]>
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> | |
e48453c3 AA |
12 | * |
13 | * For policy-specific per-blkcg data: | |
14 | * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it> | |
15 | * Arianna Avanzini <avanzini.arianna@gmail.com> | |
31e4c28d VG |
16 | */ |
17 | #include <linux/ioprio.h> | |
22084190 | 18 | #include <linux/kdev_t.h> |
9d6a986c | 19 | #include <linux/module.h> |
accee785 | 20 | #include <linux/err.h> |
9195291e | 21 | #include <linux/blkdev.h> |
52ebea74 | 22 | #include <linux/backing-dev.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
34d0f179 | 24 | #include <linux/genhd.h> |
72e06c25 | 25 | #include <linux/delay.h> |
9a9e8a26 | 26 | #include <linux/atomic.h> |
eea8f41c | 27 | #include <linux/blk-cgroup.h> |
5efd6113 | 28 | #include "blk.h" |
3e252066 | 29 | |
84c124da DS |
30 | #define MAX_KEY_LEN 100 |
31 | ||
838f13bf TH |
32 | /* |
33 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
34 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
35 | * policy [un]register operations including cgroup file additions / | |
36 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
37 | * allows grabbing it from cgroup callbacks. | |
38 | */ | |
39 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 40 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 41 | |
e48453c3 | 42 | struct blkcg blkcg_root; |
3c798398 | 43 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 44 | |
496d5e75 TH |
45 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
46 | ||
3c798398 | 47 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 48 | |
a2b1693b | 49 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 50 | const struct blkcg_policy *pol) |
a2b1693b TH |
51 | { |
52 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
53 | } | |
54 | ||
0381411e TH |
55 | /** |
56 | * blkg_free - free a blkg | |
57 | * @blkg: blkg to free | |
58 | * | |
59 | * Free @blkg which may be partially allocated. | |
60 | */ | |
3c798398 | 61 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 62 | { |
e8989fae | 63 | int i; |
549d3aa8 TH |
64 | |
65 | if (!blkg) | |
66 | return; | |
67 | ||
db613670 TH |
68 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
69 | kfree(blkg->pd[i]); | |
e8989fae | 70 | |
a051661c | 71 | blk_exit_rl(&blkg->rl); |
549d3aa8 | 72 | kfree(blkg); |
0381411e TH |
73 | } |
74 | ||
75 | /** | |
76 | * blkg_alloc - allocate a blkg | |
77 | * @blkcg: block cgroup the new blkg is associated with | |
78 | * @q: request_queue the new blkg is associated with | |
15974993 | 79 | * @gfp_mask: allocation mask to use |
0381411e | 80 | * |
e8989fae | 81 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 82 | */ |
15974993 TH |
83 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
84 | gfp_t gfp_mask) | |
0381411e | 85 | { |
3c798398 | 86 | struct blkcg_gq *blkg; |
e8989fae | 87 | int i; |
0381411e TH |
88 | |
89 | /* alloc and init base part */ | |
15974993 | 90 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
91 | if (!blkg) |
92 | return NULL; | |
93 | ||
c875f4d0 | 94 | blkg->q = q; |
e8989fae | 95 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 96 | blkg->blkcg = blkcg; |
a5049a8a | 97 | atomic_set(&blkg->refcnt, 1); |
0381411e | 98 | |
a051661c TH |
99 | /* root blkg uses @q->root_rl, init rl only for !root blkgs */ |
100 | if (blkcg != &blkcg_root) { | |
101 | if (blk_init_rl(&blkg->rl, q, gfp_mask)) | |
102 | goto err_free; | |
103 | blkg->rl.blkg = blkg; | |
104 | } | |
105 | ||
8bd435b3 | 106 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 107 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 108 | struct blkg_policy_data *pd; |
0381411e | 109 | |
a2b1693b | 110 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
111 | continue; |
112 | ||
113 | /* alloc per-policy data and attach it to blkg */ | |
15974993 | 114 | pd = kzalloc_node(pol->pd_size, gfp_mask, q->node); |
a051661c TH |
115 | if (!pd) |
116 | goto err_free; | |
549d3aa8 | 117 | |
e8989fae TH |
118 | blkg->pd[i] = pd; |
119 | pd->blkg = blkg; | |
b276a876 | 120 | pd->plid = i; |
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 | ||
16b3de66 TH |
130 | /** |
131 | * __blkg_lookup - internal version of blkg_lookup() | |
132 | * @blkcg: blkcg of interest | |
133 | * @q: request_queue of interest | |
134 | * @update_hint: whether to update lookup hint with the result or not | |
135 | * | |
136 | * This is internal version and shouldn't be used by policy | |
137 | * implementations. Looks up blkgs for the @blkcg - @q pair regardless of | |
138 | * @q's bypass state. If @update_hint is %true, the caller should be | |
139 | * holding @q->queue_lock and lookup hint is updated on success. | |
140 | */ | |
dd4a4ffc TH |
141 | struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q, |
142 | bool update_hint) | |
80fd9979 | 143 | { |
3c798398 | 144 | struct blkcg_gq *blkg; |
80fd9979 | 145 | |
a637120e TH |
146 | blkg = rcu_dereference(blkcg->blkg_hint); |
147 | if (blkg && blkg->q == q) | |
148 | return blkg; | |
149 | ||
150 | /* | |
86cde6b6 TH |
151 | * Hint didn't match. Look up from the radix tree. Note that the |
152 | * hint can only be updated under queue_lock as otherwise @blkg | |
153 | * could have already been removed from blkg_tree. The caller is | |
154 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
155 | */ |
156 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
157 | if (blkg && blkg->q == q) { |
158 | if (update_hint) { | |
159 | lockdep_assert_held(q->queue_lock); | |
160 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
161 | } | |
a637120e | 162 | return blkg; |
86cde6b6 | 163 | } |
a637120e | 164 | |
80fd9979 TH |
165 | return NULL; |
166 | } | |
167 | ||
168 | /** | |
169 | * blkg_lookup - lookup blkg for the specified blkcg - q pair | |
170 | * @blkcg: blkcg of interest | |
171 | * @q: request_queue of interest | |
172 | * | |
173 | * Lookup blkg for the @blkcg - @q pair. This function should be called | |
174 | * under RCU read lock and is guaranteed to return %NULL if @q is bypassing | |
175 | * - see blk_queue_bypass_start() for details. | |
176 | */ | |
3c798398 | 177 | struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q) |
80fd9979 TH |
178 | { |
179 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
180 | ||
181 | if (unlikely(blk_queue_bypass(q))) | |
182 | return NULL; | |
86cde6b6 | 183 | return __blkg_lookup(blkcg, q, false); |
80fd9979 TH |
184 | } |
185 | EXPORT_SYMBOL_GPL(blkg_lookup); | |
186 | ||
15974993 TH |
187 | /* |
188 | * If @new_blkg is %NULL, this function tries to allocate a new one as | |
189 | * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return. | |
190 | */ | |
86cde6b6 TH |
191 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
192 | struct request_queue *q, | |
193 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 194 | { |
3c798398 | 195 | struct blkcg_gq *blkg; |
ce7acfea | 196 | struct bdi_writeback_congested *wb_congested; |
f427d909 | 197 | int i, ret; |
5624a4e4 | 198 | |
cd1604fa TH |
199 | WARN_ON_ONCE(!rcu_read_lock_held()); |
200 | lockdep_assert_held(q->queue_lock); | |
201 | ||
7ee9c562 | 202 | /* blkg holds a reference to blkcg */ |
ec903c0c | 203 | if (!css_tryget_online(&blkcg->css)) { |
93e6d5d8 TH |
204 | ret = -EINVAL; |
205 | goto err_free_blkg; | |
15974993 | 206 | } |
cd1604fa | 207 | |
ce7acfea TH |
208 | wb_congested = wb_congested_get_create(&q->backing_dev_info, |
209 | blkcg->css.id, GFP_ATOMIC); | |
210 | if (!wb_congested) { | |
211 | ret = -ENOMEM; | |
212 | goto err_put_css; | |
213 | } | |
214 | ||
496fb780 | 215 | /* allocate */ |
15974993 TH |
216 | if (!new_blkg) { |
217 | new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC); | |
218 | if (unlikely(!new_blkg)) { | |
93e6d5d8 | 219 | ret = -ENOMEM; |
ce7acfea | 220 | goto err_put_congested; |
15974993 TH |
221 | } |
222 | } | |
223 | blkg = new_blkg; | |
ce7acfea | 224 | blkg->wb_congested = wb_congested; |
cd1604fa | 225 | |
db613670 | 226 | /* link parent */ |
3c547865 TH |
227 | if (blkcg_parent(blkcg)) { |
228 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
229 | if (WARN_ON_ONCE(!blkg->parent)) { | |
2423c9c3 | 230 | ret = -EINVAL; |
ce7acfea | 231 | goto err_put_congested; |
3c547865 TH |
232 | } |
233 | blkg_get(blkg->parent); | |
234 | } | |
235 | ||
db613670 TH |
236 | /* invoke per-policy init */ |
237 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
238 | struct blkcg_policy *pol = blkcg_policy[i]; | |
239 | ||
240 | if (blkg->pd[i] && pol->pd_init_fn) | |
241 | pol->pd_init_fn(blkg); | |
242 | } | |
243 | ||
244 | /* insert */ | |
cd1604fa | 245 | spin_lock(&blkcg->lock); |
a637120e TH |
246 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
247 | if (likely(!ret)) { | |
248 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
249 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
250 | |
251 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
252 | struct blkcg_policy *pol = blkcg_policy[i]; | |
253 | ||
254 | if (blkg->pd[i] && pol->pd_online_fn) | |
255 | pol->pd_online_fn(blkg); | |
256 | } | |
a637120e | 257 | } |
f427d909 | 258 | blkg->online = true; |
cd1604fa | 259 | spin_unlock(&blkcg->lock); |
496fb780 | 260 | |
ec13b1d6 | 261 | if (!ret) |
a637120e | 262 | return blkg; |
15974993 | 263 | |
3c547865 TH |
264 | /* @blkg failed fully initialized, use the usual release path */ |
265 | blkg_put(blkg); | |
266 | return ERR_PTR(ret); | |
267 | ||
ce7acfea TH |
268 | err_put_congested: |
269 | wb_congested_put(wb_congested); | |
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 | */ | |
ec6c676a | 359 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
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 TH |
387 | } |
388 | } | |
389 | ||
2a4fd070 TH |
390 | /* |
391 | * A group is RCU protected, but having an rcu lock does not mean that one | |
392 | * can access all the fields of blkg and assume these are valid. For | |
393 | * example, don't try to follow throtl_data and request queue links. | |
394 | * | |
395 | * Having a reference to blkg under an rcu allows accesses to only values | |
396 | * local to groups like group stats and group rate limits. | |
397 | */ | |
398 | void __blkg_release_rcu(struct rcu_head *rcu_head) | |
1adaf3dd | 399 | { |
2a4fd070 | 400 | struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head); |
db613670 TH |
401 | int i; |
402 | ||
403 | /* tell policies that this one is being freed */ | |
404 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
405 | struct blkcg_policy *pol = blkcg_policy[i]; | |
406 | ||
407 | if (blkg->pd[i] && pol->pd_exit_fn) | |
408 | pol->pd_exit_fn(blkg); | |
409 | } | |
410 | ||
3c547865 | 411 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
1adaf3dd | 412 | css_put(&blkg->blkcg->css); |
a5049a8a | 413 | if (blkg->parent) |
3c547865 | 414 | blkg_put(blkg->parent); |
1adaf3dd | 415 | |
ce7acfea TH |
416 | wb_congested_put(blkg->wb_congested); |
417 | ||
2a4fd070 | 418 | blkg_free(blkg); |
1adaf3dd | 419 | } |
2a4fd070 | 420 | EXPORT_SYMBOL_GPL(__blkg_release_rcu); |
1adaf3dd | 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 | ||
182446d0 TH |
457 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
458 | struct cftype *cftype, u64 val) | |
303a3acb | 459 | { |
182446d0 | 460 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 461 | struct blkcg_gq *blkg; |
bc0d6501 | 462 | int i; |
303a3acb | 463 | |
838f13bf | 464 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 465 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
466 | |
467 | /* | |
468 | * Note that stat reset is racy - it doesn't synchronize against | |
469 | * stat updates. This is a debug feature which shouldn't exist | |
470 | * anyway. If you get hit by a race, retry. | |
471 | */ | |
b67bfe0d | 472 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
8bd435b3 | 473 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 474 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 475 | |
a2b1693b | 476 | if (blkcg_policy_enabled(blkg->q, pol) && |
f9fcc2d3 TH |
477 | pol->pd_reset_stats_fn) |
478 | pol->pd_reset_stats_fn(blkg); | |
bc0d6501 | 479 | } |
303a3acb | 480 | } |
f0bdc8cd | 481 | |
303a3acb | 482 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 483 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
484 | return 0; |
485 | } | |
486 | ||
3c798398 | 487 | static const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 488 | { |
d3d32e69 TH |
489 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
490 | if (blkg->q->backing_dev_info.dev) | |
491 | return dev_name(blkg->q->backing_dev_info.dev); | |
492 | return NULL; | |
303a3acb DS |
493 | } |
494 | ||
d3d32e69 TH |
495 | /** |
496 | * blkcg_print_blkgs - helper for printing per-blkg data | |
497 | * @sf: seq_file to print to | |
498 | * @blkcg: blkcg of interest | |
499 | * @prfill: fill function to print out a blkg | |
500 | * @pol: policy in question | |
501 | * @data: data to be passed to @prfill | |
502 | * @show_total: to print out sum of prfill return values or not | |
503 | * | |
504 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
505 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
506 | * policy data and @data and the matching queue lock held. If @show_total |
507 | * is %true, the sum of the return values from @prfill is printed with | |
508 | * "Total" label at the end. | |
d3d32e69 TH |
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 | 520 | u64 total = 0; |
5624a4e4 | 521 | |
810ecfa7 | 522 | rcu_read_lock(); |
ee89f812 | 523 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
810ecfa7 | 524 | spin_lock_irq(blkg->q->queue_lock); |
a2b1693b | 525 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 526 | total += prfill(sf, blkg->pd[pol->plid], data); |
810ecfa7 TH |
527 | spin_unlock_irq(blkg->q->queue_lock); |
528 | } | |
529 | rcu_read_unlock(); | |
d3d32e69 TH |
530 | |
531 | if (show_total) | |
532 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
533 | } | |
829fdb50 | 534 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
535 | |
536 | /** | |
537 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
538 | * @sf: seq_file to print to | |
f95a04af | 539 | * @pd: policy private data of interest |
d3d32e69 TH |
540 | * @v: value to print |
541 | * | |
f95a04af | 542 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 543 | */ |
f95a04af | 544 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 545 | { |
f95a04af | 546 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
547 | |
548 | if (!dname) | |
549 | return 0; | |
550 | ||
551 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
552 | return v; | |
553 | } | |
829fdb50 | 554 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
555 | |
556 | /** | |
557 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
558 | * @sf: seq_file to print to | |
f95a04af | 559 | * @pd: policy private data of interest |
d3d32e69 TH |
560 | * @rwstat: rwstat to print |
561 | * | |
f95a04af | 562 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 563 | */ |
f95a04af | 564 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 565 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
566 | { |
567 | static const char *rwstr[] = { | |
568 | [BLKG_RWSTAT_READ] = "Read", | |
569 | [BLKG_RWSTAT_WRITE] = "Write", | |
570 | [BLKG_RWSTAT_SYNC] = "Sync", | |
571 | [BLKG_RWSTAT_ASYNC] = "Async", | |
572 | }; | |
f95a04af | 573 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
574 | u64 v; |
575 | int i; | |
576 | ||
577 | if (!dname) | |
578 | return 0; | |
579 | ||
580 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
581 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
582 | (unsigned long long)rwstat->cnt[i]); | |
583 | ||
584 | v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE]; | |
585 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); | |
586 | return v; | |
587 | } | |
b50da39f | 588 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 589 | |
5bc4afb1 TH |
590 | /** |
591 | * blkg_prfill_stat - prfill callback for blkg_stat | |
592 | * @sf: seq_file to print to | |
f95a04af TH |
593 | * @pd: policy private data of interest |
594 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
595 | * |
596 | * prfill callback for printing a blkg_stat. | |
597 | */ | |
f95a04af | 598 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 599 | { |
f95a04af | 600 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 601 | } |
5bc4afb1 | 602 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 603 | |
5bc4afb1 TH |
604 | /** |
605 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
606 | * @sf: seq_file to print to | |
f95a04af TH |
607 | * @pd: policy private data of interest |
608 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
609 | * |
610 | * prfill callback for printing a blkg_rwstat. | |
611 | */ | |
f95a04af TH |
612 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
613 | int off) | |
d3d32e69 | 614 | { |
f95a04af | 615 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 616 | |
f95a04af | 617 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 618 | } |
5bc4afb1 | 619 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 620 | |
16b3de66 TH |
621 | /** |
622 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
623 | * @pd: policy private data of interest | |
624 | * @off: offset to the blkg_stat in @pd | |
625 | * | |
626 | * Collect the blkg_stat specified by @off from @pd and all its online | |
627 | * descendants and return the sum. The caller must be holding the queue | |
628 | * lock for online tests. | |
629 | */ | |
630 | u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off) | |
631 | { | |
632 | struct blkcg_policy *pol = blkcg_policy[pd->plid]; | |
633 | struct blkcg_gq *pos_blkg; | |
492eb21b | 634 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 635 | u64 sum = 0; |
16b3de66 TH |
636 | |
637 | lockdep_assert_held(pd->blkg->q->queue_lock); | |
638 | ||
16b3de66 | 639 | rcu_read_lock(); |
492eb21b | 640 | blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) { |
16b3de66 TH |
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; | |
492eb21b | 667 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 668 | struct blkg_rwstat sum = { }; |
16b3de66 TH |
669 | int i; |
670 | ||
671 | lockdep_assert_held(pd->blkg->q->queue_lock); | |
672 | ||
16b3de66 | 673 | rcu_read_lock(); |
492eb21b | 674 | blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) { |
16b3de66 TH |
675 | struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol); |
676 | struct blkg_rwstat *rwstat = (void *)pos_pd + off; | |
677 | struct blkg_rwstat tmp; | |
678 | ||
679 | if (!pos_blkg->online) | |
680 | continue; | |
681 | ||
682 | tmp = blkg_rwstat_read(rwstat); | |
683 | ||
684 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
685 | sum.cnt[i] += tmp.cnt[i]; | |
686 | } | |
687 | rcu_read_unlock(); | |
688 | ||
689 | return sum; | |
690 | } | |
691 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
692 | ||
3a8b31d3 TH |
693 | /** |
694 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
695 | * @blkcg: target block cgroup | |
da8b0662 | 696 | * @pol: target policy |
3a8b31d3 TH |
697 | * @input: input string |
698 | * @ctx: blkg_conf_ctx to be filled | |
699 | * | |
700 | * Parse per-blkg config update from @input and initialize @ctx with the | |
701 | * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new | |
da8b0662 TH |
702 | * value. This function returns with RCU read lock and queue lock held and |
703 | * must be paired with blkg_conf_finish(). | |
3a8b31d3 | 704 | */ |
3c798398 TH |
705 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
706 | const char *input, struct blkg_conf_ctx *ctx) | |
da8b0662 | 707 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
34d0f179 | 708 | { |
3a8b31d3 | 709 | struct gendisk *disk; |
3c798398 | 710 | struct blkcg_gq *blkg; |
726fa694 TH |
711 | unsigned int major, minor; |
712 | unsigned long long v; | |
713 | int part, ret; | |
34d0f179 | 714 | |
726fa694 TH |
715 | if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3) |
716 | return -EINVAL; | |
3a8b31d3 | 717 | |
726fa694 | 718 | disk = get_gendisk(MKDEV(major, minor), &part); |
4bfd482e | 719 | if (!disk || part) |
726fa694 | 720 | return -EINVAL; |
e56da7e2 TH |
721 | |
722 | rcu_read_lock(); | |
4bfd482e | 723 | spin_lock_irq(disk->queue->queue_lock); |
da8b0662 | 724 | |
a2b1693b | 725 | if (blkcg_policy_enabled(disk->queue, pol)) |
3c96cb32 | 726 | blkg = blkg_lookup_create(blkcg, disk->queue); |
a2b1693b TH |
727 | else |
728 | blkg = ERR_PTR(-EINVAL); | |
e56da7e2 | 729 | |
4bfd482e TH |
730 | if (IS_ERR(blkg)) { |
731 | ret = PTR_ERR(blkg); | |
3a8b31d3 | 732 | rcu_read_unlock(); |
da8b0662 | 733 | spin_unlock_irq(disk->queue->queue_lock); |
3a8b31d3 TH |
734 | put_disk(disk); |
735 | /* | |
736 | * If queue was bypassing, we should retry. Do so after a | |
737 | * short msleep(). It isn't strictly necessary but queue | |
738 | * can be bypassing for some time and it's always nice to | |
739 | * avoid busy looping. | |
740 | */ | |
741 | if (ret == -EBUSY) { | |
742 | msleep(10); | |
743 | ret = restart_syscall(); | |
7702e8f4 | 744 | } |
726fa694 | 745 | return ret; |
062a644d | 746 | } |
3a8b31d3 TH |
747 | |
748 | ctx->disk = disk; | |
749 | ctx->blkg = blkg; | |
726fa694 TH |
750 | ctx->v = v; |
751 | return 0; | |
34d0f179 | 752 | } |
829fdb50 | 753 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 754 | |
3a8b31d3 TH |
755 | /** |
756 | * blkg_conf_finish - finish up per-blkg config update | |
757 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
758 | * | |
759 | * Finish up after per-blkg config update. This function must be paired | |
760 | * with blkg_conf_prep(). | |
761 | */ | |
829fdb50 | 762 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
da8b0662 | 763 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 764 | { |
da8b0662 | 765 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
3a8b31d3 TH |
766 | rcu_read_unlock(); |
767 | put_disk(ctx->disk); | |
34d0f179 | 768 | } |
829fdb50 | 769 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 770 | |
3c798398 | 771 | struct cftype blkcg_files[] = { |
84c124da DS |
772 | { |
773 | .name = "reset_stats", | |
3c798398 | 774 | .write_u64 = blkcg_reset_stats, |
22084190 | 775 | }, |
4baf6e33 | 776 | { } /* terminate */ |
31e4c28d VG |
777 | }; |
778 | ||
9f13ef67 | 779 | /** |
92fb9748 | 780 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 781 | * @css: css of interest |
9f13ef67 | 782 | * |
eb95419b TH |
783 | * This function is called when @css is about to go away and responsible |
784 | * for shooting down all blkgs associated with @css. blkgs should be | |
9f13ef67 TH |
785 | * removed while holding both q and blkcg locks. As blkcg lock is nested |
786 | * inside q lock, this function performs reverse double lock dancing. | |
787 | * | |
788 | * This is the blkcg counterpart of ioc_release_fn(). | |
789 | */ | |
eb95419b | 790 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 791 | { |
eb95419b | 792 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 793 | |
9f13ef67 | 794 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 795 | |
9f13ef67 | 796 | while (!hlist_empty(&blkcg->blkg_list)) { |
3c798398 TH |
797 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, |
798 | struct blkcg_gq, blkcg_node); | |
c875f4d0 | 799 | struct request_queue *q = blkg->q; |
b1c35769 | 800 | |
9f13ef67 TH |
801 | if (spin_trylock(q->queue_lock)) { |
802 | blkg_destroy(blkg); | |
803 | spin_unlock(q->queue_lock); | |
804 | } else { | |
805 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 806 | cpu_relax(); |
a5567932 | 807 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 808 | } |
9f13ef67 | 809 | } |
b1c35769 | 810 | |
9f13ef67 | 811 | spin_unlock_irq(&blkcg->lock); |
52ebea74 TH |
812 | |
813 | wb_blkcg_offline(blkcg); | |
7ee9c562 TH |
814 | } |
815 | ||
eb95419b | 816 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 817 | { |
eb95419b | 818 | struct blkcg *blkcg = css_to_blkcg(css); |
7ee9c562 | 819 | |
a322baad AA |
820 | if (blkcg != &blkcg_root) { |
821 | int i; | |
822 | ||
823 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
824 | kfree(blkcg->pd[i]); | |
67523c48 | 825 | kfree(blkcg); |
a322baad | 826 | } |
31e4c28d VG |
827 | } |
828 | ||
eb95419b TH |
829 | static struct cgroup_subsys_state * |
830 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 831 | { |
3c798398 | 832 | struct blkcg *blkcg; |
e48453c3 AA |
833 | struct cgroup_subsys_state *ret; |
834 | int i; | |
31e4c28d | 835 | |
eb95419b | 836 | if (!parent_css) { |
3c798398 | 837 | blkcg = &blkcg_root; |
31e4c28d VG |
838 | goto done; |
839 | } | |
840 | ||
31e4c28d | 841 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); |
e48453c3 AA |
842 | if (!blkcg) { |
843 | ret = ERR_PTR(-ENOMEM); | |
844 | goto free_blkcg; | |
845 | } | |
846 | ||
144232b3 TH |
847 | mutex_lock(&blkcg_pol_mutex); |
848 | ||
e48453c3 AA |
849 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { |
850 | struct blkcg_policy *pol = blkcg_policy[i]; | |
851 | struct blkcg_policy_data *cpd; | |
852 | ||
853 | /* | |
854 | * If the policy hasn't been attached yet, wait for it | |
855 | * to be attached before doing anything else. Otherwise, | |
856 | * check if the policy requires any specific per-cgroup | |
857 | * data: if it does, allocate and initialize it. | |
858 | */ | |
859 | if (!pol || !pol->cpd_size) | |
860 | continue; | |
861 | ||
862 | BUG_ON(blkcg->pd[i]); | |
863 | cpd = kzalloc(pol->cpd_size, GFP_KERNEL); | |
864 | if (!cpd) { | |
144232b3 | 865 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 AA |
866 | ret = ERR_PTR(-ENOMEM); |
867 | goto free_pd_blkcg; | |
868 | } | |
869 | blkcg->pd[i] = cpd; | |
870 | cpd->plid = i; | |
871 | pol->cpd_init_fn(blkcg); | |
872 | } | |
31e4c28d | 873 | |
144232b3 | 874 | mutex_unlock(&blkcg_pol_mutex); |
31e4c28d VG |
875 | done: |
876 | spin_lock_init(&blkcg->lock); | |
a637120e | 877 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC); |
31e4c28d | 878 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
879 | #ifdef CONFIG_CGROUP_WRITEBACK |
880 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
881 | #endif | |
31e4c28d | 882 | return &blkcg->css; |
e48453c3 AA |
883 | |
884 | free_pd_blkcg: | |
885 | for (i--; i >= 0; i--) | |
886 | kfree(blkcg->pd[i]); | |
887 | ||
888 | free_blkcg: | |
889 | kfree(blkcg); | |
890 | return ret; | |
31e4c28d VG |
891 | } |
892 | ||
5efd6113 TH |
893 | /** |
894 | * blkcg_init_queue - initialize blkcg part of request queue | |
895 | * @q: request_queue to initialize | |
896 | * | |
897 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
898 | * part of new request_queue @q. | |
899 | * | |
900 | * RETURNS: | |
901 | * 0 on success, -errno on failure. | |
902 | */ | |
903 | int blkcg_init_queue(struct request_queue *q) | |
904 | { | |
ec13b1d6 TH |
905 | struct blkcg_gq *new_blkg, *blkg; |
906 | bool preloaded; | |
907 | int ret; | |
908 | ||
909 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); | |
910 | if (!new_blkg) | |
911 | return -ENOMEM; | |
912 | ||
913 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
5efd6113 | 914 | |
ec13b1d6 TH |
915 | /* |
916 | * Make sure the root blkg exists and count the existing blkgs. As | |
917 | * @q is bypassing at this point, blkg_lookup_create() can't be | |
918 | * used. Open code insertion. | |
919 | */ | |
920 | rcu_read_lock(); | |
921 | spin_lock_irq(q->queue_lock); | |
922 | blkg = blkg_create(&blkcg_root, q, new_blkg); | |
923 | spin_unlock_irq(q->queue_lock); | |
924 | rcu_read_unlock(); | |
925 | ||
926 | if (preloaded) | |
927 | radix_tree_preload_end(); | |
928 | ||
929 | if (IS_ERR(blkg)) { | |
930 | kfree(new_blkg); | |
931 | return PTR_ERR(blkg); | |
932 | } | |
933 | ||
934 | q->root_blkg = blkg; | |
935 | q->root_rl.blkg = blkg; | |
5efd6113 | 936 | |
ec13b1d6 TH |
937 | ret = blk_throtl_init(q); |
938 | if (ret) { | |
939 | spin_lock_irq(q->queue_lock); | |
940 | blkg_destroy_all(q); | |
941 | spin_unlock_irq(q->queue_lock); | |
942 | } | |
943 | return ret; | |
5efd6113 TH |
944 | } |
945 | ||
946 | /** | |
947 | * blkcg_drain_queue - drain blkcg part of request_queue | |
948 | * @q: request_queue to drain | |
949 | * | |
950 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
951 | */ | |
952 | void blkcg_drain_queue(struct request_queue *q) | |
953 | { | |
954 | lockdep_assert_held(q->queue_lock); | |
955 | ||
0b462c89 TH |
956 | /* |
957 | * @q could be exiting and already have destroyed all blkgs as | |
958 | * indicated by NULL root_blkg. If so, don't confuse policies. | |
959 | */ | |
960 | if (!q->root_blkg) | |
961 | return; | |
962 | ||
5efd6113 TH |
963 | blk_throtl_drain(q); |
964 | } | |
965 | ||
966 | /** | |
967 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
968 | * @q: request_queue being released | |
969 | * | |
970 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
971 | */ | |
972 | void blkcg_exit_queue(struct request_queue *q) | |
973 | { | |
6d18b008 | 974 | spin_lock_irq(q->queue_lock); |
3c96cb32 | 975 | blkg_destroy_all(q); |
6d18b008 TH |
976 | spin_unlock_irq(q->queue_lock); |
977 | ||
5efd6113 TH |
978 | blk_throtl_exit(q); |
979 | } | |
980 | ||
31e4c28d VG |
981 | /* |
982 | * We cannot support shared io contexts, as we have no mean to support | |
983 | * two tasks with the same ioc in two different groups without major rework | |
984 | * of the main cic data structures. For now we allow a task to change | |
985 | * its cgroup only if it's the only owner of its ioc. | |
986 | */ | |
eb95419b TH |
987 | static int blkcg_can_attach(struct cgroup_subsys_state *css, |
988 | struct cgroup_taskset *tset) | |
31e4c28d | 989 | { |
bb9d97b6 | 990 | struct task_struct *task; |
31e4c28d VG |
991 | struct io_context *ioc; |
992 | int ret = 0; | |
993 | ||
994 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
924f0d9a | 995 | cgroup_taskset_for_each(task, tset) { |
bb9d97b6 TH |
996 | task_lock(task); |
997 | ioc = task->io_context; | |
998 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
999 | ret = -EINVAL; | |
1000 | task_unlock(task); | |
1001 | if (ret) | |
1002 | break; | |
1003 | } | |
31e4c28d VG |
1004 | return ret; |
1005 | } | |
1006 | ||
073219e9 | 1007 | struct cgroup_subsys blkio_cgrp_subsys = { |
92fb9748 TH |
1008 | .css_alloc = blkcg_css_alloc, |
1009 | .css_offline = blkcg_css_offline, | |
1010 | .css_free = blkcg_css_free, | |
3c798398 | 1011 | .can_attach = blkcg_can_attach, |
5577964e | 1012 | .legacy_cftypes = blkcg_files, |
1ced953b TH |
1013 | #ifdef CONFIG_MEMCG |
1014 | /* | |
1015 | * This ensures that, if available, memcg is automatically enabled | |
1016 | * together on the default hierarchy so that the owner cgroup can | |
1017 | * be retrieved from writeback pages. | |
1018 | */ | |
1019 | .depends_on = 1 << memory_cgrp_id, | |
1020 | #endif | |
676f7c8f | 1021 | }; |
073219e9 | 1022 | EXPORT_SYMBOL_GPL(blkio_cgrp_subsys); |
676f7c8f | 1023 | |
a2b1693b TH |
1024 | /** |
1025 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1026 | * @q: request_queue of interest | |
1027 | * @pol: blkcg policy to activate | |
1028 | * | |
1029 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1030 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1031 | * | |
1032 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1033 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1034 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1035 | * always enough for dereferencing policy data. | |
1036 | * | |
1037 | * The caller is responsible for synchronizing [de]activations and policy | |
1038 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1039 | */ | |
1040 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1041 | const struct blkcg_policy *pol) |
a2b1693b TH |
1042 | { |
1043 | LIST_HEAD(pds); | |
e48453c3 | 1044 | LIST_HEAD(cpds); |
ec13b1d6 | 1045 | struct blkcg_gq *blkg; |
e48453c3 AA |
1046 | struct blkg_policy_data *pd, *nd; |
1047 | struct blkcg_policy_data *cpd, *cnd; | |
a2b1693b TH |
1048 | int cnt = 0, ret; |
1049 | ||
1050 | if (blkcg_policy_enabled(q, pol)) | |
1051 | return 0; | |
1052 | ||
ec13b1d6 | 1053 | /* count and allocate policy_data for all existing blkgs */ |
a2b1693b | 1054 | blk_queue_bypass_start(q); |
a2b1693b | 1055 | spin_lock_irq(q->queue_lock); |
a2b1693b TH |
1056 | list_for_each_entry(blkg, &q->blkg_list, q_node) |
1057 | cnt++; | |
a2b1693b TH |
1058 | spin_unlock_irq(q->queue_lock); |
1059 | ||
e48453c3 AA |
1060 | /* |
1061 | * Allocate per-blkg and per-blkcg policy data | |
1062 | * for all existing blkgs. | |
1063 | */ | |
a2b1693b | 1064 | while (cnt--) { |
f95a04af | 1065 | pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node); |
a2b1693b TH |
1066 | if (!pd) { |
1067 | ret = -ENOMEM; | |
1068 | goto out_free; | |
1069 | } | |
1070 | list_add_tail(&pd->alloc_node, &pds); | |
e48453c3 AA |
1071 | |
1072 | if (!pol->cpd_size) | |
1073 | continue; | |
1074 | cpd = kzalloc_node(pol->cpd_size, GFP_KERNEL, q->node); | |
1075 | if (!cpd) { | |
1076 | ret = -ENOMEM; | |
1077 | goto out_free; | |
1078 | } | |
1079 | list_add_tail(&cpd->alloc_node, &cpds); | |
a2b1693b TH |
1080 | } |
1081 | ||
1082 | /* | |
e48453c3 | 1083 | * Install the allocated pds and cpds. With @q bypassing, no new blkg |
a2b1693b TH |
1084 | * should have been created while the queue lock was dropped. |
1085 | */ | |
1086 | spin_lock_irq(q->queue_lock); | |
1087 | ||
1088 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
e48453c3 AA |
1089 | if (WARN_ON(list_empty(&pds)) || |
1090 | WARN_ON(pol->cpd_size && list_empty(&cpds))) { | |
a2b1693b TH |
1091 | /* umm... this shouldn't happen, just abort */ |
1092 | ret = -ENOMEM; | |
1093 | goto out_unlock; | |
1094 | } | |
e48453c3 AA |
1095 | cpd = list_first_entry(&cpds, struct blkcg_policy_data, |
1096 | alloc_node); | |
1097 | list_del_init(&cpd->alloc_node); | |
a2b1693b TH |
1098 | pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node); |
1099 | list_del_init(&pd->alloc_node); | |
1100 | ||
1101 | /* grab blkcg lock too while installing @pd on @blkg */ | |
1102 | spin_lock(&blkg->blkcg->lock); | |
1103 | ||
e48453c3 AA |
1104 | if (!pol->cpd_size) |
1105 | goto no_cpd; | |
1106 | if (!blkg->blkcg->pd[pol->plid]) { | |
1107 | /* Per-policy per-blkcg data */ | |
1108 | blkg->blkcg->pd[pol->plid] = cpd; | |
1109 | cpd->plid = pol->plid; | |
1110 | pol->cpd_init_fn(blkg->blkcg); | |
1111 | } else { /* must free it as it has already been extracted */ | |
1112 | kfree(cpd); | |
1113 | } | |
1114 | no_cpd: | |
a2b1693b TH |
1115 | blkg->pd[pol->plid] = pd; |
1116 | pd->blkg = blkg; | |
b276a876 | 1117 | pd->plid = pol->plid; |
f9fcc2d3 | 1118 | pol->pd_init_fn(blkg); |
a2b1693b TH |
1119 | |
1120 | spin_unlock(&blkg->blkcg->lock); | |
1121 | } | |
1122 | ||
1123 | __set_bit(pol->plid, q->blkcg_pols); | |
1124 | ret = 0; | |
1125 | out_unlock: | |
1126 | spin_unlock_irq(q->queue_lock); | |
1127 | out_free: | |
1128 | blk_queue_bypass_end(q); | |
e48453c3 | 1129 | list_for_each_entry_safe(pd, nd, &pds, alloc_node) |
a2b1693b | 1130 | kfree(pd); |
e48453c3 AA |
1131 | list_for_each_entry_safe(cpd, cnd, &cpds, alloc_node) |
1132 | kfree(cpd); | |
a2b1693b TH |
1133 | return ret; |
1134 | } | |
1135 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1136 | ||
1137 | /** | |
1138 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1139 | * @q: request_queue of interest | |
1140 | * @pol: blkcg policy to deactivate | |
1141 | * | |
1142 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1143 | * blkcg_activate_policy(). | |
1144 | */ | |
1145 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1146 | const struct blkcg_policy *pol) |
a2b1693b | 1147 | { |
3c798398 | 1148 | struct blkcg_gq *blkg; |
a2b1693b TH |
1149 | |
1150 | if (!blkcg_policy_enabled(q, pol)) | |
1151 | return; | |
1152 | ||
1153 | blk_queue_bypass_start(q); | |
1154 | spin_lock_irq(q->queue_lock); | |
1155 | ||
1156 | __clear_bit(pol->plid, q->blkcg_pols); | |
1157 | ||
1158 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1159 | /* grab blkcg lock too while removing @pd from @blkg */ | |
1160 | spin_lock(&blkg->blkcg->lock); | |
1161 | ||
f427d909 TH |
1162 | if (pol->pd_offline_fn) |
1163 | pol->pd_offline_fn(blkg); | |
f9fcc2d3 TH |
1164 | if (pol->pd_exit_fn) |
1165 | pol->pd_exit_fn(blkg); | |
a2b1693b TH |
1166 | |
1167 | kfree(blkg->pd[pol->plid]); | |
1168 | blkg->pd[pol->plid] = NULL; | |
1169 | ||
1170 | spin_unlock(&blkg->blkcg->lock); | |
1171 | } | |
1172 | ||
1173 | spin_unlock_irq(q->queue_lock); | |
1174 | blk_queue_bypass_end(q); | |
1175 | } | |
1176 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1177 | ||
8bd435b3 | 1178 | /** |
3c798398 TH |
1179 | * blkcg_policy_register - register a blkcg policy |
1180 | * @pol: blkcg policy to register | |
8bd435b3 | 1181 | * |
3c798398 TH |
1182 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1183 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1184 | */ |
d5bf0291 | 1185 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1186 | { |
8bd435b3 | 1187 | int i, ret; |
e8989fae | 1188 | |
f95a04af TH |
1189 | if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data))) |
1190 | return -EINVAL; | |
1191 | ||
838f13bf | 1192 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1193 | mutex_lock(&blkcg_pol_mutex); |
1194 | ||
8bd435b3 TH |
1195 | /* find an empty slot */ |
1196 | ret = -ENOSPC; | |
1197 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1198 | if (!blkcg_policy[i]) |
8bd435b3 TH |
1199 | break; |
1200 | if (i >= BLKCG_MAX_POLS) | |
838f13bf | 1201 | goto err_unlock; |
035d10b2 | 1202 | |
8bd435b3 | 1203 | /* register and update blkgs */ |
3c798398 TH |
1204 | pol->plid = i; |
1205 | blkcg_policy[i] = pol; | |
838f13bf | 1206 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1207 | |
8bd435b3 | 1208 | /* everything is in place, add intf files for the new policy */ |
3c798398 | 1209 | if (pol->cftypes) |
2cf669a5 TH |
1210 | WARN_ON(cgroup_add_legacy_cftypes(&blkio_cgrp_subsys, |
1211 | pol->cftypes)); | |
838f13bf TH |
1212 | mutex_unlock(&blkcg_pol_register_mutex); |
1213 | return 0; | |
1214 | ||
1215 | err_unlock: | |
bc0d6501 | 1216 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1217 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1218 | return ret; |
3e252066 | 1219 | } |
3c798398 | 1220 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1221 | |
8bd435b3 | 1222 | /** |
3c798398 TH |
1223 | * blkcg_policy_unregister - unregister a blkcg policy |
1224 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1225 | * |
3c798398 | 1226 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1227 | */ |
3c798398 | 1228 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1229 | { |
838f13bf | 1230 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1231 | |
3c798398 | 1232 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1233 | goto out_unlock; |
1234 | ||
1235 | /* kill the intf files first */ | |
3c798398 | 1236 | if (pol->cftypes) |
2bb566cb | 1237 | cgroup_rm_cftypes(pol->cftypes); |
44ea53de | 1238 | |
8bd435b3 | 1239 | /* unregister and update blkgs */ |
838f13bf | 1240 | mutex_lock(&blkcg_pol_mutex); |
3c798398 | 1241 | blkcg_policy[pol->plid] = NULL; |
bc0d6501 | 1242 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1243 | out_unlock: |
1244 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1245 | } |
3c798398 | 1246 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |