blob: 75642a352a8f40595069f0833a8e7cef131eca02 [file] [log] [blame]
Vivek Goyal31e4c282009-12-03 12:59:42 -05001/*
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>
Vivek Goyal22084192009-12-03 12:59:49 -050014#include <linux/seq_file.h>
15#include <linux/kdev_t.h>
Vivek Goyal9d6a9862009-12-04 10:36:41 -050016#include <linux/module.h>
Stephen Rothwellaccee782009-12-07 19:29:39 +110017#include <linux/err.h>
Divyesh Shah91952912010-04-01 15:01:41 -070018#include <linux/blkdev.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/slab.h>
Vivek Goyal31e4c282009-12-03 12:59:42 -050020#include "blk-cgroup.h"
Gui Jianfeng34d0f172010-04-13 16:05:49 +080021#include <linux/genhd.h>
Vivek Goyal3e252062009-12-04 10:36:42 -050022
Divyesh Shah84c124d2010-04-09 08:31:19 +020023#define MAX_KEY_LEN 100
24
Vivek Goyal3e252062009-12-04 10:36:42 -050025static DEFINE_SPINLOCK(blkio_list_lock);
26static LIST_HEAD(blkio_list);
Vivek Goyalb1c35762009-12-03 12:59:47 -050027
Vivek Goyal31e4c282009-12-03 12:59:42 -050028struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT };
Vivek Goyal9d6a9862009-12-04 10:36:41 -050029EXPORT_SYMBOL_GPL(blkio_root_cgroup);
30
Ben Blum67523c42010-03-10 15:22:11 -080031static struct cgroup_subsys_state *blkiocg_create(struct cgroup_subsys *,
32 struct cgroup *);
Tejun Heobb9d97b2011-12-12 18:12:21 -080033static int blkiocg_can_attach(struct cgroup_subsys *, struct cgroup *,
34 struct cgroup_taskset *);
35static void blkiocg_attach(struct cgroup_subsys *, struct cgroup *,
36 struct cgroup_taskset *);
Ben Blum67523c42010-03-10 15:22:11 -080037static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *);
38static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *);
39
Vivek Goyal062a6442010-09-15 17:06:33 -040040/* for encoding cft->private value on file */
41#define BLKIOFILE_PRIVATE(x, val) (((x) << 16) | (val))
42/* What policy owns the file, proportional or throttle */
43#define BLKIOFILE_POLICY(val) (((val) >> 16) & 0xffff)
44#define BLKIOFILE_ATTR(val) ((val) & 0xffff)
45
Ben Blum67523c42010-03-10 15:22:11 -080046struct cgroup_subsys blkio_subsys = {
47 .name = "blkio",
48 .create = blkiocg_create,
Tejun Heobb9d97b2011-12-12 18:12:21 -080049 .can_attach = blkiocg_can_attach,
50 .attach = blkiocg_attach,
Ben Blum67523c42010-03-10 15:22:11 -080051 .destroy = blkiocg_destroy,
52 .populate = blkiocg_populate,
53#ifdef CONFIG_BLK_CGROUP
54 /* note: blkio_subsys_id is otherwise defined in blk-cgroup.h */
55 .subsys_id = blkio_subsys_id,
56#endif
57 .use_id = 1,
58 .module = THIS_MODULE,
59};
60EXPORT_SYMBOL_GPL(blkio_subsys);
61
Gui Jianfeng34d0f172010-04-13 16:05:49 +080062static inline void blkio_policy_insert_node(struct blkio_cgroup *blkcg,
63 struct blkio_policy_node *pn)
64{
65 list_add(&pn->node, &blkcg->policy_list);
66}
67
Vivek Goyal062a6442010-09-15 17:06:33 -040068static inline bool cftype_blkg_same_policy(struct cftype *cft,
69 struct blkio_group *blkg)
70{
71 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
72
73 if (blkg->plid == plid)
74 return 1;
75
76 return 0;
77}
78
79/* Determines if policy node matches cgroup file being accessed */
80static inline bool pn_matches_cftype(struct cftype *cft,
81 struct blkio_policy_node *pn)
82{
83 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
84 int fileid = BLKIOFILE_ATTR(cft->private);
85
86 return (plid == pn->plid && fileid == pn->fileid);
87}
88
Gui Jianfeng34d0f172010-04-13 16:05:49 +080089/* Must be called with blkcg->lock held */
90static inline void blkio_policy_delete_node(struct blkio_policy_node *pn)
91{
92 list_del(&pn->node);
93}
94
95/* Must be called with blkcg->lock held */
96static struct blkio_policy_node *
Vivek Goyal062a6442010-09-15 17:06:33 -040097blkio_policy_search_node(const struct blkio_cgroup *blkcg, dev_t dev,
98 enum blkio_policy_id plid, int fileid)
Gui Jianfeng34d0f172010-04-13 16:05:49 +080099{
100 struct blkio_policy_node *pn;
101
102 list_for_each_entry(pn, &blkcg->policy_list, node) {
Vivek Goyal062a6442010-09-15 17:06:33 -0400103 if (pn->dev == dev && pn->plid == plid && pn->fileid == fileid)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800104 return pn;
105 }
106
107 return NULL;
108}
109
Vivek Goyal31e4c282009-12-03 12:59:42 -0500110struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
111{
112 return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
113 struct blkio_cgroup, css);
114}
Vivek Goyal9d6a9862009-12-04 10:36:41 -0500115EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500116
Vivek Goyal70087dc2011-05-16 15:24:08 +0200117struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk)
118{
119 return container_of(task_subsys_state(tsk, blkio_subsys_id),
120 struct blkio_cgroup, css);
121}
122EXPORT_SYMBOL_GPL(task_blkio_cgroup);
123
Vivek Goyal062a6442010-09-15 17:06:33 -0400124static inline void
125blkio_update_group_weight(struct blkio_group *blkg, unsigned int weight)
126{
127 struct blkio_policy_type *blkiop;
128
129 list_for_each_entry(blkiop, &blkio_list, list) {
130 /* If this policy does not own the blkg, do not send updates */
131 if (blkiop->plid != blkg->plid)
132 continue;
133 if (blkiop->ops.blkio_update_group_weight_fn)
Vivek Goyalfe071432010-10-01 14:49:49 +0200134 blkiop->ops.blkio_update_group_weight_fn(blkg->key,
135 blkg, weight);
Vivek Goyal062a6442010-09-15 17:06:33 -0400136 }
137}
138
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400139static inline void blkio_update_group_bps(struct blkio_group *blkg, u64 bps,
140 int fileid)
141{
142 struct blkio_policy_type *blkiop;
143
144 list_for_each_entry(blkiop, &blkio_list, list) {
145
146 /* If this policy does not own the blkg, do not send updates */
147 if (blkiop->plid != blkg->plid)
148 continue;
149
150 if (fileid == BLKIO_THROTL_read_bps_device
151 && blkiop->ops.blkio_update_group_read_bps_fn)
Vivek Goyalfe071432010-10-01 14:49:49 +0200152 blkiop->ops.blkio_update_group_read_bps_fn(blkg->key,
153 blkg, bps);
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400154
155 if (fileid == BLKIO_THROTL_write_bps_device
156 && blkiop->ops.blkio_update_group_write_bps_fn)
Vivek Goyalfe071432010-10-01 14:49:49 +0200157 blkiop->ops.blkio_update_group_write_bps_fn(blkg->key,
158 blkg, bps);
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400159 }
160}
161
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400162static inline void blkio_update_group_iops(struct blkio_group *blkg,
163 unsigned int iops, int fileid)
164{
165 struct blkio_policy_type *blkiop;
166
167 list_for_each_entry(blkiop, &blkio_list, list) {
168
169 /* If this policy does not own the blkg, do not send updates */
170 if (blkiop->plid != blkg->plid)
171 continue;
172
173 if (fileid == BLKIO_THROTL_read_iops_device
174 && blkiop->ops.blkio_update_group_read_iops_fn)
Vivek Goyalfe071432010-10-01 14:49:49 +0200175 blkiop->ops.blkio_update_group_read_iops_fn(blkg->key,
176 blkg, iops);
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400177
178 if (fileid == BLKIO_THROTL_write_iops_device
179 && blkiop->ops.blkio_update_group_write_iops_fn)
Vivek Goyalfe071432010-10-01 14:49:49 +0200180 blkiop->ops.blkio_update_group_write_iops_fn(blkg->key,
181 blkg,iops);
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400182 }
183}
184
Divyesh Shah91952912010-04-01 15:01:41 -0700185/*
186 * Add to the appropriate stat variable depending on the request type.
187 * This should be called with the blkg->stats_lock held.
188 */
Divyesh Shah84c124d2010-04-09 08:31:19 +0200189static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction,
190 bool sync)
Divyesh Shah91952912010-04-01 15:01:41 -0700191{
Divyesh Shah84c124d2010-04-09 08:31:19 +0200192 if (direction)
193 stat[BLKIO_STAT_WRITE] += add;
Divyesh Shah91952912010-04-01 15:01:41 -0700194 else
Divyesh Shah84c124d2010-04-09 08:31:19 +0200195 stat[BLKIO_STAT_READ] += add;
196 if (sync)
197 stat[BLKIO_STAT_SYNC] += add;
Divyesh Shah91952912010-04-01 15:01:41 -0700198 else
Divyesh Shah84c124d2010-04-09 08:31:19 +0200199 stat[BLKIO_STAT_ASYNC] += add;
Divyesh Shah91952912010-04-01 15:01:41 -0700200}
201
Divyesh Shahcdc11842010-04-08 21:15:10 -0700202/*
203 * Decrements the appropriate stat variable if non-zero depending on the
204 * request type. Panics on value being zero.
205 * This should be called with the blkg->stats_lock held.
206 */
207static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync)
208{
209 if (direction) {
210 BUG_ON(stat[BLKIO_STAT_WRITE] == 0);
211 stat[BLKIO_STAT_WRITE]--;
212 } else {
213 BUG_ON(stat[BLKIO_STAT_READ] == 0);
214 stat[BLKIO_STAT_READ]--;
215 }
216 if (sync) {
217 BUG_ON(stat[BLKIO_STAT_SYNC] == 0);
218 stat[BLKIO_STAT_SYNC]--;
219 } else {
220 BUG_ON(stat[BLKIO_STAT_ASYNC] == 0);
221 stat[BLKIO_STAT_ASYNC]--;
222 }
223}
224
225#ifdef CONFIG_DEBUG_BLK_CGROUP
Divyesh Shah812df482010-04-08 21:15:35 -0700226/* This should be called with the blkg->stats_lock held. */
227static void blkio_set_start_group_wait_time(struct blkio_group *blkg,
228 struct blkio_group *curr_blkg)
229{
230 if (blkio_blkg_waiting(&blkg->stats))
231 return;
232 if (blkg == curr_blkg)
233 return;
234 blkg->stats.start_group_wait_time = sched_clock();
235 blkio_mark_blkg_waiting(&blkg->stats);
236}
237
238/* This should be called with the blkg->stats_lock held. */
239static void blkio_update_group_wait_time(struct blkio_group_stats *stats)
240{
241 unsigned long long now;
242
243 if (!blkio_blkg_waiting(stats))
244 return;
245
246 now = sched_clock();
247 if (time_after64(now, stats->start_group_wait_time))
248 stats->group_wait_time += now - stats->start_group_wait_time;
249 blkio_clear_blkg_waiting(stats);
250}
251
252/* This should be called with the blkg->stats_lock held. */
253static void blkio_end_empty_time(struct blkio_group_stats *stats)
254{
255 unsigned long long now;
256
257 if (!blkio_blkg_empty(stats))
258 return;
259
260 now = sched_clock();
261 if (time_after64(now, stats->start_empty_time))
262 stats->empty_time += now - stats->start_empty_time;
263 blkio_clear_blkg_empty(stats);
264}
265
266void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
267{
268 unsigned long flags;
269
270 spin_lock_irqsave(&blkg->stats_lock, flags);
271 BUG_ON(blkio_blkg_idling(&blkg->stats));
272 blkg->stats.start_idle_time = sched_clock();
273 blkio_mark_blkg_idling(&blkg->stats);
274 spin_unlock_irqrestore(&blkg->stats_lock, flags);
275}
276EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats);
277
278void blkiocg_update_idle_time_stats(struct blkio_group *blkg)
279{
280 unsigned long flags;
281 unsigned long long now;
282 struct blkio_group_stats *stats;
283
284 spin_lock_irqsave(&blkg->stats_lock, flags);
285 stats = &blkg->stats;
286 if (blkio_blkg_idling(stats)) {
287 now = sched_clock();
288 if (time_after64(now, stats->start_idle_time))
289 stats->idle_time += now - stats->start_idle_time;
290 blkio_clear_blkg_idling(stats);
291 }
292 spin_unlock_irqrestore(&blkg->stats_lock, flags);
293}
294EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats);
295
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200296void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg)
Divyesh Shahcdc11842010-04-08 21:15:10 -0700297{
298 unsigned long flags;
299 struct blkio_group_stats *stats;
300
301 spin_lock_irqsave(&blkg->stats_lock, flags);
302 stats = &blkg->stats;
303 stats->avg_queue_size_sum +=
304 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] +
305 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE];
306 stats->avg_queue_size_samples++;
Divyesh Shah812df482010-04-08 21:15:35 -0700307 blkio_update_group_wait_time(stats);
Divyesh Shahcdc11842010-04-08 21:15:10 -0700308 spin_unlock_irqrestore(&blkg->stats_lock, flags);
309}
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200310EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats);
311
Vivek Goyale5ff0822010-04-26 19:25:11 +0200312void blkiocg_set_start_empty_time(struct blkio_group *blkg)
Divyesh Shah28baf442010-04-14 11:22:38 +0200313{
314 unsigned long flags;
315 struct blkio_group_stats *stats;
316
317 spin_lock_irqsave(&blkg->stats_lock, flags);
318 stats = &blkg->stats;
319
320 if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] ||
321 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) {
322 spin_unlock_irqrestore(&blkg->stats_lock, flags);
323 return;
324 }
325
326 /*
Vivek Goyale5ff0822010-04-26 19:25:11 +0200327 * group is already marked empty. This can happen if cfqq got new
328 * request in parent group and moved to this group while being added
329 * to service tree. Just ignore the event and move on.
Divyesh Shah28baf442010-04-14 11:22:38 +0200330 */
Vivek Goyale5ff0822010-04-26 19:25:11 +0200331 if(blkio_blkg_empty(stats)) {
332 spin_unlock_irqrestore(&blkg->stats_lock, flags);
333 return;
334 }
335
Divyesh Shah28baf442010-04-14 11:22:38 +0200336 stats->start_empty_time = sched_clock();
337 blkio_mark_blkg_empty(stats);
338 spin_unlock_irqrestore(&blkg->stats_lock, flags);
339}
340EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time);
341
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200342void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
343 unsigned long dequeue)
344{
345 blkg->stats.dequeue += dequeue;
346}
347EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats);
Divyesh Shah812df482010-04-08 21:15:35 -0700348#else
349static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg,
350 struct blkio_group *curr_blkg) {}
351static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {}
Divyesh Shahcdc11842010-04-08 21:15:10 -0700352#endif
353
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200354void blkiocg_update_io_add_stats(struct blkio_group *blkg,
Divyesh Shahcdc11842010-04-08 21:15:10 -0700355 struct blkio_group *curr_blkg, bool direction,
356 bool sync)
357{
358 unsigned long flags;
359
360 spin_lock_irqsave(&blkg->stats_lock, flags);
361 blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction,
362 sync);
Divyesh Shah812df482010-04-08 21:15:35 -0700363 blkio_end_empty_time(&blkg->stats);
364 blkio_set_start_group_wait_time(blkg, curr_blkg);
Divyesh Shahcdc11842010-04-08 21:15:10 -0700365 spin_unlock_irqrestore(&blkg->stats_lock, flags);
366}
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200367EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats);
Divyesh Shahcdc11842010-04-08 21:15:10 -0700368
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200369void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
Divyesh Shahcdc11842010-04-08 21:15:10 -0700370 bool direction, bool sync)
371{
372 unsigned long flags;
373
374 spin_lock_irqsave(&blkg->stats_lock, flags);
375 blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED],
376 direction, sync);
377 spin_unlock_irqrestore(&blkg->stats_lock, flags);
378}
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200379EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats);
Divyesh Shahcdc11842010-04-08 21:15:10 -0700380
Justin TerAvest167400d2011-03-12 16:54:00 +0100381void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time,
382 unsigned long unaccounted_time)
Vivek Goyal22084192009-12-03 12:59:49 -0500383{
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700384 unsigned long flags;
385
386 spin_lock_irqsave(&blkg->stats_lock, flags);
387 blkg->stats.time += time;
Vivek Goyala23e6862011-05-19 15:38:20 -0400388#ifdef CONFIG_DEBUG_BLK_CGROUP
Justin TerAvest167400d2011-03-12 16:54:00 +0100389 blkg->stats.unaccounted_time += unaccounted_time;
Vivek Goyala23e6862011-05-19 15:38:20 -0400390#endif
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700391 spin_unlock_irqrestore(&blkg->stats_lock, flags);
Vivek Goyal22084192009-12-03 12:59:49 -0500392}
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700393EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used);
Vivek Goyal22084192009-12-03 12:59:49 -0500394
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400395/*
396 * should be called under rcu read lock or queue lock to make sure blkg pointer
397 * is valid.
398 */
Divyesh Shah84c124d2010-04-09 08:31:19 +0200399void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
400 uint64_t bytes, bool direction, bool sync)
Divyesh Shah91952912010-04-01 15:01:41 -0700401{
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400402 struct blkio_group_stats_cpu *stats_cpu;
Vivek Goyal575969a2011-05-19 15:38:29 -0400403 unsigned long flags;
404
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);
Divyesh Shah91952912010-04-01 15:01:41 -0700411
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400412 stats_cpu = this_cpu_ptr(blkg->stats_cpu);
413
Vivek Goyal575969a2011-05-19 15:38:29 -0400414 u64_stats_update_begin(&stats_cpu->syncp);
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400415 stats_cpu->sectors += bytes >> 9;
416 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED],
417 1, direction, sync);
418 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES],
419 bytes, direction, sync);
Vivek Goyal575969a2011-05-19 15:38:29 -0400420 u64_stats_update_end(&stats_cpu->syncp);
421 local_irq_restore(flags);
Divyesh Shah91952912010-04-01 15:01:41 -0700422}
Divyesh Shah84c124d2010-04-09 08:31:19 +0200423EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
Divyesh Shah91952912010-04-01 15:01:41 -0700424
Divyesh Shah84c124d2010-04-09 08:31:19 +0200425void blkiocg_update_completion_stats(struct blkio_group *blkg,
426 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
Divyesh Shah91952912010-04-01 15:01:41 -0700427{
428 struct blkio_group_stats *stats;
429 unsigned long flags;
430 unsigned long long now = sched_clock();
431
432 spin_lock_irqsave(&blkg->stats_lock, flags);
433 stats = &blkg->stats;
Divyesh Shah84c124d2010-04-09 08:31:19 +0200434 if (time_after64(now, io_start_time))
435 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
436 now - io_start_time, direction, sync);
437 if (time_after64(io_start_time, start_time))
438 blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
439 io_start_time - start_time, direction, sync);
Divyesh Shah91952912010-04-01 15:01:41 -0700440 spin_unlock_irqrestore(&blkg->stats_lock, flags);
441}
Divyesh Shah84c124d2010-04-09 08:31:19 +0200442EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
Divyesh Shah91952912010-04-01 15:01:41 -0700443
Vivek Goyal317389a2011-05-23 10:02:19 +0200444/* Merged stats are per cpu. */
Divyesh Shah812d4022010-04-08 21:14:23 -0700445void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
446 bool sync)
447{
Vivek Goyal317389a2011-05-23 10:02:19 +0200448 struct blkio_group_stats_cpu *stats_cpu;
Divyesh Shah812d4022010-04-08 21:14:23 -0700449 unsigned long flags;
450
Vivek Goyal317389a2011-05-23 10:02:19 +0200451 /*
452 * Disabling interrupts to provide mutual exclusion between two
453 * writes on same cpu. It probably is not needed for 64bit. Not
454 * optimizing that case yet.
455 */
456 local_irq_save(flags);
457
458 stats_cpu = this_cpu_ptr(blkg->stats_cpu);
459
460 u64_stats_update_begin(&stats_cpu->syncp);
461 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1,
462 direction, sync);
463 u64_stats_update_end(&stats_cpu->syncp);
464 local_irq_restore(flags);
Divyesh Shah812d4022010-04-08 21:14:23 -0700465}
466EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
467
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400468/*
469 * This function allocates the per cpu stats for blkio_group. Should be called
470 * from sleepable context as alloc_per_cpu() requires that.
471 */
472int blkio_alloc_blkg_stats(struct blkio_group *blkg)
473{
474 /* Allocate memory for per cpu stats */
475 blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu);
476 if (!blkg->stats_cpu)
477 return -ENOMEM;
478 return 0;
479}
480EXPORT_SYMBOL_GPL(blkio_alloc_blkg_stats);
481
Vivek Goyal31e4c282009-12-03 12:59:42 -0500482void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
Vivek Goyal062a6442010-09-15 17:06:33 -0400483 struct blkio_group *blkg, void *key, dev_t dev,
484 enum blkio_policy_id plid)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500485{
486 unsigned long flags;
487
488 spin_lock_irqsave(&blkcg->lock, flags);
Divyesh Shah8d2a91f2010-04-16 08:10:51 +0200489 spin_lock_init(&blkg->stats_lock);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500490 rcu_assign_pointer(blkg->key, key);
Vivek Goyalb1c35762009-12-03 12:59:47 -0500491 blkg->blkcg_id = css_id(&blkcg->css);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500492 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
Vivek Goyal062a6442010-09-15 17:06:33 -0400493 blkg->plid = plid;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500494 spin_unlock_irqrestore(&blkcg->lock, flags);
Vivek Goyal2868ef72009-12-03 12:59:48 -0500495 /* Need to take css reference ? */
496 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
Vivek Goyal22084192009-12-03 12:59:49 -0500497 blkg->dev = dev;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500498}
Vivek Goyal9d6a9862009-12-04 10:36:41 -0500499EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500500
Vivek Goyalb1c35762009-12-03 12:59:47 -0500501static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
502{
503 hlist_del_init_rcu(&blkg->blkcg_node);
504 blkg->blkcg_id = 0;
505}
506
507/*
508 * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
509 * indicating that blk_group was unhashed by the time we got to it.
510 */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500511int blkiocg_del_blkio_group(struct blkio_group *blkg)
512{
Vivek Goyalb1c35762009-12-03 12:59:47 -0500513 struct blkio_cgroup *blkcg;
514 unsigned long flags;
515 struct cgroup_subsys_state *css;
516 int ret = 1;
517
518 rcu_read_lock();
519 css = css_lookup(&blkio_subsys, blkg->blkcg_id);
Jens Axboe0f3942a2010-05-03 14:28:55 +0200520 if (css) {
521 blkcg = container_of(css, struct blkio_cgroup, css);
522 spin_lock_irqsave(&blkcg->lock, flags);
523 if (!hlist_unhashed(&blkg->blkcg_node)) {
524 __blkiocg_del_blkio_group(blkg);
525 ret = 0;
526 }
527 spin_unlock_irqrestore(&blkcg->lock, flags);
Vivek Goyalb1c35762009-12-03 12:59:47 -0500528 }
Jens Axboe0f3942a2010-05-03 14:28:55 +0200529
Vivek Goyalb1c35762009-12-03 12:59:47 -0500530 rcu_read_unlock();
531 return ret;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500532}
Vivek Goyal9d6a9862009-12-04 10:36:41 -0500533EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500534
535/* called under rcu_read_lock(). */
536struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
537{
538 struct blkio_group *blkg;
539 struct hlist_node *n;
540 void *__key;
541
542 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
543 __key = blkg->key;
544 if (__key == key)
545 return blkg;
546 }
547
548 return NULL;
549}
Vivek Goyal9d6a9862009-12-04 10:36:41 -0500550EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500551
Vivek Goyalf0bdc8c2011-05-19 15:38:30 -0400552static void blkio_reset_stats_cpu(struct blkio_group *blkg)
553{
554 struct blkio_group_stats_cpu *stats_cpu;
555 int i, j, k;
556 /*
557 * Note: On 64 bit arch this should not be an issue. This has the
558 * possibility of returning some inconsistent value on 32bit arch
559 * as 64bit update on 32bit is non atomic. Taking care of this
560 * corner case makes code very complicated, like sending IPIs to
561 * cpus, taking care of stats of offline cpus etc.
562 *
563 * reset stats is anyway more of a debug feature and this sounds a
564 * corner case. So I am not complicating the code yet until and
565 * unless this becomes a real issue.
566 */
567 for_each_possible_cpu(i) {
568 stats_cpu = per_cpu_ptr(blkg->stats_cpu, i);
569 stats_cpu->sectors = 0;
570 for(j = 0; j < BLKIO_STAT_CPU_NR; j++)
571 for (k = 0; k < BLKIO_STAT_TOTAL; k++)
572 stats_cpu->stat_arr_cpu[j][k] = 0;
573 }
574}
575
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700576static int
Divyesh Shah84c124d2010-04-09 08:31:19 +0200577blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700578{
579 struct blkio_cgroup *blkcg;
580 struct blkio_group *blkg;
Divyesh Shah812df482010-04-08 21:15:35 -0700581 struct blkio_group_stats *stats;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700582 struct hlist_node *n;
Divyesh Shahcdc11842010-04-08 21:15:10 -0700583 uint64_t queued[BLKIO_STAT_TOTAL];
584 int i;
Divyesh Shah812df482010-04-08 21:15:35 -0700585#ifdef CONFIG_DEBUG_BLK_CGROUP
586 bool idling, waiting, empty;
587 unsigned long long now = sched_clock();
588#endif
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700589
590 blkcg = cgroup_to_blkio_cgroup(cgroup);
591 spin_lock_irq(&blkcg->lock);
592 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
593 spin_lock(&blkg->stats_lock);
Divyesh Shah812df482010-04-08 21:15:35 -0700594 stats = &blkg->stats;
595#ifdef CONFIG_DEBUG_BLK_CGROUP
596 idling = blkio_blkg_idling(stats);
597 waiting = blkio_blkg_waiting(stats);
598 empty = blkio_blkg_empty(stats);
599#endif
Divyesh Shahcdc11842010-04-08 21:15:10 -0700600 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
Divyesh Shah812df482010-04-08 21:15:35 -0700601 queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i];
602 memset(stats, 0, sizeof(struct blkio_group_stats));
Divyesh Shahcdc11842010-04-08 21:15:10 -0700603 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
Divyesh Shah812df482010-04-08 21:15:35 -0700604 stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i];
605#ifdef CONFIG_DEBUG_BLK_CGROUP
606 if (idling) {
607 blkio_mark_blkg_idling(stats);
608 stats->start_idle_time = now;
609 }
610 if (waiting) {
611 blkio_mark_blkg_waiting(stats);
612 stats->start_group_wait_time = now;
613 }
614 if (empty) {
615 blkio_mark_blkg_empty(stats);
616 stats->start_empty_time = now;
617 }
618#endif
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700619 spin_unlock(&blkg->stats_lock);
Vivek Goyalf0bdc8c2011-05-19 15:38:30 -0400620
621 /* Reset Per cpu stats which don't take blkg->stats_lock */
622 blkio_reset_stats_cpu(blkg);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700623 }
Vivek Goyalf0bdc8c2011-05-19 15:38:30 -0400624
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700625 spin_unlock_irq(&blkcg->lock);
626 return 0;
627}
628
Divyesh Shah84c124d2010-04-09 08:31:19 +0200629static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
630 int chars_left, bool diskname_only)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700631{
Divyesh Shah84c124d2010-04-09 08:31:19 +0200632 snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700633 chars_left -= strlen(str);
634 if (chars_left <= 0) {
635 printk(KERN_WARNING
636 "Possibly incorrect cgroup stat display format");
637 return;
638 }
Divyesh Shah84c124d2010-04-09 08:31:19 +0200639 if (diskname_only)
640 return;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700641 switch (type) {
Divyesh Shah84c124d2010-04-09 08:31:19 +0200642 case BLKIO_STAT_READ:
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700643 strlcat(str, " Read", chars_left);
644 break;
Divyesh Shah84c124d2010-04-09 08:31:19 +0200645 case BLKIO_STAT_WRITE:
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700646 strlcat(str, " Write", chars_left);
647 break;
Divyesh Shah84c124d2010-04-09 08:31:19 +0200648 case BLKIO_STAT_SYNC:
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700649 strlcat(str, " Sync", chars_left);
650 break;
Divyesh Shah84c124d2010-04-09 08:31:19 +0200651 case BLKIO_STAT_ASYNC:
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700652 strlcat(str, " Async", chars_left);
653 break;
Divyesh Shah84c124d2010-04-09 08:31:19 +0200654 case BLKIO_STAT_TOTAL:
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700655 strlcat(str, " Total", chars_left);
656 break;
657 default:
658 strlcat(str, " Invalid", chars_left);
659 }
660}
661
Divyesh Shah84c124d2010-04-09 08:31:19 +0200662static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
663 struct cgroup_map_cb *cb, dev_t dev)
664{
665 blkio_get_key_name(0, dev, str, chars_left, true);
666 cb->fill(cb, str, val);
667 return val;
668}
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700669
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400670
671static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg,
672 enum stat_type_cpu type, enum stat_sub_type sub_type)
673{
674 int cpu;
675 struct blkio_group_stats_cpu *stats_cpu;
Vivek Goyal575969a2011-05-19 15:38:29 -0400676 u64 val = 0, tval;
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400677
678 for_each_possible_cpu(cpu) {
Vivek Goyal575969a2011-05-19 15:38:29 -0400679 unsigned int start;
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400680 stats_cpu = per_cpu_ptr(blkg->stats_cpu, cpu);
681
Vivek Goyal575969a2011-05-19 15:38:29 -0400682 do {
683 start = u64_stats_fetch_begin(&stats_cpu->syncp);
684 if (type == BLKIO_STAT_CPU_SECTORS)
685 tval = stats_cpu->sectors;
686 else
687 tval = stats_cpu->stat_arr_cpu[type][sub_type];
688 } while(u64_stats_fetch_retry(&stats_cpu->syncp, start));
689
690 val += tval;
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400691 }
692
693 return val;
694}
695
696static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg,
697 struct cgroup_map_cb *cb, dev_t dev, enum stat_type_cpu type)
698{
699 uint64_t disk_total, val;
700 char key_str[MAX_KEY_LEN];
701 enum stat_sub_type sub_type;
702
703 if (type == BLKIO_STAT_CPU_SECTORS) {
704 val = blkio_read_stat_cpu(blkg, type, 0);
705 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, dev);
706 }
707
708 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
709 sub_type++) {
710 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
711 val = blkio_read_stat_cpu(blkg, type, sub_type);
712 cb->fill(cb, key_str, val);
713 }
714
715 disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) +
716 blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE);
717
718 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
719 cb->fill(cb, key_str, disk_total);
720 return disk_total;
721}
722
Divyesh Shah84c124d2010-04-09 08:31:19 +0200723/* This should be called with blkg->stats_lock held */
724static uint64_t blkio_get_stat(struct blkio_group *blkg,
725 struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700726{
727 uint64_t disk_total;
728 char key_str[MAX_KEY_LEN];
Divyesh Shah84c124d2010-04-09 08:31:19 +0200729 enum stat_sub_type sub_type;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700730
Divyesh Shah84c124d2010-04-09 08:31:19 +0200731 if (type == BLKIO_STAT_TIME)
732 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
733 blkg->stats.time, cb, dev);
Justin TerAvest9026e522011-03-22 21:26:54 +0100734#ifdef CONFIG_DEBUG_BLK_CGROUP
Justin TerAvest167400d2011-03-12 16:54:00 +0100735 if (type == BLKIO_STAT_UNACCOUNTED_TIME)
736 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
737 blkg->stats.unaccounted_time, cb, dev);
Divyesh Shahcdc11842010-04-08 21:15:10 -0700738 if (type == BLKIO_STAT_AVG_QUEUE_SIZE) {
739 uint64_t sum = blkg->stats.avg_queue_size_sum;
740 uint64_t samples = blkg->stats.avg_queue_size_samples;
741 if (samples)
742 do_div(sum, samples);
743 else
744 sum = 0;
745 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev);
746 }
Divyesh Shah812df482010-04-08 21:15:35 -0700747 if (type == BLKIO_STAT_GROUP_WAIT_TIME)
748 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
749 blkg->stats.group_wait_time, cb, dev);
750 if (type == BLKIO_STAT_IDLE_TIME)
751 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
752 blkg->stats.idle_time, cb, dev);
753 if (type == BLKIO_STAT_EMPTY_TIME)
754 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
755 blkg->stats.empty_time, cb, dev);
Divyesh Shah84c124d2010-04-09 08:31:19 +0200756 if (type == BLKIO_STAT_DEQUEUE)
757 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
758 blkg->stats.dequeue, cb, dev);
759#endif
760
761 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
762 sub_type++) {
763 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
764 cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700765 }
Divyesh Shah84c124d2010-04-09 08:31:19 +0200766 disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
767 blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
768 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700769 cb->fill(cb, key_str, disk_total);
770 return disk_total;
771}
772
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800773static int blkio_policy_parse_and_set(char *buf,
Vivek Goyal062a6442010-09-15 17:06:33 -0400774 struct blkio_policy_node *newpn, enum blkio_policy_id plid, int fileid)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800775{
Tejun Heoece84242011-10-19 14:31:15 +0200776 struct gendisk *disk = NULL;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800777 char *s[4], *p, *major_s = NULL, *minor_s = NULL;
Wanlong Gaod11bb442011-09-21 10:22:10 +0200778 unsigned long major, minor;
Tejun Heoece84242011-10-19 14:31:15 +0200779 int i = 0, ret = -EINVAL;
780 int part;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800781 dev_t dev;
Wanlong Gaod11bb442011-09-21 10:22:10 +0200782 u64 temp;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800783
784 memset(s, 0, sizeof(s));
785
786 while ((p = strsep(&buf, " ")) != NULL) {
787 if (!*p)
788 continue;
789
790 s[i++] = p;
791
792 /* Prevent from inputing too many things */
793 if (i == 3)
794 break;
795 }
796
797 if (i != 2)
Tejun Heoece84242011-10-19 14:31:15 +0200798 goto out;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800799
800 p = strsep(&s[0], ":");
801 if (p != NULL)
802 major_s = p;
803 else
Tejun Heoece84242011-10-19 14:31:15 +0200804 goto out;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800805
806 minor_s = s[0];
807 if (!minor_s)
Tejun Heoece84242011-10-19 14:31:15 +0200808 goto out;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800809
Tejun Heoece84242011-10-19 14:31:15 +0200810 if (strict_strtoul(major_s, 10, &major))
811 goto out;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800812
Tejun Heoece84242011-10-19 14:31:15 +0200813 if (strict_strtoul(minor_s, 10, &minor))
814 goto out;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800815
816 dev = MKDEV(major, minor);
817
Tejun Heoece84242011-10-19 14:31:15 +0200818 if (strict_strtoull(s[1], 10, &temp))
819 goto out;
Wanlong Gaod11bb442011-09-21 10:22:10 +0200820
821 /* For rule removal, do not check for device presence. */
822 if (temp) {
Tejun Heoece84242011-10-19 14:31:15 +0200823 disk = get_gendisk(dev, &part);
824 if (!disk || part) {
825 ret = -ENODEV;
826 goto out;
827 }
Wanlong Gaod11bb442011-09-21 10:22:10 +0200828 }
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800829
830 newpn->dev = dev;
831
Vivek Goyal062a6442010-09-15 17:06:33 -0400832 switch (plid) {
833 case BLKIO_POLICY_PROP:
Wanlong Gaod11bb442011-09-21 10:22:10 +0200834 if ((temp < BLKIO_WEIGHT_MIN && temp > 0) ||
835 temp > BLKIO_WEIGHT_MAX)
Tejun Heoece84242011-10-19 14:31:15 +0200836 goto out;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800837
Vivek Goyal062a6442010-09-15 17:06:33 -0400838 newpn->plid = plid;
839 newpn->fileid = fileid;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400840 newpn->val.weight = temp;
841 break;
842 case BLKIO_POLICY_THROTL:
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400843 switch(fileid) {
844 case BLKIO_THROTL_read_bps_device:
845 case BLKIO_THROTL_write_bps_device:
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400846 newpn->plid = plid;
847 newpn->fileid = fileid;
Wanlong Gaod11bb442011-09-21 10:22:10 +0200848 newpn->val.bps = temp;
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400849 break;
850 case BLKIO_THROTL_read_iops_device:
851 case BLKIO_THROTL_write_iops_device:
Wanlong Gaod11bb442011-09-21 10:22:10 +0200852 if (temp > THROTL_IOPS_MAX)
Tejun Heoece84242011-10-19 14:31:15 +0200853 goto out;
Vivek Goyal9355aed2010-10-01 21:16:41 +0200854
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400855 newpn->plid = plid;
856 newpn->fileid = fileid;
Wanlong Gaod11bb442011-09-21 10:22:10 +0200857 newpn->val.iops = (unsigned int)temp;
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400858 break;
859 }
Vivek Goyal062a6442010-09-15 17:06:33 -0400860 break;
861 default:
862 BUG();
863 }
Tejun Heoece84242011-10-19 14:31:15 +0200864 ret = 0;
865out:
866 put_disk(disk);
867 return ret;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800868}
869
870unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
871 dev_t dev)
872{
873 struct blkio_policy_node *pn;
Vivek Goyala38eb632011-10-25 15:48:12 +0200874 unsigned long flags;
875 unsigned int weight;
876
877 spin_lock_irqsave(&blkcg->lock, flags);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800878
Vivek Goyal062a6442010-09-15 17:06:33 -0400879 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_PROP,
880 BLKIO_PROP_weight_device);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800881 if (pn)
Vivek Goyala38eb632011-10-25 15:48:12 +0200882 weight = pn->val.weight;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800883 else
Vivek Goyala38eb632011-10-25 15:48:12 +0200884 weight = blkcg->weight;
885
886 spin_unlock_irqrestore(&blkcg->lock, flags);
887
888 return weight;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800889}
890EXPORT_SYMBOL_GPL(blkcg_get_weight);
891
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400892uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, dev_t dev)
893{
894 struct blkio_policy_node *pn;
Vivek Goyala38eb632011-10-25 15:48:12 +0200895 unsigned long flags;
896 uint64_t bps = -1;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800897
Vivek Goyala38eb632011-10-25 15:48:12 +0200898 spin_lock_irqsave(&blkcg->lock, flags);
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400899 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
900 BLKIO_THROTL_read_bps_device);
901 if (pn)
Vivek Goyala38eb632011-10-25 15:48:12 +0200902 bps = pn->val.bps;
903 spin_unlock_irqrestore(&blkcg->lock, flags);
904
905 return bps;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400906}
907
908uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, dev_t dev)
909{
910 struct blkio_policy_node *pn;
Vivek Goyala38eb632011-10-25 15:48:12 +0200911 unsigned long flags;
912 uint64_t bps = -1;
913
914 spin_lock_irqsave(&blkcg->lock, flags);
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400915 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
916 BLKIO_THROTL_write_bps_device);
917 if (pn)
Vivek Goyala38eb632011-10-25 15:48:12 +0200918 bps = pn->val.bps;
919 spin_unlock_irqrestore(&blkcg->lock, flags);
920
921 return bps;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400922}
923
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400924unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, dev_t dev)
925{
926 struct blkio_policy_node *pn;
Vivek Goyala38eb632011-10-25 15:48:12 +0200927 unsigned long flags;
928 unsigned int iops = -1;
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400929
Vivek Goyala38eb632011-10-25 15:48:12 +0200930 spin_lock_irqsave(&blkcg->lock, flags);
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400931 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
932 BLKIO_THROTL_read_iops_device);
933 if (pn)
Vivek Goyala38eb632011-10-25 15:48:12 +0200934 iops = pn->val.iops;
935 spin_unlock_irqrestore(&blkcg->lock, flags);
936
937 return iops;
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400938}
939
940unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, dev_t dev)
941{
942 struct blkio_policy_node *pn;
Vivek Goyala38eb632011-10-25 15:48:12 +0200943 unsigned long flags;
944 unsigned int iops = -1;
945
946 spin_lock_irqsave(&blkcg->lock, flags);
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400947 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
948 BLKIO_THROTL_write_iops_device);
949 if (pn)
Vivek Goyala38eb632011-10-25 15:48:12 +0200950 iops = pn->val.iops;
951 spin_unlock_irqrestore(&blkcg->lock, flags);
952
953 return iops;
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400954}
955
Vivek Goyal062a6442010-09-15 17:06:33 -0400956/* Checks whether user asked for deleting a policy rule */
957static bool blkio_delete_rule_command(struct blkio_policy_node *pn)
958{
959 switch(pn->plid) {
960 case BLKIO_POLICY_PROP:
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400961 if (pn->val.weight == 0)
962 return 1;
963 break;
964 case BLKIO_POLICY_THROTL:
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400965 switch(pn->fileid) {
966 case BLKIO_THROTL_read_bps_device:
967 case BLKIO_THROTL_write_bps_device:
968 if (pn->val.bps == 0)
969 return 1;
970 break;
971 case BLKIO_THROTL_read_iops_device:
972 case BLKIO_THROTL_write_iops_device:
973 if (pn->val.iops == 0)
974 return 1;
975 }
Vivek Goyal062a6442010-09-15 17:06:33 -0400976 break;
977 default:
978 BUG();
979 }
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800980
Vivek Goyal062a6442010-09-15 17:06:33 -0400981 return 0;
982}
983
984static void blkio_update_policy_rule(struct blkio_policy_node *oldpn,
985 struct blkio_policy_node *newpn)
986{
987 switch(oldpn->plid) {
988 case BLKIO_POLICY_PROP:
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400989 oldpn->val.weight = newpn->val.weight;
990 break;
991 case BLKIO_POLICY_THROTL:
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400992 switch(newpn->fileid) {
993 case BLKIO_THROTL_read_bps_device:
994 case BLKIO_THROTL_write_bps_device:
995 oldpn->val.bps = newpn->val.bps;
996 break;
997 case BLKIO_THROTL_read_iops_device:
998 case BLKIO_THROTL_write_iops_device:
999 oldpn->val.iops = newpn->val.iops;
1000 }
Vivek Goyal062a6442010-09-15 17:06:33 -04001001 break;
1002 default:
1003 BUG();
1004 }
1005}
1006
1007/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001008 * Some rules/values in blkg have changed. Propagate those to respective
Vivek Goyal062a6442010-09-15 17:06:33 -04001009 * policies.
1010 */
1011static void blkio_update_blkg_policy(struct blkio_cgroup *blkcg,
1012 struct blkio_group *blkg, struct blkio_policy_node *pn)
1013{
Vivek Goyal7702e8f2010-09-15 17:06:36 -04001014 unsigned int weight, iops;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001015 u64 bps;
Vivek Goyal062a6442010-09-15 17:06:33 -04001016
1017 switch(pn->plid) {
1018 case BLKIO_POLICY_PROP:
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001019 weight = pn->val.weight ? pn->val.weight :
Vivek Goyal062a6442010-09-15 17:06:33 -04001020 blkcg->weight;
1021 blkio_update_group_weight(blkg, weight);
1022 break;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001023 case BLKIO_POLICY_THROTL:
1024 switch(pn->fileid) {
1025 case BLKIO_THROTL_read_bps_device:
1026 case BLKIO_THROTL_write_bps_device:
1027 bps = pn->val.bps ? pn->val.bps : (-1);
1028 blkio_update_group_bps(blkg, bps, pn->fileid);
1029 break;
Vivek Goyal7702e8f2010-09-15 17:06:36 -04001030 case BLKIO_THROTL_read_iops_device:
1031 case BLKIO_THROTL_write_iops_device:
1032 iops = pn->val.iops ? pn->val.iops : (-1);
1033 blkio_update_group_iops(blkg, iops, pn->fileid);
1034 break;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001035 }
1036 break;
Vivek Goyal062a6442010-09-15 17:06:33 -04001037 default:
1038 BUG();
1039 }
1040}
1041
1042/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001043 * A policy node rule has been updated. Propagate this update to all the
Vivek Goyal062a6442010-09-15 17:06:33 -04001044 * block groups which might be affected by this update.
1045 */
1046static void blkio_update_policy_node_blkg(struct blkio_cgroup *blkcg,
1047 struct blkio_policy_node *pn)
1048{
1049 struct blkio_group *blkg;
1050 struct hlist_node *n;
1051
1052 spin_lock(&blkio_list_lock);
1053 spin_lock_irq(&blkcg->lock);
1054
1055 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1056 if (pn->dev != blkg->dev || pn->plid != blkg->plid)
1057 continue;
1058 blkio_update_blkg_policy(blkcg, blkg, pn);
1059 }
1060
1061 spin_unlock_irq(&blkcg->lock);
1062 spin_unlock(&blkio_list_lock);
1063}
1064
1065static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft,
1066 const char *buffer)
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001067{
1068 int ret = 0;
1069 char *buf;
1070 struct blkio_policy_node *newpn, *pn;
1071 struct blkio_cgroup *blkcg;
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001072 int keep_newpn = 0;
Vivek Goyal062a6442010-09-15 17:06:33 -04001073 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1074 int fileid = BLKIOFILE_ATTR(cft->private);
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001075
1076 buf = kstrdup(buffer, GFP_KERNEL);
1077 if (!buf)
1078 return -ENOMEM;
1079
1080 newpn = kzalloc(sizeof(*newpn), GFP_KERNEL);
1081 if (!newpn) {
1082 ret = -ENOMEM;
1083 goto free_buf;
1084 }
1085
Vivek Goyal062a6442010-09-15 17:06:33 -04001086 ret = blkio_policy_parse_and_set(buf, newpn, plid, fileid);
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001087 if (ret)
1088 goto free_newpn;
1089
1090 blkcg = cgroup_to_blkio_cgroup(cgrp);
1091
1092 spin_lock_irq(&blkcg->lock);
1093
Vivek Goyal062a6442010-09-15 17:06:33 -04001094 pn = blkio_policy_search_node(blkcg, newpn->dev, plid, fileid);
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001095 if (!pn) {
Vivek Goyal062a6442010-09-15 17:06:33 -04001096 if (!blkio_delete_rule_command(newpn)) {
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001097 blkio_policy_insert_node(blkcg, newpn);
1098 keep_newpn = 1;
1099 }
1100 spin_unlock_irq(&blkcg->lock);
1101 goto update_io_group;
1102 }
1103
Vivek Goyal062a6442010-09-15 17:06:33 -04001104 if (blkio_delete_rule_command(newpn)) {
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001105 blkio_policy_delete_node(pn);
Vivek Goyale060f002011-10-25 15:48:12 +02001106 kfree(pn);
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001107 spin_unlock_irq(&blkcg->lock);
1108 goto update_io_group;
1109 }
1110 spin_unlock_irq(&blkcg->lock);
1111
Vivek Goyal062a6442010-09-15 17:06:33 -04001112 blkio_update_policy_rule(pn, newpn);
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001113
1114update_io_group:
Vivek Goyal062a6442010-09-15 17:06:33 -04001115 blkio_update_policy_node_blkg(blkcg, newpn);
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001116
1117free_newpn:
1118 if (!keep_newpn)
1119 kfree(newpn);
1120free_buf:
1121 kfree(buf);
1122 return ret;
1123}
1124
Vivek Goyal062a6442010-09-15 17:06:33 -04001125static void
1126blkio_print_policy_node(struct seq_file *m, struct blkio_policy_node *pn)
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001127{
Vivek Goyal062a6442010-09-15 17:06:33 -04001128 switch(pn->plid) {
1129 case BLKIO_POLICY_PROP:
1130 if (pn->fileid == BLKIO_PROP_weight_device)
1131 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001132 MINOR(pn->dev), pn->val.weight);
1133 break;
1134 case BLKIO_POLICY_THROTL:
Vivek Goyal7702e8f2010-09-15 17:06:36 -04001135 switch(pn->fileid) {
1136 case BLKIO_THROTL_read_bps_device:
1137 case BLKIO_THROTL_write_bps_device:
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001138 seq_printf(m, "%u:%u\t%llu\n", MAJOR(pn->dev),
1139 MINOR(pn->dev), pn->val.bps);
Vivek Goyal7702e8f2010-09-15 17:06:36 -04001140 break;
1141 case BLKIO_THROTL_read_iops_device:
1142 case BLKIO_THROTL_write_iops_device:
1143 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1144 MINOR(pn->dev), pn->val.iops);
1145 break;
1146 }
Vivek Goyal062a6442010-09-15 17:06:33 -04001147 break;
1148 default:
1149 BUG();
1150 }
1151}
1152
1153/* cgroup files which read their data from policy nodes end up here */
1154static void blkio_read_policy_node_files(struct cftype *cft,
1155 struct blkio_cgroup *blkcg, struct seq_file *m)
1156{
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001157 struct blkio_policy_node *pn;
1158
Jens Axboe0f3942a2010-05-03 14:28:55 +02001159 if (!list_empty(&blkcg->policy_list)) {
1160 spin_lock_irq(&blkcg->lock);
1161 list_for_each_entry(pn, &blkcg->policy_list, node) {
Vivek Goyal062a6442010-09-15 17:06:33 -04001162 if (!pn_matches_cftype(cft, pn))
1163 continue;
1164 blkio_print_policy_node(m, pn);
Jens Axboe0f3942a2010-05-03 14:28:55 +02001165 }
1166 spin_unlock_irq(&blkcg->lock);
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001167 }
Vivek Goyal062a6442010-09-15 17:06:33 -04001168}
1169
1170static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft,
1171 struct seq_file *m)
1172{
1173 struct blkio_cgroup *blkcg;
1174 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1175 int name = BLKIOFILE_ATTR(cft->private);
1176
1177 blkcg = cgroup_to_blkio_cgroup(cgrp);
1178
1179 switch(plid) {
1180 case BLKIO_POLICY_PROP:
1181 switch(name) {
1182 case BLKIO_PROP_weight_device:
1183 blkio_read_policy_node_files(cft, blkcg, m);
1184 return 0;
1185 default:
1186 BUG();
1187 }
1188 break;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001189 case BLKIO_POLICY_THROTL:
1190 switch(name){
1191 case BLKIO_THROTL_read_bps_device:
1192 case BLKIO_THROTL_write_bps_device:
Vivek Goyal7702e8f2010-09-15 17:06:36 -04001193 case BLKIO_THROTL_read_iops_device:
1194 case BLKIO_THROTL_write_iops_device:
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001195 blkio_read_policy_node_files(cft, blkcg, m);
1196 return 0;
1197 default:
1198 BUG();
1199 }
1200 break;
Vivek Goyal062a6442010-09-15 17:06:33 -04001201 default:
1202 BUG();
1203 }
1204
1205 return 0;
1206}
1207
1208static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001209 struct cftype *cft, struct cgroup_map_cb *cb,
1210 enum stat_type type, bool show_total, bool pcpu)
Vivek Goyal062a6442010-09-15 17:06:33 -04001211{
1212 struct blkio_group *blkg;
1213 struct hlist_node *n;
1214 uint64_t cgroup_total = 0;
1215
1216 rcu_read_lock();
1217 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
1218 if (blkg->dev) {
1219 if (!cftype_blkg_same_policy(cft, blkg))
1220 continue;
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001221 if (pcpu)
1222 cgroup_total += blkio_get_stat_cpu(blkg, cb,
1223 blkg->dev, type);
1224 else {
1225 spin_lock_irq(&blkg->stats_lock);
1226 cgroup_total += blkio_get_stat(blkg, cb,
1227 blkg->dev, type);
1228 spin_unlock_irq(&blkg->stats_lock);
1229 }
Vivek Goyal062a6442010-09-15 17:06:33 -04001230 }
1231 }
1232 if (show_total)
1233 cb->fill(cb, "Total", cgroup_total);
1234 rcu_read_unlock();
1235 return 0;
1236}
1237
1238/* All map kind of cgroup file get serviced by this function */
1239static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft,
1240 struct cgroup_map_cb *cb)
1241{
1242 struct blkio_cgroup *blkcg;
1243 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1244 int name = BLKIOFILE_ATTR(cft->private);
1245
1246 blkcg = cgroup_to_blkio_cgroup(cgrp);
1247
1248 switch(plid) {
1249 case BLKIO_POLICY_PROP:
1250 switch(name) {
1251 case BLKIO_PROP_time:
1252 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001253 BLKIO_STAT_TIME, 0, 0);
Vivek Goyal062a6442010-09-15 17:06:33 -04001254 case BLKIO_PROP_sectors:
1255 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001256 BLKIO_STAT_CPU_SECTORS, 0, 1);
Vivek Goyal062a6442010-09-15 17:06:33 -04001257 case BLKIO_PROP_io_service_bytes:
1258 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001259 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
Vivek Goyal062a6442010-09-15 17:06:33 -04001260 case BLKIO_PROP_io_serviced:
1261 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001262 BLKIO_STAT_CPU_SERVICED, 1, 1);
Vivek Goyal062a6442010-09-15 17:06:33 -04001263 case BLKIO_PROP_io_service_time:
1264 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001265 BLKIO_STAT_SERVICE_TIME, 1, 0);
Vivek Goyal062a6442010-09-15 17:06:33 -04001266 case BLKIO_PROP_io_wait_time:
1267 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001268 BLKIO_STAT_WAIT_TIME, 1, 0);
Vivek Goyal062a6442010-09-15 17:06:33 -04001269 case BLKIO_PROP_io_merged:
1270 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal317389a2011-05-23 10:02:19 +02001271 BLKIO_STAT_CPU_MERGED, 1, 1);
Vivek Goyal062a6442010-09-15 17:06:33 -04001272 case BLKIO_PROP_io_queued:
1273 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001274 BLKIO_STAT_QUEUED, 1, 0);
Vivek Goyal062a6442010-09-15 17:06:33 -04001275#ifdef CONFIG_DEBUG_BLK_CGROUP
Justin TerAvest9026e522011-03-22 21:26:54 +01001276 case BLKIO_PROP_unaccounted_time:
1277 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001278 BLKIO_STAT_UNACCOUNTED_TIME, 0, 0);
Vivek Goyal062a6442010-09-15 17:06:33 -04001279 case BLKIO_PROP_dequeue:
1280 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001281 BLKIO_STAT_DEQUEUE, 0, 0);
Vivek Goyal062a6442010-09-15 17:06:33 -04001282 case BLKIO_PROP_avg_queue_size:
1283 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001284 BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0);
Vivek Goyal062a6442010-09-15 17:06:33 -04001285 case BLKIO_PROP_group_wait_time:
1286 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001287 BLKIO_STAT_GROUP_WAIT_TIME, 0, 0);
Vivek Goyal062a6442010-09-15 17:06:33 -04001288 case BLKIO_PROP_idle_time:
1289 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001290 BLKIO_STAT_IDLE_TIME, 0, 0);
Vivek Goyal062a6442010-09-15 17:06:33 -04001291 case BLKIO_PROP_empty_time:
1292 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001293 BLKIO_STAT_EMPTY_TIME, 0, 0);
Vivek Goyal062a6442010-09-15 17:06:33 -04001294#endif
1295 default:
1296 BUG();
1297 }
1298 break;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001299 case BLKIO_POLICY_THROTL:
1300 switch(name){
1301 case BLKIO_THROTL_io_service_bytes:
1302 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001303 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001304 case BLKIO_THROTL_io_serviced:
1305 return blkio_read_blkg_stats(blkcg, cft, cb,
Vivek Goyal5624a4e2011-05-19 15:38:28 -04001306 BLKIO_STAT_CPU_SERVICED, 1, 1);
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001307 default:
1308 BUG();
1309 }
1310 break;
Vivek Goyal062a6442010-09-15 17:06:33 -04001311 default:
1312 BUG();
1313 }
1314
1315 return 0;
1316}
1317
1318static int blkio_weight_write(struct blkio_cgroup *blkcg, u64 val)
1319{
1320 struct blkio_group *blkg;
1321 struct hlist_node *n;
1322 struct blkio_policy_node *pn;
1323
1324 if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
1325 return -EINVAL;
1326
1327 spin_lock(&blkio_list_lock);
1328 spin_lock_irq(&blkcg->lock);
1329 blkcg->weight = (unsigned int)val;
1330
1331 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1332 pn = blkio_policy_search_node(blkcg, blkg->dev,
1333 BLKIO_POLICY_PROP, BLKIO_PROP_weight_device);
1334 if (pn)
1335 continue;
1336
1337 blkio_update_group_weight(blkg, blkcg->weight);
1338 }
1339 spin_unlock_irq(&blkcg->lock);
1340 spin_unlock(&blkio_list_lock);
1341 return 0;
1342}
1343
1344static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) {
1345 struct blkio_cgroup *blkcg;
1346 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1347 int name = BLKIOFILE_ATTR(cft->private);
1348
1349 blkcg = cgroup_to_blkio_cgroup(cgrp);
1350
1351 switch(plid) {
1352 case BLKIO_POLICY_PROP:
1353 switch(name) {
1354 case BLKIO_PROP_weight:
1355 return (u64)blkcg->weight;
1356 }
1357 break;
1358 default:
1359 BUG();
1360 }
1361 return 0;
1362}
1363
1364static int
1365blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1366{
1367 struct blkio_cgroup *blkcg;
1368 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1369 int name = BLKIOFILE_ATTR(cft->private);
1370
1371 blkcg = cgroup_to_blkio_cgroup(cgrp);
1372
1373 switch(plid) {
1374 case BLKIO_POLICY_PROP:
1375 switch(name) {
1376 case BLKIO_PROP_weight:
1377 return blkio_weight_write(blkcg, val);
1378 }
1379 break;
1380 default:
1381 BUG();
1382 }
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001383
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001384 return 0;
1385}
1386
Vivek Goyal31e4c282009-12-03 12:59:42 -05001387struct cftype blkio_files[] = {
1388 {
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001389 .name = "weight_device",
Vivek Goyal062a6442010-09-15 17:06:33 -04001390 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1391 BLKIO_PROP_weight_device),
1392 .read_seq_string = blkiocg_file_read,
1393 .write_string = blkiocg_file_write,
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001394 .max_write_len = 256,
1395 },
1396 {
Vivek Goyal31e4c282009-12-03 12:59:42 -05001397 .name = "weight",
Vivek Goyal062a6442010-09-15 17:06:33 -04001398 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1399 BLKIO_PROP_weight),
1400 .read_u64 = blkiocg_file_read_u64,
1401 .write_u64 = blkiocg_file_write_u64,
Vivek Goyal31e4c282009-12-03 12:59:42 -05001402 },
Vivek Goyal22084192009-12-03 12:59:49 -05001403 {
1404 .name = "time",
Vivek Goyal13f98252010-10-01 14:49:41 +02001405 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1406 BLKIO_PROP_time),
1407 .read_map = blkiocg_file_read_map,
Vivek Goyal22084192009-12-03 12:59:49 -05001408 },
1409 {
1410 .name = "sectors",
Vivek Goyal13f98252010-10-01 14:49:41 +02001411 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1412 BLKIO_PROP_sectors),
1413 .read_map = blkiocg_file_read_map,
Divyesh Shah303a3ac2010-04-01 15:01:24 -07001414 },
1415 {
1416 .name = "io_service_bytes",
Vivek Goyal13f98252010-10-01 14:49:41 +02001417 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1418 BLKIO_PROP_io_service_bytes),
1419 .read_map = blkiocg_file_read_map,
Divyesh Shah303a3ac2010-04-01 15:01:24 -07001420 },
1421 {
1422 .name = "io_serviced",
Vivek Goyal13f98252010-10-01 14:49:41 +02001423 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1424 BLKIO_PROP_io_serviced),
1425 .read_map = blkiocg_file_read_map,
Divyesh Shah303a3ac2010-04-01 15:01:24 -07001426 },
1427 {
1428 .name = "io_service_time",
Vivek Goyal13f98252010-10-01 14:49:41 +02001429 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1430 BLKIO_PROP_io_service_time),
1431 .read_map = blkiocg_file_read_map,
Divyesh Shah303a3ac2010-04-01 15:01:24 -07001432 },
1433 {
1434 .name = "io_wait_time",
Vivek Goyal13f98252010-10-01 14:49:41 +02001435 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1436 BLKIO_PROP_io_wait_time),
1437 .read_map = blkiocg_file_read_map,
Divyesh Shah84c124d2010-04-09 08:31:19 +02001438 },
1439 {
Divyesh Shah812d4022010-04-08 21:14:23 -07001440 .name = "io_merged",
Vivek Goyal13f98252010-10-01 14:49:41 +02001441 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1442 BLKIO_PROP_io_merged),
1443 .read_map = blkiocg_file_read_map,
Divyesh Shah812d4022010-04-08 21:14:23 -07001444 },
1445 {
Divyesh Shahcdc11842010-04-08 21:15:10 -07001446 .name = "io_queued",
Vivek Goyal13f98252010-10-01 14:49:41 +02001447 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1448 BLKIO_PROP_io_queued),
1449 .read_map = blkiocg_file_read_map,
Divyesh Shahcdc11842010-04-08 21:15:10 -07001450 },
1451 {
Divyesh Shah84c124d2010-04-09 08:31:19 +02001452 .name = "reset_stats",
1453 .write_u64 = blkiocg_reset_stats,
Vivek Goyal22084192009-12-03 12:59:49 -05001454 },
Vivek Goyal13f98252010-10-01 14:49:41 +02001455#ifdef CONFIG_BLK_DEV_THROTTLING
1456 {
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001457 .name = "throttle.read_bps_device",
1458 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1459 BLKIO_THROTL_read_bps_device),
1460 .read_seq_string = blkiocg_file_read,
1461 .write_string = blkiocg_file_write,
1462 .max_write_len = 256,
1463 },
1464
1465 {
1466 .name = "throttle.write_bps_device",
1467 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1468 BLKIO_THROTL_write_bps_device),
1469 .read_seq_string = blkiocg_file_read,
1470 .write_string = blkiocg_file_write,
1471 .max_write_len = 256,
1472 },
Vivek Goyal7702e8f2010-09-15 17:06:36 -04001473
1474 {
1475 .name = "throttle.read_iops_device",
1476 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1477 BLKIO_THROTL_read_iops_device),
1478 .read_seq_string = blkiocg_file_read,
1479 .write_string = blkiocg_file_write,
1480 .max_write_len = 256,
1481 },
1482
1483 {
1484 .name = "throttle.write_iops_device",
1485 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1486 BLKIO_THROTL_write_iops_device),
1487 .read_seq_string = blkiocg_file_read,
1488 .write_string = blkiocg_file_write,
1489 .max_write_len = 256,
1490 },
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001491 {
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001492 .name = "throttle.io_service_bytes",
1493 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1494 BLKIO_THROTL_io_service_bytes),
1495 .read_map = blkiocg_file_read_map,
1496 },
1497 {
Vivek Goyal4c9eefa2010-09-15 17:06:34 -04001498 .name = "throttle.io_serviced",
1499 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1500 BLKIO_THROTL_io_serviced),
1501 .read_map = blkiocg_file_read_map,
1502 },
Vivek Goyal13f98252010-10-01 14:49:41 +02001503#endif /* CONFIG_BLK_DEV_THROTTLING */
1504
Vivek Goyal22084192009-12-03 12:59:49 -05001505#ifdef CONFIG_DEBUG_BLK_CGROUP
Divyesh Shahcdc11842010-04-08 21:15:10 -07001506 {
1507 .name = "avg_queue_size",
Vivek Goyal062a6442010-09-15 17:06:33 -04001508 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1509 BLKIO_PROP_avg_queue_size),
1510 .read_map = blkiocg_file_read_map,
Divyesh Shahcdc11842010-04-08 21:15:10 -07001511 },
1512 {
Divyesh Shah812df482010-04-08 21:15:35 -07001513 .name = "group_wait_time",
Vivek Goyal062a6442010-09-15 17:06:33 -04001514 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1515 BLKIO_PROP_group_wait_time),
1516 .read_map = blkiocg_file_read_map,
Divyesh Shah812df482010-04-08 21:15:35 -07001517 },
1518 {
1519 .name = "idle_time",
Vivek Goyal062a6442010-09-15 17:06:33 -04001520 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1521 BLKIO_PROP_idle_time),
1522 .read_map = blkiocg_file_read_map,
Divyesh Shah812df482010-04-08 21:15:35 -07001523 },
1524 {
1525 .name = "empty_time",
Vivek Goyal062a6442010-09-15 17:06:33 -04001526 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1527 BLKIO_PROP_empty_time),
1528 .read_map = blkiocg_file_read_map,
Divyesh Shah812df482010-04-08 21:15:35 -07001529 },
1530 {
Vivek Goyal22084192009-12-03 12:59:49 -05001531 .name = "dequeue",
Vivek Goyal062a6442010-09-15 17:06:33 -04001532 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1533 BLKIO_PROP_dequeue),
1534 .read_map = blkiocg_file_read_map,
Divyesh Shahcdc11842010-04-08 21:15:10 -07001535 },
Justin TerAvest9026e522011-03-22 21:26:54 +01001536 {
1537 .name = "unaccounted_time",
1538 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1539 BLKIO_PROP_unaccounted_time),
1540 .read_map = blkiocg_file_read_map,
1541 },
Vivek Goyal22084192009-12-03 12:59:49 -05001542#endif
Vivek Goyal31e4c282009-12-03 12:59:42 -05001543};
1544
1545static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1546{
1547 return cgroup_add_files(cgroup, subsys, blkio_files,
1548 ARRAY_SIZE(blkio_files));
1549}
1550
1551static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1552{
1553 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
Vivek Goyalb1c35762009-12-03 12:59:47 -05001554 unsigned long flags;
1555 struct blkio_group *blkg;
1556 void *key;
Vivek Goyal3e252062009-12-04 10:36:42 -05001557 struct blkio_policy_type *blkiop;
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001558 struct blkio_policy_node *pn, *pntmp;
Vivek Goyal31e4c282009-12-03 12:59:42 -05001559
Vivek Goyalb1c35762009-12-03 12:59:47 -05001560 rcu_read_lock();
Jens Axboe0f3942a2010-05-03 14:28:55 +02001561 do {
1562 spin_lock_irqsave(&blkcg->lock, flags);
Vivek Goyalb1c35762009-12-03 12:59:47 -05001563
Jens Axboe0f3942a2010-05-03 14:28:55 +02001564 if (hlist_empty(&blkcg->blkg_list)) {
1565 spin_unlock_irqrestore(&blkcg->lock, flags);
1566 break;
1567 }
1568
1569 blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
1570 blkcg_node);
1571 key = rcu_dereference(blkg->key);
1572 __blkiocg_del_blkio_group(blkg);
1573
Vivek Goyalb1c35762009-12-03 12:59:47 -05001574 spin_unlock_irqrestore(&blkcg->lock, flags);
Vivek Goyalb1c35762009-12-03 12:59:47 -05001575
Jens Axboe0f3942a2010-05-03 14:28:55 +02001576 /*
1577 * This blkio_group is being unlinked as associated cgroup is
1578 * going away. Let all the IO controlling policies know about
Vivek Goyal61014e92010-10-01 14:49:44 +02001579 * this event.
Jens Axboe0f3942a2010-05-03 14:28:55 +02001580 */
1581 spin_lock(&blkio_list_lock);
Vivek Goyal61014e92010-10-01 14:49:44 +02001582 list_for_each_entry(blkiop, &blkio_list, list) {
1583 if (blkiop->plid != blkg->plid)
1584 continue;
Jens Axboe0f3942a2010-05-03 14:28:55 +02001585 blkiop->ops.blkio_unlink_group_fn(key, blkg);
Vivek Goyal61014e92010-10-01 14:49:44 +02001586 }
Jens Axboe0f3942a2010-05-03 14:28:55 +02001587 spin_unlock(&blkio_list_lock);
1588 } while (1);
Vivek Goyalb1c35762009-12-03 12:59:47 -05001589
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001590 list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) {
1591 blkio_policy_delete_node(pn);
1592 kfree(pn);
1593 }
Jens Axboe0f3942a2010-05-03 14:28:55 +02001594
Vivek Goyal31e4c282009-12-03 12:59:42 -05001595 free_css_id(&blkio_subsys, &blkcg->css);
Vivek Goyalb1c35762009-12-03 12:59:47 -05001596 rcu_read_unlock();
Ben Blum67523c42010-03-10 15:22:11 -08001597 if (blkcg != &blkio_root_cgroup)
1598 kfree(blkcg);
Vivek Goyal31e4c282009-12-03 12:59:42 -05001599}
1600
1601static struct cgroup_subsys_state *
1602blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1603{
Li Zefan03415092010-05-07 08:57:00 +02001604 struct blkio_cgroup *blkcg;
1605 struct cgroup *parent = cgroup->parent;
Vivek Goyal31e4c282009-12-03 12:59:42 -05001606
Li Zefan03415092010-05-07 08:57:00 +02001607 if (!parent) {
Vivek Goyal31e4c282009-12-03 12:59:42 -05001608 blkcg = &blkio_root_cgroup;
1609 goto done;
1610 }
1611
Vivek Goyal31e4c282009-12-03 12:59:42 -05001612 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1613 if (!blkcg)
1614 return ERR_PTR(-ENOMEM);
1615
1616 blkcg->weight = BLKIO_WEIGHT_DEFAULT;
1617done:
1618 spin_lock_init(&blkcg->lock);
1619 INIT_HLIST_HEAD(&blkcg->blkg_list);
1620
Gui Jianfeng34d0f172010-04-13 16:05:49 +08001621 INIT_LIST_HEAD(&blkcg->policy_list);
Vivek Goyal31e4c282009-12-03 12:59:42 -05001622 return &blkcg->css;
1623}
1624
1625/*
1626 * We cannot support shared io contexts, as we have no mean to support
1627 * two tasks with the same ioc in two different groups without major rework
1628 * of the main cic data structures. For now we allow a task to change
1629 * its cgroup only if it's the only owner of its ioc.
1630 */
Tejun Heobb9d97b2011-12-12 18:12:21 -08001631static int blkiocg_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
1632 struct cgroup_taskset *tset)
Vivek Goyal31e4c282009-12-03 12:59:42 -05001633{
Tejun Heobb9d97b2011-12-12 18:12:21 -08001634 struct task_struct *task;
Vivek Goyal31e4c282009-12-03 12:59:42 -05001635 struct io_context *ioc;
1636 int ret = 0;
1637
1638 /* task_lock() is needed to avoid races with exit_io_context() */
Tejun Heobb9d97b2011-12-12 18:12:21 -08001639 cgroup_taskset_for_each(task, cgrp, tset) {
1640 task_lock(task);
1641 ioc = task->io_context;
1642 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1643 ret = -EINVAL;
1644 task_unlock(task);
1645 if (ret)
1646 break;
1647 }
Vivek Goyal31e4c282009-12-03 12:59:42 -05001648 return ret;
1649}
1650
Tejun Heobb9d97b2011-12-12 18:12:21 -08001651static void blkiocg_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
1652 struct cgroup_taskset *tset)
Vivek Goyal31e4c282009-12-03 12:59:42 -05001653{
Tejun Heobb9d97b2011-12-12 18:12:21 -08001654 struct task_struct *task;
Vivek Goyal31e4c282009-12-03 12:59:42 -05001655 struct io_context *ioc;
1656
Tejun Heobb9d97b2011-12-12 18:12:21 -08001657 cgroup_taskset_for_each(task, cgrp, tset) {
Linus Torvaldsb3c9dd12012-01-15 12:24:45 -08001658 /* we don't lose anything even if ioc allocation fails */
1659 ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE);
1660 if (ioc) {
1661 ioc_cgroup_changed(ioc);
Tejun Heo11a31222012-02-07 07:51:30 +01001662 put_io_context(ioc);
Linus Torvaldsb3c9dd12012-01-15 12:24:45 -08001663 }
Tejun Heobb9d97b2011-12-12 18:12:21 -08001664 }
Vivek Goyal31e4c282009-12-03 12:59:42 -05001665}
1666
Vivek Goyal3e252062009-12-04 10:36:42 -05001667void blkio_policy_register(struct blkio_policy_type *blkiop)
1668{
1669 spin_lock(&blkio_list_lock);
1670 list_add_tail(&blkiop->list, &blkio_list);
1671 spin_unlock(&blkio_list_lock);
1672}
1673EXPORT_SYMBOL_GPL(blkio_policy_register);
1674
1675void blkio_policy_unregister(struct blkio_policy_type *blkiop)
1676{
1677 spin_lock(&blkio_list_lock);
1678 list_del_init(&blkiop->list);
1679 spin_unlock(&blkio_list_lock);
1680}
1681EXPORT_SYMBOL_GPL(blkio_policy_unregister);
Ben Blum67523c42010-03-10 15:22:11 -08001682
1683static int __init init_cgroup_blkio(void)
1684{
1685 return cgroup_load_subsys(&blkio_subsys);
1686}
1687
1688static void __exit exit_cgroup_blkio(void)
1689{
1690 cgroup_unload_subsys(&blkio_subsys);
1691}
1692
1693module_init(init_cgroup_blkio);
1694module_exit(exit_cgroup_blkio);
1695MODULE_LICENSE("GPL");