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 "blk-cgroup.h"
16 extern void cfq_unlink_blkio_group(void *, struct blkio_group
*);
18 struct blkio_cgroup blkio_root_cgroup
= { .weight
= 2*BLKIO_WEIGHT_DEFAULT
};
20 struct blkio_cgroup
*cgroup_to_blkio_cgroup(struct cgroup
*cgroup
)
22 return container_of(cgroup_subsys_state(cgroup
, blkio_subsys_id
),
23 struct blkio_cgroup
, css
);
26 void blkiocg_add_blkio_group(struct blkio_cgroup
*blkcg
,
27 struct blkio_group
*blkg
, void *key
)
31 spin_lock_irqsave(&blkcg
->lock
, flags
);
32 rcu_assign_pointer(blkg
->key
, key
);
33 blkg
->blkcg_id
= css_id(&blkcg
->css
);
34 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
35 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
38 static void __blkiocg_del_blkio_group(struct blkio_group
*blkg
)
40 hlist_del_init_rcu(&blkg
->blkcg_node
);
45 * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
46 * indicating that blk_group was unhashed by the time we got to it.
48 int blkiocg_del_blkio_group(struct blkio_group
*blkg
)
50 struct blkio_cgroup
*blkcg
;
52 struct cgroup_subsys_state
*css
;
56 css
= css_lookup(&blkio_subsys
, blkg
->blkcg_id
);
60 blkcg
= container_of(css
, struct blkio_cgroup
, css
);
61 spin_lock_irqsave(&blkcg
->lock
, flags
);
62 if (!hlist_unhashed(&blkg
->blkcg_node
)) {
63 __blkiocg_del_blkio_group(blkg
);
66 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
72 /* called under rcu_read_lock(). */
73 struct blkio_group
*blkiocg_lookup_group(struct blkio_cgroup
*blkcg
, void *key
)
75 struct blkio_group
*blkg
;
79 hlist_for_each_entry_rcu(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
) {
88 #define SHOW_FUNCTION(__VAR) \
89 static u64 blkiocg_##__VAR##_read(struct cgroup *cgroup, \
90 struct cftype *cftype) \
92 struct blkio_cgroup *blkcg; \
94 blkcg = cgroup_to_blkio_cgroup(cgroup); \
95 return (u64)blkcg->__VAR; \
98 SHOW_FUNCTION(weight
);
102 blkiocg_weight_write(struct cgroup
*cgroup
, struct cftype
*cftype
, u64 val
)
104 struct blkio_cgroup
*blkcg
;
106 if (val
< BLKIO_WEIGHT_MIN
|| val
> BLKIO_WEIGHT_MAX
)
109 blkcg
= cgroup_to_blkio_cgroup(cgroup
);
110 blkcg
->weight
= (unsigned int)val
;
114 struct cftype blkio_files
[] = {
117 .read_u64
= blkiocg_weight_read
,
118 .write_u64
= blkiocg_weight_write
,
122 static int blkiocg_populate(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
)
124 return cgroup_add_files(cgroup
, subsys
, blkio_files
,
125 ARRAY_SIZE(blkio_files
));
128 static void blkiocg_destroy(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
)
130 struct blkio_cgroup
*blkcg
= cgroup_to_blkio_cgroup(cgroup
);
132 struct blkio_group
*blkg
;
137 spin_lock_irqsave(&blkcg
->lock
, flags
);
139 if (hlist_empty(&blkcg
->blkg_list
)) {
140 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
144 blkg
= hlist_entry(blkcg
->blkg_list
.first
, struct blkio_group
,
146 key
= rcu_dereference(blkg
->key
);
147 __blkiocg_del_blkio_group(blkg
);
149 spin_unlock_irqrestore(&blkcg
->lock
, flags
);
152 * This blkio_group is being unlinked as associated cgroup is going
153 * away. Let all the IO controlling policies know about this event.
155 * Currently this is static call to one io controlling policy. Once
156 * we have more policies in place, we need some dynamic registration
157 * of callback function.
159 cfq_unlink_blkio_group(key
, blkg
);
162 free_css_id(&blkio_subsys
, &blkcg
->css
);
167 static struct cgroup_subsys_state
*
168 blkiocg_create(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
)
170 struct blkio_cgroup
*blkcg
, *parent_blkcg
;
172 if (!cgroup
->parent
) {
173 blkcg
= &blkio_root_cgroup
;
177 /* Currently we do not support hierarchy deeper than two level (0,1) */
178 parent_blkcg
= cgroup_to_blkio_cgroup(cgroup
->parent
);
179 if (css_depth(&parent_blkcg
->css
) > 0)
180 return ERR_PTR(-EINVAL
);
182 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
184 return ERR_PTR(-ENOMEM
);
186 blkcg
->weight
= BLKIO_WEIGHT_DEFAULT
;
188 spin_lock_init(&blkcg
->lock
);
189 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
195 * We cannot support shared io contexts, as we have no mean to support
196 * two tasks with the same ioc in two different groups without major rework
197 * of the main cic data structures. For now we allow a task to change
198 * its cgroup only if it's the only owner of its ioc.
200 static int blkiocg_can_attach(struct cgroup_subsys
*subsys
,
201 struct cgroup
*cgroup
, struct task_struct
*tsk
,
204 struct io_context
*ioc
;
207 /* task_lock() is needed to avoid races with exit_io_context() */
209 ioc
= tsk
->io_context
;
210 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
217 static void blkiocg_attach(struct cgroup_subsys
*subsys
, struct cgroup
*cgroup
,
218 struct cgroup
*prev
, struct task_struct
*tsk
,
221 struct io_context
*ioc
;
224 ioc
= tsk
->io_context
;
226 ioc
->cgroup_changed
= 1;
230 struct cgroup_subsys blkio_subsys
= {
232 .create
= blkiocg_create
,
233 .can_attach
= blkiocg_can_attach
,
234 .attach
= blkiocg_attach
,
235 .destroy
= blkiocg_destroy
,
236 .populate
= blkiocg_populate
,
237 .subsys_id
= blkio_subsys_id
,