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 | 14 | #include <linux/kdev_t.h> |
9d6a986c | 15 | #include <linux/module.h> |
accee785 | 16 | #include <linux/err.h> |
9195291e | 17 | #include <linux/blkdev.h> |
5a0e3ad6 | 18 | #include <linux/slab.h> |
34d0f179 | 19 | #include <linux/genhd.h> |
72e06c25 | 20 | #include <linux/delay.h> |
9a9e8a26 | 21 | #include <linux/atomic.h> |
72e06c25 | 22 | #include "blk-cgroup.h" |
5efd6113 | 23 | #include "blk.h" |
3e252066 | 24 | |
84c124da DS |
25 | #define MAX_KEY_LEN 100 |
26 | ||
bc0d6501 | 27 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 28 | |
3c798398 TH |
29 | struct blkcg blkcg_root = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT }; |
30 | EXPORT_SYMBOL_GPL(blkcg_root); | |
9d6a986c | 31 | |
3c798398 | 32 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 33 | |
3c798398 | 34 | struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup) |
31e4c28d VG |
35 | { |
36 | return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id), | |
3c798398 | 37 | struct blkcg, css); |
31e4c28d | 38 | } |
3c798398 | 39 | EXPORT_SYMBOL_GPL(cgroup_to_blkcg); |
31e4c28d | 40 | |
3c798398 | 41 | static struct blkcg *task_blkcg(struct task_struct *tsk) |
70087dc3 VG |
42 | { |
43 | return container_of(task_subsys_state(tsk, blkio_subsys_id), | |
3c798398 | 44 | struct blkcg, css); |
70087dc3 | 45 | } |
4f85cb96 | 46 | |
3c798398 | 47 | struct blkcg *bio_blkcg(struct bio *bio) |
4f85cb96 TH |
48 | { |
49 | if (bio && bio->bi_css) | |
3c798398 TH |
50 | return container_of(bio->bi_css, struct blkcg, css); |
51 | return task_blkcg(current); | |
4f85cb96 | 52 | } |
3c798398 | 53 | EXPORT_SYMBOL_GPL(bio_blkcg); |
70087dc3 | 54 | |
a2b1693b | 55 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 56 | const struct blkcg_policy *pol) |
a2b1693b TH |
57 | { |
58 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
59 | } | |
60 | ||
0381411e TH |
61 | /** |
62 | * blkg_free - free a blkg | |
63 | * @blkg: blkg to free | |
64 | * | |
65 | * Free @blkg which may be partially allocated. | |
66 | */ | |
3c798398 | 67 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 68 | { |
e8989fae | 69 | int i; |
549d3aa8 TH |
70 | |
71 | if (!blkg) | |
72 | return; | |
73 | ||
8bd435b3 | 74 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 75 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae TH |
76 | struct blkg_policy_data *pd = blkg->pd[i]; |
77 | ||
9ade5ea4 TH |
78 | if (!pd) |
79 | continue; | |
80 | ||
f9fcc2d3 TH |
81 | if (pol && pol->pd_exit_fn) |
82 | pol->pd_exit_fn(blkg); | |
9ade5ea4 | 83 | |
9ade5ea4 | 84 | kfree(pd); |
0381411e | 85 | } |
e8989fae | 86 | |
549d3aa8 | 87 | kfree(blkg); |
0381411e TH |
88 | } |
89 | ||
90 | /** | |
91 | * blkg_alloc - allocate a blkg | |
92 | * @blkcg: block cgroup the new blkg is associated with | |
93 | * @q: request_queue the new blkg is associated with | |
0381411e | 94 | * |
e8989fae | 95 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 96 | */ |
3c798398 | 97 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q) |
0381411e | 98 | { |
3c798398 | 99 | struct blkcg_gq *blkg; |
e8989fae | 100 | int i; |
0381411e TH |
101 | |
102 | /* alloc and init base part */ | |
103 | blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node); | |
104 | if (!blkg) | |
105 | return NULL; | |
106 | ||
c875f4d0 | 107 | blkg->q = q; |
e8989fae | 108 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 109 | blkg->blkcg = blkcg; |
1adaf3dd | 110 | blkg->refcnt = 1; |
0381411e | 111 | |
8bd435b3 | 112 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 113 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 114 | struct blkg_policy_data *pd; |
0381411e | 115 | |
a2b1693b | 116 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
117 | continue; |
118 | ||
119 | /* alloc per-policy data and attach it to blkg */ | |
f95a04af | 120 | pd = kzalloc_node(pol->pd_size, GFP_ATOMIC, q->node); |
e8989fae TH |
121 | if (!pd) { |
122 | blkg_free(blkg); | |
123 | return NULL; | |
124 | } | |
549d3aa8 | 125 | |
e8989fae TH |
126 | blkg->pd[i] = pd; |
127 | pd->blkg = blkg; | |
0381411e TH |
128 | } |
129 | ||
549d3aa8 | 130 | /* invoke per-policy init */ |
8bd435b3 | 131 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 132 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 133 | |
a2b1693b | 134 | if (blkcg_policy_enabled(blkg->q, pol)) |
f9fcc2d3 | 135 | pol->pd_init_fn(blkg); |
e8989fae TH |
136 | } |
137 | ||
0381411e TH |
138 | return blkg; |
139 | } | |
140 | ||
3c798398 TH |
141 | static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, |
142 | struct request_queue *q) | |
80fd9979 | 143 | { |
3c798398 | 144 | struct blkcg_gq *blkg; |
80fd9979 TH |
145 | struct hlist_node *n; |
146 | ||
147 | hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) | |
148 | if (blkg->q == q) | |
149 | return blkg; | |
150 | return NULL; | |
151 | } | |
152 | ||
153 | /** | |
154 | * blkg_lookup - lookup blkg for the specified blkcg - q pair | |
155 | * @blkcg: blkcg of interest | |
156 | * @q: request_queue of interest | |
157 | * | |
158 | * Lookup blkg for the @blkcg - @q pair. This function should be called | |
159 | * under RCU read lock and is guaranteed to return %NULL if @q is bypassing | |
160 | * - see blk_queue_bypass_start() for details. | |
161 | */ | |
3c798398 | 162 | struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q) |
80fd9979 TH |
163 | { |
164 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
165 | ||
166 | if (unlikely(blk_queue_bypass(q))) | |
167 | return NULL; | |
168 | return __blkg_lookup(blkcg, q); | |
169 | } | |
170 | EXPORT_SYMBOL_GPL(blkg_lookup); | |
171 | ||
3c798398 TH |
172 | static struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg, |
173 | struct request_queue *q) | |
cd1604fa | 174 | __releases(q->queue_lock) __acquires(q->queue_lock) |
5624a4e4 | 175 | { |
3c798398 | 176 | struct blkcg_gq *blkg; |
496fb780 | 177 | int ret; |
5624a4e4 | 178 | |
cd1604fa TH |
179 | WARN_ON_ONCE(!rcu_read_lock_held()); |
180 | lockdep_assert_held(q->queue_lock); | |
181 | ||
80fd9979 | 182 | blkg = __blkg_lookup(blkcg, q); |
cd1604fa TH |
183 | if (blkg) |
184 | return blkg; | |
185 | ||
7ee9c562 | 186 | /* blkg holds a reference to blkcg */ |
cd1604fa TH |
187 | if (!css_tryget(&blkcg->css)) |
188 | return ERR_PTR(-EINVAL); | |
189 | ||
496fb780 TH |
190 | /* allocate */ |
191 | ret = -ENOMEM; | |
1cd9e039 | 192 | blkg = blkg_alloc(blkcg, q); |
496fb780 TH |
193 | if (unlikely(!blkg)) |
194 | goto err_put; | |
cd1604fa TH |
195 | |
196 | /* insert */ | |
197 | spin_lock(&blkcg->lock); | |
31e4c28d | 198 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); |
e8989fae | 199 | list_add(&blkg->q_node, &q->blkg_list); |
cd1604fa | 200 | spin_unlock(&blkcg->lock); |
cd1604fa | 201 | return blkg; |
496fb780 TH |
202 | |
203 | err_put: | |
204 | css_put(&blkcg->css); | |
205 | return ERR_PTR(ret); | |
31e4c28d | 206 | } |
3c96cb32 | 207 | |
3c798398 TH |
208 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
209 | struct request_queue *q) | |
3c96cb32 TH |
210 | { |
211 | /* | |
212 | * This could be the first entry point of blkcg implementation and | |
213 | * we shouldn't allow anything to go through for a bypassing queue. | |
214 | */ | |
215 | if (unlikely(blk_queue_bypass(q))) | |
216 | return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY); | |
217 | return __blkg_lookup_create(blkcg, q); | |
218 | } | |
cd1604fa | 219 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 220 | |
3c798398 | 221 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a TH |
222 | { |
223 | struct request_queue *q = blkg->q; | |
3c798398 | 224 | struct blkcg *blkcg = blkg->blkcg; |
03aa264a TH |
225 | |
226 | lockdep_assert_held(q->queue_lock); | |
9f13ef67 | 227 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
228 | |
229 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 230 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 231 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
e8989fae | 232 | list_del_init(&blkg->q_node); |
9f13ef67 | 233 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 234 | |
03aa264a TH |
235 | /* |
236 | * Put the reference taken at the time of creation so that when all | |
237 | * queues are gone, group can be destroyed. | |
238 | */ | |
239 | blkg_put(blkg); | |
240 | } | |
241 | ||
9f13ef67 TH |
242 | /** |
243 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
244 | * @q: request_queue of interest | |
9f13ef67 | 245 | * |
3c96cb32 | 246 | * Destroy all blkgs associated with @q. |
9f13ef67 | 247 | */ |
3c96cb32 | 248 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 249 | { |
3c798398 | 250 | struct blkcg_gq *blkg, *n; |
72e06c25 | 251 | |
6d18b008 | 252 | lockdep_assert_held(q->queue_lock); |
72e06c25 | 253 | |
9f13ef67 | 254 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 255 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 256 | |
9f13ef67 TH |
257 | spin_lock(&blkcg->lock); |
258 | blkg_destroy(blkg); | |
259 | spin_unlock(&blkcg->lock); | |
72e06c25 TH |
260 | } |
261 | } | |
262 | ||
1adaf3dd TH |
263 | static void blkg_rcu_free(struct rcu_head *rcu_head) |
264 | { | |
3c798398 | 265 | blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head)); |
1adaf3dd TH |
266 | } |
267 | ||
3c798398 | 268 | void __blkg_release(struct blkcg_gq *blkg) |
1adaf3dd TH |
269 | { |
270 | /* release the extra blkcg reference this blkg has been holding */ | |
271 | css_put(&blkg->blkcg->css); | |
272 | ||
273 | /* | |
274 | * A group is freed in rcu manner. But having an rcu lock does not | |
275 | * mean that one can access all the fields of blkg and assume these | |
276 | * are valid. For example, don't try to follow throtl_data and | |
277 | * request queue links. | |
278 | * | |
279 | * Having a reference to blkg under an rcu allows acess to only | |
280 | * values local to groups like group stats and group rate limits | |
281 | */ | |
282 | call_rcu(&blkg->rcu_head, blkg_rcu_free); | |
283 | } | |
284 | EXPORT_SYMBOL_GPL(__blkg_release); | |
285 | ||
3c798398 TH |
286 | static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, |
287 | u64 val) | |
303a3acb | 288 | { |
3c798398 TH |
289 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
290 | struct blkcg_gq *blkg; | |
303a3acb | 291 | struct hlist_node *n; |
bc0d6501 | 292 | int i; |
303a3acb | 293 | |
bc0d6501 | 294 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 295 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
296 | |
297 | /* | |
298 | * Note that stat reset is racy - it doesn't synchronize against | |
299 | * stat updates. This is a debug feature which shouldn't exist | |
300 | * anyway. If you get hit by a race, retry. | |
301 | */ | |
303a3acb | 302 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
8bd435b3 | 303 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 304 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 305 | |
a2b1693b | 306 | if (blkcg_policy_enabled(blkg->q, pol) && |
f9fcc2d3 TH |
307 | pol->pd_reset_stats_fn) |
308 | pol->pd_reset_stats_fn(blkg); | |
bc0d6501 | 309 | } |
303a3acb | 310 | } |
f0bdc8cd | 311 | |
303a3acb | 312 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 313 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
314 | return 0; |
315 | } | |
316 | ||
3c798398 | 317 | static const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 318 | { |
d3d32e69 TH |
319 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
320 | if (blkg->q->backing_dev_info.dev) | |
321 | return dev_name(blkg->q->backing_dev_info.dev); | |
322 | return NULL; | |
303a3acb DS |
323 | } |
324 | ||
d3d32e69 TH |
325 | /** |
326 | * blkcg_print_blkgs - helper for printing per-blkg data | |
327 | * @sf: seq_file to print to | |
328 | * @blkcg: blkcg of interest | |
329 | * @prfill: fill function to print out a blkg | |
330 | * @pol: policy in question | |
331 | * @data: data to be passed to @prfill | |
332 | * @show_total: to print out sum of prfill return values or not | |
333 | * | |
334 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
335 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
336 | * policy data and @data. If @show_total is %true, the sum of the return | |
337 | * values from @prfill is printed with "Total" label at the end. | |
338 | * | |
339 | * This is to be used to construct print functions for | |
340 | * cftype->read_seq_string method. | |
341 | */ | |
3c798398 | 342 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
343 | u64 (*prfill)(struct seq_file *, |
344 | struct blkg_policy_data *, int), | |
3c798398 | 345 | const struct blkcg_policy *pol, int data, |
ec399347 | 346 | bool show_total) |
5624a4e4 | 347 | { |
3c798398 | 348 | struct blkcg_gq *blkg; |
d3d32e69 TH |
349 | struct hlist_node *n; |
350 | u64 total = 0; | |
5624a4e4 | 351 | |
d3d32e69 TH |
352 | spin_lock_irq(&blkcg->lock); |
353 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) | |
a2b1693b | 354 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 355 | total += prfill(sf, blkg->pd[pol->plid], data); |
d3d32e69 TH |
356 | spin_unlock_irq(&blkcg->lock); |
357 | ||
358 | if (show_total) | |
359 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
360 | } | |
829fdb50 | 361 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
362 | |
363 | /** | |
364 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
365 | * @sf: seq_file to print to | |
f95a04af | 366 | * @pd: policy private data of interest |
d3d32e69 TH |
367 | * @v: value to print |
368 | * | |
f95a04af | 369 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 370 | */ |
f95a04af | 371 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 372 | { |
f95a04af | 373 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
374 | |
375 | if (!dname) | |
376 | return 0; | |
377 | ||
378 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
379 | return v; | |
380 | } | |
829fdb50 | 381 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
382 | |
383 | /** | |
384 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
385 | * @sf: seq_file to print to | |
f95a04af | 386 | * @pd: policy private data of interest |
d3d32e69 TH |
387 | * @rwstat: rwstat to print |
388 | * | |
f95a04af | 389 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 390 | */ |
f95a04af | 391 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 392 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
393 | { |
394 | static const char *rwstr[] = { | |
395 | [BLKG_RWSTAT_READ] = "Read", | |
396 | [BLKG_RWSTAT_WRITE] = "Write", | |
397 | [BLKG_RWSTAT_SYNC] = "Sync", | |
398 | [BLKG_RWSTAT_ASYNC] = "Async", | |
399 | }; | |
f95a04af | 400 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
401 | u64 v; |
402 | int i; | |
403 | ||
404 | if (!dname) | |
405 | return 0; | |
406 | ||
407 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
408 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
409 | (unsigned long long)rwstat->cnt[i]); | |
410 | ||
411 | v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE]; | |
412 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); | |
413 | return v; | |
414 | } | |
415 | ||
5bc4afb1 TH |
416 | /** |
417 | * blkg_prfill_stat - prfill callback for blkg_stat | |
418 | * @sf: seq_file to print to | |
f95a04af TH |
419 | * @pd: policy private data of interest |
420 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
421 | * |
422 | * prfill callback for printing a blkg_stat. | |
423 | */ | |
f95a04af | 424 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 425 | { |
f95a04af | 426 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 427 | } |
5bc4afb1 | 428 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 429 | |
5bc4afb1 TH |
430 | /** |
431 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
432 | * @sf: seq_file to print to | |
f95a04af TH |
433 | * @pd: policy private data of interest |
434 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
435 | * |
436 | * prfill callback for printing a blkg_rwstat. | |
437 | */ | |
f95a04af TH |
438 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
439 | int off) | |
d3d32e69 | 440 | { |
f95a04af | 441 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 442 | |
f95a04af | 443 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 444 | } |
5bc4afb1 | 445 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 446 | |
3a8b31d3 TH |
447 | /** |
448 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
449 | * @blkcg: target block cgroup | |
da8b0662 | 450 | * @pol: target policy |
3a8b31d3 TH |
451 | * @input: input string |
452 | * @ctx: blkg_conf_ctx to be filled | |
453 | * | |
454 | * Parse per-blkg config update from @input and initialize @ctx with the | |
455 | * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new | |
da8b0662 TH |
456 | * value. This function returns with RCU read lock and queue lock held and |
457 | * must be paired with blkg_conf_finish(). | |
3a8b31d3 | 458 | */ |
3c798398 TH |
459 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
460 | const char *input, struct blkg_conf_ctx *ctx) | |
da8b0662 | 461 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
34d0f179 | 462 | { |
3a8b31d3 | 463 | struct gendisk *disk; |
3c798398 | 464 | struct blkcg_gq *blkg; |
726fa694 TH |
465 | unsigned int major, minor; |
466 | unsigned long long v; | |
467 | int part, ret; | |
34d0f179 | 468 | |
726fa694 TH |
469 | if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3) |
470 | return -EINVAL; | |
3a8b31d3 | 471 | |
726fa694 | 472 | disk = get_gendisk(MKDEV(major, minor), &part); |
4bfd482e | 473 | if (!disk || part) |
726fa694 | 474 | return -EINVAL; |
e56da7e2 TH |
475 | |
476 | rcu_read_lock(); | |
4bfd482e | 477 | spin_lock_irq(disk->queue->queue_lock); |
da8b0662 | 478 | |
a2b1693b | 479 | if (blkcg_policy_enabled(disk->queue, pol)) |
3c96cb32 | 480 | blkg = blkg_lookup_create(blkcg, disk->queue); |
a2b1693b TH |
481 | else |
482 | blkg = ERR_PTR(-EINVAL); | |
e56da7e2 | 483 | |
4bfd482e TH |
484 | if (IS_ERR(blkg)) { |
485 | ret = PTR_ERR(blkg); | |
3a8b31d3 | 486 | rcu_read_unlock(); |
da8b0662 | 487 | spin_unlock_irq(disk->queue->queue_lock); |
3a8b31d3 TH |
488 | put_disk(disk); |
489 | /* | |
490 | * If queue was bypassing, we should retry. Do so after a | |
491 | * short msleep(). It isn't strictly necessary but queue | |
492 | * can be bypassing for some time and it's always nice to | |
493 | * avoid busy looping. | |
494 | */ | |
495 | if (ret == -EBUSY) { | |
496 | msleep(10); | |
497 | ret = restart_syscall(); | |
7702e8f4 | 498 | } |
726fa694 | 499 | return ret; |
062a644d | 500 | } |
3a8b31d3 TH |
501 | |
502 | ctx->disk = disk; | |
503 | ctx->blkg = blkg; | |
726fa694 TH |
504 | ctx->v = v; |
505 | return 0; | |
34d0f179 | 506 | } |
829fdb50 | 507 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 508 | |
3a8b31d3 TH |
509 | /** |
510 | * blkg_conf_finish - finish up per-blkg config update | |
511 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
512 | * | |
513 | * Finish up after per-blkg config update. This function must be paired | |
514 | * with blkg_conf_prep(). | |
515 | */ | |
829fdb50 | 516 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
da8b0662 | 517 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 518 | { |
da8b0662 | 519 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
3a8b31d3 TH |
520 | rcu_read_unlock(); |
521 | put_disk(ctx->disk); | |
34d0f179 | 522 | } |
829fdb50 | 523 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 524 | |
3c798398 | 525 | struct cftype blkcg_files[] = { |
84c124da DS |
526 | { |
527 | .name = "reset_stats", | |
3c798398 | 528 | .write_u64 = blkcg_reset_stats, |
22084190 | 529 | }, |
4baf6e33 | 530 | { } /* terminate */ |
31e4c28d VG |
531 | }; |
532 | ||
9f13ef67 | 533 | /** |
3c798398 | 534 | * blkcg_pre_destroy - cgroup pre_destroy callback |
9f13ef67 TH |
535 | * @cgroup: cgroup of interest |
536 | * | |
537 | * This function is called when @cgroup is about to go away and responsible | |
538 | * for shooting down all blkgs associated with @cgroup. blkgs should be | |
539 | * removed while holding both q and blkcg locks. As blkcg lock is nested | |
540 | * inside q lock, this function performs reverse double lock dancing. | |
541 | * | |
542 | * This is the blkcg counterpart of ioc_release_fn(). | |
543 | */ | |
3c798398 | 544 | static int blkcg_pre_destroy(struct cgroup *cgroup) |
31e4c28d | 545 | { |
3c798398 | 546 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
b1c35769 | 547 | |
9f13ef67 | 548 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 549 | |
9f13ef67 | 550 | while (!hlist_empty(&blkcg->blkg_list)) { |
3c798398 TH |
551 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, |
552 | struct blkcg_gq, blkcg_node); | |
c875f4d0 | 553 | struct request_queue *q = blkg->q; |
b1c35769 | 554 | |
9f13ef67 TH |
555 | if (spin_trylock(q->queue_lock)) { |
556 | blkg_destroy(blkg); | |
557 | spin_unlock(q->queue_lock); | |
558 | } else { | |
559 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 560 | cpu_relax(); |
a5567932 | 561 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 562 | } |
9f13ef67 | 563 | } |
b1c35769 | 564 | |
9f13ef67 | 565 | spin_unlock_irq(&blkcg->lock); |
7ee9c562 TH |
566 | return 0; |
567 | } | |
568 | ||
3c798398 | 569 | static void blkcg_destroy(struct cgroup *cgroup) |
7ee9c562 | 570 | { |
3c798398 | 571 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
7ee9c562 | 572 | |
3c798398 | 573 | if (blkcg != &blkcg_root) |
67523c48 | 574 | kfree(blkcg); |
31e4c28d VG |
575 | } |
576 | ||
3c798398 | 577 | static struct cgroup_subsys_state *blkcg_create(struct cgroup *cgroup) |
31e4c28d | 578 | { |
9a9e8a26 | 579 | static atomic64_t id_seq = ATOMIC64_INIT(0); |
3c798398 | 580 | struct blkcg *blkcg; |
0341509f | 581 | struct cgroup *parent = cgroup->parent; |
31e4c28d | 582 | |
0341509f | 583 | if (!parent) { |
3c798398 | 584 | blkcg = &blkcg_root; |
31e4c28d VG |
585 | goto done; |
586 | } | |
587 | ||
31e4c28d VG |
588 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
589 | if (!blkcg) | |
590 | return ERR_PTR(-ENOMEM); | |
591 | ||
3381cb8d | 592 | blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT; |
9a9e8a26 | 593 | blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */ |
31e4c28d VG |
594 | done: |
595 | spin_lock_init(&blkcg->lock); | |
596 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
597 | ||
598 | return &blkcg->css; | |
599 | } | |
600 | ||
5efd6113 TH |
601 | /** |
602 | * blkcg_init_queue - initialize blkcg part of request queue | |
603 | * @q: request_queue to initialize | |
604 | * | |
605 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
606 | * part of new request_queue @q. | |
607 | * | |
608 | * RETURNS: | |
609 | * 0 on success, -errno on failure. | |
610 | */ | |
611 | int blkcg_init_queue(struct request_queue *q) | |
612 | { | |
613 | might_sleep(); | |
614 | ||
3c96cb32 | 615 | return blk_throtl_init(q); |
5efd6113 TH |
616 | } |
617 | ||
618 | /** | |
619 | * blkcg_drain_queue - drain blkcg part of request_queue | |
620 | * @q: request_queue to drain | |
621 | * | |
622 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
623 | */ | |
624 | void blkcg_drain_queue(struct request_queue *q) | |
625 | { | |
626 | lockdep_assert_held(q->queue_lock); | |
627 | ||
628 | blk_throtl_drain(q); | |
629 | } | |
630 | ||
631 | /** | |
632 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
633 | * @q: request_queue being released | |
634 | * | |
635 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
636 | */ | |
637 | void blkcg_exit_queue(struct request_queue *q) | |
638 | { | |
6d18b008 | 639 | spin_lock_irq(q->queue_lock); |
3c96cb32 | 640 | blkg_destroy_all(q); |
6d18b008 TH |
641 | spin_unlock_irq(q->queue_lock); |
642 | ||
5efd6113 TH |
643 | blk_throtl_exit(q); |
644 | } | |
645 | ||
31e4c28d VG |
646 | /* |
647 | * We cannot support shared io contexts, as we have no mean to support | |
648 | * two tasks with the same ioc in two different groups without major rework | |
649 | * of the main cic data structures. For now we allow a task to change | |
650 | * its cgroup only if it's the only owner of its ioc. | |
651 | */ | |
3c798398 | 652 | static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
31e4c28d | 653 | { |
bb9d97b6 | 654 | struct task_struct *task; |
31e4c28d VG |
655 | struct io_context *ioc; |
656 | int ret = 0; | |
657 | ||
658 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
bb9d97b6 TH |
659 | cgroup_taskset_for_each(task, cgrp, tset) { |
660 | task_lock(task); | |
661 | ioc = task->io_context; | |
662 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
663 | ret = -EINVAL; | |
664 | task_unlock(task); | |
665 | if (ret) | |
666 | break; | |
667 | } | |
31e4c28d VG |
668 | return ret; |
669 | } | |
670 | ||
676f7c8f TH |
671 | struct cgroup_subsys blkio_subsys = { |
672 | .name = "blkio", | |
3c798398 TH |
673 | .create = blkcg_create, |
674 | .can_attach = blkcg_can_attach, | |
675 | .pre_destroy = blkcg_pre_destroy, | |
676 | .destroy = blkcg_destroy, | |
676f7c8f | 677 | .subsys_id = blkio_subsys_id, |
3c798398 | 678 | .base_cftypes = blkcg_files, |
676f7c8f TH |
679 | .module = THIS_MODULE, |
680 | }; | |
681 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
682 | ||
a2b1693b TH |
683 | /** |
684 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
685 | * @q: request_queue of interest | |
686 | * @pol: blkcg policy to activate | |
687 | * | |
688 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
689 | * bypass mode to populate its blkgs with policy_data for @pol. | |
690 | * | |
691 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
692 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
693 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
694 | * always enough for dereferencing policy data. | |
695 | * | |
696 | * The caller is responsible for synchronizing [de]activations and policy | |
697 | * [un]registerations. Returns 0 on success, -errno on failure. | |
698 | */ | |
699 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 700 | const struct blkcg_policy *pol) |
a2b1693b TH |
701 | { |
702 | LIST_HEAD(pds); | |
3c798398 | 703 | struct blkcg_gq *blkg; |
a2b1693b TH |
704 | struct blkg_policy_data *pd, *n; |
705 | int cnt = 0, ret; | |
706 | ||
707 | if (blkcg_policy_enabled(q, pol)) | |
708 | return 0; | |
709 | ||
710 | blk_queue_bypass_start(q); | |
711 | ||
712 | /* make sure the root blkg exists and count the existing blkgs */ | |
713 | spin_lock_irq(q->queue_lock); | |
714 | ||
715 | rcu_read_lock(); | |
3c798398 | 716 | blkg = __blkg_lookup_create(&blkcg_root, q); |
a2b1693b TH |
717 | rcu_read_unlock(); |
718 | ||
719 | if (IS_ERR(blkg)) { | |
720 | ret = PTR_ERR(blkg); | |
721 | goto out_unlock; | |
722 | } | |
723 | q->root_blkg = blkg; | |
724 | ||
725 | list_for_each_entry(blkg, &q->blkg_list, q_node) | |
726 | cnt++; | |
727 | ||
728 | spin_unlock_irq(q->queue_lock); | |
729 | ||
730 | /* allocate policy_data for all existing blkgs */ | |
731 | while (cnt--) { | |
f95a04af | 732 | pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node); |
a2b1693b TH |
733 | if (!pd) { |
734 | ret = -ENOMEM; | |
735 | goto out_free; | |
736 | } | |
737 | list_add_tail(&pd->alloc_node, &pds); | |
738 | } | |
739 | ||
740 | /* | |
741 | * Install the allocated pds. With @q bypassing, no new blkg | |
742 | * should have been created while the queue lock was dropped. | |
743 | */ | |
744 | spin_lock_irq(q->queue_lock); | |
745 | ||
746 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
747 | if (WARN_ON(list_empty(&pds))) { | |
748 | /* umm... this shouldn't happen, just abort */ | |
749 | ret = -ENOMEM; | |
750 | goto out_unlock; | |
751 | } | |
752 | pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node); | |
753 | list_del_init(&pd->alloc_node); | |
754 | ||
755 | /* grab blkcg lock too while installing @pd on @blkg */ | |
756 | spin_lock(&blkg->blkcg->lock); | |
757 | ||
758 | blkg->pd[pol->plid] = pd; | |
759 | pd->blkg = blkg; | |
f9fcc2d3 | 760 | pol->pd_init_fn(blkg); |
a2b1693b TH |
761 | |
762 | spin_unlock(&blkg->blkcg->lock); | |
763 | } | |
764 | ||
765 | __set_bit(pol->plid, q->blkcg_pols); | |
766 | ret = 0; | |
767 | out_unlock: | |
768 | spin_unlock_irq(q->queue_lock); | |
769 | out_free: | |
770 | blk_queue_bypass_end(q); | |
771 | list_for_each_entry_safe(pd, n, &pds, alloc_node) | |
772 | kfree(pd); | |
773 | return ret; | |
774 | } | |
775 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
776 | ||
777 | /** | |
778 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
779 | * @q: request_queue of interest | |
780 | * @pol: blkcg policy to deactivate | |
781 | * | |
782 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
783 | * blkcg_activate_policy(). | |
784 | */ | |
785 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 786 | const struct blkcg_policy *pol) |
a2b1693b | 787 | { |
3c798398 | 788 | struct blkcg_gq *blkg; |
a2b1693b TH |
789 | |
790 | if (!blkcg_policy_enabled(q, pol)) | |
791 | return; | |
792 | ||
793 | blk_queue_bypass_start(q); | |
794 | spin_lock_irq(q->queue_lock); | |
795 | ||
796 | __clear_bit(pol->plid, q->blkcg_pols); | |
797 | ||
6d18b008 TH |
798 | /* if no policy is left, no need for blkgs - shoot them down */ |
799 | if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS)) | |
800 | blkg_destroy_all(q); | |
801 | ||
a2b1693b TH |
802 | list_for_each_entry(blkg, &q->blkg_list, q_node) { |
803 | /* grab blkcg lock too while removing @pd from @blkg */ | |
804 | spin_lock(&blkg->blkcg->lock); | |
805 | ||
f9fcc2d3 TH |
806 | if (pol->pd_exit_fn) |
807 | pol->pd_exit_fn(blkg); | |
a2b1693b TH |
808 | |
809 | kfree(blkg->pd[pol->plid]); | |
810 | blkg->pd[pol->plid] = NULL; | |
811 | ||
812 | spin_unlock(&blkg->blkcg->lock); | |
813 | } | |
814 | ||
815 | spin_unlock_irq(q->queue_lock); | |
816 | blk_queue_bypass_end(q); | |
817 | } | |
818 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
819 | ||
8bd435b3 | 820 | /** |
3c798398 TH |
821 | * blkcg_policy_register - register a blkcg policy |
822 | * @pol: blkcg policy to register | |
8bd435b3 | 823 | * |
3c798398 TH |
824 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
825 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 826 | */ |
3c798398 | 827 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 828 | { |
8bd435b3 | 829 | int i, ret; |
e8989fae | 830 | |
f95a04af TH |
831 | if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data))) |
832 | return -EINVAL; | |
833 | ||
bc0d6501 TH |
834 | mutex_lock(&blkcg_pol_mutex); |
835 | ||
8bd435b3 TH |
836 | /* find an empty slot */ |
837 | ret = -ENOSPC; | |
838 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 839 | if (!blkcg_policy[i]) |
8bd435b3 TH |
840 | break; |
841 | if (i >= BLKCG_MAX_POLS) | |
842 | goto out_unlock; | |
035d10b2 | 843 | |
8bd435b3 | 844 | /* register and update blkgs */ |
3c798398 TH |
845 | pol->plid = i; |
846 | blkcg_policy[i] = pol; | |
8bd435b3 | 847 | |
8bd435b3 | 848 | /* everything is in place, add intf files for the new policy */ |
3c798398 TH |
849 | if (pol->cftypes) |
850 | WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes)); | |
8bd435b3 TH |
851 | ret = 0; |
852 | out_unlock: | |
bc0d6501 | 853 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 854 | return ret; |
3e252066 | 855 | } |
3c798398 | 856 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 857 | |
8bd435b3 | 858 | /** |
3c798398 TH |
859 | * blkcg_policy_unregister - unregister a blkcg policy |
860 | * @pol: blkcg policy to unregister | |
8bd435b3 | 861 | * |
3c798398 | 862 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 863 | */ |
3c798398 | 864 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 865 | { |
bc0d6501 TH |
866 | mutex_lock(&blkcg_pol_mutex); |
867 | ||
3c798398 | 868 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
869 | goto out_unlock; |
870 | ||
871 | /* kill the intf files first */ | |
3c798398 TH |
872 | if (pol->cftypes) |
873 | cgroup_rm_cftypes(&blkio_subsys, pol->cftypes); | |
44ea53de | 874 | |
8bd435b3 | 875 | /* unregister and update blkgs */ |
3c798398 | 876 | blkcg_policy[pol->plid] = NULL; |
8bd435b3 | 877 | out_unlock: |
bc0d6501 | 878 | mutex_unlock(&blkcg_pol_mutex); |
3e252066 | 879 | } |
3c798398 | 880 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |