]>
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 | ||
062a644d VG |
40 | /* for encoding cft->private value on file */ |
41 | #define BLKIOFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
42 | /* What policy owns the file, proportional or throttle */ | |
43 | #define BLKIOFILE_POLICY(val) (((val) >> 16) & 0xffff) | |
44 | #define BLKIOFILE_ATTR(val) ((val) & 0xffff) | |
45 | ||
67523c48 BB |
46 | struct cgroup_subsys blkio_subsys = { |
47 | .name = "blkio", | |
48 | .create = blkiocg_create, | |
49 | .can_attach = blkiocg_can_attach, | |
50 | .attach = blkiocg_attach, | |
51 | .destroy = blkiocg_destroy, | |
52 | .populate = blkiocg_populate, | |
53 | #ifdef CONFIG_BLK_CGROUP | |
54 | /* note: blkio_subsys_id is otherwise defined in blk-cgroup.h */ | |
55 | .subsys_id = blkio_subsys_id, | |
56 | #endif | |
57 | .use_id = 1, | |
58 | .module = THIS_MODULE, | |
59 | }; | |
60 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
61 | ||
34d0f179 GJ |
62 | static inline void blkio_policy_insert_node(struct blkio_cgroup *blkcg, |
63 | struct blkio_policy_node *pn) | |
64 | { | |
65 | list_add(&pn->node, &blkcg->policy_list); | |
66 | } | |
67 | ||
062a644d VG |
68 | static inline bool cftype_blkg_same_policy(struct cftype *cft, |
69 | struct blkio_group *blkg) | |
70 | { | |
71 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
72 | ||
73 | if (blkg->plid == plid) | |
74 | return 1; | |
75 | ||
76 | return 0; | |
77 | } | |
78 | ||
79 | /* Determines if policy node matches cgroup file being accessed */ | |
80 | static inline bool pn_matches_cftype(struct cftype *cft, | |
81 | struct blkio_policy_node *pn) | |
82 | { | |
83 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
84 | int fileid = BLKIOFILE_ATTR(cft->private); | |
85 | ||
86 | return (plid == pn->plid && fileid == pn->fileid); | |
87 | } | |
88 | ||
34d0f179 GJ |
89 | /* Must be called with blkcg->lock held */ |
90 | static inline void blkio_policy_delete_node(struct blkio_policy_node *pn) | |
91 | { | |
92 | list_del(&pn->node); | |
93 | } | |
94 | ||
95 | /* Must be called with blkcg->lock held */ | |
96 | static struct blkio_policy_node * | |
062a644d VG |
97 | blkio_policy_search_node(const struct blkio_cgroup *blkcg, dev_t dev, |
98 | enum blkio_policy_id plid, int fileid) | |
34d0f179 GJ |
99 | { |
100 | struct blkio_policy_node *pn; | |
101 | ||
102 | list_for_each_entry(pn, &blkcg->policy_list, node) { | |
062a644d | 103 | if (pn->dev == dev && pn->plid == plid && pn->fileid == fileid) |
34d0f179 GJ |
104 | return pn; |
105 | } | |
106 | ||
107 | return NULL; | |
108 | } | |
109 | ||
31e4c28d VG |
110 | struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup) |
111 | { | |
112 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), | |
113 | struct blkio_cgroup, css); | |
114 | } | |
9d6a986c | 115 | EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup); |
31e4c28d | 116 | |
70087dc3 VG |
117 | struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk) |
118 | { | |
119 | return container_of(task_subsys_state(tsk, blkio_subsys_id), | |
120 | struct blkio_cgroup, css); | |
121 | } | |
122 | EXPORT_SYMBOL_GPL(task_blkio_cgroup); | |
123 | ||
062a644d VG |
124 | static inline void |
125 | blkio_update_group_weight(struct blkio_group *blkg, unsigned int weight) | |
126 | { | |
127 | struct blkio_policy_type *blkiop; | |
128 | ||
129 | list_for_each_entry(blkiop, &blkio_list, list) { | |
130 | /* If this policy does not own the blkg, do not send updates */ | |
131 | if (blkiop->plid != blkg->plid) | |
132 | continue; | |
133 | if (blkiop->ops.blkio_update_group_weight_fn) | |
fe071437 VG |
134 | blkiop->ops.blkio_update_group_weight_fn(blkg->key, |
135 | blkg, weight); | |
062a644d VG |
136 | } |
137 | } | |
138 | ||
4c9eefa1 VG |
139 | static inline void blkio_update_group_bps(struct blkio_group *blkg, u64 bps, |
140 | int fileid) | |
141 | { | |
142 | struct blkio_policy_type *blkiop; | |
143 | ||
144 | list_for_each_entry(blkiop, &blkio_list, list) { | |
145 | ||
146 | /* If this policy does not own the blkg, do not send updates */ | |
147 | if (blkiop->plid != blkg->plid) | |
148 | continue; | |
149 | ||
150 | if (fileid == BLKIO_THROTL_read_bps_device | |
151 | && blkiop->ops.blkio_update_group_read_bps_fn) | |
fe071437 VG |
152 | blkiop->ops.blkio_update_group_read_bps_fn(blkg->key, |
153 | blkg, bps); | |
4c9eefa1 VG |
154 | |
155 | if (fileid == BLKIO_THROTL_write_bps_device | |
156 | && blkiop->ops.blkio_update_group_write_bps_fn) | |
fe071437 VG |
157 | blkiop->ops.blkio_update_group_write_bps_fn(blkg->key, |
158 | blkg, bps); | |
4c9eefa1 VG |
159 | } |
160 | } | |
161 | ||
7702e8f4 VG |
162 | static inline void blkio_update_group_iops(struct blkio_group *blkg, |
163 | unsigned int iops, int fileid) | |
164 | { | |
165 | struct blkio_policy_type *blkiop; | |
166 | ||
167 | list_for_each_entry(blkiop, &blkio_list, list) { | |
168 | ||
169 | /* If this policy does not own the blkg, do not send updates */ | |
170 | if (blkiop->plid != blkg->plid) | |
171 | continue; | |
172 | ||
173 | if (fileid == BLKIO_THROTL_read_iops_device | |
174 | && blkiop->ops.blkio_update_group_read_iops_fn) | |
fe071437 VG |
175 | blkiop->ops.blkio_update_group_read_iops_fn(blkg->key, |
176 | blkg, iops); | |
7702e8f4 VG |
177 | |
178 | if (fileid == BLKIO_THROTL_write_iops_device | |
179 | && blkiop->ops.blkio_update_group_write_iops_fn) | |
fe071437 VG |
180 | blkiop->ops.blkio_update_group_write_iops_fn(blkg->key, |
181 | blkg,iops); | |
7702e8f4 VG |
182 | } |
183 | } | |
184 | ||
9195291e DS |
185 | /* |
186 | * Add to the appropriate stat variable depending on the request type. | |
187 | * This should be called with the blkg->stats_lock held. | |
188 | */ | |
84c124da DS |
189 | static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction, |
190 | bool sync) | |
9195291e | 191 | { |
84c124da DS |
192 | if (direction) |
193 | stat[BLKIO_STAT_WRITE] += add; | |
9195291e | 194 | else |
84c124da DS |
195 | stat[BLKIO_STAT_READ] += add; |
196 | if (sync) | |
197 | stat[BLKIO_STAT_SYNC] += add; | |
9195291e | 198 | else |
84c124da | 199 | stat[BLKIO_STAT_ASYNC] += add; |
9195291e DS |
200 | } |
201 | ||
cdc1184c DS |
202 | /* |
203 | * Decrements the appropriate stat variable if non-zero depending on the | |
204 | * request type. Panics on value being zero. | |
205 | * This should be called with the blkg->stats_lock held. | |
206 | */ | |
207 | static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync) | |
208 | { | |
209 | if (direction) { | |
210 | BUG_ON(stat[BLKIO_STAT_WRITE] == 0); | |
211 | stat[BLKIO_STAT_WRITE]--; | |
212 | } else { | |
213 | BUG_ON(stat[BLKIO_STAT_READ] == 0); | |
214 | stat[BLKIO_STAT_READ]--; | |
215 | } | |
216 | if (sync) { | |
217 | BUG_ON(stat[BLKIO_STAT_SYNC] == 0); | |
218 | stat[BLKIO_STAT_SYNC]--; | |
219 | } else { | |
220 | BUG_ON(stat[BLKIO_STAT_ASYNC] == 0); | |
221 | stat[BLKIO_STAT_ASYNC]--; | |
222 | } | |
223 | } | |
224 | ||
225 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
812df48d DS |
226 | /* This should be called with the blkg->stats_lock held. */ |
227 | static void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
228 | struct blkio_group *curr_blkg) | |
229 | { | |
230 | if (blkio_blkg_waiting(&blkg->stats)) | |
231 | return; | |
232 | if (blkg == curr_blkg) | |
233 | return; | |
234 | blkg->stats.start_group_wait_time = sched_clock(); | |
235 | blkio_mark_blkg_waiting(&blkg->stats); | |
236 | } | |
237 | ||
238 | /* This should be called with the blkg->stats_lock held. */ | |
239 | static void blkio_update_group_wait_time(struct blkio_group_stats *stats) | |
240 | { | |
241 | unsigned long long now; | |
242 | ||
243 | if (!blkio_blkg_waiting(stats)) | |
244 | return; | |
245 | ||
246 | now = sched_clock(); | |
247 | if (time_after64(now, stats->start_group_wait_time)) | |
248 | stats->group_wait_time += now - stats->start_group_wait_time; | |
249 | blkio_clear_blkg_waiting(stats); | |
250 | } | |
251 | ||
252 | /* This should be called with the blkg->stats_lock held. */ | |
253 | static void blkio_end_empty_time(struct blkio_group_stats *stats) | |
254 | { | |
255 | unsigned long long now; | |
256 | ||
257 | if (!blkio_blkg_empty(stats)) | |
258 | return; | |
259 | ||
260 | now = sched_clock(); | |
261 | if (time_after64(now, stats->start_empty_time)) | |
262 | stats->empty_time += now - stats->start_empty_time; | |
263 | blkio_clear_blkg_empty(stats); | |
264 | } | |
265 | ||
266 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg) | |
267 | { | |
268 | unsigned long flags; | |
269 | ||
270 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
271 | BUG_ON(blkio_blkg_idling(&blkg->stats)); | |
272 | blkg->stats.start_idle_time = sched_clock(); | |
273 | blkio_mark_blkg_idling(&blkg->stats); | |
274 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
275 | } | |
276 | EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats); | |
277 | ||
278 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg) | |
279 | { | |
280 | unsigned long flags; | |
281 | unsigned long long now; | |
282 | struct blkio_group_stats *stats; | |
283 | ||
284 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
285 | stats = &blkg->stats; | |
286 | if (blkio_blkg_idling(stats)) { | |
287 | now = sched_clock(); | |
288 | if (time_after64(now, stats->start_idle_time)) | |
289 | stats->idle_time += now - stats->start_idle_time; | |
290 | blkio_clear_blkg_idling(stats); | |
291 | } | |
292 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
293 | } | |
294 | EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats); | |
295 | ||
a11cdaa7 | 296 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg) |
cdc1184c DS |
297 | { |
298 | unsigned long flags; | |
299 | struct blkio_group_stats *stats; | |
300 | ||
301 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
302 | stats = &blkg->stats; | |
303 | stats->avg_queue_size_sum += | |
304 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] + | |
305 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]; | |
306 | stats->avg_queue_size_samples++; | |
812df48d | 307 | blkio_update_group_wait_time(stats); |
cdc1184c DS |
308 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
309 | } | |
a11cdaa7 DS |
310 | EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats); |
311 | ||
e5ff082e | 312 | void blkiocg_set_start_empty_time(struct blkio_group *blkg) |
28baf442 DS |
313 | { |
314 | unsigned long flags; | |
315 | struct blkio_group_stats *stats; | |
316 | ||
317 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
318 | stats = &blkg->stats; | |
319 | ||
320 | if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] || | |
321 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) { | |
322 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
323 | return; | |
324 | } | |
325 | ||
326 | /* | |
e5ff082e VG |
327 | * group is already marked empty. This can happen if cfqq got new |
328 | * request in parent group and moved to this group while being added | |
329 | * to service tree. Just ignore the event and move on. | |
28baf442 | 330 | */ |
e5ff082e VG |
331 | if(blkio_blkg_empty(stats)) { |
332 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
333 | return; | |
334 | } | |
335 | ||
28baf442 DS |
336 | stats->start_empty_time = sched_clock(); |
337 | blkio_mark_blkg_empty(stats); | |
338 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
339 | } | |
340 | EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time); | |
341 | ||
a11cdaa7 DS |
342 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
343 | unsigned long dequeue) | |
344 | { | |
345 | blkg->stats.dequeue += dequeue; | |
346 | } | |
347 | EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats); | |
812df48d DS |
348 | #else |
349 | static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
350 | struct blkio_group *curr_blkg) {} | |
351 | static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {} | |
cdc1184c DS |
352 | #endif |
353 | ||
a11cdaa7 | 354 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c DS |
355 | struct blkio_group *curr_blkg, bool direction, |
356 | bool sync) | |
357 | { | |
358 | unsigned long flags; | |
359 | ||
360 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
361 | blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction, | |
362 | sync); | |
812df48d DS |
363 | blkio_end_empty_time(&blkg->stats); |
364 | blkio_set_start_group_wait_time(blkg, curr_blkg); | |
cdc1184c DS |
365 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
366 | } | |
a11cdaa7 | 367 | EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats); |
cdc1184c | 368 | |
a11cdaa7 | 369 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c DS |
370 | bool direction, bool sync) |
371 | { | |
372 | unsigned long flags; | |
373 | ||
374 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
375 | blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], | |
376 | direction, sync); | |
377 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
378 | } | |
a11cdaa7 | 379 | EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats); |
cdc1184c | 380 | |
167400d3 JT |
381 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time, |
382 | unsigned long unaccounted_time) | |
22084190 | 383 | { |
303a3acb DS |
384 | unsigned long flags; |
385 | ||
386 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
387 | blkg->stats.time += time; | |
a23e6869 | 388 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 | 389 | blkg->stats.unaccounted_time += unaccounted_time; |
a23e6869 | 390 | #endif |
303a3acb | 391 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
22084190 | 392 | } |
303a3acb | 393 | EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used); |
22084190 | 394 | |
5624a4e4 VG |
395 | /* |
396 | * should be called under rcu read lock or queue lock to make sure blkg pointer | |
397 | * is valid. | |
398 | */ | |
84c124da DS |
399 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
400 | uint64_t bytes, bool direction, bool sync) | |
9195291e | 401 | { |
5624a4e4 | 402 | struct blkio_group_stats_cpu *stats_cpu; |
575969a0 VG |
403 | unsigned long flags; |
404 | ||
405 | /* | |
406 | * Disabling interrupts to provide mutual exclusion between two | |
407 | * writes on same cpu. It probably is not needed for 64bit. Not | |
408 | * optimizing that case yet. | |
409 | */ | |
410 | local_irq_save(flags); | |
9195291e | 411 | |
5624a4e4 VG |
412 | stats_cpu = this_cpu_ptr(blkg->stats_cpu); |
413 | ||
575969a0 | 414 | u64_stats_update_begin(&stats_cpu->syncp); |
5624a4e4 VG |
415 | stats_cpu->sectors += bytes >> 9; |
416 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED], | |
417 | 1, direction, sync); | |
418 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES], | |
419 | bytes, direction, sync); | |
575969a0 VG |
420 | u64_stats_update_end(&stats_cpu->syncp); |
421 | local_irq_restore(flags); | |
9195291e | 422 | } |
84c124da | 423 | EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats); |
9195291e | 424 | |
84c124da DS |
425 | void blkiocg_update_completion_stats(struct blkio_group *blkg, |
426 | uint64_t start_time, uint64_t io_start_time, bool direction, bool sync) | |
9195291e DS |
427 | { |
428 | struct blkio_group_stats *stats; | |
429 | unsigned long flags; | |
430 | unsigned long long now = sched_clock(); | |
431 | ||
432 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
433 | stats = &blkg->stats; | |
84c124da DS |
434 | if (time_after64(now, io_start_time)) |
435 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME], | |
436 | now - io_start_time, direction, sync); | |
437 | if (time_after64(io_start_time, start_time)) | |
438 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME], | |
439 | io_start_time - start_time, direction, sync); | |
9195291e DS |
440 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
441 | } | |
84c124da | 442 | EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats); |
9195291e | 443 | |
812d4026 DS |
444 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction, |
445 | bool sync) | |
446 | { | |
447 | unsigned long flags; | |
448 | ||
449 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
450 | blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_MERGED], 1, direction, | |
451 | sync); | |
452 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
453 | } | |
454 | EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats); | |
455 | ||
5624a4e4 VG |
456 | /* |
457 | * This function allocates the per cpu stats for blkio_group. Should be called | |
458 | * from sleepable context as alloc_per_cpu() requires that. | |
459 | */ | |
460 | int blkio_alloc_blkg_stats(struct blkio_group *blkg) | |
461 | { | |
462 | /* Allocate memory for per cpu stats */ | |
463 | blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu); | |
464 | if (!blkg->stats_cpu) | |
465 | return -ENOMEM; | |
466 | return 0; | |
467 | } | |
468 | EXPORT_SYMBOL_GPL(blkio_alloc_blkg_stats); | |
469 | ||
31e4c28d | 470 | void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg, |
062a644d VG |
471 | struct blkio_group *blkg, void *key, dev_t dev, |
472 | enum blkio_policy_id plid) | |
31e4c28d VG |
473 | { |
474 | unsigned long flags; | |
475 | ||
476 | spin_lock_irqsave(&blkcg->lock, flags); | |
8d2a91f8 | 477 | spin_lock_init(&blkg->stats_lock); |
31e4c28d | 478 | rcu_assign_pointer(blkg->key, key); |
b1c35769 | 479 | blkg->blkcg_id = css_id(&blkcg->css); |
31e4c28d | 480 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
062a644d | 481 | blkg->plid = plid; |
31e4c28d | 482 | spin_unlock_irqrestore(&blkcg->lock, flags); |
2868ef7b VG |
483 | /* Need to take css reference ? */ |
484 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); | |
22084190 | 485 | blkg->dev = dev; |
31e4c28d | 486 | } |
9d6a986c | 487 | EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group); |
31e4c28d | 488 | |
b1c35769 VG |
489 | static void __blkiocg_del_blkio_group(struct blkio_group *blkg) |
490 | { | |
491 | hlist_del_init_rcu(&blkg->blkcg_node); | |
492 | blkg->blkcg_id = 0; | |
493 | } | |
494 | ||
495 | /* | |
496 | * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1 | |
497 | * indicating that blk_group was unhashed by the time we got to it. | |
498 | */ | |
31e4c28d VG |
499 | int blkiocg_del_blkio_group(struct blkio_group *blkg) |
500 | { | |
b1c35769 VG |
501 | struct blkio_cgroup *blkcg; |
502 | unsigned long flags; | |
503 | struct cgroup_subsys_state *css; | |
504 | int ret = 1; | |
505 | ||
506 | rcu_read_lock(); | |
507 | css = css_lookup(&blkio_subsys, blkg->blkcg_id); | |
0f3942a3 JA |
508 | if (css) { |
509 | blkcg = container_of(css, struct blkio_cgroup, css); | |
510 | spin_lock_irqsave(&blkcg->lock, flags); | |
511 | if (!hlist_unhashed(&blkg->blkcg_node)) { | |
512 | __blkiocg_del_blkio_group(blkg); | |
513 | ret = 0; | |
514 | } | |
515 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
b1c35769 | 516 | } |
0f3942a3 | 517 | |
b1c35769 VG |
518 | rcu_read_unlock(); |
519 | return ret; | |
31e4c28d | 520 | } |
9d6a986c | 521 | EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group); |
31e4c28d VG |
522 | |
523 | /* called under rcu_read_lock(). */ | |
524 | struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) | |
525 | { | |
526 | struct blkio_group *blkg; | |
527 | struct hlist_node *n; | |
528 | void *__key; | |
529 | ||
530 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
531 | __key = blkg->key; | |
532 | if (__key == key) | |
533 | return blkg; | |
534 | } | |
535 | ||
536 | return NULL; | |
537 | } | |
9d6a986c | 538 | EXPORT_SYMBOL_GPL(blkiocg_lookup_group); |
31e4c28d | 539 | |
f0bdc8cd VG |
540 | static void blkio_reset_stats_cpu(struct blkio_group *blkg) |
541 | { | |
542 | struct blkio_group_stats_cpu *stats_cpu; | |
543 | int i, j, k; | |
544 | /* | |
545 | * Note: On 64 bit arch this should not be an issue. This has the | |
546 | * possibility of returning some inconsistent value on 32bit arch | |
547 | * as 64bit update on 32bit is non atomic. Taking care of this | |
548 | * corner case makes code very complicated, like sending IPIs to | |
549 | * cpus, taking care of stats of offline cpus etc. | |
550 | * | |
551 | * reset stats is anyway more of a debug feature and this sounds a | |
552 | * corner case. So I am not complicating the code yet until and | |
553 | * unless this becomes a real issue. | |
554 | */ | |
555 | for_each_possible_cpu(i) { | |
556 | stats_cpu = per_cpu_ptr(blkg->stats_cpu, i); | |
557 | stats_cpu->sectors = 0; | |
558 | for(j = 0; j < BLKIO_STAT_CPU_NR; j++) | |
559 | for (k = 0; k < BLKIO_STAT_TOTAL; k++) | |
560 | stats_cpu->stat_arr_cpu[j][k] = 0; | |
561 | } | |
562 | } | |
563 | ||
303a3acb | 564 | static int |
84c124da | 565 | blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val) |
303a3acb DS |
566 | { |
567 | struct blkio_cgroup *blkcg; | |
568 | struct blkio_group *blkg; | |
812df48d | 569 | struct blkio_group_stats *stats; |
303a3acb | 570 | struct hlist_node *n; |
cdc1184c DS |
571 | uint64_t queued[BLKIO_STAT_TOTAL]; |
572 | int i; | |
812df48d DS |
573 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
574 | bool idling, waiting, empty; | |
575 | unsigned long long now = sched_clock(); | |
576 | #endif | |
303a3acb DS |
577 | |
578 | blkcg = cgroup_to_blkio_cgroup(cgroup); | |
579 | spin_lock_irq(&blkcg->lock); | |
580 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
581 | spin_lock(&blkg->stats_lock); | |
812df48d DS |
582 | stats = &blkg->stats; |
583 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
584 | idling = blkio_blkg_idling(stats); | |
585 | waiting = blkio_blkg_waiting(stats); | |
586 | empty = blkio_blkg_empty(stats); | |
587 | #endif | |
cdc1184c | 588 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
812df48d DS |
589 | queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i]; |
590 | memset(stats, 0, sizeof(struct blkio_group_stats)); | |
cdc1184c | 591 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
812df48d DS |
592 | stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i]; |
593 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
594 | if (idling) { | |
595 | blkio_mark_blkg_idling(stats); | |
596 | stats->start_idle_time = now; | |
597 | } | |
598 | if (waiting) { | |
599 | blkio_mark_blkg_waiting(stats); | |
600 | stats->start_group_wait_time = now; | |
601 | } | |
602 | if (empty) { | |
603 | blkio_mark_blkg_empty(stats); | |
604 | stats->start_empty_time = now; | |
605 | } | |
606 | #endif | |
303a3acb | 607 | spin_unlock(&blkg->stats_lock); |
f0bdc8cd VG |
608 | |
609 | /* Reset Per cpu stats which don't take blkg->stats_lock */ | |
610 | blkio_reset_stats_cpu(blkg); | |
303a3acb | 611 | } |
f0bdc8cd | 612 | |
303a3acb DS |
613 | spin_unlock_irq(&blkcg->lock); |
614 | return 0; | |
615 | } | |
616 | ||
84c124da DS |
617 | static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str, |
618 | int chars_left, bool diskname_only) | |
303a3acb | 619 | { |
84c124da | 620 | snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev)); |
303a3acb DS |
621 | chars_left -= strlen(str); |
622 | if (chars_left <= 0) { | |
623 | printk(KERN_WARNING | |
624 | "Possibly incorrect cgroup stat display format"); | |
625 | return; | |
626 | } | |
84c124da DS |
627 | if (diskname_only) |
628 | return; | |
303a3acb | 629 | switch (type) { |
84c124da | 630 | case BLKIO_STAT_READ: |
303a3acb DS |
631 | strlcat(str, " Read", chars_left); |
632 | break; | |
84c124da | 633 | case BLKIO_STAT_WRITE: |
303a3acb DS |
634 | strlcat(str, " Write", chars_left); |
635 | break; | |
84c124da | 636 | case BLKIO_STAT_SYNC: |
303a3acb DS |
637 | strlcat(str, " Sync", chars_left); |
638 | break; | |
84c124da | 639 | case BLKIO_STAT_ASYNC: |
303a3acb DS |
640 | strlcat(str, " Async", chars_left); |
641 | break; | |
84c124da | 642 | case BLKIO_STAT_TOTAL: |
303a3acb DS |
643 | strlcat(str, " Total", chars_left); |
644 | break; | |
645 | default: | |
646 | strlcat(str, " Invalid", chars_left); | |
647 | } | |
648 | } | |
649 | ||
84c124da DS |
650 | static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val, |
651 | struct cgroup_map_cb *cb, dev_t dev) | |
652 | { | |
653 | blkio_get_key_name(0, dev, str, chars_left, true); | |
654 | cb->fill(cb, str, val); | |
655 | return val; | |
656 | } | |
303a3acb | 657 | |
5624a4e4 VG |
658 | |
659 | static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg, | |
660 | enum stat_type_cpu type, enum stat_sub_type sub_type) | |
661 | { | |
662 | int cpu; | |
663 | struct blkio_group_stats_cpu *stats_cpu; | |
575969a0 | 664 | u64 val = 0, tval; |
5624a4e4 VG |
665 | |
666 | for_each_possible_cpu(cpu) { | |
575969a0 | 667 | unsigned int start; |
5624a4e4 VG |
668 | stats_cpu = per_cpu_ptr(blkg->stats_cpu, cpu); |
669 | ||
575969a0 VG |
670 | do { |
671 | start = u64_stats_fetch_begin(&stats_cpu->syncp); | |
672 | if (type == BLKIO_STAT_CPU_SECTORS) | |
673 | tval = stats_cpu->sectors; | |
674 | else | |
675 | tval = stats_cpu->stat_arr_cpu[type][sub_type]; | |
676 | } while(u64_stats_fetch_retry(&stats_cpu->syncp, start)); | |
677 | ||
678 | val += tval; | |
5624a4e4 VG |
679 | } |
680 | ||
681 | return val; | |
682 | } | |
683 | ||
684 | static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg, | |
685 | struct cgroup_map_cb *cb, dev_t dev, enum stat_type_cpu type) | |
686 | { | |
687 | uint64_t disk_total, val; | |
688 | char key_str[MAX_KEY_LEN]; | |
689 | enum stat_sub_type sub_type; | |
690 | ||
691 | if (type == BLKIO_STAT_CPU_SECTORS) { | |
692 | val = blkio_read_stat_cpu(blkg, type, 0); | |
693 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, dev); | |
694 | } | |
695 | ||
696 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; | |
697 | sub_type++) { | |
698 | blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false); | |
699 | val = blkio_read_stat_cpu(blkg, type, sub_type); | |
700 | cb->fill(cb, key_str, val); | |
701 | } | |
702 | ||
703 | disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) + | |
704 | blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE); | |
705 | ||
706 | blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false); | |
707 | cb->fill(cb, key_str, disk_total); | |
708 | return disk_total; | |
709 | } | |
710 | ||
84c124da DS |
711 | /* This should be called with blkg->stats_lock held */ |
712 | static uint64_t blkio_get_stat(struct blkio_group *blkg, | |
713 | struct cgroup_map_cb *cb, dev_t dev, enum stat_type type) | |
303a3acb DS |
714 | { |
715 | uint64_t disk_total; | |
716 | char key_str[MAX_KEY_LEN]; | |
84c124da DS |
717 | enum stat_sub_type sub_type; |
718 | ||
719 | if (type == BLKIO_STAT_TIME) | |
720 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
721 | blkg->stats.time, cb, dev); | |
9026e521 | 722 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 JT |
723 | if (type == BLKIO_STAT_UNACCOUNTED_TIME) |
724 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
725 | blkg->stats.unaccounted_time, cb, dev); | |
cdc1184c DS |
726 | if (type == BLKIO_STAT_AVG_QUEUE_SIZE) { |
727 | uint64_t sum = blkg->stats.avg_queue_size_sum; | |
728 | uint64_t samples = blkg->stats.avg_queue_size_samples; | |
729 | if (samples) | |
730 | do_div(sum, samples); | |
731 | else | |
732 | sum = 0; | |
733 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev); | |
734 | } | |
812df48d DS |
735 | if (type == BLKIO_STAT_GROUP_WAIT_TIME) |
736 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
737 | blkg->stats.group_wait_time, cb, dev); | |
738 | if (type == BLKIO_STAT_IDLE_TIME) | |
739 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
740 | blkg->stats.idle_time, cb, dev); | |
741 | if (type == BLKIO_STAT_EMPTY_TIME) | |
742 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
743 | blkg->stats.empty_time, cb, dev); | |
84c124da DS |
744 | if (type == BLKIO_STAT_DEQUEUE) |
745 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
746 | blkg->stats.dequeue, cb, dev); | |
747 | #endif | |
303a3acb | 748 | |
84c124da DS |
749 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; |
750 | sub_type++) { | |
751 | blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false); | |
752 | cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]); | |
303a3acb | 753 | } |
84c124da DS |
754 | disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] + |
755 | blkg->stats.stat_arr[type][BLKIO_STAT_WRITE]; | |
756 | blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false); | |
303a3acb DS |
757 | cb->fill(cb, key_str, disk_total); |
758 | return disk_total; | |
759 | } | |
760 | ||
34d0f179 GJ |
761 | static int blkio_check_dev_num(dev_t dev) |
762 | { | |
763 | int part = 0; | |
764 | struct gendisk *disk; | |
765 | ||
766 | disk = get_gendisk(dev, &part); | |
767 | if (!disk || part) | |
768 | return -ENODEV; | |
769 | ||
770 | return 0; | |
771 | } | |
772 | ||
773 | static int blkio_policy_parse_and_set(char *buf, | |
062a644d | 774 | struct blkio_policy_node *newpn, enum blkio_policy_id plid, int fileid) |
34d0f179 GJ |
775 | { |
776 | char *s[4], *p, *major_s = NULL, *minor_s = NULL; | |
777 | int ret; | |
778 | unsigned long major, minor, temp; | |
779 | int i = 0; | |
780 | dev_t dev; | |
9355aede | 781 | u64 bps, iops; |
34d0f179 GJ |
782 | |
783 | memset(s, 0, sizeof(s)); | |
784 | ||
785 | while ((p = strsep(&buf, " ")) != NULL) { | |
786 | if (!*p) | |
787 | continue; | |
788 | ||
789 | s[i++] = p; | |
790 | ||
791 | /* Prevent from inputing too many things */ | |
792 | if (i == 3) | |
793 | break; | |
794 | } | |
795 | ||
796 | if (i != 2) | |
797 | return -EINVAL; | |
798 | ||
799 | p = strsep(&s[0], ":"); | |
800 | if (p != NULL) | |
801 | major_s = p; | |
802 | else | |
803 | return -EINVAL; | |
804 | ||
805 | minor_s = s[0]; | |
806 | if (!minor_s) | |
807 | return -EINVAL; | |
808 | ||
809 | ret = strict_strtoul(major_s, 10, &major); | |
810 | if (ret) | |
811 | return -EINVAL; | |
812 | ||
813 | ret = strict_strtoul(minor_s, 10, &minor); | |
814 | if (ret) | |
815 | return -EINVAL; | |
816 | ||
817 | dev = MKDEV(major, minor); | |
818 | ||
819 | ret = blkio_check_dev_num(dev); | |
820 | if (ret) | |
821 | return ret; | |
822 | ||
823 | newpn->dev = dev; | |
824 | ||
825 | if (s[1] == NULL) | |
826 | return -EINVAL; | |
827 | ||
062a644d VG |
828 | switch (plid) { |
829 | case BLKIO_POLICY_PROP: | |
830 | ret = strict_strtoul(s[1], 10, &temp); | |
831 | if (ret || (temp < BLKIO_WEIGHT_MIN && temp > 0) || | |
832 | temp > BLKIO_WEIGHT_MAX) | |
833 | return -EINVAL; | |
34d0f179 | 834 | |
062a644d VG |
835 | newpn->plid = plid; |
836 | newpn->fileid = fileid; | |
4c9eefa1 VG |
837 | newpn->val.weight = temp; |
838 | break; | |
839 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
840 | switch(fileid) { |
841 | case BLKIO_THROTL_read_bps_device: | |
842 | case BLKIO_THROTL_write_bps_device: | |
843 | ret = strict_strtoull(s[1], 10, &bps); | |
844 | if (ret) | |
845 | return -EINVAL; | |
4c9eefa1 | 846 | |
7702e8f4 VG |
847 | newpn->plid = plid; |
848 | newpn->fileid = fileid; | |
849 | newpn->val.bps = bps; | |
850 | break; | |
851 | case BLKIO_THROTL_read_iops_device: | |
852 | case BLKIO_THROTL_write_iops_device: | |
9355aede | 853 | ret = strict_strtoull(s[1], 10, &iops); |
7702e8f4 VG |
854 | if (ret) |
855 | return -EINVAL; | |
856 | ||
9355aede VG |
857 | if (iops > THROTL_IOPS_MAX) |
858 | return -EINVAL; | |
859 | ||
7702e8f4 VG |
860 | newpn->plid = plid; |
861 | newpn->fileid = fileid; | |
9355aede | 862 | newpn->val.iops = (unsigned int)iops; |
7702e8f4 VG |
863 | break; |
864 | } | |
062a644d VG |
865 | break; |
866 | default: | |
867 | BUG(); | |
868 | } | |
34d0f179 GJ |
869 | |
870 | return 0; | |
871 | } | |
872 | ||
873 | unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg, | |
874 | dev_t dev) | |
875 | { | |
876 | struct blkio_policy_node *pn; | |
877 | ||
062a644d VG |
878 | pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_PROP, |
879 | BLKIO_PROP_weight_device); | |
34d0f179 | 880 | if (pn) |
4c9eefa1 | 881 | return pn->val.weight; |
34d0f179 GJ |
882 | else |
883 | return blkcg->weight; | |
884 | } | |
885 | EXPORT_SYMBOL_GPL(blkcg_get_weight); | |
886 | ||
4c9eefa1 VG |
887 | uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, dev_t dev) |
888 | { | |
889 | struct blkio_policy_node *pn; | |
890 | ||
891 | pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL, | |
892 | BLKIO_THROTL_read_bps_device); | |
893 | if (pn) | |
894 | return pn->val.bps; | |
895 | else | |
896 | return -1; | |
897 | } | |
898 | ||
899 | uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, dev_t dev) | |
900 | { | |
901 | struct blkio_policy_node *pn; | |
902 | pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL, | |
903 | BLKIO_THROTL_write_bps_device); | |
904 | if (pn) | |
905 | return pn->val.bps; | |
906 | else | |
907 | return -1; | |
908 | } | |
909 | ||
7702e8f4 VG |
910 | unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, dev_t dev) |
911 | { | |
912 | struct blkio_policy_node *pn; | |
913 | ||
914 | pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL, | |
915 | BLKIO_THROTL_read_iops_device); | |
916 | if (pn) | |
917 | return pn->val.iops; | |
918 | else | |
919 | return -1; | |
920 | } | |
921 | ||
922 | unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, dev_t dev) | |
923 | { | |
924 | struct blkio_policy_node *pn; | |
925 | pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL, | |
926 | BLKIO_THROTL_write_iops_device); | |
927 | if (pn) | |
928 | return pn->val.iops; | |
929 | else | |
930 | return -1; | |
931 | } | |
932 | ||
062a644d VG |
933 | /* Checks whether user asked for deleting a policy rule */ |
934 | static bool blkio_delete_rule_command(struct blkio_policy_node *pn) | |
935 | { | |
936 | switch(pn->plid) { | |
937 | case BLKIO_POLICY_PROP: | |
4c9eefa1 VG |
938 | if (pn->val.weight == 0) |
939 | return 1; | |
940 | break; | |
941 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
942 | switch(pn->fileid) { |
943 | case BLKIO_THROTL_read_bps_device: | |
944 | case BLKIO_THROTL_write_bps_device: | |
945 | if (pn->val.bps == 0) | |
946 | return 1; | |
947 | break; | |
948 | case BLKIO_THROTL_read_iops_device: | |
949 | case BLKIO_THROTL_write_iops_device: | |
950 | if (pn->val.iops == 0) | |
951 | return 1; | |
952 | } | |
062a644d VG |
953 | break; |
954 | default: | |
955 | BUG(); | |
956 | } | |
957 | ||
958 | return 0; | |
959 | } | |
960 | ||
961 | static void blkio_update_policy_rule(struct blkio_policy_node *oldpn, | |
962 | struct blkio_policy_node *newpn) | |
963 | { | |
964 | switch(oldpn->plid) { | |
965 | case BLKIO_POLICY_PROP: | |
4c9eefa1 VG |
966 | oldpn->val.weight = newpn->val.weight; |
967 | break; | |
968 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
969 | switch(newpn->fileid) { |
970 | case BLKIO_THROTL_read_bps_device: | |
971 | case BLKIO_THROTL_write_bps_device: | |
972 | oldpn->val.bps = newpn->val.bps; | |
973 | break; | |
974 | case BLKIO_THROTL_read_iops_device: | |
975 | case BLKIO_THROTL_write_iops_device: | |
976 | oldpn->val.iops = newpn->val.iops; | |
977 | } | |
062a644d VG |
978 | break; |
979 | default: | |
980 | BUG(); | |
981 | } | |
982 | } | |
983 | ||
984 | /* | |
25985edc | 985 | * Some rules/values in blkg have changed. Propagate those to respective |
062a644d VG |
986 | * policies. |
987 | */ | |
988 | static void blkio_update_blkg_policy(struct blkio_cgroup *blkcg, | |
989 | struct blkio_group *blkg, struct blkio_policy_node *pn) | |
990 | { | |
7702e8f4 | 991 | unsigned int weight, iops; |
4c9eefa1 | 992 | u64 bps; |
062a644d VG |
993 | |
994 | switch(pn->plid) { | |
995 | case BLKIO_POLICY_PROP: | |
4c9eefa1 | 996 | weight = pn->val.weight ? pn->val.weight : |
062a644d VG |
997 | blkcg->weight; |
998 | blkio_update_group_weight(blkg, weight); | |
999 | break; | |
4c9eefa1 VG |
1000 | case BLKIO_POLICY_THROTL: |
1001 | switch(pn->fileid) { | |
1002 | case BLKIO_THROTL_read_bps_device: | |
1003 | case BLKIO_THROTL_write_bps_device: | |
1004 | bps = pn->val.bps ? pn->val.bps : (-1); | |
1005 | blkio_update_group_bps(blkg, bps, pn->fileid); | |
1006 | break; | |
7702e8f4 VG |
1007 | case BLKIO_THROTL_read_iops_device: |
1008 | case BLKIO_THROTL_write_iops_device: | |
1009 | iops = pn->val.iops ? pn->val.iops : (-1); | |
1010 | blkio_update_group_iops(blkg, iops, pn->fileid); | |
1011 | break; | |
4c9eefa1 VG |
1012 | } |
1013 | break; | |
062a644d VG |
1014 | default: |
1015 | BUG(); | |
1016 | } | |
1017 | } | |
1018 | ||
1019 | /* | |
25985edc | 1020 | * A policy node rule has been updated. Propagate this update to all the |
062a644d VG |
1021 | * block groups which might be affected by this update. |
1022 | */ | |
1023 | static void blkio_update_policy_node_blkg(struct blkio_cgroup *blkcg, | |
1024 | struct blkio_policy_node *pn) | |
1025 | { | |
1026 | struct blkio_group *blkg; | |
1027 | struct hlist_node *n; | |
1028 | ||
1029 | spin_lock(&blkio_list_lock); | |
1030 | spin_lock_irq(&blkcg->lock); | |
34d0f179 | 1031 | |
062a644d VG |
1032 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
1033 | if (pn->dev != blkg->dev || pn->plid != blkg->plid) | |
1034 | continue; | |
1035 | blkio_update_blkg_policy(blkcg, blkg, pn); | |
1036 | } | |
1037 | ||
1038 | spin_unlock_irq(&blkcg->lock); | |
1039 | spin_unlock(&blkio_list_lock); | |
1040 | } | |
34d0f179 | 1041 | |
062a644d VG |
1042 | static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft, |
1043 | const char *buffer) | |
34d0f179 GJ |
1044 | { |
1045 | int ret = 0; | |
1046 | char *buf; | |
1047 | struct blkio_policy_node *newpn, *pn; | |
1048 | struct blkio_cgroup *blkcg; | |
34d0f179 | 1049 | int keep_newpn = 0; |
062a644d VG |
1050 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); |
1051 | int fileid = BLKIOFILE_ATTR(cft->private); | |
34d0f179 GJ |
1052 | |
1053 | buf = kstrdup(buffer, GFP_KERNEL); | |
1054 | if (!buf) | |
1055 | return -ENOMEM; | |
1056 | ||
1057 | newpn = kzalloc(sizeof(*newpn), GFP_KERNEL); | |
1058 | if (!newpn) { | |
1059 | ret = -ENOMEM; | |
1060 | goto free_buf; | |
1061 | } | |
1062 | ||
062a644d | 1063 | ret = blkio_policy_parse_and_set(buf, newpn, plid, fileid); |
34d0f179 GJ |
1064 | if (ret) |
1065 | goto free_newpn; | |
1066 | ||
1067 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1068 | ||
1069 | spin_lock_irq(&blkcg->lock); | |
1070 | ||
062a644d | 1071 | pn = blkio_policy_search_node(blkcg, newpn->dev, plid, fileid); |
34d0f179 | 1072 | if (!pn) { |
062a644d | 1073 | if (!blkio_delete_rule_command(newpn)) { |
34d0f179 GJ |
1074 | blkio_policy_insert_node(blkcg, newpn); |
1075 | keep_newpn = 1; | |
1076 | } | |
1077 | spin_unlock_irq(&blkcg->lock); | |
1078 | goto update_io_group; | |
1079 | } | |
1080 | ||
062a644d | 1081 | if (blkio_delete_rule_command(newpn)) { |
34d0f179 GJ |
1082 | blkio_policy_delete_node(pn); |
1083 | spin_unlock_irq(&blkcg->lock); | |
1084 | goto update_io_group; | |
1085 | } | |
1086 | spin_unlock_irq(&blkcg->lock); | |
1087 | ||
062a644d | 1088 | blkio_update_policy_rule(pn, newpn); |
34d0f179 GJ |
1089 | |
1090 | update_io_group: | |
062a644d | 1091 | blkio_update_policy_node_blkg(blkcg, newpn); |
34d0f179 GJ |
1092 | |
1093 | free_newpn: | |
1094 | if (!keep_newpn) | |
1095 | kfree(newpn); | |
1096 | free_buf: | |
1097 | kfree(buf); | |
1098 | return ret; | |
1099 | } | |
1100 | ||
062a644d VG |
1101 | static void |
1102 | blkio_print_policy_node(struct seq_file *m, struct blkio_policy_node *pn) | |
34d0f179 | 1103 | { |
062a644d VG |
1104 | switch(pn->plid) { |
1105 | case BLKIO_POLICY_PROP: | |
1106 | if (pn->fileid == BLKIO_PROP_weight_device) | |
1107 | seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev), | |
4c9eefa1 VG |
1108 | MINOR(pn->dev), pn->val.weight); |
1109 | break; | |
1110 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
1111 | switch(pn->fileid) { |
1112 | case BLKIO_THROTL_read_bps_device: | |
1113 | case BLKIO_THROTL_write_bps_device: | |
4c9eefa1 VG |
1114 | seq_printf(m, "%u:%u\t%llu\n", MAJOR(pn->dev), |
1115 | MINOR(pn->dev), pn->val.bps); | |
7702e8f4 VG |
1116 | break; |
1117 | case BLKIO_THROTL_read_iops_device: | |
1118 | case BLKIO_THROTL_write_iops_device: | |
1119 | seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev), | |
1120 | MINOR(pn->dev), pn->val.iops); | |
1121 | break; | |
1122 | } | |
062a644d VG |
1123 | break; |
1124 | default: | |
1125 | BUG(); | |
1126 | } | |
1127 | } | |
34d0f179 | 1128 | |
062a644d VG |
1129 | /* cgroup files which read their data from policy nodes end up here */ |
1130 | static void blkio_read_policy_node_files(struct cftype *cft, | |
1131 | struct blkio_cgroup *blkcg, struct seq_file *m) | |
34d0f179 | 1132 | { |
34d0f179 | 1133 | struct blkio_policy_node *pn; |
34d0f179 | 1134 | |
0f3942a3 JA |
1135 | if (!list_empty(&blkcg->policy_list)) { |
1136 | spin_lock_irq(&blkcg->lock); | |
1137 | list_for_each_entry(pn, &blkcg->policy_list, node) { | |
062a644d VG |
1138 | if (!pn_matches_cftype(cft, pn)) |
1139 | continue; | |
1140 | blkio_print_policy_node(m, pn); | |
0f3942a3 JA |
1141 | } |
1142 | spin_unlock_irq(&blkcg->lock); | |
34d0f179 | 1143 | } |
062a644d VG |
1144 | } |
1145 | ||
1146 | static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft, | |
1147 | struct seq_file *m) | |
1148 | { | |
1149 | struct blkio_cgroup *blkcg; | |
1150 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1151 | int name = BLKIOFILE_ATTR(cft->private); | |
1152 | ||
1153 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1154 | ||
1155 | switch(plid) { | |
1156 | case BLKIO_POLICY_PROP: | |
1157 | switch(name) { | |
1158 | case BLKIO_PROP_weight_device: | |
1159 | blkio_read_policy_node_files(cft, blkcg, m); | |
1160 | return 0; | |
1161 | default: | |
1162 | BUG(); | |
1163 | } | |
1164 | break; | |
4c9eefa1 VG |
1165 | case BLKIO_POLICY_THROTL: |
1166 | switch(name){ | |
1167 | case BLKIO_THROTL_read_bps_device: | |
1168 | case BLKIO_THROTL_write_bps_device: | |
7702e8f4 VG |
1169 | case BLKIO_THROTL_read_iops_device: |
1170 | case BLKIO_THROTL_write_iops_device: | |
4c9eefa1 VG |
1171 | blkio_read_policy_node_files(cft, blkcg, m); |
1172 | return 0; | |
1173 | default: | |
1174 | BUG(); | |
1175 | } | |
1176 | break; | |
062a644d VG |
1177 | default: |
1178 | BUG(); | |
1179 | } | |
1180 | ||
1181 | return 0; | |
1182 | } | |
1183 | ||
1184 | static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg, | |
5624a4e4 VG |
1185 | struct cftype *cft, struct cgroup_map_cb *cb, |
1186 | enum stat_type type, bool show_total, bool pcpu) | |
062a644d VG |
1187 | { |
1188 | struct blkio_group *blkg; | |
1189 | struct hlist_node *n; | |
1190 | uint64_t cgroup_total = 0; | |
1191 | ||
1192 | rcu_read_lock(); | |
1193 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
1194 | if (blkg->dev) { | |
1195 | if (!cftype_blkg_same_policy(cft, blkg)) | |
1196 | continue; | |
5624a4e4 VG |
1197 | if (pcpu) |
1198 | cgroup_total += blkio_get_stat_cpu(blkg, cb, | |
1199 | blkg->dev, type); | |
1200 | else { | |
1201 | spin_lock_irq(&blkg->stats_lock); | |
1202 | cgroup_total += blkio_get_stat(blkg, cb, | |
1203 | blkg->dev, type); | |
1204 | spin_unlock_irq(&blkg->stats_lock); | |
1205 | } | |
062a644d VG |
1206 | } |
1207 | } | |
1208 | if (show_total) | |
1209 | cb->fill(cb, "Total", cgroup_total); | |
1210 | rcu_read_unlock(); | |
1211 | return 0; | |
1212 | } | |
1213 | ||
1214 | /* All map kind of cgroup file get serviced by this function */ | |
1215 | static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft, | |
1216 | struct cgroup_map_cb *cb) | |
1217 | { | |
1218 | struct blkio_cgroup *blkcg; | |
1219 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1220 | int name = BLKIOFILE_ATTR(cft->private); | |
1221 | ||
1222 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1223 | ||
1224 | switch(plid) { | |
1225 | case BLKIO_POLICY_PROP: | |
1226 | switch(name) { | |
1227 | case BLKIO_PROP_time: | |
1228 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1229 | BLKIO_STAT_TIME, 0, 0); |
062a644d VG |
1230 | case BLKIO_PROP_sectors: |
1231 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1232 | BLKIO_STAT_CPU_SECTORS, 0, 1); |
062a644d VG |
1233 | case BLKIO_PROP_io_service_bytes: |
1234 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1235 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
062a644d VG |
1236 | case BLKIO_PROP_io_serviced: |
1237 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1238 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
062a644d VG |
1239 | case BLKIO_PROP_io_service_time: |
1240 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1241 | BLKIO_STAT_SERVICE_TIME, 1, 0); |
062a644d VG |
1242 | case BLKIO_PROP_io_wait_time: |
1243 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1244 | BLKIO_STAT_WAIT_TIME, 1, 0); |
062a644d VG |
1245 | case BLKIO_PROP_io_merged: |
1246 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1247 | BLKIO_STAT_MERGED, 1, 0); |
062a644d VG |
1248 | case BLKIO_PROP_io_queued: |
1249 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1250 | BLKIO_STAT_QUEUED, 1, 0); |
062a644d | 1251 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
9026e521 JT |
1252 | case BLKIO_PROP_unaccounted_time: |
1253 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1254 | BLKIO_STAT_UNACCOUNTED_TIME, 0, 0); |
062a644d VG |
1255 | case BLKIO_PROP_dequeue: |
1256 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1257 | BLKIO_STAT_DEQUEUE, 0, 0); |
062a644d VG |
1258 | case BLKIO_PROP_avg_queue_size: |
1259 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1260 | BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0); |
062a644d VG |
1261 | case BLKIO_PROP_group_wait_time: |
1262 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1263 | BLKIO_STAT_GROUP_WAIT_TIME, 0, 0); |
062a644d VG |
1264 | case BLKIO_PROP_idle_time: |
1265 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1266 | BLKIO_STAT_IDLE_TIME, 0, 0); |
062a644d VG |
1267 | case BLKIO_PROP_empty_time: |
1268 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1269 | BLKIO_STAT_EMPTY_TIME, 0, 0); |
062a644d VG |
1270 | #endif |
1271 | default: | |
1272 | BUG(); | |
1273 | } | |
1274 | break; | |
4c9eefa1 VG |
1275 | case BLKIO_POLICY_THROTL: |
1276 | switch(name){ | |
1277 | case BLKIO_THROTL_io_service_bytes: | |
1278 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1279 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
4c9eefa1 VG |
1280 | case BLKIO_THROTL_io_serviced: |
1281 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1282 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
4c9eefa1 VG |
1283 | default: |
1284 | BUG(); | |
1285 | } | |
1286 | break; | |
062a644d VG |
1287 | default: |
1288 | BUG(); | |
1289 | } | |
1290 | ||
1291 | return 0; | |
1292 | } | |
1293 | ||
1294 | static int blkio_weight_write(struct blkio_cgroup *blkcg, u64 val) | |
1295 | { | |
1296 | struct blkio_group *blkg; | |
1297 | struct hlist_node *n; | |
1298 | struct blkio_policy_node *pn; | |
1299 | ||
1300 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) | |
1301 | return -EINVAL; | |
1302 | ||
1303 | spin_lock(&blkio_list_lock); | |
1304 | spin_lock_irq(&blkcg->lock); | |
1305 | blkcg->weight = (unsigned int)val; | |
1306 | ||
1307 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
1308 | pn = blkio_policy_search_node(blkcg, blkg->dev, | |
1309 | BLKIO_POLICY_PROP, BLKIO_PROP_weight_device); | |
1310 | if (pn) | |
1311 | continue; | |
1312 | ||
1313 | blkio_update_group_weight(blkg, blkcg->weight); | |
1314 | } | |
1315 | spin_unlock_irq(&blkcg->lock); | |
1316 | spin_unlock(&blkio_list_lock); | |
1317 | return 0; | |
1318 | } | |
1319 | ||
1320 | static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) { | |
1321 | struct blkio_cgroup *blkcg; | |
1322 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1323 | int name = BLKIOFILE_ATTR(cft->private); | |
1324 | ||
1325 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1326 | ||
1327 | switch(plid) { | |
1328 | case BLKIO_POLICY_PROP: | |
1329 | switch(name) { | |
1330 | case BLKIO_PROP_weight: | |
1331 | return (u64)blkcg->weight; | |
1332 | } | |
1333 | break; | |
1334 | default: | |
1335 | BUG(); | |
1336 | } | |
1337 | return 0; | |
1338 | } | |
1339 | ||
1340 | static int | |
1341 | blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) | |
1342 | { | |
1343 | struct blkio_cgroup *blkcg; | |
1344 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1345 | int name = BLKIOFILE_ATTR(cft->private); | |
1346 | ||
1347 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1348 | ||
1349 | switch(plid) { | |
1350 | case BLKIO_POLICY_PROP: | |
1351 | switch(name) { | |
1352 | case BLKIO_PROP_weight: | |
1353 | return blkio_weight_write(blkcg, val); | |
1354 | } | |
1355 | break; | |
1356 | default: | |
1357 | BUG(); | |
1358 | } | |
34d0f179 | 1359 | |
34d0f179 GJ |
1360 | return 0; |
1361 | } | |
1362 | ||
31e4c28d | 1363 | struct cftype blkio_files[] = { |
34d0f179 GJ |
1364 | { |
1365 | .name = "weight_device", | |
062a644d VG |
1366 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1367 | BLKIO_PROP_weight_device), | |
1368 | .read_seq_string = blkiocg_file_read, | |
1369 | .write_string = blkiocg_file_write, | |
34d0f179 GJ |
1370 | .max_write_len = 256, |
1371 | }, | |
31e4c28d VG |
1372 | { |
1373 | .name = "weight", | |
062a644d VG |
1374 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1375 | BLKIO_PROP_weight), | |
1376 | .read_u64 = blkiocg_file_read_u64, | |
1377 | .write_u64 = blkiocg_file_write_u64, | |
31e4c28d | 1378 | }, |
22084190 VG |
1379 | { |
1380 | .name = "time", | |
062a644d VG |
1381 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1382 | BLKIO_PROP_time), | |
1383 | .read_map = blkiocg_file_read_map, | |
22084190 VG |
1384 | }, |
1385 | { | |
1386 | .name = "sectors", | |
062a644d VG |
1387 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1388 | BLKIO_PROP_sectors), | |
1389 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1390 | }, |
1391 | { | |
1392 | .name = "io_service_bytes", | |
062a644d VG |
1393 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1394 | BLKIO_PROP_io_service_bytes), | |
1395 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1396 | }, |
1397 | { | |
1398 | .name = "io_serviced", | |
062a644d VG |
1399 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1400 | BLKIO_PROP_io_serviced), | |
1401 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1402 | }, |
1403 | { | |
1404 | .name = "io_service_time", | |
062a644d VG |
1405 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1406 | BLKIO_PROP_io_service_time), | |
1407 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1408 | }, |
1409 | { | |
1410 | .name = "io_wait_time", | |
062a644d VG |
1411 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1412 | BLKIO_PROP_io_wait_time), | |
1413 | .read_map = blkiocg_file_read_map, | |
84c124da | 1414 | }, |
812d4026 DS |
1415 | { |
1416 | .name = "io_merged", | |
062a644d VG |
1417 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1418 | BLKIO_PROP_io_merged), | |
1419 | .read_map = blkiocg_file_read_map, | |
812d4026 | 1420 | }, |
cdc1184c DS |
1421 | { |
1422 | .name = "io_queued", | |
062a644d VG |
1423 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1424 | BLKIO_PROP_io_queued), | |
1425 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1426 | }, |
84c124da DS |
1427 | { |
1428 | .name = "reset_stats", | |
1429 | .write_u64 = blkiocg_reset_stats, | |
22084190 | 1430 | }, |
13f98250 VG |
1431 | #ifdef CONFIG_BLK_DEV_THROTTLING |
1432 | { | |
1433 | .name = "throttle.read_bps_device", | |
1434 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1435 | BLKIO_THROTL_read_bps_device), | |
1436 | .read_seq_string = blkiocg_file_read, | |
1437 | .write_string = blkiocg_file_write, | |
1438 | .max_write_len = 256, | |
1439 | }, | |
1440 | ||
1441 | { | |
1442 | .name = "throttle.write_bps_device", | |
1443 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1444 | BLKIO_THROTL_write_bps_device), | |
1445 | .read_seq_string = blkiocg_file_read, | |
1446 | .write_string = blkiocg_file_write, | |
1447 | .max_write_len = 256, | |
1448 | }, | |
1449 | ||
1450 | { | |
1451 | .name = "throttle.read_iops_device", | |
1452 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1453 | BLKIO_THROTL_read_iops_device), | |
1454 | .read_seq_string = blkiocg_file_read, | |
1455 | .write_string = blkiocg_file_write, | |
1456 | .max_write_len = 256, | |
1457 | }, | |
1458 | ||
1459 | { | |
1460 | .name = "throttle.write_iops_device", | |
1461 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1462 | BLKIO_THROTL_write_iops_device), | |
1463 | .read_seq_string = blkiocg_file_read, | |
1464 | .write_string = blkiocg_file_write, | |
1465 | .max_write_len = 256, | |
1466 | }, | |
1467 | { | |
1468 | .name = "throttle.io_service_bytes", | |
1469 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1470 | BLKIO_THROTL_io_service_bytes), | |
1471 | .read_map = blkiocg_file_read_map, | |
1472 | }, | |
1473 | { | |
1474 | .name = "throttle.io_serviced", | |
1475 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1476 | BLKIO_THROTL_io_serviced), | |
1477 | .read_map = blkiocg_file_read_map, | |
1478 | }, | |
1479 | #endif /* CONFIG_BLK_DEV_THROTTLING */ | |
1480 | ||
22084190 | 1481 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
cdc1184c DS |
1482 | { |
1483 | .name = "avg_queue_size", | |
062a644d VG |
1484 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1485 | BLKIO_PROP_avg_queue_size), | |
1486 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1487 | }, |
812df48d DS |
1488 | { |
1489 | .name = "group_wait_time", | |
062a644d VG |
1490 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1491 | BLKIO_PROP_group_wait_time), | |
1492 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1493 | }, |
1494 | { | |
1495 | .name = "idle_time", | |
062a644d VG |
1496 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1497 | BLKIO_PROP_idle_time), | |
1498 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1499 | }, |
1500 | { | |
1501 | .name = "empty_time", | |
062a644d VG |
1502 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1503 | BLKIO_PROP_empty_time), | |
1504 | .read_map = blkiocg_file_read_map, | |
812df48d | 1505 | }, |
cdc1184c | 1506 | { |
22084190 | 1507 | .name = "dequeue", |
062a644d VG |
1508 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1509 | BLKIO_PROP_dequeue), | |
1510 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1511 | }, |
9026e521 JT |
1512 | { |
1513 | .name = "unaccounted_time", | |
1514 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, | |
1515 | BLKIO_PROP_unaccounted_time), | |
1516 | .read_map = blkiocg_file_read_map, | |
1517 | }, | |
22084190 | 1518 | #endif |
31e4c28d VG |
1519 | }; |
1520 | ||
1521 | static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1522 | { | |
1523 | return cgroup_add_files(cgroup, subsys, blkio_files, | |
1524 | ARRAY_SIZE(blkio_files)); | |
1525 | } | |
1526 | ||
1527 | static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1528 | { | |
1529 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
b1c35769 VG |
1530 | unsigned long flags; |
1531 | struct blkio_group *blkg; | |
1532 | void *key; | |
3e252066 | 1533 | struct blkio_policy_type *blkiop; |
34d0f179 | 1534 | struct blkio_policy_node *pn, *pntmp; |
b1c35769 VG |
1535 | |
1536 | rcu_read_lock(); | |
0f3942a3 JA |
1537 | do { |
1538 | spin_lock_irqsave(&blkcg->lock, flags); | |
b1c35769 | 1539 | |
0f3942a3 JA |
1540 | if (hlist_empty(&blkcg->blkg_list)) { |
1541 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
1542 | break; | |
1543 | } | |
b1c35769 | 1544 | |
0f3942a3 JA |
1545 | blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group, |
1546 | blkcg_node); | |
1547 | key = rcu_dereference(blkg->key); | |
1548 | __blkiocg_del_blkio_group(blkg); | |
31e4c28d | 1549 | |
0f3942a3 | 1550 | spin_unlock_irqrestore(&blkcg->lock, flags); |
b1c35769 | 1551 | |
0f3942a3 JA |
1552 | /* |
1553 | * This blkio_group is being unlinked as associated cgroup is | |
1554 | * going away. Let all the IO controlling policies know about | |
61014e96 | 1555 | * this event. |
0f3942a3 JA |
1556 | */ |
1557 | spin_lock(&blkio_list_lock); | |
61014e96 VG |
1558 | list_for_each_entry(blkiop, &blkio_list, list) { |
1559 | if (blkiop->plid != blkg->plid) | |
1560 | continue; | |
0f3942a3 | 1561 | blkiop->ops.blkio_unlink_group_fn(key, blkg); |
61014e96 | 1562 | } |
0f3942a3 JA |
1563 | spin_unlock(&blkio_list_lock); |
1564 | } while (1); | |
34d0f179 | 1565 | |
34d0f179 GJ |
1566 | list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) { |
1567 | blkio_policy_delete_node(pn); | |
1568 | kfree(pn); | |
1569 | } | |
0f3942a3 | 1570 | |
31e4c28d | 1571 | free_css_id(&blkio_subsys, &blkcg->css); |
b1c35769 | 1572 | rcu_read_unlock(); |
67523c48 BB |
1573 | if (blkcg != &blkio_root_cgroup) |
1574 | kfree(blkcg); | |
31e4c28d VG |
1575 | } |
1576 | ||
1577 | static struct cgroup_subsys_state * | |
1578 | blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1579 | { | |
0341509f LZ |
1580 | struct blkio_cgroup *blkcg; |
1581 | struct cgroup *parent = cgroup->parent; | |
31e4c28d | 1582 | |
0341509f | 1583 | if (!parent) { |
31e4c28d VG |
1584 | blkcg = &blkio_root_cgroup; |
1585 | goto done; | |
1586 | } | |
1587 | ||
31e4c28d VG |
1588 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
1589 | if (!blkcg) | |
1590 | return ERR_PTR(-ENOMEM); | |
1591 | ||
1592 | blkcg->weight = BLKIO_WEIGHT_DEFAULT; | |
1593 | done: | |
1594 | spin_lock_init(&blkcg->lock); | |
1595 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
1596 | ||
34d0f179 | 1597 | INIT_LIST_HEAD(&blkcg->policy_list); |
31e4c28d VG |
1598 | return &blkcg->css; |
1599 | } | |
1600 | ||
1601 | /* | |
1602 | * We cannot support shared io contexts, as we have no mean to support | |
1603 | * two tasks with the same ioc in two different groups without major rework | |
1604 | * of the main cic data structures. For now we allow a task to change | |
1605 | * its cgroup only if it's the only owner of its ioc. | |
1606 | */ | |
1607 | static int blkiocg_can_attach(struct cgroup_subsys *subsys, | |
1608 | struct cgroup *cgroup, struct task_struct *tsk, | |
1609 | bool threadgroup) | |
1610 | { | |
1611 | struct io_context *ioc; | |
1612 | int ret = 0; | |
1613 | ||
1614 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
1615 | task_lock(tsk); | |
1616 | ioc = tsk->io_context; | |
1617 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1618 | ret = -EINVAL; | |
1619 | task_unlock(tsk); | |
1620 | ||
1621 | return ret; | |
1622 | } | |
1623 | ||
1624 | static void blkiocg_attach(struct cgroup_subsys *subsys, struct cgroup *cgroup, | |
1625 | struct cgroup *prev, struct task_struct *tsk, | |
1626 | bool threadgroup) | |
1627 | { | |
1628 | struct io_context *ioc; | |
1629 | ||
1630 | task_lock(tsk); | |
1631 | ioc = tsk->io_context; | |
1632 | if (ioc) | |
1633 | ioc->cgroup_changed = 1; | |
1634 | task_unlock(tsk); | |
1635 | } | |
1636 | ||
3e252066 VG |
1637 | void blkio_policy_register(struct blkio_policy_type *blkiop) |
1638 | { | |
1639 | spin_lock(&blkio_list_lock); | |
1640 | list_add_tail(&blkiop->list, &blkio_list); | |
1641 | spin_unlock(&blkio_list_lock); | |
1642 | } | |
1643 | EXPORT_SYMBOL_GPL(blkio_policy_register); | |
1644 | ||
1645 | void blkio_policy_unregister(struct blkio_policy_type *blkiop) | |
1646 | { | |
1647 | spin_lock(&blkio_list_lock); | |
1648 | list_del_init(&blkiop->list); | |
1649 | spin_unlock(&blkio_list_lock); | |
1650 | } | |
1651 | EXPORT_SYMBOL_GPL(blkio_policy_unregister); | |
67523c48 BB |
1652 | |
1653 | static int __init init_cgroup_blkio(void) | |
1654 | { | |
1655 | return cgroup_load_subsys(&blkio_subsys); | |
1656 | } | |
1657 | ||
1658 | static void __exit exit_cgroup_blkio(void) | |
1659 | { | |
1660 | cgroup_unload_subsys(&blkio_subsys); | |
1661 | } | |
1662 | ||
1663 | module_init(init_cgroup_blkio); | |
1664 | module_exit(exit_cgroup_blkio); | |
1665 | MODULE_LICENSE("GPL"); |