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 | /** |
b5f2954d | 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 | * |
b5f2954d DZ |
261 | * Returns pointer to the looked up or created blkg on success, ERR_PTR() |
262 | * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not | |
263 | * dead and bypassing, returns ERR_PTR(-EBUSY). | |
86cde6b6 | 264 | */ |
b5f2954d DZ |
265 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
266 | struct request_queue *q) | |
3c96cb32 | 267 | { |
86cde6b6 TH |
268 | struct blkcg_gq *blkg; |
269 | ||
270 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
0d945c1f | 271 | lockdep_assert_held(&q->queue_lock); |
86cde6b6 | 272 | |
86cde6b6 TH |
273 | blkg = __blkg_lookup(blkcg, q, true); |
274 | if (blkg) | |
275 | return blkg; | |
276 | ||
3c547865 TH |
277 | /* |
278 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
b5f2954d | 279 | * non-root blkgs have access to their parents. |
3c547865 TH |
280 | */ |
281 | while (true) { | |
282 | struct blkcg *pos = blkcg; | |
283 | struct blkcg *parent = blkcg_parent(blkcg); | |
b5f2954d DZ |
284 | |
285 | while (parent && !__blkg_lookup(parent, q, false)) { | |
3c547865 TH |
286 | pos = parent; |
287 | parent = blkcg_parent(parent); | |
288 | } | |
289 | ||
d708f0d5 | 290 | blkg = blkg_create(pos, q, NULL); |
b5f2954d | 291 | if (pos == blkcg || IS_ERR(blkg)) |
3c547865 TH |
292 | return blkg; |
293 | } | |
3c96cb32 | 294 | } |
31e4c28d | 295 | |
3c798398 | 296 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 297 | { |
3c798398 | 298 | struct blkcg *blkcg = blkg->blkcg; |
77ea7338 | 299 | struct blkcg_gq *parent = blkg->parent; |
6b065462 | 300 | int i; |
03aa264a | 301 | |
0d945c1f | 302 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 303 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
304 | |
305 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 306 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 307 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 308 | |
6b065462 DZF |
309 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
310 | struct blkcg_policy *pol = blkcg_policy[i]; | |
311 | ||
312 | if (blkg->pd[i] && pol->pd_offline_fn) | |
313 | pol->pd_offline_fn(blkg->pd[i]); | |
314 | } | |
315 | ||
77ea7338 TH |
316 | if (parent) { |
317 | blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes); | |
318 | blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios); | |
319 | } | |
320 | ||
f427d909 TH |
321 | blkg->online = false; |
322 | ||
a637120e | 323 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 324 | list_del_init(&blkg->q_node); |
9f13ef67 | 325 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 326 | |
a637120e TH |
327 | /* |
328 | * Both setting lookup hint to and clearing it from @blkg are done | |
329 | * under queue_lock. If it's not pointing to @blkg now, it never | |
330 | * will. Hint assignment itself can race safely. | |
331 | */ | |
ec6c676a | 332 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
333 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
334 | ||
03aa264a TH |
335 | /* |
336 | * Put the reference taken at the time of creation so that when all | |
337 | * queues are gone, group can be destroyed. | |
338 | */ | |
b5f2954d | 339 | blkg_put(blkg); |
03aa264a TH |
340 | } |
341 | ||
9f13ef67 TH |
342 | /** |
343 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
344 | * @q: request_queue of interest | |
9f13ef67 | 345 | * |
3c96cb32 | 346 | * Destroy all blkgs associated with @q. |
9f13ef67 | 347 | */ |
3c96cb32 | 348 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 349 | { |
3c798398 | 350 | struct blkcg_gq *blkg, *n; |
72e06c25 | 351 | |
0d945c1f | 352 | spin_lock_irq(&q->queue_lock); |
9f13ef67 | 353 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 354 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 355 | |
9f13ef67 TH |
356 | spin_lock(&blkcg->lock); |
357 | blkg_destroy(blkg); | |
358 | spin_unlock(&blkcg->lock); | |
72e06c25 | 359 | } |
6fe810bd TH |
360 | |
361 | q->root_blkg = NULL; | |
0d945c1f | 362 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
363 | } |
364 | ||
b5f2954d DZ |
365 | /* |
366 | * A group is RCU protected, but having an rcu lock does not mean that one | |
367 | * can access all the fields of blkg and assume these are valid. For | |
368 | * example, don't try to follow throtl_data and request queue links. | |
369 | * | |
370 | * Having a reference to blkg under an rcu allows accesses to only values | |
371 | * local to groups like group stats and group rate limits. | |
372 | */ | |
373 | void __blkg_release_rcu(struct rcu_head *rcu_head) | |
374 | { | |
375 | struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head); | |
376 | ||
377 | /* release the blkcg and parent blkg refs this blkg has been holding */ | |
378 | css_put(&blkg->blkcg->css); | |
379 | if (blkg->parent) | |
380 | blkg_put(blkg->parent); | |
381 | ||
382 | wb_congested_put(blkg->wb_congested); | |
383 | ||
384 | blkg_free(blkg); | |
385 | } | |
386 | EXPORT_SYMBOL_GPL(__blkg_release_rcu); | |
387 | ||
182446d0 TH |
388 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
389 | struct cftype *cftype, u64 val) | |
303a3acb | 390 | { |
182446d0 | 391 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 392 | struct blkcg_gq *blkg; |
bc0d6501 | 393 | int i; |
303a3acb | 394 | |
838f13bf | 395 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 396 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
397 | |
398 | /* | |
399 | * Note that stat reset is racy - it doesn't synchronize against | |
400 | * stat updates. This is a debug feature which shouldn't exist | |
401 | * anyway. If you get hit by a race, retry. | |
402 | */ | |
b67bfe0d | 403 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
77ea7338 TH |
404 | blkg_rwstat_reset(&blkg->stat_bytes); |
405 | blkg_rwstat_reset(&blkg->stat_ios); | |
406 | ||
8bd435b3 | 407 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 408 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 409 | |
a9520cd6 TH |
410 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
411 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 412 | } |
303a3acb | 413 | } |
f0bdc8cd | 414 | |
303a3acb | 415 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 416 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
417 | return 0; |
418 | } | |
419 | ||
dd165eb3 | 420 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 421 | { |
d3d32e69 | 422 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
dc3b17cc JK |
423 | if (blkg->q->backing_dev_info->dev) |
424 | return dev_name(blkg->q->backing_dev_info->dev); | |
d3d32e69 | 425 | return NULL; |
303a3acb DS |
426 | } |
427 | ||
d3d32e69 TH |
428 | /** |
429 | * blkcg_print_blkgs - helper for printing per-blkg data | |
430 | * @sf: seq_file to print to | |
431 | * @blkcg: blkcg of interest | |
432 | * @prfill: fill function to print out a blkg | |
433 | * @pol: policy in question | |
434 | * @data: data to be passed to @prfill | |
435 | * @show_total: to print out sum of prfill return values or not | |
436 | * | |
437 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
438 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
439 | * policy data and @data and the matching queue lock held. If @show_total |
440 | * is %true, the sum of the return values from @prfill is printed with | |
441 | * "Total" label at the end. | |
d3d32e69 TH |
442 | * |
443 | * This is to be used to construct print functions for | |
444 | * cftype->read_seq_string method. | |
445 | */ | |
3c798398 | 446 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
447 | u64 (*prfill)(struct seq_file *, |
448 | struct blkg_policy_data *, int), | |
3c798398 | 449 | const struct blkcg_policy *pol, int data, |
ec399347 | 450 | bool show_total) |
5624a4e4 | 451 | { |
3c798398 | 452 | struct blkcg_gq *blkg; |
d3d32e69 | 453 | u64 total = 0; |
5624a4e4 | 454 | |
810ecfa7 | 455 | rcu_read_lock(); |
ee89f812 | 456 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 457 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 458 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 459 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 460 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
461 | } |
462 | rcu_read_unlock(); | |
d3d32e69 TH |
463 | |
464 | if (show_total) | |
465 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
466 | } | |
829fdb50 | 467 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
468 | |
469 | /** | |
470 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
471 | * @sf: seq_file to print to | |
f95a04af | 472 | * @pd: policy private data of interest |
d3d32e69 TH |
473 | * @v: value to print |
474 | * | |
f95a04af | 475 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 476 | */ |
f95a04af | 477 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 478 | { |
f95a04af | 479 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
480 | |
481 | if (!dname) | |
482 | return 0; | |
483 | ||
484 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
485 | return v; | |
486 | } | |
829fdb50 | 487 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
488 | |
489 | /** | |
490 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
491 | * @sf: seq_file to print to | |
f95a04af | 492 | * @pd: policy private data of interest |
d3d32e69 TH |
493 | * @rwstat: rwstat to print |
494 | * | |
f95a04af | 495 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 496 | */ |
f95a04af | 497 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 498 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
499 | { |
500 | static const char *rwstr[] = { | |
501 | [BLKG_RWSTAT_READ] = "Read", | |
502 | [BLKG_RWSTAT_WRITE] = "Write", | |
503 | [BLKG_RWSTAT_SYNC] = "Sync", | |
504 | [BLKG_RWSTAT_ASYNC] = "Async", | |
636620b6 | 505 | [BLKG_RWSTAT_DISCARD] = "Discard", |
d3d32e69 | 506 | }; |
f95a04af | 507 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
508 | u64 v; |
509 | int i; | |
510 | ||
511 | if (!dname) | |
512 | return 0; | |
513 | ||
514 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
515 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
24bdb8ef | 516 | (unsigned long long)atomic64_read(&rwstat->aux_cnt[i])); |
d3d32e69 | 517 | |
24bdb8ef | 518 | v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) + |
636620b6 TH |
519 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]) + |
520 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_DISCARD]); | |
d3d32e69 TH |
521 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); |
522 | return v; | |
523 | } | |
b50da39f | 524 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 525 | |
5bc4afb1 TH |
526 | /** |
527 | * blkg_prfill_stat - prfill callback for blkg_stat | |
528 | * @sf: seq_file to print to | |
f95a04af TH |
529 | * @pd: policy private data of interest |
530 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
531 | * |
532 | * prfill callback for printing a blkg_stat. | |
533 | */ | |
f95a04af | 534 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 535 | { |
f95a04af | 536 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 537 | } |
5bc4afb1 | 538 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 539 | |
5bc4afb1 TH |
540 | /** |
541 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
542 | * @sf: seq_file to print to | |
f95a04af TH |
543 | * @pd: policy private data of interest |
544 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
545 | * |
546 | * prfill callback for printing a blkg_rwstat. | |
547 | */ | |
f95a04af TH |
548 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
549 | int off) | |
d3d32e69 | 550 | { |
f95a04af | 551 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 552 | |
f95a04af | 553 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 554 | } |
5bc4afb1 | 555 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 556 | |
77ea7338 TH |
557 | static u64 blkg_prfill_rwstat_field(struct seq_file *sf, |
558 | struct blkg_policy_data *pd, int off) | |
559 | { | |
560 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off); | |
561 | ||
562 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
563 | } | |
564 | ||
565 | /** | |
566 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes | |
567 | * @sf: seq_file to print to | |
568 | * @v: unused | |
569 | * | |
570 | * To be used as cftype->seq_show to print blkg->stat_bytes. | |
571 | * cftype->private must be set to the blkcg_policy. | |
572 | */ | |
573 | int blkg_print_stat_bytes(struct seq_file *sf, void *v) | |
574 | { | |
575 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
576 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
577 | offsetof(struct blkcg_gq, stat_bytes), true); | |
578 | return 0; | |
579 | } | |
580 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes); | |
581 | ||
582 | /** | |
583 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios | |
584 | * @sf: seq_file to print to | |
585 | * @v: unused | |
586 | * | |
587 | * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private | |
588 | * must be set to the blkcg_policy. | |
589 | */ | |
590 | int blkg_print_stat_ios(struct seq_file *sf, void *v) | |
591 | { | |
592 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
593 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
594 | offsetof(struct blkcg_gq, stat_ios), true); | |
595 | return 0; | |
596 | } | |
597 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios); | |
598 | ||
599 | static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf, | |
600 | struct blkg_policy_data *pd, | |
601 | int off) | |
602 | { | |
603 | struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg, | |
604 | NULL, off); | |
605 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
606 | } | |
607 | ||
608 | /** | |
609 | * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes | |
610 | * @sf: seq_file to print to | |
611 | * @v: unused | |
612 | */ | |
613 | int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v) | |
614 | { | |
615 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
616 | blkg_prfill_rwstat_field_recursive, | |
617 | (void *)seq_cft(sf)->private, | |
618 | offsetof(struct blkcg_gq, stat_bytes), true); | |
619 | return 0; | |
620 | } | |
621 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive); | |
622 | ||
623 | /** | |
624 | * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios | |
625 | * @sf: seq_file to print to | |
626 | * @v: unused | |
627 | */ | |
628 | int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v) | |
629 | { | |
630 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
631 | blkg_prfill_rwstat_field_recursive, | |
632 | (void *)seq_cft(sf)->private, | |
633 | offsetof(struct blkcg_gq, stat_ios), true); | |
634 | return 0; | |
635 | } | |
636 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive); | |
637 | ||
16b3de66 TH |
638 | /** |
639 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
f12c74ca TH |
640 | * @blkg: blkg of interest |
641 | * @pol: blkcg_policy which contains the blkg_stat | |
642 | * @off: offset to the blkg_stat in blkg_policy_data or @blkg | |
16b3de66 | 643 | * |
f12c74ca TH |
644 | * Collect the blkg_stat specified by @blkg, @pol and @off and all its |
645 | * online descendants and their aux counts. The caller must be holding the | |
646 | * queue lock for online tests. | |
647 | * | |
648 | * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is | |
649 | * at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 650 | */ |
f12c74ca TH |
651 | u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg, |
652 | struct blkcg_policy *pol, int off) | |
16b3de66 | 653 | { |
16b3de66 | 654 | struct blkcg_gq *pos_blkg; |
492eb21b | 655 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 656 | u64 sum = 0; |
16b3de66 | 657 | |
0d945c1f | 658 | lockdep_assert_held(&blkg->q->queue_lock); |
16b3de66 | 659 | |
16b3de66 | 660 | rcu_read_lock(); |
f12c74ca TH |
661 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
662 | struct blkg_stat *stat; | |
663 | ||
664 | if (!pos_blkg->online) | |
665 | continue; | |
16b3de66 | 666 | |
f12c74ca TH |
667 | if (pol) |
668 | stat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
669 | else | |
670 | stat = (void *)blkg + off; | |
671 | ||
672 | sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
16b3de66 TH |
673 | } |
674 | rcu_read_unlock(); | |
675 | ||
676 | return sum; | |
677 | } | |
678 | EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum); | |
679 | ||
680 | /** | |
681 | * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat | |
f12c74ca TH |
682 | * @blkg: blkg of interest |
683 | * @pol: blkcg_policy which contains the blkg_rwstat | |
684 | * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg | |
16b3de66 | 685 | * |
f12c74ca TH |
686 | * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its |
687 | * online descendants and their aux counts. The caller must be holding the | |
688 | * queue lock for online tests. | |
689 | * | |
690 | * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it | |
691 | * is at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 692 | */ |
f12c74ca TH |
693 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, |
694 | struct blkcg_policy *pol, int off) | |
16b3de66 | 695 | { |
16b3de66 | 696 | struct blkcg_gq *pos_blkg; |
492eb21b | 697 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 698 | struct blkg_rwstat sum = { }; |
16b3de66 TH |
699 | int i; |
700 | ||
0d945c1f | 701 | lockdep_assert_held(&blkg->q->queue_lock); |
16b3de66 | 702 | |
16b3de66 | 703 | rcu_read_lock(); |
f12c74ca | 704 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
3a7faead | 705 | struct blkg_rwstat *rwstat; |
16b3de66 TH |
706 | |
707 | if (!pos_blkg->online) | |
708 | continue; | |
709 | ||
f12c74ca TH |
710 | if (pol) |
711 | rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
712 | else | |
713 | rwstat = (void *)pos_blkg + off; | |
714 | ||
16b3de66 | 715 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
3a7faead TH |
716 | atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) + |
717 | percpu_counter_sum_positive(&rwstat->cpu_cnt[i]), | |
718 | &sum.aux_cnt[i]); | |
16b3de66 TH |
719 | } |
720 | rcu_read_unlock(); | |
721 | ||
722 | return sum; | |
723 | } | |
724 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
725 | ||
457e490f TE |
726 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ |
727 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, | |
728 | const struct blkcg_policy *pol, | |
729 | struct request_queue *q) | |
730 | { | |
731 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
0d945c1f | 732 | lockdep_assert_held(&q->queue_lock); |
457e490f TE |
733 | |
734 | if (!blkcg_policy_enabled(q, pol)) | |
735 | return ERR_PTR(-EOPNOTSUPP); | |
457e490f TE |
736 | return __blkg_lookup(blkcg, q, true /* update_hint */); |
737 | } | |
738 | ||
3a8b31d3 TH |
739 | /** |
740 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
741 | * @blkcg: target block cgroup | |
da8b0662 | 742 | * @pol: target policy |
3a8b31d3 TH |
743 | * @input: input string |
744 | * @ctx: blkg_conf_ctx to be filled | |
745 | * | |
746 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
747 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
748 | * part of @input following MAJ:MIN. This function returns with RCU read | |
749 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 750 | */ |
3c798398 | 751 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 752 | char *input, struct blkg_conf_ctx *ctx) |
0d945c1f | 753 | __acquires(rcu) __acquires(&disk->queue->queue_lock) |
34d0f179 | 754 | { |
3a8b31d3 | 755 | struct gendisk *disk; |
457e490f | 756 | struct request_queue *q; |
3c798398 | 757 | struct blkcg_gq *blkg; |
726fa694 | 758 | unsigned int major, minor; |
36aa9e5f TH |
759 | int key_len, part, ret; |
760 | char *body; | |
34d0f179 | 761 | |
36aa9e5f | 762 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) |
726fa694 | 763 | return -EINVAL; |
3a8b31d3 | 764 | |
36aa9e5f TH |
765 | body = input + key_len; |
766 | if (!isspace(*body)) | |
767 | return -EINVAL; | |
768 | body = skip_spaces(body); | |
769 | ||
726fa694 | 770 | disk = get_gendisk(MKDEV(major, minor), &part); |
5f6c2d2b | 771 | if (!disk) |
20386ce0 | 772 | return -ENODEV; |
5f6c2d2b | 773 | if (part) { |
457e490f TE |
774 | ret = -ENODEV; |
775 | goto fail; | |
5f6c2d2b | 776 | } |
e56da7e2 | 777 | |
457e490f | 778 | q = disk->queue; |
da8b0662 | 779 | |
457e490f | 780 | rcu_read_lock(); |
0d945c1f | 781 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 782 | |
457e490f | 783 | blkg = blkg_lookup_check(blkcg, pol, q); |
4bfd482e TH |
784 | if (IS_ERR(blkg)) { |
785 | ret = PTR_ERR(blkg); | |
457e490f TE |
786 | goto fail_unlock; |
787 | } | |
788 | ||
789 | if (blkg) | |
790 | goto success; | |
791 | ||
792 | /* | |
793 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
794 | * non-root blkgs have access to their parents. | |
795 | */ | |
796 | while (true) { | |
797 | struct blkcg *pos = blkcg; | |
798 | struct blkcg *parent; | |
799 | struct blkcg_gq *new_blkg; | |
800 | ||
801 | parent = blkcg_parent(blkcg); | |
802 | while (parent && !__blkg_lookup(parent, q, false)) { | |
803 | pos = parent; | |
804 | parent = blkcg_parent(parent); | |
805 | } | |
806 | ||
807 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 808 | spin_unlock_irq(&q->queue_lock); |
3a8b31d3 | 809 | rcu_read_unlock(); |
457e490f TE |
810 | |
811 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); | |
812 | if (unlikely(!new_blkg)) { | |
813 | ret = -ENOMEM; | |
814 | goto fail; | |
7702e8f4 | 815 | } |
3a8b31d3 | 816 | |
457e490f | 817 | rcu_read_lock(); |
0d945c1f | 818 | spin_lock_irq(&q->queue_lock); |
457e490f TE |
819 | |
820 | blkg = blkg_lookup_check(pos, pol, q); | |
821 | if (IS_ERR(blkg)) { | |
822 | ret = PTR_ERR(blkg); | |
823 | goto fail_unlock; | |
824 | } | |
825 | ||
826 | if (blkg) { | |
827 | blkg_free(new_blkg); | |
828 | } else { | |
829 | blkg = blkg_create(pos, q, new_blkg); | |
830 | if (unlikely(IS_ERR(blkg))) { | |
831 | ret = PTR_ERR(blkg); | |
832 | goto fail_unlock; | |
833 | } | |
834 | } | |
835 | ||
836 | if (pos == blkcg) | |
837 | goto success; | |
838 | } | |
839 | success: | |
3a8b31d3 TH |
840 | ctx->disk = disk; |
841 | ctx->blkg = blkg; | |
36aa9e5f | 842 | ctx->body = body; |
726fa694 | 843 | return 0; |
457e490f TE |
844 | |
845 | fail_unlock: | |
0d945c1f | 846 | spin_unlock_irq(&q->queue_lock); |
457e490f TE |
847 | rcu_read_unlock(); |
848 | fail: | |
9df6c299 | 849 | put_disk_and_module(disk); |
457e490f TE |
850 | /* |
851 | * If queue was bypassing, we should retry. Do so after a | |
852 | * short msleep(). It isn't strictly necessary but queue | |
853 | * can be bypassing for some time and it's always nice to | |
854 | * avoid busy looping. | |
855 | */ | |
856 | if (ret == -EBUSY) { | |
857 | msleep(10); | |
858 | ret = restart_syscall(); | |
859 | } | |
860 | return ret; | |
34d0f179 GJ |
861 | } |
862 | ||
3a8b31d3 TH |
863 | /** |
864 | * blkg_conf_finish - finish up per-blkg config update | |
865 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
866 | * | |
867 | * Finish up after per-blkg config update. This function must be paired | |
868 | * with blkg_conf_prep(). | |
869 | */ | |
829fdb50 | 870 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
0d945c1f | 871 | __releases(&ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 872 | { |
0d945c1f | 873 | spin_unlock_irq(&ctx->disk->queue->queue_lock); |
3a8b31d3 | 874 | rcu_read_unlock(); |
9df6c299 | 875 | put_disk_and_module(ctx->disk); |
34d0f179 GJ |
876 | } |
877 | ||
2ee867dc TH |
878 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
879 | { | |
880 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
881 | struct blkcg_gq *blkg; | |
882 | ||
883 | rcu_read_lock(); | |
884 | ||
885 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
886 | const char *dname; | |
903d23f0 | 887 | char *buf; |
2ee867dc | 888 | struct blkg_rwstat rwstat; |
636620b6 | 889 | u64 rbytes, wbytes, rios, wios, dbytes, dios; |
903d23f0 JB |
890 | size_t size = seq_get_buf(sf, &buf), off = 0; |
891 | int i; | |
892 | bool has_stats = false; | |
2ee867dc TH |
893 | |
894 | dname = blkg_dev_name(blkg); | |
895 | if (!dname) | |
896 | continue; | |
897 | ||
903d23f0 JB |
898 | /* |
899 | * Hooray string manipulation, count is the size written NOT | |
900 | * INCLUDING THE \0, so size is now count+1 less than what we | |
901 | * had before, but we want to start writing the next bit from | |
902 | * the \0 so we only add count to buf. | |
903 | */ | |
904 | off += scnprintf(buf+off, size-off, "%s ", dname); | |
905 | ||
0d945c1f | 906 | spin_lock_irq(&blkg->q->queue_lock); |
2ee867dc TH |
907 | |
908 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
909 | offsetof(struct blkcg_gq, stat_bytes)); | |
910 | rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
911 | wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
636620b6 | 912 | dbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_DISCARD]); |
2ee867dc TH |
913 | |
914 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
915 | offsetof(struct blkcg_gq, stat_ios)); | |
916 | rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
917 | wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
636620b6 | 918 | dios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_DISCARD]); |
2ee867dc | 919 | |
0d945c1f | 920 | spin_unlock_irq(&blkg->q->queue_lock); |
2ee867dc | 921 | |
903d23f0 JB |
922 | if (rbytes || wbytes || rios || wios) { |
923 | has_stats = true; | |
924 | off += scnprintf(buf+off, size-off, | |
636620b6 TH |
925 | "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
926 | rbytes, wbytes, rios, wios, | |
927 | dbytes, dios); | |
903d23f0 JB |
928 | } |
929 | ||
930 | if (!blkcg_debug_stats) | |
931 | goto next; | |
932 | ||
d09d8df3 JB |
933 | if (atomic_read(&blkg->use_delay)) { |
934 | has_stats = true; | |
935 | off += scnprintf(buf+off, size-off, | |
936 | " use_delay=%d delay_nsec=%llu", | |
937 | atomic_read(&blkg->use_delay), | |
938 | (unsigned long long)atomic64_read(&blkg->delay_nsec)); | |
939 | } | |
940 | ||
903d23f0 JB |
941 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
942 | struct blkcg_policy *pol = blkcg_policy[i]; | |
943 | size_t written; | |
944 | ||
945 | if (!blkg->pd[i] || !pol->pd_stat_fn) | |
946 | continue; | |
947 | ||
948 | written = pol->pd_stat_fn(blkg->pd[i], buf+off, size-off); | |
949 | if (written) | |
950 | has_stats = true; | |
951 | off += written; | |
952 | } | |
953 | next: | |
954 | if (has_stats) { | |
955 | off += scnprintf(buf+off, size-off, "\n"); | |
956 | seq_commit(sf, off); | |
957 | } | |
2ee867dc TH |
958 | } |
959 | ||
960 | rcu_read_unlock(); | |
961 | return 0; | |
962 | } | |
963 | ||
e1f3b941 | 964 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
965 | { |
966 | .name = "stat", | |
ca0752c5 | 967 | .flags = CFTYPE_NOT_ON_ROOT, |
2ee867dc TH |
968 | .seq_show = blkcg_print_stat, |
969 | }, | |
970 | { } /* terminate */ | |
971 | }; | |
972 | ||
e1f3b941 | 973 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
974 | { |
975 | .name = "reset_stats", | |
3c798398 | 976 | .write_u64 = blkcg_reset_stats, |
22084190 | 977 | }, |
4baf6e33 | 978 | { } /* terminate */ |
31e4c28d VG |
979 | }; |
980 | ||
59b57717 DZF |
981 | /* |
982 | * blkcg destruction is a three-stage process. | |
983 | * | |
984 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
985 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
986 | * to the completion of writeback associated with the blkcg. This lets us | |
987 | * avoid punting potentially large amounts of outstanding writeback to root | |
988 | * while maintaining any ongoing policies. The next stage is triggered when | |
989 | * the nr_cgwbs count goes to zero. | |
990 | * | |
991 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
992 | * and handles the destruction of blkgs. Here the css reference held by | |
993 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
994 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
995 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
996 | * punted to the root_blkg. | |
997 | * | |
998 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
999 | * This finally frees the blkcg. | |
1000 | */ | |
1001 | ||
9f13ef67 | 1002 | /** |
92fb9748 | 1003 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 1004 | * @css: css of interest |
9f13ef67 | 1005 | * |
59b57717 DZF |
1006 | * This function is called when @css is about to go away. Here the cgwbs are |
1007 | * offlined first and only once writeback associated with the blkcg has | |
1008 | * finished do we start step 2 (see above). | |
9f13ef67 | 1009 | */ |
eb95419b | 1010 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 1011 | { |
eb95419b | 1012 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 1013 | |
59b57717 DZF |
1014 | /* this prevents anyone from attaching or migrating to this blkcg */ |
1015 | wb_blkcg_offline(blkcg); | |
1016 | ||
1017 | /* put the base cgwb reference allowing step 2 to be triggered */ | |
1018 | blkcg_cgwb_put(blkcg); | |
1019 | } | |
1020 | ||
1021 | /** | |
1022 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1023 | * @blkcg: blkcg of interest | |
1024 | * | |
1025 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1026 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1027 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1028 | * blkcg_css_free to eventually be called. | |
1029 | * | |
1030 | * This is the blkcg counterpart of ioc_release_fn(). | |
1031 | */ | |
1032 | void blkcg_destroy_blkgs(struct blkcg *blkcg) | |
1033 | { | |
9f13ef67 | 1034 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1035 | |
4c699480 JQ |
1036 | while (!hlist_empty(&blkcg->blkg_list)) { |
1037 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1038 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1039 | struct request_queue *q = blkg->q; |
1040 | ||
0d945c1f | 1041 | if (spin_trylock(&q->queue_lock)) { |
4c699480 | 1042 | blkg_destroy(blkg); |
0d945c1f | 1043 | spin_unlock(&q->queue_lock); |
4c699480 JQ |
1044 | } else { |
1045 | spin_unlock_irq(&blkcg->lock); | |
1046 | cpu_relax(); | |
1047 | spin_lock_irq(&blkcg->lock); | |
1048 | } | |
1049 | } | |
6b065462 | 1050 | |
4c699480 JQ |
1051 | spin_unlock_irq(&blkcg->lock); |
1052 | } | |
1053 | ||
eb95419b | 1054 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1055 | { |
eb95419b | 1056 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1057 | int i; |
7ee9c562 | 1058 | |
7876f930 | 1059 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1060 | |
7876f930 | 1061 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1062 | |
bc915e61 | 1063 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1064 | if (blkcg->cpd[i]) |
1065 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1066 | ||
1067 | mutex_unlock(&blkcg_pol_mutex); | |
1068 | ||
bc915e61 | 1069 | kfree(blkcg); |
31e4c28d VG |
1070 | } |
1071 | ||
eb95419b TH |
1072 | static struct cgroup_subsys_state * |
1073 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1074 | { |
3c798398 | 1075 | struct blkcg *blkcg; |
e48453c3 AA |
1076 | struct cgroup_subsys_state *ret; |
1077 | int i; | |
31e4c28d | 1078 | |
7876f930 TH |
1079 | mutex_lock(&blkcg_pol_mutex); |
1080 | ||
eb95419b | 1081 | if (!parent_css) { |
3c798398 | 1082 | blkcg = &blkcg_root; |
bc915e61 TH |
1083 | } else { |
1084 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1085 | if (!blkcg) { | |
1086 | ret = ERR_PTR(-ENOMEM); | |
4c18c9e9 | 1087 | goto unlock; |
bc915e61 | 1088 | } |
e48453c3 AA |
1089 | } |
1090 | ||
1091 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1092 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1093 | struct blkcg_policy_data *cpd; | |
1094 | ||
1095 | /* | |
1096 | * If the policy hasn't been attached yet, wait for it | |
1097 | * to be attached before doing anything else. Otherwise, | |
1098 | * check if the policy requires any specific per-cgroup | |
1099 | * data: if it does, allocate and initialize it. | |
1100 | */ | |
e4a9bde9 | 1101 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1102 | continue; |
1103 | ||
e4a9bde9 | 1104 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
1105 | if (!cpd) { |
1106 | ret = ERR_PTR(-ENOMEM); | |
1107 | goto free_pd_blkcg; | |
1108 | } | |
81437648 TH |
1109 | blkcg->cpd[i] = cpd; |
1110 | cpd->blkcg = blkcg; | |
e48453c3 | 1111 | cpd->plid = i; |
e4a9bde9 TH |
1112 | if (pol->cpd_init_fn) |
1113 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1114 | } |
31e4c28d | 1115 | |
31e4c28d | 1116 | spin_lock_init(&blkcg->lock); |
e00f4f4d | 1117 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1118 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1119 | #ifdef CONFIG_CGROUP_WRITEBACK |
1120 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
59b57717 | 1121 | refcount_set(&blkcg->cgwb_refcnt, 1); |
52ebea74 | 1122 | #endif |
7876f930 TH |
1123 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1124 | ||
1125 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1126 | return &blkcg->css; |
e48453c3 AA |
1127 | |
1128 | free_pd_blkcg: | |
1129 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1130 | if (blkcg->cpd[i]) |
1131 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
4c18c9e9 | 1132 | |
1133 | if (blkcg != &blkcg_root) | |
1134 | kfree(blkcg); | |
1135 | unlock: | |
7876f930 | 1136 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 1137 | return ret; |
31e4c28d VG |
1138 | } |
1139 | ||
5efd6113 TH |
1140 | /** |
1141 | * blkcg_init_queue - initialize blkcg part of request queue | |
1142 | * @q: request_queue to initialize | |
1143 | * | |
1144 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1145 | * part of new request_queue @q. | |
1146 | * | |
1147 | * RETURNS: | |
1148 | * 0 on success, -errno on failure. | |
1149 | */ | |
1150 | int blkcg_init_queue(struct request_queue *q) | |
1151 | { | |
d708f0d5 JA |
1152 | struct blkcg_gq *new_blkg, *blkg; |
1153 | bool preloaded; | |
ec13b1d6 TH |
1154 | int ret; |
1155 | ||
d708f0d5 JA |
1156 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
1157 | if (!new_blkg) | |
1158 | return -ENOMEM; | |
1159 | ||
1160 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1161 | ||
bea54883 | 1162 | /* Make sure the root blkg exists. */ |
ec13b1d6 | 1163 | rcu_read_lock(); |
0d945c1f | 1164 | spin_lock_irq(&q->queue_lock); |
d708f0d5 | 1165 | blkg = blkg_create(&blkcg_root, q, new_blkg); |
901932a3 JB |
1166 | if (IS_ERR(blkg)) |
1167 | goto err_unlock; | |
1168 | q->root_blkg = blkg; | |
0d945c1f | 1169 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 TH |
1170 | rcu_read_unlock(); |
1171 | ||
d708f0d5 JA |
1172 | if (preloaded) |
1173 | radix_tree_preload_end(); | |
1174 | ||
d7067512 | 1175 | ret = blk_iolatency_init(q); |
04be60b5 CH |
1176 | if (ret) |
1177 | goto err_destroy_all; | |
d7067512 | 1178 | |
ec13b1d6 | 1179 | ret = blk_throtl_init(q); |
04be60b5 CH |
1180 | if (ret) |
1181 | goto err_destroy_all; | |
1182 | return 0; | |
901932a3 | 1183 | |
04be60b5 | 1184 | err_destroy_all: |
04be60b5 | 1185 | blkg_destroy_all(q); |
04be60b5 | 1186 | return ret; |
901932a3 | 1187 | err_unlock: |
0d945c1f | 1188 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1189 | rcu_read_unlock(); |
1190 | if (preloaded) | |
1191 | radix_tree_preload_end(); | |
1192 | return PTR_ERR(blkg); | |
5efd6113 TH |
1193 | } |
1194 | ||
1195 | /** | |
1196 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1197 | * @q: request_queue to drain | |
1198 | * | |
1199 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1200 | */ | |
1201 | void blkcg_drain_queue(struct request_queue *q) | |
1202 | { | |
0d945c1f | 1203 | lockdep_assert_held(&q->queue_lock); |
5efd6113 | 1204 | |
0b462c89 TH |
1205 | /* |
1206 | * @q could be exiting and already have destroyed all blkgs as | |
1207 | * indicated by NULL root_blkg. If so, don't confuse policies. | |
1208 | */ | |
1209 | if (!q->root_blkg) | |
1210 | return; | |
1211 | ||
5efd6113 TH |
1212 | blk_throtl_drain(q); |
1213 | } | |
1214 | ||
1215 | /** | |
1216 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1217 | * @q: request_queue being released | |
1218 | * | |
1219 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1220 | */ | |
1221 | void blkcg_exit_queue(struct request_queue *q) | |
1222 | { | |
3c96cb32 | 1223 | blkg_destroy_all(q); |
5efd6113 TH |
1224 | blk_throtl_exit(q); |
1225 | } | |
1226 | ||
31e4c28d VG |
1227 | /* |
1228 | * We cannot support shared io contexts, as we have no mean to support | |
1229 | * two tasks with the same ioc in two different groups without major rework | |
1230 | * of the main cic data structures. For now we allow a task to change | |
1231 | * its cgroup only if it's the only owner of its ioc. | |
1232 | */ | |
1f7dd3e5 | 1233 | static int blkcg_can_attach(struct cgroup_taskset *tset) |
31e4c28d | 1234 | { |
bb9d97b6 | 1235 | struct task_struct *task; |
1f7dd3e5 | 1236 | struct cgroup_subsys_state *dst_css; |
31e4c28d VG |
1237 | struct io_context *ioc; |
1238 | int ret = 0; | |
1239 | ||
1240 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
1f7dd3e5 | 1241 | cgroup_taskset_for_each(task, dst_css, tset) { |
bb9d97b6 TH |
1242 | task_lock(task); |
1243 | ioc = task->io_context; | |
1244 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1245 | ret = -EINVAL; | |
1246 | task_unlock(task); | |
1247 | if (ret) | |
1248 | break; | |
1249 | } | |
31e4c28d VG |
1250 | return ret; |
1251 | } | |
1252 | ||
69d7fde5 TH |
1253 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1254 | { | |
1255 | int i; | |
1256 | ||
1257 | mutex_lock(&blkcg_pol_mutex); | |
1258 | ||
1259 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1260 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1261 | struct blkcg *blkcg; | |
1262 | ||
1263 | if (!pol || !pol->cpd_bind_fn) | |
1264 | continue; | |
1265 | ||
1266 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1267 | if (blkcg->cpd[pol->plid]) | |
1268 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1269 | } | |
1270 | mutex_unlock(&blkcg_pol_mutex); | |
1271 | } | |
1272 | ||
d09d8df3 JB |
1273 | static void blkcg_exit(struct task_struct *tsk) |
1274 | { | |
1275 | if (tsk->throttle_queue) | |
1276 | blk_put_queue(tsk->throttle_queue); | |
1277 | tsk->throttle_queue = NULL; | |
1278 | } | |
1279 | ||
c165b3e3 | 1280 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 TH |
1281 | .css_alloc = blkcg_css_alloc, |
1282 | .css_offline = blkcg_css_offline, | |
1283 | .css_free = blkcg_css_free, | |
3c798398 | 1284 | .can_attach = blkcg_can_attach, |
69d7fde5 | 1285 | .bind = blkcg_bind, |
2ee867dc | 1286 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1287 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1288 | .legacy_name = "blkio", |
d09d8df3 | 1289 | .exit = blkcg_exit, |
1ced953b TH |
1290 | #ifdef CONFIG_MEMCG |
1291 | /* | |
1292 | * This ensures that, if available, memcg is automatically enabled | |
1293 | * together on the default hierarchy so that the owner cgroup can | |
1294 | * be retrieved from writeback pages. | |
1295 | */ | |
1296 | .depends_on = 1 << memory_cgrp_id, | |
1297 | #endif | |
676f7c8f | 1298 | }; |
c165b3e3 | 1299 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1300 | |
a2b1693b TH |
1301 | /** |
1302 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1303 | * @q: request_queue of interest | |
1304 | * @pol: blkcg policy to activate | |
1305 | * | |
1306 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1307 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1308 | * | |
1309 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1310 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1311 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1312 | * always enough for dereferencing policy data. | |
1313 | * | |
1314 | * The caller is responsible for synchronizing [de]activations and policy | |
1315 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1316 | */ | |
1317 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1318 | const struct blkcg_policy *pol) |
a2b1693b | 1319 | { |
4c55f4f9 | 1320 | struct blkg_policy_data *pd_prealloc = NULL; |
ec13b1d6 | 1321 | struct blkcg_gq *blkg; |
4c55f4f9 | 1322 | int ret; |
a2b1693b TH |
1323 | |
1324 | if (blkcg_policy_enabled(q, pol)) | |
1325 | return 0; | |
1326 | ||
38dbb7dd | 1327 | if (q->mq_ops) |
bd166ef1 | 1328 | blk_mq_freeze_queue(q); |
4c55f4f9 TH |
1329 | pd_prealloc: |
1330 | if (!pd_prealloc) { | |
001bea73 | 1331 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node); |
4c55f4f9 | 1332 | if (!pd_prealloc) { |
a2b1693b | 1333 | ret = -ENOMEM; |
4c55f4f9 | 1334 | goto out_bypass_end; |
a2b1693b | 1335 | } |
a2b1693b TH |
1336 | } |
1337 | ||
0d945c1f | 1338 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1339 | |
1340 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
4c55f4f9 TH |
1341 | struct blkg_policy_data *pd; |
1342 | ||
1343 | if (blkg->pd[pol->plid]) | |
1344 | continue; | |
a2b1693b | 1345 | |
e00f4f4d | 1346 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node); |
4c55f4f9 TH |
1347 | if (!pd) |
1348 | swap(pd, pd_prealloc); | |
1349 | if (!pd) { | |
0d945c1f | 1350 | spin_unlock_irq(&q->queue_lock); |
4c55f4f9 TH |
1351 | goto pd_prealloc; |
1352 | } | |
a2b1693b TH |
1353 | |
1354 | blkg->pd[pol->plid] = pd; | |
1355 | pd->blkg = blkg; | |
b276a876 | 1356 | pd->plid = pol->plid; |
3e418710 | 1357 | if (pol->pd_init_fn) |
a9520cd6 | 1358 | pol->pd_init_fn(pd); |
a2b1693b TH |
1359 | } |
1360 | ||
1361 | __set_bit(pol->plid, q->blkcg_pols); | |
1362 | ret = 0; | |
4c55f4f9 | 1363 | |
0d945c1f | 1364 | spin_unlock_irq(&q->queue_lock); |
4c55f4f9 | 1365 | out_bypass_end: |
38dbb7dd | 1366 | if (q->mq_ops) |
bd166ef1 | 1367 | blk_mq_unfreeze_queue(q); |
001bea73 TH |
1368 | if (pd_prealloc) |
1369 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b TH |
1370 | return ret; |
1371 | } | |
1372 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1373 | ||
1374 | /** | |
1375 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1376 | * @q: request_queue of interest | |
1377 | * @pol: blkcg policy to deactivate | |
1378 | * | |
1379 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1380 | * blkcg_activate_policy(). | |
1381 | */ | |
1382 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1383 | const struct blkcg_policy *pol) |
a2b1693b | 1384 | { |
3c798398 | 1385 | struct blkcg_gq *blkg; |
a2b1693b TH |
1386 | |
1387 | if (!blkcg_policy_enabled(q, pol)) | |
1388 | return; | |
1389 | ||
38dbb7dd | 1390 | if (q->mq_ops) |
bd166ef1 | 1391 | blk_mq_freeze_queue(q); |
bd166ef1 | 1392 | |
0d945c1f | 1393 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1394 | |
1395 | __clear_bit(pol->plid, q->blkcg_pols); | |
1396 | ||
1397 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
001bea73 | 1398 | if (blkg->pd[pol->plid]) { |
6b065462 | 1399 | if (pol->pd_offline_fn) |
a9520cd6 | 1400 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1401 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1402 | blkg->pd[pol->plid] = NULL; | |
1403 | } | |
a2b1693b TH |
1404 | } |
1405 | ||
0d945c1f | 1406 | spin_unlock_irq(&q->queue_lock); |
bd166ef1 | 1407 | |
38dbb7dd | 1408 | if (q->mq_ops) |
bd166ef1 | 1409 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1410 | } |
1411 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1412 | ||
8bd435b3 | 1413 | /** |
3c798398 TH |
1414 | * blkcg_policy_register - register a blkcg policy |
1415 | * @pol: blkcg policy to register | |
8bd435b3 | 1416 | * |
3c798398 TH |
1417 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1418 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1419 | */ |
d5bf0291 | 1420 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1421 | { |
06b285bd | 1422 | struct blkcg *blkcg; |
8bd435b3 | 1423 | int i, ret; |
e8989fae | 1424 | |
838f13bf | 1425 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1426 | mutex_lock(&blkcg_pol_mutex); |
1427 | ||
8bd435b3 TH |
1428 | /* find an empty slot */ |
1429 | ret = -ENOSPC; | |
1430 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1431 | if (!blkcg_policy[i]) |
8bd435b3 | 1432 | break; |
01c5f85a JA |
1433 | if (i >= BLKCG_MAX_POLS) { |
1434 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1435 | goto err_unlock; |
01c5f85a | 1436 | } |
035d10b2 | 1437 | |
e8401073 | 1438 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1439 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1440 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1441 | goto err_unlock; | |
1442 | ||
06b285bd | 1443 | /* register @pol */ |
3c798398 | 1444 | pol->plid = i; |
06b285bd TH |
1445 | blkcg_policy[pol->plid] = pol; |
1446 | ||
1447 | /* allocate and install cpd's */ | |
e4a9bde9 | 1448 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1449 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1450 | struct blkcg_policy_data *cpd; | |
1451 | ||
e4a9bde9 | 1452 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1453 | if (!cpd) |
06b285bd | 1454 | goto err_free_cpds; |
06b285bd | 1455 | |
81437648 TH |
1456 | blkcg->cpd[pol->plid] = cpd; |
1457 | cpd->blkcg = blkcg; | |
06b285bd | 1458 | cpd->plid = pol->plid; |
81437648 | 1459 | pol->cpd_init_fn(cpd); |
06b285bd TH |
1460 | } |
1461 | } | |
1462 | ||
838f13bf | 1463 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1464 | |
8bd435b3 | 1465 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1466 | if (pol->dfl_cftypes) |
1467 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1468 | pol->dfl_cftypes)); | |
880f50e2 | 1469 | if (pol->legacy_cftypes) |
c165b3e3 | 1470 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1471 | pol->legacy_cftypes)); |
838f13bf TH |
1472 | mutex_unlock(&blkcg_pol_register_mutex); |
1473 | return 0; | |
1474 | ||
06b285bd | 1475 | err_free_cpds: |
58a9edce | 1476 | if (pol->cpd_free_fn) { |
06b285bd | 1477 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1478 | if (blkcg->cpd[pol->plid]) { |
1479 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1480 | blkcg->cpd[pol->plid] = NULL; | |
1481 | } | |
06b285bd TH |
1482 | } |
1483 | } | |
1484 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1485 | err_unlock: |
bc0d6501 | 1486 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1487 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1488 | return ret; |
3e252066 | 1489 | } |
3c798398 | 1490 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1491 | |
8bd435b3 | 1492 | /** |
3c798398 TH |
1493 | * blkcg_policy_unregister - unregister a blkcg policy |
1494 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1495 | * |
3c798398 | 1496 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1497 | */ |
3c798398 | 1498 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1499 | { |
06b285bd TH |
1500 | struct blkcg *blkcg; |
1501 | ||
838f13bf | 1502 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1503 | |
3c798398 | 1504 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1505 | goto out_unlock; |
1506 | ||
1507 | /* kill the intf files first */ | |
2ee867dc TH |
1508 | if (pol->dfl_cftypes) |
1509 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1510 | if (pol->legacy_cftypes) |
1511 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1512 | |
06b285bd | 1513 | /* remove cpds and unregister */ |
838f13bf | 1514 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1515 | |
58a9edce | 1516 | if (pol->cpd_free_fn) { |
06b285bd | 1517 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1518 | if (blkcg->cpd[pol->plid]) { |
1519 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1520 | blkcg->cpd[pol->plid] = NULL; | |
1521 | } | |
06b285bd TH |
1522 | } |
1523 | } | |
3c798398 | 1524 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1525 | |
bc0d6501 | 1526 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1527 | out_unlock: |
1528 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1529 | } |
3c798398 | 1530 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1531 | |
d09d8df3 JB |
1532 | /* |
1533 | * Scale the accumulated delay based on how long it has been since we updated | |
1534 | * the delay. We only call this when we are adding delay, in case it's been a | |
1535 | * while since we added delay, and when we are checking to see if we need to | |
1536 | * delay a task, to account for any delays that may have occurred. | |
1537 | */ | |
1538 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1539 | { | |
1540 | u64 old = atomic64_read(&blkg->delay_start); | |
1541 | ||
1542 | /* | |
1543 | * We only want to scale down every second. The idea here is that we | |
1544 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1545 | * time window. We only want to throttle tasks for recent delay that | |
1546 | * has occurred, in 1 second time windows since that's the maximum | |
1547 | * things can be throttled. We save the current delay window in | |
1548 | * blkg->last_delay so we know what amount is still left to be charged | |
1549 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1550 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1551 | * are ok with whatever is happening now, and we can take away more of | |
1552 | * the accumulated delay as we've already throttled enough that | |
1553 | * everybody is happy with their IO latencies. | |
1554 | */ | |
1555 | if (time_before64(old + NSEC_PER_SEC, now) && | |
1556 | atomic64_cmpxchg(&blkg->delay_start, old, now) == old) { | |
1557 | u64 cur = atomic64_read(&blkg->delay_nsec); | |
1558 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1559 | int cur_use = atomic_read(&blkg->use_delay); | |
1560 | ||
1561 | /* | |
1562 | * We've been unthrottled, subtract a larger chunk of our | |
1563 | * accumulated delay. | |
1564 | */ | |
1565 | if (cur_use < blkg->last_use) | |
1566 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1567 | ||
1568 | /* | |
1569 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1570 | * should only ever be growing except here where we subtract out | |
1571 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1572 | * rather not end up with negative numbers. | |
1573 | */ | |
1574 | if (unlikely(cur < sub)) { | |
1575 | atomic64_set(&blkg->delay_nsec, 0); | |
1576 | blkg->last_delay = 0; | |
1577 | } else { | |
1578 | atomic64_sub(sub, &blkg->delay_nsec); | |
1579 | blkg->last_delay = cur - sub; | |
1580 | } | |
1581 | blkg->last_use = cur_use; | |
1582 | } | |
1583 | } | |
1584 | ||
1585 | /* | |
1586 | * This is called when we want to actually walk up the hierarchy and check to | |
1587 | * see if we need to throttle, and then actually throttle if there is some | |
1588 | * accumulated delay. This should only be called upon return to user space so | |
1589 | * we're not holding some lock that would induce a priority inversion. | |
1590 | */ | |
1591 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1592 | { | |
1593 | u64 now = ktime_to_ns(ktime_get()); | |
1594 | u64 exp; | |
1595 | u64 delay_nsec = 0; | |
1596 | int tok; | |
1597 | ||
1598 | while (blkg->parent) { | |
1599 | if (atomic_read(&blkg->use_delay)) { | |
1600 | blkcg_scale_delay(blkg, now); | |
1601 | delay_nsec = max_t(u64, delay_nsec, | |
1602 | atomic64_read(&blkg->delay_nsec)); | |
1603 | } | |
1604 | blkg = blkg->parent; | |
1605 | } | |
1606 | ||
1607 | if (!delay_nsec) | |
1608 | return; | |
1609 | ||
1610 | /* | |
1611 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1612 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1613 | * delay, and we want userspace to be able to do _something_ so cap the | |
1614 | * delays at 1 second. If there's 10's of seconds worth of delay then | |
1615 | * the tasks will be delayed for 1 second for every syscall. | |
1616 | */ | |
1617 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
1618 | ||
1619 | /* | |
1620 | * TODO: the use_memdelay flag is going to be for the upcoming psi stuff | |
1621 | * that hasn't landed upstream yet. Once that stuff is in place we need | |
1622 | * to do a psi_memstall_enter/leave if memdelay is set. | |
1623 | */ | |
1624 | ||
1625 | exp = ktime_add_ns(now, delay_nsec); | |
1626 | tok = io_schedule_prepare(); | |
1627 | do { | |
1628 | __set_current_state(TASK_KILLABLE); | |
1629 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1630 | break; | |
1631 | } while (!fatal_signal_pending(current)); | |
1632 | io_schedule_finish(tok); | |
1633 | } | |
1634 | ||
1635 | /** | |
1636 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1637 | * | |
1638 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1639 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
1640 | * check to see if current->throttle_queue is set and if not this doesn't do | |
1641 | * anything. This should only ever be called by the resume code, it's not meant | |
1642 | * to be called by people willy-nilly as it will actually do the work to | |
1643 | * throttle the task if it is setup for throttling. | |
1644 | */ | |
1645 | void blkcg_maybe_throttle_current(void) | |
1646 | { | |
1647 | struct request_queue *q = current->throttle_queue; | |
1648 | struct cgroup_subsys_state *css; | |
1649 | struct blkcg *blkcg; | |
1650 | struct blkcg_gq *blkg; | |
1651 | bool use_memdelay = current->use_memdelay; | |
1652 | ||
1653 | if (!q) | |
1654 | return; | |
1655 | ||
1656 | current->throttle_queue = NULL; | |
1657 | current->use_memdelay = false; | |
1658 | ||
1659 | rcu_read_lock(); | |
1660 | css = kthread_blkcg(); | |
1661 | if (css) | |
1662 | blkcg = css_to_blkcg(css); | |
1663 | else | |
1664 | blkcg = css_to_blkcg(task_css(current, io_cgrp_id)); | |
1665 | ||
1666 | if (!blkcg) | |
1667 | goto out; | |
1668 | blkg = blkg_lookup(blkcg, q); | |
1669 | if (!blkg) | |
1670 | goto out; | |
b5f2954d DZ |
1671 | blkg = blkg_try_get(blkg); |
1672 | if (!blkg) | |
d09d8df3 JB |
1673 | goto out; |
1674 | rcu_read_unlock(); | |
d09d8df3 JB |
1675 | |
1676 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1677 | blkg_put(blkg); | |
cc7ecc25 | 1678 | blk_put_queue(q); |
d09d8df3 JB |
1679 | return; |
1680 | out: | |
1681 | rcu_read_unlock(); | |
1682 | blk_put_queue(q); | |
1683 | } | |
d09d8df3 JB |
1684 | |
1685 | /** | |
1686 | * blkcg_schedule_throttle - this task needs to check for throttling | |
1687 | * @q - the request queue IO was submitted on | |
1688 | * @use_memdelay - do we charge this to memory delay for PSI | |
1689 | * | |
1690 | * This is called by the IO controller when we know there's delay accumulated | |
1691 | * for the blkg for this task. We do not pass the blkg because there are places | |
1692 | * we call this that may not have that information, the swapping code for | |
1693 | * instance will only have a request_queue at that point. This set's the | |
1694 | * notify_resume for the task to check and see if it requires throttling before | |
1695 | * returning to user space. | |
1696 | * | |
1697 | * We will only schedule once per syscall. You can call this over and over | |
1698 | * again and it will only do the check once upon return to user space, and only | |
1699 | * throttle once. If the task needs to be throttled again it'll need to be | |
1700 | * re-set at the next time we see the task. | |
1701 | */ | |
1702 | void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) | |
1703 | { | |
1704 | if (unlikely(current->flags & PF_KTHREAD)) | |
1705 | return; | |
1706 | ||
1707 | if (!blk_get_queue(q)) | |
1708 | return; | |
1709 | ||
1710 | if (current->throttle_queue) | |
1711 | blk_put_queue(current->throttle_queue); | |
1712 | current->throttle_queue = q; | |
1713 | if (use_memdelay) | |
1714 | current->use_memdelay = use_memdelay; | |
1715 | set_notify_resume(current); | |
1716 | } | |
d09d8df3 JB |
1717 | |
1718 | /** | |
1719 | * blkcg_add_delay - add delay to this blkg | |
1720 | * @now - the current time in nanoseconds | |
1721 | * @delta - how many nanoseconds of delay to add | |
1722 | * | |
1723 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1724 | * throttle tasks if an IO controller thinks we need more throttling. | |
1725 | */ | |
1726 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1727 | { | |
1728 | blkcg_scale_delay(blkg, now); | |
1729 | atomic64_add(delta, &blkg->delay_nsec); | |
1730 | } | |
d09d8df3 | 1731 | |
903d23f0 JB |
1732 | module_param(blkcg_debug_stats, bool, 0644); |
1733 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |