blob: 1039fb9ff5f5f998628884dedab422c9b405a36c [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 */
Paul E. McKenneyec6c6762014-02-17 13:35:57 -0800339 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
Tejun Heoa6371202012-04-19 16:29:24 -0700340 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 Heo36c38fb2014-05-05 12:37:30 -0400454 /*
455 * XXX: We invoke cgroup_add/rm_cftypes() under blkcg_pol_mutex
456 * which ends up putting cgroup's internal cgroup_tree_mutex under
457 * it; however, cgroup_tree_mutex is nested above cgroup file
458 * active protection and grabbing blkcg_pol_mutex from a cgroup
459 * file operation creates a possible circular dependency. cgroup
460 * internal locking is planned to go through further simplification
461 * and this issue should go away soon. For now, let's trylock
462 * blkcg_pol_mutex and restart the write on failure.
463 *
464 * http://lkml.kernel.org/g/5363C04B.4010400@oracle.com
465 */
466 if (!mutex_trylock(&blkcg_pol_mutex))
467 return restart_syscall();
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700468 spin_lock_irq(&blkcg->lock);
Tejun Heo997a0262012-03-08 10:53:58 -0800469
470 /*
471 * Note that stat reset is racy - it doesn't synchronize against
472 * stat updates. This is a debug feature which shouldn't exist
473 * anyway. If you get hit by a race, retry.
474 */
Sasha Levinb67bfe02013-02-27 17:06:00 -0800475 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo8bd435b2012-04-13 13:11:28 -0700476 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -0700477 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heo549d3aa2012-03-05 13:15:16 -0800478
Tejun Heoa2b16932012-04-13 13:11:33 -0700479 if (blkcg_policy_enabled(blkg->q, pol) &&
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700480 pol->pd_reset_stats_fn)
481 pol->pd_reset_stats_fn(blkg);
Tejun Heobc0d6502012-04-13 13:11:26 -0700482 }
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700483 }
Vivek Goyalf0bdc8c2011-05-19 15:38:30 -0400484
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700485 spin_unlock_irq(&blkcg->lock);
Tejun Heobc0d6502012-04-13 13:11:26 -0700486 mutex_unlock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700487 return 0;
488}
489
Tejun Heo3c798392012-04-16 13:57:25 -0700490static const char *blkg_dev_name(struct blkcg_gq *blkg)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700491{
Tejun Heod3d32e62012-04-01 14:38:42 -0700492 /* some drivers (floppy) instantiate a queue w/o disk registered */
493 if (blkg->q->backing_dev_info.dev)
494 return dev_name(blkg->q->backing_dev_info.dev);
495 return NULL;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700496}
497
Tejun Heod3d32e62012-04-01 14:38:42 -0700498/**
499 * blkcg_print_blkgs - helper for printing per-blkg data
500 * @sf: seq_file to print to
501 * @blkcg: blkcg of interest
502 * @prfill: fill function to print out a blkg
503 * @pol: policy in question
504 * @data: data to be passed to @prfill
505 * @show_total: to print out sum of prfill return values or not
506 *
507 * This function invokes @prfill on each blkg of @blkcg if pd for the
508 * policy specified by @pol exists. @prfill is invoked with @sf, the
Tejun Heo810ecfa2013-01-09 08:05:13 -0800509 * policy data and @data and the matching queue lock held. If @show_total
510 * is %true, the sum of the return values from @prfill is printed with
511 * "Total" label at the end.
Tejun Heod3d32e62012-04-01 14:38:42 -0700512 *
513 * This is to be used to construct print functions for
514 * cftype->read_seq_string method.
515 */
Tejun Heo3c798392012-04-16 13:57:25 -0700516void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
Tejun Heof95a04a2012-04-16 13:57:26 -0700517 u64 (*prfill)(struct seq_file *,
518 struct blkg_policy_data *, int),
Tejun Heo3c798392012-04-16 13:57:25 -0700519 const struct blkcg_policy *pol, int data,
Tejun Heoec399342012-04-13 13:11:27 -0700520 bool show_total)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400521{
Tejun Heo3c798392012-04-16 13:57:25 -0700522 struct blkcg_gq *blkg;
Tejun Heod3d32e62012-04-01 14:38:42 -0700523 u64 total = 0;
524
Tejun Heo810ecfa2013-01-09 08:05:13 -0800525 rcu_read_lock();
Linus Torvaldsee89f812013-02-28 12:52:24 -0800526 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo810ecfa2013-01-09 08:05:13 -0800527 spin_lock_irq(blkg->q->queue_lock);
Tejun Heoa2b16932012-04-13 13:11:33 -0700528 if (blkcg_policy_enabled(blkg->q, pol))
Tejun Heof95a04a2012-04-16 13:57:26 -0700529 total += prfill(sf, blkg->pd[pol->plid], data);
Tejun Heo810ecfa2013-01-09 08:05:13 -0800530 spin_unlock_irq(blkg->q->queue_lock);
531 }
532 rcu_read_unlock();
Tejun Heod3d32e62012-04-01 14:38:42 -0700533
534 if (show_total)
535 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
536}
Tejun Heo829fdb52012-04-01 14:38:43 -0700537EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
Tejun Heod3d32e62012-04-01 14:38:42 -0700538
539/**
540 * __blkg_prfill_u64 - prfill helper for a single u64 value
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 * @v: value to print
544 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700545 * Print @v 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_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
Tejun Heod3d32e62012-04-01 14:38:42 -0700548{
Tejun Heof95a04a2012-04-16 13:57:26 -0700549 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700550
551 if (!dname)
552 return 0;
553
554 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
555 return v;
556}
Tejun Heo829fdb52012-04-01 14:38:43 -0700557EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
Tejun Heod3d32e62012-04-01 14:38:42 -0700558
559/**
560 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
561 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700562 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700563 * @rwstat: rwstat to print
564 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700565 * Print @rwstat to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700566 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700567u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
Tejun Heo829fdb52012-04-01 14:38:43 -0700568 const struct blkg_rwstat *rwstat)
Tejun Heod3d32e62012-04-01 14:38:42 -0700569{
570 static const char *rwstr[] = {
571 [BLKG_RWSTAT_READ] = "Read",
572 [BLKG_RWSTAT_WRITE] = "Write",
573 [BLKG_RWSTAT_SYNC] = "Sync",
574 [BLKG_RWSTAT_ASYNC] = "Async",
575 };
Tejun Heof95a04a2012-04-16 13:57:26 -0700576 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700577 u64 v;
578 int i;
579
580 if (!dname)
581 return 0;
582
583 for (i = 0; i < BLKG_RWSTAT_NR; i++)
584 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
585 (unsigned long long)rwstat->cnt[i]);
586
587 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
588 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
589 return v;
590}
Tejun Heob50da392013-01-09 08:05:12 -0800591EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700592
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700593/**
594 * blkg_prfill_stat - prfill callback for blkg_stat
595 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700596 * @pd: policy private data of interest
597 * @off: offset to the blkg_stat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700598 *
599 * prfill callback for printing a blkg_stat.
600 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700601u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700602{
Tejun Heof95a04a2012-04-16 13:57:26 -0700603 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
Tejun Heod3d32e62012-04-01 14:38:42 -0700604}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700605EXPORT_SYMBOL_GPL(blkg_prfill_stat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700606
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700607/**
608 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
609 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700610 * @pd: policy private data of interest
611 * @off: offset to the blkg_rwstat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700612 *
613 * prfill callback for printing a blkg_rwstat.
614 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700615u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
616 int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700617{
Tejun Heof95a04a2012-04-16 13:57:26 -0700618 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
Tejun Heod3d32e62012-04-01 14:38:42 -0700619
Tejun Heof95a04a2012-04-16 13:57:26 -0700620 return __blkg_prfill_rwstat(sf, pd, &rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700621}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700622EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700623
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700624/**
Tejun Heo16b3de62013-01-09 08:05:12 -0800625 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
626 * @pd: policy private data of interest
627 * @off: offset to the blkg_stat in @pd
628 *
629 * Collect the blkg_stat specified by @off from @pd and all its online
630 * descendants and return the sum. The caller must be holding the queue
631 * lock for online tests.
632 */
633u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off)
634{
635 struct blkcg_policy *pol = blkcg_policy[pd->plid];
636 struct blkcg_gq *pos_blkg;
Tejun Heo492eb212013-08-08 20:11:25 -0400637 struct cgroup_subsys_state *pos_css;
Tejun Heobd8815a2013-08-08 20:11:27 -0400638 u64 sum = 0;
Tejun Heo16b3de62013-01-09 08:05:12 -0800639
640 lockdep_assert_held(pd->blkg->q->queue_lock);
641
Tejun Heo16b3de62013-01-09 08:05:12 -0800642 rcu_read_lock();
Tejun Heo492eb212013-08-08 20:11:25 -0400643 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
Tejun Heo16b3de62013-01-09 08:05:12 -0800644 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
645 struct blkg_stat *stat = (void *)pos_pd + off;
646
647 if (pos_blkg->online)
648 sum += blkg_stat_read(stat);
649 }
650 rcu_read_unlock();
651
652 return sum;
653}
654EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
655
656/**
657 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
658 * @pd: policy private data of interest
659 * @off: offset to the blkg_stat in @pd
660 *
661 * Collect the blkg_rwstat specified by @off from @pd and all its online
662 * descendants and return the sum. The caller must be holding the queue
663 * lock for online tests.
664 */
665struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
666 int off)
667{
668 struct blkcg_policy *pol = blkcg_policy[pd->plid];
669 struct blkcg_gq *pos_blkg;
Tejun Heo492eb212013-08-08 20:11:25 -0400670 struct cgroup_subsys_state *pos_css;
Tejun Heobd8815a2013-08-08 20:11:27 -0400671 struct blkg_rwstat sum = { };
Tejun Heo16b3de62013-01-09 08:05:12 -0800672 int i;
673
674 lockdep_assert_held(pd->blkg->q->queue_lock);
675
Tejun Heo16b3de62013-01-09 08:05:12 -0800676 rcu_read_lock();
Tejun Heo492eb212013-08-08 20:11:25 -0400677 blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) {
Tejun Heo16b3de62013-01-09 08:05:12 -0800678 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
679 struct blkg_rwstat *rwstat = (void *)pos_pd + off;
680 struct blkg_rwstat tmp;
681
682 if (!pos_blkg->online)
683 continue;
684
685 tmp = blkg_rwstat_read(rwstat);
686
687 for (i = 0; i < BLKG_RWSTAT_NR; i++)
688 sum.cnt[i] += tmp.cnt[i];
689 }
690 rcu_read_unlock();
691
692 return sum;
693}
694EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
695
696/**
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700697 * blkg_conf_prep - parse and prepare for per-blkg config update
698 * @blkcg: target block cgroup
Tejun Heoda8b0662012-04-13 13:11:29 -0700699 * @pol: target policy
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700700 * @input: input string
701 * @ctx: blkg_conf_ctx to be filled
702 *
703 * Parse per-blkg config update from @input and initialize @ctx with the
704 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
Tejun Heoda8b0662012-04-13 13:11:29 -0700705 * value. This function returns with RCU read lock and queue lock held and
706 * must be paired with blkg_conf_finish().
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700707 */
Tejun Heo3c798392012-04-16 13:57:25 -0700708int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
709 const char *input, struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700710 __acquires(rcu) __acquires(disk->queue->queue_lock)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800711{
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700712 struct gendisk *disk;
Tejun Heo3c798392012-04-16 13:57:25 -0700713 struct blkcg_gq *blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700714 unsigned int major, minor;
715 unsigned long long v;
716 int part, ret;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800717
Tejun Heo726fa692012-04-01 14:38:43 -0700718 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
719 return -EINVAL;
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700720
Tejun Heo726fa692012-04-01 14:38:43 -0700721 disk = get_gendisk(MKDEV(major, minor), &part);
Tejun Heo4bfd4822012-03-05 13:15:08 -0800722 if (!disk || part)
Tejun Heo726fa692012-04-01 14:38:43 -0700723 return -EINVAL;
Tejun Heoe56da7e2012-03-05 13:15:07 -0800724
725 rcu_read_lock();
Tejun Heo4bfd4822012-03-05 13:15:08 -0800726 spin_lock_irq(disk->queue->queue_lock);
Tejun Heoda8b0662012-04-13 13:11:29 -0700727
Tejun Heoa2b16932012-04-13 13:11:33 -0700728 if (blkcg_policy_enabled(disk->queue, pol))
Tejun Heo3c96cb32012-04-13 13:11:34 -0700729 blkg = blkg_lookup_create(blkcg, disk->queue);
Tejun Heoa2b16932012-04-13 13:11:33 -0700730 else
731 blkg = ERR_PTR(-EINVAL);
Tejun Heoe56da7e2012-03-05 13:15:07 -0800732
Tejun Heo4bfd4822012-03-05 13:15:08 -0800733 if (IS_ERR(blkg)) {
734 ret = PTR_ERR(blkg);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700735 rcu_read_unlock();
Tejun Heoda8b0662012-04-13 13:11:29 -0700736 spin_unlock_irq(disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700737 put_disk(disk);
738 /*
739 * If queue was bypassing, we should retry. Do so after a
740 * short msleep(). It isn't strictly necessary but queue
741 * can be bypassing for some time and it's always nice to
742 * avoid busy looping.
743 */
744 if (ret == -EBUSY) {
745 msleep(10);
746 ret = restart_syscall();
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400747 }
Tejun Heo726fa692012-04-01 14:38:43 -0700748 return ret;
Vivek Goyal062a6442010-09-15 17:06:33 -0400749 }
Tejun Heoe56da7e2012-03-05 13:15:07 -0800750
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700751 ctx->disk = disk;
752 ctx->blkg = blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700753 ctx->v = v;
754 return 0;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800755}
Tejun Heo829fdb52012-04-01 14:38:43 -0700756EXPORT_SYMBOL_GPL(blkg_conf_prep);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800757
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700758/**
759 * blkg_conf_finish - finish up per-blkg config update
760 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
761 *
762 * Finish up after per-blkg config update. This function must be paired
763 * with blkg_conf_prep().
764 */
Tejun Heo829fdb52012-04-01 14:38:43 -0700765void blkg_conf_finish(struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700766 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800767{
Tejun Heoda8b0662012-04-13 13:11:29 -0700768 spin_unlock_irq(ctx->disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700769 rcu_read_unlock();
770 put_disk(ctx->disk);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800771}
Tejun Heo829fdb52012-04-01 14:38:43 -0700772EXPORT_SYMBOL_GPL(blkg_conf_finish);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800773
Tejun Heo3c798392012-04-16 13:57:25 -0700774struct cftype blkcg_files[] = {
Vivek Goyal31e4c282009-12-03 12:59:42 -0500775 {
Divyesh Shah84c124d2010-04-09 08:31:19 +0200776 .name = "reset_stats",
Tejun Heo3c798392012-04-16 13:57:25 -0700777 .write_u64 = blkcg_reset_stats,
Vivek Goyal22084192009-12-03 12:59:49 -0500778 },
Tejun Heo4baf6e32012-04-01 12:09:55 -0700779 { } /* terminate */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500780};
781
Tejun Heo9f13ef62012-03-05 13:15:21 -0800782/**
Tejun Heo92fb9742012-11-19 08:13:38 -0800783 * blkcg_css_offline - cgroup css_offline callback
Tejun Heoeb954192013-08-08 20:11:23 -0400784 * @css: css of interest
Tejun Heo9f13ef62012-03-05 13:15:21 -0800785 *
Tejun Heoeb954192013-08-08 20:11:23 -0400786 * This function is called when @css is about to go away and responsible
787 * for shooting down all blkgs associated with @css. blkgs should be
Tejun Heo9f13ef62012-03-05 13:15:21 -0800788 * removed while holding both q and blkcg locks. As blkcg lock is nested
789 * inside q lock, this function performs reverse double lock dancing.
790 *
791 * This is the blkcg counterpart of ioc_release_fn().
792 */
Tejun Heoeb954192013-08-08 20:11:23 -0400793static void blkcg_css_offline(struct cgroup_subsys_state *css)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500794{
Tejun Heoeb954192013-08-08 20:11:23 -0400795 struct blkcg *blkcg = css_to_blkcg(css);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500796
Tejun Heo9f13ef62012-03-05 13:15:21 -0800797 spin_lock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800798
Tejun Heo9f13ef62012-03-05 13:15:21 -0800799 while (!hlist_empty(&blkcg->blkg_list)) {
Tejun Heo3c798392012-04-16 13:57:25 -0700800 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
801 struct blkcg_gq, blkcg_node);
Tejun Heoc875f4d2012-03-05 13:15:22 -0800802 struct request_queue *q = blkg->q;
Vivek Goyalb1c35762009-12-03 12:59:47 -0500803
Tejun Heo9f13ef62012-03-05 13:15:21 -0800804 if (spin_trylock(q->queue_lock)) {
805 blkg_destroy(blkg);
806 spin_unlock(q->queue_lock);
807 } else {
808 spin_unlock_irq(&blkcg->lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800809 cpu_relax();
Dan Carpentera5567932012-03-29 20:57:08 +0200810 spin_lock_irq(&blkcg->lock);
Jens Axboe0f3942a2010-05-03 14:28:55 +0200811 }
Tejun Heo9f13ef62012-03-05 13:15:21 -0800812 }
Jens Axboe0f3942a2010-05-03 14:28:55 +0200813
Tejun Heo9f13ef62012-03-05 13:15:21 -0800814 spin_unlock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800815}
816
Tejun Heoeb954192013-08-08 20:11:23 -0400817static void blkcg_css_free(struct cgroup_subsys_state *css)
Tejun Heo7ee9c562012-03-05 13:15:11 -0800818{
Tejun Heoeb954192013-08-08 20:11:23 -0400819 struct blkcg *blkcg = css_to_blkcg(css);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800820
Tejun Heo3c798392012-04-16 13:57:25 -0700821 if (blkcg != &blkcg_root)
Ben Blum67523c42010-03-10 15:22:11 -0800822 kfree(blkcg);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500823}
824
Tejun Heoeb954192013-08-08 20:11:23 -0400825static struct cgroup_subsys_state *
826blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500827{
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700828 static atomic64_t id_seq = ATOMIC64_INIT(0);
Tejun Heo3c798392012-04-16 13:57:25 -0700829 struct blkcg *blkcg;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500830
Tejun Heoeb954192013-08-08 20:11:23 -0400831 if (!parent_css) {
Tejun Heo3c798392012-04-16 13:57:25 -0700832 blkcg = &blkcg_root;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500833 goto done;
834 }
835
Vivek Goyal31e4c282009-12-03 12:59:42 -0500836 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
837 if (!blkcg)
838 return ERR_PTR(-ENOMEM);
839
Tejun Heo3381cb82012-04-01 14:38:44 -0700840 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
Tejun Heoe71357e2013-01-09 08:05:10 -0800841 blkcg->cfq_leaf_weight = CFQ_WEIGHT_DEFAULT;
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700842 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500843done:
844 spin_lock_init(&blkcg->lock);
Tejun Heoa6371202012-04-19 16:29:24 -0700845 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500846 INIT_HLIST_HEAD(&blkcg->blkg_list);
847
848 return &blkcg->css;
849}
850
Tejun Heo5efd6112012-03-05 13:15:12 -0800851/**
852 * blkcg_init_queue - initialize blkcg part of request queue
853 * @q: request_queue to initialize
854 *
855 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
856 * part of new request_queue @q.
857 *
858 * RETURNS:
859 * 0 on success, -errno on failure.
860 */
861int blkcg_init_queue(struct request_queue *q)
862{
863 might_sleep();
864
Tejun Heo3c96cb32012-04-13 13:11:34 -0700865 return blk_throtl_init(q);
Tejun Heo5efd6112012-03-05 13:15:12 -0800866}
867
868/**
869 * blkcg_drain_queue - drain blkcg part of request_queue
870 * @q: request_queue to drain
871 *
872 * Called from blk_drain_queue(). Responsible for draining blkcg part.
873 */
874void blkcg_drain_queue(struct request_queue *q)
875{
876 lockdep_assert_held(q->queue_lock);
877
878 blk_throtl_drain(q);
879}
880
881/**
882 * blkcg_exit_queue - exit and release blkcg part of request_queue
883 * @q: request_queue being released
884 *
885 * Called from blk_release_queue(). Responsible for exiting blkcg part.
886 */
887void blkcg_exit_queue(struct request_queue *q)
888{
Tejun Heo6d18b002012-04-13 13:11:35 -0700889 spin_lock_irq(q->queue_lock);
Tejun Heo3c96cb32012-04-13 13:11:34 -0700890 blkg_destroy_all(q);
Tejun Heo6d18b002012-04-13 13:11:35 -0700891 spin_unlock_irq(q->queue_lock);
892
Tejun Heo5efd6112012-03-05 13:15:12 -0800893 blk_throtl_exit(q);
894}
895
Vivek Goyal31e4c282009-12-03 12:59:42 -0500896/*
897 * We cannot support shared io contexts, as we have no mean to support
898 * two tasks with the same ioc in two different groups without major rework
899 * of the main cic data structures. For now we allow a task to change
900 * its cgroup only if it's the only owner of its ioc.
901 */
Tejun Heoeb954192013-08-08 20:11:23 -0400902static int blkcg_can_attach(struct cgroup_subsys_state *css,
903 struct cgroup_taskset *tset)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500904{
Tejun Heobb9d97b2011-12-12 18:12:21 -0800905 struct task_struct *task;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500906 struct io_context *ioc;
907 int ret = 0;
908
909 /* task_lock() is needed to avoid races with exit_io_context() */
Tejun Heo924f0d92014-02-13 06:58:41 -0500910 cgroup_taskset_for_each(task, tset) {
Tejun Heobb9d97b2011-12-12 18:12:21 -0800911 task_lock(task);
912 ioc = task->io_context;
913 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
914 ret = -EINVAL;
915 task_unlock(task);
916 if (ret)
917 break;
918 }
Vivek Goyal31e4c282009-12-03 12:59:42 -0500919 return ret;
920}
921
Tejun Heo073219e2014-02-08 10:36:58 -0500922struct cgroup_subsys blkio_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -0800923 .css_alloc = blkcg_css_alloc,
924 .css_offline = blkcg_css_offline,
925 .css_free = blkcg_css_free,
Tejun Heo3c798392012-04-16 13:57:25 -0700926 .can_attach = blkcg_can_attach,
Tejun Heo3c798392012-04-16 13:57:25 -0700927 .base_cftypes = blkcg_files,
Tejun Heo676f7c82012-04-01 12:09:55 -0700928};
Tejun Heo073219e2014-02-08 10:36:58 -0500929EXPORT_SYMBOL_GPL(blkio_cgrp_subsys);
Tejun Heo676f7c82012-04-01 12:09:55 -0700930
Tejun Heo8bd435b2012-04-13 13:11:28 -0700931/**
Tejun Heoa2b16932012-04-13 13:11:33 -0700932 * blkcg_activate_policy - activate a blkcg policy on a request_queue
933 * @q: request_queue of interest
934 * @pol: blkcg policy to activate
935 *
936 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
937 * bypass mode to populate its blkgs with policy_data for @pol.
938 *
939 * Activation happens with @q bypassed, so nobody would be accessing blkgs
940 * from IO path. Update of each blkg is protected by both queue and blkcg
941 * locks so that holding either lock and testing blkcg_policy_enabled() is
942 * always enough for dereferencing policy data.
943 *
944 * The caller is responsible for synchronizing [de]activations and policy
945 * [un]registerations. Returns 0 on success, -errno on failure.
946 */
947int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700948 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -0700949{
950 LIST_HEAD(pds);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800951 struct blkcg_gq *blkg, *new_blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -0700952 struct blkg_policy_data *pd, *n;
953 int cnt = 0, ret;
Tejun Heo15974992012-06-04 20:40:52 -0700954 bool preloaded;
Tejun Heoa2b16932012-04-13 13:11:33 -0700955
956 if (blkcg_policy_enabled(q, pol))
957 return 0;
958
Tejun Heo15974992012-06-04 20:40:52 -0700959 /* preallocations for root blkg */
Tejun Heo86cde6b2013-01-09 08:05:10 -0800960 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
961 if (!new_blkg)
Tejun Heo15974992012-06-04 20:40:52 -0700962 return -ENOMEM;
963
Tejun Heoa2b16932012-04-13 13:11:33 -0700964 blk_queue_bypass_start(q);
965
Jun'ichi Nomurae5072662013-04-09 15:01:21 +0200966 preloaded = !radix_tree_preload(GFP_KERNEL);
967
Tejun Heo86cde6b2013-01-09 08:05:10 -0800968 /*
969 * Make sure the root blkg exists and count the existing blkgs. As
970 * @q is bypassing at this point, blkg_lookup_create() can't be
971 * used. Open code it.
972 */
Tejun Heoa2b16932012-04-13 13:11:33 -0700973 spin_lock_irq(q->queue_lock);
974
975 rcu_read_lock();
Tejun Heo86cde6b2013-01-09 08:05:10 -0800976 blkg = __blkg_lookup(&blkcg_root, q, false);
977 if (blkg)
978 blkg_free(new_blkg);
979 else
980 blkg = blkg_create(&blkcg_root, q, new_blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -0700981 rcu_read_unlock();
982
Tejun Heo15974992012-06-04 20:40:52 -0700983 if (preloaded)
984 radix_tree_preload_end();
985
Tejun Heoa2b16932012-04-13 13:11:33 -0700986 if (IS_ERR(blkg)) {
987 ret = PTR_ERR(blkg);
988 goto out_unlock;
989 }
Tejun Heoa2b16932012-04-13 13:11:33 -0700990
991 list_for_each_entry(blkg, &q->blkg_list, q_node)
992 cnt++;
993
994 spin_unlock_irq(q->queue_lock);
995
996 /* allocate policy_data for all existing blkgs */
997 while (cnt--) {
Tejun Heof95a04a2012-04-16 13:57:26 -0700998 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
Tejun Heoa2b16932012-04-13 13:11:33 -0700999 if (!pd) {
1000 ret = -ENOMEM;
1001 goto out_free;
1002 }
1003 list_add_tail(&pd->alloc_node, &pds);
1004 }
1005
1006 /*
1007 * Install the allocated pds. With @q bypassing, no new blkg
1008 * should have been created while the queue lock was dropped.
1009 */
1010 spin_lock_irq(q->queue_lock);
1011
1012 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1013 if (WARN_ON(list_empty(&pds))) {
1014 /* umm... this shouldn't happen, just abort */
1015 ret = -ENOMEM;
1016 goto out_unlock;
1017 }
1018 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
1019 list_del_init(&pd->alloc_node);
1020
1021 /* grab blkcg lock too while installing @pd on @blkg */
1022 spin_lock(&blkg->blkcg->lock);
1023
1024 blkg->pd[pol->plid] = pd;
1025 pd->blkg = blkg;
Tejun Heob276a872013-01-09 08:05:12 -08001026 pd->plid = pol->plid;
Tejun Heof9fcc2d2012-04-16 13:57:27 -07001027 pol->pd_init_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -07001028
1029 spin_unlock(&blkg->blkcg->lock);
1030 }
1031
1032 __set_bit(pol->plid, q->blkcg_pols);
1033 ret = 0;
1034out_unlock:
1035 spin_unlock_irq(q->queue_lock);
1036out_free:
1037 blk_queue_bypass_end(q);
1038 list_for_each_entry_safe(pd, n, &pds, alloc_node)
1039 kfree(pd);
1040 return ret;
1041}
1042EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1043
1044/**
1045 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1046 * @q: request_queue of interest
1047 * @pol: blkcg policy to deactivate
1048 *
1049 * Deactivate @pol on @q. Follows the same synchronization rules as
1050 * blkcg_activate_policy().
1051 */
1052void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -07001053 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -07001054{
Tejun Heo3c798392012-04-16 13:57:25 -07001055 struct blkcg_gq *blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -07001056
1057 if (!blkcg_policy_enabled(q, pol))
1058 return;
1059
1060 blk_queue_bypass_start(q);
1061 spin_lock_irq(q->queue_lock);
1062
1063 __clear_bit(pol->plid, q->blkcg_pols);
1064
Tejun Heo6d18b002012-04-13 13:11:35 -07001065 /* if no policy is left, no need for blkgs - shoot them down */
1066 if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
1067 blkg_destroy_all(q);
1068
Tejun Heoa2b16932012-04-13 13:11:33 -07001069 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1070 /* grab blkcg lock too while removing @pd from @blkg */
1071 spin_lock(&blkg->blkcg->lock);
1072
Tejun Heof427d902013-01-09 08:05:12 -08001073 if (pol->pd_offline_fn)
1074 pol->pd_offline_fn(blkg);
Tejun Heof9fcc2d2012-04-16 13:57:27 -07001075 if (pol->pd_exit_fn)
1076 pol->pd_exit_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -07001077
1078 kfree(blkg->pd[pol->plid]);
1079 blkg->pd[pol->plid] = NULL;
1080
1081 spin_unlock(&blkg->blkcg->lock);
1082 }
1083
1084 spin_unlock_irq(q->queue_lock);
1085 blk_queue_bypass_end(q);
1086}
1087EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1088
1089/**
Tejun Heo3c798392012-04-16 13:57:25 -07001090 * blkcg_policy_register - register a blkcg policy
1091 * @pol: blkcg policy to register
Tejun Heo8bd435b2012-04-13 13:11:28 -07001092 *
Tejun Heo3c798392012-04-16 13:57:25 -07001093 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1094 * successful registration. Returns 0 on success and -errno on failure.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001095 */
Tejun Heo3c798392012-04-16 13:57:25 -07001096int blkcg_policy_register(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001097{
Tejun Heo8bd435b2012-04-13 13:11:28 -07001098 int i, ret;
Tejun Heoe8989fa2012-03-05 13:15:20 -08001099
Tejun Heof95a04a2012-04-16 13:57:26 -07001100 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
1101 return -EINVAL;
1102
Tejun Heobc0d6502012-04-13 13:11:26 -07001103 mutex_lock(&blkcg_pol_mutex);
1104
Tejun Heo8bd435b2012-04-13 13:11:28 -07001105 /* find an empty slot */
1106 ret = -ENOSPC;
1107 for (i = 0; i < BLKCG_MAX_POLS; i++)
Tejun Heo3c798392012-04-16 13:57:25 -07001108 if (!blkcg_policy[i])
Tejun Heo8bd435b2012-04-13 13:11:28 -07001109 break;
1110 if (i >= BLKCG_MAX_POLS)
1111 goto out_unlock;
Tejun Heo035d10b2012-03-05 13:15:04 -08001112
Tejun Heo8bd435b2012-04-13 13:11:28 -07001113 /* register and update blkgs */
Tejun Heo3c798392012-04-16 13:57:25 -07001114 pol->plid = i;
1115 blkcg_policy[i] = pol;
Tejun Heo8bd435b2012-04-13 13:11:28 -07001116
Tejun Heo8bd435b2012-04-13 13:11:28 -07001117 /* everything is in place, add intf files for the new policy */
Tejun Heo3c798392012-04-16 13:57:25 -07001118 if (pol->cftypes)
Tejun Heo073219e2014-02-08 10:36:58 -05001119 WARN_ON(cgroup_add_cftypes(&blkio_cgrp_subsys, pol->cftypes));
Tejun Heo8bd435b2012-04-13 13:11:28 -07001120 ret = 0;
1121out_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -07001122 mutex_unlock(&blkcg_pol_mutex);
Tejun Heo8bd435b2012-04-13 13:11:28 -07001123 return ret;
Vivek Goyal3e252062009-12-04 10:36:42 -05001124}
Tejun Heo3c798392012-04-16 13:57:25 -07001125EXPORT_SYMBOL_GPL(blkcg_policy_register);
Vivek Goyal3e252062009-12-04 10:36:42 -05001126
Tejun Heo8bd435b2012-04-13 13:11:28 -07001127/**
Tejun Heo3c798392012-04-16 13:57:25 -07001128 * blkcg_policy_unregister - unregister a blkcg policy
1129 * @pol: blkcg policy to unregister
Tejun Heo8bd435b2012-04-13 13:11:28 -07001130 *
Tejun Heo3c798392012-04-16 13:57:25 -07001131 * Undo blkcg_policy_register(@pol). Might sleep.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001132 */
Tejun Heo3c798392012-04-16 13:57:25 -07001133void blkcg_policy_unregister(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001134{
Tejun Heobc0d6502012-04-13 13:11:26 -07001135 mutex_lock(&blkcg_pol_mutex);
1136
Tejun Heo3c798392012-04-16 13:57:25 -07001137 if (WARN_ON(blkcg_policy[pol->plid] != pol))
Tejun Heo8bd435b2012-04-13 13:11:28 -07001138 goto out_unlock;
1139
1140 /* kill the intf files first */
Tejun Heo3c798392012-04-16 13:57:25 -07001141 if (pol->cftypes)
Tejun Heo2bb566c2013-08-08 20:11:23 -04001142 cgroup_rm_cftypes(pol->cftypes);
Tejun Heo44ea53d2012-04-01 14:38:43 -07001143
Tejun Heo8bd435b2012-04-13 13:11:28 -07001144 /* unregister and update blkgs */
Tejun Heo3c798392012-04-16 13:57:25 -07001145 blkcg_policy[pol->plid] = NULL;
Tejun Heo8bd435b2012-04-13 13:11:28 -07001146out_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -07001147 mutex_unlock(&blkcg_pol_mutex);
Vivek Goyal3e252062009-12-04 10:36:42 -05001148}
Tejun Heo3c798392012-04-16 13:57:25 -07001149EXPORT_SYMBOL_GPL(blkcg_policy_unregister);