]>
Commit | Line | Data |
---|---|---|
31e4c28d VG |
1 | /* |
2 | * Common Block IO controller cgroup interface | |
3 | * | |
4 | * Based on ideas and code from CFQ, CFS and BFQ: | |
5 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> | |
6 | * | |
7 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> | |
8 | * Paolo Valente <paolo.valente@unimore.it> | |
9 | * | |
10 | * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> | |
11 | * Nauman Rafique <nauman@google.com> | |
12 | */ | |
13 | #include <linux/ioprio.h> | |
22084190 VG |
14 | #include <linux/seq_file.h> |
15 | #include <linux/kdev_t.h> | |
9d6a986c | 16 | #include <linux/module.h> |
accee785 | 17 | #include <linux/err.h> |
9195291e | 18 | #include <linux/blkdev.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
34d0f179 | 20 | #include <linux/genhd.h> |
72e06c25 TH |
21 | #include <linux/delay.h> |
22 | #include "blk-cgroup.h" | |
5efd6113 | 23 | #include "blk.h" |
3e252066 | 24 | |
84c124da DS |
25 | #define MAX_KEY_LEN 100 |
26 | ||
3e252066 VG |
27 | static DEFINE_SPINLOCK(blkio_list_lock); |
28 | static LIST_HEAD(blkio_list); | |
b1c35769 | 29 | |
923adde1 TH |
30 | static DEFINE_MUTEX(all_q_mutex); |
31 | static LIST_HEAD(all_q_list); | |
32 | ||
31e4c28d | 33 | struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT }; |
9d6a986c VG |
34 | EXPORT_SYMBOL_GPL(blkio_root_cgroup); |
35 | ||
035d10b2 TH |
36 | static struct blkio_policy_type *blkio_policy[BLKIO_NR_POLICIES]; |
37 | ||
67523c48 BB |
38 | static struct cgroup_subsys_state *blkiocg_create(struct cgroup_subsys *, |
39 | struct cgroup *); | |
bb9d97b6 TH |
40 | static int blkiocg_can_attach(struct cgroup_subsys *, struct cgroup *, |
41 | struct cgroup_taskset *); | |
42 | static void blkiocg_attach(struct cgroup_subsys *, struct cgroup *, | |
43 | struct cgroup_taskset *); | |
7ee9c562 | 44 | static int blkiocg_pre_destroy(struct cgroup_subsys *, struct cgroup *); |
67523c48 BB |
45 | static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *); |
46 | static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *); | |
47 | ||
062a644d VG |
48 | /* for encoding cft->private value on file */ |
49 | #define BLKIOFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
50 | /* What policy owns the file, proportional or throttle */ | |
51 | #define BLKIOFILE_POLICY(val) (((val) >> 16) & 0xffff) | |
52 | #define BLKIOFILE_ATTR(val) ((val) & 0xffff) | |
53 | ||
67523c48 BB |
54 | struct cgroup_subsys blkio_subsys = { |
55 | .name = "blkio", | |
56 | .create = blkiocg_create, | |
bb9d97b6 TH |
57 | .can_attach = blkiocg_can_attach, |
58 | .attach = blkiocg_attach, | |
7ee9c562 | 59 | .pre_destroy = blkiocg_pre_destroy, |
67523c48 BB |
60 | .destroy = blkiocg_destroy, |
61 | .populate = blkiocg_populate, | |
67523c48 | 62 | .subsys_id = blkio_subsys_id, |
67523c48 BB |
63 | .module = THIS_MODULE, |
64 | }; | |
65 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
66 | ||
31e4c28d VG |
67 | struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup) |
68 | { | |
69 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), | |
70 | struct blkio_cgroup, css); | |
71 | } | |
9d6a986c | 72 | EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup); |
31e4c28d | 73 | |
70087dc3 VG |
74 | struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk) |
75 | { | |
76 | return container_of(task_subsys_state(tsk, blkio_subsys_id), | |
77 | struct blkio_cgroup, css); | |
78 | } | |
79 | EXPORT_SYMBOL_GPL(task_blkio_cgroup); | |
80 | ||
062a644d VG |
81 | static inline void |
82 | blkio_update_group_weight(struct blkio_group *blkg, unsigned int weight) | |
83 | { | |
84 | struct blkio_policy_type *blkiop; | |
85 | ||
86 | list_for_each_entry(blkiop, &blkio_list, list) { | |
87 | /* If this policy does not own the blkg, do not send updates */ | |
88 | if (blkiop->plid != blkg->plid) | |
89 | continue; | |
90 | if (blkiop->ops.blkio_update_group_weight_fn) | |
ca32aefc | 91 | blkiop->ops.blkio_update_group_weight_fn(blkg->q, |
fe071437 | 92 | blkg, weight); |
062a644d VG |
93 | } |
94 | } | |
95 | ||
4c9eefa1 VG |
96 | static inline void blkio_update_group_bps(struct blkio_group *blkg, u64 bps, |
97 | int fileid) | |
98 | { | |
99 | struct blkio_policy_type *blkiop; | |
100 | ||
101 | list_for_each_entry(blkiop, &blkio_list, list) { | |
102 | ||
103 | /* If this policy does not own the blkg, do not send updates */ | |
104 | if (blkiop->plid != blkg->plid) | |
105 | continue; | |
106 | ||
107 | if (fileid == BLKIO_THROTL_read_bps_device | |
108 | && blkiop->ops.blkio_update_group_read_bps_fn) | |
ca32aefc | 109 | blkiop->ops.blkio_update_group_read_bps_fn(blkg->q, |
fe071437 | 110 | blkg, bps); |
4c9eefa1 VG |
111 | |
112 | if (fileid == BLKIO_THROTL_write_bps_device | |
113 | && blkiop->ops.blkio_update_group_write_bps_fn) | |
ca32aefc | 114 | blkiop->ops.blkio_update_group_write_bps_fn(blkg->q, |
fe071437 | 115 | blkg, bps); |
4c9eefa1 VG |
116 | } |
117 | } | |
118 | ||
7702e8f4 VG |
119 | static inline void blkio_update_group_iops(struct blkio_group *blkg, |
120 | unsigned int iops, int fileid) | |
121 | { | |
122 | struct blkio_policy_type *blkiop; | |
123 | ||
124 | list_for_each_entry(blkiop, &blkio_list, list) { | |
125 | ||
126 | /* If this policy does not own the blkg, do not send updates */ | |
127 | if (blkiop->plid != blkg->plid) | |
128 | continue; | |
129 | ||
130 | if (fileid == BLKIO_THROTL_read_iops_device | |
131 | && blkiop->ops.blkio_update_group_read_iops_fn) | |
ca32aefc | 132 | blkiop->ops.blkio_update_group_read_iops_fn(blkg->q, |
fe071437 | 133 | blkg, iops); |
7702e8f4 VG |
134 | |
135 | if (fileid == BLKIO_THROTL_write_iops_device | |
136 | && blkiop->ops.blkio_update_group_write_iops_fn) | |
ca32aefc | 137 | blkiop->ops.blkio_update_group_write_iops_fn(blkg->q, |
fe071437 | 138 | blkg,iops); |
7702e8f4 VG |
139 | } |
140 | } | |
141 | ||
9195291e DS |
142 | /* |
143 | * Add to the appropriate stat variable depending on the request type. | |
144 | * This should be called with the blkg->stats_lock held. | |
145 | */ | |
84c124da DS |
146 | static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction, |
147 | bool sync) | |
9195291e | 148 | { |
84c124da DS |
149 | if (direction) |
150 | stat[BLKIO_STAT_WRITE] += add; | |
9195291e | 151 | else |
84c124da DS |
152 | stat[BLKIO_STAT_READ] += add; |
153 | if (sync) | |
154 | stat[BLKIO_STAT_SYNC] += add; | |
9195291e | 155 | else |
84c124da | 156 | stat[BLKIO_STAT_ASYNC] += add; |
9195291e DS |
157 | } |
158 | ||
cdc1184c DS |
159 | /* |
160 | * Decrements the appropriate stat variable if non-zero depending on the | |
161 | * request type. Panics on value being zero. | |
162 | * This should be called with the blkg->stats_lock held. | |
163 | */ | |
164 | static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync) | |
165 | { | |
166 | if (direction) { | |
167 | BUG_ON(stat[BLKIO_STAT_WRITE] == 0); | |
168 | stat[BLKIO_STAT_WRITE]--; | |
169 | } else { | |
170 | BUG_ON(stat[BLKIO_STAT_READ] == 0); | |
171 | stat[BLKIO_STAT_READ]--; | |
172 | } | |
173 | if (sync) { | |
174 | BUG_ON(stat[BLKIO_STAT_SYNC] == 0); | |
175 | stat[BLKIO_STAT_SYNC]--; | |
176 | } else { | |
177 | BUG_ON(stat[BLKIO_STAT_ASYNC] == 0); | |
178 | stat[BLKIO_STAT_ASYNC]--; | |
179 | } | |
180 | } | |
181 | ||
182 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
812df48d DS |
183 | /* This should be called with the blkg->stats_lock held. */ |
184 | static void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
185 | struct blkio_group *curr_blkg) | |
186 | { | |
549d3aa8 TH |
187 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
188 | ||
189 | if (blkio_blkg_waiting(&pd->stats)) | |
812df48d DS |
190 | return; |
191 | if (blkg == curr_blkg) | |
192 | return; | |
549d3aa8 TH |
193 | pd->stats.start_group_wait_time = sched_clock(); |
194 | blkio_mark_blkg_waiting(&pd->stats); | |
812df48d DS |
195 | } |
196 | ||
197 | /* This should be called with the blkg->stats_lock held. */ | |
198 | static void blkio_update_group_wait_time(struct blkio_group_stats *stats) | |
199 | { | |
200 | unsigned long long now; | |
201 | ||
202 | if (!blkio_blkg_waiting(stats)) | |
203 | return; | |
204 | ||
205 | now = sched_clock(); | |
206 | if (time_after64(now, stats->start_group_wait_time)) | |
207 | stats->group_wait_time += now - stats->start_group_wait_time; | |
208 | blkio_clear_blkg_waiting(stats); | |
209 | } | |
210 | ||
211 | /* This should be called with the blkg->stats_lock held. */ | |
212 | static void blkio_end_empty_time(struct blkio_group_stats *stats) | |
213 | { | |
214 | unsigned long long now; | |
215 | ||
216 | if (!blkio_blkg_empty(stats)) | |
217 | return; | |
218 | ||
219 | now = sched_clock(); | |
220 | if (time_after64(now, stats->start_empty_time)) | |
221 | stats->empty_time += now - stats->start_empty_time; | |
222 | blkio_clear_blkg_empty(stats); | |
223 | } | |
224 | ||
225 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg) | |
226 | { | |
549d3aa8 | 227 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
812df48d DS |
228 | unsigned long flags; |
229 | ||
230 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 TH |
231 | BUG_ON(blkio_blkg_idling(&pd->stats)); |
232 | pd->stats.start_idle_time = sched_clock(); | |
233 | blkio_mark_blkg_idling(&pd->stats); | |
812df48d DS |
234 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
235 | } | |
236 | EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats); | |
237 | ||
238 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg) | |
239 | { | |
549d3aa8 | 240 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
812df48d DS |
241 | unsigned long flags; |
242 | unsigned long long now; | |
243 | struct blkio_group_stats *stats; | |
244 | ||
245 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 246 | stats = &pd->stats; |
812df48d DS |
247 | if (blkio_blkg_idling(stats)) { |
248 | now = sched_clock(); | |
249 | if (time_after64(now, stats->start_idle_time)) | |
250 | stats->idle_time += now - stats->start_idle_time; | |
251 | blkio_clear_blkg_idling(stats); | |
252 | } | |
253 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
254 | } | |
255 | EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats); | |
256 | ||
a11cdaa7 | 257 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg) |
cdc1184c | 258 | { |
549d3aa8 | 259 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
cdc1184c DS |
260 | unsigned long flags; |
261 | struct blkio_group_stats *stats; | |
262 | ||
263 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 264 | stats = &pd->stats; |
cdc1184c DS |
265 | stats->avg_queue_size_sum += |
266 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] + | |
267 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]; | |
268 | stats->avg_queue_size_samples++; | |
812df48d | 269 | blkio_update_group_wait_time(stats); |
cdc1184c DS |
270 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
271 | } | |
a11cdaa7 DS |
272 | EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats); |
273 | ||
e5ff082e | 274 | void blkiocg_set_start_empty_time(struct blkio_group *blkg) |
28baf442 | 275 | { |
549d3aa8 | 276 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
28baf442 DS |
277 | unsigned long flags; |
278 | struct blkio_group_stats *stats; | |
279 | ||
280 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 281 | stats = &pd->stats; |
28baf442 DS |
282 | |
283 | if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] || | |
284 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) { | |
285 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
286 | return; | |
287 | } | |
288 | ||
289 | /* | |
e5ff082e VG |
290 | * group is already marked empty. This can happen if cfqq got new |
291 | * request in parent group and moved to this group while being added | |
292 | * to service tree. Just ignore the event and move on. | |
28baf442 | 293 | */ |
e5ff082e VG |
294 | if(blkio_blkg_empty(stats)) { |
295 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
296 | return; | |
297 | } | |
298 | ||
28baf442 DS |
299 | stats->start_empty_time = sched_clock(); |
300 | blkio_mark_blkg_empty(stats); | |
301 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
302 | } | |
303 | EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time); | |
304 | ||
a11cdaa7 DS |
305 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
306 | unsigned long dequeue) | |
307 | { | |
549d3aa8 TH |
308 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
309 | ||
310 | pd->stats.dequeue += dequeue; | |
a11cdaa7 DS |
311 | } |
312 | EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats); | |
812df48d DS |
313 | #else |
314 | static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
315 | struct blkio_group *curr_blkg) {} | |
316 | static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {} | |
cdc1184c DS |
317 | #endif |
318 | ||
a11cdaa7 | 319 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c DS |
320 | struct blkio_group *curr_blkg, bool direction, |
321 | bool sync) | |
322 | { | |
549d3aa8 | 323 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
cdc1184c DS |
324 | unsigned long flags; |
325 | ||
326 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 327 | blkio_add_stat(pd->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction, |
cdc1184c | 328 | sync); |
549d3aa8 | 329 | blkio_end_empty_time(&pd->stats); |
812df48d | 330 | blkio_set_start_group_wait_time(blkg, curr_blkg); |
cdc1184c DS |
331 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
332 | } | |
a11cdaa7 | 333 | EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats); |
cdc1184c | 334 | |
a11cdaa7 | 335 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c DS |
336 | bool direction, bool sync) |
337 | { | |
549d3aa8 | 338 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
cdc1184c DS |
339 | unsigned long flags; |
340 | ||
341 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 342 | blkio_check_and_dec_stat(pd->stats.stat_arr[BLKIO_STAT_QUEUED], |
cdc1184c DS |
343 | direction, sync); |
344 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
345 | } | |
a11cdaa7 | 346 | EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats); |
cdc1184c | 347 | |
167400d3 JT |
348 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time, |
349 | unsigned long unaccounted_time) | |
22084190 | 350 | { |
549d3aa8 | 351 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
303a3acb DS |
352 | unsigned long flags; |
353 | ||
354 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 355 | pd->stats.time += time; |
a23e6869 | 356 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
549d3aa8 | 357 | pd->stats.unaccounted_time += unaccounted_time; |
a23e6869 | 358 | #endif |
303a3acb | 359 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
22084190 | 360 | } |
303a3acb | 361 | EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used); |
22084190 | 362 | |
5624a4e4 VG |
363 | /* |
364 | * should be called under rcu read lock or queue lock to make sure blkg pointer | |
365 | * is valid. | |
366 | */ | |
84c124da DS |
367 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
368 | uint64_t bytes, bool direction, bool sync) | |
9195291e | 369 | { |
549d3aa8 | 370 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
5624a4e4 | 371 | struct blkio_group_stats_cpu *stats_cpu; |
575969a0 VG |
372 | unsigned long flags; |
373 | ||
374 | /* | |
375 | * Disabling interrupts to provide mutual exclusion between two | |
376 | * writes on same cpu. It probably is not needed for 64bit. Not | |
377 | * optimizing that case yet. | |
378 | */ | |
379 | local_irq_save(flags); | |
9195291e | 380 | |
549d3aa8 | 381 | stats_cpu = this_cpu_ptr(pd->stats_cpu); |
5624a4e4 | 382 | |
575969a0 | 383 | u64_stats_update_begin(&stats_cpu->syncp); |
5624a4e4 VG |
384 | stats_cpu->sectors += bytes >> 9; |
385 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED], | |
386 | 1, direction, sync); | |
387 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES], | |
388 | bytes, direction, sync); | |
575969a0 VG |
389 | u64_stats_update_end(&stats_cpu->syncp); |
390 | local_irq_restore(flags); | |
9195291e | 391 | } |
84c124da | 392 | EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats); |
9195291e | 393 | |
84c124da DS |
394 | void blkiocg_update_completion_stats(struct blkio_group *blkg, |
395 | uint64_t start_time, uint64_t io_start_time, bool direction, bool sync) | |
9195291e | 396 | { |
549d3aa8 | 397 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
9195291e DS |
398 | struct blkio_group_stats *stats; |
399 | unsigned long flags; | |
400 | unsigned long long now = sched_clock(); | |
401 | ||
402 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 403 | stats = &pd->stats; |
84c124da DS |
404 | if (time_after64(now, io_start_time)) |
405 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME], | |
406 | now - io_start_time, direction, sync); | |
407 | if (time_after64(io_start_time, start_time)) | |
408 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME], | |
409 | io_start_time - start_time, direction, sync); | |
9195291e DS |
410 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
411 | } | |
84c124da | 412 | EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats); |
9195291e | 413 | |
317389a7 | 414 | /* Merged stats are per cpu. */ |
812d4026 DS |
415 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction, |
416 | bool sync) | |
417 | { | |
549d3aa8 | 418 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
317389a7 | 419 | struct blkio_group_stats_cpu *stats_cpu; |
812d4026 DS |
420 | unsigned long flags; |
421 | ||
317389a7 VG |
422 | /* |
423 | * Disabling interrupts to provide mutual exclusion between two | |
424 | * writes on same cpu. It probably is not needed for 64bit. Not | |
425 | * optimizing that case yet. | |
426 | */ | |
427 | local_irq_save(flags); | |
428 | ||
549d3aa8 | 429 | stats_cpu = this_cpu_ptr(pd->stats_cpu); |
317389a7 VG |
430 | |
431 | u64_stats_update_begin(&stats_cpu->syncp); | |
432 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1, | |
433 | direction, sync); | |
434 | u64_stats_update_end(&stats_cpu->syncp); | |
435 | local_irq_restore(flags); | |
812d4026 DS |
436 | } |
437 | EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats); | |
438 | ||
0381411e TH |
439 | /** |
440 | * blkg_free - free a blkg | |
441 | * @blkg: blkg to free | |
442 | * | |
443 | * Free @blkg which may be partially allocated. | |
444 | */ | |
445 | static void blkg_free(struct blkio_group *blkg) | |
446 | { | |
549d3aa8 TH |
447 | struct blkg_policy_data *pd; |
448 | ||
449 | if (!blkg) | |
450 | return; | |
451 | ||
452 | pd = blkg->pd[blkg->plid]; | |
453 | if (pd) { | |
454 | free_percpu(pd->stats_cpu); | |
455 | kfree(pd); | |
0381411e | 456 | } |
549d3aa8 | 457 | kfree(blkg); |
0381411e TH |
458 | } |
459 | ||
460 | /** | |
461 | * blkg_alloc - allocate a blkg | |
462 | * @blkcg: block cgroup the new blkg is associated with | |
463 | * @q: request_queue the new blkg is associated with | |
464 | * @pol: policy the new blkg is associated with | |
465 | * | |
466 | * Allocate a new blkg assocating @blkcg and @q for @pol. | |
467 | * | |
468 | * FIXME: Should be called with queue locked but currently isn't due to | |
469 | * percpu stat breakage. | |
470 | */ | |
471 | static struct blkio_group *blkg_alloc(struct blkio_cgroup *blkcg, | |
472 | struct request_queue *q, | |
473 | struct blkio_policy_type *pol) | |
474 | { | |
475 | struct blkio_group *blkg; | |
549d3aa8 | 476 | struct blkg_policy_data *pd; |
0381411e TH |
477 | |
478 | /* alloc and init base part */ | |
479 | blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node); | |
480 | if (!blkg) | |
481 | return NULL; | |
482 | ||
483 | spin_lock_init(&blkg->stats_lock); | |
484 | rcu_assign_pointer(blkg->q, q); | |
485 | blkg->blkcg = blkcg; | |
486 | blkg->plid = pol->plid; | |
1adaf3dd | 487 | blkg->refcnt = 1; |
0381411e TH |
488 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); |
489 | ||
549d3aa8 TH |
490 | /* alloc per-policy data and attach it to blkg */ |
491 | pd = kzalloc_node(sizeof(*pd) + pol->pdata_size, GFP_ATOMIC, | |
492 | q->node); | |
493 | if (!pd) { | |
0381411e TH |
494 | blkg_free(blkg); |
495 | return NULL; | |
496 | } | |
497 | ||
549d3aa8 TH |
498 | blkg->pd[pol->plid] = pd; |
499 | pd->blkg = blkg; | |
500 | ||
0381411e | 501 | /* broken, read comment in the callsite */ |
549d3aa8 TH |
502 | |
503 | pd->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu); | |
504 | if (!pd->stats_cpu) { | |
0381411e TH |
505 | blkg_free(blkg); |
506 | return NULL; | |
507 | } | |
508 | ||
549d3aa8 | 509 | /* invoke per-policy init */ |
0381411e TH |
510 | pol->ops.blkio_init_group_fn(blkg); |
511 | return blkg; | |
512 | } | |
513 | ||
cd1604fa TH |
514 | struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg, |
515 | struct request_queue *q, | |
516 | enum blkio_policy_id plid, | |
517 | bool for_root) | |
518 | __releases(q->queue_lock) __acquires(q->queue_lock) | |
5624a4e4 | 519 | { |
cd1604fa TH |
520 | struct blkio_policy_type *pol = blkio_policy[plid]; |
521 | struct blkio_group *blkg, *new_blkg; | |
5624a4e4 | 522 | |
cd1604fa TH |
523 | WARN_ON_ONCE(!rcu_read_lock_held()); |
524 | lockdep_assert_held(q->queue_lock); | |
525 | ||
526 | /* | |
527 | * This could be the first entry point of blkcg implementation and | |
528 | * we shouldn't allow anything to go through for a bypassing queue. | |
529 | * The following can be removed if blkg lookup is guaranteed to | |
530 | * fail on a bypassing queue. | |
531 | */ | |
532 | if (unlikely(blk_queue_bypass(q)) && !for_root) | |
533 | return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY); | |
534 | ||
535 | blkg = blkg_lookup(blkcg, q, plid); | |
536 | if (blkg) | |
537 | return blkg; | |
538 | ||
7ee9c562 | 539 | /* blkg holds a reference to blkcg */ |
cd1604fa TH |
540 | if (!css_tryget(&blkcg->css)) |
541 | return ERR_PTR(-EINVAL); | |
542 | ||
543 | /* | |
544 | * Allocate and initialize. | |
545 | * | |
546 | * FIXME: The following is broken. Percpu memory allocation | |
547 | * requires %GFP_KERNEL context and can't be performed from IO | |
548 | * path. Allocation here should inherently be atomic and the | |
549 | * following lock dancing can be removed once the broken percpu | |
550 | * allocation is fixed. | |
551 | */ | |
552 | spin_unlock_irq(q->queue_lock); | |
553 | rcu_read_unlock(); | |
554 | ||
0381411e | 555 | new_blkg = blkg_alloc(blkcg, q, pol); |
cd1604fa TH |
556 | |
557 | rcu_read_lock(); | |
558 | spin_lock_irq(q->queue_lock); | |
31e4c28d | 559 | |
cd1604fa TH |
560 | /* did bypass get turned on inbetween? */ |
561 | if (unlikely(blk_queue_bypass(q)) && !for_root) { | |
562 | blkg = ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY); | |
563 | goto out; | |
564 | } | |
565 | ||
566 | /* did someone beat us to it? */ | |
567 | blkg = blkg_lookup(blkcg, q, plid); | |
568 | if (unlikely(blkg)) | |
569 | goto out; | |
570 | ||
571 | /* did alloc fail? */ | |
0381411e | 572 | if (unlikely(!new_blkg)) { |
cd1604fa TH |
573 | blkg = ERR_PTR(-ENOMEM); |
574 | goto out; | |
575 | } | |
576 | ||
577 | /* insert */ | |
578 | spin_lock(&blkcg->lock); | |
579 | swap(blkg, new_blkg); | |
31e4c28d | 580 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
cd1604fa TH |
581 | pol->ops.blkio_link_group_fn(q, blkg); |
582 | spin_unlock(&blkcg->lock); | |
583 | out: | |
0381411e | 584 | blkg_free(new_blkg); |
cd1604fa | 585 | return blkg; |
31e4c28d | 586 | } |
cd1604fa | 587 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 588 | |
b1c35769 VG |
589 | static void __blkiocg_del_blkio_group(struct blkio_group *blkg) |
590 | { | |
591 | hlist_del_init_rcu(&blkg->blkcg_node); | |
b1c35769 VG |
592 | } |
593 | ||
594 | /* | |
595 | * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1 | |
596 | * indicating that blk_group was unhashed by the time we got to it. | |
597 | */ | |
31e4c28d VG |
598 | int blkiocg_del_blkio_group(struct blkio_group *blkg) |
599 | { | |
7ee9c562 | 600 | struct blkio_cgroup *blkcg = blkg->blkcg; |
b1c35769 | 601 | unsigned long flags; |
b1c35769 VG |
602 | int ret = 1; |
603 | ||
7ee9c562 TH |
604 | spin_lock_irqsave(&blkcg->lock, flags); |
605 | if (!hlist_unhashed(&blkg->blkcg_node)) { | |
606 | __blkiocg_del_blkio_group(blkg); | |
607 | ret = 0; | |
b1c35769 | 608 | } |
7ee9c562 | 609 | spin_unlock_irqrestore(&blkcg->lock, flags); |
0f3942a3 | 610 | |
b1c35769 | 611 | return ret; |
31e4c28d | 612 | } |
9d6a986c | 613 | EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group); |
31e4c28d VG |
614 | |
615 | /* called under rcu_read_lock(). */ | |
cd1604fa TH |
616 | struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg, |
617 | struct request_queue *q, | |
618 | enum blkio_policy_id plid) | |
31e4c28d VG |
619 | { |
620 | struct blkio_group *blkg; | |
621 | struct hlist_node *n; | |
31e4c28d | 622 | |
ca32aefc TH |
623 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) |
624 | if (blkg->q == q && blkg->plid == plid) | |
31e4c28d | 625 | return blkg; |
31e4c28d VG |
626 | return NULL; |
627 | } | |
cd1604fa | 628 | EXPORT_SYMBOL_GPL(blkg_lookup); |
31e4c28d | 629 | |
72e06c25 TH |
630 | void blkg_destroy_all(struct request_queue *q) |
631 | { | |
632 | struct blkio_policy_type *pol; | |
633 | ||
634 | while (true) { | |
635 | bool done = true; | |
636 | ||
637 | spin_lock(&blkio_list_lock); | |
638 | spin_lock_irq(q->queue_lock); | |
639 | ||
640 | /* | |
641 | * clear_queue_fn() might return with non-empty group list | |
642 | * if it raced cgroup removal and lost. cgroup removal is | |
643 | * guaranteed to make forward progress and retrying after a | |
644 | * while is enough. This ugliness is scheduled to be | |
645 | * removed after locking update. | |
646 | */ | |
647 | list_for_each_entry(pol, &blkio_list, list) | |
648 | if (!pol->ops.blkio_clear_queue_fn(q)) | |
649 | done = false; | |
650 | ||
651 | spin_unlock_irq(q->queue_lock); | |
652 | spin_unlock(&blkio_list_lock); | |
653 | ||
654 | if (done) | |
655 | break; | |
656 | ||
657 | msleep(10); /* just some random duration I like */ | |
658 | } | |
659 | } | |
660 | ||
1adaf3dd TH |
661 | static void blkg_rcu_free(struct rcu_head *rcu_head) |
662 | { | |
663 | blkg_free(container_of(rcu_head, struct blkio_group, rcu_head)); | |
664 | } | |
665 | ||
666 | void __blkg_release(struct blkio_group *blkg) | |
667 | { | |
668 | /* release the extra blkcg reference this blkg has been holding */ | |
669 | css_put(&blkg->blkcg->css); | |
670 | ||
671 | /* | |
672 | * A group is freed in rcu manner. But having an rcu lock does not | |
673 | * mean that one can access all the fields of blkg and assume these | |
674 | * are valid. For example, don't try to follow throtl_data and | |
675 | * request queue links. | |
676 | * | |
677 | * Having a reference to blkg under an rcu allows acess to only | |
678 | * values local to groups like group stats and group rate limits | |
679 | */ | |
680 | call_rcu(&blkg->rcu_head, blkg_rcu_free); | |
681 | } | |
682 | EXPORT_SYMBOL_GPL(__blkg_release); | |
683 | ||
f0bdc8cd VG |
684 | static void blkio_reset_stats_cpu(struct blkio_group *blkg) |
685 | { | |
549d3aa8 | 686 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
f0bdc8cd VG |
687 | struct blkio_group_stats_cpu *stats_cpu; |
688 | int i, j, k; | |
689 | /* | |
690 | * Note: On 64 bit arch this should not be an issue. This has the | |
691 | * possibility of returning some inconsistent value on 32bit arch | |
692 | * as 64bit update on 32bit is non atomic. Taking care of this | |
693 | * corner case makes code very complicated, like sending IPIs to | |
694 | * cpus, taking care of stats of offline cpus etc. | |
695 | * | |
696 | * reset stats is anyway more of a debug feature and this sounds a | |
697 | * corner case. So I am not complicating the code yet until and | |
698 | * unless this becomes a real issue. | |
699 | */ | |
700 | for_each_possible_cpu(i) { | |
549d3aa8 | 701 | stats_cpu = per_cpu_ptr(pd->stats_cpu, i); |
f0bdc8cd VG |
702 | stats_cpu->sectors = 0; |
703 | for(j = 0; j < BLKIO_STAT_CPU_NR; j++) | |
704 | for (k = 0; k < BLKIO_STAT_TOTAL; k++) | |
705 | stats_cpu->stat_arr_cpu[j][k] = 0; | |
706 | } | |
707 | } | |
708 | ||
303a3acb | 709 | static int |
84c124da | 710 | blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val) |
303a3acb DS |
711 | { |
712 | struct blkio_cgroup *blkcg; | |
713 | struct blkio_group *blkg; | |
812df48d | 714 | struct blkio_group_stats *stats; |
303a3acb | 715 | struct hlist_node *n; |
cdc1184c DS |
716 | uint64_t queued[BLKIO_STAT_TOTAL]; |
717 | int i; | |
812df48d DS |
718 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
719 | bool idling, waiting, empty; | |
720 | unsigned long long now = sched_clock(); | |
721 | #endif | |
303a3acb DS |
722 | |
723 | blkcg = cgroup_to_blkio_cgroup(cgroup); | |
724 | spin_lock_irq(&blkcg->lock); | |
725 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
549d3aa8 TH |
726 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
727 | ||
303a3acb | 728 | spin_lock(&blkg->stats_lock); |
549d3aa8 | 729 | stats = &pd->stats; |
812df48d DS |
730 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
731 | idling = blkio_blkg_idling(stats); | |
732 | waiting = blkio_blkg_waiting(stats); | |
733 | empty = blkio_blkg_empty(stats); | |
734 | #endif | |
cdc1184c | 735 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
812df48d DS |
736 | queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i]; |
737 | memset(stats, 0, sizeof(struct blkio_group_stats)); | |
cdc1184c | 738 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
812df48d DS |
739 | stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i]; |
740 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
741 | if (idling) { | |
742 | blkio_mark_blkg_idling(stats); | |
743 | stats->start_idle_time = now; | |
744 | } | |
745 | if (waiting) { | |
746 | blkio_mark_blkg_waiting(stats); | |
747 | stats->start_group_wait_time = now; | |
748 | } | |
749 | if (empty) { | |
750 | blkio_mark_blkg_empty(stats); | |
751 | stats->start_empty_time = now; | |
752 | } | |
753 | #endif | |
303a3acb | 754 | spin_unlock(&blkg->stats_lock); |
f0bdc8cd VG |
755 | |
756 | /* Reset Per cpu stats which don't take blkg->stats_lock */ | |
757 | blkio_reset_stats_cpu(blkg); | |
303a3acb | 758 | } |
f0bdc8cd | 759 | |
303a3acb DS |
760 | spin_unlock_irq(&blkcg->lock); |
761 | return 0; | |
762 | } | |
763 | ||
7a4dd281 TH |
764 | static void blkio_get_key_name(enum stat_sub_type type, const char *dname, |
765 | char *str, int chars_left, bool diskname_only) | |
303a3acb | 766 | { |
7a4dd281 | 767 | snprintf(str, chars_left, "%s", dname); |
303a3acb DS |
768 | chars_left -= strlen(str); |
769 | if (chars_left <= 0) { | |
770 | printk(KERN_WARNING | |
771 | "Possibly incorrect cgroup stat display format"); | |
772 | return; | |
773 | } | |
84c124da DS |
774 | if (diskname_only) |
775 | return; | |
303a3acb | 776 | switch (type) { |
84c124da | 777 | case BLKIO_STAT_READ: |
303a3acb DS |
778 | strlcat(str, " Read", chars_left); |
779 | break; | |
84c124da | 780 | case BLKIO_STAT_WRITE: |
303a3acb DS |
781 | strlcat(str, " Write", chars_left); |
782 | break; | |
84c124da | 783 | case BLKIO_STAT_SYNC: |
303a3acb DS |
784 | strlcat(str, " Sync", chars_left); |
785 | break; | |
84c124da | 786 | case BLKIO_STAT_ASYNC: |
303a3acb DS |
787 | strlcat(str, " Async", chars_left); |
788 | break; | |
84c124da | 789 | case BLKIO_STAT_TOTAL: |
303a3acb DS |
790 | strlcat(str, " Total", chars_left); |
791 | break; | |
792 | default: | |
793 | strlcat(str, " Invalid", chars_left); | |
794 | } | |
795 | } | |
796 | ||
84c124da | 797 | static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val, |
7a4dd281 | 798 | struct cgroup_map_cb *cb, const char *dname) |
84c124da | 799 | { |
7a4dd281 | 800 | blkio_get_key_name(0, dname, str, chars_left, true); |
84c124da DS |
801 | cb->fill(cb, str, val); |
802 | return val; | |
803 | } | |
303a3acb | 804 | |
5624a4e4 VG |
805 | |
806 | static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg, | |
807 | enum stat_type_cpu type, enum stat_sub_type sub_type) | |
808 | { | |
549d3aa8 | 809 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
5624a4e4 VG |
810 | int cpu; |
811 | struct blkio_group_stats_cpu *stats_cpu; | |
575969a0 | 812 | u64 val = 0, tval; |
5624a4e4 VG |
813 | |
814 | for_each_possible_cpu(cpu) { | |
575969a0 | 815 | unsigned int start; |
549d3aa8 | 816 | stats_cpu = per_cpu_ptr(pd->stats_cpu, cpu); |
5624a4e4 | 817 | |
575969a0 VG |
818 | do { |
819 | start = u64_stats_fetch_begin(&stats_cpu->syncp); | |
820 | if (type == BLKIO_STAT_CPU_SECTORS) | |
821 | tval = stats_cpu->sectors; | |
822 | else | |
823 | tval = stats_cpu->stat_arr_cpu[type][sub_type]; | |
824 | } while(u64_stats_fetch_retry(&stats_cpu->syncp, start)); | |
825 | ||
826 | val += tval; | |
5624a4e4 VG |
827 | } |
828 | ||
829 | return val; | |
830 | } | |
831 | ||
832 | static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg, | |
7a4dd281 TH |
833 | struct cgroup_map_cb *cb, const char *dname, |
834 | enum stat_type_cpu type) | |
5624a4e4 VG |
835 | { |
836 | uint64_t disk_total, val; | |
837 | char key_str[MAX_KEY_LEN]; | |
838 | enum stat_sub_type sub_type; | |
839 | ||
840 | if (type == BLKIO_STAT_CPU_SECTORS) { | |
841 | val = blkio_read_stat_cpu(blkg, type, 0); | |
7a4dd281 TH |
842 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, |
843 | dname); | |
5624a4e4 VG |
844 | } |
845 | ||
846 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; | |
847 | sub_type++) { | |
7a4dd281 TH |
848 | blkio_get_key_name(sub_type, dname, key_str, MAX_KEY_LEN, |
849 | false); | |
5624a4e4 VG |
850 | val = blkio_read_stat_cpu(blkg, type, sub_type); |
851 | cb->fill(cb, key_str, val); | |
852 | } | |
853 | ||
854 | disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) + | |
855 | blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE); | |
856 | ||
7a4dd281 TH |
857 | blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN, |
858 | false); | |
5624a4e4 VG |
859 | cb->fill(cb, key_str, disk_total); |
860 | return disk_total; | |
861 | } | |
862 | ||
84c124da DS |
863 | /* This should be called with blkg->stats_lock held */ |
864 | static uint64_t blkio_get_stat(struct blkio_group *blkg, | |
7a4dd281 TH |
865 | struct cgroup_map_cb *cb, const char *dname, |
866 | enum stat_type type) | |
303a3acb | 867 | { |
549d3aa8 | 868 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
303a3acb DS |
869 | uint64_t disk_total; |
870 | char key_str[MAX_KEY_LEN]; | |
84c124da DS |
871 | enum stat_sub_type sub_type; |
872 | ||
873 | if (type == BLKIO_STAT_TIME) | |
874 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 875 | pd->stats.time, cb, dname); |
9026e521 | 876 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 JT |
877 | if (type == BLKIO_STAT_UNACCOUNTED_TIME) |
878 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 879 | pd->stats.unaccounted_time, cb, dname); |
cdc1184c | 880 | if (type == BLKIO_STAT_AVG_QUEUE_SIZE) { |
549d3aa8 TH |
881 | uint64_t sum = pd->stats.avg_queue_size_sum; |
882 | uint64_t samples = pd->stats.avg_queue_size_samples; | |
cdc1184c DS |
883 | if (samples) |
884 | do_div(sum, samples); | |
885 | else | |
886 | sum = 0; | |
7a4dd281 TH |
887 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, |
888 | sum, cb, dname); | |
cdc1184c | 889 | } |
812df48d DS |
890 | if (type == BLKIO_STAT_GROUP_WAIT_TIME) |
891 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 892 | pd->stats.group_wait_time, cb, dname); |
812df48d DS |
893 | if (type == BLKIO_STAT_IDLE_TIME) |
894 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 895 | pd->stats.idle_time, cb, dname); |
812df48d DS |
896 | if (type == BLKIO_STAT_EMPTY_TIME) |
897 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 898 | pd->stats.empty_time, cb, dname); |
84c124da DS |
899 | if (type == BLKIO_STAT_DEQUEUE) |
900 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 901 | pd->stats.dequeue, cb, dname); |
84c124da | 902 | #endif |
303a3acb | 903 | |
84c124da DS |
904 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; |
905 | sub_type++) { | |
7a4dd281 TH |
906 | blkio_get_key_name(sub_type, dname, key_str, MAX_KEY_LEN, |
907 | false); | |
549d3aa8 | 908 | cb->fill(cb, key_str, pd->stats.stat_arr[type][sub_type]); |
303a3acb | 909 | } |
549d3aa8 TH |
910 | disk_total = pd->stats.stat_arr[type][BLKIO_STAT_READ] + |
911 | pd->stats.stat_arr[type][BLKIO_STAT_WRITE]; | |
7a4dd281 TH |
912 | blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN, |
913 | false); | |
303a3acb DS |
914 | cb->fill(cb, key_str, disk_total); |
915 | return disk_total; | |
916 | } | |
917 | ||
4bfd482e TH |
918 | static int blkio_policy_parse_and_set(char *buf, enum blkio_policy_id plid, |
919 | int fileid, struct blkio_cgroup *blkcg) | |
34d0f179 | 920 | { |
ece84241 | 921 | struct gendisk *disk = NULL; |
e56da7e2 | 922 | struct blkio_group *blkg = NULL; |
549d3aa8 | 923 | struct blkg_policy_data *pd; |
34d0f179 | 924 | char *s[4], *p, *major_s = NULL, *minor_s = NULL; |
d11bb446 | 925 | unsigned long major, minor; |
ece84241 TH |
926 | int i = 0, ret = -EINVAL; |
927 | int part; | |
34d0f179 | 928 | dev_t dev; |
d11bb446 | 929 | u64 temp; |
34d0f179 GJ |
930 | |
931 | memset(s, 0, sizeof(s)); | |
932 | ||
933 | while ((p = strsep(&buf, " ")) != NULL) { | |
934 | if (!*p) | |
935 | continue; | |
936 | ||
937 | s[i++] = p; | |
938 | ||
939 | /* Prevent from inputing too many things */ | |
940 | if (i == 3) | |
941 | break; | |
942 | } | |
943 | ||
944 | if (i != 2) | |
ece84241 | 945 | goto out; |
34d0f179 GJ |
946 | |
947 | p = strsep(&s[0], ":"); | |
948 | if (p != NULL) | |
949 | major_s = p; | |
950 | else | |
ece84241 | 951 | goto out; |
34d0f179 GJ |
952 | |
953 | minor_s = s[0]; | |
954 | if (!minor_s) | |
ece84241 | 955 | goto out; |
34d0f179 | 956 | |
ece84241 TH |
957 | if (strict_strtoul(major_s, 10, &major)) |
958 | goto out; | |
34d0f179 | 959 | |
ece84241 TH |
960 | if (strict_strtoul(minor_s, 10, &minor)) |
961 | goto out; | |
34d0f179 GJ |
962 | |
963 | dev = MKDEV(major, minor); | |
964 | ||
ece84241 TH |
965 | if (strict_strtoull(s[1], 10, &temp)) |
966 | goto out; | |
34d0f179 | 967 | |
e56da7e2 | 968 | disk = get_gendisk(dev, &part); |
4bfd482e | 969 | if (!disk || part) |
e56da7e2 | 970 | goto out; |
e56da7e2 TH |
971 | |
972 | rcu_read_lock(); | |
973 | ||
4bfd482e TH |
974 | spin_lock_irq(disk->queue->queue_lock); |
975 | blkg = blkg_lookup_create(blkcg, disk->queue, plid, false); | |
976 | spin_unlock_irq(disk->queue->queue_lock); | |
e56da7e2 | 977 | |
4bfd482e TH |
978 | if (IS_ERR(blkg)) { |
979 | ret = PTR_ERR(blkg); | |
980 | goto out_unlock; | |
d11bb446 | 981 | } |
34d0f179 | 982 | |
549d3aa8 TH |
983 | pd = blkg->pd[plid]; |
984 | ||
062a644d VG |
985 | switch (plid) { |
986 | case BLKIO_POLICY_PROP: | |
d11bb446 WG |
987 | if ((temp < BLKIO_WEIGHT_MIN && temp > 0) || |
988 | temp > BLKIO_WEIGHT_MAX) | |
e56da7e2 | 989 | goto out_unlock; |
34d0f179 | 990 | |
549d3aa8 | 991 | pd->conf.weight = temp; |
4bfd482e | 992 | blkio_update_group_weight(blkg, temp ?: blkcg->weight); |
4c9eefa1 VG |
993 | break; |
994 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
995 | switch(fileid) { |
996 | case BLKIO_THROTL_read_bps_device: | |
549d3aa8 | 997 | pd->conf.bps[READ] = temp; |
4bfd482e | 998 | blkio_update_group_bps(blkg, temp ?: -1, fileid); |
e56da7e2 | 999 | break; |
7702e8f4 | 1000 | case BLKIO_THROTL_write_bps_device: |
549d3aa8 | 1001 | pd->conf.bps[WRITE] = temp; |
4bfd482e | 1002 | blkio_update_group_bps(blkg, temp ?: -1, fileid); |
7702e8f4 VG |
1003 | break; |
1004 | case BLKIO_THROTL_read_iops_device: | |
e56da7e2 TH |
1005 | if (temp > THROTL_IOPS_MAX) |
1006 | goto out_unlock; | |
549d3aa8 | 1007 | pd->conf.iops[READ] = temp; |
4bfd482e | 1008 | blkio_update_group_iops(blkg, temp ?: -1, fileid); |
e56da7e2 | 1009 | break; |
7702e8f4 | 1010 | case BLKIO_THROTL_write_iops_device: |
d11bb446 | 1011 | if (temp > THROTL_IOPS_MAX) |
e56da7e2 | 1012 | goto out_unlock; |
549d3aa8 | 1013 | pd->conf.iops[WRITE] = temp; |
4bfd482e | 1014 | blkio_update_group_iops(blkg, temp ?: -1, fileid); |
7702e8f4 VG |
1015 | break; |
1016 | } | |
062a644d VG |
1017 | break; |
1018 | default: | |
1019 | BUG(); | |
1020 | } | |
ece84241 | 1021 | ret = 0; |
e56da7e2 TH |
1022 | out_unlock: |
1023 | rcu_read_unlock(); | |
ece84241 TH |
1024 | out: |
1025 | put_disk(disk); | |
e56da7e2 TH |
1026 | |
1027 | /* | |
1028 | * If queue was bypassing, we should retry. Do so after a short | |
1029 | * msleep(). It isn't strictly necessary but queue can be | |
1030 | * bypassing for some time and it's always nice to avoid busy | |
1031 | * looping. | |
1032 | */ | |
1033 | if (ret == -EBUSY) { | |
1034 | msleep(10); | |
1035 | return restart_syscall(); | |
1036 | } | |
ece84241 | 1037 | return ret; |
34d0f179 GJ |
1038 | } |
1039 | ||
062a644d VG |
1040 | static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft, |
1041 | const char *buffer) | |
34d0f179 GJ |
1042 | { |
1043 | int ret = 0; | |
1044 | char *buf; | |
e56da7e2 | 1045 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); |
062a644d VG |
1046 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); |
1047 | int fileid = BLKIOFILE_ATTR(cft->private); | |
34d0f179 GJ |
1048 | |
1049 | buf = kstrdup(buffer, GFP_KERNEL); | |
1050 | if (!buf) | |
1051 | return -ENOMEM; | |
1052 | ||
4bfd482e | 1053 | ret = blkio_policy_parse_and_set(buf, plid, fileid, blkcg); |
34d0f179 GJ |
1054 | kfree(buf); |
1055 | return ret; | |
1056 | } | |
1057 | ||
92616b5b VG |
1058 | static const char *blkg_dev_name(struct blkio_group *blkg) |
1059 | { | |
1060 | /* some drivers (floppy) instantiate a queue w/o disk registered */ | |
1061 | if (blkg->q->backing_dev_info.dev) | |
1062 | return dev_name(blkg->q->backing_dev_info.dev); | |
1063 | return NULL; | |
1064 | } | |
1065 | ||
4bfd482e TH |
1066 | static void blkio_print_group_conf(struct cftype *cft, struct blkio_group *blkg, |
1067 | struct seq_file *m) | |
34d0f179 | 1068 | { |
549d3aa8 | 1069 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; |
92616b5b | 1070 | const char *dname = blkg_dev_name(blkg); |
4bfd482e TH |
1071 | int fileid = BLKIOFILE_ATTR(cft->private); |
1072 | int rw = WRITE; | |
1073 | ||
92616b5b VG |
1074 | if (!dname) |
1075 | return; | |
1076 | ||
4bfd482e | 1077 | switch (blkg->plid) { |
062a644d | 1078 | case BLKIO_POLICY_PROP: |
549d3aa8 | 1079 | if (pd->conf.weight) |
7a4dd281 | 1080 | seq_printf(m, "%s\t%u\n", |
549d3aa8 | 1081 | dname, pd->conf.weight); |
4c9eefa1 VG |
1082 | break; |
1083 | case BLKIO_POLICY_THROTL: | |
4bfd482e | 1084 | switch (fileid) { |
7702e8f4 | 1085 | case BLKIO_THROTL_read_bps_device: |
4bfd482e | 1086 | rw = READ; |
7702e8f4 | 1087 | case BLKIO_THROTL_write_bps_device: |
549d3aa8 | 1088 | if (pd->conf.bps[rw]) |
7a4dd281 | 1089 | seq_printf(m, "%s\t%llu\n", |
549d3aa8 | 1090 | dname, pd->conf.bps[rw]); |
7702e8f4 VG |
1091 | break; |
1092 | case BLKIO_THROTL_read_iops_device: | |
4bfd482e | 1093 | rw = READ; |
7702e8f4 | 1094 | case BLKIO_THROTL_write_iops_device: |
549d3aa8 | 1095 | if (pd->conf.iops[rw]) |
7a4dd281 | 1096 | seq_printf(m, "%s\t%u\n", |
549d3aa8 | 1097 | dname, pd->conf.iops[rw]); |
7702e8f4 VG |
1098 | break; |
1099 | } | |
062a644d VG |
1100 | break; |
1101 | default: | |
1102 | BUG(); | |
1103 | } | |
1104 | } | |
34d0f179 | 1105 | |
062a644d | 1106 | /* cgroup files which read their data from policy nodes end up here */ |
4bfd482e TH |
1107 | static void blkio_read_conf(struct cftype *cft, struct blkio_cgroup *blkcg, |
1108 | struct seq_file *m) | |
34d0f179 | 1109 | { |
4bfd482e TH |
1110 | struct blkio_group *blkg; |
1111 | struct hlist_node *n; | |
34d0f179 | 1112 | |
4bfd482e TH |
1113 | spin_lock_irq(&blkcg->lock); |
1114 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) | |
1115 | if (BLKIOFILE_POLICY(cft->private) == blkg->plid) | |
1116 | blkio_print_group_conf(cft, blkg, m); | |
1117 | spin_unlock_irq(&blkcg->lock); | |
062a644d VG |
1118 | } |
1119 | ||
1120 | static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft, | |
1121 | struct seq_file *m) | |
1122 | { | |
1123 | struct blkio_cgroup *blkcg; | |
1124 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1125 | int name = BLKIOFILE_ATTR(cft->private); | |
1126 | ||
1127 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1128 | ||
1129 | switch(plid) { | |
1130 | case BLKIO_POLICY_PROP: | |
1131 | switch(name) { | |
1132 | case BLKIO_PROP_weight_device: | |
4bfd482e | 1133 | blkio_read_conf(cft, blkcg, m); |
062a644d VG |
1134 | return 0; |
1135 | default: | |
1136 | BUG(); | |
1137 | } | |
1138 | break; | |
4c9eefa1 VG |
1139 | case BLKIO_POLICY_THROTL: |
1140 | switch(name){ | |
1141 | case BLKIO_THROTL_read_bps_device: | |
1142 | case BLKIO_THROTL_write_bps_device: | |
7702e8f4 VG |
1143 | case BLKIO_THROTL_read_iops_device: |
1144 | case BLKIO_THROTL_write_iops_device: | |
4bfd482e | 1145 | blkio_read_conf(cft, blkcg, m); |
4c9eefa1 VG |
1146 | return 0; |
1147 | default: | |
1148 | BUG(); | |
1149 | } | |
1150 | break; | |
062a644d VG |
1151 | default: |
1152 | BUG(); | |
1153 | } | |
1154 | ||
1155 | return 0; | |
1156 | } | |
1157 | ||
1158 | static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg, | |
5624a4e4 VG |
1159 | struct cftype *cft, struct cgroup_map_cb *cb, |
1160 | enum stat_type type, bool show_total, bool pcpu) | |
062a644d VG |
1161 | { |
1162 | struct blkio_group *blkg; | |
1163 | struct hlist_node *n; | |
1164 | uint64_t cgroup_total = 0; | |
1165 | ||
1166 | rcu_read_lock(); | |
1167 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
92616b5b | 1168 | const char *dname = blkg_dev_name(blkg); |
7a4dd281 | 1169 | |
92616b5b | 1170 | if (!dname || BLKIOFILE_POLICY(cft->private) != blkg->plid) |
7a4dd281 TH |
1171 | continue; |
1172 | if (pcpu) | |
1173 | cgroup_total += blkio_get_stat_cpu(blkg, cb, dname, | |
1174 | type); | |
1175 | else { | |
1176 | spin_lock_irq(&blkg->stats_lock); | |
1177 | cgroup_total += blkio_get_stat(blkg, cb, dname, type); | |
1178 | spin_unlock_irq(&blkg->stats_lock); | |
062a644d VG |
1179 | } |
1180 | } | |
1181 | if (show_total) | |
1182 | cb->fill(cb, "Total", cgroup_total); | |
1183 | rcu_read_unlock(); | |
1184 | return 0; | |
1185 | } | |
1186 | ||
1187 | /* All map kind of cgroup file get serviced by this function */ | |
1188 | static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft, | |
1189 | struct cgroup_map_cb *cb) | |
1190 | { | |
1191 | struct blkio_cgroup *blkcg; | |
1192 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1193 | int name = BLKIOFILE_ATTR(cft->private); | |
1194 | ||
1195 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1196 | ||
1197 | switch(plid) { | |
1198 | case BLKIO_POLICY_PROP: | |
1199 | switch(name) { | |
1200 | case BLKIO_PROP_time: | |
1201 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1202 | BLKIO_STAT_TIME, 0, 0); |
062a644d VG |
1203 | case BLKIO_PROP_sectors: |
1204 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1205 | BLKIO_STAT_CPU_SECTORS, 0, 1); |
062a644d VG |
1206 | case BLKIO_PROP_io_service_bytes: |
1207 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1208 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
062a644d VG |
1209 | case BLKIO_PROP_io_serviced: |
1210 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1211 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
062a644d VG |
1212 | case BLKIO_PROP_io_service_time: |
1213 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1214 | BLKIO_STAT_SERVICE_TIME, 1, 0); |
062a644d VG |
1215 | case BLKIO_PROP_io_wait_time: |
1216 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1217 | BLKIO_STAT_WAIT_TIME, 1, 0); |
062a644d VG |
1218 | case BLKIO_PROP_io_merged: |
1219 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
317389a7 | 1220 | BLKIO_STAT_CPU_MERGED, 1, 1); |
062a644d VG |
1221 | case BLKIO_PROP_io_queued: |
1222 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1223 | BLKIO_STAT_QUEUED, 1, 0); |
062a644d | 1224 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
9026e521 JT |
1225 | case BLKIO_PROP_unaccounted_time: |
1226 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1227 | BLKIO_STAT_UNACCOUNTED_TIME, 0, 0); |
062a644d VG |
1228 | case BLKIO_PROP_dequeue: |
1229 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1230 | BLKIO_STAT_DEQUEUE, 0, 0); |
062a644d VG |
1231 | case BLKIO_PROP_avg_queue_size: |
1232 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1233 | BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0); |
062a644d VG |
1234 | case BLKIO_PROP_group_wait_time: |
1235 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1236 | BLKIO_STAT_GROUP_WAIT_TIME, 0, 0); |
062a644d VG |
1237 | case BLKIO_PROP_idle_time: |
1238 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1239 | BLKIO_STAT_IDLE_TIME, 0, 0); |
062a644d VG |
1240 | case BLKIO_PROP_empty_time: |
1241 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1242 | BLKIO_STAT_EMPTY_TIME, 0, 0); |
062a644d VG |
1243 | #endif |
1244 | default: | |
1245 | BUG(); | |
1246 | } | |
1247 | break; | |
4c9eefa1 VG |
1248 | case BLKIO_POLICY_THROTL: |
1249 | switch(name){ | |
1250 | case BLKIO_THROTL_io_service_bytes: | |
1251 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1252 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
4c9eefa1 VG |
1253 | case BLKIO_THROTL_io_serviced: |
1254 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1255 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
4c9eefa1 VG |
1256 | default: |
1257 | BUG(); | |
1258 | } | |
1259 | break; | |
062a644d VG |
1260 | default: |
1261 | BUG(); | |
1262 | } | |
1263 | ||
1264 | return 0; | |
1265 | } | |
1266 | ||
4bfd482e | 1267 | static int blkio_weight_write(struct blkio_cgroup *blkcg, int plid, u64 val) |
062a644d VG |
1268 | { |
1269 | struct blkio_group *blkg; | |
1270 | struct hlist_node *n; | |
062a644d VG |
1271 | |
1272 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) | |
1273 | return -EINVAL; | |
1274 | ||
1275 | spin_lock(&blkio_list_lock); | |
1276 | spin_lock_irq(&blkcg->lock); | |
1277 | blkcg->weight = (unsigned int)val; | |
1278 | ||
549d3aa8 TH |
1279 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
1280 | struct blkg_policy_data *pd = blkg->pd[blkg->plid]; | |
1281 | ||
1282 | if (blkg->plid == plid && !pd->conf.weight) | |
4bfd482e | 1283 | blkio_update_group_weight(blkg, blkcg->weight); |
549d3aa8 | 1284 | } |
062a644d | 1285 | |
062a644d VG |
1286 | spin_unlock_irq(&blkcg->lock); |
1287 | spin_unlock(&blkio_list_lock); | |
1288 | return 0; | |
1289 | } | |
1290 | ||
1291 | static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) { | |
1292 | struct blkio_cgroup *blkcg; | |
1293 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1294 | int name = BLKIOFILE_ATTR(cft->private); | |
1295 | ||
1296 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1297 | ||
1298 | switch(plid) { | |
1299 | case BLKIO_POLICY_PROP: | |
1300 | switch(name) { | |
1301 | case BLKIO_PROP_weight: | |
1302 | return (u64)blkcg->weight; | |
1303 | } | |
1304 | break; | |
1305 | default: | |
1306 | BUG(); | |
1307 | } | |
1308 | return 0; | |
1309 | } | |
1310 | ||
1311 | static int | |
1312 | blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) | |
1313 | { | |
1314 | struct blkio_cgroup *blkcg; | |
1315 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1316 | int name = BLKIOFILE_ATTR(cft->private); | |
1317 | ||
1318 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1319 | ||
1320 | switch(plid) { | |
1321 | case BLKIO_POLICY_PROP: | |
1322 | switch(name) { | |
1323 | case BLKIO_PROP_weight: | |
4bfd482e | 1324 | return blkio_weight_write(blkcg, plid, val); |
062a644d VG |
1325 | } |
1326 | break; | |
1327 | default: | |
1328 | BUG(); | |
1329 | } | |
34d0f179 | 1330 | |
34d0f179 GJ |
1331 | return 0; |
1332 | } | |
1333 | ||
31e4c28d | 1334 | struct cftype blkio_files[] = { |
34d0f179 GJ |
1335 | { |
1336 | .name = "weight_device", | |
062a644d VG |
1337 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1338 | BLKIO_PROP_weight_device), | |
1339 | .read_seq_string = blkiocg_file_read, | |
1340 | .write_string = blkiocg_file_write, | |
34d0f179 GJ |
1341 | .max_write_len = 256, |
1342 | }, | |
31e4c28d VG |
1343 | { |
1344 | .name = "weight", | |
062a644d VG |
1345 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1346 | BLKIO_PROP_weight), | |
1347 | .read_u64 = blkiocg_file_read_u64, | |
1348 | .write_u64 = blkiocg_file_write_u64, | |
31e4c28d | 1349 | }, |
22084190 VG |
1350 | { |
1351 | .name = "time", | |
062a644d VG |
1352 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1353 | BLKIO_PROP_time), | |
1354 | .read_map = blkiocg_file_read_map, | |
22084190 VG |
1355 | }, |
1356 | { | |
1357 | .name = "sectors", | |
062a644d VG |
1358 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1359 | BLKIO_PROP_sectors), | |
1360 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1361 | }, |
1362 | { | |
1363 | .name = "io_service_bytes", | |
062a644d VG |
1364 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1365 | BLKIO_PROP_io_service_bytes), | |
1366 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1367 | }, |
1368 | { | |
1369 | .name = "io_serviced", | |
062a644d VG |
1370 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1371 | BLKIO_PROP_io_serviced), | |
1372 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1373 | }, |
1374 | { | |
1375 | .name = "io_service_time", | |
062a644d VG |
1376 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1377 | BLKIO_PROP_io_service_time), | |
1378 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1379 | }, |
1380 | { | |
1381 | .name = "io_wait_time", | |
062a644d VG |
1382 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1383 | BLKIO_PROP_io_wait_time), | |
1384 | .read_map = blkiocg_file_read_map, | |
84c124da | 1385 | }, |
812d4026 DS |
1386 | { |
1387 | .name = "io_merged", | |
062a644d VG |
1388 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1389 | BLKIO_PROP_io_merged), | |
1390 | .read_map = blkiocg_file_read_map, | |
812d4026 | 1391 | }, |
cdc1184c DS |
1392 | { |
1393 | .name = "io_queued", | |
062a644d VG |
1394 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1395 | BLKIO_PROP_io_queued), | |
1396 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1397 | }, |
84c124da DS |
1398 | { |
1399 | .name = "reset_stats", | |
1400 | .write_u64 = blkiocg_reset_stats, | |
22084190 | 1401 | }, |
13f98250 VG |
1402 | #ifdef CONFIG_BLK_DEV_THROTTLING |
1403 | { | |
1404 | .name = "throttle.read_bps_device", | |
1405 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1406 | BLKIO_THROTL_read_bps_device), | |
1407 | .read_seq_string = blkiocg_file_read, | |
1408 | .write_string = blkiocg_file_write, | |
1409 | .max_write_len = 256, | |
1410 | }, | |
1411 | ||
1412 | { | |
1413 | .name = "throttle.write_bps_device", | |
1414 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1415 | BLKIO_THROTL_write_bps_device), | |
1416 | .read_seq_string = blkiocg_file_read, | |
1417 | .write_string = blkiocg_file_write, | |
1418 | .max_write_len = 256, | |
1419 | }, | |
1420 | ||
1421 | { | |
1422 | .name = "throttle.read_iops_device", | |
1423 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1424 | BLKIO_THROTL_read_iops_device), | |
1425 | .read_seq_string = blkiocg_file_read, | |
1426 | .write_string = blkiocg_file_write, | |
1427 | .max_write_len = 256, | |
1428 | }, | |
1429 | ||
1430 | { | |
1431 | .name = "throttle.write_iops_device", | |
1432 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1433 | BLKIO_THROTL_write_iops_device), | |
1434 | .read_seq_string = blkiocg_file_read, | |
1435 | .write_string = blkiocg_file_write, | |
1436 | .max_write_len = 256, | |
1437 | }, | |
1438 | { | |
1439 | .name = "throttle.io_service_bytes", | |
1440 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1441 | BLKIO_THROTL_io_service_bytes), | |
1442 | .read_map = blkiocg_file_read_map, | |
1443 | }, | |
1444 | { | |
1445 | .name = "throttle.io_serviced", | |
1446 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1447 | BLKIO_THROTL_io_serviced), | |
1448 | .read_map = blkiocg_file_read_map, | |
1449 | }, | |
1450 | #endif /* CONFIG_BLK_DEV_THROTTLING */ | |
1451 | ||
22084190 | 1452 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
cdc1184c DS |
1453 | { |
1454 | .name = "avg_queue_size", | |
062a644d VG |
1455 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1456 | BLKIO_PROP_avg_queue_size), | |
1457 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1458 | }, |
812df48d DS |
1459 | { |
1460 | .name = "group_wait_time", | |
062a644d VG |
1461 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1462 | BLKIO_PROP_group_wait_time), | |
1463 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1464 | }, |
1465 | { | |
1466 | .name = "idle_time", | |
062a644d VG |
1467 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1468 | BLKIO_PROP_idle_time), | |
1469 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1470 | }, |
1471 | { | |
1472 | .name = "empty_time", | |
062a644d VG |
1473 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1474 | BLKIO_PROP_empty_time), | |
1475 | .read_map = blkiocg_file_read_map, | |
812df48d | 1476 | }, |
cdc1184c | 1477 | { |
22084190 | 1478 | .name = "dequeue", |
062a644d VG |
1479 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1480 | BLKIO_PROP_dequeue), | |
1481 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1482 | }, |
9026e521 JT |
1483 | { |
1484 | .name = "unaccounted_time", | |
1485 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, | |
1486 | BLKIO_PROP_unaccounted_time), | |
1487 | .read_map = blkiocg_file_read_map, | |
1488 | }, | |
22084190 | 1489 | #endif |
31e4c28d VG |
1490 | }; |
1491 | ||
1492 | static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1493 | { | |
1494 | return cgroup_add_files(cgroup, subsys, blkio_files, | |
1495 | ARRAY_SIZE(blkio_files)); | |
1496 | } | |
1497 | ||
7ee9c562 TH |
1498 | static int blkiocg_pre_destroy(struct cgroup_subsys *subsys, |
1499 | struct cgroup *cgroup) | |
31e4c28d VG |
1500 | { |
1501 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
b1c35769 VG |
1502 | unsigned long flags; |
1503 | struct blkio_group *blkg; | |
ca32aefc | 1504 | struct request_queue *q; |
3e252066 | 1505 | struct blkio_policy_type *blkiop; |
b1c35769 VG |
1506 | |
1507 | rcu_read_lock(); | |
7ee9c562 | 1508 | |
0f3942a3 JA |
1509 | do { |
1510 | spin_lock_irqsave(&blkcg->lock, flags); | |
b1c35769 | 1511 | |
0f3942a3 JA |
1512 | if (hlist_empty(&blkcg->blkg_list)) { |
1513 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
1514 | break; | |
1515 | } | |
b1c35769 | 1516 | |
0f3942a3 JA |
1517 | blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group, |
1518 | blkcg_node); | |
ca32aefc | 1519 | q = rcu_dereference(blkg->q); |
0f3942a3 | 1520 | __blkiocg_del_blkio_group(blkg); |
31e4c28d | 1521 | |
0f3942a3 | 1522 | spin_unlock_irqrestore(&blkcg->lock, flags); |
b1c35769 | 1523 | |
0f3942a3 JA |
1524 | /* |
1525 | * This blkio_group is being unlinked as associated cgroup is | |
1526 | * going away. Let all the IO controlling policies know about | |
61014e96 | 1527 | * this event. |
0f3942a3 JA |
1528 | */ |
1529 | spin_lock(&blkio_list_lock); | |
61014e96 VG |
1530 | list_for_each_entry(blkiop, &blkio_list, list) { |
1531 | if (blkiop->plid != blkg->plid) | |
1532 | continue; | |
ca32aefc | 1533 | blkiop->ops.blkio_unlink_group_fn(q, blkg); |
61014e96 | 1534 | } |
0f3942a3 JA |
1535 | spin_unlock(&blkio_list_lock); |
1536 | } while (1); | |
34d0f179 | 1537 | |
b1c35769 | 1538 | rcu_read_unlock(); |
7ee9c562 TH |
1539 | |
1540 | return 0; | |
1541 | } | |
1542 | ||
1543 | static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1544 | { | |
1545 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
1546 | ||
67523c48 BB |
1547 | if (blkcg != &blkio_root_cgroup) |
1548 | kfree(blkcg); | |
31e4c28d VG |
1549 | } |
1550 | ||
1551 | static struct cgroup_subsys_state * | |
1552 | blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1553 | { | |
0341509f LZ |
1554 | struct blkio_cgroup *blkcg; |
1555 | struct cgroup *parent = cgroup->parent; | |
31e4c28d | 1556 | |
0341509f | 1557 | if (!parent) { |
31e4c28d VG |
1558 | blkcg = &blkio_root_cgroup; |
1559 | goto done; | |
1560 | } | |
1561 | ||
31e4c28d VG |
1562 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
1563 | if (!blkcg) | |
1564 | return ERR_PTR(-ENOMEM); | |
1565 | ||
1566 | blkcg->weight = BLKIO_WEIGHT_DEFAULT; | |
1567 | done: | |
1568 | spin_lock_init(&blkcg->lock); | |
1569 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
1570 | ||
1571 | return &blkcg->css; | |
1572 | } | |
1573 | ||
5efd6113 TH |
1574 | /** |
1575 | * blkcg_init_queue - initialize blkcg part of request queue | |
1576 | * @q: request_queue to initialize | |
1577 | * | |
1578 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1579 | * part of new request_queue @q. | |
1580 | * | |
1581 | * RETURNS: | |
1582 | * 0 on success, -errno on failure. | |
1583 | */ | |
1584 | int blkcg_init_queue(struct request_queue *q) | |
1585 | { | |
923adde1 TH |
1586 | int ret; |
1587 | ||
5efd6113 TH |
1588 | might_sleep(); |
1589 | ||
923adde1 TH |
1590 | ret = blk_throtl_init(q); |
1591 | if (ret) | |
1592 | return ret; | |
1593 | ||
1594 | mutex_lock(&all_q_mutex); | |
1595 | INIT_LIST_HEAD(&q->all_q_node); | |
1596 | list_add_tail(&q->all_q_node, &all_q_list); | |
1597 | mutex_unlock(&all_q_mutex); | |
1598 | ||
1599 | return 0; | |
5efd6113 TH |
1600 | } |
1601 | ||
1602 | /** | |
1603 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1604 | * @q: request_queue to drain | |
1605 | * | |
1606 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1607 | */ | |
1608 | void blkcg_drain_queue(struct request_queue *q) | |
1609 | { | |
1610 | lockdep_assert_held(q->queue_lock); | |
1611 | ||
1612 | blk_throtl_drain(q); | |
1613 | } | |
1614 | ||
1615 | /** | |
1616 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1617 | * @q: request_queue being released | |
1618 | * | |
1619 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1620 | */ | |
1621 | void blkcg_exit_queue(struct request_queue *q) | |
1622 | { | |
923adde1 TH |
1623 | mutex_lock(&all_q_mutex); |
1624 | list_del_init(&q->all_q_node); | |
1625 | mutex_unlock(&all_q_mutex); | |
1626 | ||
5efd6113 TH |
1627 | blk_throtl_exit(q); |
1628 | } | |
1629 | ||
31e4c28d VG |
1630 | /* |
1631 | * We cannot support shared io contexts, as we have no mean to support | |
1632 | * two tasks with the same ioc in two different groups without major rework | |
1633 | * of the main cic data structures. For now we allow a task to change | |
1634 | * its cgroup only if it's the only owner of its ioc. | |
1635 | */ | |
bb9d97b6 TH |
1636 | static int blkiocg_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
1637 | struct cgroup_taskset *tset) | |
31e4c28d | 1638 | { |
bb9d97b6 | 1639 | struct task_struct *task; |
31e4c28d VG |
1640 | struct io_context *ioc; |
1641 | int ret = 0; | |
1642 | ||
1643 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
bb9d97b6 TH |
1644 | cgroup_taskset_for_each(task, cgrp, tset) { |
1645 | task_lock(task); | |
1646 | ioc = task->io_context; | |
1647 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1648 | ret = -EINVAL; | |
1649 | task_unlock(task); | |
1650 | if (ret) | |
1651 | break; | |
1652 | } | |
31e4c28d VG |
1653 | return ret; |
1654 | } | |
1655 | ||
bb9d97b6 TH |
1656 | static void blkiocg_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
1657 | struct cgroup_taskset *tset) | |
31e4c28d | 1658 | { |
bb9d97b6 | 1659 | struct task_struct *task; |
31e4c28d VG |
1660 | struct io_context *ioc; |
1661 | ||
bb9d97b6 | 1662 | cgroup_taskset_for_each(task, cgrp, tset) { |
b3c9dd18 LT |
1663 | /* we don't lose anything even if ioc allocation fails */ |
1664 | ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE); | |
1665 | if (ioc) { | |
1666 | ioc_cgroup_changed(ioc); | |
11a3122f | 1667 | put_io_context(ioc); |
b3c9dd18 | 1668 | } |
bb9d97b6 | 1669 | } |
31e4c28d VG |
1670 | } |
1671 | ||
923adde1 TH |
1672 | static void blkcg_bypass_start(void) |
1673 | __acquires(&all_q_mutex) | |
1674 | { | |
1675 | struct request_queue *q; | |
1676 | ||
1677 | mutex_lock(&all_q_mutex); | |
1678 | ||
1679 | list_for_each_entry(q, &all_q_list, all_q_node) { | |
1680 | blk_queue_bypass_start(q); | |
1681 | blkg_destroy_all(q); | |
1682 | } | |
1683 | } | |
1684 | ||
1685 | static void blkcg_bypass_end(void) | |
1686 | __releases(&all_q_mutex) | |
1687 | { | |
1688 | struct request_queue *q; | |
1689 | ||
1690 | list_for_each_entry(q, &all_q_list, all_q_node) | |
1691 | blk_queue_bypass_end(q); | |
1692 | ||
1693 | mutex_unlock(&all_q_mutex); | |
1694 | } | |
1695 | ||
3e252066 VG |
1696 | void blkio_policy_register(struct blkio_policy_type *blkiop) |
1697 | { | |
923adde1 | 1698 | blkcg_bypass_start(); |
3e252066 | 1699 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1700 | |
1701 | BUG_ON(blkio_policy[blkiop->plid]); | |
1702 | blkio_policy[blkiop->plid] = blkiop; | |
3e252066 | 1703 | list_add_tail(&blkiop->list, &blkio_list); |
035d10b2 | 1704 | |
3e252066 | 1705 | spin_unlock(&blkio_list_lock); |
923adde1 | 1706 | blkcg_bypass_end(); |
3e252066 VG |
1707 | } |
1708 | EXPORT_SYMBOL_GPL(blkio_policy_register); | |
1709 | ||
1710 | void blkio_policy_unregister(struct blkio_policy_type *blkiop) | |
1711 | { | |
923adde1 | 1712 | blkcg_bypass_start(); |
3e252066 | 1713 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1714 | |
1715 | BUG_ON(blkio_policy[blkiop->plid] != blkiop); | |
1716 | blkio_policy[blkiop->plid] = NULL; | |
3e252066 | 1717 | list_del_init(&blkiop->list); |
035d10b2 | 1718 | |
3e252066 | 1719 | spin_unlock(&blkio_list_lock); |
923adde1 | 1720 | blkcg_bypass_end(); |
3e252066 VG |
1721 | } |
1722 | EXPORT_SYMBOL_GPL(blkio_policy_unregister); |