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 | |
e71357e1 TH |
29 | struct blkcg blkcg_root = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT, |
30 | .cfq_leaf_weight = 2 * CFQ_WEIGHT_DEFAULT, }; | |
3c798398 | 31 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 32 | |
3c798398 | 33 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 34 | |
a2b1693b | 35 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 36 | const struct blkcg_policy *pol) |
a2b1693b TH |
37 | { |
38 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
39 | } | |
40 | ||
0381411e TH |
41 | /** |
42 | * blkg_free - free a blkg | |
43 | * @blkg: blkg to free | |
44 | * | |
45 | * Free @blkg which may be partially allocated. | |
46 | */ | |
3c798398 | 47 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 48 | { |
e8989fae | 49 | int i; |
549d3aa8 TH |
50 | |
51 | if (!blkg) | |
52 | return; | |
53 | ||
8bd435b3 | 54 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 55 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae TH |
56 | struct blkg_policy_data *pd = blkg->pd[i]; |
57 | ||
9ade5ea4 TH |
58 | if (!pd) |
59 | continue; | |
60 | ||
f9fcc2d3 TH |
61 | if (pol && pol->pd_exit_fn) |
62 | pol->pd_exit_fn(blkg); | |
9ade5ea4 | 63 | |
9ade5ea4 | 64 | kfree(pd); |
0381411e | 65 | } |
e8989fae | 66 | |
a051661c | 67 | blk_exit_rl(&blkg->rl); |
549d3aa8 | 68 | kfree(blkg); |
0381411e TH |
69 | } |
70 | ||
71 | /** | |
72 | * blkg_alloc - allocate a blkg | |
73 | * @blkcg: block cgroup the new blkg is associated with | |
74 | * @q: request_queue the new blkg is associated with | |
15974993 | 75 | * @gfp_mask: allocation mask to use |
0381411e | 76 | * |
e8989fae | 77 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 78 | */ |
15974993 TH |
79 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
80 | gfp_t gfp_mask) | |
0381411e | 81 | { |
3c798398 | 82 | struct blkcg_gq *blkg; |
e8989fae | 83 | int i; |
0381411e TH |
84 | |
85 | /* alloc and init base part */ | |
15974993 | 86 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
87 | if (!blkg) |
88 | return NULL; | |
89 | ||
c875f4d0 | 90 | blkg->q = q; |
e8989fae | 91 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 92 | blkg->blkcg = blkcg; |
1adaf3dd | 93 | blkg->refcnt = 1; |
0381411e | 94 | |
a051661c TH |
95 | /* root blkg uses @q->root_rl, init rl only for !root blkgs */ |
96 | if (blkcg != &blkcg_root) { | |
97 | if (blk_init_rl(&blkg->rl, q, gfp_mask)) | |
98 | goto err_free; | |
99 | blkg->rl.blkg = blkg; | |
100 | } | |
101 | ||
8bd435b3 | 102 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 103 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 104 | struct blkg_policy_data *pd; |
0381411e | 105 | |
a2b1693b | 106 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
107 | continue; |
108 | ||
109 | /* alloc per-policy data and attach it to blkg */ | |
15974993 | 110 | pd = kzalloc_node(pol->pd_size, gfp_mask, q->node); |
a051661c TH |
111 | if (!pd) |
112 | goto err_free; | |
549d3aa8 | 113 | |
e8989fae TH |
114 | blkg->pd[i] = pd; |
115 | pd->blkg = blkg; | |
b276a876 | 116 | pd->plid = i; |
e8989fae | 117 | |
9b2ea86b | 118 | /* invoke per-policy init */ |
356d2e58 | 119 | if (pol->pd_init_fn) |
f9fcc2d3 | 120 | pol->pd_init_fn(blkg); |
e8989fae TH |
121 | } |
122 | ||
0381411e | 123 | return blkg; |
a051661c TH |
124 | |
125 | err_free: | |
126 | blkg_free(blkg); | |
127 | return NULL; | |
0381411e TH |
128 | } |
129 | ||
3c798398 | 130 | static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, |
86cde6b6 | 131 | struct request_queue *q, bool update_hint) |
80fd9979 | 132 | { |
3c798398 | 133 | struct blkcg_gq *blkg; |
80fd9979 | 134 | |
a637120e TH |
135 | blkg = rcu_dereference(blkcg->blkg_hint); |
136 | if (blkg && blkg->q == q) | |
137 | return blkg; | |
138 | ||
139 | /* | |
86cde6b6 TH |
140 | * Hint didn't match. Look up from the radix tree. Note that the |
141 | * hint can only be updated under queue_lock as otherwise @blkg | |
142 | * could have already been removed from blkg_tree. The caller is | |
143 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
144 | */ |
145 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
146 | if (blkg && blkg->q == q) { |
147 | if (update_hint) { | |
148 | lockdep_assert_held(q->queue_lock); | |
149 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
150 | } | |
a637120e | 151 | return blkg; |
86cde6b6 | 152 | } |
a637120e | 153 | |
80fd9979 TH |
154 | return NULL; |
155 | } | |
156 | ||
157 | /** | |
158 | * blkg_lookup - lookup blkg for the specified blkcg - q pair | |
159 | * @blkcg: blkcg of interest | |
160 | * @q: request_queue of interest | |
161 | * | |
162 | * Lookup blkg for the @blkcg - @q pair. This function should be called | |
163 | * under RCU read lock and is guaranteed to return %NULL if @q is bypassing | |
164 | * - see blk_queue_bypass_start() for details. | |
165 | */ | |
3c798398 | 166 | struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q) |
80fd9979 TH |
167 | { |
168 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
169 | ||
170 | if (unlikely(blk_queue_bypass(q))) | |
171 | return NULL; | |
86cde6b6 | 172 | return __blkg_lookup(blkcg, q, false); |
80fd9979 TH |
173 | } |
174 | EXPORT_SYMBOL_GPL(blkg_lookup); | |
175 | ||
15974993 TH |
176 | /* |
177 | * If @new_blkg is %NULL, this function tries to allocate a new one as | |
178 | * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return. | |
179 | */ | |
86cde6b6 TH |
180 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
181 | struct request_queue *q, | |
182 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 183 | { |
3c798398 | 184 | struct blkcg_gq *blkg; |
f427d909 | 185 | int i, ret; |
5624a4e4 | 186 | |
cd1604fa TH |
187 | WARN_ON_ONCE(!rcu_read_lock_held()); |
188 | lockdep_assert_held(q->queue_lock); | |
189 | ||
7ee9c562 | 190 | /* blkg holds a reference to blkcg */ |
15974993 | 191 | if (!css_tryget(&blkcg->css)) { |
93e6d5d8 TH |
192 | ret = -EINVAL; |
193 | goto err_free_blkg; | |
15974993 | 194 | } |
cd1604fa | 195 | |
496fb780 | 196 | /* allocate */ |
15974993 TH |
197 | if (!new_blkg) { |
198 | new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC); | |
199 | if (unlikely(!new_blkg)) { | |
93e6d5d8 TH |
200 | ret = -ENOMEM; |
201 | goto err_put_css; | |
15974993 TH |
202 | } |
203 | } | |
204 | blkg = new_blkg; | |
cd1604fa | 205 | |
3c547865 TH |
206 | /* link parent and insert */ |
207 | if (blkcg_parent(blkcg)) { | |
208 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
209 | if (WARN_ON_ONCE(!blkg->parent)) { | |
210 | blkg = ERR_PTR(-EINVAL); | |
211 | goto err_put_css; | |
212 | } | |
213 | blkg_get(blkg->parent); | |
214 | } | |
215 | ||
cd1604fa | 216 | spin_lock(&blkcg->lock); |
a637120e TH |
217 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
218 | if (likely(!ret)) { | |
219 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
220 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
221 | |
222 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
223 | struct blkcg_policy *pol = blkcg_policy[i]; | |
224 | ||
225 | if (blkg->pd[i] && pol->pd_online_fn) | |
226 | pol->pd_online_fn(blkg); | |
227 | } | |
a637120e | 228 | } |
f427d909 | 229 | blkg->online = true; |
cd1604fa | 230 | spin_unlock(&blkcg->lock); |
496fb780 | 231 | |
a637120e TH |
232 | if (!ret) |
233 | return blkg; | |
15974993 | 234 | |
3c547865 TH |
235 | /* @blkg failed fully initialized, use the usual release path */ |
236 | blkg_put(blkg); | |
237 | return ERR_PTR(ret); | |
238 | ||
93e6d5d8 | 239 | err_put_css: |
496fb780 | 240 | css_put(&blkcg->css); |
93e6d5d8 | 241 | err_free_blkg: |
15974993 | 242 | blkg_free(new_blkg); |
93e6d5d8 | 243 | return ERR_PTR(ret); |
31e4c28d | 244 | } |
3c96cb32 | 245 | |
86cde6b6 TH |
246 | /** |
247 | * blkg_lookup_create - lookup blkg, try to create one if not there | |
248 | * @blkcg: blkcg of interest | |
249 | * @q: request_queue of interest | |
250 | * | |
251 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
252 | * create one. blkg creation is performed recursively from blkcg_root such |
253 | * that all non-root blkg's have access to the parent blkg. This function | |
254 | * should be called under RCU read lock and @q->queue_lock. | |
86cde6b6 TH |
255 | * |
256 | * Returns pointer to the looked up or created blkg on success, ERR_PTR() | |
257 | * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not | |
258 | * dead and bypassing, returns ERR_PTR(-EBUSY). | |
259 | */ | |
3c798398 TH |
260 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
261 | struct request_queue *q) | |
3c96cb32 | 262 | { |
86cde6b6 TH |
263 | struct blkcg_gq *blkg; |
264 | ||
265 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
266 | lockdep_assert_held(q->queue_lock); | |
267 | ||
3c96cb32 TH |
268 | /* |
269 | * This could be the first entry point of blkcg implementation and | |
270 | * we shouldn't allow anything to go through for a bypassing queue. | |
271 | */ | |
272 | if (unlikely(blk_queue_bypass(q))) | |
3f3299d5 | 273 | return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY); |
86cde6b6 TH |
274 | |
275 | blkg = __blkg_lookup(blkcg, q, true); | |
276 | if (blkg) | |
277 | return blkg; | |
278 | ||
3c547865 TH |
279 | /* |
280 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
281 | * non-root blkgs have access to their parents. | |
282 | */ | |
283 | while (true) { | |
284 | struct blkcg *pos = blkcg; | |
285 | struct blkcg *parent = blkcg_parent(blkcg); | |
286 | ||
287 | while (parent && !__blkg_lookup(parent, q, false)) { | |
288 | pos = parent; | |
289 | parent = blkcg_parent(parent); | |
290 | } | |
291 | ||
292 | blkg = blkg_create(pos, q, NULL); | |
293 | if (pos == blkcg || IS_ERR(blkg)) | |
294 | return blkg; | |
295 | } | |
3c96cb32 | 296 | } |
cd1604fa | 297 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 298 | |
3c798398 | 299 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 300 | { |
3c798398 | 301 | struct blkcg *blkcg = blkg->blkcg; |
f427d909 | 302 | int i; |
03aa264a | 303 | |
27e1f9d1 | 304 | lockdep_assert_held(blkg->q->queue_lock); |
9f13ef67 | 305 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
306 | |
307 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 308 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 309 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 310 | |
f427d909 TH |
311 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
312 | struct blkcg_policy *pol = blkcg_policy[i]; | |
313 | ||
314 | if (blkg->pd[i] && pol->pd_offline_fn) | |
315 | pol->pd_offline_fn(blkg); | |
316 | } | |
317 | blkg->online = false; | |
318 | ||
a637120e | 319 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 320 | list_del_init(&blkg->q_node); |
9f13ef67 | 321 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 322 | |
a637120e TH |
323 | /* |
324 | * Both setting lookup hint to and clearing it from @blkg are done | |
325 | * under queue_lock. If it's not pointing to @blkg now, it never | |
326 | * will. Hint assignment itself can race safely. | |
327 | */ | |
328 | if (rcu_dereference_raw(blkcg->blkg_hint) == blkg) | |
329 | rcu_assign_pointer(blkcg->blkg_hint, NULL); | |
330 | ||
03aa264a TH |
331 | /* |
332 | * Put the reference taken at the time of creation so that when all | |
333 | * queues are gone, group can be destroyed. | |
334 | */ | |
335 | blkg_put(blkg); | |
336 | } | |
337 | ||
9f13ef67 TH |
338 | /** |
339 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
340 | * @q: request_queue of interest | |
9f13ef67 | 341 | * |
3c96cb32 | 342 | * Destroy all blkgs associated with @q. |
9f13ef67 | 343 | */ |
3c96cb32 | 344 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 345 | { |
3c798398 | 346 | struct blkcg_gq *blkg, *n; |
72e06c25 | 347 | |
6d18b008 | 348 | lockdep_assert_held(q->queue_lock); |
72e06c25 | 349 | |
9f13ef67 | 350 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 351 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 352 | |
9f13ef67 TH |
353 | spin_lock(&blkcg->lock); |
354 | blkg_destroy(blkg); | |
355 | spin_unlock(&blkcg->lock); | |
72e06c25 | 356 | } |
65635cbc JN |
357 | |
358 | /* | |
359 | * root blkg is destroyed. Just clear the pointer since | |
360 | * root_rl does not take reference on root blkg. | |
361 | */ | |
362 | q->root_blkg = NULL; | |
363 | q->root_rl.blkg = NULL; | |
72e06c25 TH |
364 | } |
365 | ||
1adaf3dd TH |
366 | static void blkg_rcu_free(struct rcu_head *rcu_head) |
367 | { | |
3c798398 | 368 | blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head)); |
1adaf3dd TH |
369 | } |
370 | ||
3c798398 | 371 | void __blkg_release(struct blkcg_gq *blkg) |
1adaf3dd | 372 | { |
3c547865 | 373 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
1adaf3dd | 374 | css_put(&blkg->blkcg->css); |
3c547865 TH |
375 | if (blkg->parent) |
376 | blkg_put(blkg->parent); | |
1adaf3dd TH |
377 | |
378 | /* | |
379 | * A group is freed in rcu manner. But having an rcu lock does not | |
380 | * mean that one can access all the fields of blkg and assume these | |
381 | * are valid. For example, don't try to follow throtl_data and | |
382 | * request queue links. | |
383 | * | |
384 | * Having a reference to blkg under an rcu allows acess to only | |
385 | * values local to groups like group stats and group rate limits | |
386 | */ | |
387 | call_rcu(&blkg->rcu_head, blkg_rcu_free); | |
388 | } | |
389 | EXPORT_SYMBOL_GPL(__blkg_release); | |
390 | ||
a051661c TH |
391 | /* |
392 | * The next function used by blk_queue_for_each_rl(). It's a bit tricky | |
393 | * because the root blkg uses @q->root_rl instead of its own rl. | |
394 | */ | |
395 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
396 | struct request_queue *q) | |
397 | { | |
398 | struct list_head *ent; | |
399 | struct blkcg_gq *blkg; | |
400 | ||
401 | /* | |
402 | * Determine the current blkg list_head. The first entry is | |
403 | * root_rl which is off @q->blkg_list and mapped to the head. | |
404 | */ | |
405 | if (rl == &q->root_rl) { | |
406 | ent = &q->blkg_list; | |
65c77fd9 JN |
407 | /* There are no more block groups, hence no request lists */ |
408 | if (list_empty(ent)) | |
409 | return NULL; | |
a051661c TH |
410 | } else { |
411 | blkg = container_of(rl, struct blkcg_gq, rl); | |
412 | ent = &blkg->q_node; | |
413 | } | |
414 | ||
415 | /* walk to the next list_head, skip root blkcg */ | |
416 | ent = ent->next; | |
417 | if (ent == &q->root_blkg->q_node) | |
418 | ent = ent->next; | |
419 | if (ent == &q->blkg_list) | |
420 | return NULL; | |
421 | ||
422 | blkg = container_of(ent, struct blkcg_gq, q_node); | |
423 | return &blkg->rl; | |
424 | } | |
425 | ||
3c798398 TH |
426 | static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, |
427 | u64 val) | |
303a3acb | 428 | { |
3c798398 TH |
429 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
430 | struct blkcg_gq *blkg; | |
303a3acb | 431 | struct hlist_node *n; |
bc0d6501 | 432 | int i; |
303a3acb | 433 | |
bc0d6501 | 434 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 435 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
436 | |
437 | /* | |
438 | * Note that stat reset is racy - it doesn't synchronize against | |
439 | * stat updates. This is a debug feature which shouldn't exist | |
440 | * anyway. If you get hit by a race, retry. | |
441 | */ | |
303a3acb | 442 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
8bd435b3 | 443 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 444 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 445 | |
a2b1693b | 446 | if (blkcg_policy_enabled(blkg->q, pol) && |
f9fcc2d3 TH |
447 | pol->pd_reset_stats_fn) |
448 | pol->pd_reset_stats_fn(blkg); | |
bc0d6501 | 449 | } |
303a3acb | 450 | } |
f0bdc8cd | 451 | |
303a3acb | 452 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 453 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
454 | return 0; |
455 | } | |
456 | ||
3c798398 | 457 | static const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 458 | { |
d3d32e69 TH |
459 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
460 | if (blkg->q->backing_dev_info.dev) | |
461 | return dev_name(blkg->q->backing_dev_info.dev); | |
462 | return NULL; | |
303a3acb DS |
463 | } |
464 | ||
d3d32e69 TH |
465 | /** |
466 | * blkcg_print_blkgs - helper for printing per-blkg data | |
467 | * @sf: seq_file to print to | |
468 | * @blkcg: blkcg of interest | |
469 | * @prfill: fill function to print out a blkg | |
470 | * @pol: policy in question | |
471 | * @data: data to be passed to @prfill | |
472 | * @show_total: to print out sum of prfill return values or not | |
473 | * | |
474 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
475 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
476 | * policy data and @data. If @show_total is %true, the sum of the return | |
477 | * values from @prfill is printed with "Total" label at the end. | |
478 | * | |
479 | * This is to be used to construct print functions for | |
480 | * cftype->read_seq_string method. | |
481 | */ | |
3c798398 | 482 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
483 | u64 (*prfill)(struct seq_file *, |
484 | struct blkg_policy_data *, int), | |
3c798398 | 485 | const struct blkcg_policy *pol, int data, |
ec399347 | 486 | bool show_total) |
5624a4e4 | 487 | { |
3c798398 | 488 | struct blkcg_gq *blkg; |
d3d32e69 TH |
489 | struct hlist_node *n; |
490 | u64 total = 0; | |
5624a4e4 | 491 | |
d3d32e69 TH |
492 | spin_lock_irq(&blkcg->lock); |
493 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) | |
a2b1693b | 494 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 495 | total += prfill(sf, blkg->pd[pol->plid], data); |
d3d32e69 TH |
496 | spin_unlock_irq(&blkcg->lock); |
497 | ||
498 | if (show_total) | |
499 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
500 | } | |
829fdb50 | 501 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
502 | |
503 | /** | |
504 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
505 | * @sf: seq_file to print to | |
f95a04af | 506 | * @pd: policy private data of interest |
d3d32e69 TH |
507 | * @v: value to print |
508 | * | |
f95a04af | 509 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 510 | */ |
f95a04af | 511 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 512 | { |
f95a04af | 513 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
514 | |
515 | if (!dname) | |
516 | return 0; | |
517 | ||
518 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
519 | return v; | |
520 | } | |
829fdb50 | 521 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
522 | |
523 | /** | |
524 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
525 | * @sf: seq_file to print to | |
f95a04af | 526 | * @pd: policy private data of interest |
d3d32e69 TH |
527 | * @rwstat: rwstat to print |
528 | * | |
f95a04af | 529 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 530 | */ |
f95a04af | 531 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 532 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
533 | { |
534 | static const char *rwstr[] = { | |
535 | [BLKG_RWSTAT_READ] = "Read", | |
536 | [BLKG_RWSTAT_WRITE] = "Write", | |
537 | [BLKG_RWSTAT_SYNC] = "Sync", | |
538 | [BLKG_RWSTAT_ASYNC] = "Async", | |
539 | }; | |
f95a04af | 540 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
541 | u64 v; |
542 | int i; | |
543 | ||
544 | if (!dname) | |
545 | return 0; | |
546 | ||
547 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
548 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
549 | (unsigned long long)rwstat->cnt[i]); | |
550 | ||
551 | v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE]; | |
552 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); | |
553 | return v; | |
554 | } | |
b50da39f | 555 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 556 | |
5bc4afb1 TH |
557 | /** |
558 | * blkg_prfill_stat - prfill callback for blkg_stat | |
559 | * @sf: seq_file to print to | |
f95a04af TH |
560 | * @pd: policy private data of interest |
561 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
562 | * |
563 | * prfill callback for printing a blkg_stat. | |
564 | */ | |
f95a04af | 565 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 566 | { |
f95a04af | 567 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 568 | } |
5bc4afb1 | 569 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 570 | |
5bc4afb1 TH |
571 | /** |
572 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
573 | * @sf: seq_file to print to | |
f95a04af TH |
574 | * @pd: policy private data of interest |
575 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
576 | * |
577 | * prfill callback for printing a blkg_rwstat. | |
578 | */ | |
f95a04af TH |
579 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
580 | int off) | |
d3d32e69 | 581 | { |
f95a04af | 582 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 583 | |
f95a04af | 584 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 585 | } |
5bc4afb1 | 586 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 587 | |
3a8b31d3 TH |
588 | /** |
589 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
590 | * @blkcg: target block cgroup | |
da8b0662 | 591 | * @pol: target policy |
3a8b31d3 TH |
592 | * @input: input string |
593 | * @ctx: blkg_conf_ctx to be filled | |
594 | * | |
595 | * Parse per-blkg config update from @input and initialize @ctx with the | |
596 | * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new | |
da8b0662 TH |
597 | * value. This function returns with RCU read lock and queue lock held and |
598 | * must be paired with blkg_conf_finish(). | |
3a8b31d3 | 599 | */ |
3c798398 TH |
600 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
601 | const char *input, struct blkg_conf_ctx *ctx) | |
da8b0662 | 602 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
34d0f179 | 603 | { |
3a8b31d3 | 604 | struct gendisk *disk; |
3c798398 | 605 | struct blkcg_gq *blkg; |
726fa694 TH |
606 | unsigned int major, minor; |
607 | unsigned long long v; | |
608 | int part, ret; | |
34d0f179 | 609 | |
726fa694 TH |
610 | if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3) |
611 | return -EINVAL; | |
3a8b31d3 | 612 | |
726fa694 | 613 | disk = get_gendisk(MKDEV(major, minor), &part); |
4bfd482e | 614 | if (!disk || part) |
726fa694 | 615 | return -EINVAL; |
e56da7e2 TH |
616 | |
617 | rcu_read_lock(); | |
4bfd482e | 618 | spin_lock_irq(disk->queue->queue_lock); |
da8b0662 | 619 | |
a2b1693b | 620 | if (blkcg_policy_enabled(disk->queue, pol)) |
3c96cb32 | 621 | blkg = blkg_lookup_create(blkcg, disk->queue); |
a2b1693b TH |
622 | else |
623 | blkg = ERR_PTR(-EINVAL); | |
e56da7e2 | 624 | |
4bfd482e TH |
625 | if (IS_ERR(blkg)) { |
626 | ret = PTR_ERR(blkg); | |
3a8b31d3 | 627 | rcu_read_unlock(); |
da8b0662 | 628 | spin_unlock_irq(disk->queue->queue_lock); |
3a8b31d3 TH |
629 | put_disk(disk); |
630 | /* | |
631 | * If queue was bypassing, we should retry. Do so after a | |
632 | * short msleep(). It isn't strictly necessary but queue | |
633 | * can be bypassing for some time and it's always nice to | |
634 | * avoid busy looping. | |
635 | */ | |
636 | if (ret == -EBUSY) { | |
637 | msleep(10); | |
638 | ret = restart_syscall(); | |
7702e8f4 | 639 | } |
726fa694 | 640 | return ret; |
062a644d | 641 | } |
3a8b31d3 TH |
642 | |
643 | ctx->disk = disk; | |
644 | ctx->blkg = blkg; | |
726fa694 TH |
645 | ctx->v = v; |
646 | return 0; | |
34d0f179 | 647 | } |
829fdb50 | 648 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 649 | |
3a8b31d3 TH |
650 | /** |
651 | * blkg_conf_finish - finish up per-blkg config update | |
652 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
653 | * | |
654 | * Finish up after per-blkg config update. This function must be paired | |
655 | * with blkg_conf_prep(). | |
656 | */ | |
829fdb50 | 657 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
da8b0662 | 658 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 659 | { |
da8b0662 | 660 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
3a8b31d3 TH |
661 | rcu_read_unlock(); |
662 | put_disk(ctx->disk); | |
34d0f179 | 663 | } |
829fdb50 | 664 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 665 | |
3c798398 | 666 | struct cftype blkcg_files[] = { |
84c124da DS |
667 | { |
668 | .name = "reset_stats", | |
3c798398 | 669 | .write_u64 = blkcg_reset_stats, |
22084190 | 670 | }, |
4baf6e33 | 671 | { } /* terminate */ |
31e4c28d VG |
672 | }; |
673 | ||
9f13ef67 | 674 | /** |
92fb9748 | 675 | * blkcg_css_offline - cgroup css_offline callback |
9f13ef67 TH |
676 | * @cgroup: cgroup of interest |
677 | * | |
678 | * This function is called when @cgroup is about to go away and responsible | |
679 | * for shooting down all blkgs associated with @cgroup. blkgs should be | |
680 | * removed while holding both q and blkcg locks. As blkcg lock is nested | |
681 | * inside q lock, this function performs reverse double lock dancing. | |
682 | * | |
683 | * This is the blkcg counterpart of ioc_release_fn(). | |
684 | */ | |
92fb9748 | 685 | static void blkcg_css_offline(struct cgroup *cgroup) |
31e4c28d | 686 | { |
3c798398 | 687 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
b1c35769 | 688 | |
9f13ef67 | 689 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 690 | |
9f13ef67 | 691 | while (!hlist_empty(&blkcg->blkg_list)) { |
3c798398 TH |
692 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, |
693 | struct blkcg_gq, blkcg_node); | |
c875f4d0 | 694 | struct request_queue *q = blkg->q; |
b1c35769 | 695 | |
9f13ef67 TH |
696 | if (spin_trylock(q->queue_lock)) { |
697 | blkg_destroy(blkg); | |
698 | spin_unlock(q->queue_lock); | |
699 | } else { | |
700 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 701 | cpu_relax(); |
a5567932 | 702 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 703 | } |
9f13ef67 | 704 | } |
b1c35769 | 705 | |
9f13ef67 | 706 | spin_unlock_irq(&blkcg->lock); |
7ee9c562 TH |
707 | } |
708 | ||
92fb9748 | 709 | static void blkcg_css_free(struct cgroup *cgroup) |
7ee9c562 | 710 | { |
3c798398 | 711 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
7ee9c562 | 712 | |
3c798398 | 713 | if (blkcg != &blkcg_root) |
67523c48 | 714 | kfree(blkcg); |
31e4c28d VG |
715 | } |
716 | ||
92fb9748 | 717 | static struct cgroup_subsys_state *blkcg_css_alloc(struct cgroup *cgroup) |
31e4c28d | 718 | { |
9a9e8a26 | 719 | static atomic64_t id_seq = ATOMIC64_INIT(0); |
3c798398 | 720 | struct blkcg *blkcg; |
0341509f | 721 | struct cgroup *parent = cgroup->parent; |
31e4c28d | 722 | |
0341509f | 723 | if (!parent) { |
3c798398 | 724 | blkcg = &blkcg_root; |
31e4c28d VG |
725 | goto done; |
726 | } | |
727 | ||
31e4c28d VG |
728 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
729 | if (!blkcg) | |
730 | return ERR_PTR(-ENOMEM); | |
731 | ||
3381cb8d | 732 | blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT; |
e71357e1 | 733 | blkcg->cfq_leaf_weight = CFQ_WEIGHT_DEFAULT; |
9a9e8a26 | 734 | blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */ |
31e4c28d VG |
735 | done: |
736 | spin_lock_init(&blkcg->lock); | |
a637120e | 737 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC); |
31e4c28d VG |
738 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
739 | ||
740 | return &blkcg->css; | |
741 | } | |
742 | ||
5efd6113 TH |
743 | /** |
744 | * blkcg_init_queue - initialize blkcg part of request queue | |
745 | * @q: request_queue to initialize | |
746 | * | |
747 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
748 | * part of new request_queue @q. | |
749 | * | |
750 | * RETURNS: | |
751 | * 0 on success, -errno on failure. | |
752 | */ | |
753 | int blkcg_init_queue(struct request_queue *q) | |
754 | { | |
755 | might_sleep(); | |
756 | ||
3c96cb32 | 757 | return blk_throtl_init(q); |
5efd6113 TH |
758 | } |
759 | ||
760 | /** | |
761 | * blkcg_drain_queue - drain blkcg part of request_queue | |
762 | * @q: request_queue to drain | |
763 | * | |
764 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
765 | */ | |
766 | void blkcg_drain_queue(struct request_queue *q) | |
767 | { | |
768 | lockdep_assert_held(q->queue_lock); | |
769 | ||
770 | blk_throtl_drain(q); | |
771 | } | |
772 | ||
773 | /** | |
774 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
775 | * @q: request_queue being released | |
776 | * | |
777 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
778 | */ | |
779 | void blkcg_exit_queue(struct request_queue *q) | |
780 | { | |
6d18b008 | 781 | spin_lock_irq(q->queue_lock); |
3c96cb32 | 782 | blkg_destroy_all(q); |
6d18b008 TH |
783 | spin_unlock_irq(q->queue_lock); |
784 | ||
5efd6113 TH |
785 | blk_throtl_exit(q); |
786 | } | |
787 | ||
31e4c28d VG |
788 | /* |
789 | * We cannot support shared io contexts, as we have no mean to support | |
790 | * two tasks with the same ioc in two different groups without major rework | |
791 | * of the main cic data structures. For now we allow a task to change | |
792 | * its cgroup only if it's the only owner of its ioc. | |
793 | */ | |
3c798398 | 794 | static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
31e4c28d | 795 | { |
bb9d97b6 | 796 | struct task_struct *task; |
31e4c28d VG |
797 | struct io_context *ioc; |
798 | int ret = 0; | |
799 | ||
800 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
bb9d97b6 TH |
801 | cgroup_taskset_for_each(task, cgrp, tset) { |
802 | task_lock(task); | |
803 | ioc = task->io_context; | |
804 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
805 | ret = -EINVAL; | |
806 | task_unlock(task); | |
807 | if (ret) | |
808 | break; | |
809 | } | |
31e4c28d VG |
810 | return ret; |
811 | } | |
812 | ||
676f7c8f TH |
813 | struct cgroup_subsys blkio_subsys = { |
814 | .name = "blkio", | |
92fb9748 TH |
815 | .css_alloc = blkcg_css_alloc, |
816 | .css_offline = blkcg_css_offline, | |
817 | .css_free = blkcg_css_free, | |
3c798398 | 818 | .can_attach = blkcg_can_attach, |
676f7c8f | 819 | .subsys_id = blkio_subsys_id, |
3c798398 | 820 | .base_cftypes = blkcg_files, |
676f7c8f | 821 | .module = THIS_MODULE, |
8c7f6edb TH |
822 | |
823 | /* | |
824 | * blkio subsystem is utterly broken in terms of hierarchy support. | |
825 | * It treats all cgroups equally regardless of where they're | |
826 | * located in the hierarchy - all cgroups are treated as if they're | |
827 | * right below the root. Fix it and remove the following. | |
828 | */ | |
829 | .broken_hierarchy = true, | |
676f7c8f TH |
830 | }; |
831 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
832 | ||
a2b1693b TH |
833 | /** |
834 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
835 | * @q: request_queue of interest | |
836 | * @pol: blkcg policy to activate | |
837 | * | |
838 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
839 | * bypass mode to populate its blkgs with policy_data for @pol. | |
840 | * | |
841 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
842 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
843 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
844 | * always enough for dereferencing policy data. | |
845 | * | |
846 | * The caller is responsible for synchronizing [de]activations and policy | |
847 | * [un]registerations. Returns 0 on success, -errno on failure. | |
848 | */ | |
849 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 850 | const struct blkcg_policy *pol) |
a2b1693b TH |
851 | { |
852 | LIST_HEAD(pds); | |
86cde6b6 | 853 | struct blkcg_gq *blkg, *new_blkg; |
a2b1693b TH |
854 | struct blkg_policy_data *pd, *n; |
855 | int cnt = 0, ret; | |
15974993 | 856 | bool preloaded; |
a2b1693b TH |
857 | |
858 | if (blkcg_policy_enabled(q, pol)) | |
859 | return 0; | |
860 | ||
15974993 | 861 | /* preallocations for root blkg */ |
86cde6b6 TH |
862 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
863 | if (!new_blkg) | |
15974993 TH |
864 | return -ENOMEM; |
865 | ||
866 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
867 | ||
a2b1693b TH |
868 | blk_queue_bypass_start(q); |
869 | ||
86cde6b6 TH |
870 | /* |
871 | * Make sure the root blkg exists and count the existing blkgs. As | |
872 | * @q is bypassing at this point, blkg_lookup_create() can't be | |
873 | * used. Open code it. | |
874 | */ | |
a2b1693b TH |
875 | spin_lock_irq(q->queue_lock); |
876 | ||
877 | rcu_read_lock(); | |
86cde6b6 TH |
878 | blkg = __blkg_lookup(&blkcg_root, q, false); |
879 | if (blkg) | |
880 | blkg_free(new_blkg); | |
881 | else | |
882 | blkg = blkg_create(&blkcg_root, q, new_blkg); | |
a2b1693b TH |
883 | rcu_read_unlock(); |
884 | ||
15974993 TH |
885 | if (preloaded) |
886 | radix_tree_preload_end(); | |
887 | ||
a2b1693b TH |
888 | if (IS_ERR(blkg)) { |
889 | ret = PTR_ERR(blkg); | |
890 | goto out_unlock; | |
891 | } | |
892 | q->root_blkg = blkg; | |
a051661c | 893 | q->root_rl.blkg = blkg; |
a2b1693b TH |
894 | |
895 | list_for_each_entry(blkg, &q->blkg_list, q_node) | |
896 | cnt++; | |
897 | ||
898 | spin_unlock_irq(q->queue_lock); | |
899 | ||
900 | /* allocate policy_data for all existing blkgs */ | |
901 | while (cnt--) { | |
f95a04af | 902 | pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node); |
a2b1693b TH |
903 | if (!pd) { |
904 | ret = -ENOMEM; | |
905 | goto out_free; | |
906 | } | |
907 | list_add_tail(&pd->alloc_node, &pds); | |
908 | } | |
909 | ||
910 | /* | |
911 | * Install the allocated pds. With @q bypassing, no new blkg | |
912 | * should have been created while the queue lock was dropped. | |
913 | */ | |
914 | spin_lock_irq(q->queue_lock); | |
915 | ||
916 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
917 | if (WARN_ON(list_empty(&pds))) { | |
918 | /* umm... this shouldn't happen, just abort */ | |
919 | ret = -ENOMEM; | |
920 | goto out_unlock; | |
921 | } | |
922 | pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node); | |
923 | list_del_init(&pd->alloc_node); | |
924 | ||
925 | /* grab blkcg lock too while installing @pd on @blkg */ | |
926 | spin_lock(&blkg->blkcg->lock); | |
927 | ||
928 | blkg->pd[pol->plid] = pd; | |
929 | pd->blkg = blkg; | |
b276a876 | 930 | pd->plid = pol->plid; |
f9fcc2d3 | 931 | pol->pd_init_fn(blkg); |
a2b1693b TH |
932 | |
933 | spin_unlock(&blkg->blkcg->lock); | |
934 | } | |
935 | ||
936 | __set_bit(pol->plid, q->blkcg_pols); | |
937 | ret = 0; | |
938 | out_unlock: | |
939 | spin_unlock_irq(q->queue_lock); | |
940 | out_free: | |
941 | blk_queue_bypass_end(q); | |
942 | list_for_each_entry_safe(pd, n, &pds, alloc_node) | |
943 | kfree(pd); | |
944 | return ret; | |
945 | } | |
946 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
947 | ||
948 | /** | |
949 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
950 | * @q: request_queue of interest | |
951 | * @pol: blkcg policy to deactivate | |
952 | * | |
953 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
954 | * blkcg_activate_policy(). | |
955 | */ | |
956 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 957 | const struct blkcg_policy *pol) |
a2b1693b | 958 | { |
3c798398 | 959 | struct blkcg_gq *blkg; |
a2b1693b TH |
960 | |
961 | if (!blkcg_policy_enabled(q, pol)) | |
962 | return; | |
963 | ||
964 | blk_queue_bypass_start(q); | |
965 | spin_lock_irq(q->queue_lock); | |
966 | ||
967 | __clear_bit(pol->plid, q->blkcg_pols); | |
968 | ||
6d18b008 TH |
969 | /* if no policy is left, no need for blkgs - shoot them down */ |
970 | if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS)) | |
971 | blkg_destroy_all(q); | |
972 | ||
a2b1693b TH |
973 | list_for_each_entry(blkg, &q->blkg_list, q_node) { |
974 | /* grab blkcg lock too while removing @pd from @blkg */ | |
975 | spin_lock(&blkg->blkcg->lock); | |
976 | ||
f427d909 TH |
977 | if (pol->pd_offline_fn) |
978 | pol->pd_offline_fn(blkg); | |
f9fcc2d3 TH |
979 | if (pol->pd_exit_fn) |
980 | pol->pd_exit_fn(blkg); | |
a2b1693b TH |
981 | |
982 | kfree(blkg->pd[pol->plid]); | |
983 | blkg->pd[pol->plid] = NULL; | |
984 | ||
985 | spin_unlock(&blkg->blkcg->lock); | |
986 | } | |
987 | ||
988 | spin_unlock_irq(q->queue_lock); | |
989 | blk_queue_bypass_end(q); | |
990 | } | |
991 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
992 | ||
8bd435b3 | 993 | /** |
3c798398 TH |
994 | * blkcg_policy_register - register a blkcg policy |
995 | * @pol: blkcg policy to register | |
8bd435b3 | 996 | * |
3c798398 TH |
997 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
998 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 999 | */ |
3c798398 | 1000 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1001 | { |
8bd435b3 | 1002 | int i, ret; |
e8989fae | 1003 | |
f95a04af TH |
1004 | if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data))) |
1005 | return -EINVAL; | |
1006 | ||
bc0d6501 TH |
1007 | mutex_lock(&blkcg_pol_mutex); |
1008 | ||
8bd435b3 TH |
1009 | /* find an empty slot */ |
1010 | ret = -ENOSPC; | |
1011 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1012 | if (!blkcg_policy[i]) |
8bd435b3 TH |
1013 | break; |
1014 | if (i >= BLKCG_MAX_POLS) | |
1015 | goto out_unlock; | |
035d10b2 | 1016 | |
8bd435b3 | 1017 | /* register and update blkgs */ |
3c798398 TH |
1018 | pol->plid = i; |
1019 | blkcg_policy[i] = pol; | |
8bd435b3 | 1020 | |
8bd435b3 | 1021 | /* everything is in place, add intf files for the new policy */ |
3c798398 TH |
1022 | if (pol->cftypes) |
1023 | WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes)); | |
8bd435b3 TH |
1024 | ret = 0; |
1025 | out_unlock: | |
bc0d6501 | 1026 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1027 | return ret; |
3e252066 | 1028 | } |
3c798398 | 1029 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1030 | |
8bd435b3 | 1031 | /** |
3c798398 TH |
1032 | * blkcg_policy_unregister - unregister a blkcg policy |
1033 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1034 | * |
3c798398 | 1035 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1036 | */ |
3c798398 | 1037 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1038 | { |
bc0d6501 TH |
1039 | mutex_lock(&blkcg_pol_mutex); |
1040 | ||
3c798398 | 1041 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1042 | goto out_unlock; |
1043 | ||
1044 | /* kill the intf files first */ | |
3c798398 TH |
1045 | if (pol->cftypes) |
1046 | cgroup_rm_cftypes(&blkio_subsys, pol->cftypes); | |
44ea53de | 1047 | |
8bd435b3 | 1048 | /* unregister and update blkgs */ |
3c798398 | 1049 | blkcg_policy[pol->plid] = NULL; |
8bd435b3 | 1050 | out_unlock: |
bc0d6501 | 1051 | mutex_unlock(&blkcg_pol_mutex); |
3e252066 | 1052 | } |
3c798398 | 1053 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |