072 }
1073
1074 spin_unlock_irq(q->queue_lock);
1075 blk_queue_bypass_end(q);
1076}
1077EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1078
1079/**
Tejun Heo3c798392012-04-16 13:57:25 -07001080 * blkcg_policy_register - register a blkcg policy
1081 * @pol: blkcg policy to register
Tejun Heo8bd435b2012-04-13 13:11:28 -07001082 *
Tejun Heo3c798392012-04-16 13:57:25 -07001083 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1084 * successful registration. Returns 0 on success and -errno on failure.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001085 */
Tejun Heo3c798392012-04-16 13:57:25 -07001086int blkcg_policy_register(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001087{
Tejun Heo8bd435b2012-04-13 13:11:28 -07001088 int i, ret;
Tejun Heoe8989fa2012-03-05 13:15:20 -08001089
Tejun Heof95a04a2012-04-16 13:57:26 -07001090 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
1091 return -EINVAL;
1092
Tejun Heobc0d6502012-04-13 13:11:26 -07001093 mutex_lock(&blkcg_pol_mutex);
1094
Tejun Heo8bd435b2012-04-13 13:11:28 -07001095 /* find an empty slot */
1096 ret = -ENOSPC;
1097 for (i = 0; i < BLKCG_MAX_POLS; i++)
Tejun Heo3c798392012-04-16 13:57:25 -07001098 if (!blkcg_policy[i])
Tejun Heo8bd435b2012-04-13 13:11:28 -07001099 break;
1100 if (i >= BLKCG_MAX_POLS)
1101 goto out_unlock;
Tejun Heo035d10b2012-03-05 13:15:04 -08001102
Tejun Heo8bd435b2012-04-13 13:11:28 -07001103 /* register and update blkgs */
Tejun Heo3c798392012-04-16 13:57:25 -07001104 pol->plid = i;
1105 blkcg_policy[i] = pol;
Tejun Heo8bd435b2012-04-13 13:11:28 -07001106
Tejun Heo8bd435b2012-04-13 13:11:28 -07001107 /* everything is in place, add intf files for the new policy */
Tejun Heo3c798392012-04-16 13:57:25 -07001108 if (pol->cftypes)
1109 WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes));
Tejun Heo8bd435b2012-04-13 13:11:28 -07001110 ret = 0;
1111out_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -07001112 mutex_unlock(&blkcg_pol_mutex);
Tejun Heo8bd435b2012-04-13 13:11:28 -07001113 return ret;
Vivek Goyal3e252062009-12-04 10:36:42 -05001114}
Tejun Heo3c798392012-04-16 13:57:25 -07001115EXPORT_SYMBOL_GPL(blkcg_policy_register);
Vivek Goyal3e252062009-12-04 10:36:42 -05001116
Tejun Heo8bd435b2012-04-13 13:11:28 -07001117/**
Tejun Heo3c798392012-04-16 13:57:25 -07001118 * blkcg_policy_unregister - unregister a blkcg policy
1119 * @pol: blkcg policy to unregister
Tejun Heo8bd435b2012-04-13 13:11:28 -07001120 *
Tejun Heo3c798392012-04-16 13:57:25 -07001121 * Undo blkcg_policy_register(@pol). Might sleep.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001122 */
Tejun Heo3c798392012-04-16 13:57:25 -07001123void blkcg_policy_unregister(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001124{
Tejun Heobc0d6502012-04-13 13:11:26 -07001125 mutex_lock(&blkcg_pol_mutex);
1126
Tejun Heo3c798392012-04-16 13:57:25 -07001127 if (WARN_ON(blkcg_policy[pol->plid] != pol))
Tejun Heo8bd435b2012-04-13 13:11:28 -07001128 goto out_unlock;
1129
1130 /* kill the intf files first */
Tejun Heo3c798392012-04-16 13:57:25 -07001131 if (pol->cftypes)
Tejun Heo2bb566c2013-08-08 20:11:23 -04001132 cgroup_rm_cftypes(pol->cftypes);
Tejun Heo44ea53d2012-04-01 14:38:43 -07001133
Tejun Heo8bd435b2012-04-13 13:11:28 -07001134 /* unregister and update blkgs */
Tejun Heo3c798392012-04-16 13:57:25 -07001135 blkcg_policy[pol->plid] = NULL;
Tejun Heo8bd435b2012-04-13 13:11:28 -07001136out_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -07001137 mutex_unlock(&blkcg_pol_mutex);
Vivek Goyal3e252062009-12-04 10:36:42 -05001138}
Tejun Heo3c798392012-04-16 13:57:25 -07001139EXPORT_SYMBOL_GPL(blkcg_policy_unregister);