]>
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 | ||
31e4c28d VG |
46 | struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup) |
47 | { | |
48 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), | |
49 | struct blkio_cgroup, css); | |
50 | } | |
9d6a986c | 51 | EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup); |
31e4c28d | 52 | |
4f85cb96 | 53 | static struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk) |
70087dc3 VG |
54 | { |
55 | return container_of(task_subsys_state(tsk, blkio_subsys_id), | |
56 | struct blkio_cgroup, css); | |
57 | } | |
4f85cb96 TH |
58 | |
59 | struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio) | |
60 | { | |
61 | if (bio && bio->bi_css) | |
62 | return container_of(bio->bi_css, struct blkio_cgroup, css); | |
63 | return task_blkio_cgroup(current); | |
64 | } | |
65 | EXPORT_SYMBOL_GPL(bio_blkio_cgroup); | |
70087dc3 | 66 | |
c1768268 TH |
67 | static inline void blkio_update_group_weight(struct blkio_group *blkg, |
68 | int plid, unsigned int weight) | |
062a644d VG |
69 | { |
70 | struct blkio_policy_type *blkiop; | |
71 | ||
72 | list_for_each_entry(blkiop, &blkio_list, list) { | |
73 | /* If this policy does not own the blkg, do not send updates */ | |
c1768268 | 74 | if (blkiop->plid != plid) |
062a644d VG |
75 | continue; |
76 | if (blkiop->ops.blkio_update_group_weight_fn) | |
ca32aefc | 77 | blkiop->ops.blkio_update_group_weight_fn(blkg->q, |
fe071437 | 78 | blkg, weight); |
062a644d VG |
79 | } |
80 | } | |
81 | ||
c1768268 | 82 | static inline void blkio_update_group_bps(struct blkio_group *blkg, int plid, |
3a8b31d3 | 83 | u64 bps, int rw) |
4c9eefa1 VG |
84 | { |
85 | struct blkio_policy_type *blkiop; | |
86 | ||
87 | list_for_each_entry(blkiop, &blkio_list, list) { | |
88 | ||
89 | /* If this policy does not own the blkg, do not send updates */ | |
c1768268 | 90 | if (blkiop->plid != plid) |
4c9eefa1 VG |
91 | continue; |
92 | ||
3a8b31d3 | 93 | if (rw == READ && blkiop->ops.blkio_update_group_read_bps_fn) |
ca32aefc | 94 | blkiop->ops.blkio_update_group_read_bps_fn(blkg->q, |
fe071437 | 95 | blkg, bps); |
4c9eefa1 | 96 | |
3a8b31d3 | 97 | if (rw == WRITE && blkiop->ops.blkio_update_group_write_bps_fn) |
ca32aefc | 98 | blkiop->ops.blkio_update_group_write_bps_fn(blkg->q, |
fe071437 | 99 | blkg, bps); |
4c9eefa1 VG |
100 | } |
101 | } | |
102 | ||
3a8b31d3 TH |
103 | static inline void blkio_update_group_iops(struct blkio_group *blkg, int plid, |
104 | u64 iops, int rw) | |
7702e8f4 VG |
105 | { |
106 | struct blkio_policy_type *blkiop; | |
107 | ||
108 | list_for_each_entry(blkiop, &blkio_list, list) { | |
109 | ||
110 | /* If this policy does not own the blkg, do not send updates */ | |
c1768268 | 111 | if (blkiop->plid != plid) |
7702e8f4 VG |
112 | continue; |
113 | ||
3a8b31d3 | 114 | if (rw == READ && blkiop->ops.blkio_update_group_read_iops_fn) |
ca32aefc | 115 | blkiop->ops.blkio_update_group_read_iops_fn(blkg->q, |
fe071437 | 116 | blkg, iops); |
7702e8f4 | 117 | |
3a8b31d3 | 118 | if (rw == WRITE && blkiop->ops.blkio_update_group_write_iops_fn) |
ca32aefc | 119 | blkiop->ops.blkio_update_group_write_iops_fn(blkg->q, |
fe071437 | 120 | blkg,iops); |
7702e8f4 VG |
121 | } |
122 | } | |
123 | ||
cdc1184c | 124 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
edf1b879 | 125 | /* This should be called with the queue_lock held. */ |
812df48d | 126 | static void blkio_set_start_group_wait_time(struct blkio_group *blkg, |
c1768268 TH |
127 | struct blkio_policy_type *pol, |
128 | struct blkio_group *curr_blkg) | |
812df48d | 129 | { |
c1768268 | 130 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
549d3aa8 TH |
131 | |
132 | if (blkio_blkg_waiting(&pd->stats)) | |
812df48d DS |
133 | return; |
134 | if (blkg == curr_blkg) | |
135 | return; | |
549d3aa8 TH |
136 | pd->stats.start_group_wait_time = sched_clock(); |
137 | blkio_mark_blkg_waiting(&pd->stats); | |
812df48d DS |
138 | } |
139 | ||
edf1b879 | 140 | /* This should be called with the queue_lock held. */ |
812df48d DS |
141 | static void blkio_update_group_wait_time(struct blkio_group_stats *stats) |
142 | { | |
143 | unsigned long long now; | |
144 | ||
145 | if (!blkio_blkg_waiting(stats)) | |
146 | return; | |
147 | ||
148 | now = sched_clock(); | |
149 | if (time_after64(now, stats->start_group_wait_time)) | |
edcb0722 TH |
150 | blkg_stat_add(&stats->group_wait_time, |
151 | now - stats->start_group_wait_time); | |
812df48d DS |
152 | blkio_clear_blkg_waiting(stats); |
153 | } | |
154 | ||
edf1b879 | 155 | /* This should be called with the queue_lock held. */ |
812df48d DS |
156 | static void blkio_end_empty_time(struct blkio_group_stats *stats) |
157 | { | |
158 | unsigned long long now; | |
159 | ||
160 | if (!blkio_blkg_empty(stats)) | |
161 | return; | |
162 | ||
163 | now = sched_clock(); | |
164 | if (time_after64(now, stats->start_empty_time)) | |
edcb0722 TH |
165 | blkg_stat_add(&stats->empty_time, |
166 | now - stats->start_empty_time); | |
812df48d DS |
167 | blkio_clear_blkg_empty(stats); |
168 | } | |
169 | ||
c1768268 TH |
170 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg, |
171 | struct blkio_policy_type *pol) | |
812df48d | 172 | { |
edf1b879 | 173 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
812df48d | 174 | |
edf1b879 TH |
175 | lockdep_assert_held(blkg->q->queue_lock); |
176 | BUG_ON(blkio_blkg_idling(stats)); | |
177 | ||
178 | stats->start_idle_time = sched_clock(); | |
179 | blkio_mark_blkg_idling(stats); | |
812df48d DS |
180 | } |
181 | EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats); | |
182 | ||
c1768268 TH |
183 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg, |
184 | struct blkio_policy_type *pol) | |
812df48d | 185 | { |
edf1b879 TH |
186 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
187 | ||
188 | lockdep_assert_held(blkg->q->queue_lock); | |
812df48d | 189 | |
812df48d | 190 | if (blkio_blkg_idling(stats)) { |
edf1b879 TH |
191 | unsigned long long now = sched_clock(); |
192 | ||
edcb0722 TH |
193 | if (time_after64(now, stats->start_idle_time)) |
194 | blkg_stat_add(&stats->idle_time, | |
195 | now - stats->start_idle_time); | |
812df48d DS |
196 | blkio_clear_blkg_idling(stats); |
197 | } | |
812df48d DS |
198 | } |
199 | EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats); | |
200 | ||
c1768268 TH |
201 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg, |
202 | struct blkio_policy_type *pol) | |
cdc1184c | 203 | { |
edf1b879 | 204 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
cdc1184c | 205 | |
edf1b879 TH |
206 | lockdep_assert_held(blkg->q->queue_lock); |
207 | ||
edcb0722 TH |
208 | blkg_stat_add(&stats->avg_queue_size_sum, |
209 | blkg_rwstat_sum(&stats->queued)); | |
210 | blkg_stat_add(&stats->avg_queue_size_samples, 1); | |
812df48d | 211 | blkio_update_group_wait_time(stats); |
cdc1184c | 212 | } |
a11cdaa7 DS |
213 | EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats); |
214 | ||
c1768268 TH |
215 | void blkiocg_set_start_empty_time(struct blkio_group *blkg, |
216 | struct blkio_policy_type *pol) | |
28baf442 | 217 | { |
edf1b879 | 218 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
28baf442 | 219 | |
edf1b879 | 220 | lockdep_assert_held(blkg->q->queue_lock); |
28baf442 | 221 | |
edcb0722 | 222 | if (blkg_rwstat_sum(&stats->queued)) |
28baf442 | 223 | return; |
28baf442 DS |
224 | |
225 | /* | |
e5ff082e VG |
226 | * group is already marked empty. This can happen if cfqq got new |
227 | * request in parent group and moved to this group while being added | |
228 | * to service tree. Just ignore the event and move on. | |
28baf442 | 229 | */ |
edf1b879 | 230 | if (blkio_blkg_empty(stats)) |
e5ff082e | 231 | return; |
e5ff082e | 232 | |
28baf442 DS |
233 | stats->start_empty_time = sched_clock(); |
234 | blkio_mark_blkg_empty(stats); | |
28baf442 DS |
235 | } |
236 | EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time); | |
237 | ||
a11cdaa7 | 238 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
c1768268 TH |
239 | struct blkio_policy_type *pol, |
240 | unsigned long dequeue) | |
a11cdaa7 | 241 | { |
c1768268 | 242 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
549d3aa8 | 243 | |
edf1b879 TH |
244 | lockdep_assert_held(blkg->q->queue_lock); |
245 | ||
edcb0722 | 246 | blkg_stat_add(&pd->stats.dequeue, dequeue); |
a11cdaa7 DS |
247 | } |
248 | EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats); | |
812df48d DS |
249 | #else |
250 | static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
c1768268 TH |
251 | struct blkio_policy_type *pol, |
252 | struct blkio_group *curr_blkg) { } | |
253 | static inline void blkio_end_empty_time(struct blkio_group_stats *stats) { } | |
cdc1184c DS |
254 | #endif |
255 | ||
a11cdaa7 | 256 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
c1768268 TH |
257 | struct blkio_policy_type *pol, |
258 | struct blkio_group *curr_blkg, bool direction, | |
259 | bool sync) | |
cdc1184c | 260 | { |
edf1b879 | 261 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
edcb0722 | 262 | int rw = (direction ? REQ_WRITE : 0) | (sync ? REQ_SYNC : 0); |
edf1b879 TH |
263 | |
264 | lockdep_assert_held(blkg->q->queue_lock); | |
265 | ||
edcb0722 | 266 | blkg_rwstat_add(&stats->queued, rw, 1); |
edf1b879 | 267 | blkio_end_empty_time(stats); |
c1768268 | 268 | blkio_set_start_group_wait_time(blkg, pol, curr_blkg); |
cdc1184c | 269 | } |
a11cdaa7 | 270 | EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats); |
cdc1184c | 271 | |
a11cdaa7 | 272 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
c1768268 TH |
273 | struct blkio_policy_type *pol, |
274 | bool direction, bool sync) | |
cdc1184c | 275 | { |
edf1b879 | 276 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
edcb0722 | 277 | int rw = (direction ? REQ_WRITE : 0) | (sync ? REQ_SYNC : 0); |
edf1b879 TH |
278 | |
279 | lockdep_assert_held(blkg->q->queue_lock); | |
cdc1184c | 280 | |
edcb0722 | 281 | blkg_rwstat_add(&stats->queued, rw, -1); |
cdc1184c | 282 | } |
a11cdaa7 | 283 | EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats); |
cdc1184c | 284 | |
c1768268 TH |
285 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
286 | struct blkio_policy_type *pol, | |
287 | unsigned long time, | |
288 | unsigned long unaccounted_time) | |
22084190 | 289 | { |
edf1b879 TH |
290 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
291 | ||
292 | lockdep_assert_held(blkg->q->queue_lock); | |
303a3acb | 293 | |
edcb0722 | 294 | blkg_stat_add(&stats->time, time); |
a23e6869 | 295 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
edcb0722 | 296 | blkg_stat_add(&stats->unaccounted_time, unaccounted_time); |
a23e6869 | 297 | #endif |
22084190 | 298 | } |
303a3acb | 299 | EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used); |
22084190 | 300 | |
5624a4e4 VG |
301 | /* |
302 | * should be called under rcu read lock or queue lock to make sure blkg pointer | |
303 | * is valid. | |
304 | */ | |
84c124da | 305 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
c1768268 TH |
306 | struct blkio_policy_type *pol, |
307 | uint64_t bytes, bool direction, bool sync) | |
9195291e | 308 | { |
edcb0722 | 309 | int rw = (direction ? REQ_WRITE : 0) | (sync ? REQ_SYNC : 0); |
c1768268 | 310 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
5624a4e4 | 311 | struct blkio_group_stats_cpu *stats_cpu; |
575969a0 VG |
312 | unsigned long flags; |
313 | ||
1cd9e039 VG |
314 | /* If per cpu stats are not allocated yet, don't do any accounting. */ |
315 | if (pd->stats_cpu == NULL) | |
316 | return; | |
317 | ||
575969a0 VG |
318 | /* |
319 | * Disabling interrupts to provide mutual exclusion between two | |
320 | * writes on same cpu. It probably is not needed for 64bit. Not | |
321 | * optimizing that case yet. | |
322 | */ | |
323 | local_irq_save(flags); | |
9195291e | 324 | |
549d3aa8 | 325 | stats_cpu = this_cpu_ptr(pd->stats_cpu); |
5624a4e4 | 326 | |
edcb0722 TH |
327 | blkg_stat_add(&stats_cpu->sectors, bytes >> 9); |
328 | blkg_rwstat_add(&stats_cpu->serviced, rw, 1); | |
329 | blkg_rwstat_add(&stats_cpu->service_bytes, rw, bytes); | |
330 | ||
575969a0 | 331 | local_irq_restore(flags); |
9195291e | 332 | } |
84c124da | 333 | EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats); |
9195291e | 334 | |
84c124da | 335 | void blkiocg_update_completion_stats(struct blkio_group *blkg, |
c1768268 TH |
336 | struct blkio_policy_type *pol, |
337 | uint64_t start_time, | |
338 | uint64_t io_start_time, bool direction, | |
339 | bool sync) | |
9195291e | 340 | { |
edf1b879 | 341 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
9195291e | 342 | unsigned long long now = sched_clock(); |
edcb0722 | 343 | int rw = (direction ? REQ_WRITE : 0) | (sync ? REQ_SYNC : 0); |
9195291e | 344 | |
edf1b879 TH |
345 | lockdep_assert_held(blkg->q->queue_lock); |
346 | ||
84c124da | 347 | if (time_after64(now, io_start_time)) |
edcb0722 | 348 | blkg_rwstat_add(&stats->service_time, rw, now - io_start_time); |
84c124da | 349 | if (time_after64(io_start_time, start_time)) |
edcb0722 TH |
350 | blkg_rwstat_add(&stats->wait_time, rw, |
351 | io_start_time - start_time); | |
9195291e | 352 | } |
84c124da | 353 | EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats); |
9195291e | 354 | |
317389a7 | 355 | /* Merged stats are per cpu. */ |
c1768268 TH |
356 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, |
357 | struct blkio_policy_type *pol, | |
358 | bool direction, bool sync) | |
812d4026 | 359 | { |
edf1b879 | 360 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
edcb0722 | 361 | int rw = (direction ? REQ_WRITE : 0) | (sync ? REQ_SYNC : 0); |
edf1b879 TH |
362 | |
363 | lockdep_assert_held(blkg->q->queue_lock); | |
812d4026 | 364 | |
edcb0722 | 365 | blkg_rwstat_add(&stats->merged, rw, 1); |
812d4026 DS |
366 | } |
367 | EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats); | |
368 | ||
1cd9e039 VG |
369 | /* |
370 | * Worker for allocating per cpu stat for blk groups. This is scheduled on | |
371 | * the system_nrt_wq once there are some groups on the alloc_list waiting | |
372 | * for allocation. | |
373 | */ | |
374 | static void blkio_stat_alloc_fn(struct work_struct *work) | |
375 | { | |
376 | static void *pcpu_stats[BLKIO_NR_POLICIES]; | |
377 | struct delayed_work *dwork = to_delayed_work(work); | |
378 | struct blkio_group *blkg; | |
379 | int i; | |
380 | bool empty = false; | |
381 | ||
382 | alloc_stats: | |
383 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { | |
384 | if (pcpu_stats[i] != NULL) | |
385 | continue; | |
386 | ||
387 | pcpu_stats[i] = alloc_percpu(struct blkio_group_stats_cpu); | |
388 | ||
389 | /* Allocation failed. Try again after some time. */ | |
390 | if (pcpu_stats[i] == NULL) { | |
391 | queue_delayed_work(system_nrt_wq, dwork, | |
392 | msecs_to_jiffies(10)); | |
393 | return; | |
394 | } | |
395 | } | |
396 | ||
397 | spin_lock_irq(&blkio_list_lock); | |
398 | spin_lock(&alloc_list_lock); | |
399 | ||
400 | /* cgroup got deleted or queue exited. */ | |
401 | if (!list_empty(&alloc_list)) { | |
402 | blkg = list_first_entry(&alloc_list, struct blkio_group, | |
403 | alloc_node); | |
404 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { | |
405 | struct blkg_policy_data *pd = blkg->pd[i]; | |
406 | ||
407 | if (blkio_policy[i] && pd && !pd->stats_cpu) | |
408 | swap(pd->stats_cpu, pcpu_stats[i]); | |
409 | } | |
410 | ||
411 | list_del_init(&blkg->alloc_node); | |
412 | } | |
413 | ||
414 | empty = list_empty(&alloc_list); | |
415 | ||
416 | spin_unlock(&alloc_list_lock); | |
417 | spin_unlock_irq(&blkio_list_lock); | |
418 | ||
419 | if (!empty) | |
420 | goto alloc_stats; | |
421 | } | |
422 | ||
0381411e TH |
423 | /** |
424 | * blkg_free - free a blkg | |
425 | * @blkg: blkg to free | |
426 | * | |
427 | * Free @blkg which may be partially allocated. | |
428 | */ | |
429 | static void blkg_free(struct blkio_group *blkg) | |
430 | { | |
e8989fae | 431 | int i; |
549d3aa8 TH |
432 | |
433 | if (!blkg) | |
434 | return; | |
435 | ||
e8989fae TH |
436 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
437 | struct blkg_policy_data *pd = blkg->pd[i]; | |
438 | ||
439 | if (pd) { | |
440 | free_percpu(pd->stats_cpu); | |
441 | kfree(pd); | |
442 | } | |
0381411e | 443 | } |
e8989fae | 444 | |
549d3aa8 | 445 | kfree(blkg); |
0381411e TH |
446 | } |
447 | ||
448 | /** | |
449 | * blkg_alloc - allocate a blkg | |
450 | * @blkcg: block cgroup the new blkg is associated with | |
451 | * @q: request_queue the new blkg is associated with | |
0381411e | 452 | * |
e8989fae | 453 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e TH |
454 | */ |
455 | static struct blkio_group *blkg_alloc(struct blkio_cgroup *blkcg, | |
e8989fae | 456 | struct request_queue *q) |
0381411e TH |
457 | { |
458 | struct blkio_group *blkg; | |
e8989fae | 459 | int i; |
0381411e TH |
460 | |
461 | /* alloc and init base part */ | |
462 | blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node); | |
463 | if (!blkg) | |
464 | return NULL; | |
465 | ||
c875f4d0 | 466 | blkg->q = q; |
e8989fae | 467 | INIT_LIST_HEAD(&blkg->q_node); |
1cd9e039 | 468 | INIT_LIST_HEAD(&blkg->alloc_node); |
0381411e | 469 | blkg->blkcg = blkcg; |
1adaf3dd | 470 | blkg->refcnt = 1; |
0381411e TH |
471 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); |
472 | ||
e8989fae TH |
473 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
474 | struct blkio_policy_type *pol = blkio_policy[i]; | |
475 | struct blkg_policy_data *pd; | |
0381411e | 476 | |
e8989fae TH |
477 | if (!pol) |
478 | continue; | |
479 | ||
480 | /* alloc per-policy data and attach it to blkg */ | |
481 | pd = kzalloc_node(sizeof(*pd) + pol->pdata_size, GFP_ATOMIC, | |
482 | q->node); | |
483 | if (!pd) { | |
484 | blkg_free(blkg); | |
485 | return NULL; | |
486 | } | |
549d3aa8 | 487 | |
e8989fae TH |
488 | blkg->pd[i] = pd; |
489 | pd->blkg = blkg; | |
0381411e TH |
490 | } |
491 | ||
549d3aa8 | 492 | /* invoke per-policy init */ |
e8989fae TH |
493 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
494 | struct blkio_policy_type *pol = blkio_policy[i]; | |
495 | ||
496 | if (pol) | |
497 | pol->ops.blkio_init_group_fn(blkg); | |
498 | } | |
499 | ||
0381411e TH |
500 | return blkg; |
501 | } | |
502 | ||
cd1604fa TH |
503 | struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg, |
504 | struct request_queue *q, | |
cd1604fa TH |
505 | bool for_root) |
506 | __releases(q->queue_lock) __acquires(q->queue_lock) | |
5624a4e4 | 507 | { |
1cd9e039 | 508 | struct blkio_group *blkg; |
5624a4e4 | 509 | |
cd1604fa TH |
510 | WARN_ON_ONCE(!rcu_read_lock_held()); |
511 | lockdep_assert_held(q->queue_lock); | |
512 | ||
513 | /* | |
514 | * This could be the first entry point of blkcg implementation and | |
515 | * we shouldn't allow anything to go through for a bypassing queue. | |
516 | * The following can be removed if blkg lookup is guaranteed to | |
517 | * fail on a bypassing queue. | |
518 | */ | |
519 | if (unlikely(blk_queue_bypass(q)) && !for_root) | |
520 | return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY); | |
521 | ||
e8989fae | 522 | blkg = blkg_lookup(blkcg, q); |
cd1604fa TH |
523 | if (blkg) |
524 | return blkg; | |
525 | ||
7ee9c562 | 526 | /* blkg holds a reference to blkcg */ |
cd1604fa TH |
527 | if (!css_tryget(&blkcg->css)) |
528 | return ERR_PTR(-EINVAL); | |
529 | ||
530 | /* | |
531 | * Allocate and initialize. | |
cd1604fa | 532 | */ |
1cd9e039 | 533 | blkg = blkg_alloc(blkcg, q); |
cd1604fa TH |
534 | |
535 | /* did alloc fail? */ | |
1cd9e039 | 536 | if (unlikely(!blkg)) { |
cd1604fa TH |
537 | blkg = ERR_PTR(-ENOMEM); |
538 | goto out; | |
539 | } | |
540 | ||
541 | /* insert */ | |
542 | spin_lock(&blkcg->lock); | |
31e4c28d | 543 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
e8989fae | 544 | list_add(&blkg->q_node, &q->blkg_list); |
cd1604fa | 545 | spin_unlock(&blkcg->lock); |
1cd9e039 VG |
546 | |
547 | spin_lock(&alloc_list_lock); | |
548 | list_add(&blkg->alloc_node, &alloc_list); | |
549 | /* Queue per cpu stat allocation from worker thread. */ | |
550 | queue_delayed_work(system_nrt_wq, &blkio_stat_alloc_work, 0); | |
551 | spin_unlock(&alloc_list_lock); | |
cd1604fa | 552 | out: |
cd1604fa | 553 | return blkg; |
31e4c28d | 554 | } |
cd1604fa | 555 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 556 | |
31e4c28d | 557 | /* called under rcu_read_lock(). */ |
cd1604fa | 558 | struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg, |
e8989fae | 559 | struct request_queue *q) |
31e4c28d VG |
560 | { |
561 | struct blkio_group *blkg; | |
562 | struct hlist_node *n; | |
31e4c28d | 563 | |
ca32aefc | 564 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) |
e8989fae | 565 | if (blkg->q == q) |
31e4c28d | 566 | return blkg; |
31e4c28d VG |
567 | return NULL; |
568 | } | |
cd1604fa | 569 | EXPORT_SYMBOL_GPL(blkg_lookup); |
31e4c28d | 570 | |
e8989fae | 571 | static void blkg_destroy(struct blkio_group *blkg) |
03aa264a TH |
572 | { |
573 | struct request_queue *q = blkg->q; | |
9f13ef67 | 574 | struct blkio_cgroup *blkcg = blkg->blkcg; |
03aa264a TH |
575 | |
576 | lockdep_assert_held(q->queue_lock); | |
9f13ef67 | 577 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
578 | |
579 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 580 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 581 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
e8989fae | 582 | list_del_init(&blkg->q_node); |
9f13ef67 | 583 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 584 | |
1cd9e039 VG |
585 | spin_lock(&alloc_list_lock); |
586 | list_del_init(&blkg->alloc_node); | |
587 | spin_unlock(&alloc_list_lock); | |
588 | ||
03aa264a TH |
589 | /* |
590 | * Put the reference taken at the time of creation so that when all | |
591 | * queues are gone, group can be destroyed. | |
592 | */ | |
593 | blkg_put(blkg); | |
594 | } | |
595 | ||
e8989fae TH |
596 | /* |
597 | * XXX: This updates blkg policy data in-place for root blkg, which is | |
598 | * necessary across elevator switch and policy registration as root blkgs | |
599 | * aren't shot down. This broken and racy implementation is temporary. | |
600 | * Eventually, blkg shoot down will be replaced by proper in-place update. | |
601 | */ | |
602 | void update_root_blkg_pd(struct request_queue *q, enum blkio_policy_id plid) | |
603 | { | |
604 | struct blkio_policy_type *pol = blkio_policy[plid]; | |
605 | struct blkio_group *blkg = blkg_lookup(&blkio_root_cgroup, q); | |
606 | struct blkg_policy_data *pd; | |
607 | ||
608 | if (!blkg) | |
609 | return; | |
610 | ||
611 | kfree(blkg->pd[plid]); | |
612 | blkg->pd[plid] = NULL; | |
613 | ||
614 | if (!pol) | |
615 | return; | |
616 | ||
617 | pd = kzalloc(sizeof(*pd) + pol->pdata_size, GFP_KERNEL); | |
618 | WARN_ON_ONCE(!pd); | |
619 | ||
620 | pd->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu); | |
621 | WARN_ON_ONCE(!pd->stats_cpu); | |
622 | ||
623 | blkg->pd[plid] = pd; | |
624 | pd->blkg = blkg; | |
625 | pol->ops.blkio_init_group_fn(blkg); | |
626 | } | |
627 | EXPORT_SYMBOL_GPL(update_root_blkg_pd); | |
628 | ||
9f13ef67 TH |
629 | /** |
630 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
631 | * @q: request_queue of interest | |
632 | * @destroy_root: whether to destroy root blkg or not | |
633 | * | |
634 | * Destroy blkgs associated with @q. If @destroy_root is %true, all are | |
635 | * destroyed; otherwise, root blkg is left alone. | |
636 | */ | |
e8989fae | 637 | void blkg_destroy_all(struct request_queue *q, bool destroy_root) |
72e06c25 | 638 | { |
03aa264a | 639 | struct blkio_group *blkg, *n; |
72e06c25 | 640 | |
9f13ef67 | 641 | spin_lock_irq(q->queue_lock); |
72e06c25 | 642 | |
9f13ef67 TH |
643 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
644 | struct blkio_cgroup *blkcg = blkg->blkcg; | |
72e06c25 | 645 | |
9f13ef67 TH |
646 | /* skip root? */ |
647 | if (!destroy_root && blkg->blkcg == &blkio_root_cgroup) | |
648 | continue; | |
72e06c25 | 649 | |
9f13ef67 TH |
650 | spin_lock(&blkcg->lock); |
651 | blkg_destroy(blkg); | |
652 | spin_unlock(&blkcg->lock); | |
72e06c25 | 653 | } |
9f13ef67 TH |
654 | |
655 | spin_unlock_irq(q->queue_lock); | |
72e06c25 | 656 | } |
03aa264a | 657 | EXPORT_SYMBOL_GPL(blkg_destroy_all); |
72e06c25 | 658 | |
1adaf3dd TH |
659 | static void blkg_rcu_free(struct rcu_head *rcu_head) |
660 | { | |
661 | blkg_free(container_of(rcu_head, struct blkio_group, rcu_head)); | |
662 | } | |
663 | ||
664 | void __blkg_release(struct blkio_group *blkg) | |
665 | { | |
666 | /* release the extra blkcg reference this blkg has been holding */ | |
667 | css_put(&blkg->blkcg->css); | |
668 | ||
669 | /* | |
670 | * A group is freed in rcu manner. But having an rcu lock does not | |
671 | * mean that one can access all the fields of blkg and assume these | |
672 | * are valid. For example, don't try to follow throtl_data and | |
673 | * request queue links. | |
674 | * | |
675 | * Having a reference to blkg under an rcu allows acess to only | |
676 | * values local to groups like group stats and group rate limits | |
677 | */ | |
678 | call_rcu(&blkg->rcu_head, blkg_rcu_free); | |
679 | } | |
680 | EXPORT_SYMBOL_GPL(__blkg_release); | |
681 | ||
c1768268 | 682 | static void blkio_reset_stats_cpu(struct blkio_group *blkg, int plid) |
f0bdc8cd | 683 | { |
c1768268 | 684 | struct blkg_policy_data *pd = blkg->pd[plid]; |
997a026c | 685 | int cpu; |
1cd9e039 VG |
686 | |
687 | if (pd->stats_cpu == NULL) | |
688 | return; | |
997a026c TH |
689 | |
690 | for_each_possible_cpu(cpu) { | |
691 | struct blkio_group_stats_cpu *sc = | |
692 | per_cpu_ptr(pd->stats_cpu, cpu); | |
693 | ||
edcb0722 TH |
694 | blkg_rwstat_reset(&sc->service_bytes); |
695 | blkg_rwstat_reset(&sc->serviced); | |
696 | blkg_stat_reset(&sc->sectors); | |
f0bdc8cd VG |
697 | } |
698 | } | |
699 | ||
303a3acb | 700 | static int |
84c124da | 701 | blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val) |
303a3acb | 702 | { |
997a026c | 703 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); |
303a3acb DS |
704 | struct blkio_group *blkg; |
705 | struct hlist_node *n; | |
303a3acb | 706 | |
e8989fae | 707 | spin_lock(&blkio_list_lock); |
303a3acb | 708 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
709 | |
710 | /* | |
711 | * Note that stat reset is racy - it doesn't synchronize against | |
712 | * stat updates. This is a debug feature which shouldn't exist | |
713 | * anyway. If you get hit by a race, retry. | |
714 | */ | |
303a3acb | 715 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
e8989fae | 716 | struct blkio_policy_type *pol; |
549d3aa8 | 717 | |
e8989fae TH |
718 | list_for_each_entry(pol, &blkio_list, list) { |
719 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; | |
997a026c TH |
720 | struct blkio_group_stats *stats = &pd->stats; |
721 | ||
722 | /* queued stats shouldn't be cleared */ | |
edcb0722 TH |
723 | blkg_rwstat_reset(&stats->merged); |
724 | blkg_rwstat_reset(&stats->service_time); | |
725 | blkg_rwstat_reset(&stats->wait_time); | |
726 | blkg_stat_reset(&stats->time); | |
812df48d | 727 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
edcb0722 TH |
728 | blkg_stat_reset(&stats->unaccounted_time); |
729 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
730 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
731 | blkg_stat_reset(&stats->dequeue); | |
732 | blkg_stat_reset(&stats->group_wait_time); | |
733 | blkg_stat_reset(&stats->idle_time); | |
734 | blkg_stat_reset(&stats->empty_time); | |
812df48d | 735 | #endif |
e8989fae TH |
736 | blkio_reset_stats_cpu(blkg, pol->plid); |
737 | } | |
303a3acb | 738 | } |
f0bdc8cd | 739 | |
303a3acb | 740 | spin_unlock_irq(&blkcg->lock); |
e8989fae | 741 | spin_unlock(&blkio_list_lock); |
303a3acb DS |
742 | return 0; |
743 | } | |
744 | ||
d3d32e69 | 745 | static const char *blkg_dev_name(struct blkio_group *blkg) |
303a3acb | 746 | { |
d3d32e69 TH |
747 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
748 | if (blkg->q->backing_dev_info.dev) | |
749 | return dev_name(blkg->q->backing_dev_info.dev); | |
750 | return NULL; | |
303a3acb DS |
751 | } |
752 | ||
d3d32e69 TH |
753 | /** |
754 | * blkcg_print_blkgs - helper for printing per-blkg data | |
755 | * @sf: seq_file to print to | |
756 | * @blkcg: blkcg of interest | |
757 | * @prfill: fill function to print out a blkg | |
758 | * @pol: policy in question | |
759 | * @data: data to be passed to @prfill | |
760 | * @show_total: to print out sum of prfill return values or not | |
761 | * | |
762 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
763 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
764 | * policy data and @data. If @show_total is %true, the sum of the return | |
765 | * values from @prfill is printed with "Total" label at the end. | |
766 | * | |
767 | * This is to be used to construct print functions for | |
768 | * cftype->read_seq_string method. | |
769 | */ | |
770 | static void blkcg_print_blkgs(struct seq_file *sf, struct blkio_cgroup *blkcg, | |
771 | u64 (*prfill)(struct seq_file *, | |
772 | struct blkg_policy_data *, int), | |
773 | int pol, int data, bool show_total) | |
5624a4e4 | 774 | { |
d3d32e69 TH |
775 | struct blkio_group *blkg; |
776 | struct hlist_node *n; | |
777 | u64 total = 0; | |
5624a4e4 | 778 | |
d3d32e69 TH |
779 | spin_lock_irq(&blkcg->lock); |
780 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) | |
781 | if (blkg->pd[pol]) | |
782 | total += prfill(sf, blkg->pd[pol], data); | |
783 | spin_unlock_irq(&blkcg->lock); | |
784 | ||
785 | if (show_total) | |
786 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
787 | } | |
788 | ||
789 | /** | |
790 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
791 | * @sf: seq_file to print to | |
792 | * @pd: policy data of interest | |
793 | * @v: value to print | |
794 | * | |
795 | * Print @v to @sf for the device assocaited with @pd. | |
796 | */ | |
797 | static u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, | |
798 | u64 v) | |
799 | { | |
800 | const char *dname = blkg_dev_name(pd->blkg); | |
801 | ||
802 | if (!dname) | |
803 | return 0; | |
804 | ||
805 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
806 | return v; | |
807 | } | |
808 | ||
809 | /** | |
810 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
811 | * @sf: seq_file to print to | |
812 | * @pd: policy data of interest | |
813 | * @rwstat: rwstat to print | |
814 | * | |
815 | * Print @rwstat to @sf for the device assocaited with @pd. | |
816 | */ | |
817 | static u64 __blkg_prfill_rwstat(struct seq_file *sf, | |
818 | struct blkg_policy_data *pd, | |
819 | const struct blkg_rwstat *rwstat) | |
820 | { | |
821 | static const char *rwstr[] = { | |
822 | [BLKG_RWSTAT_READ] = "Read", | |
823 | [BLKG_RWSTAT_WRITE] = "Write", | |
824 | [BLKG_RWSTAT_SYNC] = "Sync", | |
825 | [BLKG_RWSTAT_ASYNC] = "Async", | |
826 | }; | |
827 | const char *dname = blkg_dev_name(pd->blkg); | |
828 | u64 v; | |
829 | int i; | |
830 | ||
831 | if (!dname) | |
832 | return 0; | |
833 | ||
834 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
835 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
836 | (unsigned long long)rwstat->cnt[i]); | |
837 | ||
838 | v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE]; | |
839 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); | |
840 | return v; | |
841 | } | |
842 | ||
843 | static u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, | |
844 | int off) | |
845 | { | |
846 | return __blkg_prfill_u64(sf, pd, | |
847 | blkg_stat_read((void *)&pd->stats + off)); | |
848 | } | |
849 | ||
850 | static u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, | |
851 | int off) | |
852 | { | |
853 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)&pd->stats + off); | |
854 | ||
855 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
856 | } | |
857 | ||
858 | /* print blkg_stat specified by BLKCG_STAT_PRIV() */ | |
859 | static int blkcg_print_stat(struct cgroup *cgrp, struct cftype *cft, | |
860 | struct seq_file *sf) | |
861 | { | |
862 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); | |
863 | ||
864 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_stat, | |
865 | BLKCG_STAT_POL(cft->private), | |
866 | BLKCG_STAT_OFF(cft->private), false); | |
867 | return 0; | |
868 | } | |
869 | ||
870 | /* print blkg_rwstat specified by BLKCG_STAT_PRIV() */ | |
871 | static int blkcg_print_rwstat(struct cgroup *cgrp, struct cftype *cft, | |
872 | struct seq_file *sf) | |
873 | { | |
874 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); | |
875 | ||
876 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_rwstat, | |
877 | BLKCG_STAT_POL(cft->private), | |
878 | BLKCG_STAT_OFF(cft->private), true); | |
879 | return 0; | |
880 | } | |
881 | ||
882 | static u64 blkg_prfill_cpu_stat(struct seq_file *sf, | |
883 | struct blkg_policy_data *pd, int off) | |
884 | { | |
885 | u64 v = 0; | |
886 | int cpu; | |
1cd9e039 | 887 | |
5624a4e4 | 888 | for_each_possible_cpu(cpu) { |
d3d32e69 | 889 | struct blkio_group_stats_cpu *sc = |
edcb0722 | 890 | per_cpu_ptr(pd->stats_cpu, cpu); |
edcb0722 | 891 | |
d3d32e69 | 892 | v += blkg_stat_read((void *)sc + off); |
5624a4e4 VG |
893 | } |
894 | ||
d3d32e69 | 895 | return __blkg_prfill_u64(sf, pd, v); |
5624a4e4 VG |
896 | } |
897 | ||
d3d32e69 TH |
898 | static u64 blkg_prfill_cpu_rwstat(struct seq_file *sf, |
899 | struct blkg_policy_data *pd, int off) | |
5624a4e4 | 900 | { |
d3d32e69 TH |
901 | struct blkg_rwstat rwstat = { }, tmp; |
902 | int i, cpu; | |
903 | ||
904 | for_each_possible_cpu(cpu) { | |
905 | struct blkio_group_stats_cpu *sc = | |
906 | per_cpu_ptr(pd->stats_cpu, cpu); | |
5624a4e4 | 907 | |
d3d32e69 TH |
908 | tmp = blkg_rwstat_read((void *)sc + off); |
909 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
910 | rwstat.cnt[i] += tmp.cnt[i]; | |
5624a4e4 VG |
911 | } |
912 | ||
d3d32e69 TH |
913 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
914 | } | |
5624a4e4 | 915 | |
d3d32e69 TH |
916 | /* print per-cpu blkg_stat specified by BLKCG_STAT_PRIV() */ |
917 | static int blkcg_print_cpu_stat(struct cgroup *cgrp, struct cftype *cft, | |
918 | struct seq_file *sf) | |
919 | { | |
920 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); | |
921 | ||
922 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_cpu_stat, | |
923 | BLKCG_STAT_POL(cft->private), | |
924 | BLKCG_STAT_OFF(cft->private), false); | |
925 | return 0; | |
5624a4e4 VG |
926 | } |
927 | ||
d3d32e69 TH |
928 | /* print per-cpu blkg_rwstat specified by BLKCG_STAT_PRIV() */ |
929 | static int blkcg_print_cpu_rwstat(struct cgroup *cgrp, struct cftype *cft, | |
930 | struct seq_file *sf) | |
303a3acb | 931 | { |
d3d32e69 | 932 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); |
edcb0722 | 933 | |
d3d32e69 TH |
934 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_cpu_rwstat, |
935 | BLKCG_STAT_POL(cft->private), | |
936 | BLKCG_STAT_OFF(cft->private), true); | |
937 | return 0; | |
938 | } | |
303a3acb | 939 | |
d3d32e69 TH |
940 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
941 | static u64 blkg_prfill_avg_queue_size(struct seq_file *sf, | |
942 | struct blkg_policy_data *pd, int off) | |
943 | { | |
944 | u64 samples = blkg_stat_read(&pd->stats.avg_queue_size_samples); | |
945 | u64 v = 0; | |
c4c76a05 | 946 | |
d3d32e69 TH |
947 | if (samples) { |
948 | v = blkg_stat_read(&pd->stats.avg_queue_size_sum); | |
949 | do_div(v, samples); | |
edcb0722 | 950 | } |
d3d32e69 TH |
951 | __blkg_prfill_u64(sf, pd, v); |
952 | return 0; | |
953 | } | |
c4c76a05 | 954 | |
d3d32e69 TH |
955 | /* print avg_queue_size */ |
956 | static int blkcg_print_avg_queue_size(struct cgroup *cgrp, struct cftype *cft, | |
957 | struct seq_file *sf) | |
958 | { | |
959 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); | |
c4c76a05 | 960 | |
d3d32e69 TH |
961 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_avg_queue_size, |
962 | BLKIO_POLICY_PROP, 0, false); | |
963 | return 0; | |
303a3acb | 964 | } |
d3d32e69 | 965 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ |
303a3acb | 966 | |
3a8b31d3 TH |
967 | struct blkg_conf_ctx { |
968 | struct gendisk *disk; | |
969 | struct blkio_group *blkg; | |
970 | u64 v; | |
971 | }; | |
972 | ||
973 | /** | |
974 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
975 | * @blkcg: target block cgroup | |
976 | * @input: input string | |
977 | * @ctx: blkg_conf_ctx to be filled | |
978 | * | |
979 | * Parse per-blkg config update from @input and initialize @ctx with the | |
980 | * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new | |
981 | * value. This function returns with RCU read locked and must be paired | |
982 | * with blkg_conf_finish(). | |
983 | */ | |
984 | static int blkg_conf_prep(struct blkio_cgroup *blkcg, const char *input, | |
985 | struct blkg_conf_ctx *ctx) | |
986 | __acquires(rcu) | |
34d0f179 | 987 | { |
3a8b31d3 TH |
988 | struct gendisk *disk; |
989 | struct blkio_group *blkg; | |
990 | char *buf, *s[4], *p, *major_s, *minor_s; | |
d11bb446 | 991 | unsigned long major, minor; |
ece84241 TH |
992 | int i = 0, ret = -EINVAL; |
993 | int part; | |
34d0f179 | 994 | dev_t dev; |
d11bb446 | 995 | u64 temp; |
34d0f179 | 996 | |
3a8b31d3 TH |
997 | buf = kstrdup(input, GFP_KERNEL); |
998 | if (!buf) | |
999 | return -ENOMEM; | |
1000 | ||
34d0f179 GJ |
1001 | memset(s, 0, sizeof(s)); |
1002 | ||
1003 | while ((p = strsep(&buf, " ")) != NULL) { | |
1004 | if (!*p) | |
1005 | continue; | |
1006 | ||
1007 | s[i++] = p; | |
1008 | ||
1009 | /* Prevent from inputing too many things */ | |
1010 | if (i == 3) | |
1011 | break; | |
1012 | } | |
1013 | ||
1014 | if (i != 2) | |
ece84241 | 1015 | goto out; |
34d0f179 GJ |
1016 | |
1017 | p = strsep(&s[0], ":"); | |
1018 | if (p != NULL) | |
1019 | major_s = p; | |
1020 | else | |
ece84241 | 1021 | goto out; |
34d0f179 GJ |
1022 | |
1023 | minor_s = s[0]; | |
1024 | if (!minor_s) | |
ece84241 | 1025 | goto out; |
34d0f179 | 1026 | |
ece84241 TH |
1027 | if (strict_strtoul(major_s, 10, &major)) |
1028 | goto out; | |
34d0f179 | 1029 | |
ece84241 TH |
1030 | if (strict_strtoul(minor_s, 10, &minor)) |
1031 | goto out; | |
34d0f179 GJ |
1032 | |
1033 | dev = MKDEV(major, minor); | |
1034 | ||
ece84241 TH |
1035 | if (strict_strtoull(s[1], 10, &temp)) |
1036 | goto out; | |
34d0f179 | 1037 | |
e56da7e2 | 1038 | disk = get_gendisk(dev, &part); |
4bfd482e | 1039 | if (!disk || part) |
e56da7e2 | 1040 | goto out; |
e56da7e2 TH |
1041 | |
1042 | rcu_read_lock(); | |
1043 | ||
4bfd482e | 1044 | spin_lock_irq(disk->queue->queue_lock); |
aaec55a0 | 1045 | blkg = blkg_lookup_create(blkcg, disk->queue, false); |
4bfd482e | 1046 | spin_unlock_irq(disk->queue->queue_lock); |
e56da7e2 | 1047 | |
4bfd482e TH |
1048 | if (IS_ERR(blkg)) { |
1049 | ret = PTR_ERR(blkg); | |
3a8b31d3 TH |
1050 | rcu_read_unlock(); |
1051 | put_disk(disk); | |
1052 | /* | |
1053 | * If queue was bypassing, we should retry. Do so after a | |
1054 | * short msleep(). It isn't strictly necessary but queue | |
1055 | * can be bypassing for some time and it's always nice to | |
1056 | * avoid busy looping. | |
1057 | */ | |
1058 | if (ret == -EBUSY) { | |
1059 | msleep(10); | |
1060 | ret = restart_syscall(); | |
7702e8f4 | 1061 | } |
3a8b31d3 | 1062 | goto out; |
062a644d | 1063 | } |
3a8b31d3 TH |
1064 | |
1065 | ctx->disk = disk; | |
1066 | ctx->blkg = blkg; | |
1067 | ctx->v = temp; | |
ece84241 TH |
1068 | ret = 0; |
1069 | out: | |
3a8b31d3 | 1070 | kfree(buf); |
ece84241 | 1071 | return ret; |
34d0f179 GJ |
1072 | } |
1073 | ||
3a8b31d3 TH |
1074 | /** |
1075 | * blkg_conf_finish - finish up per-blkg config update | |
1076 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
1077 | * | |
1078 | * Finish up after per-blkg config update. This function must be paired | |
1079 | * with blkg_conf_prep(). | |
1080 | */ | |
1081 | static void blkg_conf_finish(struct blkg_conf_ctx *ctx) | |
1082 | __releases(rcu) | |
34d0f179 | 1083 | { |
3a8b31d3 TH |
1084 | rcu_read_unlock(); |
1085 | put_disk(ctx->disk); | |
34d0f179 GJ |
1086 | } |
1087 | ||
c4682aec TH |
1088 | /* for propio conf */ |
1089 | static u64 blkg_prfill_weight_device(struct seq_file *sf, | |
1090 | struct blkg_policy_data *pd, int off) | |
34d0f179 | 1091 | { |
c4682aec TH |
1092 | if (!pd->conf.weight) |
1093 | return 0; | |
1094 | return __blkg_prfill_u64(sf, pd, pd->conf.weight); | |
062a644d | 1095 | } |
34d0f179 | 1096 | |
c4682aec TH |
1097 | static int blkcg_print_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1098 | struct seq_file *sf) | |
34d0f179 | 1099 | { |
c4682aec TH |
1100 | blkcg_print_blkgs(sf, cgroup_to_blkio_cgroup(cgrp), |
1101 | blkg_prfill_weight_device, BLKIO_POLICY_PROP, 0, | |
1102 | false); | |
1103 | return 0; | |
062a644d VG |
1104 | } |
1105 | ||
c4682aec TH |
1106 | static int blkcg_print_weight(struct cgroup *cgrp, struct cftype *cft, |
1107 | struct seq_file *sf) | |
062a644d | 1108 | { |
c4682aec | 1109 | seq_printf(sf, "%u\n", cgroup_to_blkio_cgroup(cgrp)->weight); |
062a644d VG |
1110 | return 0; |
1111 | } | |
1112 | ||
3a8b31d3 TH |
1113 | static int blkcg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1114 | const char *buf) | |
1115 | { | |
1116 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1117 | struct blkg_policy_data *pd; | |
1118 | struct blkg_conf_ctx ctx; | |
1119 | int ret; | |
1120 | ||
1121 | ret = blkg_conf_prep(blkcg, buf, &ctx); | |
1122 | if (ret) | |
1123 | return ret; | |
1124 | ||
1125 | ret = -EINVAL; | |
1126 | pd = ctx.blkg->pd[BLKIO_POLICY_PROP]; | |
1127 | if (pd && (!ctx.v || (ctx.v >= BLKIO_WEIGHT_MIN && | |
1128 | ctx.v <= BLKIO_WEIGHT_MAX))) { | |
1129 | pd->conf.weight = ctx.v; | |
1130 | blkio_update_group_weight(ctx.blkg, BLKIO_POLICY_PROP, | |
1131 | ctx.v ?: blkcg->weight); | |
1132 | ret = 0; | |
1133 | } | |
1134 | ||
1135 | blkg_conf_finish(&ctx); | |
1136 | return ret; | |
1137 | } | |
1138 | ||
627f29f4 | 1139 | static int blkcg_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) |
062a644d | 1140 | { |
627f29f4 | 1141 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); |
062a644d VG |
1142 | struct blkio_group *blkg; |
1143 | struct hlist_node *n; | |
062a644d VG |
1144 | |
1145 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) | |
1146 | return -EINVAL; | |
1147 | ||
1148 | spin_lock(&blkio_list_lock); | |
1149 | spin_lock_irq(&blkcg->lock); | |
1150 | blkcg->weight = (unsigned int)val; | |
1151 | ||
549d3aa8 | 1152 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
627f29f4 | 1153 | struct blkg_policy_data *pd = blkg->pd[BLKIO_POLICY_PROP]; |
549d3aa8 | 1154 | |
627f29f4 TH |
1155 | if (pd && !pd->conf.weight) |
1156 | blkio_update_group_weight(blkg, BLKIO_POLICY_PROP, | |
1157 | blkcg->weight); | |
549d3aa8 | 1158 | } |
062a644d | 1159 | |
062a644d VG |
1160 | spin_unlock_irq(&blkcg->lock); |
1161 | spin_unlock(&blkio_list_lock); | |
1162 | return 0; | |
1163 | } | |
1164 | ||
c4682aec TH |
1165 | /* for blk-throttle conf */ |
1166 | #ifdef CONFIG_BLK_DEV_THROTTLING | |
1167 | static u64 blkg_prfill_conf_u64(struct seq_file *sf, | |
1168 | struct blkg_policy_data *pd, int off) | |
1169 | { | |
1170 | u64 v = *(u64 *)((void *)&pd->conf + off); | |
1171 | ||
1172 | if (!v) | |
1173 | return 0; | |
1174 | return __blkg_prfill_u64(sf, pd, v); | |
1175 | } | |
062a644d | 1176 | |
c4682aec TH |
1177 | static int blkcg_print_conf_u64(struct cgroup *cgrp, struct cftype *cft, |
1178 | struct seq_file *sf) | |
1179 | { | |
c4682aec TH |
1180 | blkcg_print_blkgs(sf, cgroup_to_blkio_cgroup(cgrp), |
1181 | blkg_prfill_conf_u64, BLKIO_POLICY_THROTL, | |
3a8b31d3 | 1182 | cft->private, false); |
062a644d VG |
1183 | return 0; |
1184 | } | |
3a8b31d3 TH |
1185 | |
1186 | static int blkcg_set_conf_u64(struct cgroup *cgrp, struct cftype *cft, | |
1187 | const char *buf, int rw, | |
1188 | void (*update)(struct blkio_group *, int, u64, int)) | |
1189 | { | |
1190 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1191 | struct blkg_policy_data *pd; | |
1192 | struct blkg_conf_ctx ctx; | |
1193 | int ret; | |
1194 | ||
1195 | ret = blkg_conf_prep(blkcg, buf, &ctx); | |
1196 | if (ret) | |
1197 | return ret; | |
1198 | ||
1199 | ret = -EINVAL; | |
1200 | pd = ctx.blkg->pd[BLKIO_POLICY_THROTL]; | |
1201 | if (pd) { | |
1202 | *(u64 *)((void *)&pd->conf + cft->private) = ctx.v; | |
1203 | update(ctx.blkg, BLKIO_POLICY_THROTL, ctx.v ?: -1, rw); | |
1204 | ret = 0; | |
1205 | } | |
1206 | ||
1207 | blkg_conf_finish(&ctx); | |
1208 | return ret; | |
1209 | } | |
1210 | ||
1211 | static int blkcg_set_conf_bps_r(struct cgroup *cgrp, struct cftype *cft, | |
1212 | const char *buf) | |
1213 | { | |
1214 | return blkcg_set_conf_u64(cgrp, cft, buf, READ, blkio_update_group_bps); | |
1215 | } | |
1216 | ||
1217 | static int blkcg_set_conf_bps_w(struct cgroup *cgrp, struct cftype *cft, | |
1218 | const char *buf) | |
1219 | { | |
1220 | return blkcg_set_conf_u64(cgrp, cft, buf, WRITE, blkio_update_group_bps); | |
1221 | } | |
1222 | ||
1223 | static int blkcg_set_conf_iops_r(struct cgroup *cgrp, struct cftype *cft, | |
1224 | const char *buf) | |
1225 | { | |
1226 | return blkcg_set_conf_u64(cgrp, cft, buf, READ, blkio_update_group_iops); | |
1227 | } | |
1228 | ||
1229 | static int blkcg_set_conf_iops_w(struct cgroup *cgrp, struct cftype *cft, | |
1230 | const char *buf) | |
1231 | { | |
1232 | return blkcg_set_conf_u64(cgrp, cft, buf, WRITE, blkio_update_group_iops); | |
1233 | } | |
c4682aec | 1234 | #endif |
062a644d | 1235 | |
31e4c28d | 1236 | struct cftype blkio_files[] = { |
34d0f179 GJ |
1237 | { |
1238 | .name = "weight_device", | |
c4682aec | 1239 | .read_seq_string = blkcg_print_weight_device, |
3a8b31d3 | 1240 | .write_string = blkcg_set_weight_device, |
34d0f179 GJ |
1241 | .max_write_len = 256, |
1242 | }, | |
31e4c28d VG |
1243 | { |
1244 | .name = "weight", | |
c4682aec | 1245 | .read_seq_string = blkcg_print_weight, |
627f29f4 | 1246 | .write_u64 = blkcg_set_weight, |
31e4c28d | 1247 | }, |
22084190 VG |
1248 | { |
1249 | .name = "time", | |
d3d32e69 TH |
1250 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1251 | offsetof(struct blkio_group_stats, time)), | |
1252 | .read_seq_string = blkcg_print_stat, | |
22084190 VG |
1253 | }, |
1254 | { | |
1255 | .name = "sectors", | |
d3d32e69 TH |
1256 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1257 | offsetof(struct blkio_group_stats_cpu, sectors)), | |
1258 | .read_seq_string = blkcg_print_cpu_stat, | |
303a3acb DS |
1259 | }, |
1260 | { | |
1261 | .name = "io_service_bytes", | |
d3d32e69 TH |
1262 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1263 | offsetof(struct blkio_group_stats_cpu, service_bytes)), | |
1264 | .read_seq_string = blkcg_print_cpu_rwstat, | |
303a3acb DS |
1265 | }, |
1266 | { | |
1267 | .name = "io_serviced", | |
d3d32e69 TH |
1268 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1269 | offsetof(struct blkio_group_stats_cpu, serviced)), | |
1270 | .read_seq_string = blkcg_print_cpu_rwstat, | |
303a3acb DS |
1271 | }, |
1272 | { | |
1273 | .name = "io_service_time", | |
d3d32e69 TH |
1274 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1275 | offsetof(struct blkio_group_stats, service_time)), | |
1276 | .read_seq_string = blkcg_print_rwstat, | |
303a3acb DS |
1277 | }, |
1278 | { | |
1279 | .name = "io_wait_time", | |
d3d32e69 TH |
1280 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1281 | offsetof(struct blkio_group_stats, wait_time)), | |
1282 | .read_seq_string = blkcg_print_rwstat, | |
84c124da | 1283 | }, |
812d4026 DS |
1284 | { |
1285 | .name = "io_merged", | |
d3d32e69 TH |
1286 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1287 | offsetof(struct blkio_group_stats, merged)), | |
1288 | .read_seq_string = blkcg_print_rwstat, | |
812d4026 | 1289 | }, |
cdc1184c DS |
1290 | { |
1291 | .name = "io_queued", | |
d3d32e69 TH |
1292 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1293 | offsetof(struct blkio_group_stats, queued)), | |
1294 | .read_seq_string = blkcg_print_rwstat, | |
cdc1184c | 1295 | }, |
84c124da DS |
1296 | { |
1297 | .name = "reset_stats", | |
1298 | .write_u64 = blkiocg_reset_stats, | |
22084190 | 1299 | }, |
13f98250 VG |
1300 | #ifdef CONFIG_BLK_DEV_THROTTLING |
1301 | { | |
1302 | .name = "throttle.read_bps_device", | |
3a8b31d3 | 1303 | .private = offsetof(struct blkio_group_conf, bps[READ]), |
c4682aec | 1304 | .read_seq_string = blkcg_print_conf_u64, |
3a8b31d3 | 1305 | .write_string = blkcg_set_conf_bps_r, |
13f98250 VG |
1306 | .max_write_len = 256, |
1307 | }, | |
1308 | ||
1309 | { | |
1310 | .name = "throttle.write_bps_device", | |
3a8b31d3 | 1311 | .private = offsetof(struct blkio_group_conf, bps[WRITE]), |
c4682aec | 1312 | .read_seq_string = blkcg_print_conf_u64, |
3a8b31d3 | 1313 | .write_string = blkcg_set_conf_bps_w, |
13f98250 VG |
1314 | .max_write_len = 256, |
1315 | }, | |
1316 | ||
1317 | { | |
1318 | .name = "throttle.read_iops_device", | |
3a8b31d3 | 1319 | .private = offsetof(struct blkio_group_conf, iops[READ]), |
c4682aec | 1320 | .read_seq_string = blkcg_print_conf_u64, |
3a8b31d3 | 1321 | .write_string = blkcg_set_conf_iops_r, |
13f98250 VG |
1322 | .max_write_len = 256, |
1323 | }, | |
1324 | ||
1325 | { | |
1326 | .name = "throttle.write_iops_device", | |
3a8b31d3 | 1327 | .private = offsetof(struct blkio_group_conf, iops[WRITE]), |
c4682aec | 1328 | .read_seq_string = blkcg_print_conf_u64, |
3a8b31d3 | 1329 | .write_string = blkcg_set_conf_iops_w, |
13f98250 VG |
1330 | .max_write_len = 256, |
1331 | }, | |
1332 | { | |
1333 | .name = "throttle.io_service_bytes", | |
d3d32e69 TH |
1334 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_THROTL, |
1335 | offsetof(struct blkio_group_stats_cpu, service_bytes)), | |
1336 | .read_seq_string = blkcg_print_cpu_rwstat, | |
13f98250 VG |
1337 | }, |
1338 | { | |
1339 | .name = "throttle.io_serviced", | |
d3d32e69 TH |
1340 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_THROTL, |
1341 | offsetof(struct blkio_group_stats_cpu, serviced)), | |
1342 | .read_seq_string = blkcg_print_cpu_rwstat, | |
13f98250 VG |
1343 | }, |
1344 | #endif /* CONFIG_BLK_DEV_THROTTLING */ | |
1345 | ||
22084190 | 1346 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
cdc1184c DS |
1347 | { |
1348 | .name = "avg_queue_size", | |
d3d32e69 | 1349 | .read_seq_string = blkcg_print_avg_queue_size, |
cdc1184c | 1350 | }, |
812df48d DS |
1351 | { |
1352 | .name = "group_wait_time", | |
d3d32e69 TH |
1353 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1354 | offsetof(struct blkio_group_stats, group_wait_time)), | |
1355 | .read_seq_string = blkcg_print_stat, | |
812df48d DS |
1356 | }, |
1357 | { | |
1358 | .name = "idle_time", | |
d3d32e69 TH |
1359 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1360 | offsetof(struct blkio_group_stats, idle_time)), | |
1361 | .read_seq_string = blkcg_print_stat, | |
812df48d DS |
1362 | }, |
1363 | { | |
1364 | .name = "empty_time", | |
d3d32e69 TH |
1365 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1366 | offsetof(struct blkio_group_stats, empty_time)), | |
1367 | .read_seq_string = blkcg_print_stat, | |
812df48d | 1368 | }, |
cdc1184c | 1369 | { |
22084190 | 1370 | .name = "dequeue", |
d3d32e69 TH |
1371 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1372 | offsetof(struct blkio_group_stats, dequeue)), | |
1373 | .read_seq_string = blkcg_print_stat, | |
cdc1184c | 1374 | }, |
9026e521 JT |
1375 | { |
1376 | .name = "unaccounted_time", | |
d3d32e69 TH |
1377 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
1378 | offsetof(struct blkio_group_stats, unaccounted_time)), | |
1379 | .read_seq_string = blkcg_print_stat, | |
9026e521 | 1380 | }, |
22084190 | 1381 | #endif |
4baf6e33 | 1382 | { } /* terminate */ |
31e4c28d VG |
1383 | }; |
1384 | ||
9f13ef67 TH |
1385 | /** |
1386 | * blkiocg_pre_destroy - cgroup pre_destroy callback | |
9f13ef67 TH |
1387 | * @cgroup: cgroup of interest |
1388 | * | |
1389 | * This function is called when @cgroup is about to go away and responsible | |
1390 | * for shooting down all blkgs associated with @cgroup. blkgs should be | |
1391 | * removed while holding both q and blkcg locks. As blkcg lock is nested | |
1392 | * inside q lock, this function performs reverse double lock dancing. | |
1393 | * | |
1394 | * This is the blkcg counterpart of ioc_release_fn(). | |
1395 | */ | |
959d851c | 1396 | static int blkiocg_pre_destroy(struct cgroup *cgroup) |
31e4c28d VG |
1397 | { |
1398 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
b1c35769 | 1399 | |
9f13ef67 | 1400 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1401 | |
9f13ef67 TH |
1402 | while (!hlist_empty(&blkcg->blkg_list)) { |
1403 | struct blkio_group *blkg = hlist_entry(blkcg->blkg_list.first, | |
1404 | struct blkio_group, blkcg_node); | |
c875f4d0 | 1405 | struct request_queue *q = blkg->q; |
b1c35769 | 1406 | |
9f13ef67 TH |
1407 | if (spin_trylock(q->queue_lock)) { |
1408 | blkg_destroy(blkg); | |
1409 | spin_unlock(q->queue_lock); | |
1410 | } else { | |
1411 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 1412 | cpu_relax(); |
a5567932 | 1413 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 1414 | } |
9f13ef67 | 1415 | } |
b1c35769 | 1416 | |
9f13ef67 | 1417 | spin_unlock_irq(&blkcg->lock); |
7ee9c562 TH |
1418 | return 0; |
1419 | } | |
1420 | ||
959d851c | 1421 | static void blkiocg_destroy(struct cgroup *cgroup) |
7ee9c562 TH |
1422 | { |
1423 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
1424 | ||
67523c48 BB |
1425 | if (blkcg != &blkio_root_cgroup) |
1426 | kfree(blkcg); | |
31e4c28d VG |
1427 | } |
1428 | ||
761b3ef5 | 1429 | static struct cgroup_subsys_state *blkiocg_create(struct cgroup *cgroup) |
31e4c28d | 1430 | { |
9a9e8a26 | 1431 | static atomic64_t id_seq = ATOMIC64_INIT(0); |
0341509f LZ |
1432 | struct blkio_cgroup *blkcg; |
1433 | struct cgroup *parent = cgroup->parent; | |
31e4c28d | 1434 | |
0341509f | 1435 | if (!parent) { |
31e4c28d VG |
1436 | blkcg = &blkio_root_cgroup; |
1437 | goto done; | |
1438 | } | |
1439 | ||
31e4c28d VG |
1440 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
1441 | if (!blkcg) | |
1442 | return ERR_PTR(-ENOMEM); | |
1443 | ||
1444 | blkcg->weight = BLKIO_WEIGHT_DEFAULT; | |
9a9e8a26 | 1445 | blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */ |
31e4c28d VG |
1446 | done: |
1447 | spin_lock_init(&blkcg->lock); | |
1448 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
1449 | ||
1450 | return &blkcg->css; | |
1451 | } | |
1452 | ||
5efd6113 TH |
1453 | /** |
1454 | * blkcg_init_queue - initialize blkcg part of request queue | |
1455 | * @q: request_queue to initialize | |
1456 | * | |
1457 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1458 | * part of new request_queue @q. | |
1459 | * | |
1460 | * RETURNS: | |
1461 | * 0 on success, -errno on failure. | |
1462 | */ | |
1463 | int blkcg_init_queue(struct request_queue *q) | |
1464 | { | |
923adde1 TH |
1465 | int ret; |
1466 | ||
5efd6113 TH |
1467 | might_sleep(); |
1468 | ||
923adde1 TH |
1469 | ret = blk_throtl_init(q); |
1470 | if (ret) | |
1471 | return ret; | |
1472 | ||
1473 | mutex_lock(&all_q_mutex); | |
1474 | INIT_LIST_HEAD(&q->all_q_node); | |
1475 | list_add_tail(&q->all_q_node, &all_q_list); | |
1476 | mutex_unlock(&all_q_mutex); | |
1477 | ||
1478 | return 0; | |
5efd6113 TH |
1479 | } |
1480 | ||
1481 | /** | |
1482 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1483 | * @q: request_queue to drain | |
1484 | * | |
1485 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1486 | */ | |
1487 | void blkcg_drain_queue(struct request_queue *q) | |
1488 | { | |
1489 | lockdep_assert_held(q->queue_lock); | |
1490 | ||
1491 | blk_throtl_drain(q); | |
1492 | } | |
1493 | ||
1494 | /** | |
1495 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1496 | * @q: request_queue being released | |
1497 | * | |
1498 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1499 | */ | |
1500 | void blkcg_exit_queue(struct request_queue *q) | |
1501 | { | |
923adde1 TH |
1502 | mutex_lock(&all_q_mutex); |
1503 | list_del_init(&q->all_q_node); | |
1504 | mutex_unlock(&all_q_mutex); | |
1505 | ||
e8989fae TH |
1506 | blkg_destroy_all(q, true); |
1507 | ||
5efd6113 TH |
1508 | blk_throtl_exit(q); |
1509 | } | |
1510 | ||
31e4c28d VG |
1511 | /* |
1512 | * We cannot support shared io contexts, as we have no mean to support | |
1513 | * two tasks with the same ioc in two different groups without major rework | |
1514 | * of the main cic data structures. For now we allow a task to change | |
1515 | * its cgroup only if it's the only owner of its ioc. | |
1516 | */ | |
761b3ef5 | 1517 | static int blkiocg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
31e4c28d | 1518 | { |
bb9d97b6 | 1519 | struct task_struct *task; |
31e4c28d VG |
1520 | struct io_context *ioc; |
1521 | int ret = 0; | |
1522 | ||
1523 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
bb9d97b6 TH |
1524 | cgroup_taskset_for_each(task, cgrp, tset) { |
1525 | task_lock(task); | |
1526 | ioc = task->io_context; | |
1527 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1528 | ret = -EINVAL; | |
1529 | task_unlock(task); | |
1530 | if (ret) | |
1531 | break; | |
1532 | } | |
31e4c28d VG |
1533 | return ret; |
1534 | } | |
1535 | ||
923adde1 TH |
1536 | static void blkcg_bypass_start(void) |
1537 | __acquires(&all_q_mutex) | |
1538 | { | |
1539 | struct request_queue *q; | |
1540 | ||
1541 | mutex_lock(&all_q_mutex); | |
1542 | ||
1543 | list_for_each_entry(q, &all_q_list, all_q_node) { | |
1544 | blk_queue_bypass_start(q); | |
e8989fae | 1545 | blkg_destroy_all(q, false); |
923adde1 TH |
1546 | } |
1547 | } | |
1548 | ||
1549 | static void blkcg_bypass_end(void) | |
1550 | __releases(&all_q_mutex) | |
1551 | { | |
1552 | struct request_queue *q; | |
1553 | ||
1554 | list_for_each_entry(q, &all_q_list, all_q_node) | |
1555 | blk_queue_bypass_end(q); | |
1556 | ||
1557 | mutex_unlock(&all_q_mutex); | |
1558 | } | |
1559 | ||
676f7c8f TH |
1560 | struct cgroup_subsys blkio_subsys = { |
1561 | .name = "blkio", | |
1562 | .create = blkiocg_create, | |
1563 | .can_attach = blkiocg_can_attach, | |
959d851c | 1564 | .pre_destroy = blkiocg_pre_destroy, |
676f7c8f | 1565 | .destroy = blkiocg_destroy, |
676f7c8f | 1566 | .subsys_id = blkio_subsys_id, |
4baf6e33 | 1567 | .base_cftypes = blkio_files, |
676f7c8f TH |
1568 | .module = THIS_MODULE, |
1569 | }; | |
1570 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
1571 | ||
3e252066 VG |
1572 | void blkio_policy_register(struct blkio_policy_type *blkiop) |
1573 | { | |
e8989fae TH |
1574 | struct request_queue *q; |
1575 | ||
923adde1 | 1576 | blkcg_bypass_start(); |
3e252066 | 1577 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1578 | |
1579 | BUG_ON(blkio_policy[blkiop->plid]); | |
1580 | blkio_policy[blkiop->plid] = blkiop; | |
3e252066 | 1581 | list_add_tail(&blkiop->list, &blkio_list); |
035d10b2 | 1582 | |
3e252066 | 1583 | spin_unlock(&blkio_list_lock); |
e8989fae TH |
1584 | list_for_each_entry(q, &all_q_list, all_q_node) |
1585 | update_root_blkg_pd(q, blkiop->plid); | |
923adde1 | 1586 | blkcg_bypass_end(); |
3e252066 VG |
1587 | } |
1588 | EXPORT_SYMBOL_GPL(blkio_policy_register); | |
1589 | ||
1590 | void blkio_policy_unregister(struct blkio_policy_type *blkiop) | |
1591 | { | |
e8989fae TH |
1592 | struct request_queue *q; |
1593 | ||
923adde1 | 1594 | blkcg_bypass_start(); |
3e252066 | 1595 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1596 | |
1597 | BUG_ON(blkio_policy[blkiop->plid] != blkiop); | |
1598 | blkio_policy[blkiop->plid] = NULL; | |
3e252066 | 1599 | list_del_init(&blkiop->list); |
035d10b2 | 1600 | |
3e252066 | 1601 | spin_unlock(&blkio_list_lock); |
e8989fae TH |
1602 | list_for_each_entry(q, &all_q_list, all_q_node) |
1603 | update_root_blkg_pd(q, blkiop->plid); | |
923adde1 | 1604 | blkcg_bypass_end(); |
3e252066 VG |
1605 | } |
1606 | EXPORT_SYMBOL_GPL(blkio_policy_unregister); |