mm/backing-dev.c

   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 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
  16
  17 struct backing_dev_info noop_backing_dev_info = {
  18         .name           = "noop",
  19         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
  20 };
  21
  22 static struct class *bdi_class;
  23
  24 /*
  25  * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
  26  * locking.
  27  */
  28 DEFINE_SPINLOCK(bdi_lock);
  29 LIST_HEAD(bdi_list);
  30
  31 /* bdi_wq serves all asynchronous writeback tasks */
  32 struct workqueue_struct *bdi_wq;
  33
  34 #ifdef CONFIG_DEBUG_FS
  35 #include <linux/debugfs.h>
  36 #include <linux/seq_file.h>
  37
  38 static struct dentry *bdi_debug_root;
  39
  40 static void bdi_debug_init(void)
  41 {
  42         bdi_debug_root = debugfs_create_dir("bdi", NULL);
  43 }
  44
  45 static int bdi_debug_stats_show(struct seq_file *m, void *v)
  46 {
  47         struct backing_dev_info *bdi = m->private;
  48         struct bdi_writeback *wb = &bdi->wb;
  49         unsigned long background_thresh;
  50         unsigned long dirty_thresh;
  51         unsigned long bdi_thresh;
  52         unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time;
  53         struct inode *inode;
  54
  55         nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0;
  56         spin_lock(&wb->list_lock);
  57         list_for_each_entry(inode, &wb->b_dirty, i_wb_list)
  58                 nr_dirty++;
  59         list_for_each_entry(inode, &wb->b_io, i_wb_list)
  60                 nr_io++;
  61         list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
  62                 nr_more_io++;
  63         list_for_each_entry(inode, &wb->b_dirty_time, i_wb_list)
  64                 if (inode->i_state & I_DIRTY_TIME)
  65                         nr_dirty_time++;
  66         spin_unlock(&wb->list_lock);
  67
  68         global_dirty_limits(&background_thresh, &dirty_thresh);
  69         bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
  70
  71 #define K(x) ((x) << (PAGE_SHIFT - 10))
  72         seq_printf(m,
  73                    "BdiWriteback:       %10lu kB\n"
  74                    "BdiReclaimable:     %10lu kB\n"
  75                    "BdiDirtyThresh:     %10lu kB\n"
  76                    "DirtyThresh:        %10lu kB\n"
  77                    "BackgroundThresh:   %10lu kB\n"
  78                    "BdiDirtied:         %10lu kB\n"
  79                    "BdiWritten:         %10lu kB\n"
  80                    "BdiWriteBandwidth:  %10lu kBps\n"
  81                    "b_dirty:            %10lu\n"
  82                    "b_io:               %10lu\n"
  83                    "b_more_io:          %10lu\n"
  84                    "b_dirty_time:       %10lu\n"
  85                    "bdi_list:           %10u\n"
  86                    "state:              %10lx\n",
  87                    (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
  88                    (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
  89                    K(bdi_thresh),
  90                    K(dirty_thresh),
  91                    K(background_thresh),
  92                    (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)),
  93                    (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)),
  94                    (unsigned long) K(bdi->write_bandwidth),
  95                    nr_dirty,
  96                    nr_io,
  97                    nr_more_io,
  98                    nr_dirty_time,
  99                    !list_empty(&bdi->bdi_list), bdi->state);
 100 #undef K
 101
 102         return 0;
 103 }
 104
 105 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
 106 {
 107         return single_open(file, bdi_debug_stats_show, inode->i_private);
 108 }
 109
 110 static const struct file_operations bdi_debug_stats_fops = {
 111         .open           = bdi_debug_stats_open,
 112         .read           = seq_read,
 113         .llseek         = seq_lseek,
 114         .release        = single_release,
 115 };
 116
 117 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
 118 {
 119         bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
 120         bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
 121                                                bdi, &bdi_debug_stats_fops);
 122 }
 123
 124 static void bdi_debug_unregister(struct backing_dev_info *bdi)
 125 {
 126         debugfs_remove(bdi->debug_stats);
 127         debugfs_remove(bdi->debug_dir);
 128 }
 129 #else
 130 static inline void bdi_debug_init(void)
 131 {
 132 }
 133 static inline void bdi_debug_register(struct backing_dev_info *bdi,
 134                                       const char *name)
 135 {
 136 }
 137 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
 138 {
 139 }
 140 #endif
 141
 142 static ssize_t read_ahead_kb_store(struct device *dev,
 143                                   struct device_attribute *attr,
 144                                   const char *buf, size_t count)
 145 {
 146         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 147         unsigned long read_ahead_kb;
 148         ssize_t ret;
 149
 150         ret = kstrtoul(buf, 10, &read_ahead_kb);
 151         if (ret < 0)
 152                 return ret;
 153
 154         bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
 155
 156         return count;
 157 }
 158
 159 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
 160
 161 #define BDI_SHOW(name, expr)                                            \
 162 static ssize_t name##_show(struct device *dev,                          \
 163                            struct device_attribute *attr, char *page)   \
 164 {                                                                       \
 165         struct backing_dev_info *bdi = dev_get_drvdata(dev);            \
 166                                                                         \
 167         return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);  \
 168 }                                                                       \
 169 static DEVICE_ATTR_RW(name);
 170
 171 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
 172
 173 static ssize_t min_ratio_store(struct device *dev,
 174                 struct device_attribute *attr, const char *buf, size_t count)
 175 {
 176         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 177         unsigned int ratio;
 178         ssize_t ret;
 179
 180         ret = kstrtouint(buf, 10, &ratio);
 181         if (ret < 0)
 182                 return ret;
 183
 184         ret = bdi_set_min_ratio(bdi, ratio);
 185         if (!ret)
 186                 ret = count;
 187
 188         return ret;
 189 }
 190 BDI_SHOW(min_ratio, bdi->min_ratio)
 191
 192 static ssize_t max_ratio_store(struct device *dev,
 193                 struct device_attribute *attr, const char *buf, size_t count)
 194 {
 195         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 196         unsigned int ratio;
 197         ssize_t ret;
 198
 199         ret = kstrtouint(buf, 10, &ratio);
 200         if (ret < 0)
 201                 return ret;
 202
 203         ret = bdi_set_max_ratio(bdi, ratio);
 204         if (!ret)
 205                 ret = count;
 206
 207         return ret;
 208 }
 209 BDI_SHOW(max_ratio, bdi->max_ratio)
 210
 211 static ssize_t stable_pages_required_show(struct device *dev,
 212                                           struct device_attribute *attr,
 213                                           char *page)
 214 {
 215         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 216
 217         return snprintf(page, PAGE_SIZE-1, "%d\n",
 218                         bdi_cap_stable_pages_required(bdi) ? 1 : 0);
 219 }
 220 static DEVICE_ATTR_RO(stable_pages_required);
 221
 222 static struct attribute *bdi_dev_attrs[] = {
 223         &dev_attr_read_ahead_kb.attr,
 224         &dev_attr_min_ratio.attr,
 225         &dev_attr_max_ratio.attr,
 226         &dev_attr_stable_pages_required.attr,
 227         NULL,
 228 };
 229 ATTRIBUTE_GROUPS(bdi_dev);
 230
 231 static __init int bdi_class_init(void)
 232 {
 233         bdi_class = class_create(THIS_MODULE, "bdi");
 234         if (IS_ERR(bdi_class))
 235                 return PTR_ERR(bdi_class);
 236
 237         bdi_class->dev_groups = bdi_dev_groups;
 238         bdi_debug_init();
 239         return 0;
 240 }
 241 postcore_initcall(bdi_class_init);
 242
 243 static int __init default_bdi_init(void)
 244 {
 245         int err;
 246
 247         bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
 248                                               WQ_UNBOUND | WQ_SYSFS, 0);
 249         if (!bdi_wq)
 250                 return -ENOMEM;
 251
 252         err = bdi_init(&noop_backing_dev_info);
 253
 254         return err;
 255 }
 256 subsys_initcall(default_bdi_init);
 257
 258 int bdi_has_dirty_io(struct backing_dev_info *bdi)
 259 {
 260         return wb_has_dirty_io(&bdi->wb);
 261 }
 262
 263 /*
 264  * This function is used when the first inode for this bdi 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 bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
 278 {
 279         unsigned long timeout;
 280
 281         timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
 282         spin_lock_bh(&bdi->wb_lock);
 283         if (test_bit(BDI_registered, &bdi->state))
 284                 queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout);
 285         spin_unlock_bh(&bdi->wb_lock);
 286 }
 287
 288 /*
 289  * Remove bdi from bdi_list, and ensure that it is no longer visible
 290  */
 291 static void bdi_remove_from_list(struct backing_dev_info *bdi)
 292 {
 293         spin_lock_bh(&bdi_lock);
 294         list_del_rcu(&bdi->bdi_list);
 295         spin_unlock_bh(&bdi_lock);
 296
 297         synchronize_rcu_expedited();
 298 }
 299
 300 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 301                 const char *fmt, ...)
 302 {
 303         va_list args;
 304         struct device *dev;
 305
 306         if (bdi->dev)   /* The driver needs to use separate queues per device */
 307                 return 0;
 308
 309         va_start(args, fmt);
 310         dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
 311         va_end(args);
 312         if (IS_ERR(dev))
 313                 return PTR_ERR(dev);
 314
 315         bdi->dev = dev;
 316
 317         bdi_debug_register(bdi, dev_name(dev));
 318         set_bit(BDI_registered, &bdi->state);
 319
 320         spin_lock_bh(&bdi_lock);
 321         list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
 322         spin_unlock_bh(&bdi_lock);
 323
 324         trace_writeback_bdi_register(bdi);
 325         return 0;
 326 }
 327 EXPORT_SYMBOL(bdi_register);
 328
 329 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
 330 {
 331         return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
 332 }
 333 EXPORT_SYMBOL(bdi_register_dev);
 334
 335 /*
 336  * Remove bdi from the global list and shutdown any threads we have running
 337  */
 338 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
 339 {
 340         /* Make sure nobody queues further work */
 341         spin_lock_bh(&bdi->wb_lock);
 342         if (!test_and_clear_bit(BDI_registered, &bdi->state)) {
 343                 spin_unlock_bh(&bdi->wb_lock);
 344                 return;
 345         }
 346         spin_unlock_bh(&bdi->wb_lock);
 347
 348         /*
 349          * Make sure nobody finds us on the bdi_list anymore
 350          */
 351         bdi_remove_from_list(bdi);
 352
 353         /*
 354          * Drain work list and shutdown the delayed_work.  At this point,
 355          * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
 356          * is dying and its work_list needs to be drained no matter what.
 357          */
 358         mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 359         flush_delayed_work(&bdi->wb.dwork);
 360 }
 361
 362 /*
 363  * Called when the device behind @bdi has been removed or ejected.
 364  *
 365  * We can't really do much here except for reducing the dirty ratio at
 366  * the moment.  In the future we should be able to set a flag so that
 367  * the filesystem can handle errors at mark_inode_dirty time instead
 368  * of only at writeback time.
 369  */
 370 void bdi_unregister(struct backing_dev_info *bdi)
 371 {
 372         if (WARN_ON_ONCE(!bdi->dev))
 373                 return;
 374
 375         bdi_set_min_ratio(bdi, 0);
 376 }
 377 EXPORT_SYMBOL(bdi_unregister);
 378
 379 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
 380 {
 381         memset(wb, 0, sizeof(*wb));
 382
 383         wb->bdi = bdi;
 384         wb->last_old_flush = jiffies;
 385         INIT_LIST_HEAD(&wb->b_dirty);
 386         INIT_LIST_HEAD(&wb->b_io);
 387         INIT_LIST_HEAD(&wb->b_more_io);
 388         INIT_LIST_HEAD(&wb->b_dirty_time);
 389         spin_lock_init(&wb->list_lock);
 390         INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
 391 }
 392
 393 /*
 394  * Initial write bandwidth: 100 MB/s
 395  */
 396 #define INIT_BW         (100 << (20 - PAGE_SHIFT))
 397
 398 int bdi_init(struct backing_dev_info *bdi)
 399 {
 400         int i, err;
 401
 402         bdi->dev = NULL;
 403
 404         bdi->min_ratio = 0;
 405         bdi->max_ratio = 100;
 406         bdi->max_prop_frac = FPROP_FRAC_BASE;
 407         spin_lock_init(&bdi->wb_lock);
 408         INIT_LIST_HEAD(&bdi->bdi_list);
 409         INIT_LIST_HEAD(&bdi->work_list);
 410
 411         bdi_wb_init(&bdi->wb, bdi);
 412
 413         for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
 414                 err = percpu_counter_init(&bdi->bdi_stat[i], 0, GFP_KERNEL);
 415                 if (err)
 416                         goto err;
 417         }
 418
 419         bdi->dirty_exceeded = 0;
 420
 421         bdi->bw_time_stamp = jiffies;
 422         bdi->written_stamp = 0;
 423
 424         bdi->balanced_dirty_ratelimit = INIT_BW;
 425         bdi->dirty_ratelimit = INIT_BW;
 426         bdi->write_bandwidth = INIT_BW;
 427         bdi->avg_write_bandwidth = INIT_BW;
 428
 429         err = fprop_local_init_percpu(&bdi->completions, GFP_KERNEL);
 430
 431         if (err) {
 432 err:
 433                 while (i--)
 434                         percpu_counter_destroy(&bdi->bdi_stat[i]);
 435         }
 436
 437         return err;
 438 }
 439 EXPORT_SYMBOL(bdi_init);
 440
 441 void bdi_destroy(struct backing_dev_info *bdi)
 442 {
 443         int i;
 444
 445         bdi_wb_shutdown(bdi);
 446
 447         WARN_ON(!list_empty(&bdi->work_list));
 448         WARN_ON(delayed_work_pending(&bdi->wb.dwork));
 449
 450         if (bdi->dev) {
 451                 bdi_debug_unregister(bdi);
 452                 device_unregister(bdi->dev);
 453                 bdi->dev = NULL;
 454         }
 455
 456         for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
 457                 percpu_counter_destroy(&bdi->bdi_stat[i]);
 458         fprop_local_destroy_percpu(&bdi->completions);
 459 }
 460 EXPORT_SYMBOL(bdi_destroy);
 461
 462 /*
 463  * For use from filesystems to quickly init and register a bdi associated
 464  * with dirty writeback
 465  */
 466 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name)
 467 {
 468         int err;
 469
 470         bdi->name = name;
 471         bdi->capabilities = 0;
 472         err = bdi_init(bdi);
 473         if (err)
 474                 return err;
 475
 476         err = bdi_register(bdi, NULL, "%.28s-%ld", name,
 477                            atomic_long_inc_return(&bdi_seq));
 478         if (err) {
 479                 bdi_destroy(bdi);
 480                 return err;
 481         }
 482
 483         return 0;
 484 }
 485 EXPORT_SYMBOL(bdi_setup_and_register);
 486
 487 static wait_queue_head_t congestion_wqh[2] = {
 488                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
 489                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
 490         };
 491 static atomic_t nr_bdi_congested[2];
 492
 493 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 494 {
 495         enum bdi_state bit;
 496         wait_queue_head_t *wqh = &congestion_wqh[sync];
 497
 498         bit = sync ? BDI_sync_congested : BDI_async_congested;
 499         if (test_and_clear_bit(bit, &bdi->state))
 500                 atomic_dec(&nr_bdi_congested[sync]);
 501         smp_mb__after_atomic();
 502         if (waitqueue_active(wqh))
 503                 wake_up(wqh);
 504 }
 505 EXPORT_SYMBOL(clear_bdi_congested);
 506
 507 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
 508 {
 509         enum bdi_state bit;
 510
 511         bit = sync ? BDI_sync_congested : BDI_async_congested;
 512         if (!test_and_set_bit(bit, &bdi->state))
 513                 atomic_inc(&nr_bdi_congested[sync]);
 514 }
 515 EXPORT_SYMBOL(set_bdi_congested);
 516
 517 /**
 518  * congestion_wait - wait for a backing_dev to become uncongested
 519  * @sync: SYNC or ASYNC IO
 520  * @timeout: timeout in jiffies
 521  *
 522  * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
 523  * write congestion.  If no backing_devs are congested then just wait for the
 524  * next write to be completed.
 525  */
 526 long congestion_wait(int sync, long timeout)
 527 {
 528         long ret;
 529         unsigned long start = jiffies;
 530         DEFINE_WAIT(wait);
 531         wait_queue_head_t *wqh = &congestion_wqh[sync];
 532
 533         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 534         ret = io_schedule_timeout(timeout);
 535         finish_wait(wqh, &wait);
 536
 537         trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
 538                                         jiffies_to_usecs(jiffies - start));
 539
 540         return ret;
 541 }
 542 EXPORT_SYMBOL(congestion_wait);
 543
 544 /**
 545  * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
 546  * @zone: A zone to check if it is heavily congested
 547  * @sync: SYNC or ASYNC IO
 548  * @timeout: timeout in jiffies
 549  *
 550  * In the event of a congested backing_dev (any backing_dev) and the given
 551  * @zone has experienced recent congestion, this waits for up to @timeout
 552  * jiffies for either a BDI to exit congestion of the given @sync queue
 553  * or a write to complete.
 554  *
 555  * In the absence of zone congestion, cond_resched() is called to yield
 556  * the processor if necessary but otherwise does not sleep.
 557  *
 558  * The return value is 0 if the sleep is for the full timeout. Otherwise,
 559  * it is the number of jiffies that were still remaining when the function
 560  * returned. return_value == timeout implies the function did not sleep.
 561  */
 562 long wait_iff_congested(struct zone *zone, int sync, long timeout)
 563 {
 564         long ret;
 565         unsigned long start = jiffies;
 566         DEFINE_WAIT(wait);
 567         wait_queue_head_t *wqh = &congestion_wqh[sync];
 568
 569         /*
 570          * If there is no congestion, or heavy congestion is not being
 571          * encountered in the current zone, yield if necessary instead
 572          * of sleeping on the congestion queue
 573          */
 574         if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
 575             !test_bit(ZONE_CONGESTED, &zone->flags)) {
 576                 cond_resched();
 577
 578                 /* In case we scheduled, work out time remaining */
 579                 ret = timeout - (jiffies - start);
 580                 if (ret < 0)
 581                         ret = 0;
 582
 583                 goto out;
 584         }
 585
 586         /* Sleep until uncongested or a write happens */
 587         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 588         ret = io_schedule_timeout(timeout);
 589         finish_wait(wqh, &wait);
 590
 591 out:
 592         trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
 593                                         jiffies_to_usecs(jiffies - start));
 594
 595         return ret;
 596 }
 597 EXPORT_SYMBOL(wait_iff_congested);
 598
 599 int pdflush_proc_obsolete(struct ctl_table *table, int write,
 600                         void __user *buffer, size_t *lenp, loff_t *ppos)
 601 {
 602         char kbuf[] = "0\n";
 603
 604         if (*ppos || *lenp < sizeof(kbuf)) {
 605                 *lenp = 0;
 606                 return 0;
 607         }
 608
 609         if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
 610                 return -EFAULT;
 611         printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",
 612                         table->procname);
 613
 614         *lenp = 2;
 615         *ppos += *lenp;
 616         return 2;
 617 }