Leave mangled flag only for setting nd.intent.open.flag
[linux-2.6.git] / block / blk-cgroup.c
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>
14 #include <linux/seq_file.h>
15 #include <linux/kdev_t.h>
16 #include <linux/module.h>
17 #include <linux/err.h>
18 #include "blk-cgroup.h"
19
20 static DEFINE_SPINLOCK(blkio_list_lock);
21 static LIST_HEAD(blkio_list);
22
23 struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT };
24 EXPORT_SYMBOL_GPL(blkio_root_cgroup);
25
26 struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
27 {
28         return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
29                             struct blkio_cgroup, css);
30 }
31 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
32
33 void blkiocg_update_blkio_group_stats(struct blkio_group *blkg,
34                         unsigned long time, unsigned long sectors)
35 {
36         blkg->time += time;
37         blkg->sectors += sectors;
38 }
39 EXPORT_SYMBOL_GPL(blkiocg_update_blkio_group_stats);
40
41 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
42                         struct blkio_group *blkg, void *key, dev_t dev)
43 {
44         unsigned long flags;
45
46         spin_lock_irqsave(&blkcg->lock, flags);
47         rcu_assign_pointer(blkg->key, key);
48         blkg->blkcg_id = css_id(&blkcg->css);
49         hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
50         spin_unlock_irqrestore(&blkcg->lock, flags);
51 #ifdef CONFIG_DEBUG_BLK_CGROUP
52         /* Need to take css reference ? */
53         cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
54 #endif
55         blkg->dev = dev;
56 }
57 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
58
59 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
60 {
61         hlist_del_init_rcu(&blkg->blkcg_node);
62         blkg->blkcg_id = 0;
63 }
64
65 /*
66  * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
67  * indicating that blk_group was unhashed by the time we got to it.
68  */
69 int blkiocg_del_blkio_group(struct blkio_group *blkg)
70 {
71         struct blkio_cgroup *blkcg;
72         unsigned long flags;
73         struct cgroup_subsys_state *css;
74         int ret = 1;
75
76         rcu_read_lock();
77         css = css_lookup(&blkio_subsys, blkg->blkcg_id);
78         if (!css)
79                 goto out;
80
81         blkcg = container_of(css, struct blkio_cgroup, css);
82         spin_lock_irqsave(&blkcg->lock, flags);
83         if (!hlist_unhashed(&blkg->blkcg_node)) {
84                 __blkiocg_del_blkio_group(blkg);
85                 ret = 0;
86         }
87         spin_unlock_irqrestore(&blkcg->lock, flags);
88 out:
89         rcu_read_unlock();
90         return ret;
91 }
92 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
93
94 /* called under rcu_read_lock(). */
95 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
96 {
97         struct blkio_group *blkg;
98         struct hlist_node *n;
99         void *__key;
100
101         hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
102                 __key = blkg->key;
103                 if (__key == key)
104                         return blkg;
105         }
106
107         return NULL;
108 }
109 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
110
111 #define SHOW_FUNCTION(__VAR)                                            \
112 static u64 blkiocg_##__VAR##_read(struct cgroup *cgroup,                \
113                                        struct cftype *cftype)           \
114 {                                                                       \
115         struct blkio_cgroup *blkcg;                                     \
116                                                                         \
117         blkcg = cgroup_to_blkio_cgroup(cgroup);                         \
118         return (u64)blkcg->__VAR;                                       \
119 }
120
121 SHOW_FUNCTION(weight);
122 #undef SHOW_FUNCTION
123
124 static int
125 blkiocg_weight_write(struct cgroup *cgroup, struct cftype *cftype, u64 val)
126 {
127         struct blkio_cgroup *blkcg;
128         struct blkio_group *blkg;
129         struct hlist_node *n;
130         struct blkio_policy_type *blkiop;
131
132         if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
133                 return -EINVAL;
134
135         blkcg = cgroup_to_blkio_cgroup(cgroup);
136         spin_lock(&blkio_list_lock);
137         spin_lock_irq(&blkcg->lock);
138         blkcg->weight = (unsigned int)val;
139         hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
140                 list_for_each_entry(blkiop, &blkio_list, list)
141                         blkiop->ops.blkio_update_group_weight_fn(blkg,
142                                         blkcg->weight);
143         }
144         spin_unlock_irq(&blkcg->lock);
145         spin_unlock(&blkio_list_lock);
146         return 0;
147 }
148
149 #define SHOW_FUNCTION_PER_GROUP(__VAR)                                  \
150 static int blkiocg_##__VAR##_read(struct cgroup *cgroup,                \
151                         struct cftype *cftype, struct seq_file *m)      \
152 {                                                                       \
153         struct blkio_cgroup *blkcg;                                     \
154         struct blkio_group *blkg;                                       \
155         struct hlist_node *n;                                           \
156                                                                         \
157         if (!cgroup_lock_live_group(cgroup))                            \
158                 return -ENODEV;                                         \
159                                                                         \
160         blkcg = cgroup_to_blkio_cgroup(cgroup);                         \
161         rcu_read_lock();                                                \
162         hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {\
163                 if (blkg->dev)                                          \
164                         seq_printf(m, "%u:%u %lu\n", MAJOR(blkg->dev),  \
165                                  MINOR(blkg->dev), blkg->__VAR);        \
166         }                                                               \
167         rcu_read_unlock();                                              \
168         cgroup_unlock();                                                \
169         return 0;                                                       \
170 }
171
172 SHOW_FUNCTION_PER_GROUP(time);
173 SHOW_FUNCTION_PER_GROUP(sectors);
174 #ifdef CONFIG_DEBUG_BLK_CGROUP
175 SHOW_FUNCTION_PER_GROUP(dequeue);
176 #endif
177 #undef SHOW_FUNCTION_PER_GROUP
178
179 #ifdef CONFIG_DEBUG_BLK_CGROUP
180 void blkiocg_update_blkio_group_dequeue_stats(struct blkio_group *blkg,
181                         unsigned long dequeue)
182 {
183         blkg->dequeue += dequeue;
184 }
185 EXPORT_SYMBOL_GPL(blkiocg_update_blkio_group_dequeue_stats);
186 #endif
187
188 struct cftype blkio_files[] = {
189         {
190                 .name = "weight",
191                 .read_u64 = blkiocg_weight_read,
192                 .write_u64 = blkiocg_weight_write,
193         },
194         {
195                 .name = "time",
196                 .read_seq_string = blkiocg_time_read,
197         },
198         {
199                 .name = "sectors",
200                 .read_seq_string = blkiocg_sectors_read,
201         },
202 #ifdef CONFIG_DEBUG_BLK_CGROUP
203        {
204                 .name = "dequeue",
205                 .read_seq_string = blkiocg_dequeue_read,
206        },
207 #endif
208 };
209
210 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
211 {
212         return cgroup_add_files(cgroup, subsys, blkio_files,
213                                 ARRAY_SIZE(blkio_files));
214 }
215
216 static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup)
217 {
218         struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
219         unsigned long flags;
220         struct blkio_group *blkg;
221         void *key;
222         struct blkio_policy_type *blkiop;
223
224         rcu_read_lock();
225 remove_entry:
226         spin_lock_irqsave(&blkcg->lock, flags);
227
228         if (hlist_empty(&blkcg->blkg_list)) {
229                 spin_unlock_irqrestore(&blkcg->lock, flags);
230                 goto done;
231         }
232
233         blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
234                                 blkcg_node);
235         key = rcu_dereference(blkg->key);
236         __blkiocg_del_blkio_group(blkg);
237
238         spin_unlock_irqrestore(&blkcg->lock, flags);
239
240         /*
241          * This blkio_group is being unlinked as associated cgroup is going
242          * away. Let all the IO controlling policies know about this event.
243          *
244          * Currently this is static call to one io controlling policy. Once
245          * we have more policies in place, we need some dynamic registration
246          * of callback function.
247          */
248         spin_lock(&blkio_list_lock);
249         list_for_each_entry(blkiop, &blkio_list, list)
250                 blkiop->ops.blkio_unlink_group_fn(key, blkg);
251         spin_unlock(&blkio_list_lock);
252         goto remove_entry;
253 done:
254         free_css_id(&blkio_subsys, &blkcg->css);
255         rcu_read_unlock();
256         kfree(blkcg);
257 }
258
259 static struct cgroup_subsys_state *
260 blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup)
261 {
262         struct blkio_cgroup *blkcg, *parent_blkcg;
263
264         if (!cgroup->parent) {
265                 blkcg = &blkio_root_cgroup;
266                 goto done;
267         }
268
269         /* Currently we do not support hierarchy deeper than two level (0,1) */
270         parent_blkcg = cgroup_to_blkio_cgroup(cgroup->parent);
271         if (css_depth(&parent_blkcg->css) > 0)
272                 return ERR_PTR(-EINVAL);
273
274         blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
275         if (!blkcg)
276                 return ERR_PTR(-ENOMEM);
277
278         blkcg->weight = BLKIO_WEIGHT_DEFAULT;
279 done:
280         spin_lock_init(&blkcg->lock);
281         INIT_HLIST_HEAD(&blkcg->blkg_list);
282
283         return &blkcg->css;
284 }
285
286 /*
287  * We cannot support shared io contexts, as we have no mean to support
288  * two tasks with the same ioc in two different groups without major rework
289  * of the main cic data structures.  For now we allow a task to change
290  * its cgroup only if it's the only owner of its ioc.
291  */
292 static int blkiocg_can_attach(struct cgroup_subsys *subsys,
293                                 struct cgroup *cgroup, struct task_struct *tsk,
294                                 bool threadgroup)
295 {
296         struct io_context *ioc;
297         int ret = 0;
298
299         /* task_lock() is needed to avoid races with exit_io_context() */
300         task_lock(tsk);
301         ioc = tsk->io_context;
302         if (ioc && atomic_read(&ioc->nr_tasks) > 1)
303                 ret = -EINVAL;
304         task_unlock(tsk);
305
306         return ret;
307 }
308
309 static void blkiocg_attach(struct cgroup_subsys *subsys, struct cgroup *cgroup,
310                                 struct cgroup *prev, struct task_struct *tsk,
311                                 bool threadgroup)
312 {
313         struct io_context *ioc;
314
315         task_lock(tsk);
316         ioc = tsk->io_context;
317         if (ioc)
318                 ioc->cgroup_changed = 1;
319         task_unlock(tsk);
320 }
321
322 struct cgroup_subsys blkio_subsys = {
323         .name = "blkio",
324         .create = blkiocg_create,
325         .can_attach = blkiocg_can_attach,
326         .attach = blkiocg_attach,
327         .destroy = blkiocg_destroy,
328         .populate = blkiocg_populate,
329         .subsys_id = blkio_subsys_id,
330         .use_id = 1,
331 };
332
333 void blkio_policy_register(struct blkio_policy_type *blkiop)
334 {
335         spin_lock(&blkio_list_lock);
336         list_add_tail(&blkiop->list, &blkio_list);
337         spin_unlock(&blkio_list_lock);
338 }
339 EXPORT_SYMBOL_GPL(blkio_policy_register);
340
341 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
342 {
343         spin_lock(&blkio_list_lock);
344         list_del_init(&blkiop->list);
345         spin_unlock(&blkio_list_lock);
346 }
347 EXPORT_SYMBOL_GPL(blkio_policy_unregister);