| 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 | void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) |
| 18 | { |
| 19 | } |
| 20 | EXPORT_SYMBOL(default_unplug_io_fn); |
| 21 | |
| 22 | struct backing_dev_info default_backing_dev_info = { |
| 23 | .name = "default", |
| 24 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, |
| 25 | .state = 0, |
| 26 | .capabilities = BDI_CAP_MAP_COPY, |
| 27 | .unplug_io_fn = default_unplug_io_fn, |
| 28 | }; |
| 29 | EXPORT_SYMBOL_GPL(default_backing_dev_info); |
| 30 | |
| 31 | struct backing_dev_info noop_backing_dev_info = { |
| 32 | .name = "noop", |
| 33 | }; |
| 34 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); |
| 35 | |
| 36 | static struct class *bdi_class; |
| 37 | |
| 38 | /* |
| 39 | * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as |
| 40 | * reader side protection for bdi_pending_list. bdi_list has RCU reader side |
| 41 | * locking. |
| 42 | */ |
| 43 | DEFINE_SPINLOCK(bdi_lock); |
| 44 | LIST_HEAD(bdi_list); |
| 45 | LIST_HEAD(bdi_pending_list); |
| 46 | |
| 47 | static struct task_struct *sync_supers_tsk; |
| 48 | static struct timer_list sync_supers_timer; |
| 49 | |
| 50 | static int bdi_sync_supers(void *); |
| 51 | static void sync_supers_timer_fn(unsigned long); |
| 52 | |
| 53 | static void bdi_add_default_flusher_thread(struct backing_dev_info *bdi); |
| 54 | |
| 55 | #ifdef CONFIG_DEBUG_FS |
| 56 | #include <linux/debugfs.h> |
| 57 | #include <linux/seq_file.h> |
| 58 | |
| 59 | static struct dentry *bdi_debug_root; |
| 60 | |
| 61 | static void bdi_debug_init(void) |
| 62 | { |
| 63 | bdi_debug_root = debugfs_create_dir("bdi", NULL); |
| 64 | } |
| 65 | |
| 66 | static int bdi_debug_stats_show(struct seq_file *m, void *v) |
| 67 | { |
| 68 | struct backing_dev_info *bdi = m->private; |
| 69 | struct bdi_writeback *wb = &bdi->wb; |
| 70 | unsigned long background_thresh; |
| 71 | unsigned long dirty_thresh; |
| 72 | unsigned long bdi_thresh; |
| 73 | unsigned long nr_dirty, nr_io, nr_more_io, nr_wb; |
| 74 | struct inode *inode; |
| 75 | |
| 76 | nr_wb = nr_dirty = nr_io = nr_more_io = 0; |
| 77 | spin_lock(&inode_lock); |
| 78 | list_for_each_entry(inode, &wb->b_dirty, i_list) |
| 79 | nr_dirty++; |
| 80 | list_for_each_entry(inode, &wb->b_io, i_list) |
| 81 | nr_io++; |
| 82 | list_for_each_entry(inode, &wb->b_more_io, i_list) |
| 83 | nr_more_io++; |
| 84 | spin_unlock(&inode_lock); |
| 85 | |
| 86 | get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi); |
| 87 | |
| 88 | #define K(x) ((x) << (PAGE_SHIFT - 10)) |
| 89 | seq_printf(m, |
| 90 | "BdiWriteback: %8lu kB\n" |
| 91 | "BdiReclaimable: %8lu kB\n" |
| 92 | "BdiDirtyThresh: %8lu kB\n" |
| 93 | "DirtyThresh: %8lu kB\n" |
| 94 | "BackgroundThresh: %8lu kB\n" |
| 95 | "b_dirty: %8lu\n" |
| 96 | "b_io: %8lu\n" |
| 97 | "b_more_io: %8lu\n" |
| 98 | "bdi_list: %8u\n" |
| 99 | "state: %8lx\n", |
| 100 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
| 101 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), |
| 102 | K(bdi_thresh), K(dirty_thresh), |
| 103 | K(background_thresh), nr_dirty, nr_io, nr_more_io, |
| 104 | !list_empty(&bdi->bdi_list), bdi->state); |
| 105 | #undef K |
| 106 | |
| 107 | return 0; |
| 108 | } |
| 109 | |
| 110 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) |
| 111 | { |
| 112 | return single_open(file, bdi_debug_stats_show, inode->i_private); |
| 113 | } |
| 114 | |
| 115 | static const struct file_operations bdi_debug_stats_fops = { |
| 116 | .open = bdi_debug_stats_open, |
| 117 | .read = seq_read, |
| 118 | .llseek = seq_lseek, |
| 119 | .release = single_release, |
| 120 | }; |
| 121 | |
| 122 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) |
| 123 | { |
| 124 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); |
| 125 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, |
| 126 | bdi, &bdi_debug_stats_fops); |
| 127 | } |
| 128 | |
| 129 | static void bdi_debug_unregister(struct backing_dev_info *bdi) |
| 130 | { |
| 131 | debugfs_remove(bdi->debug_stats); |
| 132 | debugfs_remove(bdi->debug_dir); |
| 133 | } |
| 134 | #else |
| 135 | static inline void bdi_debug_init(void) |
| 136 | { |
| 137 | } |
| 138 | static inline void bdi_debug_register(struct backing_dev_info *bdi, |
| 139 | const char *name) |
| 140 | { |
| 141 | } |
| 142 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) |
| 143 | { |
| 144 | } |
| 145 | #endif |
| 146 | |
| 147 | static ssize_t read_ahead_kb_store(struct device *dev, |
| 148 | struct device_attribute *attr, |
| 149 | const char *buf, size_t count) |
| 150 | { |
| 151 | struct backing_dev_info *bdi = dev_get_drvdata(dev); |
| 152 | char *end; |
| 153 | unsigned long read_ahead_kb; |
| 154 | ssize_t ret = -EINVAL; |
| 155 | |
| 156 | read_ahead_kb = simple_strtoul(buf, &end, 10); |
| 157 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { |
| 158 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); |
| 159 | ret = count; |
| 160 | } |
| 161 | return ret; |
| 162 | } |
| 163 | |
| 164 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) |
| 165 | |
| 166 | #define BDI_SHOW(name, expr) \ |
| 167 | static ssize_t name##_show(struct device *dev, \ |
| 168 | struct device_attribute *attr, char *page) \ |
| 169 | { \ |
| 170 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ |
| 171 | \ |
| 172 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ |
| 173 | } |
| 174 | |
| 175 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) |
| 176 | |
| 177 | static ssize_t min_ratio_store(struct device *dev, |
| 178 | struct device_attribute *attr, const char *buf, size_t count) |
| 179 | { |
| 180 | struct backing_dev_info *bdi = dev_get_drvdata(dev); |
| 181 | char *end; |
| 182 | unsigned int ratio; |
| 183 | ssize_t ret = -EINVAL; |
| 184 | |
| 185 | ratio = simple_strtoul(buf, &end, 10); |
| 186 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { |
| 187 | ret = bdi_set_min_ratio(bdi, ratio); |
| 188 | if (!ret) |
| 189 | ret = count; |
| 190 | } |
| 191 | return ret; |
| 192 | } |
| 193 | BDI_SHOW(min_ratio, bdi->min_ratio) |
| 194 | |
| 195 | static ssize_t max_ratio_store(struct device *dev, |
| 196 | struct device_attribute *attr, const char *buf, size_t count) |
| 197 | { |
| 198 | struct backing_dev_info *bdi = dev_get_drvdata(dev); |
| 199 | char *end; |
| 200 | unsigned int ratio; |
| 201 | ssize_t ret = -EINVAL; |
| 202 | |
| 203 | ratio = simple_strtoul(buf, &end, 10); |
| 204 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { |
| 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 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) |
| 214 | |
| 215 | static struct device_attribute bdi_dev_attrs[] = { |
| 216 | __ATTR_RW(read_ahead_kb), |
| 217 | __ATTR_RW(min_ratio), |
| 218 | __ATTR_RW(max_ratio), |
| 219 | __ATTR_NULL, |
| 220 | }; |
| 221 | |
| 222 | static __init int bdi_class_init(void) |
| 223 | { |
| 224 | bdi_class = class_create(THIS_MODULE, "bdi"); |
| 225 | if (IS_ERR(bdi_class)) |
| 226 | return PTR_ERR(bdi_class); |
| 227 | |
| 228 | bdi_class->dev_attrs = bdi_dev_attrs; |
| 229 | bdi_debug_init(); |
| 230 | return 0; |
| 231 | } |
| 232 | postcore_initcall(bdi_class_init); |
| 233 | |
| 234 | static int __init default_bdi_init(void) |
| 235 | { |
| 236 | int err; |
| 237 | |
| 238 | sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers"); |
| 239 | BUG_ON(IS_ERR(sync_supers_tsk)); |
| 240 | |
| 241 | init_timer(&sync_supers_timer); |
| 242 | setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0); |
| 243 | bdi_arm_supers_timer(); |
| 244 | |
| 245 | err = bdi_init(&default_backing_dev_info); |
| 246 | if (!err) |
| 247 | bdi_register(&default_backing_dev_info, NULL, "default"); |
| 248 | |
| 249 | return err; |
| 250 | } |
| 251 | subsys_initcall(default_bdi_init); |
| 252 | |
| 253 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
| 254 | { |
| 255 | memset(wb, 0, sizeof(*wb)); |
| 256 | |
| 257 | wb->bdi = bdi; |
| 258 | wb->last_old_flush = jiffies; |
| 259 | INIT_LIST_HEAD(&wb->b_dirty); |
| 260 | INIT_LIST_HEAD(&wb->b_io); |
| 261 | INIT_LIST_HEAD(&wb->b_more_io); |
| 262 | } |
| 263 | |
| 264 | int bdi_has_dirty_io(struct backing_dev_info *bdi) |
| 265 | { |
| 266 | return wb_has_dirty_io(&bdi->wb); |
| 267 | } |
| 268 | |
| 269 | static void bdi_flush_io(struct backing_dev_info *bdi) |
| 270 | { |
| 271 | struct writeback_control wbc = { |
| 272 | .sync_mode = WB_SYNC_NONE, |
| 273 | .older_than_this = NULL, |
| 274 | .range_cyclic = 1, |
| 275 | .nr_to_write = 1024, |
| 276 | }; |
| 277 | |
| 278 | writeback_inodes_wb(&bdi->wb, &wbc); |
| 279 | } |
| 280 | |
| 281 | /* |
| 282 | * kupdated() used to do this. We cannot do it from the bdi_forker_thread() |
| 283 | * or we risk deadlocking on ->s_umount. The longer term solution would be |
| 284 | * to implement sync_supers_bdi() or similar and simply do it from the |
| 285 | * bdi writeback thread individually. |
| 286 | */ |
| 287 | static int bdi_sync_supers(void *unused) |
| 288 | { |
| 289 | set_user_nice(current, 0); |
| 290 | |
| 291 | while (!kthread_should_stop()) { |
| 292 | set_current_state(TASK_INTERRUPTIBLE); |
| 293 | schedule(); |
| 294 | |
| 295 | /* |
| 296 | * Do this periodically, like kupdated() did before. |
| 297 | */ |
| 298 | sync_supers(); |
| 299 | } |
| 300 | |
| 301 | return 0; |
| 302 | } |
| 303 | |
| 304 | void bdi_arm_supers_timer(void) |
| 305 | { |
| 306 | unsigned long next; |
| 307 | |
| 308 | if (!dirty_writeback_interval) |
| 309 | return; |
| 310 | |
| 311 | next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies; |
| 312 | mod_timer(&sync_supers_timer, round_jiffies_up(next)); |
| 313 | } |
| 314 | |
| 315 | static void sync_supers_timer_fn(unsigned long unused) |
| 316 | { |
| 317 | wake_up_process(sync_supers_tsk); |
| 318 | bdi_arm_supers_timer(); |
| 319 | } |
| 320 | |
| 321 | static int bdi_forker_thread(void *ptr) |
| 322 | { |
| 323 | struct bdi_writeback *me = ptr; |
| 324 | |
| 325 | current->flags |= PF_FLUSHER | PF_SWAPWRITE; |
| 326 | set_freezable(); |
| 327 | |
| 328 | /* |
| 329 | * Our parent may run at a different priority, just set us to normal |
| 330 | */ |
| 331 | set_user_nice(current, 0); |
| 332 | |
| 333 | for (;;) { |
| 334 | struct backing_dev_info *bdi, *tmp; |
| 335 | struct bdi_writeback *wb; |
| 336 | |
| 337 | /* |
| 338 | * Temporary measure, we want to make sure we don't see |
| 339 | * dirty data on the default backing_dev_info |
| 340 | */ |
| 341 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) |
| 342 | wb_do_writeback(me, 0); |
| 343 | |
| 344 | spin_lock_bh(&bdi_lock); |
| 345 | |
| 346 | /* |
| 347 | * Check if any existing bdi's have dirty data without |
| 348 | * a thread registered. If so, set that up. |
| 349 | */ |
| 350 | list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) { |
| 351 | if (bdi->wb.task) |
| 352 | continue; |
| 353 | if (list_empty(&bdi->work_list) && |
| 354 | !bdi_has_dirty_io(bdi)) |
| 355 | continue; |
| 356 | |
| 357 | bdi_add_default_flusher_thread(bdi); |
| 358 | } |
| 359 | |
| 360 | set_current_state(TASK_INTERRUPTIBLE); |
| 361 | |
| 362 | if (list_empty(&bdi_pending_list)) { |
| 363 | unsigned long wait; |
| 364 | |
| 365 | spin_unlock_bh(&bdi_lock); |
| 366 | wait = msecs_to_jiffies(dirty_writeback_interval * 10); |
| 367 | if (wait) |
| 368 | schedule_timeout(wait); |
| 369 | else |
| 370 | schedule(); |
| 371 | try_to_freeze(); |
| 372 | continue; |
| 373 | } |
| 374 | |
| 375 | __set_current_state(TASK_RUNNING); |
| 376 | |
| 377 | /* |
| 378 | * This is our real job - check for pending entries in |
| 379 | * bdi_pending_list, and create the threads that got added |
| 380 | */ |
| 381 | bdi = list_entry(bdi_pending_list.next, struct backing_dev_info, |
| 382 | bdi_list); |
| 383 | list_del_init(&bdi->bdi_list); |
| 384 | spin_unlock_bh(&bdi_lock); |
| 385 | |
| 386 | wb = &bdi->wb; |
| 387 | wb->task = kthread_run(bdi_writeback_thread, wb, "flush-%s", |
| 388 | dev_name(bdi->dev)); |
| 389 | /* |
| 390 | * If thread creation fails, then readd the bdi to |
| 391 | * the pending list and force writeout of the bdi |
| 392 | * from this forker thread. That will free some memory |
| 393 | * and we can try again. |
| 394 | */ |
| 395 | if (IS_ERR(wb->task)) { |
| 396 | wb->task = NULL; |
| 397 | |
| 398 | /* |
| 399 | * Add this 'bdi' to the back, so we get |
| 400 | * a chance to flush other bdi's to free |
| 401 | * memory. |
| 402 | */ |
| 403 | spin_lock_bh(&bdi_lock); |
| 404 | list_add_tail(&bdi->bdi_list, &bdi_pending_list); |
| 405 | spin_unlock_bh(&bdi_lock); |
| 406 | |
| 407 | bdi_flush_io(bdi); |
| 408 | } |
| 409 | } |
| 410 | |
| 411 | return 0; |
| 412 | } |
| 413 | |
| 414 | static void bdi_add_to_pending(struct rcu_head *head) |
| 415 | { |
| 416 | struct backing_dev_info *bdi; |
| 417 | |
| 418 | bdi = container_of(head, struct backing_dev_info, rcu_head); |
| 419 | INIT_LIST_HEAD(&bdi->bdi_list); |
| 420 | |
| 421 | spin_lock(&bdi_lock); |
| 422 | list_add_tail(&bdi->bdi_list, &bdi_pending_list); |
| 423 | spin_unlock(&bdi_lock); |
| 424 | |
| 425 | /* |
| 426 | * We are now on the pending list, wake up bdi_forker_task() |
| 427 | * to finish the job and add us back to the active bdi_list |
| 428 | */ |
| 429 | wake_up_process(default_backing_dev_info.wb.task); |
| 430 | } |
| 431 | |
| 432 | /* |
| 433 | * Add the default flusher thread that gets created for any bdi |
| 434 | * that has dirty data pending writeout |
| 435 | */ |
| 436 | static void bdi_add_default_flusher_thread(struct backing_dev_info *bdi) |
| 437 | { |
| 438 | if (!bdi_cap_writeback_dirty(bdi)) |
| 439 | return; |
| 440 | |
| 441 | if (WARN_ON(!test_bit(BDI_registered, &bdi->state))) { |
| 442 | printk(KERN_ERR "bdi %p/%s is not registered!\n", |
| 443 | bdi, bdi->name); |
| 444 | return; |
| 445 | } |
| 446 | |
| 447 | /* |
| 448 | * Check with the helper whether to proceed adding a thread. Will only |
| 449 | * abort if we two or more simultanous calls to |
| 450 | * bdi_add_default_flusher_thread() occured, further additions will |
| 451 | * block waiting for previous additions to finish. |
| 452 | */ |
| 453 | if (!test_and_set_bit(BDI_pending, &bdi->state)) { |
| 454 | list_del_rcu(&bdi->bdi_list); |
| 455 | |
| 456 | /* |
| 457 | * We must wait for the current RCU period to end before |
| 458 | * moving to the pending list. So schedule that operation |
| 459 | * from an RCU callback. |
| 460 | */ |
| 461 | call_rcu(&bdi->rcu_head, bdi_add_to_pending); |
| 462 | } |
| 463 | } |
| 464 | |
| 465 | /* |
| 466 | * Remove bdi from bdi_list, and ensure that it is no longer visible |
| 467 | */ |
| 468 | static void bdi_remove_from_list(struct backing_dev_info *bdi) |
| 469 | { |
| 470 | spin_lock_bh(&bdi_lock); |
| 471 | list_del_rcu(&bdi->bdi_list); |
| 472 | spin_unlock_bh(&bdi_lock); |
| 473 | |
| 474 | synchronize_rcu(); |
| 475 | } |
| 476 | |
| 477 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
| 478 | const char *fmt, ...) |
| 479 | { |
| 480 | va_list args; |
| 481 | int ret = 0; |
| 482 | struct device *dev; |
| 483 | |
| 484 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
| 485 | goto exit; |
| 486 | |
| 487 | va_start(args, fmt); |
| 488 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
| 489 | va_end(args); |
| 490 | if (IS_ERR(dev)) { |
| 491 | ret = PTR_ERR(dev); |
| 492 | goto exit; |
| 493 | } |
| 494 | |
| 495 | spin_lock_bh(&bdi_lock); |
| 496 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); |
| 497 | spin_unlock_bh(&bdi_lock); |
| 498 | |
| 499 | bdi->dev = dev; |
| 500 | |
| 501 | /* |
| 502 | * Just start the forker thread for our default backing_dev_info, |
| 503 | * and add other bdi's to the list. They will get a thread created |
| 504 | * on-demand when they need it. |
| 505 | */ |
| 506 | if (bdi_cap_flush_forker(bdi)) { |
| 507 | struct bdi_writeback *wb = &bdi->wb; |
| 508 | |
| 509 | wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s", |
| 510 | dev_name(dev)); |
| 511 | if (IS_ERR(wb->task)) { |
| 512 | wb->task = NULL; |
| 513 | ret = -ENOMEM; |
| 514 | |
| 515 | bdi_remove_from_list(bdi); |
| 516 | goto exit; |
| 517 | } |
| 518 | } |
| 519 | |
| 520 | bdi_debug_register(bdi, dev_name(dev)); |
| 521 | set_bit(BDI_registered, &bdi->state); |
| 522 | trace_writeback_bdi_register(bdi); |
| 523 | exit: |
| 524 | return ret; |
| 525 | } |
| 526 | EXPORT_SYMBOL(bdi_register); |
| 527 | |
| 528 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) |
| 529 | { |
| 530 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); |
| 531 | } |
| 532 | EXPORT_SYMBOL(bdi_register_dev); |
| 533 | |
| 534 | /* |
| 535 | * Remove bdi from the global list and shutdown any threads we have running |
| 536 | */ |
| 537 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) |
| 538 | { |
| 539 | if (!bdi_cap_writeback_dirty(bdi)) |
| 540 | return; |
| 541 | |
| 542 | /* |
| 543 | * If setup is pending, wait for that to complete first |
| 544 | */ |
| 545 | wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait, |
| 546 | TASK_UNINTERRUPTIBLE); |
| 547 | |
| 548 | /* |
| 549 | * Make sure nobody finds us on the bdi_list anymore |
| 550 | */ |
| 551 | bdi_remove_from_list(bdi); |
| 552 | |
| 553 | /* |
| 554 | * Finally, kill the kernel thread. We don't need to be RCU |
| 555 | * safe anymore, since the bdi is gone from visibility. Force |
| 556 | * unfreeze of the thread before calling kthread_stop(), otherwise |
| 557 | * it would never exet if it is currently stuck in the refrigerator. |
| 558 | */ |
| 559 | if (bdi->wb.task) { |
| 560 | thaw_process(bdi->wb.task); |
| 561 | kthread_stop(bdi->wb.task); |
| 562 | } |
| 563 | } |
| 564 | |
| 565 | /* |
| 566 | * This bdi is going away now, make sure that no super_blocks point to it |
| 567 | */ |
| 568 | static void bdi_prune_sb(struct backing_dev_info *bdi) |
| 569 | { |
| 570 | struct super_block *sb; |
| 571 | |
| 572 | spin_lock(&sb_lock); |
| 573 | list_for_each_entry(sb, &super_blocks, s_list) { |
| 574 | if (sb->s_bdi == bdi) |
| 575 | sb->s_bdi = NULL; |
| 576 | } |
| 577 | spin_unlock(&sb_lock); |
| 578 | } |
| 579 | |
| 580 | void bdi_unregister(struct backing_dev_info *bdi) |
| 581 | { |
| 582 | if (bdi->dev) { |
| 583 | trace_writeback_bdi_unregister(bdi); |
| 584 | bdi_prune_sb(bdi); |
| 585 | |
| 586 | if (!bdi_cap_flush_forker(bdi)) |
| 587 | bdi_wb_shutdown(bdi); |
| 588 | bdi_debug_unregister(bdi); |
| 589 | device_unregister(bdi->dev); |
| 590 | bdi->dev = NULL; |
| 591 | } |
| 592 | } |
| 593 | EXPORT_SYMBOL(bdi_unregister); |
| 594 | |
| 595 | int bdi_init(struct backing_dev_info *bdi) |
| 596 | { |
| 597 | int i, err; |
| 598 | |
| 599 | bdi->dev = NULL; |
| 600 | |
| 601 | bdi->min_ratio = 0; |
| 602 | bdi->max_ratio = 100; |
| 603 | bdi->max_prop_frac = PROP_FRAC_BASE; |
| 604 | spin_lock_init(&bdi->wb_lock); |
| 605 | INIT_RCU_HEAD(&bdi->rcu_head); |
| 606 | INIT_LIST_HEAD(&bdi->bdi_list); |
| 607 | INIT_LIST_HEAD(&bdi->work_list); |
| 608 | |
| 609 | bdi_wb_init(&bdi->wb, bdi); |
| 610 | |
| 611 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
| 612 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); |
| 613 | if (err) |
| 614 | goto err; |
| 615 | } |
| 616 | |
| 617 | bdi->dirty_exceeded = 0; |
| 618 | err = prop_local_init_percpu(&bdi->completions); |
| 619 | |
| 620 | if (err) { |
| 621 | err: |
| 622 | while (i--) |
| 623 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
| 624 | } |
| 625 | |
| 626 | return err; |
| 627 | } |
| 628 | EXPORT_SYMBOL(bdi_init); |
| 629 | |
| 630 | void bdi_destroy(struct backing_dev_info *bdi) |
| 631 | { |
| 632 | int i; |
| 633 | |
| 634 | /* |
| 635 | * Splice our entries to the default_backing_dev_info, if this |
| 636 | * bdi disappears |
| 637 | */ |
| 638 | if (bdi_has_dirty_io(bdi)) { |
| 639 | struct bdi_writeback *dst = &default_backing_dev_info.wb; |
| 640 | |
| 641 | spin_lock(&inode_lock); |
| 642 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); |
| 643 | list_splice(&bdi->wb.b_io, &dst->b_io); |
| 644 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); |
| 645 | spin_unlock(&inode_lock); |
| 646 | } |
| 647 | |
| 648 | bdi_unregister(bdi); |
| 649 | |
| 650 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
| 651 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
| 652 | |
| 653 | prop_local_destroy_percpu(&bdi->completions); |
| 654 | } |
| 655 | EXPORT_SYMBOL(bdi_destroy); |
| 656 | |
| 657 | /* |
| 658 | * For use from filesystems to quickly init and register a bdi associated |
| 659 | * with dirty writeback |
| 660 | */ |
| 661 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name, |
| 662 | unsigned int cap) |
| 663 | { |
| 664 | char tmp[32]; |
| 665 | int err; |
| 666 | |
| 667 | bdi->name = name; |
| 668 | bdi->capabilities = cap; |
| 669 | err = bdi_init(bdi); |
| 670 | if (err) |
| 671 | return err; |
| 672 | |
| 673 | sprintf(tmp, "%.28s%s", name, "-%d"); |
| 674 | err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq)); |
| 675 | if (err) { |
| 676 | bdi_destroy(bdi); |
| 677 | return err; |
| 678 | } |
| 679 | |
| 680 | return 0; |
| 681 | } |
| 682 | EXPORT_SYMBOL(bdi_setup_and_register); |
| 683 | |
| 684 | static wait_queue_head_t congestion_wqh[2] = { |
| 685 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), |
| 686 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) |
| 687 | }; |
| 688 | |
| 689 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
| 690 | { |
| 691 | enum bdi_state bit; |
| 692 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
| 693 | |
| 694 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
| 695 | clear_bit(bit, &bdi->state); |
| 696 | smp_mb__after_clear_bit(); |
| 697 | if (waitqueue_active(wqh)) |
| 698 | wake_up(wqh); |
| 699 | } |
| 700 | EXPORT_SYMBOL(clear_bdi_congested); |
| 701 | |
| 702 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
| 703 | { |
| 704 | enum bdi_state bit; |
| 705 | |
| 706 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
| 707 | set_bit(bit, &bdi->state); |
| 708 | } |
| 709 | EXPORT_SYMBOL(set_bdi_congested); |
| 710 | |
| 711 | /** |
| 712 | * congestion_wait - wait for a backing_dev to become uncongested |
| 713 | * @sync: SYNC or ASYNC IO |
| 714 | * @timeout: timeout in jiffies |
| 715 | * |
| 716 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit |
| 717 | * write congestion. If no backing_devs are congested then just wait for the |
| 718 | * next write to be completed. |
| 719 | */ |
| 720 | long congestion_wait(int sync, long timeout) |
| 721 | { |
| 722 | long ret; |
| 723 | DEFINE_WAIT(wait); |
| 724 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
| 725 | |
| 726 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); |
| 727 | ret = io_schedule_timeout(timeout); |
| 728 | finish_wait(wqh, &wait); |
| 729 | return ret; |
| 730 | } |
| 731 | EXPORT_SYMBOL(congestion_wait); |
| 732 | |