~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/block/blk-cgroup.c

Version: ~ [ linux-6.0 ] ~ [ linux-5.19.12 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.71 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.146 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.215 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.260 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.295 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.330 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  * Common Block IO controller cgroup interface
  4  *
  5  * Based on ideas and code from CFQ, CFS and BFQ:
  6  * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
  7  *
  8  * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
  9  *                    Paolo Valente <paolo.valente@unimore.it>
 10  *
 11  * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
 12  *                    Nauman Rafique <nauman@google.com>
 13  *
 14  * For policy-specific per-blkcg data:
 15  * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it>
 16  *                    Arianna Avanzini <avanzini.arianna@gmail.com>
 17  */
 18 #include <linux/ioprio.h>
 19 #include <linux/kdev_t.h>
 20 #include <linux/module.h>
 21 #include <linux/sched/signal.h>
 22 #include <linux/err.h>
 23 #include <linux/blkdev.h>
 24 #include <linux/backing-dev.h>
 25 #include <linux/slab.h>
 26 #include <linux/delay.h>
 27 #include <linux/atomic.h>
 28 #include <linux/ctype.h>
 29 #include <linux/resume_user_mode.h>
 30 #include <linux/psi.h>
 31 #include <linux/part_stat.h>
 32 #include "blk.h"
 33 #include "blk-cgroup.h"
 34 #include "blk-ioprio.h"
 35 #include "blk-throttle.h"
 36 
 37 /*
 38  * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation.
 39  * blkcg_pol_register_mutex nests outside of it and synchronizes entire
 40  * policy [un]register operations including cgroup file additions /
 41  * removals.  Putting cgroup file registration outside blkcg_pol_mutex
 42  * allows grabbing it from cgroup callbacks.
 43  */
 44 static DEFINE_MUTEX(blkcg_pol_register_mutex);
 45 static DEFINE_MUTEX(blkcg_pol_mutex);
 46 
 47 struct blkcg blkcg_root;
 48 EXPORT_SYMBOL_GPL(blkcg_root);
 49 
 50 struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css;
 51 EXPORT_SYMBOL_GPL(blkcg_root_css);
 52 
 53 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
 54 
 55 static LIST_HEAD(all_blkcgs);           /* protected by blkcg_pol_mutex */
 56 
 57 bool blkcg_debug_stats = false;
 58 static struct workqueue_struct *blkcg_punt_bio_wq;
 59 
 60 #define BLKG_DESTROY_BATCH_SIZE  64
 61 
 62 /**
 63  * blkcg_css - find the current css
 64  *
 65  * Find the css associated with either the kthread or the current task.
 66  * This may return a dying css, so it is up to the caller to use tryget logic
 67  * to confirm it is alive and well.
 68  */
 69 static struct cgroup_subsys_state *blkcg_css(void)
 70 {
 71         struct cgroup_subsys_state *css;
 72 
 73         css = kthread_blkcg();
 74         if (css)
 75                 return css;
 76         return task_css(current, io_cgrp_id);
 77 }
 78 
 79 static bool blkcg_policy_enabled(struct request_queue *q,
 80                                  const struct blkcg_policy *pol)
 81 {
 82         return pol && test_bit(pol->plid, q->blkcg_pols);
 83 }
 84 
 85 static void blkg_free_workfn(struct work_struct *work)
 86 {
 87         struct blkcg_gq *blkg = container_of(work, struct blkcg_gq,
 88                                              free_work);
 89         int i;
 90 
 91         for (i = 0; i < BLKCG_MAX_POLS; i++)
 92                 if (blkg->pd[i])
 93                         blkcg_policy[i]->pd_free_fn(blkg->pd[i]);
 94 
 95         if (blkg->q)
 96                 blk_put_queue(blkg->q);
 97         free_percpu(blkg->iostat_cpu);
 98         percpu_ref_exit(&blkg->refcnt);
 99         kfree(blkg);
100 }
101 
102 /**
103  * blkg_free - free a blkg
104  * @blkg: blkg to free
105  *
106  * Free @blkg which may be partially allocated.
107  */
108 static void blkg_free(struct blkcg_gq *blkg)
109 {
110         if (!blkg)
111                 return;
112 
113         /*
114          * Both ->pd_free_fn() and request queue's release handler may
115          * sleep, so free us by scheduling one work func
116          */
117         INIT_WORK(&blkg->free_work, blkg_free_workfn);
118         schedule_work(&blkg->free_work);
119 }
120 
121 static void __blkg_release(struct rcu_head *rcu)
122 {
123         struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head);
124 
125         WARN_ON(!bio_list_empty(&blkg->async_bios));
126 
127         /* release the blkcg and parent blkg refs this blkg has been holding */
128         css_put(&blkg->blkcg->css);
129         if (blkg->parent)
130                 blkg_put(blkg->parent);
131         blkg_free(blkg);
132 }
133 
134 /*
135  * A group is RCU protected, but having an rcu lock does not mean that one
136  * can access all the fields of blkg and assume these are valid.  For
137  * example, don't try to follow throtl_data and request queue links.
138  *
139  * Having a reference to blkg under an rcu allows accesses to only values
140  * local to groups like group stats and group rate limits.
141  */
142 static void blkg_release(struct percpu_ref *ref)
143 {
144         struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt);
145 
146         call_rcu(&blkg->rcu_head, __blkg_release);
147 }
148 
149 static void blkg_async_bio_workfn(struct work_struct *work)
150 {
151         struct blkcg_gq *blkg = container_of(work, struct blkcg_gq,
152                                              async_bio_work);
153         struct bio_list bios = BIO_EMPTY_LIST;
154         struct bio *bio;
155         struct blk_plug plug;
156         bool need_plug = false;
157 
158         /* as long as there are pending bios, @blkg can't go away */
159         spin_lock_bh(&blkg->async_bio_lock);
160         bio_list_merge(&bios, &blkg->async_bios);
161         bio_list_init(&blkg->async_bios);
162         spin_unlock_bh(&blkg->async_bio_lock);
163 
164         /* start plug only when bio_list contains at least 2 bios */
165         if (bios.head && bios.head->bi_next) {
166                 need_plug = true;
167                 blk_start_plug(&plug);
168         }
169         while ((bio = bio_list_pop(&bios)))
170                 submit_bio(bio);
171         if (need_plug)
172                 blk_finish_plug(&plug);
173 }
174 
175 /**
176  * bio_blkcg_css - return the blkcg CSS associated with a bio
177  * @bio: target bio
178  *
179  * This returns the CSS for the blkcg associated with a bio, or %NULL if not
180  * associated. Callers are expected to either handle %NULL or know association
181  * has been done prior to calling this.
182  */
183 struct cgroup_subsys_state *bio_blkcg_css(struct bio *bio)
184 {
185         if (!bio || !bio->bi_blkg)
186                 return NULL;
187         return &bio->bi_blkg->blkcg->css;
188 }
189 EXPORT_SYMBOL_GPL(bio_blkcg_css);
190 
191 /**
192  * blkcg_parent - get the parent of a blkcg
193  * @blkcg: blkcg of interest
194  *
195  * Return the parent blkcg of @blkcg.  Can be called anytime.
196  */
197 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
198 {
199         return css_to_blkcg(blkcg->css.parent);
200 }
201 
202 /**
203  * blkg_alloc - allocate a blkg
204  * @blkcg: block cgroup the new blkg is associated with
205  * @q: request_queue the new blkg is associated with
206  * @gfp_mask: allocation mask to use
207  *
208  * Allocate a new blkg assocating @blkcg and @q.
209  */
210 static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
211                                    gfp_t gfp_mask)
212 {
213         struct blkcg_gq *blkg;
214         int i, cpu;
215 
216         /* alloc and init base part */
217         blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
218         if (!blkg)
219                 return NULL;
220 
221         if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask))
222                 goto err_free;
223 
224         blkg->iostat_cpu = alloc_percpu_gfp(struct blkg_iostat_set, gfp_mask);
225         if (!blkg->iostat_cpu)
226                 goto err_free;
227 
228         if (!blk_get_queue(q))
229                 goto err_free;
230 
231         blkg->q = q;
232         INIT_LIST_HEAD(&blkg->q_node);
233         spin_lock_init(&blkg->async_bio_lock);
234         bio_list_init(&blkg->async_bios);
235         INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn);
236         blkg->blkcg = blkcg;
237 
238         u64_stats_init(&blkg->iostat.sync);
239         for_each_possible_cpu(cpu)
240                 u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync);
241 
242         for (i = 0; i < BLKCG_MAX_POLS; i++) {
243                 struct blkcg_policy *pol = blkcg_policy[i];
244                 struct blkg_policy_data *pd;
245 
246                 if (!blkcg_policy_enabled(q, pol))
247                         continue;
248 
249                 /* alloc per-policy data and attach it to blkg */
250                 pd = pol->pd_alloc_fn(gfp_mask, q, blkcg);
251                 if (!pd)
252                         goto err_free;
253 
254                 blkg->pd[i] = pd;
255                 pd->blkg = blkg;
256                 pd->plid = i;
257         }
258 
259         return blkg;
260 
261 err_free:
262         blkg_free(blkg);
263         return NULL;
264 }
265 
266 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
267                                       struct request_queue *q, bool update_hint)
268 {
269         struct blkcg_gq *blkg;
270 
271         /*
272          * Hint didn't match.  Look up from the radix tree.  Note that the
273          * hint can only be updated under queue_lock as otherwise @blkg
274          * could have already been removed from blkg_tree.  The caller is
275          * responsible for grabbing queue_lock if @update_hint.
276          */
277         blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
278         if (blkg && blkg->q == q) {
279                 if (update_hint) {
280                         lockdep_assert_held(&q->queue_lock);
281                         rcu_assign_pointer(blkcg->blkg_hint, blkg);
282                 }
283                 return blkg;
284         }
285 
286         return NULL;
287 }
288 EXPORT_SYMBOL_GPL(blkg_lookup_slowpath);
289 
290 /*
291  * If @new_blkg is %NULL, this function tries to allocate a new one as
292  * necessary using %GFP_NOWAIT.  @new_blkg is always consumed on return.
293  */
294 static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
295                                     struct request_queue *q,
296                                     struct blkcg_gq *new_blkg)
297 {
298         struct blkcg_gq *blkg;
299         int i, ret;
300 
301         lockdep_assert_held(&q->queue_lock);
302 
303         /* request_queue is dying, do not create/recreate a blkg */
304         if (blk_queue_dying(q)) {
305                 ret = -ENODEV;
306                 goto err_free_blkg;
307         }
308 
309         /* blkg holds a reference to blkcg */
310         if (!css_tryget_online(&blkcg->css)) {
311                 ret = -ENODEV;
312                 goto err_free_blkg;
313         }
314 
315         /* allocate */
316         if (!new_blkg) {
317                 new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN);
318                 if (unlikely(!new_blkg)) {
319                         ret = -ENOMEM;
320                         goto err_put_css;
321                 }
322         }
323         blkg = new_blkg;
324 
325         /* link parent */
326         if (blkcg_parent(blkcg)) {
327                 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
328                 if (WARN_ON_ONCE(!blkg->parent)) {
329                         ret = -ENODEV;
330                         goto err_put_css;
331                 }
332                 blkg_get(blkg->parent);
333         }
334 
335         /* invoke per-policy init */
336         for (i = 0; i < BLKCG_MAX_POLS; i++) {
337                 struct blkcg_policy *pol = blkcg_policy[i];
338 
339                 if (blkg->pd[i] && pol->pd_init_fn)
340                         pol->pd_init_fn(blkg->pd[i]);
341         }
342 
343         /* insert */
344         spin_lock(&blkcg->lock);
345         ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
346         if (likely(!ret)) {
347                 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
348                 list_add(&blkg->q_node, &q->blkg_list);
349 
350                 for (i = 0; i < BLKCG_MAX_POLS; i++) {
351                         struct blkcg_policy *pol = blkcg_policy[i];
352 
353                         if (blkg->pd[i] && pol->pd_online_fn)
354                                 pol->pd_online_fn(blkg->pd[i]);
355                 }
356         }
357         blkg->online = true;
358         spin_unlock(&blkcg->lock);
359 
360         if (!ret)
361                 return blkg;
362 
363         /* @blkg failed fully initialized, use the usual release path */
364         blkg_put(blkg);
365         return ERR_PTR(ret);
366 
367 err_put_css:
368         css_put(&blkcg->css);
369 err_free_blkg:
370         blkg_free(new_blkg);
371         return ERR_PTR(ret);
372 }
373 
374 /**
375  * blkg_lookup_create - lookup blkg, try to create one if not there
376  * @blkcg: blkcg of interest
377  * @q: request_queue of interest
378  *
379  * Lookup blkg for the @blkcg - @q pair.  If it doesn't exist, try to
380  * create one.  blkg creation is performed recursively from blkcg_root such
381  * that all non-root blkg's have access to the parent blkg.  This function
382  * should be called under RCU read lock and takes @q->queue_lock.
383  *
384  * Returns the blkg or the closest blkg if blkg_create() fails as it walks
385  * down from root.
386  */
387 static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
388                 struct request_queue *q)
389 {
390         struct blkcg_gq *blkg;
391         unsigned long flags;
392 
393         WARN_ON_ONCE(!rcu_read_lock_held());
394 
395         blkg = blkg_lookup(blkcg, q);
396         if (blkg)
397                 return blkg;
398 
399         spin_lock_irqsave(&q->queue_lock, flags);
400         blkg = __blkg_lookup(blkcg, q, true);
401         if (blkg)
402                 goto found;
403 
404         /*
405          * Create blkgs walking down from blkcg_root to @blkcg, so that all
406          * non-root blkgs have access to their parents.  Returns the closest
407          * blkg to the intended blkg should blkg_create() fail.
408          */
409         while (true) {
410                 struct blkcg *pos = blkcg;
411                 struct blkcg *parent = blkcg_parent(blkcg);
412                 struct blkcg_gq *ret_blkg = q->root_blkg;
413 
414                 while (parent) {
415                         blkg = __blkg_lookup(parent, q, false);
416                         if (blkg) {
417                                 /* remember closest blkg */
418                                 ret_blkg = blkg;
419                                 break;
420                         }
421                         pos = parent;
422                         parent = blkcg_parent(parent);
423                 }
424 
425                 blkg = blkg_create(pos, q, NULL);
426                 if (IS_ERR(blkg)) {
427                         blkg = ret_blkg;
428                         break;
429                 }
430                 if (pos == blkcg)
431                         break;
432         }
433 
434 found:
435         spin_unlock_irqrestore(&q->queue_lock, flags);
436         return blkg;
437 }
438 
439 static void blkg_destroy(struct blkcg_gq *blkg)
440 {
441         struct blkcg *blkcg = blkg->blkcg;
442         int i;
443 
444         lockdep_assert_held(&blkg->q->queue_lock);
445         lockdep_assert_held(&blkcg->lock);
446 
447         /* Something wrong if we are trying to remove same group twice */
448         WARN_ON_ONCE(list_empty(&blkg->q_node));
449         WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
450 
451         for (i = 0; i < BLKCG_MAX_POLS; i++) {
452                 struct blkcg_policy *pol = blkcg_policy[i];
453 
454                 if (blkg->pd[i] && pol->pd_offline_fn)
455                         pol->pd_offline_fn(blkg->pd[i]);
456         }
457 
458         blkg->online = false;
459 
460         radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
461         list_del_init(&blkg->q_node);
462         hlist_del_init_rcu(&blkg->blkcg_node);
463 
464         /*
465          * Both setting lookup hint to and clearing it from @blkg are done
466          * under queue_lock.  If it's not pointing to @blkg now, it never
467          * will.  Hint assignment itself can race safely.
468          */
469         if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
470                 rcu_assign_pointer(blkcg->blkg_hint, NULL);
471 
472         /*
473          * Put the reference taken at the time of creation so that when all
474          * queues are gone, group can be destroyed.
475          */
476         percpu_ref_kill(&blkg->refcnt);
477 }
478 
479 /**
480  * blkg_destroy_all - destroy all blkgs associated with a request_queue
481  * @q: request_queue of interest
482  *
483  * Destroy all blkgs associated with @q.
484  */
485 static void blkg_destroy_all(struct request_queue *q)
486 {
487         struct blkcg_gq *blkg, *n;
488         int count = BLKG_DESTROY_BATCH_SIZE;
489 
490 restart:
491         spin_lock_irq(&q->queue_lock);
492         list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
493                 struct blkcg *blkcg = blkg->blkcg;
494 
495                 spin_lock(&blkcg->lock);
496                 blkg_destroy(blkg);
497                 spin_unlock(&blkcg->lock);
498 
499                 /*
500                  * in order to avoid holding the spin lock for too long, release
501                  * it when a batch of blkgs are destroyed.
502                  */
503                 if (!(--count)) {
504                         count = BLKG_DESTROY_BATCH_SIZE;
505                         spin_unlock_irq(&q->queue_lock);
506                         cond_resched();
507                         goto restart;
508                 }
509         }
510 
511         q->root_blkg = NULL;
512         spin_unlock_irq(&q->queue_lock);
513 }
514 
515 static int blkcg_reset_stats(struct cgroup_subsys_state *css,
516                              struct cftype *cftype, u64 val)
517 {
518         struct blkcg *blkcg = css_to_blkcg(css);
519         struct blkcg_gq *blkg;
520         int i, cpu;
521 
522         mutex_lock(&blkcg_pol_mutex);
523         spin_lock_irq(&blkcg->lock);
524 
525         /*
526          * Note that stat reset is racy - it doesn't synchronize against
527          * stat updates.  This is a debug feature which shouldn't exist
528          * anyway.  If you get hit by a race, retry.
529          */
530         hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
531                 for_each_possible_cpu(cpu) {
532                         struct blkg_iostat_set *bis =
533                                 per_cpu_ptr(blkg->iostat_cpu, cpu);
534                         memset(bis, 0, sizeof(*bis));
535                 }
536                 memset(&blkg->iostat, 0, sizeof(blkg->iostat));
537 
538                 for (i = 0; i < BLKCG_MAX_POLS; i++) {
539                         struct blkcg_policy *pol = blkcg_policy[i];
540 
541                         if (blkg->pd[i] && pol->pd_reset_stats_fn)
542                                 pol->pd_reset_stats_fn(blkg->pd[i]);
543                 }
544         }
545 
546         spin_unlock_irq(&blkcg->lock);
547         mutex_unlock(&blkcg_pol_mutex);
548         return 0;
549 }
550 
551 const char *blkg_dev_name(struct blkcg_gq *blkg)
552 {
553         if (!blkg->q->disk || !blkg->q->disk->bdi->dev)
554                 return NULL;
555         return bdi_dev_name(blkg->q->disk->bdi);
556 }
557 
558 /**
559  * blkcg_print_blkgs - helper for printing per-blkg data
560  * @sf: seq_file to print to
561  * @blkcg: blkcg of interest
562  * @prfill: fill function to print out a blkg
563  * @pol: policy in question
564  * @data: data to be passed to @prfill
565  * @show_total: to print out sum of prfill return values or not
566  *
567  * This function invokes @prfill on each blkg of @blkcg if pd for the
568  * policy specified by @pol exists.  @prfill is invoked with @sf, the
569  * policy data and @data and the matching queue lock held.  If @show_total
570  * is %true, the sum of the return values from @prfill is printed with
571  * "Total" label at the end.
572  *
573  * This is to be used to construct print functions for
574  * cftype->read_seq_string method.
575  */
576 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
577                        u64 (*prfill)(struct seq_file *,
578                                      struct blkg_policy_data *, int),
579                        const struct blkcg_policy *pol, int data,
580                        bool show_total)
581 {
582         struct blkcg_gq *blkg;
583         u64 total = 0;
584 
585         rcu_read_lock();
586         hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
587                 spin_lock_irq(&blkg->q->queue_lock);
588                 if (blkcg_policy_enabled(blkg->q, pol))
589                         total += prfill(sf, blkg->pd[pol->plid], data);
590                 spin_unlock_irq(&blkg->q->queue_lock);
591         }
592         rcu_read_unlock();
593 
594         if (show_total)
595                 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
596 }
597 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
598 
599 /**
600  * __blkg_prfill_u64 - prfill helper for a single u64 value
601  * @sf: seq_file to print to
602  * @pd: policy private data of interest
603  * @v: value to print
604  *
605  * Print @v to @sf for the device assocaited with @pd.
606  */
607 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
608 {
609         const char *dname = blkg_dev_name(pd->blkg);
610 
611         if (!dname)
612                 return 0;
613 
614         seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
615         return v;
616 }
617 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
618 
619 /* Performs queue bypass and policy enabled checks then looks up blkg. */
620 static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg,
621                                           const struct blkcg_policy *pol,
622                                           struct request_queue *q)
623 {
624         WARN_ON_ONCE(!rcu_read_lock_held());
625         lockdep_assert_held(&q->queue_lock);
626 
627         if (!blkcg_policy_enabled(q, pol))
628                 return ERR_PTR(-EOPNOTSUPP);
629         return __blkg_lookup(blkcg, q, true /* update_hint */);
630 }
631 
632 /**
633  * blkcg_conf_open_bdev - parse and open bdev for per-blkg config update
634  * @inputp: input string pointer
635  *
636  * Parse the device node prefix part, MAJ:MIN, of per-blkg config update
637  * from @input and get and return the matching bdev.  *@inputp is
638  * updated to point past the device node prefix.  Returns an ERR_PTR()
639  * value on error.
640  *
641  * Use this function iff blkg_conf_prep() can't be used for some reason.
642  */
643 struct block_device *blkcg_conf_open_bdev(char **inputp)
644 {
645         char *input = *inputp;
646         unsigned int major, minor;
647         struct block_device *bdev;
648         int key_len;
649 
650         if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
651                 return ERR_PTR(-EINVAL);
652 
653         input += key_len;
654         if (!isspace(*input))
655                 return ERR_PTR(-EINVAL);
656         input = skip_spaces(input);
657 
658         bdev = blkdev_get_no_open(MKDEV(major, minor));
659         if (!bdev)
660                 return ERR_PTR(-ENODEV);
661         if (bdev_is_partition(bdev)) {
662                 blkdev_put_no_open(bdev);
663                 return ERR_PTR(-ENODEV);
664         }
665 
666         *inputp = input;
667         return bdev;
668 }
669 
670 /**
671  * blkg_conf_prep - parse and prepare for per-blkg config update
672  * @blkcg: target block cgroup
673  * @pol: target policy
674  * @input: input string
675  * @ctx: blkg_conf_ctx to be filled
676  *
677  * Parse per-blkg config update from @input and initialize @ctx with the
678  * result.  @ctx->blkg points to the blkg to be updated and @ctx->body the
679  * part of @input following MAJ:MIN.  This function returns with RCU read
680  * lock and queue lock held and must be paired with blkg_conf_finish().
681  */
682 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
683                    char *input, struct blkg_conf_ctx *ctx)
684         __acquires(rcu) __acquires(&bdev->bd_queue->queue_lock)
685 {
686         struct block_device *bdev;
687         struct request_queue *q;
688         struct blkcg_gq *blkg;
689         int ret;
690 
691         bdev = blkcg_conf_open_bdev(&input);
692         if (IS_ERR(bdev))
693                 return PTR_ERR(bdev);
694 
695         q = bdev_get_queue(bdev);
696 
697         /*
698          * blkcg_deactivate_policy() requires queue to be frozen, we can grab
699          * q_usage_counter to prevent concurrent with blkcg_deactivate_policy().
700          */
701         ret = blk_queue_enter(q, 0);
702         if (ret)
703                 goto fail;
704 
705         rcu_read_lock();
706         spin_lock_irq(&q->queue_lock);
707 
708         blkg = blkg_lookup_check(blkcg, pol, q);
709         if (IS_ERR(blkg)) {
710                 ret = PTR_ERR(blkg);
711                 goto fail_unlock;
712         }
713 
714         if (blkg)
715                 goto success;
716 
717         /*
718          * Create blkgs walking down from blkcg_root to @blkcg, so that all
719          * non-root blkgs have access to their parents.
720          */
721         while (true) {
722                 struct blkcg *pos = blkcg;
723                 struct blkcg *parent;
724                 struct blkcg_gq *new_blkg;
725 
726                 parent = blkcg_parent(blkcg);
727                 while (parent && !__blkg_lookup(parent, q, false)) {
728                         pos = parent;
729                         parent = blkcg_parent(parent);
730                 }
731 
732                 /* Drop locks to do new blkg allocation with GFP_KERNEL. */
733                 spin_unlock_irq(&q->queue_lock);
734                 rcu_read_unlock();
735 
736                 new_blkg = blkg_alloc(pos, q, GFP_KERNEL);
737                 if (unlikely(!new_blkg)) {
738                         ret = -ENOMEM;
739                         goto fail_exit_queue;
740                 }
741 
742                 if (radix_tree_preload(GFP_KERNEL)) {
743                         blkg_free(new_blkg);
744                         ret = -ENOMEM;
745                         goto fail_exit_queue;
746                 }
747 
748                 rcu_read_lock();
749                 spin_lock_irq(&q->queue_lock);
750 
751                 blkg = blkg_lookup_check(pos, pol, q);
752                 if (IS_ERR(blkg)) {
753                         ret = PTR_ERR(blkg);
754                         blkg_free(new_blkg);
755                         goto fail_preloaded;
756                 }
757 
758                 if (blkg) {
759                         blkg_free(new_blkg);
760                 } else {
761                         blkg = blkg_create(pos, q, new_blkg);
762                         if (IS_ERR(blkg)) {
763                                 ret = PTR_ERR(blkg);
764                                 goto fail_preloaded;
765                         }
766                 }
767 
768                 radix_tree_preload_end();
769 
770                 if (pos == blkcg)
771                         goto success;
772         }
773 success:
774         blk_queue_exit(q);
775         ctx->bdev = bdev;
776         ctx->blkg = blkg;
777         ctx->body = input;
778         return 0;
779 
780 fail_preloaded:
781         radix_tree_preload_end();
782 fail_unlock:
783         spin_unlock_irq(&q->queue_lock);
784         rcu_read_unlock();
785 fail_exit_queue:
786         blk_queue_exit(q);
787 fail:
788         blkdev_put_no_open(bdev);
789         /*
790          * If queue was bypassing, we should retry.  Do so after a
791          * short msleep().  It isn't strictly necessary but queue
792          * can be bypassing for some time and it's always nice to
793          * avoid busy looping.
794          */
795         if (ret == -EBUSY) {
796                 msleep(10);
797                 ret = restart_syscall();
798         }
799         return ret;
800 }
801 EXPORT_SYMBOL_GPL(blkg_conf_prep);
802 
803 /**
804  * blkg_conf_finish - finish up per-blkg config update
805  * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
806  *
807  * Finish up after per-blkg config update.  This function must be paired
808  * with blkg_conf_prep().
809  */
810 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
811         __releases(&ctx->bdev->bd_queue->queue_lock) __releases(rcu)
812 {
813         spin_unlock_irq(&bdev_get_queue(ctx->bdev)->queue_lock);
814         rcu_read_unlock();
815         blkdev_put_no_open(ctx->bdev);
816 }
817 EXPORT_SYMBOL_GPL(blkg_conf_finish);
818 
819 static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src)
820 {
821         int i;
822 
823         for (i = 0; i < BLKG_IOSTAT_NR; i++) {
824                 dst->bytes[i] = src->bytes[i];
825                 dst->ios[i] = src->ios[i];
826         }
827 }
828 
829 static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src)
830 {
831         int i;
832 
833         for (i = 0; i < BLKG_IOSTAT_NR; i++) {
834                 dst->bytes[i] += src->bytes[i];
835                 dst->ios[i] += src->ios[i];
836         }
837 }
838 
839 static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src)
840 {
841         int i;
842 
843         for (i = 0; i < BLKG_IOSTAT_NR; i++) {
844                 dst->bytes[i] -= src->bytes[i];
845                 dst->ios[i] -= src->ios[i];
846         }
847 }
848 
849 static void blkcg_iostat_update(struct blkcg_gq *blkg, struct blkg_iostat *cur,
850                                 struct blkg_iostat *last)
851 {
852         struct blkg_iostat delta;
853         unsigned long flags;
854 
855         /* propagate percpu delta to global */
856         flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync);
857         blkg_iostat_set(&delta, cur);
858         blkg_iostat_sub(&delta, last);
859         blkg_iostat_add(&blkg->iostat.cur, &delta);
860         blkg_iostat_add(last, &delta);
861         u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags);
862 }
863 
864 static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu)
865 {
866         struct blkcg *blkcg = css_to_blkcg(css);
867         struct blkcg_gq *blkg;
868 
869         /* Root-level stats are sourced from system-wide IO stats */
870         if (!cgroup_parent(css->cgroup))
871                 return;
872 
873         rcu_read_lock();
874 
875         hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
876                 struct blkcg_gq *parent = blkg->parent;
877                 struct blkg_iostat_set *bisc = per_cpu_ptr(blkg->iostat_cpu, cpu);
878                 struct blkg_iostat cur;
879                 unsigned int seq;
880 
881                 /* fetch the current per-cpu values */
882                 do {
883                         seq = u64_stats_fetch_begin(&bisc->sync);
884                         blkg_iostat_set(&cur, &bisc->cur);
885                 } while (u64_stats_fetch_retry(&bisc->sync, seq));
886 
887                 blkcg_iostat_update(blkg, &cur, &bisc->last);
888 
889                 /* propagate global delta to parent (unless that's root) */
890                 if (parent && parent->parent)
891                         blkcg_iostat_update(parent, &blkg->iostat.cur,
892                                             &blkg->iostat.last);
893         }
894 
895         rcu_read_unlock();
896 }
897 
898 /*
899  * We source root cgroup stats from the system-wide stats to avoid
900  * tracking the same information twice and incurring overhead when no
901  * cgroups are defined. For that reason, cgroup_rstat_flush in
902  * blkcg_print_stat does not actually fill out the iostat in the root
903  * cgroup's blkcg_gq.
904  *
905  * However, we would like to re-use the printing code between the root and
906  * non-root cgroups to the extent possible. For that reason, we simulate
907  * flushing the root cgroup's stats by explicitly filling in the iostat
908  * with disk level statistics.
909  */
910 static void blkcg_fill_root_iostats(void)
911 {
912         struct class_dev_iter iter;
913         struct device *dev;
914 
915         class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
916         while ((dev = class_dev_iter_next(&iter))) {
917                 struct block_device *bdev = dev_to_bdev(dev);
918                 struct blkcg_gq *blkg =
919                         blk_queue_root_blkg(bdev_get_queue(bdev));
920                 struct blkg_iostat tmp;
921                 int cpu;
922                 unsigned long flags;
923 
924                 memset(&tmp, 0, sizeof(tmp));
925                 for_each_possible_cpu(cpu) {
926                         struct disk_stats *cpu_dkstats;
927 
928                         cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu);
929                         tmp.ios[BLKG_IOSTAT_READ] +=
930                                 cpu_dkstats->ios[STAT_READ];
931                         tmp.ios[BLKG_IOSTAT_WRITE] +=
932                                 cpu_dkstats->ios[STAT_WRITE];
933                         tmp.ios[BLKG_IOSTAT_DISCARD] +=
934                                 cpu_dkstats->ios[STAT_DISCARD];
935                         // convert sectors to bytes
936                         tmp.bytes[BLKG_IOSTAT_READ] +=
937                                 cpu_dkstats->sectors[STAT_READ] << 9;
938                         tmp.bytes[BLKG_IOSTAT_WRITE] +=
939                                 cpu_dkstats->sectors[STAT_WRITE] << 9;
940                         tmp.bytes[BLKG_IOSTAT_DISCARD] +=
941                                 cpu_dkstats->sectors[STAT_DISCARD] << 9;
942                 }
943 
944                 flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync);
945                 blkg_iostat_set(&blkg->iostat.cur, &tmp);
946                 u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags);
947         }
948 }
949 
950 static void blkcg_print_one_stat(struct blkcg_gq *blkg, struct seq_file *s)
951 {
952         struct blkg_iostat_set *bis = &blkg->iostat;
953         u64 rbytes, wbytes, rios, wios, dbytes, dios;
954         const char *dname;
955         unsigned seq;
956         int i;
957 
958         if (!blkg->online)
959                 return;
960 
961         dname = blkg_dev_name(blkg);
962         if (!dname)
963                 return;
964 
965         seq_printf(s, "%s ", dname);
966 
967         do {
968                 seq = u64_stats_fetch_begin(&bis->sync);
969 
970                 rbytes = bis->cur.bytes[BLKG_IOSTAT_READ];
971                 wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE];
972                 dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD];
973                 rios = bis->cur.ios[BLKG_IOSTAT_READ];
974                 wios = bis->cur.ios[BLKG_IOSTAT_WRITE];
975                 dios = bis->cur.ios[BLKG_IOSTAT_DISCARD];
976         } while (u64_stats_fetch_retry(&bis->sync, seq));
977 
978         if (rbytes || wbytes || rios || wios) {
979                 seq_printf(s, "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu",
980                         rbytes, wbytes, rios, wios,
981                         dbytes, dios);
982         }
983 
984         if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) {
985                 seq_printf(s, " use_delay=%d delay_nsec=%llu",
986                         atomic_read(&blkg->use_delay),
987                         atomic64_read(&blkg->delay_nsec));
988         }
989 
990         for (i = 0; i < BLKCG_MAX_POLS; i++) {
991                 struct blkcg_policy *pol = blkcg_policy[i];
992 
993                 if (!blkg->pd[i] || !pol->pd_stat_fn)
994                         continue;
995 
996                 pol->pd_stat_fn(blkg->pd[i], s);
997         }
998 
999         seq_puts(s, "\n");
1000 }
1001 
1002 static int blkcg_print_stat(struct seq_file *sf, void *v)
1003 {
1004         struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
1005         struct blkcg_gq *blkg;
1006 
1007         if (!seq_css(sf)->parent)
1008                 blkcg_fill_root_iostats();
1009         else
1010                 cgroup_rstat_flush(blkcg->css.cgroup);
1011 
1012         rcu_read_lock();
1013         hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
1014                 spin_lock_irq(&blkg->q->queue_lock);
1015                 blkcg_print_one_stat(blkg, sf);
1016                 spin_unlock_irq(&blkg->q->queue_lock);
1017         }
1018         rcu_read_unlock();
1019         return 0;
1020 }
1021 
1022 static struct cftype blkcg_files[] = {
1023         {
1024                 .name = "stat",
1025                 .seq_show = blkcg_print_stat,
1026         },
1027         { }     /* terminate */
1028 };
1029 
1030 static struct cftype blkcg_legacy_files[] = {
1031         {
1032                 .name = "reset_stats",
1033                 .write_u64 = blkcg_reset_stats,
1034         },
1035         { }     /* terminate */
1036 };
1037 
1038 #ifdef CONFIG_CGROUP_WRITEBACK
1039 struct list_head *blkcg_get_cgwb_list(struct cgroup_subsys_state *css)
1040 {
1041         return &css_to_blkcg(css)->cgwb_list;
1042 }
1043 #endif
1044 
1045 /*
1046  * blkcg destruction is a three-stage process.
1047  *
1048  * 1. Destruction starts.  The blkcg_css_offline() callback is invoked
1049  *    which offlines writeback.  Here we tie the next stage of blkg destruction
1050  *    to the completion of writeback associated with the blkcg.  This lets us
1051  *    avoid punting potentially large amounts of outstanding writeback to root
1052  *    while maintaining any ongoing policies.  The next stage is triggered when
1053  *    the nr_cgwbs count goes to zero.
1054  *
1055  * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called
1056  *    and handles the destruction of blkgs.  Here the css reference held by
1057  *    the blkg is put back eventually allowing blkcg_css_free() to be called.
1058  *    This work may occur in cgwb_release_workfn() on the cgwb_release
1059  *    workqueue.  Any submitted ios that fail to get the blkg ref will be
1060  *    punted to the root_blkg.
1061  *
1062  * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called.
1063  *    This finally frees the blkcg.
1064  */
1065 
1066 /**
1067  * blkcg_destroy_blkgs - responsible for shooting down blkgs
1068  * @blkcg: blkcg of interest
1069  *
1070  * blkgs should be removed while holding both q and blkcg locks.  As blkcg lock
1071  * is nested inside q lock, this function performs reverse double lock dancing.
1072  * Destroying the blkgs releases the reference held on the blkcg's css allowing
1073  * blkcg_css_free to eventually be called.
1074  *
1075  * This is the blkcg counterpart of ioc_release_fn().
1076  */
1077 static void blkcg_destroy_blkgs(struct blkcg *blkcg)
1078 {
1079         might_sleep();
1080 
1081         spin_lock_irq(&blkcg->lock);
1082 
1083         while (!hlist_empty(&blkcg->blkg_list)) {
1084                 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
1085                                                 struct blkcg_gq, blkcg_node);
1086                 struct request_queue *q = blkg->q;
1087 
1088                 if (need_resched() || !spin_trylock(&q->queue_lock)) {
1089                         /*
1090                          * Given that the system can accumulate a huge number
1091                          * of blkgs in pathological cases, check to see if we
1092                          * need to rescheduling to avoid softlockup.
1093                          */
1094                         spin_unlock_irq(&blkcg->lock);
1095                         cond_resched();
1096                         spin_lock_irq(&blkcg->lock);
1097                         continue;
1098                 }
1099 
1100                 blkg_destroy(blkg);
1101                 spin_unlock(&q->queue_lock);
1102         }
1103 
1104         spin_unlock_irq(&blkcg->lock);
1105 }
1106 
1107 /**
1108  * blkcg_pin_online - pin online state
1109  * @blkcg_css: blkcg of interest
1110  *
1111  * While pinned, a blkcg is kept online.  This is primarily used to
1112  * impedance-match blkg and cgwb lifetimes so that blkg doesn't go offline
1113  * while an associated cgwb is still active.
1114  */
1115 void blkcg_pin_online(struct cgroup_subsys_state *blkcg_css)
1116 {
1117         refcount_inc(&css_to_blkcg(blkcg_css)->online_pin);
1118 }
1119 
1120 /**
1121  * blkcg_unpin_online - unpin online state
1122  * @blkcg_css: blkcg of interest
1123  *
1124  * This is primarily used to impedance-match blkg and cgwb lifetimes so
1125  * that blkg doesn't go offline while an associated cgwb is still active.
1126  * When this count goes to zero, all active cgwbs have finished so the
1127  * blkcg can continue destruction by calling blkcg_destroy_blkgs().
1128  */
1129 void blkcg_unpin_online(struct cgroup_subsys_state *blkcg_css)
1130 {
1131         struct blkcg *blkcg = css_to_blkcg(blkcg_css);
1132 
1133         do {
1134                 if (!refcount_dec_and_test(&blkcg->online_pin))
1135                         break;
1136                 blkcg_destroy_blkgs(blkcg);
1137                 blkcg = blkcg_parent(blkcg);
1138         } while (blkcg);
1139 }
1140 
1141 /**
1142  * blkcg_css_offline - cgroup css_offline callback
1143  * @css: css of interest
1144  *
1145  * This function is called when @css is about to go away.  Here the cgwbs are
1146  * offlined first and only once writeback associated with the blkcg has
1147  * finished do we start step 2 (see above).
1148  */
1149 static void blkcg_css_offline(struct cgroup_subsys_state *css)
1150 {
1151         /* this prevents anyone from attaching or migrating to this blkcg */
1152         wb_blkcg_offline(css);
1153 
1154         /* put the base online pin allowing step 2 to be triggered */
1155         blkcg_unpin_online(css);
1156 }
1157 
1158 static void blkcg_css_free(struct cgroup_subsys_state *css)
1159 {
1160         struct blkcg *blkcg = css_to_blkcg(css);
1161         int i;
1162 
1163         mutex_lock(&blkcg_pol_mutex);
1164 
1165         list_del(&blkcg->all_blkcgs_node);
1166 
1167         for (i = 0; i < BLKCG_MAX_POLS; i++)
1168                 if (blkcg->cpd[i])
1169                         blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1170 
1171         mutex_unlock(&blkcg_pol_mutex);
1172 
1173         kfree(blkcg);
1174 }
1175 
1176 static struct cgroup_subsys_state *
1177 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
1178 {
1179         struct blkcg *blkcg;
1180         struct cgroup_subsys_state *ret;
1181         int i;
1182 
1183         mutex_lock(&blkcg_pol_mutex);
1184 
1185         if (!parent_css) {
1186                 blkcg = &blkcg_root;
1187         } else {
1188                 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1189                 if (!blkcg) {
1190                         ret = ERR_PTR(-ENOMEM);
1191                         goto unlock;
1192                 }
1193         }
1194 
1195         for (i = 0; i < BLKCG_MAX_POLS ; i++) {
1196                 struct blkcg_policy *pol = blkcg_policy[i];
1197                 struct blkcg_policy_data *cpd;
1198 
1199                 /*
1200                  * If the policy hasn't been attached yet, wait for it
1201                  * to be attached before doing anything else. Otherwise,
1202                  * check if the policy requires any specific per-cgroup
1203                  * data: if it does, allocate and initialize it.
1204                  */
1205                 if (!pol || !pol->cpd_alloc_fn)
1206                         continue;
1207 
1208                 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1209                 if (!cpd) {
1210                         ret = ERR_PTR(-ENOMEM);
1211                         goto free_pd_blkcg;
1212                 }
1213                 blkcg->cpd[i] = cpd;
1214                 cpd->blkcg = blkcg;
1215                 cpd->plid = i;
1216                 if (pol->cpd_init_fn)
1217                         pol->cpd_init_fn(cpd);
1218         }
1219 
1220         spin_lock_init(&blkcg->lock);
1221         refcount_set(&blkcg->online_pin, 1);
1222         INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN);
1223         INIT_HLIST_HEAD(&blkcg->blkg_list);
1224 #ifdef CONFIG_CGROUP_WRITEBACK
1225         INIT_LIST_HEAD(&blkcg->cgwb_list);
1226 #endif
1227         list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
1228 
1229         mutex_unlock(&blkcg_pol_mutex);
1230         return &blkcg->css;
1231 
1232 free_pd_blkcg:
1233         for (i--; i >= 0; i--)
1234                 if (blkcg->cpd[i])
1235                         blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1236 
1237         if (blkcg != &blkcg_root)
1238                 kfree(blkcg);
1239 unlock:
1240         mutex_unlock(&blkcg_pol_mutex);
1241         return ret;
1242 }
1243 
1244 static int blkcg_css_online(struct cgroup_subsys_state *css)
1245 {
1246         struct blkcg *parent = blkcg_parent(css_to_blkcg(css));
1247 
1248         /*
1249          * blkcg_pin_online() is used to delay blkcg offline so that blkgs
1250          * don't go offline while cgwbs are still active on them.  Pin the
1251          * parent so that offline always happens towards the root.
1252          */
1253         if (parent)
1254                 blkcg_pin_online(css);
1255         return 0;
1256 }
1257 
1258 /**
1259  * blkcg_init_queue - initialize blkcg part of request queue
1260  * @q: request_queue to initialize
1261  *
1262  * Called from blk_alloc_queue(). Responsible for initializing blkcg
1263  * part of new request_queue @q.
1264  *
1265  * RETURNS:
1266  * 0 on success, -errno on failure.
1267  */
1268 int blkcg_init_queue(struct request_queue *q)
1269 {
1270         struct blkcg_gq *new_blkg, *blkg;
1271         bool preloaded;
1272         int ret;
1273 
1274         INIT_LIST_HEAD(&q->blkg_list);
1275 
1276         new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
1277         if (!new_blkg)
1278                 return -ENOMEM;
1279 
1280         preloaded = !radix_tree_preload(GFP_KERNEL);
1281 
1282         /* Make sure the root blkg exists. */
1283         /* spin_lock_irq can serve as RCU read-side critical section. */
1284         spin_lock_irq(&q->queue_lock);
1285         blkg = blkg_create(&blkcg_root, q, new_blkg);
1286         if (IS_ERR(blkg))
1287                 goto err_unlock;
1288         q->root_blkg = blkg;
1289         spin_unlock_irq(&q->queue_lock);
1290 
1291         if (preloaded)
1292                 radix_tree_preload_end();
1293 
1294         ret = blk_ioprio_init(q);
1295         if (ret)
1296                 goto err_destroy_all;
1297 
1298         ret = blk_throtl_init(q);
1299         if (ret)
1300                 goto err_destroy_all;
1301 
1302         ret = blk_iolatency_init(q);
1303         if (ret) {
1304                 blk_throtl_exit(q);
1305                 blk_ioprio_exit(q);
1306                 goto err_destroy_all;
1307         }
1308 
1309         return 0;
1310 
1311 err_destroy_all:
1312         blkg_destroy_all(q);
1313         return ret;
1314 err_unlock:
1315         spin_unlock_irq(&q->queue_lock);
1316         if (preloaded)
1317                 radix_tree_preload_end();
1318         return PTR_ERR(blkg);
1319 }
1320 
1321 /**
1322  * blkcg_exit_queue - exit and release blkcg part of request_queue
1323  * @q: request_queue being released
1324  *
1325  * Called from blk_exit_queue().  Responsible for exiting blkcg part.
1326  */
1327 void blkcg_exit_queue(struct request_queue *q)
1328 {
1329         blkg_destroy_all(q);
1330         blk_throtl_exit(q);
1331 }
1332 
1333 static void blkcg_bind(struct cgroup_subsys_state *root_css)
1334 {
1335         int i;
1336 
1337         mutex_lock(&blkcg_pol_mutex);
1338 
1339         for (i = 0; i < BLKCG_MAX_POLS; i++) {
1340                 struct blkcg_policy *pol = blkcg_policy[i];
1341                 struct blkcg *blkcg;
1342 
1343                 if (!pol || !pol->cpd_bind_fn)
1344                         continue;
1345 
1346                 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node)
1347                         if (blkcg->cpd[pol->plid])
1348                                 pol->cpd_bind_fn(blkcg->cpd[pol->plid]);
1349         }
1350         mutex_unlock(&blkcg_pol_mutex);
1351 }
1352 
1353 static void blkcg_exit(struct task_struct *tsk)
1354 {
1355         if (tsk->throttle_queue)
1356                 blk_put_queue(tsk->throttle_queue);
1357         tsk->throttle_queue = NULL;
1358 }
1359 
1360 struct cgroup_subsys io_cgrp_subsys = {
1361         .css_alloc = blkcg_css_alloc,
1362         .css_online = blkcg_css_online,
1363         .css_offline = blkcg_css_offline,
1364         .css_free = blkcg_css_free,
1365         .css_rstat_flush = blkcg_rstat_flush,
1366         .bind = blkcg_bind,
1367         .dfl_cftypes = blkcg_files,
1368         .legacy_cftypes = blkcg_legacy_files,
1369         .legacy_name = "blkio",
1370         .exit = blkcg_exit,
1371 #ifdef CONFIG_MEMCG
1372         /*
1373          * This ensures that, if available, memcg is automatically enabled
1374          * together on the default hierarchy so that the owner cgroup can
1375          * be retrieved from writeback pages.
1376          */
1377         .depends_on = 1 << memory_cgrp_id,
1378 #endif
1379 };
1380 EXPORT_SYMBOL_GPL(io_cgrp_subsys);
1381 
1382 /**
1383  * blkcg_activate_policy - activate a blkcg policy on a request_queue
1384  * @q: request_queue of interest
1385  * @pol: blkcg policy to activate
1386  *
1387  * Activate @pol on @q.  Requires %GFP_KERNEL context.  @q goes through
1388  * bypass mode to populate its blkgs with policy_data for @pol.
1389  *
1390  * Activation happens with @q bypassed, so nobody would be accessing blkgs
1391  * from IO path.  Update of each blkg is protected by both queue and blkcg
1392  * locks so that holding either lock and testing blkcg_policy_enabled() is
1393  * always enough for dereferencing policy data.
1394  *
1395  * The caller is responsible for synchronizing [de]activations and policy
1396  * [un]registerations.  Returns 0 on success, -errno on failure.
1397  */
1398 int blkcg_activate_policy(struct request_queue *q,
1399                           const struct blkcg_policy *pol)
1400 {
1401         struct blkg_policy_data *pd_prealloc = NULL;
1402         struct blkcg_gq *blkg, *pinned_blkg = NULL;
1403         int ret;
1404 
1405         if (blkcg_policy_enabled(q, pol))
1406                 return 0;
1407 
1408         if (queue_is_mq(q))
1409                 blk_mq_freeze_queue(q);
1410 retry:
1411         spin_lock_irq(&q->queue_lock);
1412 
1413         /* blkg_list is pushed at the head, reverse walk to allocate parents first */
1414         list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) {
1415                 struct blkg_policy_data *pd;
1416 
1417                 if (blkg->pd[pol->plid])
1418                         continue;
1419 
1420                 /* If prealloc matches, use it; otherwise try GFP_NOWAIT */
1421                 if (blkg == pinned_blkg) {
1422                         pd = pd_prealloc;
1423                         pd_prealloc = NULL;
1424                 } else {
1425                         pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q,
1426                                               blkg->blkcg);
1427                 }
1428 
1429                 if (!pd) {
1430                         /*
1431                          * GFP_NOWAIT failed.  Free the existing one and
1432                          * prealloc for @blkg w/ GFP_KERNEL.
1433                          */
1434                         if (pinned_blkg)
1435                                 blkg_put(pinned_blkg);
1436                         blkg_get(blkg);
1437                         pinned_blkg = blkg;
1438 
1439                         spin_unlock_irq(&q->queue_lock);
1440 
1441                         if (pd_prealloc)
1442                                 pol->pd_free_fn(pd_prealloc);
1443                         pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q,
1444                                                        blkg->blkcg);
1445                         if (pd_prealloc)
1446                                 goto retry;
1447                         else
1448                                 goto enomem;
1449                 }
1450 
1451                 blkg->pd[pol->plid] = pd;
1452                 pd->blkg = blkg;
1453                 pd->plid = pol->plid;
1454         }
1455 
1456         /* all allocated, init in the same order */
1457         if (pol->pd_init_fn)
1458                 list_for_each_entry_reverse(blkg, &q->blkg_list, q_node)
1459                         pol->pd_init_fn(blkg->pd[pol->plid]);
1460 
1461         __set_bit(pol->plid, q->blkcg_pols);
1462         ret = 0;
1463 
1464         spin_unlock_irq(&q->queue_lock);
1465 out:
1466         if (queue_is_mq(q))
1467                 blk_mq_unfreeze_queue(q);
1468         if (pinned_blkg)
1469                 blkg_put(pinned_blkg);
1470         if (pd_prealloc)
1471                 pol->pd_free_fn(pd_prealloc);
1472         return ret;
1473 
1474 enomem:
1475         /* alloc failed, nothing's initialized yet, free everything */
1476         spin_lock_irq(&q->queue_lock);
1477         list_for_each_entry(blkg, &q->blkg_list, q_node) {
1478                 struct blkcg *blkcg = blkg->blkcg;
1479 
1480                 spin_lock(&blkcg->lock);
1481                 if (blkg->pd[pol->plid]) {
1482                         pol->pd_free_fn(blkg->pd[pol->plid]);
1483                         blkg->pd[pol->plid] = NULL;
1484                 }
1485                 spin_unlock(&blkcg->lock);
1486         }
1487         spin_unlock_irq(&q->queue_lock);
1488         ret = -ENOMEM;
1489         goto out;
1490 }
1491 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1492 
1493 /**
1494  * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1495  * @q: request_queue of interest
1496  * @pol: blkcg policy to deactivate
1497  *
1498  * Deactivate @pol on @q.  Follows the same synchronization rules as
1499  * blkcg_activate_policy().
1500  */
1501 void blkcg_deactivate_policy(struct request_queue *q,
1502                              const struct blkcg_policy *pol)
1503 {
1504         struct blkcg_gq *blkg;
1505 
1506         if (!blkcg_policy_enabled(q, pol))
1507                 return;
1508 
1509         if (queue_is_mq(q))
1510                 blk_mq_freeze_queue(q);
1511 
1512         spin_lock_irq(&q->queue_lock);
1513 
1514         __clear_bit(pol->plid, q->blkcg_pols);
1515 
1516         list_for_each_entry(blkg, &q->blkg_list, q_node) {
1517                 struct blkcg *blkcg = blkg->blkcg;
1518 
1519                 spin_lock(&blkcg->lock);
1520                 if (blkg->pd[pol->plid]) {
1521                         if (pol->pd_offline_fn)
1522                                 pol->pd_offline_fn(blkg->pd[pol->plid]);
1523                         pol->pd_free_fn(blkg->pd[pol->plid]);
1524                         blkg->pd[pol->plid] = NULL;
1525                 }
1526                 spin_unlock(&blkcg->lock);
1527         }
1528 
1529         spin_unlock_irq(&q->queue_lock);
1530 
1531         if (queue_is_mq(q))
1532                 blk_mq_unfreeze_queue(q);
1533 }
1534 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1535 
1536 static void blkcg_free_all_cpd(struct blkcg_policy *pol)
1537 {
1538         struct blkcg *blkcg;
1539 
1540         list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1541                 if (blkcg->cpd[pol->plid]) {
1542                         pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1543                         blkcg->cpd[pol->plid] = NULL;
1544                 }
1545         }
1546 }
1547 
1548 /**
1549  * blkcg_policy_register - register a blkcg policy
1550  * @pol: blkcg policy to register
1551  *
1552  * Register @pol with blkcg core.  Might sleep and @pol may be modified on
1553  * successful registration.  Returns 0 on success and -errno on failure.
1554  */
1555 int blkcg_policy_register(struct blkcg_policy *pol)
1556 {
1557         struct blkcg *blkcg;
1558         int i, ret;
1559 
1560         mutex_lock(&blkcg_pol_register_mutex);
1561         mutex_lock(&blkcg_pol_mutex);
1562 
1563         /* find an empty slot */
1564         ret = -ENOSPC;
1565         for (i = 0; i < BLKCG_MAX_POLS; i++)
1566                 if (!blkcg_policy[i])
1567                         break;
1568         if (i >= BLKCG_MAX_POLS) {
1569                 pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n");
1570                 goto err_unlock;
1571         }
1572 
1573         /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */
1574         if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) ||
1575                 (!pol->pd_alloc_fn ^ !pol->pd_free_fn))
1576                 goto err_unlock;
1577 
1578         /* register @pol */
1579         pol->plid = i;
1580         blkcg_policy[pol->plid] = pol;
1581 
1582         /* allocate and install cpd's */
1583         if (pol->cpd_alloc_fn) {
1584                 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1585                         struct blkcg_policy_data *cpd;
1586 
1587                         cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1588                         if (!cpd)
1589                                 goto err_free_cpds;
1590 
1591                         blkcg->cpd[pol->plid] = cpd;
1592                         cpd->blkcg = blkcg;
1593                         cpd->plid = pol->plid;
1594                         if (pol->cpd_init_fn)
1595                                 pol->cpd_init_fn(cpd);
1596                 }
1597         }
1598 
1599         mutex_unlock(&blkcg_pol_mutex);
1600 
1601         /* everything is in place, add intf files for the new policy */
1602         if (pol->dfl_cftypes)
1603                 WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys,
1604                                                pol->dfl_cftypes));
1605         if (pol->legacy_cftypes)
1606                 WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys,
1607                                                   pol->legacy_cftypes));
1608         mutex_unlock(&blkcg_pol_register_mutex);
1609         return 0;
1610 
1611 err_free_cpds:
1612         if (pol->cpd_free_fn)
1613                 blkcg_free_all_cpd(pol);
1614 
1615         blkcg_policy[pol->plid] = NULL;
1616 err_unlock:
1617         mutex_unlock(&blkcg_pol_mutex);
1618         mutex_unlock(&blkcg_pol_register_mutex);
1619         return ret;
1620 }
1621 EXPORT_SYMBOL_GPL(blkcg_policy_register);
1622 
1623 /**
1624  * blkcg_policy_unregister - unregister a blkcg policy
1625  * @pol: blkcg policy to unregister
1626  *
1627  * Undo blkcg_policy_register(@pol).  Might sleep.
1628  */
1629 void blkcg_policy_unregister(struct blkcg_policy *pol)
1630 {
1631         mutex_lock(&blkcg_pol_register_mutex);
1632 
1633         if (WARN_ON(blkcg_policy[pol->plid] != pol))
1634                 goto out_unlock;
1635 
1636         /* kill the intf files first */
1637         if (pol->dfl_cftypes)
1638                 cgroup_rm_cftypes(pol->dfl_cftypes);
1639         if (pol->legacy_cftypes)
1640                 cgroup_rm_cftypes(pol->legacy_cftypes);
1641 
1642         /* remove cpds and unregister */
1643         mutex_lock(&blkcg_pol_mutex);
1644 
1645         if (pol->cpd_free_fn)
1646                 blkcg_free_all_cpd(pol);
1647 
1648         blkcg_policy[pol->plid] = NULL;
1649 
1650         mutex_unlock(&blkcg_pol_mutex);
1651 out_unlock:
1652         mutex_unlock(&blkcg_pol_register_mutex);
1653 }
1654 EXPORT_SYMBOL_GPL(blkcg_policy_unregister);
1655 
1656 bool __blkcg_punt_bio_submit(struct bio *bio)
1657 {
1658         struct blkcg_gq *blkg = bio->bi_blkg;
1659 
1660         /* consume the flag first */
1661         bio->bi_opf &= ~REQ_CGROUP_PUNT;
1662 
1663         /* never bounce for the root cgroup */
1664         if (!blkg->parent)
1665                 return false;
1666 
1667         spin_lock_bh(&blkg->async_bio_lock);
1668         bio_list_add(&blkg->async_bios, bio);
1669         spin_unlock_bh(&blkg->async_bio_lock);
1670 
1671         queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work);
1672         return true;
1673 }
1674 
1675 /*
1676  * Scale the accumulated delay based on how long it has been since we updated
1677  * the delay.  We only call this when we are adding delay, in case it's been a
1678  * while since we added delay, and when we are checking to see if we need to
1679  * delay a task, to account for any delays that may have occurred.
1680  */
1681 static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now)
1682 {
1683         u64 old = atomic64_read(&blkg->delay_start);
1684 
1685         /* negative use_delay means no scaling, see blkcg_set_delay() */
1686         if (atomic_read(&blkg->use_delay) < 0)
1687                 return;
1688 
1689         /*
1690          * We only want to scale down every second.  The idea here is that we
1691          * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain
1692          * time window.  We only want to throttle tasks for recent delay that
1693          * has occurred, in 1 second time windows since that's the maximum
1694          * things can be throttled.  We save the current delay window in
1695          * blkg->last_delay so we know what amount is still left to be charged
1696          * to the blkg from this point onward.  blkg->last_use keeps track of
1697          * the use_delay counter.  The idea is if we're unthrottling the blkg we
1698          * are ok with whatever is happening now, and we can take away more of
1699          * the accumulated delay as we've already throttled enough that
1700          * everybody is happy with their IO latencies.
1701          */
1702         if (time_before64(old + NSEC_PER_SEC, now) &&
1703             atomic64_try_cmpxchg(&blkg->delay_start, &old, now)) {
1704                 u64 cur = atomic64_read(&blkg->delay_nsec);
1705                 u64 sub = min_t(u64, blkg->last_delay, now - old);
1706                 int cur_use = atomic_read(&blkg->use_delay);
1707 
1708                 /*
1709                  * We've been unthrottled, subtract a larger chunk of our
1710                  * accumulated delay.
1711                  */
1712                 if (cur_use < blkg->last_use)
1713                         sub = max_t(u64, sub, blkg->last_delay >> 1);
1714 
1715                 /*
1716                  * This shouldn't happen, but handle it anyway.  Our delay_nsec
1717                  * should only ever be growing except here where we subtract out
1718                  * min(last_delay, 1 second), but lord knows bugs happen and I'd
1719                  * rather not end up with negative numbers.
1720                  */
1721                 if (unlikely(cur < sub)) {
1722                         atomic64_set(&blkg->delay_nsec, 0);
1723                         blkg->last_delay = 0;
1724                 } else {
1725                         atomic64_sub(sub, &blkg->delay_nsec);
1726                         blkg->last_delay = cur - sub;
1727                 }
1728                 blkg->last_use = cur_use;
1729         }
1730 }
1731 
1732 /*
1733  * This is called when we want to actually walk up the hierarchy and check to
1734  * see if we need to throttle, and then actually throttle if there is some
1735  * accumulated delay.  This should only be called upon return to user space so
1736  * we're not holding some lock that would induce a priority inversion.
1737  */
1738 static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay)
1739 {
1740         unsigned long pflags;
1741         bool clamp;
1742         u64 now = ktime_to_ns(ktime_get());
1743         u64 exp;
1744         u64 delay_nsec = 0;
1745         int tok;
1746 
1747         while (blkg->parent) {
1748                 int use_delay = atomic_read(&blkg->use_delay);
1749 
1750                 if (use_delay) {
1751                         u64 this_delay;
1752 
1753                         blkcg_scale_delay(blkg, now);
1754                         this_delay = atomic64_read(&blkg->delay_nsec);
1755                         if (this_delay > delay_nsec) {
1756                                 delay_nsec = this_delay;
1757                                 clamp = use_delay > 0;
1758                         }
1759                 }
1760                 blkg = blkg->parent;
1761         }
1762 
1763         if (!delay_nsec)
1764                 return;
1765 
1766         /*
1767          * Let's not sleep for all eternity if we've amassed a huge delay.
1768          * Swapping or metadata IO can accumulate 10's of seconds worth of
1769          * delay, and we want userspace to be able to do _something_ so cap the
1770          * delays at 0.25s. If there's 10's of seconds worth of delay then the
1771          * tasks will be delayed for 0.25 second for every syscall. If
1772          * blkcg_set_delay() was used as indicated by negative use_delay, the
1773          * caller is responsible for regulating the range.
1774          */
1775         if (clamp)
1776                 delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC);
1777 
1778         if (use_memdelay)
1779                 psi_memstall_enter(&pflags);
1780 
1781         exp = ktime_add_ns(now, delay_nsec);
1782         tok = io_schedule_prepare();
1783         do {
1784                 __set_current_state(TASK_KILLABLE);
1785                 if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS))
1786                         break;
1787         } while (!fatal_signal_pending(current));
1788         io_schedule_finish(tok);
1789 
1790         if (use_memdelay)
1791                 psi_memstall_leave(&pflags);
1792 }
1793 
1794 /**
1795  * blkcg_maybe_throttle_current - throttle the current task if it has been marked
1796  *
1797  * This is only called if we've been marked with set_notify_resume().  Obviously
1798  * we can be set_notify_resume() for reasons other than blkcg throttling, so we
1799  * check to see if current->throttle_queue is set and if not this doesn't do
1800  * anything.  This should only ever be called by the resume code, it's not meant
1801  * to be called by people willy-nilly as it will actually do the work to
1802  * throttle the task if it is setup for throttling.
1803  */
1804 void blkcg_maybe_throttle_current(void)
1805 {
1806         struct request_queue *q = current->throttle_queue;
1807         struct blkcg *blkcg;
1808         struct blkcg_gq *blkg;
1809         bool use_memdelay = current->use_memdelay;
1810 
1811         if (!q)
1812                 return;
1813 
1814         current->throttle_queue = NULL;
1815         current->use_memdelay = false;
1816 
1817         rcu_read_lock();
1818         blkcg = css_to_blkcg(blkcg_css());
1819         if (!blkcg)
1820                 goto out;
1821         blkg = blkg_lookup(blkcg, q);
1822         if (!blkg)
1823                 goto out;
1824         if (!blkg_tryget(blkg))
1825                 goto out;
1826         rcu_read_unlock();
1827 
1828         blkcg_maybe_throttle_blkg(blkg, use_memdelay);
1829         blkg_put(blkg);
1830         blk_put_queue(q);
1831         return;
1832 out:
1833         rcu_read_unlock();
1834         blk_put_queue(q);
1835 }
1836 
1837 /**
1838  * blkcg_schedule_throttle - this task needs to check for throttling
1839  * @q: the request queue IO was submitted on
1840  * @use_memdelay: do we charge this to memory delay for PSI
1841  *
1842  * This is called by the IO controller when we know there's delay accumulated
1843  * for the blkg for this task.  We do not pass the blkg because there are places
1844  * we call this that may not have that information, the swapping code for
1845  * instance will only have a request_queue at that point.  This set's the
1846  * notify_resume for the task to check and see if it requires throttling before
1847  * returning to user space.
1848  *
1849  * We will only schedule once per syscall.  You can call this over and over
1850  * again and it will only do the check once upon return to user space, and only
1851  * throttle once.  If the task needs to be throttled again it'll need to be
1852  * re-set at the next time we see the task.
1853  */
1854 void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay)
1855 {
1856         if (unlikely(current->flags & PF_KTHREAD))
1857                 return;
1858 
1859         if (current->throttle_queue != q) {
1860                 if (!blk_get_queue(q))
1861                         return;
1862 
1863                 if (current->throttle_queue)
1864                         blk_put_queue(current->throttle_queue);
1865                 current->throttle_queue = q;
1866         }
1867 
1868         if (use_memdelay)
1869                 current->use_memdelay = use_memdelay;
1870         set_notify_resume(current);
1871 }
1872 
1873 /**
1874  * blkcg_add_delay - add delay to this blkg
1875  * @blkg: blkg of interest
1876  * @now: the current time in nanoseconds
1877  * @delta: how many nanoseconds of delay to add
1878  *
1879  * Charge @delta to the blkg's current delay accumulation.  This is used to
1880  * throttle tasks if an IO controller thinks we need more throttling.
1881  */
1882 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta)
1883 {
1884         if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))
1885                 return;
1886         blkcg_scale_delay(blkg, now);
1887         atomic64_add(delta, &blkg->delay_nsec);
1888 }
1889 
1890 /**
1891  * blkg_tryget_closest - try and get a blkg ref on the closet blkg
1892  * @bio: target bio
1893  * @css: target css
1894  *
1895  * As the failure mode here is to walk up the blkg tree, this ensure that the
1896  * blkg->parent pointers are always valid.  This returns the blkg that it ended
1897  * up taking a reference on or %NULL if no reference was taken.
1898  */
1899 static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio,
1900                 struct cgroup_subsys_state *css)
1901 {
1902         struct blkcg_gq *blkg, *ret_blkg = NULL;
1903 
1904         rcu_read_lock();
1905         blkg = blkg_lookup_create(css_to_blkcg(css),
1906                                   bdev_get_queue(bio->bi_bdev));
1907         while (blkg) {
1908                 if (blkg_tryget(blkg)) {
1909                         ret_blkg = blkg;
1910                         break;
1911                 }
1912                 blkg = blkg->parent;
1913         }
1914         rcu_read_unlock();
1915 
1916         return ret_blkg;
1917 }
1918 
1919 /**
1920  * bio_associate_blkg_from_css - associate a bio with a specified css
1921  * @bio: target bio
1922  * @css: target css
1923  *
1924  * Associate @bio with the blkg found by combining the css's blkg and the
1925  * request_queue of the @bio.  An association failure is handled by walking up
1926  * the blkg tree.  Therefore, the blkg associated can be anything between @blkg
1927  * and q->root_blkg.  This situation only happens when a cgroup is dying and
1928  * then the remaining bios will spill to the closest alive blkg.
1929  *
1930  * A reference will be taken on the blkg and will be released when @bio is
1931  * freed.
1932  */
1933 void bio_associate_blkg_from_css(struct bio *bio,
1934                                  struct cgroup_subsys_state *css)
1935 {
1936         if (bio->bi_blkg)
1937                 blkg_put(bio->bi_blkg);
1938 
1939         if (css && css->parent) {
1940                 bio->bi_blkg = blkg_tryget_closest(bio, css);
1941         } else {
1942                 blkg_get(bdev_get_queue(bio->bi_bdev)->root_blkg);
1943                 bio->bi_blkg = bdev_get_queue(bio->bi_bdev)->root_blkg;
1944         }
1945 }
1946 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css);
1947 
1948 /**
1949  * bio_associate_blkg - associate a bio with a blkg
1950  * @bio: target bio
1951  *
1952  * Associate @bio with the blkg found from the bio's css and request_queue.
1953  * If one is not found, bio_lookup_blkg() creates the blkg.  If a blkg is
1954  * already associated, the css is reused and association redone as the
1955  * request_queue may have changed.
1956  */
1957 void bio_associate_blkg(struct bio *bio)
1958 {
1959         struct cgroup_subsys_state *css;
1960 
1961         rcu_read_lock();
1962 
1963         if (bio->bi_blkg)
1964                 css = bio_blkcg_css(bio);
1965         else
1966                 css = blkcg_css();
1967 
1968         bio_associate_blkg_from_css(bio, css);
1969 
1970         rcu_read_unlock();
1971 }
1972 EXPORT_SYMBOL_GPL(bio_associate_blkg);
1973 
1974 /**
1975  * bio_clone_blkg_association - clone blkg association from src to dst bio
1976  * @dst: destination bio
1977  * @src: source bio
1978  */
1979 void bio_clone_blkg_association(struct bio *dst, struct bio *src)
1980 {
1981         if (src->bi_blkg)
1982                 bio_associate_blkg_from_css(dst, bio_blkcg_css(src));
1983 }
1984 EXPORT_SYMBOL_GPL(bio_clone_blkg_association);
1985 
1986 static int blk_cgroup_io_type(struct bio *bio)
1987 {
1988         if (op_is_discard(bio->bi_opf))
1989                 return BLKG_IOSTAT_DISCARD;
1990         if (op_is_write(bio->bi_opf))
1991                 return BLKG_IOSTAT_WRITE;
1992         return BLKG_IOSTAT_READ;
1993 }
1994 
1995 void blk_cgroup_bio_start(struct bio *bio)
1996 {
1997         int rwd = blk_cgroup_io_type(bio), cpu;
1998         struct blkg_iostat_set *bis;
1999         unsigned long flags;
2000 
2001         cpu = get_cpu();
2002         bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu);
2003         flags = u64_stats_update_begin_irqsave(&bis->sync);
2004 
2005         /*
2006          * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split
2007          * bio and we would have already accounted for the size of the bio.
2008          */
2009         if (!bio_flagged(bio, BIO_CGROUP_ACCT)) {
2010                 bio_set_flag(bio, BIO_CGROUP_ACCT);
2011                 bis->cur.bytes[rwd] += bio->bi_iter.bi_size;
2012         }
2013         bis->cur.ios[rwd]++;
2014 
2015         u64_stats_update_end_irqrestore(&bis->sync, flags);
2016         if (cgroup_subsys_on_dfl(io_cgrp_subsys))
2017                 cgroup_rstat_updated(bio->bi_blkg->blkcg->css.cgroup, cpu);
2018         put_cpu();
2019 }
2020 
2021 bool blk_cgroup_congested(void)
2022 {
2023         struct cgroup_subsys_state *css;
2024         bool ret = false;
2025 
2026         rcu_read_lock();
2027         for (css = blkcg_css(); css; css = css->parent) {
2028                 if (atomic_read(&css->cgroup->congestion_count)) {
2029                         ret = true;
2030                         break;
2031                 }
2032         }
2033         rcu_read_unlock();
2034         return ret;
2035 }
2036 
2037 static int __init blkcg_init(void)
2038 {
2039         blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio",
2040                                             WQ_MEM_RECLAIM | WQ_FREEZABLE |
2041                                             WQ_UNBOUND | WQ_SYSFS, 0);
2042         if (!blkcg_punt_bio_wq)
2043                 return -ENOMEM;
2044         return 0;
2045 }
2046 subsys_initcall(blkcg_init);
2047 
2048 module_param(blkcg_debug_stats, bool, 0644);
2049 MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not");
2050 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp