]>
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> |
34d0f179 | 20 | #include <linux/genhd.h> |
72e06c25 TH |
21 | #include <linux/delay.h> |
22 | #include "blk-cgroup.h" | |
5efd6113 | 23 | #include "blk.h" |
3e252066 | 24 | |
84c124da DS |
25 | #define MAX_KEY_LEN 100 |
26 | ||
3e252066 VG |
27 | static DEFINE_SPINLOCK(blkio_list_lock); |
28 | static LIST_HEAD(blkio_list); | |
b1c35769 | 29 | |
923adde1 TH |
30 | static DEFINE_MUTEX(all_q_mutex); |
31 | static LIST_HEAD(all_q_list); | |
32 | ||
31e4c28d | 33 | struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT }; |
9d6a986c VG |
34 | EXPORT_SYMBOL_GPL(blkio_root_cgroup); |
35 | ||
035d10b2 TH |
36 | static struct blkio_policy_type *blkio_policy[BLKIO_NR_POLICIES]; |
37 | ||
67523c48 BB |
38 | static struct cgroup_subsys_state *blkiocg_create(struct cgroup_subsys *, |
39 | struct cgroup *); | |
bb9d97b6 TH |
40 | static int blkiocg_can_attach(struct cgroup_subsys *, struct cgroup *, |
41 | struct cgroup_taskset *); | |
42 | static void blkiocg_attach(struct cgroup_subsys *, struct cgroup *, | |
43 | struct cgroup_taskset *); | |
7ee9c562 | 44 | static int blkiocg_pre_destroy(struct cgroup_subsys *, struct cgroup *); |
67523c48 BB |
45 | static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *); |
46 | static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *); | |
47 | ||
062a644d VG |
48 | /* for encoding cft->private value on file */ |
49 | #define BLKIOFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
50 | /* What policy owns the file, proportional or throttle */ | |
51 | #define BLKIOFILE_POLICY(val) (((val) >> 16) & 0xffff) | |
52 | #define BLKIOFILE_ATTR(val) ((val) & 0xffff) | |
53 | ||
67523c48 BB |
54 | struct cgroup_subsys blkio_subsys = { |
55 | .name = "blkio", | |
56 | .create = blkiocg_create, | |
bb9d97b6 TH |
57 | .can_attach = blkiocg_can_attach, |
58 | .attach = blkiocg_attach, | |
7ee9c562 | 59 | .pre_destroy = blkiocg_pre_destroy, |
67523c48 BB |
60 | .destroy = blkiocg_destroy, |
61 | .populate = blkiocg_populate, | |
67523c48 | 62 | .subsys_id = blkio_subsys_id, |
67523c48 BB |
63 | .module = THIS_MODULE, |
64 | }; | |
65 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
66 | ||
31e4c28d VG |
67 | struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup) |
68 | { | |
69 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), | |
70 | struct blkio_cgroup, css); | |
71 | } | |
9d6a986c | 72 | EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup); |
31e4c28d | 73 | |
70087dc3 VG |
74 | struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk) |
75 | { | |
76 | return container_of(task_subsys_state(tsk, blkio_subsys_id), | |
77 | struct blkio_cgroup, css); | |
78 | } | |
79 | EXPORT_SYMBOL_GPL(task_blkio_cgroup); | |
80 | ||
062a644d VG |
81 | static inline void |
82 | blkio_update_group_weight(struct blkio_group *blkg, unsigned int weight) | |
83 | { | |
84 | struct blkio_policy_type *blkiop; | |
85 | ||
86 | list_for_each_entry(blkiop, &blkio_list, list) { | |
87 | /* If this policy does not own the blkg, do not send updates */ | |
88 | if (blkiop->plid != blkg->plid) | |
89 | continue; | |
90 | if (blkiop->ops.blkio_update_group_weight_fn) | |
ca32aefc | 91 | blkiop->ops.blkio_update_group_weight_fn(blkg->q, |
fe071437 | 92 | blkg, weight); |
062a644d VG |
93 | } |
94 | } | |
95 | ||
4c9eefa1 VG |
96 | static inline void blkio_update_group_bps(struct blkio_group *blkg, u64 bps, |
97 | int fileid) | |
98 | { | |
99 | struct blkio_policy_type *blkiop; | |
100 | ||
101 | list_for_each_entry(blkiop, &blkio_list, list) { | |
102 | ||
103 | /* If this policy does not own the blkg, do not send updates */ | |
104 | if (blkiop->plid != blkg->plid) | |
105 | continue; | |
106 | ||
107 | if (fileid == BLKIO_THROTL_read_bps_device | |
108 | && blkiop->ops.blkio_update_group_read_bps_fn) | |
ca32aefc | 109 | blkiop->ops.blkio_update_group_read_bps_fn(blkg->q, |
fe071437 | 110 | blkg, bps); |
4c9eefa1 VG |
111 | |
112 | if (fileid == BLKIO_THROTL_write_bps_device | |
113 | && blkiop->ops.blkio_update_group_write_bps_fn) | |
ca32aefc | 114 | blkiop->ops.blkio_update_group_write_bps_fn(blkg->q, |
fe071437 | 115 | blkg, bps); |
4c9eefa1 VG |
116 | } |
117 | } | |
118 | ||
7702e8f4 VG |
119 | static inline void blkio_update_group_iops(struct blkio_group *blkg, |
120 | unsigned int iops, int fileid) | |
121 | { | |
122 | struct blkio_policy_type *blkiop; | |
123 | ||
124 | list_for_each_entry(blkiop, &blkio_list, list) { | |
125 | ||
126 | /* If this policy does not own the blkg, do not send updates */ | |
127 | if (blkiop->plid != blkg->plid) | |
128 | continue; | |
129 | ||
130 | if (fileid == BLKIO_THROTL_read_iops_device | |
131 | && blkiop->ops.blkio_update_group_read_iops_fn) | |
ca32aefc | 132 | blkiop->ops.blkio_update_group_read_iops_fn(blkg->q, |
fe071437 | 133 | blkg, iops); |
7702e8f4 VG |
134 | |
135 | if (fileid == BLKIO_THROTL_write_iops_device | |
136 | && blkiop->ops.blkio_update_group_write_iops_fn) | |
ca32aefc | 137 | blkiop->ops.blkio_update_group_write_iops_fn(blkg->q, |
fe071437 | 138 | blkg,iops); |
7702e8f4 VG |
139 | } |
140 | } | |
141 | ||
9195291e DS |
142 | /* |
143 | * Add to the appropriate stat variable depending on the request type. | |
144 | * This should be called with the blkg->stats_lock held. | |
145 | */ | |
84c124da DS |
146 | static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction, |
147 | bool sync) | |
9195291e | 148 | { |
84c124da DS |
149 | if (direction) |
150 | stat[BLKIO_STAT_WRITE] += add; | |
9195291e | 151 | else |
84c124da DS |
152 | stat[BLKIO_STAT_READ] += add; |
153 | if (sync) | |
154 | stat[BLKIO_STAT_SYNC] += add; | |
9195291e | 155 | else |
84c124da | 156 | stat[BLKIO_STAT_ASYNC] += add; |
9195291e DS |
157 | } |
158 | ||
cdc1184c DS |
159 | /* |
160 | * Decrements the appropriate stat variable if non-zero depending on the | |
161 | * request type. Panics on value being zero. | |
162 | * This should be called with the blkg->stats_lock held. | |
163 | */ | |
164 | static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync) | |
165 | { | |
166 | if (direction) { | |
167 | BUG_ON(stat[BLKIO_STAT_WRITE] == 0); | |
168 | stat[BLKIO_STAT_WRITE]--; | |
169 | } else { | |
170 | BUG_ON(stat[BLKIO_STAT_READ] == 0); | |
171 | stat[BLKIO_STAT_READ]--; | |
172 | } | |
173 | if (sync) { | |
174 | BUG_ON(stat[BLKIO_STAT_SYNC] == 0); | |
175 | stat[BLKIO_STAT_SYNC]--; | |
176 | } else { | |
177 | BUG_ON(stat[BLKIO_STAT_ASYNC] == 0); | |
178 | stat[BLKIO_STAT_ASYNC]--; | |
179 | } | |
180 | } | |
181 | ||
182 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
812df48d DS |
183 | /* This should be called with the blkg->stats_lock held. */ |
184 | static void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
185 | struct blkio_group *curr_blkg) | |
186 | { | |
187 | if (blkio_blkg_waiting(&blkg->stats)) | |
188 | return; | |
189 | if (blkg == curr_blkg) | |
190 | return; | |
191 | blkg->stats.start_group_wait_time = sched_clock(); | |
192 | blkio_mark_blkg_waiting(&blkg->stats); | |
193 | } | |
194 | ||
195 | /* This should be called with the blkg->stats_lock held. */ | |
196 | static void blkio_update_group_wait_time(struct blkio_group_stats *stats) | |
197 | { | |
198 | unsigned long long now; | |
199 | ||
200 | if (!blkio_blkg_waiting(stats)) | |
201 | return; | |
202 | ||
203 | now = sched_clock(); | |
204 | if (time_after64(now, stats->start_group_wait_time)) | |
205 | stats->group_wait_time += now - stats->start_group_wait_time; | |
206 | blkio_clear_blkg_waiting(stats); | |
207 | } | |
208 | ||
209 | /* This should be called with the blkg->stats_lock held. */ | |
210 | static void blkio_end_empty_time(struct blkio_group_stats *stats) | |
211 | { | |
212 | unsigned long long now; | |
213 | ||
214 | if (!blkio_blkg_empty(stats)) | |
215 | return; | |
216 | ||
217 | now = sched_clock(); | |
218 | if (time_after64(now, stats->start_empty_time)) | |
219 | stats->empty_time += now - stats->start_empty_time; | |
220 | blkio_clear_blkg_empty(stats); | |
221 | } | |
222 | ||
223 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg) | |
224 | { | |
225 | unsigned long flags; | |
226 | ||
227 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
228 | BUG_ON(blkio_blkg_idling(&blkg->stats)); | |
229 | blkg->stats.start_idle_time = sched_clock(); | |
230 | blkio_mark_blkg_idling(&blkg->stats); | |
231 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
232 | } | |
233 | EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats); | |
234 | ||
235 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg) | |
236 | { | |
237 | unsigned long flags; | |
238 | unsigned long long now; | |
239 | struct blkio_group_stats *stats; | |
240 | ||
241 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
242 | stats = &blkg->stats; | |
243 | if (blkio_blkg_idling(stats)) { | |
244 | now = sched_clock(); | |
245 | if (time_after64(now, stats->start_idle_time)) | |
246 | stats->idle_time += now - stats->start_idle_time; | |
247 | blkio_clear_blkg_idling(stats); | |
248 | } | |
249 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
250 | } | |
251 | EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats); | |
252 | ||
a11cdaa7 | 253 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg) |
cdc1184c DS |
254 | { |
255 | unsigned long flags; | |
256 | struct blkio_group_stats *stats; | |
257 | ||
258 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
259 | stats = &blkg->stats; | |
260 | stats->avg_queue_size_sum += | |
261 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] + | |
262 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]; | |
263 | stats->avg_queue_size_samples++; | |
812df48d | 264 | blkio_update_group_wait_time(stats); |
cdc1184c DS |
265 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
266 | } | |
a11cdaa7 DS |
267 | EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats); |
268 | ||
e5ff082e | 269 | void blkiocg_set_start_empty_time(struct blkio_group *blkg) |
28baf442 DS |
270 | { |
271 | unsigned long flags; | |
272 | struct blkio_group_stats *stats; | |
273 | ||
274 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
275 | stats = &blkg->stats; | |
276 | ||
277 | if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] || | |
278 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) { | |
279 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
280 | return; | |
281 | } | |
282 | ||
283 | /* | |
e5ff082e VG |
284 | * group is already marked empty. This can happen if cfqq got new |
285 | * request in parent group and moved to this group while being added | |
286 | * to service tree. Just ignore the event and move on. | |
28baf442 | 287 | */ |
e5ff082e VG |
288 | if(blkio_blkg_empty(stats)) { |
289 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
290 | return; | |
291 | } | |
292 | ||
28baf442 DS |
293 | stats->start_empty_time = sched_clock(); |
294 | blkio_mark_blkg_empty(stats); | |
295 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
296 | } | |
297 | EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time); | |
298 | ||
a11cdaa7 DS |
299 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
300 | unsigned long dequeue) | |
301 | { | |
302 | blkg->stats.dequeue += dequeue; | |
303 | } | |
304 | EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats); | |
812df48d DS |
305 | #else |
306 | static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
307 | struct blkio_group *curr_blkg) {} | |
308 | static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {} | |
cdc1184c DS |
309 | #endif |
310 | ||
a11cdaa7 | 311 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c DS |
312 | struct blkio_group *curr_blkg, bool direction, |
313 | bool sync) | |
314 | { | |
315 | unsigned long flags; | |
316 | ||
317 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
318 | blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction, | |
319 | sync); | |
812df48d DS |
320 | blkio_end_empty_time(&blkg->stats); |
321 | blkio_set_start_group_wait_time(blkg, curr_blkg); | |
cdc1184c DS |
322 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
323 | } | |
a11cdaa7 | 324 | EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats); |
cdc1184c | 325 | |
a11cdaa7 | 326 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c DS |
327 | bool direction, bool sync) |
328 | { | |
329 | unsigned long flags; | |
330 | ||
331 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
332 | blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], | |
333 | direction, sync); | |
334 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
335 | } | |
a11cdaa7 | 336 | EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats); |
cdc1184c | 337 | |
167400d3 JT |
338 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time, |
339 | unsigned long unaccounted_time) | |
22084190 | 340 | { |
303a3acb DS |
341 | unsigned long flags; |
342 | ||
343 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
344 | blkg->stats.time += time; | |
a23e6869 | 345 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 | 346 | blkg->stats.unaccounted_time += unaccounted_time; |
a23e6869 | 347 | #endif |
303a3acb | 348 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
22084190 | 349 | } |
303a3acb | 350 | EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used); |
22084190 | 351 | |
5624a4e4 VG |
352 | /* |
353 | * should be called under rcu read lock or queue lock to make sure blkg pointer | |
354 | * is valid. | |
355 | */ | |
84c124da DS |
356 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
357 | uint64_t bytes, bool direction, bool sync) | |
9195291e | 358 | { |
5624a4e4 | 359 | struct blkio_group_stats_cpu *stats_cpu; |
575969a0 VG |
360 | unsigned long flags; |
361 | ||
362 | /* | |
363 | * Disabling interrupts to provide mutual exclusion between two | |
364 | * writes on same cpu. It probably is not needed for 64bit. Not | |
365 | * optimizing that case yet. | |
366 | */ | |
367 | local_irq_save(flags); | |
9195291e | 368 | |
5624a4e4 VG |
369 | stats_cpu = this_cpu_ptr(blkg->stats_cpu); |
370 | ||
575969a0 | 371 | u64_stats_update_begin(&stats_cpu->syncp); |
5624a4e4 VG |
372 | stats_cpu->sectors += bytes >> 9; |
373 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED], | |
374 | 1, direction, sync); | |
375 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES], | |
376 | bytes, direction, sync); | |
575969a0 VG |
377 | u64_stats_update_end(&stats_cpu->syncp); |
378 | local_irq_restore(flags); | |
9195291e | 379 | } |
84c124da | 380 | EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats); |
9195291e | 381 | |
84c124da DS |
382 | void blkiocg_update_completion_stats(struct blkio_group *blkg, |
383 | uint64_t start_time, uint64_t io_start_time, bool direction, bool sync) | |
9195291e DS |
384 | { |
385 | struct blkio_group_stats *stats; | |
386 | unsigned long flags; | |
387 | unsigned long long now = sched_clock(); | |
388 | ||
389 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
390 | stats = &blkg->stats; | |
84c124da DS |
391 | if (time_after64(now, io_start_time)) |
392 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME], | |
393 | now - io_start_time, direction, sync); | |
394 | if (time_after64(io_start_time, start_time)) | |
395 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME], | |
396 | io_start_time - start_time, direction, sync); | |
9195291e DS |
397 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
398 | } | |
84c124da | 399 | EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats); |
9195291e | 400 | |
317389a7 | 401 | /* Merged stats are per cpu. */ |
812d4026 DS |
402 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction, |
403 | bool sync) | |
404 | { | |
317389a7 | 405 | struct blkio_group_stats_cpu *stats_cpu; |
812d4026 DS |
406 | unsigned long flags; |
407 | ||
317389a7 VG |
408 | /* |
409 | * Disabling interrupts to provide mutual exclusion between two | |
410 | * writes on same cpu. It probably is not needed for 64bit. Not | |
411 | * optimizing that case yet. | |
412 | */ | |
413 | local_irq_save(flags); | |
414 | ||
415 | stats_cpu = this_cpu_ptr(blkg->stats_cpu); | |
416 | ||
417 | u64_stats_update_begin(&stats_cpu->syncp); | |
418 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1, | |
419 | direction, sync); | |
420 | u64_stats_update_end(&stats_cpu->syncp); | |
421 | local_irq_restore(flags); | |
812d4026 DS |
422 | } |
423 | EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats); | |
424 | ||
0381411e TH |
425 | /** |
426 | * blkg_free - free a blkg | |
427 | * @blkg: blkg to free | |
428 | * | |
429 | * Free @blkg which may be partially allocated. | |
430 | */ | |
431 | static void blkg_free(struct blkio_group *blkg) | |
432 | { | |
433 | if (blkg) { | |
434 | free_percpu(blkg->stats_cpu); | |
435 | kfree(blkg->pd); | |
436 | kfree(blkg); | |
437 | } | |
438 | } | |
439 | ||
440 | /** | |
441 | * blkg_alloc - allocate a blkg | |
442 | * @blkcg: block cgroup the new blkg is associated with | |
443 | * @q: request_queue the new blkg is associated with | |
444 | * @pol: policy the new blkg is associated with | |
445 | * | |
446 | * Allocate a new blkg assocating @blkcg and @q for @pol. | |
447 | * | |
448 | * FIXME: Should be called with queue locked but currently isn't due to | |
449 | * percpu stat breakage. | |
450 | */ | |
451 | static struct blkio_group *blkg_alloc(struct blkio_cgroup *blkcg, | |
452 | struct request_queue *q, | |
453 | struct blkio_policy_type *pol) | |
454 | { | |
455 | struct blkio_group *blkg; | |
456 | ||
457 | /* alloc and init base part */ | |
458 | blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node); | |
459 | if (!blkg) | |
460 | return NULL; | |
461 | ||
462 | spin_lock_init(&blkg->stats_lock); | |
463 | rcu_assign_pointer(blkg->q, q); | |
464 | blkg->blkcg = blkcg; | |
465 | blkg->plid = pol->plid; | |
466 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); | |
467 | ||
468 | /* alloc per-policy data */ | |
469 | blkg->pd = kzalloc_node(sizeof(*blkg->pd) + pol->pdata_size, GFP_ATOMIC, | |
470 | q->node); | |
471 | if (!blkg->pd) { | |
472 | blkg_free(blkg); | |
473 | return NULL; | |
474 | } | |
475 | ||
476 | /* broken, read comment in the callsite */ | |
477 | blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu); | |
478 | if (!blkg->stats_cpu) { | |
479 | blkg_free(blkg); | |
480 | return NULL; | |
481 | } | |
482 | ||
483 | /* attach pd to blkg and invoke per-policy init */ | |
484 | blkg->pd->blkg = blkg; | |
485 | pol->ops.blkio_init_group_fn(blkg); | |
486 | return blkg; | |
487 | } | |
488 | ||
cd1604fa TH |
489 | struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg, |
490 | struct request_queue *q, | |
491 | enum blkio_policy_id plid, | |
492 | bool for_root) | |
493 | __releases(q->queue_lock) __acquires(q->queue_lock) | |
5624a4e4 | 494 | { |
cd1604fa TH |
495 | struct blkio_policy_type *pol = blkio_policy[plid]; |
496 | struct blkio_group *blkg, *new_blkg; | |
5624a4e4 | 497 | |
cd1604fa TH |
498 | WARN_ON_ONCE(!rcu_read_lock_held()); |
499 | lockdep_assert_held(q->queue_lock); | |
500 | ||
501 | /* | |
502 | * This could be the first entry point of blkcg implementation and | |
503 | * we shouldn't allow anything to go through for a bypassing queue. | |
504 | * The following can be removed if blkg lookup is guaranteed to | |
505 | * fail on a bypassing queue. | |
506 | */ | |
507 | if (unlikely(blk_queue_bypass(q)) && !for_root) | |
508 | return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY); | |
509 | ||
510 | blkg = blkg_lookup(blkcg, q, plid); | |
511 | if (blkg) | |
512 | return blkg; | |
513 | ||
7ee9c562 | 514 | /* blkg holds a reference to blkcg */ |
cd1604fa TH |
515 | if (!css_tryget(&blkcg->css)) |
516 | return ERR_PTR(-EINVAL); | |
517 | ||
518 | /* | |
519 | * Allocate and initialize. | |
520 | * | |
521 | * FIXME: The following is broken. Percpu memory allocation | |
522 | * requires %GFP_KERNEL context and can't be performed from IO | |
523 | * path. Allocation here should inherently be atomic and the | |
524 | * following lock dancing can be removed once the broken percpu | |
525 | * allocation is fixed. | |
526 | */ | |
527 | spin_unlock_irq(q->queue_lock); | |
528 | rcu_read_unlock(); | |
529 | ||
0381411e | 530 | new_blkg = blkg_alloc(blkcg, q, pol); |
cd1604fa TH |
531 | |
532 | rcu_read_lock(); | |
533 | spin_lock_irq(q->queue_lock); | |
31e4c28d | 534 | |
cd1604fa TH |
535 | /* did bypass get turned on inbetween? */ |
536 | if (unlikely(blk_queue_bypass(q)) && !for_root) { | |
537 | blkg = ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY); | |
538 | goto out; | |
539 | } | |
540 | ||
541 | /* did someone beat us to it? */ | |
542 | blkg = blkg_lookup(blkcg, q, plid); | |
543 | if (unlikely(blkg)) | |
544 | goto out; | |
545 | ||
546 | /* did alloc fail? */ | |
0381411e | 547 | if (unlikely(!new_blkg)) { |
cd1604fa TH |
548 | blkg = ERR_PTR(-ENOMEM); |
549 | goto out; | |
550 | } | |
551 | ||
552 | /* insert */ | |
553 | spin_lock(&blkcg->lock); | |
554 | swap(blkg, new_blkg); | |
31e4c28d | 555 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
cd1604fa TH |
556 | pol->ops.blkio_link_group_fn(q, blkg); |
557 | spin_unlock(&blkcg->lock); | |
558 | out: | |
0381411e | 559 | blkg_free(new_blkg); |
cd1604fa | 560 | return blkg; |
31e4c28d | 561 | } |
cd1604fa | 562 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 563 | |
b1c35769 VG |
564 | static void __blkiocg_del_blkio_group(struct blkio_group *blkg) |
565 | { | |
566 | hlist_del_init_rcu(&blkg->blkcg_node); | |
b1c35769 VG |
567 | } |
568 | ||
569 | /* | |
570 | * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1 | |
571 | * indicating that blk_group was unhashed by the time we got to it. | |
572 | */ | |
31e4c28d VG |
573 | int blkiocg_del_blkio_group(struct blkio_group *blkg) |
574 | { | |
7ee9c562 | 575 | struct blkio_cgroup *blkcg = blkg->blkcg; |
b1c35769 | 576 | unsigned long flags; |
b1c35769 VG |
577 | int ret = 1; |
578 | ||
7ee9c562 TH |
579 | spin_lock_irqsave(&blkcg->lock, flags); |
580 | if (!hlist_unhashed(&blkg->blkcg_node)) { | |
581 | __blkiocg_del_blkio_group(blkg); | |
582 | ret = 0; | |
b1c35769 | 583 | } |
7ee9c562 | 584 | spin_unlock_irqrestore(&blkcg->lock, flags); |
0f3942a3 | 585 | |
b1c35769 | 586 | return ret; |
31e4c28d | 587 | } |
9d6a986c | 588 | EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group); |
31e4c28d VG |
589 | |
590 | /* called under rcu_read_lock(). */ | |
cd1604fa TH |
591 | struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg, |
592 | struct request_queue *q, | |
593 | enum blkio_policy_id plid) | |
31e4c28d VG |
594 | { |
595 | struct blkio_group *blkg; | |
596 | struct hlist_node *n; | |
31e4c28d | 597 | |
ca32aefc TH |
598 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) |
599 | if (blkg->q == q && blkg->plid == plid) | |
31e4c28d | 600 | return blkg; |
31e4c28d VG |
601 | return NULL; |
602 | } | |
cd1604fa | 603 | EXPORT_SYMBOL_GPL(blkg_lookup); |
31e4c28d | 604 | |
72e06c25 TH |
605 | void blkg_destroy_all(struct request_queue *q) |
606 | { | |
607 | struct blkio_policy_type *pol; | |
608 | ||
609 | while (true) { | |
610 | bool done = true; | |
611 | ||
612 | spin_lock(&blkio_list_lock); | |
613 | spin_lock_irq(q->queue_lock); | |
614 | ||
615 | /* | |
616 | * clear_queue_fn() might return with non-empty group list | |
617 | * if it raced cgroup removal and lost. cgroup removal is | |
618 | * guaranteed to make forward progress and retrying after a | |
619 | * while is enough. This ugliness is scheduled to be | |
620 | * removed after locking update. | |
621 | */ | |
622 | list_for_each_entry(pol, &blkio_list, list) | |
623 | if (!pol->ops.blkio_clear_queue_fn(q)) | |
624 | done = false; | |
625 | ||
626 | spin_unlock_irq(q->queue_lock); | |
627 | spin_unlock(&blkio_list_lock); | |
628 | ||
629 | if (done) | |
630 | break; | |
631 | ||
632 | msleep(10); /* just some random duration I like */ | |
633 | } | |
634 | } | |
635 | ||
f0bdc8cd VG |
636 | static void blkio_reset_stats_cpu(struct blkio_group *blkg) |
637 | { | |
638 | struct blkio_group_stats_cpu *stats_cpu; | |
639 | int i, j, k; | |
640 | /* | |
641 | * Note: On 64 bit arch this should not be an issue. This has the | |
642 | * possibility of returning some inconsistent value on 32bit arch | |
643 | * as 64bit update on 32bit is non atomic. Taking care of this | |
644 | * corner case makes code very complicated, like sending IPIs to | |
645 | * cpus, taking care of stats of offline cpus etc. | |
646 | * | |
647 | * reset stats is anyway more of a debug feature and this sounds a | |
648 | * corner case. So I am not complicating the code yet until and | |
649 | * unless this becomes a real issue. | |
650 | */ | |
651 | for_each_possible_cpu(i) { | |
652 | stats_cpu = per_cpu_ptr(blkg->stats_cpu, i); | |
653 | stats_cpu->sectors = 0; | |
654 | for(j = 0; j < BLKIO_STAT_CPU_NR; j++) | |
655 | for (k = 0; k < BLKIO_STAT_TOTAL; k++) | |
656 | stats_cpu->stat_arr_cpu[j][k] = 0; | |
657 | } | |
658 | } | |
659 | ||
303a3acb | 660 | static int |
84c124da | 661 | blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val) |
303a3acb DS |
662 | { |
663 | struct blkio_cgroup *blkcg; | |
664 | struct blkio_group *blkg; | |
812df48d | 665 | struct blkio_group_stats *stats; |
303a3acb | 666 | struct hlist_node *n; |
cdc1184c DS |
667 | uint64_t queued[BLKIO_STAT_TOTAL]; |
668 | int i; | |
812df48d DS |
669 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
670 | bool idling, waiting, empty; | |
671 | unsigned long long now = sched_clock(); | |
672 | #endif | |
303a3acb DS |
673 | |
674 | blkcg = cgroup_to_blkio_cgroup(cgroup); | |
675 | spin_lock_irq(&blkcg->lock); | |
676 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
677 | spin_lock(&blkg->stats_lock); | |
812df48d DS |
678 | stats = &blkg->stats; |
679 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
680 | idling = blkio_blkg_idling(stats); | |
681 | waiting = blkio_blkg_waiting(stats); | |
682 | empty = blkio_blkg_empty(stats); | |
683 | #endif | |
cdc1184c | 684 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
812df48d DS |
685 | queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i]; |
686 | memset(stats, 0, sizeof(struct blkio_group_stats)); | |
cdc1184c | 687 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
812df48d DS |
688 | stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i]; |
689 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
690 | if (idling) { | |
691 | blkio_mark_blkg_idling(stats); | |
692 | stats->start_idle_time = now; | |
693 | } | |
694 | if (waiting) { | |
695 | blkio_mark_blkg_waiting(stats); | |
696 | stats->start_group_wait_time = now; | |
697 | } | |
698 | if (empty) { | |
699 | blkio_mark_blkg_empty(stats); | |
700 | stats->start_empty_time = now; | |
701 | } | |
702 | #endif | |
303a3acb | 703 | spin_unlock(&blkg->stats_lock); |
f0bdc8cd VG |
704 | |
705 | /* Reset Per cpu stats which don't take blkg->stats_lock */ | |
706 | blkio_reset_stats_cpu(blkg); | |
303a3acb | 707 | } |
f0bdc8cd | 708 | |
303a3acb DS |
709 | spin_unlock_irq(&blkcg->lock); |
710 | return 0; | |
711 | } | |
712 | ||
7a4dd281 TH |
713 | static void blkio_get_key_name(enum stat_sub_type type, const char *dname, |
714 | char *str, int chars_left, bool diskname_only) | |
303a3acb | 715 | { |
7a4dd281 | 716 | snprintf(str, chars_left, "%s", dname); |
303a3acb DS |
717 | chars_left -= strlen(str); |
718 | if (chars_left <= 0) { | |
719 | printk(KERN_WARNING | |
720 | "Possibly incorrect cgroup stat display format"); | |
721 | return; | |
722 | } | |
84c124da DS |
723 | if (diskname_only) |
724 | return; | |
303a3acb | 725 | switch (type) { |
84c124da | 726 | case BLKIO_STAT_READ: |
303a3acb DS |
727 | strlcat(str, " Read", chars_left); |
728 | break; | |
84c124da | 729 | case BLKIO_STAT_WRITE: |
303a3acb DS |
730 | strlcat(str, " Write", chars_left); |
731 | break; | |
84c124da | 732 | case BLKIO_STAT_SYNC: |
303a3acb DS |
733 | strlcat(str, " Sync", chars_left); |
734 | break; | |
84c124da | 735 | case BLKIO_STAT_ASYNC: |
303a3acb DS |
736 | strlcat(str, " Async", chars_left); |
737 | break; | |
84c124da | 738 | case BLKIO_STAT_TOTAL: |
303a3acb DS |
739 | strlcat(str, " Total", chars_left); |
740 | break; | |
741 | default: | |
742 | strlcat(str, " Invalid", chars_left); | |
743 | } | |
744 | } | |
745 | ||
84c124da | 746 | static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val, |
7a4dd281 | 747 | struct cgroup_map_cb *cb, const char *dname) |
84c124da | 748 | { |
7a4dd281 | 749 | blkio_get_key_name(0, dname, str, chars_left, true); |
84c124da DS |
750 | cb->fill(cb, str, val); |
751 | return val; | |
752 | } | |
303a3acb | 753 | |
5624a4e4 VG |
754 | |
755 | static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg, | |
756 | enum stat_type_cpu type, enum stat_sub_type sub_type) | |
757 | { | |
758 | int cpu; | |
759 | struct blkio_group_stats_cpu *stats_cpu; | |
575969a0 | 760 | u64 val = 0, tval; |
5624a4e4 VG |
761 | |
762 | for_each_possible_cpu(cpu) { | |
575969a0 | 763 | unsigned int start; |
5624a4e4 VG |
764 | stats_cpu = per_cpu_ptr(blkg->stats_cpu, cpu); |
765 | ||
575969a0 VG |
766 | do { |
767 | start = u64_stats_fetch_begin(&stats_cpu->syncp); | |
768 | if (type == BLKIO_STAT_CPU_SECTORS) | |
769 | tval = stats_cpu->sectors; | |
770 | else | |
771 | tval = stats_cpu->stat_arr_cpu[type][sub_type]; | |
772 | } while(u64_stats_fetch_retry(&stats_cpu->syncp, start)); | |
773 | ||
774 | val += tval; | |
5624a4e4 VG |
775 | } |
776 | ||
777 | return val; | |
778 | } | |
779 | ||
780 | static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg, | |
7a4dd281 TH |
781 | struct cgroup_map_cb *cb, const char *dname, |
782 | enum stat_type_cpu type) | |
5624a4e4 VG |
783 | { |
784 | uint64_t disk_total, val; | |
785 | char key_str[MAX_KEY_LEN]; | |
786 | enum stat_sub_type sub_type; | |
787 | ||
788 | if (type == BLKIO_STAT_CPU_SECTORS) { | |
789 | val = blkio_read_stat_cpu(blkg, type, 0); | |
7a4dd281 TH |
790 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, |
791 | dname); | |
5624a4e4 VG |
792 | } |
793 | ||
794 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; | |
795 | sub_type++) { | |
7a4dd281 TH |
796 | blkio_get_key_name(sub_type, dname, key_str, MAX_KEY_LEN, |
797 | false); | |
5624a4e4 VG |
798 | val = blkio_read_stat_cpu(blkg, type, sub_type); |
799 | cb->fill(cb, key_str, val); | |
800 | } | |
801 | ||
802 | disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) + | |
803 | blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE); | |
804 | ||
7a4dd281 TH |
805 | blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN, |
806 | false); | |
5624a4e4 VG |
807 | cb->fill(cb, key_str, disk_total); |
808 | return disk_total; | |
809 | } | |
810 | ||
84c124da DS |
811 | /* This should be called with blkg->stats_lock held */ |
812 | static uint64_t blkio_get_stat(struct blkio_group *blkg, | |
7a4dd281 TH |
813 | struct cgroup_map_cb *cb, const char *dname, |
814 | enum stat_type type) | |
303a3acb DS |
815 | { |
816 | uint64_t disk_total; | |
817 | char key_str[MAX_KEY_LEN]; | |
84c124da DS |
818 | enum stat_sub_type sub_type; |
819 | ||
820 | if (type == BLKIO_STAT_TIME) | |
821 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
7a4dd281 | 822 | blkg->stats.time, cb, dname); |
9026e521 | 823 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 JT |
824 | if (type == BLKIO_STAT_UNACCOUNTED_TIME) |
825 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
7a4dd281 | 826 | blkg->stats.unaccounted_time, cb, dname); |
cdc1184c DS |
827 | if (type == BLKIO_STAT_AVG_QUEUE_SIZE) { |
828 | uint64_t sum = blkg->stats.avg_queue_size_sum; | |
829 | uint64_t samples = blkg->stats.avg_queue_size_samples; | |
830 | if (samples) | |
831 | do_div(sum, samples); | |
832 | else | |
833 | sum = 0; | |
7a4dd281 TH |
834 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, |
835 | sum, cb, dname); | |
cdc1184c | 836 | } |
812df48d DS |
837 | if (type == BLKIO_STAT_GROUP_WAIT_TIME) |
838 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
7a4dd281 | 839 | blkg->stats.group_wait_time, cb, dname); |
812df48d DS |
840 | if (type == BLKIO_STAT_IDLE_TIME) |
841 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
7a4dd281 | 842 | blkg->stats.idle_time, cb, dname); |
812df48d DS |
843 | if (type == BLKIO_STAT_EMPTY_TIME) |
844 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
7a4dd281 | 845 | blkg->stats.empty_time, cb, dname); |
84c124da DS |
846 | if (type == BLKIO_STAT_DEQUEUE) |
847 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
7a4dd281 | 848 | blkg->stats.dequeue, cb, dname); |
84c124da | 849 | #endif |
303a3acb | 850 | |
84c124da DS |
851 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; |
852 | sub_type++) { | |
7a4dd281 TH |
853 | blkio_get_key_name(sub_type, dname, key_str, MAX_KEY_LEN, |
854 | false); | |
84c124da | 855 | cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]); |
303a3acb | 856 | } |
84c124da DS |
857 | disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] + |
858 | blkg->stats.stat_arr[type][BLKIO_STAT_WRITE]; | |
7a4dd281 TH |
859 | blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN, |
860 | false); | |
303a3acb DS |
861 | cb->fill(cb, key_str, disk_total); |
862 | return disk_total; | |
863 | } | |
864 | ||
4bfd482e TH |
865 | static int blkio_policy_parse_and_set(char *buf, enum blkio_policy_id plid, |
866 | int fileid, struct blkio_cgroup *blkcg) | |
34d0f179 | 867 | { |
ece84241 | 868 | struct gendisk *disk = NULL; |
e56da7e2 | 869 | struct blkio_group *blkg = NULL; |
34d0f179 | 870 | char *s[4], *p, *major_s = NULL, *minor_s = NULL; |
d11bb446 | 871 | unsigned long major, minor; |
ece84241 TH |
872 | int i = 0, ret = -EINVAL; |
873 | int part; | |
34d0f179 | 874 | dev_t dev; |
d11bb446 | 875 | u64 temp; |
34d0f179 GJ |
876 | |
877 | memset(s, 0, sizeof(s)); | |
878 | ||
879 | while ((p = strsep(&buf, " ")) != NULL) { | |
880 | if (!*p) | |
881 | continue; | |
882 | ||
883 | s[i++] = p; | |
884 | ||
885 | /* Prevent from inputing too many things */ | |
886 | if (i == 3) | |
887 | break; | |
888 | } | |
889 | ||
890 | if (i != 2) | |
ece84241 | 891 | goto out; |
34d0f179 GJ |
892 | |
893 | p = strsep(&s[0], ":"); | |
894 | if (p != NULL) | |
895 | major_s = p; | |
896 | else | |
ece84241 | 897 | goto out; |
34d0f179 GJ |
898 | |
899 | minor_s = s[0]; | |
900 | if (!minor_s) | |
ece84241 | 901 | goto out; |
34d0f179 | 902 | |
ece84241 TH |
903 | if (strict_strtoul(major_s, 10, &major)) |
904 | goto out; | |
34d0f179 | 905 | |
ece84241 TH |
906 | if (strict_strtoul(minor_s, 10, &minor)) |
907 | goto out; | |
34d0f179 GJ |
908 | |
909 | dev = MKDEV(major, minor); | |
910 | ||
ece84241 TH |
911 | if (strict_strtoull(s[1], 10, &temp)) |
912 | goto out; | |
34d0f179 | 913 | |
e56da7e2 | 914 | disk = get_gendisk(dev, &part); |
4bfd482e | 915 | if (!disk || part) |
e56da7e2 | 916 | goto out; |
e56da7e2 TH |
917 | |
918 | rcu_read_lock(); | |
919 | ||
4bfd482e TH |
920 | spin_lock_irq(disk->queue->queue_lock); |
921 | blkg = blkg_lookup_create(blkcg, disk->queue, plid, false); | |
922 | spin_unlock_irq(disk->queue->queue_lock); | |
e56da7e2 | 923 | |
4bfd482e TH |
924 | if (IS_ERR(blkg)) { |
925 | ret = PTR_ERR(blkg); | |
926 | goto out_unlock; | |
d11bb446 | 927 | } |
34d0f179 | 928 | |
062a644d VG |
929 | switch (plid) { |
930 | case BLKIO_POLICY_PROP: | |
d11bb446 WG |
931 | if ((temp < BLKIO_WEIGHT_MIN && temp > 0) || |
932 | temp > BLKIO_WEIGHT_MAX) | |
e56da7e2 | 933 | goto out_unlock; |
34d0f179 | 934 | |
4bfd482e TH |
935 | blkg->conf.weight = temp; |
936 | blkio_update_group_weight(blkg, temp ?: blkcg->weight); | |
4c9eefa1 VG |
937 | break; |
938 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
939 | switch(fileid) { |
940 | case BLKIO_THROTL_read_bps_device: | |
4bfd482e TH |
941 | blkg->conf.bps[READ] = temp; |
942 | blkio_update_group_bps(blkg, temp ?: -1, fileid); | |
e56da7e2 | 943 | break; |
7702e8f4 | 944 | case BLKIO_THROTL_write_bps_device: |
4bfd482e TH |
945 | blkg->conf.bps[WRITE] = temp; |
946 | blkio_update_group_bps(blkg, temp ?: -1, fileid); | |
7702e8f4 VG |
947 | break; |
948 | case BLKIO_THROTL_read_iops_device: | |
e56da7e2 TH |
949 | if (temp > THROTL_IOPS_MAX) |
950 | goto out_unlock; | |
4bfd482e TH |
951 | blkg->conf.iops[READ] = temp; |
952 | blkio_update_group_iops(blkg, temp ?: -1, fileid); | |
e56da7e2 | 953 | break; |
7702e8f4 | 954 | case BLKIO_THROTL_write_iops_device: |
d11bb446 | 955 | if (temp > THROTL_IOPS_MAX) |
e56da7e2 | 956 | goto out_unlock; |
4bfd482e TH |
957 | blkg->conf.iops[WRITE] = temp; |
958 | blkio_update_group_iops(blkg, temp ?: -1, fileid); | |
7702e8f4 VG |
959 | break; |
960 | } | |
062a644d VG |
961 | break; |
962 | default: | |
963 | BUG(); | |
964 | } | |
ece84241 | 965 | ret = 0; |
e56da7e2 TH |
966 | out_unlock: |
967 | rcu_read_unlock(); | |
ece84241 TH |
968 | out: |
969 | put_disk(disk); | |
e56da7e2 TH |
970 | |
971 | /* | |
972 | * If queue was bypassing, we should retry. Do so after a short | |
973 | * msleep(). It isn't strictly necessary but queue can be | |
974 | * bypassing for some time and it's always nice to avoid busy | |
975 | * looping. | |
976 | */ | |
977 | if (ret == -EBUSY) { | |
978 | msleep(10); | |
979 | return restart_syscall(); | |
980 | } | |
ece84241 | 981 | return ret; |
34d0f179 GJ |
982 | } |
983 | ||
062a644d VG |
984 | static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft, |
985 | const char *buffer) | |
34d0f179 GJ |
986 | { |
987 | int ret = 0; | |
988 | char *buf; | |
e56da7e2 | 989 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); |
062a644d VG |
990 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); |
991 | int fileid = BLKIOFILE_ATTR(cft->private); | |
34d0f179 GJ |
992 | |
993 | buf = kstrdup(buffer, GFP_KERNEL); | |
994 | if (!buf) | |
995 | return -ENOMEM; | |
996 | ||
4bfd482e | 997 | ret = blkio_policy_parse_and_set(buf, plid, fileid, blkcg); |
34d0f179 GJ |
998 | kfree(buf); |
999 | return ret; | |
1000 | } | |
1001 | ||
92616b5b VG |
1002 | static const char *blkg_dev_name(struct blkio_group *blkg) |
1003 | { | |
1004 | /* some drivers (floppy) instantiate a queue w/o disk registered */ | |
1005 | if (blkg->q->backing_dev_info.dev) | |
1006 | return dev_name(blkg->q->backing_dev_info.dev); | |
1007 | return NULL; | |
1008 | } | |
1009 | ||
4bfd482e TH |
1010 | static void blkio_print_group_conf(struct cftype *cft, struct blkio_group *blkg, |
1011 | struct seq_file *m) | |
34d0f179 | 1012 | { |
92616b5b | 1013 | const char *dname = blkg_dev_name(blkg); |
4bfd482e TH |
1014 | int fileid = BLKIOFILE_ATTR(cft->private); |
1015 | int rw = WRITE; | |
1016 | ||
92616b5b VG |
1017 | if (!dname) |
1018 | return; | |
1019 | ||
4bfd482e | 1020 | switch (blkg->plid) { |
062a644d | 1021 | case BLKIO_POLICY_PROP: |
4bfd482e | 1022 | if (blkg->conf.weight) |
7a4dd281 TH |
1023 | seq_printf(m, "%s\t%u\n", |
1024 | dname, blkg->conf.weight); | |
4c9eefa1 VG |
1025 | break; |
1026 | case BLKIO_POLICY_THROTL: | |
4bfd482e | 1027 | switch (fileid) { |
7702e8f4 | 1028 | case BLKIO_THROTL_read_bps_device: |
4bfd482e | 1029 | rw = READ; |
7702e8f4 | 1030 | case BLKIO_THROTL_write_bps_device: |
4bfd482e | 1031 | if (blkg->conf.bps[rw]) |
7a4dd281 TH |
1032 | seq_printf(m, "%s\t%llu\n", |
1033 | dname, blkg->conf.bps[rw]); | |
7702e8f4 VG |
1034 | break; |
1035 | case BLKIO_THROTL_read_iops_device: | |
4bfd482e | 1036 | rw = READ; |
7702e8f4 | 1037 | case BLKIO_THROTL_write_iops_device: |
4bfd482e | 1038 | if (blkg->conf.iops[rw]) |
7a4dd281 TH |
1039 | seq_printf(m, "%s\t%u\n", |
1040 | dname, blkg->conf.iops[rw]); | |
7702e8f4 VG |
1041 | break; |
1042 | } | |
062a644d VG |
1043 | break; |
1044 | default: | |
1045 | BUG(); | |
1046 | } | |
1047 | } | |
34d0f179 | 1048 | |
062a644d | 1049 | /* cgroup files which read their data from policy nodes end up here */ |
4bfd482e TH |
1050 | static void blkio_read_conf(struct cftype *cft, struct blkio_cgroup *blkcg, |
1051 | struct seq_file *m) | |
34d0f179 | 1052 | { |
4bfd482e TH |
1053 | struct blkio_group *blkg; |
1054 | struct hlist_node *n; | |
34d0f179 | 1055 | |
4bfd482e TH |
1056 | spin_lock_irq(&blkcg->lock); |
1057 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) | |
1058 | if (BLKIOFILE_POLICY(cft->private) == blkg->plid) | |
1059 | blkio_print_group_conf(cft, blkg, m); | |
1060 | spin_unlock_irq(&blkcg->lock); | |
062a644d VG |
1061 | } |
1062 | ||
1063 | static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft, | |
1064 | struct seq_file *m) | |
1065 | { | |
1066 | struct blkio_cgroup *blkcg; | |
1067 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1068 | int name = BLKIOFILE_ATTR(cft->private); | |
1069 | ||
1070 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1071 | ||
1072 | switch(plid) { | |
1073 | case BLKIO_POLICY_PROP: | |
1074 | switch(name) { | |
1075 | case BLKIO_PROP_weight_device: | |
4bfd482e | 1076 | blkio_read_conf(cft, blkcg, m); |
062a644d VG |
1077 | return 0; |
1078 | default: | |
1079 | BUG(); | |
1080 | } | |
1081 | break; | |
4c9eefa1 VG |
1082 | case BLKIO_POLICY_THROTL: |
1083 | switch(name){ | |
1084 | case BLKIO_THROTL_read_bps_device: | |
1085 | case BLKIO_THROTL_write_bps_device: | |
7702e8f4 VG |
1086 | case BLKIO_THROTL_read_iops_device: |
1087 | case BLKIO_THROTL_write_iops_device: | |
4bfd482e | 1088 | blkio_read_conf(cft, blkcg, m); |
4c9eefa1 VG |
1089 | return 0; |
1090 | default: | |
1091 | BUG(); | |
1092 | } | |
1093 | break; | |
062a644d VG |
1094 | default: |
1095 | BUG(); | |
1096 | } | |
1097 | ||
1098 | return 0; | |
1099 | } | |
1100 | ||
1101 | static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg, | |
5624a4e4 VG |
1102 | struct cftype *cft, struct cgroup_map_cb *cb, |
1103 | enum stat_type type, bool show_total, bool pcpu) | |
062a644d VG |
1104 | { |
1105 | struct blkio_group *blkg; | |
1106 | struct hlist_node *n; | |
1107 | uint64_t cgroup_total = 0; | |
1108 | ||
1109 | rcu_read_lock(); | |
1110 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
92616b5b | 1111 | const char *dname = blkg_dev_name(blkg); |
7a4dd281 | 1112 | |
92616b5b | 1113 | if (!dname || BLKIOFILE_POLICY(cft->private) != blkg->plid) |
7a4dd281 TH |
1114 | continue; |
1115 | if (pcpu) | |
1116 | cgroup_total += blkio_get_stat_cpu(blkg, cb, dname, | |
1117 | type); | |
1118 | else { | |
1119 | spin_lock_irq(&blkg->stats_lock); | |
1120 | cgroup_total += blkio_get_stat(blkg, cb, dname, type); | |
1121 | spin_unlock_irq(&blkg->stats_lock); | |
062a644d VG |
1122 | } |
1123 | } | |
1124 | if (show_total) | |
1125 | cb->fill(cb, "Total", cgroup_total); | |
1126 | rcu_read_unlock(); | |
1127 | return 0; | |
1128 | } | |
1129 | ||
1130 | /* All map kind of cgroup file get serviced by this function */ | |
1131 | static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft, | |
1132 | struct cgroup_map_cb *cb) | |
1133 | { | |
1134 | struct blkio_cgroup *blkcg; | |
1135 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1136 | int name = BLKIOFILE_ATTR(cft->private); | |
1137 | ||
1138 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1139 | ||
1140 | switch(plid) { | |
1141 | case BLKIO_POLICY_PROP: | |
1142 | switch(name) { | |
1143 | case BLKIO_PROP_time: | |
1144 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1145 | BLKIO_STAT_TIME, 0, 0); |
062a644d VG |
1146 | case BLKIO_PROP_sectors: |
1147 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1148 | BLKIO_STAT_CPU_SECTORS, 0, 1); |
062a644d VG |
1149 | case BLKIO_PROP_io_service_bytes: |
1150 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1151 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
062a644d VG |
1152 | case BLKIO_PROP_io_serviced: |
1153 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1154 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
062a644d VG |
1155 | case BLKIO_PROP_io_service_time: |
1156 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1157 | BLKIO_STAT_SERVICE_TIME, 1, 0); |
062a644d VG |
1158 | case BLKIO_PROP_io_wait_time: |
1159 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1160 | BLKIO_STAT_WAIT_TIME, 1, 0); |
062a644d VG |
1161 | case BLKIO_PROP_io_merged: |
1162 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
317389a7 | 1163 | BLKIO_STAT_CPU_MERGED, 1, 1); |
062a644d VG |
1164 | case BLKIO_PROP_io_queued: |
1165 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1166 | BLKIO_STAT_QUEUED, 1, 0); |
062a644d | 1167 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
9026e521 JT |
1168 | case BLKIO_PROP_unaccounted_time: |
1169 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1170 | BLKIO_STAT_UNACCOUNTED_TIME, 0, 0); |
062a644d VG |
1171 | case BLKIO_PROP_dequeue: |
1172 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1173 | BLKIO_STAT_DEQUEUE, 0, 0); |
062a644d VG |
1174 | case BLKIO_PROP_avg_queue_size: |
1175 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1176 | BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0); |
062a644d VG |
1177 | case BLKIO_PROP_group_wait_time: |
1178 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1179 | BLKIO_STAT_GROUP_WAIT_TIME, 0, 0); |
062a644d VG |
1180 | case BLKIO_PROP_idle_time: |
1181 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1182 | BLKIO_STAT_IDLE_TIME, 0, 0); |
062a644d VG |
1183 | case BLKIO_PROP_empty_time: |
1184 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1185 | BLKIO_STAT_EMPTY_TIME, 0, 0); |
062a644d VG |
1186 | #endif |
1187 | default: | |
1188 | BUG(); | |
1189 | } | |
1190 | break; | |
4c9eefa1 VG |
1191 | case BLKIO_POLICY_THROTL: |
1192 | switch(name){ | |
1193 | case BLKIO_THROTL_io_service_bytes: | |
1194 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1195 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
4c9eefa1 VG |
1196 | case BLKIO_THROTL_io_serviced: |
1197 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1198 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
4c9eefa1 VG |
1199 | default: |
1200 | BUG(); | |
1201 | } | |
1202 | break; | |
062a644d VG |
1203 | default: |
1204 | BUG(); | |
1205 | } | |
1206 | ||
1207 | return 0; | |
1208 | } | |
1209 | ||
4bfd482e | 1210 | static int blkio_weight_write(struct blkio_cgroup *blkcg, int plid, u64 val) |
062a644d VG |
1211 | { |
1212 | struct blkio_group *blkg; | |
1213 | struct hlist_node *n; | |
062a644d VG |
1214 | |
1215 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) | |
1216 | return -EINVAL; | |
1217 | ||
1218 | spin_lock(&blkio_list_lock); | |
1219 | spin_lock_irq(&blkcg->lock); | |
1220 | blkcg->weight = (unsigned int)val; | |
1221 | ||
4bfd482e TH |
1222 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) |
1223 | if (blkg->plid == plid && !blkg->conf.weight) | |
1224 | blkio_update_group_weight(blkg, blkcg->weight); | |
062a644d | 1225 | |
062a644d VG |
1226 | spin_unlock_irq(&blkcg->lock); |
1227 | spin_unlock(&blkio_list_lock); | |
1228 | return 0; | |
1229 | } | |
1230 | ||
1231 | static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) { | |
1232 | struct blkio_cgroup *blkcg; | |
1233 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1234 | int name = BLKIOFILE_ATTR(cft->private); | |
1235 | ||
1236 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1237 | ||
1238 | switch(plid) { | |
1239 | case BLKIO_POLICY_PROP: | |
1240 | switch(name) { | |
1241 | case BLKIO_PROP_weight: | |
1242 | return (u64)blkcg->weight; | |
1243 | } | |
1244 | break; | |
1245 | default: | |
1246 | BUG(); | |
1247 | } | |
1248 | return 0; | |
1249 | } | |
1250 | ||
1251 | static int | |
1252 | blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) | |
1253 | { | |
1254 | struct blkio_cgroup *blkcg; | |
1255 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1256 | int name = BLKIOFILE_ATTR(cft->private); | |
1257 | ||
1258 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1259 | ||
1260 | switch(plid) { | |
1261 | case BLKIO_POLICY_PROP: | |
1262 | switch(name) { | |
1263 | case BLKIO_PROP_weight: | |
4bfd482e | 1264 | return blkio_weight_write(blkcg, plid, val); |
062a644d VG |
1265 | } |
1266 | break; | |
1267 | default: | |
1268 | BUG(); | |
1269 | } | |
34d0f179 | 1270 | |
34d0f179 GJ |
1271 | return 0; |
1272 | } | |
1273 | ||
31e4c28d | 1274 | struct cftype blkio_files[] = { |
34d0f179 GJ |
1275 | { |
1276 | .name = "weight_device", | |
062a644d VG |
1277 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1278 | BLKIO_PROP_weight_device), | |
1279 | .read_seq_string = blkiocg_file_read, | |
1280 | .write_string = blkiocg_file_write, | |
34d0f179 GJ |
1281 | .max_write_len = 256, |
1282 | }, | |
31e4c28d VG |
1283 | { |
1284 | .name = "weight", | |
062a644d VG |
1285 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1286 | BLKIO_PROP_weight), | |
1287 | .read_u64 = blkiocg_file_read_u64, | |
1288 | .write_u64 = blkiocg_file_write_u64, | |
31e4c28d | 1289 | }, |
22084190 VG |
1290 | { |
1291 | .name = "time", | |
062a644d VG |
1292 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1293 | BLKIO_PROP_time), | |
1294 | .read_map = blkiocg_file_read_map, | |
22084190 VG |
1295 | }, |
1296 | { | |
1297 | .name = "sectors", | |
062a644d VG |
1298 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1299 | BLKIO_PROP_sectors), | |
1300 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1301 | }, |
1302 | { | |
1303 | .name = "io_service_bytes", | |
062a644d VG |
1304 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1305 | BLKIO_PROP_io_service_bytes), | |
1306 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1307 | }, |
1308 | { | |
1309 | .name = "io_serviced", | |
062a644d VG |
1310 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1311 | BLKIO_PROP_io_serviced), | |
1312 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1313 | }, |
1314 | { | |
1315 | .name = "io_service_time", | |
062a644d VG |
1316 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1317 | BLKIO_PROP_io_service_time), | |
1318 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1319 | }, |
1320 | { | |
1321 | .name = "io_wait_time", | |
062a644d VG |
1322 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1323 | BLKIO_PROP_io_wait_time), | |
1324 | .read_map = blkiocg_file_read_map, | |
84c124da | 1325 | }, |
812d4026 DS |
1326 | { |
1327 | .name = "io_merged", | |
062a644d VG |
1328 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1329 | BLKIO_PROP_io_merged), | |
1330 | .read_map = blkiocg_file_read_map, | |
812d4026 | 1331 | }, |
cdc1184c DS |
1332 | { |
1333 | .name = "io_queued", | |
062a644d VG |
1334 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1335 | BLKIO_PROP_io_queued), | |
1336 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1337 | }, |
84c124da DS |
1338 | { |
1339 | .name = "reset_stats", | |
1340 | .write_u64 = blkiocg_reset_stats, | |
22084190 | 1341 | }, |
13f98250 VG |
1342 | #ifdef CONFIG_BLK_DEV_THROTTLING |
1343 | { | |
1344 | .name = "throttle.read_bps_device", | |
1345 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1346 | BLKIO_THROTL_read_bps_device), | |
1347 | .read_seq_string = blkiocg_file_read, | |
1348 | .write_string = blkiocg_file_write, | |
1349 | .max_write_len = 256, | |
1350 | }, | |
1351 | ||
1352 | { | |
1353 | .name = "throttle.write_bps_device", | |
1354 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1355 | BLKIO_THROTL_write_bps_device), | |
1356 | .read_seq_string = blkiocg_file_read, | |
1357 | .write_string = blkiocg_file_write, | |
1358 | .max_write_len = 256, | |
1359 | }, | |
1360 | ||
1361 | { | |
1362 | .name = "throttle.read_iops_device", | |
1363 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1364 | BLKIO_THROTL_read_iops_device), | |
1365 | .read_seq_string = blkiocg_file_read, | |
1366 | .write_string = blkiocg_file_write, | |
1367 | .max_write_len = 256, | |
1368 | }, | |
1369 | ||
1370 | { | |
1371 | .name = "throttle.write_iops_device", | |
1372 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1373 | BLKIO_THROTL_write_iops_device), | |
1374 | .read_seq_string = blkiocg_file_read, | |
1375 | .write_string = blkiocg_file_write, | |
1376 | .max_write_len = 256, | |
1377 | }, | |
1378 | { | |
1379 | .name = "throttle.io_service_bytes", | |
1380 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1381 | BLKIO_THROTL_io_service_bytes), | |
1382 | .read_map = blkiocg_file_read_map, | |
1383 | }, | |
1384 | { | |
1385 | .name = "throttle.io_serviced", | |
1386 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1387 | BLKIO_THROTL_io_serviced), | |
1388 | .read_map = blkiocg_file_read_map, | |
1389 | }, | |
1390 | #endif /* CONFIG_BLK_DEV_THROTTLING */ | |
1391 | ||
22084190 | 1392 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
cdc1184c DS |
1393 | { |
1394 | .name = "avg_queue_size", | |
062a644d VG |
1395 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1396 | BLKIO_PROP_avg_queue_size), | |
1397 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1398 | }, |
812df48d DS |
1399 | { |
1400 | .name = "group_wait_time", | |
062a644d VG |
1401 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1402 | BLKIO_PROP_group_wait_time), | |
1403 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1404 | }, |
1405 | { | |
1406 | .name = "idle_time", | |
062a644d VG |
1407 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1408 | BLKIO_PROP_idle_time), | |
1409 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1410 | }, |
1411 | { | |
1412 | .name = "empty_time", | |
062a644d VG |
1413 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1414 | BLKIO_PROP_empty_time), | |
1415 | .read_map = blkiocg_file_read_map, | |
812df48d | 1416 | }, |
cdc1184c | 1417 | { |
22084190 | 1418 | .name = "dequeue", |
062a644d VG |
1419 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1420 | BLKIO_PROP_dequeue), | |
1421 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1422 | }, |
9026e521 JT |
1423 | { |
1424 | .name = "unaccounted_time", | |
1425 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, | |
1426 | BLKIO_PROP_unaccounted_time), | |
1427 | .read_map = blkiocg_file_read_map, | |
1428 | }, | |
22084190 | 1429 | #endif |
31e4c28d VG |
1430 | }; |
1431 | ||
1432 | static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1433 | { | |
1434 | return cgroup_add_files(cgroup, subsys, blkio_files, | |
1435 | ARRAY_SIZE(blkio_files)); | |
1436 | } | |
1437 | ||
7ee9c562 TH |
1438 | static int blkiocg_pre_destroy(struct cgroup_subsys *subsys, |
1439 | struct cgroup *cgroup) | |
31e4c28d VG |
1440 | { |
1441 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
b1c35769 VG |
1442 | unsigned long flags; |
1443 | struct blkio_group *blkg; | |
ca32aefc | 1444 | struct request_queue *q; |
3e252066 | 1445 | struct blkio_policy_type *blkiop; |
b1c35769 VG |
1446 | |
1447 | rcu_read_lock(); | |
7ee9c562 | 1448 | |
0f3942a3 JA |
1449 | do { |
1450 | spin_lock_irqsave(&blkcg->lock, flags); | |
b1c35769 | 1451 | |
0f3942a3 JA |
1452 | if (hlist_empty(&blkcg->blkg_list)) { |
1453 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
1454 | break; | |
1455 | } | |
b1c35769 | 1456 | |
0f3942a3 JA |
1457 | blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group, |
1458 | blkcg_node); | |
ca32aefc | 1459 | q = rcu_dereference(blkg->q); |
0f3942a3 | 1460 | __blkiocg_del_blkio_group(blkg); |
31e4c28d | 1461 | |
0f3942a3 | 1462 | spin_unlock_irqrestore(&blkcg->lock, flags); |
b1c35769 | 1463 | |
0f3942a3 JA |
1464 | /* |
1465 | * This blkio_group is being unlinked as associated cgroup is | |
1466 | * going away. Let all the IO controlling policies know about | |
61014e96 | 1467 | * this event. |
0f3942a3 JA |
1468 | */ |
1469 | spin_lock(&blkio_list_lock); | |
61014e96 VG |
1470 | list_for_each_entry(blkiop, &blkio_list, list) { |
1471 | if (blkiop->plid != blkg->plid) | |
1472 | continue; | |
ca32aefc | 1473 | blkiop->ops.blkio_unlink_group_fn(q, blkg); |
61014e96 | 1474 | } |
0f3942a3 JA |
1475 | spin_unlock(&blkio_list_lock); |
1476 | } while (1); | |
34d0f179 | 1477 | |
b1c35769 | 1478 | rcu_read_unlock(); |
7ee9c562 TH |
1479 | |
1480 | return 0; | |
1481 | } | |
1482 | ||
1483 | static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1484 | { | |
1485 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
1486 | ||
67523c48 BB |
1487 | if (blkcg != &blkio_root_cgroup) |
1488 | kfree(blkcg); | |
31e4c28d VG |
1489 | } |
1490 | ||
1491 | static struct cgroup_subsys_state * | |
1492 | blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1493 | { | |
0341509f LZ |
1494 | struct blkio_cgroup *blkcg; |
1495 | struct cgroup *parent = cgroup->parent; | |
31e4c28d | 1496 | |
0341509f | 1497 | if (!parent) { |
31e4c28d VG |
1498 | blkcg = &blkio_root_cgroup; |
1499 | goto done; | |
1500 | } | |
1501 | ||
31e4c28d VG |
1502 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
1503 | if (!blkcg) | |
1504 | return ERR_PTR(-ENOMEM); | |
1505 | ||
1506 | blkcg->weight = BLKIO_WEIGHT_DEFAULT; | |
1507 | done: | |
1508 | spin_lock_init(&blkcg->lock); | |
1509 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
1510 | ||
1511 | return &blkcg->css; | |
1512 | } | |
1513 | ||
5efd6113 TH |
1514 | /** |
1515 | * blkcg_init_queue - initialize blkcg part of request queue | |
1516 | * @q: request_queue to initialize | |
1517 | * | |
1518 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1519 | * part of new request_queue @q. | |
1520 | * | |
1521 | * RETURNS: | |
1522 | * 0 on success, -errno on failure. | |
1523 | */ | |
1524 | int blkcg_init_queue(struct request_queue *q) | |
1525 | { | |
923adde1 TH |
1526 | int ret; |
1527 | ||
5efd6113 TH |
1528 | might_sleep(); |
1529 | ||
923adde1 TH |
1530 | ret = blk_throtl_init(q); |
1531 | if (ret) | |
1532 | return ret; | |
1533 | ||
1534 | mutex_lock(&all_q_mutex); | |
1535 | INIT_LIST_HEAD(&q->all_q_node); | |
1536 | list_add_tail(&q->all_q_node, &all_q_list); | |
1537 | mutex_unlock(&all_q_mutex); | |
1538 | ||
1539 | return 0; | |
5efd6113 TH |
1540 | } |
1541 | ||
1542 | /** | |
1543 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1544 | * @q: request_queue to drain | |
1545 | * | |
1546 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1547 | */ | |
1548 | void blkcg_drain_queue(struct request_queue *q) | |
1549 | { | |
1550 | lockdep_assert_held(q->queue_lock); | |
1551 | ||
1552 | blk_throtl_drain(q); | |
1553 | } | |
1554 | ||
1555 | /** | |
1556 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1557 | * @q: request_queue being released | |
1558 | * | |
1559 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1560 | */ | |
1561 | void blkcg_exit_queue(struct request_queue *q) | |
1562 | { | |
923adde1 TH |
1563 | mutex_lock(&all_q_mutex); |
1564 | list_del_init(&q->all_q_node); | |
1565 | mutex_unlock(&all_q_mutex); | |
1566 | ||
5efd6113 TH |
1567 | blk_throtl_exit(q); |
1568 | } | |
1569 | ||
31e4c28d VG |
1570 | /* |
1571 | * We cannot support shared io contexts, as we have no mean to support | |
1572 | * two tasks with the same ioc in two different groups without major rework | |
1573 | * of the main cic data structures. For now we allow a task to change | |
1574 | * its cgroup only if it's the only owner of its ioc. | |
1575 | */ | |
bb9d97b6 TH |
1576 | static int blkiocg_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
1577 | struct cgroup_taskset *tset) | |
31e4c28d | 1578 | { |
bb9d97b6 | 1579 | struct task_struct *task; |
31e4c28d VG |
1580 | struct io_context *ioc; |
1581 | int ret = 0; | |
1582 | ||
1583 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
bb9d97b6 TH |
1584 | cgroup_taskset_for_each(task, cgrp, tset) { |
1585 | task_lock(task); | |
1586 | ioc = task->io_context; | |
1587 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1588 | ret = -EINVAL; | |
1589 | task_unlock(task); | |
1590 | if (ret) | |
1591 | break; | |
1592 | } | |
31e4c28d VG |
1593 | return ret; |
1594 | } | |
1595 | ||
bb9d97b6 TH |
1596 | static void blkiocg_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
1597 | struct cgroup_taskset *tset) | |
31e4c28d | 1598 | { |
bb9d97b6 | 1599 | struct task_struct *task; |
31e4c28d VG |
1600 | struct io_context *ioc; |
1601 | ||
bb9d97b6 | 1602 | cgroup_taskset_for_each(task, cgrp, tset) { |
b3c9dd18 LT |
1603 | /* we don't lose anything even if ioc allocation fails */ |
1604 | ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE); | |
1605 | if (ioc) { | |
1606 | ioc_cgroup_changed(ioc); | |
11a3122f | 1607 | put_io_context(ioc); |
b3c9dd18 | 1608 | } |
bb9d97b6 | 1609 | } |
31e4c28d VG |
1610 | } |
1611 | ||
923adde1 TH |
1612 | static void blkcg_bypass_start(void) |
1613 | __acquires(&all_q_mutex) | |
1614 | { | |
1615 | struct request_queue *q; | |
1616 | ||
1617 | mutex_lock(&all_q_mutex); | |
1618 | ||
1619 | list_for_each_entry(q, &all_q_list, all_q_node) { | |
1620 | blk_queue_bypass_start(q); | |
1621 | blkg_destroy_all(q); | |
1622 | } | |
1623 | } | |
1624 | ||
1625 | static void blkcg_bypass_end(void) | |
1626 | __releases(&all_q_mutex) | |
1627 | { | |
1628 | struct request_queue *q; | |
1629 | ||
1630 | list_for_each_entry(q, &all_q_list, all_q_node) | |
1631 | blk_queue_bypass_end(q); | |
1632 | ||
1633 | mutex_unlock(&all_q_mutex); | |
1634 | } | |
1635 | ||
3e252066 VG |
1636 | void blkio_policy_register(struct blkio_policy_type *blkiop) |
1637 | { | |
923adde1 | 1638 | blkcg_bypass_start(); |
3e252066 | 1639 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1640 | |
1641 | BUG_ON(blkio_policy[blkiop->plid]); | |
1642 | blkio_policy[blkiop->plid] = blkiop; | |
3e252066 | 1643 | list_add_tail(&blkiop->list, &blkio_list); |
035d10b2 | 1644 | |
3e252066 | 1645 | spin_unlock(&blkio_list_lock); |
923adde1 | 1646 | blkcg_bypass_end(); |
3e252066 VG |
1647 | } |
1648 | EXPORT_SYMBOL_GPL(blkio_policy_register); | |
1649 | ||
1650 | void blkio_policy_unregister(struct blkio_policy_type *blkiop) | |
1651 | { | |
923adde1 | 1652 | blkcg_bypass_start(); |
3e252066 | 1653 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1654 | |
1655 | BUG_ON(blkio_policy[blkiop->plid] != blkiop); | |
1656 | blkio_policy[blkiop->plid] = NULL; | |
3e252066 | 1657 | list_del_init(&blkiop->list); |
035d10b2 | 1658 | |
3e252066 | 1659 | spin_unlock(&blkio_list_lock); |
923adde1 | 1660 | blkcg_bypass_end(); |
3e252066 VG |
1661 | } |
1662 | EXPORT_SYMBOL_GPL(blkio_policy_unregister); |