]>
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> | |
9d6a986c | 16 | #include <linux/module.h> |
accee785 | 17 | #include <linux/err.h> |
9195291e | 18 | #include <linux/blkdev.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
31e4c28d | 20 | #include "blk-cgroup.h" |
34d0f179 | 21 | #include <linux/genhd.h> |
3e252066 | 22 | |
84c124da DS |
23 | #define MAX_KEY_LEN 100 |
24 | ||
3e252066 VG |
25 | static DEFINE_SPINLOCK(blkio_list_lock); |
26 | static LIST_HEAD(blkio_list); | |
b1c35769 | 27 | |
31e4c28d | 28 | struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT }; |
9d6a986c VG |
29 | EXPORT_SYMBOL_GPL(blkio_root_cgroup); |
30 | ||
67523c48 BB |
31 | static struct cgroup_subsys_state *blkiocg_create(struct cgroup_subsys *, |
32 | struct cgroup *); | |
33 | static int blkiocg_can_attach(struct cgroup_subsys *, struct cgroup *, | |
34 | struct task_struct *, bool); | |
35 | static void blkiocg_attach(struct cgroup_subsys *, struct cgroup *, | |
36 | struct cgroup *, struct task_struct *, bool); | |
37 | static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *); | |
38 | static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *); | |
39 | ||
40 | struct cgroup_subsys blkio_subsys = { | |
41 | .name = "blkio", | |
42 | .create = blkiocg_create, | |
43 | .can_attach = blkiocg_can_attach, | |
44 | .attach = blkiocg_attach, | |
45 | .destroy = blkiocg_destroy, | |
46 | .populate = blkiocg_populate, | |
47 | #ifdef CONFIG_BLK_CGROUP | |
48 | /* note: blkio_subsys_id is otherwise defined in blk-cgroup.h */ | |
49 | .subsys_id = blkio_subsys_id, | |
50 | #endif | |
51 | .use_id = 1, | |
52 | .module = THIS_MODULE, | |
53 | }; | |
54 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
55 | ||
34d0f179 GJ |
56 | static inline void blkio_policy_insert_node(struct blkio_cgroup *blkcg, |
57 | struct blkio_policy_node *pn) | |
58 | { | |
59 | list_add(&pn->node, &blkcg->policy_list); | |
60 | } | |
61 | ||
62 | /* Must be called with blkcg->lock held */ | |
63 | static inline void blkio_policy_delete_node(struct blkio_policy_node *pn) | |
64 | { | |
65 | list_del(&pn->node); | |
66 | } | |
67 | ||
68 | /* Must be called with blkcg->lock held */ | |
69 | static struct blkio_policy_node * | |
70 | blkio_policy_search_node(const struct blkio_cgroup *blkcg, dev_t dev) | |
71 | { | |
72 | struct blkio_policy_node *pn; | |
73 | ||
74 | list_for_each_entry(pn, &blkcg->policy_list, node) { | |
75 | if (pn->dev == dev) | |
76 | return pn; | |
77 | } | |
78 | ||
79 | return NULL; | |
80 | } | |
81 | ||
31e4c28d VG |
82 | struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup) |
83 | { | |
84 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), | |
85 | struct blkio_cgroup, css); | |
86 | } | |
9d6a986c | 87 | EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup); |
31e4c28d | 88 | |
84c124da DS |
89 | void blkio_group_init(struct blkio_group *blkg) |
90 | { | |
91 | spin_lock_init(&blkg->stats_lock); | |
92 | } | |
93 | EXPORT_SYMBOL_GPL(blkio_group_init); | |
94 | ||
9195291e DS |
95 | /* |
96 | * Add to the appropriate stat variable depending on the request type. | |
97 | * This should be called with the blkg->stats_lock held. | |
98 | */ | |
84c124da DS |
99 | static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction, |
100 | bool sync) | |
9195291e | 101 | { |
84c124da DS |
102 | if (direction) |
103 | stat[BLKIO_STAT_WRITE] += add; | |
9195291e | 104 | else |
84c124da DS |
105 | stat[BLKIO_STAT_READ] += add; |
106 | if (sync) | |
107 | stat[BLKIO_STAT_SYNC] += add; | |
9195291e | 108 | else |
84c124da | 109 | stat[BLKIO_STAT_ASYNC] += add; |
9195291e DS |
110 | } |
111 | ||
cdc1184c DS |
112 | /* |
113 | * Decrements the appropriate stat variable if non-zero depending on the | |
114 | * request type. Panics on value being zero. | |
115 | * This should be called with the blkg->stats_lock held. | |
116 | */ | |
117 | static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync) | |
118 | { | |
119 | if (direction) { | |
120 | BUG_ON(stat[BLKIO_STAT_WRITE] == 0); | |
121 | stat[BLKIO_STAT_WRITE]--; | |
122 | } else { | |
123 | BUG_ON(stat[BLKIO_STAT_READ] == 0); | |
124 | stat[BLKIO_STAT_READ]--; | |
125 | } | |
126 | if (sync) { | |
127 | BUG_ON(stat[BLKIO_STAT_SYNC] == 0); | |
128 | stat[BLKIO_STAT_SYNC]--; | |
129 | } else { | |
130 | BUG_ON(stat[BLKIO_STAT_ASYNC] == 0); | |
131 | stat[BLKIO_STAT_ASYNC]--; | |
132 | } | |
133 | } | |
134 | ||
135 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
812df48d DS |
136 | /* This should be called with the blkg->stats_lock held. */ |
137 | static void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
138 | struct blkio_group *curr_blkg) | |
139 | { | |
140 | if (blkio_blkg_waiting(&blkg->stats)) | |
141 | return; | |
142 | if (blkg == curr_blkg) | |
143 | return; | |
144 | blkg->stats.start_group_wait_time = sched_clock(); | |
145 | blkio_mark_blkg_waiting(&blkg->stats); | |
146 | } | |
147 | ||
148 | /* This should be called with the blkg->stats_lock held. */ | |
149 | static void blkio_update_group_wait_time(struct blkio_group_stats *stats) | |
150 | { | |
151 | unsigned long long now; | |
152 | ||
153 | if (!blkio_blkg_waiting(stats)) | |
154 | return; | |
155 | ||
156 | now = sched_clock(); | |
157 | if (time_after64(now, stats->start_group_wait_time)) | |
158 | stats->group_wait_time += now - stats->start_group_wait_time; | |
159 | blkio_clear_blkg_waiting(stats); | |
160 | } | |
161 | ||
162 | /* This should be called with the blkg->stats_lock held. */ | |
163 | static void blkio_end_empty_time(struct blkio_group_stats *stats) | |
164 | { | |
165 | unsigned long long now; | |
166 | ||
167 | if (!blkio_blkg_empty(stats)) | |
168 | return; | |
169 | ||
170 | now = sched_clock(); | |
171 | if (time_after64(now, stats->start_empty_time)) | |
172 | stats->empty_time += now - stats->start_empty_time; | |
173 | blkio_clear_blkg_empty(stats); | |
174 | } | |
175 | ||
176 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg) | |
177 | { | |
178 | unsigned long flags; | |
179 | ||
180 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
181 | BUG_ON(blkio_blkg_idling(&blkg->stats)); | |
182 | blkg->stats.start_idle_time = sched_clock(); | |
183 | blkio_mark_blkg_idling(&blkg->stats); | |
184 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
185 | } | |
186 | EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats); | |
187 | ||
188 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg) | |
189 | { | |
190 | unsigned long flags; | |
191 | unsigned long long now; | |
192 | struct blkio_group_stats *stats; | |
193 | ||
194 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
195 | stats = &blkg->stats; | |
196 | if (blkio_blkg_idling(stats)) { | |
197 | now = sched_clock(); | |
198 | if (time_after64(now, stats->start_idle_time)) | |
199 | stats->idle_time += now - stats->start_idle_time; | |
200 | blkio_clear_blkg_idling(stats); | |
201 | } | |
202 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
203 | } | |
204 | EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats); | |
205 | ||
a11cdaa7 | 206 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg) |
cdc1184c DS |
207 | { |
208 | unsigned long flags; | |
209 | struct blkio_group_stats *stats; | |
210 | ||
211 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
212 | stats = &blkg->stats; | |
213 | stats->avg_queue_size_sum += | |
214 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] + | |
215 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]; | |
216 | stats->avg_queue_size_samples++; | |
812df48d | 217 | blkio_update_group_wait_time(stats); |
cdc1184c DS |
218 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
219 | } | |
a11cdaa7 DS |
220 | EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats); |
221 | ||
28baf442 DS |
222 | void blkiocg_set_start_empty_time(struct blkio_group *blkg, bool ignore) |
223 | { | |
224 | unsigned long flags; | |
225 | struct blkio_group_stats *stats; | |
226 | ||
227 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
228 | stats = &blkg->stats; | |
229 | ||
230 | if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] || | |
231 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) { | |
232 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
233 | return; | |
234 | } | |
235 | ||
236 | /* | |
237 | * If ignore is set, we do not panic on the empty flag being set | |
238 | * already. This is to avoid cases where there are superfluous timeslice | |
239 | * complete events (for eg., forced_dispatch in CFQ) when no IOs are | |
240 | * served which could result in triggering the empty check incorrectly. | |
241 | */ | |
242 | BUG_ON(!ignore && blkio_blkg_empty(stats)); | |
243 | stats->start_empty_time = sched_clock(); | |
244 | blkio_mark_blkg_empty(stats); | |
245 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
246 | } | |
247 | EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time); | |
248 | ||
a11cdaa7 DS |
249 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
250 | unsigned long dequeue) | |
251 | { | |
252 | blkg->stats.dequeue += dequeue; | |
253 | } | |
254 | EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats); | |
812df48d DS |
255 | #else |
256 | static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
257 | struct blkio_group *curr_blkg) {} | |
258 | static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {} | |
cdc1184c DS |
259 | #endif |
260 | ||
a11cdaa7 | 261 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c DS |
262 | struct blkio_group *curr_blkg, bool direction, |
263 | bool sync) | |
264 | { | |
265 | unsigned long flags; | |
266 | ||
267 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
268 | blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction, | |
269 | sync); | |
812df48d DS |
270 | blkio_end_empty_time(&blkg->stats); |
271 | blkio_set_start_group_wait_time(blkg, curr_blkg); | |
cdc1184c DS |
272 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
273 | } | |
a11cdaa7 | 274 | EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats); |
cdc1184c | 275 | |
a11cdaa7 | 276 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c DS |
277 | bool direction, bool sync) |
278 | { | |
279 | unsigned long flags; | |
280 | ||
281 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
282 | blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], | |
283 | direction, sync); | |
284 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
285 | } | |
a11cdaa7 | 286 | EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats); |
cdc1184c | 287 | |
303a3acb | 288 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time) |
22084190 | 289 | { |
303a3acb DS |
290 | unsigned long flags; |
291 | ||
292 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
293 | blkg->stats.time += time; | |
294 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
22084190 | 295 | } |
303a3acb | 296 | EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used); |
22084190 | 297 | |
84c124da DS |
298 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
299 | uint64_t bytes, bool direction, bool sync) | |
9195291e DS |
300 | { |
301 | struct blkio_group_stats *stats; | |
302 | unsigned long flags; | |
303 | ||
304 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
305 | stats = &blkg->stats; | |
84c124da DS |
306 | stats->sectors += bytes >> 9; |
307 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICED], 1, direction, | |
308 | sync); | |
309 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_BYTES], bytes, | |
310 | direction, sync); | |
9195291e DS |
311 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
312 | } | |
84c124da | 313 | EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats); |
9195291e | 314 | |
84c124da DS |
315 | void blkiocg_update_completion_stats(struct blkio_group *blkg, |
316 | uint64_t start_time, uint64_t io_start_time, bool direction, bool sync) | |
9195291e DS |
317 | { |
318 | struct blkio_group_stats *stats; | |
319 | unsigned long flags; | |
320 | unsigned long long now = sched_clock(); | |
321 | ||
322 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
323 | stats = &blkg->stats; | |
84c124da DS |
324 | if (time_after64(now, io_start_time)) |
325 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME], | |
326 | now - io_start_time, direction, sync); | |
327 | if (time_after64(io_start_time, start_time)) | |
328 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME], | |
329 | io_start_time - start_time, direction, sync); | |
9195291e DS |
330 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
331 | } | |
84c124da | 332 | EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats); |
9195291e | 333 | |
812d4026 DS |
334 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction, |
335 | bool sync) | |
336 | { | |
337 | unsigned long flags; | |
338 | ||
339 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
340 | blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_MERGED], 1, direction, | |
341 | sync); | |
342 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
343 | } | |
344 | EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats); | |
345 | ||
31e4c28d | 346 | void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg, |
22084190 | 347 | struct blkio_group *blkg, void *key, dev_t dev) |
31e4c28d VG |
348 | { |
349 | unsigned long flags; | |
350 | ||
351 | spin_lock_irqsave(&blkcg->lock, flags); | |
352 | rcu_assign_pointer(blkg->key, key); | |
b1c35769 | 353 | blkg->blkcg_id = css_id(&blkcg->css); |
31e4c28d VG |
354 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
355 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
2868ef7b VG |
356 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
357 | /* Need to take css reference ? */ | |
358 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); | |
359 | #endif | |
22084190 | 360 | blkg->dev = dev; |
31e4c28d | 361 | } |
9d6a986c | 362 | EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group); |
31e4c28d | 363 | |
b1c35769 VG |
364 | static void __blkiocg_del_blkio_group(struct blkio_group *blkg) |
365 | { | |
366 | hlist_del_init_rcu(&blkg->blkcg_node); | |
367 | blkg->blkcg_id = 0; | |
368 | } | |
369 | ||
370 | /* | |
371 | * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1 | |
372 | * indicating that blk_group was unhashed by the time we got to it. | |
373 | */ | |
31e4c28d VG |
374 | int blkiocg_del_blkio_group(struct blkio_group *blkg) |
375 | { | |
b1c35769 VG |
376 | struct blkio_cgroup *blkcg; |
377 | unsigned long flags; | |
378 | struct cgroup_subsys_state *css; | |
379 | int ret = 1; | |
380 | ||
381 | rcu_read_lock(); | |
382 | css = css_lookup(&blkio_subsys, blkg->blkcg_id); | |
383 | if (!css) | |
384 | goto out; | |
385 | ||
386 | blkcg = container_of(css, struct blkio_cgroup, css); | |
387 | spin_lock_irqsave(&blkcg->lock, flags); | |
388 | if (!hlist_unhashed(&blkg->blkcg_node)) { | |
389 | __blkiocg_del_blkio_group(blkg); | |
390 | ret = 0; | |
391 | } | |
392 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
393 | out: | |
394 | rcu_read_unlock(); | |
395 | return ret; | |
31e4c28d | 396 | } |
9d6a986c | 397 | EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group); |
31e4c28d VG |
398 | |
399 | /* called under rcu_read_lock(). */ | |
400 | struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) | |
401 | { | |
402 | struct blkio_group *blkg; | |
403 | struct hlist_node *n; | |
404 | void *__key; | |
405 | ||
406 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
407 | __key = blkg->key; | |
408 | if (__key == key) | |
409 | return blkg; | |
410 | } | |
411 | ||
412 | return NULL; | |
413 | } | |
9d6a986c | 414 | EXPORT_SYMBOL_GPL(blkiocg_lookup_group); |
31e4c28d VG |
415 | |
416 | #define SHOW_FUNCTION(__VAR) \ | |
417 | static u64 blkiocg_##__VAR##_read(struct cgroup *cgroup, \ | |
418 | struct cftype *cftype) \ | |
419 | { \ | |
420 | struct blkio_cgroup *blkcg; \ | |
421 | \ | |
422 | blkcg = cgroup_to_blkio_cgroup(cgroup); \ | |
423 | return (u64)blkcg->__VAR; \ | |
424 | } | |
425 | ||
426 | SHOW_FUNCTION(weight); | |
427 | #undef SHOW_FUNCTION | |
428 | ||
429 | static int | |
430 | blkiocg_weight_write(struct cgroup *cgroup, struct cftype *cftype, u64 val) | |
431 | { | |
432 | struct blkio_cgroup *blkcg; | |
f8d461d6 VG |
433 | struct blkio_group *blkg; |
434 | struct hlist_node *n; | |
3e252066 | 435 | struct blkio_policy_type *blkiop; |
34d0f179 | 436 | struct blkio_policy_node *pn; |
31e4c28d VG |
437 | |
438 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) | |
439 | return -EINVAL; | |
440 | ||
441 | blkcg = cgroup_to_blkio_cgroup(cgroup); | |
bcf4dd43 | 442 | spin_lock(&blkio_list_lock); |
f8d461d6 | 443 | spin_lock_irq(&blkcg->lock); |
31e4c28d | 444 | blkcg->weight = (unsigned int)val; |
34d0f179 | 445 | |
3e252066 | 446 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
34d0f179 GJ |
447 | pn = blkio_policy_search_node(blkcg, blkg->dev); |
448 | ||
449 | if (pn) | |
450 | continue; | |
451 | ||
3e252066 VG |
452 | list_for_each_entry(blkiop, &blkio_list, list) |
453 | blkiop->ops.blkio_update_group_weight_fn(blkg, | |
454 | blkcg->weight); | |
3e252066 | 455 | } |
f8d461d6 | 456 | spin_unlock_irq(&blkcg->lock); |
bcf4dd43 | 457 | spin_unlock(&blkio_list_lock); |
31e4c28d VG |
458 | return 0; |
459 | } | |
460 | ||
303a3acb | 461 | static int |
84c124da | 462 | blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val) |
303a3acb DS |
463 | { |
464 | struct blkio_cgroup *blkcg; | |
465 | struct blkio_group *blkg; | |
812df48d | 466 | struct blkio_group_stats *stats; |
303a3acb | 467 | struct hlist_node *n; |
cdc1184c DS |
468 | uint64_t queued[BLKIO_STAT_TOTAL]; |
469 | int i; | |
812df48d DS |
470 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
471 | bool idling, waiting, empty; | |
472 | unsigned long long now = sched_clock(); | |
473 | #endif | |
303a3acb DS |
474 | |
475 | blkcg = cgroup_to_blkio_cgroup(cgroup); | |
476 | spin_lock_irq(&blkcg->lock); | |
477 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
478 | spin_lock(&blkg->stats_lock); | |
812df48d DS |
479 | stats = &blkg->stats; |
480 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
481 | idling = blkio_blkg_idling(stats); | |
482 | waiting = blkio_blkg_waiting(stats); | |
483 | empty = blkio_blkg_empty(stats); | |
484 | #endif | |
cdc1184c | 485 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
812df48d DS |
486 | queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i]; |
487 | memset(stats, 0, sizeof(struct blkio_group_stats)); | |
cdc1184c | 488 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
812df48d DS |
489 | stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i]; |
490 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
491 | if (idling) { | |
492 | blkio_mark_blkg_idling(stats); | |
493 | stats->start_idle_time = now; | |
494 | } | |
495 | if (waiting) { | |
496 | blkio_mark_blkg_waiting(stats); | |
497 | stats->start_group_wait_time = now; | |
498 | } | |
499 | if (empty) { | |
500 | blkio_mark_blkg_empty(stats); | |
501 | stats->start_empty_time = now; | |
502 | } | |
503 | #endif | |
303a3acb DS |
504 | spin_unlock(&blkg->stats_lock); |
505 | } | |
506 | spin_unlock_irq(&blkcg->lock); | |
507 | return 0; | |
508 | } | |
509 | ||
84c124da DS |
510 | static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str, |
511 | int chars_left, bool diskname_only) | |
303a3acb | 512 | { |
84c124da | 513 | snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev)); |
303a3acb DS |
514 | chars_left -= strlen(str); |
515 | if (chars_left <= 0) { | |
516 | printk(KERN_WARNING | |
517 | "Possibly incorrect cgroup stat display format"); | |
518 | return; | |
519 | } | |
84c124da DS |
520 | if (diskname_only) |
521 | return; | |
303a3acb | 522 | switch (type) { |
84c124da | 523 | case BLKIO_STAT_READ: |
303a3acb DS |
524 | strlcat(str, " Read", chars_left); |
525 | break; | |
84c124da | 526 | case BLKIO_STAT_WRITE: |
303a3acb DS |
527 | strlcat(str, " Write", chars_left); |
528 | break; | |
84c124da | 529 | case BLKIO_STAT_SYNC: |
303a3acb DS |
530 | strlcat(str, " Sync", chars_left); |
531 | break; | |
84c124da | 532 | case BLKIO_STAT_ASYNC: |
303a3acb DS |
533 | strlcat(str, " Async", chars_left); |
534 | break; | |
84c124da | 535 | case BLKIO_STAT_TOTAL: |
303a3acb DS |
536 | strlcat(str, " Total", chars_left); |
537 | break; | |
538 | default: | |
539 | strlcat(str, " Invalid", chars_left); | |
540 | } | |
541 | } | |
542 | ||
84c124da DS |
543 | static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val, |
544 | struct cgroup_map_cb *cb, dev_t dev) | |
545 | { | |
546 | blkio_get_key_name(0, dev, str, chars_left, true); | |
547 | cb->fill(cb, str, val); | |
548 | return val; | |
549 | } | |
303a3acb | 550 | |
84c124da DS |
551 | /* This should be called with blkg->stats_lock held */ |
552 | static uint64_t blkio_get_stat(struct blkio_group *blkg, | |
553 | struct cgroup_map_cb *cb, dev_t dev, enum stat_type type) | |
303a3acb DS |
554 | { |
555 | uint64_t disk_total; | |
556 | char key_str[MAX_KEY_LEN]; | |
84c124da DS |
557 | enum stat_sub_type sub_type; |
558 | ||
559 | if (type == BLKIO_STAT_TIME) | |
560 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
561 | blkg->stats.time, cb, dev); | |
562 | if (type == BLKIO_STAT_SECTORS) | |
563 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
564 | blkg->stats.sectors, cb, dev); | |
565 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
cdc1184c DS |
566 | if (type == BLKIO_STAT_AVG_QUEUE_SIZE) { |
567 | uint64_t sum = blkg->stats.avg_queue_size_sum; | |
568 | uint64_t samples = blkg->stats.avg_queue_size_samples; | |
569 | if (samples) | |
570 | do_div(sum, samples); | |
571 | else | |
572 | sum = 0; | |
573 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev); | |
574 | } | |
812df48d DS |
575 | if (type == BLKIO_STAT_GROUP_WAIT_TIME) |
576 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
577 | blkg->stats.group_wait_time, cb, dev); | |
578 | if (type == BLKIO_STAT_IDLE_TIME) | |
579 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
580 | blkg->stats.idle_time, cb, dev); | |
581 | if (type == BLKIO_STAT_EMPTY_TIME) | |
582 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
583 | blkg->stats.empty_time, cb, dev); | |
84c124da DS |
584 | if (type == BLKIO_STAT_DEQUEUE) |
585 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
586 | blkg->stats.dequeue, cb, dev); | |
587 | #endif | |
303a3acb | 588 | |
84c124da DS |
589 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; |
590 | sub_type++) { | |
591 | blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false); | |
592 | cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]); | |
303a3acb | 593 | } |
84c124da DS |
594 | disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] + |
595 | blkg->stats.stat_arr[type][BLKIO_STAT_WRITE]; | |
596 | blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false); | |
303a3acb DS |
597 | cb->fill(cb, key_str, disk_total); |
598 | return disk_total; | |
599 | } | |
600 | ||
84c124da | 601 | #define SHOW_FUNCTION_PER_GROUP(__VAR, type, show_total) \ |
22084190 | 602 | static int blkiocg_##__VAR##_read(struct cgroup *cgroup, \ |
303a3acb | 603 | struct cftype *cftype, struct cgroup_map_cb *cb) \ |
22084190 VG |
604 | { \ |
605 | struct blkio_cgroup *blkcg; \ | |
606 | struct blkio_group *blkg; \ | |
607 | struct hlist_node *n; \ | |
303a3acb | 608 | uint64_t cgroup_total = 0; \ |
22084190 VG |
609 | \ |
610 | if (!cgroup_lock_live_group(cgroup)) \ | |
611 | return -ENODEV; \ | |
612 | \ | |
613 | blkcg = cgroup_to_blkio_cgroup(cgroup); \ | |
614 | rcu_read_lock(); \ | |
615 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {\ | |
303a3acb DS |
616 | if (blkg->dev) { \ |
617 | spin_lock_irq(&blkg->stats_lock); \ | |
84c124da DS |
618 | cgroup_total += blkio_get_stat(blkg, cb, \ |
619 | blkg->dev, type); \ | |
303a3acb DS |
620 | spin_unlock_irq(&blkg->stats_lock); \ |
621 | } \ | |
22084190 | 622 | } \ |
303a3acb DS |
623 | if (show_total) \ |
624 | cb->fill(cb, "Total", cgroup_total); \ | |
22084190 VG |
625 | rcu_read_unlock(); \ |
626 | cgroup_unlock(); \ | |
627 | return 0; \ | |
628 | } | |
629 | ||
84c124da DS |
630 | SHOW_FUNCTION_PER_GROUP(time, BLKIO_STAT_TIME, 0); |
631 | SHOW_FUNCTION_PER_GROUP(sectors, BLKIO_STAT_SECTORS, 0); | |
632 | SHOW_FUNCTION_PER_GROUP(io_service_bytes, BLKIO_STAT_SERVICE_BYTES, 1); | |
633 | SHOW_FUNCTION_PER_GROUP(io_serviced, BLKIO_STAT_SERVICED, 1); | |
634 | SHOW_FUNCTION_PER_GROUP(io_service_time, BLKIO_STAT_SERVICE_TIME, 1); | |
635 | SHOW_FUNCTION_PER_GROUP(io_wait_time, BLKIO_STAT_WAIT_TIME, 1); | |
812d4026 | 636 | SHOW_FUNCTION_PER_GROUP(io_merged, BLKIO_STAT_MERGED, 1); |
cdc1184c | 637 | SHOW_FUNCTION_PER_GROUP(io_queued, BLKIO_STAT_QUEUED, 1); |
22084190 | 638 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
84c124da | 639 | SHOW_FUNCTION_PER_GROUP(dequeue, BLKIO_STAT_DEQUEUE, 0); |
cdc1184c | 640 | SHOW_FUNCTION_PER_GROUP(avg_queue_size, BLKIO_STAT_AVG_QUEUE_SIZE, 0); |
812df48d DS |
641 | SHOW_FUNCTION_PER_GROUP(group_wait_time, BLKIO_STAT_GROUP_WAIT_TIME, 0); |
642 | SHOW_FUNCTION_PER_GROUP(idle_time, BLKIO_STAT_IDLE_TIME, 0); | |
643 | SHOW_FUNCTION_PER_GROUP(empty_time, BLKIO_STAT_EMPTY_TIME, 0); | |
22084190 VG |
644 | #endif |
645 | #undef SHOW_FUNCTION_PER_GROUP | |
646 | ||
34d0f179 GJ |
647 | static int blkio_check_dev_num(dev_t dev) |
648 | { | |
649 | int part = 0; | |
650 | struct gendisk *disk; | |
651 | ||
652 | disk = get_gendisk(dev, &part); | |
653 | if (!disk || part) | |
654 | return -ENODEV; | |
655 | ||
656 | return 0; | |
657 | } | |
658 | ||
659 | static int blkio_policy_parse_and_set(char *buf, | |
660 | struct blkio_policy_node *newpn) | |
661 | { | |
662 | char *s[4], *p, *major_s = NULL, *minor_s = NULL; | |
663 | int ret; | |
664 | unsigned long major, minor, temp; | |
665 | int i = 0; | |
666 | dev_t dev; | |
667 | ||
668 | memset(s, 0, sizeof(s)); | |
669 | ||
670 | while ((p = strsep(&buf, " ")) != NULL) { | |
671 | if (!*p) | |
672 | continue; | |
673 | ||
674 | s[i++] = p; | |
675 | ||
676 | /* Prevent from inputing too many things */ | |
677 | if (i == 3) | |
678 | break; | |
679 | } | |
680 | ||
681 | if (i != 2) | |
682 | return -EINVAL; | |
683 | ||
684 | p = strsep(&s[0], ":"); | |
685 | if (p != NULL) | |
686 | major_s = p; | |
687 | else | |
688 | return -EINVAL; | |
689 | ||
690 | minor_s = s[0]; | |
691 | if (!minor_s) | |
692 | return -EINVAL; | |
693 | ||
694 | ret = strict_strtoul(major_s, 10, &major); | |
695 | if (ret) | |
696 | return -EINVAL; | |
697 | ||
698 | ret = strict_strtoul(minor_s, 10, &minor); | |
699 | if (ret) | |
700 | return -EINVAL; | |
701 | ||
702 | dev = MKDEV(major, minor); | |
703 | ||
704 | ret = blkio_check_dev_num(dev); | |
705 | if (ret) | |
706 | return ret; | |
707 | ||
708 | newpn->dev = dev; | |
709 | ||
710 | if (s[1] == NULL) | |
711 | return -EINVAL; | |
712 | ||
713 | ret = strict_strtoul(s[1], 10, &temp); | |
714 | if (ret || (temp < BLKIO_WEIGHT_MIN && temp > 0) || | |
715 | temp > BLKIO_WEIGHT_MAX) | |
716 | return -EINVAL; | |
717 | ||
718 | newpn->weight = temp; | |
719 | ||
720 | return 0; | |
721 | } | |
722 | ||
723 | unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg, | |
724 | dev_t dev) | |
725 | { | |
726 | struct blkio_policy_node *pn; | |
727 | ||
728 | pn = blkio_policy_search_node(blkcg, dev); | |
729 | if (pn) | |
730 | return pn->weight; | |
731 | else | |
732 | return blkcg->weight; | |
733 | } | |
734 | EXPORT_SYMBOL_GPL(blkcg_get_weight); | |
735 | ||
736 | ||
737 | static int blkiocg_weight_device_write(struct cgroup *cgrp, struct cftype *cft, | |
738 | const char *buffer) | |
739 | { | |
740 | int ret = 0; | |
741 | char *buf; | |
742 | struct blkio_policy_node *newpn, *pn; | |
743 | struct blkio_cgroup *blkcg; | |
744 | struct blkio_group *blkg; | |
745 | int keep_newpn = 0; | |
746 | struct hlist_node *n; | |
747 | struct blkio_policy_type *blkiop; | |
748 | ||
749 | buf = kstrdup(buffer, GFP_KERNEL); | |
750 | if (!buf) | |
751 | return -ENOMEM; | |
752 | ||
753 | newpn = kzalloc(sizeof(*newpn), GFP_KERNEL); | |
754 | if (!newpn) { | |
755 | ret = -ENOMEM; | |
756 | goto free_buf; | |
757 | } | |
758 | ||
759 | ret = blkio_policy_parse_and_set(buf, newpn); | |
760 | if (ret) | |
761 | goto free_newpn; | |
762 | ||
763 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
764 | ||
765 | spin_lock_irq(&blkcg->lock); | |
766 | ||
767 | pn = blkio_policy_search_node(blkcg, newpn->dev); | |
768 | if (!pn) { | |
769 | if (newpn->weight != 0) { | |
770 | blkio_policy_insert_node(blkcg, newpn); | |
771 | keep_newpn = 1; | |
772 | } | |
773 | spin_unlock_irq(&blkcg->lock); | |
774 | goto update_io_group; | |
775 | } | |
776 | ||
777 | if (newpn->weight == 0) { | |
778 | /* weight == 0 means deleteing a specific weight */ | |
779 | blkio_policy_delete_node(pn); | |
780 | spin_unlock_irq(&blkcg->lock); | |
781 | goto update_io_group; | |
782 | } | |
783 | spin_unlock_irq(&blkcg->lock); | |
784 | ||
785 | pn->weight = newpn->weight; | |
786 | ||
787 | update_io_group: | |
788 | /* update weight for each cfqg */ | |
789 | spin_lock(&blkio_list_lock); | |
790 | spin_lock_irq(&blkcg->lock); | |
791 | ||
792 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
793 | if (newpn->dev == blkg->dev) { | |
794 | list_for_each_entry(blkiop, &blkio_list, list) | |
795 | blkiop->ops.blkio_update_group_weight_fn(blkg, | |
796 | newpn->weight ? | |
797 | newpn->weight : | |
798 | blkcg->weight); | |
799 | } | |
800 | } | |
801 | ||
802 | spin_unlock_irq(&blkcg->lock); | |
803 | spin_unlock(&blkio_list_lock); | |
804 | ||
805 | free_newpn: | |
806 | if (!keep_newpn) | |
807 | kfree(newpn); | |
808 | free_buf: | |
809 | kfree(buf); | |
810 | return ret; | |
811 | } | |
812 | ||
813 | static int blkiocg_weight_device_read(struct cgroup *cgrp, struct cftype *cft, | |
814 | struct seq_file *m) | |
815 | { | |
816 | struct blkio_cgroup *blkcg; | |
817 | struct blkio_policy_node *pn; | |
818 | ||
819 | seq_printf(m, "dev\tweight\n"); | |
820 | ||
821 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
822 | if (list_empty(&blkcg->policy_list)) | |
823 | goto out; | |
824 | ||
825 | spin_lock_irq(&blkcg->lock); | |
826 | list_for_each_entry(pn, &blkcg->policy_list, node) { | |
827 | seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev), | |
828 | MINOR(pn->dev), pn->weight); | |
829 | } | |
830 | spin_unlock_irq(&blkcg->lock); | |
831 | ||
832 | out: | |
833 | return 0; | |
834 | } | |
835 | ||
31e4c28d | 836 | struct cftype blkio_files[] = { |
34d0f179 GJ |
837 | { |
838 | .name = "weight_device", | |
839 | .read_seq_string = blkiocg_weight_device_read, | |
840 | .write_string = blkiocg_weight_device_write, | |
841 | .max_write_len = 256, | |
842 | }, | |
31e4c28d VG |
843 | { |
844 | .name = "weight", | |
845 | .read_u64 = blkiocg_weight_read, | |
846 | .write_u64 = blkiocg_weight_write, | |
847 | }, | |
22084190 VG |
848 | { |
849 | .name = "time", | |
303a3acb | 850 | .read_map = blkiocg_time_read, |
22084190 VG |
851 | }, |
852 | { | |
853 | .name = "sectors", | |
303a3acb | 854 | .read_map = blkiocg_sectors_read, |
303a3acb DS |
855 | }, |
856 | { | |
857 | .name = "io_service_bytes", | |
858 | .read_map = blkiocg_io_service_bytes_read, | |
303a3acb DS |
859 | }, |
860 | { | |
861 | .name = "io_serviced", | |
862 | .read_map = blkiocg_io_serviced_read, | |
303a3acb DS |
863 | }, |
864 | { | |
865 | .name = "io_service_time", | |
866 | .read_map = blkiocg_io_service_time_read, | |
303a3acb DS |
867 | }, |
868 | { | |
869 | .name = "io_wait_time", | |
870 | .read_map = blkiocg_io_wait_time_read, | |
84c124da | 871 | }, |
812d4026 DS |
872 | { |
873 | .name = "io_merged", | |
874 | .read_map = blkiocg_io_merged_read, | |
875 | }, | |
cdc1184c DS |
876 | { |
877 | .name = "io_queued", | |
878 | .read_map = blkiocg_io_queued_read, | |
879 | }, | |
84c124da DS |
880 | { |
881 | .name = "reset_stats", | |
882 | .write_u64 = blkiocg_reset_stats, | |
22084190 VG |
883 | }, |
884 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
cdc1184c DS |
885 | { |
886 | .name = "avg_queue_size", | |
887 | .read_map = blkiocg_avg_queue_size_read, | |
888 | }, | |
812df48d DS |
889 | { |
890 | .name = "group_wait_time", | |
891 | .read_map = blkiocg_group_wait_time_read, | |
892 | }, | |
893 | { | |
894 | .name = "idle_time", | |
895 | .read_map = blkiocg_idle_time_read, | |
896 | }, | |
897 | { | |
898 | .name = "empty_time", | |
899 | .read_map = blkiocg_empty_time_read, | |
900 | }, | |
cdc1184c | 901 | { |
22084190 | 902 | .name = "dequeue", |
303a3acb | 903 | .read_map = blkiocg_dequeue_read, |
cdc1184c | 904 | }, |
22084190 | 905 | #endif |
31e4c28d VG |
906 | }; |
907 | ||
908 | static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
909 | { | |
910 | return cgroup_add_files(cgroup, subsys, blkio_files, | |
911 | ARRAY_SIZE(blkio_files)); | |
912 | } | |
913 | ||
914 | static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
915 | { | |
916 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
b1c35769 VG |
917 | unsigned long flags; |
918 | struct blkio_group *blkg; | |
919 | void *key; | |
3e252066 | 920 | struct blkio_policy_type *blkiop; |
34d0f179 | 921 | struct blkio_policy_node *pn, *pntmp; |
b1c35769 VG |
922 | |
923 | rcu_read_lock(); | |
924 | remove_entry: | |
925 | spin_lock_irqsave(&blkcg->lock, flags); | |
926 | ||
927 | if (hlist_empty(&blkcg->blkg_list)) { | |
928 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
929 | goto done; | |
930 | } | |
931 | ||
932 | blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group, | |
933 | blkcg_node); | |
934 | key = rcu_dereference(blkg->key); | |
935 | __blkiocg_del_blkio_group(blkg); | |
31e4c28d | 936 | |
b1c35769 VG |
937 | spin_unlock_irqrestore(&blkcg->lock, flags); |
938 | ||
939 | /* | |
940 | * This blkio_group is being unlinked as associated cgroup is going | |
941 | * away. Let all the IO controlling policies know about this event. | |
942 | * | |
943 | * Currently this is static call to one io controlling policy. Once | |
944 | * we have more policies in place, we need some dynamic registration | |
945 | * of callback function. | |
946 | */ | |
3e252066 VG |
947 | spin_lock(&blkio_list_lock); |
948 | list_for_each_entry(blkiop, &blkio_list, list) | |
949 | blkiop->ops.blkio_unlink_group_fn(key, blkg); | |
950 | spin_unlock(&blkio_list_lock); | |
b1c35769 | 951 | goto remove_entry; |
34d0f179 | 952 | |
b1c35769 | 953 | done: |
34d0f179 GJ |
954 | list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) { |
955 | blkio_policy_delete_node(pn); | |
956 | kfree(pn); | |
957 | } | |
31e4c28d | 958 | free_css_id(&blkio_subsys, &blkcg->css); |
b1c35769 | 959 | rcu_read_unlock(); |
67523c48 BB |
960 | if (blkcg != &blkio_root_cgroup) |
961 | kfree(blkcg); | |
31e4c28d VG |
962 | } |
963 | ||
964 | static struct cgroup_subsys_state * | |
965 | blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
966 | { | |
967 | struct blkio_cgroup *blkcg, *parent_blkcg; | |
968 | ||
969 | if (!cgroup->parent) { | |
970 | blkcg = &blkio_root_cgroup; | |
971 | goto done; | |
972 | } | |
973 | ||
974 | /* Currently we do not support hierarchy deeper than two level (0,1) */ | |
975 | parent_blkcg = cgroup_to_blkio_cgroup(cgroup->parent); | |
976 | if (css_depth(&parent_blkcg->css) > 0) | |
977 | return ERR_PTR(-EINVAL); | |
978 | ||
979 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
980 | if (!blkcg) | |
981 | return ERR_PTR(-ENOMEM); | |
982 | ||
983 | blkcg->weight = BLKIO_WEIGHT_DEFAULT; | |
984 | done: | |
985 | spin_lock_init(&blkcg->lock); | |
986 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
987 | ||
34d0f179 | 988 | INIT_LIST_HEAD(&blkcg->policy_list); |
31e4c28d VG |
989 | return &blkcg->css; |
990 | } | |
991 | ||
992 | /* | |
993 | * We cannot support shared io contexts, as we have no mean to support | |
994 | * two tasks with the same ioc in two different groups without major rework | |
995 | * of the main cic data structures. For now we allow a task to change | |
996 | * its cgroup only if it's the only owner of its ioc. | |
997 | */ | |
998 | static int blkiocg_can_attach(struct cgroup_subsys *subsys, | |
999 | struct cgroup *cgroup, struct task_struct *tsk, | |
1000 | bool threadgroup) | |
1001 | { | |
1002 | struct io_context *ioc; | |
1003 | int ret = 0; | |
1004 | ||
1005 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
1006 | task_lock(tsk); | |
1007 | ioc = tsk->io_context; | |
1008 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1009 | ret = -EINVAL; | |
1010 | task_unlock(tsk); | |
1011 | ||
1012 | return ret; | |
1013 | } | |
1014 | ||
1015 | static void blkiocg_attach(struct cgroup_subsys *subsys, struct cgroup *cgroup, | |
1016 | struct cgroup *prev, struct task_struct *tsk, | |
1017 | bool threadgroup) | |
1018 | { | |
1019 | struct io_context *ioc; | |
1020 | ||
1021 | task_lock(tsk); | |
1022 | ioc = tsk->io_context; | |
1023 | if (ioc) | |
1024 | ioc->cgroup_changed = 1; | |
1025 | task_unlock(tsk); | |
1026 | } | |
1027 | ||
3e252066 VG |
1028 | void blkio_policy_register(struct blkio_policy_type *blkiop) |
1029 | { | |
1030 | spin_lock(&blkio_list_lock); | |
1031 | list_add_tail(&blkiop->list, &blkio_list); | |
1032 | spin_unlock(&blkio_list_lock); | |
1033 | } | |
1034 | EXPORT_SYMBOL_GPL(blkio_policy_register); | |
1035 | ||
1036 | void blkio_policy_unregister(struct blkio_policy_type *blkiop) | |
1037 | { | |
1038 | spin_lock(&blkio_list_lock); | |
1039 | list_del_init(&blkiop->list); | |
1040 | spin_unlock(&blkio_list_lock); | |
1041 | } | |
1042 | EXPORT_SYMBOL_GPL(blkio_policy_unregister); | |
67523c48 BB |
1043 | |
1044 | static int __init init_cgroup_blkio(void) | |
1045 | { | |
1046 | return cgroup_load_subsys(&blkio_subsys); | |
1047 | } | |
1048 | ||
1049 | static void __exit exit_cgroup_blkio(void) | |
1050 | { | |
1051 | cgroup_unload_subsys(&blkio_subsys); | |
1052 | } | |
1053 | ||
1054 | module_init(init_cgroup_blkio); | |
1055 | module_exit(exit_cgroup_blkio); | |
1056 | MODULE_LICENSE("GPL"); |