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 | |
77ea7338 TH |
79 | blkg_rwstat_exit(&blkg->stat_ios); |
80 | blkg_rwstat_exit(&blkg->stat_bytes); | |
549d3aa8 | 81 | kfree(blkg); |
0381411e TH |
82 | } |
83 | ||
84 | /** | |
85 | * blkg_alloc - allocate a blkg | |
86 | * @blkcg: block cgroup the new blkg is associated with | |
87 | * @q: request_queue the new blkg is associated with | |
15974993 | 88 | * @gfp_mask: allocation mask to use |
0381411e | 89 | * |
e8989fae | 90 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 91 | */ |
15974993 TH |
92 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
93 | gfp_t gfp_mask) | |
0381411e | 94 | { |
3c798398 | 95 | struct blkcg_gq *blkg; |
e8989fae | 96 | int i; |
0381411e TH |
97 | |
98 | /* alloc and init base part */ | |
15974993 | 99 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
100 | if (!blkg) |
101 | return NULL; | |
102 | ||
77ea7338 TH |
103 | if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) || |
104 | blkg_rwstat_init(&blkg->stat_ios, gfp_mask)) | |
105 | goto err_free; | |
106 | ||
c875f4d0 | 107 | blkg->q = q; |
e8989fae | 108 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 109 | blkg->blkcg = blkcg; |
b5f2954d | 110 | atomic_set(&blkg->refcnt, 1); |
0381411e | 111 | |
8bd435b3 | 112 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 113 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 114 | struct blkg_policy_data *pd; |
0381411e | 115 | |
a2b1693b | 116 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
117 | continue; |
118 | ||
119 | /* alloc per-policy data and attach it to blkg */ | |
001bea73 | 120 | pd = pol->pd_alloc_fn(gfp_mask, q->node); |
a051661c TH |
121 | if (!pd) |
122 | goto err_free; | |
549d3aa8 | 123 | |
e8989fae TH |
124 | blkg->pd[i] = pd; |
125 | pd->blkg = blkg; | |
b276a876 | 126 | pd->plid = i; |
e8989fae TH |
127 | } |
128 | ||
0381411e | 129 | return blkg; |
a051661c TH |
130 | |
131 | err_free: | |
132 | blkg_free(blkg); | |
133 | return NULL; | |
0381411e TH |
134 | } |
135 | ||
24f29046 TH |
136 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
137 | struct request_queue *q, bool update_hint) | |
80fd9979 | 138 | { |
3c798398 | 139 | struct blkcg_gq *blkg; |
80fd9979 | 140 | |
a637120e | 141 | /* |
86cde6b6 TH |
142 | * Hint didn't match. Look up from the radix tree. Note that the |
143 | * hint can only be updated under queue_lock as otherwise @blkg | |
144 | * could have already been removed from blkg_tree. The caller is | |
145 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
146 | */ |
147 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
148 | if (blkg && blkg->q == q) { |
149 | if (update_hint) { | |
0d945c1f | 150 | lockdep_assert_held(&q->queue_lock); |
86cde6b6 TH |
151 | rcu_assign_pointer(blkcg->blkg_hint, blkg); |
152 | } | |
a637120e | 153 | return blkg; |
86cde6b6 | 154 | } |
a637120e | 155 | |
80fd9979 TH |
156 | return NULL; |
157 | } | |
ae118896 | 158 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); |
80fd9979 | 159 | |
15974993 | 160 | /* |
d708f0d5 JA |
161 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
162 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 163 | */ |
86cde6b6 | 164 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
d708f0d5 JA |
165 | struct request_queue *q, |
166 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 167 | { |
d708f0d5 | 168 | struct blkcg_gq *blkg; |
ce7acfea | 169 | struct bdi_writeback_congested *wb_congested; |
f427d909 | 170 | int i, ret; |
5624a4e4 | 171 | |
cd1604fa | 172 | WARN_ON_ONCE(!rcu_read_lock_held()); |
0d945c1f | 173 | lockdep_assert_held(&q->queue_lock); |
cd1604fa | 174 | |
7ee9c562 | 175 | /* blkg holds a reference to blkcg */ |
ec903c0c | 176 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 177 | ret = -ENODEV; |
93e6d5d8 | 178 | goto err_free_blkg; |
15974993 | 179 | } |
cd1604fa | 180 | |
dc3b17cc | 181 | wb_congested = wb_congested_get_create(q->backing_dev_info, |
d708f0d5 JA |
182 | blkcg->css.id, |
183 | GFP_NOWAIT | __GFP_NOWARN); | |
184 | if (!wb_congested) { | |
ce7acfea | 185 | ret = -ENOMEM; |
d708f0d5 | 186 | goto err_put_css; |
ce7acfea TH |
187 | } |
188 | ||
d708f0d5 JA |
189 | /* allocate */ |
190 | if (!new_blkg) { | |
191 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN); | |
192 | if (unlikely(!new_blkg)) { | |
193 | ret = -ENOMEM; | |
194 | goto err_put_congested; | |
15974993 TH |
195 | } |
196 | } | |
d708f0d5 JA |
197 | blkg = new_blkg; |
198 | blkg->wb_congested = wb_congested; | |
cd1604fa | 199 | |
db613670 | 200 | /* link parent */ |
3c547865 TH |
201 | if (blkcg_parent(blkcg)) { |
202 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
203 | if (WARN_ON_ONCE(!blkg->parent)) { | |
20386ce0 | 204 | ret = -ENODEV; |
d708f0d5 | 205 | goto err_put_congested; |
3c547865 TH |
206 | } |
207 | blkg_get(blkg->parent); | |
208 | } | |
209 | ||
db613670 TH |
210 | /* invoke per-policy init */ |
211 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
212 | struct blkcg_policy *pol = blkcg_policy[i]; | |
213 | ||
214 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 215 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
216 | } |
217 | ||
218 | /* insert */ | |
cd1604fa | 219 | spin_lock(&blkcg->lock); |
a637120e TH |
220 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
221 | if (likely(!ret)) { | |
222 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
223 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
224 | |
225 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
226 | struct blkcg_policy *pol = blkcg_policy[i]; | |
227 | ||
228 | if (blkg->pd[i] && pol->pd_online_fn) | |
a9520cd6 | 229 | pol->pd_online_fn(blkg->pd[i]); |
f427d909 | 230 | } |
a637120e | 231 | } |
f427d909 | 232 | blkg->online = true; |
cd1604fa | 233 | spin_unlock(&blkcg->lock); |
496fb780 | 234 | |
ec13b1d6 | 235 | if (!ret) |
a637120e | 236 | return blkg; |
15974993 | 237 | |
3c547865 TH |
238 | /* @blkg failed fully initialized, use the usual release path */ |
239 | blkg_put(blkg); | |
240 | return ERR_PTR(ret); | |
241 | ||
d708f0d5 JA |
242 | err_put_congested: |
243 | wb_congested_put(wb_congested); | |
244 | err_put_css: | |
496fb780 | 245 | css_put(&blkcg->css); |
93e6d5d8 | 246 | err_free_blkg: |
d708f0d5 | 247 | blkg_free(new_blkg); |
93e6d5d8 | 248 | return ERR_PTR(ret); |
31e4c28d | 249 | } |
3c96cb32 | 250 | |
86cde6b6 | 251 | /** |
b978962a | 252 | * __blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 TH |
253 | * @blkcg: blkcg of interest |
254 | * @q: request_queue of interest | |
255 | * | |
256 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
257 | * create one. blkg creation is performed recursively from blkcg_root such |
258 | * that all non-root blkg's have access to the parent blkg. This function | |
259 | * should be called under RCU read lock and @q->queue_lock. | |
86cde6b6 | 260 | * |
beea9da0 DZ |
261 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
262 | * down from root. | |
86cde6b6 | 263 | */ |
b978962a DZ |
264 | struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg, |
265 | struct request_queue *q) | |
3c96cb32 | 266 | { |
86cde6b6 TH |
267 | struct blkcg_gq *blkg; |
268 | ||
269 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
0d945c1f | 270 | lockdep_assert_held(&q->queue_lock); |
86cde6b6 | 271 | |
86cde6b6 TH |
272 | blkg = __blkg_lookup(blkcg, q, true); |
273 | if (blkg) | |
274 | return blkg; | |
275 | ||
3c547865 TH |
276 | /* |
277 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
278 | * non-root blkgs have access to their parents. Returns the closest |
279 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
280 | */ |
281 | while (true) { | |
282 | struct blkcg *pos = blkcg; | |
283 | struct blkcg *parent = blkcg_parent(blkcg); | |
beea9da0 DZ |
284 | struct blkcg_gq *ret_blkg = q->root_blkg; |
285 | ||
286 | while (parent) { | |
287 | blkg = __blkg_lookup(parent, q, false); | |
288 | if (blkg) { | |
289 | /* remember closest blkg */ | |
290 | ret_blkg = blkg; | |
291 | break; | |
292 | } | |
3c547865 TH |
293 | pos = parent; |
294 | parent = blkcg_parent(parent); | |
295 | } | |
296 | ||
d708f0d5 | 297 | blkg = blkg_create(pos, q, NULL); |
beea9da0 DZ |
298 | if (IS_ERR(blkg)) |
299 | return ret_blkg; | |
300 | if (pos == blkcg) | |
3c547865 TH |
301 | return blkg; |
302 | } | |
3c96cb32 | 303 | } |
31e4c28d | 304 | |
b978962a DZ |
305 | /** |
306 | * blkg_lookup_create - find or create a blkg | |
307 | * @blkcg: target block cgroup | |
308 | * @q: target request_queue | |
309 | * | |
310 | * This looks up or creates the blkg representing the unique pair | |
311 | * of the blkcg and the request_queue. | |
312 | */ | |
313 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, | |
314 | struct request_queue *q) | |
315 | { | |
316 | struct blkcg_gq *blkg = blkg_lookup(blkcg, q); | |
317 | ||
318 | if (unlikely(!blkg)) { | |
319 | spin_lock_irq(&q->queue_lock); | |
320 | blkg = __blkg_lookup_create(blkcg, q); | |
321 | spin_unlock_irq(&q->queue_lock); | |
322 | } | |
323 | ||
324 | return blkg; | |
325 | } | |
326 | ||
3c798398 | 327 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 328 | { |
3c798398 | 329 | struct blkcg *blkcg = blkg->blkcg; |
77ea7338 | 330 | struct blkcg_gq *parent = blkg->parent; |
6b065462 | 331 | int i; |
03aa264a | 332 | |
0d945c1f | 333 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 334 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
335 | |
336 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 337 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 338 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 339 | |
6b065462 DZF |
340 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
341 | struct blkcg_policy *pol = blkcg_policy[i]; | |
342 | ||
343 | if (blkg->pd[i] && pol->pd_offline_fn) | |
344 | pol->pd_offline_fn(blkg->pd[i]); | |
345 | } | |
346 | ||
77ea7338 TH |
347 | if (parent) { |
348 | blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes); | |
349 | blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios); | |
350 | } | |
351 | ||
f427d909 TH |
352 | blkg->online = false; |
353 | ||
a637120e | 354 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 355 | list_del_init(&blkg->q_node); |
9f13ef67 | 356 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 357 | |
a637120e TH |
358 | /* |
359 | * Both setting lookup hint to and clearing it from @blkg are done | |
360 | * under queue_lock. If it's not pointing to @blkg now, it never | |
361 | * will. Hint assignment itself can race safely. | |
362 | */ | |
ec6c676a | 363 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
364 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
365 | ||
03aa264a TH |
366 | /* |
367 | * Put the reference taken at the time of creation so that when all | |
368 | * queues are gone, group can be destroyed. | |
369 | */ | |
b5f2954d | 370 | blkg_put(blkg); |
03aa264a TH |
371 | } |
372 | ||
9f13ef67 TH |
373 | /** |
374 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
375 | * @q: request_queue of interest | |
9f13ef67 | 376 | * |
3c96cb32 | 377 | * Destroy all blkgs associated with @q. |
9f13ef67 | 378 | */ |
3c96cb32 | 379 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 380 | { |
3c798398 | 381 | struct blkcg_gq *blkg, *n; |
72e06c25 | 382 | |
0d945c1f | 383 | spin_lock_irq(&q->queue_lock); |
9f13ef67 | 384 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 385 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 386 | |
9f13ef67 TH |
387 | spin_lock(&blkcg->lock); |
388 | blkg_destroy(blkg); | |
389 | spin_unlock(&blkcg->lock); | |
72e06c25 | 390 | } |
6fe810bd TH |
391 | |
392 | q->root_blkg = NULL; | |
0d945c1f | 393 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
394 | } |
395 | ||
b5f2954d DZ |
396 | /* |
397 | * A group is RCU protected, but having an rcu lock does not mean that one | |
398 | * can access all the fields of blkg and assume these are valid. For | |
399 | * example, don't try to follow throtl_data and request queue links. | |
400 | * | |
401 | * Having a reference to blkg under an rcu allows accesses to only values | |
402 | * local to groups like group stats and group rate limits. | |
403 | */ | |
404 | void __blkg_release_rcu(struct rcu_head *rcu_head) | |
405 | { | |
406 | struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head); | |
407 | ||
408 | /* release the blkcg and parent blkg refs this blkg has been holding */ | |
409 | css_put(&blkg->blkcg->css); | |
410 | if (blkg->parent) | |
411 | blkg_put(blkg->parent); | |
412 | ||
413 | wb_congested_put(blkg->wb_congested); | |
414 | ||
415 | blkg_free(blkg); | |
416 | } | |
417 | EXPORT_SYMBOL_GPL(__blkg_release_rcu); | |
418 | ||
182446d0 TH |
419 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
420 | struct cftype *cftype, u64 val) | |
303a3acb | 421 | { |
182446d0 | 422 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 423 | struct blkcg_gq *blkg; |
bc0d6501 | 424 | int i; |
303a3acb | 425 | |
838f13bf | 426 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 427 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
428 | |
429 | /* | |
430 | * Note that stat reset is racy - it doesn't synchronize against | |
431 | * stat updates. This is a debug feature which shouldn't exist | |
432 | * anyway. If you get hit by a race, retry. | |
433 | */ | |
b67bfe0d | 434 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
77ea7338 TH |
435 | blkg_rwstat_reset(&blkg->stat_bytes); |
436 | blkg_rwstat_reset(&blkg->stat_ios); | |
437 | ||
8bd435b3 | 438 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 439 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 440 | |
a9520cd6 TH |
441 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
442 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 443 | } |
303a3acb | 444 | } |
f0bdc8cd | 445 | |
303a3acb | 446 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 447 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
448 | return 0; |
449 | } | |
450 | ||
dd165eb3 | 451 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 452 | { |
d3d32e69 | 453 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
dc3b17cc JK |
454 | if (blkg->q->backing_dev_info->dev) |
455 | return dev_name(blkg->q->backing_dev_info->dev); | |
d3d32e69 | 456 | return NULL; |
303a3acb DS |
457 | } |
458 | ||
d3d32e69 TH |
459 | /** |
460 | * blkcg_print_blkgs - helper for printing per-blkg data | |
461 | * @sf: seq_file to print to | |
462 | * @blkcg: blkcg of interest | |
463 | * @prfill: fill function to print out a blkg | |
464 | * @pol: policy in question | |
465 | * @data: data to be passed to @prfill | |
466 | * @show_total: to print out sum of prfill return values or not | |
467 | * | |
468 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
469 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
470 | * policy data and @data and the matching queue lock held. If @show_total |
471 | * is %true, the sum of the return values from @prfill is printed with | |
472 | * "Total" label at the end. | |
d3d32e69 TH |
473 | * |
474 | * This is to be used to construct print functions for | |
475 | * cftype->read_seq_string method. | |
476 | */ | |
3c798398 | 477 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
478 | u64 (*prfill)(struct seq_file *, |
479 | struct blkg_policy_data *, int), | |
3c798398 | 480 | const struct blkcg_policy *pol, int data, |
ec399347 | 481 | bool show_total) |
5624a4e4 | 482 | { |
3c798398 | 483 | struct blkcg_gq *blkg; |
d3d32e69 | 484 | u64 total = 0; |
5624a4e4 | 485 | |
810ecfa7 | 486 | rcu_read_lock(); |
ee89f812 | 487 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 488 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 489 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 490 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 491 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
492 | } |
493 | rcu_read_unlock(); | |
d3d32e69 TH |
494 | |
495 | if (show_total) | |
496 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
497 | } | |
829fdb50 | 498 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
499 | |
500 | /** | |
501 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
502 | * @sf: seq_file to print to | |
f95a04af | 503 | * @pd: policy private data of interest |
d3d32e69 TH |
504 | * @v: value to print |
505 | * | |
f95a04af | 506 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 507 | */ |
f95a04af | 508 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 509 | { |
f95a04af | 510 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
511 | |
512 | if (!dname) | |
513 | return 0; | |
514 | ||
515 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
516 | return v; | |
517 | } | |
829fdb50 | 518 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
519 | |
520 | /** | |
521 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
522 | * @sf: seq_file to print to | |
f95a04af | 523 | * @pd: policy private data of interest |
d3d32e69 TH |
524 | * @rwstat: rwstat to print |
525 | * | |
f95a04af | 526 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 527 | */ |
f95a04af | 528 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 529 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
530 | { |
531 | static const char *rwstr[] = { | |
532 | [BLKG_RWSTAT_READ] = "Read", | |
533 | [BLKG_RWSTAT_WRITE] = "Write", | |
534 | [BLKG_RWSTAT_SYNC] = "Sync", | |
535 | [BLKG_RWSTAT_ASYNC] = "Async", | |
636620b6 | 536 | [BLKG_RWSTAT_DISCARD] = "Discard", |
d3d32e69 | 537 | }; |
f95a04af | 538 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
539 | u64 v; |
540 | int i; | |
541 | ||
542 | if (!dname) | |
543 | return 0; | |
544 | ||
545 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
546 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
24bdb8ef | 547 | (unsigned long long)atomic64_read(&rwstat->aux_cnt[i])); |
d3d32e69 | 548 | |
24bdb8ef | 549 | v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) + |
636620b6 TH |
550 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]) + |
551 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_DISCARD]); | |
d3d32e69 TH |
552 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); |
553 | return v; | |
554 | } | |
b50da39f | 555 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 556 | |
5bc4afb1 TH |
557 | /** |
558 | * blkg_prfill_stat - prfill callback for blkg_stat | |
559 | * @sf: seq_file to print to | |
f95a04af TH |
560 | * @pd: policy private data of interest |
561 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
562 | * |
563 | * prfill callback for printing a blkg_stat. | |
564 | */ | |
f95a04af | 565 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 566 | { |
f95a04af | 567 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 568 | } |
5bc4afb1 | 569 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 570 | |
5bc4afb1 TH |
571 | /** |
572 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
573 | * @sf: seq_file to print to | |
f95a04af TH |
574 | * @pd: policy private data of interest |
575 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
576 | * |
577 | * prfill callback for printing a blkg_rwstat. | |
578 | */ | |
f95a04af TH |
579 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
580 | int off) | |
d3d32e69 | 581 | { |
f95a04af | 582 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 583 | |
f95a04af | 584 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 585 | } |
5bc4afb1 | 586 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 587 | |
77ea7338 TH |
588 | static u64 blkg_prfill_rwstat_field(struct seq_file *sf, |
589 | struct blkg_policy_data *pd, int off) | |
590 | { | |
591 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off); | |
592 | ||
593 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
594 | } | |
595 | ||
596 | /** | |
597 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes | |
598 | * @sf: seq_file to print to | |
599 | * @v: unused | |
600 | * | |
601 | * To be used as cftype->seq_show to print blkg->stat_bytes. | |
602 | * cftype->private must be set to the blkcg_policy. | |
603 | */ | |
604 | int blkg_print_stat_bytes(struct seq_file *sf, void *v) | |
605 | { | |
606 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
607 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
608 | offsetof(struct blkcg_gq, stat_bytes), true); | |
609 | return 0; | |
610 | } | |
611 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes); | |
612 | ||
613 | /** | |
614 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios | |
615 | * @sf: seq_file to print to | |
616 | * @v: unused | |
617 | * | |
618 | * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private | |
619 | * must be set to the blkcg_policy. | |
620 | */ | |
621 | int blkg_print_stat_ios(struct seq_file *sf, void *v) | |
622 | { | |
623 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
624 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
625 | offsetof(struct blkcg_gq, stat_ios), true); | |
626 | return 0; | |
627 | } | |
628 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios); | |
629 | ||
630 | static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf, | |
631 | struct blkg_policy_data *pd, | |
632 | int off) | |
633 | { | |
634 | struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg, | |
635 | NULL, off); | |
636 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
637 | } | |
638 | ||
639 | /** | |
640 | * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes | |
641 | * @sf: seq_file to print to | |
642 | * @v: unused | |
643 | */ | |
644 | int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v) | |
645 | { | |
646 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
647 | blkg_prfill_rwstat_field_recursive, | |
648 | (void *)seq_cft(sf)->private, | |
649 | offsetof(struct blkcg_gq, stat_bytes), true); | |
650 | return 0; | |
651 | } | |
652 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive); | |
653 | ||
654 | /** | |
655 | * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios | |
656 | * @sf: seq_file to print to | |
657 | * @v: unused | |
658 | */ | |
659 | int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v) | |
660 | { | |
661 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
662 | blkg_prfill_rwstat_field_recursive, | |
663 | (void *)seq_cft(sf)->private, | |
664 | offsetof(struct blkcg_gq, stat_ios), true); | |
665 | return 0; | |
666 | } | |
667 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive); | |
668 | ||
16b3de66 TH |
669 | /** |
670 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
f12c74ca TH |
671 | * @blkg: blkg of interest |
672 | * @pol: blkcg_policy which contains the blkg_stat | |
673 | * @off: offset to the blkg_stat in blkg_policy_data or @blkg | |
16b3de66 | 674 | * |
f12c74ca TH |
675 | * Collect the blkg_stat specified by @blkg, @pol and @off and all its |
676 | * online descendants and their aux counts. The caller must be holding the | |
677 | * queue lock for online tests. | |
678 | * | |
679 | * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is | |
680 | * at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 681 | */ |
f12c74ca TH |
682 | u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg, |
683 | struct blkcg_policy *pol, int off) | |
16b3de66 | 684 | { |
16b3de66 | 685 | struct blkcg_gq *pos_blkg; |
492eb21b | 686 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 687 | u64 sum = 0; |
16b3de66 | 688 | |
0d945c1f | 689 | lockdep_assert_held(&blkg->q->queue_lock); |
16b3de66 | 690 | |
16b3de66 | 691 | rcu_read_lock(); |
f12c74ca TH |
692 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
693 | struct blkg_stat *stat; | |
694 | ||
695 | if (!pos_blkg->online) | |
696 | continue; | |
16b3de66 | 697 | |
f12c74ca TH |
698 | if (pol) |
699 | stat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
700 | else | |
701 | stat = (void *)blkg + off; | |
702 | ||
703 | sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
16b3de66 TH |
704 | } |
705 | rcu_read_unlock(); | |
706 | ||
707 | return sum; | |
708 | } | |
709 | EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum); | |
710 | ||
711 | /** | |
712 | * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat | |
f12c74ca TH |
713 | * @blkg: blkg of interest |
714 | * @pol: blkcg_policy which contains the blkg_rwstat | |
715 | * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg | |
16b3de66 | 716 | * |
f12c74ca TH |
717 | * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its |
718 | * online descendants and their aux counts. The caller must be holding the | |
719 | * queue lock for online tests. | |
720 | * | |
721 | * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it | |
722 | * is at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 723 | */ |
f12c74ca TH |
724 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, |
725 | struct blkcg_policy *pol, int off) | |
16b3de66 | 726 | { |
16b3de66 | 727 | struct blkcg_gq *pos_blkg; |
492eb21b | 728 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 729 | struct blkg_rwstat sum = { }; |
16b3de66 TH |
730 | int i; |
731 | ||
0d945c1f | 732 | lockdep_assert_held(&blkg->q->queue_lock); |
16b3de66 | 733 | |
16b3de66 | 734 | rcu_read_lock(); |
f12c74ca | 735 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
3a7faead | 736 | struct blkg_rwstat *rwstat; |
16b3de66 TH |
737 | |
738 | if (!pos_blkg->online) | |
739 | continue; | |
740 | ||
f12c74ca TH |
741 | if (pol) |
742 | rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
743 | else | |
744 | rwstat = (void *)pos_blkg + off; | |
745 | ||
16b3de66 | 746 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
3a7faead TH |
747 | atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) + |
748 | percpu_counter_sum_positive(&rwstat->cpu_cnt[i]), | |
749 | &sum.aux_cnt[i]); | |
16b3de66 TH |
750 | } |
751 | rcu_read_unlock(); | |
752 | ||
753 | return sum; | |
754 | } | |
755 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
756 | ||
457e490f TE |
757 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ |
758 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, | |
759 | const struct blkcg_policy *pol, | |
760 | struct request_queue *q) | |
761 | { | |
762 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
0d945c1f | 763 | lockdep_assert_held(&q->queue_lock); |
457e490f TE |
764 | |
765 | if (!blkcg_policy_enabled(q, pol)) | |
766 | return ERR_PTR(-EOPNOTSUPP); | |
457e490f TE |
767 | return __blkg_lookup(blkcg, q, true /* update_hint */); |
768 | } | |
769 | ||
3a8b31d3 TH |
770 | /** |
771 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
772 | * @blkcg: target block cgroup | |
da8b0662 | 773 | * @pol: target policy |
3a8b31d3 TH |
774 | * @input: input string |
775 | * @ctx: blkg_conf_ctx to be filled | |
776 | * | |
777 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
778 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
779 | * part of @input following MAJ:MIN. This function returns with RCU read | |
780 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 781 | */ |
3c798398 | 782 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 783 | char *input, struct blkg_conf_ctx *ctx) |
0d945c1f | 784 | __acquires(rcu) __acquires(&disk->queue->queue_lock) |
34d0f179 | 785 | { |
3a8b31d3 | 786 | struct gendisk *disk; |
457e490f | 787 | struct request_queue *q; |
3c798398 | 788 | struct blkcg_gq *blkg; |
726fa694 | 789 | unsigned int major, minor; |
36aa9e5f TH |
790 | int key_len, part, ret; |
791 | char *body; | |
34d0f179 | 792 | |
36aa9e5f | 793 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) |
726fa694 | 794 | return -EINVAL; |
3a8b31d3 | 795 | |
36aa9e5f TH |
796 | body = input + key_len; |
797 | if (!isspace(*body)) | |
798 | return -EINVAL; | |
799 | body = skip_spaces(body); | |
800 | ||
726fa694 | 801 | disk = get_gendisk(MKDEV(major, minor), &part); |
5f6c2d2b | 802 | if (!disk) |
20386ce0 | 803 | return -ENODEV; |
5f6c2d2b | 804 | if (part) { |
457e490f TE |
805 | ret = -ENODEV; |
806 | goto fail; | |
5f6c2d2b | 807 | } |
e56da7e2 | 808 | |
457e490f | 809 | q = disk->queue; |
da8b0662 | 810 | |
457e490f | 811 | rcu_read_lock(); |
0d945c1f | 812 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 813 | |
457e490f | 814 | blkg = blkg_lookup_check(blkcg, pol, q); |
4bfd482e TH |
815 | if (IS_ERR(blkg)) { |
816 | ret = PTR_ERR(blkg); | |
457e490f TE |
817 | goto fail_unlock; |
818 | } | |
819 | ||
820 | if (blkg) | |
821 | goto success; | |
822 | ||
823 | /* | |
824 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
825 | * non-root blkgs have access to their parents. | |
826 | */ | |
827 | while (true) { | |
828 | struct blkcg *pos = blkcg; | |
829 | struct blkcg *parent; | |
830 | struct blkcg_gq *new_blkg; | |
831 | ||
832 | parent = blkcg_parent(blkcg); | |
833 | while (parent && !__blkg_lookup(parent, q, false)) { | |
834 | pos = parent; | |
835 | parent = blkcg_parent(parent); | |
836 | } | |
837 | ||
838 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 839 | spin_unlock_irq(&q->queue_lock); |
3a8b31d3 | 840 | rcu_read_unlock(); |
457e490f TE |
841 | |
842 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); | |
843 | if (unlikely(!new_blkg)) { | |
844 | ret = -ENOMEM; | |
845 | goto fail; | |
7702e8f4 | 846 | } |
3a8b31d3 | 847 | |
457e490f | 848 | rcu_read_lock(); |
0d945c1f | 849 | spin_lock_irq(&q->queue_lock); |
457e490f TE |
850 | |
851 | blkg = blkg_lookup_check(pos, pol, q); | |
852 | if (IS_ERR(blkg)) { | |
853 | ret = PTR_ERR(blkg); | |
854 | goto fail_unlock; | |
855 | } | |
856 | ||
857 | if (blkg) { | |
858 | blkg_free(new_blkg); | |
859 | } else { | |
860 | blkg = blkg_create(pos, q, new_blkg); | |
861 | if (unlikely(IS_ERR(blkg))) { | |
862 | ret = PTR_ERR(blkg); | |
863 | goto fail_unlock; | |
864 | } | |
865 | } | |
866 | ||
867 | if (pos == blkcg) | |
868 | goto success; | |
869 | } | |
870 | success: | |
3a8b31d3 TH |
871 | ctx->disk = disk; |
872 | ctx->blkg = blkg; | |
36aa9e5f | 873 | ctx->body = body; |
726fa694 | 874 | return 0; |
457e490f TE |
875 | |
876 | fail_unlock: | |
0d945c1f | 877 | spin_unlock_irq(&q->queue_lock); |
457e490f TE |
878 | rcu_read_unlock(); |
879 | fail: | |
9df6c299 | 880 | put_disk_and_module(disk); |
457e490f TE |
881 | /* |
882 | * If queue was bypassing, we should retry. Do so after a | |
883 | * short msleep(). It isn't strictly necessary but queue | |
884 | * can be bypassing for some time and it's always nice to | |
885 | * avoid busy looping. | |
886 | */ | |
887 | if (ret == -EBUSY) { | |
888 | msleep(10); | |
889 | ret = restart_syscall(); | |
890 | } | |
891 | return ret; | |
34d0f179 GJ |
892 | } |
893 | ||
3a8b31d3 TH |
894 | /** |
895 | * blkg_conf_finish - finish up per-blkg config update | |
896 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
897 | * | |
898 | * Finish up after per-blkg config update. This function must be paired | |
899 | * with blkg_conf_prep(). | |
900 | */ | |
829fdb50 | 901 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
0d945c1f | 902 | __releases(&ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 903 | { |
0d945c1f | 904 | spin_unlock_irq(&ctx->disk->queue->queue_lock); |
3a8b31d3 | 905 | rcu_read_unlock(); |
9df6c299 | 906 | put_disk_and_module(ctx->disk); |
34d0f179 GJ |
907 | } |
908 | ||
2ee867dc TH |
909 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
910 | { | |
911 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
912 | struct blkcg_gq *blkg; | |
913 | ||
914 | rcu_read_lock(); | |
915 | ||
916 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
917 | const char *dname; | |
903d23f0 | 918 | char *buf; |
2ee867dc | 919 | struct blkg_rwstat rwstat; |
636620b6 | 920 | u64 rbytes, wbytes, rios, wios, dbytes, dios; |
903d23f0 JB |
921 | size_t size = seq_get_buf(sf, &buf), off = 0; |
922 | int i; | |
923 | bool has_stats = false; | |
2ee867dc TH |
924 | |
925 | dname = blkg_dev_name(blkg); | |
926 | if (!dname) | |
927 | continue; | |
928 | ||
903d23f0 JB |
929 | /* |
930 | * Hooray string manipulation, count is the size written NOT | |
931 | * INCLUDING THE \0, so size is now count+1 less than what we | |
932 | * had before, but we want to start writing the next bit from | |
933 | * the \0 so we only add count to buf. | |
934 | */ | |
935 | off += scnprintf(buf+off, size-off, "%s ", dname); | |
936 | ||
0d945c1f | 937 | spin_lock_irq(&blkg->q->queue_lock); |
2ee867dc TH |
938 | |
939 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
940 | offsetof(struct blkcg_gq, stat_bytes)); | |
941 | rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
942 | wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
636620b6 | 943 | dbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_DISCARD]); |
2ee867dc TH |
944 | |
945 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
946 | offsetof(struct blkcg_gq, stat_ios)); | |
947 | rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
948 | wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
636620b6 | 949 | dios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_DISCARD]); |
2ee867dc | 950 | |
0d945c1f | 951 | spin_unlock_irq(&blkg->q->queue_lock); |
2ee867dc | 952 | |
903d23f0 JB |
953 | if (rbytes || wbytes || rios || wios) { |
954 | has_stats = true; | |
955 | off += scnprintf(buf+off, size-off, | |
636620b6 TH |
956 | "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
957 | rbytes, wbytes, rios, wios, | |
958 | dbytes, dios); | |
903d23f0 JB |
959 | } |
960 | ||
961 | if (!blkcg_debug_stats) | |
962 | goto next; | |
963 | ||
d09d8df3 JB |
964 | if (atomic_read(&blkg->use_delay)) { |
965 | has_stats = true; | |
966 | off += scnprintf(buf+off, size-off, | |
967 | " use_delay=%d delay_nsec=%llu", | |
968 | atomic_read(&blkg->use_delay), | |
969 | (unsigned long long)atomic64_read(&blkg->delay_nsec)); | |
970 | } | |
971 | ||
903d23f0 JB |
972 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
973 | struct blkcg_policy *pol = blkcg_policy[i]; | |
974 | size_t written; | |
975 | ||
976 | if (!blkg->pd[i] || !pol->pd_stat_fn) | |
977 | continue; | |
978 | ||
979 | written = pol->pd_stat_fn(blkg->pd[i], buf+off, size-off); | |
980 | if (written) | |
981 | has_stats = true; | |
982 | off += written; | |
983 | } | |
984 | next: | |
985 | if (has_stats) { | |
986 | off += scnprintf(buf+off, size-off, "\n"); | |
987 | seq_commit(sf, off); | |
988 | } | |
2ee867dc TH |
989 | } |
990 | ||
991 | rcu_read_unlock(); | |
992 | return 0; | |
993 | } | |
994 | ||
e1f3b941 | 995 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
996 | { |
997 | .name = "stat", | |
ca0752c5 | 998 | .flags = CFTYPE_NOT_ON_ROOT, |
2ee867dc TH |
999 | .seq_show = blkcg_print_stat, |
1000 | }, | |
1001 | { } /* terminate */ | |
1002 | }; | |
1003 | ||
e1f3b941 | 1004 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
1005 | { |
1006 | .name = "reset_stats", | |
3c798398 | 1007 | .write_u64 = blkcg_reset_stats, |
22084190 | 1008 | }, |
4baf6e33 | 1009 | { } /* terminate */ |
31e4c28d VG |
1010 | }; |
1011 | ||
59b57717 DZF |
1012 | /* |
1013 | * blkcg destruction is a three-stage process. | |
1014 | * | |
1015 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
1016 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
1017 | * to the completion of writeback associated with the blkcg. This lets us | |
1018 | * avoid punting potentially large amounts of outstanding writeback to root | |
1019 | * while maintaining any ongoing policies. The next stage is triggered when | |
1020 | * the nr_cgwbs count goes to zero. | |
1021 | * | |
1022 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
1023 | * and handles the destruction of blkgs. Here the css reference held by | |
1024 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
1025 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
1026 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
1027 | * punted to the root_blkg. | |
1028 | * | |
1029 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
1030 | * This finally frees the blkcg. | |
1031 | */ | |
1032 | ||
9f13ef67 | 1033 | /** |
92fb9748 | 1034 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 1035 | * @css: css of interest |
9f13ef67 | 1036 | * |
59b57717 DZF |
1037 | * This function is called when @css is about to go away. Here the cgwbs are |
1038 | * offlined first and only once writeback associated with the blkcg has | |
1039 | * finished do we start step 2 (see above). | |
9f13ef67 | 1040 | */ |
eb95419b | 1041 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 1042 | { |
eb95419b | 1043 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 1044 | |
59b57717 DZF |
1045 | /* this prevents anyone from attaching or migrating to this blkcg */ |
1046 | wb_blkcg_offline(blkcg); | |
1047 | ||
1048 | /* put the base cgwb reference allowing step 2 to be triggered */ | |
1049 | blkcg_cgwb_put(blkcg); | |
1050 | } | |
1051 | ||
1052 | /** | |
1053 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1054 | * @blkcg: blkcg of interest | |
1055 | * | |
1056 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1057 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1058 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1059 | * blkcg_css_free to eventually be called. | |
1060 | * | |
1061 | * This is the blkcg counterpart of ioc_release_fn(). | |
1062 | */ | |
1063 | void blkcg_destroy_blkgs(struct blkcg *blkcg) | |
1064 | { | |
9f13ef67 | 1065 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1066 | |
4c699480 JQ |
1067 | while (!hlist_empty(&blkcg->blkg_list)) { |
1068 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1069 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1070 | struct request_queue *q = blkg->q; |
1071 | ||
0d945c1f | 1072 | if (spin_trylock(&q->queue_lock)) { |
4c699480 | 1073 | blkg_destroy(blkg); |
0d945c1f | 1074 | spin_unlock(&q->queue_lock); |
4c699480 JQ |
1075 | } else { |
1076 | spin_unlock_irq(&blkcg->lock); | |
1077 | cpu_relax(); | |
1078 | spin_lock_irq(&blkcg->lock); | |
1079 | } | |
1080 | } | |
6b065462 | 1081 | |
4c699480 JQ |
1082 | spin_unlock_irq(&blkcg->lock); |
1083 | } | |
1084 | ||
eb95419b | 1085 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1086 | { |
eb95419b | 1087 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1088 | int i; |
7ee9c562 | 1089 | |
7876f930 | 1090 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1091 | |
7876f930 | 1092 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1093 | |
bc915e61 | 1094 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1095 | if (blkcg->cpd[i]) |
1096 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1097 | ||
1098 | mutex_unlock(&blkcg_pol_mutex); | |
1099 | ||
bc915e61 | 1100 | kfree(blkcg); |
31e4c28d VG |
1101 | } |
1102 | ||
eb95419b TH |
1103 | static struct cgroup_subsys_state * |
1104 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1105 | { |
3c798398 | 1106 | struct blkcg *blkcg; |
e48453c3 AA |
1107 | struct cgroup_subsys_state *ret; |
1108 | int i; | |
31e4c28d | 1109 | |
7876f930 TH |
1110 | mutex_lock(&blkcg_pol_mutex); |
1111 | ||
eb95419b | 1112 | if (!parent_css) { |
3c798398 | 1113 | blkcg = &blkcg_root; |
bc915e61 TH |
1114 | } else { |
1115 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1116 | if (!blkcg) { | |
1117 | ret = ERR_PTR(-ENOMEM); | |
4c18c9e9 | 1118 | goto unlock; |
bc915e61 | 1119 | } |
e48453c3 AA |
1120 | } |
1121 | ||
1122 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1123 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1124 | struct blkcg_policy_data *cpd; | |
1125 | ||
1126 | /* | |
1127 | * If the policy hasn't been attached yet, wait for it | |
1128 | * to be attached before doing anything else. Otherwise, | |
1129 | * check if the policy requires any specific per-cgroup | |
1130 | * data: if it does, allocate and initialize it. | |
1131 | */ | |
e4a9bde9 | 1132 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1133 | continue; |
1134 | ||
e4a9bde9 | 1135 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
1136 | if (!cpd) { |
1137 | ret = ERR_PTR(-ENOMEM); | |
1138 | goto free_pd_blkcg; | |
1139 | } | |
81437648 TH |
1140 | blkcg->cpd[i] = cpd; |
1141 | cpd->blkcg = blkcg; | |
e48453c3 | 1142 | cpd->plid = i; |
e4a9bde9 TH |
1143 | if (pol->cpd_init_fn) |
1144 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1145 | } |
31e4c28d | 1146 | |
31e4c28d | 1147 | spin_lock_init(&blkcg->lock); |
e00f4f4d | 1148 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1149 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1150 | #ifdef CONFIG_CGROUP_WRITEBACK |
1151 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
59b57717 | 1152 | refcount_set(&blkcg->cgwb_refcnt, 1); |
52ebea74 | 1153 | #endif |
7876f930 TH |
1154 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1155 | ||
1156 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1157 | return &blkcg->css; |
e48453c3 AA |
1158 | |
1159 | free_pd_blkcg: | |
1160 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1161 | if (blkcg->cpd[i]) |
1162 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
4c18c9e9 | 1163 | |
1164 | if (blkcg != &blkcg_root) | |
1165 | kfree(blkcg); | |
1166 | unlock: | |
7876f930 | 1167 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 1168 | return ret; |
31e4c28d VG |
1169 | } |
1170 | ||
5efd6113 TH |
1171 | /** |
1172 | * blkcg_init_queue - initialize blkcg part of request queue | |
1173 | * @q: request_queue to initialize | |
1174 | * | |
1175 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1176 | * part of new request_queue @q. | |
1177 | * | |
1178 | * RETURNS: | |
1179 | * 0 on success, -errno on failure. | |
1180 | */ | |
1181 | int blkcg_init_queue(struct request_queue *q) | |
1182 | { | |
d708f0d5 JA |
1183 | struct blkcg_gq *new_blkg, *blkg; |
1184 | bool preloaded; | |
ec13b1d6 TH |
1185 | int ret; |
1186 | ||
d708f0d5 JA |
1187 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
1188 | if (!new_blkg) | |
1189 | return -ENOMEM; | |
1190 | ||
1191 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1192 | ||
bea54883 | 1193 | /* Make sure the root blkg exists. */ |
ec13b1d6 | 1194 | rcu_read_lock(); |
0d945c1f | 1195 | spin_lock_irq(&q->queue_lock); |
d708f0d5 | 1196 | blkg = blkg_create(&blkcg_root, q, new_blkg); |
901932a3 JB |
1197 | if (IS_ERR(blkg)) |
1198 | goto err_unlock; | |
1199 | q->root_blkg = blkg; | |
0d945c1f | 1200 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 TH |
1201 | rcu_read_unlock(); |
1202 | ||
d708f0d5 JA |
1203 | if (preloaded) |
1204 | radix_tree_preload_end(); | |
1205 | ||
d7067512 | 1206 | ret = blk_iolatency_init(q); |
04be60b5 CH |
1207 | if (ret) |
1208 | goto err_destroy_all; | |
d7067512 | 1209 | |
ec13b1d6 | 1210 | ret = blk_throtl_init(q); |
04be60b5 CH |
1211 | if (ret) |
1212 | goto err_destroy_all; | |
1213 | return 0; | |
901932a3 | 1214 | |
04be60b5 | 1215 | err_destroy_all: |
04be60b5 | 1216 | blkg_destroy_all(q); |
04be60b5 | 1217 | return ret; |
901932a3 | 1218 | err_unlock: |
0d945c1f | 1219 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1220 | rcu_read_unlock(); |
1221 | if (preloaded) | |
1222 | radix_tree_preload_end(); | |
1223 | return PTR_ERR(blkg); | |
5efd6113 TH |
1224 | } |
1225 | ||
1226 | /** | |
1227 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1228 | * @q: request_queue to drain | |
1229 | * | |
1230 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1231 | */ | |
1232 | void blkcg_drain_queue(struct request_queue *q) | |
1233 | { | |
0d945c1f | 1234 | lockdep_assert_held(&q->queue_lock); |
5efd6113 | 1235 | |
0b462c89 TH |
1236 | /* |
1237 | * @q could be exiting and already have destroyed all blkgs as | |
1238 | * indicated by NULL root_blkg. If so, don't confuse policies. | |
1239 | */ | |
1240 | if (!q->root_blkg) | |
1241 | return; | |
1242 | ||
5efd6113 TH |
1243 | blk_throtl_drain(q); |
1244 | } | |
1245 | ||
1246 | /** | |
1247 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1248 | * @q: request_queue being released | |
1249 | * | |
1250 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1251 | */ | |
1252 | void blkcg_exit_queue(struct request_queue *q) | |
1253 | { | |
3c96cb32 | 1254 | blkg_destroy_all(q); |
5efd6113 TH |
1255 | blk_throtl_exit(q); |
1256 | } | |
1257 | ||
31e4c28d VG |
1258 | /* |
1259 | * We cannot support shared io contexts, as we have no mean to support | |
1260 | * two tasks with the same ioc in two different groups without major rework | |
1261 | * of the main cic data structures. For now we allow a task to change | |
1262 | * its cgroup only if it's the only owner of its ioc. | |
1263 | */ | |
1f7dd3e5 | 1264 | static int blkcg_can_attach(struct cgroup_taskset *tset) |
31e4c28d | 1265 | { |
bb9d97b6 | 1266 | struct task_struct *task; |
1f7dd3e5 | 1267 | struct cgroup_subsys_state *dst_css; |
31e4c28d VG |
1268 | struct io_context *ioc; |
1269 | int ret = 0; | |
1270 | ||
1271 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
1f7dd3e5 | 1272 | cgroup_taskset_for_each(task, dst_css, tset) { |
bb9d97b6 TH |
1273 | task_lock(task); |
1274 | ioc = task->io_context; | |
1275 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1276 | ret = -EINVAL; | |
1277 | task_unlock(task); | |
1278 | if (ret) | |
1279 | break; | |
1280 | } | |
31e4c28d VG |
1281 | return ret; |
1282 | } | |
1283 | ||
69d7fde5 TH |
1284 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1285 | { | |
1286 | int i; | |
1287 | ||
1288 | mutex_lock(&blkcg_pol_mutex); | |
1289 | ||
1290 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1291 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1292 | struct blkcg *blkcg; | |
1293 | ||
1294 | if (!pol || !pol->cpd_bind_fn) | |
1295 | continue; | |
1296 | ||
1297 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1298 | if (blkcg->cpd[pol->plid]) | |
1299 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1300 | } | |
1301 | mutex_unlock(&blkcg_pol_mutex); | |
1302 | } | |
1303 | ||
d09d8df3 JB |
1304 | static void blkcg_exit(struct task_struct *tsk) |
1305 | { | |
1306 | if (tsk->throttle_queue) | |
1307 | blk_put_queue(tsk->throttle_queue); | |
1308 | tsk->throttle_queue = NULL; | |
1309 | } | |
1310 | ||
c165b3e3 | 1311 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 TH |
1312 | .css_alloc = blkcg_css_alloc, |
1313 | .css_offline = blkcg_css_offline, | |
1314 | .css_free = blkcg_css_free, | |
3c798398 | 1315 | .can_attach = blkcg_can_attach, |
69d7fde5 | 1316 | .bind = blkcg_bind, |
2ee867dc | 1317 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1318 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1319 | .legacy_name = "blkio", |
d09d8df3 | 1320 | .exit = blkcg_exit, |
1ced953b TH |
1321 | #ifdef CONFIG_MEMCG |
1322 | /* | |
1323 | * This ensures that, if available, memcg is automatically enabled | |
1324 | * together on the default hierarchy so that the owner cgroup can | |
1325 | * be retrieved from writeback pages. | |
1326 | */ | |
1327 | .depends_on = 1 << memory_cgrp_id, | |
1328 | #endif | |
676f7c8f | 1329 | }; |
c165b3e3 | 1330 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1331 | |
a2b1693b TH |
1332 | /** |
1333 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1334 | * @q: request_queue of interest | |
1335 | * @pol: blkcg policy to activate | |
1336 | * | |
1337 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1338 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1339 | * | |
1340 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1341 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1342 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1343 | * always enough for dereferencing policy data. | |
1344 | * | |
1345 | * The caller is responsible for synchronizing [de]activations and policy | |
1346 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1347 | */ | |
1348 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1349 | const struct blkcg_policy *pol) |
a2b1693b | 1350 | { |
4c55f4f9 | 1351 | struct blkg_policy_data *pd_prealloc = NULL; |
ec13b1d6 | 1352 | struct blkcg_gq *blkg; |
4c55f4f9 | 1353 | int ret; |
a2b1693b TH |
1354 | |
1355 | if (blkcg_policy_enabled(q, pol)) | |
1356 | return 0; | |
1357 | ||
344e9ffc | 1358 | if (queue_is_mq(q)) |
bd166ef1 | 1359 | blk_mq_freeze_queue(q); |
4c55f4f9 TH |
1360 | pd_prealloc: |
1361 | if (!pd_prealloc) { | |
001bea73 | 1362 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node); |
4c55f4f9 | 1363 | if (!pd_prealloc) { |
a2b1693b | 1364 | ret = -ENOMEM; |
4c55f4f9 | 1365 | goto out_bypass_end; |
a2b1693b | 1366 | } |
a2b1693b TH |
1367 | } |
1368 | ||
0d945c1f | 1369 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1370 | |
1371 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
4c55f4f9 TH |
1372 | struct blkg_policy_data *pd; |
1373 | ||
1374 | if (blkg->pd[pol->plid]) | |
1375 | continue; | |
a2b1693b | 1376 | |
e00f4f4d | 1377 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node); |
4c55f4f9 TH |
1378 | if (!pd) |
1379 | swap(pd, pd_prealloc); | |
1380 | if (!pd) { | |
0d945c1f | 1381 | spin_unlock_irq(&q->queue_lock); |
4c55f4f9 TH |
1382 | goto pd_prealloc; |
1383 | } | |
a2b1693b TH |
1384 | |
1385 | blkg->pd[pol->plid] = pd; | |
1386 | pd->blkg = blkg; | |
b276a876 | 1387 | pd->plid = pol->plid; |
3e418710 | 1388 | if (pol->pd_init_fn) |
a9520cd6 | 1389 | pol->pd_init_fn(pd); |
a2b1693b TH |
1390 | } |
1391 | ||
1392 | __set_bit(pol->plid, q->blkcg_pols); | |
1393 | ret = 0; | |
4c55f4f9 | 1394 | |
0d945c1f | 1395 | spin_unlock_irq(&q->queue_lock); |
4c55f4f9 | 1396 | out_bypass_end: |
344e9ffc | 1397 | if (queue_is_mq(q)) |
bd166ef1 | 1398 | blk_mq_unfreeze_queue(q); |
001bea73 TH |
1399 | if (pd_prealloc) |
1400 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b TH |
1401 | return ret; |
1402 | } | |
1403 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1404 | ||
1405 | /** | |
1406 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1407 | * @q: request_queue of interest | |
1408 | * @pol: blkcg policy to deactivate | |
1409 | * | |
1410 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1411 | * blkcg_activate_policy(). | |
1412 | */ | |
1413 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1414 | const struct blkcg_policy *pol) |
a2b1693b | 1415 | { |
3c798398 | 1416 | struct blkcg_gq *blkg; |
a2b1693b TH |
1417 | |
1418 | if (!blkcg_policy_enabled(q, pol)) | |
1419 | return; | |
1420 | ||
344e9ffc | 1421 | if (queue_is_mq(q)) |
bd166ef1 | 1422 | blk_mq_freeze_queue(q); |
bd166ef1 | 1423 | |
0d945c1f | 1424 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1425 | |
1426 | __clear_bit(pol->plid, q->blkcg_pols); | |
1427 | ||
1428 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
001bea73 | 1429 | if (blkg->pd[pol->plid]) { |
6b065462 | 1430 | if (pol->pd_offline_fn) |
a9520cd6 | 1431 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1432 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1433 | blkg->pd[pol->plid] = NULL; | |
1434 | } | |
a2b1693b TH |
1435 | } |
1436 | ||
0d945c1f | 1437 | spin_unlock_irq(&q->queue_lock); |
bd166ef1 | 1438 | |
344e9ffc | 1439 | if (queue_is_mq(q)) |
bd166ef1 | 1440 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1441 | } |
1442 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1443 | ||
8bd435b3 | 1444 | /** |
3c798398 TH |
1445 | * blkcg_policy_register - register a blkcg policy |
1446 | * @pol: blkcg policy to register | |
8bd435b3 | 1447 | * |
3c798398 TH |
1448 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1449 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1450 | */ |
d5bf0291 | 1451 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1452 | { |
06b285bd | 1453 | struct blkcg *blkcg; |
8bd435b3 | 1454 | int i, ret; |
e8989fae | 1455 | |
838f13bf | 1456 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1457 | mutex_lock(&blkcg_pol_mutex); |
1458 | ||
8bd435b3 TH |
1459 | /* find an empty slot */ |
1460 | ret = -ENOSPC; | |
1461 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1462 | if (!blkcg_policy[i]) |
8bd435b3 | 1463 | break; |
01c5f85a JA |
1464 | if (i >= BLKCG_MAX_POLS) { |
1465 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1466 | goto err_unlock; |
01c5f85a | 1467 | } |
035d10b2 | 1468 | |
e8401073 | 1469 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1470 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1471 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1472 | goto err_unlock; | |
1473 | ||
06b285bd | 1474 | /* register @pol */ |
3c798398 | 1475 | pol->plid = i; |
06b285bd TH |
1476 | blkcg_policy[pol->plid] = pol; |
1477 | ||
1478 | /* allocate and install cpd's */ | |
e4a9bde9 | 1479 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1480 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1481 | struct blkcg_policy_data *cpd; | |
1482 | ||
e4a9bde9 | 1483 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1484 | if (!cpd) |
06b285bd | 1485 | goto err_free_cpds; |
06b285bd | 1486 | |
81437648 TH |
1487 | blkcg->cpd[pol->plid] = cpd; |
1488 | cpd->blkcg = blkcg; | |
06b285bd | 1489 | cpd->plid = pol->plid; |
81437648 | 1490 | pol->cpd_init_fn(cpd); |
06b285bd TH |
1491 | } |
1492 | } | |
1493 | ||
838f13bf | 1494 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1495 | |
8bd435b3 | 1496 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1497 | if (pol->dfl_cftypes) |
1498 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1499 | pol->dfl_cftypes)); | |
880f50e2 | 1500 | if (pol->legacy_cftypes) |
c165b3e3 | 1501 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1502 | pol->legacy_cftypes)); |
838f13bf TH |
1503 | mutex_unlock(&blkcg_pol_register_mutex); |
1504 | return 0; | |
1505 | ||
06b285bd | 1506 | err_free_cpds: |
58a9edce | 1507 | if (pol->cpd_free_fn) { |
06b285bd | 1508 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1509 | if (blkcg->cpd[pol->plid]) { |
1510 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1511 | blkcg->cpd[pol->plid] = NULL; | |
1512 | } | |
06b285bd TH |
1513 | } |
1514 | } | |
1515 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1516 | err_unlock: |
bc0d6501 | 1517 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1518 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1519 | return ret; |
3e252066 | 1520 | } |
3c798398 | 1521 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1522 | |
8bd435b3 | 1523 | /** |
3c798398 TH |
1524 | * blkcg_policy_unregister - unregister a blkcg policy |
1525 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1526 | * |
3c798398 | 1527 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1528 | */ |
3c798398 | 1529 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1530 | { |
06b285bd TH |
1531 | struct blkcg *blkcg; |
1532 | ||
838f13bf | 1533 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1534 | |
3c798398 | 1535 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1536 | goto out_unlock; |
1537 | ||
1538 | /* kill the intf files first */ | |
2ee867dc TH |
1539 | if (pol->dfl_cftypes) |
1540 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1541 | if (pol->legacy_cftypes) |
1542 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1543 | |
06b285bd | 1544 | /* remove cpds and unregister */ |
838f13bf | 1545 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1546 | |
58a9edce | 1547 | if (pol->cpd_free_fn) { |
06b285bd | 1548 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1549 | if (blkcg->cpd[pol->plid]) { |
1550 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1551 | blkcg->cpd[pol->plid] = NULL; | |
1552 | } | |
06b285bd TH |
1553 | } |
1554 | } | |
3c798398 | 1555 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1556 | |
bc0d6501 | 1557 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1558 | out_unlock: |
1559 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1560 | } |
3c798398 | 1561 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1562 | |
d09d8df3 JB |
1563 | /* |
1564 | * Scale the accumulated delay based on how long it has been since we updated | |
1565 | * the delay. We only call this when we are adding delay, in case it's been a | |
1566 | * while since we added delay, and when we are checking to see if we need to | |
1567 | * delay a task, to account for any delays that may have occurred. | |
1568 | */ | |
1569 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1570 | { | |
1571 | u64 old = atomic64_read(&blkg->delay_start); | |
1572 | ||
1573 | /* | |
1574 | * We only want to scale down every second. The idea here is that we | |
1575 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1576 | * time window. We only want to throttle tasks for recent delay that | |
1577 | * has occurred, in 1 second time windows since that's the maximum | |
1578 | * things can be throttled. We save the current delay window in | |
1579 | * blkg->last_delay so we know what amount is still left to be charged | |
1580 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1581 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1582 | * are ok with whatever is happening now, and we can take away more of | |
1583 | * the accumulated delay as we've already throttled enough that | |
1584 | * everybody is happy with their IO latencies. | |
1585 | */ | |
1586 | if (time_before64(old + NSEC_PER_SEC, now) && | |
1587 | atomic64_cmpxchg(&blkg->delay_start, old, now) == old) { | |
1588 | u64 cur = atomic64_read(&blkg->delay_nsec); | |
1589 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1590 | int cur_use = atomic_read(&blkg->use_delay); | |
1591 | ||
1592 | /* | |
1593 | * We've been unthrottled, subtract a larger chunk of our | |
1594 | * accumulated delay. | |
1595 | */ | |
1596 | if (cur_use < blkg->last_use) | |
1597 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1598 | ||
1599 | /* | |
1600 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1601 | * should only ever be growing except here where we subtract out | |
1602 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1603 | * rather not end up with negative numbers. | |
1604 | */ | |
1605 | if (unlikely(cur < sub)) { | |
1606 | atomic64_set(&blkg->delay_nsec, 0); | |
1607 | blkg->last_delay = 0; | |
1608 | } else { | |
1609 | atomic64_sub(sub, &blkg->delay_nsec); | |
1610 | blkg->last_delay = cur - sub; | |
1611 | } | |
1612 | blkg->last_use = cur_use; | |
1613 | } | |
1614 | } | |
1615 | ||
1616 | /* | |
1617 | * This is called when we want to actually walk up the hierarchy and check to | |
1618 | * see if we need to throttle, and then actually throttle if there is some | |
1619 | * accumulated delay. This should only be called upon return to user space so | |
1620 | * we're not holding some lock that would induce a priority inversion. | |
1621 | */ | |
1622 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1623 | { | |
1624 | u64 now = ktime_to_ns(ktime_get()); | |
1625 | u64 exp; | |
1626 | u64 delay_nsec = 0; | |
1627 | int tok; | |
1628 | ||
1629 | while (blkg->parent) { | |
1630 | if (atomic_read(&blkg->use_delay)) { | |
1631 | blkcg_scale_delay(blkg, now); | |
1632 | delay_nsec = max_t(u64, delay_nsec, | |
1633 | atomic64_read(&blkg->delay_nsec)); | |
1634 | } | |
1635 | blkg = blkg->parent; | |
1636 | } | |
1637 | ||
1638 | if (!delay_nsec) | |
1639 | return; | |
1640 | ||
1641 | /* | |
1642 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1643 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1644 | * delay, and we want userspace to be able to do _something_ so cap the | |
1645 | * delays at 1 second. If there's 10's of seconds worth of delay then | |
1646 | * the tasks will be delayed for 1 second for every syscall. | |
1647 | */ | |
1648 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
1649 | ||
1650 | /* | |
1651 | * TODO: the use_memdelay flag is going to be for the upcoming psi stuff | |
1652 | * that hasn't landed upstream yet. Once that stuff is in place we need | |
1653 | * to do a psi_memstall_enter/leave if memdelay is set. | |
1654 | */ | |
1655 | ||
1656 | exp = ktime_add_ns(now, delay_nsec); | |
1657 | tok = io_schedule_prepare(); | |
1658 | do { | |
1659 | __set_current_state(TASK_KILLABLE); | |
1660 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1661 | break; | |
1662 | } while (!fatal_signal_pending(current)); | |
1663 | io_schedule_finish(tok); | |
1664 | } | |
1665 | ||
1666 | /** | |
1667 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1668 | * | |
1669 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1670 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
1671 | * check to see if current->throttle_queue is set and if not this doesn't do | |
1672 | * anything. This should only ever be called by the resume code, it's not meant | |
1673 | * to be called by people willy-nilly as it will actually do the work to | |
1674 | * throttle the task if it is setup for throttling. | |
1675 | */ | |
1676 | void blkcg_maybe_throttle_current(void) | |
1677 | { | |
1678 | struct request_queue *q = current->throttle_queue; | |
1679 | struct cgroup_subsys_state *css; | |
1680 | struct blkcg *blkcg; | |
1681 | struct blkcg_gq *blkg; | |
1682 | bool use_memdelay = current->use_memdelay; | |
1683 | ||
1684 | if (!q) | |
1685 | return; | |
1686 | ||
1687 | current->throttle_queue = NULL; | |
1688 | current->use_memdelay = false; | |
1689 | ||
1690 | rcu_read_lock(); | |
1691 | css = kthread_blkcg(); | |
1692 | if (css) | |
1693 | blkcg = css_to_blkcg(css); | |
1694 | else | |
1695 | blkcg = css_to_blkcg(task_css(current, io_cgrp_id)); | |
1696 | ||
1697 | if (!blkcg) | |
1698 | goto out; | |
1699 | blkg = blkg_lookup(blkcg, q); | |
1700 | if (!blkg) | |
1701 | goto out; | |
b5f2954d DZ |
1702 | blkg = blkg_try_get(blkg); |
1703 | if (!blkg) | |
d09d8df3 JB |
1704 | goto out; |
1705 | rcu_read_unlock(); | |
d09d8df3 JB |
1706 | |
1707 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1708 | blkg_put(blkg); | |
cc7ecc25 | 1709 | blk_put_queue(q); |
d09d8df3 JB |
1710 | return; |
1711 | out: | |
1712 | rcu_read_unlock(); | |
1713 | blk_put_queue(q); | |
1714 | } | |
d09d8df3 JB |
1715 | |
1716 | /** | |
1717 | * blkcg_schedule_throttle - this task needs to check for throttling | |
1718 | * @q - the request queue IO was submitted on | |
1719 | * @use_memdelay - do we charge this to memory delay for PSI | |
1720 | * | |
1721 | * This is called by the IO controller when we know there's delay accumulated | |
1722 | * for the blkg for this task. We do not pass the blkg because there are places | |
1723 | * we call this that may not have that information, the swapping code for | |
1724 | * instance will only have a request_queue at that point. This set's the | |
1725 | * notify_resume for the task to check and see if it requires throttling before | |
1726 | * returning to user space. | |
1727 | * | |
1728 | * We will only schedule once per syscall. You can call this over and over | |
1729 | * again and it will only do the check once upon return to user space, and only | |
1730 | * throttle once. If the task needs to be throttled again it'll need to be | |
1731 | * re-set at the next time we see the task. | |
1732 | */ | |
1733 | void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) | |
1734 | { | |
1735 | if (unlikely(current->flags & PF_KTHREAD)) | |
1736 | return; | |
1737 | ||
1738 | if (!blk_get_queue(q)) | |
1739 | return; | |
1740 | ||
1741 | if (current->throttle_queue) | |
1742 | blk_put_queue(current->throttle_queue); | |
1743 | current->throttle_queue = q; | |
1744 | if (use_memdelay) | |
1745 | current->use_memdelay = use_memdelay; | |
1746 | set_notify_resume(current); | |
1747 | } | |
d09d8df3 JB |
1748 | |
1749 | /** | |
1750 | * blkcg_add_delay - add delay to this blkg | |
1751 | * @now - the current time in nanoseconds | |
1752 | * @delta - how many nanoseconds of delay to add | |
1753 | * | |
1754 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1755 | * throttle tasks if an IO controller thinks we need more throttling. | |
1756 | */ | |
1757 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1758 | { | |
1759 | blkcg_scale_delay(blkg, now); | |
1760 | atomic64_add(delta, &blkg->delay_nsec); | |
1761 | } | |
d09d8df3 | 1762 | |
903d23f0 JB |
1763 | module_param(blkcg_debug_stats, bool, 0644); |
1764 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |