]>
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> |
174cd4b1 | 20 | #include <linux/sched/signal.h> |
accee785 | 21 | #include <linux/err.h> |
9195291e | 22 | #include <linux/blkdev.h> |
52ebea74 | 23 | #include <linux/backing-dev.h> |
5a0e3ad6 | 24 | #include <linux/slab.h> |
34d0f179 | 25 | #include <linux/genhd.h> |
72e06c25 | 26 | #include <linux/delay.h> |
9a9e8a26 | 27 | #include <linux/atomic.h> |
36aa9e5f | 28 | #include <linux/ctype.h> |
eea8f41c | 29 | #include <linux/blk-cgroup.h> |
5efd6113 | 30 | #include "blk.h" |
3e252066 | 31 | |
84c124da DS |
32 | #define MAX_KEY_LEN 100 |
33 | ||
838f13bf TH |
34 | /* |
35 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
36 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
37 | * policy [un]register operations including cgroup file additions / | |
38 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
39 | * allows grabbing it from cgroup callbacks. | |
40 | */ | |
41 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 42 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 43 | |
e48453c3 | 44 | struct blkcg blkcg_root; |
3c798398 | 45 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 46 | |
496d5e75 TH |
47 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
48 | ||
3c798398 | 49 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 50 | |
7876f930 TH |
51 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
52 | ||
a2b1693b | 53 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 54 | const struct blkcg_policy *pol) |
a2b1693b TH |
55 | { |
56 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
57 | } | |
58 | ||
0381411e TH |
59 | /** |
60 | * blkg_free - free a blkg | |
61 | * @blkg: blkg to free | |
62 | * | |
63 | * Free @blkg which may be partially allocated. | |
64 | */ | |
3c798398 | 65 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 66 | { |
e8989fae | 67 | int i; |
549d3aa8 TH |
68 | |
69 | if (!blkg) | |
70 | return; | |
71 | ||
db613670 | 72 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
73 | if (blkg->pd[i]) |
74 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
e8989fae | 75 | |
994b7832 TH |
76 | if (blkg->blkcg != &blkcg_root) |
77 | blk_exit_rl(&blkg->rl); | |
77ea7338 TH |
78 | |
79 | blkg_rwstat_exit(&blkg->stat_ios); | |
80 | blkg_rwstat_exit(&blkg->stat_bytes); | |
549d3aa8 | 81 | kfree(blkg); |
0381411e TH |
82 | } |
83 | ||
84 | /** | |
85 | * blkg_alloc - allocate a blkg | |
86 | * @blkcg: block cgroup the new blkg is associated with | |
87 | * @q: request_queue the new blkg is associated with | |
15974993 | 88 | * @gfp_mask: allocation mask to use |
0381411e | 89 | * |
e8989fae | 90 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 91 | */ |
15974993 TH |
92 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
93 | gfp_t gfp_mask) | |
0381411e | 94 | { |
3c798398 | 95 | struct blkcg_gq *blkg; |
e8989fae | 96 | int i; |
0381411e TH |
97 | |
98 | /* alloc and init base part */ | |
15974993 | 99 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
100 | if (!blkg) |
101 | return NULL; | |
102 | ||
77ea7338 TH |
103 | if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) || |
104 | blkg_rwstat_init(&blkg->stat_ios, gfp_mask)) | |
105 | goto err_free; | |
106 | ||
c875f4d0 | 107 | blkg->q = q; |
e8989fae | 108 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 109 | blkg->blkcg = blkcg; |
a5049a8a | 110 | atomic_set(&blkg->refcnt, 1); |
0381411e | 111 | |
a051661c TH |
112 | /* root blkg uses @q->root_rl, init rl only for !root blkgs */ |
113 | if (blkcg != &blkcg_root) { | |
114 | if (blk_init_rl(&blkg->rl, q, gfp_mask)) | |
115 | goto err_free; | |
116 | blkg->rl.blkg = blkg; | |
117 | } | |
118 | ||
8bd435b3 | 119 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 120 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 121 | struct blkg_policy_data *pd; |
0381411e | 122 | |
a2b1693b | 123 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
124 | continue; |
125 | ||
126 | /* alloc per-policy data and attach it to blkg */ | |
001bea73 | 127 | pd = pol->pd_alloc_fn(gfp_mask, q->node); |
a051661c TH |
128 | if (!pd) |
129 | goto err_free; | |
549d3aa8 | 130 | |
e8989fae TH |
131 | blkg->pd[i] = pd; |
132 | pd->blkg = blkg; | |
b276a876 | 133 | pd->plid = i; |
e8989fae TH |
134 | } |
135 | ||
0381411e | 136 | return blkg; |
a051661c TH |
137 | |
138 | err_free: | |
139 | blkg_free(blkg); | |
140 | return NULL; | |
0381411e TH |
141 | } |
142 | ||
24f29046 TH |
143 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
144 | struct request_queue *q, bool update_hint) | |
80fd9979 | 145 | { |
3c798398 | 146 | struct blkcg_gq *blkg; |
80fd9979 | 147 | |
a637120e | 148 | /* |
86cde6b6 TH |
149 | * Hint didn't match. Look up from the radix tree. Note that the |
150 | * hint can only be updated under queue_lock as otherwise @blkg | |
151 | * could have already been removed from blkg_tree. The caller is | |
152 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
153 | */ |
154 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
155 | if (blkg && blkg->q == q) { |
156 | if (update_hint) { | |
157 | lockdep_assert_held(q->queue_lock); | |
158 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
159 | } | |
a637120e | 160 | return blkg; |
86cde6b6 | 161 | } |
a637120e | 162 | |
80fd9979 TH |
163 | return NULL; |
164 | } | |
ae118896 | 165 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); |
80fd9979 | 166 | |
15974993 | 167 | /* |
7fc6b87a TE |
168 | * If gfp mask allows blocking, this function temporarily drops rcu and queue |
169 | * locks to allocate memory. | |
15974993 | 170 | */ |
86cde6b6 | 171 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
7fc6b87a TE |
172 | struct request_queue *q, gfp_t gfp, |
173 | const struct blkcg_policy *pol) | |
5624a4e4 | 174 | { |
7fc6b87a | 175 | struct blkcg_gq *blkg = NULL; |
ce7acfea | 176 | struct bdi_writeback_congested *wb_congested; |
f427d909 | 177 | int i, ret; |
7fc6b87a TE |
178 | const bool drop_locks = gfpflags_allow_blocking(gfp); |
179 | bool preloaded = false; | |
5624a4e4 | 180 | |
cd1604fa TH |
181 | WARN_ON_ONCE(!rcu_read_lock_held()); |
182 | lockdep_assert_held(q->queue_lock); | |
183 | ||
7ee9c562 | 184 | /* blkg holds a reference to blkcg */ |
ec903c0c | 185 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 186 | ret = -ENODEV; |
93e6d5d8 | 187 | goto err_free_blkg; |
15974993 | 188 | } |
cd1604fa | 189 | |
7fc6b87a TE |
190 | if (drop_locks) { |
191 | spin_unlock_irq(q->queue_lock); | |
192 | rcu_read_unlock(); | |
193 | } | |
194 | ||
dc3b17cc | 195 | wb_congested = wb_congested_get_create(q->backing_dev_info, |
7fc6b87a TE |
196 | blkcg->css.id, gfp); |
197 | blkg = blkg_alloc(blkcg, q, gfp); | |
198 | ||
199 | if (drop_locks) { | |
200 | preloaded = !radix_tree_preload(gfp); | |
201 | rcu_read_lock(); | |
202 | spin_lock_irq(q->queue_lock); | |
203 | } | |
204 | ||
205 | if (unlikely(!wb_congested || !blkg)) { | |
ce7acfea | 206 | ret = -ENOMEM; |
7fc6b87a | 207 | goto err_put; |
ce7acfea TH |
208 | } |
209 | ||
7fc6b87a TE |
210 | blkg->wb_congested = wb_congested; |
211 | ||
212 | if (pol) { | |
213 | WARN_ON(!drop_locks); | |
214 | ||
215 | if (!blkcg_policy_enabled(q, pol)) { | |
216 | ret = -EOPNOTSUPP; | |
217 | goto err_put; | |
218 | } | |
219 | ||
220 | /* | |
221 | * This could be the first entry point of blkcg implementation | |
222 | * and we shouldn't allow anything to go through for a bypassing | |
223 | * queue. | |
224 | */ | |
225 | if (unlikely(blk_queue_bypass(q))) { | |
226 | ret = blk_queue_dying(q) ? -ENODEV : -EBUSY; | |
227 | goto err_put; | |
15974993 TH |
228 | } |
229 | } | |
cd1604fa | 230 | |
db613670 | 231 | /* link parent */ |
3c547865 TH |
232 | if (blkcg_parent(blkcg)) { |
233 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
234 | if (WARN_ON_ONCE(!blkg->parent)) { | |
20386ce0 | 235 | ret = -ENODEV; |
7fc6b87a | 236 | goto err_put; |
3c547865 TH |
237 | } |
238 | blkg_get(blkg->parent); | |
239 | } | |
240 | ||
db613670 TH |
241 | /* invoke per-policy init */ |
242 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
243 | struct blkcg_policy *pol = blkcg_policy[i]; | |
244 | ||
245 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 246 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
247 | } |
248 | ||
249 | /* insert */ | |
cd1604fa | 250 | spin_lock(&blkcg->lock); |
a637120e TH |
251 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
252 | if (likely(!ret)) { | |
253 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
254 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
255 | |
256 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
257 | struct blkcg_policy *pol = blkcg_policy[i]; | |
258 | ||
259 | if (blkg->pd[i] && pol->pd_online_fn) | |
a9520cd6 | 260 | pol->pd_online_fn(blkg->pd[i]); |
f427d909 | 261 | } |
a637120e | 262 | } |
7fc6b87a TE |
263 | |
264 | if (preloaded) | |
265 | radix_tree_preload_end(); | |
f427d909 | 266 | blkg->online = true; |
cd1604fa | 267 | spin_unlock(&blkcg->lock); |
496fb780 | 268 | |
ec13b1d6 | 269 | if (!ret) |
a637120e | 270 | return blkg; |
15974993 | 271 | |
3c547865 TH |
272 | /* @blkg failed fully initialized, use the usual release path */ |
273 | blkg_put(blkg); | |
274 | return ERR_PTR(ret); | |
275 | ||
7fc6b87a TE |
276 | err_put: |
277 | if (preloaded) | |
278 | radix_tree_preload_end(); | |
279 | if (wb_congested) | |
280 | wb_congested_put(wb_congested); | |
496fb780 | 281 | css_put(&blkcg->css); |
93e6d5d8 | 282 | err_free_blkg: |
7fc6b87a | 283 | blkg_free(blkg); |
93e6d5d8 | 284 | return ERR_PTR(ret); |
31e4c28d | 285 | } |
3c96cb32 | 286 | |
86cde6b6 | 287 | /** |
7fc6b87a | 288 | * __blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 TH |
289 | * @blkcg: blkcg of interest |
290 | * @q: request_queue of interest | |
7fc6b87a TE |
291 | * @gfp: gfp mask |
292 | * @pol: blkcg policy (optional) | |
86cde6b6 TH |
293 | * |
294 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
295 | * create one. blkg creation is performed recursively from blkcg_root such |
296 | * that all non-root blkg's have access to the parent blkg. This function | |
297 | * should be called under RCU read lock and @q->queue_lock. | |
86cde6b6 | 298 | * |
7fc6b87a TE |
299 | * When gfp mask allows blocking, rcu and queue locks may be dropped for |
300 | * allocating memory. In this case, the locks will be reacquired on return. | |
301 | * | |
86cde6b6 TH |
302 | * Returns pointer to the looked up or created blkg on success, ERR_PTR() |
303 | * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not | |
304 | * dead and bypassing, returns ERR_PTR(-EBUSY). | |
305 | */ | |
7fc6b87a TE |
306 | struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg, |
307 | struct request_queue *q, gfp_t gfp, | |
308 | const struct blkcg_policy *pol) | |
3c96cb32 | 309 | { |
86cde6b6 TH |
310 | struct blkcg_gq *blkg; |
311 | ||
312 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
313 | lockdep_assert_held(q->queue_lock); | |
314 | ||
86cde6b6 TH |
315 | blkg = __blkg_lookup(blkcg, q, true); |
316 | if (blkg) | |
317 | return blkg; | |
318 | ||
3c547865 TH |
319 | /* |
320 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
321 | * non-root blkgs have access to their parents. | |
322 | */ | |
323 | while (true) { | |
324 | struct blkcg *pos = blkcg; | |
325 | struct blkcg *parent = blkcg_parent(blkcg); | |
326 | ||
327 | while (parent && !__blkg_lookup(parent, q, false)) { | |
328 | pos = parent; | |
329 | parent = blkcg_parent(parent); | |
330 | } | |
331 | ||
7fc6b87a | 332 | blkg = blkg_create(pos, q, gfp, pol); |
3c547865 TH |
333 | if (pos == blkcg || IS_ERR(blkg)) |
334 | return blkg; | |
335 | } | |
3c96cb32 | 336 | } |
31e4c28d | 337 | |
7fc6b87a TE |
338 | /** |
339 | * blkg_lookup_create - lookup blkg, try to create one if not there | |
340 | * | |
341 | * Performs an initial queue bypass check and then passes control to | |
342 | * __blkg_lookup_create(). | |
343 | */ | |
344 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, | |
345 | struct request_queue *q, gfp_t gfp, | |
346 | const struct blkcg_policy *pol) | |
347 | { | |
348 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
349 | lockdep_assert_held(q->queue_lock); | |
350 | ||
351 | /* | |
352 | * This could be the first entry point of blkcg implementation and | |
353 | * we shouldn't allow anything to go through for a bypassing queue. | |
354 | */ | |
355 | if (unlikely(blk_queue_bypass(q))) | |
356 | return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY); | |
357 | ||
358 | return __blkg_lookup_create(blkcg, q, gfp, pol); | |
359 | } | |
360 | ||
3c798398 | 361 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 362 | { |
3c798398 | 363 | struct blkcg *blkcg = blkg->blkcg; |
77ea7338 | 364 | struct blkcg_gq *parent = blkg->parent; |
f427d909 | 365 | int i; |
03aa264a | 366 | |
27e1f9d1 | 367 | lockdep_assert_held(blkg->q->queue_lock); |
9f13ef67 | 368 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
369 | |
370 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 371 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 372 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 373 | |
f427d909 TH |
374 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
375 | struct blkcg_policy *pol = blkcg_policy[i]; | |
376 | ||
377 | if (blkg->pd[i] && pol->pd_offline_fn) | |
a9520cd6 | 378 | pol->pd_offline_fn(blkg->pd[i]); |
f427d909 | 379 | } |
77ea7338 TH |
380 | |
381 | if (parent) { | |
382 | blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes); | |
383 | blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios); | |
384 | } | |
385 | ||
f427d909 TH |
386 | blkg->online = false; |
387 | ||
a637120e | 388 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 389 | list_del_init(&blkg->q_node); |
9f13ef67 | 390 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 391 | |
a637120e TH |
392 | /* |
393 | * Both setting lookup hint to and clearing it from @blkg are done | |
394 | * under queue_lock. If it's not pointing to @blkg now, it never | |
395 | * will. Hint assignment itself can race safely. | |
396 | */ | |
ec6c676a | 397 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
398 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
399 | ||
03aa264a TH |
400 | /* |
401 | * Put the reference taken at the time of creation so that when all | |
402 | * queues are gone, group can be destroyed. | |
403 | */ | |
404 | blkg_put(blkg); | |
405 | } | |
406 | ||
9f13ef67 TH |
407 | /** |
408 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
409 | * @q: request_queue of interest | |
9f13ef67 | 410 | * |
3c96cb32 | 411 | * Destroy all blkgs associated with @q. |
9f13ef67 | 412 | */ |
3c96cb32 | 413 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 414 | { |
3c798398 | 415 | struct blkcg_gq *blkg, *n; |
72e06c25 | 416 | |
6d18b008 | 417 | lockdep_assert_held(q->queue_lock); |
72e06c25 | 418 | |
9f13ef67 | 419 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 420 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 421 | |
9f13ef67 TH |
422 | spin_lock(&blkcg->lock); |
423 | blkg_destroy(blkg); | |
424 | spin_unlock(&blkcg->lock); | |
72e06c25 | 425 | } |
6fe810bd TH |
426 | |
427 | q->root_blkg = NULL; | |
428 | q->root_rl.blkg = NULL; | |
72e06c25 TH |
429 | } |
430 | ||
2a4fd070 TH |
431 | /* |
432 | * A group is RCU protected, but having an rcu lock does not mean that one | |
433 | * can access all the fields of blkg and assume these are valid. For | |
434 | * example, don't try to follow throtl_data and request queue links. | |
435 | * | |
436 | * Having a reference to blkg under an rcu allows accesses to only values | |
437 | * local to groups like group stats and group rate limits. | |
438 | */ | |
439 | void __blkg_release_rcu(struct rcu_head *rcu_head) | |
1adaf3dd | 440 | { |
2a4fd070 | 441 | struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head); |
db613670 | 442 | |
3c547865 | 443 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
1adaf3dd | 444 | css_put(&blkg->blkcg->css); |
a5049a8a | 445 | if (blkg->parent) |
3c547865 | 446 | blkg_put(blkg->parent); |
1adaf3dd | 447 | |
ce7acfea TH |
448 | wb_congested_put(blkg->wb_congested); |
449 | ||
2a4fd070 | 450 | blkg_free(blkg); |
1adaf3dd | 451 | } |
2a4fd070 | 452 | EXPORT_SYMBOL_GPL(__blkg_release_rcu); |
1adaf3dd | 453 | |
a051661c TH |
454 | /* |
455 | * The next function used by blk_queue_for_each_rl(). It's a bit tricky | |
456 | * because the root blkg uses @q->root_rl instead of its own rl. | |
457 | */ | |
458 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
459 | struct request_queue *q) | |
460 | { | |
461 | struct list_head *ent; | |
462 | struct blkcg_gq *blkg; | |
463 | ||
464 | /* | |
465 | * Determine the current blkg list_head. The first entry is | |
466 | * root_rl which is off @q->blkg_list and mapped to the head. | |
467 | */ | |
468 | if (rl == &q->root_rl) { | |
469 | ent = &q->blkg_list; | |
65c77fd9 JN |
470 | /* There are no more block groups, hence no request lists */ |
471 | if (list_empty(ent)) | |
472 | return NULL; | |
a051661c TH |
473 | } else { |
474 | blkg = container_of(rl, struct blkcg_gq, rl); | |
475 | ent = &blkg->q_node; | |
476 | } | |
477 | ||
478 | /* walk to the next list_head, skip root blkcg */ | |
479 | ent = ent->next; | |
480 | if (ent == &q->root_blkg->q_node) | |
481 | ent = ent->next; | |
482 | if (ent == &q->blkg_list) | |
483 | return NULL; | |
484 | ||
485 | blkg = container_of(ent, struct blkcg_gq, q_node); | |
486 | return &blkg->rl; | |
487 | } | |
488 | ||
182446d0 TH |
489 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
490 | struct cftype *cftype, u64 val) | |
303a3acb | 491 | { |
182446d0 | 492 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 493 | struct blkcg_gq *blkg; |
bc0d6501 | 494 | int i; |
303a3acb | 495 | |
838f13bf | 496 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 497 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
498 | |
499 | /* | |
500 | * Note that stat reset is racy - it doesn't synchronize against | |
501 | * stat updates. This is a debug feature which shouldn't exist | |
502 | * anyway. If you get hit by a race, retry. | |
503 | */ | |
b67bfe0d | 504 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
77ea7338 TH |
505 | blkg_rwstat_reset(&blkg->stat_bytes); |
506 | blkg_rwstat_reset(&blkg->stat_ios); | |
507 | ||
8bd435b3 | 508 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 509 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 510 | |
a9520cd6 TH |
511 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
512 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 513 | } |
303a3acb | 514 | } |
f0bdc8cd | 515 | |
303a3acb | 516 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 517 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
518 | return 0; |
519 | } | |
520 | ||
dd165eb3 | 521 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 522 | { |
d3d32e69 | 523 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
dc3b17cc JK |
524 | if (blkg->q->backing_dev_info->dev) |
525 | return dev_name(blkg->q->backing_dev_info->dev); | |
d3d32e69 | 526 | return NULL; |
303a3acb | 527 | } |
dd165eb3 | 528 | EXPORT_SYMBOL_GPL(blkg_dev_name); |
303a3acb | 529 | |
d3d32e69 TH |
530 | /** |
531 | * blkcg_print_blkgs - helper for printing per-blkg data | |
532 | * @sf: seq_file to print to | |
533 | * @blkcg: blkcg of interest | |
534 | * @prfill: fill function to print out a blkg | |
535 | * @pol: policy in question | |
536 | * @data: data to be passed to @prfill | |
537 | * @show_total: to print out sum of prfill return values or not | |
538 | * | |
539 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
540 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
541 | * policy data and @data and the matching queue lock held. If @show_total |
542 | * is %true, the sum of the return values from @prfill is printed with | |
543 | * "Total" label at the end. | |
d3d32e69 TH |
544 | * |
545 | * This is to be used to construct print functions for | |
546 | * cftype->read_seq_string method. | |
547 | */ | |
3c798398 | 548 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
549 | u64 (*prfill)(struct seq_file *, |
550 | struct blkg_policy_data *, int), | |
3c798398 | 551 | const struct blkcg_policy *pol, int data, |
ec399347 | 552 | bool show_total) |
5624a4e4 | 553 | { |
3c798398 | 554 | struct blkcg_gq *blkg; |
d3d32e69 | 555 | u64 total = 0; |
5624a4e4 | 556 | |
810ecfa7 | 557 | rcu_read_lock(); |
ee89f812 | 558 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
810ecfa7 | 559 | spin_lock_irq(blkg->q->queue_lock); |
a2b1693b | 560 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 561 | total += prfill(sf, blkg->pd[pol->plid], data); |
810ecfa7 TH |
562 | spin_unlock_irq(blkg->q->queue_lock); |
563 | } | |
564 | rcu_read_unlock(); | |
d3d32e69 TH |
565 | |
566 | if (show_total) | |
567 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
568 | } | |
829fdb50 | 569 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
570 | |
571 | /** | |
572 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
573 | * @sf: seq_file to print to | |
f95a04af | 574 | * @pd: policy private data of interest |
d3d32e69 TH |
575 | * @v: value to print |
576 | * | |
f95a04af | 577 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 578 | */ |
f95a04af | 579 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 580 | { |
f95a04af | 581 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
582 | |
583 | if (!dname) | |
584 | return 0; | |
585 | ||
586 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
587 | return v; | |
588 | } | |
829fdb50 | 589 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
590 | |
591 | /** | |
592 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
593 | * @sf: seq_file to print to | |
f95a04af | 594 | * @pd: policy private data of interest |
d3d32e69 TH |
595 | * @rwstat: rwstat to print |
596 | * | |
f95a04af | 597 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 598 | */ |
f95a04af | 599 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 600 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
601 | { |
602 | static const char *rwstr[] = { | |
603 | [BLKG_RWSTAT_READ] = "Read", | |
604 | [BLKG_RWSTAT_WRITE] = "Write", | |
605 | [BLKG_RWSTAT_SYNC] = "Sync", | |
606 | [BLKG_RWSTAT_ASYNC] = "Async", | |
607 | }; | |
f95a04af | 608 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
609 | u64 v; |
610 | int i; | |
611 | ||
612 | if (!dname) | |
613 | return 0; | |
614 | ||
615 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
616 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
24bdb8ef | 617 | (unsigned long long)atomic64_read(&rwstat->aux_cnt[i])); |
d3d32e69 | 618 | |
24bdb8ef TH |
619 | v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) + |
620 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]); | |
d3d32e69 TH |
621 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); |
622 | return v; | |
623 | } | |
b50da39f | 624 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 625 | |
5bc4afb1 TH |
626 | /** |
627 | * blkg_prfill_stat - prfill callback for blkg_stat | |
628 | * @sf: seq_file to print to | |
f95a04af TH |
629 | * @pd: policy private data of interest |
630 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
631 | * |
632 | * prfill callback for printing a blkg_stat. | |
633 | */ | |
f95a04af | 634 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 635 | { |
f95a04af | 636 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 637 | } |
5bc4afb1 | 638 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 639 | |
5bc4afb1 TH |
640 | /** |
641 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
642 | * @sf: seq_file to print to | |
f95a04af TH |
643 | * @pd: policy private data of interest |
644 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
645 | * |
646 | * prfill callback for printing a blkg_rwstat. | |
647 | */ | |
f95a04af TH |
648 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
649 | int off) | |
d3d32e69 | 650 | { |
f95a04af | 651 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 652 | |
f95a04af | 653 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 654 | } |
5bc4afb1 | 655 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 656 | |
77ea7338 TH |
657 | static u64 blkg_prfill_rwstat_field(struct seq_file *sf, |
658 | struct blkg_policy_data *pd, int off) | |
659 | { | |
660 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off); | |
661 | ||
662 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
663 | } | |
664 | ||
665 | /** | |
666 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes | |
667 | * @sf: seq_file to print to | |
668 | * @v: unused | |
669 | * | |
670 | * To be used as cftype->seq_show to print blkg->stat_bytes. | |
671 | * cftype->private must be set to the blkcg_policy. | |
672 | */ | |
673 | int blkg_print_stat_bytes(struct seq_file *sf, void *v) | |
674 | { | |
675 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
676 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
677 | offsetof(struct blkcg_gq, stat_bytes), true); | |
678 | return 0; | |
679 | } | |
680 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes); | |
681 | ||
682 | /** | |
683 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios | |
684 | * @sf: seq_file to print to | |
685 | * @v: unused | |
686 | * | |
687 | * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private | |
688 | * must be set to the blkcg_policy. | |
689 | */ | |
690 | int blkg_print_stat_ios(struct seq_file *sf, void *v) | |
691 | { | |
692 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
693 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
694 | offsetof(struct blkcg_gq, stat_ios), true); | |
695 | return 0; | |
696 | } | |
697 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios); | |
698 | ||
699 | static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf, | |
700 | struct blkg_policy_data *pd, | |
701 | int off) | |
702 | { | |
703 | struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg, | |
704 | NULL, off); | |
705 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
706 | } | |
707 | ||
708 | /** | |
709 | * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes | |
710 | * @sf: seq_file to print to | |
711 | * @v: unused | |
712 | */ | |
713 | int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v) | |
714 | { | |
715 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
716 | blkg_prfill_rwstat_field_recursive, | |
717 | (void *)seq_cft(sf)->private, | |
718 | offsetof(struct blkcg_gq, stat_bytes), true); | |
719 | return 0; | |
720 | } | |
721 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive); | |
722 | ||
723 | /** | |
724 | * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios | |
725 | * @sf: seq_file to print to | |
726 | * @v: unused | |
727 | */ | |
728 | int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v) | |
729 | { | |
730 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
731 | blkg_prfill_rwstat_field_recursive, | |
732 | (void *)seq_cft(sf)->private, | |
733 | offsetof(struct blkcg_gq, stat_ios), true); | |
734 | return 0; | |
735 | } | |
736 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive); | |
737 | ||
16b3de66 TH |
738 | /** |
739 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
f12c74ca TH |
740 | * @blkg: blkg of interest |
741 | * @pol: blkcg_policy which contains the blkg_stat | |
742 | * @off: offset to the blkg_stat in blkg_policy_data or @blkg | |
16b3de66 | 743 | * |
f12c74ca TH |
744 | * Collect the blkg_stat specified by @blkg, @pol and @off and all its |
745 | * online descendants and their aux counts. The caller must be holding the | |
746 | * queue lock for online tests. | |
747 | * | |
748 | * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is | |
749 | * at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 750 | */ |
f12c74ca TH |
751 | u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg, |
752 | struct blkcg_policy *pol, int off) | |
16b3de66 | 753 | { |
16b3de66 | 754 | struct blkcg_gq *pos_blkg; |
492eb21b | 755 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 756 | u64 sum = 0; |
16b3de66 | 757 | |
f12c74ca | 758 | lockdep_assert_held(blkg->q->queue_lock); |
16b3de66 | 759 | |
16b3de66 | 760 | rcu_read_lock(); |
f12c74ca TH |
761 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
762 | struct blkg_stat *stat; | |
763 | ||
764 | if (!pos_blkg->online) | |
765 | continue; | |
16b3de66 | 766 | |
f12c74ca TH |
767 | if (pol) |
768 | stat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
769 | else | |
770 | stat = (void *)blkg + off; | |
771 | ||
772 | sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
16b3de66 TH |
773 | } |
774 | rcu_read_unlock(); | |
775 | ||
776 | return sum; | |
777 | } | |
778 | EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum); | |
779 | ||
780 | /** | |
781 | * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat | |
f12c74ca TH |
782 | * @blkg: blkg of interest |
783 | * @pol: blkcg_policy which contains the blkg_rwstat | |
784 | * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg | |
16b3de66 | 785 | * |
f12c74ca TH |
786 | * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its |
787 | * online descendants and their aux counts. The caller must be holding the | |
788 | * queue lock for online tests. | |
789 | * | |
790 | * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it | |
791 | * is at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 792 | */ |
f12c74ca TH |
793 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, |
794 | struct blkcg_policy *pol, int off) | |
16b3de66 | 795 | { |
16b3de66 | 796 | struct blkcg_gq *pos_blkg; |
492eb21b | 797 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 798 | struct blkg_rwstat sum = { }; |
16b3de66 TH |
799 | int i; |
800 | ||
f12c74ca | 801 | lockdep_assert_held(blkg->q->queue_lock); |
16b3de66 | 802 | |
16b3de66 | 803 | rcu_read_lock(); |
f12c74ca | 804 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
3a7faead | 805 | struct blkg_rwstat *rwstat; |
16b3de66 TH |
806 | |
807 | if (!pos_blkg->online) | |
808 | continue; | |
809 | ||
f12c74ca TH |
810 | if (pol) |
811 | rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
812 | else | |
813 | rwstat = (void *)pos_blkg + off; | |
814 | ||
16b3de66 | 815 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
3a7faead TH |
816 | atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) + |
817 | percpu_counter_sum_positive(&rwstat->cpu_cnt[i]), | |
818 | &sum.aux_cnt[i]); | |
16b3de66 TH |
819 | } |
820 | rcu_read_unlock(); | |
821 | ||
822 | return sum; | |
823 | } | |
824 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
825 | ||
3a8b31d3 TH |
826 | /** |
827 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
828 | * @blkcg: target block cgroup | |
da8b0662 | 829 | * @pol: target policy |
3a8b31d3 TH |
830 | * @input: input string |
831 | * @ctx: blkg_conf_ctx to be filled | |
832 | * | |
833 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
834 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
835 | * part of @input following MAJ:MIN. This function returns with RCU read | |
836 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 837 | */ |
3c798398 | 838 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 839 | char *input, struct blkg_conf_ctx *ctx) |
da8b0662 | 840 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
34d0f179 | 841 | { |
3a8b31d3 | 842 | struct gendisk *disk; |
3c798398 | 843 | struct blkcg_gq *blkg; |
39a169b6 | 844 | struct module *owner; |
726fa694 | 845 | unsigned int major, minor; |
36aa9e5f TH |
846 | int key_len, part, ret; |
847 | char *body; | |
34d0f179 | 848 | |
36aa9e5f | 849 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) |
726fa694 | 850 | return -EINVAL; |
3a8b31d3 | 851 | |
36aa9e5f TH |
852 | body = input + key_len; |
853 | if (!isspace(*body)) | |
854 | return -EINVAL; | |
855 | body = skip_spaces(body); | |
856 | ||
726fa694 | 857 | disk = get_gendisk(MKDEV(major, minor), &part); |
5f6c2d2b | 858 | if (!disk) |
20386ce0 | 859 | return -ENODEV; |
5f6c2d2b | 860 | if (part) { |
39a169b6 | 861 | owner = disk->fops->owner; |
5f6c2d2b | 862 | put_disk(disk); |
39a169b6 | 863 | module_put(owner); |
20386ce0 | 864 | return -ENODEV; |
5f6c2d2b | 865 | } |
e56da7e2 TH |
866 | |
867 | rcu_read_lock(); | |
4bfd482e | 868 | spin_lock_irq(disk->queue->queue_lock); |
da8b0662 | 869 | |
a2b1693b | 870 | if (blkcg_policy_enabled(disk->queue, pol)) |
7fc6b87a | 871 | blkg = blkg_lookup_create(blkcg, disk->queue, GFP_KERNEL, pol); |
a2b1693b | 872 | else |
20386ce0 | 873 | blkg = ERR_PTR(-EOPNOTSUPP); |
e56da7e2 | 874 | |
4bfd482e TH |
875 | if (IS_ERR(blkg)) { |
876 | ret = PTR_ERR(blkg); | |
3a8b31d3 | 877 | rcu_read_unlock(); |
da8b0662 | 878 | spin_unlock_irq(disk->queue->queue_lock); |
39a169b6 | 879 | owner = disk->fops->owner; |
3a8b31d3 | 880 | put_disk(disk); |
39a169b6 | 881 | module_put(owner); |
3a8b31d3 TH |
882 | /* |
883 | * If queue was bypassing, we should retry. Do so after a | |
884 | * short msleep(). It isn't strictly necessary but queue | |
885 | * can be bypassing for some time and it's always nice to | |
886 | * avoid busy looping. | |
887 | */ | |
888 | if (ret == -EBUSY) { | |
889 | msleep(10); | |
890 | ret = restart_syscall(); | |
7702e8f4 | 891 | } |
726fa694 | 892 | return ret; |
062a644d | 893 | } |
3a8b31d3 TH |
894 | |
895 | ctx->disk = disk; | |
896 | ctx->blkg = blkg; | |
36aa9e5f | 897 | ctx->body = body; |
726fa694 | 898 | return 0; |
34d0f179 | 899 | } |
829fdb50 | 900 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 901 | |
3a8b31d3 TH |
902 | /** |
903 | * blkg_conf_finish - finish up per-blkg config update | |
904 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
905 | * | |
906 | * Finish up after per-blkg config update. This function must be paired | |
907 | * with blkg_conf_prep(). | |
908 | */ | |
829fdb50 | 909 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
da8b0662 | 910 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 911 | { |
39a169b6 RP |
912 | struct module *owner; |
913 | ||
da8b0662 | 914 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
3a8b31d3 | 915 | rcu_read_unlock(); |
39a169b6 | 916 | owner = ctx->disk->fops->owner; |
3a8b31d3 | 917 | put_disk(ctx->disk); |
39a169b6 | 918 | module_put(owner); |
34d0f179 | 919 | } |
829fdb50 | 920 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 921 | |
2ee867dc TH |
922 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
923 | { | |
924 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
925 | struct blkcg_gq *blkg; | |
926 | ||
927 | rcu_read_lock(); | |
928 | ||
929 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
930 | const char *dname; | |
931 | struct blkg_rwstat rwstat; | |
932 | u64 rbytes, wbytes, rios, wios; | |
933 | ||
934 | dname = blkg_dev_name(blkg); | |
935 | if (!dname) | |
936 | continue; | |
937 | ||
938 | spin_lock_irq(blkg->q->queue_lock); | |
939 | ||
940 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
941 | offsetof(struct blkcg_gq, stat_bytes)); | |
942 | rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
943 | wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
944 | ||
945 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
946 | offsetof(struct blkcg_gq, stat_ios)); | |
947 | rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
948 | wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
949 | ||
950 | spin_unlock_irq(blkg->q->queue_lock); | |
951 | ||
952 | if (rbytes || wbytes || rios || wios) | |
953 | seq_printf(sf, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n", | |
954 | dname, rbytes, wbytes, rios, wios); | |
955 | } | |
956 | ||
957 | rcu_read_unlock(); | |
958 | return 0; | |
959 | } | |
960 | ||
e1f3b941 | 961 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
962 | { |
963 | .name = "stat", | |
ca0752c5 | 964 | .flags = CFTYPE_NOT_ON_ROOT, |
2ee867dc TH |
965 | .seq_show = blkcg_print_stat, |
966 | }, | |
967 | { } /* terminate */ | |
968 | }; | |
969 | ||
e1f3b941 | 970 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
971 | { |
972 | .name = "reset_stats", | |
3c798398 | 973 | .write_u64 = blkcg_reset_stats, |
22084190 | 974 | }, |
4baf6e33 | 975 | { } /* terminate */ |
31e4c28d VG |
976 | }; |
977 | ||
9f13ef67 | 978 | /** |
92fb9748 | 979 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 980 | * @css: css of interest |
9f13ef67 | 981 | * |
eb95419b TH |
982 | * This function is called when @css is about to go away and responsible |
983 | * for shooting down all blkgs associated with @css. blkgs should be | |
9f13ef67 TH |
984 | * removed while holding both q and blkcg locks. As blkcg lock is nested |
985 | * inside q lock, this function performs reverse double lock dancing. | |
986 | * | |
987 | * This is the blkcg counterpart of ioc_release_fn(). | |
988 | */ | |
eb95419b | 989 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 990 | { |
eb95419b | 991 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 992 | |
9f13ef67 | 993 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 994 | |
9f13ef67 | 995 | while (!hlist_empty(&blkcg->blkg_list)) { |
3c798398 TH |
996 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, |
997 | struct blkcg_gq, blkcg_node); | |
c875f4d0 | 998 | struct request_queue *q = blkg->q; |
b1c35769 | 999 | |
9f13ef67 TH |
1000 | if (spin_trylock(q->queue_lock)) { |
1001 | blkg_destroy(blkg); | |
1002 | spin_unlock(q->queue_lock); | |
1003 | } else { | |
1004 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 1005 | cpu_relax(); |
a5567932 | 1006 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 1007 | } |
9f13ef67 | 1008 | } |
b1c35769 | 1009 | |
9f13ef67 | 1010 | spin_unlock_irq(&blkcg->lock); |
52ebea74 TH |
1011 | |
1012 | wb_blkcg_offline(blkcg); | |
7ee9c562 TH |
1013 | } |
1014 | ||
eb95419b | 1015 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1016 | { |
eb95419b | 1017 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1018 | int i; |
7ee9c562 | 1019 | |
7876f930 | 1020 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1021 | |
7876f930 | 1022 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1023 | |
bc915e61 | 1024 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1025 | if (blkcg->cpd[i]) |
1026 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1027 | ||
1028 | mutex_unlock(&blkcg_pol_mutex); | |
1029 | ||
bc915e61 | 1030 | kfree(blkcg); |
31e4c28d VG |
1031 | } |
1032 | ||
eb95419b TH |
1033 | static struct cgroup_subsys_state * |
1034 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1035 | { |
3c798398 | 1036 | struct blkcg *blkcg; |
e48453c3 AA |
1037 | struct cgroup_subsys_state *ret; |
1038 | int i; | |
31e4c28d | 1039 | |
7876f930 TH |
1040 | mutex_lock(&blkcg_pol_mutex); |
1041 | ||
eb95419b | 1042 | if (!parent_css) { |
3c798398 | 1043 | blkcg = &blkcg_root; |
bc915e61 TH |
1044 | } else { |
1045 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1046 | if (!blkcg) { | |
1047 | ret = ERR_PTR(-ENOMEM); | |
1048 | goto free_blkcg; | |
1049 | } | |
e48453c3 AA |
1050 | } |
1051 | ||
1052 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1053 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1054 | struct blkcg_policy_data *cpd; | |
1055 | ||
1056 | /* | |
1057 | * If the policy hasn't been attached yet, wait for it | |
1058 | * to be attached before doing anything else. Otherwise, | |
1059 | * check if the policy requires any specific per-cgroup | |
1060 | * data: if it does, allocate and initialize it. | |
1061 | */ | |
e4a9bde9 | 1062 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1063 | continue; |
1064 | ||
e4a9bde9 | 1065 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
1066 | if (!cpd) { |
1067 | ret = ERR_PTR(-ENOMEM); | |
1068 | goto free_pd_blkcg; | |
1069 | } | |
81437648 TH |
1070 | blkcg->cpd[i] = cpd; |
1071 | cpd->blkcg = blkcg; | |
e48453c3 | 1072 | cpd->plid = i; |
e4a9bde9 TH |
1073 | if (pol->cpd_init_fn) |
1074 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1075 | } |
31e4c28d | 1076 | |
31e4c28d | 1077 | spin_lock_init(&blkcg->lock); |
e00f4f4d | 1078 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1079 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1080 | #ifdef CONFIG_CGROUP_WRITEBACK |
1081 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1082 | #endif | |
7876f930 TH |
1083 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1084 | ||
1085 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1086 | return &blkcg->css; |
e48453c3 AA |
1087 | |
1088 | free_pd_blkcg: | |
1089 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1090 | if (blkcg->cpd[i]) |
1091 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
e48453c3 AA |
1092 | free_blkcg: |
1093 | kfree(blkcg); | |
7876f930 | 1094 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 1095 | return ret; |
31e4c28d VG |
1096 | } |
1097 | ||
5efd6113 TH |
1098 | /** |
1099 | * blkcg_init_queue - initialize blkcg part of request queue | |
1100 | * @q: request_queue to initialize | |
1101 | * | |
1102 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1103 | * part of new request_queue @q. | |
1104 | * | |
1105 | * RETURNS: | |
1106 | * 0 on success, -errno on failure. | |
1107 | */ | |
1108 | int blkcg_init_queue(struct request_queue *q) | |
1109 | { | |
7fc6b87a | 1110 | struct blkcg_gq *blkg; |
ec13b1d6 TH |
1111 | int ret; |
1112 | ||
ec13b1d6 TH |
1113 | rcu_read_lock(); |
1114 | spin_lock_irq(q->queue_lock); | |
7fc6b87a | 1115 | blkg = __blkg_lookup_create(&blkcg_root, q, GFP_KERNEL, NULL); |
ec13b1d6 TH |
1116 | spin_unlock_irq(q->queue_lock); |
1117 | rcu_read_unlock(); | |
1118 | ||
9b54d816 | 1119 | if (IS_ERR(blkg)) |
ec13b1d6 | 1120 | return PTR_ERR(blkg); |
ec13b1d6 TH |
1121 | |
1122 | q->root_blkg = blkg; | |
1123 | q->root_rl.blkg = blkg; | |
5efd6113 | 1124 | |
ec13b1d6 TH |
1125 | ret = blk_throtl_init(q); |
1126 | if (ret) { | |
1127 | spin_lock_irq(q->queue_lock); | |
1128 | blkg_destroy_all(q); | |
1129 | spin_unlock_irq(q->queue_lock); | |
1130 | } | |
1131 | return ret; | |
5efd6113 TH |
1132 | } |
1133 | ||
1134 | /** | |
1135 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1136 | * @q: request_queue to drain | |
1137 | * | |
1138 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1139 | */ | |
1140 | void blkcg_drain_queue(struct request_queue *q) | |
1141 | { | |
1142 | lockdep_assert_held(q->queue_lock); | |
1143 | ||
0b462c89 TH |
1144 | /* |
1145 | * @q could be exiting and already have destroyed all blkgs as | |
1146 | * indicated by NULL root_blkg. If so, don't confuse policies. | |
1147 | */ | |
1148 | if (!q->root_blkg) | |
1149 | return; | |
1150 | ||
5efd6113 TH |
1151 | blk_throtl_drain(q); |
1152 | } | |
1153 | ||
1154 | /** | |
1155 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1156 | * @q: request_queue being released | |
1157 | * | |
1158 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1159 | */ | |
1160 | void blkcg_exit_queue(struct request_queue *q) | |
1161 | { | |
6d18b008 | 1162 | spin_lock_irq(q->queue_lock); |
3c96cb32 | 1163 | blkg_destroy_all(q); |
6d18b008 TH |
1164 | spin_unlock_irq(q->queue_lock); |
1165 | ||
5efd6113 TH |
1166 | blk_throtl_exit(q); |
1167 | } | |
1168 | ||
31e4c28d VG |
1169 | /* |
1170 | * We cannot support shared io contexts, as we have no mean to support | |
1171 | * two tasks with the same ioc in two different groups without major rework | |
1172 | * of the main cic data structures. For now we allow a task to change | |
1173 | * its cgroup only if it's the only owner of its ioc. | |
1174 | */ | |
1f7dd3e5 | 1175 | static int blkcg_can_attach(struct cgroup_taskset *tset) |
31e4c28d | 1176 | { |
bb9d97b6 | 1177 | struct task_struct *task; |
1f7dd3e5 | 1178 | struct cgroup_subsys_state *dst_css; |
31e4c28d VG |
1179 | struct io_context *ioc; |
1180 | int ret = 0; | |
1181 | ||
1182 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
1f7dd3e5 | 1183 | cgroup_taskset_for_each(task, dst_css, tset) { |
bb9d97b6 TH |
1184 | task_lock(task); |
1185 | ioc = task->io_context; | |
1186 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1187 | ret = -EINVAL; | |
1188 | task_unlock(task); | |
1189 | if (ret) | |
1190 | break; | |
1191 | } | |
31e4c28d VG |
1192 | return ret; |
1193 | } | |
1194 | ||
69d7fde5 TH |
1195 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1196 | { | |
1197 | int i; | |
1198 | ||
1199 | mutex_lock(&blkcg_pol_mutex); | |
1200 | ||
1201 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1202 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1203 | struct blkcg *blkcg; | |
1204 | ||
1205 | if (!pol || !pol->cpd_bind_fn) | |
1206 | continue; | |
1207 | ||
1208 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1209 | if (blkcg->cpd[pol->plid]) | |
1210 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1211 | } | |
1212 | mutex_unlock(&blkcg_pol_mutex); | |
1213 | } | |
1214 | ||
c165b3e3 | 1215 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 TH |
1216 | .css_alloc = blkcg_css_alloc, |
1217 | .css_offline = blkcg_css_offline, | |
1218 | .css_free = blkcg_css_free, | |
3c798398 | 1219 | .can_attach = blkcg_can_attach, |
69d7fde5 | 1220 | .bind = blkcg_bind, |
2ee867dc | 1221 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1222 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1223 | .legacy_name = "blkio", |
1ced953b TH |
1224 | #ifdef CONFIG_MEMCG |
1225 | /* | |
1226 | * This ensures that, if available, memcg is automatically enabled | |
1227 | * together on the default hierarchy so that the owner cgroup can | |
1228 | * be retrieved from writeback pages. | |
1229 | */ | |
1230 | .depends_on = 1 << memory_cgrp_id, | |
1231 | #endif | |
676f7c8f | 1232 | }; |
c165b3e3 | 1233 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1234 | |
a2b1693b TH |
1235 | /** |
1236 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1237 | * @q: request_queue of interest | |
1238 | * @pol: blkcg policy to activate | |
1239 | * | |
1240 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1241 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1242 | * | |
1243 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1244 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1245 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1246 | * always enough for dereferencing policy data. | |
1247 | * | |
1248 | * The caller is responsible for synchronizing [de]activations and policy | |
1249 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1250 | */ | |
1251 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1252 | const struct blkcg_policy *pol) |
a2b1693b | 1253 | { |
4c55f4f9 | 1254 | struct blkg_policy_data *pd_prealloc = NULL; |
ec13b1d6 | 1255 | struct blkcg_gq *blkg; |
4c55f4f9 | 1256 | int ret; |
a2b1693b TH |
1257 | |
1258 | if (blkcg_policy_enabled(q, pol)) | |
1259 | return 0; | |
1260 | ||
38dbb7dd | 1261 | if (q->mq_ops) |
bd166ef1 | 1262 | blk_mq_freeze_queue(q); |
38dbb7dd | 1263 | else |
bd166ef1 | 1264 | blk_queue_bypass_start(q); |
4c55f4f9 TH |
1265 | pd_prealloc: |
1266 | if (!pd_prealloc) { | |
001bea73 | 1267 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node); |
4c55f4f9 | 1268 | if (!pd_prealloc) { |
a2b1693b | 1269 | ret = -ENOMEM; |
4c55f4f9 | 1270 | goto out_bypass_end; |
a2b1693b | 1271 | } |
a2b1693b TH |
1272 | } |
1273 | ||
a2b1693b TH |
1274 | spin_lock_irq(q->queue_lock); |
1275 | ||
1276 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
4c55f4f9 TH |
1277 | struct blkg_policy_data *pd; |
1278 | ||
1279 | if (blkg->pd[pol->plid]) | |
1280 | continue; | |
a2b1693b | 1281 | |
e00f4f4d | 1282 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node); |
4c55f4f9 TH |
1283 | if (!pd) |
1284 | swap(pd, pd_prealloc); | |
1285 | if (!pd) { | |
1286 | spin_unlock_irq(q->queue_lock); | |
1287 | goto pd_prealloc; | |
1288 | } | |
a2b1693b TH |
1289 | |
1290 | blkg->pd[pol->plid] = pd; | |
1291 | pd->blkg = blkg; | |
b276a876 | 1292 | pd->plid = pol->plid; |
3e418710 | 1293 | if (pol->pd_init_fn) |
a9520cd6 | 1294 | pol->pd_init_fn(pd); |
a2b1693b TH |
1295 | } |
1296 | ||
1297 | __set_bit(pol->plid, q->blkcg_pols); | |
1298 | ret = 0; | |
4c55f4f9 | 1299 | |
a2b1693b | 1300 | spin_unlock_irq(q->queue_lock); |
4c55f4f9 | 1301 | out_bypass_end: |
38dbb7dd | 1302 | if (q->mq_ops) |
bd166ef1 | 1303 | blk_mq_unfreeze_queue(q); |
38dbb7dd | 1304 | else |
bd166ef1 | 1305 | blk_queue_bypass_end(q); |
001bea73 TH |
1306 | if (pd_prealloc) |
1307 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b TH |
1308 | return ret; |
1309 | } | |
1310 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1311 | ||
1312 | /** | |
1313 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1314 | * @q: request_queue of interest | |
1315 | * @pol: blkcg policy to deactivate | |
1316 | * | |
1317 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1318 | * blkcg_activate_policy(). | |
1319 | */ | |
1320 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1321 | const struct blkcg_policy *pol) |
a2b1693b | 1322 | { |
3c798398 | 1323 | struct blkcg_gq *blkg; |
a2b1693b TH |
1324 | |
1325 | if (!blkcg_policy_enabled(q, pol)) | |
1326 | return; | |
1327 | ||
38dbb7dd | 1328 | if (q->mq_ops) |
bd166ef1 | 1329 | blk_mq_freeze_queue(q); |
38dbb7dd | 1330 | else |
bd166ef1 JA |
1331 | blk_queue_bypass_start(q); |
1332 | ||
a2b1693b TH |
1333 | spin_lock_irq(q->queue_lock); |
1334 | ||
1335 | __clear_bit(pol->plid, q->blkcg_pols); | |
1336 | ||
1337 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1338 | /* grab blkcg lock too while removing @pd from @blkg */ | |
1339 | spin_lock(&blkg->blkcg->lock); | |
1340 | ||
001bea73 | 1341 | if (blkg->pd[pol->plid]) { |
a9520cd6 TH |
1342 | if (pol->pd_offline_fn) |
1343 | pol->pd_offline_fn(blkg->pd[pol->plid]); | |
001bea73 TH |
1344 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1345 | blkg->pd[pol->plid] = NULL; | |
1346 | } | |
a2b1693b TH |
1347 | |
1348 | spin_unlock(&blkg->blkcg->lock); | |
1349 | } | |
1350 | ||
1351 | spin_unlock_irq(q->queue_lock); | |
bd166ef1 | 1352 | |
38dbb7dd | 1353 | if (q->mq_ops) |
bd166ef1 | 1354 | blk_mq_unfreeze_queue(q); |
38dbb7dd | 1355 | else |
bd166ef1 | 1356 | blk_queue_bypass_end(q); |
a2b1693b TH |
1357 | } |
1358 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1359 | ||
8bd435b3 | 1360 | /** |
3c798398 TH |
1361 | * blkcg_policy_register - register a blkcg policy |
1362 | * @pol: blkcg policy to register | |
8bd435b3 | 1363 | * |
3c798398 TH |
1364 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1365 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1366 | */ |
d5bf0291 | 1367 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1368 | { |
06b285bd | 1369 | struct blkcg *blkcg; |
8bd435b3 | 1370 | int i, ret; |
e8989fae | 1371 | |
838f13bf | 1372 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1373 | mutex_lock(&blkcg_pol_mutex); |
1374 | ||
8bd435b3 TH |
1375 | /* find an empty slot */ |
1376 | ret = -ENOSPC; | |
1377 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1378 | if (!blkcg_policy[i]) |
8bd435b3 TH |
1379 | break; |
1380 | if (i >= BLKCG_MAX_POLS) | |
838f13bf | 1381 | goto err_unlock; |
035d10b2 | 1382 | |
06b285bd | 1383 | /* register @pol */ |
3c798398 | 1384 | pol->plid = i; |
06b285bd TH |
1385 | blkcg_policy[pol->plid] = pol; |
1386 | ||
1387 | /* allocate and install cpd's */ | |
e4a9bde9 | 1388 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1389 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1390 | struct blkcg_policy_data *cpd; | |
1391 | ||
e4a9bde9 | 1392 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1393 | if (!cpd) |
06b285bd | 1394 | goto err_free_cpds; |
06b285bd | 1395 | |
81437648 TH |
1396 | blkcg->cpd[pol->plid] = cpd; |
1397 | cpd->blkcg = blkcg; | |
06b285bd | 1398 | cpd->plid = pol->plid; |
81437648 | 1399 | pol->cpd_init_fn(cpd); |
06b285bd TH |
1400 | } |
1401 | } | |
1402 | ||
838f13bf | 1403 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1404 | |
8bd435b3 | 1405 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1406 | if (pol->dfl_cftypes) |
1407 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1408 | pol->dfl_cftypes)); | |
880f50e2 | 1409 | if (pol->legacy_cftypes) |
c165b3e3 | 1410 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1411 | pol->legacy_cftypes)); |
838f13bf TH |
1412 | mutex_unlock(&blkcg_pol_register_mutex); |
1413 | return 0; | |
1414 | ||
06b285bd | 1415 | err_free_cpds: |
e4a9bde9 | 1416 | if (pol->cpd_alloc_fn) { |
06b285bd | 1417 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1418 | if (blkcg->cpd[pol->plid]) { |
1419 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1420 | blkcg->cpd[pol->plid] = NULL; | |
1421 | } | |
06b285bd TH |
1422 | } |
1423 | } | |
1424 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1425 | err_unlock: |
bc0d6501 | 1426 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1427 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1428 | return ret; |
3e252066 | 1429 | } |
3c798398 | 1430 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1431 | |
8bd435b3 | 1432 | /** |
3c798398 TH |
1433 | * blkcg_policy_unregister - unregister a blkcg policy |
1434 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1435 | * |
3c798398 | 1436 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1437 | */ |
3c798398 | 1438 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1439 | { |
06b285bd TH |
1440 | struct blkcg *blkcg; |
1441 | ||
838f13bf | 1442 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1443 | |
3c798398 | 1444 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1445 | goto out_unlock; |
1446 | ||
1447 | /* kill the intf files first */ | |
2ee867dc TH |
1448 | if (pol->dfl_cftypes) |
1449 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1450 | if (pol->legacy_cftypes) |
1451 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1452 | |
06b285bd | 1453 | /* remove cpds and unregister */ |
838f13bf | 1454 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1455 | |
e4a9bde9 | 1456 | if (pol->cpd_alloc_fn) { |
06b285bd | 1457 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1458 | if (blkcg->cpd[pol->plid]) { |
1459 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1460 | blkcg->cpd[pol->plid] = NULL; | |
1461 | } | |
06b285bd TH |
1462 | } |
1463 | } | |
3c798398 | 1464 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1465 | |
bc0d6501 | 1466 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1467 | out_unlock: |
1468 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1469 | } |
3c798398 | 1470 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |