]>
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 | |
16b3de66 TH |
35 | static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, |
36 | struct request_queue *q, bool update_hint); | |
37 | ||
38 | /** | |
39 | * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants | |
40 | * @d_blkg: loop cursor pointing to the current descendant | |
41 | * @pos_cgrp: used for iteration | |
42 | * @p_blkg: target blkg to walk descendants of | |
43 | * | |
44 | * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU | |
45 | * read locked. If called under either blkcg or queue lock, the iteration | |
46 | * is guaranteed to include all and only online blkgs. The caller may | |
47 | * update @pos_cgrp by calling cgroup_rightmost_descendant() to skip | |
48 | * subtree. | |
49 | */ | |
50 | #define blkg_for_each_descendant_pre(d_blkg, pos_cgrp, p_blkg) \ | |
51 | cgroup_for_each_descendant_pre((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \ | |
52 | if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \ | |
53 | (p_blkg)->q, false))) | |
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 | |
a051661c | 87 | blk_exit_rl(&blkg->rl); |
549d3aa8 | 88 | kfree(blkg); |
0381411e TH |
89 | } |
90 | ||
91 | /** | |
92 | * blkg_alloc - allocate a blkg | |
93 | * @blkcg: block cgroup the new blkg is associated with | |
94 | * @q: request_queue the new blkg is associated with | |
15974993 | 95 | * @gfp_mask: allocation mask to use |
0381411e | 96 | * |
e8989fae | 97 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 98 | */ |
15974993 TH |
99 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
100 | gfp_t gfp_mask) | |
0381411e | 101 | { |
3c798398 | 102 | struct blkcg_gq *blkg; |
e8989fae | 103 | int i; |
0381411e TH |
104 | |
105 | /* alloc and init base part */ | |
15974993 | 106 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
107 | if (!blkg) |
108 | return NULL; | |
109 | ||
c875f4d0 | 110 | blkg->q = q; |
e8989fae | 111 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 112 | blkg->blkcg = blkcg; |
1adaf3dd | 113 | blkg->refcnt = 1; |
0381411e | 114 | |
a051661c TH |
115 | /* root blkg uses @q->root_rl, init rl only for !root blkgs */ |
116 | if (blkcg != &blkcg_root) { | |
117 | if (blk_init_rl(&blkg->rl, q, gfp_mask)) | |
118 | goto err_free; | |
119 | blkg->rl.blkg = blkg; | |
120 | } | |
121 | ||
8bd435b3 | 122 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 123 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 124 | struct blkg_policy_data *pd; |
0381411e | 125 | |
a2b1693b | 126 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
127 | continue; |
128 | ||
129 | /* alloc per-policy data and attach it to blkg */ | |
15974993 | 130 | pd = kzalloc_node(pol->pd_size, gfp_mask, q->node); |
a051661c TH |
131 | if (!pd) |
132 | goto err_free; | |
549d3aa8 | 133 | |
e8989fae TH |
134 | blkg->pd[i] = pd; |
135 | pd->blkg = blkg; | |
b276a876 | 136 | pd->plid = i; |
e8989fae | 137 | |
9b2ea86b | 138 | /* invoke per-policy init */ |
356d2e58 | 139 | if (pol->pd_init_fn) |
f9fcc2d3 | 140 | pol->pd_init_fn(blkg); |
e8989fae TH |
141 | } |
142 | ||
0381411e | 143 | return blkg; |
a051661c TH |
144 | |
145 | err_free: | |
146 | blkg_free(blkg); | |
147 | return NULL; | |
0381411e TH |
148 | } |
149 | ||
16b3de66 TH |
150 | /** |
151 | * __blkg_lookup - internal version of blkg_lookup() | |
152 | * @blkcg: blkcg of interest | |
153 | * @q: request_queue of interest | |
154 | * @update_hint: whether to update lookup hint with the result or not | |
155 | * | |
156 | * This is internal version and shouldn't be used by policy | |
157 | * implementations. Looks up blkgs for the @blkcg - @q pair regardless of | |
158 | * @q's bypass state. If @update_hint is %true, the caller should be | |
159 | * holding @q->queue_lock and lookup hint is updated on success. | |
160 | */ | |
3c798398 | 161 | static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, |
86cde6b6 | 162 | struct request_queue *q, bool update_hint) |
80fd9979 | 163 | { |
3c798398 | 164 | struct blkcg_gq *blkg; |
80fd9979 | 165 | |
a637120e TH |
166 | blkg = rcu_dereference(blkcg->blkg_hint); |
167 | if (blkg && blkg->q == q) | |
168 | return blkg; | |
169 | ||
170 | /* | |
86cde6b6 TH |
171 | * Hint didn't match. Look up from the radix tree. Note that the |
172 | * hint can only be updated under queue_lock as otherwise @blkg | |
173 | * could have already been removed from blkg_tree. The caller is | |
174 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
175 | */ |
176 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
177 | if (blkg && blkg->q == q) { |
178 | if (update_hint) { | |
179 | lockdep_assert_held(q->queue_lock); | |
180 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
181 | } | |
a637120e | 182 | return blkg; |
86cde6b6 | 183 | } |
a637120e | 184 | |
80fd9979 TH |
185 | return NULL; |
186 | } | |
187 | ||
188 | /** | |
189 | * blkg_lookup - lookup blkg for the specified blkcg - q pair | |
190 | * @blkcg: blkcg of interest | |
191 | * @q: request_queue of interest | |
192 | * | |
193 | * Lookup blkg for the @blkcg - @q pair. This function should be called | |
194 | * under RCU read lock and is guaranteed to return %NULL if @q is bypassing | |
195 | * - see blk_queue_bypass_start() for details. | |
196 | */ | |
3c798398 | 197 | struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q) |
80fd9979 TH |
198 | { |
199 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
200 | ||
201 | if (unlikely(blk_queue_bypass(q))) | |
202 | return NULL; | |
86cde6b6 | 203 | return __blkg_lookup(blkcg, q, false); |
80fd9979 TH |
204 | } |
205 | EXPORT_SYMBOL_GPL(blkg_lookup); | |
206 | ||
15974993 TH |
207 | /* |
208 | * If @new_blkg is %NULL, this function tries to allocate a new one as | |
209 | * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return. | |
210 | */ | |
86cde6b6 TH |
211 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
212 | struct request_queue *q, | |
213 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 214 | { |
3c798398 | 215 | struct blkcg_gq *blkg; |
f427d909 | 216 | int i, ret; |
5624a4e4 | 217 | |
cd1604fa TH |
218 | WARN_ON_ONCE(!rcu_read_lock_held()); |
219 | lockdep_assert_held(q->queue_lock); | |
220 | ||
7ee9c562 | 221 | /* blkg holds a reference to blkcg */ |
15974993 | 222 | if (!css_tryget(&blkcg->css)) { |
93e6d5d8 TH |
223 | ret = -EINVAL; |
224 | goto err_free_blkg; | |
15974993 | 225 | } |
cd1604fa | 226 | |
496fb780 | 227 | /* allocate */ |
15974993 TH |
228 | if (!new_blkg) { |
229 | new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC); | |
230 | if (unlikely(!new_blkg)) { | |
93e6d5d8 TH |
231 | ret = -ENOMEM; |
232 | goto err_put_css; | |
15974993 TH |
233 | } |
234 | } | |
235 | blkg = new_blkg; | |
cd1604fa | 236 | |
3c547865 TH |
237 | /* link parent and insert */ |
238 | if (blkcg_parent(blkcg)) { | |
239 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
240 | if (WARN_ON_ONCE(!blkg->parent)) { | |
241 | blkg = ERR_PTR(-EINVAL); | |
242 | goto err_put_css; | |
243 | } | |
244 | blkg_get(blkg->parent); | |
245 | } | |
246 | ||
cd1604fa | 247 | spin_lock(&blkcg->lock); |
a637120e TH |
248 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
249 | if (likely(!ret)) { | |
250 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
251 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
252 | |
253 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
254 | struct blkcg_policy *pol = blkcg_policy[i]; | |
255 | ||
256 | if (blkg->pd[i] && pol->pd_online_fn) | |
257 | pol->pd_online_fn(blkg); | |
258 | } | |
a637120e | 259 | } |
f427d909 | 260 | blkg->online = true; |
cd1604fa | 261 | spin_unlock(&blkcg->lock); |
496fb780 | 262 | |
a637120e TH |
263 | if (!ret) |
264 | return blkg; | |
15974993 | 265 | |
3c547865 TH |
266 | /* @blkg failed fully initialized, use the usual release path */ |
267 | blkg_put(blkg); | |
268 | return ERR_PTR(ret); | |
269 | ||
93e6d5d8 | 270 | err_put_css: |
496fb780 | 271 | css_put(&blkcg->css); |
93e6d5d8 | 272 | err_free_blkg: |
15974993 | 273 | blkg_free(new_blkg); |
93e6d5d8 | 274 | return ERR_PTR(ret); |
31e4c28d | 275 | } |
3c96cb32 | 276 | |
86cde6b6 TH |
277 | /** |
278 | * blkg_lookup_create - lookup blkg, try to create one if not there | |
279 | * @blkcg: blkcg of interest | |
280 | * @q: request_queue of interest | |
281 | * | |
282 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
283 | * create one. blkg creation is performed recursively from blkcg_root such |
284 | * that all non-root blkg's have access to the parent blkg. This function | |
285 | * should be called under RCU read lock and @q->queue_lock. | |
86cde6b6 TH |
286 | * |
287 | * Returns pointer to the looked up or created blkg on success, ERR_PTR() | |
288 | * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not | |
289 | * dead and bypassing, returns ERR_PTR(-EBUSY). | |
290 | */ | |
3c798398 TH |
291 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
292 | struct request_queue *q) | |
3c96cb32 | 293 | { |
86cde6b6 TH |
294 | struct blkcg_gq *blkg; |
295 | ||
296 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
297 | lockdep_assert_held(q->queue_lock); | |
298 | ||
3c96cb32 TH |
299 | /* |
300 | * This could be the first entry point of blkcg implementation and | |
301 | * we shouldn't allow anything to go through for a bypassing queue. | |
302 | */ | |
303 | if (unlikely(blk_queue_bypass(q))) | |
3f3299d5 | 304 | return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY); |
86cde6b6 TH |
305 | |
306 | blkg = __blkg_lookup(blkcg, q, true); | |
307 | if (blkg) | |
308 | return blkg; | |
309 | ||
3c547865 TH |
310 | /* |
311 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
312 | * non-root blkgs have access to their parents. | |
313 | */ | |
314 | while (true) { | |
315 | struct blkcg *pos = blkcg; | |
316 | struct blkcg *parent = blkcg_parent(blkcg); | |
317 | ||
318 | while (parent && !__blkg_lookup(parent, q, false)) { | |
319 | pos = parent; | |
320 | parent = blkcg_parent(parent); | |
321 | } | |
322 | ||
323 | blkg = blkg_create(pos, q, NULL); | |
324 | if (pos == blkcg || IS_ERR(blkg)) | |
325 | return blkg; | |
326 | } | |
3c96cb32 | 327 | } |
cd1604fa | 328 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 329 | |
3c798398 | 330 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 331 | { |
3c798398 | 332 | struct blkcg *blkcg = blkg->blkcg; |
f427d909 | 333 | int i; |
03aa264a | 334 | |
27e1f9d1 | 335 | lockdep_assert_held(blkg->q->queue_lock); |
9f13ef67 | 336 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
337 | |
338 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 339 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 340 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 341 | |
f427d909 TH |
342 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
343 | struct blkcg_policy *pol = blkcg_policy[i]; | |
344 | ||
345 | if (blkg->pd[i] && pol->pd_offline_fn) | |
346 | pol->pd_offline_fn(blkg); | |
347 | } | |
348 | blkg->online = false; | |
349 | ||
a637120e | 350 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 351 | list_del_init(&blkg->q_node); |
9f13ef67 | 352 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 353 | |
a637120e TH |
354 | /* |
355 | * Both setting lookup hint to and clearing it from @blkg are done | |
356 | * under queue_lock. If it's not pointing to @blkg now, it never | |
357 | * will. Hint assignment itself can race safely. | |
358 | */ | |
359 | if (rcu_dereference_raw(blkcg->blkg_hint) == blkg) | |
360 | rcu_assign_pointer(blkcg->blkg_hint, NULL); | |
361 | ||
03aa264a TH |
362 | /* |
363 | * Put the reference taken at the time of creation so that when all | |
364 | * queues are gone, group can be destroyed. | |
365 | */ | |
366 | blkg_put(blkg); | |
367 | } | |
368 | ||
9f13ef67 TH |
369 | /** |
370 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
371 | * @q: request_queue of interest | |
9f13ef67 | 372 | * |
3c96cb32 | 373 | * Destroy all blkgs associated with @q. |
9f13ef67 | 374 | */ |
3c96cb32 | 375 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 376 | { |
3c798398 | 377 | struct blkcg_gq *blkg, *n; |
72e06c25 | 378 | |
6d18b008 | 379 | lockdep_assert_held(q->queue_lock); |
72e06c25 | 380 | |
9f13ef67 | 381 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 382 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 383 | |
9f13ef67 TH |
384 | spin_lock(&blkcg->lock); |
385 | blkg_destroy(blkg); | |
386 | spin_unlock(&blkcg->lock); | |
72e06c25 | 387 | } |
65635cbc JN |
388 | |
389 | /* | |
390 | * root blkg is destroyed. Just clear the pointer since | |
391 | * root_rl does not take reference on root blkg. | |
392 | */ | |
393 | q->root_blkg = NULL; | |
394 | q->root_rl.blkg = NULL; | |
72e06c25 TH |
395 | } |
396 | ||
1adaf3dd TH |
397 | static void blkg_rcu_free(struct rcu_head *rcu_head) |
398 | { | |
3c798398 | 399 | blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head)); |
1adaf3dd TH |
400 | } |
401 | ||
3c798398 | 402 | void __blkg_release(struct blkcg_gq *blkg) |
1adaf3dd | 403 | { |
3c547865 | 404 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
1adaf3dd | 405 | css_put(&blkg->blkcg->css); |
3c547865 TH |
406 | if (blkg->parent) |
407 | blkg_put(blkg->parent); | |
1adaf3dd TH |
408 | |
409 | /* | |
410 | * A group is freed in rcu manner. But having an rcu lock does not | |
411 | * mean that one can access all the fields of blkg and assume these | |
412 | * are valid. For example, don't try to follow throtl_data and | |
413 | * request queue links. | |
414 | * | |
415 | * Having a reference to blkg under an rcu allows acess to only | |
416 | * values local to groups like group stats and group rate limits | |
417 | */ | |
418 | call_rcu(&blkg->rcu_head, blkg_rcu_free); | |
419 | } | |
420 | EXPORT_SYMBOL_GPL(__blkg_release); | |
421 | ||
a051661c TH |
422 | /* |
423 | * The next function used by blk_queue_for_each_rl(). It's a bit tricky | |
424 | * because the root blkg uses @q->root_rl instead of its own rl. | |
425 | */ | |
426 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
427 | struct request_queue *q) | |
428 | { | |
429 | struct list_head *ent; | |
430 | struct blkcg_gq *blkg; | |
431 | ||
432 | /* | |
433 | * Determine the current blkg list_head. The first entry is | |
434 | * root_rl which is off @q->blkg_list and mapped to the head. | |
435 | */ | |
436 | if (rl == &q->root_rl) { | |
437 | ent = &q->blkg_list; | |
65c77fd9 JN |
438 | /* There are no more block groups, hence no request lists */ |
439 | if (list_empty(ent)) | |
440 | return NULL; | |
a051661c TH |
441 | } else { |
442 | blkg = container_of(rl, struct blkcg_gq, rl); | |
443 | ent = &blkg->q_node; | |
444 | } | |
445 | ||
446 | /* walk to the next list_head, skip root blkcg */ | |
447 | ent = ent->next; | |
448 | if (ent == &q->root_blkg->q_node) | |
449 | ent = ent->next; | |
450 | if (ent == &q->blkg_list) | |
451 | return NULL; | |
452 | ||
453 | blkg = container_of(ent, struct blkcg_gq, q_node); | |
454 | return &blkg->rl; | |
455 | } | |
456 | ||
3c798398 TH |
457 | static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, |
458 | u64 val) | |
303a3acb | 459 | { |
3c798398 TH |
460 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
461 | struct blkcg_gq *blkg; | |
303a3acb | 462 | struct hlist_node *n; |
bc0d6501 | 463 | int i; |
303a3acb | 464 | |
bc0d6501 | 465 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 466 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
467 | |
468 | /* | |
469 | * Note that stat reset is racy - it doesn't synchronize against | |
470 | * stat updates. This is a debug feature which shouldn't exist | |
471 | * anyway. If you get hit by a race, retry. | |
472 | */ | |
303a3acb | 473 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
8bd435b3 | 474 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 475 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 476 | |
a2b1693b | 477 | if (blkcg_policy_enabled(blkg->q, pol) && |
f9fcc2d3 TH |
478 | pol->pd_reset_stats_fn) |
479 | pol->pd_reset_stats_fn(blkg); | |
bc0d6501 | 480 | } |
303a3acb | 481 | } |
f0bdc8cd | 482 | |
303a3acb | 483 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 484 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
485 | return 0; |
486 | } | |
487 | ||
3c798398 | 488 | static const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 489 | { |
d3d32e69 TH |
490 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
491 | if (blkg->q->backing_dev_info.dev) | |
492 | return dev_name(blkg->q->backing_dev_info.dev); | |
493 | return NULL; | |
303a3acb DS |
494 | } |
495 | ||
d3d32e69 TH |
496 | /** |
497 | * blkcg_print_blkgs - helper for printing per-blkg data | |
498 | * @sf: seq_file to print to | |
499 | * @blkcg: blkcg of interest | |
500 | * @prfill: fill function to print out a blkg | |
501 | * @pol: policy in question | |
502 | * @data: data to be passed to @prfill | |
503 | * @show_total: to print out sum of prfill return values or not | |
504 | * | |
505 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
506 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
507 | * policy data and @data. If @show_total is %true, the sum of the return | |
508 | * values from @prfill is printed with "Total" label at the end. | |
509 | * | |
510 | * This is to be used to construct print functions for | |
511 | * cftype->read_seq_string method. | |
512 | */ | |
3c798398 | 513 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
514 | u64 (*prfill)(struct seq_file *, |
515 | struct blkg_policy_data *, int), | |
3c798398 | 516 | const struct blkcg_policy *pol, int data, |
ec399347 | 517 | bool show_total) |
5624a4e4 | 518 | { |
3c798398 | 519 | struct blkcg_gq *blkg; |
d3d32e69 TH |
520 | struct hlist_node *n; |
521 | u64 total = 0; | |
5624a4e4 | 522 | |
d3d32e69 TH |
523 | spin_lock_irq(&blkcg->lock); |
524 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) | |
a2b1693b | 525 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 526 | total += prfill(sf, blkg->pd[pol->plid], data); |
d3d32e69 TH |
527 | spin_unlock_irq(&blkcg->lock); |
528 | ||
529 | if (show_total) | |
530 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
531 | } | |
829fdb50 | 532 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
533 | |
534 | /** | |
535 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
536 | * @sf: seq_file to print to | |
f95a04af | 537 | * @pd: policy private data of interest |
d3d32e69 TH |
538 | * @v: value to print |
539 | * | |
f95a04af | 540 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 541 | */ |
f95a04af | 542 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 543 | { |
f95a04af | 544 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
545 | |
546 | if (!dname) | |
547 | return 0; | |
548 | ||
549 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
550 | return v; | |
551 | } | |
829fdb50 | 552 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
553 | |
554 | /** | |
555 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
556 | * @sf: seq_file to print to | |
f95a04af | 557 | * @pd: policy private data of interest |
d3d32e69 TH |
558 | * @rwstat: rwstat to print |
559 | * | |
f95a04af | 560 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 561 | */ |
f95a04af | 562 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 563 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
564 | { |
565 | static const char *rwstr[] = { | |
566 | [BLKG_RWSTAT_READ] = "Read", | |
567 | [BLKG_RWSTAT_WRITE] = "Write", | |
568 | [BLKG_RWSTAT_SYNC] = "Sync", | |
569 | [BLKG_RWSTAT_ASYNC] = "Async", | |
570 | }; | |
f95a04af | 571 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
572 | u64 v; |
573 | int i; | |
574 | ||
575 | if (!dname) | |
576 | return 0; | |
577 | ||
578 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
579 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
580 | (unsigned long long)rwstat->cnt[i]); | |
581 | ||
582 | v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE]; | |
583 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); | |
584 | return v; | |
585 | } | |
b50da39f | 586 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 587 | |
5bc4afb1 TH |
588 | /** |
589 | * blkg_prfill_stat - prfill callback for blkg_stat | |
590 | * @sf: seq_file to print to | |
f95a04af TH |
591 | * @pd: policy private data of interest |
592 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
593 | * |
594 | * prfill callback for printing a blkg_stat. | |
595 | */ | |
f95a04af | 596 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 597 | { |
f95a04af | 598 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 599 | } |
5bc4afb1 | 600 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 601 | |
5bc4afb1 TH |
602 | /** |
603 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
604 | * @sf: seq_file to print to | |
f95a04af TH |
605 | * @pd: policy private data of interest |
606 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
607 | * |
608 | * prfill callback for printing a blkg_rwstat. | |
609 | */ | |
f95a04af TH |
610 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
611 | int off) | |
d3d32e69 | 612 | { |
f95a04af | 613 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 614 | |
f95a04af | 615 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 616 | } |
5bc4afb1 | 617 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 618 | |
16b3de66 TH |
619 | /** |
620 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
621 | * @pd: policy private data of interest | |
622 | * @off: offset to the blkg_stat in @pd | |
623 | * | |
624 | * Collect the blkg_stat specified by @off from @pd and all its online | |
625 | * descendants and return the sum. The caller must be holding the queue | |
626 | * lock for online tests. | |
627 | */ | |
628 | u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off) | |
629 | { | |
630 | struct blkcg_policy *pol = blkcg_policy[pd->plid]; | |
631 | struct blkcg_gq *pos_blkg; | |
632 | struct cgroup *pos_cgrp; | |
633 | u64 sum; | |
634 | ||
635 | lockdep_assert_held(pd->blkg->q->queue_lock); | |
636 | ||
637 | sum = blkg_stat_read((void *)pd + off); | |
638 | ||
639 | rcu_read_lock(); | |
640 | blkg_for_each_descendant_pre(pos_blkg, pos_cgrp, pd_to_blkg(pd)) { | |
641 | struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol); | |
642 | struct blkg_stat *stat = (void *)pos_pd + off; | |
643 | ||
644 | if (pos_blkg->online) | |
645 | sum += blkg_stat_read(stat); | |
646 | } | |
647 | rcu_read_unlock(); | |
648 | ||
649 | return sum; | |
650 | } | |
651 | EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum); | |
652 | ||
653 | /** | |
654 | * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat | |
655 | * @pd: policy private data of interest | |
656 | * @off: offset to the blkg_stat in @pd | |
657 | * | |
658 | * Collect the blkg_rwstat specified by @off from @pd and all its online | |
659 | * descendants and return the sum. The caller must be holding the queue | |
660 | * lock for online tests. | |
661 | */ | |
662 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd, | |
663 | int off) | |
664 | { | |
665 | struct blkcg_policy *pol = blkcg_policy[pd->plid]; | |
666 | struct blkcg_gq *pos_blkg; | |
667 | struct cgroup *pos_cgrp; | |
668 | struct blkg_rwstat sum; | |
669 | int i; | |
670 | ||
671 | lockdep_assert_held(pd->blkg->q->queue_lock); | |
672 | ||
673 | sum = blkg_rwstat_read((void *)pd + off); | |
674 | ||
675 | rcu_read_lock(); | |
676 | blkg_for_each_descendant_pre(pos_blkg, pos_cgrp, pd_to_blkg(pd)) { | |
677 | struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol); | |
678 | struct blkg_rwstat *rwstat = (void *)pos_pd + off; | |
679 | struct blkg_rwstat tmp; | |
680 | ||
681 | if (!pos_blkg->online) | |
682 | continue; | |
683 | ||
684 | tmp = blkg_rwstat_read(rwstat); | |
685 | ||
686 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
687 | sum.cnt[i] += tmp.cnt[i]; | |
688 | } | |
689 | rcu_read_unlock(); | |
690 | ||
691 | return sum; | |
692 | } | |
693 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
694 | ||
3a8b31d3 TH |
695 | /** |
696 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
697 | * @blkcg: target block cgroup | |
da8b0662 | 698 | * @pol: target policy |
3a8b31d3 TH |
699 | * @input: input string |
700 | * @ctx: blkg_conf_ctx to be filled | |
701 | * | |
702 | * Parse per-blkg config update from @input and initialize @ctx with the | |
703 | * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new | |
da8b0662 TH |
704 | * value. This function returns with RCU read lock and queue lock held and |
705 | * must be paired with blkg_conf_finish(). | |
3a8b31d3 | 706 | */ |
3c798398 TH |
707 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
708 | const char *input, struct blkg_conf_ctx *ctx) | |
da8b0662 | 709 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
34d0f179 | 710 | { |
3a8b31d3 | 711 | struct gendisk *disk; |
3c798398 | 712 | struct blkcg_gq *blkg; |
726fa694 TH |
713 | unsigned int major, minor; |
714 | unsigned long long v; | |
715 | int part, ret; | |
34d0f179 | 716 | |
726fa694 TH |
717 | if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3) |
718 | return -EINVAL; | |
3a8b31d3 | 719 | |
726fa694 | 720 | disk = get_gendisk(MKDEV(major, minor), &part); |
4bfd482e | 721 | if (!disk || part) |
726fa694 | 722 | return -EINVAL; |
e56da7e2 TH |
723 | |
724 | rcu_read_lock(); | |
4bfd482e | 725 | spin_lock_irq(disk->queue->queue_lock); |
da8b0662 | 726 | |
a2b1693b | 727 | if (blkcg_policy_enabled(disk->queue, pol)) |
3c96cb32 | 728 | blkg = blkg_lookup_create(blkcg, disk->queue); |
a2b1693b TH |
729 | else |
730 | blkg = ERR_PTR(-EINVAL); | |
e56da7e2 | 731 | |
4bfd482e TH |
732 | if (IS_ERR(blkg)) { |
733 | ret = PTR_ERR(blkg); | |
3a8b31d3 | 734 | rcu_read_unlock(); |
da8b0662 | 735 | spin_unlock_irq(disk->queue->queue_lock); |
3a8b31d3 TH |
736 | put_disk(disk); |
737 | /* | |
738 | * If queue was bypassing, we should retry. Do so after a | |
739 | * short msleep(). It isn't strictly necessary but queue | |
740 | * can be bypassing for some time and it's always nice to | |
741 | * avoid busy looping. | |
742 | */ | |
743 | if (ret == -EBUSY) { | |
744 | msleep(10); | |
745 | ret = restart_syscall(); | |
7702e8f4 | 746 | } |
726fa694 | 747 | return ret; |
062a644d | 748 | } |
3a8b31d3 TH |
749 | |
750 | ctx->disk = disk; | |
751 | ctx->blkg = blkg; | |
726fa694 TH |
752 | ctx->v = v; |
753 | return 0; | |
34d0f179 | 754 | } |
829fdb50 | 755 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 756 | |
3a8b31d3 TH |
757 | /** |
758 | * blkg_conf_finish - finish up per-blkg config update | |
759 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
760 | * | |
761 | * Finish up after per-blkg config update. This function must be paired | |
762 | * with blkg_conf_prep(). | |
763 | */ | |
829fdb50 | 764 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
da8b0662 | 765 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 766 | { |
da8b0662 | 767 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
3a8b31d3 TH |
768 | rcu_read_unlock(); |
769 | put_disk(ctx->disk); | |
34d0f179 | 770 | } |
829fdb50 | 771 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 772 | |
3c798398 | 773 | struct cftype blkcg_files[] = { |
84c124da DS |
774 | { |
775 | .name = "reset_stats", | |
3c798398 | 776 | .write_u64 = blkcg_reset_stats, |
22084190 | 777 | }, |
4baf6e33 | 778 | { } /* terminate */ |
31e4c28d VG |
779 | }; |
780 | ||
9f13ef67 | 781 | /** |
92fb9748 | 782 | * blkcg_css_offline - cgroup css_offline callback |
9f13ef67 TH |
783 | * @cgroup: cgroup of interest |
784 | * | |
785 | * This function is called when @cgroup is about to go away and responsible | |
786 | * for shooting down all blkgs associated with @cgroup. blkgs should be | |
787 | * removed while holding both q and blkcg locks. As blkcg lock is nested | |
788 | * inside q lock, this function performs reverse double lock dancing. | |
789 | * | |
790 | * This is the blkcg counterpart of ioc_release_fn(). | |
791 | */ | |
92fb9748 | 792 | static void blkcg_css_offline(struct cgroup *cgroup) |
31e4c28d | 793 | { |
3c798398 | 794 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
b1c35769 | 795 | |
9f13ef67 | 796 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 797 | |
9f13ef67 | 798 | while (!hlist_empty(&blkcg->blkg_list)) { |
3c798398 TH |
799 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, |
800 | struct blkcg_gq, blkcg_node); | |
c875f4d0 | 801 | struct request_queue *q = blkg->q; |
b1c35769 | 802 | |
9f13ef67 TH |
803 | if (spin_trylock(q->queue_lock)) { |
804 | blkg_destroy(blkg); | |
805 | spin_unlock(q->queue_lock); | |
806 | } else { | |
807 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 808 | cpu_relax(); |
a5567932 | 809 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 810 | } |
9f13ef67 | 811 | } |
b1c35769 | 812 | |
9f13ef67 | 813 | spin_unlock_irq(&blkcg->lock); |
7ee9c562 TH |
814 | } |
815 | ||
92fb9748 | 816 | static void blkcg_css_free(struct cgroup *cgroup) |
7ee9c562 | 817 | { |
3c798398 | 818 | struct blkcg *blkcg = cgroup_to_blkcg(cgroup); |
7ee9c562 | 819 | |
3c798398 | 820 | if (blkcg != &blkcg_root) |
67523c48 | 821 | kfree(blkcg); |
31e4c28d VG |
822 | } |
823 | ||
92fb9748 | 824 | static struct cgroup_subsys_state *blkcg_css_alloc(struct cgroup *cgroup) |
31e4c28d | 825 | { |
9a9e8a26 | 826 | static atomic64_t id_seq = ATOMIC64_INIT(0); |
3c798398 | 827 | struct blkcg *blkcg; |
0341509f | 828 | struct cgroup *parent = cgroup->parent; |
31e4c28d | 829 | |
0341509f | 830 | if (!parent) { |
3c798398 | 831 | blkcg = &blkcg_root; |
31e4c28d VG |
832 | goto done; |
833 | } | |
834 | ||
31e4c28d VG |
835 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
836 | if (!blkcg) | |
837 | return ERR_PTR(-ENOMEM); | |
838 | ||
3381cb8d | 839 | blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT; |
e71357e1 | 840 | blkcg->cfq_leaf_weight = CFQ_WEIGHT_DEFAULT; |
9a9e8a26 | 841 | blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */ |
31e4c28d VG |
842 | done: |
843 | spin_lock_init(&blkcg->lock); | |
a637120e | 844 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC); |
31e4c28d VG |
845 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
846 | ||
847 | return &blkcg->css; | |
848 | } | |
849 | ||
5efd6113 TH |
850 | /** |
851 | * blkcg_init_queue - initialize blkcg part of request queue | |
852 | * @q: request_queue to initialize | |
853 | * | |
854 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
855 | * part of new request_queue @q. | |
856 | * | |
857 | * RETURNS: | |
858 | * 0 on success, -errno on failure. | |
859 | */ | |
860 | int blkcg_init_queue(struct request_queue *q) | |
861 | { | |
862 | might_sleep(); | |
863 | ||
3c96cb32 | 864 | return blk_throtl_init(q); |
5efd6113 TH |
865 | } |
866 | ||
867 | /** | |
868 | * blkcg_drain_queue - drain blkcg part of request_queue | |
869 | * @q: request_queue to drain | |
870 | * | |
871 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
872 | */ | |
873 | void blkcg_drain_queue(struct request_queue *q) | |
874 | { | |
875 | lockdep_assert_held(q->queue_lock); | |
876 | ||
877 | blk_throtl_drain(q); | |
878 | } | |
879 | ||
880 | /** | |
881 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
882 | * @q: request_queue being released | |
883 | * | |
884 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
885 | */ | |
886 | void blkcg_exit_queue(struct request_queue *q) | |
887 | { | |
6d18b008 | 888 | spin_lock_irq(q->queue_lock); |
3c96cb32 | 889 | blkg_destroy_all(q); |
6d18b008 TH |
890 | spin_unlock_irq(q->queue_lock); |
891 | ||
5efd6113 TH |
892 | blk_throtl_exit(q); |
893 | } | |
894 | ||
31e4c28d VG |
895 | /* |
896 | * We cannot support shared io contexts, as we have no mean to support | |
897 | * two tasks with the same ioc in two different groups without major rework | |
898 | * of the main cic data structures. For now we allow a task to change | |
899 | * its cgroup only if it's the only owner of its ioc. | |
900 | */ | |
3c798398 | 901 | static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
31e4c28d | 902 | { |
bb9d97b6 | 903 | struct task_struct *task; |
31e4c28d VG |
904 | struct io_context *ioc; |
905 | int ret = 0; | |
906 | ||
907 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
bb9d97b6 TH |
908 | cgroup_taskset_for_each(task, cgrp, tset) { |
909 | task_lock(task); | |
910 | ioc = task->io_context; | |
911 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
912 | ret = -EINVAL; | |
913 | task_unlock(task); | |
914 | if (ret) | |
915 | break; | |
916 | } | |
31e4c28d VG |
917 | return ret; |
918 | } | |
919 | ||
676f7c8f TH |
920 | struct cgroup_subsys blkio_subsys = { |
921 | .name = "blkio", | |
92fb9748 TH |
922 | .css_alloc = blkcg_css_alloc, |
923 | .css_offline = blkcg_css_offline, | |
924 | .css_free = blkcg_css_free, | |
3c798398 | 925 | .can_attach = blkcg_can_attach, |
676f7c8f | 926 | .subsys_id = blkio_subsys_id, |
3c798398 | 927 | .base_cftypes = blkcg_files, |
676f7c8f | 928 | .module = THIS_MODULE, |
8c7f6edb TH |
929 | |
930 | /* | |
931 | * blkio subsystem is utterly broken in terms of hierarchy support. | |
932 | * It treats all cgroups equally regardless of where they're | |
933 | * located in the hierarchy - all cgroups are treated as if they're | |
934 | * right below the root. Fix it and remove the following. | |
935 | */ | |
936 | .broken_hierarchy = true, | |
676f7c8f TH |
937 | }; |
938 | EXPORT_SYMBOL_GPL(blkio_subsys); | |
939 | ||
a2b1693b TH |
940 | /** |
941 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
942 | * @q: request_queue of interest | |
943 | * @pol: blkcg policy to activate | |
944 | * | |
945 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
946 | * bypass mode to populate its blkgs with policy_data for @pol. | |
947 | * | |
948 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
949 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
950 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
951 | * always enough for dereferencing policy data. | |
952 | * | |
953 | * The caller is responsible for synchronizing [de]activations and policy | |
954 | * [un]registerations. Returns 0 on success, -errno on failure. | |
955 | */ | |
956 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 957 | const struct blkcg_policy *pol) |
a2b1693b TH |
958 | { |
959 | LIST_HEAD(pds); | |
86cde6b6 | 960 | struct blkcg_gq *blkg, *new_blkg; |
a2b1693b TH |
961 | struct blkg_policy_data *pd, *n; |
962 | int cnt = 0, ret; | |
15974993 | 963 | bool preloaded; |
a2b1693b TH |
964 | |
965 | if (blkcg_policy_enabled(q, pol)) | |
966 | return 0; | |
967 | ||
15974993 | 968 | /* preallocations for root blkg */ |
86cde6b6 TH |
969 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
970 | if (!new_blkg) | |
15974993 TH |
971 | return -ENOMEM; |
972 | ||
973 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
974 | ||
a2b1693b TH |
975 | blk_queue_bypass_start(q); |
976 | ||
86cde6b6 TH |
977 | /* |
978 | * Make sure the root blkg exists and count the existing blkgs. As | |
979 | * @q is bypassing at this point, blkg_lookup_create() can't be | |
980 | * used. Open code it. | |
981 | */ | |
a2b1693b TH |
982 | spin_lock_irq(q->queue_lock); |
983 | ||
984 | rcu_read_lock(); | |
86cde6b6 TH |
985 | blkg = __blkg_lookup(&blkcg_root, q, false); |
986 | if (blkg) | |
987 | blkg_free(new_blkg); | |
988 | else | |
989 | blkg = blkg_create(&blkcg_root, q, new_blkg); | |
a2b1693b TH |
990 | rcu_read_unlock(); |
991 | ||
15974993 TH |
992 | if (preloaded) |
993 | radix_tree_preload_end(); | |
994 | ||
a2b1693b TH |
995 | if (IS_ERR(blkg)) { |
996 | ret = PTR_ERR(blkg); | |
997 | goto out_unlock; | |
998 | } | |
999 | q->root_blkg = blkg; | |
a051661c | 1000 | q->root_rl.blkg = blkg; |
a2b1693b TH |
1001 | |
1002 | list_for_each_entry(blkg, &q->blkg_list, q_node) | |
1003 | cnt++; | |
1004 | ||
1005 | spin_unlock_irq(q->queue_lock); | |
1006 | ||
1007 | /* allocate policy_data for all existing blkgs */ | |
1008 | while (cnt--) { | |
f95a04af | 1009 | pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node); |
a2b1693b TH |
1010 | if (!pd) { |
1011 | ret = -ENOMEM; | |
1012 | goto out_free; | |
1013 | } | |
1014 | list_add_tail(&pd->alloc_node, &pds); | |
1015 | } | |
1016 | ||
1017 | /* | |
1018 | * Install the allocated pds. With @q bypassing, no new blkg | |
1019 | * should have been created while the queue lock was dropped. | |
1020 | */ | |
1021 | spin_lock_irq(q->queue_lock); | |
1022 | ||
1023 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1024 | if (WARN_ON(list_empty(&pds))) { | |
1025 | /* umm... this shouldn't happen, just abort */ | |
1026 | ret = -ENOMEM; | |
1027 | goto out_unlock; | |
1028 | } | |
1029 | pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node); | |
1030 | list_del_init(&pd->alloc_node); | |
1031 | ||
1032 | /* grab blkcg lock too while installing @pd on @blkg */ | |
1033 | spin_lock(&blkg->blkcg->lock); | |
1034 | ||
1035 | blkg->pd[pol->plid] = pd; | |
1036 | pd->blkg = blkg; | |
b276a876 | 1037 | pd->plid = pol->plid; |
f9fcc2d3 | 1038 | pol->pd_init_fn(blkg); |
a2b1693b TH |
1039 | |
1040 | spin_unlock(&blkg->blkcg->lock); | |
1041 | } | |
1042 | ||
1043 | __set_bit(pol->plid, q->blkcg_pols); | |
1044 | ret = 0; | |
1045 | out_unlock: | |
1046 | spin_unlock_irq(q->queue_lock); | |
1047 | out_free: | |
1048 | blk_queue_bypass_end(q); | |
1049 | list_for_each_entry_safe(pd, n, &pds, alloc_node) | |
1050 | kfree(pd); | |
1051 | return ret; | |
1052 | } | |
1053 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1054 | ||
1055 | /** | |
1056 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1057 | * @q: request_queue of interest | |
1058 | * @pol: blkcg policy to deactivate | |
1059 | * | |
1060 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1061 | * blkcg_activate_policy(). | |
1062 | */ | |
1063 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1064 | const struct blkcg_policy *pol) |
a2b1693b | 1065 | { |
3c798398 | 1066 | struct blkcg_gq *blkg; |
a2b1693b TH |
1067 | |
1068 | if (!blkcg_policy_enabled(q, pol)) | |
1069 | return; | |
1070 | ||
1071 | blk_queue_bypass_start(q); | |
1072 | spin_lock_irq(q->queue_lock); | |
1073 | ||
1074 | __clear_bit(pol->plid, q->blkcg_pols); | |
1075 | ||
6d18b008 TH |
1076 | /* if no policy is left, no need for blkgs - shoot them down */ |
1077 | if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS)) | |
1078 | blkg_destroy_all(q); | |
1079 | ||
a2b1693b TH |
1080 | list_for_each_entry(blkg, &q->blkg_list, q_node) { |
1081 | /* grab blkcg lock too while removing @pd from @blkg */ | |
1082 | spin_lock(&blkg->blkcg->lock); | |
1083 | ||
f427d909 TH |
1084 | if (pol->pd_offline_fn) |
1085 | pol->pd_offline_fn(blkg); | |
f9fcc2d3 TH |
1086 | if (pol->pd_exit_fn) |
1087 | pol->pd_exit_fn(blkg); | |
a2b1693b TH |
1088 | |
1089 | kfree(blkg->pd[pol->plid]); | |
1090 | blkg->pd[pol->plid] = NULL; | |
1091 | ||
1092 | spin_unlock(&blkg->blkcg->lock); | |
1093 | } | |
1094 | ||
1095 | spin_unlock_irq(q->queue_lock); | |
1096 | blk_queue_bypass_end(q); | |
1097 | } | |
1098 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1099 | ||
8bd435b3 | 1100 | /** |
3c798398 TH |
1101 | * blkcg_policy_register - register a blkcg policy |
1102 | * @pol: blkcg policy to register | |
8bd435b3 | 1103 | * |
3c798398 TH |
1104 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1105 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1106 | */ |
3c798398 | 1107 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1108 | { |
8bd435b3 | 1109 | int i, ret; |
e8989fae | 1110 | |
f95a04af TH |
1111 | if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data))) |
1112 | return -EINVAL; | |
1113 | ||
bc0d6501 TH |
1114 | mutex_lock(&blkcg_pol_mutex); |
1115 | ||
8bd435b3 TH |
1116 | /* find an empty slot */ |
1117 | ret = -ENOSPC; | |
1118 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1119 | if (!blkcg_policy[i]) |
8bd435b3 TH |
1120 | break; |
1121 | if (i >= BLKCG_MAX_POLS) | |
1122 | goto out_unlock; | |
035d10b2 | 1123 | |
8bd435b3 | 1124 | /* register and update blkgs */ |
3c798398 TH |
1125 | pol->plid = i; |
1126 | blkcg_policy[i] = pol; | |
8bd435b3 | 1127 | |
8bd435b3 | 1128 | /* everything is in place, add intf files for the new policy */ |
3c798398 TH |
1129 | if (pol->cftypes) |
1130 | WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes)); | |
8bd435b3 TH |
1131 | ret = 0; |
1132 | out_unlock: | |
bc0d6501 | 1133 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1134 | return ret; |
3e252066 | 1135 | } |
3c798398 | 1136 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1137 | |
8bd435b3 | 1138 | /** |
3c798398 TH |
1139 | * blkcg_policy_unregister - unregister a blkcg policy |
1140 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1141 | * |
3c798398 | 1142 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1143 | */ |
3c798398 | 1144 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1145 | { |
bc0d6501 TH |
1146 | mutex_lock(&blkcg_pol_mutex); |
1147 | ||
3c798398 | 1148 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1149 | goto out_unlock; |
1150 | ||
1151 | /* kill the intf files first */ | |
3c798398 TH |
1152 | if (pol->cftypes) |
1153 | cgroup_rm_cftypes(&blkio_subsys, pol->cftypes); | |
44ea53de | 1154 | |
8bd435b3 | 1155 | /* unregister and update blkgs */ |
3c798398 | 1156 | blkcg_policy[pol->plid] = NULL; |
8bd435b3 | 1157 | out_unlock: |
bc0d6501 | 1158 | mutex_unlock(&blkcg_pol_mutex); |
3e252066 | 1159 | } |
3c798398 | 1160 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |