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