]>
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 | |
7876f930 TH |
49 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
50 | ||
a2b1693b | 51 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 52 | const struct blkcg_policy *pol) |
a2b1693b TH |
53 | { |
54 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
55 | } | |
56 | ||
0381411e TH |
57 | /** |
58 | * blkg_free - free a blkg | |
59 | * @blkg: blkg to free | |
60 | * | |
61 | * Free @blkg which may be partially allocated. | |
62 | */ | |
3c798398 | 63 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 64 | { |
e8989fae | 65 | int i; |
549d3aa8 TH |
66 | |
67 | if (!blkg) | |
68 | return; | |
69 | ||
db613670 | 70 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
71 | if (blkg->pd[i]) |
72 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
e8989fae | 73 | |
994b7832 TH |
74 | if (blkg->blkcg != &blkcg_root) |
75 | blk_exit_rl(&blkg->rl); | |
549d3aa8 | 76 | kfree(blkg); |
0381411e TH |
77 | } |
78 | ||
79 | /** | |
80 | * blkg_alloc - allocate a blkg | |
81 | * @blkcg: block cgroup the new blkg is associated with | |
82 | * @q: request_queue the new blkg is associated with | |
15974993 | 83 | * @gfp_mask: allocation mask to use |
0381411e | 84 | * |
e8989fae | 85 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 86 | */ |
15974993 TH |
87 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
88 | gfp_t gfp_mask) | |
0381411e | 89 | { |
3c798398 | 90 | struct blkcg_gq *blkg; |
e8989fae | 91 | int i; |
0381411e TH |
92 | |
93 | /* alloc and init base part */ | |
15974993 | 94 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
95 | if (!blkg) |
96 | return NULL; | |
97 | ||
c875f4d0 | 98 | blkg->q = q; |
e8989fae | 99 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 100 | blkg->blkcg = blkcg; |
a5049a8a | 101 | atomic_set(&blkg->refcnt, 1); |
0381411e | 102 | |
a051661c TH |
103 | /* root blkg uses @q->root_rl, init rl only for !root blkgs */ |
104 | if (blkcg != &blkcg_root) { | |
105 | if (blk_init_rl(&blkg->rl, q, gfp_mask)) | |
106 | goto err_free; | |
107 | blkg->rl.blkg = blkg; | |
108 | } | |
109 | ||
8bd435b3 | 110 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 111 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 112 | struct blkg_policy_data *pd; |
0381411e | 113 | |
a2b1693b | 114 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
115 | continue; |
116 | ||
117 | /* alloc per-policy data and attach it to blkg */ | |
001bea73 | 118 | pd = pol->pd_alloc_fn(gfp_mask, q->node); |
a051661c TH |
119 | if (!pd) |
120 | goto err_free; | |
549d3aa8 | 121 | |
e8989fae TH |
122 | blkg->pd[i] = pd; |
123 | pd->blkg = blkg; | |
b276a876 | 124 | pd->plid = i; |
e8989fae TH |
125 | } |
126 | ||
0381411e | 127 | return blkg; |
a051661c TH |
128 | |
129 | err_free: | |
130 | blkg_free(blkg); | |
131 | return NULL; | |
0381411e TH |
132 | } |
133 | ||
16b3de66 TH |
134 | /** |
135 | * __blkg_lookup - internal version of blkg_lookup() | |
136 | * @blkcg: blkcg of interest | |
137 | * @q: request_queue of interest | |
138 | * @update_hint: whether to update lookup hint with the result or not | |
139 | * | |
140 | * This is internal version and shouldn't be used by policy | |
141 | * implementations. Looks up blkgs for the @blkcg - @q pair regardless of | |
142 | * @q's bypass state. If @update_hint is %true, the caller should be | |
143 | * holding @q->queue_lock and lookup hint is updated on success. | |
144 | */ | |
dd4a4ffc TH |
145 | struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q, |
146 | bool update_hint) | |
80fd9979 | 147 | { |
3c798398 | 148 | struct blkcg_gq *blkg; |
80fd9979 | 149 | |
a637120e TH |
150 | blkg = rcu_dereference(blkcg->blkg_hint); |
151 | if (blkg && blkg->q == q) | |
152 | return blkg; | |
153 | ||
154 | /* | |
86cde6b6 TH |
155 | * Hint didn't match. Look up from the radix tree. Note that the |
156 | * hint can only be updated under queue_lock as otherwise @blkg | |
157 | * could have already been removed from blkg_tree. The caller is | |
158 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
159 | */ |
160 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
161 | if (blkg && blkg->q == q) { |
162 | if (update_hint) { | |
163 | lockdep_assert_held(q->queue_lock); | |
164 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
165 | } | |
a637120e | 166 | return blkg; |
86cde6b6 | 167 | } |
a637120e | 168 | |
80fd9979 TH |
169 | return NULL; |
170 | } | |
171 | ||
172 | /** | |
173 | * blkg_lookup - lookup blkg for the specified blkcg - q pair | |
174 | * @blkcg: blkcg of interest | |
175 | * @q: request_queue of interest | |
176 | * | |
177 | * Lookup blkg for the @blkcg - @q pair. This function should be called | |
178 | * under RCU read lock and is guaranteed to return %NULL if @q is bypassing | |
179 | * - see blk_queue_bypass_start() for details. | |
180 | */ | |
3c798398 | 181 | struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q) |
80fd9979 TH |
182 | { |
183 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
184 | ||
185 | if (unlikely(blk_queue_bypass(q))) | |
186 | return NULL; | |
86cde6b6 | 187 | return __blkg_lookup(blkcg, q, false); |
80fd9979 TH |
188 | } |
189 | EXPORT_SYMBOL_GPL(blkg_lookup); | |
190 | ||
15974993 TH |
191 | /* |
192 | * If @new_blkg is %NULL, this function tries to allocate a new one as | |
d93a11f1 | 193 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. |
15974993 | 194 | */ |
86cde6b6 TH |
195 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
196 | struct request_queue *q, | |
197 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 198 | { |
3c798398 | 199 | struct blkcg_gq *blkg; |
ce7acfea | 200 | struct bdi_writeback_congested *wb_congested; |
f427d909 | 201 | int i, ret; |
5624a4e4 | 202 | |
cd1604fa TH |
203 | WARN_ON_ONCE(!rcu_read_lock_held()); |
204 | lockdep_assert_held(q->queue_lock); | |
205 | ||
7ee9c562 | 206 | /* blkg holds a reference to blkcg */ |
ec903c0c | 207 | if (!css_tryget_online(&blkcg->css)) { |
93e6d5d8 TH |
208 | ret = -EINVAL; |
209 | goto err_free_blkg; | |
15974993 | 210 | } |
cd1604fa | 211 | |
ce7acfea | 212 | wb_congested = wb_congested_get_create(&q->backing_dev_info, |
d93a11f1 | 213 | blkcg->css.id, GFP_NOWAIT); |
ce7acfea TH |
214 | if (!wb_congested) { |
215 | ret = -ENOMEM; | |
216 | goto err_put_css; | |
217 | } | |
218 | ||
496fb780 | 219 | /* allocate */ |
15974993 | 220 | if (!new_blkg) { |
d93a11f1 | 221 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT); |
15974993 | 222 | if (unlikely(!new_blkg)) { |
93e6d5d8 | 223 | ret = -ENOMEM; |
ce7acfea | 224 | goto err_put_congested; |
15974993 TH |
225 | } |
226 | } | |
227 | blkg = new_blkg; | |
ce7acfea | 228 | blkg->wb_congested = wb_congested; |
cd1604fa | 229 | |
db613670 | 230 | /* link parent */ |
3c547865 TH |
231 | if (blkcg_parent(blkcg)) { |
232 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
233 | if (WARN_ON_ONCE(!blkg->parent)) { | |
2423c9c3 | 234 | ret = -EINVAL; |
ce7acfea | 235 | goto err_put_congested; |
3c547865 TH |
236 | } |
237 | blkg_get(blkg->parent); | |
238 | } | |
239 | ||
db613670 TH |
240 | /* invoke per-policy init */ |
241 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
242 | struct blkcg_policy *pol = blkcg_policy[i]; | |
243 | ||
244 | if (blkg->pd[i] && pol->pd_init_fn) | |
245 | pol->pd_init_fn(blkg); | |
246 | } | |
247 | ||
248 | /* insert */ | |
cd1604fa | 249 | spin_lock(&blkcg->lock); |
a637120e TH |
250 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
251 | if (likely(!ret)) { | |
252 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
253 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
254 | |
255 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
256 | struct blkcg_policy *pol = blkcg_policy[i]; | |
257 | ||
258 | if (blkg->pd[i] && pol->pd_online_fn) | |
259 | pol->pd_online_fn(blkg); | |
260 | } | |
a637120e | 261 | } |
f427d909 | 262 | blkg->online = true; |
cd1604fa | 263 | spin_unlock(&blkcg->lock); |
496fb780 | 264 | |
ec13b1d6 | 265 | if (!ret) |
a637120e | 266 | return blkg; |
15974993 | 267 | |
3c547865 TH |
268 | /* @blkg failed fully initialized, use the usual release path */ |
269 | blkg_put(blkg); | |
270 | return ERR_PTR(ret); | |
271 | ||
ce7acfea TH |
272 | err_put_congested: |
273 | wb_congested_put(wb_congested); | |
93e6d5d8 | 274 | err_put_css: |
496fb780 | 275 | css_put(&blkcg->css); |
93e6d5d8 | 276 | err_free_blkg: |
15974993 | 277 | blkg_free(new_blkg); |
93e6d5d8 | 278 | return ERR_PTR(ret); |
31e4c28d | 279 | } |
3c96cb32 | 280 | |
86cde6b6 TH |
281 | /** |
282 | * blkg_lookup_create - lookup blkg, try to create one if not there | |
283 | * @blkcg: blkcg of interest | |
284 | * @q: request_queue of interest | |
285 | * | |
286 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
287 | * create one. blkg creation is performed recursively from blkcg_root such |
288 | * that all non-root blkg's have access to the parent blkg. This function | |
289 | * should be called under RCU read lock and @q->queue_lock. | |
86cde6b6 TH |
290 | * |
291 | * Returns pointer to the looked up or created blkg on success, ERR_PTR() | |
292 | * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not | |
293 | * dead and bypassing, returns ERR_PTR(-EBUSY). | |
294 | */ | |
3c798398 TH |
295 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
296 | struct request_queue *q) | |
3c96cb32 | 297 | { |
86cde6b6 TH |
298 | struct blkcg_gq *blkg; |
299 | ||
300 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
301 | lockdep_assert_held(q->queue_lock); | |
302 | ||
3c96cb32 TH |
303 | /* |
304 | * This could be the first entry point of blkcg implementation and | |
305 | * we shouldn't allow anything to go through for a bypassing queue. | |
306 | */ | |
307 | if (unlikely(blk_queue_bypass(q))) | |
3f3299d5 | 308 | return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY); |
86cde6b6 TH |
309 | |
310 | blkg = __blkg_lookup(blkcg, q, true); | |
311 | if (blkg) | |
312 | return blkg; | |
313 | ||
3c547865 TH |
314 | /* |
315 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
316 | * non-root blkgs have access to their parents. | |
317 | */ | |
318 | while (true) { | |
319 | struct blkcg *pos = blkcg; | |
320 | struct blkcg *parent = blkcg_parent(blkcg); | |
321 | ||
322 | while (parent && !__blkg_lookup(parent, q, false)) { | |
323 | pos = parent; | |
324 | parent = blkcg_parent(parent); | |
325 | } | |
326 | ||
327 | blkg = blkg_create(pos, q, NULL); | |
328 | if (pos == blkcg || IS_ERR(blkg)) | |
329 | return blkg; | |
330 | } | |
3c96cb32 | 331 | } |
cd1604fa | 332 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 333 | |
3c798398 | 334 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 335 | { |
3c798398 | 336 | struct blkcg *blkcg = blkg->blkcg; |
f427d909 | 337 | int i; |
03aa264a | 338 | |
27e1f9d1 | 339 | lockdep_assert_held(blkg->q->queue_lock); |
9f13ef67 | 340 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
341 | |
342 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 343 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 344 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 345 | |
f427d909 TH |
346 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
347 | struct blkcg_policy *pol = blkcg_policy[i]; | |
348 | ||
349 | if (blkg->pd[i] && pol->pd_offline_fn) | |
350 | pol->pd_offline_fn(blkg); | |
351 | } | |
352 | blkg->online = false; | |
353 | ||
a637120e | 354 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 355 | list_del_init(&blkg->q_node); |
9f13ef67 | 356 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 357 | |
a637120e TH |
358 | /* |
359 | * Both setting lookup hint to and clearing it from @blkg are done | |
360 | * under queue_lock. If it's not pointing to @blkg now, it never | |
361 | * will. Hint assignment itself can race safely. | |
362 | */ | |
ec6c676a | 363 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
364 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
365 | ||
03aa264a TH |
366 | /* |
367 | * Put the reference taken at the time of creation so that when all | |
368 | * queues are gone, group can be destroyed. | |
369 | */ | |
370 | blkg_put(blkg); | |
371 | } | |
372 | ||
9f13ef67 TH |
373 | /** |
374 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
375 | * @q: request_queue of interest | |
9f13ef67 | 376 | * |
3c96cb32 | 377 | * Destroy all blkgs associated with @q. |
9f13ef67 | 378 | */ |
3c96cb32 | 379 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 380 | { |
3c798398 | 381 | struct blkcg_gq *blkg, *n; |
72e06c25 | 382 | |
6d18b008 | 383 | lockdep_assert_held(q->queue_lock); |
72e06c25 | 384 | |
9f13ef67 | 385 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 386 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 387 | |
9f13ef67 TH |
388 | spin_lock(&blkcg->lock); |
389 | blkg_destroy(blkg); | |
390 | spin_unlock(&blkcg->lock); | |
72e06c25 TH |
391 | } |
392 | } | |
393 | ||
2a4fd070 TH |
394 | /* |
395 | * A group is RCU protected, but having an rcu lock does not mean that one | |
396 | * can access all the fields of blkg and assume these are valid. For | |
397 | * example, don't try to follow throtl_data and request queue links. | |
398 | * | |
399 | * Having a reference to blkg under an rcu allows accesses to only values | |
400 | * local to groups like group stats and group rate limits. | |
401 | */ | |
402 | void __blkg_release_rcu(struct rcu_head *rcu_head) | |
1adaf3dd | 403 | { |
2a4fd070 | 404 | struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head); |
db613670 TH |
405 | int i; |
406 | ||
407 | /* tell policies that this one is being freed */ | |
408 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
409 | struct blkcg_policy *pol = blkcg_policy[i]; | |
410 | ||
411 | if (blkg->pd[i] && pol->pd_exit_fn) | |
412 | pol->pd_exit_fn(blkg); | |
413 | } | |
414 | ||
3c547865 | 415 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
1adaf3dd | 416 | css_put(&blkg->blkcg->css); |
a5049a8a | 417 | if (blkg->parent) |
3c547865 | 418 | blkg_put(blkg->parent); |
1adaf3dd | 419 | |
ce7acfea TH |
420 | wb_congested_put(blkg->wb_congested); |
421 | ||
2a4fd070 | 422 | blkg_free(blkg); |
1adaf3dd | 423 | } |
2a4fd070 | 424 | EXPORT_SYMBOL_GPL(__blkg_release_rcu); |
1adaf3dd | 425 | |
a051661c TH |
426 | /* |
427 | * The next function used by blk_queue_for_each_rl(). It's a bit tricky | |
428 | * because the root blkg uses @q->root_rl instead of its own rl. | |
429 | */ | |
430 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
431 | struct request_queue *q) | |
432 | { | |
433 | struct list_head *ent; | |
434 | struct blkcg_gq *blkg; | |
435 | ||
436 | /* | |
437 | * Determine the current blkg list_head. The first entry is | |
438 | * root_rl which is off @q->blkg_list and mapped to the head. | |
439 | */ | |
440 | if (rl == &q->root_rl) { | |
441 | ent = &q->blkg_list; | |
65c77fd9 JN |
442 | /* There are no more block groups, hence no request lists */ |
443 | if (list_empty(ent)) | |
444 | return NULL; | |
a051661c TH |
445 | } else { |
446 | blkg = container_of(rl, struct blkcg_gq, rl); | |
447 | ent = &blkg->q_node; | |
448 | } | |
449 | ||
450 | /* walk to the next list_head, skip root blkcg */ | |
451 | ent = ent->next; | |
452 | if (ent == &q->root_blkg->q_node) | |
453 | ent = ent->next; | |
454 | if (ent == &q->blkg_list) | |
455 | return NULL; | |
456 | ||
457 | blkg = container_of(ent, struct blkcg_gq, q_node); | |
458 | return &blkg->rl; | |
459 | } | |
460 | ||
182446d0 TH |
461 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
462 | struct cftype *cftype, u64 val) | |
303a3acb | 463 | { |
182446d0 | 464 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 465 | struct blkcg_gq *blkg; |
bc0d6501 | 466 | int i; |
303a3acb | 467 | |
838f13bf | 468 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 469 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
470 | |
471 | /* | |
472 | * Note that stat reset is racy - it doesn't synchronize against | |
473 | * stat updates. This is a debug feature which shouldn't exist | |
474 | * anyway. If you get hit by a race, retry. | |
475 | */ | |
b67bfe0d | 476 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
8bd435b3 | 477 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 478 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 479 | |
a2b1693b | 480 | if (blkcg_policy_enabled(blkg->q, pol) && |
f9fcc2d3 TH |
481 | pol->pd_reset_stats_fn) |
482 | pol->pd_reset_stats_fn(blkg); | |
bc0d6501 | 483 | } |
303a3acb | 484 | } |
f0bdc8cd | 485 | |
303a3acb | 486 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 487 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
488 | return 0; |
489 | } | |
490 | ||
3c798398 | 491 | static const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 492 | { |
d3d32e69 TH |
493 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
494 | if (blkg->q->backing_dev_info.dev) | |
495 | return dev_name(blkg->q->backing_dev_info.dev); | |
496 | return NULL; | |
303a3acb DS |
497 | } |
498 | ||
d3d32e69 TH |
499 | /** |
500 | * blkcg_print_blkgs - helper for printing per-blkg data | |
501 | * @sf: seq_file to print to | |
502 | * @blkcg: blkcg of interest | |
503 | * @prfill: fill function to print out a blkg | |
504 | * @pol: policy in question | |
505 | * @data: data to be passed to @prfill | |
506 | * @show_total: to print out sum of prfill return values or not | |
507 | * | |
508 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
509 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
510 | * policy data and @data and the matching queue lock held. If @show_total |
511 | * is %true, the sum of the return values from @prfill is printed with | |
512 | * "Total" label at the end. | |
d3d32e69 TH |
513 | * |
514 | * This is to be used to construct print functions for | |
515 | * cftype->read_seq_string method. | |
516 | */ | |
3c798398 | 517 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
518 | u64 (*prfill)(struct seq_file *, |
519 | struct blkg_policy_data *, int), | |
3c798398 | 520 | const struct blkcg_policy *pol, int data, |
ec399347 | 521 | bool show_total) |
5624a4e4 | 522 | { |
3c798398 | 523 | struct blkcg_gq *blkg; |
d3d32e69 | 524 | u64 total = 0; |
5624a4e4 | 525 | |
810ecfa7 | 526 | rcu_read_lock(); |
ee89f812 | 527 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
810ecfa7 | 528 | spin_lock_irq(blkg->q->queue_lock); |
a2b1693b | 529 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 530 | total += prfill(sf, blkg->pd[pol->plid], data); |
810ecfa7 TH |
531 | spin_unlock_irq(blkg->q->queue_lock); |
532 | } | |
533 | rcu_read_unlock(); | |
d3d32e69 TH |
534 | |
535 | if (show_total) | |
536 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
537 | } | |
829fdb50 | 538 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
539 | |
540 | /** | |
541 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
542 | * @sf: seq_file to print to | |
f95a04af | 543 | * @pd: policy private data of interest |
d3d32e69 TH |
544 | * @v: value to print |
545 | * | |
f95a04af | 546 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 547 | */ |
f95a04af | 548 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 549 | { |
f95a04af | 550 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
551 | |
552 | if (!dname) | |
553 | return 0; | |
554 | ||
555 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
556 | return v; | |
557 | } | |
829fdb50 | 558 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
559 | |
560 | /** | |
561 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
562 | * @sf: seq_file to print to | |
f95a04af | 563 | * @pd: policy private data of interest |
d3d32e69 TH |
564 | * @rwstat: rwstat to print |
565 | * | |
f95a04af | 566 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 567 | */ |
f95a04af | 568 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 569 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
570 | { |
571 | static const char *rwstr[] = { | |
572 | [BLKG_RWSTAT_READ] = "Read", | |
573 | [BLKG_RWSTAT_WRITE] = "Write", | |
574 | [BLKG_RWSTAT_SYNC] = "Sync", | |
575 | [BLKG_RWSTAT_ASYNC] = "Async", | |
576 | }; | |
f95a04af | 577 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
578 | u64 v; |
579 | int i; | |
580 | ||
581 | if (!dname) | |
582 | return 0; | |
583 | ||
584 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
585 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
586 | (unsigned long long)rwstat->cnt[i]); | |
587 | ||
588 | v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE]; | |
589 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); | |
590 | return v; | |
591 | } | |
b50da39f | 592 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 593 | |
5bc4afb1 TH |
594 | /** |
595 | * blkg_prfill_stat - prfill callback for blkg_stat | |
596 | * @sf: seq_file to print to | |
f95a04af TH |
597 | * @pd: policy private data of interest |
598 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
599 | * |
600 | * prfill callback for printing a blkg_stat. | |
601 | */ | |
f95a04af | 602 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 603 | { |
f95a04af | 604 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 605 | } |
5bc4afb1 | 606 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 607 | |
5bc4afb1 TH |
608 | /** |
609 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
610 | * @sf: seq_file to print to | |
f95a04af TH |
611 | * @pd: policy private data of interest |
612 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
613 | * |
614 | * prfill callback for printing a blkg_rwstat. | |
615 | */ | |
f95a04af TH |
616 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
617 | int off) | |
d3d32e69 | 618 | { |
f95a04af | 619 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 620 | |
f95a04af | 621 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 622 | } |
5bc4afb1 | 623 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 624 | |
16b3de66 TH |
625 | /** |
626 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
627 | * @pd: policy private data of interest | |
628 | * @off: offset to the blkg_stat in @pd | |
629 | * | |
630 | * Collect the blkg_stat specified by @off from @pd and all its online | |
631 | * descendants and return the sum. The caller must be holding the queue | |
632 | * lock for online tests. | |
633 | */ | |
634 | u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off) | |
635 | { | |
636 | struct blkcg_policy *pol = blkcg_policy[pd->plid]; | |
637 | struct blkcg_gq *pos_blkg; | |
492eb21b | 638 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 639 | u64 sum = 0; |
16b3de66 TH |
640 | |
641 | lockdep_assert_held(pd->blkg->q->queue_lock); | |
642 | ||
16b3de66 | 643 | rcu_read_lock(); |
492eb21b | 644 | blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) { |
16b3de66 TH |
645 | struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol); |
646 | struct blkg_stat *stat = (void *)pos_pd + off; | |
647 | ||
648 | if (pos_blkg->online) | |
649 | sum += blkg_stat_read(stat); | |
650 | } | |
651 | rcu_read_unlock(); | |
652 | ||
653 | return sum; | |
654 | } | |
655 | EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum); | |
656 | ||
657 | /** | |
658 | * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat | |
659 | * @pd: policy private data of interest | |
660 | * @off: offset to the blkg_stat in @pd | |
661 | * | |
662 | * Collect the blkg_rwstat specified by @off from @pd and all its online | |
663 | * descendants and return the sum. The caller must be holding the queue | |
664 | * lock for online tests. | |
665 | */ | |
666 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd, | |
667 | int off) | |
668 | { | |
669 | struct blkcg_policy *pol = blkcg_policy[pd->plid]; | |
670 | struct blkcg_gq *pos_blkg; | |
492eb21b | 671 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 672 | struct blkg_rwstat sum = { }; |
16b3de66 TH |
673 | int i; |
674 | ||
675 | lockdep_assert_held(pd->blkg->q->queue_lock); | |
676 | ||
16b3de66 | 677 | rcu_read_lock(); |
492eb21b | 678 | blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) { |
16b3de66 TH |
679 | struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol); |
680 | struct blkg_rwstat *rwstat = (void *)pos_pd + off; | |
681 | struct blkg_rwstat tmp; | |
682 | ||
683 | if (!pos_blkg->online) | |
684 | continue; | |
685 | ||
686 | tmp = blkg_rwstat_read(rwstat); | |
687 | ||
688 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
689 | sum.cnt[i] += tmp.cnt[i]; | |
690 | } | |
691 | rcu_read_unlock(); | |
692 | ||
693 | return sum; | |
694 | } | |
695 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
696 | ||
3a8b31d3 TH |
697 | /** |
698 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
699 | * @blkcg: target block cgroup | |
da8b0662 | 700 | * @pol: target policy |
3a8b31d3 TH |
701 | * @input: input string |
702 | * @ctx: blkg_conf_ctx to be filled | |
703 | * | |
704 | * Parse per-blkg config update from @input and initialize @ctx with the | |
705 | * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new | |
da8b0662 TH |
706 | * value. This function returns with RCU read lock and queue lock held and |
707 | * must be paired with blkg_conf_finish(). | |
3a8b31d3 | 708 | */ |
3c798398 TH |
709 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
710 | const char *input, struct blkg_conf_ctx *ctx) | |
da8b0662 | 711 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
34d0f179 | 712 | { |
3a8b31d3 | 713 | struct gendisk *disk; |
3c798398 | 714 | struct blkcg_gq *blkg; |
726fa694 TH |
715 | unsigned int major, minor; |
716 | unsigned long long v; | |
717 | int part, ret; | |
34d0f179 | 718 | |
726fa694 TH |
719 | if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3) |
720 | return -EINVAL; | |
3a8b31d3 | 721 | |
726fa694 | 722 | disk = get_gendisk(MKDEV(major, minor), &part); |
5f6c2d2b | 723 | if (!disk) |
726fa694 | 724 | return -EINVAL; |
5f6c2d2b TH |
725 | if (part) { |
726 | put_disk(disk); | |
727 | return -EINVAL; | |
728 | } | |
e56da7e2 TH |
729 | |
730 | rcu_read_lock(); | |
4bfd482e | 731 | spin_lock_irq(disk->queue->queue_lock); |
da8b0662 | 732 | |
a2b1693b | 733 | if (blkcg_policy_enabled(disk->queue, pol)) |
3c96cb32 | 734 | blkg = blkg_lookup_create(blkcg, disk->queue); |
a2b1693b TH |
735 | else |
736 | blkg = ERR_PTR(-EINVAL); | |
e56da7e2 | 737 | |
4bfd482e TH |
738 | if (IS_ERR(blkg)) { |
739 | ret = PTR_ERR(blkg); | |
3a8b31d3 | 740 | rcu_read_unlock(); |
da8b0662 | 741 | spin_unlock_irq(disk->queue->queue_lock); |
3a8b31d3 TH |
742 | put_disk(disk); |
743 | /* | |
744 | * If queue was bypassing, we should retry. Do so after a | |
745 | * short msleep(). It isn't strictly necessary but queue | |
746 | * can be bypassing for some time and it's always nice to | |
747 | * avoid busy looping. | |
748 | */ | |
749 | if (ret == -EBUSY) { | |
750 | msleep(10); | |
751 | ret = restart_syscall(); | |
7702e8f4 | 752 | } |
726fa694 | 753 | return ret; |
062a644d | 754 | } |
3a8b31d3 TH |
755 | |
756 | ctx->disk = disk; | |
757 | ctx->blkg = blkg; | |
726fa694 TH |
758 | ctx->v = v; |
759 | return 0; | |
34d0f179 | 760 | } |
829fdb50 | 761 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 762 | |
3a8b31d3 TH |
763 | /** |
764 | * blkg_conf_finish - finish up per-blkg config update | |
765 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
766 | * | |
767 | * Finish up after per-blkg config update. This function must be paired | |
768 | * with blkg_conf_prep(). | |
769 | */ | |
829fdb50 | 770 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
da8b0662 | 771 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 772 | { |
da8b0662 | 773 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
3a8b31d3 TH |
774 | rcu_read_unlock(); |
775 | put_disk(ctx->disk); | |
34d0f179 | 776 | } |
829fdb50 | 777 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 778 | |
3c798398 | 779 | struct cftype blkcg_files[] = { |
84c124da DS |
780 | { |
781 | .name = "reset_stats", | |
3c798398 | 782 | .write_u64 = blkcg_reset_stats, |
22084190 | 783 | }, |
4baf6e33 | 784 | { } /* terminate */ |
31e4c28d VG |
785 | }; |
786 | ||
9f13ef67 | 787 | /** |
92fb9748 | 788 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 789 | * @css: css of interest |
9f13ef67 | 790 | * |
eb95419b TH |
791 | * This function is called when @css is about to go away and responsible |
792 | * for shooting down all blkgs associated with @css. blkgs should be | |
9f13ef67 TH |
793 | * removed while holding both q and blkcg locks. As blkcg lock is nested |
794 | * inside q lock, this function performs reverse double lock dancing. | |
795 | * | |
796 | * This is the blkcg counterpart of ioc_release_fn(). | |
797 | */ | |
eb95419b | 798 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 799 | { |
eb95419b | 800 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 801 | |
9f13ef67 | 802 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 803 | |
9f13ef67 | 804 | while (!hlist_empty(&blkcg->blkg_list)) { |
3c798398 TH |
805 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, |
806 | struct blkcg_gq, blkcg_node); | |
c875f4d0 | 807 | struct request_queue *q = blkg->q; |
b1c35769 | 808 | |
9f13ef67 TH |
809 | if (spin_trylock(q->queue_lock)) { |
810 | blkg_destroy(blkg); | |
811 | spin_unlock(q->queue_lock); | |
812 | } else { | |
813 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 814 | cpu_relax(); |
a5567932 | 815 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 816 | } |
9f13ef67 | 817 | } |
b1c35769 | 818 | |
9f13ef67 | 819 | spin_unlock_irq(&blkcg->lock); |
52ebea74 TH |
820 | |
821 | wb_blkcg_offline(blkcg); | |
7ee9c562 TH |
822 | } |
823 | ||
eb95419b | 824 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 825 | { |
eb95419b | 826 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 827 | int i; |
7ee9c562 | 828 | |
7876f930 TH |
829 | mutex_lock(&blkcg_pol_mutex); |
830 | list_del(&blkcg->all_blkcgs_node); | |
831 | mutex_unlock(&blkcg_pol_mutex); | |
832 | ||
bc915e61 TH |
833 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
834 | kfree(blkcg->pd[i]); | |
835 | kfree(blkcg); | |
31e4c28d VG |
836 | } |
837 | ||
eb95419b TH |
838 | static struct cgroup_subsys_state * |
839 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 840 | { |
3c798398 | 841 | struct blkcg *blkcg; |
e48453c3 AA |
842 | struct cgroup_subsys_state *ret; |
843 | int i; | |
31e4c28d | 844 | |
7876f930 TH |
845 | mutex_lock(&blkcg_pol_mutex); |
846 | ||
eb95419b | 847 | if (!parent_css) { |
3c798398 | 848 | blkcg = &blkcg_root; |
bc915e61 TH |
849 | } else { |
850 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
851 | if (!blkcg) { | |
852 | ret = ERR_PTR(-ENOMEM); | |
853 | goto free_blkcg; | |
854 | } | |
e48453c3 AA |
855 | } |
856 | ||
857 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
858 | struct blkcg_policy *pol = blkcg_policy[i]; | |
859 | struct blkcg_policy_data *cpd; | |
860 | ||
861 | /* | |
862 | * If the policy hasn't been attached yet, wait for it | |
863 | * to be attached before doing anything else. Otherwise, | |
864 | * check if the policy requires any specific per-cgroup | |
865 | * data: if it does, allocate and initialize it. | |
866 | */ | |
867 | if (!pol || !pol->cpd_size) | |
868 | continue; | |
869 | ||
870 | BUG_ON(blkcg->pd[i]); | |
871 | cpd = kzalloc(pol->cpd_size, GFP_KERNEL); | |
872 | if (!cpd) { | |
873 | ret = ERR_PTR(-ENOMEM); | |
874 | goto free_pd_blkcg; | |
875 | } | |
876 | blkcg->pd[i] = cpd; | |
877 | cpd->plid = i; | |
878 | pol->cpd_init_fn(blkcg); | |
879 | } | |
31e4c28d | 880 | |
31e4c28d | 881 | spin_lock_init(&blkcg->lock); |
d93a11f1 | 882 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT); |
31e4c28d | 883 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
884 | #ifdef CONFIG_CGROUP_WRITEBACK |
885 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
886 | #endif | |
7876f930 TH |
887 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
888 | ||
889 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 890 | return &blkcg->css; |
e48453c3 AA |
891 | |
892 | free_pd_blkcg: | |
893 | for (i--; i >= 0; i--) | |
894 | kfree(blkcg->pd[i]); | |
e48453c3 AA |
895 | free_blkcg: |
896 | kfree(blkcg); | |
7876f930 | 897 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 898 | return ret; |
31e4c28d VG |
899 | } |
900 | ||
5efd6113 TH |
901 | /** |
902 | * blkcg_init_queue - initialize blkcg part of request queue | |
903 | * @q: request_queue to initialize | |
904 | * | |
905 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
906 | * part of new request_queue @q. | |
907 | * | |
908 | * RETURNS: | |
909 | * 0 on success, -errno on failure. | |
910 | */ | |
911 | int blkcg_init_queue(struct request_queue *q) | |
912 | { | |
ec13b1d6 TH |
913 | struct blkcg_gq *new_blkg, *blkg; |
914 | bool preloaded; | |
915 | int ret; | |
916 | ||
917 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); | |
918 | if (!new_blkg) | |
919 | return -ENOMEM; | |
920 | ||
921 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
5efd6113 | 922 | |
ec13b1d6 TH |
923 | /* |
924 | * Make sure the root blkg exists and count the existing blkgs. As | |
925 | * @q is bypassing at this point, blkg_lookup_create() can't be | |
926 | * used. Open code insertion. | |
927 | */ | |
928 | rcu_read_lock(); | |
929 | spin_lock_irq(q->queue_lock); | |
930 | blkg = blkg_create(&blkcg_root, q, new_blkg); | |
931 | spin_unlock_irq(q->queue_lock); | |
932 | rcu_read_unlock(); | |
933 | ||
934 | if (preloaded) | |
935 | radix_tree_preload_end(); | |
936 | ||
937 | if (IS_ERR(blkg)) { | |
994b7832 | 938 | blkg_free(new_blkg); |
ec13b1d6 TH |
939 | return PTR_ERR(blkg); |
940 | } | |
941 | ||
942 | q->root_blkg = blkg; | |
943 | q->root_rl.blkg = blkg; | |
5efd6113 | 944 | |
ec13b1d6 TH |
945 | ret = blk_throtl_init(q); |
946 | if (ret) { | |
947 | spin_lock_irq(q->queue_lock); | |
948 | blkg_destroy_all(q); | |
949 | spin_unlock_irq(q->queue_lock); | |
950 | } | |
951 | return ret; | |
5efd6113 TH |
952 | } |
953 | ||
954 | /** | |
955 | * blkcg_drain_queue - drain blkcg part of request_queue | |
956 | * @q: request_queue to drain | |
957 | * | |
958 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
959 | */ | |
960 | void blkcg_drain_queue(struct request_queue *q) | |
961 | { | |
962 | lockdep_assert_held(q->queue_lock); | |
963 | ||
0b462c89 TH |
964 | /* |
965 | * @q could be exiting and already have destroyed all blkgs as | |
966 | * indicated by NULL root_blkg. If so, don't confuse policies. | |
967 | */ | |
968 | if (!q->root_blkg) | |
969 | return; | |
970 | ||
5efd6113 TH |
971 | blk_throtl_drain(q); |
972 | } | |
973 | ||
974 | /** | |
975 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
976 | * @q: request_queue being released | |
977 | * | |
978 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
979 | */ | |
980 | void blkcg_exit_queue(struct request_queue *q) | |
981 | { | |
6d18b008 | 982 | spin_lock_irq(q->queue_lock); |
3c96cb32 | 983 | blkg_destroy_all(q); |
6d18b008 TH |
984 | spin_unlock_irq(q->queue_lock); |
985 | ||
5efd6113 TH |
986 | blk_throtl_exit(q); |
987 | } | |
988 | ||
31e4c28d VG |
989 | /* |
990 | * We cannot support shared io contexts, as we have no mean to support | |
991 | * two tasks with the same ioc in two different groups without major rework | |
992 | * of the main cic data structures. For now we allow a task to change | |
993 | * its cgroup only if it's the only owner of its ioc. | |
994 | */ | |
eb95419b TH |
995 | static int blkcg_can_attach(struct cgroup_subsys_state *css, |
996 | struct cgroup_taskset *tset) | |
31e4c28d | 997 | { |
bb9d97b6 | 998 | struct task_struct *task; |
31e4c28d VG |
999 | struct io_context *ioc; |
1000 | int ret = 0; | |
1001 | ||
1002 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
924f0d9a | 1003 | cgroup_taskset_for_each(task, tset) { |
bb9d97b6 TH |
1004 | task_lock(task); |
1005 | ioc = task->io_context; | |
1006 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1007 | ret = -EINVAL; | |
1008 | task_unlock(task); | |
1009 | if (ret) | |
1010 | break; | |
1011 | } | |
31e4c28d VG |
1012 | return ret; |
1013 | } | |
1014 | ||
073219e9 | 1015 | struct cgroup_subsys blkio_cgrp_subsys = { |
92fb9748 TH |
1016 | .css_alloc = blkcg_css_alloc, |
1017 | .css_offline = blkcg_css_offline, | |
1018 | .css_free = blkcg_css_free, | |
3c798398 | 1019 | .can_attach = blkcg_can_attach, |
5577964e | 1020 | .legacy_cftypes = blkcg_files, |
1ced953b TH |
1021 | #ifdef CONFIG_MEMCG |
1022 | /* | |
1023 | * This ensures that, if available, memcg is automatically enabled | |
1024 | * together on the default hierarchy so that the owner cgroup can | |
1025 | * be retrieved from writeback pages. | |
1026 | */ | |
1027 | .depends_on = 1 << memory_cgrp_id, | |
1028 | #endif | |
676f7c8f | 1029 | }; |
073219e9 | 1030 | EXPORT_SYMBOL_GPL(blkio_cgrp_subsys); |
676f7c8f | 1031 | |
a2b1693b TH |
1032 | /** |
1033 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1034 | * @q: request_queue of interest | |
1035 | * @pol: blkcg policy to activate | |
1036 | * | |
1037 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1038 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1039 | * | |
1040 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1041 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1042 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1043 | * always enough for dereferencing policy data. | |
1044 | * | |
1045 | * The caller is responsible for synchronizing [de]activations and policy | |
1046 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1047 | */ | |
1048 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1049 | const struct blkcg_policy *pol) |
a2b1693b | 1050 | { |
4c55f4f9 | 1051 | struct blkg_policy_data *pd_prealloc = NULL; |
ec13b1d6 | 1052 | struct blkcg_gq *blkg; |
4c55f4f9 | 1053 | int ret; |
a2b1693b TH |
1054 | |
1055 | if (blkcg_policy_enabled(q, pol)) | |
1056 | return 0; | |
1057 | ||
1058 | blk_queue_bypass_start(q); | |
4c55f4f9 TH |
1059 | pd_prealloc: |
1060 | if (!pd_prealloc) { | |
001bea73 | 1061 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node); |
4c55f4f9 | 1062 | if (!pd_prealloc) { |
a2b1693b | 1063 | ret = -ENOMEM; |
4c55f4f9 | 1064 | goto out_bypass_end; |
a2b1693b | 1065 | } |
a2b1693b TH |
1066 | } |
1067 | ||
a2b1693b TH |
1068 | spin_lock_irq(q->queue_lock); |
1069 | ||
1070 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
4c55f4f9 TH |
1071 | struct blkg_policy_data *pd; |
1072 | ||
1073 | if (blkg->pd[pol->plid]) | |
1074 | continue; | |
a2b1693b | 1075 | |
001bea73 | 1076 | pd = pol->pd_alloc_fn(GFP_NOWAIT, q->node); |
4c55f4f9 TH |
1077 | if (!pd) |
1078 | swap(pd, pd_prealloc); | |
1079 | if (!pd) { | |
1080 | spin_unlock_irq(q->queue_lock); | |
1081 | goto pd_prealloc; | |
1082 | } | |
a2b1693b TH |
1083 | |
1084 | blkg->pd[pol->plid] = pd; | |
1085 | pd->blkg = blkg; | |
b276a876 | 1086 | pd->plid = pol->plid; |
3e418710 TH |
1087 | if (pol->pd_init_fn) |
1088 | pol->pd_init_fn(blkg); | |
a2b1693b TH |
1089 | } |
1090 | ||
1091 | __set_bit(pol->plid, q->blkcg_pols); | |
1092 | ret = 0; | |
4c55f4f9 | 1093 | |
a2b1693b | 1094 | spin_unlock_irq(q->queue_lock); |
4c55f4f9 | 1095 | out_bypass_end: |
a2b1693b | 1096 | blk_queue_bypass_end(q); |
001bea73 TH |
1097 | if (pd_prealloc) |
1098 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b TH |
1099 | return ret; |
1100 | } | |
1101 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1102 | ||
1103 | /** | |
1104 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1105 | * @q: request_queue of interest | |
1106 | * @pol: blkcg policy to deactivate | |
1107 | * | |
1108 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1109 | * blkcg_activate_policy(). | |
1110 | */ | |
1111 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1112 | const struct blkcg_policy *pol) |
a2b1693b | 1113 | { |
3c798398 | 1114 | struct blkcg_gq *blkg; |
a2b1693b TH |
1115 | |
1116 | if (!blkcg_policy_enabled(q, pol)) | |
1117 | return; | |
1118 | ||
1119 | blk_queue_bypass_start(q); | |
1120 | spin_lock_irq(q->queue_lock); | |
1121 | ||
1122 | __clear_bit(pol->plid, q->blkcg_pols); | |
1123 | ||
1124 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1125 | /* grab blkcg lock too while removing @pd from @blkg */ | |
1126 | spin_lock(&blkg->blkcg->lock); | |
1127 | ||
f427d909 TH |
1128 | if (pol->pd_offline_fn) |
1129 | pol->pd_offline_fn(blkg); | |
f9fcc2d3 TH |
1130 | if (pol->pd_exit_fn) |
1131 | pol->pd_exit_fn(blkg); | |
a2b1693b | 1132 | |
001bea73 TH |
1133 | if (blkg->pd[pol->plid]) { |
1134 | pol->pd_free_fn(blkg->pd[pol->plid]); | |
1135 | blkg->pd[pol->plid] = NULL; | |
1136 | } | |
a2b1693b TH |
1137 | |
1138 | spin_unlock(&blkg->blkcg->lock); | |
1139 | } | |
1140 | ||
1141 | spin_unlock_irq(q->queue_lock); | |
1142 | blk_queue_bypass_end(q); | |
1143 | } | |
1144 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1145 | ||
8bd435b3 | 1146 | /** |
3c798398 TH |
1147 | * blkcg_policy_register - register a blkcg policy |
1148 | * @pol: blkcg policy to register | |
8bd435b3 | 1149 | * |
3c798398 TH |
1150 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1151 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1152 | */ |
d5bf0291 | 1153 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1154 | { |
06b285bd | 1155 | struct blkcg *blkcg; |
8bd435b3 | 1156 | int i, ret; |
e8989fae | 1157 | |
838f13bf | 1158 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1159 | mutex_lock(&blkcg_pol_mutex); |
1160 | ||
8bd435b3 TH |
1161 | /* find an empty slot */ |
1162 | ret = -ENOSPC; | |
1163 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1164 | if (!blkcg_policy[i]) |
8bd435b3 TH |
1165 | break; |
1166 | if (i >= BLKCG_MAX_POLS) | |
838f13bf | 1167 | goto err_unlock; |
035d10b2 | 1168 | |
06b285bd | 1169 | /* register @pol */ |
3c798398 | 1170 | pol->plid = i; |
06b285bd TH |
1171 | blkcg_policy[pol->plid] = pol; |
1172 | ||
1173 | /* allocate and install cpd's */ | |
1174 | if (pol->cpd_size) { | |
1175 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1176 | struct blkcg_policy_data *cpd; | |
1177 | ||
1178 | cpd = kzalloc(pol->cpd_size, GFP_KERNEL); | |
1179 | if (!cpd) { | |
1180 | mutex_unlock(&blkcg_pol_mutex); | |
1181 | goto err_free_cpds; | |
1182 | } | |
1183 | ||
1184 | blkcg->pd[pol->plid] = cpd; | |
1185 | cpd->plid = pol->plid; | |
1186 | pol->cpd_init_fn(blkcg); | |
1187 | } | |
1188 | } | |
1189 | ||
838f13bf | 1190 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1191 | |
8bd435b3 | 1192 | /* everything is in place, add intf files for the new policy */ |
3c798398 | 1193 | if (pol->cftypes) |
2cf669a5 TH |
1194 | WARN_ON(cgroup_add_legacy_cftypes(&blkio_cgrp_subsys, |
1195 | pol->cftypes)); | |
838f13bf TH |
1196 | mutex_unlock(&blkcg_pol_register_mutex); |
1197 | return 0; | |
1198 | ||
06b285bd TH |
1199 | err_free_cpds: |
1200 | if (pol->cpd_size) { | |
1201 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1202 | kfree(blkcg->pd[pol->plid]); | |
1203 | blkcg->pd[pol->plid] = NULL; | |
1204 | } | |
1205 | } | |
1206 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1207 | err_unlock: |
bc0d6501 | 1208 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1209 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1210 | return ret; |
3e252066 | 1211 | } |
3c798398 | 1212 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1213 | |
8bd435b3 | 1214 | /** |
3c798398 TH |
1215 | * blkcg_policy_unregister - unregister a blkcg policy |
1216 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1217 | * |
3c798398 | 1218 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1219 | */ |
3c798398 | 1220 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1221 | { |
06b285bd TH |
1222 | struct blkcg *blkcg; |
1223 | ||
838f13bf | 1224 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1225 | |
3c798398 | 1226 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1227 | goto out_unlock; |
1228 | ||
1229 | /* kill the intf files first */ | |
3c798398 | 1230 | if (pol->cftypes) |
2bb566cb | 1231 | cgroup_rm_cftypes(pol->cftypes); |
44ea53de | 1232 | |
06b285bd | 1233 | /* remove cpds and unregister */ |
838f13bf | 1234 | mutex_lock(&blkcg_pol_mutex); |
06b285bd TH |
1235 | |
1236 | if (pol->cpd_size) { | |
1237 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1238 | kfree(blkcg->pd[pol->plid]); | |
1239 | blkcg->pd[pol->plid] = NULL; | |
1240 | } | |
1241 | } | |
3c798398 | 1242 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1243 | |
bc0d6501 | 1244 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1245 | out_unlock: |
1246 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1247 | } |
3c798398 | 1248 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |