| 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 | |
| 14 | void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) |
| 15 | { |
| 16 | } |
| 17 | EXPORT_SYMBOL(default_unplug_io_fn); |
| 18 | |
| 19 | struct backing_dev_info default_backing_dev_info = { |
| 20 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, |
| 21 | .state = 0, |
| 22 | .capabilities = BDI_CAP_MAP_COPY, |
| 23 | .unplug_io_fn = default_unplug_io_fn, |
| 24 | }; |
| 25 | EXPORT_SYMBOL_GPL(default_backing_dev_info); |
| 26 | |
| 27 | static struct class *bdi_class; |
| 28 | DEFINE_SPINLOCK(bdi_lock); |
| 29 | LIST_HEAD(bdi_list); |
| 30 | LIST_HEAD(bdi_pending_list); |
| 31 | |
| 32 | static struct task_struct *sync_supers_tsk; |
| 33 | static struct timer_list sync_supers_timer; |
| 34 | |
| 35 | static int bdi_sync_supers(void *); |
| 36 | static void sync_supers_timer_fn(unsigned long); |
| 37 | static void arm_supers_timer(void); |
| 38 | |
| 39 | static void bdi_add_default_flusher_task(struct backing_dev_info *bdi); |
| 40 | |
| 41 | #ifdef CONFIG_DEBUG_FS |
| 42 | #include <linux/debugfs.h> |
| 43 | #include <linux/seq_file.h> |
| 44 | |
| 45 | static struct dentry *bdi_debug_root; |
| 46 | |
| 47 | static void bdi_debug_init(void) |
| 48 | { |
| 49 | bdi_debug_root = debugfs_create_dir("bdi", NULL); |
| 50 | } |
| 51 | |
| 52 | static int bdi_debug_stats_show(struct seq_file *m, void *v) |
| 53 | { |
| 54 | struct backing_dev_info *bdi = m->private; |
| 55 | struct bdi_writeback *wb; |
| 56 | unsigned long background_thresh; |
| 57 | unsigned long dirty_thresh; |
| 58 | unsigned long bdi_thresh; |
| 59 | unsigned long nr_dirty, nr_io, nr_more_io, nr_wb; |
| 60 | struct inode *inode; |
| 61 | |
| 62 | /* |
| 63 | * inode lock is enough here, the bdi->wb_list is protected by |
| 64 | * RCU on the reader side |
| 65 | */ |
| 66 | nr_wb = nr_dirty = nr_io = nr_more_io = 0; |
| 67 | spin_lock(&inode_lock); |
| 68 | list_for_each_entry(wb, &bdi->wb_list, list) { |
| 69 | nr_wb++; |
| 70 | list_for_each_entry(inode, &wb->b_dirty, i_list) |
| 71 | nr_dirty++; |
| 72 | list_for_each_entry(inode, &wb->b_io, i_list) |
| 73 | nr_io++; |
| 74 | list_for_each_entry(inode, &wb->b_more_io, i_list) |
| 75 | nr_more_io++; |
| 76 | } |
| 77 | spin_unlock(&inode_lock); |
| 78 | |
| 79 | get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi); |
| 80 | |
| 81 | #define K(x) ((x) << (PAGE_SHIFT - 10)) |
| 82 | seq_printf(m, |
| 83 | "BdiWriteback: %8lu kB\n" |
| 84 | "BdiReclaimable: %8lu kB\n" |
| 85 | "BdiDirtyThresh: %8lu kB\n" |
| 86 | "DirtyThresh: %8lu kB\n" |
| 87 | "BackgroundThresh: %8lu kB\n" |
| 88 | "WriteBack threads:%8lu\n" |
| 89 | "b_dirty: %8lu\n" |
| 90 | "b_io: %8lu\n" |
| 91 | "b_more_io: %8lu\n" |
| 92 | "bdi_list: %8u\n" |
| 93 | "state: %8lx\n" |
| 94 | "wb_mask: %8lx\n" |
| 95 | "wb_list: %8u\n" |
| 96 | "wb_cnt: %8u\n", |
| 97 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
| 98 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), |
| 99 | K(bdi_thresh), K(dirty_thresh), |
| 100 | K(background_thresh), nr_wb, nr_dirty, nr_io, nr_more_io, |
| 101 | !list_empty(&bdi->bdi_list), bdi->state, bdi->wb_mask, |
| 102 | !list_empty(&bdi->wb_list), bdi->wb_cnt); |
| 103 | #undef K |
| 104 | |
| 105 | return 0; |
| 106 | } |
| 107 | |
| 108 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) |
| 109 | { |
| 110 | return single_open(file, bdi_debug_stats_show, inode->i_private); |
| 111 | } |
| 112 | |
| 113 | static const struct file_operations bdi_debug_stats_fops = { |
| 114 | .open = bdi_debug_stats_open, |
| 115 | .read = seq_read, |
| 116 | .llseek = seq_lseek, |
| 117 | .release = single_release, |
| 118 | }; |
| 119 | |
| 120 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) |
| 121 | { |
| 122 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); |
| 123 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, |
| 124 | bdi, &bdi_debug_stats_fops); |
| 125 | } |
| 126 | |
| 127 | static void bdi_debug_unregister(struct backing_dev_info *bdi) |
| 128 | { |
| 129 | debugfs_remove(bdi->debug_stats); |
| 130 | debugfs_remove(bdi->debug_dir); |
| 131 | } |
| 132 | #else |
| 133 | static inline void bdi_debug_init(void) |
| 134 | { |
| 135 | } |
| 136 | static inline void bdi_debug_register(struct backing_dev_info *bdi, |
| 137 | const char *name) |
| 138 | { |
| 139 | } |
| 140 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) |
| 141 | { |
| 142 | } |
| 143 | #endif |
| 144 | |
| 145 | static ssize_t read_ahead_kb_store(struct device *dev, |
| 146 | struct device_attribute *attr, |
| 147 | const char *buf, size_t count) |
| 148 | { |
| 149 | struct backing_dev_info *bdi = dev_get_drvdata(dev); |
| 150 | char *end; |
| 151 | unsigned long read_ahead_kb; |
| 152 | ssize_t ret = -EINVAL; |
| 153 | |
| 154 | read_ahead_kb = simple_strtoul(buf, &end, 10); |
| 155 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { |
| 156 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); |
| 157 | ret = count; |
| 158 | } |
| 159 | return ret; |
| 160 | } |
| 161 | |
| 162 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) |
| 163 | |
| 164 | #define BDI_SHOW(name, expr) \ |
| 165 | static ssize_t name##_show(struct device *dev, \ |
| 166 | struct device_attribute *attr, char *page) \ |
| 167 | { \ |
| 168 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ |
| 169 | \ |
| 170 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ |
| 171 | } |
| 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 | char *end; |
| 180 | unsigned int ratio; |
| 181 | ssize_t ret = -EINVAL; |
| 182 | |
| 183 | ratio = simple_strtoul(buf, &end, 10); |
| 184 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { |
| 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 | char *end; |
| 198 | unsigned int ratio; |
| 199 | ssize_t ret = -EINVAL; |
| 200 | |
| 201 | ratio = simple_strtoul(buf, &end, 10); |
| 202 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { |
| 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 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) |
| 212 | |
| 213 | static struct device_attribute bdi_dev_attrs[] = { |
| 214 | __ATTR_RW(read_ahead_kb), |
| 215 | __ATTR_RW(min_ratio), |
| 216 | __ATTR_RW(max_ratio), |
| 217 | __ATTR_NULL, |
| 218 | }; |
| 219 | |
| 220 | static __init int bdi_class_init(void) |
| 221 | { |
| 222 | bdi_class = class_create(THIS_MODULE, "bdi"); |
| 223 | bdi_class->dev_attrs = bdi_dev_attrs; |
| 224 | bdi_debug_init(); |
| 225 | return 0; |
| 226 | } |
| 227 | postcore_initcall(bdi_class_init); |
| 228 | |
| 229 | static int __init default_bdi_init(void) |
| 230 | { |
| 231 | int err; |
| 232 | |
| 233 | sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers"); |
| 234 | BUG_ON(IS_ERR(sync_supers_tsk)); |
| 235 | |
| 236 | init_timer(&sync_supers_timer); |
| 237 | setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0); |
| 238 | arm_supers_timer(); |
| 239 | |
| 240 | err = bdi_init(&default_backing_dev_info); |
| 241 | if (!err) |
| 242 | bdi_register(&default_backing_dev_info, NULL, "default"); |
| 243 | |
| 244 | return err; |
| 245 | } |
| 246 | subsys_initcall(default_bdi_init); |
| 247 | |
| 248 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
| 249 | { |
| 250 | memset(wb, 0, sizeof(*wb)); |
| 251 | |
| 252 | wb->bdi = bdi; |
| 253 | wb->last_old_flush = jiffies; |
| 254 | INIT_LIST_HEAD(&wb->b_dirty); |
| 255 | INIT_LIST_HEAD(&wb->b_io); |
| 256 | INIT_LIST_HEAD(&wb->b_more_io); |
| 257 | } |
| 258 | |
| 259 | static void bdi_task_init(struct backing_dev_info *bdi, |
| 260 | struct bdi_writeback *wb) |
| 261 | { |
| 262 | struct task_struct *tsk = current; |
| 263 | |
| 264 | spin_lock(&bdi->wb_lock); |
| 265 | list_add_tail_rcu(&wb->list, &bdi->wb_list); |
| 266 | spin_unlock(&bdi->wb_lock); |
| 267 | |
| 268 | tsk->flags |= PF_FLUSHER | PF_SWAPWRITE; |
| 269 | set_freezable(); |
| 270 | |
| 271 | /* |
| 272 | * Our parent may run at a different priority, just set us to normal |
| 273 | */ |
| 274 | set_user_nice(tsk, 0); |
| 275 | } |
| 276 | |
| 277 | static int bdi_start_fn(void *ptr) |
| 278 | { |
| 279 | struct bdi_writeback *wb = ptr; |
| 280 | struct backing_dev_info *bdi = wb->bdi; |
| 281 | int ret; |
| 282 | |
| 283 | /* |
| 284 | * Add us to the active bdi_list |
| 285 | */ |
| 286 | spin_lock(&bdi_lock); |
| 287 | list_add(&bdi->bdi_list, &bdi_list); |
| 288 | spin_unlock(&bdi_lock); |
| 289 | |
| 290 | bdi_task_init(bdi, wb); |
| 291 | |
| 292 | /* |
| 293 | * Clear pending bit and wakeup anybody waiting to tear us down |
| 294 | */ |
| 295 | clear_bit(BDI_pending, &bdi->state); |
| 296 | smp_mb__after_clear_bit(); |
| 297 | wake_up_bit(&bdi->state, BDI_pending); |
| 298 | |
| 299 | ret = bdi_writeback_task(wb); |
| 300 | |
| 301 | /* |
| 302 | * Remove us from the list |
| 303 | */ |
| 304 | spin_lock(&bdi->wb_lock); |
| 305 | list_del_rcu(&wb->list); |
| 306 | spin_unlock(&bdi->wb_lock); |
| 307 | |
| 308 | /* |
| 309 | * Flush any work that raced with us exiting. No new work |
| 310 | * will be added, since this bdi isn't discoverable anymore. |
| 311 | */ |
| 312 | if (!list_empty(&bdi->work_list)) |
| 313 | wb_do_writeback(wb, 1); |
| 314 | |
| 315 | wb->task = NULL; |
| 316 | return ret; |
| 317 | } |
| 318 | |
| 319 | int bdi_has_dirty_io(struct backing_dev_info *bdi) |
| 320 | { |
| 321 | return wb_has_dirty_io(&bdi->wb); |
| 322 | } |
| 323 | |
| 324 | static void bdi_flush_io(struct backing_dev_info *bdi) |
| 325 | { |
| 326 | struct writeback_control wbc = { |
| 327 | .bdi = bdi, |
| 328 | .sync_mode = WB_SYNC_NONE, |
| 329 | .older_than_this = NULL, |
| 330 | .range_cyclic = 1, |
| 331 | .nr_to_write = 1024, |
| 332 | }; |
| 333 | |
| 334 | writeback_inodes_wbc(&wbc); |
| 335 | } |
| 336 | |
| 337 | /* |
| 338 | * kupdated() used to do this. We cannot do it from the bdi_forker_task() |
| 339 | * or we risk deadlocking on ->s_umount. The longer term solution would be |
| 340 | * to implement sync_supers_bdi() or similar and simply do it from the |
| 341 | * bdi writeback tasks individually. |
| 342 | */ |
| 343 | static int bdi_sync_supers(void *unused) |
| 344 | { |
| 345 | set_user_nice(current, 0); |
| 346 | |
| 347 | while (!kthread_should_stop()) { |
| 348 | set_current_state(TASK_INTERRUPTIBLE); |
| 349 | schedule(); |
| 350 | |
| 351 | /* |
| 352 | * Do this periodically, like kupdated() did before. |
| 353 | */ |
| 354 | sync_supers(); |
| 355 | } |
| 356 | |
| 357 | return 0; |
| 358 | } |
| 359 | |
| 360 | static void arm_supers_timer(void) |
| 361 | { |
| 362 | unsigned long next; |
| 363 | |
| 364 | next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies; |
| 365 | mod_timer(&sync_supers_timer, round_jiffies_up(next)); |
| 366 | } |
| 367 | |
| 368 | static void sync_supers_timer_fn(unsigned long unused) |
| 369 | { |
| 370 | wake_up_process(sync_supers_tsk); |
| 371 | arm_supers_timer(); |
| 372 | } |
| 373 | |
| 374 | static int bdi_forker_task(void *ptr) |
| 375 | { |
| 376 | struct bdi_writeback *me = ptr; |
| 377 | |
| 378 | bdi_task_init(me->bdi, me); |
| 379 | |
| 380 | for (;;) { |
| 381 | struct backing_dev_info *bdi, *tmp; |
| 382 | struct bdi_writeback *wb; |
| 383 | |
| 384 | /* |
| 385 | * Temporary measure, we want to make sure we don't see |
| 386 | * dirty data on the default backing_dev_info |
| 387 | */ |
| 388 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) |
| 389 | wb_do_writeback(me, 0); |
| 390 | |
| 391 | spin_lock(&bdi_lock); |
| 392 | |
| 393 | /* |
| 394 | * Check if any existing bdi's have dirty data without |
| 395 | * a thread registered. If so, set that up. |
| 396 | */ |
| 397 | list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) { |
| 398 | if (bdi->wb.task) |
| 399 | continue; |
| 400 | if (list_empty(&bdi->work_list) && |
| 401 | !bdi_has_dirty_io(bdi)) |
| 402 | continue; |
| 403 | |
| 404 | bdi_add_default_flusher_task(bdi); |
| 405 | } |
| 406 | |
| 407 | set_current_state(TASK_INTERRUPTIBLE); |
| 408 | |
| 409 | if (list_empty(&bdi_pending_list)) { |
| 410 | unsigned long wait; |
| 411 | |
| 412 | spin_unlock(&bdi_lock); |
| 413 | wait = msecs_to_jiffies(dirty_writeback_interval * 10); |
| 414 | schedule_timeout(wait); |
| 415 | try_to_freeze(); |
| 416 | continue; |
| 417 | } |
| 418 | |
| 419 | __set_current_state(TASK_RUNNING); |
| 420 | |
| 421 | /* |
| 422 | * This is our real job - check for pending entries in |
| 423 | * bdi_pending_list, and create the tasks that got added |
| 424 | */ |
| 425 | bdi = list_entry(bdi_pending_list.next, struct backing_dev_info, |
| 426 | bdi_list); |
| 427 | list_del_init(&bdi->bdi_list); |
| 428 | spin_unlock(&bdi_lock); |
| 429 | |
| 430 | wb = &bdi->wb; |
| 431 | wb->task = kthread_run(bdi_start_fn, wb, "flush-%s", |
| 432 | dev_name(bdi->dev)); |
| 433 | /* |
| 434 | * If task creation fails, then readd the bdi to |
| 435 | * the pending list and force writeout of the bdi |
| 436 | * from this forker thread. That will free some memory |
| 437 | * and we can try again. |
| 438 | */ |
| 439 | if (IS_ERR(wb->task)) { |
| 440 | wb->task = NULL; |
| 441 | |
| 442 | /* |
| 443 | * Add this 'bdi' to the back, so we get |
| 444 | * a chance to flush other bdi's to free |
| 445 | * memory. |
| 446 | */ |
| 447 | spin_lock(&bdi_lock); |
| 448 | list_add_tail(&bdi->bdi_list, &bdi_pending_list); |
| 449 | spin_unlock(&bdi_lock); |
| 450 | |
| 451 | bdi_flush_io(bdi); |
| 452 | } |
| 453 | } |
| 454 | |
| 455 | return 0; |
| 456 | } |
| 457 | |
| 458 | /* |
| 459 | * Add the default flusher task that gets created for any bdi |
| 460 | * that has dirty data pending writeout |
| 461 | */ |
| 462 | void static bdi_add_default_flusher_task(struct backing_dev_info *bdi) |
| 463 | { |
| 464 | if (!bdi_cap_writeback_dirty(bdi)) |
| 465 | return; |
| 466 | |
| 467 | /* |
| 468 | * Check with the helper whether to proceed adding a task. Will only |
| 469 | * abort if we two or more simultanous calls to |
| 470 | * bdi_add_default_flusher_task() occured, further additions will block |
| 471 | * waiting for previous additions to finish. |
| 472 | */ |
| 473 | if (!test_and_set_bit(BDI_pending, &bdi->state)) { |
| 474 | list_move_tail(&bdi->bdi_list, &bdi_pending_list); |
| 475 | |
| 476 | /* |
| 477 | * We are now on the pending list, wake up bdi_forker_task() |
| 478 | * to finish the job and add us back to the active bdi_list |
| 479 | */ |
| 480 | wake_up_process(default_backing_dev_info.wb.task); |
| 481 | } |
| 482 | } |
| 483 | |
| 484 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
| 485 | const char *fmt, ...) |
| 486 | { |
| 487 | va_list args; |
| 488 | int ret = 0; |
| 489 | struct device *dev; |
| 490 | |
| 491 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
| 492 | goto exit; |
| 493 | |
| 494 | va_start(args, fmt); |
| 495 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
| 496 | va_end(args); |
| 497 | if (IS_ERR(dev)) { |
| 498 | ret = PTR_ERR(dev); |
| 499 | goto exit; |
| 500 | } |
| 501 | |
| 502 | spin_lock(&bdi_lock); |
| 503 | list_add_tail(&bdi->bdi_list, &bdi_list); |
| 504 | spin_unlock(&bdi_lock); |
| 505 | |
| 506 | bdi->dev = dev; |
| 507 | |
| 508 | /* |
| 509 | * Just start the forker thread for our default backing_dev_info, |
| 510 | * and add other bdi's to the list. They will get a thread created |
| 511 | * on-demand when they need it. |
| 512 | */ |
| 513 | if (bdi_cap_flush_forker(bdi)) { |
| 514 | struct bdi_writeback *wb = &bdi->wb; |
| 515 | |
| 516 | wb->task = kthread_run(bdi_forker_task, wb, "bdi-%s", |
| 517 | dev_name(dev)); |
| 518 | if (IS_ERR(wb->task)) { |
| 519 | wb->task = NULL; |
| 520 | ret = -ENOMEM; |
| 521 | |
| 522 | spin_lock(&bdi_lock); |
| 523 | list_del(&bdi->bdi_list); |
| 524 | spin_unlock(&bdi_lock); |
| 525 | goto exit; |
| 526 | } |
| 527 | } |
| 528 | |
| 529 | bdi_debug_register(bdi, dev_name(dev)); |
| 530 | exit: |
| 531 | return ret; |
| 532 | } |
| 533 | EXPORT_SYMBOL(bdi_register); |
| 534 | |
| 535 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) |
| 536 | { |
| 537 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); |
| 538 | } |
| 539 | EXPORT_SYMBOL(bdi_register_dev); |
| 540 | |
| 541 | /* |
| 542 | * Remove bdi from the global list and shutdown any threads we have running |
| 543 | */ |
| 544 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) |
| 545 | { |
| 546 | struct bdi_writeback *wb; |
| 547 | |
| 548 | if (!bdi_cap_writeback_dirty(bdi)) |
| 549 | return; |
| 550 | |
| 551 | /* |
| 552 | * If setup is pending, wait for that to complete first |
| 553 | */ |
| 554 | wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait, |
| 555 | TASK_UNINTERRUPTIBLE); |
| 556 | |
| 557 | /* |
| 558 | * Make sure nobody finds us on the bdi_list anymore |
| 559 | */ |
| 560 | spin_lock(&bdi_lock); |
| 561 | list_del(&bdi->bdi_list); |
| 562 | spin_unlock(&bdi_lock); |
| 563 | |
| 564 | /* |
| 565 | * Finally, kill the kernel threads. We don't need to be RCU |
| 566 | * safe anymore, since the bdi is gone from visibility. |
| 567 | */ |
| 568 | list_for_each_entry(wb, &bdi->wb_list, list) |
| 569 | kthread_stop(wb->task); |
| 570 | } |
| 571 | |
| 572 | void bdi_unregister(struct backing_dev_info *bdi) |
| 573 | { |
| 574 | if (bdi->dev) { |
| 575 | if (!bdi_cap_flush_forker(bdi)) |
| 576 | bdi_wb_shutdown(bdi); |
| 577 | bdi_debug_unregister(bdi); |
| 578 | device_unregister(bdi->dev); |
| 579 | bdi->dev = NULL; |
| 580 | } |
| 581 | } |
| 582 | EXPORT_SYMBOL(bdi_unregister); |
| 583 | |
| 584 | int bdi_init(struct backing_dev_info *bdi) |
| 585 | { |
| 586 | int i, err; |
| 587 | |
| 588 | bdi->dev = NULL; |
| 589 | |
| 590 | bdi->min_ratio = 0; |
| 591 | bdi->max_ratio = 100; |
| 592 | bdi->max_prop_frac = PROP_FRAC_BASE; |
| 593 | spin_lock_init(&bdi->wb_lock); |
| 594 | INIT_LIST_HEAD(&bdi->bdi_list); |
| 595 | INIT_LIST_HEAD(&bdi->wb_list); |
| 596 | INIT_LIST_HEAD(&bdi->work_list); |
| 597 | |
| 598 | bdi_wb_init(&bdi->wb, bdi); |
| 599 | |
| 600 | /* |
| 601 | * Just one thread support for now, hard code mask and count |
| 602 | */ |
| 603 | bdi->wb_mask = 1; |
| 604 | bdi->wb_cnt = 1; |
| 605 | |
| 606 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
| 607 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); |
| 608 | if (err) |
| 609 | goto err; |
| 610 | } |
| 611 | |
| 612 | bdi->dirty_exceeded = 0; |
| 613 | err = prop_local_init_percpu(&bdi->completions); |
| 614 | |
| 615 | if (err) { |
| 616 | err: |
| 617 | while (i--) |
| 618 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
| 619 | } |
| 620 | |
| 621 | return err; |
| 622 | } |
| 623 | EXPORT_SYMBOL(bdi_init); |
| 624 | |
| 625 | void bdi_destroy(struct backing_dev_info *bdi) |
| 626 | { |
| 627 | int i; |
| 628 | |
| 629 | WARN_ON(bdi_has_dirty_io(bdi)); |
| 630 | |
| 631 | bdi_unregister(bdi); |
| 632 | |
| 633 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
| 634 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
| 635 | |
| 636 | prop_local_destroy_percpu(&bdi->completions); |
| 637 | } |
| 638 | EXPORT_SYMBOL(bdi_destroy); |
| 639 | |
| 640 | static wait_queue_head_t congestion_wqh[2] = { |
| 641 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), |
| 642 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) |
| 643 | }; |
| 644 | |
| 645 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
| 646 | { |
| 647 | enum bdi_state bit; |
| 648 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
| 649 | |
| 650 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
| 651 | clear_bit(bit, &bdi->state); |
| 652 | smp_mb__after_clear_bit(); |
| 653 | if (waitqueue_active(wqh)) |
| 654 | wake_up(wqh); |
| 655 | } |
| 656 | EXPORT_SYMBOL(clear_bdi_congested); |
| 657 | |
| 658 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
| 659 | { |
| 660 | enum bdi_state bit; |
| 661 | |
| 662 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
| 663 | set_bit(bit, &bdi->state); |
| 664 | } |
| 665 | EXPORT_SYMBOL(set_bdi_congested); |
| 666 | |
| 667 | /** |
| 668 | * congestion_wait - wait for a backing_dev to become uncongested |
| 669 | * @sync: SYNC or ASYNC IO |
| 670 | * @timeout: timeout in jiffies |
| 671 | * |
| 672 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit |
| 673 | * write congestion. If no backing_devs are congested then just wait for the |
| 674 | * next write to be completed. |
| 675 | */ |
| 676 | long congestion_wait(int sync, long timeout) |
| 677 | { |
| 678 | long ret; |
| 679 | DEFINE_WAIT(wait); |
| 680 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
| 681 | |
| 682 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); |
| 683 | ret = io_schedule_timeout(timeout); |
| 684 | finish_wait(wqh, &wait); |
| 685 | return ret; |
| 686 | } |
| 687 | EXPORT_SYMBOL(congestion_wait); |
| 688 | |