Commit | Line | Data |
---|---|---|
3dcf60bc | 1 | // SPDX-License-Identifier: GPL-2.0 |
31e4c28d VG |
2 | /* |
3 | * Common Block IO controller cgroup interface | |
4 | * | |
5 | * Based on ideas and code from CFQ, CFS and BFQ: | |
6 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> | |
7 | * | |
8 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> | |
9 | * Paolo Valente <paolo.valente@unimore.it> | |
10 | * | |
11 | * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> | |
12 | * Nauman Rafique <nauman@google.com> | |
e48453c3 AA |
13 | * |
14 | * For policy-specific per-blkcg data: | |
15 | * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it> | |
16 | * Arianna Avanzini <avanzini.arianna@gmail.com> | |
31e4c28d VG |
17 | */ |
18 | #include <linux/ioprio.h> | |
22084190 | 19 | #include <linux/kdev_t.h> |
9d6a986c | 20 | #include <linux/module.h> |
174cd4b1 | 21 | #include <linux/sched/signal.h> |
accee785 | 22 | #include <linux/err.h> |
9195291e | 23 | #include <linux/blkdev.h> |
52ebea74 | 24 | #include <linux/backing-dev.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
34d0f179 | 26 | #include <linux/genhd.h> |
72e06c25 | 27 | #include <linux/delay.h> |
9a9e8a26 | 28 | #include <linux/atomic.h> |
36aa9e5f | 29 | #include <linux/ctype.h> |
eea8f41c | 30 | #include <linux/blk-cgroup.h> |
d09d8df3 | 31 | #include <linux/tracehook.h> |
5efd6113 | 32 | #include "blk.h" |
3e252066 | 33 | |
84c124da DS |
34 | #define MAX_KEY_LEN 100 |
35 | ||
838f13bf TH |
36 | /* |
37 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
38 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
39 | * policy [un]register operations including cgroup file additions / | |
40 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
41 | * allows grabbing it from cgroup callbacks. | |
42 | */ | |
43 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 44 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 45 | |
e48453c3 | 46 | struct blkcg blkcg_root; |
3c798398 | 47 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 48 | |
496d5e75 TH |
49 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
50 | ||
3c798398 | 51 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 52 | |
7876f930 TH |
53 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
54 | ||
903d23f0 JB |
55 | static bool blkcg_debug_stats = false; |
56 | ||
a2b1693b | 57 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 58 | const struct blkcg_policy *pol) |
a2b1693b TH |
59 | { |
60 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
61 | } | |
62 | ||
0381411e TH |
63 | /** |
64 | * blkg_free - free a blkg | |
65 | * @blkg: blkg to free | |
66 | * | |
67 | * Free @blkg which may be partially allocated. | |
68 | */ | |
3c798398 | 69 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 70 | { |
e8989fae | 71 | int i; |
549d3aa8 TH |
72 | |
73 | if (!blkg) | |
74 | return; | |
75 | ||
db613670 | 76 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
77 | if (blkg->pd[i]) |
78 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
e8989fae | 79 | |
77ea7338 TH |
80 | blkg_rwstat_exit(&blkg->stat_ios); |
81 | blkg_rwstat_exit(&blkg->stat_bytes); | |
549d3aa8 | 82 | kfree(blkg); |
0381411e TH |
83 | } |
84 | ||
7fcf2b03 DZ |
85 | static void __blkg_release(struct rcu_head *rcu) |
86 | { | |
87 | struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head); | |
88 | ||
89 | percpu_ref_exit(&blkg->refcnt); | |
90 | ||
91 | /* release the blkcg and parent blkg refs this blkg has been holding */ | |
92 | css_put(&blkg->blkcg->css); | |
93 | if (blkg->parent) | |
94 | blkg_put(blkg->parent); | |
95 | ||
96 | wb_congested_put(blkg->wb_congested); | |
97 | ||
98 | blkg_free(blkg); | |
99 | } | |
100 | ||
101 | /* | |
102 | * A group is RCU protected, but having an rcu lock does not mean that one | |
103 | * can access all the fields of blkg and assume these are valid. For | |
104 | * example, don't try to follow throtl_data and request queue links. | |
105 | * | |
106 | * Having a reference to blkg under an rcu allows accesses to only values | |
107 | * local to groups like group stats and group rate limits. | |
108 | */ | |
109 | static void blkg_release(struct percpu_ref *ref) | |
110 | { | |
111 | struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt); | |
112 | ||
113 | call_rcu(&blkg->rcu_head, __blkg_release); | |
114 | } | |
115 | ||
0381411e TH |
116 | /** |
117 | * blkg_alloc - allocate a blkg | |
118 | * @blkcg: block cgroup the new blkg is associated with | |
119 | * @q: request_queue the new blkg is associated with | |
15974993 | 120 | * @gfp_mask: allocation mask to use |
0381411e | 121 | * |
e8989fae | 122 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 123 | */ |
15974993 TH |
124 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
125 | gfp_t gfp_mask) | |
0381411e | 126 | { |
3c798398 | 127 | struct blkcg_gq *blkg; |
e8989fae | 128 | int i; |
0381411e TH |
129 | |
130 | /* alloc and init base part */ | |
15974993 | 131 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
132 | if (!blkg) |
133 | return NULL; | |
134 | ||
77ea7338 TH |
135 | if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) || |
136 | blkg_rwstat_init(&blkg->stat_ios, gfp_mask)) | |
137 | goto err_free; | |
138 | ||
c875f4d0 | 139 | blkg->q = q; |
e8989fae | 140 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 141 | blkg->blkcg = blkcg; |
0381411e | 142 | |
8bd435b3 | 143 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 144 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 145 | struct blkg_policy_data *pd; |
0381411e | 146 | |
a2b1693b | 147 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
148 | continue; |
149 | ||
150 | /* alloc per-policy data and attach it to blkg */ | |
001bea73 | 151 | pd = pol->pd_alloc_fn(gfp_mask, q->node); |
a051661c TH |
152 | if (!pd) |
153 | goto err_free; | |
549d3aa8 | 154 | |
e8989fae TH |
155 | blkg->pd[i] = pd; |
156 | pd->blkg = blkg; | |
b276a876 | 157 | pd->plid = i; |
e8989fae TH |
158 | } |
159 | ||
0381411e | 160 | return blkg; |
a051661c TH |
161 | |
162 | err_free: | |
163 | blkg_free(blkg); | |
164 | return NULL; | |
0381411e TH |
165 | } |
166 | ||
24f29046 TH |
167 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
168 | struct request_queue *q, bool update_hint) | |
80fd9979 | 169 | { |
3c798398 | 170 | struct blkcg_gq *blkg; |
80fd9979 | 171 | |
a637120e | 172 | /* |
86cde6b6 TH |
173 | * Hint didn't match. Look up from the radix tree. Note that the |
174 | * hint can only be updated under queue_lock as otherwise @blkg | |
175 | * could have already been removed from blkg_tree. The caller is | |
176 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
177 | */ |
178 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
179 | if (blkg && blkg->q == q) { |
180 | if (update_hint) { | |
0d945c1f | 181 | lockdep_assert_held(&q->queue_lock); |
86cde6b6 TH |
182 | rcu_assign_pointer(blkcg->blkg_hint, blkg); |
183 | } | |
a637120e | 184 | return blkg; |
86cde6b6 | 185 | } |
a637120e | 186 | |
80fd9979 TH |
187 | return NULL; |
188 | } | |
ae118896 | 189 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); |
80fd9979 | 190 | |
15974993 | 191 | /* |
d708f0d5 JA |
192 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
193 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 194 | */ |
86cde6b6 | 195 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
d708f0d5 JA |
196 | struct request_queue *q, |
197 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 198 | { |
d708f0d5 | 199 | struct blkcg_gq *blkg; |
ce7acfea | 200 | struct bdi_writeback_congested *wb_congested; |
f427d909 | 201 | int i, ret; |
5624a4e4 | 202 | |
cd1604fa | 203 | WARN_ON_ONCE(!rcu_read_lock_held()); |
0d945c1f | 204 | lockdep_assert_held(&q->queue_lock); |
cd1604fa | 205 | |
0273ac34 DZ |
206 | /* request_queue is dying, do not create/recreate a blkg */ |
207 | if (blk_queue_dying(q)) { | |
208 | ret = -ENODEV; | |
209 | goto err_free_blkg; | |
210 | } | |
211 | ||
7ee9c562 | 212 | /* blkg holds a reference to blkcg */ |
ec903c0c | 213 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 214 | ret = -ENODEV; |
93e6d5d8 | 215 | goto err_free_blkg; |
15974993 | 216 | } |
cd1604fa | 217 | |
dc3b17cc | 218 | wb_congested = wb_congested_get_create(q->backing_dev_info, |
d708f0d5 JA |
219 | blkcg->css.id, |
220 | GFP_NOWAIT | __GFP_NOWARN); | |
221 | if (!wb_congested) { | |
ce7acfea | 222 | ret = -ENOMEM; |
d708f0d5 | 223 | goto err_put_css; |
ce7acfea TH |
224 | } |
225 | ||
d708f0d5 JA |
226 | /* allocate */ |
227 | if (!new_blkg) { | |
228 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN); | |
229 | if (unlikely(!new_blkg)) { | |
230 | ret = -ENOMEM; | |
231 | goto err_put_congested; | |
15974993 TH |
232 | } |
233 | } | |
d708f0d5 JA |
234 | blkg = new_blkg; |
235 | blkg->wb_congested = wb_congested; | |
cd1604fa | 236 | |
db613670 | 237 | /* link parent */ |
3c547865 TH |
238 | if (blkcg_parent(blkcg)) { |
239 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
240 | if (WARN_ON_ONCE(!blkg->parent)) { | |
20386ce0 | 241 | ret = -ENODEV; |
d708f0d5 | 242 | goto err_put_congested; |
3c547865 TH |
243 | } |
244 | blkg_get(blkg->parent); | |
245 | } | |
246 | ||
7fcf2b03 DZ |
247 | ret = percpu_ref_init(&blkg->refcnt, blkg_release, 0, |
248 | GFP_NOWAIT | __GFP_NOWARN); | |
249 | if (ret) | |
250 | goto err_cancel_ref; | |
251 | ||
db613670 TH |
252 | /* invoke per-policy init */ |
253 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
254 | struct blkcg_policy *pol = blkcg_policy[i]; | |
255 | ||
256 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 257 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
258 | } |
259 | ||
260 | /* insert */ | |
cd1604fa | 261 | spin_lock(&blkcg->lock); |
a637120e TH |
262 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
263 | if (likely(!ret)) { | |
264 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
265 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
266 | |
267 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
268 | struct blkcg_policy *pol = blkcg_policy[i]; | |
269 | ||
270 | if (blkg->pd[i] && pol->pd_online_fn) | |
a9520cd6 | 271 | pol->pd_online_fn(blkg->pd[i]); |
f427d909 | 272 | } |
a637120e | 273 | } |
f427d909 | 274 | blkg->online = true; |
cd1604fa | 275 | spin_unlock(&blkcg->lock); |
496fb780 | 276 | |
ec13b1d6 | 277 | if (!ret) |
a637120e | 278 | return blkg; |
15974993 | 279 | |
3c547865 TH |
280 | /* @blkg failed fully initialized, use the usual release path */ |
281 | blkg_put(blkg); | |
282 | return ERR_PTR(ret); | |
283 | ||
7fcf2b03 DZ |
284 | err_cancel_ref: |
285 | percpu_ref_exit(&blkg->refcnt); | |
d708f0d5 JA |
286 | err_put_congested: |
287 | wb_congested_put(wb_congested); | |
288 | err_put_css: | |
496fb780 | 289 | css_put(&blkcg->css); |
93e6d5d8 | 290 | err_free_blkg: |
d708f0d5 | 291 | blkg_free(new_blkg); |
93e6d5d8 | 292 | return ERR_PTR(ret); |
31e4c28d | 293 | } |
3c96cb32 | 294 | |
86cde6b6 | 295 | /** |
b978962a | 296 | * __blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 TH |
297 | * @blkcg: blkcg of interest |
298 | * @q: request_queue of interest | |
299 | * | |
300 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
301 | * create one. blkg creation is performed recursively from blkcg_root such |
302 | * that all non-root blkg's have access to the parent blkg. This function | |
303 | * should be called under RCU read lock and @q->queue_lock. | |
86cde6b6 | 304 | * |
beea9da0 DZ |
305 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
306 | * down from root. | |
86cde6b6 | 307 | */ |
b978962a DZ |
308 | struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg, |
309 | struct request_queue *q) | |
3c96cb32 | 310 | { |
86cde6b6 TH |
311 | struct blkcg_gq *blkg; |
312 | ||
313 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
0d945c1f | 314 | lockdep_assert_held(&q->queue_lock); |
86cde6b6 | 315 | |
86cde6b6 TH |
316 | blkg = __blkg_lookup(blkcg, q, true); |
317 | if (blkg) | |
318 | return blkg; | |
319 | ||
3c547865 TH |
320 | /* |
321 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
322 | * non-root blkgs have access to their parents. Returns the closest |
323 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
324 | */ |
325 | while (true) { | |
326 | struct blkcg *pos = blkcg; | |
327 | struct blkcg *parent = blkcg_parent(blkcg); | |
beea9da0 DZ |
328 | struct blkcg_gq *ret_blkg = q->root_blkg; |
329 | ||
330 | while (parent) { | |
331 | blkg = __blkg_lookup(parent, q, false); | |
332 | if (blkg) { | |
333 | /* remember closest blkg */ | |
334 | ret_blkg = blkg; | |
335 | break; | |
336 | } | |
3c547865 TH |
337 | pos = parent; |
338 | parent = blkcg_parent(parent); | |
339 | } | |
340 | ||
d708f0d5 | 341 | blkg = blkg_create(pos, q, NULL); |
beea9da0 DZ |
342 | if (IS_ERR(blkg)) |
343 | return ret_blkg; | |
344 | if (pos == blkcg) | |
3c547865 TH |
345 | return blkg; |
346 | } | |
3c96cb32 | 347 | } |
31e4c28d | 348 | |
b978962a DZ |
349 | /** |
350 | * blkg_lookup_create - find or create a blkg | |
351 | * @blkcg: target block cgroup | |
352 | * @q: target request_queue | |
353 | * | |
354 | * This looks up or creates the blkg representing the unique pair | |
355 | * of the blkcg and the request_queue. | |
356 | */ | |
357 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, | |
358 | struct request_queue *q) | |
359 | { | |
360 | struct blkcg_gq *blkg = blkg_lookup(blkcg, q); | |
361 | ||
362 | if (unlikely(!blkg)) { | |
3a762de5 ML |
363 | unsigned long flags; |
364 | ||
365 | spin_lock_irqsave(&q->queue_lock, flags); | |
b978962a | 366 | blkg = __blkg_lookup_create(blkcg, q); |
3a762de5 | 367 | spin_unlock_irqrestore(&q->queue_lock, flags); |
b978962a DZ |
368 | } |
369 | ||
370 | return blkg; | |
371 | } | |
372 | ||
3c798398 | 373 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 374 | { |
3c798398 | 375 | struct blkcg *blkcg = blkg->blkcg; |
77ea7338 | 376 | struct blkcg_gq *parent = blkg->parent; |
6b065462 | 377 | int i; |
03aa264a | 378 | |
0d945c1f | 379 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 380 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
381 | |
382 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 383 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 384 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 385 | |
6b065462 DZF |
386 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
387 | struct blkcg_policy *pol = blkcg_policy[i]; | |
388 | ||
389 | if (blkg->pd[i] && pol->pd_offline_fn) | |
390 | pol->pd_offline_fn(blkg->pd[i]); | |
391 | } | |
392 | ||
77ea7338 TH |
393 | if (parent) { |
394 | blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes); | |
395 | blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios); | |
396 | } | |
397 | ||
f427d909 TH |
398 | blkg->online = false; |
399 | ||
a637120e | 400 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 401 | list_del_init(&blkg->q_node); |
9f13ef67 | 402 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 403 | |
a637120e TH |
404 | /* |
405 | * Both setting lookup hint to and clearing it from @blkg are done | |
406 | * under queue_lock. If it's not pointing to @blkg now, it never | |
407 | * will. Hint assignment itself can race safely. | |
408 | */ | |
ec6c676a | 409 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
410 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
411 | ||
03aa264a TH |
412 | /* |
413 | * Put the reference taken at the time of creation so that when all | |
414 | * queues are gone, group can be destroyed. | |
415 | */ | |
7fcf2b03 | 416 | percpu_ref_kill(&blkg->refcnt); |
03aa264a TH |
417 | } |
418 | ||
9f13ef67 TH |
419 | /** |
420 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
421 | * @q: request_queue of interest | |
9f13ef67 | 422 | * |
3c96cb32 | 423 | * Destroy all blkgs associated with @q. |
9f13ef67 | 424 | */ |
3c96cb32 | 425 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 426 | { |
3c798398 | 427 | struct blkcg_gq *blkg, *n; |
72e06c25 | 428 | |
0d945c1f | 429 | spin_lock_irq(&q->queue_lock); |
9f13ef67 | 430 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 431 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 432 | |
9f13ef67 TH |
433 | spin_lock(&blkcg->lock); |
434 | blkg_destroy(blkg); | |
435 | spin_unlock(&blkcg->lock); | |
72e06c25 | 436 | } |
6fe810bd TH |
437 | |
438 | q->root_blkg = NULL; | |
0d945c1f | 439 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
440 | } |
441 | ||
182446d0 TH |
442 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
443 | struct cftype *cftype, u64 val) | |
303a3acb | 444 | { |
182446d0 | 445 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 446 | struct blkcg_gq *blkg; |
bc0d6501 | 447 | int i; |
303a3acb | 448 | |
838f13bf | 449 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 450 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
451 | |
452 | /* | |
453 | * Note that stat reset is racy - it doesn't synchronize against | |
454 | * stat updates. This is a debug feature which shouldn't exist | |
455 | * anyway. If you get hit by a race, retry. | |
456 | */ | |
b67bfe0d | 457 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
77ea7338 TH |
458 | blkg_rwstat_reset(&blkg->stat_bytes); |
459 | blkg_rwstat_reset(&blkg->stat_ios); | |
460 | ||
8bd435b3 | 461 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 462 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 463 | |
a9520cd6 TH |
464 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
465 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 466 | } |
303a3acb | 467 | } |
f0bdc8cd | 468 | |
303a3acb | 469 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 470 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
471 | return 0; |
472 | } | |
473 | ||
dd165eb3 | 474 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 475 | { |
d3d32e69 | 476 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
dc3b17cc JK |
477 | if (blkg->q->backing_dev_info->dev) |
478 | return dev_name(blkg->q->backing_dev_info->dev); | |
d3d32e69 | 479 | return NULL; |
303a3acb DS |
480 | } |
481 | ||
d3d32e69 TH |
482 | /** |
483 | * blkcg_print_blkgs - helper for printing per-blkg data | |
484 | * @sf: seq_file to print to | |
485 | * @blkcg: blkcg of interest | |
486 | * @prfill: fill function to print out a blkg | |
487 | * @pol: policy in question | |
488 | * @data: data to be passed to @prfill | |
489 | * @show_total: to print out sum of prfill return values or not | |
490 | * | |
491 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
492 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
493 | * policy data and @data and the matching queue lock held. If @show_total |
494 | * is %true, the sum of the return values from @prfill is printed with | |
495 | * "Total" label at the end. | |
d3d32e69 TH |
496 | * |
497 | * This is to be used to construct print functions for | |
498 | * cftype->read_seq_string method. | |
499 | */ | |
3c798398 | 500 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
501 | u64 (*prfill)(struct seq_file *, |
502 | struct blkg_policy_data *, int), | |
3c798398 | 503 | const struct blkcg_policy *pol, int data, |
ec399347 | 504 | bool show_total) |
5624a4e4 | 505 | { |
3c798398 | 506 | struct blkcg_gq *blkg; |
d3d32e69 | 507 | u64 total = 0; |
5624a4e4 | 508 | |
810ecfa7 | 509 | rcu_read_lock(); |
ee89f812 | 510 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 511 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 512 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 513 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 514 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
515 | } |
516 | rcu_read_unlock(); | |
d3d32e69 TH |
517 | |
518 | if (show_total) | |
519 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
520 | } | |
829fdb50 | 521 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
522 | |
523 | /** | |
524 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
525 | * @sf: seq_file to print to | |
f95a04af | 526 | * @pd: policy private data of interest |
d3d32e69 TH |
527 | * @v: value to print |
528 | * | |
f95a04af | 529 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 530 | */ |
f95a04af | 531 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 532 | { |
f95a04af | 533 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
534 | |
535 | if (!dname) | |
536 | return 0; | |
537 | ||
538 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
539 | return v; | |
540 | } | |
829fdb50 | 541 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
542 | |
543 | /** | |
544 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
545 | * @sf: seq_file to print to | |
f95a04af | 546 | * @pd: policy private data of interest |
d3d32e69 TH |
547 | * @rwstat: rwstat to print |
548 | * | |
f95a04af | 549 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 550 | */ |
f95a04af | 551 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 552 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
553 | { |
554 | static const char *rwstr[] = { | |
555 | [BLKG_RWSTAT_READ] = "Read", | |
556 | [BLKG_RWSTAT_WRITE] = "Write", | |
557 | [BLKG_RWSTAT_SYNC] = "Sync", | |
558 | [BLKG_RWSTAT_ASYNC] = "Async", | |
636620b6 | 559 | [BLKG_RWSTAT_DISCARD] = "Discard", |
d3d32e69 | 560 | }; |
f95a04af | 561 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
562 | u64 v; |
563 | int i; | |
564 | ||
565 | if (!dname) | |
566 | return 0; | |
567 | ||
568 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
569 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
24bdb8ef | 570 | (unsigned long long)atomic64_read(&rwstat->aux_cnt[i])); |
d3d32e69 | 571 | |
24bdb8ef | 572 | v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) + |
636620b6 TH |
573 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]) + |
574 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_DISCARD]); | |
d3d32e69 TH |
575 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); |
576 | return v; | |
577 | } | |
b50da39f | 578 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 579 | |
5bc4afb1 TH |
580 | /** |
581 | * blkg_prfill_stat - prfill callback for blkg_stat | |
582 | * @sf: seq_file to print to | |
f95a04af TH |
583 | * @pd: policy private data of interest |
584 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
585 | * |
586 | * prfill callback for printing a blkg_stat. | |
587 | */ | |
f95a04af | 588 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 589 | { |
f95a04af | 590 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 591 | } |
5bc4afb1 | 592 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 593 | |
5bc4afb1 TH |
594 | /** |
595 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
596 | * @sf: seq_file to print to | |
f95a04af TH |
597 | * @pd: policy private data of interest |
598 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
599 | * |
600 | * prfill callback for printing a blkg_rwstat. | |
601 | */ | |
f95a04af TH |
602 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
603 | int off) | |
d3d32e69 | 604 | { |
f95a04af | 605 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 606 | |
f95a04af | 607 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 608 | } |
5bc4afb1 | 609 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 610 | |
77ea7338 TH |
611 | static u64 blkg_prfill_rwstat_field(struct seq_file *sf, |
612 | struct blkg_policy_data *pd, int off) | |
613 | { | |
614 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off); | |
615 | ||
616 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
617 | } | |
618 | ||
619 | /** | |
620 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes | |
621 | * @sf: seq_file to print to | |
622 | * @v: unused | |
623 | * | |
624 | * To be used as cftype->seq_show to print blkg->stat_bytes. | |
625 | * cftype->private must be set to the blkcg_policy. | |
626 | */ | |
627 | int blkg_print_stat_bytes(struct seq_file *sf, void *v) | |
628 | { | |
629 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
630 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
631 | offsetof(struct blkcg_gq, stat_bytes), true); | |
632 | return 0; | |
633 | } | |
634 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes); | |
635 | ||
636 | /** | |
637 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios | |
638 | * @sf: seq_file to print to | |
639 | * @v: unused | |
640 | * | |
641 | * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private | |
642 | * must be set to the blkcg_policy. | |
643 | */ | |
644 | int blkg_print_stat_ios(struct seq_file *sf, void *v) | |
645 | { | |
646 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
647 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
648 | offsetof(struct blkcg_gq, stat_ios), true); | |
649 | return 0; | |
650 | } | |
651 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios); | |
652 | ||
653 | static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf, | |
654 | struct blkg_policy_data *pd, | |
655 | int off) | |
656 | { | |
657 | struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg, | |
658 | NULL, off); | |
659 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
660 | } | |
661 | ||
662 | /** | |
663 | * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes | |
664 | * @sf: seq_file to print to | |
665 | * @v: unused | |
666 | */ | |
667 | int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v) | |
668 | { | |
669 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
670 | blkg_prfill_rwstat_field_recursive, | |
671 | (void *)seq_cft(sf)->private, | |
672 | offsetof(struct blkcg_gq, stat_bytes), true); | |
673 | return 0; | |
674 | } | |
675 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive); | |
676 | ||
677 | /** | |
678 | * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios | |
679 | * @sf: seq_file to print to | |
680 | * @v: unused | |
681 | */ | |
682 | int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v) | |
683 | { | |
684 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
685 | blkg_prfill_rwstat_field_recursive, | |
686 | (void *)seq_cft(sf)->private, | |
687 | offsetof(struct blkcg_gq, stat_ios), true); | |
688 | return 0; | |
689 | } | |
690 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive); | |
691 | ||
16b3de66 TH |
692 | /** |
693 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
f12c74ca TH |
694 | * @blkg: blkg of interest |
695 | * @pol: blkcg_policy which contains the blkg_stat | |
696 | * @off: offset to the blkg_stat in blkg_policy_data or @blkg | |
16b3de66 | 697 | * |
f12c74ca TH |
698 | * Collect the blkg_stat specified by @blkg, @pol and @off and all its |
699 | * online descendants and their aux counts. The caller must be holding the | |
700 | * queue lock for online tests. | |
701 | * | |
702 | * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is | |
703 | * at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 704 | */ |
f12c74ca TH |
705 | u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg, |
706 | struct blkcg_policy *pol, int off) | |
16b3de66 | 707 | { |
16b3de66 | 708 | struct blkcg_gq *pos_blkg; |
492eb21b | 709 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 710 | u64 sum = 0; |
16b3de66 | 711 | |
0d945c1f | 712 | lockdep_assert_held(&blkg->q->queue_lock); |
16b3de66 | 713 | |
16b3de66 | 714 | rcu_read_lock(); |
f12c74ca TH |
715 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
716 | struct blkg_stat *stat; | |
717 | ||
718 | if (!pos_blkg->online) | |
719 | continue; | |
16b3de66 | 720 | |
f12c74ca TH |
721 | if (pol) |
722 | stat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
723 | else | |
724 | stat = (void *)blkg + off; | |
725 | ||
726 | sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
16b3de66 TH |
727 | } |
728 | rcu_read_unlock(); | |
729 | ||
730 | return sum; | |
731 | } | |
732 | EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum); | |
733 | ||
734 | /** | |
735 | * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat | |
f12c74ca TH |
736 | * @blkg: blkg of interest |
737 | * @pol: blkcg_policy which contains the blkg_rwstat | |
738 | * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg | |
16b3de66 | 739 | * |
f12c74ca TH |
740 | * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its |
741 | * online descendants and their aux counts. The caller must be holding the | |
742 | * queue lock for online tests. | |
743 | * | |
744 | * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it | |
745 | * is at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 746 | */ |
f12c74ca TH |
747 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, |
748 | struct blkcg_policy *pol, int off) | |
16b3de66 | 749 | { |
16b3de66 | 750 | struct blkcg_gq *pos_blkg; |
492eb21b | 751 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 752 | struct blkg_rwstat sum = { }; |
16b3de66 TH |
753 | int i; |
754 | ||
0d945c1f | 755 | lockdep_assert_held(&blkg->q->queue_lock); |
16b3de66 | 756 | |
16b3de66 | 757 | rcu_read_lock(); |
f12c74ca | 758 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
3a7faead | 759 | struct blkg_rwstat *rwstat; |
16b3de66 TH |
760 | |
761 | if (!pos_blkg->online) | |
762 | continue; | |
763 | ||
f12c74ca TH |
764 | if (pol) |
765 | rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
766 | else | |
767 | rwstat = (void *)pos_blkg + off; | |
768 | ||
16b3de66 | 769 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
3a7faead TH |
770 | atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) + |
771 | percpu_counter_sum_positive(&rwstat->cpu_cnt[i]), | |
772 | &sum.aux_cnt[i]); | |
16b3de66 TH |
773 | } |
774 | rcu_read_unlock(); | |
775 | ||
776 | return sum; | |
777 | } | |
778 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
779 | ||
457e490f TE |
780 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ |
781 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, | |
782 | const struct blkcg_policy *pol, | |
783 | struct request_queue *q) | |
784 | { | |
785 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
0d945c1f | 786 | lockdep_assert_held(&q->queue_lock); |
457e490f TE |
787 | |
788 | if (!blkcg_policy_enabled(q, pol)) | |
789 | return ERR_PTR(-EOPNOTSUPP); | |
457e490f TE |
790 | return __blkg_lookup(blkcg, q, true /* update_hint */); |
791 | } | |
792 | ||
3a8b31d3 TH |
793 | /** |
794 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
795 | * @blkcg: target block cgroup | |
da8b0662 | 796 | * @pol: target policy |
3a8b31d3 TH |
797 | * @input: input string |
798 | * @ctx: blkg_conf_ctx to be filled | |
799 | * | |
800 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
801 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
802 | * part of @input following MAJ:MIN. This function returns with RCU read | |
803 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 804 | */ |
3c798398 | 805 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 806 | char *input, struct blkg_conf_ctx *ctx) |
0d945c1f | 807 | __acquires(rcu) __acquires(&disk->queue->queue_lock) |
34d0f179 | 808 | { |
3a8b31d3 | 809 | struct gendisk *disk; |
457e490f | 810 | struct request_queue *q; |
3c798398 | 811 | struct blkcg_gq *blkg; |
726fa694 | 812 | unsigned int major, minor; |
36aa9e5f TH |
813 | int key_len, part, ret; |
814 | char *body; | |
34d0f179 | 815 | |
36aa9e5f | 816 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) |
726fa694 | 817 | return -EINVAL; |
3a8b31d3 | 818 | |
36aa9e5f TH |
819 | body = input + key_len; |
820 | if (!isspace(*body)) | |
821 | return -EINVAL; | |
822 | body = skip_spaces(body); | |
823 | ||
726fa694 | 824 | disk = get_gendisk(MKDEV(major, minor), &part); |
5f6c2d2b | 825 | if (!disk) |
20386ce0 | 826 | return -ENODEV; |
5f6c2d2b | 827 | if (part) { |
457e490f TE |
828 | ret = -ENODEV; |
829 | goto fail; | |
5f6c2d2b | 830 | } |
e56da7e2 | 831 | |
457e490f | 832 | q = disk->queue; |
da8b0662 | 833 | |
457e490f | 834 | rcu_read_lock(); |
0d945c1f | 835 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 836 | |
457e490f | 837 | blkg = blkg_lookup_check(blkcg, pol, q); |
4bfd482e TH |
838 | if (IS_ERR(blkg)) { |
839 | ret = PTR_ERR(blkg); | |
457e490f TE |
840 | goto fail_unlock; |
841 | } | |
842 | ||
843 | if (blkg) | |
844 | goto success; | |
845 | ||
846 | /* | |
847 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
848 | * non-root blkgs have access to their parents. | |
849 | */ | |
850 | while (true) { | |
851 | struct blkcg *pos = blkcg; | |
852 | struct blkcg *parent; | |
853 | struct blkcg_gq *new_blkg; | |
854 | ||
855 | parent = blkcg_parent(blkcg); | |
856 | while (parent && !__blkg_lookup(parent, q, false)) { | |
857 | pos = parent; | |
858 | parent = blkcg_parent(parent); | |
859 | } | |
860 | ||
861 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 862 | spin_unlock_irq(&q->queue_lock); |
3a8b31d3 | 863 | rcu_read_unlock(); |
457e490f TE |
864 | |
865 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); | |
866 | if (unlikely(!new_blkg)) { | |
867 | ret = -ENOMEM; | |
868 | goto fail; | |
7702e8f4 | 869 | } |
3a8b31d3 | 870 | |
457e490f | 871 | rcu_read_lock(); |
0d945c1f | 872 | spin_lock_irq(&q->queue_lock); |
457e490f TE |
873 | |
874 | blkg = blkg_lookup_check(pos, pol, q); | |
875 | if (IS_ERR(blkg)) { | |
876 | ret = PTR_ERR(blkg); | |
877 | goto fail_unlock; | |
878 | } | |
879 | ||
880 | if (blkg) { | |
881 | blkg_free(new_blkg); | |
882 | } else { | |
883 | blkg = blkg_create(pos, q, new_blkg); | |
884 | if (unlikely(IS_ERR(blkg))) { | |
885 | ret = PTR_ERR(blkg); | |
886 | goto fail_unlock; | |
887 | } | |
888 | } | |
889 | ||
890 | if (pos == blkcg) | |
891 | goto success; | |
892 | } | |
893 | success: | |
3a8b31d3 TH |
894 | ctx->disk = disk; |
895 | ctx->blkg = blkg; | |
36aa9e5f | 896 | ctx->body = body; |
726fa694 | 897 | return 0; |
457e490f TE |
898 | |
899 | fail_unlock: | |
0d945c1f | 900 | spin_unlock_irq(&q->queue_lock); |
457e490f TE |
901 | rcu_read_unlock(); |
902 | fail: | |
9df6c299 | 903 | put_disk_and_module(disk); |
457e490f TE |
904 | /* |
905 | * If queue was bypassing, we should retry. Do so after a | |
906 | * short msleep(). It isn't strictly necessary but queue | |
907 | * can be bypassing for some time and it's always nice to | |
908 | * avoid busy looping. | |
909 | */ | |
910 | if (ret == -EBUSY) { | |
911 | msleep(10); | |
912 | ret = restart_syscall(); | |
913 | } | |
914 | return ret; | |
34d0f179 GJ |
915 | } |
916 | ||
3a8b31d3 TH |
917 | /** |
918 | * blkg_conf_finish - finish up per-blkg config update | |
919 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
920 | * | |
921 | * Finish up after per-blkg config update. This function must be paired | |
922 | * with blkg_conf_prep(). | |
923 | */ | |
829fdb50 | 924 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
0d945c1f | 925 | __releases(&ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 926 | { |
0d945c1f | 927 | spin_unlock_irq(&ctx->disk->queue->queue_lock); |
3a8b31d3 | 928 | rcu_read_unlock(); |
9df6c299 | 929 | put_disk_and_module(ctx->disk); |
34d0f179 GJ |
930 | } |
931 | ||
2ee867dc TH |
932 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
933 | { | |
934 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
935 | struct blkcg_gq *blkg; | |
936 | ||
937 | rcu_read_lock(); | |
938 | ||
939 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
940 | const char *dname; | |
903d23f0 | 941 | char *buf; |
2ee867dc | 942 | struct blkg_rwstat rwstat; |
636620b6 | 943 | u64 rbytes, wbytes, rios, wios, dbytes, dios; |
903d23f0 JB |
944 | size_t size = seq_get_buf(sf, &buf), off = 0; |
945 | int i; | |
946 | bool has_stats = false; | |
2ee867dc TH |
947 | |
948 | dname = blkg_dev_name(blkg); | |
949 | if (!dname) | |
950 | continue; | |
951 | ||
903d23f0 JB |
952 | /* |
953 | * Hooray string manipulation, count is the size written NOT | |
954 | * INCLUDING THE \0, so size is now count+1 less than what we | |
955 | * had before, but we want to start writing the next bit from | |
956 | * the \0 so we only add count to buf. | |
957 | */ | |
958 | off += scnprintf(buf+off, size-off, "%s ", dname); | |
959 | ||
0d945c1f | 960 | spin_lock_irq(&blkg->q->queue_lock); |
2ee867dc TH |
961 | |
962 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
963 | offsetof(struct blkcg_gq, stat_bytes)); | |
964 | rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
965 | wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
636620b6 | 966 | dbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_DISCARD]); |
2ee867dc TH |
967 | |
968 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
969 | offsetof(struct blkcg_gq, stat_ios)); | |
970 | rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
971 | wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
636620b6 | 972 | dios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_DISCARD]); |
2ee867dc | 973 | |
0d945c1f | 974 | spin_unlock_irq(&blkg->q->queue_lock); |
2ee867dc | 975 | |
903d23f0 JB |
976 | if (rbytes || wbytes || rios || wios) { |
977 | has_stats = true; | |
978 | off += scnprintf(buf+off, size-off, | |
636620b6 TH |
979 | "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
980 | rbytes, wbytes, rios, wios, | |
981 | dbytes, dios); | |
903d23f0 JB |
982 | } |
983 | ||
984 | if (!blkcg_debug_stats) | |
985 | goto next; | |
986 | ||
d09d8df3 JB |
987 | if (atomic_read(&blkg->use_delay)) { |
988 | has_stats = true; | |
989 | off += scnprintf(buf+off, size-off, | |
990 | " use_delay=%d delay_nsec=%llu", | |
991 | atomic_read(&blkg->use_delay), | |
992 | (unsigned long long)atomic64_read(&blkg->delay_nsec)); | |
993 | } | |
994 | ||
903d23f0 JB |
995 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
996 | struct blkcg_policy *pol = blkcg_policy[i]; | |
997 | size_t written; | |
998 | ||
999 | if (!blkg->pd[i] || !pol->pd_stat_fn) | |
1000 | continue; | |
1001 | ||
1002 | written = pol->pd_stat_fn(blkg->pd[i], buf+off, size-off); | |
1003 | if (written) | |
1004 | has_stats = true; | |
1005 | off += written; | |
1006 | } | |
1007 | next: | |
1008 | if (has_stats) { | |
1009 | off += scnprintf(buf+off, size-off, "\n"); | |
1010 | seq_commit(sf, off); | |
1011 | } | |
2ee867dc TH |
1012 | } |
1013 | ||
1014 | rcu_read_unlock(); | |
1015 | return 0; | |
1016 | } | |
1017 | ||
e1f3b941 | 1018 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
1019 | { |
1020 | .name = "stat", | |
ca0752c5 | 1021 | .flags = CFTYPE_NOT_ON_ROOT, |
2ee867dc TH |
1022 | .seq_show = blkcg_print_stat, |
1023 | }, | |
1024 | { } /* terminate */ | |
1025 | }; | |
1026 | ||
e1f3b941 | 1027 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
1028 | { |
1029 | .name = "reset_stats", | |
3c798398 | 1030 | .write_u64 = blkcg_reset_stats, |
22084190 | 1031 | }, |
4baf6e33 | 1032 | { } /* terminate */ |
31e4c28d VG |
1033 | }; |
1034 | ||
59b57717 DZF |
1035 | /* |
1036 | * blkcg destruction is a three-stage process. | |
1037 | * | |
1038 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
1039 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
1040 | * to the completion of writeback associated with the blkcg. This lets us | |
1041 | * avoid punting potentially large amounts of outstanding writeback to root | |
1042 | * while maintaining any ongoing policies. The next stage is triggered when | |
1043 | * the nr_cgwbs count goes to zero. | |
1044 | * | |
1045 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
1046 | * and handles the destruction of blkgs. Here the css reference held by | |
1047 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
1048 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
1049 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
1050 | * punted to the root_blkg. | |
1051 | * | |
1052 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
1053 | * This finally frees the blkcg. | |
1054 | */ | |
1055 | ||
9f13ef67 | 1056 | /** |
92fb9748 | 1057 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 1058 | * @css: css of interest |
9f13ef67 | 1059 | * |
59b57717 DZF |
1060 | * This function is called when @css is about to go away. Here the cgwbs are |
1061 | * offlined first and only once writeback associated with the blkcg has | |
1062 | * finished do we start step 2 (see above). | |
9f13ef67 | 1063 | */ |
eb95419b | 1064 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 1065 | { |
eb95419b | 1066 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 1067 | |
59b57717 DZF |
1068 | /* this prevents anyone from attaching or migrating to this blkcg */ |
1069 | wb_blkcg_offline(blkcg); | |
1070 | ||
1071 | /* put the base cgwb reference allowing step 2 to be triggered */ | |
1072 | blkcg_cgwb_put(blkcg); | |
1073 | } | |
1074 | ||
1075 | /** | |
1076 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1077 | * @blkcg: blkcg of interest | |
1078 | * | |
1079 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1080 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1081 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1082 | * blkcg_css_free to eventually be called. | |
1083 | * | |
1084 | * This is the blkcg counterpart of ioc_release_fn(). | |
1085 | */ | |
1086 | void blkcg_destroy_blkgs(struct blkcg *blkcg) | |
1087 | { | |
9f13ef67 | 1088 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1089 | |
4c699480 JQ |
1090 | while (!hlist_empty(&blkcg->blkg_list)) { |
1091 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1092 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1093 | struct request_queue *q = blkg->q; |
1094 | ||
0d945c1f | 1095 | if (spin_trylock(&q->queue_lock)) { |
4c699480 | 1096 | blkg_destroy(blkg); |
0d945c1f | 1097 | spin_unlock(&q->queue_lock); |
4c699480 JQ |
1098 | } else { |
1099 | spin_unlock_irq(&blkcg->lock); | |
1100 | cpu_relax(); | |
1101 | spin_lock_irq(&blkcg->lock); | |
1102 | } | |
1103 | } | |
6b065462 | 1104 | |
4c699480 JQ |
1105 | spin_unlock_irq(&blkcg->lock); |
1106 | } | |
1107 | ||
eb95419b | 1108 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1109 | { |
eb95419b | 1110 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1111 | int i; |
7ee9c562 | 1112 | |
7876f930 | 1113 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1114 | |
7876f930 | 1115 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1116 | |
bc915e61 | 1117 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1118 | if (blkcg->cpd[i]) |
1119 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1120 | ||
1121 | mutex_unlock(&blkcg_pol_mutex); | |
1122 | ||
bc915e61 | 1123 | kfree(blkcg); |
31e4c28d VG |
1124 | } |
1125 | ||
eb95419b TH |
1126 | static struct cgroup_subsys_state * |
1127 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1128 | { |
3c798398 | 1129 | struct blkcg *blkcg; |
e48453c3 AA |
1130 | struct cgroup_subsys_state *ret; |
1131 | int i; | |
31e4c28d | 1132 | |
7876f930 TH |
1133 | mutex_lock(&blkcg_pol_mutex); |
1134 | ||
eb95419b | 1135 | if (!parent_css) { |
3c798398 | 1136 | blkcg = &blkcg_root; |
bc915e61 TH |
1137 | } else { |
1138 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1139 | if (!blkcg) { | |
1140 | ret = ERR_PTR(-ENOMEM); | |
4c18c9e9 | 1141 | goto unlock; |
bc915e61 | 1142 | } |
e48453c3 AA |
1143 | } |
1144 | ||
1145 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1146 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1147 | struct blkcg_policy_data *cpd; | |
1148 | ||
1149 | /* | |
1150 | * If the policy hasn't been attached yet, wait for it | |
1151 | * to be attached before doing anything else. Otherwise, | |
1152 | * check if the policy requires any specific per-cgroup | |
1153 | * data: if it does, allocate and initialize it. | |
1154 | */ | |
e4a9bde9 | 1155 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1156 | continue; |
1157 | ||
e4a9bde9 | 1158 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
1159 | if (!cpd) { |
1160 | ret = ERR_PTR(-ENOMEM); | |
1161 | goto free_pd_blkcg; | |
1162 | } | |
81437648 TH |
1163 | blkcg->cpd[i] = cpd; |
1164 | cpd->blkcg = blkcg; | |
e48453c3 | 1165 | cpd->plid = i; |
e4a9bde9 TH |
1166 | if (pol->cpd_init_fn) |
1167 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1168 | } |
31e4c28d | 1169 | |
31e4c28d | 1170 | spin_lock_init(&blkcg->lock); |
e00f4f4d | 1171 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1172 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1173 | #ifdef CONFIG_CGROUP_WRITEBACK |
1174 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
59b57717 | 1175 | refcount_set(&blkcg->cgwb_refcnt, 1); |
52ebea74 | 1176 | #endif |
7876f930 TH |
1177 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1178 | ||
1179 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1180 | return &blkcg->css; |
e48453c3 AA |
1181 | |
1182 | free_pd_blkcg: | |
1183 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1184 | if (blkcg->cpd[i]) |
1185 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
4c18c9e9 | 1186 | |
1187 | if (blkcg != &blkcg_root) | |
1188 | kfree(blkcg); | |
1189 | unlock: | |
7876f930 | 1190 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 1191 | return ret; |
31e4c28d VG |
1192 | } |
1193 | ||
5efd6113 TH |
1194 | /** |
1195 | * blkcg_init_queue - initialize blkcg part of request queue | |
1196 | * @q: request_queue to initialize | |
1197 | * | |
1198 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1199 | * part of new request_queue @q. | |
1200 | * | |
1201 | * RETURNS: | |
1202 | * 0 on success, -errno on failure. | |
1203 | */ | |
1204 | int blkcg_init_queue(struct request_queue *q) | |
1205 | { | |
d708f0d5 JA |
1206 | struct blkcg_gq *new_blkg, *blkg; |
1207 | bool preloaded; | |
ec13b1d6 TH |
1208 | int ret; |
1209 | ||
d708f0d5 JA |
1210 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
1211 | if (!new_blkg) | |
1212 | return -ENOMEM; | |
1213 | ||
1214 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1215 | ||
bea54883 | 1216 | /* Make sure the root blkg exists. */ |
ec13b1d6 | 1217 | rcu_read_lock(); |
0d945c1f | 1218 | spin_lock_irq(&q->queue_lock); |
d708f0d5 | 1219 | blkg = blkg_create(&blkcg_root, q, new_blkg); |
901932a3 JB |
1220 | if (IS_ERR(blkg)) |
1221 | goto err_unlock; | |
1222 | q->root_blkg = blkg; | |
0d945c1f | 1223 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 TH |
1224 | rcu_read_unlock(); |
1225 | ||
d708f0d5 JA |
1226 | if (preloaded) |
1227 | radix_tree_preload_end(); | |
1228 | ||
d7067512 | 1229 | ret = blk_iolatency_init(q); |
04be60b5 CH |
1230 | if (ret) |
1231 | goto err_destroy_all; | |
d7067512 | 1232 | |
ec13b1d6 | 1233 | ret = blk_throtl_init(q); |
04be60b5 CH |
1234 | if (ret) |
1235 | goto err_destroy_all; | |
1236 | return 0; | |
901932a3 | 1237 | |
04be60b5 | 1238 | err_destroy_all: |
04be60b5 | 1239 | blkg_destroy_all(q); |
04be60b5 | 1240 | return ret; |
901932a3 | 1241 | err_unlock: |
0d945c1f | 1242 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1243 | rcu_read_unlock(); |
1244 | if (preloaded) | |
1245 | radix_tree_preload_end(); | |
1246 | return PTR_ERR(blkg); | |
5efd6113 TH |
1247 | } |
1248 | ||
1249 | /** | |
1250 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1251 | * @q: request_queue to drain | |
1252 | * | |
1253 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1254 | */ | |
1255 | void blkcg_drain_queue(struct request_queue *q) | |
1256 | { | |
0d945c1f | 1257 | lockdep_assert_held(&q->queue_lock); |
5efd6113 | 1258 | |
0b462c89 TH |
1259 | /* |
1260 | * @q could be exiting and already have destroyed all blkgs as | |
1261 | * indicated by NULL root_blkg. If so, don't confuse policies. | |
1262 | */ | |
1263 | if (!q->root_blkg) | |
1264 | return; | |
1265 | ||
5efd6113 TH |
1266 | blk_throtl_drain(q); |
1267 | } | |
1268 | ||
1269 | /** | |
1270 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1271 | * @q: request_queue being released | |
1272 | * | |
7585d508 | 1273 | * Called from blk_exit_queue(). Responsible for exiting blkcg part. |
5efd6113 TH |
1274 | */ |
1275 | void blkcg_exit_queue(struct request_queue *q) | |
1276 | { | |
3c96cb32 | 1277 | blkg_destroy_all(q); |
5efd6113 TH |
1278 | blk_throtl_exit(q); |
1279 | } | |
1280 | ||
31e4c28d VG |
1281 | /* |
1282 | * We cannot support shared io contexts, as we have no mean to support | |
1283 | * two tasks with the same ioc in two different groups without major rework | |
1284 | * of the main cic data structures. For now we allow a task to change | |
1285 | * its cgroup only if it's the only owner of its ioc. | |
1286 | */ | |
1f7dd3e5 | 1287 | static int blkcg_can_attach(struct cgroup_taskset *tset) |
31e4c28d | 1288 | { |
bb9d97b6 | 1289 | struct task_struct *task; |
1f7dd3e5 | 1290 | struct cgroup_subsys_state *dst_css; |
31e4c28d VG |
1291 | struct io_context *ioc; |
1292 | int ret = 0; | |
1293 | ||
1294 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
1f7dd3e5 | 1295 | cgroup_taskset_for_each(task, dst_css, tset) { |
bb9d97b6 TH |
1296 | task_lock(task); |
1297 | ioc = task->io_context; | |
1298 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1299 | ret = -EINVAL; | |
1300 | task_unlock(task); | |
1301 | if (ret) | |
1302 | break; | |
1303 | } | |
31e4c28d VG |
1304 | return ret; |
1305 | } | |
1306 | ||
69d7fde5 TH |
1307 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1308 | { | |
1309 | int i; | |
1310 | ||
1311 | mutex_lock(&blkcg_pol_mutex); | |
1312 | ||
1313 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1314 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1315 | struct blkcg *blkcg; | |
1316 | ||
1317 | if (!pol || !pol->cpd_bind_fn) | |
1318 | continue; | |
1319 | ||
1320 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1321 | if (blkcg->cpd[pol->plid]) | |
1322 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1323 | } | |
1324 | mutex_unlock(&blkcg_pol_mutex); | |
1325 | } | |
1326 | ||
d09d8df3 JB |
1327 | static void blkcg_exit(struct task_struct *tsk) |
1328 | { | |
1329 | if (tsk->throttle_queue) | |
1330 | blk_put_queue(tsk->throttle_queue); | |
1331 | tsk->throttle_queue = NULL; | |
1332 | } | |
1333 | ||
c165b3e3 | 1334 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 TH |
1335 | .css_alloc = blkcg_css_alloc, |
1336 | .css_offline = blkcg_css_offline, | |
1337 | .css_free = blkcg_css_free, | |
3c798398 | 1338 | .can_attach = blkcg_can_attach, |
69d7fde5 | 1339 | .bind = blkcg_bind, |
2ee867dc | 1340 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1341 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1342 | .legacy_name = "blkio", |
d09d8df3 | 1343 | .exit = blkcg_exit, |
1ced953b TH |
1344 | #ifdef CONFIG_MEMCG |
1345 | /* | |
1346 | * This ensures that, if available, memcg is automatically enabled | |
1347 | * together on the default hierarchy so that the owner cgroup can | |
1348 | * be retrieved from writeback pages. | |
1349 | */ | |
1350 | .depends_on = 1 << memory_cgrp_id, | |
1351 | #endif | |
676f7c8f | 1352 | }; |
c165b3e3 | 1353 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1354 | |
a2b1693b TH |
1355 | /** |
1356 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1357 | * @q: request_queue of interest | |
1358 | * @pol: blkcg policy to activate | |
1359 | * | |
1360 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1361 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1362 | * | |
1363 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1364 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1365 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1366 | * always enough for dereferencing policy data. | |
1367 | * | |
1368 | * The caller is responsible for synchronizing [de]activations and policy | |
1369 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1370 | */ | |
1371 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1372 | const struct blkcg_policy *pol) |
a2b1693b | 1373 | { |
4c55f4f9 | 1374 | struct blkg_policy_data *pd_prealloc = NULL; |
ec13b1d6 | 1375 | struct blkcg_gq *blkg; |
4c55f4f9 | 1376 | int ret; |
a2b1693b TH |
1377 | |
1378 | if (blkcg_policy_enabled(q, pol)) | |
1379 | return 0; | |
1380 | ||
344e9ffc | 1381 | if (queue_is_mq(q)) |
bd166ef1 | 1382 | blk_mq_freeze_queue(q); |
4c55f4f9 TH |
1383 | pd_prealloc: |
1384 | if (!pd_prealloc) { | |
001bea73 | 1385 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node); |
4c55f4f9 | 1386 | if (!pd_prealloc) { |
a2b1693b | 1387 | ret = -ENOMEM; |
4c55f4f9 | 1388 | goto out_bypass_end; |
a2b1693b | 1389 | } |
a2b1693b TH |
1390 | } |
1391 | ||
0d945c1f | 1392 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1393 | |
1394 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
4c55f4f9 TH |
1395 | struct blkg_policy_data *pd; |
1396 | ||
1397 | if (blkg->pd[pol->plid]) | |
1398 | continue; | |
a2b1693b | 1399 | |
e00f4f4d | 1400 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node); |
4c55f4f9 TH |
1401 | if (!pd) |
1402 | swap(pd, pd_prealloc); | |
1403 | if (!pd) { | |
0d945c1f | 1404 | spin_unlock_irq(&q->queue_lock); |
4c55f4f9 TH |
1405 | goto pd_prealloc; |
1406 | } | |
a2b1693b TH |
1407 | |
1408 | blkg->pd[pol->plid] = pd; | |
1409 | pd->blkg = blkg; | |
b276a876 | 1410 | pd->plid = pol->plid; |
3e418710 | 1411 | if (pol->pd_init_fn) |
a9520cd6 | 1412 | pol->pd_init_fn(pd); |
a2b1693b TH |
1413 | } |
1414 | ||
1415 | __set_bit(pol->plid, q->blkcg_pols); | |
1416 | ret = 0; | |
4c55f4f9 | 1417 | |
0d945c1f | 1418 | spin_unlock_irq(&q->queue_lock); |
4c55f4f9 | 1419 | out_bypass_end: |
344e9ffc | 1420 | if (queue_is_mq(q)) |
bd166ef1 | 1421 | blk_mq_unfreeze_queue(q); |
001bea73 TH |
1422 | if (pd_prealloc) |
1423 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b TH |
1424 | return ret; |
1425 | } | |
1426 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1427 | ||
1428 | /** | |
1429 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1430 | * @q: request_queue of interest | |
1431 | * @pol: blkcg policy to deactivate | |
1432 | * | |
1433 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1434 | * blkcg_activate_policy(). | |
1435 | */ | |
1436 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1437 | const struct blkcg_policy *pol) |
a2b1693b | 1438 | { |
3c798398 | 1439 | struct blkcg_gq *blkg; |
a2b1693b TH |
1440 | |
1441 | if (!blkcg_policy_enabled(q, pol)) | |
1442 | return; | |
1443 | ||
344e9ffc | 1444 | if (queue_is_mq(q)) |
bd166ef1 | 1445 | blk_mq_freeze_queue(q); |
bd166ef1 | 1446 | |
0d945c1f | 1447 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1448 | |
1449 | __clear_bit(pol->plid, q->blkcg_pols); | |
1450 | ||
1451 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
001bea73 | 1452 | if (blkg->pd[pol->plid]) { |
6b065462 | 1453 | if (pol->pd_offline_fn) |
a9520cd6 | 1454 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1455 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1456 | blkg->pd[pol->plid] = NULL; | |
1457 | } | |
a2b1693b TH |
1458 | } |
1459 | ||
0d945c1f | 1460 | spin_unlock_irq(&q->queue_lock); |
bd166ef1 | 1461 | |
344e9ffc | 1462 | if (queue_is_mq(q)) |
bd166ef1 | 1463 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1464 | } |
1465 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1466 | ||
8bd435b3 | 1467 | /** |
3c798398 TH |
1468 | * blkcg_policy_register - register a blkcg policy |
1469 | * @pol: blkcg policy to register | |
8bd435b3 | 1470 | * |
3c798398 TH |
1471 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1472 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1473 | */ |
d5bf0291 | 1474 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1475 | { |
06b285bd | 1476 | struct blkcg *blkcg; |
8bd435b3 | 1477 | int i, ret; |
e8989fae | 1478 | |
838f13bf | 1479 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1480 | mutex_lock(&blkcg_pol_mutex); |
1481 | ||
8bd435b3 TH |
1482 | /* find an empty slot */ |
1483 | ret = -ENOSPC; | |
1484 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1485 | if (!blkcg_policy[i]) |
8bd435b3 | 1486 | break; |
01c5f85a JA |
1487 | if (i >= BLKCG_MAX_POLS) { |
1488 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1489 | goto err_unlock; |
01c5f85a | 1490 | } |
035d10b2 | 1491 | |
e8401073 | 1492 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1493 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1494 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1495 | goto err_unlock; | |
1496 | ||
06b285bd | 1497 | /* register @pol */ |
3c798398 | 1498 | pol->plid = i; |
06b285bd TH |
1499 | blkcg_policy[pol->plid] = pol; |
1500 | ||
1501 | /* allocate and install cpd's */ | |
e4a9bde9 | 1502 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1503 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1504 | struct blkcg_policy_data *cpd; | |
1505 | ||
e4a9bde9 | 1506 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1507 | if (!cpd) |
06b285bd | 1508 | goto err_free_cpds; |
06b285bd | 1509 | |
81437648 TH |
1510 | blkcg->cpd[pol->plid] = cpd; |
1511 | cpd->blkcg = blkcg; | |
06b285bd | 1512 | cpd->plid = pol->plid; |
81437648 | 1513 | pol->cpd_init_fn(cpd); |
06b285bd TH |
1514 | } |
1515 | } | |
1516 | ||
838f13bf | 1517 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1518 | |
8bd435b3 | 1519 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1520 | if (pol->dfl_cftypes) |
1521 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1522 | pol->dfl_cftypes)); | |
880f50e2 | 1523 | if (pol->legacy_cftypes) |
c165b3e3 | 1524 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1525 | pol->legacy_cftypes)); |
838f13bf TH |
1526 | mutex_unlock(&blkcg_pol_register_mutex); |
1527 | return 0; | |
1528 | ||
06b285bd | 1529 | err_free_cpds: |
58a9edce | 1530 | if (pol->cpd_free_fn) { |
06b285bd | 1531 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1532 | if (blkcg->cpd[pol->plid]) { |
1533 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1534 | blkcg->cpd[pol->plid] = NULL; | |
1535 | } | |
06b285bd TH |
1536 | } |
1537 | } | |
1538 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1539 | err_unlock: |
bc0d6501 | 1540 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1541 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1542 | return ret; |
3e252066 | 1543 | } |
3c798398 | 1544 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1545 | |
8bd435b3 | 1546 | /** |
3c798398 TH |
1547 | * blkcg_policy_unregister - unregister a blkcg policy |
1548 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1549 | * |
3c798398 | 1550 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1551 | */ |
3c798398 | 1552 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1553 | { |
06b285bd TH |
1554 | struct blkcg *blkcg; |
1555 | ||
838f13bf | 1556 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1557 | |
3c798398 | 1558 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1559 | goto out_unlock; |
1560 | ||
1561 | /* kill the intf files first */ | |
2ee867dc TH |
1562 | if (pol->dfl_cftypes) |
1563 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1564 | if (pol->legacy_cftypes) |
1565 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1566 | |
06b285bd | 1567 | /* remove cpds and unregister */ |
838f13bf | 1568 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1569 | |
58a9edce | 1570 | if (pol->cpd_free_fn) { |
06b285bd | 1571 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1572 | if (blkcg->cpd[pol->plid]) { |
1573 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1574 | blkcg->cpd[pol->plid] = NULL; | |
1575 | } | |
06b285bd TH |
1576 | } |
1577 | } | |
3c798398 | 1578 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1579 | |
bc0d6501 | 1580 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1581 | out_unlock: |
1582 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1583 | } |
3c798398 | 1584 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1585 | |
d09d8df3 JB |
1586 | /* |
1587 | * Scale the accumulated delay based on how long it has been since we updated | |
1588 | * the delay. We only call this when we are adding delay, in case it's been a | |
1589 | * while since we added delay, and when we are checking to see if we need to | |
1590 | * delay a task, to account for any delays that may have occurred. | |
1591 | */ | |
1592 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1593 | { | |
1594 | u64 old = atomic64_read(&blkg->delay_start); | |
1595 | ||
1596 | /* | |
1597 | * We only want to scale down every second. The idea here is that we | |
1598 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1599 | * time window. We only want to throttle tasks for recent delay that | |
1600 | * has occurred, in 1 second time windows since that's the maximum | |
1601 | * things can be throttled. We save the current delay window in | |
1602 | * blkg->last_delay so we know what amount is still left to be charged | |
1603 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1604 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1605 | * are ok with whatever is happening now, and we can take away more of | |
1606 | * the accumulated delay as we've already throttled enough that | |
1607 | * everybody is happy with their IO latencies. | |
1608 | */ | |
1609 | if (time_before64(old + NSEC_PER_SEC, now) && | |
1610 | atomic64_cmpxchg(&blkg->delay_start, old, now) == old) { | |
1611 | u64 cur = atomic64_read(&blkg->delay_nsec); | |
1612 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1613 | int cur_use = atomic_read(&blkg->use_delay); | |
1614 | ||
1615 | /* | |
1616 | * We've been unthrottled, subtract a larger chunk of our | |
1617 | * accumulated delay. | |
1618 | */ | |
1619 | if (cur_use < blkg->last_use) | |
1620 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1621 | ||
1622 | /* | |
1623 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1624 | * should only ever be growing except here where we subtract out | |
1625 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1626 | * rather not end up with negative numbers. | |
1627 | */ | |
1628 | if (unlikely(cur < sub)) { | |
1629 | atomic64_set(&blkg->delay_nsec, 0); | |
1630 | blkg->last_delay = 0; | |
1631 | } else { | |
1632 | atomic64_sub(sub, &blkg->delay_nsec); | |
1633 | blkg->last_delay = cur - sub; | |
1634 | } | |
1635 | blkg->last_use = cur_use; | |
1636 | } | |
1637 | } | |
1638 | ||
1639 | /* | |
1640 | * This is called when we want to actually walk up the hierarchy and check to | |
1641 | * see if we need to throttle, and then actually throttle if there is some | |
1642 | * accumulated delay. This should only be called upon return to user space so | |
1643 | * we're not holding some lock that would induce a priority inversion. | |
1644 | */ | |
1645 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1646 | { | |
1647 | u64 now = ktime_to_ns(ktime_get()); | |
1648 | u64 exp; | |
1649 | u64 delay_nsec = 0; | |
1650 | int tok; | |
1651 | ||
1652 | while (blkg->parent) { | |
1653 | if (atomic_read(&blkg->use_delay)) { | |
1654 | blkcg_scale_delay(blkg, now); | |
1655 | delay_nsec = max_t(u64, delay_nsec, | |
1656 | atomic64_read(&blkg->delay_nsec)); | |
1657 | } | |
1658 | blkg = blkg->parent; | |
1659 | } | |
1660 | ||
1661 | if (!delay_nsec) | |
1662 | return; | |
1663 | ||
1664 | /* | |
1665 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1666 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1667 | * delay, and we want userspace to be able to do _something_ so cap the | |
1668 | * delays at 1 second. If there's 10's of seconds worth of delay then | |
1669 | * the tasks will be delayed for 1 second for every syscall. | |
1670 | */ | |
1671 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
1672 | ||
1673 | /* | |
1674 | * TODO: the use_memdelay flag is going to be for the upcoming psi stuff | |
1675 | * that hasn't landed upstream yet. Once that stuff is in place we need | |
1676 | * to do a psi_memstall_enter/leave if memdelay is set. | |
1677 | */ | |
1678 | ||
1679 | exp = ktime_add_ns(now, delay_nsec); | |
1680 | tok = io_schedule_prepare(); | |
1681 | do { | |
1682 | __set_current_state(TASK_KILLABLE); | |
1683 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1684 | break; | |
1685 | } while (!fatal_signal_pending(current)); | |
1686 | io_schedule_finish(tok); | |
1687 | } | |
1688 | ||
1689 | /** | |
1690 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1691 | * | |
1692 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1693 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
1694 | * check to see if current->throttle_queue is set and if not this doesn't do | |
1695 | * anything. This should only ever be called by the resume code, it's not meant | |
1696 | * to be called by people willy-nilly as it will actually do the work to | |
1697 | * throttle the task if it is setup for throttling. | |
1698 | */ | |
1699 | void blkcg_maybe_throttle_current(void) | |
1700 | { | |
1701 | struct request_queue *q = current->throttle_queue; | |
1702 | struct cgroup_subsys_state *css; | |
1703 | struct blkcg *blkcg; | |
1704 | struct blkcg_gq *blkg; | |
1705 | bool use_memdelay = current->use_memdelay; | |
1706 | ||
1707 | if (!q) | |
1708 | return; | |
1709 | ||
1710 | current->throttle_queue = NULL; | |
1711 | current->use_memdelay = false; | |
1712 | ||
1713 | rcu_read_lock(); | |
1714 | css = kthread_blkcg(); | |
1715 | if (css) | |
1716 | blkcg = css_to_blkcg(css); | |
1717 | else | |
1718 | blkcg = css_to_blkcg(task_css(current, io_cgrp_id)); | |
1719 | ||
1720 | if (!blkcg) | |
1721 | goto out; | |
1722 | blkg = blkg_lookup(blkcg, q); | |
1723 | if (!blkg) | |
1724 | goto out; | |
7754f669 | 1725 | if (!blkg_tryget(blkg)) |
d09d8df3 JB |
1726 | goto out; |
1727 | rcu_read_unlock(); | |
d09d8df3 JB |
1728 | |
1729 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1730 | blkg_put(blkg); | |
cc7ecc25 | 1731 | blk_put_queue(q); |
d09d8df3 JB |
1732 | return; |
1733 | out: | |
1734 | rcu_read_unlock(); | |
1735 | blk_put_queue(q); | |
1736 | } | |
d09d8df3 JB |
1737 | |
1738 | /** | |
1739 | * blkcg_schedule_throttle - this task needs to check for throttling | |
537d71b3 BVA |
1740 | * @q: the request queue IO was submitted on |
1741 | * @use_memdelay: do we charge this to memory delay for PSI | |
d09d8df3 JB |
1742 | * |
1743 | * This is called by the IO controller when we know there's delay accumulated | |
1744 | * for the blkg for this task. We do not pass the blkg because there are places | |
1745 | * we call this that may not have that information, the swapping code for | |
1746 | * instance will only have a request_queue at that point. This set's the | |
1747 | * notify_resume for the task to check and see if it requires throttling before | |
1748 | * returning to user space. | |
1749 | * | |
1750 | * We will only schedule once per syscall. You can call this over and over | |
1751 | * again and it will only do the check once upon return to user space, and only | |
1752 | * throttle once. If the task needs to be throttled again it'll need to be | |
1753 | * re-set at the next time we see the task. | |
1754 | */ | |
1755 | void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) | |
1756 | { | |
1757 | if (unlikely(current->flags & PF_KTHREAD)) | |
1758 | return; | |
1759 | ||
1760 | if (!blk_get_queue(q)) | |
1761 | return; | |
1762 | ||
1763 | if (current->throttle_queue) | |
1764 | blk_put_queue(current->throttle_queue); | |
1765 | current->throttle_queue = q; | |
1766 | if (use_memdelay) | |
1767 | current->use_memdelay = use_memdelay; | |
1768 | set_notify_resume(current); | |
1769 | } | |
d09d8df3 JB |
1770 | |
1771 | /** | |
1772 | * blkcg_add_delay - add delay to this blkg | |
537d71b3 BVA |
1773 | * @blkg: blkg of interest |
1774 | * @now: the current time in nanoseconds | |
1775 | * @delta: how many nanoseconds of delay to add | |
d09d8df3 JB |
1776 | * |
1777 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1778 | * throttle tasks if an IO controller thinks we need more throttling. | |
1779 | */ | |
1780 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1781 | { | |
1782 | blkcg_scale_delay(blkg, now); | |
1783 | atomic64_add(delta, &blkg->delay_nsec); | |
1784 | } | |
d09d8df3 | 1785 | |
903d23f0 JB |
1786 | module_param(blkcg_debug_stats, bool, 0644); |
1787 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |