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 | 21 | #include <linux/delay.h> |
9a9e8a26 | 22 | #include <linux/atomic.h> |
72e06c25 | 23 | #include "blk-cgroup.h" |
5efd6113 | 24 | #include "blk.h" |
3e252066 | 25 | |
84c124da DS |
26 | #define MAX_KEY_LEN 100 |
27 | ||
3e252066 VG |
28 | static DEFINE_SPINLOCK(blkio_list_lock); |
29 | static LIST_HEAD(blkio_list); | |
b1c35769 | 30 | |
923adde1 TH |
31 | static DEFINE_MUTEX(all_q_mutex); |
32 | static LIST_HEAD(all_q_list); | |
33 | ||
1cd9e039 VG |
34 | /* List of groups pending per cpu stats allocation */ |
35 | static DEFINE_SPINLOCK(alloc_list_lock); | |
36 | static LIST_HEAD(alloc_list); | |
37 | ||
38 | static void blkio_stat_alloc_fn(struct work_struct *); | |
39 | static DECLARE_DELAYED_WORK(blkio_stat_alloc_work, blkio_stat_alloc_fn); | |
40 | ||
31e4c28d | 41 | struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT }; |
9d6a986c VG |
42 | EXPORT_SYMBOL_GPL(blkio_root_cgroup); |
43 | ||
035d10b2 TH |
44 | static struct blkio_policy_type *blkio_policy[BLKIO_NR_POLICIES]; |
45 | ||
67523c48 BB |
46 | static struct cgroup_subsys_state *blkiocg_create(struct cgroup_subsys *, |
47 | struct cgroup *); | |
bb9d97b6 TH |
48 | static int blkiocg_can_attach(struct cgroup_subsys *, struct cgroup *, |
49 | struct cgroup_taskset *); | |
7ee9c562 | 50 | static int blkiocg_pre_destroy(struct cgroup_subsys *, struct cgroup *); |
67523c48 BB |
51 | static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *); |
52 | static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *); | |
53 | ||
062a644d VG |
54 | /* for encoding cft->private value on file */ |
55 | #define BLKIOFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
56 | /* What policy owns the file, proportional or throttle */ | |
57 | #define BLKIOFILE_POLICY(val) (((val) >> 16) & 0xffff) | |
58 | #define BLKIOFILE_ATTR(val) ((val) & 0xffff) | |
59 | ||
67523c48 BB |
60 | struct cgroup_subsys blkio_subsys = { |
61 | .name = "blkio", | |
62 | .create = blkiocg_create, | |
bb9d97b6 | 63 | .can_attach = blkiocg_can_attach, |
7ee9c562 | 64 | .pre_destroy = blkiocg_pre_destroy, |
67523c48 BB |
65 | .destroy = blkiocg_destroy, |
66 | .populate = blkiocg_populate, | |
67523c48 | 67 | .subsys_id = blkio_subsys_id, |
67523c48 BB |
68 | .module = THIS_MODULE, |
69 | }; | |
70 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
71 | ||
31e4c28d VG |
72 | struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup) |
73 | { | |
74 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), | |
75 | struct blkio_cgroup, css); | |
76 | } | |
9d6a986c | 77 | EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup); |
31e4c28d | 78 | |
4f85cb96 | 79 | static struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk) |
70087dc3 VG |
80 | { |
81 | return container_of(task_subsys_state(tsk, blkio_subsys_id), | |
82 | struct blkio_cgroup, css); | |
83 | } | |
4f85cb96 TH |
84 | |
85 | struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio) | |
86 | { | |
87 | if (bio && bio->bi_css) | |
88 | return container_of(bio->bi_css, struct blkio_cgroup, css); | |
89 | return task_blkio_cgroup(current); | |
90 | } | |
91 | EXPORT_SYMBOL_GPL(bio_blkio_cgroup); | |
70087dc3 | 92 | |
c1768268 TH |
93 | static inline void blkio_update_group_weight(struct blkio_group *blkg, |
94 | int plid, unsigned int weight) | |
062a644d VG |
95 | { |
96 | struct blkio_policy_type *blkiop; | |
97 | ||
98 | list_for_each_entry(blkiop, &blkio_list, list) { | |
99 | /* If this policy does not own the blkg, do not send updates */ | |
c1768268 | 100 | if (blkiop->plid != plid) |
062a644d VG |
101 | continue; |
102 | if (blkiop->ops.blkio_update_group_weight_fn) | |
ca32aefc | 103 | blkiop->ops.blkio_update_group_weight_fn(blkg->q, |
fe071437 | 104 | blkg, weight); |
062a644d VG |
105 | } |
106 | } | |
107 | ||
c1768268 TH |
108 | static inline void blkio_update_group_bps(struct blkio_group *blkg, int plid, |
109 | u64 bps, int fileid) | |
4c9eefa1 VG |
110 | { |
111 | struct blkio_policy_type *blkiop; | |
112 | ||
113 | list_for_each_entry(blkiop, &blkio_list, list) { | |
114 | ||
115 | /* If this policy does not own the blkg, do not send updates */ | |
c1768268 | 116 | if (blkiop->plid != plid) |
4c9eefa1 VG |
117 | continue; |
118 | ||
119 | if (fileid == BLKIO_THROTL_read_bps_device | |
120 | && blkiop->ops.blkio_update_group_read_bps_fn) | |
ca32aefc | 121 | blkiop->ops.blkio_update_group_read_bps_fn(blkg->q, |
fe071437 | 122 | blkg, bps); |
4c9eefa1 VG |
123 | |
124 | if (fileid == BLKIO_THROTL_write_bps_device | |
125 | && blkiop->ops.blkio_update_group_write_bps_fn) | |
ca32aefc | 126 | blkiop->ops.blkio_update_group_write_bps_fn(blkg->q, |
fe071437 | 127 | blkg, bps); |
4c9eefa1 VG |
128 | } |
129 | } | |
130 | ||
7702e8f4 | 131 | static inline void blkio_update_group_iops(struct blkio_group *blkg, |
c1768268 TH |
132 | int plid, unsigned int iops, |
133 | int fileid) | |
7702e8f4 VG |
134 | { |
135 | struct blkio_policy_type *blkiop; | |
136 | ||
137 | list_for_each_entry(blkiop, &blkio_list, list) { | |
138 | ||
139 | /* If this policy does not own the blkg, do not send updates */ | |
c1768268 | 140 | if (blkiop->plid != plid) |
7702e8f4 VG |
141 | continue; |
142 | ||
143 | if (fileid == BLKIO_THROTL_read_iops_device | |
144 | && blkiop->ops.blkio_update_group_read_iops_fn) | |
ca32aefc | 145 | blkiop->ops.blkio_update_group_read_iops_fn(blkg->q, |
fe071437 | 146 | blkg, iops); |
7702e8f4 VG |
147 | |
148 | if (fileid == BLKIO_THROTL_write_iops_device | |
149 | && blkiop->ops.blkio_update_group_write_iops_fn) | |
ca32aefc | 150 | blkiop->ops.blkio_update_group_write_iops_fn(blkg->q, |
fe071437 | 151 | blkg,iops); |
7702e8f4 VG |
152 | } |
153 | } | |
154 | ||
9195291e DS |
155 | /* |
156 | * Add to the appropriate stat variable depending on the request type. | |
edf1b879 | 157 | * This should be called with queue_lock held. |
9195291e | 158 | */ |
84c124da DS |
159 | static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction, |
160 | bool sync) | |
9195291e | 161 | { |
84c124da DS |
162 | if (direction) |
163 | stat[BLKIO_STAT_WRITE] += add; | |
9195291e | 164 | else |
84c124da DS |
165 | stat[BLKIO_STAT_READ] += add; |
166 | if (sync) | |
167 | stat[BLKIO_STAT_SYNC] += add; | |
9195291e | 168 | else |
84c124da | 169 | stat[BLKIO_STAT_ASYNC] += add; |
9195291e DS |
170 | } |
171 | ||
cdc1184c DS |
172 | /* |
173 | * Decrements the appropriate stat variable if non-zero depending on the | |
174 | * request type. Panics on value being zero. | |
edf1b879 | 175 | * This should be called with the queue_lock held. |
cdc1184c DS |
176 | */ |
177 | static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync) | |
178 | { | |
179 | if (direction) { | |
180 | BUG_ON(stat[BLKIO_STAT_WRITE] == 0); | |
181 | stat[BLKIO_STAT_WRITE]--; | |
182 | } else { | |
183 | BUG_ON(stat[BLKIO_STAT_READ] == 0); | |
184 | stat[BLKIO_STAT_READ]--; | |
185 | } | |
186 | if (sync) { | |
187 | BUG_ON(stat[BLKIO_STAT_SYNC] == 0); | |
188 | stat[BLKIO_STAT_SYNC]--; | |
189 | } else { | |
190 | BUG_ON(stat[BLKIO_STAT_ASYNC] == 0); | |
191 | stat[BLKIO_STAT_ASYNC]--; | |
192 | } | |
193 | } | |
194 | ||
195 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
edf1b879 | 196 | /* This should be called with the queue_lock held. */ |
812df48d | 197 | static void blkio_set_start_group_wait_time(struct blkio_group *blkg, |
c1768268 TH |
198 | struct blkio_policy_type *pol, |
199 | struct blkio_group *curr_blkg) | |
812df48d | 200 | { |
c1768268 | 201 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
549d3aa8 TH |
202 | |
203 | if (blkio_blkg_waiting(&pd->stats)) | |
812df48d DS |
204 | return; |
205 | if (blkg == curr_blkg) | |
206 | return; | |
549d3aa8 TH |
207 | pd->stats.start_group_wait_time = sched_clock(); |
208 | blkio_mark_blkg_waiting(&pd->stats); | |
812df48d DS |
209 | } |
210 | ||
edf1b879 | 211 | /* This should be called with the queue_lock held. */ |
812df48d DS |
212 | static void blkio_update_group_wait_time(struct blkio_group_stats *stats) |
213 | { | |
214 | unsigned long long now; | |
215 | ||
216 | if (!blkio_blkg_waiting(stats)) | |
217 | return; | |
218 | ||
219 | now = sched_clock(); | |
220 | if (time_after64(now, stats->start_group_wait_time)) | |
221 | stats->group_wait_time += now - stats->start_group_wait_time; | |
222 | blkio_clear_blkg_waiting(stats); | |
223 | } | |
224 | ||
edf1b879 | 225 | /* This should be called with the queue_lock held. */ |
812df48d DS |
226 | static void blkio_end_empty_time(struct blkio_group_stats *stats) |
227 | { | |
228 | unsigned long long now; | |
229 | ||
230 | if (!blkio_blkg_empty(stats)) | |
231 | return; | |
232 | ||
233 | now = sched_clock(); | |
234 | if (time_after64(now, stats->start_empty_time)) | |
235 | stats->empty_time += now - stats->start_empty_time; | |
236 | blkio_clear_blkg_empty(stats); | |
237 | } | |
238 | ||
c1768268 TH |
239 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg, |
240 | struct blkio_policy_type *pol) | |
812df48d | 241 | { |
edf1b879 | 242 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
812df48d | 243 | |
edf1b879 TH |
244 | lockdep_assert_held(blkg->q->queue_lock); |
245 | BUG_ON(blkio_blkg_idling(stats)); | |
246 | ||
247 | stats->start_idle_time = sched_clock(); | |
248 | blkio_mark_blkg_idling(stats); | |
812df48d DS |
249 | } |
250 | EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats); | |
251 | ||
c1768268 TH |
252 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg, |
253 | struct blkio_policy_type *pol) | |
812df48d | 254 | { |
edf1b879 TH |
255 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
256 | ||
257 | lockdep_assert_held(blkg->q->queue_lock); | |
812df48d | 258 | |
812df48d | 259 | if (blkio_blkg_idling(stats)) { |
edf1b879 TH |
260 | unsigned long long now = sched_clock(); |
261 | ||
262 | if (time_after64(now, stats->start_idle_time)) { | |
263 | u64_stats_update_begin(&stats->syncp); | |
812df48d | 264 | stats->idle_time += now - stats->start_idle_time; |
edf1b879 TH |
265 | u64_stats_update_end(&stats->syncp); |
266 | } | |
812df48d DS |
267 | blkio_clear_blkg_idling(stats); |
268 | } | |
812df48d DS |
269 | } |
270 | EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats); | |
271 | ||
c1768268 TH |
272 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg, |
273 | struct blkio_policy_type *pol) | |
cdc1184c | 274 | { |
edf1b879 | 275 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
cdc1184c | 276 | |
edf1b879 TH |
277 | lockdep_assert_held(blkg->q->queue_lock); |
278 | ||
279 | u64_stats_update_begin(&stats->syncp); | |
cdc1184c DS |
280 | stats->avg_queue_size_sum += |
281 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] + | |
282 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]; | |
283 | stats->avg_queue_size_samples++; | |
812df48d | 284 | blkio_update_group_wait_time(stats); |
edf1b879 | 285 | u64_stats_update_end(&stats->syncp); |
cdc1184c | 286 | } |
a11cdaa7 DS |
287 | EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats); |
288 | ||
c1768268 TH |
289 | void blkiocg_set_start_empty_time(struct blkio_group *blkg, |
290 | struct blkio_policy_type *pol) | |
28baf442 | 291 | { |
edf1b879 | 292 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
28baf442 | 293 | |
edf1b879 | 294 | lockdep_assert_held(blkg->q->queue_lock); |
28baf442 DS |
295 | |
296 | if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] || | |
edf1b879 | 297 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) |
28baf442 | 298 | return; |
28baf442 DS |
299 | |
300 | /* | |
e5ff082e VG |
301 | * group is already marked empty. This can happen if cfqq got new |
302 | * request in parent group and moved to this group while being added | |
303 | * to service tree. Just ignore the event and move on. | |
28baf442 | 304 | */ |
edf1b879 | 305 | if (blkio_blkg_empty(stats)) |
e5ff082e | 306 | return; |
e5ff082e | 307 | |
28baf442 DS |
308 | stats->start_empty_time = sched_clock(); |
309 | blkio_mark_blkg_empty(stats); | |
28baf442 DS |
310 | } |
311 | EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time); | |
312 | ||
a11cdaa7 | 313 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
c1768268 TH |
314 | struct blkio_policy_type *pol, |
315 | unsigned long dequeue) | |
a11cdaa7 | 316 | { |
c1768268 | 317 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
549d3aa8 | 318 | |
edf1b879 TH |
319 | lockdep_assert_held(blkg->q->queue_lock); |
320 | ||
549d3aa8 | 321 | pd->stats.dequeue += dequeue; |
a11cdaa7 DS |
322 | } |
323 | EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats); | |
812df48d DS |
324 | #else |
325 | static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
c1768268 TH |
326 | struct blkio_policy_type *pol, |
327 | struct blkio_group *curr_blkg) { } | |
328 | static inline void blkio_end_empty_time(struct blkio_group_stats *stats) { } | |
cdc1184c DS |
329 | #endif |
330 | ||
a11cdaa7 | 331 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
c1768268 TH |
332 | struct blkio_policy_type *pol, |
333 | struct blkio_group *curr_blkg, bool direction, | |
334 | bool sync) | |
cdc1184c | 335 | { |
edf1b879 TH |
336 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
337 | ||
338 | lockdep_assert_held(blkg->q->queue_lock); | |
339 | ||
340 | u64_stats_update_begin(&stats->syncp); | |
341 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_QUEUED], 1, direction, sync); | |
342 | blkio_end_empty_time(stats); | |
343 | u64_stats_update_end(&stats->syncp); | |
cdc1184c | 344 | |
c1768268 | 345 | blkio_set_start_group_wait_time(blkg, pol, curr_blkg); |
cdc1184c | 346 | } |
a11cdaa7 | 347 | EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats); |
cdc1184c | 348 | |
a11cdaa7 | 349 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
c1768268 TH |
350 | struct blkio_policy_type *pol, |
351 | bool direction, bool sync) | |
cdc1184c | 352 | { |
edf1b879 TH |
353 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
354 | ||
355 | lockdep_assert_held(blkg->q->queue_lock); | |
cdc1184c | 356 | |
edf1b879 TH |
357 | u64_stats_update_begin(&stats->syncp); |
358 | blkio_check_and_dec_stat(stats->stat_arr[BLKIO_STAT_QUEUED], direction, | |
359 | sync); | |
360 | u64_stats_update_end(&stats->syncp); | |
cdc1184c | 361 | } |
a11cdaa7 | 362 | EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats); |
cdc1184c | 363 | |
c1768268 TH |
364 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
365 | struct blkio_policy_type *pol, | |
366 | unsigned long time, | |
367 | unsigned long unaccounted_time) | |
22084190 | 368 | { |
edf1b879 TH |
369 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
370 | ||
371 | lockdep_assert_held(blkg->q->queue_lock); | |
303a3acb | 372 | |
edf1b879 TH |
373 | u64_stats_update_begin(&stats->syncp); |
374 | stats->time += time; | |
a23e6869 | 375 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
edf1b879 | 376 | stats->unaccounted_time += unaccounted_time; |
a23e6869 | 377 | #endif |
edf1b879 | 378 | u64_stats_update_end(&stats->syncp); |
22084190 | 379 | } |
303a3acb | 380 | EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used); |
22084190 | 381 | |
5624a4e4 VG |
382 | /* |
383 | * should be called under rcu read lock or queue lock to make sure blkg pointer | |
384 | * is valid. | |
385 | */ | |
84c124da | 386 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
c1768268 TH |
387 | struct blkio_policy_type *pol, |
388 | uint64_t bytes, bool direction, bool sync) | |
9195291e | 389 | { |
c1768268 | 390 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
5624a4e4 | 391 | struct blkio_group_stats_cpu *stats_cpu; |
575969a0 VG |
392 | unsigned long flags; |
393 | ||
1cd9e039 VG |
394 | /* If per cpu stats are not allocated yet, don't do any accounting. */ |
395 | if (pd->stats_cpu == NULL) | |
396 | return; | |
397 | ||
575969a0 VG |
398 | /* |
399 | * Disabling interrupts to provide mutual exclusion between two | |
400 | * writes on same cpu. It probably is not needed for 64bit. Not | |
401 | * optimizing that case yet. | |
402 | */ | |
403 | local_irq_save(flags); | |
9195291e | 404 | |
549d3aa8 | 405 | stats_cpu = this_cpu_ptr(pd->stats_cpu); |
5624a4e4 | 406 | |
575969a0 | 407 | u64_stats_update_begin(&stats_cpu->syncp); |
5624a4e4 VG |
408 | stats_cpu->sectors += bytes >> 9; |
409 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED], | |
410 | 1, direction, sync); | |
411 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES], | |
412 | bytes, direction, sync); | |
575969a0 VG |
413 | u64_stats_update_end(&stats_cpu->syncp); |
414 | local_irq_restore(flags); | |
9195291e | 415 | } |
84c124da | 416 | EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats); |
9195291e | 417 | |
84c124da | 418 | void blkiocg_update_completion_stats(struct blkio_group *blkg, |
c1768268 TH |
419 | struct blkio_policy_type *pol, |
420 | uint64_t start_time, | |
421 | uint64_t io_start_time, bool direction, | |
422 | bool sync) | |
9195291e | 423 | { |
edf1b879 | 424 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
9195291e DS |
425 | unsigned long long now = sched_clock(); |
426 | ||
edf1b879 TH |
427 | lockdep_assert_held(blkg->q->queue_lock); |
428 | ||
429 | u64_stats_update_begin(&stats->syncp); | |
84c124da DS |
430 | if (time_after64(now, io_start_time)) |
431 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME], | |
432 | now - io_start_time, direction, sync); | |
433 | if (time_after64(io_start_time, start_time)) | |
434 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME], | |
435 | io_start_time - start_time, direction, sync); | |
edf1b879 | 436 | u64_stats_update_end(&stats->syncp); |
9195291e | 437 | } |
84c124da | 438 | EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats); |
9195291e | 439 | |
317389a7 | 440 | /* Merged stats are per cpu. */ |
c1768268 TH |
441 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, |
442 | struct blkio_policy_type *pol, | |
443 | bool direction, bool sync) | |
812d4026 | 444 | { |
edf1b879 TH |
445 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
446 | ||
447 | lockdep_assert_held(blkg->q->queue_lock); | |
812d4026 | 448 | |
edf1b879 | 449 | u64_stats_update_begin(&stats->syncp); |
5fe224d2 | 450 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_MERGED], 1, direction, sync); |
edf1b879 | 451 | u64_stats_update_end(&stats->syncp); |
812d4026 DS |
452 | } |
453 | EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats); | |
454 | ||
1cd9e039 VG |
455 | /* |
456 | * Worker for allocating per cpu stat for blk groups. This is scheduled on | |
457 | * the system_nrt_wq once there are some groups on the alloc_list waiting | |
458 | * for allocation. | |
459 | */ | |
460 | static void blkio_stat_alloc_fn(struct work_struct *work) | |
461 | { | |
462 | static void *pcpu_stats[BLKIO_NR_POLICIES]; | |
463 | struct delayed_work *dwork = to_delayed_work(work); | |
464 | struct blkio_group *blkg; | |
465 | int i; | |
466 | bool empty = false; | |
467 | ||
468 | alloc_stats: | |
469 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { | |
470 | if (pcpu_stats[i] != NULL) | |
471 | continue; | |
472 | ||
473 | pcpu_stats[i] = alloc_percpu(struct blkio_group_stats_cpu); | |
474 | ||
475 | /* Allocation failed. Try again after some time. */ | |
476 | if (pcpu_stats[i] == NULL) { | |
477 | queue_delayed_work(system_nrt_wq, dwork, | |
478 | msecs_to_jiffies(10)); | |
479 | return; | |
480 | } | |
481 | } | |
482 | ||
483 | spin_lock_irq(&blkio_list_lock); | |
484 | spin_lock(&alloc_list_lock); | |
485 | ||
486 | /* cgroup got deleted or queue exited. */ | |
487 | if (!list_empty(&alloc_list)) { | |
488 | blkg = list_first_entry(&alloc_list, struct blkio_group, | |
489 | alloc_node); | |
490 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { | |
491 | struct blkg_policy_data *pd = blkg->pd[i]; | |
492 | ||
493 | if (blkio_policy[i] && pd && !pd->stats_cpu) | |
494 | swap(pd->stats_cpu, pcpu_stats[i]); | |
495 | } | |
496 | ||
497 | list_del_init(&blkg->alloc_node); | |
498 | } | |
499 | ||
500 | empty = list_empty(&alloc_list); | |
501 | ||
502 | spin_unlock(&alloc_list_lock); | |
503 | spin_unlock_irq(&blkio_list_lock); | |
504 | ||
505 | if (!empty) | |
506 | goto alloc_stats; | |
507 | } | |
508 | ||
0381411e TH |
509 | /** |
510 | * blkg_free - free a blkg | |
511 | * @blkg: blkg to free | |
512 | * | |
513 | * Free @blkg which may be partially allocated. | |
514 | */ | |
515 | static void blkg_free(struct blkio_group *blkg) | |
516 | { | |
e8989fae | 517 | int i; |
549d3aa8 TH |
518 | |
519 | if (!blkg) | |
520 | return; | |
521 | ||
e8989fae TH |
522 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
523 | struct blkg_policy_data *pd = blkg->pd[i]; | |
524 | ||
525 | if (pd) { | |
526 | free_percpu(pd->stats_cpu); | |
527 | kfree(pd); | |
528 | } | |
0381411e | 529 | } |
e8989fae | 530 | |
549d3aa8 | 531 | kfree(blkg); |
0381411e TH |
532 | } |
533 | ||
534 | /** | |
535 | * blkg_alloc - allocate a blkg | |
536 | * @blkcg: block cgroup the new blkg is associated with | |
537 | * @q: request_queue the new blkg is associated with | |
0381411e | 538 | * |
e8989fae | 539 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e TH |
540 | */ |
541 | static struct blkio_group *blkg_alloc(struct blkio_cgroup *blkcg, | |
e8989fae | 542 | struct request_queue *q) |
0381411e TH |
543 | { |
544 | struct blkio_group *blkg; | |
e8989fae | 545 | int i; |
0381411e TH |
546 | |
547 | /* alloc and init base part */ | |
548 | blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node); | |
549 | if (!blkg) | |
550 | return NULL; | |
551 | ||
c875f4d0 | 552 | blkg->q = q; |
e8989fae | 553 | INIT_LIST_HEAD(&blkg->q_node); |
1cd9e039 | 554 | INIT_LIST_HEAD(&blkg->alloc_node); |
0381411e | 555 | blkg->blkcg = blkcg; |
1adaf3dd | 556 | blkg->refcnt = 1; |
0381411e TH |
557 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); |
558 | ||
e8989fae TH |
559 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
560 | struct blkio_policy_type *pol = blkio_policy[i]; | |
561 | struct blkg_policy_data *pd; | |
0381411e | 562 | |
e8989fae TH |
563 | if (!pol) |
564 | continue; | |
565 | ||
566 | /* alloc per-policy data and attach it to blkg */ | |
567 | pd = kzalloc_node(sizeof(*pd) + pol->pdata_size, GFP_ATOMIC, | |
568 | q->node); | |
569 | if (!pd) { | |
570 | blkg_free(blkg); | |
571 | return NULL; | |
572 | } | |
549d3aa8 | 573 | |
e8989fae TH |
574 | blkg->pd[i] = pd; |
575 | pd->blkg = blkg; | |
0381411e TH |
576 | } |
577 | ||
549d3aa8 | 578 | /* invoke per-policy init */ |
e8989fae TH |
579 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
580 | struct blkio_policy_type *pol = blkio_policy[i]; | |
581 | ||
582 | if (pol) | |
583 | pol->ops.blkio_init_group_fn(blkg); | |
584 | } | |
585 | ||
0381411e TH |
586 | return blkg; |
587 | } | |
588 | ||
cd1604fa TH |
589 | struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg, |
590 | struct request_queue *q, | |
591 | enum blkio_policy_id plid, | |
592 | bool for_root) | |
593 | __releases(q->queue_lock) __acquires(q->queue_lock) | |
5624a4e4 | 594 | { |
1cd9e039 | 595 | struct blkio_group *blkg; |
5624a4e4 | 596 | |
cd1604fa TH |
597 | WARN_ON_ONCE(!rcu_read_lock_held()); |
598 | lockdep_assert_held(q->queue_lock); | |
599 | ||
600 | /* | |
601 | * This could be the first entry point of blkcg implementation and | |
602 | * we shouldn't allow anything to go through for a bypassing queue. | |
603 | * The following can be removed if blkg lookup is guaranteed to | |
604 | * fail on a bypassing queue. | |
605 | */ | |
606 | if (unlikely(blk_queue_bypass(q)) && !for_root) | |
607 | return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY); | |
608 | ||
e8989fae | 609 | blkg = blkg_lookup(blkcg, q); |
cd1604fa TH |
610 | if (blkg) |
611 | return blkg; | |
612 | ||
7ee9c562 | 613 | /* blkg holds a reference to blkcg */ |
cd1604fa TH |
614 | if (!css_tryget(&blkcg->css)) |
615 | return ERR_PTR(-EINVAL); | |
616 | ||
617 | /* | |
618 | * Allocate and initialize. | |
cd1604fa | 619 | */ |
1cd9e039 | 620 | blkg = blkg_alloc(blkcg, q); |
cd1604fa TH |
621 | |
622 | /* did alloc fail? */ | |
1cd9e039 | 623 | if (unlikely(!blkg)) { |
cd1604fa TH |
624 | blkg = ERR_PTR(-ENOMEM); |
625 | goto out; | |
626 | } | |
627 | ||
628 | /* insert */ | |
629 | spin_lock(&blkcg->lock); | |
31e4c28d | 630 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
e8989fae | 631 | list_add(&blkg->q_node, &q->blkg_list); |
cd1604fa | 632 | spin_unlock(&blkcg->lock); |
1cd9e039 VG |
633 | |
634 | spin_lock(&alloc_list_lock); | |
635 | list_add(&blkg->alloc_node, &alloc_list); | |
636 | /* Queue per cpu stat allocation from worker thread. */ | |
637 | queue_delayed_work(system_nrt_wq, &blkio_stat_alloc_work, 0); | |
638 | spin_unlock(&alloc_list_lock); | |
cd1604fa | 639 | out: |
cd1604fa | 640 | return blkg; |
31e4c28d | 641 | } |
cd1604fa | 642 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 643 | |
31e4c28d | 644 | /* called under rcu_read_lock(). */ |
cd1604fa | 645 | struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg, |
e8989fae | 646 | struct request_queue *q) |
31e4c28d VG |
647 | { |
648 | struct blkio_group *blkg; | |
649 | struct hlist_node *n; | |
31e4c28d | 650 | |
ca32aefc | 651 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) |
e8989fae | 652 | if (blkg->q == q) |
31e4c28d | 653 | return blkg; |
31e4c28d VG |
654 | return NULL; |
655 | } | |
cd1604fa | 656 | EXPORT_SYMBOL_GPL(blkg_lookup); |
31e4c28d | 657 | |
e8989fae | 658 | static void blkg_destroy(struct blkio_group *blkg) |
03aa264a TH |
659 | { |
660 | struct request_queue *q = blkg->q; | |
9f13ef67 | 661 | struct blkio_cgroup *blkcg = blkg->blkcg; |
03aa264a TH |
662 | |
663 | lockdep_assert_held(q->queue_lock); | |
9f13ef67 | 664 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
665 | |
666 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 667 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 668 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
e8989fae | 669 | list_del_init(&blkg->q_node); |
9f13ef67 | 670 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 671 | |
1cd9e039 VG |
672 | spin_lock(&alloc_list_lock); |
673 | list_del_init(&blkg->alloc_node); | |
674 | spin_unlock(&alloc_list_lock); | |
675 | ||
03aa264a TH |
676 | /* |
677 | * Put the reference taken at the time of creation so that when all | |
678 | * queues are gone, group can be destroyed. | |
679 | */ | |
680 | blkg_put(blkg); | |
681 | } | |
682 | ||
e8989fae TH |
683 | /* |
684 | * XXX: This updates blkg policy data in-place for root blkg, which is | |
685 | * necessary across elevator switch and policy registration as root blkgs | |
686 | * aren't shot down. This broken and racy implementation is temporary. | |
687 | * Eventually, blkg shoot down will be replaced by proper in-place update. | |
688 | */ | |
689 | void update_root_blkg_pd(struct request_queue *q, enum blkio_policy_id plid) | |
690 | { | |
691 | struct blkio_policy_type *pol = blkio_policy[plid]; | |
692 | struct blkio_group *blkg = blkg_lookup(&blkio_root_cgroup, q); | |
693 | struct blkg_policy_data *pd; | |
694 | ||
695 | if (!blkg) | |
696 | return; | |
697 | ||
698 | kfree(blkg->pd[plid]); | |
699 | blkg->pd[plid] = NULL; | |
700 | ||
701 | if (!pol) | |
702 | return; | |
703 | ||
704 | pd = kzalloc(sizeof(*pd) + pol->pdata_size, GFP_KERNEL); | |
705 | WARN_ON_ONCE(!pd); | |
706 | ||
707 | pd->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu); | |
708 | WARN_ON_ONCE(!pd->stats_cpu); | |
709 | ||
710 | blkg->pd[plid] = pd; | |
711 | pd->blkg = blkg; | |
712 | pol->ops.blkio_init_group_fn(blkg); | |
713 | } | |
714 | EXPORT_SYMBOL_GPL(update_root_blkg_pd); | |
715 | ||
9f13ef67 TH |
716 | /** |
717 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
718 | * @q: request_queue of interest | |
719 | * @destroy_root: whether to destroy root blkg or not | |
720 | * | |
721 | * Destroy blkgs associated with @q. If @destroy_root is %true, all are | |
722 | * destroyed; otherwise, root blkg is left alone. | |
723 | */ | |
e8989fae | 724 | void blkg_destroy_all(struct request_queue *q, bool destroy_root) |
72e06c25 | 725 | { |
03aa264a | 726 | struct blkio_group *blkg, *n; |
72e06c25 | 727 | |
9f13ef67 | 728 | spin_lock_irq(q->queue_lock); |
72e06c25 | 729 | |
9f13ef67 TH |
730 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
731 | struct blkio_cgroup *blkcg = blkg->blkcg; | |
72e06c25 | 732 | |
9f13ef67 TH |
733 | /* skip root? */ |
734 | if (!destroy_root && blkg->blkcg == &blkio_root_cgroup) | |
735 | continue; | |
72e06c25 | 736 | |
9f13ef67 TH |
737 | spin_lock(&blkcg->lock); |
738 | blkg_destroy(blkg); | |
739 | spin_unlock(&blkcg->lock); | |
72e06c25 | 740 | } |
9f13ef67 TH |
741 | |
742 | spin_unlock_irq(q->queue_lock); | |
72e06c25 | 743 | } |
03aa264a | 744 | EXPORT_SYMBOL_GPL(blkg_destroy_all); |
72e06c25 | 745 | |
1adaf3dd TH |
746 | static void blkg_rcu_free(struct rcu_head *rcu_head) |
747 | { | |
748 | blkg_free(container_of(rcu_head, struct blkio_group, rcu_head)); | |
749 | } | |
750 | ||
751 | void __blkg_release(struct blkio_group *blkg) | |
752 | { | |
753 | /* release the extra blkcg reference this blkg has been holding */ | |
754 | css_put(&blkg->blkcg->css); | |
755 | ||
756 | /* | |
757 | * A group is freed in rcu manner. But having an rcu lock does not | |
758 | * mean that one can access all the fields of blkg and assume these | |
759 | * are valid. For example, don't try to follow throtl_data and | |
760 | * request queue links. | |
761 | * | |
762 | * Having a reference to blkg under an rcu allows acess to only | |
763 | * values local to groups like group stats and group rate limits | |
764 | */ | |
765 | call_rcu(&blkg->rcu_head, blkg_rcu_free); | |
766 | } | |
767 | EXPORT_SYMBOL_GPL(__blkg_release); | |
768 | ||
c1768268 | 769 | static void blkio_reset_stats_cpu(struct blkio_group *blkg, int plid) |
f0bdc8cd | 770 | { |
c1768268 | 771 | struct blkg_policy_data *pd = blkg->pd[plid]; |
997a026c | 772 | int cpu; |
1cd9e039 VG |
773 | |
774 | if (pd->stats_cpu == NULL) | |
775 | return; | |
997a026c TH |
776 | |
777 | for_each_possible_cpu(cpu) { | |
778 | struct blkio_group_stats_cpu *sc = | |
779 | per_cpu_ptr(pd->stats_cpu, cpu); | |
780 | ||
781 | sc->sectors = 0; | |
782 | memset(sc->stat_arr_cpu, 0, sizeof(sc->stat_arr_cpu)); | |
f0bdc8cd VG |
783 | } |
784 | } | |
785 | ||
303a3acb | 786 | static int |
84c124da | 787 | blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val) |
303a3acb | 788 | { |
997a026c | 789 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); |
303a3acb DS |
790 | struct blkio_group *blkg; |
791 | struct hlist_node *n; | |
cdc1184c | 792 | int i; |
303a3acb | 793 | |
e8989fae | 794 | spin_lock(&blkio_list_lock); |
303a3acb | 795 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
796 | |
797 | /* | |
798 | * Note that stat reset is racy - it doesn't synchronize against | |
799 | * stat updates. This is a debug feature which shouldn't exist | |
800 | * anyway. If you get hit by a race, retry. | |
801 | */ | |
303a3acb | 802 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
e8989fae | 803 | struct blkio_policy_type *pol; |
549d3aa8 | 804 | |
e8989fae TH |
805 | list_for_each_entry(pol, &blkio_list, list) { |
806 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; | |
997a026c TH |
807 | struct blkio_group_stats *stats = &pd->stats; |
808 | ||
809 | /* queued stats shouldn't be cleared */ | |
810 | for (i = 0; i < ARRAY_SIZE(stats->stat_arr); i++) | |
811 | if (i != BLKIO_STAT_QUEUED) | |
812 | memset(stats->stat_arr[i], 0, | |
813 | sizeof(stats->stat_arr[i])); | |
814 | stats->time = 0; | |
812df48d | 815 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
997a026c TH |
816 | memset((void *)stats + BLKG_STATS_DEBUG_CLEAR_START, 0, |
817 | BLKG_STATS_DEBUG_CLEAR_SIZE); | |
812df48d | 818 | #endif |
e8989fae TH |
819 | blkio_reset_stats_cpu(blkg, pol->plid); |
820 | } | |
303a3acb | 821 | } |
f0bdc8cd | 822 | |
303a3acb | 823 | spin_unlock_irq(&blkcg->lock); |
e8989fae | 824 | spin_unlock(&blkio_list_lock); |
303a3acb DS |
825 | return 0; |
826 | } | |
827 | ||
7a4dd281 TH |
828 | static void blkio_get_key_name(enum stat_sub_type type, const char *dname, |
829 | char *str, int chars_left, bool diskname_only) | |
303a3acb | 830 | { |
7a4dd281 | 831 | snprintf(str, chars_left, "%s", dname); |
303a3acb DS |
832 | chars_left -= strlen(str); |
833 | if (chars_left <= 0) { | |
834 | printk(KERN_WARNING | |
835 | "Possibly incorrect cgroup stat display format"); | |
836 | return; | |
837 | } | |
84c124da DS |
838 | if (diskname_only) |
839 | return; | |
303a3acb | 840 | switch (type) { |
84c124da | 841 | case BLKIO_STAT_READ: |
303a3acb DS |
842 | strlcat(str, " Read", chars_left); |
843 | break; | |
84c124da | 844 | case BLKIO_STAT_WRITE: |
303a3acb DS |
845 | strlcat(str, " Write", chars_left); |
846 | break; | |
84c124da | 847 | case BLKIO_STAT_SYNC: |
303a3acb DS |
848 | strlcat(str, " Sync", chars_left); |
849 | break; | |
84c124da | 850 | case BLKIO_STAT_ASYNC: |
303a3acb DS |
851 | strlcat(str, " Async", chars_left); |
852 | break; | |
84c124da | 853 | case BLKIO_STAT_TOTAL: |
303a3acb DS |
854 | strlcat(str, " Total", chars_left); |
855 | break; | |
856 | default: | |
857 | strlcat(str, " Invalid", chars_left); | |
858 | } | |
859 | } | |
860 | ||
c1768268 | 861 | static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg, int plid, |
5624a4e4 VG |
862 | enum stat_type_cpu type, enum stat_sub_type sub_type) |
863 | { | |
c1768268 | 864 | struct blkg_policy_data *pd = blkg->pd[plid]; |
5624a4e4 VG |
865 | int cpu; |
866 | struct blkio_group_stats_cpu *stats_cpu; | |
575969a0 | 867 | u64 val = 0, tval; |
5624a4e4 | 868 | |
1cd9e039 VG |
869 | if (pd->stats_cpu == NULL) |
870 | return val; | |
871 | ||
5624a4e4 | 872 | for_each_possible_cpu(cpu) { |
575969a0 | 873 | unsigned int start; |
549d3aa8 | 874 | stats_cpu = per_cpu_ptr(pd->stats_cpu, cpu); |
5624a4e4 | 875 | |
575969a0 VG |
876 | do { |
877 | start = u64_stats_fetch_begin(&stats_cpu->syncp); | |
878 | if (type == BLKIO_STAT_CPU_SECTORS) | |
879 | tval = stats_cpu->sectors; | |
880 | else | |
881 | tval = stats_cpu->stat_arr_cpu[type][sub_type]; | |
882 | } while(u64_stats_fetch_retry(&stats_cpu->syncp, start)); | |
883 | ||
884 | val += tval; | |
5624a4e4 VG |
885 | } |
886 | ||
887 | return val; | |
888 | } | |
889 | ||
c1768268 | 890 | static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg, int plid, |
7a4dd281 TH |
891 | struct cgroup_map_cb *cb, const char *dname, |
892 | enum stat_type_cpu type) | |
5624a4e4 VG |
893 | { |
894 | uint64_t disk_total, val; | |
895 | char key_str[MAX_KEY_LEN]; | |
896 | enum stat_sub_type sub_type; | |
897 | ||
898 | if (type == BLKIO_STAT_CPU_SECTORS) { | |
c1768268 | 899 | val = blkio_read_stat_cpu(blkg, plid, type, 0); |
c4c76a05 TH |
900 | blkio_get_key_name(0, dname, key_str, MAX_KEY_LEN, true); |
901 | cb->fill(cb, key_str, val); | |
902 | return val; | |
5624a4e4 VG |
903 | } |
904 | ||
905 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; | |
906 | sub_type++) { | |
7a4dd281 TH |
907 | blkio_get_key_name(sub_type, dname, key_str, MAX_KEY_LEN, |
908 | false); | |
c1768268 | 909 | val = blkio_read_stat_cpu(blkg, plid, type, sub_type); |
5624a4e4 VG |
910 | cb->fill(cb, key_str, val); |
911 | } | |
912 | ||
c1768268 TH |
913 | disk_total = blkio_read_stat_cpu(blkg, plid, type, BLKIO_STAT_READ) + |
914 | blkio_read_stat_cpu(blkg, plid, type, BLKIO_STAT_WRITE); | |
5624a4e4 | 915 | |
7a4dd281 TH |
916 | blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN, |
917 | false); | |
5624a4e4 VG |
918 | cb->fill(cb, key_str, disk_total); |
919 | return disk_total; | |
920 | } | |
921 | ||
c1768268 | 922 | static uint64_t blkio_get_stat(struct blkio_group *blkg, int plid, |
7a4dd281 TH |
923 | struct cgroup_map_cb *cb, const char *dname, |
924 | enum stat_type type) | |
303a3acb | 925 | { |
c4c76a05 TH |
926 | struct blkio_group_stats *stats = &blkg->pd[plid]->stats; |
927 | uint64_t v = 0, disk_total = 0; | |
303a3acb | 928 | char key_str[MAX_KEY_LEN]; |
edf1b879 | 929 | unsigned int sync_start; |
c4c76a05 | 930 | int st; |
84c124da | 931 | |
c4c76a05 | 932 | if (type >= BLKIO_STAT_ARR_NR) { |
edf1b879 TH |
933 | do { |
934 | sync_start = u64_stats_fetch_begin(&stats->syncp); | |
935 | switch (type) { | |
936 | case BLKIO_STAT_TIME: | |
937 | v = stats->time; | |
938 | break; | |
9026e521 | 939 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
edf1b879 TH |
940 | case BLKIO_STAT_UNACCOUNTED_TIME: |
941 | v = stats->unaccounted_time; | |
942 | break; | |
943 | case BLKIO_STAT_AVG_QUEUE_SIZE: { | |
944 | uint64_t samples = stats->avg_queue_size_samples; | |
c4c76a05 | 945 | |
edf1b879 TH |
946 | if (samples) { |
947 | v = stats->avg_queue_size_sum; | |
948 | do_div(v, samples); | |
949 | } | |
950 | break; | |
c4c76a05 | 951 | } |
edf1b879 TH |
952 | case BLKIO_STAT_IDLE_TIME: |
953 | v = stats->idle_time; | |
954 | break; | |
955 | case BLKIO_STAT_EMPTY_TIME: | |
956 | v = stats->empty_time; | |
957 | break; | |
958 | case BLKIO_STAT_DEQUEUE: | |
959 | v = stats->dequeue; | |
960 | break; | |
961 | case BLKIO_STAT_GROUP_WAIT_TIME: | |
962 | v = stats->group_wait_time; | |
963 | break; | |
84c124da | 964 | #endif |
edf1b879 TH |
965 | default: |
966 | WARN_ON_ONCE(1); | |
967 | } | |
968 | } while (u64_stats_fetch_retry(&stats->syncp, sync_start)); | |
303a3acb | 969 | |
c4c76a05 TH |
970 | blkio_get_key_name(0, dname, key_str, MAX_KEY_LEN, true); |
971 | cb->fill(cb, key_str, v); | |
972 | return v; | |
303a3acb | 973 | } |
c4c76a05 TH |
974 | |
975 | for (st = BLKIO_STAT_READ; st < BLKIO_STAT_TOTAL; st++) { | |
edf1b879 TH |
976 | do { |
977 | sync_start = u64_stats_fetch_begin(&stats->syncp); | |
978 | v = stats->stat_arr[type][st]; | |
979 | } while (u64_stats_fetch_retry(&stats->syncp, sync_start)); | |
c4c76a05 TH |
980 | |
981 | blkio_get_key_name(st, dname, key_str, MAX_KEY_LEN, false); | |
982 | cb->fill(cb, key_str, v); | |
983 | if (st == BLKIO_STAT_READ || st == BLKIO_STAT_WRITE) | |
984 | disk_total += v; | |
985 | } | |
986 | ||
7a4dd281 TH |
987 | blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN, |
988 | false); | |
303a3acb DS |
989 | cb->fill(cb, key_str, disk_total); |
990 | return disk_total; | |
991 | } | |
992 | ||
4bfd482e TH |
993 | static int blkio_policy_parse_and_set(char *buf, enum blkio_policy_id plid, |
994 | int fileid, struct blkio_cgroup *blkcg) | |
34d0f179 | 995 | { |
ece84241 | 996 | struct gendisk *disk = NULL; |
e56da7e2 | 997 | struct blkio_group *blkg = NULL; |
549d3aa8 | 998 | struct blkg_policy_data *pd; |
34d0f179 | 999 | char *s[4], *p, *major_s = NULL, *minor_s = NULL; |
d11bb446 | 1000 | unsigned long major, minor; |
ece84241 TH |
1001 | int i = 0, ret = -EINVAL; |
1002 | int part; | |
34d0f179 | 1003 | dev_t dev; |
d11bb446 | 1004 | u64 temp; |
34d0f179 GJ |
1005 | |
1006 | memset(s, 0, sizeof(s)); | |
1007 | ||
1008 | while ((p = strsep(&buf, " ")) != NULL) { | |
1009 | if (!*p) | |
1010 | continue; | |
1011 | ||
1012 | s[i++] = p; | |
1013 | ||
1014 | /* Prevent from inputing too many things */ | |
1015 | if (i == 3) | |
1016 | break; | |
1017 | } | |
1018 | ||
1019 | if (i != 2) | |
ece84241 | 1020 | goto out; |
34d0f179 GJ |
1021 | |
1022 | p = strsep(&s[0], ":"); | |
1023 | if (p != NULL) | |
1024 | major_s = p; | |
1025 | else | |
ece84241 | 1026 | goto out; |
34d0f179 GJ |
1027 | |
1028 | minor_s = s[0]; | |
1029 | if (!minor_s) | |
ece84241 | 1030 | goto out; |
34d0f179 | 1031 | |
ece84241 TH |
1032 | if (strict_strtoul(major_s, 10, &major)) |
1033 | goto out; | |
34d0f179 | 1034 | |
ece84241 TH |
1035 | if (strict_strtoul(minor_s, 10, &minor)) |
1036 | goto out; | |
34d0f179 GJ |
1037 | |
1038 | dev = MKDEV(major, minor); | |
1039 | ||
ece84241 TH |
1040 | if (strict_strtoull(s[1], 10, &temp)) |
1041 | goto out; | |
34d0f179 | 1042 | |
e56da7e2 | 1043 | disk = get_gendisk(dev, &part); |
4bfd482e | 1044 | if (!disk || part) |
e56da7e2 | 1045 | goto out; |
e56da7e2 TH |
1046 | |
1047 | rcu_read_lock(); | |
1048 | ||
4bfd482e TH |
1049 | spin_lock_irq(disk->queue->queue_lock); |
1050 | blkg = blkg_lookup_create(blkcg, disk->queue, plid, false); | |
1051 | spin_unlock_irq(disk->queue->queue_lock); | |
e56da7e2 | 1052 | |
4bfd482e TH |
1053 | if (IS_ERR(blkg)) { |
1054 | ret = PTR_ERR(blkg); | |
1055 | goto out_unlock; | |
d11bb446 | 1056 | } |
34d0f179 | 1057 | |
549d3aa8 TH |
1058 | pd = blkg->pd[plid]; |
1059 | ||
062a644d VG |
1060 | switch (plid) { |
1061 | case BLKIO_POLICY_PROP: | |
d11bb446 WG |
1062 | if ((temp < BLKIO_WEIGHT_MIN && temp > 0) || |
1063 | temp > BLKIO_WEIGHT_MAX) | |
e56da7e2 | 1064 | goto out_unlock; |
34d0f179 | 1065 | |
549d3aa8 | 1066 | pd->conf.weight = temp; |
c1768268 | 1067 | blkio_update_group_weight(blkg, plid, temp ?: blkcg->weight); |
4c9eefa1 VG |
1068 | break; |
1069 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
1070 | switch(fileid) { |
1071 | case BLKIO_THROTL_read_bps_device: | |
549d3aa8 | 1072 | pd->conf.bps[READ] = temp; |
c1768268 | 1073 | blkio_update_group_bps(blkg, plid, temp ?: -1, fileid); |
e56da7e2 | 1074 | break; |
7702e8f4 | 1075 | case BLKIO_THROTL_write_bps_device: |
549d3aa8 | 1076 | pd->conf.bps[WRITE] = temp; |
c1768268 | 1077 | blkio_update_group_bps(blkg, plid, temp ?: -1, fileid); |
7702e8f4 VG |
1078 | break; |
1079 | case BLKIO_THROTL_read_iops_device: | |
e56da7e2 TH |
1080 | if (temp > THROTL_IOPS_MAX) |
1081 | goto out_unlock; | |
549d3aa8 | 1082 | pd->conf.iops[READ] = temp; |
c1768268 | 1083 | blkio_update_group_iops(blkg, plid, temp ?: -1, fileid); |
e56da7e2 | 1084 | break; |
7702e8f4 | 1085 | case BLKIO_THROTL_write_iops_device: |
d11bb446 | 1086 | if (temp > THROTL_IOPS_MAX) |
e56da7e2 | 1087 | goto out_unlock; |
549d3aa8 | 1088 | pd->conf.iops[WRITE] = temp; |
c1768268 | 1089 | blkio_update_group_iops(blkg, plid, temp ?: -1, fileid); |
7702e8f4 VG |
1090 | break; |
1091 | } | |
062a644d VG |
1092 | break; |
1093 | default: | |
1094 | BUG(); | |
1095 | } | |
ece84241 | 1096 | ret = 0; |
e56da7e2 TH |
1097 | out_unlock: |
1098 | rcu_read_unlock(); | |
ece84241 TH |
1099 | out: |
1100 | put_disk(disk); | |
e56da7e2 TH |
1101 | |
1102 | /* | |
1103 | * If queue was bypassing, we should retry. Do so after a short | |
1104 | * msleep(). It isn't strictly necessary but queue can be | |
1105 | * bypassing for some time and it's always nice to avoid busy | |
1106 | * looping. | |
1107 | */ | |
1108 | if (ret == -EBUSY) { | |
1109 | msleep(10); | |
1110 | return restart_syscall(); | |
1111 | } | |
ece84241 | 1112 | return ret; |
34d0f179 GJ |
1113 | } |
1114 | ||
062a644d VG |
1115 | static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft, |
1116 | const char *buffer) | |
34d0f179 GJ |
1117 | { |
1118 | int ret = 0; | |
1119 | char *buf; | |
e56da7e2 | 1120 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); |
062a644d VG |
1121 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); |
1122 | int fileid = BLKIOFILE_ATTR(cft->private); | |
34d0f179 GJ |
1123 | |
1124 | buf = kstrdup(buffer, GFP_KERNEL); | |
1125 | if (!buf) | |
1126 | return -ENOMEM; | |
1127 | ||
4bfd482e | 1128 | ret = blkio_policy_parse_and_set(buf, plid, fileid, blkcg); |
34d0f179 GJ |
1129 | kfree(buf); |
1130 | return ret; | |
1131 | } | |
1132 | ||
92616b5b VG |
1133 | static const char *blkg_dev_name(struct blkio_group *blkg) |
1134 | { | |
1135 | /* some drivers (floppy) instantiate a queue w/o disk registered */ | |
1136 | if (blkg->q->backing_dev_info.dev) | |
1137 | return dev_name(blkg->q->backing_dev_info.dev); | |
1138 | return NULL; | |
1139 | } | |
1140 | ||
4bfd482e TH |
1141 | static void blkio_print_group_conf(struct cftype *cft, struct blkio_group *blkg, |
1142 | struct seq_file *m) | |
34d0f179 | 1143 | { |
c1768268 | 1144 | int plid = BLKIOFILE_POLICY(cft->private); |
4bfd482e | 1145 | int fileid = BLKIOFILE_ATTR(cft->private); |
c1768268 TH |
1146 | struct blkg_policy_data *pd = blkg->pd[plid]; |
1147 | const char *dname = blkg_dev_name(blkg); | |
4bfd482e TH |
1148 | int rw = WRITE; |
1149 | ||
92616b5b VG |
1150 | if (!dname) |
1151 | return; | |
1152 | ||
c1768268 | 1153 | switch (plid) { |
062a644d | 1154 | case BLKIO_POLICY_PROP: |
549d3aa8 | 1155 | if (pd->conf.weight) |
7a4dd281 | 1156 | seq_printf(m, "%s\t%u\n", |
549d3aa8 | 1157 | dname, pd->conf.weight); |
4c9eefa1 VG |
1158 | break; |
1159 | case BLKIO_POLICY_THROTL: | |
4bfd482e | 1160 | switch (fileid) { |
7702e8f4 | 1161 | case BLKIO_THROTL_read_bps_device: |
4bfd482e | 1162 | rw = READ; |
7702e8f4 | 1163 | case BLKIO_THROTL_write_bps_device: |
549d3aa8 | 1164 | if (pd->conf.bps[rw]) |
7a4dd281 | 1165 | seq_printf(m, "%s\t%llu\n", |
549d3aa8 | 1166 | dname, pd->conf.bps[rw]); |
7702e8f4 VG |
1167 | break; |
1168 | case BLKIO_THROTL_read_iops_device: | |
4bfd482e | 1169 | rw = READ; |
7702e8f4 | 1170 | case BLKIO_THROTL_write_iops_device: |
549d3aa8 | 1171 | if (pd->conf.iops[rw]) |
7a4dd281 | 1172 | seq_printf(m, "%s\t%u\n", |
549d3aa8 | 1173 | dname, pd->conf.iops[rw]); |
7702e8f4 VG |
1174 | break; |
1175 | } | |
062a644d VG |
1176 | break; |
1177 | default: | |
1178 | BUG(); | |
1179 | } | |
1180 | } | |
34d0f179 | 1181 | |
062a644d | 1182 | /* cgroup files which read their data from policy nodes end up here */ |
4bfd482e TH |
1183 | static void blkio_read_conf(struct cftype *cft, struct blkio_cgroup *blkcg, |
1184 | struct seq_file *m) | |
34d0f179 | 1185 | { |
4bfd482e TH |
1186 | struct blkio_group *blkg; |
1187 | struct hlist_node *n; | |
34d0f179 | 1188 | |
4bfd482e TH |
1189 | spin_lock_irq(&blkcg->lock); |
1190 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) | |
e8989fae | 1191 | blkio_print_group_conf(cft, blkg, m); |
4bfd482e | 1192 | spin_unlock_irq(&blkcg->lock); |
062a644d VG |
1193 | } |
1194 | ||
1195 | static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft, | |
1196 | struct seq_file *m) | |
1197 | { | |
1198 | struct blkio_cgroup *blkcg; | |
1199 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1200 | int name = BLKIOFILE_ATTR(cft->private); | |
1201 | ||
1202 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1203 | ||
1204 | switch(plid) { | |
1205 | case BLKIO_POLICY_PROP: | |
1206 | switch(name) { | |
1207 | case BLKIO_PROP_weight_device: | |
4bfd482e | 1208 | blkio_read_conf(cft, blkcg, m); |
062a644d VG |
1209 | return 0; |
1210 | default: | |
1211 | BUG(); | |
1212 | } | |
1213 | break; | |
4c9eefa1 VG |
1214 | case BLKIO_POLICY_THROTL: |
1215 | switch(name){ | |
1216 | case BLKIO_THROTL_read_bps_device: | |
1217 | case BLKIO_THROTL_write_bps_device: | |
7702e8f4 VG |
1218 | case BLKIO_THROTL_read_iops_device: |
1219 | case BLKIO_THROTL_write_iops_device: | |
4bfd482e | 1220 | blkio_read_conf(cft, blkcg, m); |
4c9eefa1 VG |
1221 | return 0; |
1222 | default: | |
1223 | BUG(); | |
1224 | } | |
1225 | break; | |
062a644d VG |
1226 | default: |
1227 | BUG(); | |
1228 | } | |
1229 | ||
1230 | return 0; | |
1231 | } | |
1232 | ||
1233 | static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg, | |
5624a4e4 VG |
1234 | struct cftype *cft, struct cgroup_map_cb *cb, |
1235 | enum stat_type type, bool show_total, bool pcpu) | |
062a644d VG |
1236 | { |
1237 | struct blkio_group *blkg; | |
1238 | struct hlist_node *n; | |
1239 | uint64_t cgroup_total = 0; | |
1240 | ||
c875f4d0 TH |
1241 | spin_lock_irq(&blkcg->lock); |
1242 | ||
1243 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
92616b5b | 1244 | const char *dname = blkg_dev_name(blkg); |
c1768268 | 1245 | int plid = BLKIOFILE_POLICY(cft->private); |
7a4dd281 | 1246 | |
e8989fae | 1247 | if (!dname) |
7a4dd281 | 1248 | continue; |
edf1b879 | 1249 | if (pcpu) |
c1768268 TH |
1250 | cgroup_total += blkio_get_stat_cpu(blkg, plid, |
1251 | cb, dname, type); | |
edf1b879 | 1252 | else |
c1768268 TH |
1253 | cgroup_total += blkio_get_stat(blkg, plid, |
1254 | cb, dname, type); | |
062a644d VG |
1255 | } |
1256 | if (show_total) | |
1257 | cb->fill(cb, "Total", cgroup_total); | |
c875f4d0 TH |
1258 | |
1259 | spin_unlock_irq(&blkcg->lock); | |
062a644d VG |
1260 | return 0; |
1261 | } | |
1262 | ||
1263 | /* All map kind of cgroup file get serviced by this function */ | |
1264 | static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft, | |
1265 | struct cgroup_map_cb *cb) | |
1266 | { | |
1267 | struct blkio_cgroup *blkcg; | |
1268 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1269 | int name = BLKIOFILE_ATTR(cft->private); | |
1270 | ||
1271 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1272 | ||
1273 | switch(plid) { | |
1274 | case BLKIO_POLICY_PROP: | |
1275 | switch(name) { | |
1276 | case BLKIO_PROP_time: | |
1277 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1278 | BLKIO_STAT_TIME, 0, 0); |
062a644d VG |
1279 | case BLKIO_PROP_sectors: |
1280 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1281 | BLKIO_STAT_CPU_SECTORS, 0, 1); |
062a644d VG |
1282 | case BLKIO_PROP_io_service_bytes: |
1283 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1284 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
062a644d VG |
1285 | case BLKIO_PROP_io_serviced: |
1286 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1287 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
062a644d VG |
1288 | case BLKIO_PROP_io_service_time: |
1289 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1290 | BLKIO_STAT_SERVICE_TIME, 1, 0); |
062a644d VG |
1291 | case BLKIO_PROP_io_wait_time: |
1292 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1293 | BLKIO_STAT_WAIT_TIME, 1, 0); |
062a644d VG |
1294 | case BLKIO_PROP_io_merged: |
1295 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5fe224d2 | 1296 | BLKIO_STAT_MERGED, 1, 0); |
062a644d VG |
1297 | case BLKIO_PROP_io_queued: |
1298 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1299 | BLKIO_STAT_QUEUED, 1, 0); |
062a644d | 1300 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
9026e521 JT |
1301 | case BLKIO_PROP_unaccounted_time: |
1302 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1303 | BLKIO_STAT_UNACCOUNTED_TIME, 0, 0); |
062a644d VG |
1304 | case BLKIO_PROP_dequeue: |
1305 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1306 | BLKIO_STAT_DEQUEUE, 0, 0); |
062a644d VG |
1307 | case BLKIO_PROP_avg_queue_size: |
1308 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1309 | BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0); |
062a644d VG |
1310 | case BLKIO_PROP_group_wait_time: |
1311 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1312 | BLKIO_STAT_GROUP_WAIT_TIME, 0, 0); |
062a644d VG |
1313 | case BLKIO_PROP_idle_time: |
1314 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1315 | BLKIO_STAT_IDLE_TIME, 0, 0); |
062a644d VG |
1316 | case BLKIO_PROP_empty_time: |
1317 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1318 | BLKIO_STAT_EMPTY_TIME, 0, 0); |
062a644d VG |
1319 | #endif |
1320 | default: | |
1321 | BUG(); | |
1322 | } | |
1323 | break; | |
4c9eefa1 VG |
1324 | case BLKIO_POLICY_THROTL: |
1325 | switch(name){ | |
1326 | case BLKIO_THROTL_io_service_bytes: | |
1327 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1328 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
4c9eefa1 VG |
1329 | case BLKIO_THROTL_io_serviced: |
1330 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1331 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
4c9eefa1 VG |
1332 | default: |
1333 | BUG(); | |
1334 | } | |
1335 | break; | |
062a644d VG |
1336 | default: |
1337 | BUG(); | |
1338 | } | |
1339 | ||
1340 | return 0; | |
1341 | } | |
1342 | ||
4bfd482e | 1343 | static int blkio_weight_write(struct blkio_cgroup *blkcg, int plid, u64 val) |
062a644d VG |
1344 | { |
1345 | struct blkio_group *blkg; | |
1346 | struct hlist_node *n; | |
062a644d VG |
1347 | |
1348 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) | |
1349 | return -EINVAL; | |
1350 | ||
1351 | spin_lock(&blkio_list_lock); | |
1352 | spin_lock_irq(&blkcg->lock); | |
1353 | blkcg->weight = (unsigned int)val; | |
1354 | ||
549d3aa8 | 1355 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
e8989fae | 1356 | struct blkg_policy_data *pd = blkg->pd[plid]; |
549d3aa8 | 1357 | |
e8989fae | 1358 | if (!pd->conf.weight) |
c1768268 | 1359 | blkio_update_group_weight(blkg, plid, blkcg->weight); |
549d3aa8 | 1360 | } |
062a644d | 1361 | |
062a644d VG |
1362 | spin_unlock_irq(&blkcg->lock); |
1363 | spin_unlock(&blkio_list_lock); | |
1364 | return 0; | |
1365 | } | |
1366 | ||
1367 | static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) { | |
1368 | struct blkio_cgroup *blkcg; | |
1369 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1370 | int name = BLKIOFILE_ATTR(cft->private); | |
1371 | ||
1372 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1373 | ||
1374 | switch(plid) { | |
1375 | case BLKIO_POLICY_PROP: | |
1376 | switch(name) { | |
1377 | case BLKIO_PROP_weight: | |
1378 | return (u64)blkcg->weight; | |
1379 | } | |
1380 | break; | |
1381 | default: | |
1382 | BUG(); | |
1383 | } | |
1384 | return 0; | |
1385 | } | |
1386 | ||
1387 | static int | |
1388 | blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) | |
1389 | { | |
1390 | struct blkio_cgroup *blkcg; | |
1391 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1392 | int name = BLKIOFILE_ATTR(cft->private); | |
1393 | ||
1394 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1395 | ||
1396 | switch(plid) { | |
1397 | case BLKIO_POLICY_PROP: | |
1398 | switch(name) { | |
1399 | case BLKIO_PROP_weight: | |
4bfd482e | 1400 | return blkio_weight_write(blkcg, plid, val); |
062a644d VG |
1401 | } |
1402 | break; | |
1403 | default: | |
1404 | BUG(); | |
1405 | } | |
34d0f179 | 1406 | |
34d0f179 GJ |
1407 | return 0; |
1408 | } | |
1409 | ||
31e4c28d | 1410 | struct cftype blkio_files[] = { |
34d0f179 GJ |
1411 | { |
1412 | .name = "weight_device", | |
062a644d VG |
1413 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1414 | BLKIO_PROP_weight_device), | |
1415 | .read_seq_string = blkiocg_file_read, | |
1416 | .write_string = blkiocg_file_write, | |
34d0f179 GJ |
1417 | .max_write_len = 256, |
1418 | }, | |
31e4c28d VG |
1419 | { |
1420 | .name = "weight", | |
062a644d VG |
1421 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1422 | BLKIO_PROP_weight), | |
1423 | .read_u64 = blkiocg_file_read_u64, | |
1424 | .write_u64 = blkiocg_file_write_u64, | |
31e4c28d | 1425 | }, |
22084190 VG |
1426 | { |
1427 | .name = "time", | |
062a644d VG |
1428 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1429 | BLKIO_PROP_time), | |
1430 | .read_map = blkiocg_file_read_map, | |
22084190 VG |
1431 | }, |
1432 | { | |
1433 | .name = "sectors", | |
062a644d VG |
1434 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1435 | BLKIO_PROP_sectors), | |
1436 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1437 | }, |
1438 | { | |
1439 | .name = "io_service_bytes", | |
062a644d VG |
1440 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1441 | BLKIO_PROP_io_service_bytes), | |
1442 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1443 | }, |
1444 | { | |
1445 | .name = "io_serviced", | |
062a644d VG |
1446 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1447 | BLKIO_PROP_io_serviced), | |
1448 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1449 | }, |
1450 | { | |
1451 | .name = "io_service_time", | |
062a644d VG |
1452 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1453 | BLKIO_PROP_io_service_time), | |
1454 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1455 | }, |
1456 | { | |
1457 | .name = "io_wait_time", | |
062a644d VG |
1458 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1459 | BLKIO_PROP_io_wait_time), | |
1460 | .read_map = blkiocg_file_read_map, | |
84c124da | 1461 | }, |
812d4026 DS |
1462 | { |
1463 | .name = "io_merged", | |
062a644d VG |
1464 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1465 | BLKIO_PROP_io_merged), | |
1466 | .read_map = blkiocg_file_read_map, | |
812d4026 | 1467 | }, |
cdc1184c DS |
1468 | { |
1469 | .name = "io_queued", | |
062a644d VG |
1470 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1471 | BLKIO_PROP_io_queued), | |
1472 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1473 | }, |
84c124da DS |
1474 | { |
1475 | .name = "reset_stats", | |
1476 | .write_u64 = blkiocg_reset_stats, | |
22084190 | 1477 | }, |
13f98250 VG |
1478 | #ifdef CONFIG_BLK_DEV_THROTTLING |
1479 | { | |
1480 | .name = "throttle.read_bps_device", | |
1481 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1482 | BLKIO_THROTL_read_bps_device), | |
1483 | .read_seq_string = blkiocg_file_read, | |
1484 | .write_string = blkiocg_file_write, | |
1485 | .max_write_len = 256, | |
1486 | }, | |
1487 | ||
1488 | { | |
1489 | .name = "throttle.write_bps_device", | |
1490 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1491 | BLKIO_THROTL_write_bps_device), | |
1492 | .read_seq_string = blkiocg_file_read, | |
1493 | .write_string = blkiocg_file_write, | |
1494 | .max_write_len = 256, | |
1495 | }, | |
1496 | ||
1497 | { | |
1498 | .name = "throttle.read_iops_device", | |
1499 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1500 | BLKIO_THROTL_read_iops_device), | |
1501 | .read_seq_string = blkiocg_file_read, | |
1502 | .write_string = blkiocg_file_write, | |
1503 | .max_write_len = 256, | |
1504 | }, | |
1505 | ||
1506 | { | |
1507 | .name = "throttle.write_iops_device", | |
1508 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1509 | BLKIO_THROTL_write_iops_device), | |
1510 | .read_seq_string = blkiocg_file_read, | |
1511 | .write_string = blkiocg_file_write, | |
1512 | .max_write_len = 256, | |
1513 | }, | |
1514 | { | |
1515 | .name = "throttle.io_service_bytes", | |
1516 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1517 | BLKIO_THROTL_io_service_bytes), | |
1518 | .read_map = blkiocg_file_read_map, | |
1519 | }, | |
1520 | { | |
1521 | .name = "throttle.io_serviced", | |
1522 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1523 | BLKIO_THROTL_io_serviced), | |
1524 | .read_map = blkiocg_file_read_map, | |
1525 | }, | |
1526 | #endif /* CONFIG_BLK_DEV_THROTTLING */ | |
1527 | ||
22084190 | 1528 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
cdc1184c DS |
1529 | { |
1530 | .name = "avg_queue_size", | |
062a644d VG |
1531 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1532 | BLKIO_PROP_avg_queue_size), | |
1533 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1534 | }, |
812df48d DS |
1535 | { |
1536 | .name = "group_wait_time", | |
062a644d VG |
1537 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1538 | BLKIO_PROP_group_wait_time), | |
1539 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1540 | }, |
1541 | { | |
1542 | .name = "idle_time", | |
062a644d VG |
1543 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1544 | BLKIO_PROP_idle_time), | |
1545 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1546 | }, |
1547 | { | |
1548 | .name = "empty_time", | |
062a644d VG |
1549 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1550 | BLKIO_PROP_empty_time), | |
1551 | .read_map = blkiocg_file_read_map, | |
812df48d | 1552 | }, |
cdc1184c | 1553 | { |
22084190 | 1554 | .name = "dequeue", |
062a644d VG |
1555 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1556 | BLKIO_PROP_dequeue), | |
1557 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1558 | }, |
9026e521 JT |
1559 | { |
1560 | .name = "unaccounted_time", | |
1561 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, | |
1562 | BLKIO_PROP_unaccounted_time), | |
1563 | .read_map = blkiocg_file_read_map, | |
1564 | }, | |
22084190 | 1565 | #endif |
31e4c28d VG |
1566 | }; |
1567 | ||
1568 | static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1569 | { | |
1570 | return cgroup_add_files(cgroup, subsys, blkio_files, | |
1571 | ARRAY_SIZE(blkio_files)); | |
1572 | } | |
1573 | ||
9f13ef67 TH |
1574 | /** |
1575 | * blkiocg_pre_destroy - cgroup pre_destroy callback | |
1576 | * @subsys: cgroup subsys | |
1577 | * @cgroup: cgroup of interest | |
1578 | * | |
1579 | * This function is called when @cgroup is about to go away and responsible | |
1580 | * for shooting down all blkgs associated with @cgroup. blkgs should be | |
1581 | * removed while holding both q and blkcg locks. As blkcg lock is nested | |
1582 | * inside q lock, this function performs reverse double lock dancing. | |
1583 | * | |
1584 | * This is the blkcg counterpart of ioc_release_fn(). | |
1585 | */ | |
7ee9c562 TH |
1586 | static int blkiocg_pre_destroy(struct cgroup_subsys *subsys, |
1587 | struct cgroup *cgroup) | |
31e4c28d VG |
1588 | { |
1589 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
b1c35769 | 1590 | |
9f13ef67 | 1591 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1592 | |
9f13ef67 TH |
1593 | while (!hlist_empty(&blkcg->blkg_list)) { |
1594 | struct blkio_group *blkg = hlist_entry(blkcg->blkg_list.first, | |
1595 | struct blkio_group, blkcg_node); | |
c875f4d0 | 1596 | struct request_queue *q = blkg->q; |
b1c35769 | 1597 | |
9f13ef67 TH |
1598 | if (spin_trylock(q->queue_lock)) { |
1599 | blkg_destroy(blkg); | |
1600 | spin_unlock(q->queue_lock); | |
1601 | } else { | |
1602 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 1603 | cpu_relax(); |
9f13ef67 | 1604 | spin_lock(&blkcg->lock); |
0f3942a3 | 1605 | } |
9f13ef67 | 1606 | } |
b1c35769 | 1607 | |
9f13ef67 | 1608 | spin_unlock_irq(&blkcg->lock); |
7ee9c562 TH |
1609 | return 0; |
1610 | } | |
1611 | ||
1612 | static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1613 | { | |
1614 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
1615 | ||
67523c48 BB |
1616 | if (blkcg != &blkio_root_cgroup) |
1617 | kfree(blkcg); | |
31e4c28d VG |
1618 | } |
1619 | ||
1620 | static struct cgroup_subsys_state * | |
1621 | blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1622 | { | |
9a9e8a26 | 1623 | static atomic64_t id_seq = ATOMIC64_INIT(0); |
0341509f LZ |
1624 | struct blkio_cgroup *blkcg; |
1625 | struct cgroup *parent = cgroup->parent; | |
31e4c28d | 1626 | |
0341509f | 1627 | if (!parent) { |
31e4c28d VG |
1628 | blkcg = &blkio_root_cgroup; |
1629 | goto done; | |
1630 | } | |
1631 | ||
31e4c28d VG |
1632 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
1633 | if (!blkcg) | |
1634 | return ERR_PTR(-ENOMEM); | |
1635 | ||
1636 | blkcg->weight = BLKIO_WEIGHT_DEFAULT; | |
9a9e8a26 | 1637 | blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */ |
31e4c28d VG |
1638 | done: |
1639 | spin_lock_init(&blkcg->lock); | |
1640 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
1641 | ||
1642 | return &blkcg->css; | |
1643 | } | |
1644 | ||
5efd6113 TH |
1645 | /** |
1646 | * blkcg_init_queue - initialize blkcg part of request queue | |
1647 | * @q: request_queue to initialize | |
1648 | * | |
1649 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1650 | * part of new request_queue @q. | |
1651 | * | |
1652 | * RETURNS: | |
1653 | * 0 on success, -errno on failure. | |
1654 | */ | |
1655 | int blkcg_init_queue(struct request_queue *q) | |
1656 | { | |
923adde1 TH |
1657 | int ret; |
1658 | ||
5efd6113 TH |
1659 | might_sleep(); |
1660 | ||
923adde1 TH |
1661 | ret = blk_throtl_init(q); |
1662 | if (ret) | |
1663 | return ret; | |
1664 | ||
1665 | mutex_lock(&all_q_mutex); | |
1666 | INIT_LIST_HEAD(&q->all_q_node); | |
1667 | list_add_tail(&q->all_q_node, &all_q_list); | |
1668 | mutex_unlock(&all_q_mutex); | |
1669 | ||
1670 | return 0; | |
5efd6113 TH |
1671 | } |
1672 | ||
1673 | /** | |
1674 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1675 | * @q: request_queue to drain | |
1676 | * | |
1677 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1678 | */ | |
1679 | void blkcg_drain_queue(struct request_queue *q) | |
1680 | { | |
1681 | lockdep_assert_held(q->queue_lock); | |
1682 | ||
1683 | blk_throtl_drain(q); | |
1684 | } | |
1685 | ||
1686 | /** | |
1687 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1688 | * @q: request_queue being released | |
1689 | * | |
1690 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1691 | */ | |
1692 | void blkcg_exit_queue(struct request_queue *q) | |
1693 | { | |
923adde1 TH |
1694 | mutex_lock(&all_q_mutex); |
1695 | list_del_init(&q->all_q_node); | |
1696 | mutex_unlock(&all_q_mutex); | |
1697 | ||
e8989fae TH |
1698 | blkg_destroy_all(q, true); |
1699 | ||
5efd6113 TH |
1700 | blk_throtl_exit(q); |
1701 | } | |
1702 | ||
31e4c28d VG |
1703 | /* |
1704 | * We cannot support shared io contexts, as we have no mean to support | |
1705 | * two tasks with the same ioc in two different groups without major rework | |
1706 | * of the main cic data structures. For now we allow a task to change | |
1707 | * its cgroup only if it's the only owner of its ioc. | |
1708 | */ | |
bb9d97b6 TH |
1709 | static int blkiocg_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
1710 | struct cgroup_taskset *tset) | |
31e4c28d | 1711 | { |
bb9d97b6 | 1712 | struct task_struct *task; |
31e4c28d VG |
1713 | struct io_context *ioc; |
1714 | int ret = 0; | |
1715 | ||
1716 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
bb9d97b6 TH |
1717 | cgroup_taskset_for_each(task, cgrp, tset) { |
1718 | task_lock(task); | |
1719 | ioc = task->io_context; | |
1720 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1721 | ret = -EINVAL; | |
1722 | task_unlock(task); | |
1723 | if (ret) | |
1724 | break; | |
1725 | } | |
31e4c28d VG |
1726 | return ret; |
1727 | } | |
1728 | ||
923adde1 TH |
1729 | static void blkcg_bypass_start(void) |
1730 | __acquires(&all_q_mutex) | |
1731 | { | |
1732 | struct request_queue *q; | |
1733 | ||
1734 | mutex_lock(&all_q_mutex); | |
1735 | ||
1736 | list_for_each_entry(q, &all_q_list, all_q_node) { | |
1737 | blk_queue_bypass_start(q); | |
e8989fae | 1738 | blkg_destroy_all(q, false); |
923adde1 TH |
1739 | } |
1740 | } | |
1741 | ||
1742 | static void blkcg_bypass_end(void) | |
1743 | __releases(&all_q_mutex) | |
1744 | { | |
1745 | struct request_queue *q; | |
1746 | ||
1747 | list_for_each_entry(q, &all_q_list, all_q_node) | |
1748 | blk_queue_bypass_end(q); | |
1749 | ||
1750 | mutex_unlock(&all_q_mutex); | |
1751 | } | |
1752 | ||
3e252066 VG |
1753 | void blkio_policy_register(struct blkio_policy_type *blkiop) |
1754 | { | |
e8989fae TH |
1755 | struct request_queue *q; |
1756 | ||
923adde1 | 1757 | blkcg_bypass_start(); |
3e252066 | 1758 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1759 | |
1760 | BUG_ON(blkio_policy[blkiop->plid]); | |
1761 | blkio_policy[blkiop->plid] = blkiop; | |
3e252066 | 1762 | list_add_tail(&blkiop->list, &blkio_list); |
035d10b2 | 1763 | |
3e252066 | 1764 | spin_unlock(&blkio_list_lock); |
e8989fae TH |
1765 | list_for_each_entry(q, &all_q_list, all_q_node) |
1766 | update_root_blkg_pd(q, blkiop->plid); | |
923adde1 | 1767 | blkcg_bypass_end(); |
3e252066 VG |
1768 | } |
1769 | EXPORT_SYMBOL_GPL(blkio_policy_register); | |
1770 | ||
1771 | void blkio_policy_unregister(struct blkio_policy_type *blkiop) | |
1772 | { | |
e8989fae TH |
1773 | struct request_queue *q; |
1774 | ||
923adde1 | 1775 | blkcg_bypass_start(); |
3e252066 | 1776 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1777 | |
1778 | BUG_ON(blkio_policy[blkiop->plid] != blkiop); | |
1779 | blkio_policy[blkiop->plid] = NULL; | |
3e252066 | 1780 | list_del_init(&blkiop->list); |
035d10b2 | 1781 | |
3e252066 | 1782 | spin_unlock(&blkio_list_lock); |
e8989fae TH |
1783 | list_for_each_entry(q, &all_q_list, all_q_node) |
1784 | update_root_blkg_pd(q, blkiop->plid); | |
923adde1 | 1785 | blkcg_bypass_end(); |
3e252066 VG |
1786 | } |
1787 | EXPORT_SYMBOL_GPL(blkio_policy_unregister); |