]>
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" | |
3e252066 | 23 | |
84c124da DS |
24 | #define MAX_KEY_LEN 100 |
25 | ||
3e252066 VG |
26 | static DEFINE_SPINLOCK(blkio_list_lock); |
27 | static LIST_HEAD(blkio_list); | |
b1c35769 | 28 | |
31e4c28d | 29 | struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT }; |
9d6a986c VG |
30 | EXPORT_SYMBOL_GPL(blkio_root_cgroup); |
31 | ||
67523c48 BB |
32 | static struct cgroup_subsys_state *blkiocg_create(struct cgroup_subsys *, |
33 | struct cgroup *); | |
bb9d97b6 TH |
34 | static int blkiocg_can_attach(struct cgroup_subsys *, struct cgroup *, |
35 | struct cgroup_taskset *); | |
36 | static void blkiocg_attach(struct cgroup_subsys *, struct cgroup *, | |
37 | struct cgroup_taskset *); | |
67523c48 BB |
38 | static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *); |
39 | static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *); | |
40 | ||
062a644d VG |
41 | /* for encoding cft->private value on file */ |
42 | #define BLKIOFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
43 | /* What policy owns the file, proportional or throttle */ | |
44 | #define BLKIOFILE_POLICY(val) (((val) >> 16) & 0xffff) | |
45 | #define BLKIOFILE_ATTR(val) ((val) & 0xffff) | |
46 | ||
67523c48 BB |
47 | struct cgroup_subsys blkio_subsys = { |
48 | .name = "blkio", | |
49 | .create = blkiocg_create, | |
bb9d97b6 TH |
50 | .can_attach = blkiocg_can_attach, |
51 | .attach = blkiocg_attach, | |
67523c48 BB |
52 | .destroy = blkiocg_destroy, |
53 | .populate = blkiocg_populate, | |
67523c48 | 54 | .subsys_id = blkio_subsys_id, |
67523c48 BB |
55 | .use_id = 1, |
56 | .module = THIS_MODULE, | |
57 | }; | |
58 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
59 | ||
34d0f179 GJ |
60 | static inline void blkio_policy_insert_node(struct blkio_cgroup *blkcg, |
61 | struct blkio_policy_node *pn) | |
62 | { | |
63 | list_add(&pn->node, &blkcg->policy_list); | |
64 | } | |
65 | ||
062a644d VG |
66 | static inline bool cftype_blkg_same_policy(struct cftype *cft, |
67 | struct blkio_group *blkg) | |
68 | { | |
69 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
70 | ||
71 | if (blkg->plid == plid) | |
72 | return 1; | |
73 | ||
74 | return 0; | |
75 | } | |
76 | ||
77 | /* Determines if policy node matches cgroup file being accessed */ | |
78 | static inline bool pn_matches_cftype(struct cftype *cft, | |
79 | struct blkio_policy_node *pn) | |
80 | { | |
81 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
82 | int fileid = BLKIOFILE_ATTR(cft->private); | |
83 | ||
84 | return (plid == pn->plid && fileid == pn->fileid); | |
85 | } | |
86 | ||
34d0f179 GJ |
87 | /* Must be called with blkcg->lock held */ |
88 | static inline void blkio_policy_delete_node(struct blkio_policy_node *pn) | |
89 | { | |
90 | list_del(&pn->node); | |
91 | } | |
92 | ||
93 | /* Must be called with blkcg->lock held */ | |
94 | static struct blkio_policy_node * | |
062a644d VG |
95 | blkio_policy_search_node(const struct blkio_cgroup *blkcg, dev_t dev, |
96 | enum blkio_policy_id plid, int fileid) | |
34d0f179 GJ |
97 | { |
98 | struct blkio_policy_node *pn; | |
99 | ||
100 | list_for_each_entry(pn, &blkcg->policy_list, node) { | |
062a644d | 101 | if (pn->dev == dev && pn->plid == plid && pn->fileid == fileid) |
34d0f179 GJ |
102 | return pn; |
103 | } | |
104 | ||
105 | return NULL; | |
106 | } | |
107 | ||
31e4c28d VG |
108 | struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup) |
109 | { | |
110 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), | |
111 | struct blkio_cgroup, css); | |
112 | } | |
9d6a986c | 113 | EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup); |
31e4c28d | 114 | |
70087dc3 VG |
115 | struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk) |
116 | { | |
117 | return container_of(task_subsys_state(tsk, blkio_subsys_id), | |
118 | struct blkio_cgroup, css); | |
119 | } | |
120 | EXPORT_SYMBOL_GPL(task_blkio_cgroup); | |
121 | ||
062a644d VG |
122 | static inline void |
123 | blkio_update_group_weight(struct blkio_group *blkg, unsigned int weight) | |
124 | { | |
125 | struct blkio_policy_type *blkiop; | |
126 | ||
127 | list_for_each_entry(blkiop, &blkio_list, list) { | |
128 | /* If this policy does not own the blkg, do not send updates */ | |
129 | if (blkiop->plid != blkg->plid) | |
130 | continue; | |
131 | if (blkiop->ops.blkio_update_group_weight_fn) | |
ca32aefc | 132 | blkiop->ops.blkio_update_group_weight_fn(blkg->q, |
fe071437 | 133 | blkg, weight); |
062a644d VG |
134 | } |
135 | } | |
136 | ||
4c9eefa1 VG |
137 | static inline void blkio_update_group_bps(struct blkio_group *blkg, u64 bps, |
138 | int fileid) | |
139 | { | |
140 | struct blkio_policy_type *blkiop; | |
141 | ||
142 | list_for_each_entry(blkiop, &blkio_list, list) { | |
143 | ||
144 | /* If this policy does not own the blkg, do not send updates */ | |
145 | if (blkiop->plid != blkg->plid) | |
146 | continue; | |
147 | ||
148 | if (fileid == BLKIO_THROTL_read_bps_device | |
149 | && blkiop->ops.blkio_update_group_read_bps_fn) | |
ca32aefc | 150 | blkiop->ops.blkio_update_group_read_bps_fn(blkg->q, |
fe071437 | 151 | blkg, bps); |
4c9eefa1 VG |
152 | |
153 | if (fileid == BLKIO_THROTL_write_bps_device | |
154 | && blkiop->ops.blkio_update_group_write_bps_fn) | |
ca32aefc | 155 | blkiop->ops.blkio_update_group_write_bps_fn(blkg->q, |
fe071437 | 156 | blkg, bps); |
4c9eefa1 VG |
157 | } |
158 | } | |
159 | ||
7702e8f4 VG |
160 | static inline void blkio_update_group_iops(struct blkio_group *blkg, |
161 | unsigned int iops, int fileid) | |
162 | { | |
163 | struct blkio_policy_type *blkiop; | |
164 | ||
165 | list_for_each_entry(blkiop, &blkio_list, list) { | |
166 | ||
167 | /* If this policy does not own the blkg, do not send updates */ | |
168 | if (blkiop->plid != blkg->plid) | |
169 | continue; | |
170 | ||
171 | if (fileid == BLKIO_THROTL_read_iops_device | |
172 | && blkiop->ops.blkio_update_group_read_iops_fn) | |
ca32aefc | 173 | blkiop->ops.blkio_update_group_read_iops_fn(blkg->q, |
fe071437 | 174 | blkg, iops); |
7702e8f4 VG |
175 | |
176 | if (fileid == BLKIO_THROTL_write_iops_device | |
177 | && blkiop->ops.blkio_update_group_write_iops_fn) | |
ca32aefc | 178 | blkiop->ops.blkio_update_group_write_iops_fn(blkg->q, |
fe071437 | 179 | blkg,iops); |
7702e8f4 VG |
180 | } |
181 | } | |
182 | ||
9195291e DS |
183 | /* |
184 | * Add to the appropriate stat variable depending on the request type. | |
185 | * This should be called with the blkg->stats_lock held. | |
186 | */ | |
84c124da DS |
187 | static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction, |
188 | bool sync) | |
9195291e | 189 | { |
84c124da DS |
190 | if (direction) |
191 | stat[BLKIO_STAT_WRITE] += add; | |
9195291e | 192 | else |
84c124da DS |
193 | stat[BLKIO_STAT_READ] += add; |
194 | if (sync) | |
195 | stat[BLKIO_STAT_SYNC] += add; | |
9195291e | 196 | else |
84c124da | 197 | stat[BLKIO_STAT_ASYNC] += add; |
9195291e DS |
198 | } |
199 | ||
cdc1184c DS |
200 | /* |
201 | * Decrements the appropriate stat variable if non-zero depending on the | |
202 | * request type. Panics on value being zero. | |
203 | * This should be called with the blkg->stats_lock held. | |
204 | */ | |
205 | static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync) | |
206 | { | |
207 | if (direction) { | |
208 | BUG_ON(stat[BLKIO_STAT_WRITE] == 0); | |
209 | stat[BLKIO_STAT_WRITE]--; | |
210 | } else { | |
211 | BUG_ON(stat[BLKIO_STAT_READ] == 0); | |
212 | stat[BLKIO_STAT_READ]--; | |
213 | } | |
214 | if (sync) { | |
215 | BUG_ON(stat[BLKIO_STAT_SYNC] == 0); | |
216 | stat[BLKIO_STAT_SYNC]--; | |
217 | } else { | |
218 | BUG_ON(stat[BLKIO_STAT_ASYNC] == 0); | |
219 | stat[BLKIO_STAT_ASYNC]--; | |
220 | } | |
221 | } | |
222 | ||
223 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
812df48d DS |
224 | /* This should be called with the blkg->stats_lock held. */ |
225 | static void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
226 | struct blkio_group *curr_blkg) | |
227 | { | |
228 | if (blkio_blkg_waiting(&blkg->stats)) | |
229 | return; | |
230 | if (blkg == curr_blkg) | |
231 | return; | |
232 | blkg->stats.start_group_wait_time = sched_clock(); | |
233 | blkio_mark_blkg_waiting(&blkg->stats); | |
234 | } | |
235 | ||
236 | /* This should be called with the blkg->stats_lock held. */ | |
237 | static void blkio_update_group_wait_time(struct blkio_group_stats *stats) | |
238 | { | |
239 | unsigned long long now; | |
240 | ||
241 | if (!blkio_blkg_waiting(stats)) | |
242 | return; | |
243 | ||
244 | now = sched_clock(); | |
245 | if (time_after64(now, stats->start_group_wait_time)) | |
246 | stats->group_wait_time += now - stats->start_group_wait_time; | |
247 | blkio_clear_blkg_waiting(stats); | |
248 | } | |
249 | ||
250 | /* This should be called with the blkg->stats_lock held. */ | |
251 | static void blkio_end_empty_time(struct blkio_group_stats *stats) | |
252 | { | |
253 | unsigned long long now; | |
254 | ||
255 | if (!blkio_blkg_empty(stats)) | |
256 | return; | |
257 | ||
258 | now = sched_clock(); | |
259 | if (time_after64(now, stats->start_empty_time)) | |
260 | stats->empty_time += now - stats->start_empty_time; | |
261 | blkio_clear_blkg_empty(stats); | |
262 | } | |
263 | ||
264 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg) | |
265 | { | |
266 | unsigned long flags; | |
267 | ||
268 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
269 | BUG_ON(blkio_blkg_idling(&blkg->stats)); | |
270 | blkg->stats.start_idle_time = sched_clock(); | |
271 | blkio_mark_blkg_idling(&blkg->stats); | |
272 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
273 | } | |
274 | EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats); | |
275 | ||
276 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg) | |
277 | { | |
278 | unsigned long flags; | |
279 | unsigned long long now; | |
280 | struct blkio_group_stats *stats; | |
281 | ||
282 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
283 | stats = &blkg->stats; | |
284 | if (blkio_blkg_idling(stats)) { | |
285 | now = sched_clock(); | |
286 | if (time_after64(now, stats->start_idle_time)) | |
287 | stats->idle_time += now - stats->start_idle_time; | |
288 | blkio_clear_blkg_idling(stats); | |
289 | } | |
290 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
291 | } | |
292 | EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats); | |
293 | ||
a11cdaa7 | 294 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg) |
cdc1184c DS |
295 | { |
296 | unsigned long flags; | |
297 | struct blkio_group_stats *stats; | |
298 | ||
299 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
300 | stats = &blkg->stats; | |
301 | stats->avg_queue_size_sum += | |
302 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] + | |
303 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]; | |
304 | stats->avg_queue_size_samples++; | |
812df48d | 305 | blkio_update_group_wait_time(stats); |
cdc1184c DS |
306 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
307 | } | |
a11cdaa7 DS |
308 | EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats); |
309 | ||
e5ff082e | 310 | void blkiocg_set_start_empty_time(struct blkio_group *blkg) |
28baf442 DS |
311 | { |
312 | unsigned long flags; | |
313 | struct blkio_group_stats *stats; | |
314 | ||
315 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
316 | stats = &blkg->stats; | |
317 | ||
318 | if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] || | |
319 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) { | |
320 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
321 | return; | |
322 | } | |
323 | ||
324 | /* | |
e5ff082e VG |
325 | * group is already marked empty. This can happen if cfqq got new |
326 | * request in parent group and moved to this group while being added | |
327 | * to service tree. Just ignore the event and move on. | |
28baf442 | 328 | */ |
e5ff082e VG |
329 | if(blkio_blkg_empty(stats)) { |
330 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
331 | return; | |
332 | } | |
333 | ||
28baf442 DS |
334 | stats->start_empty_time = sched_clock(); |
335 | blkio_mark_blkg_empty(stats); | |
336 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
337 | } | |
338 | EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time); | |
339 | ||
a11cdaa7 DS |
340 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
341 | unsigned long dequeue) | |
342 | { | |
343 | blkg->stats.dequeue += dequeue; | |
344 | } | |
345 | EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats); | |
812df48d DS |
346 | #else |
347 | static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
348 | struct blkio_group *curr_blkg) {} | |
349 | static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {} | |
cdc1184c DS |
350 | #endif |
351 | ||
a11cdaa7 | 352 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c DS |
353 | struct blkio_group *curr_blkg, bool direction, |
354 | bool sync) | |
355 | { | |
356 | unsigned long flags; | |
357 | ||
358 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
359 | blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction, | |
360 | sync); | |
812df48d DS |
361 | blkio_end_empty_time(&blkg->stats); |
362 | blkio_set_start_group_wait_time(blkg, curr_blkg); | |
cdc1184c DS |
363 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
364 | } | |
a11cdaa7 | 365 | EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats); |
cdc1184c | 366 | |
a11cdaa7 | 367 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c DS |
368 | bool direction, bool sync) |
369 | { | |
370 | unsigned long flags; | |
371 | ||
372 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
373 | blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], | |
374 | direction, sync); | |
375 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
376 | } | |
a11cdaa7 | 377 | EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats); |
cdc1184c | 378 | |
167400d3 JT |
379 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time, |
380 | unsigned long unaccounted_time) | |
22084190 | 381 | { |
303a3acb DS |
382 | unsigned long flags; |
383 | ||
384 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
385 | blkg->stats.time += time; | |
a23e6869 | 386 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 | 387 | blkg->stats.unaccounted_time += unaccounted_time; |
a23e6869 | 388 | #endif |
303a3acb | 389 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
22084190 | 390 | } |
303a3acb | 391 | EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used); |
22084190 | 392 | |
5624a4e4 VG |
393 | /* |
394 | * should be called under rcu read lock or queue lock to make sure blkg pointer | |
395 | * is valid. | |
396 | */ | |
84c124da DS |
397 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
398 | uint64_t bytes, bool direction, bool sync) | |
9195291e | 399 | { |
5624a4e4 | 400 | struct blkio_group_stats_cpu *stats_cpu; |
575969a0 VG |
401 | unsigned long flags; |
402 | ||
403 | /* | |
404 | * Disabling interrupts to provide mutual exclusion between two | |
405 | * writes on same cpu. It probably is not needed for 64bit. Not | |
406 | * optimizing that case yet. | |
407 | */ | |
408 | local_irq_save(flags); | |
9195291e | 409 | |
5624a4e4 VG |
410 | stats_cpu = this_cpu_ptr(blkg->stats_cpu); |
411 | ||
575969a0 | 412 | u64_stats_update_begin(&stats_cpu->syncp); |
5624a4e4 VG |
413 | stats_cpu->sectors += bytes >> 9; |
414 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED], | |
415 | 1, direction, sync); | |
416 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES], | |
417 | bytes, direction, sync); | |
575969a0 VG |
418 | u64_stats_update_end(&stats_cpu->syncp); |
419 | local_irq_restore(flags); | |
9195291e | 420 | } |
84c124da | 421 | EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats); |
9195291e | 422 | |
84c124da DS |
423 | void blkiocg_update_completion_stats(struct blkio_group *blkg, |
424 | uint64_t start_time, uint64_t io_start_time, bool direction, bool sync) | |
9195291e DS |
425 | { |
426 | struct blkio_group_stats *stats; | |
427 | unsigned long flags; | |
428 | unsigned long long now = sched_clock(); | |
429 | ||
430 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
431 | stats = &blkg->stats; | |
84c124da DS |
432 | if (time_after64(now, io_start_time)) |
433 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME], | |
434 | now - io_start_time, direction, sync); | |
435 | if (time_after64(io_start_time, start_time)) | |
436 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME], | |
437 | io_start_time - start_time, direction, sync); | |
9195291e DS |
438 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
439 | } | |
84c124da | 440 | EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats); |
9195291e | 441 | |
317389a7 | 442 | /* Merged stats are per cpu. */ |
812d4026 DS |
443 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction, |
444 | bool sync) | |
445 | { | |
317389a7 | 446 | struct blkio_group_stats_cpu *stats_cpu; |
812d4026 DS |
447 | unsigned long flags; |
448 | ||
317389a7 VG |
449 | /* |
450 | * Disabling interrupts to provide mutual exclusion between two | |
451 | * writes on same cpu. It probably is not needed for 64bit. Not | |
452 | * optimizing that case yet. | |
453 | */ | |
454 | local_irq_save(flags); | |
455 | ||
456 | stats_cpu = this_cpu_ptr(blkg->stats_cpu); | |
457 | ||
458 | u64_stats_update_begin(&stats_cpu->syncp); | |
459 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1, | |
460 | direction, sync); | |
461 | u64_stats_update_end(&stats_cpu->syncp); | |
462 | local_irq_restore(flags); | |
812d4026 DS |
463 | } |
464 | EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats); | |
465 | ||
5624a4e4 VG |
466 | /* |
467 | * This function allocates the per cpu stats for blkio_group. Should be called | |
468 | * from sleepable context as alloc_per_cpu() requires that. | |
469 | */ | |
470 | int blkio_alloc_blkg_stats(struct blkio_group *blkg) | |
471 | { | |
472 | /* Allocate memory for per cpu stats */ | |
473 | blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu); | |
474 | if (!blkg->stats_cpu) | |
475 | return -ENOMEM; | |
476 | return 0; | |
477 | } | |
478 | EXPORT_SYMBOL_GPL(blkio_alloc_blkg_stats); | |
479 | ||
31e4c28d | 480 | void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg, |
ca32aefc | 481 | struct blkio_group *blkg, struct request_queue *q, dev_t dev, |
062a644d | 482 | enum blkio_policy_id plid) |
31e4c28d VG |
483 | { |
484 | unsigned long flags; | |
485 | ||
486 | spin_lock_irqsave(&blkcg->lock, flags); | |
8d2a91f8 | 487 | spin_lock_init(&blkg->stats_lock); |
ca32aefc | 488 | rcu_assign_pointer(blkg->q, q); |
b1c35769 | 489 | blkg->blkcg_id = css_id(&blkcg->css); |
31e4c28d | 490 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
062a644d | 491 | blkg->plid = plid; |
31e4c28d | 492 | spin_unlock_irqrestore(&blkcg->lock, flags); |
2868ef7b VG |
493 | /* Need to take css reference ? */ |
494 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); | |
22084190 | 495 | blkg->dev = dev; |
31e4c28d | 496 | } |
9d6a986c | 497 | EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group); |
31e4c28d | 498 | |
b1c35769 VG |
499 | static void __blkiocg_del_blkio_group(struct blkio_group *blkg) |
500 | { | |
501 | hlist_del_init_rcu(&blkg->blkcg_node); | |
502 | blkg->blkcg_id = 0; | |
503 | } | |
504 | ||
505 | /* | |
506 | * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1 | |
507 | * indicating that blk_group was unhashed by the time we got to it. | |
508 | */ | |
31e4c28d VG |
509 | int blkiocg_del_blkio_group(struct blkio_group *blkg) |
510 | { | |
b1c35769 VG |
511 | struct blkio_cgroup *blkcg; |
512 | unsigned long flags; | |
513 | struct cgroup_subsys_state *css; | |
514 | int ret = 1; | |
515 | ||
516 | rcu_read_lock(); | |
517 | css = css_lookup(&blkio_subsys, blkg->blkcg_id); | |
0f3942a3 JA |
518 | if (css) { |
519 | blkcg = container_of(css, struct blkio_cgroup, css); | |
520 | spin_lock_irqsave(&blkcg->lock, flags); | |
521 | if (!hlist_unhashed(&blkg->blkcg_node)) { | |
522 | __blkiocg_del_blkio_group(blkg); | |
523 | ret = 0; | |
524 | } | |
525 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
b1c35769 | 526 | } |
0f3942a3 | 527 | |
b1c35769 VG |
528 | rcu_read_unlock(); |
529 | return ret; | |
31e4c28d | 530 | } |
9d6a986c | 531 | EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group); |
31e4c28d VG |
532 | |
533 | /* called under rcu_read_lock(). */ | |
ca32aefc TH |
534 | struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, |
535 | struct request_queue *q, | |
536 | enum blkio_policy_id plid) | |
31e4c28d VG |
537 | { |
538 | struct blkio_group *blkg; | |
539 | struct hlist_node *n; | |
31e4c28d | 540 | |
ca32aefc TH |
541 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) |
542 | if (blkg->q == q && blkg->plid == plid) | |
31e4c28d | 543 | return blkg; |
31e4c28d VG |
544 | return NULL; |
545 | } | |
9d6a986c | 546 | EXPORT_SYMBOL_GPL(blkiocg_lookup_group); |
31e4c28d | 547 | |
72e06c25 TH |
548 | void blkg_destroy_all(struct request_queue *q) |
549 | { | |
550 | struct blkio_policy_type *pol; | |
551 | ||
552 | while (true) { | |
553 | bool done = true; | |
554 | ||
555 | spin_lock(&blkio_list_lock); | |
556 | spin_lock_irq(q->queue_lock); | |
557 | ||
558 | /* | |
559 | * clear_queue_fn() might return with non-empty group list | |
560 | * if it raced cgroup removal and lost. cgroup removal is | |
561 | * guaranteed to make forward progress and retrying after a | |
562 | * while is enough. This ugliness is scheduled to be | |
563 | * removed after locking update. | |
564 | */ | |
565 | list_for_each_entry(pol, &blkio_list, list) | |
566 | if (!pol->ops.blkio_clear_queue_fn(q)) | |
567 | done = false; | |
568 | ||
569 | spin_unlock_irq(q->queue_lock); | |
570 | spin_unlock(&blkio_list_lock); | |
571 | ||
572 | if (done) | |
573 | break; | |
574 | ||
575 | msleep(10); /* just some random duration I like */ | |
576 | } | |
577 | } | |
578 | ||
f0bdc8cd VG |
579 | static void blkio_reset_stats_cpu(struct blkio_group *blkg) |
580 | { | |
581 | struct blkio_group_stats_cpu *stats_cpu; | |
582 | int i, j, k; | |
583 | /* | |
584 | * Note: On 64 bit arch this should not be an issue. This has the | |
585 | * possibility of returning some inconsistent value on 32bit arch | |
586 | * as 64bit update on 32bit is non atomic. Taking care of this | |
587 | * corner case makes code very complicated, like sending IPIs to | |
588 | * cpus, taking care of stats of offline cpus etc. | |
589 | * | |
590 | * reset stats is anyway more of a debug feature and this sounds a | |
591 | * corner case. So I am not complicating the code yet until and | |
592 | * unless this becomes a real issue. | |
593 | */ | |
594 | for_each_possible_cpu(i) { | |
595 | stats_cpu = per_cpu_ptr(blkg->stats_cpu, i); | |
596 | stats_cpu->sectors = 0; | |
597 | for(j = 0; j < BLKIO_STAT_CPU_NR; j++) | |
598 | for (k = 0; k < BLKIO_STAT_TOTAL; k++) | |
599 | stats_cpu->stat_arr_cpu[j][k] = 0; | |
600 | } | |
601 | } | |
602 | ||
303a3acb | 603 | static int |
84c124da | 604 | blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val) |
303a3acb DS |
605 | { |
606 | struct blkio_cgroup *blkcg; | |
607 | struct blkio_group *blkg; | |
812df48d | 608 | struct blkio_group_stats *stats; |
303a3acb | 609 | struct hlist_node *n; |
cdc1184c DS |
610 | uint64_t queued[BLKIO_STAT_TOTAL]; |
611 | int i; | |
812df48d DS |
612 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
613 | bool idling, waiting, empty; | |
614 | unsigned long long now = sched_clock(); | |
615 | #endif | |
303a3acb DS |
616 | |
617 | blkcg = cgroup_to_blkio_cgroup(cgroup); | |
618 | spin_lock_irq(&blkcg->lock); | |
619 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
620 | spin_lock(&blkg->stats_lock); | |
812df48d DS |
621 | stats = &blkg->stats; |
622 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
623 | idling = blkio_blkg_idling(stats); | |
624 | waiting = blkio_blkg_waiting(stats); | |
625 | empty = blkio_blkg_empty(stats); | |
626 | #endif | |
cdc1184c | 627 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
812df48d DS |
628 | queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i]; |
629 | memset(stats, 0, sizeof(struct blkio_group_stats)); | |
cdc1184c | 630 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
812df48d DS |
631 | stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i]; |
632 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
633 | if (idling) { | |
634 | blkio_mark_blkg_idling(stats); | |
635 | stats->start_idle_time = now; | |
636 | } | |
637 | if (waiting) { | |
638 | blkio_mark_blkg_waiting(stats); | |
639 | stats->start_group_wait_time = now; | |
640 | } | |
641 | if (empty) { | |
642 | blkio_mark_blkg_empty(stats); | |
643 | stats->start_empty_time = now; | |
644 | } | |
645 | #endif | |
303a3acb | 646 | spin_unlock(&blkg->stats_lock); |
f0bdc8cd VG |
647 | |
648 | /* Reset Per cpu stats which don't take blkg->stats_lock */ | |
649 | blkio_reset_stats_cpu(blkg); | |
303a3acb | 650 | } |
f0bdc8cd | 651 | |
303a3acb DS |
652 | spin_unlock_irq(&blkcg->lock); |
653 | return 0; | |
654 | } | |
655 | ||
84c124da DS |
656 | static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str, |
657 | int chars_left, bool diskname_only) | |
303a3acb | 658 | { |
84c124da | 659 | snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev)); |
303a3acb DS |
660 | chars_left -= strlen(str); |
661 | if (chars_left <= 0) { | |
662 | printk(KERN_WARNING | |
663 | "Possibly incorrect cgroup stat display format"); | |
664 | return; | |
665 | } | |
84c124da DS |
666 | if (diskname_only) |
667 | return; | |
303a3acb | 668 | switch (type) { |
84c124da | 669 | case BLKIO_STAT_READ: |
303a3acb DS |
670 | strlcat(str, " Read", chars_left); |
671 | break; | |
84c124da | 672 | case BLKIO_STAT_WRITE: |
303a3acb DS |
673 | strlcat(str, " Write", chars_left); |
674 | break; | |
84c124da | 675 | case BLKIO_STAT_SYNC: |
303a3acb DS |
676 | strlcat(str, " Sync", chars_left); |
677 | break; | |
84c124da | 678 | case BLKIO_STAT_ASYNC: |
303a3acb DS |
679 | strlcat(str, " Async", chars_left); |
680 | break; | |
84c124da | 681 | case BLKIO_STAT_TOTAL: |
303a3acb DS |
682 | strlcat(str, " Total", chars_left); |
683 | break; | |
684 | default: | |
685 | strlcat(str, " Invalid", chars_left); | |
686 | } | |
687 | } | |
688 | ||
84c124da DS |
689 | static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val, |
690 | struct cgroup_map_cb *cb, dev_t dev) | |
691 | { | |
692 | blkio_get_key_name(0, dev, str, chars_left, true); | |
693 | cb->fill(cb, str, val); | |
694 | return val; | |
695 | } | |
303a3acb | 696 | |
5624a4e4 VG |
697 | |
698 | static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg, | |
699 | enum stat_type_cpu type, enum stat_sub_type sub_type) | |
700 | { | |
701 | int cpu; | |
702 | struct blkio_group_stats_cpu *stats_cpu; | |
575969a0 | 703 | u64 val = 0, tval; |
5624a4e4 VG |
704 | |
705 | for_each_possible_cpu(cpu) { | |
575969a0 | 706 | unsigned int start; |
5624a4e4 VG |
707 | stats_cpu = per_cpu_ptr(blkg->stats_cpu, cpu); |
708 | ||
575969a0 VG |
709 | do { |
710 | start = u64_stats_fetch_begin(&stats_cpu->syncp); | |
711 | if (type == BLKIO_STAT_CPU_SECTORS) | |
712 | tval = stats_cpu->sectors; | |
713 | else | |
714 | tval = stats_cpu->stat_arr_cpu[type][sub_type]; | |
715 | } while(u64_stats_fetch_retry(&stats_cpu->syncp, start)); | |
716 | ||
717 | val += tval; | |
5624a4e4 VG |
718 | } |
719 | ||
720 | return val; | |
721 | } | |
722 | ||
723 | static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg, | |
724 | struct cgroup_map_cb *cb, dev_t dev, enum stat_type_cpu type) | |
725 | { | |
726 | uint64_t disk_total, val; | |
727 | char key_str[MAX_KEY_LEN]; | |
728 | enum stat_sub_type sub_type; | |
729 | ||
730 | if (type == BLKIO_STAT_CPU_SECTORS) { | |
731 | val = blkio_read_stat_cpu(blkg, type, 0); | |
732 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, dev); | |
733 | } | |
734 | ||
735 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; | |
736 | sub_type++) { | |
737 | blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false); | |
738 | val = blkio_read_stat_cpu(blkg, type, sub_type); | |
739 | cb->fill(cb, key_str, val); | |
740 | } | |
741 | ||
742 | disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) + | |
743 | blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE); | |
744 | ||
745 | blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false); | |
746 | cb->fill(cb, key_str, disk_total); | |
747 | return disk_total; | |
748 | } | |
749 | ||
84c124da DS |
750 | /* This should be called with blkg->stats_lock held */ |
751 | static uint64_t blkio_get_stat(struct blkio_group *blkg, | |
752 | struct cgroup_map_cb *cb, dev_t dev, enum stat_type type) | |
303a3acb DS |
753 | { |
754 | uint64_t disk_total; | |
755 | char key_str[MAX_KEY_LEN]; | |
84c124da DS |
756 | enum stat_sub_type sub_type; |
757 | ||
758 | if (type == BLKIO_STAT_TIME) | |
759 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
760 | blkg->stats.time, cb, dev); | |
9026e521 | 761 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 JT |
762 | if (type == BLKIO_STAT_UNACCOUNTED_TIME) |
763 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
764 | blkg->stats.unaccounted_time, cb, dev); | |
cdc1184c DS |
765 | if (type == BLKIO_STAT_AVG_QUEUE_SIZE) { |
766 | uint64_t sum = blkg->stats.avg_queue_size_sum; | |
767 | uint64_t samples = blkg->stats.avg_queue_size_samples; | |
768 | if (samples) | |
769 | do_div(sum, samples); | |
770 | else | |
771 | sum = 0; | |
772 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev); | |
773 | } | |
812df48d DS |
774 | if (type == BLKIO_STAT_GROUP_WAIT_TIME) |
775 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
776 | blkg->stats.group_wait_time, cb, dev); | |
777 | if (type == BLKIO_STAT_IDLE_TIME) | |
778 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
779 | blkg->stats.idle_time, cb, dev); | |
780 | if (type == BLKIO_STAT_EMPTY_TIME) | |
781 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
782 | blkg->stats.empty_time, cb, dev); | |
84c124da DS |
783 | if (type == BLKIO_STAT_DEQUEUE) |
784 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
785 | blkg->stats.dequeue, cb, dev); | |
786 | #endif | |
303a3acb | 787 | |
84c124da DS |
788 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; |
789 | sub_type++) { | |
790 | blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false); | |
791 | cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]); | |
303a3acb | 792 | } |
84c124da DS |
793 | disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] + |
794 | blkg->stats.stat_arr[type][BLKIO_STAT_WRITE]; | |
795 | blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false); | |
303a3acb DS |
796 | cb->fill(cb, key_str, disk_total); |
797 | return disk_total; | |
798 | } | |
799 | ||
34d0f179 | 800 | static int blkio_policy_parse_and_set(char *buf, |
062a644d | 801 | struct blkio_policy_node *newpn, enum blkio_policy_id plid, int fileid) |
34d0f179 | 802 | { |
ece84241 | 803 | struct gendisk *disk = NULL; |
34d0f179 | 804 | char *s[4], *p, *major_s = NULL, *minor_s = NULL; |
d11bb446 | 805 | unsigned long major, minor; |
ece84241 TH |
806 | int i = 0, ret = -EINVAL; |
807 | int part; | |
34d0f179 | 808 | dev_t dev; |
d11bb446 | 809 | u64 temp; |
34d0f179 GJ |
810 | |
811 | memset(s, 0, sizeof(s)); | |
812 | ||
813 | while ((p = strsep(&buf, " ")) != NULL) { | |
814 | if (!*p) | |
815 | continue; | |
816 | ||
817 | s[i++] = p; | |
818 | ||
819 | /* Prevent from inputing too many things */ | |
820 | if (i == 3) | |
821 | break; | |
822 | } | |
823 | ||
824 | if (i != 2) | |
ece84241 | 825 | goto out; |
34d0f179 GJ |
826 | |
827 | p = strsep(&s[0], ":"); | |
828 | if (p != NULL) | |
829 | major_s = p; | |
830 | else | |
ece84241 | 831 | goto out; |
34d0f179 GJ |
832 | |
833 | minor_s = s[0]; | |
834 | if (!minor_s) | |
ece84241 | 835 | goto out; |
34d0f179 | 836 | |
ece84241 TH |
837 | if (strict_strtoul(major_s, 10, &major)) |
838 | goto out; | |
34d0f179 | 839 | |
ece84241 TH |
840 | if (strict_strtoul(minor_s, 10, &minor)) |
841 | goto out; | |
34d0f179 GJ |
842 | |
843 | dev = MKDEV(major, minor); | |
844 | ||
ece84241 TH |
845 | if (strict_strtoull(s[1], 10, &temp)) |
846 | goto out; | |
34d0f179 | 847 | |
d11bb446 WG |
848 | /* For rule removal, do not check for device presence. */ |
849 | if (temp) { | |
ece84241 TH |
850 | disk = get_gendisk(dev, &part); |
851 | if (!disk || part) { | |
852 | ret = -ENODEV; | |
853 | goto out; | |
854 | } | |
d11bb446 | 855 | } |
34d0f179 | 856 | |
d11bb446 | 857 | newpn->dev = dev; |
34d0f179 | 858 | |
062a644d VG |
859 | switch (plid) { |
860 | case BLKIO_POLICY_PROP: | |
d11bb446 WG |
861 | if ((temp < BLKIO_WEIGHT_MIN && temp > 0) || |
862 | temp > BLKIO_WEIGHT_MAX) | |
ece84241 | 863 | goto out; |
34d0f179 | 864 | |
062a644d VG |
865 | newpn->plid = plid; |
866 | newpn->fileid = fileid; | |
4c9eefa1 VG |
867 | newpn->val.weight = temp; |
868 | break; | |
869 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
870 | switch(fileid) { |
871 | case BLKIO_THROTL_read_bps_device: | |
872 | case BLKIO_THROTL_write_bps_device: | |
7702e8f4 VG |
873 | newpn->plid = plid; |
874 | newpn->fileid = fileid; | |
d11bb446 | 875 | newpn->val.bps = temp; |
7702e8f4 VG |
876 | break; |
877 | case BLKIO_THROTL_read_iops_device: | |
878 | case BLKIO_THROTL_write_iops_device: | |
d11bb446 | 879 | if (temp > THROTL_IOPS_MAX) |
ece84241 | 880 | goto out; |
9355aede | 881 | |
7702e8f4 VG |
882 | newpn->plid = plid; |
883 | newpn->fileid = fileid; | |
d11bb446 | 884 | newpn->val.iops = (unsigned int)temp; |
7702e8f4 VG |
885 | break; |
886 | } | |
062a644d VG |
887 | break; |
888 | default: | |
889 | BUG(); | |
890 | } | |
ece84241 TH |
891 | ret = 0; |
892 | out: | |
893 | put_disk(disk); | |
894 | return ret; | |
34d0f179 GJ |
895 | } |
896 | ||
897 | unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg, | |
898 | dev_t dev) | |
899 | { | |
900 | struct blkio_policy_node *pn; | |
a38eb630 VG |
901 | unsigned long flags; |
902 | unsigned int weight; | |
903 | ||
904 | spin_lock_irqsave(&blkcg->lock, flags); | |
34d0f179 | 905 | |
062a644d VG |
906 | pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_PROP, |
907 | BLKIO_PROP_weight_device); | |
34d0f179 | 908 | if (pn) |
a38eb630 | 909 | weight = pn->val.weight; |
34d0f179 | 910 | else |
a38eb630 VG |
911 | weight = blkcg->weight; |
912 | ||
913 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
914 | ||
915 | return weight; | |
34d0f179 GJ |
916 | } |
917 | EXPORT_SYMBOL_GPL(blkcg_get_weight); | |
918 | ||
4c9eefa1 VG |
919 | uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, dev_t dev) |
920 | { | |
921 | struct blkio_policy_node *pn; | |
a38eb630 VG |
922 | unsigned long flags; |
923 | uint64_t bps = -1; | |
4c9eefa1 | 924 | |
a38eb630 | 925 | spin_lock_irqsave(&blkcg->lock, flags); |
4c9eefa1 VG |
926 | pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL, |
927 | BLKIO_THROTL_read_bps_device); | |
928 | if (pn) | |
a38eb630 VG |
929 | bps = pn->val.bps; |
930 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
931 | ||
932 | return bps; | |
4c9eefa1 VG |
933 | } |
934 | ||
935 | uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, dev_t dev) | |
936 | { | |
937 | struct blkio_policy_node *pn; | |
a38eb630 VG |
938 | unsigned long flags; |
939 | uint64_t bps = -1; | |
940 | ||
941 | spin_lock_irqsave(&blkcg->lock, flags); | |
4c9eefa1 VG |
942 | pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL, |
943 | BLKIO_THROTL_write_bps_device); | |
944 | if (pn) | |
a38eb630 VG |
945 | bps = pn->val.bps; |
946 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
947 | ||
948 | return bps; | |
4c9eefa1 VG |
949 | } |
950 | ||
7702e8f4 VG |
951 | unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, dev_t dev) |
952 | { | |
953 | struct blkio_policy_node *pn; | |
a38eb630 VG |
954 | unsigned long flags; |
955 | unsigned int iops = -1; | |
7702e8f4 | 956 | |
a38eb630 | 957 | spin_lock_irqsave(&blkcg->lock, flags); |
7702e8f4 VG |
958 | pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL, |
959 | BLKIO_THROTL_read_iops_device); | |
960 | if (pn) | |
a38eb630 VG |
961 | iops = pn->val.iops; |
962 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
963 | ||
964 | return iops; | |
7702e8f4 VG |
965 | } |
966 | ||
967 | unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, dev_t dev) | |
968 | { | |
969 | struct blkio_policy_node *pn; | |
a38eb630 VG |
970 | unsigned long flags; |
971 | unsigned int iops = -1; | |
972 | ||
973 | spin_lock_irqsave(&blkcg->lock, flags); | |
7702e8f4 VG |
974 | pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL, |
975 | BLKIO_THROTL_write_iops_device); | |
976 | if (pn) | |
a38eb630 VG |
977 | iops = pn->val.iops; |
978 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
979 | ||
980 | return iops; | |
7702e8f4 VG |
981 | } |
982 | ||
062a644d VG |
983 | /* Checks whether user asked for deleting a policy rule */ |
984 | static bool blkio_delete_rule_command(struct blkio_policy_node *pn) | |
985 | { | |
986 | switch(pn->plid) { | |
987 | case BLKIO_POLICY_PROP: | |
4c9eefa1 VG |
988 | if (pn->val.weight == 0) |
989 | return 1; | |
990 | break; | |
991 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
992 | switch(pn->fileid) { |
993 | case BLKIO_THROTL_read_bps_device: | |
994 | case BLKIO_THROTL_write_bps_device: | |
995 | if (pn->val.bps == 0) | |
996 | return 1; | |
997 | break; | |
998 | case BLKIO_THROTL_read_iops_device: | |
999 | case BLKIO_THROTL_write_iops_device: | |
1000 | if (pn->val.iops == 0) | |
1001 | return 1; | |
1002 | } | |
062a644d VG |
1003 | break; |
1004 | default: | |
1005 | BUG(); | |
1006 | } | |
1007 | ||
1008 | return 0; | |
1009 | } | |
1010 | ||
1011 | static void blkio_update_policy_rule(struct blkio_policy_node *oldpn, | |
1012 | struct blkio_policy_node *newpn) | |
1013 | { | |
1014 | switch(oldpn->plid) { | |
1015 | case BLKIO_POLICY_PROP: | |
4c9eefa1 VG |
1016 | oldpn->val.weight = newpn->val.weight; |
1017 | break; | |
1018 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
1019 | switch(newpn->fileid) { |
1020 | case BLKIO_THROTL_read_bps_device: | |
1021 | case BLKIO_THROTL_write_bps_device: | |
1022 | oldpn->val.bps = newpn->val.bps; | |
1023 | break; | |
1024 | case BLKIO_THROTL_read_iops_device: | |
1025 | case BLKIO_THROTL_write_iops_device: | |
1026 | oldpn->val.iops = newpn->val.iops; | |
1027 | } | |
062a644d VG |
1028 | break; |
1029 | default: | |
1030 | BUG(); | |
1031 | } | |
1032 | } | |
1033 | ||
1034 | /* | |
25985edc | 1035 | * Some rules/values in blkg have changed. Propagate those to respective |
062a644d VG |
1036 | * policies. |
1037 | */ | |
1038 | static void blkio_update_blkg_policy(struct blkio_cgroup *blkcg, | |
1039 | struct blkio_group *blkg, struct blkio_policy_node *pn) | |
1040 | { | |
7702e8f4 | 1041 | unsigned int weight, iops; |
4c9eefa1 | 1042 | u64 bps; |
062a644d VG |
1043 | |
1044 | switch(pn->plid) { | |
1045 | case BLKIO_POLICY_PROP: | |
4c9eefa1 | 1046 | weight = pn->val.weight ? pn->val.weight : |
062a644d VG |
1047 | blkcg->weight; |
1048 | blkio_update_group_weight(blkg, weight); | |
1049 | break; | |
4c9eefa1 VG |
1050 | case BLKIO_POLICY_THROTL: |
1051 | switch(pn->fileid) { | |
1052 | case BLKIO_THROTL_read_bps_device: | |
1053 | case BLKIO_THROTL_write_bps_device: | |
1054 | bps = pn->val.bps ? pn->val.bps : (-1); | |
1055 | blkio_update_group_bps(blkg, bps, pn->fileid); | |
1056 | break; | |
7702e8f4 VG |
1057 | case BLKIO_THROTL_read_iops_device: |
1058 | case BLKIO_THROTL_write_iops_device: | |
1059 | iops = pn->val.iops ? pn->val.iops : (-1); | |
1060 | blkio_update_group_iops(blkg, iops, pn->fileid); | |
1061 | break; | |
4c9eefa1 VG |
1062 | } |
1063 | break; | |
062a644d VG |
1064 | default: |
1065 | BUG(); | |
1066 | } | |
1067 | } | |
1068 | ||
1069 | /* | |
25985edc | 1070 | * A policy node rule has been updated. Propagate this update to all the |
062a644d VG |
1071 | * block groups which might be affected by this update. |
1072 | */ | |
1073 | static void blkio_update_policy_node_blkg(struct blkio_cgroup *blkcg, | |
1074 | struct blkio_policy_node *pn) | |
1075 | { | |
1076 | struct blkio_group *blkg; | |
1077 | struct hlist_node *n; | |
1078 | ||
1079 | spin_lock(&blkio_list_lock); | |
1080 | spin_lock_irq(&blkcg->lock); | |
34d0f179 | 1081 | |
062a644d VG |
1082 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
1083 | if (pn->dev != blkg->dev || pn->plid != blkg->plid) | |
1084 | continue; | |
1085 | blkio_update_blkg_policy(blkcg, blkg, pn); | |
1086 | } | |
1087 | ||
1088 | spin_unlock_irq(&blkcg->lock); | |
1089 | spin_unlock(&blkio_list_lock); | |
1090 | } | |
34d0f179 | 1091 | |
062a644d VG |
1092 | static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft, |
1093 | const char *buffer) | |
34d0f179 GJ |
1094 | { |
1095 | int ret = 0; | |
1096 | char *buf; | |
1097 | struct blkio_policy_node *newpn, *pn; | |
1098 | struct blkio_cgroup *blkcg; | |
34d0f179 | 1099 | int keep_newpn = 0; |
062a644d VG |
1100 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); |
1101 | int fileid = BLKIOFILE_ATTR(cft->private); | |
34d0f179 GJ |
1102 | |
1103 | buf = kstrdup(buffer, GFP_KERNEL); | |
1104 | if (!buf) | |
1105 | return -ENOMEM; | |
1106 | ||
1107 | newpn = kzalloc(sizeof(*newpn), GFP_KERNEL); | |
1108 | if (!newpn) { | |
1109 | ret = -ENOMEM; | |
1110 | goto free_buf; | |
1111 | } | |
1112 | ||
062a644d | 1113 | ret = blkio_policy_parse_and_set(buf, newpn, plid, fileid); |
34d0f179 GJ |
1114 | if (ret) |
1115 | goto free_newpn; | |
1116 | ||
1117 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1118 | ||
1119 | spin_lock_irq(&blkcg->lock); | |
1120 | ||
062a644d | 1121 | pn = blkio_policy_search_node(blkcg, newpn->dev, plid, fileid); |
34d0f179 | 1122 | if (!pn) { |
062a644d | 1123 | if (!blkio_delete_rule_command(newpn)) { |
34d0f179 GJ |
1124 | blkio_policy_insert_node(blkcg, newpn); |
1125 | keep_newpn = 1; | |
1126 | } | |
1127 | spin_unlock_irq(&blkcg->lock); | |
1128 | goto update_io_group; | |
1129 | } | |
1130 | ||
062a644d | 1131 | if (blkio_delete_rule_command(newpn)) { |
34d0f179 | 1132 | blkio_policy_delete_node(pn); |
e060f00b | 1133 | kfree(pn); |
34d0f179 GJ |
1134 | spin_unlock_irq(&blkcg->lock); |
1135 | goto update_io_group; | |
1136 | } | |
1137 | spin_unlock_irq(&blkcg->lock); | |
1138 | ||
062a644d | 1139 | blkio_update_policy_rule(pn, newpn); |
34d0f179 GJ |
1140 | |
1141 | update_io_group: | |
062a644d | 1142 | blkio_update_policy_node_blkg(blkcg, newpn); |
34d0f179 GJ |
1143 | |
1144 | free_newpn: | |
1145 | if (!keep_newpn) | |
1146 | kfree(newpn); | |
1147 | free_buf: | |
1148 | kfree(buf); | |
1149 | return ret; | |
1150 | } | |
1151 | ||
062a644d VG |
1152 | static void |
1153 | blkio_print_policy_node(struct seq_file *m, struct blkio_policy_node *pn) | |
34d0f179 | 1154 | { |
062a644d VG |
1155 | switch(pn->plid) { |
1156 | case BLKIO_POLICY_PROP: | |
1157 | if (pn->fileid == BLKIO_PROP_weight_device) | |
1158 | seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev), | |
4c9eefa1 VG |
1159 | MINOR(pn->dev), pn->val.weight); |
1160 | break; | |
1161 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
1162 | switch(pn->fileid) { |
1163 | case BLKIO_THROTL_read_bps_device: | |
1164 | case BLKIO_THROTL_write_bps_device: | |
4c9eefa1 VG |
1165 | seq_printf(m, "%u:%u\t%llu\n", MAJOR(pn->dev), |
1166 | MINOR(pn->dev), pn->val.bps); | |
7702e8f4 VG |
1167 | break; |
1168 | case BLKIO_THROTL_read_iops_device: | |
1169 | case BLKIO_THROTL_write_iops_device: | |
1170 | seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev), | |
1171 | MINOR(pn->dev), pn->val.iops); | |
1172 | break; | |
1173 | } | |
062a644d VG |
1174 | break; |
1175 | default: | |
1176 | BUG(); | |
1177 | } | |
1178 | } | |
34d0f179 | 1179 | |
062a644d VG |
1180 | /* cgroup files which read their data from policy nodes end up here */ |
1181 | static void blkio_read_policy_node_files(struct cftype *cft, | |
1182 | struct blkio_cgroup *blkcg, struct seq_file *m) | |
34d0f179 | 1183 | { |
34d0f179 | 1184 | struct blkio_policy_node *pn; |
34d0f179 | 1185 | |
0f3942a3 JA |
1186 | if (!list_empty(&blkcg->policy_list)) { |
1187 | spin_lock_irq(&blkcg->lock); | |
1188 | list_for_each_entry(pn, &blkcg->policy_list, node) { | |
062a644d VG |
1189 | if (!pn_matches_cftype(cft, pn)) |
1190 | continue; | |
1191 | blkio_print_policy_node(m, pn); | |
0f3942a3 JA |
1192 | } |
1193 | spin_unlock_irq(&blkcg->lock); | |
34d0f179 | 1194 | } |
062a644d VG |
1195 | } |
1196 | ||
1197 | static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft, | |
1198 | struct seq_file *m) | |
1199 | { | |
1200 | struct blkio_cgroup *blkcg; | |
1201 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1202 | int name = BLKIOFILE_ATTR(cft->private); | |
1203 | ||
1204 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1205 | ||
1206 | switch(plid) { | |
1207 | case BLKIO_POLICY_PROP: | |
1208 | switch(name) { | |
1209 | case BLKIO_PROP_weight_device: | |
1210 | blkio_read_policy_node_files(cft, blkcg, m); | |
1211 | return 0; | |
1212 | default: | |
1213 | BUG(); | |
1214 | } | |
1215 | break; | |
4c9eefa1 VG |
1216 | case BLKIO_POLICY_THROTL: |
1217 | switch(name){ | |
1218 | case BLKIO_THROTL_read_bps_device: | |
1219 | case BLKIO_THROTL_write_bps_device: | |
7702e8f4 VG |
1220 | case BLKIO_THROTL_read_iops_device: |
1221 | case BLKIO_THROTL_write_iops_device: | |
4c9eefa1 VG |
1222 | blkio_read_policy_node_files(cft, blkcg, m); |
1223 | return 0; | |
1224 | default: | |
1225 | BUG(); | |
1226 | } | |
1227 | break; | |
062a644d VG |
1228 | default: |
1229 | BUG(); | |
1230 | } | |
1231 | ||
1232 | return 0; | |
1233 | } | |
1234 | ||
1235 | static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg, | |
5624a4e4 VG |
1236 | struct cftype *cft, struct cgroup_map_cb *cb, |
1237 | enum stat_type type, bool show_total, bool pcpu) | |
062a644d VG |
1238 | { |
1239 | struct blkio_group *blkg; | |
1240 | struct hlist_node *n; | |
1241 | uint64_t cgroup_total = 0; | |
1242 | ||
1243 | rcu_read_lock(); | |
1244 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
1245 | if (blkg->dev) { | |
1246 | if (!cftype_blkg_same_policy(cft, blkg)) | |
1247 | continue; | |
5624a4e4 VG |
1248 | if (pcpu) |
1249 | cgroup_total += blkio_get_stat_cpu(blkg, cb, | |
1250 | blkg->dev, type); | |
1251 | else { | |
1252 | spin_lock_irq(&blkg->stats_lock); | |
1253 | cgroup_total += blkio_get_stat(blkg, cb, | |
1254 | blkg->dev, type); | |
1255 | spin_unlock_irq(&blkg->stats_lock); | |
1256 | } | |
062a644d VG |
1257 | } |
1258 | } | |
1259 | if (show_total) | |
1260 | cb->fill(cb, "Total", cgroup_total); | |
1261 | rcu_read_unlock(); | |
1262 | return 0; | |
1263 | } | |
1264 | ||
1265 | /* All map kind of cgroup file get serviced by this function */ | |
1266 | static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft, | |
1267 | struct cgroup_map_cb *cb) | |
1268 | { | |
1269 | struct blkio_cgroup *blkcg; | |
1270 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1271 | int name = BLKIOFILE_ATTR(cft->private); | |
1272 | ||
1273 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1274 | ||
1275 | switch(plid) { | |
1276 | case BLKIO_POLICY_PROP: | |
1277 | switch(name) { | |
1278 | case BLKIO_PROP_time: | |
1279 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1280 | BLKIO_STAT_TIME, 0, 0); |
062a644d VG |
1281 | case BLKIO_PROP_sectors: |
1282 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1283 | BLKIO_STAT_CPU_SECTORS, 0, 1); |
062a644d VG |
1284 | case BLKIO_PROP_io_service_bytes: |
1285 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1286 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
062a644d VG |
1287 | case BLKIO_PROP_io_serviced: |
1288 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1289 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
062a644d VG |
1290 | case BLKIO_PROP_io_service_time: |
1291 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1292 | BLKIO_STAT_SERVICE_TIME, 1, 0); |
062a644d VG |
1293 | case BLKIO_PROP_io_wait_time: |
1294 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1295 | BLKIO_STAT_WAIT_TIME, 1, 0); |
062a644d VG |
1296 | case BLKIO_PROP_io_merged: |
1297 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
317389a7 | 1298 | BLKIO_STAT_CPU_MERGED, 1, 1); |
062a644d VG |
1299 | case BLKIO_PROP_io_queued: |
1300 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1301 | BLKIO_STAT_QUEUED, 1, 0); |
062a644d | 1302 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
9026e521 JT |
1303 | case BLKIO_PROP_unaccounted_time: |
1304 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1305 | BLKIO_STAT_UNACCOUNTED_TIME, 0, 0); |
062a644d VG |
1306 | case BLKIO_PROP_dequeue: |
1307 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1308 | BLKIO_STAT_DEQUEUE, 0, 0); |
062a644d VG |
1309 | case BLKIO_PROP_avg_queue_size: |
1310 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1311 | BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0); |
062a644d VG |
1312 | case BLKIO_PROP_group_wait_time: |
1313 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1314 | BLKIO_STAT_GROUP_WAIT_TIME, 0, 0); |
062a644d VG |
1315 | case BLKIO_PROP_idle_time: |
1316 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1317 | BLKIO_STAT_IDLE_TIME, 0, 0); |
062a644d VG |
1318 | case BLKIO_PROP_empty_time: |
1319 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1320 | BLKIO_STAT_EMPTY_TIME, 0, 0); |
062a644d VG |
1321 | #endif |
1322 | default: | |
1323 | BUG(); | |
1324 | } | |
1325 | break; | |
4c9eefa1 VG |
1326 | case BLKIO_POLICY_THROTL: |
1327 | switch(name){ | |
1328 | case BLKIO_THROTL_io_service_bytes: | |
1329 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1330 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
4c9eefa1 VG |
1331 | case BLKIO_THROTL_io_serviced: |
1332 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1333 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
4c9eefa1 VG |
1334 | default: |
1335 | BUG(); | |
1336 | } | |
1337 | break; | |
062a644d VG |
1338 | default: |
1339 | BUG(); | |
1340 | } | |
1341 | ||
1342 | return 0; | |
1343 | } | |
1344 | ||
1345 | static int blkio_weight_write(struct blkio_cgroup *blkcg, u64 val) | |
1346 | { | |
1347 | struct blkio_group *blkg; | |
1348 | struct hlist_node *n; | |
1349 | struct blkio_policy_node *pn; | |
1350 | ||
1351 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) | |
1352 | return -EINVAL; | |
1353 | ||
1354 | spin_lock(&blkio_list_lock); | |
1355 | spin_lock_irq(&blkcg->lock); | |
1356 | blkcg->weight = (unsigned int)val; | |
1357 | ||
1358 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
1359 | pn = blkio_policy_search_node(blkcg, blkg->dev, | |
1360 | BLKIO_POLICY_PROP, BLKIO_PROP_weight_device); | |
1361 | if (pn) | |
1362 | continue; | |
1363 | ||
1364 | blkio_update_group_weight(blkg, blkcg->weight); | |
1365 | } | |
1366 | spin_unlock_irq(&blkcg->lock); | |
1367 | spin_unlock(&blkio_list_lock); | |
1368 | return 0; | |
1369 | } | |
1370 | ||
1371 | static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) { | |
1372 | struct blkio_cgroup *blkcg; | |
1373 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1374 | int name = BLKIOFILE_ATTR(cft->private); | |
1375 | ||
1376 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1377 | ||
1378 | switch(plid) { | |
1379 | case BLKIO_POLICY_PROP: | |
1380 | switch(name) { | |
1381 | case BLKIO_PROP_weight: | |
1382 | return (u64)blkcg->weight; | |
1383 | } | |
1384 | break; | |
1385 | default: | |
1386 | BUG(); | |
1387 | } | |
1388 | return 0; | |
1389 | } | |
1390 | ||
1391 | static int | |
1392 | blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) | |
1393 | { | |
1394 | struct blkio_cgroup *blkcg; | |
1395 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1396 | int name = BLKIOFILE_ATTR(cft->private); | |
1397 | ||
1398 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1399 | ||
1400 | switch(plid) { | |
1401 | case BLKIO_POLICY_PROP: | |
1402 | switch(name) { | |
1403 | case BLKIO_PROP_weight: | |
1404 | return blkio_weight_write(blkcg, val); | |
1405 | } | |
1406 | break; | |
1407 | default: | |
1408 | BUG(); | |
1409 | } | |
34d0f179 | 1410 | |
34d0f179 GJ |
1411 | return 0; |
1412 | } | |
1413 | ||
31e4c28d | 1414 | struct cftype blkio_files[] = { |
34d0f179 GJ |
1415 | { |
1416 | .name = "weight_device", | |
062a644d VG |
1417 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1418 | BLKIO_PROP_weight_device), | |
1419 | .read_seq_string = blkiocg_file_read, | |
1420 | .write_string = blkiocg_file_write, | |
34d0f179 GJ |
1421 | .max_write_len = 256, |
1422 | }, | |
31e4c28d VG |
1423 | { |
1424 | .name = "weight", | |
062a644d VG |
1425 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1426 | BLKIO_PROP_weight), | |
1427 | .read_u64 = blkiocg_file_read_u64, | |
1428 | .write_u64 = blkiocg_file_write_u64, | |
31e4c28d | 1429 | }, |
22084190 VG |
1430 | { |
1431 | .name = "time", | |
062a644d VG |
1432 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1433 | BLKIO_PROP_time), | |
1434 | .read_map = blkiocg_file_read_map, | |
22084190 VG |
1435 | }, |
1436 | { | |
1437 | .name = "sectors", | |
062a644d VG |
1438 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1439 | BLKIO_PROP_sectors), | |
1440 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1441 | }, |
1442 | { | |
1443 | .name = "io_service_bytes", | |
062a644d VG |
1444 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1445 | BLKIO_PROP_io_service_bytes), | |
1446 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1447 | }, |
1448 | { | |
1449 | .name = "io_serviced", | |
062a644d VG |
1450 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1451 | BLKIO_PROP_io_serviced), | |
1452 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1453 | }, |
1454 | { | |
1455 | .name = "io_service_time", | |
062a644d VG |
1456 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1457 | BLKIO_PROP_io_service_time), | |
1458 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1459 | }, |
1460 | { | |
1461 | .name = "io_wait_time", | |
062a644d VG |
1462 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1463 | BLKIO_PROP_io_wait_time), | |
1464 | .read_map = blkiocg_file_read_map, | |
84c124da | 1465 | }, |
812d4026 DS |
1466 | { |
1467 | .name = "io_merged", | |
062a644d VG |
1468 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1469 | BLKIO_PROP_io_merged), | |
1470 | .read_map = blkiocg_file_read_map, | |
812d4026 | 1471 | }, |
cdc1184c DS |
1472 | { |
1473 | .name = "io_queued", | |
062a644d VG |
1474 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1475 | BLKIO_PROP_io_queued), | |
1476 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1477 | }, |
84c124da DS |
1478 | { |
1479 | .name = "reset_stats", | |
1480 | .write_u64 = blkiocg_reset_stats, | |
22084190 | 1481 | }, |
13f98250 VG |
1482 | #ifdef CONFIG_BLK_DEV_THROTTLING |
1483 | { | |
1484 | .name = "throttle.read_bps_device", | |
1485 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1486 | BLKIO_THROTL_read_bps_device), | |
1487 | .read_seq_string = blkiocg_file_read, | |
1488 | .write_string = blkiocg_file_write, | |
1489 | .max_write_len = 256, | |
1490 | }, | |
1491 | ||
1492 | { | |
1493 | .name = "throttle.write_bps_device", | |
1494 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1495 | BLKIO_THROTL_write_bps_device), | |
1496 | .read_seq_string = blkiocg_file_read, | |
1497 | .write_string = blkiocg_file_write, | |
1498 | .max_write_len = 256, | |
1499 | }, | |
1500 | ||
1501 | { | |
1502 | .name = "throttle.read_iops_device", | |
1503 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1504 | BLKIO_THROTL_read_iops_device), | |
1505 | .read_seq_string = blkiocg_file_read, | |
1506 | .write_string = blkiocg_file_write, | |
1507 | .max_write_len = 256, | |
1508 | }, | |
1509 | ||
1510 | { | |
1511 | .name = "throttle.write_iops_device", | |
1512 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1513 | BLKIO_THROTL_write_iops_device), | |
1514 | .read_seq_string = blkiocg_file_read, | |
1515 | .write_string = blkiocg_file_write, | |
1516 | .max_write_len = 256, | |
1517 | }, | |
1518 | { | |
1519 | .name = "throttle.io_service_bytes", | |
1520 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1521 | BLKIO_THROTL_io_service_bytes), | |
1522 | .read_map = blkiocg_file_read_map, | |
1523 | }, | |
1524 | { | |
1525 | .name = "throttle.io_serviced", | |
1526 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1527 | BLKIO_THROTL_io_serviced), | |
1528 | .read_map = blkiocg_file_read_map, | |
1529 | }, | |
1530 | #endif /* CONFIG_BLK_DEV_THROTTLING */ | |
1531 | ||
22084190 | 1532 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
cdc1184c DS |
1533 | { |
1534 | .name = "avg_queue_size", | |
062a644d VG |
1535 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1536 | BLKIO_PROP_avg_queue_size), | |
1537 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1538 | }, |
812df48d DS |
1539 | { |
1540 | .name = "group_wait_time", | |
062a644d VG |
1541 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1542 | BLKIO_PROP_group_wait_time), | |
1543 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1544 | }, |
1545 | { | |
1546 | .name = "idle_time", | |
062a644d VG |
1547 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1548 | BLKIO_PROP_idle_time), | |
1549 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1550 | }, |
1551 | { | |
1552 | .name = "empty_time", | |
062a644d VG |
1553 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1554 | BLKIO_PROP_empty_time), | |
1555 | .read_map = blkiocg_file_read_map, | |
812df48d | 1556 | }, |
cdc1184c | 1557 | { |
22084190 | 1558 | .name = "dequeue", |
062a644d VG |
1559 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1560 | BLKIO_PROP_dequeue), | |
1561 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1562 | }, |
9026e521 JT |
1563 | { |
1564 | .name = "unaccounted_time", | |
1565 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, | |
1566 | BLKIO_PROP_unaccounted_time), | |
1567 | .read_map = blkiocg_file_read_map, | |
1568 | }, | |
22084190 | 1569 | #endif |
31e4c28d VG |
1570 | }; |
1571 | ||
1572 | static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1573 | { | |
1574 | return cgroup_add_files(cgroup, subsys, blkio_files, | |
1575 | ARRAY_SIZE(blkio_files)); | |
1576 | } | |
1577 | ||
1578 | static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1579 | { | |
1580 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
b1c35769 VG |
1581 | unsigned long flags; |
1582 | struct blkio_group *blkg; | |
ca32aefc | 1583 | struct request_queue *q; |
3e252066 | 1584 | struct blkio_policy_type *blkiop; |
34d0f179 | 1585 | struct blkio_policy_node *pn, *pntmp; |
b1c35769 VG |
1586 | |
1587 | rcu_read_lock(); | |
0f3942a3 JA |
1588 | do { |
1589 | spin_lock_irqsave(&blkcg->lock, flags); | |
b1c35769 | 1590 | |
0f3942a3 JA |
1591 | if (hlist_empty(&blkcg->blkg_list)) { |
1592 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
1593 | break; | |
1594 | } | |
b1c35769 | 1595 | |
0f3942a3 JA |
1596 | blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group, |
1597 | blkcg_node); | |
ca32aefc | 1598 | q = rcu_dereference(blkg->q); |
0f3942a3 | 1599 | __blkiocg_del_blkio_group(blkg); |
31e4c28d | 1600 | |
0f3942a3 | 1601 | spin_unlock_irqrestore(&blkcg->lock, flags); |
b1c35769 | 1602 | |
0f3942a3 JA |
1603 | /* |
1604 | * This blkio_group is being unlinked as associated cgroup is | |
1605 | * going away. Let all the IO controlling policies know about | |
61014e96 | 1606 | * this event. |
0f3942a3 JA |
1607 | */ |
1608 | spin_lock(&blkio_list_lock); | |
61014e96 VG |
1609 | list_for_each_entry(blkiop, &blkio_list, list) { |
1610 | if (blkiop->plid != blkg->plid) | |
1611 | continue; | |
ca32aefc | 1612 | blkiop->ops.blkio_unlink_group_fn(q, blkg); |
61014e96 | 1613 | } |
0f3942a3 JA |
1614 | spin_unlock(&blkio_list_lock); |
1615 | } while (1); | |
34d0f179 | 1616 | |
34d0f179 GJ |
1617 | list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) { |
1618 | blkio_policy_delete_node(pn); | |
1619 | kfree(pn); | |
1620 | } | |
0f3942a3 | 1621 | |
31e4c28d | 1622 | free_css_id(&blkio_subsys, &blkcg->css); |
b1c35769 | 1623 | rcu_read_unlock(); |
67523c48 BB |
1624 | if (blkcg != &blkio_root_cgroup) |
1625 | kfree(blkcg); | |
31e4c28d VG |
1626 | } |
1627 | ||
1628 | static struct cgroup_subsys_state * | |
1629 | blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1630 | { | |
0341509f LZ |
1631 | struct blkio_cgroup *blkcg; |
1632 | struct cgroup *parent = cgroup->parent; | |
31e4c28d | 1633 | |
0341509f | 1634 | if (!parent) { |
31e4c28d VG |
1635 | blkcg = &blkio_root_cgroup; |
1636 | goto done; | |
1637 | } | |
1638 | ||
31e4c28d VG |
1639 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
1640 | if (!blkcg) | |
1641 | return ERR_PTR(-ENOMEM); | |
1642 | ||
1643 | blkcg->weight = BLKIO_WEIGHT_DEFAULT; | |
1644 | done: | |
1645 | spin_lock_init(&blkcg->lock); | |
1646 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
1647 | ||
34d0f179 | 1648 | INIT_LIST_HEAD(&blkcg->policy_list); |
31e4c28d VG |
1649 | return &blkcg->css; |
1650 | } | |
1651 | ||
1652 | /* | |
1653 | * We cannot support shared io contexts, as we have no mean to support | |
1654 | * two tasks with the same ioc in two different groups without major rework | |
1655 | * of the main cic data structures. For now we allow a task to change | |
1656 | * its cgroup only if it's the only owner of its ioc. | |
1657 | */ | |
bb9d97b6 TH |
1658 | static int blkiocg_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
1659 | struct cgroup_taskset *tset) | |
31e4c28d | 1660 | { |
bb9d97b6 | 1661 | struct task_struct *task; |
31e4c28d VG |
1662 | struct io_context *ioc; |
1663 | int ret = 0; | |
1664 | ||
1665 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
bb9d97b6 TH |
1666 | cgroup_taskset_for_each(task, cgrp, tset) { |
1667 | task_lock(task); | |
1668 | ioc = task->io_context; | |
1669 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1670 | ret = -EINVAL; | |
1671 | task_unlock(task); | |
1672 | if (ret) | |
1673 | break; | |
1674 | } | |
31e4c28d VG |
1675 | return ret; |
1676 | } | |
1677 | ||
bb9d97b6 TH |
1678 | static void blkiocg_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
1679 | struct cgroup_taskset *tset) | |
31e4c28d | 1680 | { |
bb9d97b6 | 1681 | struct task_struct *task; |
31e4c28d VG |
1682 | struct io_context *ioc; |
1683 | ||
bb9d97b6 | 1684 | cgroup_taskset_for_each(task, cgrp, tset) { |
b3c9dd18 LT |
1685 | /* we don't lose anything even if ioc allocation fails */ |
1686 | ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE); | |
1687 | if (ioc) { | |
1688 | ioc_cgroup_changed(ioc); | |
11a3122f | 1689 | put_io_context(ioc); |
b3c9dd18 | 1690 | } |
bb9d97b6 | 1691 | } |
31e4c28d VG |
1692 | } |
1693 | ||
3e252066 VG |
1694 | void blkio_policy_register(struct blkio_policy_type *blkiop) |
1695 | { | |
1696 | spin_lock(&blkio_list_lock); | |
1697 | list_add_tail(&blkiop->list, &blkio_list); | |
1698 | spin_unlock(&blkio_list_lock); | |
1699 | } | |
1700 | EXPORT_SYMBOL_GPL(blkio_policy_register); | |
1701 | ||
1702 | void blkio_policy_unregister(struct blkio_policy_type *blkiop) | |
1703 | { | |
1704 | spin_lock(&blkio_list_lock); | |
1705 | list_del_init(&blkiop->list); | |
1706 | spin_unlock(&blkio_list_lock); | |
1707 | } | |
1708 | EXPORT_SYMBOL_GPL(blkio_policy_unregister); |