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> |
d09d8df3 | 30 | #include <linux/tracehook.h> |
5efd6113 | 31 | #include "blk.h" |
3e252066 | 32 | |
84c124da DS |
33 | #define MAX_KEY_LEN 100 |
34 | ||
838f13bf TH |
35 | /* |
36 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
37 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
38 | * policy [un]register operations including cgroup file additions / | |
39 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
40 | * allows grabbing it from cgroup callbacks. | |
41 | */ | |
42 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 43 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 44 | |
e48453c3 | 45 | struct blkcg blkcg_root; |
3c798398 | 46 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 47 | |
496d5e75 TH |
48 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
49 | ||
3c798398 | 50 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 51 | |
7876f930 TH |
52 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
53 | ||
903d23f0 JB |
54 | static bool blkcg_debug_stats = false; |
55 | ||
a2b1693b | 56 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 57 | const struct blkcg_policy *pol) |
a2b1693b TH |
58 | { |
59 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
60 | } | |
61 | ||
0381411e TH |
62 | /** |
63 | * blkg_free - free a blkg | |
64 | * @blkg: blkg to free | |
65 | * | |
66 | * Free @blkg which may be partially allocated. | |
67 | */ | |
3c798398 | 68 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 69 | { |
e8989fae | 70 | int i; |
549d3aa8 TH |
71 | |
72 | if (!blkg) | |
73 | return; | |
74 | ||
db613670 | 75 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
76 | if (blkg->pd[i]) |
77 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
e8989fae | 78 | |
994b7832 | 79 | if (blkg->blkcg != &blkcg_root) |
b425e504 | 80 | blk_exit_rl(blkg->q, &blkg->rl); |
77ea7338 TH |
81 | |
82 | blkg_rwstat_exit(&blkg->stat_ios); | |
83 | blkg_rwstat_exit(&blkg->stat_bytes); | |
549d3aa8 | 84 | kfree(blkg); |
0381411e TH |
85 | } |
86 | ||
87 | /** | |
88 | * blkg_alloc - allocate a blkg | |
89 | * @blkcg: block cgroup the new blkg is associated with | |
90 | * @q: request_queue the new blkg is associated with | |
15974993 | 91 | * @gfp_mask: allocation mask to use |
0381411e | 92 | * |
e8989fae | 93 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 94 | */ |
15974993 TH |
95 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
96 | gfp_t gfp_mask) | |
0381411e | 97 | { |
3c798398 | 98 | struct blkcg_gq *blkg; |
e8989fae | 99 | int i; |
0381411e TH |
100 | |
101 | /* alloc and init base part */ | |
15974993 | 102 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
103 | if (!blkg) |
104 | return NULL; | |
105 | ||
77ea7338 TH |
106 | if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) || |
107 | blkg_rwstat_init(&blkg->stat_ios, gfp_mask)) | |
108 | goto err_free; | |
109 | ||
c875f4d0 | 110 | blkg->q = q; |
e8989fae | 111 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 112 | blkg->blkcg = blkcg; |
a5049a8a | 113 | atomic_set(&blkg->refcnt, 1); |
0381411e | 114 | |
a051661c TH |
115 | /* root blkg uses @q->root_rl, init rl only for !root blkgs */ |
116 | if (blkcg != &blkcg_root) { | |
117 | if (blk_init_rl(&blkg->rl, q, gfp_mask)) | |
118 | goto err_free; | |
119 | blkg->rl.blkg = blkg; | |
120 | } | |
121 | ||
8bd435b3 | 122 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 123 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 124 | struct blkg_policy_data *pd; |
0381411e | 125 | |
a2b1693b | 126 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
127 | continue; |
128 | ||
129 | /* alloc per-policy data and attach it to blkg */ | |
001bea73 | 130 | pd = pol->pd_alloc_fn(gfp_mask, q->node); |
a051661c TH |
131 | if (!pd) |
132 | goto err_free; | |
549d3aa8 | 133 | |
e8989fae TH |
134 | blkg->pd[i] = pd; |
135 | pd->blkg = blkg; | |
b276a876 | 136 | pd->plid = i; |
e8989fae TH |
137 | } |
138 | ||
0381411e | 139 | return blkg; |
a051661c TH |
140 | |
141 | err_free: | |
142 | blkg_free(blkg); | |
143 | return NULL; | |
0381411e TH |
144 | } |
145 | ||
24f29046 TH |
146 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
147 | struct request_queue *q, bool update_hint) | |
80fd9979 | 148 | { |
3c798398 | 149 | struct blkcg_gq *blkg; |
80fd9979 | 150 | |
a637120e | 151 | /* |
86cde6b6 TH |
152 | * Hint didn't match. Look up from the radix tree. Note that the |
153 | * hint can only be updated under queue_lock as otherwise @blkg | |
154 | * could have already been removed from blkg_tree. The caller is | |
155 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
156 | */ |
157 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
158 | if (blkg && blkg->q == q) { |
159 | if (update_hint) { | |
160 | lockdep_assert_held(q->queue_lock); | |
161 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
162 | } | |
a637120e | 163 | return blkg; |
86cde6b6 | 164 | } |
a637120e | 165 | |
80fd9979 TH |
166 | return NULL; |
167 | } | |
ae118896 | 168 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); |
80fd9979 | 169 | |
15974993 | 170 | /* |
d708f0d5 JA |
171 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
172 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 173 | */ |
86cde6b6 | 174 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
d708f0d5 JA |
175 | struct request_queue *q, |
176 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 177 | { |
d708f0d5 | 178 | struct blkcg_gq *blkg; |
ce7acfea | 179 | struct bdi_writeback_congested *wb_congested; |
f427d909 | 180 | int i, ret; |
5624a4e4 | 181 | |
cd1604fa TH |
182 | WARN_ON_ONCE(!rcu_read_lock_held()); |
183 | lockdep_assert_held(q->queue_lock); | |
184 | ||
7ee9c562 | 185 | /* blkg holds a reference to blkcg */ |
ec903c0c | 186 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 187 | ret = -ENODEV; |
93e6d5d8 | 188 | goto err_free_blkg; |
15974993 | 189 | } |
cd1604fa | 190 | |
dc3b17cc | 191 | wb_congested = wb_congested_get_create(q->backing_dev_info, |
d708f0d5 JA |
192 | blkcg->css.id, |
193 | GFP_NOWAIT | __GFP_NOWARN); | |
194 | if (!wb_congested) { | |
ce7acfea | 195 | ret = -ENOMEM; |
d708f0d5 | 196 | goto err_put_css; |
ce7acfea TH |
197 | } |
198 | ||
d708f0d5 JA |
199 | /* allocate */ |
200 | if (!new_blkg) { | |
201 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN); | |
202 | if (unlikely(!new_blkg)) { | |
203 | ret = -ENOMEM; | |
204 | goto err_put_congested; | |
15974993 TH |
205 | } |
206 | } | |
d708f0d5 JA |
207 | blkg = new_blkg; |
208 | blkg->wb_congested = wb_congested; | |
cd1604fa | 209 | |
db613670 | 210 | /* link parent */ |
3c547865 TH |
211 | if (blkcg_parent(blkcg)) { |
212 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
213 | if (WARN_ON_ONCE(!blkg->parent)) { | |
20386ce0 | 214 | ret = -ENODEV; |
d708f0d5 | 215 | goto err_put_congested; |
3c547865 TH |
216 | } |
217 | blkg_get(blkg->parent); | |
218 | } | |
219 | ||
db613670 TH |
220 | /* invoke per-policy init */ |
221 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
222 | struct blkcg_policy *pol = blkcg_policy[i]; | |
223 | ||
224 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 225 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
226 | } |
227 | ||
228 | /* insert */ | |
cd1604fa | 229 | spin_lock(&blkcg->lock); |
a637120e TH |
230 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
231 | if (likely(!ret)) { | |
232 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
233 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
234 | |
235 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
236 | struct blkcg_policy *pol = blkcg_policy[i]; | |
237 | ||
238 | if (blkg->pd[i] && pol->pd_online_fn) | |
a9520cd6 | 239 | pol->pd_online_fn(blkg->pd[i]); |
f427d909 | 240 | } |
a637120e | 241 | } |
f427d909 | 242 | blkg->online = true; |
cd1604fa | 243 | spin_unlock(&blkcg->lock); |
496fb780 | 244 | |
ec13b1d6 | 245 | if (!ret) |
a637120e | 246 | return blkg; |
15974993 | 247 | |
3c547865 TH |
248 | /* @blkg failed fully initialized, use the usual release path */ |
249 | blkg_put(blkg); | |
250 | return ERR_PTR(ret); | |
251 | ||
d708f0d5 JA |
252 | err_put_congested: |
253 | wb_congested_put(wb_congested); | |
254 | err_put_css: | |
496fb780 | 255 | css_put(&blkcg->css); |
93e6d5d8 | 256 | err_free_blkg: |
d708f0d5 | 257 | blkg_free(new_blkg); |
93e6d5d8 | 258 | return ERR_PTR(ret); |
31e4c28d | 259 | } |
3c96cb32 | 260 | |
86cde6b6 | 261 | /** |
d708f0d5 | 262 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 TH |
263 | * @blkcg: blkcg of interest |
264 | * @q: request_queue of interest | |
265 | * | |
266 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
267 | * create one. blkg creation is performed recursively from blkcg_root such |
268 | * that all non-root blkg's have access to the parent blkg. This function | |
269 | * should be called under RCU read lock and @q->queue_lock. | |
86cde6b6 TH |
270 | * |
271 | * Returns pointer to the looked up or created blkg on success, ERR_PTR() | |
272 | * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not | |
273 | * dead and bypassing, returns ERR_PTR(-EBUSY). | |
274 | */ | |
d708f0d5 JA |
275 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
276 | struct request_queue *q) | |
3c96cb32 | 277 | { |
86cde6b6 TH |
278 | struct blkcg_gq *blkg; |
279 | ||
280 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
281 | lockdep_assert_held(q->queue_lock); | |
282 | ||
d708f0d5 JA |
283 | /* |
284 | * This could be the first entry point of blkcg implementation and | |
285 | * we shouldn't allow anything to go through for a bypassing queue. | |
286 | */ | |
287 | if (unlikely(blk_queue_bypass(q))) | |
288 | return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY); | |
289 | ||
86cde6b6 TH |
290 | blkg = __blkg_lookup(blkcg, q, true); |
291 | if (blkg) | |
292 | return blkg; | |
293 | ||
3c547865 TH |
294 | /* |
295 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
296 | * non-root blkgs have access to their parents. | |
297 | */ | |
298 | while (true) { | |
299 | struct blkcg *pos = blkcg; | |
300 | struct blkcg *parent = blkcg_parent(blkcg); | |
301 | ||
302 | while (parent && !__blkg_lookup(parent, q, false)) { | |
303 | pos = parent; | |
304 | parent = blkcg_parent(parent); | |
305 | } | |
306 | ||
d708f0d5 | 307 | blkg = blkg_create(pos, q, NULL); |
3c547865 TH |
308 | if (pos == blkcg || IS_ERR(blkg)) |
309 | return blkg; | |
310 | } | |
3c96cb32 | 311 | } |
31e4c28d | 312 | |
4c699480 JQ |
313 | static void blkg_pd_offline(struct blkcg_gq *blkg) |
314 | { | |
315 | int i; | |
316 | ||
317 | lockdep_assert_held(blkg->q->queue_lock); | |
318 | lockdep_assert_held(&blkg->blkcg->lock); | |
319 | ||
320 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
321 | struct blkcg_policy *pol = blkcg_policy[i]; | |
322 | ||
323 | if (blkg->pd[i] && !blkg->pd[i]->offline && | |
324 | pol->pd_offline_fn) { | |
325 | pol->pd_offline_fn(blkg->pd[i]); | |
326 | blkg->pd[i]->offline = true; | |
327 | } | |
328 | } | |
329 | } | |
330 | ||
3c798398 | 331 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 332 | { |
3c798398 | 333 | struct blkcg *blkcg = blkg->blkcg; |
77ea7338 | 334 | struct blkcg_gq *parent = blkg->parent; |
03aa264a | 335 | |
27e1f9d1 | 336 | lockdep_assert_held(blkg->q->queue_lock); |
9f13ef67 | 337 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
338 | |
339 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 340 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 341 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 342 | |
77ea7338 TH |
343 | if (parent) { |
344 | blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes); | |
345 | blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios); | |
346 | } | |
347 | ||
f427d909 TH |
348 | blkg->online = false; |
349 | ||
a637120e | 350 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 351 | list_del_init(&blkg->q_node); |
9f13ef67 | 352 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 353 | |
a637120e TH |
354 | /* |
355 | * Both setting lookup hint to and clearing it from @blkg are done | |
356 | * under queue_lock. If it's not pointing to @blkg now, it never | |
357 | * will. Hint assignment itself can race safely. | |
358 | */ | |
ec6c676a | 359 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
360 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
361 | ||
03aa264a TH |
362 | /* |
363 | * Put the reference taken at the time of creation so that when all | |
364 | * queues are gone, group can be destroyed. | |
365 | */ | |
366 | blkg_put(blkg); | |
367 | } | |
368 | ||
9f13ef67 TH |
369 | /** |
370 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
371 | * @q: request_queue of interest | |
9f13ef67 | 372 | * |
3c96cb32 | 373 | * Destroy all blkgs associated with @q. |
9f13ef67 | 374 | */ |
3c96cb32 | 375 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 376 | { |
3c798398 | 377 | struct blkcg_gq *blkg, *n; |
72e06c25 | 378 | |
6d18b008 | 379 | lockdep_assert_held(q->queue_lock); |
72e06c25 | 380 | |
9f13ef67 | 381 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 382 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 383 | |
9f13ef67 | 384 | spin_lock(&blkcg->lock); |
4c699480 | 385 | blkg_pd_offline(blkg); |
9f13ef67 TH |
386 | blkg_destroy(blkg); |
387 | spin_unlock(&blkcg->lock); | |
72e06c25 | 388 | } |
6fe810bd TH |
389 | |
390 | q->root_blkg = NULL; | |
391 | q->root_rl.blkg = NULL; | |
72e06c25 TH |
392 | } |
393 | ||
2a4fd070 TH |
394 | /* |
395 | * A group is RCU protected, but having an rcu lock does not mean that one | |
396 | * can access all the fields of blkg and assume these are valid. For | |
397 | * example, don't try to follow throtl_data and request queue links. | |
398 | * | |
399 | * Having a reference to blkg under an rcu allows accesses to only values | |
400 | * local to groups like group stats and group rate limits. | |
401 | */ | |
402 | void __blkg_release_rcu(struct rcu_head *rcu_head) | |
1adaf3dd | 403 | { |
2a4fd070 | 404 | struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head); |
db613670 | 405 | |
3c547865 | 406 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
1adaf3dd | 407 | css_put(&blkg->blkcg->css); |
a5049a8a | 408 | if (blkg->parent) |
3c547865 | 409 | blkg_put(blkg->parent); |
1adaf3dd | 410 | |
ce7acfea TH |
411 | wb_congested_put(blkg->wb_congested); |
412 | ||
2a4fd070 | 413 | blkg_free(blkg); |
1adaf3dd | 414 | } |
2a4fd070 | 415 | EXPORT_SYMBOL_GPL(__blkg_release_rcu); |
1adaf3dd | 416 | |
a051661c TH |
417 | /* |
418 | * The next function used by blk_queue_for_each_rl(). It's a bit tricky | |
419 | * because the root blkg uses @q->root_rl instead of its own rl. | |
420 | */ | |
421 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
422 | struct request_queue *q) | |
423 | { | |
424 | struct list_head *ent; | |
425 | struct blkcg_gq *blkg; | |
426 | ||
427 | /* | |
428 | * Determine the current blkg list_head. The first entry is | |
429 | * root_rl which is off @q->blkg_list and mapped to the head. | |
430 | */ | |
431 | if (rl == &q->root_rl) { | |
432 | ent = &q->blkg_list; | |
65c77fd9 JN |
433 | /* There are no more block groups, hence no request lists */ |
434 | if (list_empty(ent)) | |
435 | return NULL; | |
a051661c TH |
436 | } else { |
437 | blkg = container_of(rl, struct blkcg_gq, rl); | |
438 | ent = &blkg->q_node; | |
439 | } | |
440 | ||
441 | /* walk to the next list_head, skip root blkcg */ | |
442 | ent = ent->next; | |
443 | if (ent == &q->root_blkg->q_node) | |
444 | ent = ent->next; | |
445 | if (ent == &q->blkg_list) | |
446 | return NULL; | |
447 | ||
448 | blkg = container_of(ent, struct blkcg_gq, q_node); | |
449 | return &blkg->rl; | |
450 | } | |
451 | ||
182446d0 TH |
452 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
453 | struct cftype *cftype, u64 val) | |
303a3acb | 454 | { |
182446d0 | 455 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 456 | struct blkcg_gq *blkg; |
bc0d6501 | 457 | int i; |
303a3acb | 458 | |
838f13bf | 459 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 460 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
461 | |
462 | /* | |
463 | * Note that stat reset is racy - it doesn't synchronize against | |
464 | * stat updates. This is a debug feature which shouldn't exist | |
465 | * anyway. If you get hit by a race, retry. | |
466 | */ | |
b67bfe0d | 467 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
77ea7338 TH |
468 | blkg_rwstat_reset(&blkg->stat_bytes); |
469 | blkg_rwstat_reset(&blkg->stat_ios); | |
470 | ||
8bd435b3 | 471 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 472 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 473 | |
a9520cd6 TH |
474 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
475 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 476 | } |
303a3acb | 477 | } |
f0bdc8cd | 478 | |
303a3acb | 479 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 480 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
481 | return 0; |
482 | } | |
483 | ||
dd165eb3 | 484 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 485 | { |
d3d32e69 | 486 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
dc3b17cc JK |
487 | if (blkg->q->backing_dev_info->dev) |
488 | return dev_name(blkg->q->backing_dev_info->dev); | |
d3d32e69 | 489 | return NULL; |
303a3acb | 490 | } |
dd165eb3 | 491 | EXPORT_SYMBOL_GPL(blkg_dev_name); |
303a3acb | 492 | |
d3d32e69 TH |
493 | /** |
494 | * blkcg_print_blkgs - helper for printing per-blkg data | |
495 | * @sf: seq_file to print to | |
496 | * @blkcg: blkcg of interest | |
497 | * @prfill: fill function to print out a blkg | |
498 | * @pol: policy in question | |
499 | * @data: data to be passed to @prfill | |
500 | * @show_total: to print out sum of prfill return values or not | |
501 | * | |
502 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
503 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
504 | * policy data and @data and the matching queue lock held. If @show_total |
505 | * is %true, the sum of the return values from @prfill is printed with | |
506 | * "Total" label at the end. | |
d3d32e69 TH |
507 | * |
508 | * This is to be used to construct print functions for | |
509 | * cftype->read_seq_string method. | |
510 | */ | |
3c798398 | 511 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
512 | u64 (*prfill)(struct seq_file *, |
513 | struct blkg_policy_data *, int), | |
3c798398 | 514 | const struct blkcg_policy *pol, int data, |
ec399347 | 515 | bool show_total) |
5624a4e4 | 516 | { |
3c798398 | 517 | struct blkcg_gq *blkg; |
d3d32e69 | 518 | u64 total = 0; |
5624a4e4 | 519 | |
810ecfa7 | 520 | rcu_read_lock(); |
ee89f812 | 521 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
810ecfa7 | 522 | spin_lock_irq(blkg->q->queue_lock); |
a2b1693b | 523 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 524 | total += prfill(sf, blkg->pd[pol->plid], data); |
810ecfa7 TH |
525 | spin_unlock_irq(blkg->q->queue_lock); |
526 | } | |
527 | rcu_read_unlock(); | |
d3d32e69 TH |
528 | |
529 | if (show_total) | |
530 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
531 | } | |
829fdb50 | 532 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
533 | |
534 | /** | |
535 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
536 | * @sf: seq_file to print to | |
f95a04af | 537 | * @pd: policy private data of interest |
d3d32e69 TH |
538 | * @v: value to print |
539 | * | |
f95a04af | 540 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 541 | */ |
f95a04af | 542 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 543 | { |
f95a04af | 544 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
545 | |
546 | if (!dname) | |
547 | return 0; | |
548 | ||
549 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
550 | return v; | |
551 | } | |
829fdb50 | 552 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
553 | |
554 | /** | |
555 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
556 | * @sf: seq_file to print to | |
f95a04af | 557 | * @pd: policy private data of interest |
d3d32e69 TH |
558 | * @rwstat: rwstat to print |
559 | * | |
f95a04af | 560 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 561 | */ |
f95a04af | 562 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 563 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
564 | { |
565 | static const char *rwstr[] = { | |
566 | [BLKG_RWSTAT_READ] = "Read", | |
567 | [BLKG_RWSTAT_WRITE] = "Write", | |
568 | [BLKG_RWSTAT_SYNC] = "Sync", | |
569 | [BLKG_RWSTAT_ASYNC] = "Async", | |
636620b6 | 570 | [BLKG_RWSTAT_DISCARD] = "Discard", |
d3d32e69 | 571 | }; |
f95a04af | 572 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
573 | u64 v; |
574 | int i; | |
575 | ||
576 | if (!dname) | |
577 | return 0; | |
578 | ||
579 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
580 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
24bdb8ef | 581 | (unsigned long long)atomic64_read(&rwstat->aux_cnt[i])); |
d3d32e69 | 582 | |
24bdb8ef | 583 | v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) + |
636620b6 TH |
584 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]) + |
585 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_DISCARD]); | |
d3d32e69 TH |
586 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); |
587 | return v; | |
588 | } | |
b50da39f | 589 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 590 | |
5bc4afb1 TH |
591 | /** |
592 | * blkg_prfill_stat - prfill callback for blkg_stat | |
593 | * @sf: seq_file to print to | |
f95a04af TH |
594 | * @pd: policy private data of interest |
595 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
596 | * |
597 | * prfill callback for printing a blkg_stat. | |
598 | */ | |
f95a04af | 599 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 600 | { |
f95a04af | 601 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 602 | } |
5bc4afb1 | 603 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 604 | |
5bc4afb1 TH |
605 | /** |
606 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
607 | * @sf: seq_file to print to | |
f95a04af TH |
608 | * @pd: policy private data of interest |
609 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
610 | * |
611 | * prfill callback for printing a blkg_rwstat. | |
612 | */ | |
f95a04af TH |
613 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
614 | int off) | |
d3d32e69 | 615 | { |
f95a04af | 616 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 617 | |
f95a04af | 618 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 619 | } |
5bc4afb1 | 620 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 621 | |
77ea7338 TH |
622 | static u64 blkg_prfill_rwstat_field(struct seq_file *sf, |
623 | struct blkg_policy_data *pd, int off) | |
624 | { | |
625 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off); | |
626 | ||
627 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
628 | } | |
629 | ||
630 | /** | |
631 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes | |
632 | * @sf: seq_file to print to | |
633 | * @v: unused | |
634 | * | |
635 | * To be used as cftype->seq_show to print blkg->stat_bytes. | |
636 | * cftype->private must be set to the blkcg_policy. | |
637 | */ | |
638 | int blkg_print_stat_bytes(struct seq_file *sf, void *v) | |
639 | { | |
640 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
641 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
642 | offsetof(struct blkcg_gq, stat_bytes), true); | |
643 | return 0; | |
644 | } | |
645 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes); | |
646 | ||
647 | /** | |
648 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios | |
649 | * @sf: seq_file to print to | |
650 | * @v: unused | |
651 | * | |
652 | * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private | |
653 | * must be set to the blkcg_policy. | |
654 | */ | |
655 | int blkg_print_stat_ios(struct seq_file *sf, void *v) | |
656 | { | |
657 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
658 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
659 | offsetof(struct blkcg_gq, stat_ios), true); | |
660 | return 0; | |
661 | } | |
662 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios); | |
663 | ||
664 | static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf, | |
665 | struct blkg_policy_data *pd, | |
666 | int off) | |
667 | { | |
668 | struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg, | |
669 | NULL, off); | |
670 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
671 | } | |
672 | ||
673 | /** | |
674 | * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes | |
675 | * @sf: seq_file to print to | |
676 | * @v: unused | |
677 | */ | |
678 | int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v) | |
679 | { | |
680 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
681 | blkg_prfill_rwstat_field_recursive, | |
682 | (void *)seq_cft(sf)->private, | |
683 | offsetof(struct blkcg_gq, stat_bytes), true); | |
684 | return 0; | |
685 | } | |
686 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive); | |
687 | ||
688 | /** | |
689 | * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios | |
690 | * @sf: seq_file to print to | |
691 | * @v: unused | |
692 | */ | |
693 | int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v) | |
694 | { | |
695 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
696 | blkg_prfill_rwstat_field_recursive, | |
697 | (void *)seq_cft(sf)->private, | |
698 | offsetof(struct blkcg_gq, stat_ios), true); | |
699 | return 0; | |
700 | } | |
701 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive); | |
702 | ||
16b3de66 TH |
703 | /** |
704 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
f12c74ca TH |
705 | * @blkg: blkg of interest |
706 | * @pol: blkcg_policy which contains the blkg_stat | |
707 | * @off: offset to the blkg_stat in blkg_policy_data or @blkg | |
16b3de66 | 708 | * |
f12c74ca TH |
709 | * Collect the blkg_stat specified by @blkg, @pol and @off and all its |
710 | * online descendants and their aux counts. The caller must be holding the | |
711 | * queue lock for online tests. | |
712 | * | |
713 | * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is | |
714 | * at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 715 | */ |
f12c74ca TH |
716 | u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg, |
717 | struct blkcg_policy *pol, int off) | |
16b3de66 | 718 | { |
16b3de66 | 719 | struct blkcg_gq *pos_blkg; |
492eb21b | 720 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 721 | u64 sum = 0; |
16b3de66 | 722 | |
f12c74ca | 723 | lockdep_assert_held(blkg->q->queue_lock); |
16b3de66 | 724 | |
16b3de66 | 725 | rcu_read_lock(); |
f12c74ca TH |
726 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
727 | struct blkg_stat *stat; | |
728 | ||
729 | if (!pos_blkg->online) | |
730 | continue; | |
16b3de66 | 731 | |
f12c74ca TH |
732 | if (pol) |
733 | stat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
734 | else | |
735 | stat = (void *)blkg + off; | |
736 | ||
737 | sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
16b3de66 TH |
738 | } |
739 | rcu_read_unlock(); | |
740 | ||
741 | return sum; | |
742 | } | |
743 | EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum); | |
744 | ||
745 | /** | |
746 | * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat | |
f12c74ca TH |
747 | * @blkg: blkg of interest |
748 | * @pol: blkcg_policy which contains the blkg_rwstat | |
749 | * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg | |
16b3de66 | 750 | * |
f12c74ca TH |
751 | * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its |
752 | * online descendants and their aux counts. The caller must be holding the | |
753 | * queue lock for online tests. | |
754 | * | |
755 | * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it | |
756 | * is at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 757 | */ |
f12c74ca TH |
758 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, |
759 | struct blkcg_policy *pol, int off) | |
16b3de66 | 760 | { |
16b3de66 | 761 | struct blkcg_gq *pos_blkg; |
492eb21b | 762 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 763 | struct blkg_rwstat sum = { }; |
16b3de66 TH |
764 | int i; |
765 | ||
f12c74ca | 766 | lockdep_assert_held(blkg->q->queue_lock); |
16b3de66 | 767 | |
16b3de66 | 768 | rcu_read_lock(); |
f12c74ca | 769 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
3a7faead | 770 | struct blkg_rwstat *rwstat; |
16b3de66 TH |
771 | |
772 | if (!pos_blkg->online) | |
773 | continue; | |
774 | ||
f12c74ca TH |
775 | if (pol) |
776 | rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
777 | else | |
778 | rwstat = (void *)pos_blkg + off; | |
779 | ||
16b3de66 | 780 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
3a7faead TH |
781 | atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) + |
782 | percpu_counter_sum_positive(&rwstat->cpu_cnt[i]), | |
783 | &sum.aux_cnt[i]); | |
16b3de66 TH |
784 | } |
785 | rcu_read_unlock(); | |
786 | ||
787 | return sum; | |
788 | } | |
789 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
790 | ||
457e490f TE |
791 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ |
792 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, | |
793 | const struct blkcg_policy *pol, | |
794 | struct request_queue *q) | |
795 | { | |
796 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
797 | lockdep_assert_held(q->queue_lock); | |
798 | ||
799 | if (!blkcg_policy_enabled(q, pol)) | |
800 | return ERR_PTR(-EOPNOTSUPP); | |
801 | ||
802 | /* | |
803 | * This could be the first entry point of blkcg implementation and | |
804 | * we shouldn't allow anything to go through for a bypassing queue. | |
805 | */ | |
806 | if (unlikely(blk_queue_bypass(q))) | |
807 | return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY); | |
808 | ||
809 | return __blkg_lookup(blkcg, q, true /* update_hint */); | |
810 | } | |
811 | ||
3a8b31d3 TH |
812 | /** |
813 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
814 | * @blkcg: target block cgroup | |
da8b0662 | 815 | * @pol: target policy |
3a8b31d3 TH |
816 | * @input: input string |
817 | * @ctx: blkg_conf_ctx to be filled | |
818 | * | |
819 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
820 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
821 | * part of @input following MAJ:MIN. This function returns with RCU read | |
822 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 823 | */ |
3c798398 | 824 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 825 | char *input, struct blkg_conf_ctx *ctx) |
da8b0662 | 826 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
34d0f179 | 827 | { |
3a8b31d3 | 828 | struct gendisk *disk; |
457e490f | 829 | struct request_queue *q; |
3c798398 | 830 | struct blkcg_gq *blkg; |
726fa694 | 831 | unsigned int major, minor; |
36aa9e5f TH |
832 | int key_len, part, ret; |
833 | char *body; | |
34d0f179 | 834 | |
36aa9e5f | 835 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) |
726fa694 | 836 | return -EINVAL; |
3a8b31d3 | 837 | |
36aa9e5f TH |
838 | body = input + key_len; |
839 | if (!isspace(*body)) | |
840 | return -EINVAL; | |
841 | body = skip_spaces(body); | |
842 | ||
726fa694 | 843 | disk = get_gendisk(MKDEV(major, minor), &part); |
5f6c2d2b | 844 | if (!disk) |
20386ce0 | 845 | return -ENODEV; |
5f6c2d2b | 846 | if (part) { |
457e490f TE |
847 | ret = -ENODEV; |
848 | goto fail; | |
5f6c2d2b | 849 | } |
e56da7e2 | 850 | |
457e490f | 851 | q = disk->queue; |
da8b0662 | 852 | |
457e490f TE |
853 | rcu_read_lock(); |
854 | spin_lock_irq(q->queue_lock); | |
e56da7e2 | 855 | |
457e490f | 856 | blkg = blkg_lookup_check(blkcg, pol, q); |
4bfd482e TH |
857 | if (IS_ERR(blkg)) { |
858 | ret = PTR_ERR(blkg); | |
457e490f TE |
859 | goto fail_unlock; |
860 | } | |
861 | ||
862 | if (blkg) | |
863 | goto success; | |
864 | ||
865 | /* | |
866 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
867 | * non-root blkgs have access to their parents. | |
868 | */ | |
869 | while (true) { | |
870 | struct blkcg *pos = blkcg; | |
871 | struct blkcg *parent; | |
872 | struct blkcg_gq *new_blkg; | |
873 | ||
874 | parent = blkcg_parent(blkcg); | |
875 | while (parent && !__blkg_lookup(parent, q, false)) { | |
876 | pos = parent; | |
877 | parent = blkcg_parent(parent); | |
878 | } | |
879 | ||
880 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
881 | spin_unlock_irq(q->queue_lock); | |
3a8b31d3 | 882 | rcu_read_unlock(); |
457e490f TE |
883 | |
884 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); | |
885 | if (unlikely(!new_blkg)) { | |
886 | ret = -ENOMEM; | |
887 | goto fail; | |
7702e8f4 | 888 | } |
3a8b31d3 | 889 | |
457e490f TE |
890 | rcu_read_lock(); |
891 | spin_lock_irq(q->queue_lock); | |
892 | ||
893 | blkg = blkg_lookup_check(pos, pol, q); | |
894 | if (IS_ERR(blkg)) { | |
895 | ret = PTR_ERR(blkg); | |
896 | goto fail_unlock; | |
897 | } | |
898 | ||
899 | if (blkg) { | |
900 | blkg_free(new_blkg); | |
901 | } else { | |
902 | blkg = blkg_create(pos, q, new_blkg); | |
903 | if (unlikely(IS_ERR(blkg))) { | |
904 | ret = PTR_ERR(blkg); | |
905 | goto fail_unlock; | |
906 | } | |
907 | } | |
908 | ||
909 | if (pos == blkcg) | |
910 | goto success; | |
911 | } | |
912 | success: | |
3a8b31d3 TH |
913 | ctx->disk = disk; |
914 | ctx->blkg = blkg; | |
36aa9e5f | 915 | ctx->body = body; |
726fa694 | 916 | return 0; |
457e490f TE |
917 | |
918 | fail_unlock: | |
919 | spin_unlock_irq(q->queue_lock); | |
920 | rcu_read_unlock(); | |
921 | fail: | |
9df6c299 | 922 | put_disk_and_module(disk); |
457e490f TE |
923 | /* |
924 | * If queue was bypassing, we should retry. Do so after a | |
925 | * short msleep(). It isn't strictly necessary but queue | |
926 | * can be bypassing for some time and it's always nice to | |
927 | * avoid busy looping. | |
928 | */ | |
929 | if (ret == -EBUSY) { | |
930 | msleep(10); | |
931 | ret = restart_syscall(); | |
932 | } | |
933 | return ret; | |
34d0f179 | 934 | } |
829fdb50 | 935 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 936 | |
3a8b31d3 TH |
937 | /** |
938 | * blkg_conf_finish - finish up per-blkg config update | |
939 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
940 | * | |
941 | * Finish up after per-blkg config update. This function must be paired | |
942 | * with blkg_conf_prep(). | |
943 | */ | |
829fdb50 | 944 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
da8b0662 | 945 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 946 | { |
da8b0662 | 947 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
3a8b31d3 | 948 | rcu_read_unlock(); |
9df6c299 | 949 | put_disk_and_module(ctx->disk); |
34d0f179 | 950 | } |
829fdb50 | 951 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 952 | |
2ee867dc TH |
953 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
954 | { | |
955 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
956 | struct blkcg_gq *blkg; | |
957 | ||
958 | rcu_read_lock(); | |
959 | ||
960 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
961 | const char *dname; | |
903d23f0 | 962 | char *buf; |
2ee867dc | 963 | struct blkg_rwstat rwstat; |
636620b6 | 964 | u64 rbytes, wbytes, rios, wios, dbytes, dios; |
903d23f0 JB |
965 | size_t size = seq_get_buf(sf, &buf), off = 0; |
966 | int i; | |
967 | bool has_stats = false; | |
2ee867dc TH |
968 | |
969 | dname = blkg_dev_name(blkg); | |
970 | if (!dname) | |
971 | continue; | |
972 | ||
903d23f0 JB |
973 | /* |
974 | * Hooray string manipulation, count is the size written NOT | |
975 | * INCLUDING THE \0, so size is now count+1 less than what we | |
976 | * had before, but we want to start writing the next bit from | |
977 | * the \0 so we only add count to buf. | |
978 | */ | |
979 | off += scnprintf(buf+off, size-off, "%s ", dname); | |
980 | ||
2ee867dc TH |
981 | spin_lock_irq(blkg->q->queue_lock); |
982 | ||
983 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
984 | offsetof(struct blkcg_gq, stat_bytes)); | |
985 | rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
986 | wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
636620b6 | 987 | dbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_DISCARD]); |
2ee867dc TH |
988 | |
989 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
990 | offsetof(struct blkcg_gq, stat_ios)); | |
991 | rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
992 | wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
636620b6 | 993 | dios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_DISCARD]); |
2ee867dc TH |
994 | |
995 | spin_unlock_irq(blkg->q->queue_lock); | |
996 | ||
903d23f0 JB |
997 | if (rbytes || wbytes || rios || wios) { |
998 | has_stats = true; | |
999 | off += scnprintf(buf+off, size-off, | |
636620b6 TH |
1000 | "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
1001 | rbytes, wbytes, rios, wios, | |
1002 | dbytes, dios); | |
903d23f0 JB |
1003 | } |
1004 | ||
1005 | if (!blkcg_debug_stats) | |
1006 | goto next; | |
1007 | ||
d09d8df3 JB |
1008 | if (atomic_read(&blkg->use_delay)) { |
1009 | has_stats = true; | |
1010 | off += scnprintf(buf+off, size-off, | |
1011 | " use_delay=%d delay_nsec=%llu", | |
1012 | atomic_read(&blkg->use_delay), | |
1013 | (unsigned long long)atomic64_read(&blkg->delay_nsec)); | |
1014 | } | |
1015 | ||
903d23f0 JB |
1016 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
1017 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1018 | size_t written; | |
1019 | ||
1020 | if (!blkg->pd[i] || !pol->pd_stat_fn) | |
1021 | continue; | |
1022 | ||
1023 | written = pol->pd_stat_fn(blkg->pd[i], buf+off, size-off); | |
1024 | if (written) | |
1025 | has_stats = true; | |
1026 | off += written; | |
1027 | } | |
1028 | next: | |
1029 | if (has_stats) { | |
1030 | off += scnprintf(buf+off, size-off, "\n"); | |
1031 | seq_commit(sf, off); | |
1032 | } | |
2ee867dc TH |
1033 | } |
1034 | ||
1035 | rcu_read_unlock(); | |
1036 | return 0; | |
1037 | } | |
1038 | ||
e1f3b941 | 1039 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
1040 | { |
1041 | .name = "stat", | |
ca0752c5 | 1042 | .flags = CFTYPE_NOT_ON_ROOT, |
2ee867dc TH |
1043 | .seq_show = blkcg_print_stat, |
1044 | }, | |
1045 | { } /* terminate */ | |
1046 | }; | |
1047 | ||
e1f3b941 | 1048 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
1049 | { |
1050 | .name = "reset_stats", | |
3c798398 | 1051 | .write_u64 = blkcg_reset_stats, |
22084190 | 1052 | }, |
4baf6e33 | 1053 | { } /* terminate */ |
31e4c28d VG |
1054 | }; |
1055 | ||
9f13ef67 | 1056 | /** |
92fb9748 | 1057 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 1058 | * @css: css of interest |
9f13ef67 | 1059 | * |
eb95419b | 1060 | * This function is called when @css is about to go away and responsible |
4c699480 JQ |
1061 | * for offlining all blkgs pd and killing all wbs associated with @css. |
1062 | * blkgs pd offline should be done while holding both q and blkcg locks. | |
1063 | * As blkcg lock is nested inside q lock, this function performs reverse | |
1064 | * double lock dancing. | |
9f13ef67 TH |
1065 | * |
1066 | * This is the blkcg counterpart of ioc_release_fn(). | |
1067 | */ | |
eb95419b | 1068 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 1069 | { |
eb95419b | 1070 | struct blkcg *blkcg = css_to_blkcg(css); |
4c699480 | 1071 | struct blkcg_gq *blkg; |
b1c35769 | 1072 | |
9f13ef67 | 1073 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1074 | |
4c699480 | 1075 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
c875f4d0 | 1076 | struct request_queue *q = blkg->q; |
b1c35769 | 1077 | |
9f13ef67 | 1078 | if (spin_trylock(q->queue_lock)) { |
4c699480 | 1079 | blkg_pd_offline(blkg); |
9f13ef67 TH |
1080 | spin_unlock(q->queue_lock); |
1081 | } else { | |
1082 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 1083 | cpu_relax(); |
a5567932 | 1084 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 1085 | } |
9f13ef67 | 1086 | } |
b1c35769 | 1087 | |
9f13ef67 | 1088 | spin_unlock_irq(&blkcg->lock); |
52ebea74 TH |
1089 | |
1090 | wb_blkcg_offline(blkcg); | |
7ee9c562 TH |
1091 | } |
1092 | ||
4c699480 JQ |
1093 | /** |
1094 | * blkcg_destroy_all_blkgs - destroy all blkgs associated with a blkcg | |
1095 | * @blkcg: blkcg of interest | |
1096 | * | |
1097 | * This function is called when blkcg css is about to free and responsible for | |
1098 | * destroying all blkgs associated with @blkcg. | |
1099 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1100 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1101 | */ | |
1102 | static void blkcg_destroy_all_blkgs(struct blkcg *blkcg) | |
1103 | { | |
1104 | spin_lock_irq(&blkcg->lock); | |
1105 | while (!hlist_empty(&blkcg->blkg_list)) { | |
1106 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
1107 | struct blkcg_gq, | |
1108 | blkcg_node); | |
1109 | struct request_queue *q = blkg->q; | |
1110 | ||
1111 | if (spin_trylock(q->queue_lock)) { | |
1112 | blkg_destroy(blkg); | |
1113 | spin_unlock(q->queue_lock); | |
1114 | } else { | |
1115 | spin_unlock_irq(&blkcg->lock); | |
1116 | cpu_relax(); | |
1117 | spin_lock_irq(&blkcg->lock); | |
1118 | } | |
1119 | } | |
1120 | spin_unlock_irq(&blkcg->lock); | |
1121 | } | |
1122 | ||
eb95419b | 1123 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1124 | { |
eb95419b | 1125 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1126 | int i; |
7ee9c562 | 1127 | |
4c699480 JQ |
1128 | blkcg_destroy_all_blkgs(blkcg); |
1129 | ||
7876f930 | 1130 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1131 | |
7876f930 | 1132 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1133 | |
bc915e61 | 1134 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1135 | if (blkcg->cpd[i]) |
1136 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1137 | ||
1138 | mutex_unlock(&blkcg_pol_mutex); | |
1139 | ||
bc915e61 | 1140 | kfree(blkcg); |
31e4c28d VG |
1141 | } |
1142 | ||
eb95419b TH |
1143 | static struct cgroup_subsys_state * |
1144 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1145 | { |
3c798398 | 1146 | struct blkcg *blkcg; |
e48453c3 AA |
1147 | struct cgroup_subsys_state *ret; |
1148 | int i; | |
31e4c28d | 1149 | |
7876f930 TH |
1150 | mutex_lock(&blkcg_pol_mutex); |
1151 | ||
eb95419b | 1152 | if (!parent_css) { |
3c798398 | 1153 | blkcg = &blkcg_root; |
bc915e61 TH |
1154 | } else { |
1155 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1156 | if (!blkcg) { | |
1157 | ret = ERR_PTR(-ENOMEM); | |
4c18c9e9 | 1158 | goto unlock; |
bc915e61 | 1159 | } |
e48453c3 AA |
1160 | } |
1161 | ||
1162 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1163 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1164 | struct blkcg_policy_data *cpd; | |
1165 | ||
1166 | /* | |
1167 | * If the policy hasn't been attached yet, wait for it | |
1168 | * to be attached before doing anything else. Otherwise, | |
1169 | * check if the policy requires any specific per-cgroup | |
1170 | * data: if it does, allocate and initialize it. | |
1171 | */ | |
e4a9bde9 | 1172 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1173 | continue; |
1174 | ||
e4a9bde9 | 1175 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
1176 | if (!cpd) { |
1177 | ret = ERR_PTR(-ENOMEM); | |
1178 | goto free_pd_blkcg; | |
1179 | } | |
81437648 TH |
1180 | blkcg->cpd[i] = cpd; |
1181 | cpd->blkcg = blkcg; | |
e48453c3 | 1182 | cpd->plid = i; |
e4a9bde9 TH |
1183 | if (pol->cpd_init_fn) |
1184 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1185 | } |
31e4c28d | 1186 | |
31e4c28d | 1187 | spin_lock_init(&blkcg->lock); |
e00f4f4d | 1188 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1189 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1190 | #ifdef CONFIG_CGROUP_WRITEBACK |
1191 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1192 | #endif | |
7876f930 TH |
1193 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1194 | ||
1195 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1196 | return &blkcg->css; |
e48453c3 AA |
1197 | |
1198 | free_pd_blkcg: | |
1199 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1200 | if (blkcg->cpd[i]) |
1201 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
4c18c9e9 | 1202 | |
1203 | if (blkcg != &blkcg_root) | |
1204 | kfree(blkcg); | |
1205 | unlock: | |
7876f930 | 1206 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 1207 | return ret; |
31e4c28d VG |
1208 | } |
1209 | ||
5efd6113 TH |
1210 | /** |
1211 | * blkcg_init_queue - initialize blkcg part of request queue | |
1212 | * @q: request_queue to initialize | |
1213 | * | |
1214 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1215 | * part of new request_queue @q. | |
1216 | * | |
1217 | * RETURNS: | |
1218 | * 0 on success, -errno on failure. | |
1219 | */ | |
1220 | int blkcg_init_queue(struct request_queue *q) | |
1221 | { | |
d708f0d5 JA |
1222 | struct blkcg_gq *new_blkg, *blkg; |
1223 | bool preloaded; | |
ec13b1d6 TH |
1224 | int ret; |
1225 | ||
d708f0d5 JA |
1226 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
1227 | if (!new_blkg) | |
1228 | return -ENOMEM; | |
1229 | ||
1230 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1231 | ||
bea54883 | 1232 | /* Make sure the root blkg exists. */ |
ec13b1d6 TH |
1233 | rcu_read_lock(); |
1234 | spin_lock_irq(q->queue_lock); | |
d708f0d5 | 1235 | blkg = blkg_create(&blkcg_root, q, new_blkg); |
901932a3 JB |
1236 | if (IS_ERR(blkg)) |
1237 | goto err_unlock; | |
1238 | q->root_blkg = blkg; | |
1239 | q->root_rl.blkg = blkg; | |
ec13b1d6 TH |
1240 | spin_unlock_irq(q->queue_lock); |
1241 | rcu_read_unlock(); | |
1242 | ||
d708f0d5 JA |
1243 | if (preloaded) |
1244 | radix_tree_preload_end(); | |
1245 | ||
d7067512 JB |
1246 | ret = blk_iolatency_init(q); |
1247 | if (ret) { | |
1248 | spin_lock_irq(q->queue_lock); | |
1249 | blkg_destroy_all(q); | |
1250 | spin_unlock_irq(q->queue_lock); | |
1251 | return ret; | |
1252 | } | |
1253 | ||
ec13b1d6 TH |
1254 | ret = blk_throtl_init(q); |
1255 | if (ret) { | |
1256 | spin_lock_irq(q->queue_lock); | |
1257 | blkg_destroy_all(q); | |
1258 | spin_unlock_irq(q->queue_lock); | |
1259 | } | |
1260 | return ret; | |
901932a3 JB |
1261 | |
1262 | err_unlock: | |
1263 | spin_unlock_irq(q->queue_lock); | |
1264 | rcu_read_unlock(); | |
1265 | if (preloaded) | |
1266 | radix_tree_preload_end(); | |
1267 | return PTR_ERR(blkg); | |
5efd6113 TH |
1268 | } |
1269 | ||
1270 | /** | |
1271 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1272 | * @q: request_queue to drain | |
1273 | * | |
1274 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1275 | */ | |
1276 | void blkcg_drain_queue(struct request_queue *q) | |
1277 | { | |
1278 | lockdep_assert_held(q->queue_lock); | |
1279 | ||
0b462c89 TH |
1280 | /* |
1281 | * @q could be exiting and already have destroyed all blkgs as | |
1282 | * indicated by NULL root_blkg. If so, don't confuse policies. | |
1283 | */ | |
1284 | if (!q->root_blkg) | |
1285 | return; | |
1286 | ||
5efd6113 TH |
1287 | blk_throtl_drain(q); |
1288 | } | |
1289 | ||
1290 | /** | |
1291 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1292 | * @q: request_queue being released | |
1293 | * | |
1294 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1295 | */ | |
1296 | void blkcg_exit_queue(struct request_queue *q) | |
1297 | { | |
6d18b008 | 1298 | spin_lock_irq(q->queue_lock); |
3c96cb32 | 1299 | blkg_destroy_all(q); |
6d18b008 TH |
1300 | spin_unlock_irq(q->queue_lock); |
1301 | ||
5efd6113 TH |
1302 | blk_throtl_exit(q); |
1303 | } | |
1304 | ||
31e4c28d VG |
1305 | /* |
1306 | * We cannot support shared io contexts, as we have no mean to support | |
1307 | * two tasks with the same ioc in two different groups without major rework | |
1308 | * of the main cic data structures. For now we allow a task to change | |
1309 | * its cgroup only if it's the only owner of its ioc. | |
1310 | */ | |
1f7dd3e5 | 1311 | static int blkcg_can_attach(struct cgroup_taskset *tset) |
31e4c28d | 1312 | { |
bb9d97b6 | 1313 | struct task_struct *task; |
1f7dd3e5 | 1314 | struct cgroup_subsys_state *dst_css; |
31e4c28d VG |
1315 | struct io_context *ioc; |
1316 | int ret = 0; | |
1317 | ||
1318 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
1f7dd3e5 | 1319 | cgroup_taskset_for_each(task, dst_css, tset) { |
bb9d97b6 TH |
1320 | task_lock(task); |
1321 | ioc = task->io_context; | |
1322 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1323 | ret = -EINVAL; | |
1324 | task_unlock(task); | |
1325 | if (ret) | |
1326 | break; | |
1327 | } | |
31e4c28d VG |
1328 | return ret; |
1329 | } | |
1330 | ||
69d7fde5 TH |
1331 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1332 | { | |
1333 | int i; | |
1334 | ||
1335 | mutex_lock(&blkcg_pol_mutex); | |
1336 | ||
1337 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1338 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1339 | struct blkcg *blkcg; | |
1340 | ||
1341 | if (!pol || !pol->cpd_bind_fn) | |
1342 | continue; | |
1343 | ||
1344 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1345 | if (blkcg->cpd[pol->plid]) | |
1346 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1347 | } | |
1348 | mutex_unlock(&blkcg_pol_mutex); | |
1349 | } | |
1350 | ||
d09d8df3 JB |
1351 | static void blkcg_exit(struct task_struct *tsk) |
1352 | { | |
1353 | if (tsk->throttle_queue) | |
1354 | blk_put_queue(tsk->throttle_queue); | |
1355 | tsk->throttle_queue = NULL; | |
1356 | } | |
1357 | ||
c165b3e3 | 1358 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 TH |
1359 | .css_alloc = blkcg_css_alloc, |
1360 | .css_offline = blkcg_css_offline, | |
1361 | .css_free = blkcg_css_free, | |
3c798398 | 1362 | .can_attach = blkcg_can_attach, |
69d7fde5 | 1363 | .bind = blkcg_bind, |
2ee867dc | 1364 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1365 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1366 | .legacy_name = "blkio", |
d09d8df3 | 1367 | .exit = blkcg_exit, |
1ced953b TH |
1368 | #ifdef CONFIG_MEMCG |
1369 | /* | |
1370 | * This ensures that, if available, memcg is automatically enabled | |
1371 | * together on the default hierarchy so that the owner cgroup can | |
1372 | * be retrieved from writeback pages. | |
1373 | */ | |
1374 | .depends_on = 1 << memory_cgrp_id, | |
1375 | #endif | |
676f7c8f | 1376 | }; |
c165b3e3 | 1377 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1378 | |
a2b1693b TH |
1379 | /** |
1380 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1381 | * @q: request_queue of interest | |
1382 | * @pol: blkcg policy to activate | |
1383 | * | |
1384 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1385 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1386 | * | |
1387 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1388 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1389 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1390 | * always enough for dereferencing policy data. | |
1391 | * | |
1392 | * The caller is responsible for synchronizing [de]activations and policy | |
1393 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1394 | */ | |
1395 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1396 | const struct blkcg_policy *pol) |
a2b1693b | 1397 | { |
4c55f4f9 | 1398 | struct blkg_policy_data *pd_prealloc = NULL; |
ec13b1d6 | 1399 | struct blkcg_gq *blkg; |
4c55f4f9 | 1400 | int ret; |
a2b1693b TH |
1401 | |
1402 | if (blkcg_policy_enabled(q, pol)) | |
1403 | return 0; | |
1404 | ||
38dbb7dd | 1405 | if (q->mq_ops) |
bd166ef1 | 1406 | blk_mq_freeze_queue(q); |
38dbb7dd | 1407 | else |
bd166ef1 | 1408 | blk_queue_bypass_start(q); |
4c55f4f9 TH |
1409 | pd_prealloc: |
1410 | if (!pd_prealloc) { | |
001bea73 | 1411 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node); |
4c55f4f9 | 1412 | if (!pd_prealloc) { |
a2b1693b | 1413 | ret = -ENOMEM; |
4c55f4f9 | 1414 | goto out_bypass_end; |
a2b1693b | 1415 | } |
a2b1693b TH |
1416 | } |
1417 | ||
a2b1693b TH |
1418 | spin_lock_irq(q->queue_lock); |
1419 | ||
1420 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
4c55f4f9 TH |
1421 | struct blkg_policy_data *pd; |
1422 | ||
1423 | if (blkg->pd[pol->plid]) | |
1424 | continue; | |
a2b1693b | 1425 | |
e00f4f4d | 1426 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node); |
4c55f4f9 TH |
1427 | if (!pd) |
1428 | swap(pd, pd_prealloc); | |
1429 | if (!pd) { | |
1430 | spin_unlock_irq(q->queue_lock); | |
1431 | goto pd_prealloc; | |
1432 | } | |
a2b1693b TH |
1433 | |
1434 | blkg->pd[pol->plid] = pd; | |
1435 | pd->blkg = blkg; | |
b276a876 | 1436 | pd->plid = pol->plid; |
3e418710 | 1437 | if (pol->pd_init_fn) |
a9520cd6 | 1438 | pol->pd_init_fn(pd); |
a2b1693b TH |
1439 | } |
1440 | ||
1441 | __set_bit(pol->plid, q->blkcg_pols); | |
1442 | ret = 0; | |
4c55f4f9 | 1443 | |
a2b1693b | 1444 | spin_unlock_irq(q->queue_lock); |
4c55f4f9 | 1445 | out_bypass_end: |
38dbb7dd | 1446 | if (q->mq_ops) |
bd166ef1 | 1447 | blk_mq_unfreeze_queue(q); |
38dbb7dd | 1448 | else |
bd166ef1 | 1449 | blk_queue_bypass_end(q); |
001bea73 TH |
1450 | if (pd_prealloc) |
1451 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b TH |
1452 | return ret; |
1453 | } | |
1454 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1455 | ||
1456 | /** | |
1457 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1458 | * @q: request_queue of interest | |
1459 | * @pol: blkcg policy to deactivate | |
1460 | * | |
1461 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1462 | * blkcg_activate_policy(). | |
1463 | */ | |
1464 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1465 | const struct blkcg_policy *pol) |
a2b1693b | 1466 | { |
3c798398 | 1467 | struct blkcg_gq *blkg; |
a2b1693b TH |
1468 | |
1469 | if (!blkcg_policy_enabled(q, pol)) | |
1470 | return; | |
1471 | ||
38dbb7dd | 1472 | if (q->mq_ops) |
bd166ef1 | 1473 | blk_mq_freeze_queue(q); |
38dbb7dd | 1474 | else |
bd166ef1 JA |
1475 | blk_queue_bypass_start(q); |
1476 | ||
a2b1693b TH |
1477 | spin_lock_irq(q->queue_lock); |
1478 | ||
1479 | __clear_bit(pol->plid, q->blkcg_pols); | |
1480 | ||
1481 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
001bea73 | 1482 | if (blkg->pd[pol->plid]) { |
4c699480 JQ |
1483 | if (!blkg->pd[pol->plid]->offline && |
1484 | pol->pd_offline_fn) { | |
a9520cd6 | 1485 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
4c699480 JQ |
1486 | blkg->pd[pol->plid]->offline = true; |
1487 | } | |
001bea73 TH |
1488 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1489 | blkg->pd[pol->plid] = NULL; | |
1490 | } | |
a2b1693b TH |
1491 | } |
1492 | ||
1493 | spin_unlock_irq(q->queue_lock); | |
bd166ef1 | 1494 | |
38dbb7dd | 1495 | if (q->mq_ops) |
bd166ef1 | 1496 | blk_mq_unfreeze_queue(q); |
38dbb7dd | 1497 | else |
bd166ef1 | 1498 | blk_queue_bypass_end(q); |
a2b1693b TH |
1499 | } |
1500 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1501 | ||
8bd435b3 | 1502 | /** |
3c798398 TH |
1503 | * blkcg_policy_register - register a blkcg policy |
1504 | * @pol: blkcg policy to register | |
8bd435b3 | 1505 | * |
3c798398 TH |
1506 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1507 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1508 | */ |
d5bf0291 | 1509 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1510 | { |
06b285bd | 1511 | struct blkcg *blkcg; |
8bd435b3 | 1512 | int i, ret; |
e8989fae | 1513 | |
838f13bf | 1514 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1515 | mutex_lock(&blkcg_pol_mutex); |
1516 | ||
8bd435b3 TH |
1517 | /* find an empty slot */ |
1518 | ret = -ENOSPC; | |
1519 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1520 | if (!blkcg_policy[i]) |
8bd435b3 TH |
1521 | break; |
1522 | if (i >= BLKCG_MAX_POLS) | |
838f13bf | 1523 | goto err_unlock; |
035d10b2 | 1524 | |
e8401073 | 1525 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1526 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1527 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1528 | goto err_unlock; | |
1529 | ||
06b285bd | 1530 | /* register @pol */ |
3c798398 | 1531 | pol->plid = i; |
06b285bd TH |
1532 | blkcg_policy[pol->plid] = pol; |
1533 | ||
1534 | /* allocate and install cpd's */ | |
e4a9bde9 | 1535 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1536 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1537 | struct blkcg_policy_data *cpd; | |
1538 | ||
e4a9bde9 | 1539 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1540 | if (!cpd) |
06b285bd | 1541 | goto err_free_cpds; |
06b285bd | 1542 | |
81437648 TH |
1543 | blkcg->cpd[pol->plid] = cpd; |
1544 | cpd->blkcg = blkcg; | |
06b285bd | 1545 | cpd->plid = pol->plid; |
81437648 | 1546 | pol->cpd_init_fn(cpd); |
06b285bd TH |
1547 | } |
1548 | } | |
1549 | ||
838f13bf | 1550 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1551 | |
8bd435b3 | 1552 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1553 | if (pol->dfl_cftypes) |
1554 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1555 | pol->dfl_cftypes)); | |
880f50e2 | 1556 | if (pol->legacy_cftypes) |
c165b3e3 | 1557 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1558 | pol->legacy_cftypes)); |
838f13bf TH |
1559 | mutex_unlock(&blkcg_pol_register_mutex); |
1560 | return 0; | |
1561 | ||
06b285bd | 1562 | err_free_cpds: |
58a9edce | 1563 | if (pol->cpd_free_fn) { |
06b285bd | 1564 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1565 | if (blkcg->cpd[pol->plid]) { |
1566 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1567 | blkcg->cpd[pol->plid] = NULL; | |
1568 | } | |
06b285bd TH |
1569 | } |
1570 | } | |
1571 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1572 | err_unlock: |
bc0d6501 | 1573 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1574 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1575 | return ret; |
3e252066 | 1576 | } |
3c798398 | 1577 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1578 | |
8bd435b3 | 1579 | /** |
3c798398 TH |
1580 | * blkcg_policy_unregister - unregister a blkcg policy |
1581 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1582 | * |
3c798398 | 1583 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1584 | */ |
3c798398 | 1585 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1586 | { |
06b285bd TH |
1587 | struct blkcg *blkcg; |
1588 | ||
838f13bf | 1589 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1590 | |
3c798398 | 1591 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1592 | goto out_unlock; |
1593 | ||
1594 | /* kill the intf files first */ | |
2ee867dc TH |
1595 | if (pol->dfl_cftypes) |
1596 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1597 | if (pol->legacy_cftypes) |
1598 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1599 | |
06b285bd | 1600 | /* remove cpds and unregister */ |
838f13bf | 1601 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1602 | |
58a9edce | 1603 | if (pol->cpd_free_fn) { |
06b285bd | 1604 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1605 | if (blkcg->cpd[pol->plid]) { |
1606 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1607 | blkcg->cpd[pol->plid] = NULL; | |
1608 | } | |
06b285bd TH |
1609 | } |
1610 | } | |
3c798398 | 1611 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1612 | |
bc0d6501 | 1613 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1614 | out_unlock: |
1615 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1616 | } |
3c798398 | 1617 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1618 | |
d09d8df3 JB |
1619 | /* |
1620 | * Scale the accumulated delay based on how long it has been since we updated | |
1621 | * the delay. We only call this when we are adding delay, in case it's been a | |
1622 | * while since we added delay, and when we are checking to see if we need to | |
1623 | * delay a task, to account for any delays that may have occurred. | |
1624 | */ | |
1625 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1626 | { | |
1627 | u64 old = atomic64_read(&blkg->delay_start); | |
1628 | ||
1629 | /* | |
1630 | * We only want to scale down every second. The idea here is that we | |
1631 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1632 | * time window. We only want to throttle tasks for recent delay that | |
1633 | * has occurred, in 1 second time windows since that's the maximum | |
1634 | * things can be throttled. We save the current delay window in | |
1635 | * blkg->last_delay so we know what amount is still left to be charged | |
1636 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1637 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1638 | * are ok with whatever is happening now, and we can take away more of | |
1639 | * the accumulated delay as we've already throttled enough that | |
1640 | * everybody is happy with their IO latencies. | |
1641 | */ | |
1642 | if (time_before64(old + NSEC_PER_SEC, now) && | |
1643 | atomic64_cmpxchg(&blkg->delay_start, old, now) == old) { | |
1644 | u64 cur = atomic64_read(&blkg->delay_nsec); | |
1645 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1646 | int cur_use = atomic_read(&blkg->use_delay); | |
1647 | ||
1648 | /* | |
1649 | * We've been unthrottled, subtract a larger chunk of our | |
1650 | * accumulated delay. | |
1651 | */ | |
1652 | if (cur_use < blkg->last_use) | |
1653 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1654 | ||
1655 | /* | |
1656 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1657 | * should only ever be growing except here where we subtract out | |
1658 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1659 | * rather not end up with negative numbers. | |
1660 | */ | |
1661 | if (unlikely(cur < sub)) { | |
1662 | atomic64_set(&blkg->delay_nsec, 0); | |
1663 | blkg->last_delay = 0; | |
1664 | } else { | |
1665 | atomic64_sub(sub, &blkg->delay_nsec); | |
1666 | blkg->last_delay = cur - sub; | |
1667 | } | |
1668 | blkg->last_use = cur_use; | |
1669 | } | |
1670 | } | |
1671 | ||
1672 | /* | |
1673 | * This is called when we want to actually walk up the hierarchy and check to | |
1674 | * see if we need to throttle, and then actually throttle if there is some | |
1675 | * accumulated delay. This should only be called upon return to user space so | |
1676 | * we're not holding some lock that would induce a priority inversion. | |
1677 | */ | |
1678 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1679 | { | |
1680 | u64 now = ktime_to_ns(ktime_get()); | |
1681 | u64 exp; | |
1682 | u64 delay_nsec = 0; | |
1683 | int tok; | |
1684 | ||
1685 | while (blkg->parent) { | |
1686 | if (atomic_read(&blkg->use_delay)) { | |
1687 | blkcg_scale_delay(blkg, now); | |
1688 | delay_nsec = max_t(u64, delay_nsec, | |
1689 | atomic64_read(&blkg->delay_nsec)); | |
1690 | } | |
1691 | blkg = blkg->parent; | |
1692 | } | |
1693 | ||
1694 | if (!delay_nsec) | |
1695 | return; | |
1696 | ||
1697 | /* | |
1698 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1699 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1700 | * delay, and we want userspace to be able to do _something_ so cap the | |
1701 | * delays at 1 second. If there's 10's of seconds worth of delay then | |
1702 | * the tasks will be delayed for 1 second for every syscall. | |
1703 | */ | |
1704 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
1705 | ||
1706 | /* | |
1707 | * TODO: the use_memdelay flag is going to be for the upcoming psi stuff | |
1708 | * that hasn't landed upstream yet. Once that stuff is in place we need | |
1709 | * to do a psi_memstall_enter/leave if memdelay is set. | |
1710 | */ | |
1711 | ||
1712 | exp = ktime_add_ns(now, delay_nsec); | |
1713 | tok = io_schedule_prepare(); | |
1714 | do { | |
1715 | __set_current_state(TASK_KILLABLE); | |
1716 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1717 | break; | |
1718 | } while (!fatal_signal_pending(current)); | |
1719 | io_schedule_finish(tok); | |
1720 | } | |
1721 | ||
1722 | /** | |
1723 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1724 | * | |
1725 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1726 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
1727 | * check to see if current->throttle_queue is set and if not this doesn't do | |
1728 | * anything. This should only ever be called by the resume code, it's not meant | |
1729 | * to be called by people willy-nilly as it will actually do the work to | |
1730 | * throttle the task if it is setup for throttling. | |
1731 | */ | |
1732 | void blkcg_maybe_throttle_current(void) | |
1733 | { | |
1734 | struct request_queue *q = current->throttle_queue; | |
1735 | struct cgroup_subsys_state *css; | |
1736 | struct blkcg *blkcg; | |
1737 | struct blkcg_gq *blkg; | |
1738 | bool use_memdelay = current->use_memdelay; | |
1739 | ||
1740 | if (!q) | |
1741 | return; | |
1742 | ||
1743 | current->throttle_queue = NULL; | |
1744 | current->use_memdelay = false; | |
1745 | ||
1746 | rcu_read_lock(); | |
1747 | css = kthread_blkcg(); | |
1748 | if (css) | |
1749 | blkcg = css_to_blkcg(css); | |
1750 | else | |
1751 | blkcg = css_to_blkcg(task_css(current, io_cgrp_id)); | |
1752 | ||
1753 | if (!blkcg) | |
1754 | goto out; | |
1755 | blkg = blkg_lookup(blkcg, q); | |
1756 | if (!blkg) | |
1757 | goto out; | |
1758 | blkg = blkg_try_get(blkg); | |
1759 | if (!blkg) | |
1760 | goto out; | |
1761 | rcu_read_unlock(); | |
d09d8df3 JB |
1762 | |
1763 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1764 | blkg_put(blkg); | |
cc7ecc25 | 1765 | blk_put_queue(q); |
d09d8df3 JB |
1766 | return; |
1767 | out: | |
1768 | rcu_read_unlock(); | |
1769 | blk_put_queue(q); | |
1770 | } | |
1771 | EXPORT_SYMBOL_GPL(blkcg_maybe_throttle_current); | |
1772 | ||
1773 | /** | |
1774 | * blkcg_schedule_throttle - this task needs to check for throttling | |
1775 | * @q - the request queue IO was submitted on | |
1776 | * @use_memdelay - do we charge this to memory delay for PSI | |
1777 | * | |
1778 | * This is called by the IO controller when we know there's delay accumulated | |
1779 | * for the blkg for this task. We do not pass the blkg because there are places | |
1780 | * we call this that may not have that information, the swapping code for | |
1781 | * instance will only have a request_queue at that point. This set's the | |
1782 | * notify_resume for the task to check and see if it requires throttling before | |
1783 | * returning to user space. | |
1784 | * | |
1785 | * We will only schedule once per syscall. You can call this over and over | |
1786 | * again and it will only do the check once upon return to user space, and only | |
1787 | * throttle once. If the task needs to be throttled again it'll need to be | |
1788 | * re-set at the next time we see the task. | |
1789 | */ | |
1790 | void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) | |
1791 | { | |
1792 | if (unlikely(current->flags & PF_KTHREAD)) | |
1793 | return; | |
1794 | ||
1795 | if (!blk_get_queue(q)) | |
1796 | return; | |
1797 | ||
1798 | if (current->throttle_queue) | |
1799 | blk_put_queue(current->throttle_queue); | |
1800 | current->throttle_queue = q; | |
1801 | if (use_memdelay) | |
1802 | current->use_memdelay = use_memdelay; | |
1803 | set_notify_resume(current); | |
1804 | } | |
1805 | EXPORT_SYMBOL_GPL(blkcg_schedule_throttle); | |
1806 | ||
1807 | /** | |
1808 | * blkcg_add_delay - add delay to this blkg | |
1809 | * @now - the current time in nanoseconds | |
1810 | * @delta - how many nanoseconds of delay to add | |
1811 | * | |
1812 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1813 | * throttle tasks if an IO controller thinks we need more throttling. | |
1814 | */ | |
1815 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1816 | { | |
1817 | blkcg_scale_delay(blkg, now); | |
1818 | atomic64_add(delta, &blkg->delay_nsec); | |
1819 | } | |
1820 | EXPORT_SYMBOL_GPL(blkcg_add_delay); | |
1821 | ||
903d23f0 JB |
1822 | module_param(blkcg_debug_stats, bool, 0644); |
1823 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |