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"
18 #include "cfq-iosched.h"
20 struct blkio_cgroup blkio_root_cgroup
= { .weight
= 2*BLKIO_WEIGHT_DEFAULT
};
21 EXPORT_SYMBOL_GPL(blkio_root_cgroup
);
23 bool blkiocg_css_tryget(struct blkio_cgroup
*blkcg
)
25 if (!css_tryget(&blkcg
->css
))
29 EXPORT_SYMBOL_GPL(blkiocg_css_tryget
);
31 void blkiocg_css_put(struct blkio_cgroup
*blkcg
)
35 EXPORT_SYMBOL_GPL(blkiocg_css_put
);
37 struct blkio_cgroup
*cgroup_to_blkio_cgroup(struct cgroup
*cgroup
)
39 return container_of(cgroup_subsys_state(cgroup
, blkio_subsys_id
),
40 struct blkio_cgroup
, css
);
42 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup
);
44 void blkiocg_update_blkio_group_stats(struct blkio_group
*blkg
,
45 unsigned long time
, unsigned long sectors
)
48 blkg
->sectors
+= sectors
;
50 EXPORT_SYMBOL_GPL(blkiocg_update_blkio_group_stats
);
52 void blkiocg_add_blkio_group(struct blkio_cgroup
*blkcg
,
53 struct blkio_group
*blkg
, void *key
, dev_t dev
)
57 spin_lock_irqsave(&blkcg
->lock
, flags
);
58 rcu_assign_pointer(blkg
->key
, key
);
59 blkg
->blkcg_id
= css_id(&blkcg
->css
);
60 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
61 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
62 #ifdef CONFIG_DEBUG_BLK_CGROUP
63 /* Need to take css reference ? */
64 cgroup_path(blkcg
->css
.cgroup
, blkg
->path
, sizeof(blkg
->path
));
68 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group
);
70 static void __blkiocg_del_blkio_group(struct blkio_group
*blkg
)
72 hlist_del_init_rcu(&blkg
->blkcg_node
);
77 * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
78 * indicating that blk_group was unhashed by the time we got to it.
80 int blkiocg_del_blkio_group(struct blkio_group
*blkg
)
82 struct blkio_cgroup
*blkcg
;
84 struct cgroup_subsys_state
*css
;
88 css
= css_lookup(&blkio_subsys
, blkg
->blkcg_id
);
92 blkcg
= container_of(css
, struct blkio_cgroup
, css
);
93 spin_lock_irqsave(&blkcg
->lock
, flags
);
94 if (!hlist_unhashed(&blkg
->blkcg_node
)) {
95 __blkiocg_del_blkio_group(blkg
);
98 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
103 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group
);
105 /* called under rcu_read_lock(). */
106 struct blkio_group
*blkiocg_lookup_group(struct blkio_cgroup
*blkcg
, void *key
)
108 struct blkio_group
*blkg
;
109 struct hlist_node
*n
;
112 hlist_for_each_entry_rcu(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
) {
120 EXPORT_SYMBOL_GPL(blkiocg_lookup_group
);
122 #define SHOW_FUNCTION(__VAR) \
123 static u64 blkiocg_##__VAR##_read(struct cgroup *cgroup, \
124 struct cftype *cftype) \
126 struct blkio_cgroup *blkcg; \
128 blkcg = cgroup_to_blkio_cgroup(cgroup); \
129 return (u64)blkcg->__VAR; \
132 SHOW_FUNCTION(weight
);
136 blkiocg_weight_write(struct cgroup
*cgroup
, struct cftype
*cftype
, u64 val
)
138 struct blkio_cgroup
*blkcg
;
139 struct blkio_group
*blkg
;
140 struct hlist_node
*n
;
142 if (val
< BLKIO_WEIGHT_MIN
|| val
> BLKIO_WEIGHT_MAX
)
145 blkcg
= cgroup_to_blkio_cgroup(cgroup
);
146 spin_lock_irq(&blkcg
->lock
);
147 blkcg
->weight
= (unsigned int)val
;
148 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
)
149 cfq_update_blkio_group_weight(blkg
, blkcg
->weight
);
150 spin_unlock_irq(&blkcg
->lock
);
154 #define SHOW_FUNCTION_PER_GROUP(__VAR) \
155 static int blkiocg_##__VAR##_read(struct cgroup *cgroup, \
156 struct cftype *cftype, struct seq_file *m) \
158 struct blkio_cgroup *blkcg; \
159 struct blkio_group *blkg; \
160 struct hlist_node *n; \
162 if (!cgroup_lock_live_group(cgroup)) \
165 blkcg = cgroup_to_blkio_cgroup(cgroup); \
167 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {\
169 seq_printf(m, "%u:%u %lu\n", MAJOR(blkg->dev), \
170 MINOR(blkg->dev), blkg->__VAR); \
177 SHOW_FUNCTION_PER_GROUP(time
);
178 SHOW_FUNCTION_PER_GROUP(sectors
);
179 #ifdef CONFIG_DEBUG_BLK_CGROUP
180 SHOW_FUNCTION_PER_GROUP(dequeue
);
182 #undef SHOW_FUNCTION_PER_GROUP
184 #ifdef CONFIG_DEBUG_BLK_CGROUP
185 void blkiocg_update_blkio_group_dequeue_stats(struct blkio_group
*blkg
,
186 unsigned long dequeue
)
188 blkg
->dequeue
+= dequeue
;
190 EXPORT_SYMBOL_GPL(blkiocg_update_blkio_group_dequeue_stats
);
193 struct cftype blkio_files
[] = {
196 .read_u64
= blkiocg_weight_read
,
197 .write_u64
= blkiocg_weight_write
,
201 .read_seq_string
= blkiocg_time_read
,
205 .read_seq_string
= blkiocg_sectors_read
,
207 #ifdef CONFIG_DEBUG_BLK_CGROUP
210 .read_seq_string
= blkiocg_dequeue_read
,
215 static int blkiocg_populate(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
)
217 return cgroup_add_files(cgroup
, subsys
, blkio_files
,
218 ARRAY_SIZE(blkio_files
));
221 static void blkiocg_destroy(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
)
223 struct blkio_cgroup
*blkcg
= cgroup_to_blkio_cgroup(cgroup
);
225 struct blkio_group
*blkg
;
230 spin_lock_irqsave(&blkcg
->lock
, flags
);
232 if (hlist_empty(&blkcg
->blkg_list
)) {
233 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
237 blkg
= hlist_entry(blkcg
->blkg_list
.first
, struct blkio_group
,
239 key
= rcu_dereference(blkg
->key
);
240 __blkiocg_del_blkio_group(blkg
);
242 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
245 * This blkio_group is being unlinked as associated cgroup is going
246 * away. Let all the IO controlling policies know about this event.
248 * Currently this is static call to one io controlling policy. Once
249 * we have more policies in place, we need some dynamic registration
250 * of callback function.
252 cfq_unlink_blkio_group(key
, blkg
);
255 free_css_id(&blkio_subsys
, &blkcg
->css
);
260 static struct cgroup_subsys_state
*
261 blkiocg_create(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
)
263 struct blkio_cgroup
*blkcg
, *parent_blkcg
;
265 if (!cgroup
->parent
) {
266 blkcg
= &blkio_root_cgroup
;
270 /* Currently we do not support hierarchy deeper than two level (0,1) */
271 parent_blkcg
= cgroup_to_blkio_cgroup(cgroup
->parent
);
272 if (css_depth(&parent_blkcg
->css
) > 0)
273 return ERR_PTR(-EINVAL
);
275 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
277 return ERR_PTR(-ENOMEM
);
279 blkcg
->weight
= BLKIO_WEIGHT_DEFAULT
;
281 spin_lock_init(&blkcg
->lock
);
282 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
288 * We cannot support shared io contexts, as we have no mean to support
289 * two tasks with the same ioc in two different groups without major rework
290 * of the main cic data structures. For now we allow a task to change
291 * its cgroup only if it's the only owner of its ioc.
293 static int blkiocg_can_attach(struct cgroup_subsys
*subsys
,
294 struct cgroup
*cgroup
, struct task_struct
*tsk
,
297 struct io_context
*ioc
;
300 /* task_lock() is needed to avoid races with exit_io_context() */
302 ioc
= tsk
->io_context
;
303 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
310 static void blkiocg_attach(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
,
311 struct cgroup
*prev
, struct task_struct
*tsk
,
314 struct io_context
*ioc
;
317 ioc
= tsk
->io_context
;
319 ioc
->cgroup_changed
= 1;
323 struct cgroup_subsys blkio_subsys
= {
325 .create
= blkiocg_create
,
326 .can_attach
= blkiocg_can_attach
,
327 .attach
= blkiocg_attach
,
328 .destroy
= blkiocg_destroy
,
329 .populate
= blkiocg_populate
,
330 .subsys_id
= blkio_subsys_id
,