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

TOMOYO Linux Cross Reference
Linux/mm/backing-dev.c

Version: ~ [ linux-5.11-rc3 ] ~ [ linux-5.10.7 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.89 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.167 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.215 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.251 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.251 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 
  2 #include <linux/wait.h>
  3 #include <linux/backing-dev.h>
  4 #include <linux/kthread.h>
  5 #include <linux/freezer.h>
  6 #include <linux/fs.h>
  7 #include <linux/pagemap.h>
  8 #include <linux/mm.h>
  9 #include <linux/sched.h>
 10 #include <linux/module.h>
 11 #include <linux/writeback.h>
 12 #include <linux/device.h>
 13 #include <trace/events/writeback.h>
 14 
 15 struct backing_dev_info noop_backing_dev_info = {
 16         .name           = "noop",
 17         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
 18 };
 19 EXPORT_SYMBOL_GPL(noop_backing_dev_info);
 20 
 21 static struct class *bdi_class;
 22 
 23 /*
 24  * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
 25  * locking.
 26  */
 27 DEFINE_SPINLOCK(bdi_lock);
 28 LIST_HEAD(bdi_list);
 29 
 30 /* bdi_wq serves all asynchronous writeback tasks */
 31 struct workqueue_struct *bdi_wq;
 32 
 33 #ifdef CONFIG_DEBUG_FS
 34 #include <linux/debugfs.h>
 35 #include <linux/seq_file.h>
 36 
 37 static struct dentry *bdi_debug_root;
 38 
 39 static void bdi_debug_init(void)
 40 {
 41         bdi_debug_root = debugfs_create_dir("bdi", NULL);
 42 }
 43 
 44 static int bdi_debug_stats_show(struct seq_file *m, void *v)
 45 {
 46         struct backing_dev_info *bdi = m->private;
 47         struct bdi_writeback *wb = &bdi->wb;
 48         unsigned long background_thresh;
 49         unsigned long dirty_thresh;
 50         unsigned long wb_thresh;
 51         unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time;
 52         struct inode *inode;
 53 
 54         nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0;
 55         spin_lock(&wb->list_lock);
 56         list_for_each_entry(inode, &wb->b_dirty, i_io_list)
 57                 nr_dirty++;
 58         list_for_each_entry(inode, &wb->b_io, i_io_list)
 59                 nr_io++;
 60         list_for_each_entry(inode, &wb->b_more_io, i_io_list)
 61                 nr_more_io++;
 62         list_for_each_entry(inode, &wb->b_dirty_time, i_io_list)
 63                 if (inode->i_state & I_DIRTY_TIME)
 64                         nr_dirty_time++;
 65         spin_unlock(&wb->list_lock);
 66 
 67         global_dirty_limits(&background_thresh, &dirty_thresh);
 68         wb_thresh = wb_calc_thresh(wb, dirty_thresh);
 69 
 70 #define K(x) ((x) << (PAGE_SHIFT - 10))
 71         seq_printf(m,
 72                    "BdiWriteback:       %10lu kB\n"
 73                    "BdiReclaimable:     %10lu kB\n"
 74                    "BdiDirtyThresh:     %10lu kB\n"
 75                    "DirtyThresh:        %10lu kB\n"
 76                    "BackgroundThresh:   %10lu kB\n"
 77                    "BdiDirtied:         %10lu kB\n"
 78                    "BdiWritten:         %10lu kB\n"
 79                    "BdiWriteBandwidth:  %10lu kBps\n"
 80                    "b_dirty:            %10lu\n"
 81                    "b_io:               %10lu\n"
 82                    "b_more_io:          %10lu\n"
 83                    "b_dirty_time:       %10lu\n"
 84                    "bdi_list:           %10u\n"
 85                    "state:              %10lx\n",
 86                    (unsigned long) K(wb_stat(wb, WB_WRITEBACK)),
 87                    (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)),
 88                    K(wb_thresh),
 89                    K(dirty_thresh),
 90                    K(background_thresh),
 91                    (unsigned long) K(wb_stat(wb, WB_DIRTIED)),
 92                    (unsigned long) K(wb_stat(wb, WB_WRITTEN)),
 93                    (unsigned long) K(wb->write_bandwidth),
 94                    nr_dirty,
 95                    nr_io,
 96                    nr_more_io,
 97                    nr_dirty_time,
 98                    !list_empty(&bdi->bdi_list), bdi->wb.state);
 99 #undef K
100 
101         return 0;
102 }
103 DEFINE_SHOW_ATTRIBUTE(bdi_debug_stats);
104 
105 static int bdi_debug_register(struct backing_dev_info *bdi, const char *name)
106 {
107         if (!bdi_debug_root)
108                 return -ENOMEM;
109 
110         bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
111         if (!bdi->debug_dir)
112                 return -ENOMEM;
113 
114         bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
115                                                bdi, &bdi_debug_stats_fops);
116         if (!bdi->debug_stats) {
117                 debugfs_remove(bdi->debug_dir);
118                 bdi->debug_dir = NULL;
119                 return -ENOMEM;
120         }
121 
122         return 0;
123 }
124 
125 static void bdi_debug_unregister(struct backing_dev_info *bdi)
126 {
127         debugfs_remove(bdi->debug_stats);
128         debugfs_remove(bdi->debug_dir);
129 }
130 #else
131 static inline void bdi_debug_init(void)
132 {
133 }
134 static inline int bdi_debug_register(struct backing_dev_info *bdi,
135                                       const char *name)
136 {
137         return 0;
138 }
139 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
140 {
141 }
142 #endif
143 
144 static ssize_t read_ahead_kb_store(struct device *dev,
145                                   struct device_attribute *attr,
146                                   const char *buf, size_t count)
147 {
148         struct backing_dev_info *bdi = dev_get_drvdata(dev);
149         unsigned long read_ahead_kb;
150         ssize_t ret;
151 
152         ret = kstrtoul(buf, 10, &read_ahead_kb);
153         if (ret < 0)
154                 return ret;
155 
156         bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
157 
158         return count;
159 }
160 
161 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
162 
163 #define BDI_SHOW(name, expr)                                            \
164 static ssize_t name##_show(struct device *dev,                          \
165                            struct device_attribute *attr, char *page)   \
166 {                                                                       \
167         struct backing_dev_info *bdi = dev_get_drvdata(dev);            \
168                                                                         \
169         return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);  \
170 }                                                                       \
171 static DEVICE_ATTR_RW(name);
172 
173 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
174 
175 static ssize_t min_ratio_store(struct device *dev,
176                 struct device_attribute *attr, const char *buf, size_t count)
177 {
178         struct backing_dev_info *bdi = dev_get_drvdata(dev);
179         unsigned int ratio;
180         ssize_t ret;
181 
182         ret = kstrtouint(buf, 10, &ratio);
183         if (ret < 0)
184                 return ret;
185 
186         ret = bdi_set_min_ratio(bdi, ratio);
187         if (!ret)
188                 ret = count;
189 
190         return ret;
191 }
192 BDI_SHOW(min_ratio, bdi->min_ratio)
193 
194 static ssize_t max_ratio_store(struct device *dev,
195                 struct device_attribute *attr, const char *buf, size_t count)
196 {
197         struct backing_dev_info *bdi = dev_get_drvdata(dev);
198         unsigned int ratio;
199         ssize_t ret;
200 
201         ret = kstrtouint(buf, 10, &ratio);
202         if (ret < 0)
203                 return ret;
204 
205         ret = bdi_set_max_ratio(bdi, ratio);
206         if (!ret)
207                 ret = count;
208 
209         return ret;
210 }
211 BDI_SHOW(max_ratio, bdi->max_ratio)
212 
213 static ssize_t stable_pages_required_show(struct device *dev,
214                                           struct device_attribute *attr,
215                                           char *page)
216 {
217         struct backing_dev_info *bdi = dev_get_drvdata(dev);
218 
219         return snprintf(page, PAGE_SIZE-1, "%d\n",
220                         bdi_cap_stable_pages_required(bdi) ? 1 : 0);
221 }
222 static DEVICE_ATTR_RO(stable_pages_required);
223 
224 static struct attribute *bdi_dev_attrs[] = {
225         &dev_attr_read_ahead_kb.attr,
226         &dev_attr_min_ratio.attr,
227         &dev_attr_max_ratio.attr,
228         &dev_attr_stable_pages_required.attr,
229         NULL,
230 };
231 ATTRIBUTE_GROUPS(bdi_dev);
232 
233 static __init int bdi_class_init(void)
234 {
235         bdi_class = class_create(THIS_MODULE, "bdi");
236         if (IS_ERR(bdi_class))
237                 return PTR_ERR(bdi_class);
238 
239         bdi_class->dev_groups = bdi_dev_groups;
240         bdi_debug_init();
241 
242         return 0;
243 }
244 postcore_initcall(bdi_class_init);
245 
246 static int bdi_init(struct backing_dev_info *bdi);
247 
248 static int __init default_bdi_init(void)
249 {
250         int err;
251 
252         bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
253                                               WQ_UNBOUND | WQ_SYSFS, 0);
254         if (!bdi_wq)
255                 return -ENOMEM;
256 
257         err = bdi_init(&noop_backing_dev_info);
258 
259         return err;
260 }
261 subsys_initcall(default_bdi_init);
262 
263 /*
264  * This function is used when the first inode for this wb is marked dirty. It
265  * wakes-up the corresponding bdi thread which should then take care of the
266  * periodic background write-out of dirty inodes. Since the write-out would
267  * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
268  * set up a timer which wakes the bdi thread up later.
269  *
270  * Note, we wouldn't bother setting up the timer, but this function is on the
271  * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
272  * by delaying the wake-up.
273  *
274  * We have to be careful not to postpone flush work if it is scheduled for
275  * earlier. Thus we use queue_delayed_work().
276  */
277 void wb_wakeup_delayed(struct bdi_writeback *wb)
278 {
279         unsigned long timeout;
280 
281         timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
282         spin_lock_bh(&wb->work_lock);
283         if (test_bit(WB_registered, &wb->state))
284                 queue_delayed_work(bdi_wq, &wb->dwork, timeout);
285         spin_unlock_bh(&wb->work_lock);
286 }
287 
288 /*
289  * Initial write bandwidth: 100 MB/s
290  */
291 #define INIT_BW         (100 << (20 - PAGE_SHIFT))
292 
293 static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
294                    int blkcg_id, gfp_t gfp)
295 {
296         int i, err;
297 
298         memset(wb, 0, sizeof(*wb));
299 
300         if (wb != &bdi->wb)
301                 bdi_get(bdi);
302         wb->bdi = bdi;
303         wb->last_old_flush = jiffies;
304         INIT_LIST_HEAD(&wb->b_dirty);
305         INIT_LIST_HEAD(&wb->b_io);
306         INIT_LIST_HEAD(&wb->b_more_io);
307         INIT_LIST_HEAD(&wb->b_dirty_time);
308         spin_lock_init(&wb->list_lock);
309 
310         wb->bw_time_stamp = jiffies;
311         wb->balanced_dirty_ratelimit = INIT_BW;
312         wb->dirty_ratelimit = INIT_BW;
313         wb->write_bandwidth = INIT_BW;
314         wb->avg_write_bandwidth = INIT_BW;
315 
316         spin_lock_init(&wb->work_lock);
317         INIT_LIST_HEAD(&wb->work_list);
318         INIT_DELAYED_WORK(&wb->dwork, wb_workfn);
319         wb->dirty_sleep = jiffies;
320 
321         wb->congested = wb_congested_get_create(bdi, blkcg_id, gfp);
322         if (!wb->congested) {
323                 err = -ENOMEM;
324                 goto out_put_bdi;
325         }
326 
327         err = fprop_local_init_percpu(&wb->completions, gfp);
328         if (err)
329                 goto out_put_cong;
330 
331         for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
332                 err = percpu_counter_init(&wb->stat[i], 0, gfp);
333                 if (err)
334                         goto out_destroy_stat;
335         }
336 
337         return 0;
338 
339 out_destroy_stat:
340         while (i--)
341                 percpu_counter_destroy(&wb->stat[i]);
342         fprop_local_destroy_percpu(&wb->completions);
343 out_put_cong:
344         wb_congested_put(wb->congested);
345 out_put_bdi:
346         if (wb != &bdi->wb)
347                 bdi_put(bdi);
348         return err;
349 }
350 
351 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
352 
353 /*
354  * Remove bdi from the global list and shutdown any threads we have running
355  */
356 static void wb_shutdown(struct bdi_writeback *wb)
357 {
358         /* Make sure nobody queues further work */
359         spin_lock_bh(&wb->work_lock);
360         if (!test_and_clear_bit(WB_registered, &wb->state)) {
361                 spin_unlock_bh(&wb->work_lock);
362                 return;
363         }
364         spin_unlock_bh(&wb->work_lock);
365 
366         cgwb_remove_from_bdi_list(wb);
367         /*
368          * Drain work list and shutdown the delayed_work.  !WB_registered
369          * tells wb_workfn() that @wb is dying and its work_list needs to
370          * be drained no matter what.
371          */
372         mod_delayed_work(bdi_wq, &wb->dwork, 0);
373         flush_delayed_work(&wb->dwork);
374         WARN_ON(!list_empty(&wb->work_list));
375 }
376 
377 static void wb_exit(struct bdi_writeback *wb)
378 {
379         int i;
380 
381         WARN_ON(delayed_work_pending(&wb->dwork));
382 
383         for (i = 0; i < NR_WB_STAT_ITEMS; i++)
384                 percpu_counter_destroy(&wb->stat[i]);
385 
386         fprop_local_destroy_percpu(&wb->completions);
387         wb_congested_put(wb->congested);
388         if (wb != &wb->bdi->wb)
389                 bdi_put(wb->bdi);
390 }
391 
392 #ifdef CONFIG_CGROUP_WRITEBACK
393 
394 #include <linux/memcontrol.h>
395 
396 /*
397  * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree,
398  * blkcg->cgwb_list, and memcg->cgwb_list.  bdi->cgwb_tree is also RCU
399  * protected.
400  */
401 static DEFINE_SPINLOCK(cgwb_lock);
402 static struct workqueue_struct *cgwb_release_wq;
403 
404 /**
405  * wb_congested_get_create - get or create a wb_congested
406  * @bdi: associated bdi
407  * @blkcg_id: ID of the associated blkcg
408  * @gfp: allocation mask
409  *
410  * Look up the wb_congested for @blkcg_id on @bdi.  If missing, create one.
411  * The returned wb_congested has its reference count incremented.  Returns
412  * NULL on failure.
413  */
414 struct bdi_writeback_congested *
415 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
416 {
417         struct bdi_writeback_congested *new_congested = NULL, *congested;
418         struct rb_node **node, *parent;
419         unsigned long flags;
420 retry:
421         spin_lock_irqsave(&cgwb_lock, flags);
422 
423         node = &bdi->cgwb_congested_tree.rb_node;
424         parent = NULL;
425 
426         while (*node != NULL) {
427                 parent = *node;
428                 congested = rb_entry(parent, struct bdi_writeback_congested,
429                                      rb_node);
430                 if (congested->blkcg_id < blkcg_id)
431                         node = &parent->rb_left;
432                 else if (congested->blkcg_id > blkcg_id)
433                         node = &parent->rb_right;
434                 else
435                         goto found;
436         }
437 
438         if (new_congested) {
439                 /* !found and storage for new one already allocated, insert */
440                 congested = new_congested;
441                 new_congested = NULL;
442                 rb_link_node(&congested->rb_node, parent, node);
443                 rb_insert_color(&congested->rb_node, &bdi->cgwb_congested_tree);
444                 goto found;
445         }
446 
447         spin_unlock_irqrestore(&cgwb_lock, flags);
448 
449         /* allocate storage for new one and retry */
450         new_congested = kzalloc(sizeof(*new_congested), gfp);
451         if (!new_congested)
452                 return NULL;
453 
454         atomic_set(&new_congested->refcnt, 0);
455         new_congested->__bdi = bdi;
456         new_congested->blkcg_id = blkcg_id;
457         goto retry;
458 
459 found:
460         atomic_inc(&congested->refcnt);
461         spin_unlock_irqrestore(&cgwb_lock, flags);
462         kfree(new_congested);
463         return congested;
464 }
465 
466 /**
467  * wb_congested_put - put a wb_congested
468  * @congested: wb_congested to put
469  *
470  * Put @congested and destroy it if the refcnt reaches zero.
471  */
472 void wb_congested_put(struct bdi_writeback_congested *congested)
473 {
474         unsigned long flags;
475 
476         local_irq_save(flags);
477         if (!atomic_dec_and_lock(&congested->refcnt, &cgwb_lock)) {
478                 local_irq_restore(flags);
479                 return;
480         }
481 
482         /* bdi might already have been destroyed leaving @congested unlinked */
483         if (congested->__bdi) {
484                 rb_erase(&congested->rb_node,
485                          &congested->__bdi->cgwb_congested_tree);
486                 congested->__bdi = NULL;
487         }
488 
489         spin_unlock_irqrestore(&cgwb_lock, flags);
490         kfree(congested);
491 }
492 
493 static void cgwb_release_workfn(struct work_struct *work)
494 {
495         struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
496                                                 release_work);
497 
498         mutex_lock(&wb->bdi->cgwb_release_mutex);
499         wb_shutdown(wb);
500 
501         css_put(wb->memcg_css);
502         css_put(wb->blkcg_css);
503         mutex_unlock(&wb->bdi->cgwb_release_mutex);
504 
505         fprop_local_destroy_percpu(&wb->memcg_completions);
506         percpu_ref_exit(&wb->refcnt);
507         wb_exit(wb);
508         kfree_rcu(wb, rcu);
509 }
510 
511 static void cgwb_release(struct percpu_ref *refcnt)
512 {
513         struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
514                                                 refcnt);
515         queue_work(cgwb_release_wq, &wb->release_work);
516 }
517 
518 static void cgwb_kill(struct bdi_writeback *wb)
519 {
520         lockdep_assert_held(&cgwb_lock);
521 
522         WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
523         list_del(&wb->memcg_node);
524         list_del(&wb->blkcg_node);
525         percpu_ref_kill(&wb->refcnt);
526 }
527 
528 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
529 {
530         spin_lock_irq(&cgwb_lock);
531         list_del_rcu(&wb->bdi_node);
532         spin_unlock_irq(&cgwb_lock);
533 }
534 
535 static int cgwb_create(struct backing_dev_info *bdi,
536                        struct cgroup_subsys_state *memcg_css, gfp_t gfp)
537 {
538         struct mem_cgroup *memcg;
539         struct cgroup_subsys_state *blkcg_css;
540         struct blkcg *blkcg;
541         struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
542         struct bdi_writeback *wb;
543         unsigned long flags;
544         int ret = 0;
545 
546         memcg = mem_cgroup_from_css(memcg_css);
547         blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
548         blkcg = css_to_blkcg(blkcg_css);
549         memcg_cgwb_list = mem_cgroup_cgwb_list(memcg);
550         blkcg_cgwb_list = &blkcg->cgwb_list;
551 
552         /* look up again under lock and discard on blkcg mismatch */
553         spin_lock_irqsave(&cgwb_lock, flags);
554         wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
555         if (wb && wb->blkcg_css != blkcg_css) {
556                 cgwb_kill(wb);
557                 wb = NULL;
558         }
559         spin_unlock_irqrestore(&cgwb_lock, flags);
560         if (wb)
561                 goto out_put;
562 
563         /* need to create a new one */
564         wb = kmalloc(sizeof(*wb), gfp);
565         if (!wb) {
566                 ret = -ENOMEM;
567                 goto out_put;
568         }
569 
570         ret = wb_init(wb, bdi, blkcg_css->id, gfp);
571         if (ret)
572                 goto err_free;
573 
574         ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
575         if (ret)
576                 goto err_wb_exit;
577 
578         ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
579         if (ret)
580                 goto err_ref_exit;
581 
582         wb->memcg_css = memcg_css;
583         wb->blkcg_css = blkcg_css;
584         INIT_WORK(&wb->release_work, cgwb_release_workfn);
585         set_bit(WB_registered, &wb->state);
586 
587         /*
588          * The root wb determines the registered state of the whole bdi and
589          * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
590          * whether they're still online.  Don't link @wb if any is dead.
591          * See wb_memcg_offline() and wb_blkcg_offline().
592          */
593         ret = -ENODEV;
594         spin_lock_irqsave(&cgwb_lock, flags);
595         if (test_bit(WB_registered, &bdi->wb.state) &&
596             blkcg_cgwb_list->next && memcg_cgwb_list->next) {
597                 /* we might have raced another instance of this function */
598                 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
599                 if (!ret) {
600                         list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
601                         list_add(&wb->memcg_node, memcg_cgwb_list);
602                         list_add(&wb->blkcg_node, blkcg_cgwb_list);
603                         css_get(memcg_css);
604                         css_get(blkcg_css);
605                 }
606         }
607         spin_unlock_irqrestore(&cgwb_lock, flags);
608         if (ret) {
609                 if (ret == -EEXIST)
610                         ret = 0;
611                 goto err_fprop_exit;
612         }
613         goto out_put;
614 
615 err_fprop_exit:
616         fprop_local_destroy_percpu(&wb->memcg_completions);
617 err_ref_exit:
618         percpu_ref_exit(&wb->refcnt);
619 err_wb_exit:
620         wb_exit(wb);
621 err_free:
622         kfree(wb);
623 out_put:
624         css_put(blkcg_css);
625         return ret;
626 }
627 
628 /**
629  * wb_get_create - get wb for a given memcg, create if necessary
630  * @bdi: target bdi
631  * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
632  * @gfp: allocation mask to use
633  *
634  * Try to get the wb for @memcg_css on @bdi.  If it doesn't exist, try to
635  * create one.  The returned wb has its refcount incremented.
636  *
637  * This function uses css_get() on @memcg_css and thus expects its refcnt
638  * to be positive on invocation.  IOW, rcu_read_lock() protection on
639  * @memcg_css isn't enough.  try_get it before calling this function.
640  *
641  * A wb is keyed by its associated memcg.  As blkcg implicitly enables
642  * memcg on the default hierarchy, memcg association is guaranteed to be
643  * more specific (equal or descendant to the associated blkcg) and thus can
644  * identify both the memcg and blkcg associations.
645  *
646  * Because the blkcg associated with a memcg may change as blkcg is enabled
647  * and disabled closer to root in the hierarchy, each wb keeps track of
648  * both the memcg and blkcg associated with it and verifies the blkcg on
649  * each lookup.  On mismatch, the existing wb is discarded and a new one is
650  * created.
651  */
652 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
653                                     struct cgroup_subsys_state *memcg_css,
654                                     gfp_t gfp)
655 {
656         struct bdi_writeback *wb;
657 
658         might_sleep_if(gfpflags_allow_blocking(gfp));
659 
660         if (!memcg_css->parent)
661                 return &bdi->wb;
662 
663         do {
664                 rcu_read_lock();
665                 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
666                 if (wb) {
667                         struct cgroup_subsys_state *blkcg_css;
668 
669                         /* see whether the blkcg association has changed */
670                         blkcg_css = cgroup_get_e_css(memcg_css->cgroup,
671                                                      &io_cgrp_subsys);
672                         if (unlikely(wb->blkcg_css != blkcg_css ||
673                                      !wb_tryget(wb)))
674                                 wb = NULL;
675                         css_put(blkcg_css);
676                 }
677                 rcu_read_unlock();
678         } while (!wb && !cgwb_create(bdi, memcg_css, gfp));
679 
680         return wb;
681 }
682 
683 static int cgwb_bdi_init(struct backing_dev_info *bdi)
684 {
685         int ret;
686 
687         INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
688         bdi->cgwb_congested_tree = RB_ROOT;
689         mutex_init(&bdi->cgwb_release_mutex);
690 
691         ret = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL);
692         if (!ret) {
693                 bdi->wb.memcg_css = &root_mem_cgroup->css;
694                 bdi->wb.blkcg_css = blkcg_root_css;
695         }
696         return ret;
697 }
698 
699 static void cgwb_bdi_unregister(struct backing_dev_info *bdi)
700 {
701         struct radix_tree_iter iter;
702         void **slot;
703         struct bdi_writeback *wb;
704 
705         WARN_ON(test_bit(WB_registered, &bdi->wb.state));
706 
707         spin_lock_irq(&cgwb_lock);
708         radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
709                 cgwb_kill(*slot);
710         spin_unlock_irq(&cgwb_lock);
711 
712         mutex_lock(&bdi->cgwb_release_mutex);
713         spin_lock_irq(&cgwb_lock);
714         while (!list_empty(&bdi->wb_list)) {
715                 wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
716                                       bdi_node);
717                 spin_unlock_irq(&cgwb_lock);
718                 wb_shutdown(wb);
719                 spin_lock_irq(&cgwb_lock);
720         }
721         spin_unlock_irq(&cgwb_lock);
722         mutex_unlock(&bdi->cgwb_release_mutex);
723 }
724 
725 /**
726  * wb_memcg_offline - kill all wb's associated with a memcg being offlined
727  * @memcg: memcg being offlined
728  *
729  * Also prevents creation of any new wb's associated with @memcg.
730  */
731 void wb_memcg_offline(struct mem_cgroup *memcg)
732 {
733         struct list_head *memcg_cgwb_list = mem_cgroup_cgwb_list(memcg);
734         struct bdi_writeback *wb, *next;
735 
736         spin_lock_irq(&cgwb_lock);
737         list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
738                 cgwb_kill(wb);
739         memcg_cgwb_list->next = NULL;   /* prevent new wb's */
740         spin_unlock_irq(&cgwb_lock);
741 }
742 
743 /**
744  * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
745  * @blkcg: blkcg being offlined
746  *
747  * Also prevents creation of any new wb's associated with @blkcg.
748  */
749 void wb_blkcg_offline(struct blkcg *blkcg)
750 {
751         struct bdi_writeback *wb, *next;
752 
753         spin_lock_irq(&cgwb_lock);
754         list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node)
755                 cgwb_kill(wb);
756         blkcg->cgwb_list.next = NULL;   /* prevent new wb's */
757         spin_unlock_irq(&cgwb_lock);
758 }
759 
760 static void cgwb_bdi_exit(struct backing_dev_info *bdi)
761 {
762         struct rb_node *rbn;
763 
764         spin_lock_irq(&cgwb_lock);
765         while ((rbn = rb_first(&bdi->cgwb_congested_tree))) {
766                 struct bdi_writeback_congested *congested =
767                         rb_entry(rbn, struct bdi_writeback_congested, rb_node);
768 
769                 rb_erase(rbn, &bdi->cgwb_congested_tree);
770                 congested->__bdi = NULL;        /* mark @congested unlinked */
771         }
772         spin_unlock_irq(&cgwb_lock);
773 }
774 
775 static void cgwb_bdi_register(struct backing_dev_info *bdi)
776 {
777         spin_lock_irq(&cgwb_lock);
778         list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
779         spin_unlock_irq(&cgwb_lock);
780 }
781 
782 static int __init cgwb_init(void)
783 {
784         /*
785          * There can be many concurrent release work items overwhelming
786          * system_wq.  Put them in a separate wq and limit concurrency.
787          * There's no point in executing many of these in parallel.
788          */
789         cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1);
790         if (!cgwb_release_wq)
791                 return -ENOMEM;
792 
793         return 0;
794 }
795 subsys_initcall(cgwb_init);
796 
797 #else   /* CONFIG_CGROUP_WRITEBACK */
798 
799 static int cgwb_bdi_init(struct backing_dev_info *bdi)
800 {
801         int err;
802 
803         bdi->wb_congested = kzalloc(sizeof(*bdi->wb_congested), GFP_KERNEL);
804         if (!bdi->wb_congested)
805                 return -ENOMEM;
806 
807         atomic_set(&bdi->wb_congested->refcnt, 1);
808 
809         err = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL);
810         if (err) {
811                 wb_congested_put(bdi->wb_congested);
812                 return err;
813         }
814         return 0;
815 }
816 
817 static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { }
818 
819 static void cgwb_bdi_exit(struct backing_dev_info *bdi)
820 {
821         wb_congested_put(bdi->wb_congested);
822 }
823 
824 static void cgwb_bdi_register(struct backing_dev_info *bdi)
825 {
826         list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
827 }
828 
829 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
830 {
831         list_del_rcu(&wb->bdi_node);
832 }
833 
834 #endif  /* CONFIG_CGROUP_WRITEBACK */
835 
836 static int bdi_init(struct backing_dev_info *bdi)
837 {
838         int ret;
839 
840         bdi->dev = NULL;
841 
842         kref_init(&bdi->refcnt);
843         bdi->min_ratio = 0;
844         bdi->max_ratio = 100;
845         bdi->max_prop_frac = FPROP_FRAC_BASE;
846         INIT_LIST_HEAD(&bdi->bdi_list);
847         INIT_LIST_HEAD(&bdi->wb_list);
848         init_waitqueue_head(&bdi->wb_waitq);
849 
850         ret = cgwb_bdi_init(bdi);
851 
852         return ret;
853 }
854 
855 struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id)
856 {
857         struct backing_dev_info *bdi;
858 
859         bdi = kmalloc_node(sizeof(struct backing_dev_info),
860                            gfp_mask | __GFP_ZERO, node_id);
861         if (!bdi)
862                 return NULL;
863 
864         if (bdi_init(bdi)) {
865                 kfree(bdi);
866                 return NULL;
867         }
868         return bdi;
869 }
870 EXPORT_SYMBOL(bdi_alloc_node);
871 
872 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args)
873 {
874         struct device *dev;
875 
876         if (bdi->dev)   /* The driver needs to use separate queues per device */
877                 return 0;
878 
879         dev = device_create_vargs(bdi_class, NULL, MKDEV(0, 0), bdi, fmt, args);
880         if (IS_ERR(dev))
881                 return PTR_ERR(dev);
882 
883         cgwb_bdi_register(bdi);
884         bdi->dev = dev;
885 
886         bdi_debug_register(bdi, dev_name(dev));
887         set_bit(WB_registered, &bdi->wb.state);
888 
889         spin_lock_bh(&bdi_lock);
890         list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
891         spin_unlock_bh(&bdi_lock);
892 
893         trace_writeback_bdi_register(bdi);
894         return 0;
895 }
896 EXPORT_SYMBOL(bdi_register_va);
897 
898 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...)
899 {
900         va_list args;
901         int ret;
902 
903         va_start(args, fmt);
904         ret = bdi_register_va(bdi, fmt, args);
905         va_end(args);
906         return ret;
907 }
908 EXPORT_SYMBOL(bdi_register);
909 
910 int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner)
911 {
912         int rc;
913 
914         rc = bdi_register(bdi, "%u:%u", MAJOR(owner->devt), MINOR(owner->devt));
915         if (rc)
916                 return rc;
917         /* Leaking owner reference... */
918         WARN_ON(bdi->owner);
919         bdi->owner = owner;
920         get_device(owner);
921         return 0;
922 }
923 EXPORT_SYMBOL(bdi_register_owner);
924 
925 /*
926  * Remove bdi from bdi_list, and ensure that it is no longer visible
927  */
928 static void bdi_remove_from_list(struct backing_dev_info *bdi)
929 {
930         spin_lock_bh(&bdi_lock);
931         list_del_rcu(&bdi->bdi_list);
932         spin_unlock_bh(&bdi_lock);
933 
934         synchronize_rcu_expedited();
935 }
936 
937 void bdi_unregister(struct backing_dev_info *bdi)
938 {
939         /* make sure nobody finds us on the bdi_list anymore */
940         bdi_remove_from_list(bdi);
941         wb_shutdown(&bdi->wb);
942         cgwb_bdi_unregister(bdi);
943 
944         if (bdi->dev) {
945                 bdi_debug_unregister(bdi);
946                 device_unregister(bdi->dev);
947                 bdi->dev = NULL;
948         }
949 
950         if (bdi->owner) {
951                 put_device(bdi->owner);
952                 bdi->owner = NULL;
953         }
954 }
955 
956 static void release_bdi(struct kref *ref)
957 {
958         struct backing_dev_info *bdi =
959                         container_of(ref, struct backing_dev_info, refcnt);
960 
961         if (test_bit(WB_registered, &bdi->wb.state))
962                 bdi_unregister(bdi);
963         WARN_ON_ONCE(bdi->dev);
964         wb_exit(&bdi->wb);
965         cgwb_bdi_exit(bdi);
966         kfree(bdi);
967 }
968 
969 void bdi_put(struct backing_dev_info *bdi)
970 {
971         kref_put(&bdi->refcnt, release_bdi);
972 }
973 EXPORT_SYMBOL(bdi_put);
974 
975 static wait_queue_head_t congestion_wqh[2] = {
976                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
977                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
978         };
979 static atomic_t nr_wb_congested[2];
980 
981 void clear_wb_congested(struct bdi_writeback_congested *congested, int sync)
982 {
983         wait_queue_head_t *wqh = &congestion_wqh[sync];
984         enum wb_congested_state bit;
985 
986         bit = sync ? WB_sync_congested : WB_async_congested;
987         if (test_and_clear_bit(bit, &congested->state))
988                 atomic_dec(&nr_wb_congested[sync]);
989         smp_mb__after_atomic();
990         if (waitqueue_active(wqh))
991                 wake_up(wqh);
992 }
993 EXPORT_SYMBOL(clear_wb_congested);
994 
995 void set_wb_congested(struct bdi_writeback_congested *congested, int sync)
996 {
997         enum wb_congested_state bit;
998 
999         bit = sync ? WB_sync_congested : WB_async_congested;
1000         if (!test_and_set_bit(bit, &congested->state))
1001                 atomic_inc(&nr_wb_congested[sync]);
1002 }
1003 EXPORT_SYMBOL(set_wb_congested);
1004 
1005 /**
1006  * congestion_wait - wait for a backing_dev to become uncongested
1007  * @sync: SYNC or ASYNC IO
1008  * @timeout: timeout in jiffies
1009  *
1010  * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
1011  * write congestion.  If no backing_devs are congested then just wait for the
1012  * next write to be completed.
1013  */
1014 long congestion_wait(int sync, long timeout)
1015 {
1016         long ret;
1017         unsigned long start = jiffies;
1018         DEFINE_WAIT(wait);
1019         wait_queue_head_t *wqh = &congestion_wqh[sync];
1020 
1021         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1022         ret = io_schedule_timeout(timeout);
1023         finish_wait(wqh, &wait);
1024 
1025         trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
1026                                         jiffies_to_usecs(jiffies - start));
1027 
1028         return ret;
1029 }
1030 EXPORT_SYMBOL(congestion_wait);
1031 
1032 /**
1033  * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a pgdat to complete writes
1034  * @sync: SYNC or ASYNC IO
1035  * @timeout: timeout in jiffies
1036  *
1037  * In the event of a congested backing_dev (any backing_dev) this waits
1038  * for up to @timeout jiffies for either a BDI to exit congestion of the
1039  * given @sync queue or a write to complete.
1040  *
1041  * The return value is 0 if the sleep is for the full timeout. Otherwise,
1042  * it is the number of jiffies that were still remaining when the function
1043  * returned. return_value == timeout implies the function did not sleep.
1044  */
1045 long wait_iff_congested(int sync, long timeout)
1046 {
1047         long ret;
1048         unsigned long start = jiffies;
1049         DEFINE_WAIT(wait);
1050         wait_queue_head_t *wqh = &congestion_wqh[sync];
1051 
1052         /*
1053          * If there is no congestion, yield if necessary instead
1054          * of sleeping on the congestion queue
1055          */
1056         if (atomic_read(&nr_wb_congested[sync]) == 0) {
1057                 cond_resched();
1058 
1059                 /* In case we scheduled, work out time remaining */
1060                 ret = timeout - (jiffies - start);
1061                 if (ret < 0)
1062                         ret = 0;
1063 
1064                 goto out;
1065         }
1066 
1067         /* Sleep until uncongested or a write happens */
1068         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1069         ret = io_schedule_timeout(timeout);
1070         finish_wait(wqh, &wait);
1071 
1072 out:
1073         trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
1074                                         jiffies_to_usecs(jiffies - start));
1075 
1076         return ret;
1077 }
1078 EXPORT_SYMBOL(wait_iff_congested);
1079 

~ [ 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