blob: b8858fb0cafa34e491f6ad0e719a16ba3cd8d015 [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 Heo3c798392012-04-16 13:57:25 -070029struct blkcg blkcg_root = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT };
30EXPORT_SYMBOL_GPL(blkcg_root);
Vivek Goyal9d6a9862009-12-04 10:36:41 -050031
Tejun Heo3c798392012-04-16 13:57:25 -070032static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
Tejun Heo035d10b2012-03-05 13:15:04 -080033
Tejun Heoa2b16932012-04-13 13:11:33 -070034static bool blkcg_policy_enabled(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -070035 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -070036{
37 return pol && test_bit(pol->plid, q->blkcg_pols);
38}
39
Tejun Heo03814112012-03-05 13:15:14 -080040/**
41 * blkg_free - free a blkg
42 * @blkg: blkg to free
43 *
44 * Free @blkg which may be partially allocated.
45 */
Tejun Heo3c798392012-04-16 13:57:25 -070046static void blkg_free(struct blkcg_gq *blkg)
Tejun Heo03814112012-03-05 13:15:14 -080047{
Tejun Heoe8989fa2012-03-05 13:15:20 -080048 int i;
Tejun Heo549d3aa2012-03-05 13:15:16 -080049
50 if (!blkg)
51 return;
52
Tejun Heo8bd435b2012-04-13 13:11:28 -070053 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -070054 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heoe8989fa2012-03-05 13:15:20 -080055 struct blkg_policy_data *pd = blkg->pd[i];
56
Tejun Heo9ade5ea2012-04-01 14:38:44 -070057 if (!pd)
58 continue;
59
Tejun Heof9fcc2d2012-04-16 13:57:27 -070060 if (pol && pol->pd_exit_fn)
61 pol->pd_exit_fn(blkg);
Tejun Heo9ade5ea2012-04-01 14:38:44 -070062
Tejun Heo9ade5ea2012-04-01 14:38:44 -070063 kfree(pd);
Tejun Heo03814112012-03-05 13:15:14 -080064 }
Tejun Heoe8989fa2012-03-05 13:15:20 -080065
Tejun Heoa0516612012-06-26 15:05:44 -070066 blk_exit_rl(&blkg->rl);
Tejun Heo549d3aa2012-03-05 13:15:16 -080067 kfree(blkg);
Tejun Heo03814112012-03-05 13:15:14 -080068}
69
70/**
71 * blkg_alloc - allocate a blkg
72 * @blkcg: block cgroup the new blkg is associated with
73 * @q: request_queue the new blkg is associated with
Tejun Heo15974992012-06-04 20:40:52 -070074 * @gfp_mask: allocation mask to use
Tejun Heo03814112012-03-05 13:15:14 -080075 *
Tejun Heoe8989fa2012-03-05 13:15:20 -080076 * Allocate a new blkg assocating @blkcg and @q.
Tejun Heo03814112012-03-05 13:15:14 -080077 */
Tejun Heo15974992012-06-04 20:40:52 -070078static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
79 gfp_t gfp_mask)
Tejun Heo03814112012-03-05 13:15:14 -080080{
Tejun Heo3c798392012-04-16 13:57:25 -070081 struct blkcg_gq *blkg;
Tejun Heoe8989fa2012-03-05 13:15:20 -080082 int i;
Tejun Heo03814112012-03-05 13:15:14 -080083
84 /* alloc and init base part */
Tejun Heo15974992012-06-04 20:40:52 -070085 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
Tejun Heo03814112012-03-05 13:15:14 -080086 if (!blkg)
87 return NULL;
88
Tejun Heoc875f4d2012-03-05 13:15:22 -080089 blkg->q = q;
Tejun Heoe8989fa2012-03-05 13:15:20 -080090 INIT_LIST_HEAD(&blkg->q_node);
Tejun Heo03814112012-03-05 13:15:14 -080091 blkg->blkcg = blkcg;
Tejun Heo1adaf3d2012-03-05 13:15:15 -080092 blkg->refcnt = 1;
Tejun Heo03814112012-03-05 13:15:14 -080093
Tejun Heoa0516612012-06-26 15:05:44 -070094 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
95 if (blkcg != &blkcg_root) {
96 if (blk_init_rl(&blkg->rl, q, gfp_mask))
97 goto err_free;
98 blkg->rl.blkg = blkg;
99 }
100
Tejun Heo8bd435b2012-04-13 13:11:28 -0700101 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -0700102 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heoe8989fa2012-03-05 13:15:20 -0800103 struct blkg_policy_data *pd;
Tejun Heo03814112012-03-05 13:15:14 -0800104
Tejun Heoa2b16932012-04-13 13:11:33 -0700105 if (!blkcg_policy_enabled(q, pol))
Tejun Heoe8989fa2012-03-05 13:15:20 -0800106 continue;
Tejun Heo549d3aa2012-03-05 13:15:16 -0800107
Tejun Heoe8989fa2012-03-05 13:15:20 -0800108 /* alloc per-policy data and attach it to blkg */
Tejun Heo15974992012-06-04 20:40:52 -0700109 pd = kzalloc_node(pol->pd_size, gfp_mask, q->node);
Tejun Heoa0516612012-06-26 15:05:44 -0700110 if (!pd)
111 goto err_free;
Tejun Heo549d3aa2012-03-05 13:15:16 -0800112
Tejun Heoe8989fa2012-03-05 13:15:20 -0800113 blkg->pd[i] = pd;
114 pd->blkg = blkg;
Tejun Heo03814112012-03-05 13:15:14 -0800115
Tejun Heo9b2ea862012-06-04 15:21:00 +0900116 /* invoke per-policy init */
Tejun Heoa2b16932012-04-13 13:11:33 -0700117 if (blkcg_policy_enabled(blkg->q, pol))
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700118 pol->pd_init_fn(blkg);
Tejun Heoe8989fa2012-03-05 13:15:20 -0800119 }
120
Tejun Heo03814112012-03-05 13:15:14 -0800121 return blkg;
Tejun Heoa0516612012-06-26 15:05:44 -0700122
123err_free:
124 blkg_free(blkg);
125 return NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800126}
127
Tejun Heo3c798392012-04-16 13:57:25 -0700128static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
129 struct request_queue *q)
Tejun Heo80fd9972012-04-13 14:50:53 -0700130{
Tejun Heo3c798392012-04-16 13:57:25 -0700131 struct blkcg_gq *blkg;
Tejun Heo80fd9972012-04-13 14:50:53 -0700132
Tejun Heoa6371202012-04-19 16:29:24 -0700133 blkg = rcu_dereference(blkcg->blkg_hint);
134 if (blkg && blkg->q == q)
135 return blkg;
136
137 /*
138 * Hint didn't match. Look up from the radix tree. Note that we
139 * may not be holding queue_lock and thus are not sure whether
140 * @blkg from blkg_tree has already been removed or not, so we
141 * can't update hint to the lookup result. Leave it to the caller.
142 */
143 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
144 if (blkg && blkg->q == q)
145 return blkg;
146
Tejun Heo80fd9972012-04-13 14:50:53 -0700147 return NULL;
148}
149
150/**
151 * blkg_lookup - lookup blkg for the specified blkcg - q pair
152 * @blkcg: blkcg of interest
153 * @q: request_queue of interest
154 *
155 * Lookup blkg for the @blkcg - @q pair. This function should be called
156 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
157 * - see blk_queue_bypass_start() for details.
158 */
Tejun Heo3c798392012-04-16 13:57:25 -0700159struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
Tejun Heo80fd9972012-04-13 14:50:53 -0700160{
161 WARN_ON_ONCE(!rcu_read_lock_held());
162
163 if (unlikely(blk_queue_bypass(q)))
164 return NULL;
165 return __blkg_lookup(blkcg, q);
166}
167EXPORT_SYMBOL_GPL(blkg_lookup);
168
Tejun Heo15974992012-06-04 20:40:52 -0700169/*
170 * If @new_blkg is %NULL, this function tries to allocate a new one as
171 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
172 */
Tejun Heo3c798392012-04-16 13:57:25 -0700173static struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg,
Tejun Heo15974992012-06-04 20:40:52 -0700174 struct request_queue *q,
175 struct blkcg_gq *new_blkg)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400176{
Tejun Heo3c798392012-04-16 13:57:25 -0700177 struct blkcg_gq *blkg;
Tejun Heo496fb782012-04-19 16:29:23 -0700178 int ret;
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400179
Tejun Heocd1604f2012-03-05 13:15:06 -0800180 WARN_ON_ONCE(!rcu_read_lock_held());
181 lockdep_assert_held(q->queue_lock);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500182
Tejun Heoa6371202012-04-19 16:29:24 -0700183 /* lookup and update hint on success, see __blkg_lookup() for details */
Tejun Heo80fd9972012-04-13 14:50:53 -0700184 blkg = __blkg_lookup(blkcg, q);
Tejun Heoa6371202012-04-19 16:29:24 -0700185 if (blkg) {
186 rcu_assign_pointer(blkcg->blkg_hint, blkg);
Tejun Heo15974992012-06-04 20:40:52 -0700187 goto out_free;
Tejun Heoa6371202012-04-19 16:29:24 -0700188 }
Tejun Heocd1604f2012-03-05 13:15:06 -0800189
Tejun Heo7ee9c562012-03-05 13:15:11 -0800190 /* blkg holds a reference to blkcg */
Tejun Heo15974992012-06-04 20:40:52 -0700191 if (!css_tryget(&blkcg->css)) {
192 blkg = ERR_PTR(-EINVAL);
193 goto out_free;
194 }
Tejun Heocd1604f2012-03-05 13:15:06 -0800195
Tejun Heo496fb782012-04-19 16:29:23 -0700196 /* allocate */
Tejun Heo15974992012-06-04 20:40:52 -0700197 if (!new_blkg) {
198 new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
199 if (unlikely(!new_blkg)) {
200 blkg = ERR_PTR(-ENOMEM);
201 goto out_put;
202 }
203 }
204 blkg = new_blkg;
Tejun Heocd1604f2012-03-05 13:15:06 -0800205
206 /* insert */
Tejun Heoa6371202012-04-19 16:29:24 -0700207 spin_lock(&blkcg->lock);
208 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
209 if (likely(!ret)) {
210 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
211 list_add(&blkg->q_node, &q->blkg_list);
212 }
213 spin_unlock(&blkcg->lock);
214
Tejun Heoa6371202012-04-19 16:29:24 -0700215 if (!ret)
216 return blkg;
Tejun Heo15974992012-06-04 20:40:52 -0700217
218 blkg = ERR_PTR(ret);
219out_put:
Tejun Heo496fb782012-04-19 16:29:23 -0700220 css_put(&blkcg->css);
Tejun Heo15974992012-06-04 20:40:52 -0700221out_free:
222 blkg_free(new_blkg);
223 return blkg;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500224}
Tejun Heo3c96cb32012-04-13 13:11:34 -0700225
Tejun Heo3c798392012-04-16 13:57:25 -0700226struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
227 struct request_queue *q)
Tejun Heo3c96cb32012-04-13 13:11:34 -0700228{
229 /*
230 * This could be the first entry point of blkcg implementation and
231 * we shouldn't allow anything to go through for a bypassing queue.
232 */
233 if (unlikely(blk_queue_bypass(q)))
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100234 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
Tejun Heo15974992012-06-04 20:40:52 -0700235 return __blkg_lookup_create(blkcg, q, NULL);
Tejun Heo3c96cb32012-04-13 13:11:34 -0700236}
Tejun Heocd1604f2012-03-05 13:15:06 -0800237EXPORT_SYMBOL_GPL(blkg_lookup_create);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500238
Tejun Heo3c798392012-04-16 13:57:25 -0700239static void blkg_destroy(struct blkcg_gq *blkg)
Tejun Heo72e06c22012-03-05 13:15:00 -0800240{
Tejun Heo3c798392012-04-16 13:57:25 -0700241 struct blkcg *blkcg = blkg->blkcg;
Tejun Heo03aa2642012-03-05 13:15:19 -0800242
Tejun Heo27e1f9d2012-06-05 13:36:44 +0200243 lockdep_assert_held(blkg->q->queue_lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800244 lockdep_assert_held(&blkcg->lock);
Tejun Heo03aa2642012-03-05 13:15:19 -0800245
246 /* Something wrong if we are trying to remove same group twice */
Tejun Heoe8989fa2012-03-05 13:15:20 -0800247 WARN_ON_ONCE(list_empty(&blkg->q_node));
Tejun Heo9f13ef62012-03-05 13:15:21 -0800248 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
Tejun Heoa6371202012-04-19 16:29:24 -0700249
250 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
Tejun Heoe8989fa2012-03-05 13:15:20 -0800251 list_del_init(&blkg->q_node);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800252 hlist_del_init_rcu(&blkg->blkcg_node);
Tejun Heo03aa2642012-03-05 13:15:19 -0800253
Tejun Heo03aa2642012-03-05 13:15:19 -0800254 /*
Tejun Heoa6371202012-04-19 16:29:24 -0700255 * Both setting lookup hint to and clearing it from @blkg are done
256 * under queue_lock. If it's not pointing to @blkg now, it never
257 * will. Hint assignment itself can race safely.
258 */
259 if (rcu_dereference_raw(blkcg->blkg_hint) == blkg)
260 rcu_assign_pointer(blkcg->blkg_hint, NULL);
261
262 /*
Tejun Heo03aa2642012-03-05 13:15:19 -0800263 * Put the reference taken at the time of creation so that when all
264 * queues are gone, group can be destroyed.
265 */
266 blkg_put(blkg);
267}
268
Tejun Heo9f13ef62012-03-05 13:15:21 -0800269/**
270 * blkg_destroy_all - destroy all blkgs associated with a request_queue
271 * @q: request_queue of interest
Tejun Heo9f13ef62012-03-05 13:15:21 -0800272 *
Tejun Heo3c96cb32012-04-13 13:11:34 -0700273 * Destroy all blkgs associated with @q.
Tejun Heo9f13ef62012-03-05 13:15:21 -0800274 */
Tejun Heo3c96cb32012-04-13 13:11:34 -0700275static void blkg_destroy_all(struct request_queue *q)
Tejun Heo03aa2642012-03-05 13:15:19 -0800276{
Tejun Heo3c798392012-04-16 13:57:25 -0700277 struct blkcg_gq *blkg, *n;
Tejun Heo72e06c22012-03-05 13:15:00 -0800278
Tejun Heo6d18b002012-04-13 13:11:35 -0700279 lockdep_assert_held(q->queue_lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800280
Tejun Heo9f13ef62012-03-05 13:15:21 -0800281 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
Tejun Heo3c798392012-04-16 13:57:25 -0700282 struct blkcg *blkcg = blkg->blkcg;
Tejun Heo72e06c22012-03-05 13:15:00 -0800283
Tejun Heo9f13ef62012-03-05 13:15:21 -0800284 spin_lock(&blkcg->lock);
285 blkg_destroy(blkg);
286 spin_unlock(&blkcg->lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800287 }
Jun'ichi Nomura65635cb2012-10-17 17:45:36 +0900288
289 /*
290 * root blkg is destroyed. Just clear the pointer since
291 * root_rl does not take reference on root blkg.
292 */
293 q->root_blkg = NULL;
294 q->root_rl.blkg = NULL;
Tejun Heo72e06c22012-03-05 13:15:00 -0800295}
296
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800297static void blkg_rcu_free(struct rcu_head *rcu_head)
298{
Tejun Heo3c798392012-04-16 13:57:25 -0700299 blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head));
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800300}
301
Tejun Heo3c798392012-04-16 13:57:25 -0700302void __blkg_release(struct blkcg_gq *blkg)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800303{
304 /* release the extra blkcg reference this blkg has been holding */
305 css_put(&blkg->blkcg->css);
306
307 /*
308 * A group is freed in rcu manner. But having an rcu lock does not
309 * mean that one can access all the fields of blkg and assume these
310 * are valid. For example, don't try to follow throtl_data and
311 * request queue links.
312 *
313 * Having a reference to blkg under an rcu allows acess to only
314 * values local to groups like group stats and group rate limits
315 */
316 call_rcu(&blkg->rcu_head, blkg_rcu_free);
317}
318EXPORT_SYMBOL_GPL(__blkg_release);
319
Tejun Heoa0516612012-06-26 15:05:44 -0700320/*
321 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
322 * because the root blkg uses @q->root_rl instead of its own rl.
323 */
324struct request_list *__blk_queue_next_rl(struct request_list *rl,
325 struct request_queue *q)
326{
327 struct list_head *ent;
328 struct blkcg_gq *blkg;
329
330 /*
331 * Determine the current blkg list_head. The first entry is
332 * root_rl which is off @q->blkg_list and mapped to the head.
333 */
334 if (rl == &q->root_rl) {
335 ent = &q->blkg_list;
Jun'ichi Nomura65c77fd2012-10-22 10:15:37 +0900336 /* There are no more block groups, hence no request lists */
337 if (list_empty(ent))
338 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -0700339 } else {
340 blkg = container_of(rl, struct blkcg_gq, rl);
341 ent = &blkg->q_node;
342 }
343
344 /* walk to the next list_head, skip root blkcg */
345 ent = ent->next;
346 if (ent == &q->root_blkg->q_node)
347 ent = ent->next;
348 if (ent == &q->blkg_list)
349 return NULL;
350
351 blkg = container_of(ent, struct blkcg_gq, q_node);
352 return &blkg->rl;
353}
354
Tejun Heo3c798392012-04-16 13:57:25 -0700355static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype,
356 u64 val)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700357{
Tejun Heo3c798392012-04-16 13:57:25 -0700358 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
359 struct blkcg_gq *blkg;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700360 struct hlist_node *n;
Tejun Heobc0d6502012-04-13 13:11:26 -0700361 int i;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700362
Tejun Heobc0d6502012-04-13 13:11:26 -0700363 mutex_lock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700364 spin_lock_irq(&blkcg->lock);
Tejun Heo997a0262012-03-08 10:53:58 -0800365
366 /*
367 * Note that stat reset is racy - it doesn't synchronize against
368 * stat updates. This is a debug feature which shouldn't exist
369 * anyway. If you get hit by a race, retry.
370 */
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700371 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
Tejun Heo8bd435b2012-04-13 13:11:28 -0700372 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -0700373 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heo549d3aa2012-03-05 13:15:16 -0800374
Tejun Heoa2b16932012-04-13 13:11:33 -0700375 if (blkcg_policy_enabled(blkg->q, pol) &&
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700376 pol->pd_reset_stats_fn)
377 pol->pd_reset_stats_fn(blkg);
Tejun Heobc0d6502012-04-13 13:11:26 -0700378 }
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700379 }
Vivek Goyalf0bdc8c2011-05-19 15:38:30 -0400380
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700381 spin_unlock_irq(&blkcg->lock);
Tejun Heobc0d6502012-04-13 13:11:26 -0700382 mutex_unlock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700383 return 0;
384}
385
Tejun Heo3c798392012-04-16 13:57:25 -0700386static const char *blkg_dev_name(struct blkcg_gq *blkg)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700387{
Tejun Heod3d32e62012-04-01 14:38:42 -0700388 /* some drivers (floppy) instantiate a queue w/o disk registered */
389 if (blkg->q->backing_dev_info.dev)
390 return dev_name(blkg->q->backing_dev_info.dev);
391 return NULL;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700392}
393
Tejun Heod3d32e62012-04-01 14:38:42 -0700394/**
395 * blkcg_print_blkgs - helper for printing per-blkg data
396 * @sf: seq_file to print to
397 * @blkcg: blkcg of interest
398 * @prfill: fill function to print out a blkg
399 * @pol: policy in question
400 * @data: data to be passed to @prfill
401 * @show_total: to print out sum of prfill return values or not
402 *
403 * This function invokes @prfill on each blkg of @blkcg if pd for the
404 * policy specified by @pol exists. @prfill is invoked with @sf, the
405 * policy data and @data. If @show_total is %true, the sum of the return
406 * values from @prfill is printed with "Total" label at the end.
407 *
408 * This is to be used to construct print functions for
409 * cftype->read_seq_string method.
410 */
Tejun Heo3c798392012-04-16 13:57:25 -0700411void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
Tejun Heof95a04a2012-04-16 13:57:26 -0700412 u64 (*prfill)(struct seq_file *,
413 struct blkg_policy_data *, int),
Tejun Heo3c798392012-04-16 13:57:25 -0700414 const struct blkcg_policy *pol, int data,
Tejun Heoec399342012-04-13 13:11:27 -0700415 bool show_total)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400416{
Tejun Heo3c798392012-04-16 13:57:25 -0700417 struct blkcg_gq *blkg;
Tejun Heod3d32e62012-04-01 14:38:42 -0700418 struct hlist_node *n;
419 u64 total = 0;
420
421 spin_lock_irq(&blkcg->lock);
422 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node)
Tejun Heoa2b16932012-04-13 13:11:33 -0700423 if (blkcg_policy_enabled(blkg->q, pol))
Tejun Heof95a04a2012-04-16 13:57:26 -0700424 total += prfill(sf, blkg->pd[pol->plid], data);
Tejun Heod3d32e62012-04-01 14:38:42 -0700425 spin_unlock_irq(&blkcg->lock);
426
427 if (show_total)
428 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
429}
Tejun Heo829fdb52012-04-01 14:38:43 -0700430EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
Tejun Heod3d32e62012-04-01 14:38:42 -0700431
432/**
433 * __blkg_prfill_u64 - prfill helper for a single u64 value
434 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700435 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700436 * @v: value to print
437 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700438 * Print @v to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700439 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700440u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
Tejun Heod3d32e62012-04-01 14:38:42 -0700441{
Tejun Heof95a04a2012-04-16 13:57:26 -0700442 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700443
444 if (!dname)
445 return 0;
446
447 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
448 return v;
449}
Tejun Heo829fdb52012-04-01 14:38:43 -0700450EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
Tejun Heod3d32e62012-04-01 14:38:42 -0700451
452/**
453 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
454 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700455 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700456 * @rwstat: rwstat to print
457 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700458 * Print @rwstat to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700459 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700460u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
Tejun Heo829fdb52012-04-01 14:38:43 -0700461 const struct blkg_rwstat *rwstat)
Tejun Heod3d32e62012-04-01 14:38:42 -0700462{
463 static const char *rwstr[] = {
464 [BLKG_RWSTAT_READ] = "Read",
465 [BLKG_RWSTAT_WRITE] = "Write",
466 [BLKG_RWSTAT_SYNC] = "Sync",
467 [BLKG_RWSTAT_ASYNC] = "Async",
468 };
Tejun Heof95a04a2012-04-16 13:57:26 -0700469 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700470 u64 v;
471 int i;
472
473 if (!dname)
474 return 0;
475
476 for (i = 0; i < BLKG_RWSTAT_NR; i++)
477 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
478 (unsigned long long)rwstat->cnt[i]);
479
480 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
481 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
482 return v;
483}
484
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700485/**
486 * blkg_prfill_stat - prfill callback for blkg_stat
487 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700488 * @pd: policy private data of interest
489 * @off: offset to the blkg_stat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700490 *
491 * prfill callback for printing a blkg_stat.
492 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700493u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700494{
Tejun Heof95a04a2012-04-16 13:57:26 -0700495 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
Tejun Heod3d32e62012-04-01 14:38:42 -0700496}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700497EXPORT_SYMBOL_GPL(blkg_prfill_stat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700498
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700499/**
500 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
501 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700502 * @pd: policy private data of interest
503 * @off: offset to the blkg_rwstat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700504 *
505 * prfill callback for printing a blkg_rwstat.
506 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700507u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
508 int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700509{
Tejun Heof95a04a2012-04-16 13:57:26 -0700510 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
Tejun Heod3d32e62012-04-01 14:38:42 -0700511
Tejun Heof95a04a2012-04-16 13:57:26 -0700512 return __blkg_prfill_rwstat(sf, pd, &rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700513}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700514EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700515
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700516/**
517 * blkg_conf_prep - parse and prepare for per-blkg config update
518 * @blkcg: target block cgroup
Tejun Heoda8b0662012-04-13 13:11:29 -0700519 * @pol: target policy
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700520 * @input: input string
521 * @ctx: blkg_conf_ctx to be filled
522 *
523 * Parse per-blkg config update from @input and initialize @ctx with the
524 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
Tejun Heoda8b0662012-04-13 13:11:29 -0700525 * value. This function returns with RCU read lock and queue lock held and
526 * must be paired with blkg_conf_finish().
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700527 */
Tejun Heo3c798392012-04-16 13:57:25 -0700528int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
529 const char *input, struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700530 __acquires(rcu) __acquires(disk->queue->queue_lock)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800531{
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700532 struct gendisk *disk;
Tejun Heo3c798392012-04-16 13:57:25 -0700533 struct blkcg_gq *blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700534 unsigned int major, minor;
535 unsigned long long v;
536 int part, ret;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800537
Tejun Heo726fa692012-04-01 14:38:43 -0700538 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
539 return -EINVAL;
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700540
Tejun Heo726fa692012-04-01 14:38:43 -0700541 disk = get_gendisk(MKDEV(major, minor), &part);
Tejun Heo4bfd4822012-03-05 13:15:08 -0800542 if (!disk || part)
Tejun Heo726fa692012-04-01 14:38:43 -0700543 return -EINVAL;
Tejun Heoe56da7e2012-03-05 13:15:07 -0800544
545 rcu_read_lock();
Tejun Heo4bfd4822012-03-05 13:15:08 -0800546 spin_lock_irq(disk->queue->queue_lock);
Tejun Heoda8b0662012-04-13 13:11:29 -0700547
Tejun Heoa2b16932012-04-13 13:11:33 -0700548 if (blkcg_policy_enabled(disk->queue, pol))
Tejun Heo3c96cb32012-04-13 13:11:34 -0700549 blkg = blkg_lookup_create(blkcg, disk->queue);
Tejun Heoa2b16932012-04-13 13:11:33 -0700550 else
551 blkg = ERR_PTR(-EINVAL);
Tejun Heoe56da7e2012-03-05 13:15:07 -0800552
Tejun Heo4bfd4822012-03-05 13:15:08 -0800553 if (IS_ERR(blkg)) {
554 ret = PTR_ERR(blkg);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700555 rcu_read_unlock();
Tejun Heoda8b0662012-04-13 13:11:29 -0700556 spin_unlock_irq(disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700557 put_disk(disk);
558 /*
559 * If queue was bypassing, we should retry. Do so after a
560 * short msleep(). It isn't strictly necessary but queue
561 * can be bypassing for some time and it's always nice to
562 * avoid busy looping.
563 */
564 if (ret == -EBUSY) {
565 msleep(10);
566 ret = restart_syscall();
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400567 }
Tejun Heo726fa692012-04-01 14:38:43 -0700568 return ret;
Vivek Goyal062a6442010-09-15 17:06:33 -0400569 }
Tejun Heoe56da7e2012-03-05 13:15:07 -0800570
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700571 ctx->disk = disk;
572 ctx->blkg = blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700573 ctx->v = v;
574 return 0;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800575}
Tejun Heo829fdb52012-04-01 14:38:43 -0700576EXPORT_SYMBOL_GPL(blkg_conf_prep);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800577
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700578/**
579 * blkg_conf_finish - finish up per-blkg config update
580 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
581 *
582 * Finish up after per-blkg config update. This function must be paired
583 * with blkg_conf_prep().
584 */
Tejun Heo829fdb52012-04-01 14:38:43 -0700585void blkg_conf_finish(struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700586 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800587{
Tejun Heoda8b0662012-04-13 13:11:29 -0700588 spin_unlock_irq(ctx->disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700589 rcu_read_unlock();
590 put_disk(ctx->disk);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800591}
Tejun Heo829fdb52012-04-01 14:38:43 -0700592EXPORT_SYMBOL_GPL(blkg_conf_finish);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800593
Tejun Heo3c798392012-04-16 13:57:25 -0700594struct cftype blkcg_files[] = {
Vivek Goyal31e4c282009-12-03 12:59:42 -0500595 {
Divyesh Shah84c124d2010-04-09 08:31:19 +0200596 .name = "reset_stats",
Tejun Heo3c798392012-04-16 13:57:25 -0700597 .write_u64 = blkcg_reset_stats,
Vivek Goyal22084192009-12-03 12:59:49 -0500598 },
Tejun Heo4baf6e32012-04-01 12:09:55 -0700599 { } /* terminate */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500600};
601
Tejun Heo9f13ef62012-03-05 13:15:21 -0800602/**
Tejun Heo92fb9742012-11-19 08:13:38 -0800603 * blkcg_css_offline - cgroup css_offline callback
Tejun Heo9f13ef62012-03-05 13:15:21 -0800604 * @cgroup: cgroup of interest
605 *
606 * This function is called when @cgroup is about to go away and responsible
607 * for shooting down all blkgs associated with @cgroup. blkgs should be
608 * removed while holding both q and blkcg locks. As blkcg lock is nested
609 * inside q lock, this function performs reverse double lock dancing.
610 *
611 * This is the blkcg counterpart of ioc_release_fn().
612 */
Tejun Heo92fb9742012-11-19 08:13:38 -0800613static void blkcg_css_offline(struct cgroup *cgroup)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500614{
Tejun Heo3c798392012-04-16 13:57:25 -0700615 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500616
Tejun Heo9f13ef62012-03-05 13:15:21 -0800617 spin_lock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800618
Tejun Heo9f13ef62012-03-05 13:15:21 -0800619 while (!hlist_empty(&blkcg->blkg_list)) {
Tejun Heo3c798392012-04-16 13:57:25 -0700620 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
621 struct blkcg_gq, blkcg_node);
Tejun Heoc875f4d2012-03-05 13:15:22 -0800622 struct request_queue *q = blkg->q;
Vivek Goyalb1c35762009-12-03 12:59:47 -0500623
Tejun Heo9f13ef62012-03-05 13:15:21 -0800624 if (spin_trylock(q->queue_lock)) {
625 blkg_destroy(blkg);
626 spin_unlock(q->queue_lock);
627 } else {
628 spin_unlock_irq(&blkcg->lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800629 cpu_relax();
Dan Carpentera5567932012-03-29 20:57:08 +0200630 spin_lock_irq(&blkcg->lock);
Jens Axboe0f3942a2010-05-03 14:28:55 +0200631 }
Tejun Heo9f13ef62012-03-05 13:15:21 -0800632 }
Jens Axboe0f3942a2010-05-03 14:28:55 +0200633
Tejun Heo9f13ef62012-03-05 13:15:21 -0800634 spin_unlock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800635}
636
Tejun Heo92fb9742012-11-19 08:13:38 -0800637static void blkcg_css_free(struct cgroup *cgroup)
Tejun Heo7ee9c562012-03-05 13:15:11 -0800638{
Tejun Heo3c798392012-04-16 13:57:25 -0700639 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800640
Tejun Heo3c798392012-04-16 13:57:25 -0700641 if (blkcg != &blkcg_root)
Ben Blum67523c42010-03-10 15:22:11 -0800642 kfree(blkcg);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500643}
644
Tejun Heo92fb9742012-11-19 08:13:38 -0800645static struct cgroup_subsys_state *blkcg_css_alloc(struct cgroup *cgroup)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500646{
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700647 static atomic64_t id_seq = ATOMIC64_INIT(0);
Tejun Heo3c798392012-04-16 13:57:25 -0700648 struct blkcg *blkcg;
Li Zefan03415092010-05-07 08:57:00 +0200649 struct cgroup *parent = cgroup->parent;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500650
Li Zefan03415092010-05-07 08:57:00 +0200651 if (!parent) {
Tejun Heo3c798392012-04-16 13:57:25 -0700652 blkcg = &blkcg_root;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500653 goto done;
654 }
655
Vivek Goyal31e4c282009-12-03 12:59:42 -0500656 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
657 if (!blkcg)
658 return ERR_PTR(-ENOMEM);
659
Tejun Heo3381cb82012-04-01 14:38:44 -0700660 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700661 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500662done:
663 spin_lock_init(&blkcg->lock);
Tejun Heoa6371202012-04-19 16:29:24 -0700664 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500665 INIT_HLIST_HEAD(&blkcg->blkg_list);
666
667 return &blkcg->css;
668}
669
Tejun Heo5efd6112012-03-05 13:15:12 -0800670/**
671 * blkcg_init_queue - initialize blkcg part of request queue
672 * @q: request_queue to initialize
673 *
674 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
675 * part of new request_queue @q.
676 *
677 * RETURNS:
678 * 0 on success, -errno on failure.
679 */
680int blkcg_init_queue(struct request_queue *q)
681{
682 might_sleep();
683
Tejun Heo3c96cb32012-04-13 13:11:34 -0700684 return blk_throtl_init(q);
Tejun Heo5efd6112012-03-05 13:15:12 -0800685}
686
687/**
688 * blkcg_drain_queue - drain blkcg part of request_queue
689 * @q: request_queue to drain
690 *
691 * Called from blk_drain_queue(). Responsible for draining blkcg part.
692 */
693void blkcg_drain_queue(struct request_queue *q)
694{
695 lockdep_assert_held(q->queue_lock);
696
697 blk_throtl_drain(q);
698}
699
700/**
701 * blkcg_exit_queue - exit and release blkcg part of request_queue
702 * @q: request_queue being released
703 *
704 * Called from blk_release_queue(). Responsible for exiting blkcg part.
705 */
706void blkcg_exit_queue(struct request_queue *q)
707{
Tejun Heo6d18b002012-04-13 13:11:35 -0700708 spin_lock_irq(q->queue_lock);
Tejun Heo3c96cb32012-04-13 13:11:34 -0700709 blkg_destroy_all(q);
Tejun Heo6d18b002012-04-13 13:11:35 -0700710 spin_unlock_irq(q->queue_lock);
711
Tejun Heo5efd6112012-03-05 13:15:12 -0800712 blk_throtl_exit(q);
713}
714
Vivek Goyal31e4c282009-12-03 12:59:42 -0500715/*
716 * We cannot support shared io contexts, as we have no mean to support
717 * two tasks with the same ioc in two different groups without major rework
718 * of the main cic data structures. For now we allow a task to change
719 * its cgroup only if it's the only owner of its ioc.
720 */
Tejun Heo3c798392012-04-16 13:57:25 -0700721static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500722{
Tejun Heobb9d97b2011-12-12 18:12:21 -0800723 struct task_struct *task;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500724 struct io_context *ioc;
725 int ret = 0;
726
727 /* task_lock() is needed to avoid races with exit_io_context() */
Tejun Heobb9d97b2011-12-12 18:12:21 -0800728 cgroup_taskset_for_each(task, cgrp, tset) {
729 task_lock(task);
730 ioc = task->io_context;
731 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
732 ret = -EINVAL;
733 task_unlock(task);
734 if (ret)
735 break;
736 }
Vivek Goyal31e4c282009-12-03 12:59:42 -0500737 return ret;
738}
739
Tejun Heo676f7c82012-04-01 12:09:55 -0700740struct cgroup_subsys blkio_subsys = {
741 .name = "blkio",
Tejun Heo92fb9742012-11-19 08:13:38 -0800742 .css_alloc = blkcg_css_alloc,
743 .css_offline = blkcg_css_offline,
744 .css_free = blkcg_css_free,
Tejun Heo3c798392012-04-16 13:57:25 -0700745 .can_attach = blkcg_can_attach,
Tejun Heo676f7c82012-04-01 12:09:55 -0700746 .subsys_id = blkio_subsys_id,
Tejun Heo3c798392012-04-16 13:57:25 -0700747 .base_cftypes = blkcg_files,
Tejun Heo676f7c82012-04-01 12:09:55 -0700748 .module = THIS_MODULE,
Tejun Heo8c7f6ed2012-09-13 12:20:58 -0700749
750 /*
751 * blkio subsystem is utterly broken in terms of hierarchy support.
752 * It treats all cgroups equally regardless of where they're
753 * located in the hierarchy - all cgroups are treated as if they're
754 * right below the root. Fix it and remove the following.
755 */
756 .broken_hierarchy = true,
Tejun Heo676f7c82012-04-01 12:09:55 -0700757};
758EXPORT_SYMBOL_GPL(blkio_subsys);
759
Tejun Heo8bd435b2012-04-13 13:11:28 -0700760/**
Tejun Heoa2b16932012-04-13 13:11:33 -0700761 * blkcg_activate_policy - activate a blkcg policy on a request_queue
762 * @q: request_queue of interest
763 * @pol: blkcg policy to activate
764 *
765 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
766 * bypass mode to populate its blkgs with policy_data for @pol.
767 *
768 * Activation happens with @q bypassed, so nobody would be accessing blkgs
769 * from IO path. Update of each blkg is protected by both queue and blkcg
770 * locks so that holding either lock and testing blkcg_policy_enabled() is
771 * always enough for dereferencing policy data.
772 *
773 * The caller is responsible for synchronizing [de]activations and policy
774 * [un]registerations. Returns 0 on success, -errno on failure.
775 */
776int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700777 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -0700778{
779 LIST_HEAD(pds);
Tejun Heo3c798392012-04-16 13:57:25 -0700780 struct blkcg_gq *blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -0700781 struct blkg_policy_data *pd, *n;
782 int cnt = 0, ret;
Tejun Heo15974992012-06-04 20:40:52 -0700783 bool preloaded;
Tejun Heoa2b16932012-04-13 13:11:33 -0700784
785 if (blkcg_policy_enabled(q, pol))
786 return 0;
787
Tejun Heo15974992012-06-04 20:40:52 -0700788 /* preallocations for root blkg */
789 blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
790 if (!blkg)
791 return -ENOMEM;
792
793 preloaded = !radix_tree_preload(GFP_KERNEL);
794
Tejun Heoa2b16932012-04-13 13:11:33 -0700795 blk_queue_bypass_start(q);
796
797 /* make sure the root blkg exists and count the existing blkgs */
798 spin_lock_irq(q->queue_lock);
799
800 rcu_read_lock();
Tejun Heo15974992012-06-04 20:40:52 -0700801 blkg = __blkg_lookup_create(&blkcg_root, q, blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -0700802 rcu_read_unlock();
803
Tejun Heo15974992012-06-04 20:40:52 -0700804 if (preloaded)
805 radix_tree_preload_end();
806
Tejun Heoa2b16932012-04-13 13:11:33 -0700807 if (IS_ERR(blkg)) {
808 ret = PTR_ERR(blkg);
809 goto out_unlock;
810 }
811 q->root_blkg = blkg;
Tejun Heoa0516612012-06-26 15:05:44 -0700812 q->root_rl.blkg = blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -0700813
814 list_for_each_entry(blkg, &q->blkg_list, q_node)
815 cnt++;
816
817 spin_unlock_irq(q->queue_lock);
818
819 /* allocate policy_data for all existing blkgs */
820 while (cnt--) {
Tejun Heof95a04a2012-04-16 13:57:26 -0700821 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
Tejun Heoa2b16932012-04-13 13:11:33 -0700822 if (!pd) {
823 ret = -ENOMEM;
824 goto out_free;
825 }
826 list_add_tail(&pd->alloc_node, &pds);
827 }
828
829 /*
830 * Install the allocated pds. With @q bypassing, no new blkg
831 * should have been created while the queue lock was dropped.
832 */
833 spin_lock_irq(q->queue_lock);
834
835 list_for_each_entry(blkg, &q->blkg_list, q_node) {
836 if (WARN_ON(list_empty(&pds))) {
837 /* umm... this shouldn't happen, just abort */
838 ret = -ENOMEM;
839 goto out_unlock;
840 }
841 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
842 list_del_init(&pd->alloc_node);
843
844 /* grab blkcg lock too while installing @pd on @blkg */
845 spin_lock(&blkg->blkcg->lock);
846
847 blkg->pd[pol->plid] = pd;
848 pd->blkg = blkg;
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700849 pol->pd_init_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -0700850
851 spin_unlock(&blkg->blkcg->lock);
852 }
853
854 __set_bit(pol->plid, q->blkcg_pols);
855 ret = 0;
856out_unlock:
857 spin_unlock_irq(q->queue_lock);
858out_free:
859 blk_queue_bypass_end(q);
860 list_for_each_entry_safe(pd, n, &pds, alloc_node)
861 kfree(pd);
862 return ret;
863}
864EXPORT_SYMBOL_GPL(blkcg_activate_policy);
865
866/**
867 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
868 * @q: request_queue of interest
869 * @pol: blkcg policy to deactivate
870 *
871 * Deactivate @pol on @q. Follows the same synchronization rules as
872 * blkcg_activate_policy().
873 */
874void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700875 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -0700876{
Tejun Heo3c798392012-04-16 13:57:25 -0700877 struct blkcg_gq *blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -0700878
879 if (!blkcg_policy_enabled(q, pol))
880 return;
881
882 blk_queue_bypass_start(q);
883 spin_lock_irq(q->queue_lock);
884
885 __clear_bit(pol->plid, q->blkcg_pols);
886
Tejun Heo6d18b002012-04-13 13:11:35 -0700887 /* if no policy is left, no need for blkgs - shoot them down */
888 if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
889 blkg_destroy_all(q);
890
Tejun Heoa2b16932012-04-13 13:11:33 -0700891 list_for_each_entry(blkg, &q->blkg_list, q_node) {
892 /* grab blkcg lock too while removing @pd from @blkg */
893 spin_lock(&blkg->blkcg->lock);
894
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700895 if (pol->pd_exit_fn)
896 pol->pd_exit_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -0700897
898 kfree(blkg->pd[pol->plid]);
899 blkg->pd[pol->plid] = NULL;
900
901 spin_unlock(&blkg->blkcg->lock);
902 }
903
904 spin_unlock_irq(q->queue_lock);
905 blk_queue_bypass_end(q);
906}
907EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
908
909/**
Tejun Heo3c798392012-04-16 13:57:25 -0700910 * blkcg_policy_register - register a blkcg policy
911 * @pol: blkcg policy to register
Tejun Heo8bd435b2012-04-13 13:11:28 -0700912 *
Tejun Heo3c798392012-04-16 13:57:25 -0700913 * Register @pol with blkcg core. Might sleep and @pol may be modified on
914 * successful registration. Returns 0 on success and -errno on failure.
Tejun Heo8bd435b2012-04-13 13:11:28 -0700915 */
Tejun Heo3c798392012-04-16 13:57:25 -0700916int blkcg_policy_register(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -0500917{
Tejun Heo8bd435b2012-04-13 13:11:28 -0700918 int i, ret;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800919
Tejun Heof95a04a2012-04-16 13:57:26 -0700920 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
921 return -EINVAL;
922
Tejun Heobc0d6502012-04-13 13:11:26 -0700923 mutex_lock(&blkcg_pol_mutex);
924
Tejun Heo8bd435b2012-04-13 13:11:28 -0700925 /* find an empty slot */
926 ret = -ENOSPC;
927 for (i = 0; i < BLKCG_MAX_POLS; i++)
Tejun Heo3c798392012-04-16 13:57:25 -0700928 if (!blkcg_policy[i])
Tejun Heo8bd435b2012-04-13 13:11:28 -0700929 break;
930 if (i >= BLKCG_MAX_POLS)
931 goto out_unlock;
Tejun Heo035d10b2012-03-05 13:15:04 -0800932
Tejun Heo8bd435b2012-04-13 13:11:28 -0700933 /* register and update blkgs */
Tejun Heo3c798392012-04-16 13:57:25 -0700934 pol->plid = i;
935 blkcg_policy[i] = pol;
Tejun Heo8bd435b2012-04-13 13:11:28 -0700936
Tejun Heo8bd435b2012-04-13 13:11:28 -0700937 /* everything is in place, add intf files for the new policy */
Tejun Heo3c798392012-04-16 13:57:25 -0700938 if (pol->cftypes)
939 WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes));
Tejun Heo8bd435b2012-04-13 13:11:28 -0700940 ret = 0;
941out_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -0700942 mutex_unlock(&blkcg_pol_mutex);
Tejun Heo8bd435b2012-04-13 13:11:28 -0700943 return ret;
Vivek Goyal3e252062009-12-04 10:36:42 -0500944}
Tejun Heo3c798392012-04-16 13:57:25 -0700945EXPORT_SYMBOL_GPL(blkcg_policy_register);
Vivek Goyal3e252062009-12-04 10:36:42 -0500946
Tejun Heo8bd435b2012-04-13 13:11:28 -0700947/**
Tejun Heo3c798392012-04-16 13:57:25 -0700948 * blkcg_policy_unregister - unregister a blkcg policy
949 * @pol: blkcg policy to unregister
Tejun Heo8bd435b2012-04-13 13:11:28 -0700950 *
Tejun Heo3c798392012-04-16 13:57:25 -0700951 * Undo blkcg_policy_register(@pol). Might sleep.
Tejun Heo8bd435b2012-04-13 13:11:28 -0700952 */
Tejun Heo3c798392012-04-16 13:57:25 -0700953void blkcg_policy_unregister(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -0500954{
Tejun Heobc0d6502012-04-13 13:11:26 -0700955 mutex_lock(&blkcg_pol_mutex);
956
Tejun Heo3c798392012-04-16 13:57:25 -0700957 if (WARN_ON(blkcg_policy[pol->plid] != pol))
Tejun Heo8bd435b2012-04-13 13:11:28 -0700958 goto out_unlock;
959
960 /* kill the intf files first */
Tejun Heo3c798392012-04-16 13:57:25 -0700961 if (pol->cftypes)
962 cgroup_rm_cftypes(&blkio_subsys, pol->cftypes);
Tejun Heo44ea53d2012-04-01 14:38:43 -0700963
Tejun Heo8bd435b2012-04-13 13:11:28 -0700964 /* unregister and update blkgs */
Tejun Heo3c798392012-04-16 13:57:25 -0700965 blkcg_policy[pol->plid] = NULL;
Tejun Heo8bd435b2012-04-13 13:11:28 -0700966out_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -0700967 mutex_unlock(&blkcg_pol_mutex);
Vivek Goyal3e252062009-12-04 10:36:42 -0500968}
Tejun Heo3c798392012-04-16 13:57:25 -0700969EXPORT_SYMBOL_GPL(blkcg_policy_unregister);