]>
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 | ||
c1768268 TH |
81 | static inline void blkio_update_group_weight(struct blkio_group *blkg, |
82 | int plid, unsigned int weight) | |
062a644d VG |
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 */ | |
c1768268 | 88 | if (blkiop->plid != plid) |
062a644d VG |
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 | ||
c1768268 TH |
96 | static inline void blkio_update_group_bps(struct blkio_group *blkg, int plid, |
97 | u64 bps, int fileid) | |
4c9eefa1 VG |
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 */ | |
c1768268 | 104 | if (blkiop->plid != plid) |
4c9eefa1 VG |
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 | 119 | static inline void blkio_update_group_iops(struct blkio_group *blkg, |
c1768268 TH |
120 | int plid, unsigned int iops, |
121 | int fileid) | |
7702e8f4 VG |
122 | { |
123 | struct blkio_policy_type *blkiop; | |
124 | ||
125 | list_for_each_entry(blkiop, &blkio_list, list) { | |
126 | ||
127 | /* If this policy does not own the blkg, do not send updates */ | |
c1768268 | 128 | if (blkiop->plid != plid) |
7702e8f4 VG |
129 | continue; |
130 | ||
131 | if (fileid == BLKIO_THROTL_read_iops_device | |
132 | && blkiop->ops.blkio_update_group_read_iops_fn) | |
ca32aefc | 133 | blkiop->ops.blkio_update_group_read_iops_fn(blkg->q, |
fe071437 | 134 | blkg, iops); |
7702e8f4 VG |
135 | |
136 | if (fileid == BLKIO_THROTL_write_iops_device | |
137 | && blkiop->ops.blkio_update_group_write_iops_fn) | |
ca32aefc | 138 | blkiop->ops.blkio_update_group_write_iops_fn(blkg->q, |
fe071437 | 139 | blkg,iops); |
7702e8f4 VG |
140 | } |
141 | } | |
142 | ||
9195291e DS |
143 | /* |
144 | * Add to the appropriate stat variable depending on the request type. | |
145 | * This should be called with the blkg->stats_lock held. | |
146 | */ | |
84c124da DS |
147 | static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction, |
148 | bool sync) | |
9195291e | 149 | { |
84c124da DS |
150 | if (direction) |
151 | stat[BLKIO_STAT_WRITE] += add; | |
9195291e | 152 | else |
84c124da DS |
153 | stat[BLKIO_STAT_READ] += add; |
154 | if (sync) | |
155 | stat[BLKIO_STAT_SYNC] += add; | |
9195291e | 156 | else |
84c124da | 157 | stat[BLKIO_STAT_ASYNC] += add; |
9195291e DS |
158 | } |
159 | ||
cdc1184c DS |
160 | /* |
161 | * Decrements the appropriate stat variable if non-zero depending on the | |
162 | * request type. Panics on value being zero. | |
163 | * This should be called with the blkg->stats_lock held. | |
164 | */ | |
165 | static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync) | |
166 | { | |
167 | if (direction) { | |
168 | BUG_ON(stat[BLKIO_STAT_WRITE] == 0); | |
169 | stat[BLKIO_STAT_WRITE]--; | |
170 | } else { | |
171 | BUG_ON(stat[BLKIO_STAT_READ] == 0); | |
172 | stat[BLKIO_STAT_READ]--; | |
173 | } | |
174 | if (sync) { | |
175 | BUG_ON(stat[BLKIO_STAT_SYNC] == 0); | |
176 | stat[BLKIO_STAT_SYNC]--; | |
177 | } else { | |
178 | BUG_ON(stat[BLKIO_STAT_ASYNC] == 0); | |
179 | stat[BLKIO_STAT_ASYNC]--; | |
180 | } | |
181 | } | |
182 | ||
183 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
812df48d DS |
184 | /* This should be called with the blkg->stats_lock held. */ |
185 | static void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
c1768268 TH |
186 | struct blkio_policy_type *pol, |
187 | struct blkio_group *curr_blkg) | |
812df48d | 188 | { |
c1768268 | 189 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
549d3aa8 TH |
190 | |
191 | if (blkio_blkg_waiting(&pd->stats)) | |
812df48d DS |
192 | return; |
193 | if (blkg == curr_blkg) | |
194 | return; | |
549d3aa8 TH |
195 | pd->stats.start_group_wait_time = sched_clock(); |
196 | blkio_mark_blkg_waiting(&pd->stats); | |
812df48d DS |
197 | } |
198 | ||
199 | /* This should be called with the blkg->stats_lock held. */ | |
200 | static void blkio_update_group_wait_time(struct blkio_group_stats *stats) | |
201 | { | |
202 | unsigned long long now; | |
203 | ||
204 | if (!blkio_blkg_waiting(stats)) | |
205 | return; | |
206 | ||
207 | now = sched_clock(); | |
208 | if (time_after64(now, stats->start_group_wait_time)) | |
209 | stats->group_wait_time += now - stats->start_group_wait_time; | |
210 | blkio_clear_blkg_waiting(stats); | |
211 | } | |
212 | ||
213 | /* This should be called with the blkg->stats_lock held. */ | |
214 | static void blkio_end_empty_time(struct blkio_group_stats *stats) | |
215 | { | |
216 | unsigned long long now; | |
217 | ||
218 | if (!blkio_blkg_empty(stats)) | |
219 | return; | |
220 | ||
221 | now = sched_clock(); | |
222 | if (time_after64(now, stats->start_empty_time)) | |
223 | stats->empty_time += now - stats->start_empty_time; | |
224 | blkio_clear_blkg_empty(stats); | |
225 | } | |
226 | ||
c1768268 TH |
227 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg, |
228 | struct blkio_policy_type *pol) | |
812df48d | 229 | { |
c1768268 | 230 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
812df48d DS |
231 | unsigned long flags; |
232 | ||
233 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 TH |
234 | BUG_ON(blkio_blkg_idling(&pd->stats)); |
235 | pd->stats.start_idle_time = sched_clock(); | |
236 | blkio_mark_blkg_idling(&pd->stats); | |
812df48d DS |
237 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
238 | } | |
239 | EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats); | |
240 | ||
c1768268 TH |
241 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg, |
242 | struct blkio_policy_type *pol) | |
812df48d | 243 | { |
c1768268 | 244 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
812df48d DS |
245 | unsigned long flags; |
246 | unsigned long long now; | |
247 | struct blkio_group_stats *stats; | |
248 | ||
249 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 250 | stats = &pd->stats; |
812df48d DS |
251 | if (blkio_blkg_idling(stats)) { |
252 | now = sched_clock(); | |
253 | if (time_after64(now, stats->start_idle_time)) | |
254 | stats->idle_time += now - stats->start_idle_time; | |
255 | blkio_clear_blkg_idling(stats); | |
256 | } | |
257 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
258 | } | |
259 | EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats); | |
260 | ||
c1768268 TH |
261 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg, |
262 | struct blkio_policy_type *pol) | |
cdc1184c | 263 | { |
c1768268 | 264 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
cdc1184c DS |
265 | unsigned long flags; |
266 | struct blkio_group_stats *stats; | |
267 | ||
268 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 269 | stats = &pd->stats; |
cdc1184c DS |
270 | stats->avg_queue_size_sum += |
271 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] + | |
272 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]; | |
273 | stats->avg_queue_size_samples++; | |
812df48d | 274 | blkio_update_group_wait_time(stats); |
cdc1184c DS |
275 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
276 | } | |
a11cdaa7 DS |
277 | EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats); |
278 | ||
c1768268 TH |
279 | void blkiocg_set_start_empty_time(struct blkio_group *blkg, |
280 | struct blkio_policy_type *pol) | |
28baf442 | 281 | { |
c1768268 | 282 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
28baf442 DS |
283 | unsigned long flags; |
284 | struct blkio_group_stats *stats; | |
285 | ||
286 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 287 | stats = &pd->stats; |
28baf442 DS |
288 | |
289 | if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] || | |
290 | stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) { | |
291 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
292 | return; | |
293 | } | |
294 | ||
295 | /* | |
e5ff082e VG |
296 | * group is already marked empty. This can happen if cfqq got new |
297 | * request in parent group and moved to this group while being added | |
298 | * to service tree. Just ignore the event and move on. | |
28baf442 | 299 | */ |
e5ff082e VG |
300 | if(blkio_blkg_empty(stats)) { |
301 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
302 | return; | |
303 | } | |
304 | ||
28baf442 DS |
305 | stats->start_empty_time = sched_clock(); |
306 | blkio_mark_blkg_empty(stats); | |
307 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
308 | } | |
309 | EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time); | |
310 | ||
a11cdaa7 | 311 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
c1768268 TH |
312 | struct blkio_policy_type *pol, |
313 | unsigned long dequeue) | |
a11cdaa7 | 314 | { |
c1768268 | 315 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
549d3aa8 TH |
316 | |
317 | pd->stats.dequeue += dequeue; | |
a11cdaa7 DS |
318 | } |
319 | EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats); | |
812df48d DS |
320 | #else |
321 | static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg, | |
c1768268 TH |
322 | struct blkio_policy_type *pol, |
323 | struct blkio_group *curr_blkg) { } | |
324 | static inline void blkio_end_empty_time(struct blkio_group_stats *stats) { } | |
cdc1184c DS |
325 | #endif |
326 | ||
a11cdaa7 | 327 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
c1768268 TH |
328 | struct blkio_policy_type *pol, |
329 | struct blkio_group *curr_blkg, bool direction, | |
330 | bool sync) | |
cdc1184c | 331 | { |
c1768268 | 332 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
cdc1184c DS |
333 | unsigned long flags; |
334 | ||
335 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 336 | blkio_add_stat(pd->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction, |
cdc1184c | 337 | sync); |
549d3aa8 | 338 | blkio_end_empty_time(&pd->stats); |
c1768268 | 339 | blkio_set_start_group_wait_time(blkg, pol, curr_blkg); |
cdc1184c DS |
340 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
341 | } | |
a11cdaa7 | 342 | EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats); |
cdc1184c | 343 | |
a11cdaa7 | 344 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
c1768268 TH |
345 | struct blkio_policy_type *pol, |
346 | bool direction, bool sync) | |
cdc1184c | 347 | { |
c1768268 | 348 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
cdc1184c DS |
349 | unsigned long flags; |
350 | ||
351 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 352 | blkio_check_and_dec_stat(pd->stats.stat_arr[BLKIO_STAT_QUEUED], |
cdc1184c DS |
353 | direction, sync); |
354 | spin_unlock_irqrestore(&blkg->stats_lock, flags); | |
355 | } | |
a11cdaa7 | 356 | EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats); |
cdc1184c | 357 | |
c1768268 TH |
358 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
359 | struct blkio_policy_type *pol, | |
360 | unsigned long time, | |
361 | unsigned long unaccounted_time) | |
22084190 | 362 | { |
c1768268 | 363 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
303a3acb DS |
364 | unsigned long flags; |
365 | ||
366 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 367 | pd->stats.time += time; |
a23e6869 | 368 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
549d3aa8 | 369 | pd->stats.unaccounted_time += unaccounted_time; |
a23e6869 | 370 | #endif |
303a3acb | 371 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
22084190 | 372 | } |
303a3acb | 373 | EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used); |
22084190 | 374 | |
5624a4e4 VG |
375 | /* |
376 | * should be called under rcu read lock or queue lock to make sure blkg pointer | |
377 | * is valid. | |
378 | */ | |
84c124da | 379 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
c1768268 TH |
380 | struct blkio_policy_type *pol, |
381 | uint64_t bytes, bool direction, bool sync) | |
9195291e | 382 | { |
c1768268 | 383 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
5624a4e4 | 384 | struct blkio_group_stats_cpu *stats_cpu; |
575969a0 VG |
385 | unsigned long flags; |
386 | ||
387 | /* | |
388 | * Disabling interrupts to provide mutual exclusion between two | |
389 | * writes on same cpu. It probably is not needed for 64bit. Not | |
390 | * optimizing that case yet. | |
391 | */ | |
392 | local_irq_save(flags); | |
9195291e | 393 | |
549d3aa8 | 394 | stats_cpu = this_cpu_ptr(pd->stats_cpu); |
5624a4e4 | 395 | |
575969a0 | 396 | u64_stats_update_begin(&stats_cpu->syncp); |
5624a4e4 VG |
397 | stats_cpu->sectors += bytes >> 9; |
398 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED], | |
399 | 1, direction, sync); | |
400 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES], | |
401 | bytes, direction, sync); | |
575969a0 VG |
402 | u64_stats_update_end(&stats_cpu->syncp); |
403 | local_irq_restore(flags); | |
9195291e | 404 | } |
84c124da | 405 | EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats); |
9195291e | 406 | |
84c124da | 407 | void blkiocg_update_completion_stats(struct blkio_group *blkg, |
c1768268 TH |
408 | struct blkio_policy_type *pol, |
409 | uint64_t start_time, | |
410 | uint64_t io_start_time, bool direction, | |
411 | bool sync) | |
9195291e | 412 | { |
c1768268 | 413 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
9195291e DS |
414 | struct blkio_group_stats *stats; |
415 | unsigned long flags; | |
416 | unsigned long long now = sched_clock(); | |
417 | ||
418 | spin_lock_irqsave(&blkg->stats_lock, flags); | |
549d3aa8 | 419 | stats = &pd->stats; |
84c124da DS |
420 | if (time_after64(now, io_start_time)) |
421 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME], | |
422 | now - io_start_time, direction, sync); | |
423 | if (time_after64(io_start_time, start_time)) | |
424 | blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME], | |
425 | io_start_time - start_time, direction, sync); | |
9195291e DS |
426 | spin_unlock_irqrestore(&blkg->stats_lock, flags); |
427 | } | |
84c124da | 428 | EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats); |
9195291e | 429 | |
317389a7 | 430 | /* Merged stats are per cpu. */ |
c1768268 TH |
431 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, |
432 | struct blkio_policy_type *pol, | |
433 | bool direction, bool sync) | |
812d4026 | 434 | { |
c1768268 | 435 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
317389a7 | 436 | struct blkio_group_stats_cpu *stats_cpu; |
812d4026 DS |
437 | unsigned long flags; |
438 | ||
317389a7 VG |
439 | /* |
440 | * Disabling interrupts to provide mutual exclusion between two | |
441 | * writes on same cpu. It probably is not needed for 64bit. Not | |
442 | * optimizing that case yet. | |
443 | */ | |
444 | local_irq_save(flags); | |
445 | ||
549d3aa8 | 446 | stats_cpu = this_cpu_ptr(pd->stats_cpu); |
317389a7 VG |
447 | |
448 | u64_stats_update_begin(&stats_cpu->syncp); | |
449 | blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1, | |
450 | direction, sync); | |
451 | u64_stats_update_end(&stats_cpu->syncp); | |
452 | local_irq_restore(flags); | |
812d4026 DS |
453 | } |
454 | EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats); | |
455 | ||
0381411e TH |
456 | /** |
457 | * blkg_free - free a blkg | |
458 | * @blkg: blkg to free | |
459 | * | |
460 | * Free @blkg which may be partially allocated. | |
461 | */ | |
462 | static void blkg_free(struct blkio_group *blkg) | |
463 | { | |
e8989fae | 464 | int i; |
549d3aa8 TH |
465 | |
466 | if (!blkg) | |
467 | return; | |
468 | ||
e8989fae TH |
469 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
470 | struct blkg_policy_data *pd = blkg->pd[i]; | |
471 | ||
472 | if (pd) { | |
473 | free_percpu(pd->stats_cpu); | |
474 | kfree(pd); | |
475 | } | |
0381411e | 476 | } |
e8989fae | 477 | |
549d3aa8 | 478 | kfree(blkg); |
0381411e TH |
479 | } |
480 | ||
481 | /** | |
482 | * blkg_alloc - allocate a blkg | |
483 | * @blkcg: block cgroup the new blkg is associated with | |
484 | * @q: request_queue the new blkg is associated with | |
0381411e | 485 | * |
e8989fae | 486 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e TH |
487 | * |
488 | * FIXME: Should be called with queue locked but currently isn't due to | |
489 | * percpu stat breakage. | |
490 | */ | |
491 | static struct blkio_group *blkg_alloc(struct blkio_cgroup *blkcg, | |
e8989fae | 492 | struct request_queue *q) |
0381411e TH |
493 | { |
494 | struct blkio_group *blkg; | |
e8989fae | 495 | int i; |
0381411e TH |
496 | |
497 | /* alloc and init base part */ | |
498 | blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node); | |
499 | if (!blkg) | |
500 | return NULL; | |
501 | ||
502 | spin_lock_init(&blkg->stats_lock); | |
503 | rcu_assign_pointer(blkg->q, q); | |
e8989fae | 504 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 505 | blkg->blkcg = blkcg; |
1adaf3dd | 506 | blkg->refcnt = 1; |
0381411e TH |
507 | cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path)); |
508 | ||
e8989fae TH |
509 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
510 | struct blkio_policy_type *pol = blkio_policy[i]; | |
511 | struct blkg_policy_data *pd; | |
0381411e | 512 | |
e8989fae TH |
513 | if (!pol) |
514 | continue; | |
515 | ||
516 | /* alloc per-policy data and attach it to blkg */ | |
517 | pd = kzalloc_node(sizeof(*pd) + pol->pdata_size, GFP_ATOMIC, | |
518 | q->node); | |
519 | if (!pd) { | |
520 | blkg_free(blkg); | |
521 | return NULL; | |
522 | } | |
549d3aa8 | 523 | |
e8989fae TH |
524 | blkg->pd[i] = pd; |
525 | pd->blkg = blkg; | |
549d3aa8 | 526 | |
e8989fae TH |
527 | /* broken, read comment in the callsite */ |
528 | pd->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu); | |
529 | if (!pd->stats_cpu) { | |
530 | blkg_free(blkg); | |
531 | return NULL; | |
532 | } | |
0381411e TH |
533 | } |
534 | ||
549d3aa8 | 535 | /* invoke per-policy init */ |
e8989fae TH |
536 | for (i = 0; i < BLKIO_NR_POLICIES; i++) { |
537 | struct blkio_policy_type *pol = blkio_policy[i]; | |
538 | ||
539 | if (pol) | |
540 | pol->ops.blkio_init_group_fn(blkg); | |
541 | } | |
542 | ||
0381411e TH |
543 | return blkg; |
544 | } | |
545 | ||
cd1604fa TH |
546 | struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg, |
547 | struct request_queue *q, | |
548 | enum blkio_policy_id plid, | |
549 | bool for_root) | |
550 | __releases(q->queue_lock) __acquires(q->queue_lock) | |
5624a4e4 | 551 | { |
cd1604fa | 552 | struct blkio_group *blkg, *new_blkg; |
5624a4e4 | 553 | |
cd1604fa TH |
554 | WARN_ON_ONCE(!rcu_read_lock_held()); |
555 | lockdep_assert_held(q->queue_lock); | |
556 | ||
557 | /* | |
558 | * This could be the first entry point of blkcg implementation and | |
559 | * we shouldn't allow anything to go through for a bypassing queue. | |
560 | * The following can be removed if blkg lookup is guaranteed to | |
561 | * fail on a bypassing queue. | |
562 | */ | |
563 | if (unlikely(blk_queue_bypass(q)) && !for_root) | |
564 | return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY); | |
565 | ||
e8989fae | 566 | blkg = blkg_lookup(blkcg, q); |
cd1604fa TH |
567 | if (blkg) |
568 | return blkg; | |
569 | ||
7ee9c562 | 570 | /* blkg holds a reference to blkcg */ |
cd1604fa TH |
571 | if (!css_tryget(&blkcg->css)) |
572 | return ERR_PTR(-EINVAL); | |
573 | ||
574 | /* | |
575 | * Allocate and initialize. | |
576 | * | |
577 | * FIXME: The following is broken. Percpu memory allocation | |
578 | * requires %GFP_KERNEL context and can't be performed from IO | |
579 | * path. Allocation here should inherently be atomic and the | |
580 | * following lock dancing can be removed once the broken percpu | |
581 | * allocation is fixed. | |
582 | */ | |
583 | spin_unlock_irq(q->queue_lock); | |
584 | rcu_read_unlock(); | |
585 | ||
e8989fae | 586 | new_blkg = blkg_alloc(blkcg, q); |
cd1604fa TH |
587 | |
588 | rcu_read_lock(); | |
589 | spin_lock_irq(q->queue_lock); | |
31e4c28d | 590 | |
cd1604fa TH |
591 | /* did bypass get turned on inbetween? */ |
592 | if (unlikely(blk_queue_bypass(q)) && !for_root) { | |
593 | blkg = ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY); | |
594 | goto out; | |
595 | } | |
596 | ||
597 | /* did someone beat us to it? */ | |
e8989fae | 598 | blkg = blkg_lookup(blkcg, q); |
cd1604fa TH |
599 | if (unlikely(blkg)) |
600 | goto out; | |
601 | ||
602 | /* did alloc fail? */ | |
0381411e | 603 | if (unlikely(!new_blkg)) { |
cd1604fa TH |
604 | blkg = ERR_PTR(-ENOMEM); |
605 | goto out; | |
606 | } | |
607 | ||
608 | /* insert */ | |
609 | spin_lock(&blkcg->lock); | |
610 | swap(blkg, new_blkg); | |
03aa264a | 611 | |
31e4c28d | 612 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
e8989fae TH |
613 | list_add(&blkg->q_node, &q->blkg_list); |
614 | q->nr_blkgs++; | |
03aa264a | 615 | |
cd1604fa TH |
616 | spin_unlock(&blkcg->lock); |
617 | out: | |
0381411e | 618 | blkg_free(new_blkg); |
cd1604fa | 619 | return blkg; |
31e4c28d | 620 | } |
cd1604fa | 621 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 622 | |
b1c35769 VG |
623 | static void __blkiocg_del_blkio_group(struct blkio_group *blkg) |
624 | { | |
625 | hlist_del_init_rcu(&blkg->blkcg_node); | |
b1c35769 VG |
626 | } |
627 | ||
628 | /* | |
629 | * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1 | |
630 | * indicating that blk_group was unhashed by the time we got to it. | |
631 | */ | |
31e4c28d VG |
632 | int blkiocg_del_blkio_group(struct blkio_group *blkg) |
633 | { | |
7ee9c562 | 634 | struct blkio_cgroup *blkcg = blkg->blkcg; |
b1c35769 | 635 | unsigned long flags; |
b1c35769 VG |
636 | int ret = 1; |
637 | ||
7ee9c562 TH |
638 | spin_lock_irqsave(&blkcg->lock, flags); |
639 | if (!hlist_unhashed(&blkg->blkcg_node)) { | |
640 | __blkiocg_del_blkio_group(blkg); | |
641 | ret = 0; | |
b1c35769 | 642 | } |
7ee9c562 | 643 | spin_unlock_irqrestore(&blkcg->lock, flags); |
0f3942a3 | 644 | |
b1c35769 | 645 | return ret; |
31e4c28d | 646 | } |
9d6a986c | 647 | EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group); |
31e4c28d VG |
648 | |
649 | /* called under rcu_read_lock(). */ | |
cd1604fa | 650 | struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg, |
e8989fae | 651 | struct request_queue *q) |
31e4c28d VG |
652 | { |
653 | struct blkio_group *blkg; | |
654 | struct hlist_node *n; | |
31e4c28d | 655 | |
ca32aefc | 656 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) |
e8989fae | 657 | if (blkg->q == q) |
31e4c28d | 658 | return blkg; |
31e4c28d VG |
659 | return NULL; |
660 | } | |
cd1604fa | 661 | EXPORT_SYMBOL_GPL(blkg_lookup); |
31e4c28d | 662 | |
e8989fae | 663 | static void blkg_destroy(struct blkio_group *blkg) |
03aa264a TH |
664 | { |
665 | struct request_queue *q = blkg->q; | |
666 | ||
667 | lockdep_assert_held(q->queue_lock); | |
668 | ||
669 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae TH |
670 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
671 | list_del_init(&blkg->q_node); | |
03aa264a | 672 | |
e8989fae TH |
673 | WARN_ON_ONCE(q->nr_blkgs <= 0); |
674 | q->nr_blkgs--; | |
03aa264a TH |
675 | |
676 | /* | |
677 | * Put the reference taken at the time of creation so that when all | |
678 | * queues are gone, group can be destroyed. | |
679 | */ | |
680 | blkg_put(blkg); | |
681 | } | |
682 | ||
e8989fae TH |
683 | /* |
684 | * XXX: This updates blkg policy data in-place for root blkg, which is | |
685 | * necessary across elevator switch and policy registration as root blkgs | |
686 | * aren't shot down. This broken and racy implementation is temporary. | |
687 | * Eventually, blkg shoot down will be replaced by proper in-place update. | |
688 | */ | |
689 | void update_root_blkg_pd(struct request_queue *q, enum blkio_policy_id plid) | |
690 | { | |
691 | struct blkio_policy_type *pol = blkio_policy[plid]; | |
692 | struct blkio_group *blkg = blkg_lookup(&blkio_root_cgroup, q); | |
693 | struct blkg_policy_data *pd; | |
694 | ||
695 | if (!blkg) | |
696 | return; | |
697 | ||
698 | kfree(blkg->pd[plid]); | |
699 | blkg->pd[plid] = NULL; | |
700 | ||
701 | if (!pol) | |
702 | return; | |
703 | ||
704 | pd = kzalloc(sizeof(*pd) + pol->pdata_size, GFP_KERNEL); | |
705 | WARN_ON_ONCE(!pd); | |
706 | ||
707 | pd->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu); | |
708 | WARN_ON_ONCE(!pd->stats_cpu); | |
709 | ||
710 | blkg->pd[plid] = pd; | |
711 | pd->blkg = blkg; | |
712 | pol->ops.blkio_init_group_fn(blkg); | |
713 | } | |
714 | EXPORT_SYMBOL_GPL(update_root_blkg_pd); | |
715 | ||
716 | void blkg_destroy_all(struct request_queue *q, bool destroy_root) | |
72e06c25 | 717 | { |
03aa264a | 718 | struct blkio_group *blkg, *n; |
72e06c25 TH |
719 | |
720 | while (true) { | |
721 | bool done = true; | |
722 | ||
72e06c25 TH |
723 | spin_lock_irq(q->queue_lock); |
724 | ||
e8989fae | 725 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
03aa264a TH |
726 | /* skip root? */ |
727 | if (!destroy_root && blkg->blkcg == &blkio_root_cgroup) | |
728 | continue; | |
729 | ||
730 | /* | |
731 | * If cgroup removal path got to blk_group first | |
732 | * and removed it from cgroup list, then it will | |
733 | * take care of destroying cfqg also. | |
734 | */ | |
735 | if (!blkiocg_del_blkio_group(blkg)) | |
e8989fae | 736 | blkg_destroy(blkg); |
03aa264a | 737 | else |
72e06c25 | 738 | done = false; |
03aa264a | 739 | } |
72e06c25 TH |
740 | |
741 | spin_unlock_irq(q->queue_lock); | |
72e06c25 | 742 | |
03aa264a TH |
743 | /* |
744 | * Group list may not be empty if we raced cgroup removal | |
745 | * and lost. cgroup removal is guaranteed to make forward | |
746 | * progress and retrying after a while is enough. This | |
747 | * ugliness is scheduled to be removed after locking | |
748 | * update. | |
749 | */ | |
72e06c25 TH |
750 | if (done) |
751 | break; | |
752 | ||
753 | msleep(10); /* just some random duration I like */ | |
754 | } | |
755 | } | |
03aa264a | 756 | EXPORT_SYMBOL_GPL(blkg_destroy_all); |
72e06c25 | 757 | |
1adaf3dd TH |
758 | static void blkg_rcu_free(struct rcu_head *rcu_head) |
759 | { | |
760 | blkg_free(container_of(rcu_head, struct blkio_group, rcu_head)); | |
761 | } | |
762 | ||
763 | void __blkg_release(struct blkio_group *blkg) | |
764 | { | |
765 | /* release the extra blkcg reference this blkg has been holding */ | |
766 | css_put(&blkg->blkcg->css); | |
767 | ||
768 | /* | |
769 | * A group is freed in rcu manner. But having an rcu lock does not | |
770 | * mean that one can access all the fields of blkg and assume these | |
771 | * are valid. For example, don't try to follow throtl_data and | |
772 | * request queue links. | |
773 | * | |
774 | * Having a reference to blkg under an rcu allows acess to only | |
775 | * values local to groups like group stats and group rate limits | |
776 | */ | |
777 | call_rcu(&blkg->rcu_head, blkg_rcu_free); | |
778 | } | |
779 | EXPORT_SYMBOL_GPL(__blkg_release); | |
780 | ||
c1768268 | 781 | static void blkio_reset_stats_cpu(struct blkio_group *blkg, int plid) |
f0bdc8cd | 782 | { |
c1768268 | 783 | struct blkg_policy_data *pd = blkg->pd[plid]; |
f0bdc8cd VG |
784 | struct blkio_group_stats_cpu *stats_cpu; |
785 | int i, j, k; | |
786 | /* | |
787 | * Note: On 64 bit arch this should not be an issue. This has the | |
788 | * possibility of returning some inconsistent value on 32bit arch | |
789 | * as 64bit update on 32bit is non atomic. Taking care of this | |
790 | * corner case makes code very complicated, like sending IPIs to | |
791 | * cpus, taking care of stats of offline cpus etc. | |
792 | * | |
793 | * reset stats is anyway more of a debug feature and this sounds a | |
794 | * corner case. So I am not complicating the code yet until and | |
795 | * unless this becomes a real issue. | |
796 | */ | |
797 | for_each_possible_cpu(i) { | |
549d3aa8 | 798 | stats_cpu = per_cpu_ptr(pd->stats_cpu, i); |
f0bdc8cd VG |
799 | stats_cpu->sectors = 0; |
800 | for(j = 0; j < BLKIO_STAT_CPU_NR; j++) | |
801 | for (k = 0; k < BLKIO_STAT_TOTAL; k++) | |
802 | stats_cpu->stat_arr_cpu[j][k] = 0; | |
803 | } | |
804 | } | |
805 | ||
303a3acb | 806 | static int |
84c124da | 807 | blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val) |
303a3acb DS |
808 | { |
809 | struct blkio_cgroup *blkcg; | |
810 | struct blkio_group *blkg; | |
812df48d | 811 | struct blkio_group_stats *stats; |
303a3acb | 812 | struct hlist_node *n; |
cdc1184c DS |
813 | uint64_t queued[BLKIO_STAT_TOTAL]; |
814 | int i; | |
812df48d DS |
815 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
816 | bool idling, waiting, empty; | |
817 | unsigned long long now = sched_clock(); | |
818 | #endif | |
303a3acb DS |
819 | |
820 | blkcg = cgroup_to_blkio_cgroup(cgroup); | |
e8989fae | 821 | spin_lock(&blkio_list_lock); |
303a3acb DS |
822 | spin_lock_irq(&blkcg->lock); |
823 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
e8989fae | 824 | struct blkio_policy_type *pol; |
549d3aa8 | 825 | |
e8989fae TH |
826 | list_for_each_entry(pol, &blkio_list, list) { |
827 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; | |
828 | ||
829 | spin_lock(&blkg->stats_lock); | |
830 | stats = &pd->stats; | |
812df48d | 831 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
e8989fae TH |
832 | idling = blkio_blkg_idling(stats); |
833 | waiting = blkio_blkg_waiting(stats); | |
834 | empty = blkio_blkg_empty(stats); | |
812df48d | 835 | #endif |
e8989fae TH |
836 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) |
837 | queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i]; | |
838 | memset(stats, 0, sizeof(struct blkio_group_stats)); | |
839 | for (i = 0; i < BLKIO_STAT_TOTAL; i++) | |
840 | stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i]; | |
812df48d | 841 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
e8989fae TH |
842 | if (idling) { |
843 | blkio_mark_blkg_idling(stats); | |
844 | stats->start_idle_time = now; | |
845 | } | |
846 | if (waiting) { | |
847 | blkio_mark_blkg_waiting(stats); | |
848 | stats->start_group_wait_time = now; | |
849 | } | |
850 | if (empty) { | |
851 | blkio_mark_blkg_empty(stats); | |
852 | stats->start_empty_time = now; | |
853 | } | |
812df48d | 854 | #endif |
e8989fae | 855 | spin_unlock(&blkg->stats_lock); |
f0bdc8cd | 856 | |
e8989fae TH |
857 | /* Reset Per cpu stats which don't take blkg->stats_lock */ |
858 | blkio_reset_stats_cpu(blkg, pol->plid); | |
859 | } | |
303a3acb | 860 | } |
f0bdc8cd | 861 | |
303a3acb | 862 | spin_unlock_irq(&blkcg->lock); |
e8989fae | 863 | spin_unlock(&blkio_list_lock); |
303a3acb DS |
864 | return 0; |
865 | } | |
866 | ||
7a4dd281 TH |
867 | static void blkio_get_key_name(enum stat_sub_type type, const char *dname, |
868 | char *str, int chars_left, bool diskname_only) | |
303a3acb | 869 | { |
7a4dd281 | 870 | snprintf(str, chars_left, "%s", dname); |
303a3acb DS |
871 | chars_left -= strlen(str); |
872 | if (chars_left <= 0) { | |
873 | printk(KERN_WARNING | |
874 | "Possibly incorrect cgroup stat display format"); | |
875 | return; | |
876 | } | |
84c124da DS |
877 | if (diskname_only) |
878 | return; | |
303a3acb | 879 | switch (type) { |
84c124da | 880 | case BLKIO_STAT_READ: |
303a3acb DS |
881 | strlcat(str, " Read", chars_left); |
882 | break; | |
84c124da | 883 | case BLKIO_STAT_WRITE: |
303a3acb DS |
884 | strlcat(str, " Write", chars_left); |
885 | break; | |
84c124da | 886 | case BLKIO_STAT_SYNC: |
303a3acb DS |
887 | strlcat(str, " Sync", chars_left); |
888 | break; | |
84c124da | 889 | case BLKIO_STAT_ASYNC: |
303a3acb DS |
890 | strlcat(str, " Async", chars_left); |
891 | break; | |
84c124da | 892 | case BLKIO_STAT_TOTAL: |
303a3acb DS |
893 | strlcat(str, " Total", chars_left); |
894 | break; | |
895 | default: | |
896 | strlcat(str, " Invalid", chars_left); | |
897 | } | |
898 | } | |
899 | ||
84c124da | 900 | static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val, |
7a4dd281 | 901 | struct cgroup_map_cb *cb, const char *dname) |
84c124da | 902 | { |
7a4dd281 | 903 | blkio_get_key_name(0, dname, str, chars_left, true); |
84c124da DS |
904 | cb->fill(cb, str, val); |
905 | return val; | |
906 | } | |
303a3acb | 907 | |
5624a4e4 | 908 | |
c1768268 | 909 | static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg, int plid, |
5624a4e4 VG |
910 | enum stat_type_cpu type, enum stat_sub_type sub_type) |
911 | { | |
c1768268 | 912 | struct blkg_policy_data *pd = blkg->pd[plid]; |
5624a4e4 VG |
913 | int cpu; |
914 | struct blkio_group_stats_cpu *stats_cpu; | |
575969a0 | 915 | u64 val = 0, tval; |
5624a4e4 VG |
916 | |
917 | for_each_possible_cpu(cpu) { | |
575969a0 | 918 | unsigned int start; |
549d3aa8 | 919 | stats_cpu = per_cpu_ptr(pd->stats_cpu, cpu); |
5624a4e4 | 920 | |
575969a0 VG |
921 | do { |
922 | start = u64_stats_fetch_begin(&stats_cpu->syncp); | |
923 | if (type == BLKIO_STAT_CPU_SECTORS) | |
924 | tval = stats_cpu->sectors; | |
925 | else | |
926 | tval = stats_cpu->stat_arr_cpu[type][sub_type]; | |
927 | } while(u64_stats_fetch_retry(&stats_cpu->syncp, start)); | |
928 | ||
929 | val += tval; | |
5624a4e4 VG |
930 | } |
931 | ||
932 | return val; | |
933 | } | |
934 | ||
c1768268 | 935 | static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg, int plid, |
7a4dd281 TH |
936 | struct cgroup_map_cb *cb, const char *dname, |
937 | enum stat_type_cpu type) | |
5624a4e4 VG |
938 | { |
939 | uint64_t disk_total, val; | |
940 | char key_str[MAX_KEY_LEN]; | |
941 | enum stat_sub_type sub_type; | |
942 | ||
943 | if (type == BLKIO_STAT_CPU_SECTORS) { | |
c1768268 | 944 | val = blkio_read_stat_cpu(blkg, plid, type, 0); |
7a4dd281 TH |
945 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, |
946 | dname); | |
5624a4e4 VG |
947 | } |
948 | ||
949 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; | |
950 | sub_type++) { | |
7a4dd281 TH |
951 | blkio_get_key_name(sub_type, dname, key_str, MAX_KEY_LEN, |
952 | false); | |
c1768268 | 953 | val = blkio_read_stat_cpu(blkg, plid, type, sub_type); |
5624a4e4 VG |
954 | cb->fill(cb, key_str, val); |
955 | } | |
956 | ||
c1768268 TH |
957 | disk_total = blkio_read_stat_cpu(blkg, plid, type, BLKIO_STAT_READ) + |
958 | blkio_read_stat_cpu(blkg, plid, type, BLKIO_STAT_WRITE); | |
5624a4e4 | 959 | |
7a4dd281 TH |
960 | blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN, |
961 | false); | |
5624a4e4 VG |
962 | cb->fill(cb, key_str, disk_total); |
963 | return disk_total; | |
964 | } | |
965 | ||
84c124da | 966 | /* This should be called with blkg->stats_lock held */ |
c1768268 | 967 | static uint64_t blkio_get_stat(struct blkio_group *blkg, int plid, |
7a4dd281 TH |
968 | struct cgroup_map_cb *cb, const char *dname, |
969 | enum stat_type type) | |
303a3acb | 970 | { |
c1768268 | 971 | struct blkg_policy_data *pd = blkg->pd[plid]; |
303a3acb DS |
972 | uint64_t disk_total; |
973 | char key_str[MAX_KEY_LEN]; | |
84c124da DS |
974 | enum stat_sub_type sub_type; |
975 | ||
976 | if (type == BLKIO_STAT_TIME) | |
977 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 978 | pd->stats.time, cb, dname); |
9026e521 | 979 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 JT |
980 | if (type == BLKIO_STAT_UNACCOUNTED_TIME) |
981 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 982 | pd->stats.unaccounted_time, cb, dname); |
cdc1184c | 983 | if (type == BLKIO_STAT_AVG_QUEUE_SIZE) { |
549d3aa8 TH |
984 | uint64_t sum = pd->stats.avg_queue_size_sum; |
985 | uint64_t samples = pd->stats.avg_queue_size_samples; | |
cdc1184c DS |
986 | if (samples) |
987 | do_div(sum, samples); | |
988 | else | |
989 | sum = 0; | |
7a4dd281 TH |
990 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, |
991 | sum, cb, dname); | |
cdc1184c | 992 | } |
812df48d DS |
993 | if (type == BLKIO_STAT_GROUP_WAIT_TIME) |
994 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 995 | pd->stats.group_wait_time, cb, dname); |
812df48d DS |
996 | if (type == BLKIO_STAT_IDLE_TIME) |
997 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 998 | pd->stats.idle_time, cb, dname); |
812df48d DS |
999 | if (type == BLKIO_STAT_EMPTY_TIME) |
1000 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 1001 | pd->stats.empty_time, cb, dname); |
84c124da DS |
1002 | if (type == BLKIO_STAT_DEQUEUE) |
1003 | return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, | |
549d3aa8 | 1004 | pd->stats.dequeue, cb, dname); |
84c124da | 1005 | #endif |
303a3acb | 1006 | |
84c124da DS |
1007 | for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL; |
1008 | sub_type++) { | |
7a4dd281 TH |
1009 | blkio_get_key_name(sub_type, dname, key_str, MAX_KEY_LEN, |
1010 | false); | |
549d3aa8 | 1011 | cb->fill(cb, key_str, pd->stats.stat_arr[type][sub_type]); |
303a3acb | 1012 | } |
549d3aa8 TH |
1013 | disk_total = pd->stats.stat_arr[type][BLKIO_STAT_READ] + |
1014 | pd->stats.stat_arr[type][BLKIO_STAT_WRITE]; | |
7a4dd281 TH |
1015 | blkio_get_key_name(BLKIO_STAT_TOTAL, dname, key_str, MAX_KEY_LEN, |
1016 | false); | |
303a3acb DS |
1017 | cb->fill(cb, key_str, disk_total); |
1018 | return disk_total; | |
1019 | } | |
1020 | ||
4bfd482e TH |
1021 | static int blkio_policy_parse_and_set(char *buf, enum blkio_policy_id plid, |
1022 | int fileid, struct blkio_cgroup *blkcg) | |
34d0f179 | 1023 | { |
ece84241 | 1024 | struct gendisk *disk = NULL; |
e56da7e2 | 1025 | struct blkio_group *blkg = NULL; |
549d3aa8 | 1026 | struct blkg_policy_data *pd; |
34d0f179 | 1027 | char *s[4], *p, *major_s = NULL, *minor_s = NULL; |
d11bb446 | 1028 | unsigned long major, minor; |
ece84241 TH |
1029 | int i = 0, ret = -EINVAL; |
1030 | int part; | |
34d0f179 | 1031 | dev_t dev; |
d11bb446 | 1032 | u64 temp; |
34d0f179 GJ |
1033 | |
1034 | memset(s, 0, sizeof(s)); | |
1035 | ||
1036 | while ((p = strsep(&buf, " ")) != NULL) { | |
1037 | if (!*p) | |
1038 | continue; | |
1039 | ||
1040 | s[i++] = p; | |
1041 | ||
1042 | /* Prevent from inputing too many things */ | |
1043 | if (i == 3) | |
1044 | break; | |
1045 | } | |
1046 | ||
1047 | if (i != 2) | |
ece84241 | 1048 | goto out; |
34d0f179 GJ |
1049 | |
1050 | p = strsep(&s[0], ":"); | |
1051 | if (p != NULL) | |
1052 | major_s = p; | |
1053 | else | |
ece84241 | 1054 | goto out; |
34d0f179 GJ |
1055 | |
1056 | minor_s = s[0]; | |
1057 | if (!minor_s) | |
ece84241 | 1058 | goto out; |
34d0f179 | 1059 | |
ece84241 TH |
1060 | if (strict_strtoul(major_s, 10, &major)) |
1061 | goto out; | |
34d0f179 | 1062 | |
ece84241 TH |
1063 | if (strict_strtoul(minor_s, 10, &minor)) |
1064 | goto out; | |
34d0f179 GJ |
1065 | |
1066 | dev = MKDEV(major, minor); | |
1067 | ||
ece84241 TH |
1068 | if (strict_strtoull(s[1], 10, &temp)) |
1069 | goto out; | |
34d0f179 | 1070 | |
e56da7e2 | 1071 | disk = get_gendisk(dev, &part); |
4bfd482e | 1072 | if (!disk || part) |
e56da7e2 | 1073 | goto out; |
e56da7e2 TH |
1074 | |
1075 | rcu_read_lock(); | |
1076 | ||
4bfd482e TH |
1077 | spin_lock_irq(disk->queue->queue_lock); |
1078 | blkg = blkg_lookup_create(blkcg, disk->queue, plid, false); | |
1079 | spin_unlock_irq(disk->queue->queue_lock); | |
e56da7e2 | 1080 | |
4bfd482e TH |
1081 | if (IS_ERR(blkg)) { |
1082 | ret = PTR_ERR(blkg); | |
1083 | goto out_unlock; | |
d11bb446 | 1084 | } |
34d0f179 | 1085 | |
549d3aa8 TH |
1086 | pd = blkg->pd[plid]; |
1087 | ||
062a644d VG |
1088 | switch (plid) { |
1089 | case BLKIO_POLICY_PROP: | |
d11bb446 WG |
1090 | if ((temp < BLKIO_WEIGHT_MIN && temp > 0) || |
1091 | temp > BLKIO_WEIGHT_MAX) | |
e56da7e2 | 1092 | goto out_unlock; |
34d0f179 | 1093 | |
549d3aa8 | 1094 | pd->conf.weight = temp; |
c1768268 | 1095 | blkio_update_group_weight(blkg, plid, temp ?: blkcg->weight); |
4c9eefa1 VG |
1096 | break; |
1097 | case BLKIO_POLICY_THROTL: | |
7702e8f4 VG |
1098 | switch(fileid) { |
1099 | case BLKIO_THROTL_read_bps_device: | |
549d3aa8 | 1100 | pd->conf.bps[READ] = temp; |
c1768268 | 1101 | blkio_update_group_bps(blkg, plid, temp ?: -1, fileid); |
e56da7e2 | 1102 | break; |
7702e8f4 | 1103 | case BLKIO_THROTL_write_bps_device: |
549d3aa8 | 1104 | pd->conf.bps[WRITE] = temp; |
c1768268 | 1105 | blkio_update_group_bps(blkg, plid, temp ?: -1, fileid); |
7702e8f4 VG |
1106 | break; |
1107 | case BLKIO_THROTL_read_iops_device: | |
e56da7e2 TH |
1108 | if (temp > THROTL_IOPS_MAX) |
1109 | goto out_unlock; | |
549d3aa8 | 1110 | pd->conf.iops[READ] = temp; |
c1768268 | 1111 | blkio_update_group_iops(blkg, plid, temp ?: -1, fileid); |
e56da7e2 | 1112 | break; |
7702e8f4 | 1113 | case BLKIO_THROTL_write_iops_device: |
d11bb446 | 1114 | if (temp > THROTL_IOPS_MAX) |
e56da7e2 | 1115 | goto out_unlock; |
549d3aa8 | 1116 | pd->conf.iops[WRITE] = temp; |
c1768268 | 1117 | blkio_update_group_iops(blkg, plid, temp ?: -1, fileid); |
7702e8f4 VG |
1118 | break; |
1119 | } | |
062a644d VG |
1120 | break; |
1121 | default: | |
1122 | BUG(); | |
1123 | } | |
ece84241 | 1124 | ret = 0; |
e56da7e2 TH |
1125 | out_unlock: |
1126 | rcu_read_unlock(); | |
ece84241 TH |
1127 | out: |
1128 | put_disk(disk); | |
e56da7e2 TH |
1129 | |
1130 | /* | |
1131 | * If queue was bypassing, we should retry. Do so after a short | |
1132 | * msleep(). It isn't strictly necessary but queue can be | |
1133 | * bypassing for some time and it's always nice to avoid busy | |
1134 | * looping. | |
1135 | */ | |
1136 | if (ret == -EBUSY) { | |
1137 | msleep(10); | |
1138 | return restart_syscall(); | |
1139 | } | |
ece84241 | 1140 | return ret; |
34d0f179 GJ |
1141 | } |
1142 | ||
062a644d VG |
1143 | static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft, |
1144 | const char *buffer) | |
34d0f179 GJ |
1145 | { |
1146 | int ret = 0; | |
1147 | char *buf; | |
e56da7e2 | 1148 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); |
062a644d VG |
1149 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); |
1150 | int fileid = BLKIOFILE_ATTR(cft->private); | |
34d0f179 GJ |
1151 | |
1152 | buf = kstrdup(buffer, GFP_KERNEL); | |
1153 | if (!buf) | |
1154 | return -ENOMEM; | |
1155 | ||
4bfd482e | 1156 | ret = blkio_policy_parse_and_set(buf, plid, fileid, blkcg); |
34d0f179 GJ |
1157 | kfree(buf); |
1158 | return ret; | |
1159 | } | |
1160 | ||
92616b5b VG |
1161 | static const char *blkg_dev_name(struct blkio_group *blkg) |
1162 | { | |
1163 | /* some drivers (floppy) instantiate a queue w/o disk registered */ | |
1164 | if (blkg->q->backing_dev_info.dev) | |
1165 | return dev_name(blkg->q->backing_dev_info.dev); | |
1166 | return NULL; | |
1167 | } | |
1168 | ||
4bfd482e TH |
1169 | static void blkio_print_group_conf(struct cftype *cft, struct blkio_group *blkg, |
1170 | struct seq_file *m) | |
34d0f179 | 1171 | { |
c1768268 | 1172 | int plid = BLKIOFILE_POLICY(cft->private); |
4bfd482e | 1173 | int fileid = BLKIOFILE_ATTR(cft->private); |
c1768268 TH |
1174 | struct blkg_policy_data *pd = blkg->pd[plid]; |
1175 | const char *dname = blkg_dev_name(blkg); | |
4bfd482e TH |
1176 | int rw = WRITE; |
1177 | ||
92616b5b VG |
1178 | if (!dname) |
1179 | return; | |
1180 | ||
c1768268 | 1181 | switch (plid) { |
062a644d | 1182 | case BLKIO_POLICY_PROP: |
549d3aa8 | 1183 | if (pd->conf.weight) |
7a4dd281 | 1184 | seq_printf(m, "%s\t%u\n", |
549d3aa8 | 1185 | dname, pd->conf.weight); |
4c9eefa1 VG |
1186 | break; |
1187 | case BLKIO_POLICY_THROTL: | |
4bfd482e | 1188 | switch (fileid) { |
7702e8f4 | 1189 | case BLKIO_THROTL_read_bps_device: |
4bfd482e | 1190 | rw = READ; |
7702e8f4 | 1191 | case BLKIO_THROTL_write_bps_device: |
549d3aa8 | 1192 | if (pd->conf.bps[rw]) |
7a4dd281 | 1193 | seq_printf(m, "%s\t%llu\n", |
549d3aa8 | 1194 | dname, pd->conf.bps[rw]); |
7702e8f4 VG |
1195 | break; |
1196 | case BLKIO_THROTL_read_iops_device: | |
4bfd482e | 1197 | rw = READ; |
7702e8f4 | 1198 | case BLKIO_THROTL_write_iops_device: |
549d3aa8 | 1199 | if (pd->conf.iops[rw]) |
7a4dd281 | 1200 | seq_printf(m, "%s\t%u\n", |
549d3aa8 | 1201 | dname, pd->conf.iops[rw]); |
7702e8f4 VG |
1202 | break; |
1203 | } | |
062a644d VG |
1204 | break; |
1205 | default: | |
1206 | BUG(); | |
1207 | } | |
1208 | } | |
34d0f179 | 1209 | |
062a644d | 1210 | /* cgroup files which read their data from policy nodes end up here */ |
4bfd482e TH |
1211 | static void blkio_read_conf(struct cftype *cft, struct blkio_cgroup *blkcg, |
1212 | struct seq_file *m) | |
34d0f179 | 1213 | { |
4bfd482e TH |
1214 | struct blkio_group *blkg; |
1215 | struct hlist_node *n; | |
34d0f179 | 1216 | |
4bfd482e TH |
1217 | spin_lock_irq(&blkcg->lock); |
1218 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) | |
e8989fae | 1219 | blkio_print_group_conf(cft, blkg, m); |
4bfd482e | 1220 | spin_unlock_irq(&blkcg->lock); |
062a644d VG |
1221 | } |
1222 | ||
1223 | static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft, | |
1224 | struct seq_file *m) | |
1225 | { | |
1226 | struct blkio_cgroup *blkcg; | |
1227 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1228 | int name = BLKIOFILE_ATTR(cft->private); | |
1229 | ||
1230 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1231 | ||
1232 | switch(plid) { | |
1233 | case BLKIO_POLICY_PROP: | |
1234 | switch(name) { | |
1235 | case BLKIO_PROP_weight_device: | |
4bfd482e | 1236 | blkio_read_conf(cft, blkcg, m); |
062a644d VG |
1237 | return 0; |
1238 | default: | |
1239 | BUG(); | |
1240 | } | |
1241 | break; | |
4c9eefa1 VG |
1242 | case BLKIO_POLICY_THROTL: |
1243 | switch(name){ | |
1244 | case BLKIO_THROTL_read_bps_device: | |
1245 | case BLKIO_THROTL_write_bps_device: | |
7702e8f4 VG |
1246 | case BLKIO_THROTL_read_iops_device: |
1247 | case BLKIO_THROTL_write_iops_device: | |
4bfd482e | 1248 | blkio_read_conf(cft, blkcg, m); |
4c9eefa1 VG |
1249 | return 0; |
1250 | default: | |
1251 | BUG(); | |
1252 | } | |
1253 | break; | |
062a644d VG |
1254 | default: |
1255 | BUG(); | |
1256 | } | |
1257 | ||
1258 | return 0; | |
1259 | } | |
1260 | ||
1261 | static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg, | |
5624a4e4 VG |
1262 | struct cftype *cft, struct cgroup_map_cb *cb, |
1263 | enum stat_type type, bool show_total, bool pcpu) | |
062a644d VG |
1264 | { |
1265 | struct blkio_group *blkg; | |
1266 | struct hlist_node *n; | |
1267 | uint64_t cgroup_total = 0; | |
1268 | ||
1269 | rcu_read_lock(); | |
1270 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
92616b5b | 1271 | const char *dname = blkg_dev_name(blkg); |
c1768268 | 1272 | int plid = BLKIOFILE_POLICY(cft->private); |
7a4dd281 | 1273 | |
e8989fae | 1274 | if (!dname) |
7a4dd281 | 1275 | continue; |
c1768268 TH |
1276 | if (pcpu) { |
1277 | cgroup_total += blkio_get_stat_cpu(blkg, plid, | |
1278 | cb, dname, type); | |
1279 | } else { | |
7a4dd281 | 1280 | spin_lock_irq(&blkg->stats_lock); |
c1768268 TH |
1281 | cgroup_total += blkio_get_stat(blkg, plid, |
1282 | cb, dname, type); | |
7a4dd281 | 1283 | spin_unlock_irq(&blkg->stats_lock); |
062a644d VG |
1284 | } |
1285 | } | |
1286 | if (show_total) | |
1287 | cb->fill(cb, "Total", cgroup_total); | |
1288 | rcu_read_unlock(); | |
1289 | return 0; | |
1290 | } | |
1291 | ||
1292 | /* All map kind of cgroup file get serviced by this function */ | |
1293 | static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft, | |
1294 | struct cgroup_map_cb *cb) | |
1295 | { | |
1296 | struct blkio_cgroup *blkcg; | |
1297 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1298 | int name = BLKIOFILE_ATTR(cft->private); | |
1299 | ||
1300 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1301 | ||
1302 | switch(plid) { | |
1303 | case BLKIO_POLICY_PROP: | |
1304 | switch(name) { | |
1305 | case BLKIO_PROP_time: | |
1306 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1307 | BLKIO_STAT_TIME, 0, 0); |
062a644d VG |
1308 | case BLKIO_PROP_sectors: |
1309 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1310 | BLKIO_STAT_CPU_SECTORS, 0, 1); |
062a644d VG |
1311 | case BLKIO_PROP_io_service_bytes: |
1312 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1313 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
062a644d VG |
1314 | case BLKIO_PROP_io_serviced: |
1315 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1316 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
062a644d VG |
1317 | case BLKIO_PROP_io_service_time: |
1318 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1319 | BLKIO_STAT_SERVICE_TIME, 1, 0); |
062a644d VG |
1320 | case BLKIO_PROP_io_wait_time: |
1321 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1322 | BLKIO_STAT_WAIT_TIME, 1, 0); |
062a644d VG |
1323 | case BLKIO_PROP_io_merged: |
1324 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
317389a7 | 1325 | BLKIO_STAT_CPU_MERGED, 1, 1); |
062a644d VG |
1326 | case BLKIO_PROP_io_queued: |
1327 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1328 | BLKIO_STAT_QUEUED, 1, 0); |
062a644d | 1329 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
9026e521 JT |
1330 | case BLKIO_PROP_unaccounted_time: |
1331 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1332 | BLKIO_STAT_UNACCOUNTED_TIME, 0, 0); |
062a644d VG |
1333 | case BLKIO_PROP_dequeue: |
1334 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1335 | BLKIO_STAT_DEQUEUE, 0, 0); |
062a644d VG |
1336 | case BLKIO_PROP_avg_queue_size: |
1337 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1338 | BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0); |
062a644d VG |
1339 | case BLKIO_PROP_group_wait_time: |
1340 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1341 | BLKIO_STAT_GROUP_WAIT_TIME, 0, 0); |
062a644d VG |
1342 | case BLKIO_PROP_idle_time: |
1343 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1344 | BLKIO_STAT_IDLE_TIME, 0, 0); |
062a644d VG |
1345 | case BLKIO_PROP_empty_time: |
1346 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1347 | BLKIO_STAT_EMPTY_TIME, 0, 0); |
062a644d VG |
1348 | #endif |
1349 | default: | |
1350 | BUG(); | |
1351 | } | |
1352 | break; | |
4c9eefa1 VG |
1353 | case BLKIO_POLICY_THROTL: |
1354 | switch(name){ | |
1355 | case BLKIO_THROTL_io_service_bytes: | |
1356 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1357 | BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1); |
4c9eefa1 VG |
1358 | case BLKIO_THROTL_io_serviced: |
1359 | return blkio_read_blkg_stats(blkcg, cft, cb, | |
5624a4e4 | 1360 | BLKIO_STAT_CPU_SERVICED, 1, 1); |
4c9eefa1 VG |
1361 | default: |
1362 | BUG(); | |
1363 | } | |
1364 | break; | |
062a644d VG |
1365 | default: |
1366 | BUG(); | |
1367 | } | |
1368 | ||
1369 | return 0; | |
1370 | } | |
1371 | ||
4bfd482e | 1372 | static int blkio_weight_write(struct blkio_cgroup *blkcg, int plid, u64 val) |
062a644d VG |
1373 | { |
1374 | struct blkio_group *blkg; | |
1375 | struct hlist_node *n; | |
062a644d VG |
1376 | |
1377 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) | |
1378 | return -EINVAL; | |
1379 | ||
1380 | spin_lock(&blkio_list_lock); | |
1381 | spin_lock_irq(&blkcg->lock); | |
1382 | blkcg->weight = (unsigned int)val; | |
1383 | ||
549d3aa8 | 1384 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
e8989fae | 1385 | struct blkg_policy_data *pd = blkg->pd[plid]; |
549d3aa8 | 1386 | |
e8989fae | 1387 | if (!pd->conf.weight) |
c1768268 | 1388 | blkio_update_group_weight(blkg, plid, blkcg->weight); |
549d3aa8 | 1389 | } |
062a644d | 1390 | |
062a644d VG |
1391 | spin_unlock_irq(&blkcg->lock); |
1392 | spin_unlock(&blkio_list_lock); | |
1393 | return 0; | |
1394 | } | |
1395 | ||
1396 | static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) { | |
1397 | struct blkio_cgroup *blkcg; | |
1398 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1399 | int name = BLKIOFILE_ATTR(cft->private); | |
1400 | ||
1401 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1402 | ||
1403 | switch(plid) { | |
1404 | case BLKIO_POLICY_PROP: | |
1405 | switch(name) { | |
1406 | case BLKIO_PROP_weight: | |
1407 | return (u64)blkcg->weight; | |
1408 | } | |
1409 | break; | |
1410 | default: | |
1411 | BUG(); | |
1412 | } | |
1413 | return 0; | |
1414 | } | |
1415 | ||
1416 | static int | |
1417 | blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) | |
1418 | { | |
1419 | struct blkio_cgroup *blkcg; | |
1420 | enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private); | |
1421 | int name = BLKIOFILE_ATTR(cft->private); | |
1422 | ||
1423 | blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1424 | ||
1425 | switch(plid) { | |
1426 | case BLKIO_POLICY_PROP: | |
1427 | switch(name) { | |
1428 | case BLKIO_PROP_weight: | |
4bfd482e | 1429 | return blkio_weight_write(blkcg, plid, val); |
062a644d VG |
1430 | } |
1431 | break; | |
1432 | default: | |
1433 | BUG(); | |
1434 | } | |
34d0f179 | 1435 | |
34d0f179 GJ |
1436 | return 0; |
1437 | } | |
1438 | ||
31e4c28d | 1439 | struct cftype blkio_files[] = { |
34d0f179 GJ |
1440 | { |
1441 | .name = "weight_device", | |
062a644d VG |
1442 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1443 | BLKIO_PROP_weight_device), | |
1444 | .read_seq_string = blkiocg_file_read, | |
1445 | .write_string = blkiocg_file_write, | |
34d0f179 GJ |
1446 | .max_write_len = 256, |
1447 | }, | |
31e4c28d VG |
1448 | { |
1449 | .name = "weight", | |
062a644d VG |
1450 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1451 | BLKIO_PROP_weight), | |
1452 | .read_u64 = blkiocg_file_read_u64, | |
1453 | .write_u64 = blkiocg_file_write_u64, | |
31e4c28d | 1454 | }, |
22084190 VG |
1455 | { |
1456 | .name = "time", | |
062a644d VG |
1457 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1458 | BLKIO_PROP_time), | |
1459 | .read_map = blkiocg_file_read_map, | |
22084190 VG |
1460 | }, |
1461 | { | |
1462 | .name = "sectors", | |
062a644d VG |
1463 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1464 | BLKIO_PROP_sectors), | |
1465 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1466 | }, |
1467 | { | |
1468 | .name = "io_service_bytes", | |
062a644d VG |
1469 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1470 | BLKIO_PROP_io_service_bytes), | |
1471 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1472 | }, |
1473 | { | |
1474 | .name = "io_serviced", | |
062a644d VG |
1475 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1476 | BLKIO_PROP_io_serviced), | |
1477 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1478 | }, |
1479 | { | |
1480 | .name = "io_service_time", | |
062a644d VG |
1481 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1482 | BLKIO_PROP_io_service_time), | |
1483 | .read_map = blkiocg_file_read_map, | |
303a3acb DS |
1484 | }, |
1485 | { | |
1486 | .name = "io_wait_time", | |
062a644d VG |
1487 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1488 | BLKIO_PROP_io_wait_time), | |
1489 | .read_map = blkiocg_file_read_map, | |
84c124da | 1490 | }, |
812d4026 DS |
1491 | { |
1492 | .name = "io_merged", | |
062a644d VG |
1493 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1494 | BLKIO_PROP_io_merged), | |
1495 | .read_map = blkiocg_file_read_map, | |
812d4026 | 1496 | }, |
cdc1184c DS |
1497 | { |
1498 | .name = "io_queued", | |
062a644d VG |
1499 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1500 | BLKIO_PROP_io_queued), | |
1501 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1502 | }, |
84c124da DS |
1503 | { |
1504 | .name = "reset_stats", | |
1505 | .write_u64 = blkiocg_reset_stats, | |
22084190 | 1506 | }, |
13f98250 VG |
1507 | #ifdef CONFIG_BLK_DEV_THROTTLING |
1508 | { | |
1509 | .name = "throttle.read_bps_device", | |
1510 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1511 | BLKIO_THROTL_read_bps_device), | |
1512 | .read_seq_string = blkiocg_file_read, | |
1513 | .write_string = blkiocg_file_write, | |
1514 | .max_write_len = 256, | |
1515 | }, | |
1516 | ||
1517 | { | |
1518 | .name = "throttle.write_bps_device", | |
1519 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1520 | BLKIO_THROTL_write_bps_device), | |
1521 | .read_seq_string = blkiocg_file_read, | |
1522 | .write_string = blkiocg_file_write, | |
1523 | .max_write_len = 256, | |
1524 | }, | |
1525 | ||
1526 | { | |
1527 | .name = "throttle.read_iops_device", | |
1528 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1529 | BLKIO_THROTL_read_iops_device), | |
1530 | .read_seq_string = blkiocg_file_read, | |
1531 | .write_string = blkiocg_file_write, | |
1532 | .max_write_len = 256, | |
1533 | }, | |
1534 | ||
1535 | { | |
1536 | .name = "throttle.write_iops_device", | |
1537 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1538 | BLKIO_THROTL_write_iops_device), | |
1539 | .read_seq_string = blkiocg_file_read, | |
1540 | .write_string = blkiocg_file_write, | |
1541 | .max_write_len = 256, | |
1542 | }, | |
1543 | { | |
1544 | .name = "throttle.io_service_bytes", | |
1545 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1546 | BLKIO_THROTL_io_service_bytes), | |
1547 | .read_map = blkiocg_file_read_map, | |
1548 | }, | |
1549 | { | |
1550 | .name = "throttle.io_serviced", | |
1551 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL, | |
1552 | BLKIO_THROTL_io_serviced), | |
1553 | .read_map = blkiocg_file_read_map, | |
1554 | }, | |
1555 | #endif /* CONFIG_BLK_DEV_THROTTLING */ | |
1556 | ||
22084190 | 1557 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
cdc1184c DS |
1558 | { |
1559 | .name = "avg_queue_size", | |
062a644d VG |
1560 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1561 | BLKIO_PROP_avg_queue_size), | |
1562 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1563 | }, |
812df48d DS |
1564 | { |
1565 | .name = "group_wait_time", | |
062a644d VG |
1566 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1567 | BLKIO_PROP_group_wait_time), | |
1568 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1569 | }, |
1570 | { | |
1571 | .name = "idle_time", | |
062a644d VG |
1572 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1573 | BLKIO_PROP_idle_time), | |
1574 | .read_map = blkiocg_file_read_map, | |
812df48d DS |
1575 | }, |
1576 | { | |
1577 | .name = "empty_time", | |
062a644d VG |
1578 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1579 | BLKIO_PROP_empty_time), | |
1580 | .read_map = blkiocg_file_read_map, | |
812df48d | 1581 | }, |
cdc1184c | 1582 | { |
22084190 | 1583 | .name = "dequeue", |
062a644d VG |
1584 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, |
1585 | BLKIO_PROP_dequeue), | |
1586 | .read_map = blkiocg_file_read_map, | |
cdc1184c | 1587 | }, |
9026e521 JT |
1588 | { |
1589 | .name = "unaccounted_time", | |
1590 | .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP, | |
1591 | BLKIO_PROP_unaccounted_time), | |
1592 | .read_map = blkiocg_file_read_map, | |
1593 | }, | |
22084190 | 1594 | #endif |
31e4c28d VG |
1595 | }; |
1596 | ||
1597 | static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1598 | { | |
1599 | return cgroup_add_files(cgroup, subsys, blkio_files, | |
1600 | ARRAY_SIZE(blkio_files)); | |
1601 | } | |
1602 | ||
7ee9c562 TH |
1603 | static int blkiocg_pre_destroy(struct cgroup_subsys *subsys, |
1604 | struct cgroup *cgroup) | |
31e4c28d VG |
1605 | { |
1606 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
b1c35769 VG |
1607 | unsigned long flags; |
1608 | struct blkio_group *blkg; | |
ca32aefc | 1609 | struct request_queue *q; |
b1c35769 VG |
1610 | |
1611 | rcu_read_lock(); | |
7ee9c562 | 1612 | |
0f3942a3 JA |
1613 | do { |
1614 | spin_lock_irqsave(&blkcg->lock, flags); | |
b1c35769 | 1615 | |
0f3942a3 JA |
1616 | if (hlist_empty(&blkcg->blkg_list)) { |
1617 | spin_unlock_irqrestore(&blkcg->lock, flags); | |
1618 | break; | |
1619 | } | |
b1c35769 | 1620 | |
0f3942a3 JA |
1621 | blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group, |
1622 | blkcg_node); | |
ca32aefc | 1623 | q = rcu_dereference(blkg->q); |
0f3942a3 | 1624 | __blkiocg_del_blkio_group(blkg); |
31e4c28d | 1625 | |
0f3942a3 | 1626 | spin_unlock_irqrestore(&blkcg->lock, flags); |
b1c35769 | 1627 | |
0f3942a3 JA |
1628 | /* |
1629 | * This blkio_group is being unlinked as associated cgroup is | |
1630 | * going away. Let all the IO controlling policies know about | |
61014e96 | 1631 | * this event. |
0f3942a3 JA |
1632 | */ |
1633 | spin_lock(&blkio_list_lock); | |
03aa264a | 1634 | spin_lock_irqsave(q->queue_lock, flags); |
e8989fae | 1635 | blkg_destroy(blkg); |
03aa264a | 1636 | spin_unlock_irqrestore(q->queue_lock, flags); |
0f3942a3 JA |
1637 | spin_unlock(&blkio_list_lock); |
1638 | } while (1); | |
34d0f179 | 1639 | |
b1c35769 | 1640 | rcu_read_unlock(); |
7ee9c562 TH |
1641 | |
1642 | return 0; | |
1643 | } | |
1644 | ||
1645 | static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1646 | { | |
1647 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
1648 | ||
67523c48 BB |
1649 | if (blkcg != &blkio_root_cgroup) |
1650 | kfree(blkcg); | |
31e4c28d VG |
1651 | } |
1652 | ||
1653 | static struct cgroup_subsys_state * | |
1654 | blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup) | |
1655 | { | |
0341509f LZ |
1656 | struct blkio_cgroup *blkcg; |
1657 | struct cgroup *parent = cgroup->parent; | |
31e4c28d | 1658 | |
0341509f | 1659 | if (!parent) { |
31e4c28d VG |
1660 | blkcg = &blkio_root_cgroup; |
1661 | goto done; | |
1662 | } | |
1663 | ||
31e4c28d VG |
1664 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
1665 | if (!blkcg) | |
1666 | return ERR_PTR(-ENOMEM); | |
1667 | ||
1668 | blkcg->weight = BLKIO_WEIGHT_DEFAULT; | |
1669 | done: | |
1670 | spin_lock_init(&blkcg->lock); | |
1671 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
1672 | ||
1673 | return &blkcg->css; | |
1674 | } | |
1675 | ||
5efd6113 TH |
1676 | /** |
1677 | * blkcg_init_queue - initialize blkcg part of request queue | |
1678 | * @q: request_queue to initialize | |
1679 | * | |
1680 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1681 | * part of new request_queue @q. | |
1682 | * | |
1683 | * RETURNS: | |
1684 | * 0 on success, -errno on failure. | |
1685 | */ | |
1686 | int blkcg_init_queue(struct request_queue *q) | |
1687 | { | |
923adde1 TH |
1688 | int ret; |
1689 | ||
5efd6113 TH |
1690 | might_sleep(); |
1691 | ||
923adde1 TH |
1692 | ret = blk_throtl_init(q); |
1693 | if (ret) | |
1694 | return ret; | |
1695 | ||
1696 | mutex_lock(&all_q_mutex); | |
1697 | INIT_LIST_HEAD(&q->all_q_node); | |
1698 | list_add_tail(&q->all_q_node, &all_q_list); | |
1699 | mutex_unlock(&all_q_mutex); | |
1700 | ||
1701 | return 0; | |
5efd6113 TH |
1702 | } |
1703 | ||
1704 | /** | |
1705 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1706 | * @q: request_queue to drain | |
1707 | * | |
1708 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1709 | */ | |
1710 | void blkcg_drain_queue(struct request_queue *q) | |
1711 | { | |
1712 | lockdep_assert_held(q->queue_lock); | |
1713 | ||
1714 | blk_throtl_drain(q); | |
1715 | } | |
1716 | ||
1717 | /** | |
1718 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1719 | * @q: request_queue being released | |
1720 | * | |
1721 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1722 | */ | |
1723 | void blkcg_exit_queue(struct request_queue *q) | |
1724 | { | |
923adde1 TH |
1725 | mutex_lock(&all_q_mutex); |
1726 | list_del_init(&q->all_q_node); | |
1727 | mutex_unlock(&all_q_mutex); | |
1728 | ||
e8989fae TH |
1729 | blkg_destroy_all(q, true); |
1730 | ||
5efd6113 TH |
1731 | blk_throtl_exit(q); |
1732 | } | |
1733 | ||
31e4c28d VG |
1734 | /* |
1735 | * We cannot support shared io contexts, as we have no mean to support | |
1736 | * two tasks with the same ioc in two different groups without major rework | |
1737 | * of the main cic data structures. For now we allow a task to change | |
1738 | * its cgroup only if it's the only owner of its ioc. | |
1739 | */ | |
bb9d97b6 TH |
1740 | static int blkiocg_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
1741 | struct cgroup_taskset *tset) | |
31e4c28d | 1742 | { |
bb9d97b6 | 1743 | struct task_struct *task; |
31e4c28d VG |
1744 | struct io_context *ioc; |
1745 | int ret = 0; | |
1746 | ||
1747 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
bb9d97b6 TH |
1748 | cgroup_taskset_for_each(task, cgrp, tset) { |
1749 | task_lock(task); | |
1750 | ioc = task->io_context; | |
1751 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1752 | ret = -EINVAL; | |
1753 | task_unlock(task); | |
1754 | if (ret) | |
1755 | break; | |
1756 | } | |
31e4c28d VG |
1757 | return ret; |
1758 | } | |
1759 | ||
bb9d97b6 TH |
1760 | static void blkiocg_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
1761 | struct cgroup_taskset *tset) | |
31e4c28d | 1762 | { |
bb9d97b6 | 1763 | struct task_struct *task; |
31e4c28d VG |
1764 | struct io_context *ioc; |
1765 | ||
bb9d97b6 | 1766 | cgroup_taskset_for_each(task, cgrp, tset) { |
b3c9dd18 LT |
1767 | /* we don't lose anything even if ioc allocation fails */ |
1768 | ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE); | |
1769 | if (ioc) { | |
1770 | ioc_cgroup_changed(ioc); | |
11a3122f | 1771 | put_io_context(ioc); |
b3c9dd18 | 1772 | } |
bb9d97b6 | 1773 | } |
31e4c28d VG |
1774 | } |
1775 | ||
923adde1 TH |
1776 | static void blkcg_bypass_start(void) |
1777 | __acquires(&all_q_mutex) | |
1778 | { | |
1779 | struct request_queue *q; | |
1780 | ||
1781 | mutex_lock(&all_q_mutex); | |
1782 | ||
1783 | list_for_each_entry(q, &all_q_list, all_q_node) { | |
1784 | blk_queue_bypass_start(q); | |
e8989fae | 1785 | blkg_destroy_all(q, false); |
923adde1 TH |
1786 | } |
1787 | } | |
1788 | ||
1789 | static void blkcg_bypass_end(void) | |
1790 | __releases(&all_q_mutex) | |
1791 | { | |
1792 | struct request_queue *q; | |
1793 | ||
1794 | list_for_each_entry(q, &all_q_list, all_q_node) | |
1795 | blk_queue_bypass_end(q); | |
1796 | ||
1797 | mutex_unlock(&all_q_mutex); | |
1798 | } | |
1799 | ||
3e252066 VG |
1800 | void blkio_policy_register(struct blkio_policy_type *blkiop) |
1801 | { | |
e8989fae TH |
1802 | struct request_queue *q; |
1803 | ||
923adde1 | 1804 | blkcg_bypass_start(); |
3e252066 | 1805 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1806 | |
1807 | BUG_ON(blkio_policy[blkiop->plid]); | |
1808 | blkio_policy[blkiop->plid] = blkiop; | |
3e252066 | 1809 | list_add_tail(&blkiop->list, &blkio_list); |
035d10b2 | 1810 | |
3e252066 | 1811 | spin_unlock(&blkio_list_lock); |
e8989fae TH |
1812 | list_for_each_entry(q, &all_q_list, all_q_node) |
1813 | update_root_blkg_pd(q, blkiop->plid); | |
923adde1 | 1814 | blkcg_bypass_end(); |
3e252066 VG |
1815 | } |
1816 | EXPORT_SYMBOL_GPL(blkio_policy_register); | |
1817 | ||
1818 | void blkio_policy_unregister(struct blkio_policy_type *blkiop) | |
1819 | { | |
e8989fae TH |
1820 | struct request_queue *q; |
1821 | ||
923adde1 | 1822 | blkcg_bypass_start(); |
3e252066 | 1823 | spin_lock(&blkio_list_lock); |
035d10b2 TH |
1824 | |
1825 | BUG_ON(blkio_policy[blkiop->plid] != blkiop); | |
1826 | blkio_policy[blkiop->plid] = NULL; | |
3e252066 | 1827 | list_del_init(&blkiop->list); |
035d10b2 | 1828 | |
3e252066 | 1829 | spin_unlock(&blkio_list_lock); |
e8989fae TH |
1830 | list_for_each_entry(q, &all_q_list, all_q_node) |
1831 | update_root_blkg_pd(q, blkiop->plid); | |
923adde1 | 1832 | blkcg_bypass_end(); |
3e252066 VG |
1833 | } |
1834 | EXPORT_SYMBOL_GPL(blkio_policy_unregister); |