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 | |
a2b1693b | 34 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 35 | const struct blkcg_policy *pol) |
a2b1693b TH |
36 | { |
37 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
38 | } | |
39 | ||
0381411e TH |
40 | /** |
41 | * blkg_free - free a blkg | |
42 | * @blkg: blkg to free | |
43 | * | |
44 | * Free @blkg which may be partially allocated. | |
45 | */ | |
3c798398 | 46 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 47 | { |
e8989fae | 48 | int i; |
549d3aa8 TH |
49 | |
50 | if (!blkg) | |
51 | return; | |
52 | ||
8bd435b3 | 53 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 54 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae TH |
55 | struct blkg_policy_data *pd = blkg->pd[i]; |
56 | ||
9ade5ea4 TH |
57 | if (!pd) |
58 | continue; | |
59 | ||
f9fcc2d3 TH |
60 | if (pol && pol->pd_exit_fn) |
61 | pol->pd_exit_fn(blkg); | |
9ade5ea4 | 62 | |
9ade5ea4 | 63 | kfree(pd); |
0381411e | 64 | } |
e8989fae | 65 | |
a051661c | 66 | blk_exit_rl(&blkg->rl); |
549d3aa8 | 67 | kfree(blkg); |
0381411e TH |
68 | } |
69 | ||
70 | /** | |
71 | * blkg_alloc - allocate a blkg | |
72 | * @blkcg: block cgroup the new blkg is associated with | |
73 | * @q: request_queue the new blkg is associated with | |
15974993 | 74 | * @gfp_mask: allocation mask to use |
0381411e | 75 | * |
e8989fae | 76 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 77 | */ |
15974993 TH |
78 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
79 | gfp_t gfp_mask) | |
0381411e | 80 | { |
3c798398 | 81 | struct blkcg_gq *blkg; |
e8989fae | 82 | int i; |
0381411e TH |
83 | |
84 | /* alloc and init base part */ | |
15974993 | 85 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
86 | if (!blkg) |
87 | return NULL; | |
88 | ||
c875f4d0 | 89 | blkg->q = q; |
e8989fae | 90 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 91 | blkg->blkcg = blkcg; |
1adaf3dd | 92 | blkg->refcnt = 1; |
0381411e | 93 | |
a051661c TH |
94 | /* root blkg uses @q->root_rl, init rl only for !root blkgs */ |
95 | if (blkcg != &blkcg_root) { | |
96 | if (blk_init_rl(&blkg->rl, q, gfp_mask)) | |
97 | goto err_free; | |
98 | blkg->rl.blkg = blkg; | |
99 | } | |
100 | ||
8bd435b3 | 101 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 102 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 103 | struct blkg_policy_data *pd; |
0381411e | 104 | |
a2b1693b | 105 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
106 | continue; |
107 | ||
108 | /* alloc per-policy data and attach it to blkg */ | |
15974993 | 109 | pd = kzalloc_node(pol->pd_size, gfp_mask, q->node); |
a051661c TH |
110 | if (!pd) |
111 | goto err_free; | |
549d3aa8 | 112 | |
e8989fae TH |
113 | blkg->pd[i] = pd; |
114 | pd->blkg = blkg; | |
e8989fae | 115 | |
9b2ea86b | 116 | /* invoke per-policy init */ |
356d2e58 | 117 | if (pol->pd_init_fn) |
f9fcc2d3 | 118 | pol->pd_init_fn(blkg); |
e8989fae TH |
119 | } |
120 | ||
0381411e | 121 | return blkg; |
a051661c TH |
122 | |
123 | err_free: | |
124 | blkg_free(blkg); | |
125 | return NULL; | |
0381411e TH |
126 | } |
127 | ||
3c798398 | 128 | static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, |
86cde6b6 | 129 | struct request_queue *q, bool update_hint) |
80fd9979 | 130 | { |
3c798398 | 131 | struct blkcg_gq *blkg; |
80fd9979 | 132 | |
a637120e TH |
133 | blkg = rcu_dereference(blkcg->blkg_hint); |
134 | if (blkg && blkg->q == q) | |
135 | return blkg; | |
136 | ||
137 | /* | |
86cde6b6 TH |
138 | * Hint didn't match. Look up from the radix tree. Note that the |
139 | * hint can only be updated under queue_lock as otherwise @blkg | |
140 | * could have already been removed from blkg_tree. The caller is | |
141 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
142 | */ |
143 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
144 | if (blkg && blkg->q == q) { |
145 | if (update_hint) { | |
146 | lockdep_assert_held(q->queue_lock); | |
147 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
148 | } | |
a637120e | 149 | return blkg; |
86cde6b6 | 150 | } |
a637120e | 151 | |
80fd9979 TH |
152 | return NULL; |
153 | } | |
154 | ||
155 | /** | |
156 | * blkg_lookup - lookup blkg for the specified blkcg - q pair | |
157 | * @blkcg: blkcg of interest | |
158 | * @q: request_queue of interest | |
159 | * | |
160 | * Lookup blkg for the @blkcg - @q pair. This function should be called | |
161 | * under RCU read lock and is guaranteed to return %NULL if @q is bypassing | |
162 | * - see blk_queue_bypass_start() for details. | |
163 | */ | |
3c798398 | 164 | struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q) |
80fd9979 TH |
165 | { |
166 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
167 | ||
168 | if (unlikely(blk_queue_bypass(q))) | |
169 | return NULL; | |
86cde6b6 | 170 | return __blkg_lookup(blkcg, q, false); |
80fd9979 TH |
171 | } |
172 | EXPORT_SYMBOL_GPL(blkg_lookup); | |
173 | ||
15974993 TH |
174 | /* |
175 | * If @new_blkg is %NULL, this function tries to allocate a new one as | |
176 | * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return. | |
177 | */ | |
86cde6b6 TH |
178 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
179 | struct request_queue *q, | |
180 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 181 | { |
3c798398 | 182 | struct blkcg_gq *blkg; |
496fb780 | 183 | int ret; |
5624a4e4 | 184 | |
cd1604fa TH |
185 | WARN_ON_ONCE(!rcu_read_lock_held()); |
186 | lockdep_assert_held(q->queue_lock); | |
187 | ||
7ee9c562 | 188 | /* blkg holds a reference to blkcg */ |
15974993 TH |
189 | if (!css_tryget(&blkcg->css)) { |
190 | blkg = ERR_PTR(-EINVAL); | |
191 | goto out_free; | |
192 | } | |
cd1604fa | 193 | |
496fb780 | 194 | /* allocate */ |
15974993 TH |
195 | if (!new_blkg) { |
196 | new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC); | |
197 | if (unlikely(!new_blkg)) { | |
198 | blkg = ERR_PTR(-ENOMEM); | |
199 | goto out_put; | |
200 | } | |
201 | } | |
202 | blkg = new_blkg; | |
cd1604fa TH |
203 | |
204 | /* insert */ | |
205 | spin_lock(&blkcg->lock); | |
a637120e TH |
206 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
207 | if (likely(!ret)) { | |
208 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
209 | list_add(&blkg->q_node, &q->blkg_list); | |
210 | } | |
cd1604fa | 211 | spin_unlock(&blkcg->lock); |
496fb780 | 212 | |
a637120e TH |
213 | if (!ret) |
214 | return blkg; | |
15974993 TH |
215 | |
216 | blkg = ERR_PTR(ret); | |
217 | out_put: | |
496fb780 | 218 | css_put(&blkcg->css); |
15974993 TH |
219 | out_free: |
220 | blkg_free(new_blkg); | |
221 | return blkg; | |
31e4c28d | 222 | } |
3c96cb32 | 223 | |
86cde6b6 TH |
224 | /** |
225 | * blkg_lookup_create - lookup blkg, try to create one if not there | |
226 | * @blkcg: blkcg of interest | |
227 | * @q: request_queue of interest | |
228 | * | |
229 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
230 | * create one. This function should be called under RCU read lock and | |
231 | * @q->queue_lock. | |
232 | * | |
233 | * Returns pointer to the looked up or created blkg on success, ERR_PTR() | |
234 | * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not | |
235 | * dead and bypassing, returns ERR_PTR(-EBUSY). | |
236 | */ | |
3c798398 TH |
237 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
238 | struct request_queue *q) | |
3c96cb32 | 239 | { |
86cde6b6 TH |
240 | struct blkcg_gq *blkg; |
241 | ||
242 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
243 | lockdep_assert_held(q->queue_lock); | |
244 | ||
3c96cb32 TH |
245 | /* |
246 | * This could be the first entry point of blkcg implementation and | |
247 | * we shouldn't allow anything to go through for a bypassing queue. | |
248 | */ | |
249 | if (unlikely(blk_queue_bypass(q))) | |
3f3299d5 | 250 | return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY); |
86cde6b6 TH |
251 | |
252 | blkg = __blkg_lookup(blkcg, q, true); | |
253 | if (blkg) | |
254 | return blkg; | |
255 | ||
256 | return blkg_create(blkcg, q, NULL); | |
3c96cb32 | 257 | } |
cd1604fa | 258 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 259 | |
3c798398 | 260 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 261 | { |
3c798398 | 262 | struct blkcg *blkcg = blkg->blkcg; |
03aa264a | 263 | |
27e1f9d1 | 264 | lockdep_assert_held(blkg->q->queue_lock); |
9f13ef67 | 265 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
266 | |
267 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 268 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 269 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e TH |
270 | |
271 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); | |
e8989fae | 272 | list_del_init(&blkg->q_node); |
9f13ef67 | 273 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 274 | |
a637120e TH |
275 | /* |
276 | * Both setting lookup hint to and clearing it from @blkg are done | |
277 | * under queue_lock. If it's not pointing to @blkg now, it never | |
278 | * will. Hint assignment itself can race safely. | |
279 | */ | |
280 | if (rcu_dereference_raw(blkcg->blkg_hint) == blkg) | |
281 | rcu_assign_pointer(blkcg->blkg_hint, NULL); | |
282 | ||
03aa264a TH |
283 | /* |
284 | * Put the reference taken at the time of creation so that when all | |
285 | * queues are gone, group can be destroyed. | |
286 | */ | |
287 | blkg_put(blkg); | |
288 | } | |
289 | ||
9f13ef67 TH |
290 | /** |
291 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
292 | * @q: request_queue of interest | |
9f13ef67 | 293 | * |
3c96cb32 | 294 | * Destroy all blkgs associated with @q. |
9f13ef67 | 295 | */ |
3c96cb32 | 296 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 297 | { |
3c798398 | 298 | struct blkcg_gq *blkg, *n; |
72e06c25 | 299 | |
6d18b008 | 300 | lockdep_assert_held(q->queue_lock); |
72e06c25 | 301 | |
9f13ef67 | 302 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 303 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 304 | |
9f13ef67 TH |
305 | spin_lock(&blkcg->lock); |
306 | blkg_destroy(blkg); | |
307 | spin_unlock(&blkcg->lock); | |
72e06c25 | 308 | } |
65635cbc JN |
309 | |
310 | /* | |
311 | * root blkg is destroyed. Just clear the pointer since | |
312 | * root_rl does not take reference on root blkg. | |
313 | */ | |
314 | q->root_blkg = NULL; | |
315 | q->root_rl.blkg = NULL; | |
72e06c25 TH |
316 | } |
317 | ||
1adaf3dd TH |
318 | static void blkg_rcu_free(struct rcu_head *rcu_head) |
319 | { | |
3c798398 | 320 | blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head)); |
1adaf3dd TH |
321 | } |
322 | ||
3c798398 | 323 | void __blkg_release(struct blkcg_gq *blkg) |
1adaf3dd TH |
324 | { |
325 | /* release the extra blkcg reference this blkg has been holding */ | |
326 | css_put(&blkg->blkcg->css); | |
327 | ||
328 | /* | |
329 | * A group is freed in rcu manner. But having an rcu lock does not | |
330 | * mean that one can access all the fields of blkg and assume these | |
331 | * are valid. For example, don't try to follow throtl_data and | |
332 | * request queue links. | |
333 | * | |
334 | * Having a reference to blkg under an rcu allows acess to only | |
335 | * values local to groups like group stats and group rate limits | |
336 | */ | |
337 | call_rcu(&blkg->rcu_head, blkg_rcu_free); | |
338 | } | |
339 | EXPORT_SYMBOL_GPL(__blkg_release); | |
340 | ||
a051661c TH |
341 | /* |
342 | * The next function used by blk_queue_for_each_rl(). It's a bit tricky | |
343 | * because the root blkg uses @q->root_rl instead of its own rl. | |
344 | */ | |
345 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
346 | struct request_queue *q) | |
347 | { | |
348 | struct list_head *ent; | |
349 | struct blkcg_gq *blkg; | |
350 | ||
351 | /* | |
352 | * Determine the current blkg list_head. The first entry is | |
353 | * root_rl which is off @q->blkg_list and mapped to the head. | |
354 | */ | |
355 | if (rl == &q->root_rl) { | |
356 | ent = &q->blkg_list; | |
65c77fd9 JN |
357 | /* There are no more block groups, hence no request lists */ |
358 | if (list_empty(ent)) | |
359 | return NULL; | |
a051661c TH |
360 | } else { |
361 | blkg = container_of(rl, struct blkcg_gq, rl); | |
362 | ent = &blkg->q_node; | |
363 | } | |
364 | ||
365 | /* walk to the next list_head, skip root blkcg */ | |
366 | ent = ent->next; | |
367 | if (ent == &q->root_blkg->q_node) | |
368 | ent = ent->next; | |
369 | if (ent == &q->blkg_list) | |
370 | return NULL; | |
371 | ||
372 | blkg = container_of(ent, struct blkcg_gq, q_node); | |
373 | return &blkg->rl; | |
374 | } | |
375 | ||
3c798398 TH |
376 | static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, |
377 | u64 val) | |
303a3acb | 378 | { |
3c798398 TH |
379 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
380 | struct blkcg_gq *blkg; | |
303a3acb | 381 | struct hlist_node *n; |
bc0d6501 | 382 | int i; |
303a3acb | 383 | |
bc0d6501 | 384 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 385 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
386 | |
387 | /* | |
388 | * Note that stat reset is racy - it doesn't synchronize against | |
389 | * stat updates. This is a debug feature which shouldn't exist | |
390 | * anyway. If you get hit by a race, retry. | |
391 | */ | |
303a3acb | 392 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
8bd435b3 | 393 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 394 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 395 | |
a2b1693b | 396 | if (blkcg_policy_enabled(blkg->q, pol) && |
f9fcc2d3 TH |
397 | pol->pd_reset_stats_fn) |
398 | pol->pd_reset_stats_fn(blkg); | |
bc0d6501 | 399 | } |
303a3acb | 400 | } |
f0bdc8cd | 401 | |
303a3acb | 402 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 403 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
404 | return 0; |
405 | } | |
406 | ||
3c798398 | 407 | static const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 408 | { |
d3d32e69 TH |
409 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
410 | if (blkg->q->backing_dev_info.dev) | |
411 | return dev_name(blkg->q->backing_dev_info.dev); | |
412 | return NULL; | |
303a3acb DS |
413 | } |
414 | ||
d3d32e69 TH |
415 | /** |
416 | * blkcg_print_blkgs - helper for printing per-blkg data | |
417 | * @sf: seq_file to print to | |
418 | * @blkcg: blkcg of interest | |
419 | * @prfill: fill function to print out a blkg | |
420 | * @pol: policy in question | |
421 | * @data: data to be passed to @prfill | |
422 | * @show_total: to print out sum of prfill return values or not | |
423 | * | |
424 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
425 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
426 | * policy data and @data. If @show_total is %true, the sum of the return | |
427 | * values from @prfill is printed with "Total" label at the end. | |
428 | * | |
429 | * This is to be used to construct print functions for | |
430 | * cftype->read_seq_string method. | |
431 | */ | |
3c798398 | 432 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
433 | u64 (*prfill)(struct seq_file *, |
434 | struct blkg_policy_data *, int), | |
3c798398 | 435 | const struct blkcg_policy *pol, int data, |
ec399347 | 436 | bool show_total) |
5624a4e4 | 437 | { |
3c798398 | 438 | struct blkcg_gq *blkg; |
d3d32e69 TH |
439 | struct hlist_node *n; |
440 | u64 total = 0; | |
5624a4e4 | 441 | |
d3d32e69 TH |
442 | spin_lock_irq(&blkcg->lock); |
443 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) | |
a2b1693b | 444 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 445 | total += prfill(sf, blkg->pd[pol->plid], data); |
d3d32e69 TH |
446 | spin_unlock_irq(&blkcg->lock); |
447 | ||
448 | if (show_total) | |
449 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
450 | } | |
829fdb50 | 451 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
452 | |
453 | /** | |
454 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
455 | * @sf: seq_file to print to | |
f95a04af | 456 | * @pd: policy private data of interest |
d3d32e69 TH |
457 | * @v: value to print |
458 | * | |
f95a04af | 459 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 460 | */ |
f95a04af | 461 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 462 | { |
f95a04af | 463 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
464 | |
465 | if (!dname) | |
466 | return 0; | |
467 | ||
468 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
469 | return v; | |
470 | } | |
829fdb50 | 471 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
472 | |
473 | /** | |
474 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
475 | * @sf: seq_file to print to | |
f95a04af | 476 | * @pd: policy private data of interest |
d3d32e69 TH |
477 | * @rwstat: rwstat to print |
478 | * | |
f95a04af | 479 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 480 | */ |
f95a04af | 481 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 482 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
483 | { |
484 | static const char *rwstr[] = { | |
485 | [BLKG_RWSTAT_READ] = "Read", | |
486 | [BLKG_RWSTAT_WRITE] = "Write", | |
487 | [BLKG_RWSTAT_SYNC] = "Sync", | |
488 | [BLKG_RWSTAT_ASYNC] = "Async", | |
489 | }; | |
f95a04af | 490 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
491 | u64 v; |
492 | int i; | |
493 | ||
494 | if (!dname) | |
495 | return 0; | |
496 | ||
497 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
498 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
499 | (unsigned long long)rwstat->cnt[i]); | |
500 | ||
501 | v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE]; | |
502 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); | |
503 | return v; | |
504 | } | |
505 | ||
5bc4afb1 TH |
506 | /** |
507 | * blkg_prfill_stat - prfill callback for blkg_stat | |
508 | * @sf: seq_file to print to | |
f95a04af TH |
509 | * @pd: policy private data of interest |
510 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
511 | * |
512 | * prfill callback for printing a blkg_stat. | |
513 | */ | |
f95a04af | 514 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 515 | { |
f95a04af | 516 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 517 | } |
5bc4afb1 | 518 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 519 | |
5bc4afb1 TH |
520 | /** |
521 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
522 | * @sf: seq_file to print to | |
f95a04af TH |
523 | * @pd: policy private data of interest |
524 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
525 | * |
526 | * prfill callback for printing a blkg_rwstat. | |
527 | */ | |
f95a04af TH |
528 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
529 | int off) | |
d3d32e69 | 530 | { |
f95a04af | 531 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 532 | |
f95a04af | 533 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 534 | } |
5bc4afb1 | 535 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 536 | |
3a8b31d3 TH |
537 | /** |
538 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
539 | * @blkcg: target block cgroup | |
da8b0662 | 540 | * @pol: target policy |
3a8b31d3 TH |
541 | * @input: input string |
542 | * @ctx: blkg_conf_ctx to be filled | |
543 | * | |
544 | * Parse per-blkg config update from @input and initialize @ctx with the | |
545 | * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new | |
da8b0662 TH |
546 | * value. This function returns with RCU read lock and queue lock held and |
547 | * must be paired with blkg_conf_finish(). | |
3a8b31d3 | 548 | */ |
3c798398 TH |
549 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
550 | const char *input, struct blkg_conf_ctx *ctx) | |
da8b0662 | 551 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
34d0f179 | 552 | { |
3a8b31d3 | 553 | struct gendisk *disk; |
3c798398 | 554 | struct blkcg_gq *blkg; |
726fa694 TH |
555 | unsigned int major, minor; |
556 | unsigned long long v; | |
557 | int part, ret; | |
34d0f179 | 558 | |
726fa694 TH |
559 | if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3) |
560 | return -EINVAL; | |
3a8b31d3 | 561 | |
726fa694 | 562 | disk = get_gendisk(MKDEV(major, minor), &part); |
4bfd482e | 563 | if (!disk || part) |
726fa694 | 564 | return -EINVAL; |
e56da7e2 TH |
565 | |
566 | rcu_read_lock(); | |
4bfd482e | 567 | spin_lock_irq(disk->queue->queue_lock); |
da8b0662 | 568 | |
a2b1693b | 569 | if (blkcg_policy_enabled(disk->queue, pol)) |
3c96cb32 | 570 | blkg = blkg_lookup_create(blkcg, disk->queue); |
a2b1693b TH |
571 | else |
572 | blkg = ERR_PTR(-EINVAL); | |
e56da7e2 | 573 | |
4bfd482e TH |
574 | if (IS_ERR(blkg)) { |
575 | ret = PTR_ERR(blkg); | |
3a8b31d3 | 576 | rcu_read_unlock(); |
da8b0662 | 577 | spin_unlock_irq(disk->queue->queue_lock); |
3a8b31d3 TH |
578 | put_disk(disk); |
579 | /* | |
580 | * If queue was bypassing, we should retry. Do so after a | |
581 | * short msleep(). It isn't strictly necessary but queue | |
582 | * can be bypassing for some time and it's always nice to | |
583 | * avoid busy looping. | |
584 | */ | |
585 | if (ret == -EBUSY) { | |
586 | msleep(10); | |
587 | ret = restart_syscall(); | |
7702e8f4 | 588 | } |
726fa694 | 589 | return ret; |
062a644d | 590 | } |
3a8b31d3 TH |
591 | |
592 | ctx->disk = disk; | |
593 | ctx->blkg = blkg; | |
726fa694 TH |
594 | ctx->v = v; |
595 | return 0; | |
34d0f179 | 596 | } |
829fdb50 | 597 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 598 | |
3a8b31d3 TH |
599 | /** |
600 | * blkg_conf_finish - finish up per-blkg config update | |
601 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
602 | * | |
603 | * Finish up after per-blkg config update. This function must be paired | |
604 | * with blkg_conf_prep(). | |
605 | */ | |
829fdb50 | 606 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
da8b0662 | 607 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 608 | { |
da8b0662 | 609 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
3a8b31d3 TH |
610 | rcu_read_unlock(); |
611 | put_disk(ctx->disk); | |
34d0f179 | 612 | } |
829fdb50 | 613 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 614 | |
3c798398 | 615 | struct cftype blkcg_files[] = { |
84c124da DS |
616 | { |
617 | .name = "reset_stats", | |
3c798398 | 618 | .write_u64 = blkcg_reset_stats, |
22084190 | 619 | }, |
4baf6e33 | 620 | { } /* terminate */ |
31e4c28d VG |
621 | }; |
622 | ||
9f13ef67 | 623 | /** |
92fb9748 | 624 | * blkcg_css_offline - cgroup css_offline callback |
9f13ef67 TH |
625 | * @cgroup: cgroup of interest |
626 | * | |
627 | * This function is called when @cgroup is about to go away and responsible | |
628 | * for shooting down all blkgs associated with @cgroup. blkgs should be | |
629 | * removed while holding both q and blkcg locks. As blkcg lock is nested | |
630 | * inside q lock, this function performs reverse double lock dancing. | |
631 | * | |
632 | * This is the blkcg counterpart of ioc_release_fn(). | |
633 | */ | |
92fb9748 | 634 | static void blkcg_css_offline(struct cgroup *cgroup) |
31e4c28d | 635 | { |
3c798398 | 636 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
b1c35769 | 637 | |
9f13ef67 | 638 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 639 | |
9f13ef67 | 640 | while (!hlist_empty(&blkcg->blkg_list)) { |
3c798398 TH |
641 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, |
642 | struct blkcg_gq, blkcg_node); | |
c875f4d0 | 643 | struct request_queue *q = blkg->q; |
b1c35769 | 644 | |
9f13ef67 TH |
645 | if (spin_trylock(q->queue_lock)) { |
646 | blkg_destroy(blkg); | |
647 | spin_unlock(q->queue_lock); | |
648 | } else { | |
649 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 650 | cpu_relax(); |
a5567932 | 651 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 652 | } |
9f13ef67 | 653 | } |
b1c35769 | 654 | |
9f13ef67 | 655 | spin_unlock_irq(&blkcg->lock); |
7ee9c562 TH |
656 | } |
657 | ||
92fb9748 | 658 | static void blkcg_css_free(struct cgroup *cgroup) |
7ee9c562 | 659 | { |
3c798398 | 660 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
7ee9c562 | 661 | |
3c798398 | 662 | if (blkcg != &blkcg_root) |
67523c48 | 663 | kfree(blkcg); |
31e4c28d VG |
664 | } |
665 | ||
92fb9748 | 666 | static struct cgroup_subsys_state *blkcg_css_alloc(struct cgroup *cgroup) |
31e4c28d | 667 | { |
9a9e8a26 | 668 | static atomic64_t id_seq = ATOMIC64_INIT(0); |
3c798398 | 669 | struct blkcg *blkcg; |
0341509f | 670 | struct cgroup *parent = cgroup->parent; |
31e4c28d | 671 | |
0341509f | 672 | if (!parent) { |
3c798398 | 673 | blkcg = &blkcg_root; |
31e4c28d VG |
674 | goto done; |
675 | } | |
676 | ||
31e4c28d VG |
677 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
678 | if (!blkcg) | |
679 | return ERR_PTR(-ENOMEM); | |
680 | ||
3381cb8d | 681 | blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT; |
9a9e8a26 | 682 | blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */ |
31e4c28d VG |
683 | done: |
684 | spin_lock_init(&blkcg->lock); | |
a637120e | 685 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC); |
31e4c28d VG |
686 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
687 | ||
688 | return &blkcg->css; | |
689 | } | |
690 | ||
5efd6113 TH |
691 | /** |
692 | * blkcg_init_queue - initialize blkcg part of request queue | |
693 | * @q: request_queue to initialize | |
694 | * | |
695 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
696 | * part of new request_queue @q. | |
697 | * | |
698 | * RETURNS: | |
699 | * 0 on success, -errno on failure. | |
700 | */ | |
701 | int blkcg_init_queue(struct request_queue *q) | |
702 | { | |
703 | might_sleep(); | |
704 | ||
3c96cb32 | 705 | return blk_throtl_init(q); |
5efd6113 TH |
706 | } |
707 | ||
708 | /** | |
709 | * blkcg_drain_queue - drain blkcg part of request_queue | |
710 | * @q: request_queue to drain | |
711 | * | |
712 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
713 | */ | |
714 | void blkcg_drain_queue(struct request_queue *q) | |
715 | { | |
716 | lockdep_assert_held(q->queue_lock); | |
717 | ||
718 | blk_throtl_drain(q); | |
719 | } | |
720 | ||
721 | /** | |
722 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
723 | * @q: request_queue being released | |
724 | * | |
725 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
726 | */ | |
727 | void blkcg_exit_queue(struct request_queue *q) | |
728 | { | |
6d18b008 | 729 | spin_lock_irq(q->queue_lock); |
3c96cb32 | 730 | blkg_destroy_all(q); |
6d18b008 TH |
731 | spin_unlock_irq(q->queue_lock); |
732 | ||
5efd6113 TH |
733 | blk_throtl_exit(q); |
734 | } | |
735 | ||
31e4c28d VG |
736 | /* |
737 | * We cannot support shared io contexts, as we have no mean to support | |
738 | * two tasks with the same ioc in two different groups without major rework | |
739 | * of the main cic data structures. For now we allow a task to change | |
740 | * its cgroup only if it's the only owner of its ioc. | |
741 | */ | |
3c798398 | 742 | static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
31e4c28d | 743 | { |
bb9d97b6 | 744 | struct task_struct *task; |
31e4c28d VG |
745 | struct io_context *ioc; |
746 | int ret = 0; | |
747 | ||
748 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
bb9d97b6 TH |
749 | cgroup_taskset_for_each(task, cgrp, tset) { |
750 | task_lock(task); | |
751 | ioc = task->io_context; | |
752 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
753 | ret = -EINVAL; | |
754 | task_unlock(task); | |
755 | if (ret) | |
756 | break; | |
757 | } | |
31e4c28d VG |
758 | return ret; |
759 | } | |
760 | ||
676f7c8f TH |
761 | struct cgroup_subsys blkio_subsys = { |
762 | .name = "blkio", | |
92fb9748 TH |
763 | .css_alloc = blkcg_css_alloc, |
764 | .css_offline = blkcg_css_offline, | |
765 | .css_free = blkcg_css_free, | |
3c798398 | 766 | .can_attach = blkcg_can_attach, |
676f7c8f | 767 | .subsys_id = blkio_subsys_id, |
3c798398 | 768 | .base_cftypes = blkcg_files, |
676f7c8f | 769 | .module = THIS_MODULE, |
8c7f6edb TH |
770 | |
771 | /* | |
772 | * blkio subsystem is utterly broken in terms of hierarchy support. | |
773 | * It treats all cgroups equally regardless of where they're | |
774 | * located in the hierarchy - all cgroups are treated as if they're | |
775 | * right below the root. Fix it and remove the following. | |
776 | */ | |
777 | .broken_hierarchy = true, | |
676f7c8f TH |
778 | }; |
779 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
780 | ||
a2b1693b TH |
781 | /** |
782 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
783 | * @q: request_queue of interest | |
784 | * @pol: blkcg policy to activate | |
785 | * | |
786 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
787 | * bypass mode to populate its blkgs with policy_data for @pol. | |
788 | * | |
789 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
790 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
791 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
792 | * always enough for dereferencing policy data. | |
793 | * | |
794 | * The caller is responsible for synchronizing [de]activations and policy | |
795 | * [un]registerations. Returns 0 on success, -errno on failure. | |
796 | */ | |
797 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 798 | const struct blkcg_policy *pol) |
a2b1693b TH |
799 | { |
800 | LIST_HEAD(pds); | |
86cde6b6 | 801 | struct blkcg_gq *blkg, *new_blkg; |
a2b1693b TH |
802 | struct blkg_policy_data *pd, *n; |
803 | int cnt = 0, ret; | |
15974993 | 804 | bool preloaded; |
a2b1693b TH |
805 | |
806 | if (blkcg_policy_enabled(q, pol)) | |
807 | return 0; | |
808 | ||
15974993 | 809 | /* preallocations for root blkg */ |
86cde6b6 TH |
810 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
811 | if (!new_blkg) | |
15974993 TH |
812 | return -ENOMEM; |
813 | ||
814 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
815 | ||
a2b1693b TH |
816 | blk_queue_bypass_start(q); |
817 | ||
86cde6b6 TH |
818 | /* |
819 | * Make sure the root blkg exists and count the existing blkgs. As | |
820 | * @q is bypassing at this point, blkg_lookup_create() can't be | |
821 | * used. Open code it. | |
822 | */ | |
a2b1693b TH |
823 | spin_lock_irq(q->queue_lock); |
824 | ||
825 | rcu_read_lock(); | |
86cde6b6 TH |
826 | blkg = __blkg_lookup(&blkcg_root, q, false); |
827 | if (blkg) | |
828 | blkg_free(new_blkg); | |
829 | else | |
830 | blkg = blkg_create(&blkcg_root, q, new_blkg); | |
a2b1693b TH |
831 | rcu_read_unlock(); |
832 | ||
15974993 TH |
833 | if (preloaded) |
834 | radix_tree_preload_end(); | |
835 | ||
a2b1693b TH |
836 | if (IS_ERR(blkg)) { |
837 | ret = PTR_ERR(blkg); | |
838 | goto out_unlock; | |
839 | } | |
840 | q->root_blkg = blkg; | |
a051661c | 841 | q->root_rl.blkg = blkg; |
a2b1693b TH |
842 | |
843 | list_for_each_entry(blkg, &q->blkg_list, q_node) | |
844 | cnt++; | |
845 | ||
846 | spin_unlock_irq(q->queue_lock); | |
847 | ||
848 | /* allocate policy_data for all existing blkgs */ | |
849 | while (cnt--) { | |
f95a04af | 850 | pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node); |
a2b1693b TH |
851 | if (!pd) { |
852 | ret = -ENOMEM; | |
853 | goto out_free; | |
854 | } | |
855 | list_add_tail(&pd->alloc_node, &pds); | |
856 | } | |
857 | ||
858 | /* | |
859 | * Install the allocated pds. With @q bypassing, no new blkg | |
860 | * should have been created while the queue lock was dropped. | |
861 | */ | |
862 | spin_lock_irq(q->queue_lock); | |
863 | ||
864 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
865 | if (WARN_ON(list_empty(&pds))) { | |
866 | /* umm... this shouldn't happen, just abort */ | |
867 | ret = -ENOMEM; | |
868 | goto out_unlock; | |
869 | } | |
870 | pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node); | |
871 | list_del_init(&pd->alloc_node); | |
872 | ||
873 | /* grab blkcg lock too while installing @pd on @blkg */ | |
874 | spin_lock(&blkg->blkcg->lock); | |
875 | ||
876 | blkg->pd[pol->plid] = pd; | |
877 | pd->blkg = blkg; | |
f9fcc2d3 | 878 | pol->pd_init_fn(blkg); |
a2b1693b TH |
879 | |
880 | spin_unlock(&blkg->blkcg->lock); | |
881 | } | |
882 | ||
883 | __set_bit(pol->plid, q->blkcg_pols); | |
884 | ret = 0; | |
885 | out_unlock: | |
886 | spin_unlock_irq(q->queue_lock); | |
887 | out_free: | |
888 | blk_queue_bypass_end(q); | |
889 | list_for_each_entry_safe(pd, n, &pds, alloc_node) | |
890 | kfree(pd); | |
891 | return ret; | |
892 | } | |
893 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
894 | ||
895 | /** | |
896 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
897 | * @q: request_queue of interest | |
898 | * @pol: blkcg policy to deactivate | |
899 | * | |
900 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
901 | * blkcg_activate_policy(). | |
902 | */ | |
903 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 904 | const struct blkcg_policy *pol) |
a2b1693b | 905 | { |
3c798398 | 906 | struct blkcg_gq *blkg; |
a2b1693b TH |
907 | |
908 | if (!blkcg_policy_enabled(q, pol)) | |
909 | return; | |
910 | ||
911 | blk_queue_bypass_start(q); | |
912 | spin_lock_irq(q->queue_lock); | |
913 | ||
914 | __clear_bit(pol->plid, q->blkcg_pols); | |
915 | ||
6d18b008 TH |
916 | /* if no policy is left, no need for blkgs - shoot them down */ |
917 | if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS)) | |
918 | blkg_destroy_all(q); | |
919 | ||
a2b1693b TH |
920 | list_for_each_entry(blkg, &q->blkg_list, q_node) { |
921 | /* grab blkcg lock too while removing @pd from @blkg */ | |
922 | spin_lock(&blkg->blkcg->lock); | |
923 | ||
f9fcc2d3 TH |
924 | if (pol->pd_exit_fn) |
925 | pol->pd_exit_fn(blkg); | |
a2b1693b TH |
926 | |
927 | kfree(blkg->pd[pol->plid]); | |
928 | blkg->pd[pol->plid] = NULL; | |
929 | ||
930 | spin_unlock(&blkg->blkcg->lock); | |
931 | } | |
932 | ||
933 | spin_unlock_irq(q->queue_lock); | |
934 | blk_queue_bypass_end(q); | |
935 | } | |
936 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
937 | ||
8bd435b3 | 938 | /** |
3c798398 TH |
939 | * blkcg_policy_register - register a blkcg policy |
940 | * @pol: blkcg policy to register | |
8bd435b3 | 941 | * |
3c798398 TH |
942 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
943 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 944 | */ |
3c798398 | 945 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 946 | { |
8bd435b3 | 947 | int i, ret; |
e8989fae | 948 | |
f95a04af TH |
949 | if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data))) |
950 | return -EINVAL; | |
951 | ||
bc0d6501 TH |
952 | mutex_lock(&blkcg_pol_mutex); |
953 | ||
8bd435b3 TH |
954 | /* find an empty slot */ |
955 | ret = -ENOSPC; | |
956 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 957 | if (!blkcg_policy[i]) |
8bd435b3 TH |
958 | break; |
959 | if (i >= BLKCG_MAX_POLS) | |
960 | goto out_unlock; | |
035d10b2 | 961 | |
8bd435b3 | 962 | /* register and update blkgs */ |
3c798398 TH |
963 | pol->plid = i; |
964 | blkcg_policy[i] = pol; | |
8bd435b3 | 965 | |
8bd435b3 | 966 | /* everything is in place, add intf files for the new policy */ |
3c798398 TH |
967 | if (pol->cftypes) |
968 | WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes)); | |
8bd435b3 TH |
969 | ret = 0; |
970 | out_unlock: | |
bc0d6501 | 971 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 972 | return ret; |
3e252066 | 973 | } |
3c798398 | 974 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 975 | |
8bd435b3 | 976 | /** |
3c798398 TH |
977 | * blkcg_policy_unregister - unregister a blkcg policy |
978 | * @pol: blkcg policy to unregister | |
8bd435b3 | 979 | * |
3c798398 | 980 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 981 | */ |
3c798398 | 982 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 983 | { |
bc0d6501 TH |
984 | mutex_lock(&blkcg_pol_mutex); |
985 | ||
3c798398 | 986 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
987 | goto out_unlock; |
988 | ||
989 | /* kill the intf files first */ | |
3c798398 TH |
990 | if (pol->cftypes) |
991 | cgroup_rm_cftypes(&blkio_subsys, pol->cftypes); | |
44ea53de | 992 | |
8bd435b3 | 993 | /* unregister and update blkgs */ |
3c798398 | 994 | blkcg_policy[pol->plid] = NULL; |
8bd435b3 | 995 | out_unlock: |
bc0d6501 | 996 | mutex_unlock(&blkcg_pol_mutex); |
3e252066 | 997 | } |
3c798398 | 998 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |