]>
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 | 76 | if (blkg->blkcg != &blkcg_root) |
b425e504 | 77 | blk_exit_rl(blkg->q, &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 | /* |
d708f0d5 JA |
168 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
169 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 170 | */ |
86cde6b6 | 171 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
d708f0d5 JA |
172 | struct request_queue *q, |
173 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 174 | { |
d708f0d5 | 175 | struct blkcg_gq *blkg; |
ce7acfea | 176 | struct bdi_writeback_congested *wb_congested; |
f427d909 | 177 | int i, ret; |
5624a4e4 | 178 | |
cd1604fa TH |
179 | WARN_ON_ONCE(!rcu_read_lock_held()); |
180 | lockdep_assert_held(q->queue_lock); | |
181 | ||
7ee9c562 | 182 | /* blkg holds a reference to blkcg */ |
ec903c0c | 183 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 184 | ret = -ENODEV; |
93e6d5d8 | 185 | goto err_free_blkg; |
15974993 | 186 | } |
cd1604fa | 187 | |
dc3b17cc | 188 | wb_congested = wb_congested_get_create(q->backing_dev_info, |
d708f0d5 JA |
189 | blkcg->css.id, |
190 | GFP_NOWAIT | __GFP_NOWARN); | |
191 | if (!wb_congested) { | |
ce7acfea | 192 | ret = -ENOMEM; |
d708f0d5 | 193 | goto err_put_css; |
ce7acfea TH |
194 | } |
195 | ||
d708f0d5 JA |
196 | /* allocate */ |
197 | if (!new_blkg) { | |
198 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN); | |
199 | if (unlikely(!new_blkg)) { | |
200 | ret = -ENOMEM; | |
201 | goto err_put_congested; | |
15974993 TH |
202 | } |
203 | } | |
d708f0d5 JA |
204 | blkg = new_blkg; |
205 | blkg->wb_congested = wb_congested; | |
cd1604fa | 206 | |
db613670 | 207 | /* link parent */ |
3c547865 TH |
208 | if (blkcg_parent(blkcg)) { |
209 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
210 | if (WARN_ON_ONCE(!blkg->parent)) { | |
20386ce0 | 211 | ret = -ENODEV; |
d708f0d5 | 212 | goto err_put_congested; |
3c547865 TH |
213 | } |
214 | blkg_get(blkg->parent); | |
215 | } | |
216 | ||
db613670 TH |
217 | /* invoke per-policy init */ |
218 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
219 | struct blkcg_policy *pol = blkcg_policy[i]; | |
220 | ||
221 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 222 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
223 | } |
224 | ||
225 | /* insert */ | |
cd1604fa | 226 | spin_lock(&blkcg->lock); |
a637120e TH |
227 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
228 | if (likely(!ret)) { | |
229 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
230 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
231 | |
232 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
233 | struct blkcg_policy *pol = blkcg_policy[i]; | |
234 | ||
235 | if (blkg->pd[i] && pol->pd_online_fn) | |
a9520cd6 | 236 | pol->pd_online_fn(blkg->pd[i]); |
f427d909 | 237 | } |
a637120e | 238 | } |
f427d909 | 239 | blkg->online = true; |
cd1604fa | 240 | spin_unlock(&blkcg->lock); |
496fb780 | 241 | |
ec13b1d6 | 242 | if (!ret) |
a637120e | 243 | return blkg; |
15974993 | 244 | |
3c547865 TH |
245 | /* @blkg failed fully initialized, use the usual release path */ |
246 | blkg_put(blkg); | |
247 | return ERR_PTR(ret); | |
248 | ||
d708f0d5 JA |
249 | err_put_congested: |
250 | wb_congested_put(wb_congested); | |
251 | err_put_css: | |
496fb780 | 252 | css_put(&blkcg->css); |
93e6d5d8 | 253 | err_free_blkg: |
d708f0d5 | 254 | blkg_free(new_blkg); |
93e6d5d8 | 255 | return ERR_PTR(ret); |
31e4c28d | 256 | } |
3c96cb32 | 257 | |
86cde6b6 | 258 | /** |
d708f0d5 | 259 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 TH |
260 | * @blkcg: blkcg of interest |
261 | * @q: request_queue of interest | |
262 | * | |
263 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
264 | * create one. blkg creation is performed recursively from blkcg_root such |
265 | * that all non-root blkg's have access to the parent blkg. This function | |
266 | * should be called under RCU read lock and @q->queue_lock. | |
86cde6b6 TH |
267 | * |
268 | * Returns pointer to the looked up or created blkg on success, ERR_PTR() | |
269 | * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not | |
270 | * dead and bypassing, returns ERR_PTR(-EBUSY). | |
271 | */ | |
d708f0d5 JA |
272 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
273 | struct request_queue *q) | |
3c96cb32 | 274 | { |
86cde6b6 TH |
275 | struct blkcg_gq *blkg; |
276 | ||
277 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
278 | lockdep_assert_held(q->queue_lock); | |
279 | ||
d708f0d5 JA |
280 | /* |
281 | * This could be the first entry point of blkcg implementation and | |
282 | * we shouldn't allow anything to go through for a bypassing queue. | |
283 | */ | |
284 | if (unlikely(blk_queue_bypass(q))) | |
285 | return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY); | |
286 | ||
86cde6b6 TH |
287 | blkg = __blkg_lookup(blkcg, q, true); |
288 | if (blkg) | |
289 | return blkg; | |
290 | ||
3c547865 TH |
291 | /* |
292 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
293 | * non-root blkgs have access to their parents. | |
294 | */ | |
295 | while (true) { | |
296 | struct blkcg *pos = blkcg; | |
297 | struct blkcg *parent = blkcg_parent(blkcg); | |
298 | ||
299 | while (parent && !__blkg_lookup(parent, q, false)) { | |
300 | pos = parent; | |
301 | parent = blkcg_parent(parent); | |
302 | } | |
303 | ||
d708f0d5 | 304 | blkg = blkg_create(pos, q, NULL); |
3c547865 TH |
305 | if (pos == blkcg || IS_ERR(blkg)) |
306 | return blkg; | |
307 | } | |
3c96cb32 | 308 | } |
31e4c28d | 309 | |
3c798398 | 310 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 311 | { |
3c798398 | 312 | struct blkcg *blkcg = blkg->blkcg; |
77ea7338 | 313 | struct blkcg_gq *parent = blkg->parent; |
f427d909 | 314 | int i; |
03aa264a | 315 | |
27e1f9d1 | 316 | lockdep_assert_held(blkg->q->queue_lock); |
9f13ef67 | 317 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
318 | |
319 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 320 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 321 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 322 | |
f427d909 TH |
323 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
324 | struct blkcg_policy *pol = blkcg_policy[i]; | |
325 | ||
326 | if (blkg->pd[i] && pol->pd_offline_fn) | |
a9520cd6 | 327 | pol->pd_offline_fn(blkg->pd[i]); |
f427d909 | 328 | } |
77ea7338 TH |
329 | |
330 | if (parent) { | |
331 | blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes); | |
332 | blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios); | |
333 | } | |
334 | ||
f427d909 TH |
335 | blkg->online = false; |
336 | ||
a637120e | 337 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 338 | list_del_init(&blkg->q_node); |
9f13ef67 | 339 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 340 | |
a637120e TH |
341 | /* |
342 | * Both setting lookup hint to and clearing it from @blkg are done | |
343 | * under queue_lock. If it's not pointing to @blkg now, it never | |
344 | * will. Hint assignment itself can race safely. | |
345 | */ | |
ec6c676a | 346 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
347 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
348 | ||
03aa264a TH |
349 | /* |
350 | * Put the reference taken at the time of creation so that when all | |
351 | * queues are gone, group can be destroyed. | |
352 | */ | |
353 | blkg_put(blkg); | |
354 | } | |
355 | ||
9f13ef67 TH |
356 | /** |
357 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
358 | * @q: request_queue of interest | |
9f13ef67 | 359 | * |
3c96cb32 | 360 | * Destroy all blkgs associated with @q. |
9f13ef67 | 361 | */ |
3c96cb32 | 362 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 363 | { |
3c798398 | 364 | struct blkcg_gq *blkg, *n; |
72e06c25 | 365 | |
6d18b008 | 366 | lockdep_assert_held(q->queue_lock); |
72e06c25 | 367 | |
9f13ef67 | 368 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 369 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 370 | |
9f13ef67 TH |
371 | spin_lock(&blkcg->lock); |
372 | blkg_destroy(blkg); | |
373 | spin_unlock(&blkcg->lock); | |
72e06c25 | 374 | } |
6fe810bd TH |
375 | |
376 | q->root_blkg = NULL; | |
377 | q->root_rl.blkg = NULL; | |
72e06c25 TH |
378 | } |
379 | ||
2a4fd070 TH |
380 | /* |
381 | * A group is RCU protected, but having an rcu lock does not mean that one | |
382 | * can access all the fields of blkg and assume these are valid. For | |
383 | * example, don't try to follow throtl_data and request queue links. | |
384 | * | |
385 | * Having a reference to blkg under an rcu allows accesses to only values | |
386 | * local to groups like group stats and group rate limits. | |
387 | */ | |
388 | void __blkg_release_rcu(struct rcu_head *rcu_head) | |
1adaf3dd | 389 | { |
2a4fd070 | 390 | struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head); |
db613670 | 391 | |
3c547865 | 392 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
1adaf3dd | 393 | css_put(&blkg->blkcg->css); |
a5049a8a | 394 | if (blkg->parent) |
3c547865 | 395 | blkg_put(blkg->parent); |
1adaf3dd | 396 | |
ce7acfea TH |
397 | wb_congested_put(blkg->wb_congested); |
398 | ||
2a4fd070 | 399 | blkg_free(blkg); |
1adaf3dd | 400 | } |
2a4fd070 | 401 | EXPORT_SYMBOL_GPL(__blkg_release_rcu); |
1adaf3dd | 402 | |
a051661c TH |
403 | /* |
404 | * The next function used by blk_queue_for_each_rl(). It's a bit tricky | |
405 | * because the root blkg uses @q->root_rl instead of its own rl. | |
406 | */ | |
407 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
408 | struct request_queue *q) | |
409 | { | |
410 | struct list_head *ent; | |
411 | struct blkcg_gq *blkg; | |
412 | ||
413 | /* | |
414 | * Determine the current blkg list_head. The first entry is | |
415 | * root_rl which is off @q->blkg_list and mapped to the head. | |
416 | */ | |
417 | if (rl == &q->root_rl) { | |
418 | ent = &q->blkg_list; | |
65c77fd9 JN |
419 | /* There are no more block groups, hence no request lists */ |
420 | if (list_empty(ent)) | |
421 | return NULL; | |
a051661c TH |
422 | } else { |
423 | blkg = container_of(rl, struct blkcg_gq, rl); | |
424 | ent = &blkg->q_node; | |
425 | } | |
426 | ||
427 | /* walk to the next list_head, skip root blkcg */ | |
428 | ent = ent->next; | |
429 | if (ent == &q->root_blkg->q_node) | |
430 | ent = ent->next; | |
431 | if (ent == &q->blkg_list) | |
432 | return NULL; | |
433 | ||
434 | blkg = container_of(ent, struct blkcg_gq, q_node); | |
435 | return &blkg->rl; | |
436 | } | |
437 | ||
182446d0 TH |
438 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
439 | struct cftype *cftype, u64 val) | |
303a3acb | 440 | { |
182446d0 | 441 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 442 | struct blkcg_gq *blkg; |
bc0d6501 | 443 | int i; |
303a3acb | 444 | |
838f13bf | 445 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 446 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
447 | |
448 | /* | |
449 | * Note that stat reset is racy - it doesn't synchronize against | |
450 | * stat updates. This is a debug feature which shouldn't exist | |
451 | * anyway. If you get hit by a race, retry. | |
452 | */ | |
b67bfe0d | 453 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
77ea7338 TH |
454 | blkg_rwstat_reset(&blkg->stat_bytes); |
455 | blkg_rwstat_reset(&blkg->stat_ios); | |
456 | ||
8bd435b3 | 457 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 458 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 459 | |
a9520cd6 TH |
460 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
461 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 462 | } |
303a3acb | 463 | } |
f0bdc8cd | 464 | |
303a3acb | 465 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 466 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
467 | return 0; |
468 | } | |
469 | ||
dd165eb3 | 470 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 471 | { |
d3d32e69 | 472 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
dc3b17cc JK |
473 | if (blkg->q->backing_dev_info->dev) |
474 | return dev_name(blkg->q->backing_dev_info->dev); | |
d3d32e69 | 475 | return NULL; |
303a3acb | 476 | } |
dd165eb3 | 477 | EXPORT_SYMBOL_GPL(blkg_dev_name); |
303a3acb | 478 | |
d3d32e69 TH |
479 | /** |
480 | * blkcg_print_blkgs - helper for printing per-blkg data | |
481 | * @sf: seq_file to print to | |
482 | * @blkcg: blkcg of interest | |
483 | * @prfill: fill function to print out a blkg | |
484 | * @pol: policy in question | |
485 | * @data: data to be passed to @prfill | |
486 | * @show_total: to print out sum of prfill return values or not | |
487 | * | |
488 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
489 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
490 | * policy data and @data and the matching queue lock held. If @show_total |
491 | * is %true, the sum of the return values from @prfill is printed with | |
492 | * "Total" label at the end. | |
d3d32e69 TH |
493 | * |
494 | * This is to be used to construct print functions for | |
495 | * cftype->read_seq_string method. | |
496 | */ | |
3c798398 | 497 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
498 | u64 (*prfill)(struct seq_file *, |
499 | struct blkg_policy_data *, int), | |
3c798398 | 500 | const struct blkcg_policy *pol, int data, |
ec399347 | 501 | bool show_total) |
5624a4e4 | 502 | { |
3c798398 | 503 | struct blkcg_gq *blkg; |
d3d32e69 | 504 | u64 total = 0; |
5624a4e4 | 505 | |
810ecfa7 | 506 | rcu_read_lock(); |
ee89f812 | 507 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
810ecfa7 | 508 | spin_lock_irq(blkg->q->queue_lock); |
a2b1693b | 509 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 510 | total += prfill(sf, blkg->pd[pol->plid], data); |
810ecfa7 TH |
511 | spin_unlock_irq(blkg->q->queue_lock); |
512 | } | |
513 | rcu_read_unlock(); | |
d3d32e69 TH |
514 | |
515 | if (show_total) | |
516 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
517 | } | |
829fdb50 | 518 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
519 | |
520 | /** | |
521 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
522 | * @sf: seq_file to print to | |
f95a04af | 523 | * @pd: policy private data of interest |
d3d32e69 TH |
524 | * @v: value to print |
525 | * | |
f95a04af | 526 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 527 | */ |
f95a04af | 528 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 529 | { |
f95a04af | 530 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
531 | |
532 | if (!dname) | |
533 | return 0; | |
534 | ||
535 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
536 | return v; | |
537 | } | |
829fdb50 | 538 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
539 | |
540 | /** | |
541 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
542 | * @sf: seq_file to print to | |
f95a04af | 543 | * @pd: policy private data of interest |
d3d32e69 TH |
544 | * @rwstat: rwstat to print |
545 | * | |
f95a04af | 546 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 547 | */ |
f95a04af | 548 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 549 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
550 | { |
551 | static const char *rwstr[] = { | |
552 | [BLKG_RWSTAT_READ] = "Read", | |
553 | [BLKG_RWSTAT_WRITE] = "Write", | |
554 | [BLKG_RWSTAT_SYNC] = "Sync", | |
555 | [BLKG_RWSTAT_ASYNC] = "Async", | |
556 | }; | |
f95a04af | 557 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
558 | u64 v; |
559 | int i; | |
560 | ||
561 | if (!dname) | |
562 | return 0; | |
563 | ||
564 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
565 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
24bdb8ef | 566 | (unsigned long long)atomic64_read(&rwstat->aux_cnt[i])); |
d3d32e69 | 567 | |
24bdb8ef TH |
568 | v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) + |
569 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]); | |
d3d32e69 TH |
570 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); |
571 | return v; | |
572 | } | |
b50da39f | 573 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 574 | |
5bc4afb1 TH |
575 | /** |
576 | * blkg_prfill_stat - prfill callback for blkg_stat | |
577 | * @sf: seq_file to print to | |
f95a04af TH |
578 | * @pd: policy private data of interest |
579 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
580 | * |
581 | * prfill callback for printing a blkg_stat. | |
582 | */ | |
f95a04af | 583 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 584 | { |
f95a04af | 585 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 586 | } |
5bc4afb1 | 587 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 588 | |
5bc4afb1 TH |
589 | /** |
590 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
591 | * @sf: seq_file to print to | |
f95a04af TH |
592 | * @pd: policy private data of interest |
593 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
594 | * |
595 | * prfill callback for printing a blkg_rwstat. | |
596 | */ | |
f95a04af TH |
597 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
598 | int off) | |
d3d32e69 | 599 | { |
f95a04af | 600 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 601 | |
f95a04af | 602 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 603 | } |
5bc4afb1 | 604 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 605 | |
77ea7338 TH |
606 | static u64 blkg_prfill_rwstat_field(struct seq_file *sf, |
607 | struct blkg_policy_data *pd, int off) | |
608 | { | |
609 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off); | |
610 | ||
611 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
612 | } | |
613 | ||
614 | /** | |
615 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes | |
616 | * @sf: seq_file to print to | |
617 | * @v: unused | |
618 | * | |
619 | * To be used as cftype->seq_show to print blkg->stat_bytes. | |
620 | * cftype->private must be set to the blkcg_policy. | |
621 | */ | |
622 | int blkg_print_stat_bytes(struct seq_file *sf, void *v) | |
623 | { | |
624 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
625 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
626 | offsetof(struct blkcg_gq, stat_bytes), true); | |
627 | return 0; | |
628 | } | |
629 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes); | |
630 | ||
631 | /** | |
632 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios | |
633 | * @sf: seq_file to print to | |
634 | * @v: unused | |
635 | * | |
636 | * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private | |
637 | * must be set to the blkcg_policy. | |
638 | */ | |
639 | int blkg_print_stat_ios(struct seq_file *sf, void *v) | |
640 | { | |
641 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
642 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
643 | offsetof(struct blkcg_gq, stat_ios), true); | |
644 | return 0; | |
645 | } | |
646 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios); | |
647 | ||
648 | static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf, | |
649 | struct blkg_policy_data *pd, | |
650 | int off) | |
651 | { | |
652 | struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg, | |
653 | NULL, off); | |
654 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
655 | } | |
656 | ||
657 | /** | |
658 | * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes | |
659 | * @sf: seq_file to print to | |
660 | * @v: unused | |
661 | */ | |
662 | int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v) | |
663 | { | |
664 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
665 | blkg_prfill_rwstat_field_recursive, | |
666 | (void *)seq_cft(sf)->private, | |
667 | offsetof(struct blkcg_gq, stat_bytes), true); | |
668 | return 0; | |
669 | } | |
670 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive); | |
671 | ||
672 | /** | |
673 | * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios | |
674 | * @sf: seq_file to print to | |
675 | * @v: unused | |
676 | */ | |
677 | int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v) | |
678 | { | |
679 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
680 | blkg_prfill_rwstat_field_recursive, | |
681 | (void *)seq_cft(sf)->private, | |
682 | offsetof(struct blkcg_gq, stat_ios), true); | |
683 | return 0; | |
684 | } | |
685 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive); | |
686 | ||
16b3de66 TH |
687 | /** |
688 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
f12c74ca TH |
689 | * @blkg: blkg of interest |
690 | * @pol: blkcg_policy which contains the blkg_stat | |
691 | * @off: offset to the blkg_stat in blkg_policy_data or @blkg | |
16b3de66 | 692 | * |
f12c74ca TH |
693 | * Collect the blkg_stat specified by @blkg, @pol and @off and all its |
694 | * online descendants and their aux counts. The caller must be holding the | |
695 | * queue lock for online tests. | |
696 | * | |
697 | * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is | |
698 | * at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 699 | */ |
f12c74ca TH |
700 | u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg, |
701 | struct blkcg_policy *pol, int off) | |
16b3de66 | 702 | { |
16b3de66 | 703 | struct blkcg_gq *pos_blkg; |
492eb21b | 704 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 705 | u64 sum = 0; |
16b3de66 | 706 | |
f12c74ca | 707 | lockdep_assert_held(blkg->q->queue_lock); |
16b3de66 | 708 | |
16b3de66 | 709 | rcu_read_lock(); |
f12c74ca TH |
710 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
711 | struct blkg_stat *stat; | |
712 | ||
713 | if (!pos_blkg->online) | |
714 | continue; | |
16b3de66 | 715 | |
f12c74ca TH |
716 | if (pol) |
717 | stat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
718 | else | |
719 | stat = (void *)blkg + off; | |
720 | ||
721 | sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
16b3de66 TH |
722 | } |
723 | rcu_read_unlock(); | |
724 | ||
725 | return sum; | |
726 | } | |
727 | EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum); | |
728 | ||
729 | /** | |
730 | * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat | |
f12c74ca TH |
731 | * @blkg: blkg of interest |
732 | * @pol: blkcg_policy which contains the blkg_rwstat | |
733 | * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg | |
16b3de66 | 734 | * |
f12c74ca TH |
735 | * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its |
736 | * online descendants and their aux counts. The caller must be holding the | |
737 | * queue lock for online tests. | |
738 | * | |
739 | * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it | |
740 | * is at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 741 | */ |
f12c74ca TH |
742 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, |
743 | struct blkcg_policy *pol, int off) | |
16b3de66 | 744 | { |
16b3de66 | 745 | struct blkcg_gq *pos_blkg; |
492eb21b | 746 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 747 | struct blkg_rwstat sum = { }; |
16b3de66 TH |
748 | int i; |
749 | ||
f12c74ca | 750 | lockdep_assert_held(blkg->q->queue_lock); |
16b3de66 | 751 | |
16b3de66 | 752 | rcu_read_lock(); |
f12c74ca | 753 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
3a7faead | 754 | struct blkg_rwstat *rwstat; |
16b3de66 TH |
755 | |
756 | if (!pos_blkg->online) | |
757 | continue; | |
758 | ||
f12c74ca TH |
759 | if (pol) |
760 | rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
761 | else | |
762 | rwstat = (void *)pos_blkg + off; | |
763 | ||
16b3de66 | 764 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
3a7faead TH |
765 | atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) + |
766 | percpu_counter_sum_positive(&rwstat->cpu_cnt[i]), | |
767 | &sum.aux_cnt[i]); | |
16b3de66 TH |
768 | } |
769 | rcu_read_unlock(); | |
770 | ||
771 | return sum; | |
772 | } | |
773 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
774 | ||
457e490f TE |
775 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ |
776 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, | |
777 | const struct blkcg_policy *pol, | |
778 | struct request_queue *q) | |
779 | { | |
780 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
781 | lockdep_assert_held(q->queue_lock); | |
782 | ||
783 | if (!blkcg_policy_enabled(q, pol)) | |
784 | return ERR_PTR(-EOPNOTSUPP); | |
785 | ||
786 | /* | |
787 | * This could be the first entry point of blkcg implementation and | |
788 | * we shouldn't allow anything to go through for a bypassing queue. | |
789 | */ | |
790 | if (unlikely(blk_queue_bypass(q))) | |
791 | return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY); | |
792 | ||
793 | return __blkg_lookup(blkcg, q, true /* update_hint */); | |
794 | } | |
795 | ||
3a8b31d3 TH |
796 | /** |
797 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
798 | * @blkcg: target block cgroup | |
da8b0662 | 799 | * @pol: target policy |
3a8b31d3 TH |
800 | * @input: input string |
801 | * @ctx: blkg_conf_ctx to be filled | |
802 | * | |
803 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
804 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
805 | * part of @input following MAJ:MIN. This function returns with RCU read | |
806 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 807 | */ |
3c798398 | 808 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 809 | char *input, struct blkg_conf_ctx *ctx) |
da8b0662 | 810 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
34d0f179 | 811 | { |
3a8b31d3 | 812 | struct gendisk *disk; |
457e490f | 813 | struct request_queue *q; |
3c798398 | 814 | struct blkcg_gq *blkg; |
726fa694 | 815 | unsigned int major, minor; |
36aa9e5f TH |
816 | int key_len, part, ret; |
817 | char *body; | |
34d0f179 | 818 | |
36aa9e5f | 819 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) |
726fa694 | 820 | return -EINVAL; |
3a8b31d3 | 821 | |
36aa9e5f TH |
822 | body = input + key_len; |
823 | if (!isspace(*body)) | |
824 | return -EINVAL; | |
825 | body = skip_spaces(body); | |
826 | ||
726fa694 | 827 | disk = get_gendisk(MKDEV(major, minor), &part); |
5f6c2d2b | 828 | if (!disk) |
20386ce0 | 829 | return -ENODEV; |
5f6c2d2b | 830 | if (part) { |
457e490f TE |
831 | ret = -ENODEV; |
832 | goto fail; | |
5f6c2d2b | 833 | } |
e56da7e2 | 834 | |
457e490f | 835 | q = disk->queue; |
da8b0662 | 836 | |
457e490f TE |
837 | rcu_read_lock(); |
838 | spin_lock_irq(q->queue_lock); | |
e56da7e2 | 839 | |
457e490f | 840 | blkg = blkg_lookup_check(blkcg, pol, q); |
4bfd482e TH |
841 | if (IS_ERR(blkg)) { |
842 | ret = PTR_ERR(blkg); | |
457e490f TE |
843 | goto fail_unlock; |
844 | } | |
845 | ||
846 | if (blkg) | |
847 | goto success; | |
848 | ||
849 | /* | |
850 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
851 | * non-root blkgs have access to their parents. | |
852 | */ | |
853 | while (true) { | |
854 | struct blkcg *pos = blkcg; | |
855 | struct blkcg *parent; | |
856 | struct blkcg_gq *new_blkg; | |
857 | ||
858 | parent = blkcg_parent(blkcg); | |
859 | while (parent && !__blkg_lookup(parent, q, false)) { | |
860 | pos = parent; | |
861 | parent = blkcg_parent(parent); | |
862 | } | |
863 | ||
864 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
865 | spin_unlock_irq(q->queue_lock); | |
3a8b31d3 | 866 | rcu_read_unlock(); |
457e490f TE |
867 | |
868 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); | |
869 | if (unlikely(!new_blkg)) { | |
870 | ret = -ENOMEM; | |
871 | goto fail; | |
7702e8f4 | 872 | } |
3a8b31d3 | 873 | |
457e490f TE |
874 | rcu_read_lock(); |
875 | spin_lock_irq(q->queue_lock); | |
876 | ||
877 | blkg = blkg_lookup_check(pos, pol, q); | |
878 | if (IS_ERR(blkg)) { | |
879 | ret = PTR_ERR(blkg); | |
880 | goto fail_unlock; | |
881 | } | |
882 | ||
883 | if (blkg) { | |
884 | blkg_free(new_blkg); | |
885 | } else { | |
886 | blkg = blkg_create(pos, q, new_blkg); | |
887 | if (unlikely(IS_ERR(blkg))) { | |
888 | ret = PTR_ERR(blkg); | |
889 | goto fail_unlock; | |
890 | } | |
891 | } | |
892 | ||
893 | if (pos == blkcg) | |
894 | goto success; | |
895 | } | |
896 | success: | |
3a8b31d3 TH |
897 | ctx->disk = disk; |
898 | ctx->blkg = blkg; | |
36aa9e5f | 899 | ctx->body = body; |
726fa694 | 900 | return 0; |
457e490f TE |
901 | |
902 | fail_unlock: | |
903 | spin_unlock_irq(q->queue_lock); | |
904 | rcu_read_unlock(); | |
905 | fail: | |
9df6c299 | 906 | put_disk_and_module(disk); |
457e490f TE |
907 | /* |
908 | * If queue was bypassing, we should retry. Do so after a | |
909 | * short msleep(). It isn't strictly necessary but queue | |
910 | * can be bypassing for some time and it's always nice to | |
911 | * avoid busy looping. | |
912 | */ | |
913 | if (ret == -EBUSY) { | |
914 | msleep(10); | |
915 | ret = restart_syscall(); | |
916 | } | |
917 | return ret; | |
34d0f179 | 918 | } |
829fdb50 | 919 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 920 | |
3a8b31d3 TH |
921 | /** |
922 | * blkg_conf_finish - finish up per-blkg config update | |
923 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
924 | * | |
925 | * Finish up after per-blkg config update. This function must be paired | |
926 | * with blkg_conf_prep(). | |
927 | */ | |
829fdb50 | 928 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
da8b0662 | 929 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 930 | { |
da8b0662 | 931 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
3a8b31d3 | 932 | rcu_read_unlock(); |
9df6c299 | 933 | put_disk_and_module(ctx->disk); |
34d0f179 | 934 | } |
829fdb50 | 935 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 936 | |
2ee867dc TH |
937 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
938 | { | |
939 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
940 | struct blkcg_gq *blkg; | |
941 | ||
942 | rcu_read_lock(); | |
943 | ||
944 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
945 | const char *dname; | |
946 | struct blkg_rwstat rwstat; | |
947 | u64 rbytes, wbytes, rios, wios; | |
948 | ||
949 | dname = blkg_dev_name(blkg); | |
950 | if (!dname) | |
951 | continue; | |
952 | ||
953 | spin_lock_irq(blkg->q->queue_lock); | |
954 | ||
955 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
956 | offsetof(struct blkcg_gq, stat_bytes)); | |
957 | rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
958 | wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
959 | ||
960 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
961 | offsetof(struct blkcg_gq, stat_ios)); | |
962 | rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
963 | wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
964 | ||
965 | spin_unlock_irq(blkg->q->queue_lock); | |
966 | ||
967 | if (rbytes || wbytes || rios || wios) | |
968 | seq_printf(sf, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n", | |
969 | dname, rbytes, wbytes, rios, wios); | |
970 | } | |
971 | ||
972 | rcu_read_unlock(); | |
973 | return 0; | |
974 | } | |
975 | ||
e1f3b941 | 976 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
977 | { |
978 | .name = "stat", | |
ca0752c5 | 979 | .flags = CFTYPE_NOT_ON_ROOT, |
2ee867dc TH |
980 | .seq_show = blkcg_print_stat, |
981 | }, | |
982 | { } /* terminate */ | |
983 | }; | |
984 | ||
e1f3b941 | 985 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
986 | { |
987 | .name = "reset_stats", | |
3c798398 | 988 | .write_u64 = blkcg_reset_stats, |
22084190 | 989 | }, |
4baf6e33 | 990 | { } /* terminate */ |
31e4c28d VG |
991 | }; |
992 | ||
9f13ef67 | 993 | /** |
92fb9748 | 994 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 995 | * @css: css of interest |
9f13ef67 | 996 | * |
eb95419b TH |
997 | * This function is called when @css is about to go away and responsible |
998 | * for shooting down all blkgs associated with @css. blkgs should be | |
9f13ef67 TH |
999 | * removed while holding both q and blkcg locks. As blkcg lock is nested |
1000 | * inside q lock, this function performs reverse double lock dancing. | |
1001 | * | |
1002 | * This is the blkcg counterpart of ioc_release_fn(). | |
1003 | */ | |
eb95419b | 1004 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 1005 | { |
eb95419b | 1006 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 1007 | |
9f13ef67 | 1008 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1009 | |
9f13ef67 | 1010 | while (!hlist_empty(&blkcg->blkg_list)) { |
3c798398 TH |
1011 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, |
1012 | struct blkcg_gq, blkcg_node); | |
c875f4d0 | 1013 | struct request_queue *q = blkg->q; |
b1c35769 | 1014 | |
9f13ef67 TH |
1015 | if (spin_trylock(q->queue_lock)) { |
1016 | blkg_destroy(blkg); | |
1017 | spin_unlock(q->queue_lock); | |
1018 | } else { | |
1019 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 1020 | cpu_relax(); |
a5567932 | 1021 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 1022 | } |
9f13ef67 | 1023 | } |
b1c35769 | 1024 | |
9f13ef67 | 1025 | spin_unlock_irq(&blkcg->lock); |
52ebea74 TH |
1026 | |
1027 | wb_blkcg_offline(blkcg); | |
7ee9c562 TH |
1028 | } |
1029 | ||
eb95419b | 1030 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1031 | { |
eb95419b | 1032 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1033 | int i; |
7ee9c562 | 1034 | |
7876f930 | 1035 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1036 | |
7876f930 | 1037 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1038 | |
bc915e61 | 1039 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1040 | if (blkcg->cpd[i]) |
1041 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1042 | ||
1043 | mutex_unlock(&blkcg_pol_mutex); | |
1044 | ||
bc915e61 | 1045 | kfree(blkcg); |
31e4c28d VG |
1046 | } |
1047 | ||
eb95419b TH |
1048 | static struct cgroup_subsys_state * |
1049 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1050 | { |
3c798398 | 1051 | struct blkcg *blkcg; |
e48453c3 AA |
1052 | struct cgroup_subsys_state *ret; |
1053 | int i; | |
31e4c28d | 1054 | |
7876f930 TH |
1055 | mutex_lock(&blkcg_pol_mutex); |
1056 | ||
eb95419b | 1057 | if (!parent_css) { |
3c798398 | 1058 | blkcg = &blkcg_root; |
bc915e61 TH |
1059 | } else { |
1060 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1061 | if (!blkcg) { | |
1062 | ret = ERR_PTR(-ENOMEM); | |
4c18c9e9 | 1063 | goto unlock; |
bc915e61 | 1064 | } |
e48453c3 AA |
1065 | } |
1066 | ||
1067 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1068 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1069 | struct blkcg_policy_data *cpd; | |
1070 | ||
1071 | /* | |
1072 | * If the policy hasn't been attached yet, wait for it | |
1073 | * to be attached before doing anything else. Otherwise, | |
1074 | * check if the policy requires any specific per-cgroup | |
1075 | * data: if it does, allocate and initialize it. | |
1076 | */ | |
e4a9bde9 | 1077 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1078 | continue; |
1079 | ||
e4a9bde9 | 1080 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
1081 | if (!cpd) { |
1082 | ret = ERR_PTR(-ENOMEM); | |
1083 | goto free_pd_blkcg; | |
1084 | } | |
81437648 TH |
1085 | blkcg->cpd[i] = cpd; |
1086 | cpd->blkcg = blkcg; | |
e48453c3 | 1087 | cpd->plid = i; |
e4a9bde9 TH |
1088 | if (pol->cpd_init_fn) |
1089 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1090 | } |
31e4c28d | 1091 | |
31e4c28d | 1092 | spin_lock_init(&blkcg->lock); |
e00f4f4d | 1093 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1094 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1095 | #ifdef CONFIG_CGROUP_WRITEBACK |
1096 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1097 | #endif | |
7876f930 TH |
1098 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1099 | ||
1100 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1101 | return &blkcg->css; |
e48453c3 AA |
1102 | |
1103 | free_pd_blkcg: | |
1104 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1105 | if (blkcg->cpd[i]) |
1106 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
4c18c9e9 | 1107 | |
1108 | if (blkcg != &blkcg_root) | |
1109 | kfree(blkcg); | |
1110 | unlock: | |
7876f930 | 1111 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 1112 | return ret; |
31e4c28d VG |
1113 | } |
1114 | ||
5efd6113 TH |
1115 | /** |
1116 | * blkcg_init_queue - initialize blkcg part of request queue | |
1117 | * @q: request_queue to initialize | |
1118 | * | |
1119 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1120 | * part of new request_queue @q. | |
1121 | * | |
1122 | * RETURNS: | |
1123 | * 0 on success, -errno on failure. | |
1124 | */ | |
1125 | int blkcg_init_queue(struct request_queue *q) | |
1126 | { | |
d708f0d5 JA |
1127 | struct blkcg_gq *new_blkg, *blkg; |
1128 | bool preloaded; | |
ec13b1d6 TH |
1129 | int ret; |
1130 | ||
d708f0d5 JA |
1131 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
1132 | if (!new_blkg) | |
1133 | return -ENOMEM; | |
1134 | ||
1135 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1136 | ||
1137 | /* | |
1138 | * Make sure the root blkg exists and count the existing blkgs. As | |
1139 | * @q is bypassing at this point, blkg_lookup_create() can't be | |
1140 | * used. Open code insertion. | |
1141 | */ | |
ec13b1d6 TH |
1142 | rcu_read_lock(); |
1143 | spin_lock_irq(q->queue_lock); | |
d708f0d5 | 1144 | blkg = blkg_create(&blkcg_root, q, new_blkg); |
ec13b1d6 TH |
1145 | spin_unlock_irq(q->queue_lock); |
1146 | rcu_read_unlock(); | |
1147 | ||
d708f0d5 JA |
1148 | if (preloaded) |
1149 | radix_tree_preload_end(); | |
1150 | ||
9b54d816 | 1151 | if (IS_ERR(blkg)) |
ec13b1d6 | 1152 | return PTR_ERR(blkg); |
ec13b1d6 TH |
1153 | |
1154 | q->root_blkg = blkg; | |
1155 | q->root_rl.blkg = blkg; | |
5efd6113 | 1156 | |
ec13b1d6 TH |
1157 | ret = blk_throtl_init(q); |
1158 | if (ret) { | |
1159 | spin_lock_irq(q->queue_lock); | |
1160 | blkg_destroy_all(q); | |
1161 | spin_unlock_irq(q->queue_lock); | |
1162 | } | |
1163 | return ret; | |
5efd6113 TH |
1164 | } |
1165 | ||
1166 | /** | |
1167 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1168 | * @q: request_queue to drain | |
1169 | * | |
1170 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1171 | */ | |
1172 | void blkcg_drain_queue(struct request_queue *q) | |
1173 | { | |
1174 | lockdep_assert_held(q->queue_lock); | |
1175 | ||
0b462c89 TH |
1176 | /* |
1177 | * @q could be exiting and already have destroyed all blkgs as | |
1178 | * indicated by NULL root_blkg. If so, don't confuse policies. | |
1179 | */ | |
1180 | if (!q->root_blkg) | |
1181 | return; | |
1182 | ||
5efd6113 TH |
1183 | blk_throtl_drain(q); |
1184 | } | |
1185 | ||
1186 | /** | |
1187 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1188 | * @q: request_queue being released | |
1189 | * | |
1190 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1191 | */ | |
1192 | void blkcg_exit_queue(struct request_queue *q) | |
1193 | { | |
6d18b008 | 1194 | spin_lock_irq(q->queue_lock); |
3c96cb32 | 1195 | blkg_destroy_all(q); |
6d18b008 TH |
1196 | spin_unlock_irq(q->queue_lock); |
1197 | ||
5efd6113 TH |
1198 | blk_throtl_exit(q); |
1199 | } | |
1200 | ||
31e4c28d VG |
1201 | /* |
1202 | * We cannot support shared io contexts, as we have no mean to support | |
1203 | * two tasks with the same ioc in two different groups without major rework | |
1204 | * of the main cic data structures. For now we allow a task to change | |
1205 | * its cgroup only if it's the only owner of its ioc. | |
1206 | */ | |
1f7dd3e5 | 1207 | static int blkcg_can_attach(struct cgroup_taskset *tset) |
31e4c28d | 1208 | { |
bb9d97b6 | 1209 | struct task_struct *task; |
1f7dd3e5 | 1210 | struct cgroup_subsys_state *dst_css; |
31e4c28d VG |
1211 | struct io_context *ioc; |
1212 | int ret = 0; | |
1213 | ||
1214 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
1f7dd3e5 | 1215 | cgroup_taskset_for_each(task, dst_css, tset) { |
bb9d97b6 TH |
1216 | task_lock(task); |
1217 | ioc = task->io_context; | |
1218 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1219 | ret = -EINVAL; | |
1220 | task_unlock(task); | |
1221 | if (ret) | |
1222 | break; | |
1223 | } | |
31e4c28d VG |
1224 | return ret; |
1225 | } | |
1226 | ||
69d7fde5 TH |
1227 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1228 | { | |
1229 | int i; | |
1230 | ||
1231 | mutex_lock(&blkcg_pol_mutex); | |
1232 | ||
1233 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1234 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1235 | struct blkcg *blkcg; | |
1236 | ||
1237 | if (!pol || !pol->cpd_bind_fn) | |
1238 | continue; | |
1239 | ||
1240 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1241 | if (blkcg->cpd[pol->plid]) | |
1242 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1243 | } | |
1244 | mutex_unlock(&blkcg_pol_mutex); | |
1245 | } | |
1246 | ||
c165b3e3 | 1247 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 TH |
1248 | .css_alloc = blkcg_css_alloc, |
1249 | .css_offline = blkcg_css_offline, | |
1250 | .css_free = blkcg_css_free, | |
3c798398 | 1251 | .can_attach = blkcg_can_attach, |
69d7fde5 | 1252 | .bind = blkcg_bind, |
2ee867dc | 1253 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1254 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1255 | .legacy_name = "blkio", |
1ced953b TH |
1256 | #ifdef CONFIG_MEMCG |
1257 | /* | |
1258 | * This ensures that, if available, memcg is automatically enabled | |
1259 | * together on the default hierarchy so that the owner cgroup can | |
1260 | * be retrieved from writeback pages. | |
1261 | */ | |
1262 | .depends_on = 1 << memory_cgrp_id, | |
1263 | #endif | |
676f7c8f | 1264 | }; |
c165b3e3 | 1265 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1266 | |
a2b1693b TH |
1267 | /** |
1268 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1269 | * @q: request_queue of interest | |
1270 | * @pol: blkcg policy to activate | |
1271 | * | |
1272 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1273 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1274 | * | |
1275 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1276 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1277 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1278 | * always enough for dereferencing policy data. | |
1279 | * | |
1280 | * The caller is responsible for synchronizing [de]activations and policy | |
1281 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1282 | */ | |
1283 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1284 | const struct blkcg_policy *pol) |
a2b1693b | 1285 | { |
4c55f4f9 | 1286 | struct blkg_policy_data *pd_prealloc = NULL; |
ec13b1d6 | 1287 | struct blkcg_gq *blkg; |
4c55f4f9 | 1288 | int ret; |
a2b1693b TH |
1289 | |
1290 | if (blkcg_policy_enabled(q, pol)) | |
1291 | return 0; | |
1292 | ||
38dbb7dd | 1293 | if (q->mq_ops) |
bd166ef1 | 1294 | blk_mq_freeze_queue(q); |
38dbb7dd | 1295 | else |
bd166ef1 | 1296 | blk_queue_bypass_start(q); |
4c55f4f9 TH |
1297 | pd_prealloc: |
1298 | if (!pd_prealloc) { | |
001bea73 | 1299 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node); |
4c55f4f9 | 1300 | if (!pd_prealloc) { |
a2b1693b | 1301 | ret = -ENOMEM; |
4c55f4f9 | 1302 | goto out_bypass_end; |
a2b1693b | 1303 | } |
a2b1693b TH |
1304 | } |
1305 | ||
a2b1693b TH |
1306 | spin_lock_irq(q->queue_lock); |
1307 | ||
1308 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
4c55f4f9 TH |
1309 | struct blkg_policy_data *pd; |
1310 | ||
1311 | if (blkg->pd[pol->plid]) | |
1312 | continue; | |
a2b1693b | 1313 | |
e00f4f4d | 1314 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node); |
4c55f4f9 TH |
1315 | if (!pd) |
1316 | swap(pd, pd_prealloc); | |
1317 | if (!pd) { | |
1318 | spin_unlock_irq(q->queue_lock); | |
1319 | goto pd_prealloc; | |
1320 | } | |
a2b1693b TH |
1321 | |
1322 | blkg->pd[pol->plid] = pd; | |
1323 | pd->blkg = blkg; | |
b276a876 | 1324 | pd->plid = pol->plid; |
3e418710 | 1325 | if (pol->pd_init_fn) |
a9520cd6 | 1326 | pol->pd_init_fn(pd); |
a2b1693b TH |
1327 | } |
1328 | ||
1329 | __set_bit(pol->plid, q->blkcg_pols); | |
1330 | ret = 0; | |
4c55f4f9 | 1331 | |
a2b1693b | 1332 | spin_unlock_irq(q->queue_lock); |
4c55f4f9 | 1333 | out_bypass_end: |
38dbb7dd | 1334 | if (q->mq_ops) |
bd166ef1 | 1335 | blk_mq_unfreeze_queue(q); |
38dbb7dd | 1336 | else |
bd166ef1 | 1337 | blk_queue_bypass_end(q); |
001bea73 TH |
1338 | if (pd_prealloc) |
1339 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b TH |
1340 | return ret; |
1341 | } | |
1342 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1343 | ||
1344 | /** | |
1345 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1346 | * @q: request_queue of interest | |
1347 | * @pol: blkcg policy to deactivate | |
1348 | * | |
1349 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1350 | * blkcg_activate_policy(). | |
1351 | */ | |
1352 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1353 | const struct blkcg_policy *pol) |
a2b1693b | 1354 | { |
3c798398 | 1355 | struct blkcg_gq *blkg; |
a2b1693b TH |
1356 | |
1357 | if (!blkcg_policy_enabled(q, pol)) | |
1358 | return; | |
1359 | ||
38dbb7dd | 1360 | if (q->mq_ops) |
bd166ef1 | 1361 | blk_mq_freeze_queue(q); |
38dbb7dd | 1362 | else |
bd166ef1 JA |
1363 | blk_queue_bypass_start(q); |
1364 | ||
a2b1693b TH |
1365 | spin_lock_irq(q->queue_lock); |
1366 | ||
1367 | __clear_bit(pol->plid, q->blkcg_pols); | |
1368 | ||
1369 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1370 | /* grab blkcg lock too while removing @pd from @blkg */ | |
1371 | spin_lock(&blkg->blkcg->lock); | |
1372 | ||
001bea73 | 1373 | if (blkg->pd[pol->plid]) { |
a9520cd6 TH |
1374 | if (pol->pd_offline_fn) |
1375 | pol->pd_offline_fn(blkg->pd[pol->plid]); | |
001bea73 TH |
1376 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1377 | blkg->pd[pol->plid] = NULL; | |
1378 | } | |
a2b1693b TH |
1379 | |
1380 | spin_unlock(&blkg->blkcg->lock); | |
1381 | } | |
1382 | ||
1383 | spin_unlock_irq(q->queue_lock); | |
bd166ef1 | 1384 | |
38dbb7dd | 1385 | if (q->mq_ops) |
bd166ef1 | 1386 | blk_mq_unfreeze_queue(q); |
38dbb7dd | 1387 | else |
bd166ef1 | 1388 | blk_queue_bypass_end(q); |
a2b1693b TH |
1389 | } |
1390 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1391 | ||
8bd435b3 | 1392 | /** |
3c798398 TH |
1393 | * blkcg_policy_register - register a blkcg policy |
1394 | * @pol: blkcg policy to register | |
8bd435b3 | 1395 | * |
3c798398 TH |
1396 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1397 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1398 | */ |
d5bf0291 | 1399 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1400 | { |
06b285bd | 1401 | struct blkcg *blkcg; |
8bd435b3 | 1402 | int i, ret; |
e8989fae | 1403 | |
838f13bf | 1404 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1405 | mutex_lock(&blkcg_pol_mutex); |
1406 | ||
8bd435b3 TH |
1407 | /* find an empty slot */ |
1408 | ret = -ENOSPC; | |
1409 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1410 | if (!blkcg_policy[i]) |
8bd435b3 TH |
1411 | break; |
1412 | if (i >= BLKCG_MAX_POLS) | |
838f13bf | 1413 | goto err_unlock; |
035d10b2 | 1414 | |
e8401073 | 1415 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1416 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1417 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1418 | goto err_unlock; | |
1419 | ||
06b285bd | 1420 | /* register @pol */ |
3c798398 | 1421 | pol->plid = i; |
06b285bd TH |
1422 | blkcg_policy[pol->plid] = pol; |
1423 | ||
1424 | /* allocate and install cpd's */ | |
e4a9bde9 | 1425 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1426 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1427 | struct blkcg_policy_data *cpd; | |
1428 | ||
e4a9bde9 | 1429 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1430 | if (!cpd) |
06b285bd | 1431 | goto err_free_cpds; |
06b285bd | 1432 | |
81437648 TH |
1433 | blkcg->cpd[pol->plid] = cpd; |
1434 | cpd->blkcg = blkcg; | |
06b285bd | 1435 | cpd->plid = pol->plid; |
81437648 | 1436 | pol->cpd_init_fn(cpd); |
06b285bd TH |
1437 | } |
1438 | } | |
1439 | ||
838f13bf | 1440 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1441 | |
8bd435b3 | 1442 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1443 | if (pol->dfl_cftypes) |
1444 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1445 | pol->dfl_cftypes)); | |
880f50e2 | 1446 | if (pol->legacy_cftypes) |
c165b3e3 | 1447 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1448 | pol->legacy_cftypes)); |
838f13bf TH |
1449 | mutex_unlock(&blkcg_pol_register_mutex); |
1450 | return 0; | |
1451 | ||
06b285bd | 1452 | err_free_cpds: |
58a9edce | 1453 | if (pol->cpd_free_fn) { |
06b285bd | 1454 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1455 | if (blkcg->cpd[pol->plid]) { |
1456 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1457 | blkcg->cpd[pol->plid] = NULL; | |
1458 | } | |
06b285bd TH |
1459 | } |
1460 | } | |
1461 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1462 | err_unlock: |
bc0d6501 | 1463 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1464 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1465 | return ret; |
3e252066 | 1466 | } |
3c798398 | 1467 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1468 | |
8bd435b3 | 1469 | /** |
3c798398 TH |
1470 | * blkcg_policy_unregister - unregister a blkcg policy |
1471 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1472 | * |
3c798398 | 1473 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1474 | */ |
3c798398 | 1475 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1476 | { |
06b285bd TH |
1477 | struct blkcg *blkcg; |
1478 | ||
838f13bf | 1479 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1480 | |
3c798398 | 1481 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1482 | goto out_unlock; |
1483 | ||
1484 | /* kill the intf files first */ | |
2ee867dc TH |
1485 | if (pol->dfl_cftypes) |
1486 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1487 | if (pol->legacy_cftypes) |
1488 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1489 | |
06b285bd | 1490 | /* remove cpds and unregister */ |
838f13bf | 1491 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1492 | |
58a9edce | 1493 | if (pol->cpd_free_fn) { |
06b285bd | 1494 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1495 | if (blkcg->cpd[pol->plid]) { |
1496 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1497 | blkcg->cpd[pol->plid] = NULL; | |
1498 | } | |
06b285bd TH |
1499 | } |
1500 | } | |
3c798398 | 1501 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1502 | |
bc0d6501 | 1503 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1504 | out_unlock: |
1505 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1506 | } |
3c798398 | 1507 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |