| 1 | /* |
| 2 | * fs/fs-writeback.c |
| 3 | * |
| 4 | * Copyright (C) 2002, Linus Torvalds. |
| 5 | * |
| 6 | * Contains all the functions related to writing back and waiting |
| 7 | * upon dirty inodes against superblocks, and writing back dirty |
| 8 | * pages against inodes. ie: data writeback. Writeout of the |
| 9 | * inode itself is not handled here. |
| 10 | * |
| 11 | * 10Apr2002 Andrew Morton |
| 12 | * Split out of fs/inode.c |
| 13 | * Additions for address_space-based writeback |
| 14 | */ |
| 15 | |
| 16 | #include <linux/kernel.h> |
| 17 | #include <linux/export.h> |
| 18 | #include <linux/spinlock.h> |
| 19 | #include <linux/slab.h> |
| 20 | #include <linux/sched.h> |
| 21 | #include <linux/fs.h> |
| 22 | #include <linux/mm.h> |
| 23 | #include <linux/pagemap.h> |
| 24 | #include <linux/kthread.h> |
| 25 | #include <linux/writeback.h> |
| 26 | #include <linux/blkdev.h> |
| 27 | #include <linux/backing-dev.h> |
| 28 | #include <linux/tracepoint.h> |
| 29 | #include <linux/device.h> |
| 30 | #include "internal.h" |
| 31 | |
| 32 | /* |
| 33 | * 4MB minimal write chunk size |
| 34 | */ |
| 35 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) |
| 36 | |
| 37 | /* |
| 38 | * Passed into wb_writeback(), essentially a subset of writeback_control |
| 39 | */ |
| 40 | struct wb_writeback_work { |
| 41 | long nr_pages; |
| 42 | struct super_block *sb; |
| 43 | /* |
| 44 | * Write only inodes dirtied before this time. Don't forget to set |
| 45 | * older_than_this_is_set when you set this. |
| 46 | */ |
| 47 | unsigned long older_than_this; |
| 48 | enum writeback_sync_modes sync_mode; |
| 49 | unsigned int tagged_writepages:1; |
| 50 | unsigned int for_kupdate:1; |
| 51 | unsigned int range_cyclic:1; |
| 52 | unsigned int for_background:1; |
| 53 | unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ |
| 54 | unsigned int older_than_this_is_set:1; |
| 55 | enum wb_reason reason; /* why was writeback initiated? */ |
| 56 | |
| 57 | struct list_head list; /* pending work list */ |
| 58 | struct completion *done; /* set if the caller waits */ |
| 59 | }; |
| 60 | |
| 61 | /** |
| 62 | * writeback_in_progress - determine whether there is writeback in progress |
| 63 | * @bdi: the device's backing_dev_info structure. |
| 64 | * |
| 65 | * Determine whether there is writeback waiting to be handled against a |
| 66 | * backing device. |
| 67 | */ |
| 68 | int writeback_in_progress(struct backing_dev_info *bdi) |
| 69 | { |
| 70 | return test_bit(BDI_writeback_running, &bdi->state); |
| 71 | } |
| 72 | EXPORT_SYMBOL(writeback_in_progress); |
| 73 | |
| 74 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
| 75 | { |
| 76 | struct super_block *sb = inode->i_sb; |
| 77 | |
| 78 | if (sb_is_blkdev_sb(sb)) |
| 79 | return inode->i_mapping->backing_dev_info; |
| 80 | |
| 81 | return sb->s_bdi; |
| 82 | } |
| 83 | |
| 84 | static inline struct inode *wb_inode(struct list_head *head) |
| 85 | { |
| 86 | return list_entry(head, struct inode, i_wb_list); |
| 87 | } |
| 88 | |
| 89 | /* |
| 90 | * Include the creation of the trace points after defining the |
| 91 | * wb_writeback_work structure and inline functions so that the definition |
| 92 | * remains local to this file. |
| 93 | */ |
| 94 | #define CREATE_TRACE_POINTS |
| 95 | #include <trace/events/writeback.h> |
| 96 | |
| 97 | static void bdi_queue_work(struct backing_dev_info *bdi, |
| 98 | struct wb_writeback_work *work) |
| 99 | { |
| 100 | trace_writeback_queue(bdi, work); |
| 101 | |
| 102 | spin_lock_bh(&bdi->wb_lock); |
| 103 | list_add_tail(&work->list, &bdi->work_list); |
| 104 | spin_unlock_bh(&bdi->wb_lock); |
| 105 | |
| 106 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
| 107 | } |
| 108 | |
| 109 | static void |
| 110 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
| 111 | bool range_cyclic, enum wb_reason reason) |
| 112 | { |
| 113 | struct wb_writeback_work *work; |
| 114 | |
| 115 | /* |
| 116 | * This is WB_SYNC_NONE writeback, so if allocation fails just |
| 117 | * wakeup the thread for old dirty data writeback |
| 118 | */ |
| 119 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
| 120 | if (!work) { |
| 121 | trace_writeback_nowork(bdi); |
| 122 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
| 123 | return; |
| 124 | } |
| 125 | |
| 126 | work->sync_mode = WB_SYNC_NONE; |
| 127 | work->nr_pages = nr_pages; |
| 128 | work->range_cyclic = range_cyclic; |
| 129 | work->reason = reason; |
| 130 | |
| 131 | bdi_queue_work(bdi, work); |
| 132 | } |
| 133 | |
| 134 | /** |
| 135 | * bdi_start_writeback - start writeback |
| 136 | * @bdi: the backing device to write from |
| 137 | * @nr_pages: the number of pages to write |
| 138 | * @reason: reason why some writeback work was initiated |
| 139 | * |
| 140 | * Description: |
| 141 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only |
| 142 | * started when this function returns, we make no guarantees on |
| 143 | * completion. Caller need not hold sb s_umount semaphore. |
| 144 | * |
| 145 | */ |
| 146 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
| 147 | enum wb_reason reason) |
| 148 | { |
| 149 | __bdi_start_writeback(bdi, nr_pages, true, reason); |
| 150 | } |
| 151 | |
| 152 | /** |
| 153 | * bdi_start_background_writeback - start background writeback |
| 154 | * @bdi: the backing device to write from |
| 155 | * |
| 156 | * Description: |
| 157 | * This makes sure WB_SYNC_NONE background writeback happens. When |
| 158 | * this function returns, it is only guaranteed that for given BDI |
| 159 | * some IO is happening if we are over background dirty threshold. |
| 160 | * Caller need not hold sb s_umount semaphore. |
| 161 | */ |
| 162 | void bdi_start_background_writeback(struct backing_dev_info *bdi) |
| 163 | { |
| 164 | /* |
| 165 | * We just wake up the flusher thread. It will perform background |
| 166 | * writeback as soon as there is no other work to do. |
| 167 | */ |
| 168 | trace_writeback_wake_background(bdi); |
| 169 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
| 170 | } |
| 171 | |
| 172 | /* |
| 173 | * Remove the inode from the writeback list it is on. |
| 174 | */ |
| 175 | void inode_wb_list_del(struct inode *inode) |
| 176 | { |
| 177 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
| 178 | |
| 179 | spin_lock(&bdi->wb.list_lock); |
| 180 | list_del_init(&inode->i_wb_list); |
| 181 | spin_unlock(&bdi->wb.list_lock); |
| 182 | } |
| 183 | |
| 184 | /* |
| 185 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the |
| 186 | * furthest end of its superblock's dirty-inode list. |
| 187 | * |
| 188 | * Before stamping the inode's ->dirtied_when, we check to see whether it is |
| 189 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
| 190 | * the case then the inode must have been redirtied while it was being written |
| 191 | * out and we don't reset its dirtied_when. |
| 192 | */ |
| 193 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
| 194 | { |
| 195 | assert_spin_locked(&wb->list_lock); |
| 196 | if (!list_empty(&wb->b_dirty)) { |
| 197 | struct inode *tail; |
| 198 | |
| 199 | tail = wb_inode(wb->b_dirty.next); |
| 200 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
| 201 | inode->dirtied_when = jiffies; |
| 202 | } |
| 203 | list_move(&inode->i_wb_list, &wb->b_dirty); |
| 204 | } |
| 205 | |
| 206 | /* |
| 207 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
| 208 | */ |
| 209 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
| 210 | { |
| 211 | assert_spin_locked(&wb->list_lock); |
| 212 | list_move(&inode->i_wb_list, &wb->b_more_io); |
| 213 | } |
| 214 | |
| 215 | static void inode_sync_complete(struct inode *inode) |
| 216 | { |
| 217 | inode->i_state &= ~I_SYNC; |
| 218 | /* If inode is clean an unused, put it into LRU now... */ |
| 219 | inode_add_lru(inode); |
| 220 | /* Waiters must see I_SYNC cleared before being woken up */ |
| 221 | smp_mb(); |
| 222 | wake_up_bit(&inode->i_state, __I_SYNC); |
| 223 | } |
| 224 | |
| 225 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
| 226 | { |
| 227 | bool ret = time_after(inode->dirtied_when, t); |
| 228 | #ifndef CONFIG_64BIT |
| 229 | /* |
| 230 | * For inodes being constantly redirtied, dirtied_when can get stuck. |
| 231 | * It _appears_ to be in the future, but is actually in distant past. |
| 232 | * This test is necessary to prevent such wrapped-around relative times |
| 233 | * from permanently stopping the whole bdi writeback. |
| 234 | */ |
| 235 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); |
| 236 | #endif |
| 237 | return ret; |
| 238 | } |
| 239 | |
| 240 | /* |
| 241 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
| 242 | * @delaying_queue to @dispatch_queue. |
| 243 | */ |
| 244 | static int move_expired_inodes(struct list_head *delaying_queue, |
| 245 | struct list_head *dispatch_queue, |
| 246 | struct wb_writeback_work *work) |
| 247 | { |
| 248 | LIST_HEAD(tmp); |
| 249 | struct list_head *pos, *node; |
| 250 | struct super_block *sb = NULL; |
| 251 | struct inode *inode; |
| 252 | int do_sb_sort = 0; |
| 253 | int moved = 0; |
| 254 | |
| 255 | WARN_ON_ONCE(!work->older_than_this_is_set); |
| 256 | while (!list_empty(delaying_queue)) { |
| 257 | inode = wb_inode(delaying_queue->prev); |
| 258 | if (inode_dirtied_after(inode, work->older_than_this)) |
| 259 | break; |
| 260 | list_move(&inode->i_wb_list, &tmp); |
| 261 | moved++; |
| 262 | if (sb_is_blkdev_sb(inode->i_sb)) |
| 263 | continue; |
| 264 | if (sb && sb != inode->i_sb) |
| 265 | do_sb_sort = 1; |
| 266 | sb = inode->i_sb; |
| 267 | } |
| 268 | |
| 269 | /* just one sb in list, splice to dispatch_queue and we're done */ |
| 270 | if (!do_sb_sort) { |
| 271 | list_splice(&tmp, dispatch_queue); |
| 272 | goto out; |
| 273 | } |
| 274 | |
| 275 | /* Move inodes from one superblock together */ |
| 276 | while (!list_empty(&tmp)) { |
| 277 | sb = wb_inode(tmp.prev)->i_sb; |
| 278 | list_for_each_prev_safe(pos, node, &tmp) { |
| 279 | inode = wb_inode(pos); |
| 280 | if (inode->i_sb == sb) |
| 281 | list_move(&inode->i_wb_list, dispatch_queue); |
| 282 | } |
| 283 | } |
| 284 | out: |
| 285 | return moved; |
| 286 | } |
| 287 | |
| 288 | /* |
| 289 | * Queue all expired dirty inodes for io, eldest first. |
| 290 | * Before |
| 291 | * newly dirtied b_dirty b_io b_more_io |
| 292 | * =============> gf edc BA |
| 293 | * After |
| 294 | * newly dirtied b_dirty b_io b_more_io |
| 295 | * =============> g fBAedc |
| 296 | * | |
| 297 | * +--> dequeue for IO |
| 298 | */ |
| 299 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
| 300 | { |
| 301 | int moved; |
| 302 | assert_spin_locked(&wb->list_lock); |
| 303 | list_splice_init(&wb->b_more_io, &wb->b_io); |
| 304 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); |
| 305 | trace_writeback_queue_io(wb, work, moved); |
| 306 | } |
| 307 | |
| 308 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
| 309 | { |
| 310 | int ret; |
| 311 | |
| 312 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { |
| 313 | trace_writeback_write_inode_start(inode, wbc); |
| 314 | ret = inode->i_sb->s_op->write_inode(inode, wbc); |
| 315 | trace_writeback_write_inode(inode, wbc); |
| 316 | return ret; |
| 317 | } |
| 318 | return 0; |
| 319 | } |
| 320 | |
| 321 | /* |
| 322 | * Wait for writeback on an inode to complete. Called with i_lock held. |
| 323 | * Caller must make sure inode cannot go away when we drop i_lock. |
| 324 | */ |
| 325 | static void __inode_wait_for_writeback(struct inode *inode) |
| 326 | __releases(inode->i_lock) |
| 327 | __acquires(inode->i_lock) |
| 328 | { |
| 329 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); |
| 330 | wait_queue_head_t *wqh; |
| 331 | |
| 332 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
| 333 | while (inode->i_state & I_SYNC) { |
| 334 | spin_unlock(&inode->i_lock); |
| 335 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
| 336 | spin_lock(&inode->i_lock); |
| 337 | } |
| 338 | } |
| 339 | |
| 340 | /* |
| 341 | * Wait for writeback on an inode to complete. Caller must have inode pinned. |
| 342 | */ |
| 343 | void inode_wait_for_writeback(struct inode *inode) |
| 344 | { |
| 345 | spin_lock(&inode->i_lock); |
| 346 | __inode_wait_for_writeback(inode); |
| 347 | spin_unlock(&inode->i_lock); |
| 348 | } |
| 349 | |
| 350 | /* |
| 351 | * Sleep until I_SYNC is cleared. This function must be called with i_lock |
| 352 | * held and drops it. It is aimed for callers not holding any inode reference |
| 353 | * so once i_lock is dropped, inode can go away. |
| 354 | */ |
| 355 | static void inode_sleep_on_writeback(struct inode *inode) |
| 356 | __releases(inode->i_lock) |
| 357 | { |
| 358 | DEFINE_WAIT(wait); |
| 359 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
| 360 | int sleep; |
| 361 | |
| 362 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); |
| 363 | sleep = inode->i_state & I_SYNC; |
| 364 | spin_unlock(&inode->i_lock); |
| 365 | if (sleep) |
| 366 | schedule(); |
| 367 | finish_wait(wqh, &wait); |
| 368 | } |
| 369 | |
| 370 | /* |
| 371 | * Find proper writeback list for the inode depending on its current state and |
| 372 | * possibly also change of its state while we were doing writeback. Here we |
| 373 | * handle things such as livelock prevention or fairness of writeback among |
| 374 | * inodes. This function can be called only by flusher thread - noone else |
| 375 | * processes all inodes in writeback lists and requeueing inodes behind flusher |
| 376 | * thread's back can have unexpected consequences. |
| 377 | */ |
| 378 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, |
| 379 | struct writeback_control *wbc) |
| 380 | { |
| 381 | if (inode->i_state & I_FREEING) |
| 382 | return; |
| 383 | |
| 384 | /* |
| 385 | * Sync livelock prevention. Each inode is tagged and synced in one |
| 386 | * shot. If still dirty, it will be redirty_tail()'ed below. Update |
| 387 | * the dirty time to prevent enqueue and sync it again. |
| 388 | */ |
| 389 | if ((inode->i_state & I_DIRTY) && |
| 390 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) |
| 391 | inode->dirtied_when = jiffies; |
| 392 | |
| 393 | if (wbc->pages_skipped) { |
| 394 | /* |
| 395 | * writeback is not making progress due to locked |
| 396 | * buffers. Skip this inode for now. |
| 397 | */ |
| 398 | redirty_tail(inode, wb); |
| 399 | return; |
| 400 | } |
| 401 | |
| 402 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
| 403 | /* |
| 404 | * We didn't write back all the pages. nfs_writepages() |
| 405 | * sometimes bales out without doing anything. |
| 406 | */ |
| 407 | if (wbc->nr_to_write <= 0) { |
| 408 | /* Slice used up. Queue for next turn. */ |
| 409 | requeue_io(inode, wb); |
| 410 | } else { |
| 411 | /* |
| 412 | * Writeback blocked by something other than |
| 413 | * congestion. Delay the inode for some time to |
| 414 | * avoid spinning on the CPU (100% iowait) |
| 415 | * retrying writeback of the dirty page/inode |
| 416 | * that cannot be performed immediately. |
| 417 | */ |
| 418 | redirty_tail(inode, wb); |
| 419 | } |
| 420 | } else if (inode->i_state & I_DIRTY) { |
| 421 | /* |
| 422 | * Filesystems can dirty the inode during writeback operations, |
| 423 | * such as delayed allocation during submission or metadata |
| 424 | * updates after data IO completion. |
| 425 | */ |
| 426 | redirty_tail(inode, wb); |
| 427 | } else { |
| 428 | /* The inode is clean. Remove from writeback lists. */ |
| 429 | list_del_init(&inode->i_wb_list); |
| 430 | } |
| 431 | } |
| 432 | |
| 433 | /* |
| 434 | * Write out an inode and its dirty pages. Do not update the writeback list |
| 435 | * linkage. That is left to the caller. The caller is also responsible for |
| 436 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. |
| 437 | */ |
| 438 | static int |
| 439 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
| 440 | { |
| 441 | struct address_space *mapping = inode->i_mapping; |
| 442 | long nr_to_write = wbc->nr_to_write; |
| 443 | unsigned dirty; |
| 444 | int ret; |
| 445 | |
| 446 | WARN_ON(!(inode->i_state & I_SYNC)); |
| 447 | |
| 448 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
| 449 | |
| 450 | ret = do_writepages(mapping, wbc); |
| 451 | |
| 452 | /* |
| 453 | * Make sure to wait on the data before writing out the metadata. |
| 454 | * This is important for filesystems that modify metadata on data |
| 455 | * I/O completion. We don't do it for sync(2) writeback because it has a |
| 456 | * separate, external IO completion path and ->sync_fs for guaranteeing |
| 457 | * inode metadata is written back correctly. |
| 458 | */ |
| 459 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
| 460 | int err = filemap_fdatawait(mapping); |
| 461 | if (ret == 0) |
| 462 | ret = err; |
| 463 | } |
| 464 | |
| 465 | /* |
| 466 | * Some filesystems may redirty the inode during the writeback |
| 467 | * due to delalloc, clear dirty metadata flags right before |
| 468 | * write_inode() |
| 469 | */ |
| 470 | spin_lock(&inode->i_lock); |
| 471 | /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ |
| 472 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
| 473 | inode->i_state &= ~I_DIRTY_PAGES; |
| 474 | dirty = inode->i_state & I_DIRTY; |
| 475 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); |
| 476 | spin_unlock(&inode->i_lock); |
| 477 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
| 478 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
| 479 | int err = write_inode(inode, wbc); |
| 480 | if (ret == 0) |
| 481 | ret = err; |
| 482 | } |
| 483 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
| 484 | return ret; |
| 485 | } |
| 486 | |
| 487 | /* |
| 488 | * Write out an inode's dirty pages. Either the caller has an active reference |
| 489 | * on the inode or the inode has I_WILL_FREE set. |
| 490 | * |
| 491 | * This function is designed to be called for writing back one inode which |
| 492 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() |
| 493 | * and does more profound writeback list handling in writeback_sb_inodes(). |
| 494 | */ |
| 495 | static int |
| 496 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, |
| 497 | struct writeback_control *wbc) |
| 498 | { |
| 499 | int ret = 0; |
| 500 | |
| 501 | spin_lock(&inode->i_lock); |
| 502 | if (!atomic_read(&inode->i_count)) |
| 503 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); |
| 504 | else |
| 505 | WARN_ON(inode->i_state & I_WILL_FREE); |
| 506 | |
| 507 | if (inode->i_state & I_SYNC) { |
| 508 | if (wbc->sync_mode != WB_SYNC_ALL) |
| 509 | goto out; |
| 510 | /* |
| 511 | * It's a data-integrity sync. We must wait. Since callers hold |
| 512 | * inode reference or inode has I_WILL_FREE set, it cannot go |
| 513 | * away under us. |
| 514 | */ |
| 515 | __inode_wait_for_writeback(inode); |
| 516 | } |
| 517 | WARN_ON(inode->i_state & I_SYNC); |
| 518 | /* |
| 519 | * Skip inode if it is clean and we have no outstanding writeback in |
| 520 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this |
| 521 | * function since flusher thread may be doing for example sync in |
| 522 | * parallel and if we move the inode, it could get skipped. So here we |
| 523 | * make sure inode is on some writeback list and leave it there unless |
| 524 | * we have completely cleaned the inode. |
| 525 | */ |
| 526 | if (!(inode->i_state & I_DIRTY) && |
| 527 | (wbc->sync_mode != WB_SYNC_ALL || |
| 528 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) |
| 529 | goto out; |
| 530 | inode->i_state |= I_SYNC; |
| 531 | spin_unlock(&inode->i_lock); |
| 532 | |
| 533 | ret = __writeback_single_inode(inode, wbc); |
| 534 | |
| 535 | spin_lock(&wb->list_lock); |
| 536 | spin_lock(&inode->i_lock); |
| 537 | /* |
| 538 | * If inode is clean, remove it from writeback lists. Otherwise don't |
| 539 | * touch it. See comment above for explanation. |
| 540 | */ |
| 541 | if (!(inode->i_state & I_DIRTY)) |
| 542 | list_del_init(&inode->i_wb_list); |
| 543 | spin_unlock(&wb->list_lock); |
| 544 | inode_sync_complete(inode); |
| 545 | out: |
| 546 | spin_unlock(&inode->i_lock); |
| 547 | return ret; |
| 548 | } |
| 549 | |
| 550 | static long writeback_chunk_size(struct backing_dev_info *bdi, |
| 551 | struct wb_writeback_work *work) |
| 552 | { |
| 553 | long pages; |
| 554 | |
| 555 | /* |
| 556 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty |
| 557 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX |
| 558 | * here avoids calling into writeback_inodes_wb() more than once. |
| 559 | * |
| 560 | * The intended call sequence for WB_SYNC_ALL writeback is: |
| 561 | * |
| 562 | * wb_writeback() |
| 563 | * writeback_sb_inodes() <== called only once |
| 564 | * write_cache_pages() <== called once for each inode |
| 565 | * (quickly) tag currently dirty pages |
| 566 | * (maybe slowly) sync all tagged pages |
| 567 | */ |
| 568 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) |
| 569 | pages = LONG_MAX; |
| 570 | else { |
| 571 | pages = min(bdi->avg_write_bandwidth / 2, |
| 572 | global_dirty_limit / DIRTY_SCOPE); |
| 573 | pages = min(pages, work->nr_pages); |
| 574 | pages = round_down(pages + MIN_WRITEBACK_PAGES, |
| 575 | MIN_WRITEBACK_PAGES); |
| 576 | } |
| 577 | |
| 578 | return pages; |
| 579 | } |
| 580 | |
| 581 | /* |
| 582 | * Write a portion of b_io inodes which belong to @sb. |
| 583 | * |
| 584 | * Return the number of pages and/or inodes written. |
| 585 | */ |
| 586 | static long writeback_sb_inodes(struct super_block *sb, |
| 587 | struct bdi_writeback *wb, |
| 588 | struct wb_writeback_work *work) |
| 589 | { |
| 590 | struct writeback_control wbc = { |
| 591 | .sync_mode = work->sync_mode, |
| 592 | .tagged_writepages = work->tagged_writepages, |
| 593 | .for_kupdate = work->for_kupdate, |
| 594 | .for_background = work->for_background, |
| 595 | .for_sync = work->for_sync, |
| 596 | .range_cyclic = work->range_cyclic, |
| 597 | .range_start = 0, |
| 598 | .range_end = LLONG_MAX, |
| 599 | }; |
| 600 | unsigned long start_time = jiffies; |
| 601 | long write_chunk; |
| 602 | long wrote = 0; /* count both pages and inodes */ |
| 603 | |
| 604 | while (!list_empty(&wb->b_io)) { |
| 605 | struct inode *inode = wb_inode(wb->b_io.prev); |
| 606 | |
| 607 | if (inode->i_sb != sb) { |
| 608 | if (work->sb) { |
| 609 | /* |
| 610 | * We only want to write back data for this |
| 611 | * superblock, move all inodes not belonging |
| 612 | * to it back onto the dirty list. |
| 613 | */ |
| 614 | redirty_tail(inode, wb); |
| 615 | continue; |
| 616 | } |
| 617 | |
| 618 | /* |
| 619 | * The inode belongs to a different superblock. |
| 620 | * Bounce back to the caller to unpin this and |
| 621 | * pin the next superblock. |
| 622 | */ |
| 623 | break; |
| 624 | } |
| 625 | |
| 626 | /* |
| 627 | * Don't bother with new inodes or inodes being freed, first |
| 628 | * kind does not need periodic writeout yet, and for the latter |
| 629 | * kind writeout is handled by the freer. |
| 630 | */ |
| 631 | spin_lock(&inode->i_lock); |
| 632 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
| 633 | spin_unlock(&inode->i_lock); |
| 634 | redirty_tail(inode, wb); |
| 635 | continue; |
| 636 | } |
| 637 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
| 638 | /* |
| 639 | * If this inode is locked for writeback and we are not |
| 640 | * doing writeback-for-data-integrity, move it to |
| 641 | * b_more_io so that writeback can proceed with the |
| 642 | * other inodes on s_io. |
| 643 | * |
| 644 | * We'll have another go at writing back this inode |
| 645 | * when we completed a full scan of b_io. |
| 646 | */ |
| 647 | spin_unlock(&inode->i_lock); |
| 648 | requeue_io(inode, wb); |
| 649 | trace_writeback_sb_inodes_requeue(inode); |
| 650 | continue; |
| 651 | } |
| 652 | spin_unlock(&wb->list_lock); |
| 653 | |
| 654 | /* |
| 655 | * We already requeued the inode if it had I_SYNC set and we |
| 656 | * are doing WB_SYNC_NONE writeback. So this catches only the |
| 657 | * WB_SYNC_ALL case. |
| 658 | */ |
| 659 | if (inode->i_state & I_SYNC) { |
| 660 | /* Wait for I_SYNC. This function drops i_lock... */ |
| 661 | inode_sleep_on_writeback(inode); |
| 662 | /* Inode may be gone, start again */ |
| 663 | spin_lock(&wb->list_lock); |
| 664 | continue; |
| 665 | } |
| 666 | inode->i_state |= I_SYNC; |
| 667 | spin_unlock(&inode->i_lock); |
| 668 | |
| 669 | write_chunk = writeback_chunk_size(wb->bdi, work); |
| 670 | wbc.nr_to_write = write_chunk; |
| 671 | wbc.pages_skipped = 0; |
| 672 | |
| 673 | /* |
| 674 | * We use I_SYNC to pin the inode in memory. While it is set |
| 675 | * evict_inode() will wait so the inode cannot be freed. |
| 676 | */ |
| 677 | __writeback_single_inode(inode, &wbc); |
| 678 | |
| 679 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
| 680 | wrote += write_chunk - wbc.nr_to_write; |
| 681 | spin_lock(&wb->list_lock); |
| 682 | spin_lock(&inode->i_lock); |
| 683 | if (!(inode->i_state & I_DIRTY)) |
| 684 | wrote++; |
| 685 | requeue_inode(inode, wb, &wbc); |
| 686 | inode_sync_complete(inode); |
| 687 | spin_unlock(&inode->i_lock); |
| 688 | cond_resched_lock(&wb->list_lock); |
| 689 | /* |
| 690 | * bail out to wb_writeback() often enough to check |
| 691 | * background threshold and other termination conditions. |
| 692 | */ |
| 693 | if (wrote) { |
| 694 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
| 695 | break; |
| 696 | if (work->nr_pages <= 0) |
| 697 | break; |
| 698 | } |
| 699 | } |
| 700 | return wrote; |
| 701 | } |
| 702 | |
| 703 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
| 704 | struct wb_writeback_work *work) |
| 705 | { |
| 706 | unsigned long start_time = jiffies; |
| 707 | long wrote = 0; |
| 708 | |
| 709 | while (!list_empty(&wb->b_io)) { |
| 710 | struct inode *inode = wb_inode(wb->b_io.prev); |
| 711 | struct super_block *sb = inode->i_sb; |
| 712 | |
| 713 | if (!grab_super_passive(sb)) { |
| 714 | /* |
| 715 | * grab_super_passive() may fail consistently due to |
| 716 | * s_umount being grabbed by someone else. Don't use |
| 717 | * requeue_io() to avoid busy retrying the inode/sb. |
| 718 | */ |
| 719 | redirty_tail(inode, wb); |
| 720 | continue; |
| 721 | } |
| 722 | wrote += writeback_sb_inodes(sb, wb, work); |
| 723 | drop_super(sb); |
| 724 | |
| 725 | /* refer to the same tests at the end of writeback_sb_inodes */ |
| 726 | if (wrote) { |
| 727 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
| 728 | break; |
| 729 | if (work->nr_pages <= 0) |
| 730 | break; |
| 731 | } |
| 732 | } |
| 733 | /* Leave any unwritten inodes on b_io */ |
| 734 | return wrote; |
| 735 | } |
| 736 | |
| 737 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
| 738 | enum wb_reason reason) |
| 739 | { |
| 740 | struct wb_writeback_work work = { |
| 741 | .nr_pages = nr_pages, |
| 742 | .sync_mode = WB_SYNC_NONE, |
| 743 | .range_cyclic = 1, |
| 744 | .reason = reason, |
| 745 | .older_than_this = jiffies, |
| 746 | .older_than_this_is_set = 1, |
| 747 | }; |
| 748 | |
| 749 | spin_lock(&wb->list_lock); |
| 750 | if (list_empty(&wb->b_io)) |
| 751 | queue_io(wb, &work); |
| 752 | __writeback_inodes_wb(wb, &work); |
| 753 | spin_unlock(&wb->list_lock); |
| 754 | |
| 755 | return nr_pages - work.nr_pages; |
| 756 | } |
| 757 | |
| 758 | static bool over_bground_thresh(struct backing_dev_info *bdi) |
| 759 | { |
| 760 | unsigned long background_thresh, dirty_thresh; |
| 761 | |
| 762 | global_dirty_limits(&background_thresh, &dirty_thresh); |
| 763 | |
| 764 | if (global_page_state(NR_FILE_DIRTY) + |
| 765 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) |
| 766 | return true; |
| 767 | |
| 768 | if (bdi_stat(bdi, BDI_RECLAIMABLE) > |
| 769 | bdi_dirty_limit(bdi, background_thresh)) |
| 770 | return true; |
| 771 | |
| 772 | return false; |
| 773 | } |
| 774 | |
| 775 | /* |
| 776 | * Called under wb->list_lock. If there are multiple wb per bdi, |
| 777 | * only the flusher working on the first wb should do it. |
| 778 | */ |
| 779 | static void wb_update_bandwidth(struct bdi_writeback *wb, |
| 780 | unsigned long start_time) |
| 781 | { |
| 782 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); |
| 783 | } |
| 784 | |
| 785 | /* |
| 786 | * Explicit flushing or periodic writeback of "old" data. |
| 787 | * |
| 788 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
| 789 | * dirtying-time in the inode's address_space. So this periodic writeback code |
| 790 | * just walks the superblock inode list, writing back any inodes which are |
| 791 | * older than a specific point in time. |
| 792 | * |
| 793 | * Try to run once per dirty_writeback_interval. But if a writeback event |
| 794 | * takes longer than a dirty_writeback_interval interval, then leave a |
| 795 | * one-second gap. |
| 796 | * |
| 797 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
| 798 | * all dirty pages if they are all attached to "old" mappings. |
| 799 | */ |
| 800 | static long wb_writeback(struct bdi_writeback *wb, |
| 801 | struct wb_writeback_work *work) |
| 802 | { |
| 803 | unsigned long wb_start = jiffies; |
| 804 | long nr_pages = work->nr_pages; |
| 805 | struct inode *inode; |
| 806 | long progress; |
| 807 | |
| 808 | if (!work->older_than_this_is_set) { |
| 809 | work->older_than_this = jiffies; |
| 810 | work->older_than_this_is_set = 1; |
| 811 | } |
| 812 | |
| 813 | spin_lock(&wb->list_lock); |
| 814 | for (;;) { |
| 815 | /* |
| 816 | * Stop writeback when nr_pages has been consumed |
| 817 | */ |
| 818 | if (work->nr_pages <= 0) |
| 819 | break; |
| 820 | |
| 821 | /* |
| 822 | * Background writeout and kupdate-style writeback may |
| 823 | * run forever. Stop them if there is other work to do |
| 824 | * so that e.g. sync can proceed. They'll be restarted |
| 825 | * after the other works are all done. |
| 826 | */ |
| 827 | if ((work->for_background || work->for_kupdate) && |
| 828 | !list_empty(&wb->bdi->work_list)) |
| 829 | break; |
| 830 | |
| 831 | /* |
| 832 | * For background writeout, stop when we are below the |
| 833 | * background dirty threshold |
| 834 | */ |
| 835 | if (work->for_background && !over_bground_thresh(wb->bdi)) |
| 836 | break; |
| 837 | |
| 838 | /* |
| 839 | * Kupdate and background works are special and we want to |
| 840 | * include all inodes that need writing. Livelock avoidance is |
| 841 | * handled by these works yielding to any other work so we are |
| 842 | * safe. |
| 843 | */ |
| 844 | if (work->for_kupdate) { |
| 845 | work->older_than_this = jiffies - |
| 846 | msecs_to_jiffies(dirty_expire_interval * 10); |
| 847 | } else if (work->for_background) |
| 848 | work->older_than_this = jiffies; |
| 849 | |
| 850 | trace_writeback_start(wb->bdi, work); |
| 851 | if (list_empty(&wb->b_io)) |
| 852 | queue_io(wb, work); |
| 853 | if (work->sb) |
| 854 | progress = writeback_sb_inodes(work->sb, wb, work); |
| 855 | else |
| 856 | progress = __writeback_inodes_wb(wb, work); |
| 857 | trace_writeback_written(wb->bdi, work); |
| 858 | |
| 859 | wb_update_bandwidth(wb, wb_start); |
| 860 | |
| 861 | /* |
| 862 | * Did we write something? Try for more |
| 863 | * |
| 864 | * Dirty inodes are moved to b_io for writeback in batches. |
| 865 | * The completion of the current batch does not necessarily |
| 866 | * mean the overall work is done. So we keep looping as long |
| 867 | * as made some progress on cleaning pages or inodes. |
| 868 | */ |
| 869 | if (progress) |
| 870 | continue; |
| 871 | /* |
| 872 | * No more inodes for IO, bail |
| 873 | */ |
| 874 | if (list_empty(&wb->b_more_io)) |
| 875 | break; |
| 876 | /* |
| 877 | * Nothing written. Wait for some inode to |
| 878 | * become available for writeback. Otherwise |
| 879 | * we'll just busyloop. |
| 880 | */ |
| 881 | if (!list_empty(&wb->b_more_io)) { |
| 882 | trace_writeback_wait(wb->bdi, work); |
| 883 | inode = wb_inode(wb->b_more_io.prev); |
| 884 | spin_lock(&inode->i_lock); |
| 885 | spin_unlock(&wb->list_lock); |
| 886 | /* This function drops i_lock... */ |
| 887 | inode_sleep_on_writeback(inode); |
| 888 | spin_lock(&wb->list_lock); |
| 889 | } |
| 890 | } |
| 891 | spin_unlock(&wb->list_lock); |
| 892 | |
| 893 | return nr_pages - work->nr_pages; |
| 894 | } |
| 895 | |
| 896 | /* |
| 897 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
| 898 | */ |
| 899 | static struct wb_writeback_work * |
| 900 | get_next_work_item(struct backing_dev_info *bdi) |
| 901 | { |
| 902 | struct wb_writeback_work *work = NULL; |
| 903 | |
| 904 | spin_lock_bh(&bdi->wb_lock); |
| 905 | if (!list_empty(&bdi->work_list)) { |
| 906 | work = list_entry(bdi->work_list.next, |
| 907 | struct wb_writeback_work, list); |
| 908 | list_del_init(&work->list); |
| 909 | } |
| 910 | spin_unlock_bh(&bdi->wb_lock); |
| 911 | return work; |
| 912 | } |
| 913 | |
| 914 | /* |
| 915 | * Add in the number of potentially dirty inodes, because each inode |
| 916 | * write can dirty pagecache in the underlying blockdev. |
| 917 | */ |
| 918 | static unsigned long get_nr_dirty_pages(void) |
| 919 | { |
| 920 | return global_page_state(NR_FILE_DIRTY) + |
| 921 | global_page_state(NR_UNSTABLE_NFS) + |
| 922 | get_nr_dirty_inodes(); |
| 923 | } |
| 924 | |
| 925 | static long wb_check_background_flush(struct bdi_writeback *wb) |
| 926 | { |
| 927 | if (over_bground_thresh(wb->bdi)) { |
| 928 | |
| 929 | struct wb_writeback_work work = { |
| 930 | .nr_pages = LONG_MAX, |
| 931 | .sync_mode = WB_SYNC_NONE, |
| 932 | .for_background = 1, |
| 933 | .range_cyclic = 1, |
| 934 | .reason = WB_REASON_BACKGROUND, |
| 935 | }; |
| 936 | |
| 937 | return wb_writeback(wb, &work); |
| 938 | } |
| 939 | |
| 940 | return 0; |
| 941 | } |
| 942 | |
| 943 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
| 944 | { |
| 945 | unsigned long expired; |
| 946 | long nr_pages; |
| 947 | |
| 948 | /* |
| 949 | * When set to zero, disable periodic writeback |
| 950 | */ |
| 951 | if (!dirty_writeback_interval) |
| 952 | return 0; |
| 953 | |
| 954 | expired = wb->last_old_flush + |
| 955 | msecs_to_jiffies(dirty_writeback_interval * 10); |
| 956 | if (time_before(jiffies, expired)) |
| 957 | return 0; |
| 958 | |
| 959 | wb->last_old_flush = jiffies; |
| 960 | nr_pages = get_nr_dirty_pages(); |
| 961 | |
| 962 | if (nr_pages) { |
| 963 | struct wb_writeback_work work = { |
| 964 | .nr_pages = nr_pages, |
| 965 | .sync_mode = WB_SYNC_NONE, |
| 966 | .for_kupdate = 1, |
| 967 | .range_cyclic = 1, |
| 968 | .reason = WB_REASON_PERIODIC, |
| 969 | }; |
| 970 | |
| 971 | return wb_writeback(wb, &work); |
| 972 | } |
| 973 | |
| 974 | return 0; |
| 975 | } |
| 976 | |
| 977 | /* |
| 978 | * Retrieve work items and do the writeback they describe |
| 979 | */ |
| 980 | static long wb_do_writeback(struct bdi_writeback *wb) |
| 981 | { |
| 982 | struct backing_dev_info *bdi = wb->bdi; |
| 983 | struct wb_writeback_work *work; |
| 984 | long wrote = 0; |
| 985 | |
| 986 | set_bit(BDI_writeback_running, &wb->bdi->state); |
| 987 | while ((work = get_next_work_item(bdi)) != NULL) { |
| 988 | |
| 989 | trace_writeback_exec(bdi, work); |
| 990 | |
| 991 | wrote += wb_writeback(wb, work); |
| 992 | |
| 993 | /* |
| 994 | * Notify the caller of completion if this is a synchronous |
| 995 | * work item, otherwise just free it. |
| 996 | */ |
| 997 | if (work->done) |
| 998 | complete(work->done); |
| 999 | else |
| 1000 | kfree(work); |
| 1001 | } |
| 1002 | |
| 1003 | /* |
| 1004 | * Check for periodic writeback, kupdated() style |
| 1005 | */ |
| 1006 | wrote += wb_check_old_data_flush(wb); |
| 1007 | wrote += wb_check_background_flush(wb); |
| 1008 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
| 1009 | |
| 1010 | return wrote; |
| 1011 | } |
| 1012 | |
| 1013 | /* |
| 1014 | * Handle writeback of dirty data for the device backed by this bdi. Also |
| 1015 | * reschedules periodically and does kupdated style flushing. |
| 1016 | */ |
| 1017 | void bdi_writeback_workfn(struct work_struct *work) |
| 1018 | { |
| 1019 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
| 1020 | struct bdi_writeback, dwork); |
| 1021 | struct backing_dev_info *bdi = wb->bdi; |
| 1022 | long pages_written; |
| 1023 | |
| 1024 | set_worker_desc("flush-%s", dev_name(bdi->dev)); |
| 1025 | current->flags |= PF_SWAPWRITE; |
| 1026 | |
| 1027 | if (likely(!current_is_workqueue_rescuer() || |
| 1028 | list_empty(&bdi->bdi_list))) { |
| 1029 | /* |
| 1030 | * The normal path. Keep writing back @bdi until its |
| 1031 | * work_list is empty. Note that this path is also taken |
| 1032 | * if @bdi is shutting down even when we're running off the |
| 1033 | * rescuer as work_list needs to be drained. |
| 1034 | */ |
| 1035 | do { |
| 1036 | pages_written = wb_do_writeback(wb); |
| 1037 | trace_writeback_pages_written(pages_written); |
| 1038 | } while (!list_empty(&bdi->work_list)); |
| 1039 | } else { |
| 1040 | /* |
| 1041 | * bdi_wq can't get enough workers and we're running off |
| 1042 | * the emergency worker. Don't hog it. Hopefully, 1024 is |
| 1043 | * enough for efficient IO. |
| 1044 | */ |
| 1045 | pages_written = writeback_inodes_wb(&bdi->wb, 1024, |
| 1046 | WB_REASON_FORKER_THREAD); |
| 1047 | trace_writeback_pages_written(pages_written); |
| 1048 | } |
| 1049 | |
| 1050 | if (!list_empty(&bdi->work_list) || |
| 1051 | (wb_has_dirty_io(wb) && dirty_writeback_interval)) |
| 1052 | queue_delayed_work(bdi_wq, &wb->dwork, |
| 1053 | msecs_to_jiffies(dirty_writeback_interval * 10)); |
| 1054 | |
| 1055 | current->flags &= ~PF_SWAPWRITE; |
| 1056 | } |
| 1057 | |
| 1058 | /* |
| 1059 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
| 1060 | * the whole world. |
| 1061 | */ |
| 1062 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
| 1063 | { |
| 1064 | struct backing_dev_info *bdi; |
| 1065 | |
| 1066 | if (!nr_pages) |
| 1067 | nr_pages = get_nr_dirty_pages(); |
| 1068 | |
| 1069 | rcu_read_lock(); |
| 1070 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
| 1071 | if (!bdi_has_dirty_io(bdi)) |
| 1072 | continue; |
| 1073 | __bdi_start_writeback(bdi, nr_pages, false, reason); |
| 1074 | } |
| 1075 | rcu_read_unlock(); |
| 1076 | } |
| 1077 | |
| 1078 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
| 1079 | { |
| 1080 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { |
| 1081 | struct dentry *dentry; |
| 1082 | const char *name = "?"; |
| 1083 | |
| 1084 | dentry = d_find_alias(inode); |
| 1085 | if (dentry) { |
| 1086 | spin_lock(&dentry->d_lock); |
| 1087 | name = (const char *) dentry->d_name.name; |
| 1088 | } |
| 1089 | printk(KERN_DEBUG |
| 1090 | "%s(%d): dirtied inode %lu (%s) on %s\n", |
| 1091 | current->comm, task_pid_nr(current), inode->i_ino, |
| 1092 | name, inode->i_sb->s_id); |
| 1093 | if (dentry) { |
| 1094 | spin_unlock(&dentry->d_lock); |
| 1095 | dput(dentry); |
| 1096 | } |
| 1097 | } |
| 1098 | } |
| 1099 | |
| 1100 | /** |
| 1101 | * __mark_inode_dirty - internal function |
| 1102 | * @inode: inode to mark |
| 1103 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) |
| 1104 | * Mark an inode as dirty. Callers should use mark_inode_dirty or |
| 1105 | * mark_inode_dirty_sync. |
| 1106 | * |
| 1107 | * Put the inode on the super block's dirty list. |
| 1108 | * |
| 1109 | * CAREFUL! We mark it dirty unconditionally, but move it onto the |
| 1110 | * dirty list only if it is hashed or if it refers to a blockdev. |
| 1111 | * If it was not hashed, it will never be added to the dirty list |
| 1112 | * even if it is later hashed, as it will have been marked dirty already. |
| 1113 | * |
| 1114 | * In short, make sure you hash any inodes _before_ you start marking |
| 1115 | * them dirty. |
| 1116 | * |
| 1117 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
| 1118 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of |
| 1119 | * the kernel-internal blockdev inode represents the dirtying time of the |
| 1120 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use |
| 1121 | * page->mapping->host, so the page-dirtying time is recorded in the internal |
| 1122 | * blockdev inode. |
| 1123 | */ |
| 1124 | void __mark_inode_dirty(struct inode *inode, int flags) |
| 1125 | { |
| 1126 | struct super_block *sb = inode->i_sb; |
| 1127 | struct backing_dev_info *bdi = NULL; |
| 1128 | |
| 1129 | /* |
| 1130 | * Don't do this for I_DIRTY_PAGES - that doesn't actually |
| 1131 | * dirty the inode itself |
| 1132 | */ |
| 1133 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
| 1134 | trace_writeback_dirty_inode_start(inode, flags); |
| 1135 | |
| 1136 | if (sb->s_op->dirty_inode) |
| 1137 | sb->s_op->dirty_inode(inode, flags); |
| 1138 | |
| 1139 | trace_writeback_dirty_inode(inode, flags); |
| 1140 | } |
| 1141 | |
| 1142 | /* |
| 1143 | * make sure that changes are seen by all cpus before we test i_state |
| 1144 | * -- mikulas |
| 1145 | */ |
| 1146 | smp_mb(); |
| 1147 | |
| 1148 | /* avoid the locking if we can */ |
| 1149 | if ((inode->i_state & flags) == flags) |
| 1150 | return; |
| 1151 | |
| 1152 | if (unlikely(block_dump)) |
| 1153 | block_dump___mark_inode_dirty(inode); |
| 1154 | |
| 1155 | spin_lock(&inode->i_lock); |
| 1156 | if ((inode->i_state & flags) != flags) { |
| 1157 | const int was_dirty = inode->i_state & I_DIRTY; |
| 1158 | |
| 1159 | inode->i_state |= flags; |
| 1160 | |
| 1161 | /* |
| 1162 | * If the inode is being synced, just update its dirty state. |
| 1163 | * The unlocker will place the inode on the appropriate |
| 1164 | * superblock list, based upon its state. |
| 1165 | */ |
| 1166 | if (inode->i_state & I_SYNC) |
| 1167 | goto out_unlock_inode; |
| 1168 | |
| 1169 | /* |
| 1170 | * Only add valid (hashed) inodes to the superblock's |
| 1171 | * dirty list. Add blockdev inodes as well. |
| 1172 | */ |
| 1173 | if (!S_ISBLK(inode->i_mode)) { |
| 1174 | if (inode_unhashed(inode)) |
| 1175 | goto out_unlock_inode; |
| 1176 | } |
| 1177 | if (inode->i_state & I_FREEING) |
| 1178 | goto out_unlock_inode; |
| 1179 | |
| 1180 | /* |
| 1181 | * If the inode was already on b_dirty/b_io/b_more_io, don't |
| 1182 | * reposition it (that would break b_dirty time-ordering). |
| 1183 | */ |
| 1184 | if (!was_dirty) { |
| 1185 | bool wakeup_bdi = false; |
| 1186 | bdi = inode_to_bdi(inode); |
| 1187 | |
| 1188 | spin_unlock(&inode->i_lock); |
| 1189 | spin_lock(&bdi->wb.list_lock); |
| 1190 | if (bdi_cap_writeback_dirty(bdi)) { |
| 1191 | WARN(!test_bit(BDI_registered, &bdi->state), |
| 1192 | "bdi-%s not registered\n", bdi->name); |
| 1193 | |
| 1194 | /* |
| 1195 | * If this is the first dirty inode for this |
| 1196 | * bdi, we have to wake-up the corresponding |
| 1197 | * bdi thread to make sure background |
| 1198 | * write-back happens later. |
| 1199 | */ |
| 1200 | if (!wb_has_dirty_io(&bdi->wb)) |
| 1201 | wakeup_bdi = true; |
| 1202 | } |
| 1203 | |
| 1204 | inode->dirtied_when = jiffies; |
| 1205 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
| 1206 | spin_unlock(&bdi->wb.list_lock); |
| 1207 | |
| 1208 | if (wakeup_bdi) |
| 1209 | bdi_wakeup_thread_delayed(bdi); |
| 1210 | return; |
| 1211 | } |
| 1212 | } |
| 1213 | out_unlock_inode: |
| 1214 | spin_unlock(&inode->i_lock); |
| 1215 | |
| 1216 | } |
| 1217 | EXPORT_SYMBOL(__mark_inode_dirty); |
| 1218 | |
| 1219 | static void wait_sb_inodes(struct super_block *sb) |
| 1220 | { |
| 1221 | struct inode *inode, *old_inode = NULL; |
| 1222 | |
| 1223 | /* |
| 1224 | * We need to be protected against the filesystem going from |
| 1225 | * r/o to r/w or vice versa. |
| 1226 | */ |
| 1227 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 1228 | |
| 1229 | spin_lock(&inode_sb_list_lock); |
| 1230 | |
| 1231 | /* |
| 1232 | * Data integrity sync. Must wait for all pages under writeback, |
| 1233 | * because there may have been pages dirtied before our sync |
| 1234 | * call, but which had writeout started before we write it out. |
| 1235 | * In which case, the inode may not be on the dirty list, but |
| 1236 | * we still have to wait for that writeout. |
| 1237 | */ |
| 1238 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
| 1239 | struct address_space *mapping = inode->i_mapping; |
| 1240 | |
| 1241 | spin_lock(&inode->i_lock); |
| 1242 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || |
| 1243 | (mapping->nrpages == 0)) { |
| 1244 | spin_unlock(&inode->i_lock); |
| 1245 | continue; |
| 1246 | } |
| 1247 | __iget(inode); |
| 1248 | spin_unlock(&inode->i_lock); |
| 1249 | spin_unlock(&inode_sb_list_lock); |
| 1250 | |
| 1251 | /* |
| 1252 | * We hold a reference to 'inode' so it couldn't have been |
| 1253 | * removed from s_inodes list while we dropped the |
| 1254 | * inode_sb_list_lock. We cannot iput the inode now as we can |
| 1255 | * be holding the last reference and we cannot iput it under |
| 1256 | * inode_sb_list_lock. So we keep the reference and iput it |
| 1257 | * later. |
| 1258 | */ |
| 1259 | iput(old_inode); |
| 1260 | old_inode = inode; |
| 1261 | |
| 1262 | filemap_fdatawait(mapping); |
| 1263 | |
| 1264 | cond_resched(); |
| 1265 | |
| 1266 | spin_lock(&inode_sb_list_lock); |
| 1267 | } |
| 1268 | spin_unlock(&inode_sb_list_lock); |
| 1269 | iput(old_inode); |
| 1270 | } |
| 1271 | |
| 1272 | /** |
| 1273 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
| 1274 | * @sb: the superblock |
| 1275 | * @nr: the number of pages to write |
| 1276 | * @reason: reason why some writeback work initiated |
| 1277 | * |
| 1278 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 1279 | * on how many (if any) will be written, and this function does not wait |
| 1280 | * for IO completion of submitted IO. |
| 1281 | */ |
| 1282 | void writeback_inodes_sb_nr(struct super_block *sb, |
| 1283 | unsigned long nr, |
| 1284 | enum wb_reason reason) |
| 1285 | { |
| 1286 | DECLARE_COMPLETION_ONSTACK(done); |
| 1287 | struct wb_writeback_work work = { |
| 1288 | .sb = sb, |
| 1289 | .sync_mode = WB_SYNC_NONE, |
| 1290 | .tagged_writepages = 1, |
| 1291 | .done = &done, |
| 1292 | .nr_pages = nr, |
| 1293 | .reason = reason, |
| 1294 | }; |
| 1295 | |
| 1296 | if (sb->s_bdi == &noop_backing_dev_info) |
| 1297 | return; |
| 1298 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 1299 | bdi_queue_work(sb->s_bdi, &work); |
| 1300 | wait_for_completion(&done); |
| 1301 | } |
| 1302 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
| 1303 | |
| 1304 | /** |
| 1305 | * writeback_inodes_sb - writeback dirty inodes from given super_block |
| 1306 | * @sb: the superblock |
| 1307 | * @reason: reason why some writeback work was initiated |
| 1308 | * |
| 1309 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 1310 | * on how many (if any) will be written, and this function does not wait |
| 1311 | * for IO completion of submitted IO. |
| 1312 | */ |
| 1313 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
| 1314 | { |
| 1315 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
| 1316 | } |
| 1317 | EXPORT_SYMBOL(writeback_inodes_sb); |
| 1318 | |
| 1319 | /** |
| 1320 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
| 1321 | * @sb: the superblock |
| 1322 | * @nr: the number of pages to write |
| 1323 | * @reason: the reason of writeback |
| 1324 | * |
| 1325 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
| 1326 | * Returns 1 if writeback was started, 0 if not. |
| 1327 | */ |
| 1328 | int try_to_writeback_inodes_sb_nr(struct super_block *sb, |
| 1329 | unsigned long nr, |
| 1330 | enum wb_reason reason) |
| 1331 | { |
| 1332 | if (writeback_in_progress(sb->s_bdi)) |
| 1333 | return 1; |
| 1334 | |
| 1335 | if (!down_read_trylock(&sb->s_umount)) |
| 1336 | return 0; |
| 1337 | |
| 1338 | writeback_inodes_sb_nr(sb, nr, reason); |
| 1339 | up_read(&sb->s_umount); |
| 1340 | return 1; |
| 1341 | } |
| 1342 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
| 1343 | |
| 1344 | /** |
| 1345 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
| 1346 | * @sb: the superblock |
| 1347 | * @reason: reason why some writeback work was initiated |
| 1348 | * |
| 1349 | * Implement by try_to_writeback_inodes_sb_nr() |
| 1350 | * Returns 1 if writeback was started, 0 if not. |
| 1351 | */ |
| 1352 | int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
| 1353 | { |
| 1354 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
| 1355 | } |
| 1356 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
| 1357 | |
| 1358 | /** |
| 1359 | * sync_inodes_sb - sync sb inode pages |
| 1360 | * @sb: the superblock |
| 1361 | * @older_than_this: timestamp |
| 1362 | * |
| 1363 | * This function writes and waits on any dirty inode belonging to this |
| 1364 | * superblock that has been dirtied before given timestamp. |
| 1365 | */ |
| 1366 | void sync_inodes_sb(struct super_block *sb, unsigned long older_than_this) |
| 1367 | { |
| 1368 | DECLARE_COMPLETION_ONSTACK(done); |
| 1369 | struct wb_writeback_work work = { |
| 1370 | .sb = sb, |
| 1371 | .sync_mode = WB_SYNC_ALL, |
| 1372 | .nr_pages = LONG_MAX, |
| 1373 | .older_than_this = older_than_this, |
| 1374 | .older_than_this_is_set = 1, |
| 1375 | .range_cyclic = 0, |
| 1376 | .done = &done, |
| 1377 | .reason = WB_REASON_SYNC, |
| 1378 | .for_sync = 1, |
| 1379 | }; |
| 1380 | |
| 1381 | /* Nothing to do? */ |
| 1382 | if (sb->s_bdi == &noop_backing_dev_info) |
| 1383 | return; |
| 1384 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 1385 | |
| 1386 | bdi_queue_work(sb->s_bdi, &work); |
| 1387 | wait_for_completion(&done); |
| 1388 | |
| 1389 | wait_sb_inodes(sb); |
| 1390 | } |
| 1391 | EXPORT_SYMBOL(sync_inodes_sb); |
| 1392 | |
| 1393 | /** |
| 1394 | * write_inode_now - write an inode to disk |
| 1395 | * @inode: inode to write to disk |
| 1396 | * @sync: whether the write should be synchronous or not |
| 1397 | * |
| 1398 | * This function commits an inode to disk immediately if it is dirty. This is |
| 1399 | * primarily needed by knfsd. |
| 1400 | * |
| 1401 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
| 1402 | */ |
| 1403 | int write_inode_now(struct inode *inode, int sync) |
| 1404 | { |
| 1405 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
| 1406 | struct writeback_control wbc = { |
| 1407 | .nr_to_write = LONG_MAX, |
| 1408 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
| 1409 | .range_start = 0, |
| 1410 | .range_end = LLONG_MAX, |
| 1411 | }; |
| 1412 | |
| 1413 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) |
| 1414 | wbc.nr_to_write = 0; |
| 1415 | |
| 1416 | might_sleep(); |
| 1417 | return writeback_single_inode(inode, wb, &wbc); |
| 1418 | } |
| 1419 | EXPORT_SYMBOL(write_inode_now); |
| 1420 | |
| 1421 | /** |
| 1422 | * sync_inode - write an inode and its pages to disk. |
| 1423 | * @inode: the inode to sync |
| 1424 | * @wbc: controls the writeback mode |
| 1425 | * |
| 1426 | * sync_inode() will write an inode and its pages to disk. It will also |
| 1427 | * correctly update the inode on its superblock's dirty inode lists and will |
| 1428 | * update inode->i_state. |
| 1429 | * |
| 1430 | * The caller must have a ref on the inode. |
| 1431 | */ |
| 1432 | int sync_inode(struct inode *inode, struct writeback_control *wbc) |
| 1433 | { |
| 1434 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
| 1435 | } |
| 1436 | EXPORT_SYMBOL(sync_inode); |
| 1437 | |
| 1438 | /** |
| 1439 | * sync_inode_metadata - write an inode to disk |
| 1440 | * @inode: the inode to sync |
| 1441 | * @wait: wait for I/O to complete. |
| 1442 | * |
| 1443 | * Write an inode to disk and adjust its dirty state after completion. |
| 1444 | * |
| 1445 | * Note: only writes the actual inode, no associated data or other metadata. |
| 1446 | */ |
| 1447 | int sync_inode_metadata(struct inode *inode, int wait) |
| 1448 | { |
| 1449 | struct writeback_control wbc = { |
| 1450 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, |
| 1451 | .nr_to_write = 0, /* metadata-only */ |
| 1452 | }; |
| 1453 | |
| 1454 | return sync_inode(inode, &wbc); |
| 1455 | } |
| 1456 | EXPORT_SYMBOL(sync_inode_metadata); |