]>
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> |
accee785 | 20 | #include <linux/err.h> |
9195291e | 21 | #include <linux/blkdev.h> |
52ebea74 | 22 | #include <linux/backing-dev.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
34d0f179 | 24 | #include <linux/genhd.h> |
72e06c25 | 25 | #include <linux/delay.h> |
9a9e8a26 | 26 | #include <linux/atomic.h> |
eea8f41c | 27 | #include <linux/blk-cgroup.h> |
5efd6113 | 28 | #include "blk.h" |
3e252066 | 29 | |
84c124da DS |
30 | #define MAX_KEY_LEN 100 |
31 | ||
838f13bf TH |
32 | /* |
33 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
34 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
35 | * policy [un]register operations including cgroup file additions / | |
36 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
37 | * allows grabbing it from cgroup callbacks. | |
38 | */ | |
39 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 40 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 41 | |
e48453c3 | 42 | struct blkcg blkcg_root; |
3c798398 | 43 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 44 | |
496d5e75 TH |
45 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
46 | ||
3c798398 | 47 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 48 | |
7876f930 TH |
49 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
50 | ||
a2b1693b | 51 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 52 | const struct blkcg_policy *pol) |
a2b1693b TH |
53 | { |
54 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
55 | } | |
56 | ||
0381411e TH |
57 | /** |
58 | * blkg_free - free a blkg | |
59 | * @blkg: blkg to free | |
60 | * | |
61 | * Free @blkg which may be partially allocated. | |
62 | */ | |
3c798398 | 63 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 64 | { |
e8989fae | 65 | int i; |
549d3aa8 TH |
66 | |
67 | if (!blkg) | |
68 | return; | |
69 | ||
db613670 | 70 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
71 | if (blkg->pd[i]) |
72 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
e8989fae | 73 | |
994b7832 TH |
74 | if (blkg->blkcg != &blkcg_root) |
75 | blk_exit_rl(&blkg->rl); | |
77ea7338 TH |
76 | |
77 | blkg_rwstat_exit(&blkg->stat_ios); | |
78 | blkg_rwstat_exit(&blkg->stat_bytes); | |
549d3aa8 | 79 | kfree(blkg); |
0381411e TH |
80 | } |
81 | ||
82 | /** | |
83 | * blkg_alloc - allocate a blkg | |
84 | * @blkcg: block cgroup the new blkg is associated with | |
85 | * @q: request_queue the new blkg is associated with | |
15974993 | 86 | * @gfp_mask: allocation mask to use |
0381411e | 87 | * |
e8989fae | 88 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 89 | */ |
15974993 TH |
90 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
91 | gfp_t gfp_mask) | |
0381411e | 92 | { |
3c798398 | 93 | struct blkcg_gq *blkg; |
e8989fae | 94 | int i; |
0381411e TH |
95 | |
96 | /* alloc and init base part */ | |
15974993 | 97 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
98 | if (!blkg) |
99 | return NULL; | |
100 | ||
77ea7338 TH |
101 | if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) || |
102 | blkg_rwstat_init(&blkg->stat_ios, gfp_mask)) | |
103 | goto err_free; | |
104 | ||
c875f4d0 | 105 | blkg->q = q; |
e8989fae | 106 | INIT_LIST_HEAD(&blkg->q_node); |
0381411e | 107 | blkg->blkcg = blkcg; |
a5049a8a | 108 | atomic_set(&blkg->refcnt, 1); |
0381411e | 109 | |
a051661c TH |
110 | /* root blkg uses @q->root_rl, init rl only for !root blkgs */ |
111 | if (blkcg != &blkcg_root) { | |
112 | if (blk_init_rl(&blkg->rl, q, gfp_mask)) | |
113 | goto err_free; | |
114 | blkg->rl.blkg = blkg; | |
115 | } | |
116 | ||
8bd435b3 | 117 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 118 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 119 | struct blkg_policy_data *pd; |
0381411e | 120 | |
a2b1693b | 121 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
122 | continue; |
123 | ||
124 | /* alloc per-policy data and attach it to blkg */ | |
001bea73 | 125 | pd = pol->pd_alloc_fn(gfp_mask, q->node); |
a051661c TH |
126 | if (!pd) |
127 | goto err_free; | |
549d3aa8 | 128 | |
e8989fae TH |
129 | blkg->pd[i] = pd; |
130 | pd->blkg = blkg; | |
b276a876 | 131 | pd->plid = i; |
e8989fae TH |
132 | } |
133 | ||
0381411e | 134 | return blkg; |
a051661c TH |
135 | |
136 | err_free: | |
137 | blkg_free(blkg); | |
138 | return NULL; | |
0381411e TH |
139 | } |
140 | ||
24f29046 TH |
141 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
142 | struct request_queue *q, bool update_hint) | |
80fd9979 | 143 | { |
3c798398 | 144 | struct blkcg_gq *blkg; |
80fd9979 | 145 | |
a637120e | 146 | /* |
86cde6b6 TH |
147 | * Hint didn't match. Look up from the radix tree. Note that the |
148 | * hint can only be updated under queue_lock as otherwise @blkg | |
149 | * could have already been removed from blkg_tree. The caller is | |
150 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
151 | */ |
152 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
153 | if (blkg && blkg->q == q) { |
154 | if (update_hint) { | |
155 | lockdep_assert_held(q->queue_lock); | |
156 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
157 | } | |
a637120e | 158 | return blkg; |
86cde6b6 | 159 | } |
a637120e | 160 | |
80fd9979 TH |
161 | return NULL; |
162 | } | |
ae118896 | 163 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); |
80fd9979 | 164 | |
15974993 TH |
165 | /* |
166 | * If @new_blkg is %NULL, this function tries to allocate a new one as | |
d93a11f1 | 167 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. |
15974993 | 168 | */ |
86cde6b6 TH |
169 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
170 | struct request_queue *q, | |
171 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 172 | { |
3c798398 | 173 | struct blkcg_gq *blkg; |
ce7acfea | 174 | struct bdi_writeback_congested *wb_congested; |
f427d909 | 175 | int i, ret; |
5624a4e4 | 176 | |
cd1604fa TH |
177 | WARN_ON_ONCE(!rcu_read_lock_held()); |
178 | lockdep_assert_held(q->queue_lock); | |
179 | ||
7ee9c562 | 180 | /* blkg holds a reference to blkcg */ |
ec903c0c | 181 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 182 | ret = -ENODEV; |
93e6d5d8 | 183 | goto err_free_blkg; |
15974993 | 184 | } |
cd1604fa | 185 | |
ce7acfea | 186 | wb_congested = wb_congested_get_create(&q->backing_dev_info, |
d93a11f1 | 187 | blkcg->css.id, GFP_NOWAIT); |
ce7acfea TH |
188 | if (!wb_congested) { |
189 | ret = -ENOMEM; | |
190 | goto err_put_css; | |
191 | } | |
192 | ||
496fb780 | 193 | /* allocate */ |
15974993 | 194 | if (!new_blkg) { |
d93a11f1 | 195 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT); |
15974993 | 196 | if (unlikely(!new_blkg)) { |
93e6d5d8 | 197 | ret = -ENOMEM; |
ce7acfea | 198 | goto err_put_congested; |
15974993 TH |
199 | } |
200 | } | |
201 | blkg = new_blkg; | |
ce7acfea | 202 | blkg->wb_congested = wb_congested; |
cd1604fa | 203 | |
db613670 | 204 | /* link parent */ |
3c547865 TH |
205 | if (blkcg_parent(blkcg)) { |
206 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
207 | if (WARN_ON_ONCE(!blkg->parent)) { | |
20386ce0 | 208 | ret = -ENODEV; |
ce7acfea | 209 | goto err_put_congested; |
3c547865 TH |
210 | } |
211 | blkg_get(blkg->parent); | |
212 | } | |
213 | ||
db613670 TH |
214 | /* invoke per-policy init */ |
215 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
216 | struct blkcg_policy *pol = blkcg_policy[i]; | |
217 | ||
218 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 219 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
220 | } |
221 | ||
222 | /* insert */ | |
cd1604fa | 223 | spin_lock(&blkcg->lock); |
a637120e TH |
224 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
225 | if (likely(!ret)) { | |
226 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
227 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
228 | |
229 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
230 | struct blkcg_policy *pol = blkcg_policy[i]; | |
231 | ||
232 | if (blkg->pd[i] && pol->pd_online_fn) | |
a9520cd6 | 233 | pol->pd_online_fn(blkg->pd[i]); |
f427d909 | 234 | } |
a637120e | 235 | } |
f427d909 | 236 | blkg->online = true; |
cd1604fa | 237 | spin_unlock(&blkcg->lock); |
496fb780 | 238 | |
ec13b1d6 | 239 | if (!ret) |
a637120e | 240 | return blkg; |
15974993 | 241 | |
3c547865 TH |
242 | /* @blkg failed fully initialized, use the usual release path */ |
243 | blkg_put(blkg); | |
244 | return ERR_PTR(ret); | |
245 | ||
ce7acfea TH |
246 | err_put_congested: |
247 | wb_congested_put(wb_congested); | |
93e6d5d8 | 248 | err_put_css: |
496fb780 | 249 | css_put(&blkcg->css); |
93e6d5d8 | 250 | err_free_blkg: |
15974993 | 251 | blkg_free(new_blkg); |
93e6d5d8 | 252 | return ERR_PTR(ret); |
31e4c28d | 253 | } |
3c96cb32 | 254 | |
86cde6b6 TH |
255 | /** |
256 | * blkg_lookup_create - lookup blkg, try to create one if not there | |
257 | * @blkcg: blkcg of interest | |
258 | * @q: request_queue of interest | |
259 | * | |
260 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
261 | * create one. blkg creation is performed recursively from blkcg_root such |
262 | * that all non-root blkg's have access to the parent blkg. This function | |
263 | * should be called under RCU read lock and @q->queue_lock. | |
86cde6b6 TH |
264 | * |
265 | * Returns pointer to the looked up or created blkg on success, ERR_PTR() | |
266 | * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not | |
267 | * dead and bypassing, returns ERR_PTR(-EBUSY). | |
268 | */ | |
3c798398 TH |
269 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
270 | struct request_queue *q) | |
3c96cb32 | 271 | { |
86cde6b6 TH |
272 | struct blkcg_gq *blkg; |
273 | ||
274 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
275 | lockdep_assert_held(q->queue_lock); | |
276 | ||
3c96cb32 TH |
277 | /* |
278 | * This could be the first entry point of blkcg implementation and | |
279 | * we shouldn't allow anything to go through for a bypassing queue. | |
280 | */ | |
281 | if (unlikely(blk_queue_bypass(q))) | |
20386ce0 | 282 | return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY); |
86cde6b6 TH |
283 | |
284 | blkg = __blkg_lookup(blkcg, q, true); | |
285 | if (blkg) | |
286 | return blkg; | |
287 | ||
3c547865 TH |
288 | /* |
289 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
290 | * non-root blkgs have access to their parents. | |
291 | */ | |
292 | while (true) { | |
293 | struct blkcg *pos = blkcg; | |
294 | struct blkcg *parent = blkcg_parent(blkcg); | |
295 | ||
296 | while (parent && !__blkg_lookup(parent, q, false)) { | |
297 | pos = parent; | |
298 | parent = blkcg_parent(parent); | |
299 | } | |
300 | ||
301 | blkg = blkg_create(pos, q, NULL); | |
302 | if (pos == blkcg || IS_ERR(blkg)) | |
303 | return blkg; | |
304 | } | |
3c96cb32 | 305 | } |
31e4c28d | 306 | |
3c798398 | 307 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 308 | { |
3c798398 | 309 | struct blkcg *blkcg = blkg->blkcg; |
77ea7338 | 310 | struct blkcg_gq *parent = blkg->parent; |
f427d909 | 311 | int i; |
03aa264a | 312 | |
27e1f9d1 | 313 | lockdep_assert_held(blkg->q->queue_lock); |
9f13ef67 | 314 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
315 | |
316 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 317 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 318 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 319 | |
f427d909 TH |
320 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
321 | struct blkcg_policy *pol = blkcg_policy[i]; | |
322 | ||
323 | if (blkg->pd[i] && pol->pd_offline_fn) | |
a9520cd6 | 324 | pol->pd_offline_fn(blkg->pd[i]); |
f427d909 | 325 | } |
77ea7338 TH |
326 | |
327 | if (parent) { | |
328 | blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes); | |
329 | blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios); | |
330 | } | |
331 | ||
f427d909 TH |
332 | blkg->online = false; |
333 | ||
a637120e | 334 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 335 | list_del_init(&blkg->q_node); |
9f13ef67 | 336 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 337 | |
a637120e TH |
338 | /* |
339 | * Both setting lookup hint to and clearing it from @blkg are done | |
340 | * under queue_lock. If it's not pointing to @blkg now, it never | |
341 | * will. Hint assignment itself can race safely. | |
342 | */ | |
ec6c676a | 343 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
344 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
345 | ||
03aa264a TH |
346 | /* |
347 | * Put the reference taken at the time of creation so that when all | |
348 | * queues are gone, group can be destroyed. | |
349 | */ | |
350 | blkg_put(blkg); | |
351 | } | |
352 | ||
9f13ef67 TH |
353 | /** |
354 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
355 | * @q: request_queue of interest | |
9f13ef67 | 356 | * |
3c96cb32 | 357 | * Destroy all blkgs associated with @q. |
9f13ef67 | 358 | */ |
3c96cb32 | 359 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 360 | { |
3c798398 | 361 | struct blkcg_gq *blkg, *n; |
72e06c25 | 362 | |
6d18b008 | 363 | lockdep_assert_held(q->queue_lock); |
72e06c25 | 364 | |
9f13ef67 | 365 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 366 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 367 | |
9f13ef67 TH |
368 | spin_lock(&blkcg->lock); |
369 | blkg_destroy(blkg); | |
370 | spin_unlock(&blkcg->lock); | |
72e06c25 TH |
371 | } |
372 | } | |
373 | ||
2a4fd070 TH |
374 | /* |
375 | * A group is RCU protected, but having an rcu lock does not mean that one | |
376 | * can access all the fields of blkg and assume these are valid. For | |
377 | * example, don't try to follow throtl_data and request queue links. | |
378 | * | |
379 | * Having a reference to blkg under an rcu allows accesses to only values | |
380 | * local to groups like group stats and group rate limits. | |
381 | */ | |
382 | void __blkg_release_rcu(struct rcu_head *rcu_head) | |
1adaf3dd | 383 | { |
2a4fd070 | 384 | struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head); |
db613670 | 385 | |
3c547865 | 386 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
1adaf3dd | 387 | css_put(&blkg->blkcg->css); |
a5049a8a | 388 | if (blkg->parent) |
3c547865 | 389 | blkg_put(blkg->parent); |
1adaf3dd | 390 | |
ce7acfea TH |
391 | wb_congested_put(blkg->wb_congested); |
392 | ||
2a4fd070 | 393 | blkg_free(blkg); |
1adaf3dd | 394 | } |
2a4fd070 | 395 | EXPORT_SYMBOL_GPL(__blkg_release_rcu); |
1adaf3dd | 396 | |
a051661c TH |
397 | /* |
398 | * The next function used by blk_queue_for_each_rl(). It's a bit tricky | |
399 | * because the root blkg uses @q->root_rl instead of its own rl. | |
400 | */ | |
401 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
402 | struct request_queue *q) | |
403 | { | |
404 | struct list_head *ent; | |
405 | struct blkcg_gq *blkg; | |
406 | ||
407 | /* | |
408 | * Determine the current blkg list_head. The first entry is | |
409 | * root_rl which is off @q->blkg_list and mapped to the head. | |
410 | */ | |
411 | if (rl == &q->root_rl) { | |
412 | ent = &q->blkg_list; | |
65c77fd9 JN |
413 | /* There are no more block groups, hence no request lists */ |
414 | if (list_empty(ent)) | |
415 | return NULL; | |
a051661c TH |
416 | } else { |
417 | blkg = container_of(rl, struct blkcg_gq, rl); | |
418 | ent = &blkg->q_node; | |
419 | } | |
420 | ||
421 | /* walk to the next list_head, skip root blkcg */ | |
422 | ent = ent->next; | |
423 | if (ent == &q->root_blkg->q_node) | |
424 | ent = ent->next; | |
425 | if (ent == &q->blkg_list) | |
426 | return NULL; | |
427 | ||
428 | blkg = container_of(ent, struct blkcg_gq, q_node); | |
429 | return &blkg->rl; | |
430 | } | |
431 | ||
182446d0 TH |
432 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
433 | struct cftype *cftype, u64 val) | |
303a3acb | 434 | { |
182446d0 | 435 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 436 | struct blkcg_gq *blkg; |
bc0d6501 | 437 | int i; |
303a3acb | 438 | |
838f13bf | 439 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 440 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
441 | |
442 | /* | |
443 | * Note that stat reset is racy - it doesn't synchronize against | |
444 | * stat updates. This is a debug feature which shouldn't exist | |
445 | * anyway. If you get hit by a race, retry. | |
446 | */ | |
b67bfe0d | 447 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
77ea7338 TH |
448 | blkg_rwstat_reset(&blkg->stat_bytes); |
449 | blkg_rwstat_reset(&blkg->stat_ios); | |
450 | ||
8bd435b3 | 451 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 452 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 453 | |
a9520cd6 TH |
454 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
455 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 456 | } |
303a3acb | 457 | } |
f0bdc8cd | 458 | |
303a3acb | 459 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 460 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
461 | return 0; |
462 | } | |
463 | ||
3c798398 | 464 | static const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 465 | { |
d3d32e69 TH |
466 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
467 | if (blkg->q->backing_dev_info.dev) | |
468 | return dev_name(blkg->q->backing_dev_info.dev); | |
469 | return NULL; | |
303a3acb DS |
470 | } |
471 | ||
d3d32e69 TH |
472 | /** |
473 | * blkcg_print_blkgs - helper for printing per-blkg data | |
474 | * @sf: seq_file to print to | |
475 | * @blkcg: blkcg of interest | |
476 | * @prfill: fill function to print out a blkg | |
477 | * @pol: policy in question | |
478 | * @data: data to be passed to @prfill | |
479 | * @show_total: to print out sum of prfill return values or not | |
480 | * | |
481 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
482 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
483 | * policy data and @data and the matching queue lock held. If @show_total |
484 | * is %true, the sum of the return values from @prfill is printed with | |
485 | * "Total" label at the end. | |
d3d32e69 TH |
486 | * |
487 | * This is to be used to construct print functions for | |
488 | * cftype->read_seq_string method. | |
489 | */ | |
3c798398 | 490 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
491 | u64 (*prfill)(struct seq_file *, |
492 | struct blkg_policy_data *, int), | |
3c798398 | 493 | const struct blkcg_policy *pol, int data, |
ec399347 | 494 | bool show_total) |
5624a4e4 | 495 | { |
3c798398 | 496 | struct blkcg_gq *blkg; |
d3d32e69 | 497 | u64 total = 0; |
5624a4e4 | 498 | |
810ecfa7 | 499 | rcu_read_lock(); |
ee89f812 | 500 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
810ecfa7 | 501 | spin_lock_irq(blkg->q->queue_lock); |
a2b1693b | 502 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 503 | total += prfill(sf, blkg->pd[pol->plid], data); |
810ecfa7 TH |
504 | spin_unlock_irq(blkg->q->queue_lock); |
505 | } | |
506 | rcu_read_unlock(); | |
d3d32e69 TH |
507 | |
508 | if (show_total) | |
509 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
510 | } | |
829fdb50 | 511 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
512 | |
513 | /** | |
514 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
515 | * @sf: seq_file to print to | |
f95a04af | 516 | * @pd: policy private data of interest |
d3d32e69 TH |
517 | * @v: value to print |
518 | * | |
f95a04af | 519 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 520 | */ |
f95a04af | 521 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 522 | { |
f95a04af | 523 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
524 | |
525 | if (!dname) | |
526 | return 0; | |
527 | ||
528 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
529 | return v; | |
530 | } | |
829fdb50 | 531 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 TH |
532 | |
533 | /** | |
534 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
535 | * @sf: seq_file to print to | |
f95a04af | 536 | * @pd: policy private data of interest |
d3d32e69 TH |
537 | * @rwstat: rwstat to print |
538 | * | |
f95a04af | 539 | * Print @rwstat to @sf for the device assocaited with @pd. |
d3d32e69 | 540 | */ |
f95a04af | 541 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
829fdb50 | 542 | const struct blkg_rwstat *rwstat) |
d3d32e69 TH |
543 | { |
544 | static const char *rwstr[] = { | |
545 | [BLKG_RWSTAT_READ] = "Read", | |
546 | [BLKG_RWSTAT_WRITE] = "Write", | |
547 | [BLKG_RWSTAT_SYNC] = "Sync", | |
548 | [BLKG_RWSTAT_ASYNC] = "Async", | |
549 | }; | |
f95a04af | 550 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
551 | u64 v; |
552 | int i; | |
553 | ||
554 | if (!dname) | |
555 | return 0; | |
556 | ||
557 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
558 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
24bdb8ef | 559 | (unsigned long long)atomic64_read(&rwstat->aux_cnt[i])); |
d3d32e69 | 560 | |
24bdb8ef TH |
561 | v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) + |
562 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]); | |
d3d32e69 TH |
563 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); |
564 | return v; | |
565 | } | |
b50da39f | 566 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); |
d3d32e69 | 567 | |
5bc4afb1 TH |
568 | /** |
569 | * blkg_prfill_stat - prfill callback for blkg_stat | |
570 | * @sf: seq_file to print to | |
f95a04af TH |
571 | * @pd: policy private data of interest |
572 | * @off: offset to the blkg_stat in @pd | |
5bc4afb1 TH |
573 | * |
574 | * prfill callback for printing a blkg_stat. | |
575 | */ | |
f95a04af | 576 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) |
d3d32e69 | 577 | { |
f95a04af | 578 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); |
d3d32e69 | 579 | } |
5bc4afb1 | 580 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); |
d3d32e69 | 581 | |
5bc4afb1 TH |
582 | /** |
583 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
584 | * @sf: seq_file to print to | |
f95a04af TH |
585 | * @pd: policy private data of interest |
586 | * @off: offset to the blkg_rwstat in @pd | |
5bc4afb1 TH |
587 | * |
588 | * prfill callback for printing a blkg_rwstat. | |
589 | */ | |
f95a04af TH |
590 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, |
591 | int off) | |
d3d32e69 | 592 | { |
f95a04af | 593 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); |
d3d32e69 | 594 | |
f95a04af | 595 | return __blkg_prfill_rwstat(sf, pd, &rwstat); |
d3d32e69 | 596 | } |
5bc4afb1 | 597 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); |
d3d32e69 | 598 | |
77ea7338 TH |
599 | static u64 blkg_prfill_rwstat_field(struct seq_file *sf, |
600 | struct blkg_policy_data *pd, int off) | |
601 | { | |
602 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off); | |
603 | ||
604 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
605 | } | |
606 | ||
607 | /** | |
608 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes | |
609 | * @sf: seq_file to print to | |
610 | * @v: unused | |
611 | * | |
612 | * To be used as cftype->seq_show to print blkg->stat_bytes. | |
613 | * cftype->private must be set to the blkcg_policy. | |
614 | */ | |
615 | int blkg_print_stat_bytes(struct seq_file *sf, void *v) | |
616 | { | |
617 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
618 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
619 | offsetof(struct blkcg_gq, stat_bytes), true); | |
620 | return 0; | |
621 | } | |
622 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes); | |
623 | ||
624 | /** | |
625 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios | |
626 | * @sf: seq_file to print to | |
627 | * @v: unused | |
628 | * | |
629 | * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private | |
630 | * must be set to the blkcg_policy. | |
631 | */ | |
632 | int blkg_print_stat_ios(struct seq_file *sf, void *v) | |
633 | { | |
634 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
635 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
636 | offsetof(struct blkcg_gq, stat_ios), true); | |
637 | return 0; | |
638 | } | |
639 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios); | |
640 | ||
641 | static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf, | |
642 | struct blkg_policy_data *pd, | |
643 | int off) | |
644 | { | |
645 | struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg, | |
646 | NULL, off); | |
647 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
648 | } | |
649 | ||
650 | /** | |
651 | * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes | |
652 | * @sf: seq_file to print to | |
653 | * @v: unused | |
654 | */ | |
655 | int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v) | |
656 | { | |
657 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
658 | blkg_prfill_rwstat_field_recursive, | |
659 | (void *)seq_cft(sf)->private, | |
660 | offsetof(struct blkcg_gq, stat_bytes), true); | |
661 | return 0; | |
662 | } | |
663 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive); | |
664 | ||
665 | /** | |
666 | * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios | |
667 | * @sf: seq_file to print to | |
668 | * @v: unused | |
669 | */ | |
670 | int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v) | |
671 | { | |
672 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
673 | blkg_prfill_rwstat_field_recursive, | |
674 | (void *)seq_cft(sf)->private, | |
675 | offsetof(struct blkcg_gq, stat_ios), true); | |
676 | return 0; | |
677 | } | |
678 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive); | |
679 | ||
16b3de66 TH |
680 | /** |
681 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
f12c74ca TH |
682 | * @blkg: blkg of interest |
683 | * @pol: blkcg_policy which contains the blkg_stat | |
684 | * @off: offset to the blkg_stat in blkg_policy_data or @blkg | |
16b3de66 | 685 | * |
f12c74ca TH |
686 | * Collect the blkg_stat 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_stat is at @off bytes into @blkg; otherwise, it is | |
691 | * at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 692 | */ |
f12c74ca TH |
693 | u64 blkg_stat_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 | u64 sum = 0; |
16b3de66 | 699 | |
f12c74ca | 700 | lockdep_assert_held(blkg->q->queue_lock); |
16b3de66 | 701 | |
16b3de66 | 702 | rcu_read_lock(); |
f12c74ca TH |
703 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
704 | struct blkg_stat *stat; | |
705 | ||
706 | if (!pos_blkg->online) | |
707 | continue; | |
16b3de66 | 708 | |
f12c74ca TH |
709 | if (pol) |
710 | stat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
711 | else | |
712 | stat = (void *)blkg + off; | |
713 | ||
714 | sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
16b3de66 TH |
715 | } |
716 | rcu_read_unlock(); | |
717 | ||
718 | return sum; | |
719 | } | |
720 | EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum); | |
721 | ||
722 | /** | |
723 | * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat | |
f12c74ca TH |
724 | * @blkg: blkg of interest |
725 | * @pol: blkcg_policy which contains the blkg_rwstat | |
726 | * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg | |
16b3de66 | 727 | * |
f12c74ca TH |
728 | * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its |
729 | * online descendants and their aux counts. The caller must be holding the | |
730 | * queue lock for online tests. | |
731 | * | |
732 | * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it | |
733 | * is at @off bytes into @blkg's blkg_policy_data of the policy. | |
16b3de66 | 734 | */ |
f12c74ca TH |
735 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, |
736 | struct blkcg_policy *pol, int off) | |
16b3de66 | 737 | { |
16b3de66 | 738 | struct blkcg_gq *pos_blkg; |
492eb21b | 739 | struct cgroup_subsys_state *pos_css; |
bd8815a6 | 740 | struct blkg_rwstat sum = { }; |
16b3de66 TH |
741 | int i; |
742 | ||
f12c74ca | 743 | lockdep_assert_held(blkg->q->queue_lock); |
16b3de66 | 744 | |
16b3de66 | 745 | rcu_read_lock(); |
f12c74ca | 746 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { |
3a7faead | 747 | struct blkg_rwstat *rwstat; |
16b3de66 TH |
748 | |
749 | if (!pos_blkg->online) | |
750 | continue; | |
751 | ||
f12c74ca TH |
752 | if (pol) |
753 | rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
754 | else | |
755 | rwstat = (void *)pos_blkg + off; | |
756 | ||
16b3de66 | 757 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
3a7faead TH |
758 | atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) + |
759 | percpu_counter_sum_positive(&rwstat->cpu_cnt[i]), | |
760 | &sum.aux_cnt[i]); | |
16b3de66 TH |
761 | } |
762 | rcu_read_unlock(); | |
763 | ||
764 | return sum; | |
765 | } | |
766 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
767 | ||
3a8b31d3 TH |
768 | /** |
769 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
770 | * @blkcg: target block cgroup | |
da8b0662 | 771 | * @pol: target policy |
3a8b31d3 TH |
772 | * @input: input string |
773 | * @ctx: blkg_conf_ctx to be filled | |
774 | * | |
775 | * Parse per-blkg config update from @input and initialize @ctx with the | |
776 | * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new | |
da8b0662 TH |
777 | * value. This function returns with RCU read lock and queue lock held and |
778 | * must be paired with blkg_conf_finish(). | |
3a8b31d3 | 779 | */ |
3c798398 TH |
780 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
781 | const char *input, struct blkg_conf_ctx *ctx) | |
da8b0662 | 782 | __acquires(rcu) __acquires(disk->queue->queue_lock) |
34d0f179 | 783 | { |
3a8b31d3 | 784 | struct gendisk *disk; |
3c798398 | 785 | struct blkcg_gq *blkg; |
726fa694 TH |
786 | unsigned int major, minor; |
787 | unsigned long long v; | |
788 | int part, ret; | |
34d0f179 | 789 | |
726fa694 TH |
790 | if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3) |
791 | return -EINVAL; | |
3a8b31d3 | 792 | |
726fa694 | 793 | disk = get_gendisk(MKDEV(major, minor), &part); |
5f6c2d2b | 794 | if (!disk) |
20386ce0 | 795 | return -ENODEV; |
5f6c2d2b TH |
796 | if (part) { |
797 | put_disk(disk); | |
20386ce0 | 798 | return -ENODEV; |
5f6c2d2b | 799 | } |
e56da7e2 TH |
800 | |
801 | rcu_read_lock(); | |
4bfd482e | 802 | spin_lock_irq(disk->queue->queue_lock); |
da8b0662 | 803 | |
a2b1693b | 804 | if (blkcg_policy_enabled(disk->queue, pol)) |
3c96cb32 | 805 | blkg = blkg_lookup_create(blkcg, disk->queue); |
a2b1693b | 806 | else |
20386ce0 | 807 | blkg = ERR_PTR(-EOPNOTSUPP); |
e56da7e2 | 808 | |
4bfd482e TH |
809 | if (IS_ERR(blkg)) { |
810 | ret = PTR_ERR(blkg); | |
3a8b31d3 | 811 | rcu_read_unlock(); |
da8b0662 | 812 | spin_unlock_irq(disk->queue->queue_lock); |
3a8b31d3 TH |
813 | put_disk(disk); |
814 | /* | |
815 | * If queue was bypassing, we should retry. Do so after a | |
816 | * short msleep(). It isn't strictly necessary but queue | |
817 | * can be bypassing for some time and it's always nice to | |
818 | * avoid busy looping. | |
819 | */ | |
820 | if (ret == -EBUSY) { | |
821 | msleep(10); | |
822 | ret = restart_syscall(); | |
7702e8f4 | 823 | } |
726fa694 | 824 | return ret; |
062a644d | 825 | } |
3a8b31d3 TH |
826 | |
827 | ctx->disk = disk; | |
828 | ctx->blkg = blkg; | |
726fa694 TH |
829 | ctx->v = v; |
830 | return 0; | |
34d0f179 | 831 | } |
829fdb50 | 832 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 833 | |
3a8b31d3 TH |
834 | /** |
835 | * blkg_conf_finish - finish up per-blkg config update | |
836 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
837 | * | |
838 | * Finish up after per-blkg config update. This function must be paired | |
839 | * with blkg_conf_prep(). | |
840 | */ | |
829fdb50 | 841 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
da8b0662 | 842 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 843 | { |
da8b0662 | 844 | spin_unlock_irq(ctx->disk->queue->queue_lock); |
3a8b31d3 TH |
845 | rcu_read_unlock(); |
846 | put_disk(ctx->disk); | |
34d0f179 | 847 | } |
829fdb50 | 848 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 849 | |
3c798398 | 850 | struct cftype blkcg_files[] = { |
84c124da DS |
851 | { |
852 | .name = "reset_stats", | |
3c798398 | 853 | .write_u64 = blkcg_reset_stats, |
22084190 | 854 | }, |
4baf6e33 | 855 | { } /* terminate */ |
31e4c28d VG |
856 | }; |
857 | ||
9f13ef67 | 858 | /** |
92fb9748 | 859 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 860 | * @css: css of interest |
9f13ef67 | 861 | * |
eb95419b TH |
862 | * This function is called when @css is about to go away and responsible |
863 | * for shooting down all blkgs associated with @css. blkgs should be | |
9f13ef67 TH |
864 | * removed while holding both q and blkcg locks. As blkcg lock is nested |
865 | * inside q lock, this function performs reverse double lock dancing. | |
866 | * | |
867 | * This is the blkcg counterpart of ioc_release_fn(). | |
868 | */ | |
eb95419b | 869 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 870 | { |
eb95419b | 871 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 872 | |
9f13ef67 | 873 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 874 | |
9f13ef67 | 875 | while (!hlist_empty(&blkcg->blkg_list)) { |
3c798398 TH |
876 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, |
877 | struct blkcg_gq, blkcg_node); | |
c875f4d0 | 878 | struct request_queue *q = blkg->q; |
b1c35769 | 879 | |
9f13ef67 TH |
880 | if (spin_trylock(q->queue_lock)) { |
881 | blkg_destroy(blkg); | |
882 | spin_unlock(q->queue_lock); | |
883 | } else { | |
884 | spin_unlock_irq(&blkcg->lock); | |
9f13ef67 | 885 | cpu_relax(); |
a5567932 | 886 | spin_lock_irq(&blkcg->lock); |
0f3942a3 | 887 | } |
9f13ef67 | 888 | } |
b1c35769 | 889 | |
9f13ef67 | 890 | spin_unlock_irq(&blkcg->lock); |
52ebea74 TH |
891 | |
892 | wb_blkcg_offline(blkcg); | |
7ee9c562 TH |
893 | } |
894 | ||
eb95419b | 895 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 896 | { |
eb95419b | 897 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 898 | int i; |
7ee9c562 | 899 | |
7876f930 | 900 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 901 | |
7876f930 | 902 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 903 | |
bc915e61 | 904 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
905 | if (blkcg->cpd[i]) |
906 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
907 | ||
908 | mutex_unlock(&blkcg_pol_mutex); | |
909 | ||
bc915e61 | 910 | kfree(blkcg); |
31e4c28d VG |
911 | } |
912 | ||
eb95419b TH |
913 | static struct cgroup_subsys_state * |
914 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 915 | { |
3c798398 | 916 | struct blkcg *blkcg; |
e48453c3 AA |
917 | struct cgroup_subsys_state *ret; |
918 | int i; | |
31e4c28d | 919 | |
7876f930 TH |
920 | mutex_lock(&blkcg_pol_mutex); |
921 | ||
eb95419b | 922 | if (!parent_css) { |
3c798398 | 923 | blkcg = &blkcg_root; |
bc915e61 TH |
924 | } else { |
925 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
926 | if (!blkcg) { | |
927 | ret = ERR_PTR(-ENOMEM); | |
928 | goto free_blkcg; | |
929 | } | |
e48453c3 AA |
930 | } |
931 | ||
932 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
933 | struct blkcg_policy *pol = blkcg_policy[i]; | |
934 | struct blkcg_policy_data *cpd; | |
935 | ||
936 | /* | |
937 | * If the policy hasn't been attached yet, wait for it | |
938 | * to be attached before doing anything else. Otherwise, | |
939 | * check if the policy requires any specific per-cgroup | |
940 | * data: if it does, allocate and initialize it. | |
941 | */ | |
e4a9bde9 | 942 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
943 | continue; |
944 | ||
e4a9bde9 | 945 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
946 | if (!cpd) { |
947 | ret = ERR_PTR(-ENOMEM); | |
948 | goto free_pd_blkcg; | |
949 | } | |
81437648 TH |
950 | blkcg->cpd[i] = cpd; |
951 | cpd->blkcg = blkcg; | |
e48453c3 | 952 | cpd->plid = i; |
e4a9bde9 TH |
953 | if (pol->cpd_init_fn) |
954 | pol->cpd_init_fn(cpd); | |
e48453c3 | 955 | } |
31e4c28d | 956 | |
31e4c28d | 957 | spin_lock_init(&blkcg->lock); |
d93a11f1 | 958 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT); |
31e4c28d | 959 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
960 | #ifdef CONFIG_CGROUP_WRITEBACK |
961 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
962 | #endif | |
7876f930 TH |
963 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
964 | ||
965 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 966 | return &blkcg->css; |
e48453c3 AA |
967 | |
968 | free_pd_blkcg: | |
969 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
970 | if (blkcg->cpd[i]) |
971 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
e48453c3 AA |
972 | free_blkcg: |
973 | kfree(blkcg); | |
7876f930 | 974 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 975 | return ret; |
31e4c28d VG |
976 | } |
977 | ||
5efd6113 TH |
978 | /** |
979 | * blkcg_init_queue - initialize blkcg part of request queue | |
980 | * @q: request_queue to initialize | |
981 | * | |
982 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
983 | * part of new request_queue @q. | |
984 | * | |
985 | * RETURNS: | |
986 | * 0 on success, -errno on failure. | |
987 | */ | |
988 | int blkcg_init_queue(struct request_queue *q) | |
989 | { | |
ec13b1d6 TH |
990 | struct blkcg_gq *new_blkg, *blkg; |
991 | bool preloaded; | |
992 | int ret; | |
993 | ||
994 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); | |
995 | if (!new_blkg) | |
996 | return -ENOMEM; | |
997 | ||
998 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
5efd6113 | 999 | |
ec13b1d6 TH |
1000 | /* |
1001 | * Make sure the root blkg exists and count the existing blkgs. As | |
1002 | * @q is bypassing at this point, blkg_lookup_create() can't be | |
1003 | * used. Open code insertion. | |
1004 | */ | |
1005 | rcu_read_lock(); | |
1006 | spin_lock_irq(q->queue_lock); | |
1007 | blkg = blkg_create(&blkcg_root, q, new_blkg); | |
1008 | spin_unlock_irq(q->queue_lock); | |
1009 | rcu_read_unlock(); | |
1010 | ||
1011 | if (preloaded) | |
1012 | radix_tree_preload_end(); | |
1013 | ||
1014 | if (IS_ERR(blkg)) { | |
994b7832 | 1015 | blkg_free(new_blkg); |
ec13b1d6 TH |
1016 | return PTR_ERR(blkg); |
1017 | } | |
1018 | ||
1019 | q->root_blkg = blkg; | |
1020 | q->root_rl.blkg = blkg; | |
5efd6113 | 1021 | |
ec13b1d6 TH |
1022 | ret = blk_throtl_init(q); |
1023 | if (ret) { | |
1024 | spin_lock_irq(q->queue_lock); | |
1025 | blkg_destroy_all(q); | |
1026 | spin_unlock_irq(q->queue_lock); | |
1027 | } | |
1028 | return ret; | |
5efd6113 TH |
1029 | } |
1030 | ||
1031 | /** | |
1032 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1033 | * @q: request_queue to drain | |
1034 | * | |
1035 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1036 | */ | |
1037 | void blkcg_drain_queue(struct request_queue *q) | |
1038 | { | |
1039 | lockdep_assert_held(q->queue_lock); | |
1040 | ||
0b462c89 TH |
1041 | /* |
1042 | * @q could be exiting and already have destroyed all blkgs as | |
1043 | * indicated by NULL root_blkg. If so, don't confuse policies. | |
1044 | */ | |
1045 | if (!q->root_blkg) | |
1046 | return; | |
1047 | ||
5efd6113 TH |
1048 | blk_throtl_drain(q); |
1049 | } | |
1050 | ||
1051 | /** | |
1052 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1053 | * @q: request_queue being released | |
1054 | * | |
1055 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1056 | */ | |
1057 | void blkcg_exit_queue(struct request_queue *q) | |
1058 | { | |
6d18b008 | 1059 | spin_lock_irq(q->queue_lock); |
3c96cb32 | 1060 | blkg_destroy_all(q); |
6d18b008 TH |
1061 | spin_unlock_irq(q->queue_lock); |
1062 | ||
5efd6113 TH |
1063 | blk_throtl_exit(q); |
1064 | } | |
1065 | ||
31e4c28d VG |
1066 | /* |
1067 | * We cannot support shared io contexts, as we have no mean to support | |
1068 | * two tasks with the same ioc in two different groups without major rework | |
1069 | * of the main cic data structures. For now we allow a task to change | |
1070 | * its cgroup only if it's the only owner of its ioc. | |
1071 | */ | |
eb95419b TH |
1072 | static int blkcg_can_attach(struct cgroup_subsys_state *css, |
1073 | struct cgroup_taskset *tset) | |
31e4c28d | 1074 | { |
bb9d97b6 | 1075 | struct task_struct *task; |
31e4c28d VG |
1076 | struct io_context *ioc; |
1077 | int ret = 0; | |
1078 | ||
1079 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
924f0d9a | 1080 | cgroup_taskset_for_each(task, tset) { |
bb9d97b6 TH |
1081 | task_lock(task); |
1082 | ioc = task->io_context; | |
1083 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1084 | ret = -EINVAL; | |
1085 | task_unlock(task); | |
1086 | if (ret) | |
1087 | break; | |
1088 | } | |
31e4c28d VG |
1089 | return ret; |
1090 | } | |
1091 | ||
073219e9 | 1092 | struct cgroup_subsys blkio_cgrp_subsys = { |
92fb9748 TH |
1093 | .css_alloc = blkcg_css_alloc, |
1094 | .css_offline = blkcg_css_offline, | |
1095 | .css_free = blkcg_css_free, | |
3c798398 | 1096 | .can_attach = blkcg_can_attach, |
5577964e | 1097 | .legacy_cftypes = blkcg_files, |
1ced953b TH |
1098 | #ifdef CONFIG_MEMCG |
1099 | /* | |
1100 | * This ensures that, if available, memcg is automatically enabled | |
1101 | * together on the default hierarchy so that the owner cgroup can | |
1102 | * be retrieved from writeback pages. | |
1103 | */ | |
1104 | .depends_on = 1 << memory_cgrp_id, | |
1105 | #endif | |
676f7c8f | 1106 | }; |
073219e9 | 1107 | EXPORT_SYMBOL_GPL(blkio_cgrp_subsys); |
676f7c8f | 1108 | |
a2b1693b TH |
1109 | /** |
1110 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1111 | * @q: request_queue of interest | |
1112 | * @pol: blkcg policy to activate | |
1113 | * | |
1114 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1115 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1116 | * | |
1117 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1118 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1119 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1120 | * always enough for dereferencing policy data. | |
1121 | * | |
1122 | * The caller is responsible for synchronizing [de]activations and policy | |
1123 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1124 | */ | |
1125 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1126 | const struct blkcg_policy *pol) |
a2b1693b | 1127 | { |
4c55f4f9 | 1128 | struct blkg_policy_data *pd_prealloc = NULL; |
ec13b1d6 | 1129 | struct blkcg_gq *blkg; |
4c55f4f9 | 1130 | int ret; |
a2b1693b TH |
1131 | |
1132 | if (blkcg_policy_enabled(q, pol)) | |
1133 | return 0; | |
1134 | ||
1135 | blk_queue_bypass_start(q); | |
4c55f4f9 TH |
1136 | pd_prealloc: |
1137 | if (!pd_prealloc) { | |
001bea73 | 1138 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node); |
4c55f4f9 | 1139 | if (!pd_prealloc) { |
a2b1693b | 1140 | ret = -ENOMEM; |
4c55f4f9 | 1141 | goto out_bypass_end; |
a2b1693b | 1142 | } |
a2b1693b TH |
1143 | } |
1144 | ||
a2b1693b TH |
1145 | spin_lock_irq(q->queue_lock); |
1146 | ||
1147 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
4c55f4f9 TH |
1148 | struct blkg_policy_data *pd; |
1149 | ||
1150 | if (blkg->pd[pol->plid]) | |
1151 | continue; | |
a2b1693b | 1152 | |
001bea73 | 1153 | pd = pol->pd_alloc_fn(GFP_NOWAIT, q->node); |
4c55f4f9 TH |
1154 | if (!pd) |
1155 | swap(pd, pd_prealloc); | |
1156 | if (!pd) { | |
1157 | spin_unlock_irq(q->queue_lock); | |
1158 | goto pd_prealloc; | |
1159 | } | |
a2b1693b TH |
1160 | |
1161 | blkg->pd[pol->plid] = pd; | |
1162 | pd->blkg = blkg; | |
b276a876 | 1163 | pd->plid = pol->plid; |
3e418710 | 1164 | if (pol->pd_init_fn) |
a9520cd6 | 1165 | pol->pd_init_fn(pd); |
a2b1693b TH |
1166 | } |
1167 | ||
1168 | __set_bit(pol->plid, q->blkcg_pols); | |
1169 | ret = 0; | |
4c55f4f9 | 1170 | |
a2b1693b | 1171 | spin_unlock_irq(q->queue_lock); |
4c55f4f9 | 1172 | out_bypass_end: |
a2b1693b | 1173 | blk_queue_bypass_end(q); |
001bea73 TH |
1174 | if (pd_prealloc) |
1175 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b TH |
1176 | return ret; |
1177 | } | |
1178 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1179 | ||
1180 | /** | |
1181 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1182 | * @q: request_queue of interest | |
1183 | * @pol: blkcg policy to deactivate | |
1184 | * | |
1185 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1186 | * blkcg_activate_policy(). | |
1187 | */ | |
1188 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1189 | const struct blkcg_policy *pol) |
a2b1693b | 1190 | { |
3c798398 | 1191 | struct blkcg_gq *blkg; |
a2b1693b TH |
1192 | |
1193 | if (!blkcg_policy_enabled(q, pol)) | |
1194 | return; | |
1195 | ||
1196 | blk_queue_bypass_start(q); | |
1197 | spin_lock_irq(q->queue_lock); | |
1198 | ||
1199 | __clear_bit(pol->plid, q->blkcg_pols); | |
1200 | ||
1201 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1202 | /* grab blkcg lock too while removing @pd from @blkg */ | |
1203 | spin_lock(&blkg->blkcg->lock); | |
1204 | ||
001bea73 | 1205 | if (blkg->pd[pol->plid]) { |
a9520cd6 TH |
1206 | if (pol->pd_offline_fn) |
1207 | pol->pd_offline_fn(blkg->pd[pol->plid]); | |
001bea73 TH |
1208 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1209 | blkg->pd[pol->plid] = NULL; | |
1210 | } | |
a2b1693b TH |
1211 | |
1212 | spin_unlock(&blkg->blkcg->lock); | |
1213 | } | |
1214 | ||
1215 | spin_unlock_irq(q->queue_lock); | |
1216 | blk_queue_bypass_end(q); | |
1217 | } | |
1218 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1219 | ||
8bd435b3 | 1220 | /** |
3c798398 TH |
1221 | * blkcg_policy_register - register a blkcg policy |
1222 | * @pol: blkcg policy to register | |
8bd435b3 | 1223 | * |
3c798398 TH |
1224 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1225 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1226 | */ |
d5bf0291 | 1227 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1228 | { |
06b285bd | 1229 | struct blkcg *blkcg; |
8bd435b3 | 1230 | int i, ret; |
e8989fae | 1231 | |
838f13bf | 1232 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1233 | mutex_lock(&blkcg_pol_mutex); |
1234 | ||
8bd435b3 TH |
1235 | /* find an empty slot */ |
1236 | ret = -ENOSPC; | |
1237 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1238 | if (!blkcg_policy[i]) |
8bd435b3 TH |
1239 | break; |
1240 | if (i >= BLKCG_MAX_POLS) | |
838f13bf | 1241 | goto err_unlock; |
035d10b2 | 1242 | |
06b285bd | 1243 | /* register @pol */ |
3c798398 | 1244 | pol->plid = i; |
06b285bd TH |
1245 | blkcg_policy[pol->plid] = pol; |
1246 | ||
1247 | /* allocate and install cpd's */ | |
e4a9bde9 | 1248 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1249 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1250 | struct blkcg_policy_data *cpd; | |
1251 | ||
e4a9bde9 | 1252 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
06b285bd TH |
1253 | if (!cpd) { |
1254 | mutex_unlock(&blkcg_pol_mutex); | |
1255 | goto err_free_cpds; | |
1256 | } | |
1257 | ||
81437648 TH |
1258 | blkcg->cpd[pol->plid] = cpd; |
1259 | cpd->blkcg = blkcg; | |
06b285bd | 1260 | cpd->plid = pol->plid; |
81437648 | 1261 | pol->cpd_init_fn(cpd); |
06b285bd TH |
1262 | } |
1263 | } | |
1264 | ||
838f13bf | 1265 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1266 | |
8bd435b3 | 1267 | /* everything is in place, add intf files for the new policy */ |
3c798398 | 1268 | if (pol->cftypes) |
2cf669a5 TH |
1269 | WARN_ON(cgroup_add_legacy_cftypes(&blkio_cgrp_subsys, |
1270 | pol->cftypes)); | |
838f13bf TH |
1271 | mutex_unlock(&blkcg_pol_register_mutex); |
1272 | return 0; | |
1273 | ||
06b285bd | 1274 | err_free_cpds: |
e4a9bde9 | 1275 | if (pol->cpd_alloc_fn) { |
06b285bd | 1276 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1277 | if (blkcg->cpd[pol->plid]) { |
1278 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1279 | blkcg->cpd[pol->plid] = NULL; | |
1280 | } | |
06b285bd TH |
1281 | } |
1282 | } | |
1283 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1284 | err_unlock: |
bc0d6501 | 1285 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1286 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1287 | return ret; |
3e252066 | 1288 | } |
3c798398 | 1289 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1290 | |
8bd435b3 | 1291 | /** |
3c798398 TH |
1292 | * blkcg_policy_unregister - unregister a blkcg policy |
1293 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1294 | * |
3c798398 | 1295 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1296 | */ |
3c798398 | 1297 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1298 | { |
06b285bd TH |
1299 | struct blkcg *blkcg; |
1300 | ||
838f13bf | 1301 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1302 | |
3c798398 | 1303 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1304 | goto out_unlock; |
1305 | ||
1306 | /* kill the intf files first */ | |
3c798398 | 1307 | if (pol->cftypes) |
2bb566cb | 1308 | cgroup_rm_cftypes(pol->cftypes); |
44ea53de | 1309 | |
06b285bd | 1310 | /* remove cpds and unregister */ |
838f13bf | 1311 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1312 | |
e4a9bde9 | 1313 | if (pol->cpd_alloc_fn) { |
06b285bd | 1314 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1315 | if (blkcg->cpd[pol->plid]) { |
1316 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1317 | blkcg->cpd[pol->plid] = NULL; | |
1318 | } | |
06b285bd TH |
1319 | } |
1320 | } | |
3c798398 | 1321 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1322 | |
bc0d6501 | 1323 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1324 | out_unlock: |
1325 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1326 | } |
3c798398 | 1327 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |