]>
Commit | Line | Data |
---|---|---|
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> | |
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> | |
16 | */ | |
17 | #include <linux/ioprio.h> | |
18 | #include <linux/kdev_t.h> | |
19 | #include <linux/module.h> | |
20 | #include <linux/sched/signal.h> | |
21 | #include <linux/err.h> | |
22 | #include <linux/blkdev.h> | |
23 | #include <linux/backing-dev.h> | |
24 | #include <linux/slab.h> | |
25 | #include <linux/genhd.h> | |
26 | #include <linux/delay.h> | |
27 | #include <linux/atomic.h> | |
28 | #include <linux/ctype.h> | |
29 | #include <linux/blk-cgroup.h> | |
30 | #include "blk.h" | |
31 | ||
32 | #define MAX_KEY_LEN 100 | |
33 | ||
34 | /* | |
35 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
36 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
37 | * policy [un]register operations including cgroup file additions / | |
38 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
39 | * allows grabbing it from cgroup callbacks. | |
40 | */ | |
41 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
42 | static DEFINE_MUTEX(blkcg_pol_mutex); | |
43 | ||
44 | struct blkcg blkcg_root; | |
45 | EXPORT_SYMBOL_GPL(blkcg_root); | |
46 | ||
47 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; | |
48 | ||
49 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; | |
50 | ||
51 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ | |
52 | ||
53 | static bool blkcg_policy_enabled(struct request_queue *q, | |
54 | const struct blkcg_policy *pol) | |
55 | { | |
56 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
57 | } | |
58 | ||
59 | /** | |
60 | * blkg_free - free a blkg | |
61 | * @blkg: blkg to free | |
62 | * | |
63 | * Free @blkg which may be partially allocated. | |
64 | */ | |
65 | static void blkg_free(struct blkcg_gq *blkg) | |
66 | { | |
67 | int i; | |
68 | ||
69 | if (!blkg) | |
70 | return; | |
71 | ||
72 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
73 | if (blkg->pd[i]) | |
74 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
75 | ||
76 | if (blkg->blkcg != &blkcg_root) | |
77 | blk_exit_rl(blkg->q, &blkg->rl); | |
78 | ||
79 | blkg_rwstat_exit(&blkg->stat_ios); | |
80 | blkg_rwstat_exit(&blkg->stat_bytes); | |
81 | kfree(blkg); | |
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 | |
88 | * @gfp_mask: allocation mask to use | |
89 | * | |
90 | * Allocate a new blkg assocating @blkcg and @q. | |
91 | */ | |
92 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, | |
93 | gfp_t gfp_mask) | |
94 | { | |
95 | struct blkcg_gq *blkg; | |
96 | int i; | |
97 | ||
98 | /* alloc and init base part */ | |
99 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); | |
100 | if (!blkg) | |
101 | return NULL; | |
102 | ||
103 | if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) || | |
104 | blkg_rwstat_init(&blkg->stat_ios, gfp_mask)) | |
105 | goto err_free; | |
106 | ||
107 | blkg->q = q; | |
108 | INIT_LIST_HEAD(&blkg->q_node); | |
109 | blkg->blkcg = blkcg; | |
110 | atomic_set(&blkg->refcnt, 1); | |
111 | ||
112 | /* root blkg uses @q->root_rl, init rl only for !root blkgs */ | |
113 | if (blkcg != &blkcg_root) { | |
114 | if (blk_init_rl(&blkg->rl, q, gfp_mask)) | |
115 | goto err_free; | |
116 | blkg->rl.blkg = blkg; | |
117 | } | |
118 | ||
119 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
120 | struct blkcg_policy *pol = blkcg_policy[i]; | |
121 | struct blkg_policy_data *pd; | |
122 | ||
123 | if (!blkcg_policy_enabled(q, pol)) | |
124 | continue; | |
125 | ||
126 | /* alloc per-policy data and attach it to blkg */ | |
127 | pd = pol->pd_alloc_fn(gfp_mask, q->node); | |
128 | if (!pd) | |
129 | goto err_free; | |
130 | ||
131 | blkg->pd[i] = pd; | |
132 | pd->blkg = blkg; | |
133 | pd->plid = i; | |
134 | } | |
135 | ||
136 | return blkg; | |
137 | ||
138 | err_free: | |
139 | blkg_free(blkg); | |
140 | return NULL; | |
141 | } | |
142 | ||
143 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, | |
144 | struct request_queue *q, bool update_hint) | |
145 | { | |
146 | struct blkcg_gq *blkg; | |
147 | ||
148 | /* | |
149 | * Hint didn't match. Look up from the radix tree. Note that the | |
150 | * hint can only be updated under queue_lock as otherwise @blkg | |
151 | * could have already been removed from blkg_tree. The caller is | |
152 | * responsible for grabbing queue_lock if @update_hint. | |
153 | */ | |
154 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
155 | if (blkg && blkg->q == q) { | |
156 | if (update_hint) { | |
157 | lockdep_assert_held(q->queue_lock); | |
158 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
159 | } | |
160 | return blkg; | |
161 | } | |
162 | ||
163 | return NULL; | |
164 | } | |
165 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); | |
166 | ||
167 | /* | |
168 | * If @new_blkg is %NULL, this function tries to allocate a new one as | |
169 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
170 | */ | |
171 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, | |
172 | struct request_queue *q, | |
173 | struct blkcg_gq *new_blkg) | |
174 | { | |
175 | struct blkcg_gq *blkg; | |
176 | struct bdi_writeback_congested *wb_congested; | |
177 | int i, ret; | |
178 | ||
179 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
180 | lockdep_assert_held(q->queue_lock); | |
181 | ||
182 | /* blkg holds a reference to blkcg */ | |
183 | if (!css_tryget_online(&blkcg->css)) { | |
184 | ret = -ENODEV; | |
185 | goto err_free_blkg; | |
186 | } | |
187 | ||
188 | wb_congested = wb_congested_get_create(q->backing_dev_info, | |
189 | blkcg->css.id, | |
190 | GFP_NOWAIT | __GFP_NOWARN); | |
191 | if (!wb_congested) { | |
192 | ret = -ENOMEM; | |
193 | goto err_put_css; | |
194 | } | |
195 | ||
196 | /* allocate */ | |
197 | if (!new_blkg) { | |
198 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN); | |
199 | if (unlikely(!new_blkg)) { | |
200 | ret = -ENOMEM; | |
201 | goto err_put_congested; | |
202 | } | |
203 | } | |
204 | blkg = new_blkg; | |
205 | blkg->wb_congested = wb_congested; | |
206 | ||
207 | /* link parent */ | |
208 | if (blkcg_parent(blkcg)) { | |
209 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
210 | if (WARN_ON_ONCE(!blkg->parent)) { | |
211 | ret = -ENODEV; | |
212 | goto err_put_congested; | |
213 | } | |
214 | blkg_get(blkg->parent); | |
215 | } | |
216 | ||
217 | /* invoke per-policy init */ | |
218 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
219 | struct blkcg_policy *pol = blkcg_policy[i]; | |
220 | ||
221 | if (blkg->pd[i] && pol->pd_init_fn) | |
222 | pol->pd_init_fn(blkg->pd[i]); | |
223 | } | |
224 | ||
225 | /* insert */ | |
226 | spin_lock(&blkcg->lock); | |
227 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); | |
228 | if (likely(!ret)) { | |
229 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
230 | list_add(&blkg->q_node, &q->blkg_list); | |
231 | ||
232 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
233 | struct blkcg_policy *pol = blkcg_policy[i]; | |
234 | ||
235 | if (blkg->pd[i] && pol->pd_online_fn) | |
236 | pol->pd_online_fn(blkg->pd[i]); | |
237 | } | |
238 | } | |
239 | blkg->online = true; | |
240 | spin_unlock(&blkcg->lock); | |
241 | ||
242 | if (!ret) | |
243 | return blkg; | |
244 | ||
245 | /* @blkg failed fully initialized, use the usual release path */ | |
246 | blkg_put(blkg); | |
247 | return ERR_PTR(ret); | |
248 | ||
249 | err_put_congested: | |
250 | wb_congested_put(wb_congested); | |
251 | err_put_css: | |
252 | css_put(&blkcg->css); | |
253 | err_free_blkg: | |
254 | blkg_free(new_blkg); | |
255 | return ERR_PTR(ret); | |
256 | } | |
257 | ||
258 | /** | |
259 | * blkg_lookup_create - lookup blkg, try to create one if not there | |
260 | * @blkcg: blkcg of interest | |
261 | * @q: request_queue of interest | |
262 | * | |
263 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
264 | * create one. blkg creation is performed recursively from blkcg_root such | |
265 | * that all non-root blkg's have access to the parent blkg. This function | |
266 | * should be called under RCU read lock and @q->queue_lock. | |
267 | * | |
268 | * Returns pointer to the looked up or created blkg on success, ERR_PTR() | |
269 | * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not | |
270 | * dead and bypassing, returns ERR_PTR(-EBUSY). | |
271 | */ | |
272 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, | |
273 | struct request_queue *q) | |
274 | { | |
275 | struct blkcg_gq *blkg; | |
276 | ||
277 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
278 | lockdep_assert_held(q->queue_lock); | |
279 | ||
280 | /* | |
281 | * This could be the first entry point of blkcg implementation and | |
282 | * we shouldn't allow anything to go through for a bypassing queue. | |
283 | */ | |
284 | if (unlikely(blk_queue_bypass(q))) | |
285 | return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY); | |
286 | ||
287 | blkg = __blkg_lookup(blkcg, q, true); | |
288 | if (blkg) | |
289 | return blkg; | |
290 | ||
291 | /* | |
292 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
293 | * non-root blkgs have access to their parents. | |
294 | */ | |
295 | while (true) { | |
296 | struct blkcg *pos = blkcg; | |
297 | struct blkcg *parent = blkcg_parent(blkcg); | |
298 | ||
299 | while (parent && !__blkg_lookup(parent, q, false)) { | |
300 | pos = parent; | |
301 | parent = blkcg_parent(parent); | |
302 | } | |
303 | ||
304 | blkg = blkg_create(pos, q, NULL); | |
305 | if (pos == blkcg || IS_ERR(blkg)) | |
306 | return blkg; | |
307 | } | |
308 | } | |
309 | ||
310 | static void blkg_destroy(struct blkcg_gq *blkg) | |
311 | { | |
312 | struct blkcg *blkcg = blkg->blkcg; | |
313 | struct blkcg_gq *parent = blkg->parent; | |
314 | int i; | |
315 | ||
316 | lockdep_assert_held(blkg->q->queue_lock); | |
317 | lockdep_assert_held(&blkcg->lock); | |
318 | ||
319 | /* Something wrong if we are trying to remove same group twice */ | |
320 | WARN_ON_ONCE(list_empty(&blkg->q_node)); | |
321 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); | |
322 | ||
323 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
324 | struct blkcg_policy *pol = blkcg_policy[i]; | |
325 | ||
326 | if (blkg->pd[i] && pol->pd_offline_fn) | |
327 | pol->pd_offline_fn(blkg->pd[i]); | |
328 | } | |
329 | ||
330 | if (parent) { | |
331 | blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes); | |
332 | blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios); | |
333 | } | |
334 | ||
335 | blkg->online = false; | |
336 | ||
337 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); | |
338 | list_del_init(&blkg->q_node); | |
339 | hlist_del_init_rcu(&blkg->blkcg_node); | |
340 | ||
341 | /* | |
342 | * Both setting lookup hint to and clearing it from @blkg are done | |
343 | * under queue_lock. If it's not pointing to @blkg now, it never | |
344 | * will. Hint assignment itself can race safely. | |
345 | */ | |
346 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) | |
347 | rcu_assign_pointer(blkcg->blkg_hint, NULL); | |
348 | ||
349 | /* | |
350 | * Put the reference taken at the time of creation so that when all | |
351 | * queues are gone, group can be destroyed. | |
352 | */ | |
353 | blkg_put(blkg); | |
354 | } | |
355 | ||
356 | /** | |
357 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
358 | * @q: request_queue of interest | |
359 | * | |
360 | * Destroy all blkgs associated with @q. | |
361 | */ | |
362 | static void blkg_destroy_all(struct request_queue *q) | |
363 | { | |
364 | struct blkcg_gq *blkg, *n; | |
365 | ||
366 | lockdep_assert_held(q->queue_lock); | |
367 | ||
368 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { | |
369 | struct blkcg *blkcg = blkg->blkcg; | |
370 | ||
371 | spin_lock(&blkcg->lock); | |
372 | blkg_destroy(blkg); | |
373 | spin_unlock(&blkcg->lock); | |
374 | } | |
375 | ||
376 | q->root_blkg = NULL; | |
377 | q->root_rl.blkg = NULL; | |
378 | } | |
379 | ||
380 | /* | |
381 | * A group is RCU protected, but having an rcu lock does not mean that one | |
382 | * can access all the fields of blkg and assume these are valid. For | |
383 | * example, don't try to follow throtl_data and request queue links. | |
384 | * | |
385 | * Having a reference to blkg under an rcu allows accesses to only values | |
386 | * local to groups like group stats and group rate limits. | |
387 | */ | |
388 | void __blkg_release_rcu(struct rcu_head *rcu_head) | |
389 | { | |
390 | struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head); | |
391 | ||
392 | /* release the blkcg and parent blkg refs this blkg has been holding */ | |
393 | css_put(&blkg->blkcg->css); | |
394 | if (blkg->parent) | |
395 | blkg_put(blkg->parent); | |
396 | ||
397 | wb_congested_put(blkg->wb_congested); | |
398 | ||
399 | blkg_free(blkg); | |
400 | } | |
401 | EXPORT_SYMBOL_GPL(__blkg_release_rcu); | |
402 | ||
403 | /* | |
404 | * The next function used by blk_queue_for_each_rl(). It's a bit tricky | |
405 | * because the root blkg uses @q->root_rl instead of its own rl. | |
406 | */ | |
407 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
408 | struct request_queue *q) | |
409 | { | |
410 | struct list_head *ent; | |
411 | struct blkcg_gq *blkg; | |
412 | ||
413 | /* | |
414 | * Determine the current blkg list_head. The first entry is | |
415 | * root_rl which is off @q->blkg_list and mapped to the head. | |
416 | */ | |
417 | if (rl == &q->root_rl) { | |
418 | ent = &q->blkg_list; | |
419 | /* There are no more block groups, hence no request lists */ | |
420 | if (list_empty(ent)) | |
421 | return NULL; | |
422 | } else { | |
423 | blkg = container_of(rl, struct blkcg_gq, rl); | |
424 | ent = &blkg->q_node; | |
425 | } | |
426 | ||
427 | /* walk to the next list_head, skip root blkcg */ | |
428 | ent = ent->next; | |
429 | if (ent == &q->root_blkg->q_node) | |
430 | ent = ent->next; | |
431 | if (ent == &q->blkg_list) | |
432 | return NULL; | |
433 | ||
434 | blkg = container_of(ent, struct blkcg_gq, q_node); | |
435 | return &blkg->rl; | |
436 | } | |
437 | ||
438 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, | |
439 | struct cftype *cftype, u64 val) | |
440 | { | |
441 | struct blkcg *blkcg = css_to_blkcg(css); | |
442 | struct blkcg_gq *blkg; | |
443 | int i; | |
444 | ||
445 | mutex_lock(&blkcg_pol_mutex); | |
446 | spin_lock_irq(&blkcg->lock); | |
447 | ||
448 | /* | |
449 | * Note that stat reset is racy - it doesn't synchronize against | |
450 | * stat updates. This is a debug feature which shouldn't exist | |
451 | * anyway. If you get hit by a race, retry. | |
452 | */ | |
453 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { | |
454 | blkg_rwstat_reset(&blkg->stat_bytes); | |
455 | blkg_rwstat_reset(&blkg->stat_ios); | |
456 | ||
457 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
458 | struct blkcg_policy *pol = blkcg_policy[i]; | |
459 | ||
460 | if (blkg->pd[i] && pol->pd_reset_stats_fn) | |
461 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
462 | } | |
463 | } | |
464 | ||
465 | spin_unlock_irq(&blkcg->lock); | |
466 | mutex_unlock(&blkcg_pol_mutex); | |
467 | return 0; | |
468 | } | |
469 | ||
470 | const char *blkg_dev_name(struct blkcg_gq *blkg) | |
471 | { | |
472 | /* some drivers (floppy) instantiate a queue w/o disk registered */ | |
473 | if (blkg->q->backing_dev_info->dev) | |
474 | return dev_name(blkg->q->backing_dev_info->dev); | |
475 | return NULL; | |
476 | } | |
477 | EXPORT_SYMBOL_GPL(blkg_dev_name); | |
478 | ||
479 | /** | |
480 | * blkcg_print_blkgs - helper for printing per-blkg data | |
481 | * @sf: seq_file to print to | |
482 | * @blkcg: blkcg of interest | |
483 | * @prfill: fill function to print out a blkg | |
484 | * @pol: policy in question | |
485 | * @data: data to be passed to @prfill | |
486 | * @show_total: to print out sum of prfill return values or not | |
487 | * | |
488 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
489 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
490 | * policy data and @data and the matching queue lock held. If @show_total | |
491 | * is %true, the sum of the return values from @prfill is printed with | |
492 | * "Total" label at the end. | |
493 | * | |
494 | * This is to be used to construct print functions for | |
495 | * cftype->read_seq_string method. | |
496 | */ | |
497 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, | |
498 | u64 (*prfill)(struct seq_file *, | |
499 | struct blkg_policy_data *, int), | |
500 | const struct blkcg_policy *pol, int data, | |
501 | bool show_total) | |
502 | { | |
503 | struct blkcg_gq *blkg; | |
504 | u64 total = 0; | |
505 | ||
506 | rcu_read_lock(); | |
507 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
508 | spin_lock_irq(blkg->q->queue_lock); | |
509 | if (blkcg_policy_enabled(blkg->q, pol)) | |
510 | total += prfill(sf, blkg->pd[pol->plid], data); | |
511 | spin_unlock_irq(blkg->q->queue_lock); | |
512 | } | |
513 | rcu_read_unlock(); | |
514 | ||
515 | if (show_total) | |
516 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
517 | } | |
518 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); | |
519 | ||
520 | /** | |
521 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
522 | * @sf: seq_file to print to | |
523 | * @pd: policy private data of interest | |
524 | * @v: value to print | |
525 | * | |
526 | * Print @v to @sf for the device assocaited with @pd. | |
527 | */ | |
528 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) | |
529 | { | |
530 | const char *dname = blkg_dev_name(pd->blkg); | |
531 | ||
532 | if (!dname) | |
533 | return 0; | |
534 | ||
535 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
536 | return v; | |
537 | } | |
538 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); | |
539 | ||
540 | /** | |
541 | * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat | |
542 | * @sf: seq_file to print to | |
543 | * @pd: policy private data of interest | |
544 | * @rwstat: rwstat to print | |
545 | * | |
546 | * Print @rwstat to @sf for the device assocaited with @pd. | |
547 | */ | |
548 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, | |
549 | const struct blkg_rwstat *rwstat) | |
550 | { | |
551 | static const char *rwstr[] = { | |
552 | [BLKG_RWSTAT_READ] = "Read", | |
553 | [BLKG_RWSTAT_WRITE] = "Write", | |
554 | [BLKG_RWSTAT_SYNC] = "Sync", | |
555 | [BLKG_RWSTAT_ASYNC] = "Async", | |
556 | }; | |
557 | const char *dname = blkg_dev_name(pd->blkg); | |
558 | u64 v; | |
559 | int i; | |
560 | ||
561 | if (!dname) | |
562 | return 0; | |
563 | ||
564 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
565 | seq_printf(sf, "%s %s %llu\n", dname, rwstr[i], | |
566 | (unsigned long long)atomic64_read(&rwstat->aux_cnt[i])); | |
567 | ||
568 | v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) + | |
569 | atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]); | |
570 | seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v); | |
571 | return v; | |
572 | } | |
573 | EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat); | |
574 | ||
575 | /** | |
576 | * blkg_prfill_stat - prfill callback for blkg_stat | |
577 | * @sf: seq_file to print to | |
578 | * @pd: policy private data of interest | |
579 | * @off: offset to the blkg_stat in @pd | |
580 | * | |
581 | * prfill callback for printing a blkg_stat. | |
582 | */ | |
583 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off) | |
584 | { | |
585 | return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off)); | |
586 | } | |
587 | EXPORT_SYMBOL_GPL(blkg_prfill_stat); | |
588 | ||
589 | /** | |
590 | * blkg_prfill_rwstat - prfill callback for blkg_rwstat | |
591 | * @sf: seq_file to print to | |
592 | * @pd: policy private data of interest | |
593 | * @off: offset to the blkg_rwstat in @pd | |
594 | * | |
595 | * prfill callback for printing a blkg_rwstat. | |
596 | */ | |
597 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, | |
598 | int off) | |
599 | { | |
600 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off); | |
601 | ||
602 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
603 | } | |
604 | EXPORT_SYMBOL_GPL(blkg_prfill_rwstat); | |
605 | ||
606 | static u64 blkg_prfill_rwstat_field(struct seq_file *sf, | |
607 | struct blkg_policy_data *pd, int off) | |
608 | { | |
609 | struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off); | |
610 | ||
611 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
612 | } | |
613 | ||
614 | /** | |
615 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes | |
616 | * @sf: seq_file to print to | |
617 | * @v: unused | |
618 | * | |
619 | * To be used as cftype->seq_show to print blkg->stat_bytes. | |
620 | * cftype->private must be set to the blkcg_policy. | |
621 | */ | |
622 | int blkg_print_stat_bytes(struct seq_file *sf, void *v) | |
623 | { | |
624 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
625 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
626 | offsetof(struct blkcg_gq, stat_bytes), true); | |
627 | return 0; | |
628 | } | |
629 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes); | |
630 | ||
631 | /** | |
632 | * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios | |
633 | * @sf: seq_file to print to | |
634 | * @v: unused | |
635 | * | |
636 | * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private | |
637 | * must be set to the blkcg_policy. | |
638 | */ | |
639 | int blkg_print_stat_ios(struct seq_file *sf, void *v) | |
640 | { | |
641 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
642 | blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private, | |
643 | offsetof(struct blkcg_gq, stat_ios), true); | |
644 | return 0; | |
645 | } | |
646 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios); | |
647 | ||
648 | static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf, | |
649 | struct blkg_policy_data *pd, | |
650 | int off) | |
651 | { | |
652 | struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg, | |
653 | NULL, off); | |
654 | return __blkg_prfill_rwstat(sf, pd, &rwstat); | |
655 | } | |
656 | ||
657 | /** | |
658 | * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes | |
659 | * @sf: seq_file to print to | |
660 | * @v: unused | |
661 | */ | |
662 | int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v) | |
663 | { | |
664 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
665 | blkg_prfill_rwstat_field_recursive, | |
666 | (void *)seq_cft(sf)->private, | |
667 | offsetof(struct blkcg_gq, stat_bytes), true); | |
668 | return 0; | |
669 | } | |
670 | EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive); | |
671 | ||
672 | /** | |
673 | * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios | |
674 | * @sf: seq_file to print to | |
675 | * @v: unused | |
676 | */ | |
677 | int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v) | |
678 | { | |
679 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
680 | blkg_prfill_rwstat_field_recursive, | |
681 | (void *)seq_cft(sf)->private, | |
682 | offsetof(struct blkcg_gq, stat_ios), true); | |
683 | return 0; | |
684 | } | |
685 | EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive); | |
686 | ||
687 | /** | |
688 | * blkg_stat_recursive_sum - collect hierarchical blkg_stat | |
689 | * @blkg: blkg of interest | |
690 | * @pol: blkcg_policy which contains the blkg_stat | |
691 | * @off: offset to the blkg_stat in blkg_policy_data or @blkg | |
692 | * | |
693 | * Collect the blkg_stat specified by @blkg, @pol and @off and all its | |
694 | * online descendants and their aux counts. The caller must be holding the | |
695 | * queue lock for online tests. | |
696 | * | |
697 | * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is | |
698 | * at @off bytes into @blkg's blkg_policy_data of the policy. | |
699 | */ | |
700 | u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg, | |
701 | struct blkcg_policy *pol, int off) | |
702 | { | |
703 | struct blkcg_gq *pos_blkg; | |
704 | struct cgroup_subsys_state *pos_css; | |
705 | u64 sum = 0; | |
706 | ||
707 | lockdep_assert_held(blkg->q->queue_lock); | |
708 | ||
709 | rcu_read_lock(); | |
710 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { | |
711 | struct blkg_stat *stat; | |
712 | ||
713 | if (!pos_blkg->online) | |
714 | continue; | |
715 | ||
716 | if (pol) | |
717 | stat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
718 | else | |
719 | stat = (void *)blkg + off; | |
720 | ||
721 | sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
722 | } | |
723 | rcu_read_unlock(); | |
724 | ||
725 | return sum; | |
726 | } | |
727 | EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum); | |
728 | ||
729 | /** | |
730 | * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat | |
731 | * @blkg: blkg of interest | |
732 | * @pol: blkcg_policy which contains the blkg_rwstat | |
733 | * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg | |
734 | * | |
735 | * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its | |
736 | * online descendants and their aux counts. The caller must be holding the | |
737 | * queue lock for online tests. | |
738 | * | |
739 | * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it | |
740 | * is at @off bytes into @blkg's blkg_policy_data of the policy. | |
741 | */ | |
742 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, | |
743 | struct blkcg_policy *pol, int off) | |
744 | { | |
745 | struct blkcg_gq *pos_blkg; | |
746 | struct cgroup_subsys_state *pos_css; | |
747 | struct blkg_rwstat sum = { }; | |
748 | int i; | |
749 | ||
750 | lockdep_assert_held(blkg->q->queue_lock); | |
751 | ||
752 | rcu_read_lock(); | |
753 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { | |
754 | struct blkg_rwstat *rwstat; | |
755 | ||
756 | if (!pos_blkg->online) | |
757 | continue; | |
758 | ||
759 | if (pol) | |
760 | rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off; | |
761 | else | |
762 | rwstat = (void *)pos_blkg + off; | |
763 | ||
764 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
765 | atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) + | |
766 | percpu_counter_sum_positive(&rwstat->cpu_cnt[i]), | |
767 | &sum.aux_cnt[i]); | |
768 | } | |
769 | rcu_read_unlock(); | |
770 | ||
771 | return sum; | |
772 | } | |
773 | EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum); | |
774 | ||
775 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ | |
776 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, | |
777 | const struct blkcg_policy *pol, | |
778 | struct request_queue *q) | |
779 | { | |
780 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
781 | lockdep_assert_held(q->queue_lock); | |
782 | ||
783 | if (!blkcg_policy_enabled(q, pol)) | |
784 | return ERR_PTR(-EOPNOTSUPP); | |
785 | ||
786 | /* | |
787 | * This could be the first entry point of blkcg implementation and | |
788 | * we shouldn't allow anything to go through for a bypassing queue. | |
789 | */ | |
790 | if (unlikely(blk_queue_bypass(q))) | |
791 | return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY); | |
792 | ||
793 | return __blkg_lookup(blkcg, q, true /* update_hint */); | |
794 | } | |
795 | ||
796 | /** | |
797 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
798 | * @blkcg: target block cgroup | |
799 | * @pol: target policy | |
800 | * @input: input string | |
801 | * @ctx: blkg_conf_ctx to be filled | |
802 | * | |
803 | * Parse per-blkg config update from @input and initialize @ctx with the | |
804 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the | |
805 | * part of @input following MAJ:MIN. This function returns with RCU read | |
806 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
807 | */ | |
808 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, | |
809 | char *input, struct blkg_conf_ctx *ctx) | |
810 | __acquires(rcu) __acquires(disk->queue->queue_lock) | |
811 | { | |
812 | struct gendisk *disk; | |
813 | struct request_queue *q; | |
814 | struct blkcg_gq *blkg; | |
815 | struct module *owner; | |
816 | unsigned int major, minor; | |
817 | int key_len, part, ret; | |
818 | char *body; | |
819 | ||
820 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) | |
821 | return -EINVAL; | |
822 | ||
823 | body = input + key_len; | |
824 | if (!isspace(*body)) | |
825 | return -EINVAL; | |
826 | body = skip_spaces(body); | |
827 | ||
828 | disk = get_gendisk(MKDEV(major, minor), &part); | |
829 | if (!disk) | |
830 | return -ENODEV; | |
831 | if (part) { | |
832 | ret = -ENODEV; | |
833 | goto fail; | |
834 | } | |
835 | ||
836 | q = disk->queue; | |
837 | ||
838 | rcu_read_lock(); | |
839 | spin_lock_irq(q->queue_lock); | |
840 | ||
841 | blkg = blkg_lookup_check(blkcg, pol, q); | |
842 | if (IS_ERR(blkg)) { | |
843 | ret = PTR_ERR(blkg); | |
844 | goto fail_unlock; | |
845 | } | |
846 | ||
847 | if (blkg) | |
848 | goto success; | |
849 | ||
850 | /* | |
851 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
852 | * non-root blkgs have access to their parents. | |
853 | */ | |
854 | while (true) { | |
855 | struct blkcg *pos = blkcg; | |
856 | struct blkcg *parent; | |
857 | struct blkcg_gq *new_blkg; | |
858 | ||
859 | parent = blkcg_parent(blkcg); | |
860 | while (parent && !__blkg_lookup(parent, q, false)) { | |
861 | pos = parent; | |
862 | parent = blkcg_parent(parent); | |
863 | } | |
864 | ||
865 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
866 | spin_unlock_irq(q->queue_lock); | |
867 | rcu_read_unlock(); | |
868 | ||
869 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); | |
870 | if (unlikely(!new_blkg)) { | |
871 | ret = -ENOMEM; | |
872 | goto fail; | |
873 | } | |
874 | ||
875 | rcu_read_lock(); | |
876 | spin_lock_irq(q->queue_lock); | |
877 | ||
878 | blkg = blkg_lookup_check(pos, pol, q); | |
879 | if (IS_ERR(blkg)) { | |
880 | ret = PTR_ERR(blkg); | |
881 | goto fail_unlock; | |
882 | } | |
883 | ||
884 | if (blkg) { | |
885 | blkg_free(new_blkg); | |
886 | } else { | |
887 | blkg = blkg_create(pos, q, new_blkg); | |
888 | if (unlikely(IS_ERR(blkg))) { | |
889 | ret = PTR_ERR(blkg); | |
890 | goto fail_unlock; | |
891 | } | |
892 | } | |
893 | ||
894 | if (pos == blkcg) | |
895 | goto success; | |
896 | } | |
897 | success: | |
898 | ctx->disk = disk; | |
899 | ctx->blkg = blkg; | |
900 | ctx->body = body; | |
901 | return 0; | |
902 | ||
903 | fail_unlock: | |
904 | spin_unlock_irq(q->queue_lock); | |
905 | rcu_read_unlock(); | |
906 | fail: | |
907 | owner = disk->fops->owner; | |
908 | put_disk(disk); | |
909 | module_put(owner); | |
910 | /* | |
911 | * If queue was bypassing, we should retry. Do so after a | |
912 | * short msleep(). It isn't strictly necessary but queue | |
913 | * can be bypassing for some time and it's always nice to | |
914 | * avoid busy looping. | |
915 | */ | |
916 | if (ret == -EBUSY) { | |
917 | msleep(10); | |
918 | ret = restart_syscall(); | |
919 | } | |
920 | return ret; | |
921 | } | |
922 | EXPORT_SYMBOL_GPL(blkg_conf_prep); | |
923 | ||
924 | /** | |
925 | * blkg_conf_finish - finish up per-blkg config update | |
926 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
927 | * | |
928 | * Finish up after per-blkg config update. This function must be paired | |
929 | * with blkg_conf_prep(). | |
930 | */ | |
931 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) | |
932 | __releases(ctx->disk->queue->queue_lock) __releases(rcu) | |
933 | { | |
934 | struct module *owner; | |
935 | ||
936 | spin_unlock_irq(ctx->disk->queue->queue_lock); | |
937 | rcu_read_unlock(); | |
938 | owner = ctx->disk->fops->owner; | |
939 | put_disk(ctx->disk); | |
940 | module_put(owner); | |
941 | } | |
942 | EXPORT_SYMBOL_GPL(blkg_conf_finish); | |
943 | ||
944 | static int blkcg_print_stat(struct seq_file *sf, void *v) | |
945 | { | |
946 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
947 | struct blkcg_gq *blkg; | |
948 | ||
949 | rcu_read_lock(); | |
950 | ||
951 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
952 | const char *dname; | |
953 | struct blkg_rwstat rwstat; | |
954 | u64 rbytes, wbytes, rios, wios; | |
955 | ||
956 | dname = blkg_dev_name(blkg); | |
957 | if (!dname) | |
958 | continue; | |
959 | ||
960 | spin_lock_irq(blkg->q->queue_lock); | |
961 | ||
962 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
963 | offsetof(struct blkcg_gq, stat_bytes)); | |
964 | rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
965 | wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
966 | ||
967 | rwstat = blkg_rwstat_recursive_sum(blkg, NULL, | |
968 | offsetof(struct blkcg_gq, stat_ios)); | |
969 | rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]); | |
970 | wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]); | |
971 | ||
972 | spin_unlock_irq(blkg->q->queue_lock); | |
973 | ||
974 | if (rbytes || wbytes || rios || wios) | |
975 | seq_printf(sf, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n", | |
976 | dname, rbytes, wbytes, rios, wios); | |
977 | } | |
978 | ||
979 | rcu_read_unlock(); | |
980 | return 0; | |
981 | } | |
982 | ||
983 | static struct cftype blkcg_files[] = { | |
984 | { | |
985 | .name = "stat", | |
986 | .flags = CFTYPE_NOT_ON_ROOT, | |
987 | .seq_show = blkcg_print_stat, | |
988 | }, | |
989 | { } /* terminate */ | |
990 | }; | |
991 | ||
992 | static struct cftype blkcg_legacy_files[] = { | |
993 | { | |
994 | .name = "reset_stats", | |
995 | .write_u64 = blkcg_reset_stats, | |
996 | }, | |
997 | { } /* terminate */ | |
998 | }; | |
999 | ||
1000 | /** | |
1001 | * blkcg_css_offline - cgroup css_offline callback | |
1002 | * @css: css of interest | |
1003 | * | |
1004 | * This function is called when @css is about to go away and responsible | |
1005 | * for shooting down all blkgs associated with @css. blkgs should be | |
1006 | * removed while holding both q and blkcg locks. As blkcg lock is nested | |
1007 | * inside q lock, this function performs reverse double lock dancing. | |
1008 | * | |
1009 | * This is the blkcg counterpart of ioc_release_fn(). | |
1010 | */ | |
1011 | static void blkcg_css_offline(struct cgroup_subsys_state *css) | |
1012 | { | |
1013 | struct blkcg *blkcg = css_to_blkcg(css); | |
1014 | ||
1015 | spin_lock_irq(&blkcg->lock); | |
1016 | ||
1017 | while (!hlist_empty(&blkcg->blkg_list)) { | |
1018 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
1019 | struct blkcg_gq, blkcg_node); | |
1020 | struct request_queue *q = blkg->q; | |
1021 | ||
1022 | if (spin_trylock(q->queue_lock)) { | |
1023 | blkg_destroy(blkg); | |
1024 | spin_unlock(q->queue_lock); | |
1025 | } else { | |
1026 | spin_unlock_irq(&blkcg->lock); | |
1027 | cpu_relax(); | |
1028 | spin_lock_irq(&blkcg->lock); | |
1029 | } | |
1030 | } | |
1031 | ||
1032 | spin_unlock_irq(&blkcg->lock); | |
1033 | ||
1034 | wb_blkcg_offline(blkcg); | |
1035 | } | |
1036 | ||
1037 | static void blkcg_css_free(struct cgroup_subsys_state *css) | |
1038 | { | |
1039 | struct blkcg *blkcg = css_to_blkcg(css); | |
1040 | int i; | |
1041 | ||
1042 | mutex_lock(&blkcg_pol_mutex); | |
1043 | ||
1044 | list_del(&blkcg->all_blkcgs_node); | |
1045 | ||
1046 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
1047 | if (blkcg->cpd[i]) | |
1048 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1049 | ||
1050 | mutex_unlock(&blkcg_pol_mutex); | |
1051 | ||
1052 | kfree(blkcg); | |
1053 | } | |
1054 | ||
1055 | static struct cgroup_subsys_state * | |
1056 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
1057 | { | |
1058 | struct blkcg *blkcg; | |
1059 | struct cgroup_subsys_state *ret; | |
1060 | int i; | |
1061 | ||
1062 | mutex_lock(&blkcg_pol_mutex); | |
1063 | ||
1064 | if (!parent_css) { | |
1065 | blkcg = &blkcg_root; | |
1066 | } else { | |
1067 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1068 | if (!blkcg) { | |
1069 | ret = ERR_PTR(-ENOMEM); | |
1070 | goto unlock; | |
1071 | } | |
1072 | } | |
1073 | ||
1074 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1075 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1076 | struct blkcg_policy_data *cpd; | |
1077 | ||
1078 | /* | |
1079 | * If the policy hasn't been attached yet, wait for it | |
1080 | * to be attached before doing anything else. Otherwise, | |
1081 | * check if the policy requires any specific per-cgroup | |
1082 | * data: if it does, allocate and initialize it. | |
1083 | */ | |
1084 | if (!pol || !pol->cpd_alloc_fn) | |
1085 | continue; | |
1086 | ||
1087 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); | |
1088 | if (!cpd) { | |
1089 | ret = ERR_PTR(-ENOMEM); | |
1090 | goto free_pd_blkcg; | |
1091 | } | |
1092 | blkcg->cpd[i] = cpd; | |
1093 | cpd->blkcg = blkcg; | |
1094 | cpd->plid = i; | |
1095 | if (pol->cpd_init_fn) | |
1096 | pol->cpd_init_fn(cpd); | |
1097 | } | |
1098 | ||
1099 | spin_lock_init(&blkcg->lock); | |
1100 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); | |
1101 | INIT_HLIST_HEAD(&blkcg->blkg_list); | |
1102 | #ifdef CONFIG_CGROUP_WRITEBACK | |
1103 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1104 | #endif | |
1105 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); | |
1106 | ||
1107 | mutex_unlock(&blkcg_pol_mutex); | |
1108 | return &blkcg->css; | |
1109 | ||
1110 | free_pd_blkcg: | |
1111 | for (i--; i >= 0; i--) | |
1112 | if (blkcg->cpd[i]) | |
1113 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1114 | ||
1115 | if (blkcg != &blkcg_root) | |
1116 | kfree(blkcg); | |
1117 | unlock: | |
1118 | mutex_unlock(&blkcg_pol_mutex); | |
1119 | return ret; | |
1120 | } | |
1121 | ||
1122 | /** | |
1123 | * blkcg_init_queue - initialize blkcg part of request queue | |
1124 | * @q: request_queue to initialize | |
1125 | * | |
1126 | * Called from blk_alloc_queue_node(). Responsible for initializing blkcg | |
1127 | * part of new request_queue @q. | |
1128 | * | |
1129 | * RETURNS: | |
1130 | * 0 on success, -errno on failure. | |
1131 | */ | |
1132 | int blkcg_init_queue(struct request_queue *q) | |
1133 | { | |
1134 | struct blkcg_gq *new_blkg, *blkg; | |
1135 | bool preloaded; | |
1136 | int ret; | |
1137 | ||
1138 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); | |
1139 | if (!new_blkg) | |
1140 | return -ENOMEM; | |
1141 | ||
1142 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1143 | ||
1144 | /* | |
1145 | * Make sure the root blkg exists and count the existing blkgs. As | |
1146 | * @q is bypassing at this point, blkg_lookup_create() can't be | |
1147 | * used. Open code insertion. | |
1148 | */ | |
1149 | rcu_read_lock(); | |
1150 | spin_lock_irq(q->queue_lock); | |
1151 | blkg = blkg_create(&blkcg_root, q, new_blkg); | |
1152 | spin_unlock_irq(q->queue_lock); | |
1153 | rcu_read_unlock(); | |
1154 | ||
1155 | if (preloaded) | |
1156 | radix_tree_preload_end(); | |
1157 | ||
1158 | if (IS_ERR(blkg)) | |
1159 | return PTR_ERR(blkg); | |
1160 | ||
1161 | q->root_blkg = blkg; | |
1162 | q->root_rl.blkg = blkg; | |
1163 | ||
1164 | ret = blk_throtl_init(q); | |
1165 | if (ret) { | |
1166 | spin_lock_irq(q->queue_lock); | |
1167 | blkg_destroy_all(q); | |
1168 | spin_unlock_irq(q->queue_lock); | |
1169 | } | |
1170 | return ret; | |
1171 | } | |
1172 | ||
1173 | /** | |
1174 | * blkcg_drain_queue - drain blkcg part of request_queue | |
1175 | * @q: request_queue to drain | |
1176 | * | |
1177 | * Called from blk_drain_queue(). Responsible for draining blkcg part. | |
1178 | */ | |
1179 | void blkcg_drain_queue(struct request_queue *q) | |
1180 | { | |
1181 | lockdep_assert_held(q->queue_lock); | |
1182 | ||
1183 | /* | |
1184 | * @q could be exiting and already have destroyed all blkgs as | |
1185 | * indicated by NULL root_blkg. If so, don't confuse policies. | |
1186 | */ | |
1187 | if (!q->root_blkg) | |
1188 | return; | |
1189 | ||
1190 | blk_throtl_drain(q); | |
1191 | } | |
1192 | ||
1193 | /** | |
1194 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1195 | * @q: request_queue being released | |
1196 | * | |
1197 | * Called from blk_release_queue(). Responsible for exiting blkcg part. | |
1198 | */ | |
1199 | void blkcg_exit_queue(struct request_queue *q) | |
1200 | { | |
1201 | spin_lock_irq(q->queue_lock); | |
1202 | blkg_destroy_all(q); | |
1203 | spin_unlock_irq(q->queue_lock); | |
1204 | ||
1205 | blk_throtl_exit(q); | |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * We cannot support shared io contexts, as we have no mean to support | |
1210 | * two tasks with the same ioc in two different groups without major rework | |
1211 | * of the main cic data structures. For now we allow a task to change | |
1212 | * its cgroup only if it's the only owner of its ioc. | |
1213 | */ | |
1214 | static int blkcg_can_attach(struct cgroup_taskset *tset) | |
1215 | { | |
1216 | struct task_struct *task; | |
1217 | struct cgroup_subsys_state *dst_css; | |
1218 | struct io_context *ioc; | |
1219 | int ret = 0; | |
1220 | ||
1221 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
1222 | cgroup_taskset_for_each(task, dst_css, tset) { | |
1223 | task_lock(task); | |
1224 | ioc = task->io_context; | |
1225 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1226 | ret = -EINVAL; | |
1227 | task_unlock(task); | |
1228 | if (ret) | |
1229 | break; | |
1230 | } | |
1231 | return ret; | |
1232 | } | |
1233 | ||
1234 | static void blkcg_bind(struct cgroup_subsys_state *root_css) | |
1235 | { | |
1236 | int i; | |
1237 | ||
1238 | mutex_lock(&blkcg_pol_mutex); | |
1239 | ||
1240 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1241 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1242 | struct blkcg *blkcg; | |
1243 | ||
1244 | if (!pol || !pol->cpd_bind_fn) | |
1245 | continue; | |
1246 | ||
1247 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1248 | if (blkcg->cpd[pol->plid]) | |
1249 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1250 | } | |
1251 | mutex_unlock(&blkcg_pol_mutex); | |
1252 | } | |
1253 | ||
1254 | struct cgroup_subsys io_cgrp_subsys = { | |
1255 | .css_alloc = blkcg_css_alloc, | |
1256 | .css_offline = blkcg_css_offline, | |
1257 | .css_free = blkcg_css_free, | |
1258 | .can_attach = blkcg_can_attach, | |
1259 | .bind = blkcg_bind, | |
1260 | .dfl_cftypes = blkcg_files, | |
1261 | .legacy_cftypes = blkcg_legacy_files, | |
1262 | .legacy_name = "blkio", | |
1263 | #ifdef CONFIG_MEMCG | |
1264 | /* | |
1265 | * This ensures that, if available, memcg is automatically enabled | |
1266 | * together on the default hierarchy so that the owner cgroup can | |
1267 | * be retrieved from writeback pages. | |
1268 | */ | |
1269 | .depends_on = 1 << memory_cgrp_id, | |
1270 | #endif | |
1271 | }; | |
1272 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); | |
1273 | ||
1274 | /** | |
1275 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1276 | * @q: request_queue of interest | |
1277 | * @pol: blkcg policy to activate | |
1278 | * | |
1279 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1280 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1281 | * | |
1282 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1283 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1284 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1285 | * always enough for dereferencing policy data. | |
1286 | * | |
1287 | * The caller is responsible for synchronizing [de]activations and policy | |
1288 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1289 | */ | |
1290 | int blkcg_activate_policy(struct request_queue *q, | |
1291 | const struct blkcg_policy *pol) | |
1292 | { | |
1293 | struct blkg_policy_data *pd_prealloc = NULL; | |
1294 | struct blkcg_gq *blkg; | |
1295 | int ret; | |
1296 | ||
1297 | if (blkcg_policy_enabled(q, pol)) | |
1298 | return 0; | |
1299 | ||
1300 | if (q->mq_ops) | |
1301 | blk_mq_freeze_queue(q); | |
1302 | else | |
1303 | blk_queue_bypass_start(q); | |
1304 | pd_prealloc: | |
1305 | if (!pd_prealloc) { | |
1306 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node); | |
1307 | if (!pd_prealloc) { | |
1308 | ret = -ENOMEM; | |
1309 | goto out_bypass_end; | |
1310 | } | |
1311 | } | |
1312 | ||
1313 | spin_lock_irq(q->queue_lock); | |
1314 | ||
1315 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1316 | struct blkg_policy_data *pd; | |
1317 | ||
1318 | if (blkg->pd[pol->plid]) | |
1319 | continue; | |
1320 | ||
1321 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node); | |
1322 | if (!pd) | |
1323 | swap(pd, pd_prealloc); | |
1324 | if (!pd) { | |
1325 | spin_unlock_irq(q->queue_lock); | |
1326 | goto pd_prealloc; | |
1327 | } | |
1328 | ||
1329 | blkg->pd[pol->plid] = pd; | |
1330 | pd->blkg = blkg; | |
1331 | pd->plid = pol->plid; | |
1332 | if (pol->pd_init_fn) | |
1333 | pol->pd_init_fn(pd); | |
1334 | } | |
1335 | ||
1336 | __set_bit(pol->plid, q->blkcg_pols); | |
1337 | ret = 0; | |
1338 | ||
1339 | spin_unlock_irq(q->queue_lock); | |
1340 | out_bypass_end: | |
1341 | if (q->mq_ops) | |
1342 | blk_mq_unfreeze_queue(q); | |
1343 | else | |
1344 | blk_queue_bypass_end(q); | |
1345 | if (pd_prealloc) | |
1346 | pol->pd_free_fn(pd_prealloc); | |
1347 | return ret; | |
1348 | } | |
1349 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1350 | ||
1351 | /** | |
1352 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1353 | * @q: request_queue of interest | |
1354 | * @pol: blkcg policy to deactivate | |
1355 | * | |
1356 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1357 | * blkcg_activate_policy(). | |
1358 | */ | |
1359 | void blkcg_deactivate_policy(struct request_queue *q, | |
1360 | const struct blkcg_policy *pol) | |
1361 | { | |
1362 | struct blkcg_gq *blkg; | |
1363 | ||
1364 | if (!blkcg_policy_enabled(q, pol)) | |
1365 | return; | |
1366 | ||
1367 | if (q->mq_ops) | |
1368 | blk_mq_freeze_queue(q); | |
1369 | else | |
1370 | blk_queue_bypass_start(q); | |
1371 | ||
1372 | spin_lock_irq(q->queue_lock); | |
1373 | ||
1374 | __clear_bit(pol->plid, q->blkcg_pols); | |
1375 | ||
1376 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1377 | /* grab blkcg lock too while removing @pd from @blkg */ | |
1378 | spin_lock(&blkg->blkcg->lock); | |
1379 | ||
1380 | if (blkg->pd[pol->plid]) { | |
1381 | if (pol->pd_offline_fn) | |
1382 | pol->pd_offline_fn(blkg->pd[pol->plid]); | |
1383 | pol->pd_free_fn(blkg->pd[pol->plid]); | |
1384 | blkg->pd[pol->plid] = NULL; | |
1385 | } | |
1386 | ||
1387 | spin_unlock(&blkg->blkcg->lock); | |
1388 | } | |
1389 | ||
1390 | spin_unlock_irq(q->queue_lock); | |
1391 | ||
1392 | if (q->mq_ops) | |
1393 | blk_mq_unfreeze_queue(q); | |
1394 | else | |
1395 | blk_queue_bypass_end(q); | |
1396 | } | |
1397 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1398 | ||
1399 | /** | |
1400 | * blkcg_policy_register - register a blkcg policy | |
1401 | * @pol: blkcg policy to register | |
1402 | * | |
1403 | * Register @pol with blkcg core. Might sleep and @pol may be modified on | |
1404 | * successful registration. Returns 0 on success and -errno on failure. | |
1405 | */ | |
1406 | int blkcg_policy_register(struct blkcg_policy *pol) | |
1407 | { | |
1408 | struct blkcg *blkcg; | |
1409 | int i, ret; | |
1410 | ||
1411 | mutex_lock(&blkcg_pol_register_mutex); | |
1412 | mutex_lock(&blkcg_pol_mutex); | |
1413 | ||
1414 | /* find an empty slot */ | |
1415 | ret = -ENOSPC; | |
1416 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
1417 | if (!blkcg_policy[i]) | |
1418 | break; | |
1419 | if (i >= BLKCG_MAX_POLS) | |
1420 | goto err_unlock; | |
1421 | ||
1422 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ | |
1423 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1424 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1425 | goto err_unlock; | |
1426 | ||
1427 | /* register @pol */ | |
1428 | pol->plid = i; | |
1429 | blkcg_policy[pol->plid] = pol; | |
1430 | ||
1431 | /* allocate and install cpd's */ | |
1432 | if (pol->cpd_alloc_fn) { | |
1433 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1434 | struct blkcg_policy_data *cpd; | |
1435 | ||
1436 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); | |
1437 | if (!cpd) | |
1438 | goto err_free_cpds; | |
1439 | ||
1440 | blkcg->cpd[pol->plid] = cpd; | |
1441 | cpd->blkcg = blkcg; | |
1442 | cpd->plid = pol->plid; | |
1443 | pol->cpd_init_fn(cpd); | |
1444 | } | |
1445 | } | |
1446 | ||
1447 | mutex_unlock(&blkcg_pol_mutex); | |
1448 | ||
1449 | /* everything is in place, add intf files for the new policy */ | |
1450 | if (pol->dfl_cftypes) | |
1451 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1452 | pol->dfl_cftypes)); | |
1453 | if (pol->legacy_cftypes) | |
1454 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, | |
1455 | pol->legacy_cftypes)); | |
1456 | mutex_unlock(&blkcg_pol_register_mutex); | |
1457 | return 0; | |
1458 | ||
1459 | err_free_cpds: | |
1460 | if (pol->cpd_free_fn) { | |
1461 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1462 | if (blkcg->cpd[pol->plid]) { | |
1463 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1464 | blkcg->cpd[pol->plid] = NULL; | |
1465 | } | |
1466 | } | |
1467 | } | |
1468 | blkcg_policy[pol->plid] = NULL; | |
1469 | err_unlock: | |
1470 | mutex_unlock(&blkcg_pol_mutex); | |
1471 | mutex_unlock(&blkcg_pol_register_mutex); | |
1472 | return ret; | |
1473 | } | |
1474 | EXPORT_SYMBOL_GPL(blkcg_policy_register); | |
1475 | ||
1476 | /** | |
1477 | * blkcg_policy_unregister - unregister a blkcg policy | |
1478 | * @pol: blkcg policy to unregister | |
1479 | * | |
1480 | * Undo blkcg_policy_register(@pol). Might sleep. | |
1481 | */ | |
1482 | void blkcg_policy_unregister(struct blkcg_policy *pol) | |
1483 | { | |
1484 | struct blkcg *blkcg; | |
1485 | ||
1486 | mutex_lock(&blkcg_pol_register_mutex); | |
1487 | ||
1488 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) | |
1489 | goto out_unlock; | |
1490 | ||
1491 | /* kill the intf files first */ | |
1492 | if (pol->dfl_cftypes) | |
1493 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
1494 | if (pol->legacy_cftypes) | |
1495 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
1496 | ||
1497 | /* remove cpds and unregister */ | |
1498 | mutex_lock(&blkcg_pol_mutex); | |
1499 | ||
1500 | if (pol->cpd_free_fn) { | |
1501 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1502 | if (blkcg->cpd[pol->plid]) { | |
1503 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1504 | blkcg->cpd[pol->plid] = NULL; | |
1505 | } | |
1506 | } | |
1507 | } | |
1508 | blkcg_policy[pol->plid] = NULL; | |
1509 | ||
1510 | mutex_unlock(&blkcg_pol_mutex); | |
1511 | out_unlock: | |
1512 | mutex_unlock(&blkcg_pol_register_mutex); | |
1513 | } | |
1514 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |