2 * Common Block IO controller cgroup interface
4 * Based on ideas and code from CFQ, CFS and BFQ:
5 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
7 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
8 * Paolo Valente <paolo.valente@unimore.it>
10 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
11 * Nauman Rafique <nauman@google.com>
13 #include <linux/ioprio.h>
14 #include <linux/seq_file.h>
15 #include <linux/kdev_t.h>
16 #include <linux/module.h>
17 #include "blk-cgroup.h"
19 static DEFINE_SPINLOCK(blkio_list_lock
);
20 static LIST_HEAD(blkio_list
);
22 struct blkio_cgroup blkio_root_cgroup
= { .weight
= 2*BLKIO_WEIGHT_DEFAULT
};
23 EXPORT_SYMBOL_GPL(blkio_root_cgroup
);
25 bool blkiocg_css_tryget(struct blkio_cgroup
*blkcg
)
27 if (!css_tryget(&blkcg
->css
))
31 EXPORT_SYMBOL_GPL(blkiocg_css_tryget
);
33 void blkiocg_css_put(struct blkio_cgroup
*blkcg
)
37 EXPORT_SYMBOL_GPL(blkiocg_css_put
);
39 struct blkio_cgroup
*cgroup_to_blkio_cgroup(struct cgroup
*cgroup
)
41 return container_of(cgroup_subsys_state(cgroup
, blkio_subsys_id
),
42 struct blkio_cgroup
, css
);
44 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup
);
46 void blkiocg_update_blkio_group_stats(struct blkio_group
*blkg
,
47 unsigned long time
, unsigned long sectors
)
50 blkg
->sectors
+= sectors
;
52 EXPORT_SYMBOL_GPL(blkiocg_update_blkio_group_stats
);
54 void blkiocg_add_blkio_group(struct blkio_cgroup
*blkcg
,
55 struct blkio_group
*blkg
, void *key
, dev_t dev
)
59 spin_lock_irqsave(&blkcg
->lock
, flags
);
60 rcu_assign_pointer(blkg
->key
, key
);
61 blkg
->blkcg_id
= css_id(&blkcg
->css
);
62 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
63 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
64 #ifdef CONFIG_DEBUG_BLK_CGROUP
65 /* Need to take css reference ? */
66 cgroup_path(blkcg
->css
.cgroup
, blkg
->path
, sizeof(blkg
->path
));
70 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group
);
72 static void __blkiocg_del_blkio_group(struct blkio_group
*blkg
)
74 hlist_del_init_rcu(&blkg
->blkcg_node
);
79 * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
80 * indicating that blk_group was unhashed by the time we got to it.
82 int blkiocg_del_blkio_group(struct blkio_group
*blkg
)
84 struct blkio_cgroup
*blkcg
;
86 struct cgroup_subsys_state
*css
;
90 css
= css_lookup(&blkio_subsys
, blkg
->blkcg_id
);
94 blkcg
= container_of(css
, struct blkio_cgroup
, css
);
95 spin_lock_irqsave(&blkcg
->lock
, flags
);
96 if (!hlist_unhashed(&blkg
->blkcg_node
)) {
97 __blkiocg_del_blkio_group(blkg
);
100 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
105 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group
);
107 /* called under rcu_read_lock(). */
108 struct blkio_group
*blkiocg_lookup_group(struct blkio_cgroup
*blkcg
, void *key
)
110 struct blkio_group
*blkg
;
111 struct hlist_node
*n
;
114 hlist_for_each_entry_rcu(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
) {
122 EXPORT_SYMBOL_GPL(blkiocg_lookup_group
);
124 #define SHOW_FUNCTION(__VAR) \
125 static u64 blkiocg_##__VAR##_read(struct cgroup *cgroup, \
126 struct cftype *cftype) \
128 struct blkio_cgroup *blkcg; \
130 blkcg = cgroup_to_blkio_cgroup(cgroup); \
131 return (u64)blkcg->__VAR; \
134 SHOW_FUNCTION(weight
);
138 blkiocg_weight_write(struct cgroup
*cgroup
, struct cftype
*cftype
, u64 val
)
140 struct blkio_cgroup
*blkcg
;
141 struct blkio_group
*blkg
;
142 struct hlist_node
*n
;
143 struct blkio_policy_type
*blkiop
;
145 if (val
< BLKIO_WEIGHT_MIN
|| val
> BLKIO_WEIGHT_MAX
)
148 blkcg
= cgroup_to_blkio_cgroup(cgroup
);
149 spin_lock_irq(&blkcg
->lock
);
150 blkcg
->weight
= (unsigned int)val
;
151 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
) {
152 spin_lock(&blkio_list_lock
);
153 list_for_each_entry(blkiop
, &blkio_list
, list
)
154 blkiop
->ops
.blkio_update_group_weight_fn(blkg
,
156 spin_unlock(&blkio_list_lock
);
158 spin_unlock_irq(&blkcg
->lock
);
162 #define SHOW_FUNCTION_PER_GROUP(__VAR) \
163 static int blkiocg_##__VAR##_read(struct cgroup *cgroup, \
164 struct cftype *cftype, struct seq_file *m) \
166 struct blkio_cgroup *blkcg; \
167 struct blkio_group *blkg; \
168 struct hlist_node *n; \
170 if (!cgroup_lock_live_group(cgroup)) \
173 blkcg = cgroup_to_blkio_cgroup(cgroup); \
175 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {\
177 seq_printf(m, "%u:%u %lu\n", MAJOR(blkg->dev), \
178 MINOR(blkg->dev), blkg->__VAR); \
185 SHOW_FUNCTION_PER_GROUP(time
);
186 SHOW_FUNCTION_PER_GROUP(sectors
);
187 #ifdef CONFIG_DEBUG_BLK_CGROUP
188 SHOW_FUNCTION_PER_GROUP(dequeue
);
190 #undef SHOW_FUNCTION_PER_GROUP
192 #ifdef CONFIG_DEBUG_BLK_CGROUP
193 void blkiocg_update_blkio_group_dequeue_stats(struct blkio_group
*blkg
,
194 unsigned long dequeue
)
196 blkg
->dequeue
+= dequeue
;
198 EXPORT_SYMBOL_GPL(blkiocg_update_blkio_group_dequeue_stats
);
201 struct cftype blkio_files
[] = {
204 .read_u64
= blkiocg_weight_read
,
205 .write_u64
= blkiocg_weight_write
,
209 .read_seq_string
= blkiocg_time_read
,
213 .read_seq_string
= blkiocg_sectors_read
,
215 #ifdef CONFIG_DEBUG_BLK_CGROUP
218 .read_seq_string
= blkiocg_dequeue_read
,
223 static int blkiocg_populate(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
)
225 return cgroup_add_files(cgroup
, subsys
, blkio_files
,
226 ARRAY_SIZE(blkio_files
));
229 static void blkiocg_destroy(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
)
231 struct blkio_cgroup
*blkcg
= cgroup_to_blkio_cgroup(cgroup
);
233 struct blkio_group
*blkg
;
235 struct blkio_policy_type
*blkiop
;
239 spin_lock_irqsave(&blkcg
->lock
, flags
);
241 if (hlist_empty(&blkcg
->blkg_list
)) {
242 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
246 blkg
= hlist_entry(blkcg
->blkg_list
.first
, struct blkio_group
,
248 key
= rcu_dereference(blkg
->key
);
249 __blkiocg_del_blkio_group(blkg
);
251 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
254 * This blkio_group is being unlinked as associated cgroup is going
255 * away. Let all the IO controlling policies know about this event.
257 * Currently this is static call to one io controlling policy. Once
258 * we have more policies in place, we need some dynamic registration
259 * of callback function.
261 spin_lock(&blkio_list_lock
);
262 list_for_each_entry(blkiop
, &blkio_list
, list
)
263 blkiop
->ops
.blkio_unlink_group_fn(key
, blkg
);
264 spin_unlock(&blkio_list_lock
);
267 free_css_id(&blkio_subsys
, &blkcg
->css
);
272 static struct cgroup_subsys_state
*
273 blkiocg_create(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
)
275 struct blkio_cgroup
*blkcg
, *parent_blkcg
;
277 if (!cgroup
->parent
) {
278 blkcg
= &blkio_root_cgroup
;
282 /* Currently we do not support hierarchy deeper than two level (0,1) */
283 parent_blkcg
= cgroup_to_blkio_cgroup(cgroup
->parent
);
284 if (css_depth(&parent_blkcg
->css
) > 0)
285 return ERR_PTR(-EINVAL
);
287 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
289 return ERR_PTR(-ENOMEM
);
291 blkcg
->weight
= BLKIO_WEIGHT_DEFAULT
;
293 spin_lock_init(&blkcg
->lock
);
294 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
300 * We cannot support shared io contexts, as we have no mean to support
301 * two tasks with the same ioc in two different groups without major rework
302 * of the main cic data structures. For now we allow a task to change
303 * its cgroup only if it's the only owner of its ioc.
305 static int blkiocg_can_attach(struct cgroup_subsys
*subsys
,
306 struct cgroup
*cgroup
, struct task_struct
*tsk
,
309 struct io_context
*ioc
;
312 /* task_lock() is needed to avoid races with exit_io_context() */
314 ioc
= tsk
->io_context
;
315 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
322 static void blkiocg_attach(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
,
323 struct cgroup
*prev
, struct task_struct
*tsk
,
326 struct io_context
*ioc
;
329 ioc
= tsk
->io_context
;
331 ioc
->cgroup_changed
= 1;
335 struct cgroup_subsys blkio_subsys
= {
337 .create
= blkiocg_create
,
338 .can_attach
= blkiocg_can_attach
,
339 .attach
= blkiocg_attach
,
340 .destroy
= blkiocg_destroy
,
341 .populate
= blkiocg_populate
,
342 .subsys_id
= blkio_subsys_id
,
346 void blkio_policy_register(struct blkio_policy_type
*blkiop
)
348 spin_lock(&blkio_list_lock
);
349 list_add_tail(&blkiop
->list
, &blkio_list
);
350 spin_unlock(&blkio_list_lock
);
352 EXPORT_SYMBOL_GPL(blkio_policy_register
);
354 void blkio_policy_unregister(struct blkio_policy_type
*blkiop
)
356 spin_lock(&blkio_list_lock
);
357 list_del_init(&blkiop
->list
);
358 spin_unlock(&blkio_list_lock
);
360 EXPORT_SYMBOL_GPL(blkio_policy_unregister
);