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 | ||
bc0d6501 | 32 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 33 | |
e48453c3 | 34 | struct blkcg blkcg_root; |
3c798398 | 35 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 36 | |
496d5e75 TH |
37 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
38 | ||
3c798398 | 39 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 40 | |
a2b1693b | 41 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 42 | const struct blkcg_policy *pol) |
a2b1693b TH |
43 | { |
44 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
45 | } | |
46 | ||
0381411e TH |
47 | /** |
48 | * blkg_free - free a blkg | |
49 | * @blkg: blkg to free | |
50 | * | |
51 | * Free @blkg which may be partially allocated. | |
52 | */ | |
3c798398 | 53 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 54 | { |
e8989fae | 55 | int i; |
549d3aa8 TH |
56 | |
57 | if (!blkg) | |
58 | return; | |
59 | ||
db613670 TH |
60 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
61 | kfree(blkg->pd[i]); | |
e8989fae | 62 | |
a051661c | 63 | blk_exit_rl(&blkg->rl); |
549d3aa8 | 64 | kfree(blkg); |
0381411e TH |
65 | } |
66 | ||
67 | /** | |
68 | * blkg_alloc - allocate a blkg | |
69 | * @blkcg: block cgroup the new blkg is associated with | |
70 | * @q: request_queue the new blkg is associated with | |
15974993 | 71 | * @gfp_mask: allocation mask to use |
0381411e | 72 | * |
e8989fae | 73 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 74 | */ |
15974993 TH |
75 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
76 | gfp_t gfp_mask) | |
0381411e | 77 | { |
3c798398 | 78 | struct blkcg_gq *blkg; |
e8989fae | 79 | int i; |
0381411e TH |
80 | |
81 | /* alloc and init base part */ | |
15974993 | 82 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
83 | if (!blkg) |
84 | return NULL; | |
85 | ||
c875f4d0 | 86 | blkg->q = q; |
e8989fae | 87 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 88 | blkg->blkcg = blkcg; |
a5049a8a | 89 | atomic_set(&blkg->refcnt, 1); |
0381411e | 90 | |
a051661c TH |
91 | /* root blkg uses @q->root_rl, init rl only for !root blkgs */ |
92 | if (blkcg != &blkcg_root) { | |
93 | if (blk_init_rl(&blkg->rl, q, gfp_mask)) | |
94 | goto err_free; | |
95 | blkg->rl.blkg = blkg; | |
96 | } | |
97 | ||
8bd435b3 | 98 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 99 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 100 | struct blkg_policy_data *pd; |
0381411e | 101 | |
a2b1693b | 102 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
103 | continue; |
104 | ||
105 | /* alloc per-policy data and attach it to blkg */ | |
15974993 | 106 | pd = kzalloc_node(pol->pd_size, gfp_mask, q->node); |
a051661c TH |
107 | if (!pd) |
108 | goto err_free; | |
549d3aa8 | 109 | |
e8989fae TH |
110 | blkg->pd[i] = pd; |
111 | pd->blkg = blkg; | |
b276a876 | 112 | pd->plid = i; |
e8989fae TH |
113 | } |
114 | ||
0381411e | 115 | return blkg; |
a051661c TH |
116 | |
117 | err_free: | |
118 | blkg_free(blkg); | |
119 | return NULL; | |
0381411e TH |
120 | } |
121 | ||
16b3de66 TH |
122 | /** |
123 | * __blkg_lookup - internal version of blkg_lookup() | |
124 | * @blkcg: blkcg of interest | |
125 | * @q: request_queue of interest | |
126 | * @update_hint: whether to update lookup hint with the result or not | |
127 | * | |
128 | * This is internal version and shouldn't be used by policy | |
129 | * implementations. Looks up blkgs for the @blkcg - @q pair regardless of | |
130 | * @q's bypass state. If @update_hint is %true, the caller should be | |
131 | * holding @q->queue_lock and lookup hint is updated on success. | |
132 | */ | |
dd4a4ffc TH |
133 | struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q, |
134 | bool update_hint) | |
80fd9979 | 135 | { |
3c798398 | 136 | struct blkcg_gq *blkg; |
80fd9979 | 137 | |
a637120e TH |
138 | blkg = rcu_dereference(blkcg->blkg_hint); |
139 | if (blkg && blkg->q == q) | |
140 | return blkg; | |
141 | ||
142 | /* | |
86cde6b6 TH |
143 | * Hint didn't match. Look up from the radix tree. Note that the |
144 | * hint can only be updated under queue_lock as otherwise @blkg | |
145 | * could have already been removed from blkg_tree. The caller is | |
146 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
147 | */ |
148 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
149 | if (blkg && blkg->q == q) { |
150 | if (update_hint) { | |
151 | lockdep_assert_held(q->queue_lock); | |
152 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
153 | } | |
a637120e | 154 | return blkg; |
86cde6b6 | 155 | } |
a637120e | 156 | |
80fd9979 TH |
157 | return NULL; |
158 | } | |
159 | ||
160 | /** | |
161 | * blkg_lookup - lookup blkg for the specified blkcg - q pair | |
162 | * @blkcg: blkcg of interest | |
163 | * @q: request_queue of interest | |
164 | * | |
165 | * Lookup blkg for the @blkcg - @q pair. This function should be called | |
166 | * under RCU read lock and is guaranteed to return %NULL if @q is bypassing | |
167 | * - see blk_queue_bypass_start() for details. | |
168 | */ | |
3c798398 | 169 | struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q) |
80fd9979 TH |
170 | { |
171 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
172 | ||
173 | if (unlikely(blk_queue_bypass(q))) | |
174 | return NULL; | |
86cde6b6 | 175 | return __blkg_lookup(blkcg, q, false); |
80fd9979 TH |
176 | } |
177 | EXPORT_SYMBOL_GPL(blkg_lookup); | |
178 | ||
15974993 TH |
179 | /* |
180 | * If @new_blkg is %NULL, this function tries to allocate a new one as | |
181 | * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return. | |
182 | */ | |
86cde6b6 TH |
183 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
184 | struct request_queue *q, | |
185 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 186 | { |
3c798398 | 187 | struct blkcg_gq *blkg; |
ce7acfea | 188 | struct bdi_writeback_congested *wb_congested; |
f427d909 | 189 | int i, ret; |
5624a4e4 | 190 | |
cd1604fa TH |
191 | WARN_ON_ONCE(!rcu_read_lock_held()); |
192 | lockdep_assert_held(q->queue_lock); | |
193 | ||
7ee9c562 | 194 | /* blkg holds a reference to blkcg */ |
ec903c0c | 195 | if (!css_tryget_online(&blkcg->css)) { |
93e6d5d8 TH |
196 | ret = -EINVAL; |
197 | goto err_free_blkg; | |
15974993 | 198 | } |
cd1604fa | 199 | |
ce7acfea TH |
200 | wb_congested = wb_congested_get_create(&q->backing_dev_info, |
201 | blkcg->css.id, GFP_ATOMIC); | |
202 | if (!wb_congested) { | |
203 | ret = -ENOMEM; | |
204 | goto err_put_css; | |
205 | } | |
206 | ||
496fb780 | 207 | /* allocate */ |
15974993 TH |
208 | if (!new_blkg) { |
209 | new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC); | |
210 | if (unlikely(!new_blkg)) { | |
93e6d5d8 | 211 | ret = -ENOMEM; |
ce7acfea | 212 | goto err_put_congested; |
15974993 TH |
213 | } |
214 | } | |
215 | blkg = new_blkg; | |
ce7acfea | 216 | blkg->wb_congested = wb_congested; |
cd1604fa | 217 | |
db613670 | 218 | /* link parent */ |
3c547865 TH |
219 | if (blkcg_parent(blkcg)) { |
220 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
221 | if (WARN_ON_ONCE(!blkg->parent)) { | |
2423c9c3 | 222 | ret = -EINVAL; |
ce7acfea | 223 | goto err_put_congested; |
3c547865 TH |
224 | } |
225 | blkg_get(blkg->parent); | |
226 | } | |
227 | ||
db613670 TH |
228 | /* invoke per-policy init */ |
229 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
230 | struct blkcg_policy *pol = blkcg_policy[i]; | |
231 | ||
232 | if (blkg->pd[i] && pol->pd_init_fn) | |
233 | pol->pd_init_fn(blkg); | |
234 | } | |
235 | ||
236 | /* insert */ | |
cd1604fa | 237 | spin_lock(&blkcg->lock); |
a637120e TH |
238 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
239 | if (likely(!ret)) { | |
240 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
241 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
242 | |
243 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
244 | struct blkcg_policy *pol = blkcg_policy[i]; | |
245 | ||
246 | if (blkg->pd[i] && pol->pd_online_fn) | |
247 | pol->pd_online_fn(blkg); | |
248 | } | |
a637120e | 249 | } |
f427d909 | 250 | blkg->online = true; |
cd1604fa | 251 | spin_unlock(&blkcg->lock); |
496fb780 | 252 | |
ec13b1d6 | 253 | if (!ret) |
a637120e | 254 | return blkg; |
15974993 | 255 | |
3c547865 TH |
256 | /* @blkg failed fully initialized, use the usual release path */ |
257 | blkg_put(blkg); | |
258 | return ERR_PTR(ret); | |
259 | ||
ce7acfea TH |
260 | err_put_congested: |
261 | wb_congested_put(wb_congested); | |
93e6d5d8 | 262 | err_put_css: |
496fb780 | 263 | css_put(&blkcg->css); |
93e6d5d8 | 264 | err_free_blkg: |
15974993 | 265 | blkg_free(new_blkg); |
93e6d5d8 | 266 | return ERR_PTR(ret); |
31e4c28d | 267 | } |
3c96cb32 | 268 | |
86cde6b6 TH |
269 | /** |
270 | * blkg_lookup_create - lookup blkg, try to create one if not there | |
271 | * @blkcg: blkcg of interest | |
272 | * @q: request_queue of interest | |
273 | * | |
274 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
275 | * create one. blkg creation is performed recursively from blkcg_root such |
276 | * that all non-root blkg's have access to the parent blkg. This function | |
277 | * should be called under RCU read lock and @q->queue_lock. | |
86cde6b6 TH |
278 | * |
279 | * Returns pointer to the looked up or created blkg on success, ERR_PTR() | |
280 | * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not | |
281 | * dead and bypassing, returns ERR_PTR(-EBUSY). | |
282 | */ | |
3c798398 TH |
283 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
284 | struct request_queue *q) | |
3c96cb32 | 285 | { |
86cde6b6 TH |
286 | struct blkcg_gq *blkg; |
287 | ||
288 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
289 | lockdep_assert_held(q->queue_lock); | |
290 | ||
3c96cb32 TH |
291 | /* |
292 | * This could be the first entry point of blkcg implementation and | |
293 | * we shouldn't allow anything to go through for a bypassing queue. | |
294 | */ | |
295 | if (unlikely(blk_queue_bypass(q))) | |
3f3299d5 | 296 | return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY); |
86cde6b6 TH |
297 | |
298 | blkg = __blkg_lookup(blkcg, q, true); | |
299 | if (blkg) | |
300 | return blkg; | |
301 | ||
3c547865 TH |
302 | /* |
303 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
304 | * non-root blkgs have access to their parents. | |
305 | */ | |
306 | while (true) { | |
307 | struct blkcg *pos = blkcg; | |
308 | struct blkcg *parent = blkcg_parent(blkcg); | |
309 | ||
310 | while (parent && !__blkg_lookup(parent, q, false)) { | |
311 | pos = parent; | |
312 | parent = blkcg_parent(parent); | |
313 | } | |
314 | ||
315 | blkg = blkg_create(pos, q, NULL); | |
316 | if (pos == blkcg || IS_ERR(blkg)) | |
317 | return blkg; | |
318 | } | |
3c96cb32 | 319 | } |
cd1604fa | 320 | EXPORT_SYMBOL_GPL(blkg_lookup_create); |
31e4c28d | 321 | |
3c798398 | 322 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 323 | { |
3c798398 | 324 | struct blkcg *blkcg = blkg->blkcg; |
f427d909 | 325 | int i; |
03aa264a | 326 | |
27e1f9d1 | 327 | lockdep_assert_held(blkg->q->queue_lock); |
9f13ef67 | 328 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
329 | |
330 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 331 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 332 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 333 | |
f427d909 TH |
334 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
335 | struct blkcg_policy *pol = blkcg_policy[i]; | |
336 | ||
337 | if (blkg->pd[i] && pol->pd_offline_fn) | |
338 | pol->pd_offline_fn(blkg); | |
339 | } | |
340 | blkg->online = false; | |
341 | ||
a637120e | 342 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 343 | list_del_init(&blkg->q_node); |
9f13ef67 | 344 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 345 | |
a637120e TH |
346 | /* |
347 | * Both setting lookup hint to and clearing it from @blkg are done | |
348 | * under queue_lock. If it's not pointing to @blkg now, it never | |
349 | * will. Hint assignment itself can race safely. | |
350 | */ | |
ec6c676a | 351 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
352 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
353 | ||
03aa264a TH |
354 | /* |
355 | * Put the reference taken at the time of creation so that when all | |
356 | * queues are gone, group can be destroyed. | |
357 | */ | |
358 | blkg_put(blkg); | |
359 | } | |
360 | ||
9f13ef67 TH |
361 | /** |
362 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
363 | * @q: request_queue of interest | |
9f13ef67 | 364 | * |
3c96cb32 | 365 | * Destroy all blkgs associated with @q. |
9f13ef67 | 366 | */ |
3c96cb32 | 367 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 368 | { |
3c798398 | 369 | struct blkcg_gq *blkg, *n; |
72e06c25 | 370 | |
6d18b008 | 371 | lockdep_assert_held(q->queue_lock); |
72e06c25 | 372 | |
9f13ef67 | 373 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 374 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 375 | |
9f13ef67 TH |
376 | spin_lock(&blkcg->lock); |
377 | blkg_destroy(blkg); | |
378 | spin_unlock(&blkcg->lock); | |
72e06c25 TH |
379 | } |
380 | } | |
381 | ||
2a4fd070 TH |
382 | /* |
383 | * A group is RCU protected, but having an rcu lock does not mean that one | |
384 | * can access all the fields of blkg and assume these are valid. For | |
385 | * example, don't try to follow throtl_data and request queue links. | |
386 | * | |
387 | * Having a reference to blkg under an rcu allows accesses to only values | |
388 | * local to groups like group stats and group rate limits. | |
389 | */ | |
390 | void __blkg_release_rcu(struct rcu_head *rcu_head) | |
1adaf3dd | 391 | { |
2a4fd070 | 392 | struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head); |
db613670 TH |
393 | int i; |
394 | ||
395 | /* tell policies that this one is being freed */ | |
396 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
397 | struct blkcg_policy *pol = blkcg_policy[i]; | |
398 | ||
399 | if (blkg->pd[i] && pol->pd_exit_fn) | |
400 | pol->pd_exit_fn(blkg); | |
401 | } | |
402 | ||
3c547865 | 403 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
1adaf3dd | 404 | css_put(&blkg->blkcg->css); |
a5049a8a | 405 | if (blkg->parent) |
3c547865 | 406 | blkg_put(blkg->parent); |
1adaf3dd | 407 | |
ce7acfea TH |
408 | wb_congested_put(blkg->wb_congested); |
409 | ||
2a4fd070 | 410 | blkg_free(blkg); |
1adaf3dd | 411 | } |
2a4fd070 | 412 | EXPORT_SYMBOL_GPL(__blkg_release_rcu); |
1adaf3dd | 413 | |
a051661c TH |
414 | /* |
415 | * The next function used by blk_queue_for_each_rl(). It's a bit tricky | |
416 | * because the root blkg uses @q->root_rl instead of its own rl. | |
417 | */ | |
418 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
419 | struct request_queue *q) | |
420 | { | |
421 | struct list_head *ent; | |
422 | struct blkcg_gq *blkg; | |
423 | ||
424 | /* | |
425 | * Determine the current blkg list_head. The first entry is | |
426 | * root_rl which is off @q->blkg_list and mapped to the head. | |
427 | */ | |
428 | if (rl == &q->root_rl) { | |
429 | ent = &q->blkg_list; | |
65c77fd9 JN |
430 | /* There are no more block groups, hence no request lists */ |
431 | if (list_empty(ent)) | |
432 | return NULL; | |
a051661c TH |
433 | } else { |
434 | blkg = container_of(rl, struct blkcg_gq, rl); | |
435 | ent = &blkg->q_node; | |
436 | } | |
437 | ||
438 | /* walk to the next list_head, skip root blkcg */ | |
439 | ent = ent->next; | |
440 | if (ent == &q->root_blkg->q_node) | |
441 | ent = ent->next; | |
442 | if (ent == &q->blkg_list) | |
443 | return NULL; | |
444 | ||
445 | blkg = container_of(ent, struct blkcg_gq, q_node); | |
446 | return &blkg->rl; | |
447 | } | |
448 | ||
182446d0 TH |
449 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
450 | struct cftype *cftype, u64 val) | |
303a3acb | 451 | { |
182446d0 | 452 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 453 | struct blkcg_gq *blkg; |
bc0d6501 | 454 | int i; |
303a3acb | 455 | |
36c38fb7 TH |
456 | /* |
457 | * XXX: We invoke cgroup_add/rm_cftypes() under blkcg_pol_mutex | |
458 | * which ends up putting cgroup's internal cgroup_tree_mutex under | |
459 | * it; however, cgroup_tree_mutex is nested above cgroup file | |
460 | * active protection and grabbing blkcg_pol_mutex from a cgroup | |
461 | * file operation creates a possible circular dependency. cgroup | |
462 | * internal locking is planned to go through further simplification | |
463 | * and this issue should go away soon. For now, let's trylock | |
464 | * blkcg_pol_mutex and restart the write on failure. | |
465 | * | |
466 | * http://lkml.kernel.org/g/5363C04B.4010400@oracle.com | |
467 | */ | |
468 | if (!mutex_trylock(&blkcg_pol_mutex)) | |
469 | return restart_syscall(); | |
303a3acb | 470 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
471 | |
472 | /* | |
473 | * Note that stat reset is racy - it doesn't synchronize against | |
474 | * stat updates. This is a debug feature which shouldn't exist | |
475 | * anyway. If you get hit by a race, retry. | |
476 | */ | |
b67bfe0d | 477 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
8bd435b3 | 478 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 479 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 480 | |
a2b1693b | 481 | if (blkcg_policy_enabled(blkg->q, pol) && |
f9fcc2d3 TH |
482 | pol->pd_reset_stats_fn) |
483 | pol->pd_reset_stats_fn(blkg); | |
bc0d6501 | 484 | } |
303a3acb | 485 | } |
f0bdc8cd | 486 | |
303a3acb | 487 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 488 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
489 | return 0; |
490 | } | |
491 | ||
3c798398 | 492 | static const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 493 | { |
d3d32e69 TH |
494 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
495 | if (blkg->q->backing_dev_info.dev) | |
496 | return dev_name(blkg->q->backing_dev_info.dev); | |
497 | return NULL; | |
303a3acb DS |
498 | } |
499 | ||
d3d32e69 TH |
500 | /** |
501 | * blkcg_print_blkgs - helper for printing per-blkg data | |
502 | * @sf: seq_file to print to | |
503 | * @blkcg: blkcg of interest | |
504 | * @prfill: fill function to print out a blkg | |
505 | * @pol: policy in question | |
506 | * @data: data to be passed to @prfill | |
507 | * @show_total: to print out sum of prfill return values or not | |
508 | * | |
509 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
510 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
511 | * policy data and @data and the matching queue lock held. If @show_total |
512 | * is %true, the sum of the return values from @prfill is printed with | |
513 | * "Total" label at the end. | |
d3d32e69 TH |
514 | * |
515 | * This is to be used to construct print functions for | |
516 | * cftype->read_seq_string method. | |
517 | */ | |
3c798398 | 518 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
519 | u64 (*prfill)(struct seq_file *, |
520 | struct blkg_policy_data *, int), | |
3c798398 | 521 | const struct blkcg_policy *pol, int data, |
ec399347 | 522 | bool show_total) |
5624a4e4 | 523 | { |
3c798398 | 524 | struct blkcg_gq *blkg; |
d3d32e69 | 525 | u64 total = 0; |
5624a4e4 | 526 | |
810ecfa7 | 527 | rcu_read_lock(); |
ee89f812 | 528 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
810ecfa7 | 529 | spin_lock_irq(blkg->q->queue_lock); |
a2b1693b | 530 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 531 | total += prfill(sf, blkg->pd[pol->plid], data); |
810ecfa7 TH |
532 | spin_unlock_irq(blkg->q->queue_lock); |
533 | } | |
534 | rcu_read_unlock(); | |
d3d32e69 TH |
535 | |
536 | if (show_total) | |
537 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
538 | } | |
829fdb50 | 539 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
540 | |
541 | /** | |
542 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
543 | * @sf: seq_file to print to | |
f95a04af | 544 | * @pd: policy private data of interest |
d3d32e69 TH |
545 | * @v: value to print |
546 | * | |
f95a04af | 547 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 548 | */ |
f95a04af | 549 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 550 | { |
f95a04af | 551 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
552 | |
553 | if (!dname) | |
554 | return 0; | |
555 | ||
556 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
557 | return v; | |
558 | } | |
829fdb50 | 559 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
560 | |
561 | /** | |
562 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
563 | * @sf: seq_file to print to | |
f95a04af | 564 | * @pd: policy private data of interest |
d3d32e69 TH |
565 | * @rwstat: rwstat to print |
566 | * | |
f95a04af | 567 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 568 | */ |
f95a04af | 569 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 570 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
571 | { |
572 | static const char *rwstr[] = { | |
573 | [BLKG_RWSTAT_READ] = "Read", | |
574 | [BLKG_RWSTAT_WRITE] = "Write", | |
575 | [BLKG_RWSTAT_SYNC] = "Sync", | |
576 | [BLKG_RWSTAT_ASYNC] = "Async", | |
577 | }; | |
f95a04af | 578 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
579 | u64 v; |
580 | int i; | |
581 | ||
582 | if (!dname) | |
583 | return 0; | |
584 | ||
585 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
586 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
587 | (unsigned long long)rwstat->cnt[i]); | |
588 | ||
589 | v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE]; | |
590 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); | |
591 | return v; | |
592 | } | |
b50da39f | 593 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 594 | |
5bc4afb1 TH |
595 | /** |
596 | * blkg_prfill_stat - prfill callback for blkg_stat | |
597 | * @sf: seq_file to print to | |
f95a04af TH |
598 | * @pd: policy private data of interest |
599 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
600 | * |
601 | * prfill callback for printing a blkg_stat. | |
602 | */ | |
f95a04af | 603 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 604 | { |
f95a04af | 605 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 606 | } |
5bc4afb1 | 607 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 608 | |
5bc4afb1 TH |
609 | /** |
610 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
611 | * @sf: seq_file to print to | |
f95a04af TH |
612 | * @pd: policy private data of interest |
613 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
614 | * |
615 | * prfill callback for printing a blkg_rwstat. | |
616 | */ | |
f95a04af TH |
617 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
618 | int off) | |
d3d32e69 | 619 | { |
f95a04af | 620 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 621 | |
f95a04af | 622 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 623 | } |
5bc4afb1 | 624 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 625 | |
16b3de66 TH |
626 | /** |
627 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
628 | * @pd: policy private data of interest | |
629 | * @off: offset to the blkg_stat in @pd | |
630 | * | |
631 | * Collect the blkg_stat specified by @off from @pd and all its online | |
632 | * descendants and return the sum. The caller must be holding the queue | |
633 | * lock for online tests. | |
634 | */ | |
635 | u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off) | |
636 | { | |
637 | struct blkcg_policy *pol = blkcg_policy[pd->plid]; | |
638 | struct blkcg_gq *pos_blkg; | |
492eb21b | 639 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 640 | u64 sum = 0; |
16b3de66 TH |
641 | |
642 | lockdep_assert_held(pd->blkg->q->queue_lock); | |
643 | ||
16b3de66 | 644 | rcu_read_lock(); |
492eb21b | 645 | blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) { |
16b3de66 TH |
646 | struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol); |
647 | struct blkg_stat *stat = (void *)pos_pd + off; | |
648 | ||
649 | if (pos_blkg->online) | |
650 | sum += blkg_stat_read(stat); | |
651 | } | |
652 | rcu_read_unlock(); | |
653 | ||
654 | return sum; | |
655 | } | |
656 | EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum); | |
657 | ||
658 | /** | |
659 | * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat | |
660 | * @pd: policy private data of interest | |
661 | * @off: offset to the blkg_stat in @pd | |
662 | * | |
663 | * Collect the blkg_rwstat specified by @off from @pd and all its online | |
664 | * descendants and return the sum. The caller must be holding the queue | |
665 | * lock for online tests. | |
666 | */ | |
667 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd, | |
668 | int off) | |
669 | { | |
670 | struct blkcg_policy *pol = blkcg_policy[pd->plid]; | |
671 | struct blkcg_gq *pos_blkg; | |
492eb21b | 672 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 673 | struct blkg_rwstat sum = { }; |
16b3de66 TH |
674 | int i; |
675 | ||
676 | lockdep_assert_held(pd->blkg->q->queue_lock); | |
677 | ||
16b3de66 | 678 | rcu_read_lock(); |
492eb21b | 679 | blkg_for_each_descendant_pre(pos_blkg, pos_css, pd_to_blkg(pd)) { |
16b3de66 TH |
680 | struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol); |
681 | struct blkg_rwstat *rwstat = (void *)pos_pd + off; | |
682 | struct blkg_rwstat tmp; | |
683 | ||
684 | if (!pos_blkg->online) | |
685 | continue; | |
686 | ||
687 | tmp = blkg_rwstat_read(rwstat); | |
688 | ||
689 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
690 | sum.cnt[i] += tmp.cnt[i]; | |
691 | } | |
692 | rcu_read_unlock(); | |
693 | ||
694 | return sum; | |
695 | } | |
696 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
697 | ||
3a8b31d3 TH |
698 | /** |
699 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
700 | * @blkcg: target block cgroup | |
da8b0662 | 701 | * @pol: target policy |
3a8b31d3 TH |
702 | * @input: input string |
703 | * @ctx: blkg_conf_ctx to be filled | |
704 | * | |
705 | * Parse per-blkg config update from @input and initialize @ctx with the | |
706 | * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new | |
da8b0662 TH |
707 | * value. This function returns with RCU read lock and queue lock held and |
708 | * must be paired with blkg_conf_finish(). | |
3a8b31d3 | 709 | */ |
3c798398 TH |
710 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
711 | const char *input, struct blkg_conf_ctx *ctx) | |
da8b0662 | 712 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
34d0f179 | 713 | { |
3a8b31d3 | 714 | struct gendisk *disk; |
3c798398 | 715 | struct blkcg_gq *blkg; |
726fa694 TH |
716 | unsigned int major, minor; |
717 | unsigned long long v; | |
718 | int part, ret; | |
34d0f179 | 719 | |
726fa694 TH |
720 | if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3) |
721 | return -EINVAL; | |
3a8b31d3 | 722 | |
726fa694 | 723 | disk = get_gendisk(MKDEV(major, minor), &part); |
4bfd482e | 724 | if (!disk || part) |
726fa694 | 725 | return -EINVAL; |
e56da7e2 TH |
726 | |
727 | rcu_read_lock(); | |
4bfd482e | 728 | spin_lock_irq(disk->queue->queue_lock); |
da8b0662 | 729 | |
a2b1693b | 730 | if (blkcg_policy_enabled(disk->queue, pol)) |
3c96cb32 | 731 | blkg = blkg_lookup_create(blkcg, disk->queue); |
a2b1693b TH |
732 | else |
733 | blkg = ERR_PTR(-EINVAL); | |
e56da7e2 | 734 | |
4bfd482e TH |
735 | if (IS_ERR(blkg)) { |
736 | ret = PTR_ERR(blkg); | |
3a8b31d3 | 737 | rcu_read_unlock(); |
da8b0662 | 738 | spin_unlock_irq(disk->queue->queue_lock); |
3a8b31d3 TH |
739 | put_disk(disk); |
740 | /* | |
741 | * If queue was bypassing, we should retry. Do so after a | |
742 | * short msleep(). It isn't strictly necessary but queue | |
743 | * can be bypassing for some time and it's always nice to | |
744 | * avoid busy looping. | |
745 | */ | |
746 | if (ret == -EBUSY) { | |
747 | msleep(10); | |
748 | ret = restart_syscall(); | |
7702e8f4 | 749 | } |
726fa694 | 750 | return ret; |
062a644d | 751 | } |
3a8b31d3 TH |
752 | |
753 | ctx->disk = disk; | |
754 | ctx->blkg = blkg; | |
726fa694 TH |
755 | ctx->v = v; |
756 | return 0; | |
34d0f179 | 757 | } |
829fdb50 | 758 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 759 | |
3a8b31d3 TH |
760 | /** |
761 | * blkg_conf_finish - finish up per-blkg config update | |
762 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
763 | * | |
764 | * Finish up after per-blkg config update. This function must be paired | |
765 | * with blkg_conf_prep(). | |
766 | */ | |
829fdb50 | 767 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
da8b0662 | 768 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 769 | { |
da8b0662 | 770 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
3a8b31d3 TH |
771 | rcu_read_unlock(); |
772 | put_disk(ctx->disk); | |
34d0f179 | 773 | } |
829fdb50 | 774 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 775 | |
3c798398 | 776 | struct cftype blkcg_files[] = { |
84c124da DS |
777 | { |
778 | .name = "reset_stats", | |
3c798398 | 779 | .write_u64 = blkcg_reset_stats, |
22084190 | 780 | }, |
4baf6e33 | 781 | { } /* terminate */ |
31e4c28d VG |
782 | }; |
783 | ||
9f13ef67 | 784 | /** |
92fb9748 | 785 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 786 | * @css: css of interest |
9f13ef67 | 787 | * |
eb95419b TH |
788 | * This function is called when @css is about to go away and responsible |
789 | * for shooting down all blkgs associated with @css. blkgs should be | |
9f13ef67 TH |
790 | * removed while holding both q and blkcg locks. As blkcg lock is nested |
791 | * inside q lock, this function performs reverse double lock dancing. | |
792 | * | |
793 | * This is the blkcg counterpart of ioc_release_fn(). | |
794 | */ | |
eb95419b | 795 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 796 | { |
eb95419b | 797 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 798 | |
9f13ef67 | 799 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 800 | |
9f13ef67 | 801 | while (!hlist_empty(&blkcg->blkg_list)) { |
3c798398 TH |
802 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, |
803 | struct blkcg_gq, blkcg_node); | |
c875f4d0 | 804 | struct request_queue *q = blkg->q; |
b1c35769 | 805 | |
9f13ef67 TH |
806 | if (spin_trylock(q->queue_lock)) { |
807 | blkg_destroy(blkg); | |
808 | spin_unlock(q->queue_lock); | |
809 | } else { | |
810 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 811 | cpu_relax(); |
a5567932 | 812 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 813 | } |
9f13ef67 | 814 | } |
b1c35769 | 815 | |
9f13ef67 | 816 | spin_unlock_irq(&blkcg->lock); |
52ebea74 TH |
817 | |
818 | wb_blkcg_offline(blkcg); | |
7ee9c562 TH |
819 | } |
820 | ||
eb95419b | 821 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 822 | { |
eb95419b | 823 | struct blkcg *blkcg = css_to_blkcg(css); |
7ee9c562 | 824 | |
3c798398 | 825 | if (blkcg != &blkcg_root) |
67523c48 | 826 | kfree(blkcg); |
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 | ||
847 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
848 | struct blkcg_policy *pol = blkcg_policy[i]; | |
849 | struct blkcg_policy_data *cpd; | |
850 | ||
851 | /* | |
852 | * If the policy hasn't been attached yet, wait for it | |
853 | * to be attached before doing anything else. Otherwise, | |
854 | * check if the policy requires any specific per-cgroup | |
855 | * data: if it does, allocate and initialize it. | |
856 | */ | |
857 | if (!pol || !pol->cpd_size) | |
858 | continue; | |
859 | ||
860 | BUG_ON(blkcg->pd[i]); | |
861 | cpd = kzalloc(pol->cpd_size, GFP_KERNEL); | |
862 | if (!cpd) { | |
863 | ret = ERR_PTR(-ENOMEM); | |
864 | goto free_pd_blkcg; | |
865 | } | |
866 | blkcg->pd[i] = cpd; | |
867 | cpd->plid = i; | |
868 | pol->cpd_init_fn(blkcg); | |
869 | } | |
31e4c28d | 870 | |
31e4c28d VG |
871 | done: |
872 | spin_lock_init(&blkcg->lock); | |
a637120e | 873 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC); |
31e4c28d | 874 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
875 | #ifdef CONFIG_CGROUP_WRITEBACK |
876 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
877 | #endif | |
31e4c28d | 878 | return &blkcg->css; |
e48453c3 AA |
879 | |
880 | free_pd_blkcg: | |
881 | for (i--; i >= 0; i--) | |
882 | kfree(blkcg->pd[i]); | |
883 | ||
884 | free_blkcg: | |
885 | kfree(blkcg); | |
886 | return ret; | |
31e4c28d VG |
887 | } |
888 | ||
5efd6113 TH |
889 | /** |
890 | * blkcg_init_queue - initialize blkcg part of request queue | |
891 | * @q: request_queue to initialize | |
892 | * | |
893 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
894 | * part of new request_queue @q. | |
895 | * | |
896 | * RETURNS: | |
897 | * 0 on success, -errno on failure. | |
898 | */ | |
899 | int blkcg_init_queue(struct request_queue *q) | |
900 | { | |
ec13b1d6 TH |
901 | struct blkcg_gq *new_blkg, *blkg; |
902 | bool preloaded; | |
903 | int ret; | |
904 | ||
905 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); | |
906 | if (!new_blkg) | |
907 | return -ENOMEM; | |
908 | ||
909 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
5efd6113 | 910 | |
ec13b1d6 TH |
911 | /* |
912 | * Make sure the root blkg exists and count the existing blkgs. As | |
913 | * @q is bypassing at this point, blkg_lookup_create() can't be | |
914 | * used. Open code insertion. | |
915 | */ | |
916 | rcu_read_lock(); | |
917 | spin_lock_irq(q->queue_lock); | |
918 | blkg = blkg_create(&blkcg_root, q, new_blkg); | |
919 | spin_unlock_irq(q->queue_lock); | |
920 | rcu_read_unlock(); | |
921 | ||
922 | if (preloaded) | |
923 | radix_tree_preload_end(); | |
924 | ||
925 | if (IS_ERR(blkg)) { | |
926 | kfree(new_blkg); | |
927 | return PTR_ERR(blkg); | |
928 | } | |
929 | ||
930 | q->root_blkg = blkg; | |
931 | q->root_rl.blkg = blkg; | |
5efd6113 | 932 | |
ec13b1d6 TH |
933 | ret = blk_throtl_init(q); |
934 | if (ret) { | |
935 | spin_lock_irq(q->queue_lock); | |
936 | blkg_destroy_all(q); | |
937 | spin_unlock_irq(q->queue_lock); | |
938 | } | |
939 | return ret; | |
5efd6113 TH |
940 | } |
941 | ||
942 | /** | |
943 | * blkcg_drain_queue - drain blkcg part of request_queue | |
944 | * @q: request_queue to drain | |
945 | * | |
946 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
947 | */ | |
948 | void blkcg_drain_queue(struct request_queue *q) | |
949 | { | |
950 | lockdep_assert_held(q->queue_lock); | |
951 | ||
0b462c89 TH |
952 | /* |
953 | * @q could be exiting and already have destroyed all blkgs as | |
954 | * indicated by NULL root_blkg. If so, don't confuse policies. | |
955 | */ | |
956 | if (!q->root_blkg) | |
957 | return; | |
958 | ||
5efd6113 TH |
959 | blk_throtl_drain(q); |
960 | } | |
961 | ||
962 | /** | |
963 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
964 | * @q: request_queue being released | |
965 | * | |
966 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
967 | */ | |
968 | void blkcg_exit_queue(struct request_queue *q) | |
969 | { | |
6d18b008 | 970 | spin_lock_irq(q->queue_lock); |
3c96cb32 | 971 | blkg_destroy_all(q); |
6d18b008 TH |
972 | spin_unlock_irq(q->queue_lock); |
973 | ||
5efd6113 TH |
974 | blk_throtl_exit(q); |
975 | } | |
976 | ||
31e4c28d VG |
977 | /* |
978 | * We cannot support shared io contexts, as we have no mean to support | |
979 | * two tasks with the same ioc in two different groups without major rework | |
980 | * of the main cic data structures. For now we allow a task to change | |
981 | * its cgroup only if it's the only owner of its ioc. | |
982 | */ | |
eb95419b TH |
983 | static int blkcg_can_attach(struct cgroup_subsys_state *css, |
984 | struct cgroup_taskset *tset) | |
31e4c28d | 985 | { |
bb9d97b6 | 986 | struct task_struct *task; |
31e4c28d VG |
987 | struct io_context *ioc; |
988 | int ret = 0; | |
989 | ||
990 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
924f0d9a | 991 | cgroup_taskset_for_each(task, tset) { |
bb9d97b6 TH |
992 | task_lock(task); |
993 | ioc = task->io_context; | |
994 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
995 | ret = -EINVAL; | |
996 | task_unlock(task); | |
997 | if (ret) | |
998 | break; | |
999 | } | |
31e4c28d VG |
1000 | return ret; |
1001 | } | |
1002 | ||
073219e9 | 1003 | struct cgroup_subsys blkio_cgrp_subsys = { |
92fb9748 TH |
1004 | .css_alloc = blkcg_css_alloc, |
1005 | .css_offline = blkcg_css_offline, | |
1006 | .css_free = blkcg_css_free, | |
3c798398 | 1007 | .can_attach = blkcg_can_attach, |
5577964e | 1008 | .legacy_cftypes = blkcg_files, |
1ced953b TH |
1009 | #ifdef CONFIG_MEMCG |
1010 | /* | |
1011 | * This ensures that, if available, memcg is automatically enabled | |
1012 | * together on the default hierarchy so that the owner cgroup can | |
1013 | * be retrieved from writeback pages. | |
1014 | */ | |
1015 | .depends_on = 1 << memory_cgrp_id, | |
1016 | #endif | |
676f7c8f | 1017 | }; |
073219e9 | 1018 | EXPORT_SYMBOL_GPL(blkio_cgrp_subsys); |
676f7c8f | 1019 | |
a2b1693b TH |
1020 | /** |
1021 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1022 | * @q: request_queue of interest | |
1023 | * @pol: blkcg policy to activate | |
1024 | * | |
1025 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1026 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1027 | * | |
1028 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1029 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1030 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1031 | * always enough for dereferencing policy data. | |
1032 | * | |
1033 | * The caller is responsible for synchronizing [de]activations and policy | |
1034 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1035 | */ | |
1036 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1037 | const struct blkcg_policy *pol) |
a2b1693b TH |
1038 | { |
1039 | LIST_HEAD(pds); | |
e48453c3 | 1040 | LIST_HEAD(cpds); |
ec13b1d6 | 1041 | struct blkcg_gq *blkg; |
e48453c3 AA |
1042 | struct blkg_policy_data *pd, *nd; |
1043 | struct blkcg_policy_data *cpd, *cnd; | |
a2b1693b TH |
1044 | int cnt = 0, ret; |
1045 | ||
1046 | if (blkcg_policy_enabled(q, pol)) | |
1047 | return 0; | |
1048 | ||
ec13b1d6 | 1049 | /* count and allocate policy_data for all existing blkgs */ |
a2b1693b | 1050 | blk_queue_bypass_start(q); |
a2b1693b | 1051 | spin_lock_irq(q->queue_lock); |
a2b1693b TH |
1052 | list_for_each_entry(blkg, &q->blkg_list, q_node) |
1053 | cnt++; | |
a2b1693b TH |
1054 | spin_unlock_irq(q->queue_lock); |
1055 | ||
e48453c3 AA |
1056 | /* |
1057 | * Allocate per-blkg and per-blkcg policy data | |
1058 | * for all existing blkgs. | |
1059 | */ | |
a2b1693b | 1060 | while (cnt--) { |
f95a04af | 1061 | pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node); |
a2b1693b TH |
1062 | if (!pd) { |
1063 | ret = -ENOMEM; | |
1064 | goto out_free; | |
1065 | } | |
1066 | list_add_tail(&pd->alloc_node, &pds); | |
e48453c3 AA |
1067 | |
1068 | if (!pol->cpd_size) | |
1069 | continue; | |
1070 | cpd = kzalloc_node(pol->cpd_size, GFP_KERNEL, q->node); | |
1071 | if (!cpd) { | |
1072 | ret = -ENOMEM; | |
1073 | goto out_free; | |
1074 | } | |
1075 | list_add_tail(&cpd->alloc_node, &cpds); | |
a2b1693b TH |
1076 | } |
1077 | ||
1078 | /* | |
e48453c3 | 1079 | * Install the allocated pds and cpds. With @q bypassing, no new blkg |
a2b1693b TH |
1080 | * should have been created while the queue lock was dropped. |
1081 | */ | |
1082 | spin_lock_irq(q->queue_lock); | |
1083 | ||
1084 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
e48453c3 AA |
1085 | if (WARN_ON(list_empty(&pds)) || |
1086 | WARN_ON(pol->cpd_size && list_empty(&cpds))) { | |
a2b1693b TH |
1087 | /* umm... this shouldn't happen, just abort */ |
1088 | ret = -ENOMEM; | |
1089 | goto out_unlock; | |
1090 | } | |
e48453c3 AA |
1091 | cpd = list_first_entry(&cpds, struct blkcg_policy_data, |
1092 | alloc_node); | |
1093 | list_del_init(&cpd->alloc_node); | |
a2b1693b TH |
1094 | pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node); |
1095 | list_del_init(&pd->alloc_node); | |
1096 | ||
1097 | /* grab blkcg lock too while installing @pd on @blkg */ | |
1098 | spin_lock(&blkg->blkcg->lock); | |
1099 | ||
e48453c3 AA |
1100 | if (!pol->cpd_size) |
1101 | goto no_cpd; | |
1102 | if (!blkg->blkcg->pd[pol->plid]) { | |
1103 | /* Per-policy per-blkcg data */ | |
1104 | blkg->blkcg->pd[pol->plid] = cpd; | |
1105 | cpd->plid = pol->plid; | |
1106 | pol->cpd_init_fn(blkg->blkcg); | |
1107 | } else { /* must free it as it has already been extracted */ | |
1108 | kfree(cpd); | |
1109 | } | |
1110 | no_cpd: | |
a2b1693b TH |
1111 | blkg->pd[pol->plid] = pd; |
1112 | pd->blkg = blkg; | |
b276a876 | 1113 | pd->plid = pol->plid; |
f9fcc2d3 | 1114 | pol->pd_init_fn(blkg); |
a2b1693b TH |
1115 | |
1116 | spin_unlock(&blkg->blkcg->lock); | |
1117 | } | |
1118 | ||
1119 | __set_bit(pol->plid, q->blkcg_pols); | |
1120 | ret = 0; | |
1121 | out_unlock: | |
1122 | spin_unlock_irq(q->queue_lock); | |
1123 | out_free: | |
1124 | blk_queue_bypass_end(q); | |
e48453c3 | 1125 | list_for_each_entry_safe(pd, nd, &pds, alloc_node) |
a2b1693b | 1126 | kfree(pd); |
e48453c3 AA |
1127 | list_for_each_entry_safe(cpd, cnd, &cpds, alloc_node) |
1128 | kfree(cpd); | |
a2b1693b TH |
1129 | return ret; |
1130 | } | |
1131 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1132 | ||
1133 | /** | |
1134 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1135 | * @q: request_queue of interest | |
1136 | * @pol: blkcg policy to deactivate | |
1137 | * | |
1138 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1139 | * blkcg_activate_policy(). | |
1140 | */ | |
1141 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1142 | const struct blkcg_policy *pol) |
a2b1693b | 1143 | { |
3c798398 | 1144 | struct blkcg_gq *blkg; |
a2b1693b TH |
1145 | |
1146 | if (!blkcg_policy_enabled(q, pol)) | |
1147 | return; | |
1148 | ||
1149 | blk_queue_bypass_start(q); | |
1150 | spin_lock_irq(q->queue_lock); | |
1151 | ||
1152 | __clear_bit(pol->plid, q->blkcg_pols); | |
1153 | ||
1154 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1155 | /* grab blkcg lock too while removing @pd from @blkg */ | |
1156 | spin_lock(&blkg->blkcg->lock); | |
1157 | ||
f427d909 TH |
1158 | if (pol->pd_offline_fn) |
1159 | pol->pd_offline_fn(blkg); | |
f9fcc2d3 TH |
1160 | if (pol->pd_exit_fn) |
1161 | pol->pd_exit_fn(blkg); | |
a2b1693b TH |
1162 | |
1163 | kfree(blkg->pd[pol->plid]); | |
1164 | blkg->pd[pol->plid] = NULL; | |
e48453c3 AA |
1165 | kfree(blkg->blkcg->pd[pol->plid]); |
1166 | blkg->blkcg->pd[pol->plid] = NULL; | |
a2b1693b TH |
1167 | |
1168 | spin_unlock(&blkg->blkcg->lock); | |
1169 | } | |
1170 | ||
1171 | spin_unlock_irq(q->queue_lock); | |
1172 | blk_queue_bypass_end(q); | |
1173 | } | |
1174 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1175 | ||
8bd435b3 | 1176 | /** |
3c798398 TH |
1177 | * blkcg_policy_register - register a blkcg policy |
1178 | * @pol: blkcg policy to register | |
8bd435b3 | 1179 | * |
3c798398 TH |
1180 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1181 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1182 | */ |
d5bf0291 | 1183 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1184 | { |
8bd435b3 | 1185 | int i, ret; |
e8989fae | 1186 | |
f95a04af TH |
1187 | if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data))) |
1188 | return -EINVAL; | |
1189 | ||
bc0d6501 TH |
1190 | mutex_lock(&blkcg_pol_mutex); |
1191 | ||
8bd435b3 TH |
1192 | /* find an empty slot */ |
1193 | ret = -ENOSPC; | |
1194 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1195 | if (!blkcg_policy[i]) |
8bd435b3 TH |
1196 | break; |
1197 | if (i >= BLKCG_MAX_POLS) | |
1198 | goto out_unlock; | |
035d10b2 | 1199 | |
8bd435b3 | 1200 | /* register and update blkgs */ |
3c798398 TH |
1201 | pol->plid = i; |
1202 | blkcg_policy[i] = pol; | |
8bd435b3 | 1203 | |
8bd435b3 | 1204 | /* everything is in place, add intf files for the new policy */ |
3c798398 | 1205 | if (pol->cftypes) |
2cf669a5 TH |
1206 | WARN_ON(cgroup_add_legacy_cftypes(&blkio_cgrp_subsys, |
1207 | pol->cftypes)); | |
8bd435b3 TH |
1208 | ret = 0; |
1209 | out_unlock: | |
bc0d6501 | 1210 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1211 | return ret; |
3e252066 | 1212 | } |
3c798398 | 1213 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1214 | |
8bd435b3 | 1215 | /** |
3c798398 TH |
1216 | * blkcg_policy_unregister - unregister a blkcg policy |
1217 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1218 | * |
3c798398 | 1219 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1220 | */ |
3c798398 | 1221 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1222 | { |
bc0d6501 TH |
1223 | mutex_lock(&blkcg_pol_mutex); |
1224 | ||
3c798398 | 1225 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1226 | goto out_unlock; |
1227 | ||
1228 | /* kill the intf files first */ | |
3c798398 | 1229 | if (pol->cftypes) |
2bb566cb | 1230 | cgroup_rm_cftypes(pol->cftypes); |
44ea53de | 1231 | |
8bd435b3 | 1232 | /* unregister and update blkgs */ |
3c798398 | 1233 | blkcg_policy[pol->plid] = NULL; |
8bd435b3 | 1234 | out_unlock: |
bc0d6501 | 1235 | mutex_unlock(&blkcg_pol_mutex); |
3e252066 | 1236 | } |
3c798398 | 1237 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |