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 default_backing_dev_info = {
  18         .name           = "default",
  19         .ra_pages       = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
  20 };
  21 EXPORT_SYMBOL_GPL(default_backing_dev_info);
  22
  23 struct backing_dev_info noop_backing_dev_info = {
  24         .name           = "noop",
  25         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
  26 };
  27
  28 static struct class *bdi_class;
  29
  30 /*
  31  * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
  32  * locking.
  33  */
  34 DEFINE_SPINLOCK(bdi_lock);
  35 LIST_HEAD(bdi_list);
  36
  37 /* bdi_wq serves all asynchronous writeback tasks */
  38 struct workqueue_struct *bdi_wq;
  39
  40 #ifdef CONFIG_DEBUG_FS
  41 #include <linux/debugfs.h>
  42 #include <linux/seq_file.h>
  43
  44 static struct dentry *bdi_debug_root;
  45
  46 static void bdi_debug_init(void)
  47 {
  48         bdi_debug_root = debugfs_create_dir("bdi", NULL);
  49 }
  50
  51 static int bdi_debug_stats_show(struct seq_file *m, void *v)
  52 {
  53         struct backing_dev_info *bdi = m->private;
  54         struct bdi_writeback *wb = &bdi->wb;
  55         unsigned long background_thresh;
  56         unsigned long dirty_thresh;
  57         unsigned long bdi_thresh;
  58         unsigned long nr_dirty, nr_io, nr_more_io;
  59         struct inode *inode;
  60
  61         nr_dirty = nr_io = nr_more_io = 0;
  62         spin_lock(&wb->list_lock);
  63         list_for_each_entry(inode, &wb->b_dirty, i_wb_list)
  64                 nr_dirty++;
  65         list_for_each_entry(inode, &wb->b_io, i_wb_list)
  66                 nr_io++;
  67         list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
  68                 nr_more_io++;
  69         spin_unlock(&wb->list_lock);
  70
  71         global_dirty_limits(&background_thresh, &dirty_thresh);
  72         bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
  73
  74 #define K(x) ((x) << (PAGE_SHIFT - 10))
  75         seq_printf(m,
  76                    "BdiWriteback:       %10lu kB\n"
  77                    "BdiReclaimable:     %10lu kB\n"
  78                    "BdiDirtyThresh:     %10lu kB\n"
  79                    "DirtyThresh:        %10lu kB\n"
  80                    "BackgroundThresh:   %10lu kB\n"
  81                    "BdiDirtied:         %10lu kB\n"
  82                    "BdiWritten:         %10lu kB\n"
  83                    "BdiWriteBandwidth:  %10lu kBps\n"
  84                    "b_dirty:            %10lu\n"
  85                    "b_io:               %10lu\n"
  86                    "b_more_io:          %10lu\n"
  87                    "bdi_list:           %10u\n"
  88                    "state:              %10lx\n",
  89                    (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
  90                    (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
  91                    K(bdi_thresh),
  92                    K(dirty_thresh),
  93                    K(background_thresh),
  94                    (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)),
  95                    (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)),
  96                    (unsigned long) K(bdi->write_bandwidth),
  97                    nr_dirty,
  98                    nr_io,
  99                    nr_more_io,
 100                    !list_empty(&bdi->bdi_list), bdi->state);
 101 #undef K
 102
 103         return 0;
 104 }
 105
 106 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
 107 {
 108         return single_open(file, bdi_debug_stats_show, inode->i_private);
 109 }
 110
 111 static const struct file_operations bdi_debug_stats_fops = {
 112         .open           = bdi_debug_stats_open,
 113         .read           = seq_read,
 114         .llseek         = seq_lseek,
 115         .release        = single_release,
 116 };
 117
 118 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
 119 {
 120         bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
 121         bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
 122                                                bdi, &bdi_debug_stats_fops);
 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 void bdi_debug_register(struct backing_dev_info *bdi,
 135                                       const char *name)
 136 {
 137 }
 138 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
 139 {
 140 }
 141 #endif
 142
 143 static ssize_t read_ahead_kb_store(struct device *dev,
 144                                   struct device_attribute *attr,
 145                                   const char *buf, size_t count)
 146 {
 147         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 148         unsigned long read_ahead_kb;
 149         ssize_t ret;
 150
 151         ret = kstrtoul(buf, 10, &read_ahead_kb);
 152         if (ret < 0)
 153                 return ret;
 154
 155         bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
 156
 157         return count;
 158 }
 159
 160 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
 161
 162 #define BDI_SHOW(name, expr)                                            \
 163 static ssize_t name##_show(struct device *dev,                          \
 164                            struct device_attribute *attr, char *page)   \
 165 {                                                                       \
 166         struct backing_dev_info *bdi = dev_get_drvdata(dev);            \
 167                                                                         \
 168         return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);  \
 169 }                                                                       \
 170 static DEVICE_ATTR_RW(name);
 171
 172 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
 173
 174 static ssize_t min_ratio_store(struct device *dev,
 175                 struct device_attribute *attr, const char *buf, size_t count)
 176 {
 177         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 178         unsigned int ratio;
 179         ssize_t ret;
 180
 181         ret = kstrtouint(buf, 10, &ratio);
 182         if (ret < 0)
 183                 return ret;
 184
 185         ret = bdi_set_min_ratio(bdi, ratio);
 186         if (!ret)
 187                 ret = count;
 188
 189         return ret;
 190 }
 191 BDI_SHOW(min_ratio, bdi->min_ratio)
 192
 193 static ssize_t max_ratio_store(struct device *dev,
 194                 struct device_attribute *attr, const char *buf, size_t count)
 195 {
 196         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 197         unsigned int ratio;
 198         ssize_t ret;
 199
 200         ret = kstrtouint(buf, 10, &ratio);
 201         if (ret < 0)
 202                 return ret;
 203
 204         ret = bdi_set_max_ratio(bdi, ratio);
 205         if (!ret)
 206                 ret = count;
 207
 208         return ret;
 209 }
 210 BDI_SHOW(max_ratio, bdi->max_ratio)
 211
 212 static ssize_t stable_pages_required_show(struct device *dev,
 213                                           struct device_attribute *attr,
 214                                           char *page)
 215 {
 216         struct backing_dev_info *bdi = dev_get_drvdata(dev);
 217
 218         return snprintf(page, PAGE_SIZE-1, "%d\n",
 219                         bdi_cap_stable_pages_required(bdi) ? 1 : 0);
 220 }
 221 static DEVICE_ATTR_RO(stable_pages_required);
 222
 223 static struct attribute *bdi_dev_attrs[] = {
 224         &dev_attr_read_ahead_kb.attr,
 225         &dev_attr_min_ratio.attr,
 226         &dev_attr_max_ratio.attr,
 227         &dev_attr_stable_pages_required.attr,
 228         NULL,
 229 };
 230 ATTRIBUTE_GROUPS(bdi_dev);
 231
 232 static __init int bdi_class_init(void)
 233 {
 234         bdi_class = class_create(THIS_MODULE, "bdi");
 235         if (IS_ERR(bdi_class))
 236                 return PTR_ERR(bdi_class);
 237
 238         bdi_class->dev_groups = bdi_dev_groups;
 239         bdi_debug_init();
 240         return 0;
 241 }
 242 postcore_initcall(bdi_class_init);
 243
 244 static int __init default_bdi_init(void)
 245 {
 246         int err;
 247
 248         bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
 249                                               WQ_UNBOUND | WQ_SYSFS, 0);
 250         if (!bdi_wq)
 251                 return -ENOMEM;
 252
 253         err = bdi_init(&default_backing_dev_info);
 254         if (!err)
 255                 bdi_register(&default_backing_dev_info, NULL, "default");
 256         err = bdi_init(&noop_backing_dev_info);
 257
 258         return err;
 259 }
 260 subsys_initcall(default_bdi_init);
 261
 262 int bdi_has_dirty_io(struct backing_dev_info *bdi)
 263 {
 264         return wb_has_dirty_io(&bdi->wb);
 265 }
 266
 267 /*
 268  * This function is used when the first inode for this bdi is marked dirty. It
 269  * wakes-up the corresponding bdi thread which should then take care of the
 270  * periodic background write-out of dirty inodes. Since the write-out would
 271  * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
 272  * set up a timer which wakes the bdi thread up later.
 273  *
 274  * Note, we wouldn't bother setting up the timer, but this function is on the
 275  * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
 276  * by delaying the wake-up.
 277  *
 278  * We have to be careful not to postpone flush work if it is scheduled for
 279  * earlier. Thus we use queue_delayed_work().
 280  */
 281 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
 282 {
 283         unsigned long timeout;
 284
 285         timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
 286         spin_lock_bh(&bdi->wb_lock);
 287         if (test_bit(BDI_registered, &bdi->state))
 288                 queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout);
 289         spin_unlock_bh(&bdi->wb_lock);
 290 }
 291
 292 /*
 293  * Remove bdi from bdi_list, and ensure that it is no longer visible
 294  */
 295 static void bdi_remove_from_list(struct backing_dev_info *bdi)
 296 {
 297         spin_lock_bh(&bdi_lock);
 298         list_del_rcu(&bdi->bdi_list);
 299         spin_unlock_bh(&bdi_lock);
 300
 301         synchronize_rcu_expedited();
 302 }
 303
 304 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 305                 const char *fmt, ...)
 306 {
 307         va_list args;
 308         struct device *dev;
 309
 310         if (bdi->dev)   /* The driver needs to use separate queues per device */
 311                 return 0;
 312
 313         va_start(args, fmt);
 314         dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
 315         va_end(args);
 316         if (IS_ERR(dev))
 317                 return PTR_ERR(dev);
 318
 319         bdi->dev = dev;
 320
 321         bdi_debug_register(bdi, dev_name(dev));
 322         set_bit(BDI_registered, &bdi->state);
 323
 324         spin_lock_bh(&bdi_lock);
 325         list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
 326         spin_unlock_bh(&bdi_lock);
 327
 328         trace_writeback_bdi_register(bdi);
 329         return 0;
 330 }
 331 EXPORT_SYMBOL(bdi_register);
 332
 333 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
 334 {
 335         return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
 336 }
 337 EXPORT_SYMBOL(bdi_register_dev);
 338
 339 /*
 340  * Remove bdi from the global list and shutdown any threads we have running
 341  */
 342 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
 343 {
 344         /* Make sure nobody queues further work */
 345         spin_lock_bh(&bdi->wb_lock);
 346         if (!test_and_clear_bit(BDI_registered, &bdi->state)) {
 347                 spin_unlock_bh(&bdi->wb_lock);
 348                 return;
 349         }
 350         spin_unlock_bh(&bdi->wb_lock);
 351
 352         /*
 353          * Make sure nobody finds us on the bdi_list anymore
 354          */
 355         bdi_remove_from_list(bdi);
 356
 357         /*
 358          * Drain work list and shutdown the delayed_work.  At this point,
 359          * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
 360          * is dying and its work_list needs to be drained no matter what.
 361          */
 362         mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 363         flush_delayed_work(&bdi->wb.dwork);
 364 }
 365
 366 /*
 367  * Called when the device behind @bdi has been removed or ejected.
 368  *
 369  * We can't really do much here except for reducing the dirty ratio at
 370  * the moment.  In the future we should be able to set a flag so that
 371  * the filesystem can handle errors at mark_inode_dirty time instead
 372  * of only at writeback time.
 373  */
 374 void bdi_unregister(struct backing_dev_info *bdi)
 375 {
 376         if (WARN_ON_ONCE(!bdi->dev))
 377                 return;
 378
 379         bdi_set_min_ratio(bdi, 0);
 380 }
 381 EXPORT_SYMBOL(bdi_unregister);
 382
 383 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
 384 {
 385         memset(wb, 0, sizeof(*wb));
 386
 387         wb->bdi = bdi;
 388         wb->last_old_flush = jiffies;
 389         INIT_LIST_HEAD(&wb->b_dirty);
 390         INIT_LIST_HEAD(&wb->b_io);
 391         INIT_LIST_HEAD(&wb->b_more_io);
 392         spin_lock_init(&wb->list_lock);
 393         INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
 394 }
 395
 396 /*
 397  * Initial write bandwidth: 100 MB/s
 398  */
 399 #define INIT_BW         (100 << (20 - PAGE_SHIFT))
 400
 401 int bdi_init(struct backing_dev_info *bdi)
 402 {
 403         int i, err;
 404
 405         bdi->dev = NULL;
 406
 407         bdi->min_ratio = 0;
 408         bdi->max_ratio = 100;
 409         bdi->max_prop_frac = FPROP_FRAC_BASE;
 410         spin_lock_init(&bdi->wb_lock);
 411         INIT_LIST_HEAD(&bdi->bdi_list);
 412         INIT_LIST_HEAD(&bdi->work_list);
 413
 414         bdi_wb_init(&bdi->wb, bdi);
 415
 416         for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
 417                 err = percpu_counter_init(&bdi->bdi_stat[i], 0, GFP_KERNEL);
 418                 if (err)
 419                         goto err;
 420         }
 421
 422         bdi->dirty_exceeded = 0;
 423
 424         bdi->bw_time_stamp = jiffies;
 425         bdi->written_stamp = 0;
 426
 427         bdi->balanced_dirty_ratelimit = INIT_BW;
 428         bdi->dirty_ratelimit = INIT_BW;
 429         bdi->write_bandwidth = INIT_BW;
 430         bdi->avg_write_bandwidth = INIT_BW;
 431
 432         err = fprop_local_init_percpu(&bdi->completions, GFP_KERNEL);
 433
 434         if (err) {
 435 err:
 436                 while (i--)
 437                         percpu_counter_destroy(&bdi->bdi_stat[i]);
 438         }
 439
 440         return err;
 441 }
 442 EXPORT_SYMBOL(bdi_init);
 443
 444 void bdi_destroy(struct backing_dev_info *bdi)
 445 {
 446         int i;
 447
 448         bdi_wb_shutdown(bdi);
 449
 450         WARN_ON(!list_empty(&bdi->work_list));
 451         WARN_ON(delayed_work_pending(&bdi->wb.dwork));
 452
 453         if (bdi->dev) {
 454                 bdi_debug_unregister(bdi);
 455                 device_unregister(bdi->dev);
 456                 bdi->dev = NULL;
 457         }
 458
 459         for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
 460                 percpu_counter_destroy(&bdi->bdi_stat[i]);
 461         fprop_local_destroy_percpu(&bdi->completions);
 462 }
 463 EXPORT_SYMBOL(bdi_destroy);
 464
 465 /*
 466  * For use from filesystems to quickly init and register a bdi associated
 467  * with dirty writeback
 468  */
 469 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name)
 470 {
 471         int err;
 472
 473         bdi->name = name;
 474         bdi->capabilities = 0;
 475         err = bdi_init(bdi);
 476         if (err)
 477                 return err;
 478
 479         err = bdi_register(bdi, NULL, "%.28s-%ld", name,
 480                            atomic_long_inc_return(&bdi_seq));
 481         if (err) {
 482                 bdi_destroy(bdi);
 483                 return err;
 484         }
 485
 486         return 0;
 487 }
 488 EXPORT_SYMBOL(bdi_setup_and_register);
 489
 490 static wait_queue_head_t congestion_wqh[2] = {
 491                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
 492                 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
 493         };
 494 static atomic_t nr_bdi_congested[2];
 495
 496 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 497 {
 498         enum bdi_state bit;
 499         wait_queue_head_t *wqh = &congestion_wqh[sync];
 500
 501         bit = sync ? BDI_sync_congested : BDI_async_congested;
 502         if (test_and_clear_bit(bit, &bdi->state))
 503                 atomic_dec(&nr_bdi_congested[sync]);
 504         smp_mb__after_atomic();
 505         if (waitqueue_active(wqh))
 506                 wake_up(wqh);
 507 }
 508 EXPORT_SYMBOL(clear_bdi_congested);
 509
 510 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
 511 {
 512         enum bdi_state bit;
 513
 514         bit = sync ? BDI_sync_congested : BDI_async_congested;
 515         if (!test_and_set_bit(bit, &bdi->state))
 516                 atomic_inc(&nr_bdi_congested[sync]);
 517 }
 518 EXPORT_SYMBOL(set_bdi_congested);
 519
 520 /**
 521  * congestion_wait - wait for a backing_dev to become uncongested
 522  * @sync: SYNC or ASYNC IO
 523  * @timeout: timeout in jiffies
 524  *
 525  * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
 526  * write congestion.  If no backing_devs are congested then just wait for the
 527  * next write to be completed.
 528  */
 529 long congestion_wait(int sync, long timeout)
 530 {
 531         long ret;
 532         unsigned long start = jiffies;
 533         DEFINE_WAIT(wait);
 534         wait_queue_head_t *wqh = &congestion_wqh[sync];
 535
 536         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 537         ret = io_schedule_timeout(timeout);
 538         finish_wait(wqh, &wait);
 539
 540         trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
 541                                         jiffies_to_usecs(jiffies - start));
 542
 543         return ret;
 544 }
 545 EXPORT_SYMBOL(congestion_wait);
 546
 547 /**
 548  * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
 549  * @zone: A zone to check if it is heavily congested
 550  * @sync: SYNC or ASYNC IO
 551  * @timeout: timeout in jiffies
 552  *
 553  * In the event of a congested backing_dev (any backing_dev) and the given
 554  * @zone has experienced recent congestion, this waits for up to @timeout
 555  * jiffies for either a BDI to exit congestion of the given @sync queue
 556  * or a write to complete.
 557  *
 558  * In the absence of zone congestion, cond_resched() is called to yield
 559  * the processor if necessary but otherwise does not sleep.
 560  *
 561  * The return value is 0 if the sleep is for the full timeout. Otherwise,
 562  * it is the number of jiffies that were still remaining when the function
 563  * returned. return_value == timeout implies the function did not sleep.
 564  */
 565 long wait_iff_congested(struct zone *zone, int sync, long timeout)
 566 {
 567         long ret;
 568         unsigned long start = jiffies;
 569         DEFINE_WAIT(wait);
 570         wait_queue_head_t *wqh = &congestion_wqh[sync];
 571
 572         /*
 573          * If there is no congestion, or heavy congestion is not being
 574          * encountered in the current zone, yield if necessary instead
 575          * of sleeping on the congestion queue
 576          */
 577         if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
 578             !test_bit(ZONE_CONGESTED, &zone->flags)) {
 579                 cond_resched();
 580
 581                 /* In case we scheduled, work out time remaining */
 582                 ret = timeout - (jiffies - start);
 583                 if (ret < 0)
 584                         ret = 0;
 585
 586                 goto out;
 587         }
 588
 589         /* Sleep until uncongested or a write happens */
 590         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 591         ret = io_schedule_timeout(timeout);
 592         finish_wait(wqh, &wait);
 593
 594 out:
 595         trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
 596                                         jiffies_to_usecs(jiffies - start));
 597
 598         return ret;
 599 }
 600 EXPORT_SYMBOL(wait_iff_congested);
 601
 602 int pdflush_proc_obsolete(struct ctl_table *table, int write,
 603                         void __user *buffer, size_t *lenp, loff_t *ppos)
 604 {
 605         char kbuf[] = "0\n";
 606
 607         if (*ppos || *lenp < sizeof(kbuf)) {
 608                 *lenp = 0;
 609                 return 0;
 610         }
 611
 612         if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
 613                 return -EFAULT;
 614         printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",
 615                         table->procname);
 616
 617         *lenp = 2;
 618         *ppos += *lenp;
 619         return 2;
 620 }