Commit | Line | Data |
---|---|---|
31e4c28d VG |
1 | /* |
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> | |
22084190 VG |
14 | #include <linux/seq_file.h> |
15 | #include <linux/kdev_t.h> | |
31e4c28d VG |
16 | #include "blk-cgroup.h" |
17 | ||
b1c35769 | 18 | extern void cfq_unlink_blkio_group(void *, struct blkio_group *); |
f8d461d6 | 19 | extern void cfq_update_blkio_group_weight(struct blkio_group *, unsigned int); |
b1c35769 | 20 | |
31e4c28d VG |
21 | struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT }; |
22 | ||
23 | struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup) | |
24 | { | |
25 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), | |
26 | struct blkio_cgroup, css); | |
27 | } | |
28 | ||
22084190 VG |
29 | void blkiocg_update_blkio_group_stats(struct blkio_group *blkg, |
30 | unsigned long time, unsigned long sectors) | |
31 | { | |
32 | blkg->time += time; | |
33 | blkg->sectors += sectors; | |
34 | } | |
35 | ||
31e4c28d | 36 | void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg, |
22084190 | 37 | struct blkio_group *blkg, void *key, dev_t dev) |
31e4c28d VG |
38 | { |
39 | unsigned long flags; | |
40 | ||
41 | spin_lock_irqsave(&blkcg->lock, flags); | |
42 | rcu_assign_pointer(blkg->key, key); | |
b1c35769 | 43 | blkg->blkcg_id = css_id(&blkcg->css); |
31e4c28d VG |
44 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
45 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
2868ef7b VG |
46 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
47 | /* Need to take css reference ? */ | |
48 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); | |
49 | #endif | |
22084190 | 50 | blkg->dev = dev; |
31e4c28d VG |
51 | } |
52 | ||
b1c35769 VG |
53 | static void __blkiocg_del_blkio_group(struct blkio_group *blkg) |
54 | { | |
55 | hlist_del_init_rcu(&blkg->blkcg_node); | |
56 | blkg->blkcg_id = 0; | |
57 | } | |
58 | ||
59 | /* | |
60 | * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1 | |
61 | * indicating that blk_group was unhashed by the time we got to it. | |
62 | */ | |
31e4c28d VG |
63 | int blkiocg_del_blkio_group(struct blkio_group *blkg) |
64 | { | |
b1c35769 VG |
65 | struct blkio_cgroup *blkcg; |
66 | unsigned long flags; | |
67 | struct cgroup_subsys_state *css; | |
68 | int ret = 1; | |
69 | ||
70 | rcu_read_lock(); | |
71 | css = css_lookup(&blkio_subsys, blkg->blkcg_id); | |
72 | if (!css) | |
73 | goto out; | |
74 | ||
75 | blkcg = container_of(css, struct blkio_cgroup, css); | |
76 | spin_lock_irqsave(&blkcg->lock, flags); | |
77 | if (!hlist_unhashed(&blkg->blkcg_node)) { | |
78 | __blkiocg_del_blkio_group(blkg); | |
79 | ret = 0; | |
80 | } | |
81 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
82 | out: | |
83 | rcu_read_unlock(); | |
84 | return ret; | |
31e4c28d VG |
85 | } |
86 | ||
87 | /* called under rcu_read_lock(). */ | |
88 | struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) | |
89 | { | |
90 | struct blkio_group *blkg; | |
91 | struct hlist_node *n; | |
92 | void *__key; | |
93 | ||
94 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
95 | __key = blkg->key; | |
96 | if (__key == key) | |
97 | return blkg; | |
98 | } | |
99 | ||
100 | return NULL; | |
101 | } | |
102 | ||
103 | #define SHOW_FUNCTION(__VAR) \ | |
104 | static u64 blkiocg_##__VAR##_read(struct cgroup *cgroup, \ | |
105 | struct cftype *cftype) \ | |
106 | { \ | |
107 | struct blkio_cgroup *blkcg; \ | |
108 | \ | |
109 | blkcg = cgroup_to_blkio_cgroup(cgroup); \ | |
110 | return (u64)blkcg->__VAR; \ | |
111 | } | |
112 | ||
113 | SHOW_FUNCTION(weight); | |
114 | #undef SHOW_FUNCTION | |
115 | ||
116 | static int | |
117 | blkiocg_weight_write(struct cgroup *cgroup, struct cftype *cftype, u64 val) | |
118 | { | |
119 | struct blkio_cgroup *blkcg; | |
f8d461d6 VG |
120 | struct blkio_group *blkg; |
121 | struct hlist_node *n; | |
31e4c28d VG |
122 | |
123 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) | |
124 | return -EINVAL; | |
125 | ||
126 | blkcg = cgroup_to_blkio_cgroup(cgroup); | |
f8d461d6 | 127 | spin_lock_irq(&blkcg->lock); |
31e4c28d | 128 | blkcg->weight = (unsigned int)val; |
f8d461d6 VG |
129 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) |
130 | cfq_update_blkio_group_weight(blkg, blkcg->weight); | |
131 | spin_unlock_irq(&blkcg->lock); | |
31e4c28d VG |
132 | return 0; |
133 | } | |
134 | ||
22084190 VG |
135 | #define SHOW_FUNCTION_PER_GROUP(__VAR) \ |
136 | static int blkiocg_##__VAR##_read(struct cgroup *cgroup, \ | |
137 | struct cftype *cftype, struct seq_file *m) \ | |
138 | { \ | |
139 | struct blkio_cgroup *blkcg; \ | |
140 | struct blkio_group *blkg; \ | |
141 | struct hlist_node *n; \ | |
142 | \ | |
143 | if (!cgroup_lock_live_group(cgroup)) \ | |
144 | return -ENODEV; \ | |
145 | \ | |
146 | blkcg = cgroup_to_blkio_cgroup(cgroup); \ | |
147 | rcu_read_lock(); \ | |
148 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {\ | |
149 | if (blkg->dev) \ | |
150 | seq_printf(m, "%u:%u %lu\n", MAJOR(blkg->dev), \ | |
151 | MINOR(blkg->dev), blkg->__VAR); \ | |
152 | } \ | |
153 | rcu_read_unlock(); \ | |
154 | cgroup_unlock(); \ | |
155 | return 0; \ | |
156 | } | |
157 | ||
158 | SHOW_FUNCTION_PER_GROUP(time); | |
159 | SHOW_FUNCTION_PER_GROUP(sectors); | |
160 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
161 | SHOW_FUNCTION_PER_GROUP(dequeue); | |
162 | #endif | |
163 | #undef SHOW_FUNCTION_PER_GROUP | |
164 | ||
165 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
166 | void blkiocg_update_blkio_group_dequeue_stats(struct blkio_group *blkg, | |
167 | unsigned long dequeue) | |
168 | { | |
169 | blkg->dequeue += dequeue; | |
170 | } | |
171 | #endif | |
172 | ||
31e4c28d VG |
173 | struct cftype blkio_files[] = { |
174 | { | |
175 | .name = "weight", | |
176 | .read_u64 = blkiocg_weight_read, | |
177 | .write_u64 = blkiocg_weight_write, | |
178 | }, | |
22084190 VG |
179 | { |
180 | .name = "time", | |
181 | .read_seq_string = blkiocg_time_read, | |
182 | }, | |
183 | { | |
184 | .name = "sectors", | |
185 | .read_seq_string = blkiocg_sectors_read, | |
186 | }, | |
187 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
188 | { | |
189 | .name = "dequeue", | |
190 | .read_seq_string = blkiocg_dequeue_read, | |
191 | }, | |
192 | #endif | |
31e4c28d VG |
193 | }; |
194 | ||
195 | static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
196 | { | |
197 | return cgroup_add_files(cgroup, subsys, blkio_files, | |
198 | ARRAY_SIZE(blkio_files)); | |
199 | } | |
200 | ||
201 | static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
202 | { | |
203 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
b1c35769 VG |
204 | unsigned long flags; |
205 | struct blkio_group *blkg; | |
206 | void *key; | |
207 | ||
208 | rcu_read_lock(); | |
209 | remove_entry: | |
210 | spin_lock_irqsave(&blkcg->lock, flags); | |
211 | ||
212 | if (hlist_empty(&blkcg->blkg_list)) { | |
213 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
214 | goto done; | |
215 | } | |
216 | ||
217 | blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group, | |
218 | blkcg_node); | |
219 | key = rcu_dereference(blkg->key); | |
220 | __blkiocg_del_blkio_group(blkg); | |
31e4c28d | 221 | |
b1c35769 VG |
222 | spin_unlock_irqrestore(&blkcg->lock, flags); |
223 | ||
224 | /* | |
225 | * This blkio_group is being unlinked as associated cgroup is going | |
226 | * away. Let all the IO controlling policies know about this event. | |
227 | * | |
228 | * Currently this is static call to one io controlling policy. Once | |
229 | * we have more policies in place, we need some dynamic registration | |
230 | * of callback function. | |
231 | */ | |
232 | cfq_unlink_blkio_group(key, blkg); | |
233 | goto remove_entry; | |
234 | done: | |
31e4c28d | 235 | free_css_id(&blkio_subsys, &blkcg->css); |
b1c35769 | 236 | rcu_read_unlock(); |
31e4c28d VG |
237 | kfree(blkcg); |
238 | } | |
239 | ||
240 | static struct cgroup_subsys_state * | |
241 | blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
242 | { | |
243 | struct blkio_cgroup *blkcg, *parent_blkcg; | |
244 | ||
245 | if (!cgroup->parent) { | |
246 | blkcg = &blkio_root_cgroup; | |
247 | goto done; | |
248 | } | |
249 | ||
250 | /* Currently we do not support hierarchy deeper than two level (0,1) */ | |
251 | parent_blkcg = cgroup_to_blkio_cgroup(cgroup->parent); | |
252 | if (css_depth(&parent_blkcg->css) > 0) | |
253 | return ERR_PTR(-EINVAL); | |
254 | ||
255 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
256 | if (!blkcg) | |
257 | return ERR_PTR(-ENOMEM); | |
258 | ||
259 | blkcg->weight = BLKIO_WEIGHT_DEFAULT; | |
260 | done: | |
261 | spin_lock_init(&blkcg->lock); | |
262 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
263 | ||
264 | return &blkcg->css; | |
265 | } | |
266 | ||
267 | /* | |
268 | * We cannot support shared io contexts, as we have no mean to support | |
269 | * two tasks with the same ioc in two different groups without major rework | |
270 | * of the main cic data structures. For now we allow a task to change | |
271 | * its cgroup only if it's the only owner of its ioc. | |
272 | */ | |
273 | static int blkiocg_can_attach(struct cgroup_subsys *subsys, | |
274 | struct cgroup *cgroup, struct task_struct *tsk, | |
275 | bool threadgroup) | |
276 | { | |
277 | struct io_context *ioc; | |
278 | int ret = 0; | |
279 | ||
280 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
281 | task_lock(tsk); | |
282 | ioc = tsk->io_context; | |
283 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
284 | ret = -EINVAL; | |
285 | task_unlock(tsk); | |
286 | ||
287 | return ret; | |
288 | } | |
289 | ||
290 | static void blkiocg_attach(struct cgroup_subsys *subsys, struct cgroup *cgroup, | |
291 | struct cgroup *prev, struct task_struct *tsk, | |
292 | bool threadgroup) | |
293 | { | |
294 | struct io_context *ioc; | |
295 | ||
296 | task_lock(tsk); | |
297 | ioc = tsk->io_context; | |
298 | if (ioc) | |
299 | ioc->cgroup_changed = 1; | |
300 | task_unlock(tsk); | |
301 | } | |
302 | ||
303 | struct cgroup_subsys blkio_subsys = { | |
304 | .name = "blkio", | |
305 | .create = blkiocg_create, | |
306 | .can_attach = blkiocg_can_attach, | |
307 | .attach = blkiocg_attach, | |
308 | .destroy = blkiocg_destroy, | |
309 | .populate = blkiocg_populate, | |
310 | .subsys_id = blkio_subsys_id, | |
311 | .use_id = 1, | |
312 | }; |