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;
  53         struct inode *inode;
  54
  55         nr_dirty = nr_io = nr_more_io = 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         spin_unlock(&wb->list_lock);
  64
  65         global_dirty_limits(&background_thresh, &dirty_thresh);
  66         bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
  67
  68 #define K(x) ((x) << (PAGE_SHIFT - 10))
  69         seq_printf(m,
  70                    "BdiWriteback:       %10lu kB\n"
  71                    "BdiReclaimable:     %10lu kB\n"
  72                    "BdiDirtyThresh:     %10lu kB\n"
  73                    "DirtyThresh:        %10lu kB\n"
  74                    "BackgroundThresh:   %10lu kB\n"
  75                    "BdiDirtied:         %10lu kB\n"
  76                    "BdiWritten:         %10lu kB\n"
  77                    "BdiWriteBandwidth:  %10lu kBps\n"
  78                    "b_dirty:            %10lu\n"
  79                    "b_io:               %10lu\n"
  80                    "b_more_io:          %10lu\n"
  81                    "bdi_list:           %10u\n"
  82                    "state:              %10lx\n",
  83                    (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
  84                    (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
  85                    K(bdi_thresh),
  86                    K(dirty_thresh),
  87                    K(background_thresh),
  88                    (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)),
  89                    (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)),
  90                    (unsigned long) K(bdi->write_bandwidth),
  91                    nr_dirty,
  92                    nr_io,
  93                    nr_more_io,
  94                    !list_empty(&bdi->bdi_list), bdi->state);
  95 #undef K
  96
  97         return 0;
  98 }
  99
 100 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
 101 {
 102         return single_open(file, bdi_debug_stats_show, inode->i_private);
 103 }
 104
 105 static const struct file_operations bdi_debug_stats_fops = {
 106         .open           = bdi_debug_stats_open,
 107         .read           = seq_read,
 108         .llseek         = seq_lseek,
 109         .release        = single_release,
 110 };
 111
 112 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
 113 {
 114         bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
 115         bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
 116                                                bdi, &bdi_debug_stats_fops);
 117 }
 118
 119 static void bdi_debug_unregister(struct backing_dev_info *bdi)
 120 {
 121         debugfs_remove(bdi->debug_stats);
 122         debugfs_remove(bdi->debug_dir);
 123 }
 124 #else
 125 static inline void bdi_debug_init(void)
 126 {
 127 }
 128 static inline void bdi_debug_register(struct backing_dev_info *bdi,
 129                                       const char *name)
 130 {
 131 }
 132 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
 133 {
 134 }
 135 #endif
 136
 137 static ssize_t read_ahead_kb_store(struct device *dev,
 138                                   struct device_attribute *attr,
 139                                   const char *buf, size_t count)
 140 {
 141         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 142         unsigned long read_ahead_kb;
 143         ssize_t ret;
 144
 145         ret = kstrtoul(buf, 10, &read_ahead_kb);
 146         if (ret < 0)
 147                 return ret;
 148
 149         bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
 150
 151         return count;
 152 }
 153
 154 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
 155
 156 #define BDI_SHOW(name, expr)                                            \
 157 static ssize_t name##_show(struct device *dev,                          \
 158                            struct device_attribute *attr, char *page)   \
 159 {                                                                       \
 160         struct backing_dev_info *bdi = dev_get_drvdata(dev);            \
 161                                                                         \
 162         return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);  \
 163 }                                                                       \
 164 static DEVICE_ATTR_RW(name);
 165
 166 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
 167
 168 static ssize_t min_ratio_store(struct device *dev,
 169                 struct device_attribute *attr, const char *buf, size_t count)
 170 {
 171         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 172         unsigned int ratio;
 173         ssize_t ret;
 174
 175         ret = kstrtouint(buf, 10, &ratio);
 176         if (ret < 0)
 177                 return ret;
 178
 179         ret = bdi_set_min_ratio(bdi, ratio);
 180         if (!ret)
 181                 ret = count;
 182
 183         return ret;
 184 }
 185 BDI_SHOW(min_ratio, bdi->min_ratio)
 186
 187 static ssize_t max_ratio_store(struct device *dev,
 188                 struct device_attribute *attr, const char *buf, size_t count)
 189 {
 190         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 191         unsigned int ratio;
 192         ssize_t ret;
 193
 194         ret = kstrtouint(buf, 10, &ratio);
 195         if (ret < 0)
 196                 return ret;
 197
 198         ret = bdi_set_max_ratio(bdi, ratio);
 199         if (!ret)
 200                 ret = count;
 201
 202         return ret;
 203 }
 204 BDI_SHOW(max_ratio, bdi->max_ratio)
 205
 206 static ssize_t stable_pages_required_show(struct device *dev,
 207                                           struct device_attribute *attr,
 208                                           char *page)
 209 {
 210         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 211
 212         return snprintf(page, PAGE_SIZE-1, "%d\n",
 213                         bdi_cap_stable_pages_required(bdi) ? 1 : 0);
 214 }
 215 static DEVICE_ATTR_RO(stable_pages_required);
 216
 217 static struct attribute *bdi_dev_attrs[] = {
 218         &dev_attr_read_ahead_kb.attr,
 219         &dev_attr_min_ratio.attr,
 220         &dev_attr_max_ratio.attr,
 221         &dev_attr_stable_pages_required.attr,
 222         NULL,
 223 };
 224 ATTRIBUTE_GROUPS(bdi_dev);
 225
 226 static __init int bdi_class_init(void)
 227 {
 228         bdi_class = class_create(THIS_MODULE, "bdi");
 229         if (IS_ERR(bdi_class))
 230                 return PTR_ERR(bdi_class);
 231
 232         bdi_class->dev_groups = bdi_dev_groups;
 233         bdi_debug_init();
 234         return 0;
 235 }
 236 postcore_initcall(bdi_class_init);
 237
 238 static int __init default_bdi_init(void)
 239 {
 240         int err;
 241
 242         bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
 243                                               WQ_UNBOUND | WQ_SYSFS, 0);
 244         if (!bdi_wq)
 245                 return -ENOMEM;
 246
 247         err = bdi_init(&noop_backing_dev_info);
 248
 249         return err;
 250 }
 251 subsys_initcall(default_bdi_init);
 252
 253 int bdi_has_dirty_io(struct backing_dev_info *bdi)
 254 {
 255         return wb_has_dirty_io(&bdi->wb);
 256 }
 257
 258 /*
 259  * This function is used when the first inode for this bdi is marked dirty. It
 260  * wakes-up the corresponding bdi thread which should then take care of the
 261  * periodic background write-out of dirty inodes. Since the write-out would
 262  * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
 263  * set up a timer which wakes the bdi thread up later.
 264  *
 265  * Note, we wouldn't bother setting up the timer, but this function is on the
 266  * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
 267  * by delaying the wake-up.
 268  *
 269  * We have to be careful not to postpone flush work if it is scheduled for
 270  * earlier. Thus we use queue_delayed_work().
 271  */
 272 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
 273 {
 274         unsigned long timeout;
 275
 276         timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
 277         spin_lock_bh(&bdi->wb_lock);
 278         if (test_bit(BDI_registered, &bdi->state))
 279                 queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout);
 280         spin_unlock_bh(&bdi->wb_lock);
 281 }
 282
 283 /*
 284  * Remove bdi from bdi_list, and ensure that it is no longer visible
 285  */
 286 static void bdi_remove_from_list(struct backing_dev_info *bdi)
 287 {
 288         spin_lock_bh(&bdi_lock);
 289         list_del_rcu(&bdi->bdi_list);
 290         spin_unlock_bh(&bdi_lock);
 291
 292         synchronize_rcu_expedited();
 293 }
 294
 295 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 296                 const char *fmt, ...)
 297 {
 298         va_list args;
 299         struct device *dev;
 300
 301         if (bdi->dev)   /* The driver needs to use separate queues per device */
 302                 return 0;
 303
 304         va_start(args, fmt);
 305         dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
 306         va_end(args);
 307         if (IS_ERR(dev))
 308                 return PTR_ERR(dev);
 309
 310         bdi->dev = dev;
 311
 312         bdi_debug_register(bdi, dev_name(dev));
 313         set_bit(BDI_registered, &bdi->state);
 314
 315         spin_lock_bh(&bdi_lock);
 316         list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
 317         spin_unlock_bh(&bdi_lock);
 318
 319         trace_writeback_bdi_register(bdi);
 320         return 0;
 321 }
 322 EXPORT_SYMBOL(bdi_register);
 323
 324 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
 325 {
 326         return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
 327 }
 328 EXPORT_SYMBOL(bdi_register_dev);
 329
 330 /*
 331  * Remove bdi from the global list and shutdown any threads we have running
 332  */
 333 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
 334 {
 335         /* Make sure nobody queues further work */
 336         spin_lock_bh(&bdi->wb_lock);
 337         if (!test_and_clear_bit(BDI_registered, &bdi->state)) {
 338                 spin_unlock_bh(&bdi->wb_lock);
 339                 return;
 340         }
 341         spin_unlock_bh(&bdi->wb_lock);
 342
 343         /*
 344          * Make sure nobody finds us on the bdi_list anymore
 345          */
 346         bdi_remove_from_list(bdi);
 347
 348         /*
 349          * Drain work list and shutdown the delayed_work.  At this point,
 350          * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
 351          * is dying and its work_list needs to be drained no matter what.
 352          */
 353         mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 354         flush_delayed_work(&bdi->wb.dwork);
 355 }
 356
 357 /*
 358  * Called when the device behind @bdi has been removed or ejected.
 359  *
 360  * We can't really do much here except for reducing the dirty ratio at
 361  * the moment.  In the future we should be able to set a flag so that
 362  * the filesystem can handle errors at mark_inode_dirty time instead
 363  * of only at writeback time.
 364  */
 365 void bdi_unregister(struct backing_dev_info *bdi)
 366 {
 367         if (WARN_ON_ONCE(!bdi->dev))
 368                 return;
 369
 370         bdi_set_min_ratio(bdi, 0);
 371 }
 372 EXPORT_SYMBOL(bdi_unregister);
 373
 374 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
 375 {
 376         memset(wb, 0, sizeof(*wb));
 377
 378         wb->bdi = bdi;
 379         wb->last_old_flush = jiffies;
 380         INIT_LIST_HEAD(&wb->b_dirty);
 381         INIT_LIST_HEAD(&wb->b_io);
 382         INIT_LIST_HEAD(&wb->b_more_io);
 383         spin_lock_init(&wb->list_lock);
 384         INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
 385 }
 386
 387 /*
 388  * Initial write bandwidth: 100 MB/s
 389  */
 390 #define INIT_BW         (100 << (20 - PAGE_SHIFT))
 391
 392 int bdi_init(struct backing_dev_info *bdi)
 393 {
 394         int i, err;
 395
 396         bdi->dev = NULL;
 397
 398         bdi->min_ratio = 0;
 399         bdi->max_ratio = 100;
 400         bdi->max_prop_frac = FPROP_FRAC_BASE;
 401         spin_lock_init(&bdi->wb_lock);
 402         INIT_LIST_HEAD(&bdi->bdi_list);
 403         INIT_LIST_HEAD(&bdi->work_list);
 404
 405         bdi_wb_init(&bdi->wb, bdi);
 406
 407         for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
 408                 err = percpu_counter_init(&bdi->bdi_stat[i], 0, GFP_KERNEL);
 409                 if (err)
 410                         goto err;
 411         }
 412
 413         bdi->dirty_exceeded = 0;
 414
 415         bdi->bw_time_stamp = jiffies;
 416         bdi->written_stamp = 0;
 417
 418         bdi->balanced_dirty_ratelimit = INIT_BW;
 419         bdi->dirty_ratelimit = INIT_BW;
 420         bdi->write_bandwidth = INIT_BW;
 421         bdi->avg_write_bandwidth = INIT_BW;
 422
 423         err = fprop_local_init_percpu(&bdi->completions, GFP_KERNEL);
 424
 425         if (err) {
 426 err:
 427                 while (i--)
 428                         percpu_counter_destroy(&bdi->bdi_stat[i]);
 429         }
 430
 431         return err;
 432 }
 433 EXPORT_SYMBOL(bdi_init);
 434
 435 void bdi_destroy(struct backing_dev_info *bdi)
 436 {
 437         int i;
 438
 439         bdi_wb_shutdown(bdi);
 440
 441         WARN_ON(!list_empty(&bdi->work_list));
 442         WARN_ON(delayed_work_pending(&bdi->wb.dwork));
 443
 444         if (bdi->dev) {
 445                 bdi_debug_unregister(bdi);
 446                 device_unregister(bdi->dev);
 447                 bdi->dev = NULL;
 448         }
 449
 450         for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
 451                 percpu_counter_destroy(&bdi->bdi_stat[i]);
 452         fprop_local_destroy_percpu(&bdi->completions);
 453 }
 454 EXPORT_SYMBOL(bdi_destroy);
 455
 456 /*
 457  * For use from filesystems to quickly init and register a bdi associated
 458  * with dirty writeback
 459  */
 460 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name)
 461 {
 462         int err;
 463
 464         bdi->name = name;
 465         bdi->capabilities = 0;
 466         err = bdi_init(bdi);
 467         if (err)
 468                 return err;
 469
 470         err = bdi_register(bdi, NULL, "%.28s-%ld", name,
 471                            atomic_long_inc_return(&bdi_seq));
 472         if (err) {
 473                 bdi_destroy(bdi);
 474                 return err;
 475         }
 476
 477         return 0;
 478 }
 479 EXPORT_SYMBOL(bdi_setup_and_register);
 480
 481 static wait_queue_head_t congestion_wqh[2] = {
 482                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
 483                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
 484         };
 485 static atomic_t nr_bdi_congested[2];
 486
 487 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 488 {
 489         enum bdi_state bit;
 490         wait_queue_head_t *wqh = &congestion_wqh[sync];
 491
 492         bit = sync ? BDI_sync_congested : BDI_async_congested;
 493         if (test_and_clear_bit(bit, &bdi->state))
 494                 atomic_dec(&nr_bdi_congested[sync]);
 495         smp_mb__after_atomic();
 496         if (waitqueue_active(wqh))
 497                 wake_up(wqh);
 498 }
 499 EXPORT_SYMBOL(clear_bdi_congested);
 500
 501 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
 502 {
 503         enum bdi_state bit;
 504
 505         bit = sync ? BDI_sync_congested : BDI_async_congested;
 506         if (!test_and_set_bit(bit, &bdi->state))
 507                 atomic_inc(&nr_bdi_congested[sync]);
 508 }
 509 EXPORT_SYMBOL(set_bdi_congested);
 510
 511 /**
 512  * congestion_wait - wait for a backing_dev to become uncongested
 513  * @sync: SYNC or ASYNC IO
 514  * @timeout: timeout in jiffies
 515  *
 516  * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
 517  * write congestion.  If no backing_devs are congested then just wait for the
 518  * next write to be completed.
 519  */
 520 long congestion_wait(int sync, long timeout)
 521 {
 522         long ret;
 523         unsigned long start = jiffies;
 524         DEFINE_WAIT(wait);
 525         wait_queue_head_t *wqh = &congestion_wqh[sync];
 526
 527         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 528         ret = io_schedule_timeout(timeout);
 529         finish_wait(wqh, &wait);
 530
 531         trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
 532                                         jiffies_to_usecs(jiffies - start));
 533
 534         return ret;
 535 }
 536 EXPORT_SYMBOL(congestion_wait);
 537
 538 /**
 539  * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
 540  * @zone: A zone to check if it is heavily congested
 541  * @sync: SYNC or ASYNC IO
 542  * @timeout: timeout in jiffies
 543  *
 544  * In the event of a congested backing_dev (any backing_dev) and the given
 545  * @zone has experienced recent congestion, this waits for up to @timeout
 546  * jiffies for either a BDI to exit congestion of the given @sync queue
 547  * or a write to complete.
 548  *
 549  * In the absence of zone congestion, cond_resched() is called to yield
 550  * the processor if necessary but otherwise does not sleep.
 551  *
 552  * The return value is 0 if the sleep is for the full timeout. Otherwise,
 553  * it is the number of jiffies that were still remaining when the function
 554  * returned. return_value == timeout implies the function did not sleep.
 555  */
 556 long wait_iff_congested(struct zone *zone, int sync, long timeout)
 557 {
 558         long ret;
 559         unsigned long start = jiffies;
 560         DEFINE_WAIT(wait);
 561         wait_queue_head_t *wqh = &congestion_wqh[sync];
 562
 563         /*
 564          * If there is no congestion, or heavy congestion is not being
 565          * encountered in the current zone, yield if necessary instead
 566          * of sleeping on the congestion queue
 567          */
 568         if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
 569             !test_bit(ZONE_CONGESTED, &zone->flags)) {
 570                 cond_resched();
 571
 572                 /* In case we scheduled, work out time remaining */
 573                 ret = timeout - (jiffies - start);
 574                 if (ret < 0)
 575                         ret = 0;
 576
 577                 goto out;
 578         }
 579
 580         /* Sleep until uncongested or a write happens */
 581         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 582         ret = io_schedule_timeout(timeout);
 583         finish_wait(wqh, &wait);
 584
 585 out:
 586         trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
 587                                         jiffies_to_usecs(jiffies - start));
 588
 589         return ret;
 590 }
 591 EXPORT_SYMBOL(wait_iff_congested);
 592
 593 int pdflush_proc_obsolete(struct ctl_table *table, int write,
 594                         void __user *buffer, size_t *lenp, loff_t *ppos)
 595 {
 596         char kbuf[] = "0\n";
 597
 598         if (*ppos || *lenp < sizeof(kbuf)) {
 599                 *lenp = 0;
 600                 return 0;
 601         }
 602
 603         if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
 604                 return -EFAULT;
 605         printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",
 606                         table->procname);
 607
 608         *lenp = 2;
 609         *ppos += *lenp;
 610         return 2;
 611 }