]>
Commit | Line | Data |
---|---|---|
31e4c28d VG |
1 | /* |
2 | * Common Block IO controller cgroup interface | |
3 | * | |
4 | * Based on ideas and code from CFQ, CFS and BFQ: | |
5 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> | |
6 | * | |
7 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> | |
8 | * Paolo Valente <paolo.valente@unimore.it> | |
9 | * | |
10 | * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> | |
11 | * Nauman Rafique <nauman@google.com> | |
12 | */ | |
13 | #include <linux/ioprio.h> | |
22084190 VG |
14 | #include <linux/seq_file.h> |
15 | #include <linux/kdev_t.h> | |
9d6a986c | 16 | #include <linux/module.h> |
accee785 | 17 | #include <linux/err.h> |
9195291e | 18 | #include <linux/blkdev.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
34d0f179 | 20 | #include <linux/genhd.h> |
72e06c25 TH |
21 | #include <linux/delay.h> |
22 | #include "blk-cgroup.h" | |
5efd6113 | 23 | #include "blk.h" |
3e252066 | 24 | |
84c124da DS |
25 | #define MAX_KEY_LEN 100 |
26 | ||
3e252066 VG |
27 | static DEFINE_SPINLOCK(blkio_list_lock); |
28 | static LIST_HEAD(blkio_list); | |
b1c35769 | 29 | |
923adde1 TH |
30 | static DEFINE_MUTEX(all_q_mutex); |
31 | static LIST_HEAD(all_q_list); | |
32 | ||
1cd9e039 VG |
33 | /* List of groups pending per cpu stats allocation */ |
34 | static DEFINE_SPINLOCK(alloc_list_lock); | |
35 | static LIST_HEAD(alloc_list); | |
36 | ||
37 | static void blkio_stat_alloc_fn(struct work_struct *); | |
38 | static DECLARE_DELAYED_WORK(blkio_stat_alloc_work, blkio_stat_alloc_fn); | |
39 | ||
31e4c28d | 40 | struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT }; |
9d6a986c VG |
41 | EXPORT_SYMBOL_GPL(blkio_root_cgroup); |
42 | ||
035d10b2 TH |
43 | static struct blkio_policy_type *blkio_policy[BLKIO_NR_POLICIES]; |
44 | ||
67523c48 BB |
45 | static struct cgroup_subsys_state *blkiocg_create(struct cgroup_subsys *, |
46 | struct cgroup *); | |
bb9d97b6 TH |
47 | static int blkiocg_can_attach(struct cgroup_subsys *, struct cgroup *, |
48 | struct cgroup_taskset *); | |
49 | static void blkiocg_attach(struct cgroup_subsys *, struct cgroup *, | |
50 | struct cgroup_taskset *); | |
7ee9c562 | 51 | static int blkiocg_pre_destroy(struct cgroup_subsys *, struct cgroup *); |
67523c48 BB |
52 | static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *); |
53 | static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *); | |
54 | ||
062a644d VG |
55 | /* for encoding cft->private value on file */ |
56 | #define BLKIOFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
57 | /* What policy owns the file, proportional or throttle */ | |
58 | #define BLKIOFILE_POLICY(val) (((val) >> 16) & 0xffff) | |
59 | #define BLKIOFILE_ATTR(val) ((val) & 0xffff) | |
60 | ||
67523c48 BB |
61 | struct cgroup_subsys blkio_subsys = { |
62 | .name = "blkio", | |
63 | .create = blkiocg_create, | |
bb9d97b6 TH |
64 | .can_attach = blkiocg_can_attach, |
65 | .attach = blkiocg_attach, | |
7ee9c562 | 66 | .pre_destroy = blkiocg_pre_destroy, |
67523c48 BB |
67 | .destroy = blkiocg_destroy, |
68 | .populate = blkiocg_populate, | |
67523c48 | 69 | .subsys_id = blkio_subsys_id, |
67523c48 BB |
70 | .module = THIS_MODULE, |
71 | }; | |
72 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
73 | ||
31e4c28d VG |
74 | struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup) |
75 | { | |
76 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), | |
77 | struct blkio_cgroup, css); | |
78 | } | |
9d6a986c | 79 | EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup); |
31e4c28d | 80 | |
4f85cb96 | 81 | static struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk) |
70087dc3 VG |
82 | { |
83 | return container_of(task_subsys_state(tsk, blkio_subsys_id), | |
84 | struct blkio_cgroup, css); | |
85 | } | |
4f85cb96 TH |
86 | |
87 | struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio) | |
88 | { | |
89 | if (bio && bio->bi_css) | |
90 | return container_of(bio->bi_css, struct blkio_cgroup, css); | |
91 | return task_blkio_cgroup(current); | |
92 | } | |
93 | EXPORT_SYMBOL_GPL(bio_blkio_cgroup); | |
70087dc3 | 94 | |
c1768268 TH |
95 | static inline void blkio_update_group_weight(struct blkio_group *blkg, |
96 | int plid, unsigned int weight) | |
062a644d VG |
97 | { |
98 | struct blkio_policy_type *blkiop; | |
99 | ||
100 | list_for_each_entry(blkiop, &blkio_list, list) { | |
101 | /* If this policy does not own the blkg, do not send updates */ | |
c1768268 | 102 | if (blkiop->plid != plid) |
062a644d VG |
103 | continue; |
104 | if (blkiop->ops.blkio_update_group_weight_fn) | |
ca32aefc | 105 | blkiop->ops.blkio_update_group_weight_fn(blkg->q, |
fe071437 | 106 | blkg, weight); |
062a644d VG |
107 | } |
108 | } | |
109 | ||
c1768268 TH |
110 | static inline void blkio_update_group_bps(struct blkio_group *blkg, int plid, |
111 | u64 bps, int fileid) | |
4c9eefa1 VG |
112 | { |
113 | struct blkio_policy_type *blkiop; | |
114 | ||
115 | list_for_each_entry(blkiop, &blkio_list, list) { | |
116 | ||
117 | /* If this policy does not own the blkg, do not send updates */ | |
c1768268 | 118 | if (blkiop->plid != plid) |
4c9eefa1 VG |
119 | continue; |
120 | ||
121 | if (fileid == BLKIO_THROTL_read_bps_device | |
122 | && blkiop->ops.blkio_update_group_read_bps_fn) | |
ca32aefc | 123 | blkiop->ops.blkio_update_group_read_bps_fn(blkg->q, |
fe071437 | 124 | blkg, bps); |
4c9eefa1 VG |
125 | |
126 | if (fileid == BLKIO_THROTL_write_bps_device | |
127 | && blkiop->ops.blkio_update_group_write_bps_fn) | |
ca32aefc | 128 | blkiop->ops.blkio_update_group_write_bps_fn(blkg->q, |
fe071437 | 129 | blkg, bps); |
4c9eefa1 VG |
130 | } |
131 | } | |
132 | ||
7702e8f4 | 133 | static inline void blkio_update_group_iops(struct blkio_group *blkg, |
c1768268 TH |
134 | int plid, unsigned int iops, |
135 | int fileid) | |
7702e8f4 VG |
136 | { |
137 | struct blkio_policy_type *blkiop; | |
138 | ||
139 | list_for_each_entry(blkiop, &blkio_list, list) { | |
140 | ||
141 | /* If this policy does not own the blkg, do not send updates */ | |
c1768268 | 142 | if (blkiop->plid != plid) |
7702e8f4 VG |
143 | continue; |
144 | ||
145 | if (fileid == BLKIO_THROTL_read_iops_device | |
146 | && blkiop->ops.blkio_update_group_read_iops_fn) | |
ca32aefc | 147 | blkiop->ops.blkio_update_group_read_iops_fn(blkg->q, |
fe071437 | 148 | blkg, iops); |
7702e8f4 VG |
149 | |
150 | if (fileid == BLKIO_THROTL_write_iops_device | |
151 | && blkiop->ops.blkio_update_group_write_iops_fn) | |
ca32aefc | 152 | blkiop->ops.blkio_update_group_write_iops_fn(blkg->q, |
fe071437 | 153 | blkg,iops); |
7702e8f4 VG |
154 | } |
155 | } | |
156 | ||
9195291e DS |
157 | /* |
158 | * Add to the appropriate stat variable depending on the request type. | |
159 | * This should be called with the blkg->stats_lock held. | |
160 | */ | |
84c124da DS |
161 | static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction, |
162 | bool sync) | |
9195291e | 163 | { |
84c124da DS |
164 | if (direction) |
165 | stat[BLKIO_STAT_WRITE] += add; | |
9195291e | 166 | else |
84c124da DS |
167 | stat[BLKIO_STAT_READ] += add; |
168 | if (sync) | |
169 | stat[BLKIO_STAT_SYNC] += add; | |
9195291e | 170 | else |
84c124da | 171 | stat[BLKIO_STAT_ASYNC] += add; |
9195291e DS |
172 | } |
173 | ||
cdc1184c DS |
174 | /* |
175 | * Decrements the appropriate stat variable if non-zero depending on the | |
176 | * request type. Panics on value being zero. | |
177 | * This should be called with the blkg->stats_lock held. | |
178 | */ | |
179 | static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync) | |
180 | { | |
181 | if (direction) { | |
182 | BUG_ON(stat[BLKIO_STAT_WRITE] == 0); | |
183 | stat[BLKIO_STAT_WRITE]--; | |
184 | } else { | |
185 | BUG_ON(stat[BLKIO_STAT_READ] == 0); | |
186 | stat[BLKIO_STAT_READ]--; | |
187 | } | |
188 | if (sync) { | |
189 | BUG_ON(stat[BLKIO_STAT_SYNC] == 0); | |
190 | stat[BLKIO_STAT_SYNC]--; | |
191 | } else { | |
192 | BUG_ON(stat[BLKIO_STAT_ASYNC] == 0); | |
193 | stat[BLKIO_STAT_ASYNC]--; | |
194 | } | |
195 | } | |
196 | ||
197 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
812df48d DS |
198 | /* This should be called with the blkg->stats_lock held. */ |
199 | static void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
c1768268 TH |
200 | struct blkio_policy_type *pol, |
201 | struct blkio_group *curr_blkg) | |
812df48d | 202 | { |
c1768268 | 203 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
549d3aa8 TH |
204 | |
205 | if (blkio_blkg_waiting(&pd->stats)) | |
812df48d DS |
206 | return; |
207 | if (blkg == curr_blkg) | |
208 | return; | |
549d3aa8 TH |
209 | pd->stats.start_group_wait_time = sched_clock(); |
210 | blkio_mark_blkg_waiting(&pd->stats); | |
812df48d DS |
211 | } |
212 | ||
213 | /* This should be called with the blkg->stats_lock held. */ | |
214 | static void blkio_update_group_wait_time(struct blkio_group_stats *stats) | |
215 | { | |
216 | unsigned long long now; | |
217 | ||
218 | if (!blkio_blkg_waiting(stats)) | |
219 | return; | |
220 | ||
221 | now = sched_clock(); | |
222 | if (time_after64(now, stats->start_group_wait_time)) | |
223 | stats->group_wait_time += now - stats->start_group_wait_time; | |
224 | blkio_clear_blkg_waiting(stats); | |
225 | } | |
226 | ||
227 | /* This should be called with the blkg->stats_lock held. */ | |
228 | static void blkio_end_empty_time(struct blkio_group_stats *stats) | |
229 | { | |
230 | unsigned long long now; | |
231 | ||
232 | if (!blkio_blkg_empty(stats)) | |
233 | return; | |
234 | ||
235 | now = sched_clock(); | |
236 | if (time_after64(now, stats->start_empty_time)) | |
237 | stats->empty_time += now - stats->start_empty_time; | |
238 | blkio_clear_blkg_empty(stats); | |
239 | } | |
240 | ||
c1768268 TH |
241 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg, |
242 | struct blkio_policy_type *pol) | |
812df48d | 243 | { |
c1768268 | 244 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
812df48d DS |
245 | unsigned long flags; |
246 | ||
247 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 TH |
248 | BUG_ON(blkio_blkg_idling(&pd->stats)); |
249 | pd->stats.start_idle_time = sched_clock(); | |
250 | blkio_mark_blkg_idling(&pd->stats); | |
812df48d DS |
251 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
252 | } | |
253 | EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats); | |
254 | ||
c1768268 TH |
255 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg, |
256 | struct blkio_policy_type *pol) | |
812df48d | 257 | { |
c1768268 | 258 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
812df48d DS |
259 | unsigned long flags; |
260 | unsigned long long now; | |
261 | struct blkio_group_stats *stats; | |
262 | ||
263 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 264 | stats = &pd->stats; |
812df48d DS |
265 | if (blkio_blkg_idling(stats)) { |
266 | now = sched_clock(); | |
267 | if (time_after64(now, stats->start_idle_time)) | |
268 | stats->idle_time += now - stats->start_idle_time; | |
269 | blkio_clear_blkg_idling(stats); | |
270 | } | |
271 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
272 | } | |
273 | EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats); | |
274 | ||
c1768268 TH |
275 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg, |
276 | struct blkio_policy_type *pol) | |
cdc1184c | 277 | { |
c1768268 | 278 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
cdc1184c DS |
279 | unsigned long flags; |
280 | struct blkio_group_stats *stats; | |
281 | ||
282 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 283 | stats = &pd->stats; |
cdc1184c DS |
284 | stats->avg_queue_size_sum += |
285 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] + | |
286 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]; | |
287 | stats->avg_queue_size_samples++; | |
812df48d | 288 | blkio_update_group_wait_time(stats); |
cdc1184c DS |
289 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
290 | } | |
a11cdaa7 DS |
291 | EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats); |
292 | ||
c1768268 TH |
293 | void blkiocg_set_start_empty_time(struct blkio_group *blkg, |
294 | struct blkio_policy_type *pol) | |
28baf442 | 295 | { |
c1768268 | 296 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
28baf442 DS |
297 | unsigned long flags; |
298 | struct blkio_group_stats *stats; | |
299 | ||
300 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 301 | stats = &pd->stats; |
28baf442 DS |
302 | |
303 | if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] || | |
304 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) { | |
305 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
306 | return; | |
307 | } | |
308 | ||
309 | /* | |
e5ff082e VG |
310 | * group is already marked empty. This can happen if cfqq got new |
311 | * request in parent group and moved to this group while being added | |
312 | * to service tree. Just ignore the event and move on. | |
28baf442 | 313 | */ |
e5ff082e VG |
314 | if(blkio_blkg_empty(stats)) { |
315 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
316 | return; | |
317 | } | |
318 | ||
28baf442 DS |
319 | stats->start_empty_time = sched_clock(); |
320 | blkio_mark_blkg_empty(stats); | |
321 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
322 | } | |
323 | EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time); | |
324 | ||
a11cdaa7 | 325 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
c1768268 TH |
326 | struct blkio_policy_type *pol, |
327 | unsigned long dequeue) | |
a11cdaa7 | 328 | { |
c1768268 | 329 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
549d3aa8 TH |
330 | |
331 | pd->stats.dequeue += dequeue; | |
a11cdaa7 DS |
332 | } |
333 | EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats); | |
812df48d DS |
334 | #else |
335 | static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
c1768268 TH |
336 | struct blkio_policy_type *pol, |
337 | struct blkio_group *curr_blkg) { } | |
338 | static inline void blkio_end_empty_time(struct blkio_group_stats *stats) { } | |
cdc1184c DS |
339 | #endif |
340 | ||
a11cdaa7 | 341 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
c1768268 TH |
342 | struct blkio_policy_type *pol, |
343 | struct blkio_group *curr_blkg, bool direction, | |
344 | bool sync) | |
cdc1184c | 345 | { |
c1768268 | 346 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
cdc1184c DS |
347 | unsigned long flags; |
348 | ||
349 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 350 | blkio_add_stat(pd->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction, |
cdc1184c | 351 | sync); |
549d3aa8 | 352 | blkio_end_empty_time(&pd->stats); |
c1768268 | 353 | blkio_set_start_group_wait_time(blkg, pol, curr_blkg); |
cdc1184c DS |
354 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
355 | } | |
a11cdaa7 | 356 | EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats); |
cdc1184c | 357 | |
a11cdaa7 | 358 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
c1768268 TH |
359 | struct blkio_policy_type *pol, |
360 | bool direction, bool sync) | |
cdc1184c | 361 | { |
c1768268 | 362 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
cdc1184c DS |
363 | unsigned long flags; |
364 | ||
365 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 366 | blkio_check_and_dec_stat(pd->stats.stat_arr[BLKIO_STAT_QUEUED], |
cdc1184c DS |
367 | direction, sync); |
368 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
369 | } | |
a11cdaa7 | 370 | EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats); |
cdc1184c | 371 | |
c1768268 TH |
372 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
373 | struct blkio_policy_type *pol, | |
374 | unsigned long time, | |
375 | unsigned long unaccounted_time) | |
22084190 | 376 | { |
c1768268 | 377 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
303a3acb DS |
378 | unsigned long flags; |
379 | ||
380 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 381 | pd->stats.time += time; |
a23e6869 | 382 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
549d3aa8 | 383 | pd->stats.unaccounted_time += unaccounted_time; |
a23e6869 | 384 | #endif |
303a3acb | 385 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
22084190 | 386 | } |
303a3acb | 387 | EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used); |
22084190 | 388 | |
5624a4e4 VG |
389 | /* |
390 | * should be called under rcu read lock or queue lock to make sure blkg pointer | |
391 | * is valid. | |
392 | */ | |
84c124da | 393 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
c1768268 TH |
394 | struct blkio_policy_type *pol, |
395 | uint64_t bytes, bool direction, bool sync) | |
9195291e | 396 | { |
c1768268 | 397 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
5624a4e4 | 398 | struct blkio_group_stats_cpu *stats_cpu; |
575969a0 VG |
399 | unsigned long flags; |
400 | ||
1cd9e039 VG |
401 | /* If per cpu stats are not allocated yet, don't do any accounting. */ |
402 | if (pd->stats_cpu == NULL) | |
403 | return; | |
404 | ||
575969a0 VG |
405 | /* |
406 | * Disabling interrupts to provide mutual exclusion between two | |
407 | * writes on same cpu. It probably is not needed for 64bit. Not | |
408 | * optimizing that case yet. | |
409 | */ | |
410 | local_irq_save(flags); | |
9195291e | 411 | |
549d3aa8 | 412 | stats_cpu = this_cpu_ptr(pd->stats_cpu); |
5624a4e4 | 413 | |
575969a0 | 414 | u64_stats_update_begin(&stats_cpu->syncp); |
5624a4e4 VG |
415 | stats_cpu->sectors += bytes >> 9; |
416 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED], | |
417 | 1, direction, sync); | |
418 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES], | |
419 | bytes, direction, sync); | |
575969a0 VG |
420 | u64_stats_update_end(&stats_cpu->syncp); |
421 | local_irq_restore(flags); | |
9195291e | 422 | } |
84c124da | 423 | EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats); |
9195291e | 424 | |
84c124da | 425 | void blkiocg_update_completion_stats(struct blkio_group *blkg, |
c1768268 TH |
426 | struct blkio_policy_type *pol, |
427 | uint64_t start_time, | |
428 | uint64_t io_start_time, bool direction, | |
429 | bool sync) | |
9195291e | 430 | { |
c1768268 | 431 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
9195291e DS |
432 | struct blkio_group_stats *stats; |
433 | unsigned long flags; | |
434 | unsigned long long now = sched_clock(); | |
435 | ||
436 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 437 | stats = &pd->stats; |
84c124da DS |
438 | if (time_after64(now, io_start_time)) |
439 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME], | |
440 | now - io_start_time, direction, sync); | |
441 | if (time_after64(io_start_time, start_time)) | |
442 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME], | |
443 | io_start_time - start_time, direction, sync); | |
9195291e DS |
444 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
445 | } | |
84c124da | 446 | EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats); |
9195291e | 447 | |
317389a7 | 448 | /* Merged stats are per cpu. */ |
c1768268 TH |
449 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, |
450 | struct blkio_policy_type *pol, | |
451 | bool direction, bool sync) | |
812d4026 | 452 | { |
c1768268 | 453 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
5fe224d2 | 454 | struct blkio_group_stats *stats; |
812d4026 DS |
455 | unsigned long flags; |
456 | ||
5fe224d2 TH |
457 | spin_lock_irqsave(&blkg->stats_lock, flags); |
458 | stats = &pd->stats; | |
459 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_MERGED], 1, direction, sync); | |
460 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
812d4026 DS |
461 | } |
462 | EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats); | |
463 | ||
1cd9e039 VG |
464 | /* |
465 | * Worker for allocating per cpu stat for blk groups. This is scheduled on | |
466 | * the system_nrt_wq once there are some groups on the alloc_list waiting | |
467 | * for allocation. | |
468 | */ | |
469 | static void blkio_stat_alloc_fn(struct work_struct *work) | |
470 | { | |
471 | static void *pcpu_stats[BLKIO_NR_POLICIES]; | |
472 | struct delayed_work *dwork = to_delayed_work(work); | |
473 | struct blkio_group *blkg; | |
474 | int i; | |
475 | bool empty = false; | |
476 | ||
477 | alloc_stats: | |
478 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { | |
479 | if (pcpu_stats[i] != NULL) | |
480 | continue; | |
481 | ||
482 | pcpu_stats[i] = alloc_percpu(struct blkio_group_stats_cpu); | |
483 | ||
484 | /* Allocation failed. Try again after some time. */ | |
485 | if (pcpu_stats[i] == NULL) { | |
486 | queue_delayed_work(system_nrt_wq, dwork, | |
487 | msecs_to_jiffies(10)); | |
488 | return; | |
489 | } | |
490 | } | |
491 | ||
492 | spin_lock_irq(&blkio_list_lock); | |
493 | spin_lock(&alloc_list_lock); | |
494 | ||
495 | /* cgroup got deleted or queue exited. */ | |
496 | if (!list_empty(&alloc_list)) { | |
497 | blkg = list_first_entry(&alloc_list, struct blkio_group, | |
498 | alloc_node); | |
499 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { | |
500 | struct blkg_policy_data *pd = blkg->pd[i]; | |
501 | ||
502 | if (blkio_policy[i] && pd && !pd->stats_cpu) | |
503 | swap(pd->stats_cpu, pcpu_stats[i]); | |
504 | } | |
505 | ||
506 | list_del_init(&blkg->alloc_node); | |
507 | } | |
508 | ||
509 | empty = list_empty(&alloc_list); | |
510 | ||
511 | spin_unlock(&alloc_list_lock); | |
512 | spin_unlock_irq(&blkio_list_lock); | |
513 | ||
514 | if (!empty) | |
515 | goto alloc_stats; | |
516 | } | |
517 | ||
0381411e TH |
518 | /** |
519 | * blkg_free - free a blkg | |
520 | * @blkg: blkg to free | |
521 | * | |
522 | * Free @blkg which may be partially allocated. | |
523 | */ | |
524 | static void blkg_free(struct blkio_group *blkg) | |
525 | { | |
e8989fae | 526 | int i; |
549d3aa8 TH |
527 | |
528 | if (!blkg) | |
529 | return; | |
530 | ||
e8989fae TH |
531 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
532 | struct blkg_policy_data *pd = blkg->pd[i]; | |
533 | ||
534 | if (pd) { | |
535 | free_percpu(pd->stats_cpu); | |
536 | kfree(pd); | |
537 | } | |
0381411e | 538 | } |
e8989fae | 539 | |
549d3aa8 | 540 | kfree(blkg); |
0381411e TH |
541 | } |
542 | ||
543 | /** | |
544 | * blkg_alloc - allocate a blkg | |
545 | * @blkcg: block cgroup the new blkg is associated with | |
546 | * @q: request_queue the new blkg is associated with | |
0381411e | 547 | * |
e8989fae | 548 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e TH |
549 | */ |
550 | static struct blkio_group *blkg_alloc(struct blkio_cgroup *blkcg, | |
e8989fae | 551 | struct request_queue *q) |
0381411e TH |
552 | { |
553 | struct blkio_group *blkg; | |
e8989fae | 554 | int i; |
0381411e TH |
555 | |
556 | /* alloc and init base part */ | |
557 | blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node); | |
558 | if (!blkg) | |
559 | return NULL; | |
560 | ||
561 | spin_lock_init(&blkg->stats_lock); | |
c875f4d0 | 562 | blkg->q = q; |
e8989fae | 563 | INIT_LIST_HEAD(&blkg->q_node); |
1cd9e039 | 564 | INIT_LIST_HEAD(&blkg->alloc_node); |
0381411e | 565 | blkg->blkcg = blkcg; |
1adaf3dd | 566 | blkg->refcnt = 1; |
0381411e TH |
567 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); |
568 | ||
e8989fae TH |
569 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
570 | struct blkio_policy_type *pol = blkio_policy[i]; | |
571 | struct blkg_policy_data *pd; | |
0381411e | 572 | |
e8989fae TH |
573 | if (!pol) |
574 | continue; | |
575 | ||
576 | /* alloc per-policy data and attach it to blkg */ | |
577 | pd = kzalloc_node(sizeof(*pd) + pol->pdata_size, GFP_ATOMIC, | |
578 | q->node); | |
579 | if (!pd) { | |
580 | blkg_free(blkg); | |
581 | return NULL; | |
582 | } | |
549d3aa8 | 583 | |
e8989fae TH |
584 | blkg->pd[i] = pd; |
585 | pd->blkg = blkg; | |
0381411e TH |
586 | } |
587 | ||
549d3aa8 | 588 | /* invoke per-policy init */ |
e8989fae TH |
589 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
590 | struct blkio_policy_type *pol = blkio_policy[i]; | |
591 | ||
592 | if (pol) | |
593 | pol->ops.blkio_init_group_fn(blkg); | |
594 | } | |
595 | ||
0381411e TH |
596 | return blkg; |
597 | } | |
598 | ||
cd1604fa TH |
599 | struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg, |
600 | struct request_queue *q, | |
601 | enum blkio_policy_id plid, | |
602 | bool for_root) | |
603 | __releases(q->queue_lock) __acquires(q->queue_lock) | |
5624a4e4 | 604 | { |
1cd9e039 | 605 | struct blkio_group *blkg; |
5624a4e4 | 606 | |
cd1604fa TH |
607 | WARN_ON_ONCE(!rcu_read_lock_held()); |
608 | lockdep_assert_held(q->queue_lock); | |
609 | ||
610 | /* | |
611 | * This could be the first entry point of blkcg implementation and | |
612 | * we shouldn't allow anything to go through for a bypassing queue. | |
613 | * The following can be removed if blkg lookup is guaranteed to | |
614 | * fail on a bypassing queue. | |
615 | */ | |
616 | if (unlikely(blk_queue_bypass(q)) && !for_root) | |
617 | return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY); | |
618 | ||
e8989fae | 619 | blkg = blkg_lookup(blkcg, q); |
cd1604fa TH |
620 | if (blkg) |
621 | return blkg; | |
622 | ||
7ee9c562 | 623 | /* blkg holds a reference to blkcg */ |
cd1604fa TH |
624 | if (!css_tryget(&blkcg->css)) |
625 | return ERR_PTR(-EINVAL); | |
626 | ||
627 | /* | |
628 | * Allocate and initialize. | |
cd1604fa | 629 | */ |
1cd9e039 | 630 | blkg = blkg_alloc(blkcg, q); |
cd1604fa TH |
631 | |
632 | /* did alloc fail? */ | |
1cd9e039 | 633 | if (unlikely(!blkg)) { |
cd1604fa TH |
634 | blkg = ERR_PTR(-ENOMEM); |
635 | goto out; | |
636 | } | |
637 | ||
638 | /* insert */ | |
639 | spin_lock(&blkcg->lock); | |
31e4c28d | 640 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
e8989fae | 641 | list_add(&blkg->q_node, &q->blkg_list); |
cd1604fa | 642 | spin_unlock(&blkcg->lock); |
1cd9e039 VG |
643 | |
644 | spin_lock(&alloc_list_lock); | |
645 | list_add(&blkg->alloc_node, &alloc_list); | |
646 | /* Queue per cpu stat allocation from worker thread. */ | |
647 | queue_delayed_work(system_nrt_wq, &blkio_stat_alloc_work, 0); | |
648 | spin_unlock(&alloc_list_lock); | |
cd1604fa | 649 | out: |
cd1604fa | 650 | return blkg; |
31e4c28d | 651 | } |
cd1604fa | 652 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 653 | |
31e4c28d | 654 | /* called under rcu_read_lock(). */ |
cd1604fa | 655 | struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg, |
e8989fae | 656 | struct request_queue *q) |
31e4c28d VG |
657 | { |
658 | struct blkio_group *blkg; | |
659 | struct hlist_node *n; | |
31e4c28d | 660 | |
ca32aefc | 661 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) |
e8989fae | 662 | if (blkg->q == q) |
31e4c28d | 663 | return blkg; |
31e4c28d VG |
664 | return NULL; |
665 | } | |
cd1604fa | 666 | EXPORT_SYMBOL_GPL(blkg_lookup); |
31e4c28d | 667 | |
e8989fae | 668 | static void blkg_destroy(struct blkio_group *blkg) |
03aa264a TH |
669 | { |
670 | struct request_queue *q = blkg->q; | |
9f13ef67 | 671 | struct blkio_cgroup *blkcg = blkg->blkcg; |
03aa264a TH |
672 | |
673 | lockdep_assert_held(q->queue_lock); | |
9f13ef67 | 674 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
675 | |
676 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 677 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 678 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
e8989fae | 679 | list_del_init(&blkg->q_node); |
9f13ef67 | 680 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 681 | |
1cd9e039 VG |
682 | spin_lock(&alloc_list_lock); |
683 | list_del_init(&blkg->alloc_node); | |
684 | spin_unlock(&alloc_list_lock); | |
685 | ||
03aa264a TH |
686 | /* |
687 | * Put the reference taken at the time of creation so that when all | |
688 | * queues are gone, group can be destroyed. | |
689 | */ | |
690 | blkg_put(blkg); | |
691 | } | |
692 | ||
e8989fae TH |
693 | /* |
694 | * XXX: This updates blkg policy data in-place for root blkg, which is | |
695 | * necessary across elevator switch and policy registration as root blkgs | |
696 | * aren't shot down. This broken and racy implementation is temporary. | |
697 | * Eventually, blkg shoot down will be replaced by proper in-place update. | |
698 | */ | |
699 | void update_root_blkg_pd(struct request_queue *q, enum blkio_policy_id plid) | |
700 | { | |
701 | struct blkio_policy_type *pol = blkio_policy[plid]; | |
702 | struct blkio_group *blkg = blkg_lookup(&blkio_root_cgroup, q); | |
703 | struct blkg_policy_data *pd; | |
704 | ||
705 | if (!blkg) | |
706 | return; | |
707 | ||
708 | kfree(blkg->pd[plid]); | |
709 | blkg->pd[plid] = NULL; | |
710 | ||
711 | if (!pol) | |
712 | return; | |
713 | ||
714 | pd = kzalloc(sizeof(*pd) + pol->pdata_size, GFP_KERNEL); | |
715 | WARN_ON_ONCE(!pd); | |
716 | ||
717 | pd->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu); | |
718 | WARN_ON_ONCE(!pd->stats_cpu); | |
719 | ||
720 | blkg->pd[plid] = pd; | |
721 | pd->blkg = blkg; | |
722 | pol->ops.blkio_init_group_fn(blkg); | |
723 | } | |
724 | EXPORT_SYMBOL_GPL(update_root_blkg_pd); | |
725 | ||
9f13ef67 TH |
726 | /** |
727 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
728 | * @q: request_queue of interest | |
729 | * @destroy_root: whether to destroy root blkg or not | |
730 | * | |
731 | * Destroy blkgs associated with @q. If @destroy_root is %true, all are | |
732 | * destroyed; otherwise, root blkg is left alone. | |
733 | */ | |
e8989fae | 734 | void blkg_destroy_all(struct request_queue *q, bool destroy_root) |
72e06c25 | 735 | { |
03aa264a | 736 | struct blkio_group *blkg, *n; |
72e06c25 | 737 | |
9f13ef67 | 738 | spin_lock_irq(q->queue_lock); |
72e06c25 | 739 | |
9f13ef67 TH |
740 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
741 | struct blkio_cgroup *blkcg = blkg->blkcg; | |
72e06c25 | 742 | |
9f13ef67 TH |
743 | /* skip root? */ |
744 | if (!destroy_root && blkg->blkcg == &blkio_root_cgroup) | |
745 | continue; | |
72e06c25 | 746 | |
9f13ef67 TH |
747 | spin_lock(&blkcg->lock); |
748 | blkg_destroy(blkg); | |
749 | spin_unlock(&blkcg->lock); | |
72e06c25 | 750 | } |
9f13ef67 TH |
751 | |
752 | spin_unlock_irq(q->queue_lock); | |
72e06c25 | 753 | } |
03aa264a | 754 | EXPORT_SYMBOL_GPL(blkg_destroy_all); |
72e06c25 | 755 | |
1adaf3dd TH |
756 | static void blkg_rcu_free(struct rcu_head *rcu_head) |
757 | { | |
758 | blkg_free(container_of(rcu_head, struct blkio_group, rcu_head)); | |
759 | } | |
760 | ||
761 | void __blkg_release(struct blkio_group *blkg) | |
762 | { | |
763 | /* release the extra blkcg reference this blkg has been holding */ | |
764 | css_put(&blkg->blkcg->css); | |
765 | ||
766 | /* | |
767 | * A group is freed in rcu manner. But having an rcu lock does not | |
768 | * mean that one can access all the fields of blkg and assume these | |
769 | * are valid. For example, don't try to follow throtl_data and | |
770 | * request queue links. | |
771 | * | |
772 | * Having a reference to blkg under an rcu allows acess to only | |
773 | * values local to groups like group stats and group rate limits | |
774 | */ | |
775 | call_rcu(&blkg->rcu_head, blkg_rcu_free); | |
776 | } | |
777 | EXPORT_SYMBOL_GPL(__blkg_release); | |
778 | ||
c1768268 | 779 | static void blkio_reset_stats_cpu(struct blkio_group *blkg, int plid) |
f0bdc8cd | 780 | { |
c1768268 | 781 | struct blkg_policy_data *pd = blkg->pd[plid]; |
f0bdc8cd VG |
782 | struct blkio_group_stats_cpu *stats_cpu; |
783 | int i, j, k; | |
1cd9e039 VG |
784 | |
785 | if (pd->stats_cpu == NULL) | |
786 | return; | |
f0bdc8cd VG |
787 | /* |
788 | * Note: On 64 bit arch this should not be an issue. This has the | |
789 | * possibility of returning some inconsistent value on 32bit arch | |
790 | * as 64bit update on 32bit is non atomic. Taking care of this | |
791 | * corner case makes code very complicated, like sending IPIs to | |
792 | * cpus, taking care of stats of offline cpus etc. | |
793 | * | |
794 | * reset stats is anyway more of a debug feature and this sounds a | |
795 | * corner case. So I am not complicating the code yet until and | |
796 | * unless this becomes a real issue. | |
797 | */ | |
798 | for_each_possible_cpu(i) { | |
549d3aa8 | 799 | stats_cpu = per_cpu_ptr(pd->stats_cpu, i); |
f0bdc8cd VG |
800 | stats_cpu->sectors = 0; |
801 | for(j = 0; j < BLKIO_STAT_CPU_NR; j++) | |
802 | for (k = 0; k < BLKIO_STAT_TOTAL; k++) | |
803 | stats_cpu->stat_arr_cpu[j][k] = 0; | |
804 | } | |
805 | } | |
806 | ||
303a3acb | 807 | static int |
84c124da | 808 | blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val) |
303a3acb DS |
809 | { |
810 | struct blkio_cgroup *blkcg; | |
811 | struct blkio_group *blkg; | |
812df48d | 812 | struct blkio_group_stats *stats; |
303a3acb | 813 | struct hlist_node *n; |
cdc1184c DS |
814 | uint64_t queued[BLKIO_STAT_TOTAL]; |
815 | int i; | |
812df48d DS |
816 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
817 | bool idling, waiting, empty; | |
818 | unsigned long long now = sched_clock(); | |
819 | #endif | |
303a3acb DS |
820 | |
821 | blkcg = cgroup_to_blkio_cgroup(cgroup); | |
e8989fae | 822 | spin_lock(&blkio_list_lock); |
303a3acb DS |
823 | spin_lock_irq(&blkcg->lock); |
824 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
e8989fae | 825 | struct blkio_policy_type *pol; |
549d3aa8 | 826 | |
e8989fae TH |
827 | list_for_each_entry(pol, &blkio_list, list) { |
828 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; | |
829 | ||
830 | spin_lock(&blkg->stats_lock); | |
831 | stats = &pd->stats; | |
812df48d | 832 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
e8989fae TH |
833 | idling = blkio_blkg_idling(stats); |
834 | waiting = blkio_blkg_waiting(stats); | |
835 | empty = blkio_blkg_empty(stats); | |
812df48d | 836 | #endif |
e8989fae TH |
837 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
838 | queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i]; | |
839 | memset(stats, 0, sizeof(struct blkio_group_stats)); | |
840 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) | |
841 | stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i]; | |
812df48d | 842 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
e8989fae TH |
843 | if (idling) { |
844 | blkio_mark_blkg_idling(stats); | |
845 | stats->start_idle_time = now; | |
846 | } | |
847 | if (waiting) { | |
848 | blkio_mark_blkg_waiting(stats); | |
849 | stats->start_group_wait_time = now; | |
850 | } | |
851 | if (empty) { | |
852 | blkio_mark_blkg_empty(stats); | |
853 | stats->start_empty_time = now; | |
854 | } | |
812df48d | 855 | #endif |
e8989fae | 856 | spin_unlock(&blkg->stats_lock); |
f0bdc8cd | 857 | |
e8989fae TH |
858 | /* Reset Per cpu stats which don't take blkg->stats_lock */ |
859 | blkio_reset_stats_cpu(blkg, pol->plid); | |
860 | } | |
303a3acb | 861 | } |
f0bdc8cd | 862 | |
303a3acb | 863 | spin_unlock_irq(&blkcg->lock); |
e8989fae | 864 | spin_unlock(&blkio_list_lock); |
303a3acb DS |
865 | return 0; |
866 | } | |
867 | ||
7a4dd281 TH |
868 | static void blkio_get_key_name(enum stat_sub_type type, const char *dname, |
869 | char *str, int chars_left, bool diskname_only) | |
303a3acb | 870 | { |
7a4dd281 | 871 | snprintf(str, chars_left, "%s", dname); |
303a3acb DS |
872 | chars_left -= strlen(str); |
873 | if (chars_left <= 0) { | |
874 | printk(KERN_WARNING | |
875 | "Possibly incorrect cgroup stat display format"); | |
876 | return; | |
877 | } | |
84c124da DS |
878 | if (diskname_only) |
879 | return; | |
303a3acb | 880 | switch (type) { |
84c124da | 881 | case BLKIO_STAT_READ: |
303a3acb DS |
882 | strlcat(str, " Read", chars_left); |
883 | break; | |
84c124da | 884 | case BLKIO_STAT_WRITE: |
303a3acb DS |
885 | strlcat(str, " Write", chars_left); |
886 | break; | |
84c124da | 887 | case BLKIO_STAT_SYNC: |
303a3acb DS |
888 | strlcat(str, " Sync", chars_left); |
889 | break; | |
84c124da | 890 | case BLKIO_STAT_ASYNC: |
303a3acb DS |
891 | strlcat(str, " Async", chars_left); |
892 | break; | |
84c124da | 893 | case BLKIO_STAT_TOTAL: |
303a3acb DS |
894 | strlcat(str, " Total", chars_left); |
895 | break; | |
896 | default: | |
897 | strlcat(str, " Invalid", chars_left); | |
898 | } | |
899 | } | |
900 | ||
84c124da | 901 | static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val, |
7a4dd281 | 902 | struct cgroup_map_cb *cb, const char *dname) |
84c124da | 903 | { |
7a4dd281 | 904 | blkio_get_key_name(0, dname, str, chars_left, true); |
84c124da DS |
905 | cb->fill(cb, str, val); |
906 | return val; | |
907 | } | |
303a3acb | 908 | |
5624a4e4 | 909 | |
c1768268 | 910 | static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg, int plid, |
5624a4e4 VG |
911 | enum stat_type_cpu type, enum stat_sub_type sub_type) |
912 | { | |
c1768268 | 913 | struct blkg_policy_data *pd = blkg->pd[plid]; |
5624a4e4 VG |
914 | int cpu; |
915 | struct blkio_group_stats_cpu *stats_cpu; | |
575969a0 | 916 | u64 val = 0, tval; |
5624a4e4 | 917 | |
1cd9e039 VG |
918 | if (pd->stats_cpu == NULL) |
919 | return val; | |
920 | ||
5624a4e4 | 921 | for_each_possible_cpu(cpu) { |
575969a0 | 922 | unsigned int start; |
549d3aa8 | 923 | stats_cpu = per_cpu_ptr(pd->stats_cpu, cpu); |
5624a4e4 | 924 | |
575969a0 VG |
925 | do { |
926 | start = u64_stats_fetch_begin(&stats_cpu->syncp); | |
927 | if (type == BLKIO_STAT_CPU_SECTORS) | |
928 | tval = stats_cpu->sectors; | |
929 | else | |
930 | tval = stats_cpu->stat_arr_cpu[type][sub_type]; | |
931 | } while(u64_stats_fetch_retry(&stats_cpu->syncp, start)); | |
932 | ||
933 | val += tval; | |
5624a4e4 VG |
934 | } |
935 | ||
936 | return val; | |
937 | } | |
938 | ||
c1768268 | 939 | static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg, int plid, |
7a4dd281 TH |
940 | struct cgroup_map_cb *cb, const char *dname, |
941 | enum stat_type_cpu type) | |
5624a4e4 VG |
942 | { |
943 | uint64_t disk_total, val; | |
944 | char key_str[MAX_KEY_LEN]; | |
945 | enum stat_sub_type sub_type; | |
946 | ||
947 | if (type == BLKIO_STAT_CPU_SECTORS) { | |
c1768268 | 948 | val = blkio_read_stat_cpu(blkg, plid, type, 0); |
7a4dd281 TH |
949 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, |
950 | dname); | |
5624a4e4 VG |
951 | } |
952 | ||
953 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; | |
954 | sub_type++) { | |
7a4dd281 TH |
955 | blkio_get_key_name(sub_type, dname, key_str, MAX_KEY_LEN, |
956 | false); | |
c1768268 | 957 | val = blkio_read_stat_cpu(blkg, plid, type, sub_type); |
5624a4e4 VG |
958 | cb->fill(cb, key_str, val); |
959 | } | |
960 | ||
c1768268 TH |
961 | disk_total = blkio_read_stat_cpu(blkg, plid, type, BLKIO_STAT_READ) + |
962 | blkio_read_stat_cpu(blkg, plid, type, BLKIO_STAT_WRITE); | |
5624a4e4 | 963 | |
7a4dd281 TH |
964 | blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN, |
965 | false); | |
5624a4e4 VG |
966 | cb->fill(cb, key_str, disk_total); |
967 | return disk_total; | |
968 | } | |
969 | ||
84c124da | 970 | /* This should be called with blkg->stats_lock held */ |
c1768268 | 971 | static uint64_t blkio_get_stat(struct blkio_group *blkg, int plid, |
7a4dd281 TH |
972 | struct cgroup_map_cb *cb, const char *dname, |
973 | enum stat_type type) | |
303a3acb | 974 | { |
c1768268 | 975 | struct blkg_policy_data *pd = blkg->pd[plid]; |
303a3acb DS |
976 | uint64_t disk_total; |
977 | char key_str[MAX_KEY_LEN]; | |
84c124da DS |
978 | enum stat_sub_type sub_type; |
979 | ||
980 | if (type == BLKIO_STAT_TIME) | |
981 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 982 | pd->stats.time, cb, dname); |
9026e521 | 983 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 JT |
984 | if (type == BLKIO_STAT_UNACCOUNTED_TIME) |
985 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 986 | pd->stats.unaccounted_time, cb, dname); |
cdc1184c | 987 | if (type == BLKIO_STAT_AVG_QUEUE_SIZE) { |
549d3aa8 TH |
988 | uint64_t sum = pd->stats.avg_queue_size_sum; |
989 | uint64_t samples = pd->stats.avg_queue_size_samples; | |
cdc1184c DS |
990 | if (samples) |
991 | do_div(sum, samples); | |
992 | else | |
993 | sum = 0; | |
7a4dd281 TH |
994 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, |
995 | sum, cb, dname); | |
cdc1184c | 996 | } |
812df48d DS |
997 | if (type == BLKIO_STAT_GROUP_WAIT_TIME) |
998 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 999 | pd->stats.group_wait_time, cb, dname); |
812df48d DS |
1000 | if (type == BLKIO_STAT_IDLE_TIME) |
1001 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 1002 | pd->stats.idle_time, cb, dname); |
812df48d DS |
1003 | if (type == BLKIO_STAT_EMPTY_TIME) |
1004 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 1005 | pd->stats.empty_time, cb, dname); |
84c124da DS |
1006 | if (type == BLKIO_STAT_DEQUEUE) |
1007 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 1008 | pd->stats.dequeue, cb, dname); |
84c124da | 1009 | #endif |
303a3acb | 1010 | |
84c124da DS |
1011 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; |
1012 | sub_type++) { | |
7a4dd281 TH |
1013 | blkio_get_key_name(sub_type, dname, key_str, MAX_KEY_LEN, |
1014 | false); | |
549d3aa8 | 1015 | cb->fill(cb, key_str, pd->stats.stat_arr[type][sub_type]); |
303a3acb | 1016 | } |
549d3aa8 TH |
1017 | disk_total = pd->stats.stat_arr[type][BLKIO_STAT_READ] + |
1018 | pd->stats.stat_arr[type][BLKIO_STAT_WRITE]; | |
7a4dd281 TH |
1019 | blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN, |
1020 | false); | |
303a3acb DS |
1021 | cb->fill(cb, key_str, disk_total); |
1022 | return disk_total; | |
1023 | } | |
1024 | ||
4bfd482e TH |
1025 | static int blkio_policy_parse_and_set(char *buf, enum blkio_policy_id plid, |
1026 | int fileid, struct blkio_cgroup *blkcg) | |
34d0f179 | 1027 | { |
ece84241 | 1028 | struct gendisk *disk = NULL; |
e56da7e2 | 1029 | struct blkio_group *blkg = NULL; |
549d3aa8 | 1030 | struct blkg_policy_data *pd; |
34d0f179 | 1031 | char *s[4], *p, *major_s = NULL, *minor_s = NULL; |
d11bb446 | 1032 | unsigned long major, minor; |
ece84241 TH |
1033 | int i = 0, ret = -EINVAL; |
1034 | int part; | |
34d0f179 | 1035 | dev_t dev; |
d11bb446 | 1036 | u64 temp; |
34d0f179 GJ |
1037 | |
1038 | memset(s, 0, sizeof(s)); | |
1039 | ||
1040 | while ((p = strsep(&buf, " ")) != NULL) { | |
1041 | if (!*p) | |
1042 | continue; | |
1043 | ||
1044 | s[i++] = p; | |
1045 | ||
1046 | /* Prevent from inputing too many things */ | |
1047 | if (i == 3) | |
1048 | break; | |
1049 | } | |
1050 | ||
1051 | if (i != 2) | |
ece84241 | 1052 | goto out; |
34d0f179 GJ |
1053 | |
1054 | p = strsep(&s[0], ":"); | |
1055 | if (p != NULL) | |
1056 | major_s = p; | |
1057 | else | |
ece84241 | 1058 | goto out; |
34d0f179 GJ |
1059 | |
1060 | minor_s = s[0]; | |
1061 | if (!minor_s) | |
ece84241 | 1062 | goto out; |
34d0f179 | 1063 | |
ece84241 TH |
1064 | if (strict_strtoul(major_s, 10, &major)) |
1065 | goto out; | |
34d0f179 | 1066 | |
ece84241 TH |
1067 | if (strict_strtoul(minor_s, 10, &minor)) |
1068 | goto out; | |
34d0f179 GJ |
1069 | |
1070 | dev = MKDEV(major, minor); | |
1071 | ||
ece84241 TH |
1072 | if (strict_strtoull(s[1], 10, &temp)) |
1073 | goto out; | |
34d0f179 | 1074 | |
e56da7e2 | 1075 | disk = get_gendisk(dev, &part); |
4bfd482e | 1076 | if (!disk || part) |
e56da7e2 | 1077 | goto out; |
e56da7e2 TH |
1078 | |
1079 | rcu_read_lock(); | |
1080 | ||
4bfd482e TH |
1081 | spin_lock_irq(disk->queue->queue_lock); |
1082 | blkg = blkg_lookup_create(blkcg, disk->queue, plid, false); | |
1083 | spin_unlock_irq(disk->queue->queue_lock); | |
e56da7e2 | 1084 | |
4bfd482e TH |
1085 | if (IS_ERR(blkg)) { |
1086 | ret = PTR_ERR(blkg); | |
1087 | goto out_unlock; | |
d11bb446 | 1088 | } |
34d0f179 | 1089 | |
549d3aa8 TH |
1090 | pd = blkg->pd[plid]; |
1091 | ||
062a644d VG |
1092 | switch (plid) { |
1093 | case BLKIO_POLICY_PROP: | |
d11bb446 WG |
1094 | if ((temp < BLKIO_WEIGHT_MIN && temp > 0) || |
1095 | temp > BLKIO_WEIGHT_MAX) | |
e56da7e2 | 1096 | goto out_unlock; |
34d0f179 | 1097 | |
549d3aa8 | 1098 | pd->conf.weight = temp; |
c1768268 | 1099 | blkio_update_group_weight(blkg, plid, temp ?: blkcg->weight); |
4c9eefa1 VG |
1100 | break; |
1101 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
1102 | switch(fileid) { |
1103 | case BLKIO_THROTL_read_bps_device: | |
549d3aa8 | 1104 | pd->conf.bps[READ] = temp; |
c1768268 | 1105 | blkio_update_group_bps(blkg, plid, temp ?: -1, fileid); |
e56da7e2 | 1106 | break; |
7702e8f4 | 1107 | case BLKIO_THROTL_write_bps_device: |
549d3aa8 | 1108 | pd->conf.bps[WRITE] = temp; |
c1768268 | 1109 | blkio_update_group_bps(blkg, plid, temp ?: -1, fileid); |
7702e8f4 VG |
1110 | break; |
1111 | case BLKIO_THROTL_read_iops_device: | |
e56da7e2 TH |
1112 | if (temp > THROTL_IOPS_MAX) |
1113 | goto out_unlock; | |
549d3aa8 | 1114 | pd->conf.iops[READ] = temp; |
c1768268 | 1115 | blkio_update_group_iops(blkg, plid, temp ?: -1, fileid); |
e56da7e2 | 1116 | break; |
7702e8f4 | 1117 | case BLKIO_THROTL_write_iops_device: |
d11bb446 | 1118 | if (temp > THROTL_IOPS_MAX) |
e56da7e2 | 1119 | goto out_unlock; |
549d3aa8 | 1120 | pd->conf.iops[WRITE] = temp; |
c1768268 | 1121 | blkio_update_group_iops(blkg, plid, temp ?: -1, fileid); |
7702e8f4 VG |
1122 | break; |
1123 | } | |
062a644d VG |
1124 | break; |
1125 | default: | |
1126 | BUG(); | |
1127 | } | |
ece84241 | 1128 | ret = 0; |
e56da7e2 TH |
1129 | out_unlock: |
1130 | rcu_read_unlock(); | |
ece84241 TH |
1131 | out: |
1132 | put_disk(disk); | |
e56da7e2 TH |
1133 | |
1134 | /* | |
1135 | * If queue was bypassing, we should retry. Do so after a short | |
1136 | * msleep(). It isn't strictly necessary but queue can be | |
1137 | * bypassing for some time and it's always nice to avoid busy | |
1138 | * looping. | |
1139 | */ | |
1140 | if (ret == -EBUSY) { | |
1141 | msleep(10); | |
1142 | return restart_syscall(); | |
1143 | } | |
ece84241 | 1144 | return ret; |
34d0f179 GJ |
1145 | } |
1146 | ||
062a644d VG |
1147 | static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft, |
1148 | const char *buffer) | |
34d0f179 GJ |
1149 | { |
1150 | int ret = 0; | |
1151 | char *buf; | |
e56da7e2 | 1152 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); |
062a644d VG |
1153 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); |
1154 | int fileid = BLKIOFILE_ATTR(cft->private); | |
34d0f179 GJ |
1155 | |
1156 | buf = kstrdup(buffer, GFP_KERNEL); | |
1157 | if (!buf) | |
1158 | return -ENOMEM; | |
1159 | ||
4bfd482e | 1160 | ret = blkio_policy_parse_and_set(buf, plid, fileid, blkcg); |
34d0f179 GJ |
1161 | kfree(buf); |
1162 | return ret; | |
1163 | } | |
1164 | ||
92616b5b VG |
1165 | static const char *blkg_dev_name(struct blkio_group *blkg) |
1166 | { | |
1167 | /* some drivers (floppy) instantiate a queue w/o disk registered */ | |
1168 | if (blkg->q->backing_dev_info.dev) | |
1169 | return dev_name(blkg->q->backing_dev_info.dev); | |
1170 | return NULL; | |
1171 | } | |
1172 | ||
4bfd482e TH |
1173 | static void blkio_print_group_conf(struct cftype *cft, struct blkio_group *blkg, |
1174 | struct seq_file *m) | |
34d0f179 | 1175 | { |
c1768268 | 1176 | int plid = BLKIOFILE_POLICY(cft->private); |
4bfd482e | 1177 | int fileid = BLKIOFILE_ATTR(cft->private); |
c1768268 TH |
1178 | struct blkg_policy_data *pd = blkg->pd[plid]; |
1179 | const char *dname = blkg_dev_name(blkg); | |
4bfd482e TH |
1180 | int rw = WRITE; |
1181 | ||
92616b5b VG |
1182 | if (!dname) |
1183 | return; | |
1184 | ||
c1768268 | 1185 | switch (plid) { |
062a644d | 1186 | case BLKIO_POLICY_PROP: |
549d3aa8 | 1187 | if (pd->conf.weight) |
7a4dd281 | 1188 | seq_printf(m, "%s\t%u\n", |
549d3aa8 | 1189 | dname, pd->conf.weight); |
4c9eefa1 VG |
1190 | break; |
1191 | case BLKIO_POLICY_THROTL: | |
4bfd482e | 1192 | switch (fileid) { |
7702e8f4 | 1193 | case BLKIO_THROTL_read_bps_device: |
4bfd482e | 1194 | rw = READ; |
7702e8f4 | 1195 | case BLKIO_THROTL_write_bps_device: |
549d3aa8 | 1196 | if (pd->conf.bps[rw]) |
7a4dd281 | 1197 | seq_printf(m, "%s\t%llu\n", |
549d3aa8 | 1198 | dname, pd->conf.bps[rw]); |
7702e8f4 VG |
1199 | break; |
1200 | case BLKIO_THROTL_read_iops_device: | |
4bfd482e | 1201 | rw = READ; |
7702e8f4 | 1202 | case BLKIO_THROTL_write_iops_device: |
549d3aa8 | 1203 | if (pd->conf.iops[rw]) |
7a4dd281 | 1204 | seq_printf(m, "%s\t%u\n", |
549d3aa8 | 1205 | dname, pd->conf.iops[rw]); |
7702e8f4 VG |
1206 | break; |
1207 | } | |
062a644d VG |
1208 | break; |
1209 | default: | |
1210 | BUG(); | |
1211 | } | |
1212 | } | |
34d0f179 | 1213 | |
062a644d | 1214 | /* cgroup files which read their data from policy nodes end up here */ |
4bfd482e TH |
1215 | static void blkio_read_conf(struct cftype *cft, struct blkio_cgroup *blkcg, |
1216 | struct seq_file *m) | |
34d0f179 | 1217 | { |
4bfd482e TH |
1218 | struct blkio_group *blkg; |
1219 | struct hlist_node *n; | |
34d0f179 | 1220 | |
4bfd482e TH |
1221 | spin_lock_irq(&blkcg->lock); |
1222 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) | |
e8989fae | 1223 | blkio_print_group_conf(cft, blkg, m); |
4bfd482e | 1224 | spin_unlock_irq(&blkcg->lock); |
062a644d VG |
1225 | } |
1226 | ||
1227 | static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft, | |
1228 | struct seq_file *m) | |
1229 | { | |
1230 | struct blkio_cgroup *blkcg; | |
1231 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1232 | int name = BLKIOFILE_ATTR(cft->private); | |
1233 | ||
1234 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1235 | ||
1236 | switch(plid) { | |
1237 | case BLKIO_POLICY_PROP: | |
1238 | switch(name) { | |
1239 | case BLKIO_PROP_weight_device: | |
4bfd482e | 1240 | blkio_read_conf(cft, blkcg, m); |
062a644d VG |
1241 | return 0; |
1242 | default: | |
1243 | BUG(); | |
1244 | } | |
1245 | break; | |
4c9eefa1 VG |
1246 | case BLKIO_POLICY_THROTL: |
1247 | switch(name){ | |
1248 | case BLKIO_THROTL_read_bps_device: | |
1249 | case BLKIO_THROTL_write_bps_device: | |
7702e8f4 VG |
1250 | case BLKIO_THROTL_read_iops_device: |
1251 | case BLKIO_THROTL_write_iops_device: | |
4bfd482e | 1252 | blkio_read_conf(cft, blkcg, m); |
4c9eefa1 VG |
1253 | return 0; |
1254 | default: | |
1255 | BUG(); | |
1256 | } | |
1257 | break; | |
062a644d VG |
1258 | default: |
1259 | BUG(); | |
1260 | } | |
1261 | ||
1262 | return 0; | |
1263 | } | |
1264 | ||
1265 | static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg, | |
5624a4e4 VG |
1266 | struct cftype *cft, struct cgroup_map_cb *cb, |
1267 | enum stat_type type, bool show_total, bool pcpu) | |
062a644d VG |
1268 | { |
1269 | struct blkio_group *blkg; | |
1270 | struct hlist_node *n; | |
1271 | uint64_t cgroup_total = 0; | |
1272 | ||
c875f4d0 TH |
1273 | spin_lock_irq(&blkcg->lock); |
1274 | ||
1275 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
92616b5b | 1276 | const char *dname = blkg_dev_name(blkg); |
c1768268 | 1277 | int plid = BLKIOFILE_POLICY(cft->private); |
7a4dd281 | 1278 | |
e8989fae | 1279 | if (!dname) |
7a4dd281 | 1280 | continue; |
c1768268 TH |
1281 | if (pcpu) { |
1282 | cgroup_total += blkio_get_stat_cpu(blkg, plid, | |
1283 | cb, dname, type); | |
1284 | } else { | |
c875f4d0 | 1285 | spin_lock(&blkg->stats_lock); |
c1768268 TH |
1286 | cgroup_total += blkio_get_stat(blkg, plid, |
1287 | cb, dname, type); | |
c875f4d0 | 1288 | spin_unlock(&blkg->stats_lock); |
062a644d VG |
1289 | } |
1290 | } | |
1291 | if (show_total) | |
1292 | cb->fill(cb, "Total", cgroup_total); | |
c875f4d0 TH |
1293 | |
1294 | spin_unlock_irq(&blkcg->lock); | |
062a644d VG |
1295 | return 0; |
1296 | } | |
1297 | ||
1298 | /* All map kind of cgroup file get serviced by this function */ | |
1299 | static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft, | |
1300 | struct cgroup_map_cb *cb) | |
1301 | { | |
1302 | struct blkio_cgroup *blkcg; | |
1303 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1304 | int name = BLKIOFILE_ATTR(cft->private); | |
1305 | ||
1306 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1307 | ||
1308 | switch(plid) { | |
1309 | case BLKIO_POLICY_PROP: | |
1310 | switch(name) { | |
1311 | case BLKIO_PROP_time: | |
1312 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1313 | BLKIO_STAT_TIME, 0, 0); |
062a644d VG |
1314 | case BLKIO_PROP_sectors: |
1315 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1316 | BLKIO_STAT_CPU_SECTORS, 0, 1); |
062a644d VG |
1317 | case BLKIO_PROP_io_service_bytes: |
1318 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1319 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
062a644d VG |
1320 | case BLKIO_PROP_io_serviced: |
1321 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1322 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
062a644d VG |
1323 | case BLKIO_PROP_io_service_time: |
1324 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1325 | BLKIO_STAT_SERVICE_TIME, 1, 0); |
062a644d VG |
1326 | case BLKIO_PROP_io_wait_time: |
1327 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1328 | BLKIO_STAT_WAIT_TIME, 1, 0); |
062a644d VG |
1329 | case BLKIO_PROP_io_merged: |
1330 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5fe224d2 | 1331 | BLKIO_STAT_MERGED, 1, 0); |
062a644d VG |
1332 | case BLKIO_PROP_io_queued: |
1333 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1334 | BLKIO_STAT_QUEUED, 1, 0); |
062a644d | 1335 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
9026e521 JT |
1336 | case BLKIO_PROP_unaccounted_time: |
1337 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1338 | BLKIO_STAT_UNACCOUNTED_TIME, 0, 0); |
062a644d VG |
1339 | case BLKIO_PROP_dequeue: |
1340 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1341 | BLKIO_STAT_DEQUEUE, 0, 0); |
062a644d VG |
1342 | case BLKIO_PROP_avg_queue_size: |
1343 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1344 | BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0); |
062a644d VG |
1345 | case BLKIO_PROP_group_wait_time: |
1346 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1347 | BLKIO_STAT_GROUP_WAIT_TIME, 0, 0); |
062a644d VG |
1348 | case BLKIO_PROP_idle_time: |
1349 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1350 | BLKIO_STAT_IDLE_TIME, 0, 0); |
062a644d VG |
1351 | case BLKIO_PROP_empty_time: |
1352 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1353 | BLKIO_STAT_EMPTY_TIME, 0, 0); |
062a644d VG |
1354 | #endif |
1355 | default: | |
1356 | BUG(); | |
1357 | } | |
1358 | break; | |
4c9eefa1 VG |
1359 | case BLKIO_POLICY_THROTL: |
1360 | switch(name){ | |
1361 | case BLKIO_THROTL_io_service_bytes: | |
1362 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1363 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
4c9eefa1 VG |
1364 | case BLKIO_THROTL_io_serviced: |
1365 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1366 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
4c9eefa1 VG |
1367 | default: |
1368 | BUG(); | |
1369 | } | |
1370 | break; | |
062a644d VG |
1371 | default: |
1372 | BUG(); | |
1373 | } | |
1374 | ||
1375 | return 0; | |
1376 | } | |
1377 | ||
4bfd482e | 1378 | static int blkio_weight_write(struct blkio_cgroup *blkcg, int plid, u64 val) |
062a644d VG |
1379 | { |
1380 | struct blkio_group *blkg; | |
1381 | struct hlist_node *n; | |
062a644d VG |
1382 | |
1383 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) | |
1384 | return -EINVAL; | |
1385 | ||
1386 | spin_lock(&blkio_list_lock); | |
1387 | spin_lock_irq(&blkcg->lock); | |
1388 | blkcg->weight = (unsigned int)val; | |
1389 | ||
549d3aa8 | 1390 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
e8989fae | 1391 | struct blkg_policy_data *pd = blkg->pd[plid]; |
549d3aa8 | 1392 | |
e8989fae | 1393 | if (!pd->conf.weight) |
c1768268 | 1394 | blkio_update_group_weight(blkg, plid, blkcg->weight); |
549d3aa8 | 1395 | } |
062a644d | 1396 | |
062a644d VG |
1397 | spin_unlock_irq(&blkcg->lock); |
1398 | spin_unlock(&blkio_list_lock); | |
1399 | return 0; | |
1400 | } | |
1401 | ||
1402 | static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) { | |
1403 | struct blkio_cgroup *blkcg; | |
1404 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1405 | int name = BLKIOFILE_ATTR(cft->private); | |
1406 | ||
1407 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1408 | ||
1409 | switch(plid) { | |
1410 | case BLKIO_POLICY_PROP: | |
1411 | switch(name) { | |
1412 | case BLKIO_PROP_weight: | |
1413 | return (u64)blkcg->weight; | |
1414 | } | |
1415 | break; | |
1416 | default: | |
1417 | BUG(); | |
1418 | } | |
1419 | return 0; | |
1420 | } | |
1421 | ||
1422 | static int | |
1423 | blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) | |
1424 | { | |
1425 | struct blkio_cgroup *blkcg; | |
1426 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1427 | int name = BLKIOFILE_ATTR(cft->private); | |
1428 | ||
1429 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1430 | ||
1431 | switch(plid) { | |
1432 | case BLKIO_POLICY_PROP: | |
1433 | switch(name) { | |
1434 | case BLKIO_PROP_weight: | |
4bfd482e | 1435 | return blkio_weight_write(blkcg, plid, val); |
062a644d VG |
1436 | } |
1437 | break; | |
1438 | default: | |
1439 | BUG(); | |
1440 | } | |
34d0f179 | 1441 | |
34d0f179 GJ |
1442 | return 0; |
1443 | } | |
1444 | ||
31e4c28d | 1445 | struct cftype blkio_files[] = { |
34d0f179 GJ |
1446 | { |
1447 | .name = "weight_device", | |
062a644d VG |
1448 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1449 | BLKIO_PROP_weight_device), | |
1450 | .read_seq_string = blkiocg_file_read, | |
1451 | .write_string = blkiocg_file_write, | |
34d0f179 GJ |
1452 | .max_write_len = 256, |
1453 | }, | |
31e4c28d VG |
1454 | { |
1455 | .name = "weight", | |
062a644d VG |
1456 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1457 | BLKIO_PROP_weight), | |
1458 | .read_u64 = blkiocg_file_read_u64, | |
1459 | .write_u64 = blkiocg_file_write_u64, | |
31e4c28d | 1460 | }, |
22084190 VG |
1461 | { |
1462 | .name = "time", | |
062a644d VG |
1463 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1464 | BLKIO_PROP_time), | |
1465 | .read_map = blkiocg_file_read_map, | |
22084190 VG |
1466 | }, |
1467 | { | |
1468 | .name = "sectors", | |
062a644d VG |
1469 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1470 | BLKIO_PROP_sectors), | |
1471 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1472 | }, |
1473 | { | |
1474 | .name = "io_service_bytes", | |
062a644d VG |
1475 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1476 | BLKIO_PROP_io_service_bytes), | |
1477 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1478 | }, |
1479 | { | |
1480 | .name = "io_serviced", | |
062a644d VG |
1481 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1482 | BLKIO_PROP_io_serviced), | |
1483 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1484 | }, |
1485 | { | |
1486 | .name = "io_service_time", | |
062a644d VG |
1487 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1488 | BLKIO_PROP_io_service_time), | |
1489 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1490 | }, |
1491 | { | |
1492 | .name = "io_wait_time", | |
062a644d VG |
1493 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1494 | BLKIO_PROP_io_wait_time), | |
1495 | .read_map = blkiocg_file_read_map, | |
84c124da | 1496 | }, |
812d4026 DS |
1497 | { |
1498 | .name = "io_merged", | |
062a644d VG |
1499 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1500 | BLKIO_PROP_io_merged), | |
1501 | .read_map = blkiocg_file_read_map, | |
812d4026 | 1502 | }, |
cdc1184c DS |
1503 | { |
1504 | .name = "io_queued", | |
062a644d VG |
1505 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1506 | BLKIO_PROP_io_queued), | |
1507 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1508 | }, |
84c124da DS |
1509 | { |
1510 | .name = "reset_stats", | |
1511 | .write_u64 = blkiocg_reset_stats, | |
22084190 | 1512 | }, |
13f98250 VG |
1513 | #ifdef CONFIG_BLK_DEV_THROTTLING |
1514 | { | |
1515 | .name = "throttle.read_bps_device", | |
1516 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1517 | BLKIO_THROTL_read_bps_device), | |
1518 | .read_seq_string = blkiocg_file_read, | |
1519 | .write_string = blkiocg_file_write, | |
1520 | .max_write_len = 256, | |
1521 | }, | |
1522 | ||
1523 | { | |
1524 | .name = "throttle.write_bps_device", | |
1525 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1526 | BLKIO_THROTL_write_bps_device), | |
1527 | .read_seq_string = blkiocg_file_read, | |
1528 | .write_string = blkiocg_file_write, | |
1529 | .max_write_len = 256, | |
1530 | }, | |
1531 | ||
1532 | { | |
1533 | .name = "throttle.read_iops_device", | |
1534 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1535 | BLKIO_THROTL_read_iops_device), | |
1536 | .read_seq_string = blkiocg_file_read, | |
1537 | .write_string = blkiocg_file_write, | |
1538 | .max_write_len = 256, | |
1539 | }, | |
1540 | ||
1541 | { | |
1542 | .name = "throttle.write_iops_device", | |
1543 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1544 | BLKIO_THROTL_write_iops_device), | |
1545 | .read_seq_string = blkiocg_file_read, | |
1546 | .write_string = blkiocg_file_write, | |
1547 | .max_write_len = 256, | |
1548 | }, | |
1549 | { | |
1550 | .name = "throttle.io_service_bytes", | |
1551 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1552 | BLKIO_THROTL_io_service_bytes), | |
1553 | .read_map = blkiocg_file_read_map, | |
1554 | }, | |
1555 | { | |
1556 | .name = "throttle.io_serviced", | |
1557 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1558 | BLKIO_THROTL_io_serviced), | |
1559 | .read_map = blkiocg_file_read_map, | |
1560 | }, | |
1561 | #endif /* CONFIG_BLK_DEV_THROTTLING */ | |
1562 | ||
22084190 | 1563 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
cdc1184c DS |
1564 | { |
1565 | .name = "avg_queue_size", | |
062a644d VG |
1566 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1567 | BLKIO_PROP_avg_queue_size), | |
1568 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1569 | }, |
812df48d DS |
1570 | { |
1571 | .name = "group_wait_time", | |
062a644d VG |
1572 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1573 | BLKIO_PROP_group_wait_time), | |
1574 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1575 | }, |
1576 | { | |
1577 | .name = "idle_time", | |
062a644d VG |
1578 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1579 | BLKIO_PROP_idle_time), | |
1580 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1581 | }, |
1582 | { | |
1583 | .name = "empty_time", | |
062a644d VG |
1584 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1585 | BLKIO_PROP_empty_time), | |
1586 | .read_map = blkiocg_file_read_map, | |
812df48d | 1587 | }, |
cdc1184c | 1588 | { |
22084190 | 1589 | .name = "dequeue", |
062a644d VG |
1590 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1591 | BLKIO_PROP_dequeue), | |
1592 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1593 | }, |
9026e521 JT |
1594 | { |
1595 | .name = "unaccounted_time", | |
1596 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, | |
1597 | BLKIO_PROP_unaccounted_time), | |
1598 | .read_map = blkiocg_file_read_map, | |
1599 | }, | |
22084190 | 1600 | #endif |
31e4c28d VG |
1601 | }; |
1602 | ||
1603 | static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1604 | { | |
1605 | return cgroup_add_files(cgroup, subsys, blkio_files, | |
1606 | ARRAY_SIZE(blkio_files)); | |
1607 | } | |
1608 | ||
9f13ef67 TH |
1609 | /** |
1610 | * blkiocg_pre_destroy - cgroup pre_destroy callback | |
1611 | * @subsys: cgroup subsys | |
1612 | * @cgroup: cgroup of interest | |
1613 | * | |
1614 | * This function is called when @cgroup is about to go away and responsible | |
1615 | * for shooting down all blkgs associated with @cgroup. blkgs should be | |
1616 | * removed while holding both q and blkcg locks. As blkcg lock is nested | |
1617 | * inside q lock, this function performs reverse double lock dancing. | |
1618 | * | |
1619 | * This is the blkcg counterpart of ioc_release_fn(). | |
1620 | */ | |
7ee9c562 TH |
1621 | static int blkiocg_pre_destroy(struct cgroup_subsys *subsys, |
1622 | struct cgroup *cgroup) | |
31e4c28d VG |
1623 | { |
1624 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
b1c35769 | 1625 | |
9f13ef67 | 1626 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1627 | |
9f13ef67 TH |
1628 | while (!hlist_empty(&blkcg->blkg_list)) { |
1629 | struct blkio_group *blkg = hlist_entry(blkcg->blkg_list.first, | |
1630 | struct blkio_group, blkcg_node); | |
c875f4d0 | 1631 | struct request_queue *q = blkg->q; |
b1c35769 | 1632 | |
9f13ef67 TH |
1633 | if (spin_trylock(q->queue_lock)) { |
1634 | blkg_destroy(blkg); | |
1635 | spin_unlock(q->queue_lock); | |
1636 | } else { | |
1637 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 1638 | cpu_relax(); |
9f13ef67 | 1639 | spin_lock(&blkcg->lock); |
0f3942a3 | 1640 | } |
9f13ef67 | 1641 | } |
b1c35769 | 1642 | |
9f13ef67 | 1643 | spin_unlock_irq(&blkcg->lock); |
7ee9c562 TH |
1644 | return 0; |
1645 | } | |
1646 | ||
1647 | static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1648 | { | |
1649 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
1650 | ||
67523c48 BB |
1651 | if (blkcg != &blkio_root_cgroup) |
1652 | kfree(blkcg); | |
31e4c28d VG |
1653 | } |
1654 | ||
1655 | static struct cgroup_subsys_state * | |
1656 | blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1657 | { | |
0341509f LZ |
1658 | struct blkio_cgroup *blkcg; |
1659 | struct cgroup *parent = cgroup->parent; | |
31e4c28d | 1660 | |
0341509f | 1661 | if (!parent) { |
31e4c28d VG |
1662 | blkcg = &blkio_root_cgroup; |
1663 | goto done; | |
1664 | } | |
1665 | ||
31e4c28d VG |
1666 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
1667 | if (!blkcg) | |
1668 | return ERR_PTR(-ENOMEM); | |
1669 | ||
1670 | blkcg->weight = BLKIO_WEIGHT_DEFAULT; | |
1671 | done: | |
1672 | spin_lock_init(&blkcg->lock); | |
1673 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
1674 | ||
1675 | return &blkcg->css; | |
1676 | } | |
1677 | ||
5efd6113 TH |
1678 | /** |
1679 | * blkcg_init_queue - initialize blkcg part of request queue | |
1680 | * @q: request_queue to initialize | |
1681 | * | |
1682 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1683 | * part of new request_queue @q. | |
1684 | * | |
1685 | * RETURNS: | |
1686 | * 0 on success, -errno on failure. | |
1687 | */ | |
1688 | int blkcg_init_queue(struct request_queue *q) | |
1689 | { | |
923adde1 TH |
1690 | int ret; |
1691 | ||
5efd6113 TH |
1692 | might_sleep(); |
1693 | ||
923adde1 TH |
1694 | ret = blk_throtl_init(q); |
1695 | if (ret) | |
1696 | return ret; | |
1697 | ||
1698 | mutex_lock(&all_q_mutex); | |
1699 | INIT_LIST_HEAD(&q->all_q_node); | |
1700 | list_add_tail(&q->all_q_node, &all_q_list); | |
1701 | mutex_unlock(&all_q_mutex); | |
1702 | ||
1703 | return 0; | |
5efd6113 TH |
1704 | } |
1705 | ||
1706 | /** | |
1707 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1708 | * @q: request_queue to drain | |
1709 | * | |
1710 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1711 | */ | |
1712 | void blkcg_drain_queue(struct request_queue *q) | |
1713 | { | |
1714 | lockdep_assert_held(q->queue_lock); | |
1715 | ||
1716 | blk_throtl_drain(q); | |
1717 | } | |
1718 | ||
1719 | /** | |
1720 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1721 | * @q: request_queue being released | |
1722 | * | |
1723 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1724 | */ | |
1725 | void blkcg_exit_queue(struct request_queue *q) | |
1726 | { | |
923adde1 TH |
1727 | mutex_lock(&all_q_mutex); |
1728 | list_del_init(&q->all_q_node); | |
1729 | mutex_unlock(&all_q_mutex); | |
1730 | ||
e8989fae TH |
1731 | blkg_destroy_all(q, true); |
1732 | ||
5efd6113 TH |
1733 | blk_throtl_exit(q); |
1734 | } | |
1735 | ||
31e4c28d VG |
1736 | /* |
1737 | * We cannot support shared io contexts, as we have no mean to support | |
1738 | * two tasks with the same ioc in two different groups without major rework | |
1739 | * of the main cic data structures. For now we allow a task to change | |
1740 | * its cgroup only if it's the only owner of its ioc. | |
1741 | */ | |
bb9d97b6 TH |
1742 | static int blkiocg_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
1743 | struct cgroup_taskset *tset) | |
31e4c28d | 1744 | { |
bb9d97b6 | 1745 | struct task_struct *task; |
31e4c28d VG |
1746 | struct io_context *ioc; |
1747 | int ret = 0; | |
1748 | ||
1749 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
bb9d97b6 TH |
1750 | cgroup_taskset_for_each(task, cgrp, tset) { |
1751 | task_lock(task); | |
1752 | ioc = task->io_context; | |
1753 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1754 | ret = -EINVAL; | |
1755 | task_unlock(task); | |
1756 | if (ret) | |
1757 | break; | |
1758 | } | |
31e4c28d VG |
1759 | return ret; |
1760 | } | |
1761 | ||
bb9d97b6 TH |
1762 | static void blkiocg_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
1763 | struct cgroup_taskset *tset) | |
31e4c28d | 1764 | { |
bb9d97b6 | 1765 | struct task_struct *task; |
31e4c28d VG |
1766 | struct io_context *ioc; |
1767 | ||
bb9d97b6 | 1768 | cgroup_taskset_for_each(task, cgrp, tset) { |
b3c9dd18 LT |
1769 | /* we don't lose anything even if ioc allocation fails */ |
1770 | ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE); | |
1771 | if (ioc) { | |
1772 | ioc_cgroup_changed(ioc); | |
11a3122f | 1773 | put_io_context(ioc); |
b3c9dd18 | 1774 | } |
bb9d97b6 | 1775 | } |
31e4c28d VG |
1776 | } |
1777 | ||
923adde1 TH |
1778 | static void blkcg_bypass_start(void) |
1779 | __acquires(&all_q_mutex) | |
1780 | { | |
1781 | struct request_queue *q; | |
1782 | ||
1783 | mutex_lock(&all_q_mutex); | |
1784 | ||
1785 | list_for_each_entry(q, &all_q_list, all_q_node) { | |
1786 | blk_queue_bypass_start(q); | |
e8989fae | 1787 | blkg_destroy_all(q, false); |
923adde1 TH |
1788 | } |
1789 | } | |
1790 | ||
1791 | static void blkcg_bypass_end(void) | |
1792 | __releases(&all_q_mutex) | |
1793 | { | |
1794 | struct request_queue *q; | |
1795 | ||
1796 | list_for_each_entry(q, &all_q_list, all_q_node) | |
1797 | blk_queue_bypass_end(q); | |
1798 | ||
1799 | mutex_unlock(&all_q_mutex); | |
1800 | } | |
1801 | ||
3e252066 VG |
1802 | void blkio_policy_register(struct blkio_policy_type *blkiop) |
1803 | { | |
e8989fae TH |
1804 | struct request_queue *q; |
1805 | ||
923adde1 | 1806 | blkcg_bypass_start(); |
3e252066 | 1807 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1808 | |
1809 | BUG_ON(blkio_policy[blkiop->plid]); | |
1810 | blkio_policy[blkiop->plid] = blkiop; | |
3e252066 | 1811 | list_add_tail(&blkiop->list, &blkio_list); |
035d10b2 | 1812 | |
3e252066 | 1813 | spin_unlock(&blkio_list_lock); |
e8989fae TH |
1814 | list_for_each_entry(q, &all_q_list, all_q_node) |
1815 | update_root_blkg_pd(q, blkiop->plid); | |
923adde1 | 1816 | blkcg_bypass_end(); |
3e252066 VG |
1817 | } |
1818 | EXPORT_SYMBOL_GPL(blkio_policy_register); | |
1819 | ||
1820 | void blkio_policy_unregister(struct blkio_policy_type *blkiop) | |
1821 | { | |
e8989fae TH |
1822 | struct request_queue *q; |
1823 | ||
923adde1 | 1824 | blkcg_bypass_start(); |
3e252066 | 1825 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1826 | |
1827 | BUG_ON(blkio_policy[blkiop->plid] != blkiop); | |
1828 | blkio_policy[blkiop->plid] = NULL; | |
3e252066 | 1829 | list_del_init(&blkiop->list); |
035d10b2 | 1830 | |
3e252066 | 1831 | spin_unlock(&blkio_list_lock); |
e8989fae TH |
1832 | list_for_each_entry(q, &all_q_list, all_q_node) |
1833 | update_root_blkg_pd(q, blkiop->plid); | |
923adde1 | 1834 | blkcg_bypass_end(); |
3e252066 VG |
1835 | } |
1836 | EXPORT_SYMBOL_GPL(blkio_policy_unregister); |