11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/buffer.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 2002 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95 91da177e4SLinus Torvalds * 101da177e4SLinus Torvalds * Removed a lot of unnecessary code and simplified things now that 111da177e4SLinus Torvalds * the buffer cache isn't our primary cache - Andrew Tridgell 12/96 121da177e4SLinus Torvalds * 131da177e4SLinus Torvalds * Speed up hash, lru, and free list operations. Use gfp() for allocating 141da177e4SLinus Torvalds * hash table, use SLAB cache for buffer heads. SMP threading. -DaveM 151da177e4SLinus Torvalds * 161da177e4SLinus Torvalds * Added 32k buffer block sizes - these are required older ARM systems. - RMK 171da177e4SLinus Torvalds * 181da177e4SLinus Torvalds * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de> 191da177e4SLinus Torvalds */ 201da177e4SLinus Torvalds 211da177e4SLinus Torvalds #include <linux/kernel.h> 22f361bf4aSIngo Molnar #include <linux/sched/signal.h> 231da177e4SLinus Torvalds #include <linux/syscalls.h> 241da177e4SLinus Torvalds #include <linux/fs.h> 25ae259a9cSChristoph Hellwig #include <linux/iomap.h> 261da177e4SLinus Torvalds #include <linux/mm.h> 271da177e4SLinus Torvalds #include <linux/percpu.h> 281da177e4SLinus Torvalds #include <linux/slab.h> 2916f7e0feSRandy Dunlap #include <linux/capability.h> 301da177e4SLinus Torvalds #include <linux/blkdev.h> 311da177e4SLinus Torvalds #include <linux/file.h> 321da177e4SLinus Torvalds #include <linux/quotaops.h> 331da177e4SLinus Torvalds #include <linux/highmem.h> 34630d9c47SPaul Gortmaker #include <linux/export.h> 35bafc0dbaSTejun Heo #include <linux/backing-dev.h> 361da177e4SLinus Torvalds #include <linux/writeback.h> 371da177e4SLinus Torvalds #include <linux/hash.h> 381da177e4SLinus Torvalds #include <linux/suspend.h> 391da177e4SLinus Torvalds #include <linux/buffer_head.h> 4055e829afSAndrew Morton #include <linux/task_io_accounting_ops.h> 411da177e4SLinus Torvalds #include <linux/bio.h> 421da177e4SLinus Torvalds #include <linux/cpu.h> 431da177e4SLinus Torvalds #include <linux/bitops.h> 441da177e4SLinus Torvalds #include <linux/mpage.h> 45fb1c8f93SIngo Molnar #include <linux/bit_spinlock.h> 4629f3ad7dSJan Kara #include <linux/pagevec.h> 47f745c6f5SShakeel Butt #include <linux/sched/mm.h> 485305cb83STejun Heo #include <trace/events/block.h> 491da177e4SLinus Torvalds 501da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); 512a222ca9SMike Christie static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh, 528e8f9298SJens Axboe enum rw_hint hint, struct writeback_control *wbc); 531da177e4SLinus Torvalds 541da177e4SLinus Torvalds #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) 551da177e4SLinus Torvalds 56f0059afdSTejun Heo inline void touch_buffer(struct buffer_head *bh) 57f0059afdSTejun Heo { 585305cb83STejun Heo trace_block_touch_buffer(bh); 59f0059afdSTejun Heo mark_page_accessed(bh->b_page); 60f0059afdSTejun Heo } 61f0059afdSTejun Heo EXPORT_SYMBOL(touch_buffer); 62f0059afdSTejun Heo 63fc9b52cdSHarvey Harrison void __lock_buffer(struct buffer_head *bh) 641da177e4SLinus Torvalds { 6574316201SNeilBrown wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); 661da177e4SLinus Torvalds } 671da177e4SLinus Torvalds EXPORT_SYMBOL(__lock_buffer); 681da177e4SLinus Torvalds 69fc9b52cdSHarvey Harrison void unlock_buffer(struct buffer_head *bh) 701da177e4SLinus Torvalds { 7151b07fc3SNick Piggin clear_bit_unlock(BH_Lock, &bh->b_state); 724e857c58SPeter Zijlstra smp_mb__after_atomic(); 731da177e4SLinus Torvalds wake_up_bit(&bh->b_state, BH_Lock); 741da177e4SLinus Torvalds } 751fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(unlock_buffer); 761da177e4SLinus Torvalds 771da177e4SLinus Torvalds /* 78b4597226SMel Gorman * Returns if the page has dirty or writeback buffers. If all the buffers 79b4597226SMel Gorman * are unlocked and clean then the PageDirty information is stale. If 80b4597226SMel Gorman * any of the pages are locked, it is assumed they are locked for IO. 81b4597226SMel Gorman */ 82b4597226SMel Gorman void buffer_check_dirty_writeback(struct page *page, 83b4597226SMel Gorman bool *dirty, bool *writeback) 84b4597226SMel Gorman { 85b4597226SMel Gorman struct buffer_head *head, *bh; 86b4597226SMel Gorman *dirty = false; 87b4597226SMel Gorman *writeback = false; 88b4597226SMel Gorman 89b4597226SMel Gorman BUG_ON(!PageLocked(page)); 90b4597226SMel Gorman 91b4597226SMel Gorman if (!page_has_buffers(page)) 92b4597226SMel Gorman return; 93b4597226SMel Gorman 94b4597226SMel Gorman if (PageWriteback(page)) 95b4597226SMel Gorman *writeback = true; 96b4597226SMel Gorman 97b4597226SMel Gorman head = page_buffers(page); 98b4597226SMel Gorman bh = head; 99b4597226SMel Gorman do { 100b4597226SMel Gorman if (buffer_locked(bh)) 101b4597226SMel Gorman *writeback = true; 102b4597226SMel Gorman 103b4597226SMel Gorman if (buffer_dirty(bh)) 104b4597226SMel Gorman *dirty = true; 105b4597226SMel Gorman 106b4597226SMel Gorman bh = bh->b_this_page; 107b4597226SMel Gorman } while (bh != head); 108b4597226SMel Gorman } 109b4597226SMel Gorman EXPORT_SYMBOL(buffer_check_dirty_writeback); 110b4597226SMel Gorman 111b4597226SMel Gorman /* 1121da177e4SLinus Torvalds * Block until a buffer comes unlocked. This doesn't stop it 1131da177e4SLinus Torvalds * from becoming locked again - you have to lock it yourself 1141da177e4SLinus Torvalds * if you want to preserve its state. 1151da177e4SLinus Torvalds */ 1161da177e4SLinus Torvalds void __wait_on_buffer(struct buffer_head * bh) 1171da177e4SLinus Torvalds { 11874316201SNeilBrown wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); 1191da177e4SLinus Torvalds } 1201fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__wait_on_buffer); 1211da177e4SLinus Torvalds 1221da177e4SLinus Torvalds static void 1231da177e4SLinus Torvalds __clear_page_buffers(struct page *page) 1241da177e4SLinus Torvalds { 1251da177e4SLinus Torvalds ClearPagePrivate(page); 1264c21e2f2SHugh Dickins set_page_private(page, 0); 12709cbfeafSKirill A. Shutemov put_page(page); 1281da177e4SLinus Torvalds } 1291da177e4SLinus Torvalds 130b744c2acSRobert Elliott static void buffer_io_error(struct buffer_head *bh, char *msg) 1311da177e4SLinus Torvalds { 132432f16e6SRobert Elliott if (!test_bit(BH_Quiet, &bh->b_state)) 133432f16e6SRobert Elliott printk_ratelimited(KERN_ERR 134a1c6f057SDmitry Monakhov "Buffer I/O error on dev %pg, logical block %llu%s\n", 135a1c6f057SDmitry Monakhov bh->b_bdev, (unsigned long long)bh->b_blocknr, msg); 1361da177e4SLinus Torvalds } 1371da177e4SLinus Torvalds 1381da177e4SLinus Torvalds /* 13968671f35SDmitry Monakhov * End-of-IO handler helper function which does not touch the bh after 14068671f35SDmitry Monakhov * unlocking it. 14168671f35SDmitry Monakhov * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but 14268671f35SDmitry Monakhov * a race there is benign: unlock_buffer() only use the bh's address for 14368671f35SDmitry Monakhov * hashing after unlocking the buffer, so it doesn't actually touch the bh 14468671f35SDmitry Monakhov * itself. 1451da177e4SLinus Torvalds */ 14668671f35SDmitry Monakhov static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) 1471da177e4SLinus Torvalds { 1481da177e4SLinus Torvalds if (uptodate) { 1491da177e4SLinus Torvalds set_buffer_uptodate(bh); 1501da177e4SLinus Torvalds } else { 15170246286SChristoph Hellwig /* This happens, due to failed read-ahead attempts. */ 1521da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1531da177e4SLinus Torvalds } 1541da177e4SLinus Torvalds unlock_buffer(bh); 15568671f35SDmitry Monakhov } 15668671f35SDmitry Monakhov 15768671f35SDmitry Monakhov /* 15868671f35SDmitry Monakhov * Default synchronous end-of-IO handler.. Just mark it up-to-date and 15968671f35SDmitry Monakhov * unlock the buffer. This is what ll_rw_block uses too. 16068671f35SDmitry Monakhov */ 16168671f35SDmitry Monakhov void end_buffer_read_sync(struct buffer_head *bh, int uptodate) 16268671f35SDmitry Monakhov { 16368671f35SDmitry Monakhov __end_buffer_read_notouch(bh, uptodate); 1641da177e4SLinus Torvalds put_bh(bh); 1651da177e4SLinus Torvalds } 1661fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_read_sync); 1671da177e4SLinus Torvalds 1681da177e4SLinus Torvalds void end_buffer_write_sync(struct buffer_head *bh, int uptodate) 1691da177e4SLinus Torvalds { 1701da177e4SLinus Torvalds if (uptodate) { 1711da177e4SLinus Torvalds set_buffer_uptodate(bh); 1721da177e4SLinus Torvalds } else { 173b744c2acSRobert Elliott buffer_io_error(bh, ", lost sync page write"); 17487354e5dSJeff Layton mark_buffer_write_io_error(bh); 1751da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1761da177e4SLinus Torvalds } 1771da177e4SLinus Torvalds unlock_buffer(bh); 1781da177e4SLinus Torvalds put_bh(bh); 1791da177e4SLinus Torvalds } 1801fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_write_sync); 1811da177e4SLinus Torvalds 1821da177e4SLinus Torvalds /* 1831da177e4SLinus Torvalds * Various filesystems appear to want __find_get_block to be non-blocking. 1841da177e4SLinus Torvalds * But it's the page lock which protects the buffers. To get around this, 1851da177e4SLinus Torvalds * we get exclusion from try_to_free_buffers with the blockdev mapping's 1861da177e4SLinus Torvalds * private_lock. 1871da177e4SLinus Torvalds * 188b93b0163SMatthew Wilcox * Hack idea: for the blockdev mapping, private_lock contention 1891da177e4SLinus Torvalds * may be quite high. This code could TryLock the page, and if that 190b93b0163SMatthew Wilcox * succeeds, there is no need to take private_lock. 1911da177e4SLinus Torvalds */ 1921da177e4SLinus Torvalds static struct buffer_head * 193385fd4c5SCoywolf Qi Hunt __find_get_block_slow(struct block_device *bdev, sector_t block) 1941da177e4SLinus Torvalds { 1951da177e4SLinus Torvalds struct inode *bd_inode = bdev->bd_inode; 1961da177e4SLinus Torvalds struct address_space *bd_mapping = bd_inode->i_mapping; 1971da177e4SLinus Torvalds struct buffer_head *ret = NULL; 1981da177e4SLinus Torvalds pgoff_t index; 1991da177e4SLinus Torvalds struct buffer_head *bh; 2001da177e4SLinus Torvalds struct buffer_head *head; 2011da177e4SLinus Torvalds struct page *page; 2021da177e4SLinus Torvalds int all_mapped = 1; 20343636c80STetsuo Handa static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1); 2041da177e4SLinus Torvalds 20509cbfeafSKirill A. Shutemov index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); 2062457aec6SMel Gorman page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED); 2071da177e4SLinus Torvalds if (!page) 2081da177e4SLinus Torvalds goto out; 2091da177e4SLinus Torvalds 2101da177e4SLinus Torvalds spin_lock(&bd_mapping->private_lock); 2111da177e4SLinus Torvalds if (!page_has_buffers(page)) 2121da177e4SLinus Torvalds goto out_unlock; 2131da177e4SLinus Torvalds head = page_buffers(page); 2141da177e4SLinus Torvalds bh = head; 2151da177e4SLinus Torvalds do { 21697f76d3dSNikanth Karthikesan if (!buffer_mapped(bh)) 21797f76d3dSNikanth Karthikesan all_mapped = 0; 21897f76d3dSNikanth Karthikesan else if (bh->b_blocknr == block) { 2191da177e4SLinus Torvalds ret = bh; 2201da177e4SLinus Torvalds get_bh(bh); 2211da177e4SLinus Torvalds goto out_unlock; 2221da177e4SLinus Torvalds } 2231da177e4SLinus Torvalds bh = bh->b_this_page; 2241da177e4SLinus Torvalds } while (bh != head); 2251da177e4SLinus Torvalds 2261da177e4SLinus Torvalds /* we might be here because some of the buffers on this page are 2271da177e4SLinus Torvalds * not mapped. This is due to various races between 2281da177e4SLinus Torvalds * file io on the block device and getblk. It gets dealt with 2291da177e4SLinus Torvalds * elsewhere, don't buffer_error if we had some unmapped buffers 2301da177e4SLinus Torvalds */ 23143636c80STetsuo Handa ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE); 23243636c80STetsuo Handa if (all_mapped && __ratelimit(&last_warned)) { 23343636c80STetsuo Handa printk("__find_get_block_slow() failed. block=%llu, " 23443636c80STetsuo Handa "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, " 23543636c80STetsuo Handa "device %pg blocksize: %d\n", 236205f87f6SBadari Pulavarty (unsigned long long)block, 23743636c80STetsuo Handa (unsigned long long)bh->b_blocknr, 23843636c80STetsuo Handa bh->b_state, bh->b_size, bdev, 23972a2ebd8STao Ma 1 << bd_inode->i_blkbits); 2401da177e4SLinus Torvalds } 2411da177e4SLinus Torvalds out_unlock: 2421da177e4SLinus Torvalds spin_unlock(&bd_mapping->private_lock); 24309cbfeafSKirill A. Shutemov put_page(page); 2441da177e4SLinus Torvalds out: 2451da177e4SLinus Torvalds return ret; 2461da177e4SLinus Torvalds } 2471da177e4SLinus Torvalds 2481da177e4SLinus Torvalds /* 2491da177e4SLinus Torvalds * I/O completion handler for block_read_full_page() - pages 2501da177e4SLinus Torvalds * which come unlocked at the end of I/O. 2511da177e4SLinus Torvalds */ 2521da177e4SLinus Torvalds static void end_buffer_async_read(struct buffer_head *bh, int uptodate) 2531da177e4SLinus Torvalds { 2541da177e4SLinus Torvalds unsigned long flags; 255a3972203SNick Piggin struct buffer_head *first; 2561da177e4SLinus Torvalds struct buffer_head *tmp; 2571da177e4SLinus Torvalds struct page *page; 2581da177e4SLinus Torvalds int page_uptodate = 1; 2591da177e4SLinus Torvalds 2601da177e4SLinus Torvalds BUG_ON(!buffer_async_read(bh)); 2611da177e4SLinus Torvalds 2621da177e4SLinus Torvalds page = bh->b_page; 2631da177e4SLinus Torvalds if (uptodate) { 2641da177e4SLinus Torvalds set_buffer_uptodate(bh); 2651da177e4SLinus Torvalds } else { 2661da177e4SLinus Torvalds clear_buffer_uptodate(bh); 267b744c2acSRobert Elliott buffer_io_error(bh, ", async page read"); 2681da177e4SLinus Torvalds SetPageError(page); 2691da177e4SLinus Torvalds } 2701da177e4SLinus Torvalds 2711da177e4SLinus Torvalds /* 2721da177e4SLinus Torvalds * Be _very_ careful from here on. Bad things can happen if 2731da177e4SLinus Torvalds * two buffer heads end IO at almost the same time and both 2741da177e4SLinus Torvalds * decide that the page is now completely done. 2751da177e4SLinus Torvalds */ 276a3972203SNick Piggin first = page_buffers(page); 277a3972203SNick Piggin local_irq_save(flags); 278a3972203SNick Piggin bit_spin_lock(BH_Uptodate_Lock, &first->b_state); 2791da177e4SLinus Torvalds clear_buffer_async_read(bh); 2801da177e4SLinus Torvalds unlock_buffer(bh); 2811da177e4SLinus Torvalds tmp = bh; 2821da177e4SLinus Torvalds do { 2831da177e4SLinus Torvalds if (!buffer_uptodate(tmp)) 2841da177e4SLinus Torvalds page_uptodate = 0; 2851da177e4SLinus Torvalds if (buffer_async_read(tmp)) { 2861da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 2871da177e4SLinus Torvalds goto still_busy; 2881da177e4SLinus Torvalds } 2891da177e4SLinus Torvalds tmp = tmp->b_this_page; 2901da177e4SLinus Torvalds } while (tmp != bh); 291a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 292a3972203SNick Piggin local_irq_restore(flags); 2931da177e4SLinus Torvalds 2941da177e4SLinus Torvalds /* 2951da177e4SLinus Torvalds * If none of the buffers had errors and they are all 2961da177e4SLinus Torvalds * uptodate then we can set the page uptodate. 2971da177e4SLinus Torvalds */ 2981da177e4SLinus Torvalds if (page_uptodate && !PageError(page)) 2991da177e4SLinus Torvalds SetPageUptodate(page); 3001da177e4SLinus Torvalds unlock_page(page); 3011da177e4SLinus Torvalds return; 3021da177e4SLinus Torvalds 3031da177e4SLinus Torvalds still_busy: 304a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 305a3972203SNick Piggin local_irq_restore(flags); 3061da177e4SLinus Torvalds return; 3071da177e4SLinus Torvalds } 3081da177e4SLinus Torvalds 3091da177e4SLinus Torvalds /* 3101da177e4SLinus Torvalds * Completion handler for block_write_full_page() - pages which are unlocked 3111da177e4SLinus Torvalds * during I/O, and which have PageWriteback cleared upon I/O completion. 3121da177e4SLinus Torvalds */ 31335c80d5fSChris Mason void end_buffer_async_write(struct buffer_head *bh, int uptodate) 3141da177e4SLinus Torvalds { 3151da177e4SLinus Torvalds unsigned long flags; 316a3972203SNick Piggin struct buffer_head *first; 3171da177e4SLinus Torvalds struct buffer_head *tmp; 3181da177e4SLinus Torvalds struct page *page; 3191da177e4SLinus Torvalds 3201da177e4SLinus Torvalds BUG_ON(!buffer_async_write(bh)); 3211da177e4SLinus Torvalds 3221da177e4SLinus Torvalds page = bh->b_page; 3231da177e4SLinus Torvalds if (uptodate) { 3241da177e4SLinus Torvalds set_buffer_uptodate(bh); 3251da177e4SLinus Torvalds } else { 326b744c2acSRobert Elliott buffer_io_error(bh, ", lost async page write"); 32787354e5dSJeff Layton mark_buffer_write_io_error(bh); 3281da177e4SLinus Torvalds clear_buffer_uptodate(bh); 3291da177e4SLinus Torvalds SetPageError(page); 3301da177e4SLinus Torvalds } 3311da177e4SLinus Torvalds 332a3972203SNick Piggin first = page_buffers(page); 333a3972203SNick Piggin local_irq_save(flags); 334a3972203SNick Piggin bit_spin_lock(BH_Uptodate_Lock, &first->b_state); 335a3972203SNick Piggin 3361da177e4SLinus Torvalds clear_buffer_async_write(bh); 3371da177e4SLinus Torvalds unlock_buffer(bh); 3381da177e4SLinus Torvalds tmp = bh->b_this_page; 3391da177e4SLinus Torvalds while (tmp != bh) { 3401da177e4SLinus Torvalds if (buffer_async_write(tmp)) { 3411da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 3421da177e4SLinus Torvalds goto still_busy; 3431da177e4SLinus Torvalds } 3441da177e4SLinus Torvalds tmp = tmp->b_this_page; 3451da177e4SLinus Torvalds } 346a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 347a3972203SNick Piggin local_irq_restore(flags); 3481da177e4SLinus Torvalds end_page_writeback(page); 3491da177e4SLinus Torvalds return; 3501da177e4SLinus Torvalds 3511da177e4SLinus Torvalds still_busy: 352a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 353a3972203SNick Piggin local_irq_restore(flags); 3541da177e4SLinus Torvalds return; 3551da177e4SLinus Torvalds } 3561fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_async_write); 3571da177e4SLinus Torvalds 3581da177e4SLinus Torvalds /* 3591da177e4SLinus Torvalds * If a page's buffers are under async readin (end_buffer_async_read 3601da177e4SLinus Torvalds * completion) then there is a possibility that another thread of 3611da177e4SLinus Torvalds * control could lock one of the buffers after it has completed 3621da177e4SLinus Torvalds * but while some of the other buffers have not completed. This 3631da177e4SLinus Torvalds * locked buffer would confuse end_buffer_async_read() into not unlocking 3641da177e4SLinus Torvalds * the page. So the absence of BH_Async_Read tells end_buffer_async_read() 3651da177e4SLinus Torvalds * that this buffer is not under async I/O. 3661da177e4SLinus Torvalds * 3671da177e4SLinus Torvalds * The page comes unlocked when it has no locked buffer_async buffers 3681da177e4SLinus Torvalds * left. 3691da177e4SLinus Torvalds * 3701da177e4SLinus Torvalds * PageLocked prevents anyone starting new async I/O reads any of 3711da177e4SLinus Torvalds * the buffers. 3721da177e4SLinus Torvalds * 3731da177e4SLinus Torvalds * PageWriteback is used to prevent simultaneous writeout of the same 3741da177e4SLinus Torvalds * page. 3751da177e4SLinus Torvalds * 3761da177e4SLinus Torvalds * PageLocked prevents anyone from starting writeback of a page which is 3771da177e4SLinus Torvalds * under read I/O (PageWriteback is only ever set against a locked page). 3781da177e4SLinus Torvalds */ 3791da177e4SLinus Torvalds static void mark_buffer_async_read(struct buffer_head *bh) 3801da177e4SLinus Torvalds { 3811da177e4SLinus Torvalds bh->b_end_io = end_buffer_async_read; 3821da177e4SLinus Torvalds set_buffer_async_read(bh); 3831da177e4SLinus Torvalds } 3841da177e4SLinus Torvalds 3851fe72eaaSH Hartley Sweeten static void mark_buffer_async_write_endio(struct buffer_head *bh, 38635c80d5fSChris Mason bh_end_io_t *handler) 38735c80d5fSChris Mason { 38835c80d5fSChris Mason bh->b_end_io = handler; 38935c80d5fSChris Mason set_buffer_async_write(bh); 39035c80d5fSChris Mason } 39135c80d5fSChris Mason 3921da177e4SLinus Torvalds void mark_buffer_async_write(struct buffer_head *bh) 3931da177e4SLinus Torvalds { 39435c80d5fSChris Mason mark_buffer_async_write_endio(bh, end_buffer_async_write); 3951da177e4SLinus Torvalds } 3961da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_async_write); 3971da177e4SLinus Torvalds 3981da177e4SLinus Torvalds 3991da177e4SLinus Torvalds /* 4001da177e4SLinus Torvalds * fs/buffer.c contains helper functions for buffer-backed address space's 4011da177e4SLinus Torvalds * fsync functions. A common requirement for buffer-based filesystems is 4021da177e4SLinus Torvalds * that certain data from the backing blockdev needs to be written out for 4031da177e4SLinus Torvalds * a successful fsync(). For example, ext2 indirect blocks need to be 4041da177e4SLinus Torvalds * written back and waited upon before fsync() returns. 4051da177e4SLinus Torvalds * 4061da177e4SLinus Torvalds * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), 4071da177e4SLinus Torvalds * inode_has_buffers() and invalidate_inode_buffers() are provided for the 4081da177e4SLinus Torvalds * management of a list of dependent buffers at ->i_mapping->private_list. 4091da177e4SLinus Torvalds * 4101da177e4SLinus Torvalds * Locking is a little subtle: try_to_free_buffers() will remove buffers 4111da177e4SLinus Torvalds * from their controlling inode's queue when they are being freed. But 4121da177e4SLinus Torvalds * try_to_free_buffers() will be operating against the *blockdev* mapping 4131da177e4SLinus Torvalds * at the time, not against the S_ISREG file which depends on those buffers. 4141da177e4SLinus Torvalds * So the locking for private_list is via the private_lock in the address_space 4151da177e4SLinus Torvalds * which backs the buffers. Which is different from the address_space 4161da177e4SLinus Torvalds * against which the buffers are listed. So for a particular address_space, 4171da177e4SLinus Torvalds * mapping->private_lock does *not* protect mapping->private_list! In fact, 4181da177e4SLinus Torvalds * mapping->private_list will always be protected by the backing blockdev's 4191da177e4SLinus Torvalds * ->private_lock. 4201da177e4SLinus Torvalds * 4211da177e4SLinus Torvalds * Which introduces a requirement: all buffers on an address_space's 4221da177e4SLinus Torvalds * ->private_list must be from the same address_space: the blockdev's. 4231da177e4SLinus Torvalds * 4241da177e4SLinus Torvalds * address_spaces which do not place buffers at ->private_list via these 4251da177e4SLinus Torvalds * utility functions are free to use private_lock and private_list for 4261da177e4SLinus Torvalds * whatever they want. The only requirement is that list_empty(private_list) 4271da177e4SLinus Torvalds * be true at clear_inode() time. 4281da177e4SLinus Torvalds * 4291da177e4SLinus Torvalds * FIXME: clear_inode should not call invalidate_inode_buffers(). The 4301da177e4SLinus Torvalds * filesystems should do that. invalidate_inode_buffers() should just go 4311da177e4SLinus Torvalds * BUG_ON(!list_empty). 4321da177e4SLinus Torvalds * 4331da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should 4341da177e4SLinus Torvalds * take an address_space, not an inode. And it should be called 4351da177e4SLinus Torvalds * mark_buffer_dirty_fsync() to clearly define why those buffers are being 4361da177e4SLinus Torvalds * queued up. 4371da177e4SLinus Torvalds * 4381da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the 4391da177e4SLinus Torvalds * list if it is already on a list. Because if the buffer is on a list, 4401da177e4SLinus Torvalds * it *must* already be on the right one. If not, the filesystem is being 4411da177e4SLinus Torvalds * silly. This will save a ton of locking. But first we have to ensure 4421da177e4SLinus Torvalds * that buffers are taken *off* the old inode's list when they are freed 4431da177e4SLinus Torvalds * (presumably in truncate). That requires careful auditing of all 4441da177e4SLinus Torvalds * filesystems (do it inside bforget()). It could also be done by bringing 4451da177e4SLinus Torvalds * b_inode back. 4461da177e4SLinus Torvalds */ 4471da177e4SLinus Torvalds 4481da177e4SLinus Torvalds /* 4491da177e4SLinus Torvalds * The buffer's backing address_space's private_lock must be held 4501da177e4SLinus Torvalds */ 451dbacefc9SThomas Petazzoni static void __remove_assoc_queue(struct buffer_head *bh) 4521da177e4SLinus Torvalds { 4531da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 45458ff407bSJan Kara WARN_ON(!bh->b_assoc_map); 45558ff407bSJan Kara bh->b_assoc_map = NULL; 4561da177e4SLinus Torvalds } 4571da177e4SLinus Torvalds 4581da177e4SLinus Torvalds int inode_has_buffers(struct inode *inode) 4591da177e4SLinus Torvalds { 4601da177e4SLinus Torvalds return !list_empty(&inode->i_data.private_list); 4611da177e4SLinus Torvalds } 4621da177e4SLinus Torvalds 4631da177e4SLinus Torvalds /* 4641da177e4SLinus Torvalds * osync is designed to support O_SYNC io. It waits synchronously for 4651da177e4SLinus Torvalds * all already-submitted IO to complete, but does not queue any new 4661da177e4SLinus Torvalds * writes to the disk. 4671da177e4SLinus Torvalds * 4681da177e4SLinus Torvalds * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as 4691da177e4SLinus Torvalds * you dirty the buffers, and then use osync_inode_buffers to wait for 4701da177e4SLinus Torvalds * completion. Any other dirty buffers which are not yet queued for 4711da177e4SLinus Torvalds * write will not be flushed to disk by the osync. 4721da177e4SLinus Torvalds */ 4731da177e4SLinus Torvalds static int osync_buffers_list(spinlock_t *lock, struct list_head *list) 4741da177e4SLinus Torvalds { 4751da177e4SLinus Torvalds struct buffer_head *bh; 4761da177e4SLinus Torvalds struct list_head *p; 4771da177e4SLinus Torvalds int err = 0; 4781da177e4SLinus Torvalds 4791da177e4SLinus Torvalds spin_lock(lock); 4801da177e4SLinus Torvalds repeat: 4811da177e4SLinus Torvalds list_for_each_prev(p, list) { 4821da177e4SLinus Torvalds bh = BH_ENTRY(p); 4831da177e4SLinus Torvalds if (buffer_locked(bh)) { 4841da177e4SLinus Torvalds get_bh(bh); 4851da177e4SLinus Torvalds spin_unlock(lock); 4861da177e4SLinus Torvalds wait_on_buffer(bh); 4871da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 4881da177e4SLinus Torvalds err = -EIO; 4891da177e4SLinus Torvalds brelse(bh); 4901da177e4SLinus Torvalds spin_lock(lock); 4911da177e4SLinus Torvalds goto repeat; 4921da177e4SLinus Torvalds } 4931da177e4SLinus Torvalds } 4941da177e4SLinus Torvalds spin_unlock(lock); 4951da177e4SLinus Torvalds return err; 4961da177e4SLinus Torvalds } 4971da177e4SLinus Torvalds 49808fdc8a0SMateusz Guzik void emergency_thaw_bdev(struct super_block *sb) 499c2d75438SEric Sandeen { 500c2d75438SEric Sandeen while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb)) 501a1c6f057SDmitry Monakhov printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev); 502c2d75438SEric Sandeen } 50301a05b33SAl Viro 5041da177e4SLinus Torvalds /** 50578a4a50aSRandy Dunlap * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers 50667be2dd1SMartin Waitz * @mapping: the mapping which wants those buffers written 5071da177e4SLinus Torvalds * 5081da177e4SLinus Torvalds * Starts I/O against the buffers at mapping->private_list, and waits upon 5091da177e4SLinus Torvalds * that I/O. 5101da177e4SLinus Torvalds * 51167be2dd1SMartin Waitz * Basically, this is a convenience function for fsync(). 51267be2dd1SMartin Waitz * @mapping is a file or directory which needs those buffers to be written for 51367be2dd1SMartin Waitz * a successful fsync(). 5141da177e4SLinus Torvalds */ 5151da177e4SLinus Torvalds int sync_mapping_buffers(struct address_space *mapping) 5161da177e4SLinus Torvalds { 517252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 5181da177e4SLinus Torvalds 5191da177e4SLinus Torvalds if (buffer_mapping == NULL || list_empty(&mapping->private_list)) 5201da177e4SLinus Torvalds return 0; 5211da177e4SLinus Torvalds 5221da177e4SLinus Torvalds return fsync_buffers_list(&buffer_mapping->private_lock, 5231da177e4SLinus Torvalds &mapping->private_list); 5241da177e4SLinus Torvalds } 5251da177e4SLinus Torvalds EXPORT_SYMBOL(sync_mapping_buffers); 5261da177e4SLinus Torvalds 5271da177e4SLinus Torvalds /* 5281da177e4SLinus Torvalds * Called when we've recently written block `bblock', and it is known that 5291da177e4SLinus Torvalds * `bblock' was for a buffer_boundary() buffer. This means that the block at 5301da177e4SLinus Torvalds * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's 5311da177e4SLinus Torvalds * dirty, schedule it for IO. So that indirects merge nicely with their data. 5321da177e4SLinus Torvalds */ 5331da177e4SLinus Torvalds void write_boundary_block(struct block_device *bdev, 5341da177e4SLinus Torvalds sector_t bblock, unsigned blocksize) 5351da177e4SLinus Torvalds { 5361da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); 5371da177e4SLinus Torvalds if (bh) { 5381da177e4SLinus Torvalds if (buffer_dirty(bh)) 539dfec8a14SMike Christie ll_rw_block(REQ_OP_WRITE, 0, 1, &bh); 5401da177e4SLinus Torvalds put_bh(bh); 5411da177e4SLinus Torvalds } 5421da177e4SLinus Torvalds } 5431da177e4SLinus Torvalds 5441da177e4SLinus Torvalds void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) 5451da177e4SLinus Torvalds { 5461da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 5471da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 5481da177e4SLinus Torvalds 5491da177e4SLinus Torvalds mark_buffer_dirty(bh); 550252aa6f5SRafael Aquini if (!mapping->private_data) { 551252aa6f5SRafael Aquini mapping->private_data = buffer_mapping; 5521da177e4SLinus Torvalds } else { 553252aa6f5SRafael Aquini BUG_ON(mapping->private_data != buffer_mapping); 5541da177e4SLinus Torvalds } 555535ee2fbSJan Kara if (!bh->b_assoc_map) { 5561da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 5571da177e4SLinus Torvalds list_move_tail(&bh->b_assoc_buffers, 5581da177e4SLinus Torvalds &mapping->private_list); 55958ff407bSJan Kara bh->b_assoc_map = mapping; 5601da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 5611da177e4SLinus Torvalds } 5621da177e4SLinus Torvalds } 5631da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_dirty_inode); 5641da177e4SLinus Torvalds 5651da177e4SLinus Torvalds /* 566ec82e1c1SMatthew Wilcox * Mark the page dirty, and set it dirty in the page cache, and mark the inode 567787d2214SNick Piggin * dirty. 568787d2214SNick Piggin * 569787d2214SNick Piggin * If warn is true, then emit a warning if the page is not uptodate and has 570787d2214SNick Piggin * not been truncated. 571c4843a75SGreg Thelen * 57281f8c3a4SJohannes Weiner * The caller must hold lock_page_memcg(). 573787d2214SNick Piggin */ 574f82b3764SMatthew Wilcox void __set_page_dirty(struct page *page, struct address_space *mapping, 57562cccb8cSJohannes Weiner int warn) 576787d2214SNick Piggin { 577227d53b3SKOSAKI Motohiro unsigned long flags; 578227d53b3SKOSAKI Motohiro 579b93b0163SMatthew Wilcox xa_lock_irqsave(&mapping->i_pages, flags); 580787d2214SNick Piggin if (page->mapping) { /* Race with truncate? */ 581787d2214SNick Piggin WARN_ON_ONCE(warn && !PageUptodate(page)); 58262cccb8cSJohannes Weiner account_page_dirtied(page, mapping); 583ec82e1c1SMatthew Wilcox __xa_set_mark(&mapping->i_pages, page_index(page), 584ec82e1c1SMatthew Wilcox PAGECACHE_TAG_DIRTY); 585787d2214SNick Piggin } 586b93b0163SMatthew Wilcox xa_unlock_irqrestore(&mapping->i_pages, flags); 587787d2214SNick Piggin } 588f82b3764SMatthew Wilcox EXPORT_SYMBOL_GPL(__set_page_dirty); 589787d2214SNick Piggin 590787d2214SNick Piggin /* 5911da177e4SLinus Torvalds * Add a page to the dirty page list. 5921da177e4SLinus Torvalds * 5931da177e4SLinus Torvalds * It is a sad fact of life that this function is called from several places 5941da177e4SLinus Torvalds * deeply under spinlocking. It may not sleep. 5951da177e4SLinus Torvalds * 5961da177e4SLinus Torvalds * If the page has buffers, the uptodate buffers are set dirty, to preserve 5971da177e4SLinus Torvalds * dirty-state coherency between the page and the buffers. It the page does 5981da177e4SLinus Torvalds * not have buffers then when they are later attached they will all be set 5991da177e4SLinus Torvalds * dirty. 6001da177e4SLinus Torvalds * 6011da177e4SLinus Torvalds * The buffers are dirtied before the page is dirtied. There's a small race 6021da177e4SLinus Torvalds * window in which a writepage caller may see the page cleanness but not the 6031da177e4SLinus Torvalds * buffer dirtiness. That's fine. If this code were to set the page dirty 6041da177e4SLinus Torvalds * before the buffers, a concurrent writepage caller could clear the page dirty 6051da177e4SLinus Torvalds * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean 6061da177e4SLinus Torvalds * page on the dirty page list. 6071da177e4SLinus Torvalds * 6081da177e4SLinus Torvalds * We use private_lock to lock against try_to_free_buffers while using the 6091da177e4SLinus Torvalds * page's buffer list. Also use this to protect against clean buffers being 6101da177e4SLinus Torvalds * added to the page after it was set dirty. 6111da177e4SLinus Torvalds * 6121da177e4SLinus Torvalds * FIXME: may need to call ->reservepage here as well. That's rather up to the 6131da177e4SLinus Torvalds * address_space though. 6141da177e4SLinus Torvalds */ 6151da177e4SLinus Torvalds int __set_page_dirty_buffers(struct page *page) 6161da177e4SLinus Torvalds { 617a8e7d49aSLinus Torvalds int newly_dirty; 618787d2214SNick Piggin struct address_space *mapping = page_mapping(page); 619ebf7a227SNick Piggin 620ebf7a227SNick Piggin if (unlikely(!mapping)) 621ebf7a227SNick Piggin return !TestSetPageDirty(page); 6221da177e4SLinus Torvalds 6231da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 6241da177e4SLinus Torvalds if (page_has_buffers(page)) { 6251da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 6261da177e4SLinus Torvalds struct buffer_head *bh = head; 6271da177e4SLinus Torvalds 6281da177e4SLinus Torvalds do { 6291da177e4SLinus Torvalds set_buffer_dirty(bh); 6301da177e4SLinus Torvalds bh = bh->b_this_page; 6311da177e4SLinus Torvalds } while (bh != head); 6321da177e4SLinus Torvalds } 633c4843a75SGreg Thelen /* 63481f8c3a4SJohannes Weiner * Lock out page->mem_cgroup migration to keep PageDirty 63581f8c3a4SJohannes Weiner * synchronized with per-memcg dirty page counters. 636c4843a75SGreg Thelen */ 63762cccb8cSJohannes Weiner lock_page_memcg(page); 638a8e7d49aSLinus Torvalds newly_dirty = !TestSetPageDirty(page); 6391da177e4SLinus Torvalds spin_unlock(&mapping->private_lock); 6401da177e4SLinus Torvalds 641a8e7d49aSLinus Torvalds if (newly_dirty) 64262cccb8cSJohannes Weiner __set_page_dirty(page, mapping, 1); 643c4843a75SGreg Thelen 64462cccb8cSJohannes Weiner unlock_page_memcg(page); 645c4843a75SGreg Thelen 646c4843a75SGreg Thelen if (newly_dirty) 647c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 648c4843a75SGreg Thelen 649a8e7d49aSLinus Torvalds return newly_dirty; 6501da177e4SLinus Torvalds } 6511da177e4SLinus Torvalds EXPORT_SYMBOL(__set_page_dirty_buffers); 6521da177e4SLinus Torvalds 6531da177e4SLinus Torvalds /* 6541da177e4SLinus Torvalds * Write out and wait upon a list of buffers. 6551da177e4SLinus Torvalds * 6561da177e4SLinus Torvalds * We have conflicting pressures: we want to make sure that all 6571da177e4SLinus Torvalds * initially dirty buffers get waited on, but that any subsequently 6581da177e4SLinus Torvalds * dirtied buffers don't. After all, we don't want fsync to last 6591da177e4SLinus Torvalds * forever if somebody is actively writing to the file. 6601da177e4SLinus Torvalds * 6611da177e4SLinus Torvalds * Do this in two main stages: first we copy dirty buffers to a 6621da177e4SLinus Torvalds * temporary inode list, queueing the writes as we go. Then we clean 6631da177e4SLinus Torvalds * up, waiting for those writes to complete. 6641da177e4SLinus Torvalds * 6651da177e4SLinus Torvalds * During this second stage, any subsequent updates to the file may end 6661da177e4SLinus Torvalds * up refiling the buffer on the original inode's dirty list again, so 6671da177e4SLinus Torvalds * there is a chance we will end up with a buffer queued for write but 6681da177e4SLinus Torvalds * not yet completed on that list. So, as a final cleanup we go through 6691da177e4SLinus Torvalds * the osync code to catch these locked, dirty buffers without requeuing 6701da177e4SLinus Torvalds * any newly dirty buffers for write. 6711da177e4SLinus Torvalds */ 6721da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) 6731da177e4SLinus Torvalds { 6741da177e4SLinus Torvalds struct buffer_head *bh; 6751da177e4SLinus Torvalds struct list_head tmp; 6767eaceaccSJens Axboe struct address_space *mapping; 6771da177e4SLinus Torvalds int err = 0, err2; 6784ee2491eSJens Axboe struct blk_plug plug; 6791da177e4SLinus Torvalds 6801da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 6814ee2491eSJens Axboe blk_start_plug(&plug); 6821da177e4SLinus Torvalds 6831da177e4SLinus Torvalds spin_lock(lock); 6841da177e4SLinus Torvalds while (!list_empty(list)) { 6851da177e4SLinus Torvalds bh = BH_ENTRY(list->next); 686535ee2fbSJan Kara mapping = bh->b_assoc_map; 68758ff407bSJan Kara __remove_assoc_queue(bh); 688535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 689535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 690535ee2fbSJan Kara smp_mb(); 6911da177e4SLinus Torvalds if (buffer_dirty(bh) || buffer_locked(bh)) { 6921da177e4SLinus Torvalds list_add(&bh->b_assoc_buffers, &tmp); 693535ee2fbSJan Kara bh->b_assoc_map = mapping; 6941da177e4SLinus Torvalds if (buffer_dirty(bh)) { 6951da177e4SLinus Torvalds get_bh(bh); 6961da177e4SLinus Torvalds spin_unlock(lock); 6971da177e4SLinus Torvalds /* 6981da177e4SLinus Torvalds * Ensure any pending I/O completes so that 6999cb569d6SChristoph Hellwig * write_dirty_buffer() actually writes the 7009cb569d6SChristoph Hellwig * current contents - it is a noop if I/O is 7019cb569d6SChristoph Hellwig * still in flight on potentially older 7029cb569d6SChristoph Hellwig * contents. 7031da177e4SLinus Torvalds */ 70470fd7614SChristoph Hellwig write_dirty_buffer(bh, REQ_SYNC); 7059cf6b720SJens Axboe 7069cf6b720SJens Axboe /* 7079cf6b720SJens Axboe * Kick off IO for the previous mapping. Note 7089cf6b720SJens Axboe * that we will not run the very last mapping, 7099cf6b720SJens Axboe * wait_on_buffer() will do that for us 7109cf6b720SJens Axboe * through sync_buffer(). 7119cf6b720SJens Axboe */ 7121da177e4SLinus Torvalds brelse(bh); 7131da177e4SLinus Torvalds spin_lock(lock); 7141da177e4SLinus Torvalds } 7151da177e4SLinus Torvalds } 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 7184ee2491eSJens Axboe spin_unlock(lock); 7194ee2491eSJens Axboe blk_finish_plug(&plug); 7204ee2491eSJens Axboe spin_lock(lock); 7214ee2491eSJens Axboe 7221da177e4SLinus Torvalds while (!list_empty(&tmp)) { 7231da177e4SLinus Torvalds bh = BH_ENTRY(tmp.prev); 7241da177e4SLinus Torvalds get_bh(bh); 725535ee2fbSJan Kara mapping = bh->b_assoc_map; 726535ee2fbSJan Kara __remove_assoc_queue(bh); 727535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 728535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 729535ee2fbSJan Kara smp_mb(); 730535ee2fbSJan Kara if (buffer_dirty(bh)) { 731535ee2fbSJan Kara list_add(&bh->b_assoc_buffers, 732e3892296SJan Kara &mapping->private_list); 733535ee2fbSJan Kara bh->b_assoc_map = mapping; 734535ee2fbSJan Kara } 7351da177e4SLinus Torvalds spin_unlock(lock); 7361da177e4SLinus Torvalds wait_on_buffer(bh); 7371da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 7381da177e4SLinus Torvalds err = -EIO; 7391da177e4SLinus Torvalds brelse(bh); 7401da177e4SLinus Torvalds spin_lock(lock); 7411da177e4SLinus Torvalds } 7421da177e4SLinus Torvalds 7431da177e4SLinus Torvalds spin_unlock(lock); 7441da177e4SLinus Torvalds err2 = osync_buffers_list(lock, list); 7451da177e4SLinus Torvalds if (err) 7461da177e4SLinus Torvalds return err; 7471da177e4SLinus Torvalds else 7481da177e4SLinus Torvalds return err2; 7491da177e4SLinus Torvalds } 7501da177e4SLinus Torvalds 7511da177e4SLinus Torvalds /* 7521da177e4SLinus Torvalds * Invalidate any and all dirty buffers on a given inode. We are 7531da177e4SLinus Torvalds * probably unmounting the fs, but that doesn't mean we have already 7541da177e4SLinus Torvalds * done a sync(). Just drop the buffers from the inode list. 7551da177e4SLinus Torvalds * 7561da177e4SLinus Torvalds * NOTE: we take the inode's blockdev's mapping's private_lock. Which 7571da177e4SLinus Torvalds * assumes that all the buffers are against the blockdev. Not true 7581da177e4SLinus Torvalds * for reiserfs. 7591da177e4SLinus Torvalds */ 7601da177e4SLinus Torvalds void invalidate_inode_buffers(struct inode *inode) 7611da177e4SLinus Torvalds { 7621da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 7631da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 7641da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 765252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 7661da177e4SLinus Torvalds 7671da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 7681da177e4SLinus Torvalds while (!list_empty(list)) 7691da177e4SLinus Torvalds __remove_assoc_queue(BH_ENTRY(list->next)); 7701da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds } 77352b19ac9SJan Kara EXPORT_SYMBOL(invalidate_inode_buffers); 7741da177e4SLinus Torvalds 7751da177e4SLinus Torvalds /* 7761da177e4SLinus Torvalds * Remove any clean buffers from the inode's buffer list. This is called 7771da177e4SLinus Torvalds * when we're trying to free the inode itself. Those buffers can pin it. 7781da177e4SLinus Torvalds * 7791da177e4SLinus Torvalds * Returns true if all buffers were removed. 7801da177e4SLinus Torvalds */ 7811da177e4SLinus Torvalds int remove_inode_buffers(struct inode *inode) 7821da177e4SLinus Torvalds { 7831da177e4SLinus Torvalds int ret = 1; 7841da177e4SLinus Torvalds 7851da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 7861da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 7871da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 788252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 7891da177e4SLinus Torvalds 7901da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 7911da177e4SLinus Torvalds while (!list_empty(list)) { 7921da177e4SLinus Torvalds struct buffer_head *bh = BH_ENTRY(list->next); 7931da177e4SLinus Torvalds if (buffer_dirty(bh)) { 7941da177e4SLinus Torvalds ret = 0; 7951da177e4SLinus Torvalds break; 7961da177e4SLinus Torvalds } 7971da177e4SLinus Torvalds __remove_assoc_queue(bh); 7981da177e4SLinus Torvalds } 7991da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8001da177e4SLinus Torvalds } 8011da177e4SLinus Torvalds return ret; 8021da177e4SLinus Torvalds } 8031da177e4SLinus Torvalds 8041da177e4SLinus Torvalds /* 8051da177e4SLinus Torvalds * Create the appropriate buffers when given a page for data area and 8061da177e4SLinus Torvalds * the size of each buffer.. Use the bh->b_this_page linked list to 8071da177e4SLinus Torvalds * follow the buffers created. Return NULL if unable to create more 8081da177e4SLinus Torvalds * buffers. 8091da177e4SLinus Torvalds * 8101da177e4SLinus Torvalds * The retry flag is used to differentiate async IO (paging, swapping) 8111da177e4SLinus Torvalds * which may not fail from ordinary buffer allocations. 8121da177e4SLinus Torvalds */ 8131da177e4SLinus Torvalds struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, 814640ab98fSJens Axboe bool retry) 8151da177e4SLinus Torvalds { 8161da177e4SLinus Torvalds struct buffer_head *bh, *head; 817f745c6f5SShakeel Butt gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT; 8181da177e4SLinus Torvalds long offset; 819f745c6f5SShakeel Butt struct mem_cgroup *memcg; 8201da177e4SLinus Torvalds 821640ab98fSJens Axboe if (retry) 822640ab98fSJens Axboe gfp |= __GFP_NOFAIL; 823640ab98fSJens Axboe 824f745c6f5SShakeel Butt memcg = get_mem_cgroup_from_page(page); 825f745c6f5SShakeel Butt memalloc_use_memcg(memcg); 826f745c6f5SShakeel Butt 8271da177e4SLinus Torvalds head = NULL; 8281da177e4SLinus Torvalds offset = PAGE_SIZE; 8291da177e4SLinus Torvalds while ((offset -= size) >= 0) { 830640ab98fSJens Axboe bh = alloc_buffer_head(gfp); 8311da177e4SLinus Torvalds if (!bh) 8321da177e4SLinus Torvalds goto no_grow; 8331da177e4SLinus Torvalds 8341da177e4SLinus Torvalds bh->b_this_page = head; 8351da177e4SLinus Torvalds bh->b_blocknr = -1; 8361da177e4SLinus Torvalds head = bh; 8371da177e4SLinus Torvalds 8381da177e4SLinus Torvalds bh->b_size = size; 8391da177e4SLinus Torvalds 8401da177e4SLinus Torvalds /* Link the buffer to its page */ 8411da177e4SLinus Torvalds set_bh_page(bh, page, offset); 8421da177e4SLinus Torvalds } 843f745c6f5SShakeel Butt out: 844f745c6f5SShakeel Butt memalloc_unuse_memcg(); 845f745c6f5SShakeel Butt mem_cgroup_put(memcg); 8461da177e4SLinus Torvalds return head; 8471da177e4SLinus Torvalds /* 8481da177e4SLinus Torvalds * In case anything failed, we just free everything we got. 8491da177e4SLinus Torvalds */ 8501da177e4SLinus Torvalds no_grow: 8511da177e4SLinus Torvalds if (head) { 8521da177e4SLinus Torvalds do { 8531da177e4SLinus Torvalds bh = head; 8541da177e4SLinus Torvalds head = head->b_this_page; 8551da177e4SLinus Torvalds free_buffer_head(bh); 8561da177e4SLinus Torvalds } while (head); 8571da177e4SLinus Torvalds } 8581da177e4SLinus Torvalds 859f745c6f5SShakeel Butt goto out; 8601da177e4SLinus Torvalds } 8611da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(alloc_page_buffers); 8621da177e4SLinus Torvalds 8631da177e4SLinus Torvalds static inline void 8641da177e4SLinus Torvalds link_dev_buffers(struct page *page, struct buffer_head *head) 8651da177e4SLinus Torvalds { 8661da177e4SLinus Torvalds struct buffer_head *bh, *tail; 8671da177e4SLinus Torvalds 8681da177e4SLinus Torvalds bh = head; 8691da177e4SLinus Torvalds do { 8701da177e4SLinus Torvalds tail = bh; 8711da177e4SLinus Torvalds bh = bh->b_this_page; 8721da177e4SLinus Torvalds } while (bh); 8731da177e4SLinus Torvalds tail->b_this_page = head; 8741da177e4SLinus Torvalds attach_page_buffers(page, head); 8751da177e4SLinus Torvalds } 8761da177e4SLinus Torvalds 877bbec0270SLinus Torvalds static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) 878bbec0270SLinus Torvalds { 879bbec0270SLinus Torvalds sector_t retval = ~((sector_t)0); 880bbec0270SLinus Torvalds loff_t sz = i_size_read(bdev->bd_inode); 881bbec0270SLinus Torvalds 882bbec0270SLinus Torvalds if (sz) { 883bbec0270SLinus Torvalds unsigned int sizebits = blksize_bits(size); 884bbec0270SLinus Torvalds retval = (sz >> sizebits); 885bbec0270SLinus Torvalds } 886bbec0270SLinus Torvalds return retval; 887bbec0270SLinus Torvalds } 888bbec0270SLinus Torvalds 8891da177e4SLinus Torvalds /* 8901da177e4SLinus Torvalds * Initialise the state of a blockdev page's buffers. 8911da177e4SLinus Torvalds */ 892676ce6d5SHugh Dickins static sector_t 8931da177e4SLinus Torvalds init_page_buffers(struct page *page, struct block_device *bdev, 8941da177e4SLinus Torvalds sector_t block, int size) 8951da177e4SLinus Torvalds { 8961da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 8971da177e4SLinus Torvalds struct buffer_head *bh = head; 8981da177e4SLinus Torvalds int uptodate = PageUptodate(page); 899bbec0270SLinus Torvalds sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size); 9001da177e4SLinus Torvalds 9011da177e4SLinus Torvalds do { 9021da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 90301950a34SEric Biggers bh->b_end_io = NULL; 90401950a34SEric Biggers bh->b_private = NULL; 9051da177e4SLinus Torvalds bh->b_bdev = bdev; 9061da177e4SLinus Torvalds bh->b_blocknr = block; 9071da177e4SLinus Torvalds if (uptodate) 9081da177e4SLinus Torvalds set_buffer_uptodate(bh); 909080399aaSJeff Moyer if (block < end_block) 9101da177e4SLinus Torvalds set_buffer_mapped(bh); 9111da177e4SLinus Torvalds } 9121da177e4SLinus Torvalds block++; 9131da177e4SLinus Torvalds bh = bh->b_this_page; 9141da177e4SLinus Torvalds } while (bh != head); 915676ce6d5SHugh Dickins 916676ce6d5SHugh Dickins /* 917676ce6d5SHugh Dickins * Caller needs to validate requested block against end of device. 918676ce6d5SHugh Dickins */ 919676ce6d5SHugh Dickins return end_block; 9201da177e4SLinus Torvalds } 9211da177e4SLinus Torvalds 9221da177e4SLinus Torvalds /* 9231da177e4SLinus Torvalds * Create the page-cache page that contains the requested block. 9241da177e4SLinus Torvalds * 925676ce6d5SHugh Dickins * This is used purely for blockdev mappings. 9261da177e4SLinus Torvalds */ 927676ce6d5SHugh Dickins static int 9281da177e4SLinus Torvalds grow_dev_page(struct block_device *bdev, sector_t block, 9293b5e6454SGioh Kim pgoff_t index, int size, int sizebits, gfp_t gfp) 9301da177e4SLinus Torvalds { 9311da177e4SLinus Torvalds struct inode *inode = bdev->bd_inode; 9321da177e4SLinus Torvalds struct page *page; 9331da177e4SLinus Torvalds struct buffer_head *bh; 934676ce6d5SHugh Dickins sector_t end_block; 935676ce6d5SHugh Dickins int ret = 0; /* Will call free_more_memory() */ 93684235de3SJohannes Weiner gfp_t gfp_mask; 9371da177e4SLinus Torvalds 938c62d2555SMichal Hocko gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp; 9393b5e6454SGioh Kim 94084235de3SJohannes Weiner /* 94184235de3SJohannes Weiner * XXX: __getblk_slow() can not really deal with failure and 94284235de3SJohannes Weiner * will endlessly loop on improvised global reclaim. Prefer 94384235de3SJohannes Weiner * looping in the allocator rather than here, at least that 94484235de3SJohannes Weiner * code knows what it's doing. 94584235de3SJohannes Weiner */ 94684235de3SJohannes Weiner gfp_mask |= __GFP_NOFAIL; 94784235de3SJohannes Weiner 94884235de3SJohannes Weiner page = find_or_create_page(inode->i_mapping, index, gfp_mask); 9491da177e4SLinus Torvalds 950e827f923SEric Sesterhenn BUG_ON(!PageLocked(page)); 9511da177e4SLinus Torvalds 9521da177e4SLinus Torvalds if (page_has_buffers(page)) { 9531da177e4SLinus Torvalds bh = page_buffers(page); 9541da177e4SLinus Torvalds if (bh->b_size == size) { 955676ce6d5SHugh Dickins end_block = init_page_buffers(page, bdev, 956f2d5a944SAnton Altaparmakov (sector_t)index << sizebits, 957f2d5a944SAnton Altaparmakov size); 958676ce6d5SHugh Dickins goto done; 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds if (!try_to_free_buffers(page)) 9611da177e4SLinus Torvalds goto failed; 9621da177e4SLinus Torvalds } 9631da177e4SLinus Torvalds 9641da177e4SLinus Torvalds /* 9651da177e4SLinus Torvalds * Allocate some buffers for this page 9661da177e4SLinus Torvalds */ 96794dc24c0SJens Axboe bh = alloc_page_buffers(page, size, true); 9681da177e4SLinus Torvalds 9691da177e4SLinus Torvalds /* 9701da177e4SLinus Torvalds * Link the page to the buffers and initialise them. Take the 9711da177e4SLinus Torvalds * lock to be atomic wrt __find_get_block(), which does not 9721da177e4SLinus Torvalds * run under the page lock. 9731da177e4SLinus Torvalds */ 9741da177e4SLinus Torvalds spin_lock(&inode->i_mapping->private_lock); 9751da177e4SLinus Torvalds link_dev_buffers(page, bh); 976f2d5a944SAnton Altaparmakov end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits, 977f2d5a944SAnton Altaparmakov size); 9781da177e4SLinus Torvalds spin_unlock(&inode->i_mapping->private_lock); 979676ce6d5SHugh Dickins done: 980676ce6d5SHugh Dickins ret = (block < end_block) ? 1 : -ENXIO; 9811da177e4SLinus Torvalds failed: 9821da177e4SLinus Torvalds unlock_page(page); 98309cbfeafSKirill A. Shutemov put_page(page); 984676ce6d5SHugh Dickins return ret; 9851da177e4SLinus Torvalds } 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds /* 9881da177e4SLinus Torvalds * Create buffers for the specified block device block's page. If 9891da177e4SLinus Torvalds * that page was dirty, the buffers are set dirty also. 9901da177e4SLinus Torvalds */ 991858119e1SArjan van de Ven static int 9923b5e6454SGioh Kim grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp) 9931da177e4SLinus Torvalds { 9941da177e4SLinus Torvalds pgoff_t index; 9951da177e4SLinus Torvalds int sizebits; 9961da177e4SLinus Torvalds 9971da177e4SLinus Torvalds sizebits = -1; 9981da177e4SLinus Torvalds do { 9991da177e4SLinus Torvalds sizebits++; 10001da177e4SLinus Torvalds } while ((size << sizebits) < PAGE_SIZE); 10011da177e4SLinus Torvalds 10021da177e4SLinus Torvalds index = block >> sizebits; 10031da177e4SLinus Torvalds 1004e5657933SAndrew Morton /* 1005e5657933SAndrew Morton * Check for a block which wants to lie outside our maximum possible 1006e5657933SAndrew Morton * pagecache index. (this comparison is done using sector_t types). 1007e5657933SAndrew Morton */ 1008e5657933SAndrew Morton if (unlikely(index != block >> sizebits)) { 1009e5657933SAndrew Morton printk(KERN_ERR "%s: requested out-of-range block %llu for " 1010a1c6f057SDmitry Monakhov "device %pg\n", 10118e24eea7SHarvey Harrison __func__, (unsigned long long)block, 1012a1c6f057SDmitry Monakhov bdev); 1013e5657933SAndrew Morton return -EIO; 1014e5657933SAndrew Morton } 1015676ce6d5SHugh Dickins 10161da177e4SLinus Torvalds /* Create a page with the proper size buffers.. */ 10173b5e6454SGioh Kim return grow_dev_page(bdev, block, index, size, sizebits, gfp); 10181da177e4SLinus Torvalds } 10191da177e4SLinus Torvalds 10200026ba40SEric Biggers static struct buffer_head * 10213b5e6454SGioh Kim __getblk_slow(struct block_device *bdev, sector_t block, 10223b5e6454SGioh Kim unsigned size, gfp_t gfp) 10231da177e4SLinus Torvalds { 10241da177e4SLinus Torvalds /* Size must be multiple of hard sectorsize */ 1025e1defc4fSMartin K. Petersen if (unlikely(size & (bdev_logical_block_size(bdev)-1) || 10261da177e4SLinus Torvalds (size < 512 || size > PAGE_SIZE))) { 10271da177e4SLinus Torvalds printk(KERN_ERR "getblk(): invalid block size %d requested\n", 10281da177e4SLinus Torvalds size); 1029e1defc4fSMartin K. Petersen printk(KERN_ERR "logical block size: %d\n", 1030e1defc4fSMartin K. Petersen bdev_logical_block_size(bdev)); 10311da177e4SLinus Torvalds 10321da177e4SLinus Torvalds dump_stack(); 10331da177e4SLinus Torvalds return NULL; 10341da177e4SLinus Torvalds } 10351da177e4SLinus Torvalds 1036676ce6d5SHugh Dickins for (;;) { 1037676ce6d5SHugh Dickins struct buffer_head *bh; 1038676ce6d5SHugh Dickins int ret; 1039676ce6d5SHugh Dickins 10401da177e4SLinus Torvalds bh = __find_get_block(bdev, block, size); 10411da177e4SLinus Torvalds if (bh) 10421da177e4SLinus Torvalds return bh; 10431da177e4SLinus Torvalds 10443b5e6454SGioh Kim ret = grow_buffers(bdev, block, size, gfp); 1045676ce6d5SHugh Dickins if (ret < 0) 104691f68c89SJeff Moyer return NULL; 1047676ce6d5SHugh Dickins } 10481da177e4SLinus Torvalds } 10491da177e4SLinus Torvalds 10501da177e4SLinus Torvalds /* 10511da177e4SLinus Torvalds * The relationship between dirty buffers and dirty pages: 10521da177e4SLinus Torvalds * 10531da177e4SLinus Torvalds * Whenever a page has any dirty buffers, the page's dirty bit is set, and 1054ec82e1c1SMatthew Wilcox * the page is tagged dirty in the page cache. 10551da177e4SLinus Torvalds * 10561da177e4SLinus Torvalds * At all times, the dirtiness of the buffers represents the dirtiness of 10571da177e4SLinus Torvalds * subsections of the page. If the page has buffers, the page dirty bit is 10581da177e4SLinus Torvalds * merely a hint about the true dirty state. 10591da177e4SLinus Torvalds * 10601da177e4SLinus Torvalds * When a page is set dirty in its entirety, all its buffers are marked dirty 10611da177e4SLinus Torvalds * (if the page has buffers). 10621da177e4SLinus Torvalds * 10631da177e4SLinus Torvalds * When a buffer is marked dirty, its page is dirtied, but the page's other 10641da177e4SLinus Torvalds * buffers are not. 10651da177e4SLinus Torvalds * 10661da177e4SLinus Torvalds * Also. When blockdev buffers are explicitly read with bread(), they 10671da177e4SLinus Torvalds * individually become uptodate. But their backing page remains not 10681da177e4SLinus Torvalds * uptodate - even if all of its buffers are uptodate. A subsequent 10691da177e4SLinus Torvalds * block_read_full_page() against that page will discover all the uptodate 10701da177e4SLinus Torvalds * buffers, will set the page uptodate and will perform no I/O. 10711da177e4SLinus Torvalds */ 10721da177e4SLinus Torvalds 10731da177e4SLinus Torvalds /** 10741da177e4SLinus Torvalds * mark_buffer_dirty - mark a buffer_head as needing writeout 107567be2dd1SMartin Waitz * @bh: the buffer_head to mark dirty 10761da177e4SLinus Torvalds * 1077ec82e1c1SMatthew Wilcox * mark_buffer_dirty() will set the dirty bit against the buffer, then set 1078ec82e1c1SMatthew Wilcox * its backing page dirty, then tag the page as dirty in the page cache 1079ec82e1c1SMatthew Wilcox * and then attach the address_space's inode to its superblock's dirty 10801da177e4SLinus Torvalds * inode list. 10811da177e4SLinus Torvalds * 10821da177e4SLinus Torvalds * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock, 1083b93b0163SMatthew Wilcox * i_pages lock and mapping->host->i_lock. 10841da177e4SLinus Torvalds */ 1085fc9b52cdSHarvey Harrison void mark_buffer_dirty(struct buffer_head *bh) 10861da177e4SLinus Torvalds { 1087787d2214SNick Piggin WARN_ON_ONCE(!buffer_uptodate(bh)); 10881be62dc1SLinus Torvalds 10895305cb83STejun Heo trace_block_dirty_buffer(bh); 10905305cb83STejun Heo 10911be62dc1SLinus Torvalds /* 10921be62dc1SLinus Torvalds * Very *carefully* optimize the it-is-already-dirty case. 10931be62dc1SLinus Torvalds * 10941be62dc1SLinus Torvalds * Don't let the final "is it dirty" escape to before we 10951be62dc1SLinus Torvalds * perhaps modified the buffer. 10961be62dc1SLinus Torvalds */ 10971be62dc1SLinus Torvalds if (buffer_dirty(bh)) { 10981be62dc1SLinus Torvalds smp_mb(); 10991be62dc1SLinus Torvalds if (buffer_dirty(bh)) 11001be62dc1SLinus Torvalds return; 11011be62dc1SLinus Torvalds } 11021be62dc1SLinus Torvalds 1103a8e7d49aSLinus Torvalds if (!test_set_buffer_dirty(bh)) { 1104a8e7d49aSLinus Torvalds struct page *page = bh->b_page; 1105c4843a75SGreg Thelen struct address_space *mapping = NULL; 1106c4843a75SGreg Thelen 110762cccb8cSJohannes Weiner lock_page_memcg(page); 11088e9d78edSLinus Torvalds if (!TestSetPageDirty(page)) { 1109c4843a75SGreg Thelen mapping = page_mapping(page); 11108e9d78edSLinus Torvalds if (mapping) 111162cccb8cSJohannes Weiner __set_page_dirty(page, mapping, 0); 11128e9d78edSLinus Torvalds } 111362cccb8cSJohannes Weiner unlock_page_memcg(page); 1114c4843a75SGreg Thelen if (mapping) 1115c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 1116a8e7d49aSLinus Torvalds } 11171da177e4SLinus Torvalds } 11181fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(mark_buffer_dirty); 11191da177e4SLinus Torvalds 112087354e5dSJeff Layton void mark_buffer_write_io_error(struct buffer_head *bh) 112187354e5dSJeff Layton { 112287354e5dSJeff Layton set_buffer_write_io_error(bh); 112387354e5dSJeff Layton /* FIXME: do we need to set this in both places? */ 112487354e5dSJeff Layton if (bh->b_page && bh->b_page->mapping) 112587354e5dSJeff Layton mapping_set_error(bh->b_page->mapping, -EIO); 112687354e5dSJeff Layton if (bh->b_assoc_map) 112787354e5dSJeff Layton mapping_set_error(bh->b_assoc_map, -EIO); 112887354e5dSJeff Layton } 112987354e5dSJeff Layton EXPORT_SYMBOL(mark_buffer_write_io_error); 113087354e5dSJeff Layton 11311da177e4SLinus Torvalds /* 11321da177e4SLinus Torvalds * Decrement a buffer_head's reference count. If all buffers against a page 11331da177e4SLinus Torvalds * have zero reference count, are clean and unlocked, and if the page is clean 11341da177e4SLinus Torvalds * and unlocked then try_to_free_buffers() may strip the buffers from the page 11351da177e4SLinus Torvalds * in preparation for freeing it (sometimes, rarely, buffers are removed from 11361da177e4SLinus Torvalds * a page but it ends up not being freed, and buffers may later be reattached). 11371da177e4SLinus Torvalds */ 11381da177e4SLinus Torvalds void __brelse(struct buffer_head * buf) 11391da177e4SLinus Torvalds { 11401da177e4SLinus Torvalds if (atomic_read(&buf->b_count)) { 11411da177e4SLinus Torvalds put_bh(buf); 11421da177e4SLinus Torvalds return; 11431da177e4SLinus Torvalds } 11445c752ad9SArjan van de Ven WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); 11451da177e4SLinus Torvalds } 11461fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__brelse); 11471da177e4SLinus Torvalds 11481da177e4SLinus Torvalds /* 11491da177e4SLinus Torvalds * bforget() is like brelse(), except it discards any 11501da177e4SLinus Torvalds * potentially dirty data. 11511da177e4SLinus Torvalds */ 11521da177e4SLinus Torvalds void __bforget(struct buffer_head *bh) 11531da177e4SLinus Torvalds { 11541da177e4SLinus Torvalds clear_buffer_dirty(bh); 1155535ee2fbSJan Kara if (bh->b_assoc_map) { 11561da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 11571da177e4SLinus Torvalds 11581da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 11591da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 116058ff407bSJan Kara bh->b_assoc_map = NULL; 11611da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 11621da177e4SLinus Torvalds } 11631da177e4SLinus Torvalds __brelse(bh); 11641da177e4SLinus Torvalds } 11651fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__bforget); 11661da177e4SLinus Torvalds 11671da177e4SLinus Torvalds static struct buffer_head *__bread_slow(struct buffer_head *bh) 11681da177e4SLinus Torvalds { 11691da177e4SLinus Torvalds lock_buffer(bh); 11701da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 11711da177e4SLinus Torvalds unlock_buffer(bh); 11721da177e4SLinus Torvalds return bh; 11731da177e4SLinus Torvalds } else { 11741da177e4SLinus Torvalds get_bh(bh); 11751da177e4SLinus Torvalds bh->b_end_io = end_buffer_read_sync; 11762a222ca9SMike Christie submit_bh(REQ_OP_READ, 0, bh); 11771da177e4SLinus Torvalds wait_on_buffer(bh); 11781da177e4SLinus Torvalds if (buffer_uptodate(bh)) 11791da177e4SLinus Torvalds return bh; 11801da177e4SLinus Torvalds } 11811da177e4SLinus Torvalds brelse(bh); 11821da177e4SLinus Torvalds return NULL; 11831da177e4SLinus Torvalds } 11841da177e4SLinus Torvalds 11851da177e4SLinus Torvalds /* 11861da177e4SLinus Torvalds * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). 11871da177e4SLinus Torvalds * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their 11881da177e4SLinus Torvalds * refcount elevated by one when they're in an LRU. A buffer can only appear 11891da177e4SLinus Torvalds * once in a particular CPU's LRU. A single buffer can be present in multiple 11901da177e4SLinus Torvalds * CPU's LRUs at the same time. 11911da177e4SLinus Torvalds * 11921da177e4SLinus Torvalds * This is a transparent caching front-end to sb_bread(), sb_getblk() and 11931da177e4SLinus Torvalds * sb_find_get_block(). 11941da177e4SLinus Torvalds * 11951da177e4SLinus Torvalds * The LRUs themselves only need locking against invalidate_bh_lrus. We use 11961da177e4SLinus Torvalds * a local interrupt disable for that. 11971da177e4SLinus Torvalds */ 11981da177e4SLinus Torvalds 119986cf78d7SSebastien Buisson #define BH_LRU_SIZE 16 12001da177e4SLinus Torvalds 12011da177e4SLinus Torvalds struct bh_lru { 12021da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12031da177e4SLinus Torvalds }; 12041da177e4SLinus Torvalds 12051da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; 12061da177e4SLinus Torvalds 12071da177e4SLinus Torvalds #ifdef CONFIG_SMP 12081da177e4SLinus Torvalds #define bh_lru_lock() local_irq_disable() 12091da177e4SLinus Torvalds #define bh_lru_unlock() local_irq_enable() 12101da177e4SLinus Torvalds #else 12111da177e4SLinus Torvalds #define bh_lru_lock() preempt_disable() 12121da177e4SLinus Torvalds #define bh_lru_unlock() preempt_enable() 12131da177e4SLinus Torvalds #endif 12141da177e4SLinus Torvalds 12151da177e4SLinus Torvalds static inline void check_irqs_on(void) 12161da177e4SLinus Torvalds { 12171da177e4SLinus Torvalds #ifdef irqs_disabled 12181da177e4SLinus Torvalds BUG_ON(irqs_disabled()); 12191da177e4SLinus Torvalds #endif 12201da177e4SLinus Torvalds } 12211da177e4SLinus Torvalds 12221da177e4SLinus Torvalds /* 1223241f01fbSEric Biggers * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is 1224241f01fbSEric Biggers * inserted at the front, and the buffer_head at the back if any is evicted. 1225241f01fbSEric Biggers * Or, if already in the LRU it is moved to the front. 12261da177e4SLinus Torvalds */ 12271da177e4SLinus Torvalds static void bh_lru_install(struct buffer_head *bh) 12281da177e4SLinus Torvalds { 1229241f01fbSEric Biggers struct buffer_head *evictee = bh; 1230241f01fbSEric Biggers struct bh_lru *b; 1231241f01fbSEric Biggers int i; 12321da177e4SLinus Torvalds 12331da177e4SLinus Torvalds check_irqs_on(); 12341da177e4SLinus Torvalds bh_lru_lock(); 1235241f01fbSEric Biggers 1236241f01fbSEric Biggers b = this_cpu_ptr(&bh_lrus); 1237241f01fbSEric Biggers for (i = 0; i < BH_LRU_SIZE; i++) { 1238241f01fbSEric Biggers swap(evictee, b->bhs[i]); 1239241f01fbSEric Biggers if (evictee == bh) { 1240241f01fbSEric Biggers bh_lru_unlock(); 1241241f01fbSEric Biggers return; 1242241f01fbSEric Biggers } 1243241f01fbSEric Biggers } 12441da177e4SLinus Torvalds 12451da177e4SLinus Torvalds get_bh(bh); 12461da177e4SLinus Torvalds bh_lru_unlock(); 1247241f01fbSEric Biggers brelse(evictee); 12481da177e4SLinus Torvalds } 12491da177e4SLinus Torvalds 12501da177e4SLinus Torvalds /* 12511da177e4SLinus Torvalds * Look up the bh in this cpu's LRU. If it's there, move it to the head. 12521da177e4SLinus Torvalds */ 1253858119e1SArjan van de Ven static struct buffer_head * 12543991d3bdSTomasz Kvarsin lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) 12551da177e4SLinus Torvalds { 12561da177e4SLinus Torvalds struct buffer_head *ret = NULL; 12573991d3bdSTomasz Kvarsin unsigned int i; 12581da177e4SLinus Torvalds 12591da177e4SLinus Torvalds check_irqs_on(); 12601da177e4SLinus Torvalds bh_lru_lock(); 12611da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 1262c7b92516SChristoph Lameter struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); 12631da177e4SLinus Torvalds 12649470dd5dSZach Brown if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && 12659470dd5dSZach Brown bh->b_size == size) { 12661da177e4SLinus Torvalds if (i) { 12671da177e4SLinus Torvalds while (i) { 1268c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[i], 1269c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[i - 1])); 12701da177e4SLinus Torvalds i--; 12711da177e4SLinus Torvalds } 1272c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[0], bh); 12731da177e4SLinus Torvalds } 12741da177e4SLinus Torvalds get_bh(bh); 12751da177e4SLinus Torvalds ret = bh; 12761da177e4SLinus Torvalds break; 12771da177e4SLinus Torvalds } 12781da177e4SLinus Torvalds } 12791da177e4SLinus Torvalds bh_lru_unlock(); 12801da177e4SLinus Torvalds return ret; 12811da177e4SLinus Torvalds } 12821da177e4SLinus Torvalds 12831da177e4SLinus Torvalds /* 12841da177e4SLinus Torvalds * Perform a pagecache lookup for the matching buffer. If it's there, refresh 12851da177e4SLinus Torvalds * it in the LRU and mark it as accessed. If it is not present then return 12861da177e4SLinus Torvalds * NULL 12871da177e4SLinus Torvalds */ 12881da177e4SLinus Torvalds struct buffer_head * 12893991d3bdSTomasz Kvarsin __find_get_block(struct block_device *bdev, sector_t block, unsigned size) 12901da177e4SLinus Torvalds { 12911da177e4SLinus Torvalds struct buffer_head *bh = lookup_bh_lru(bdev, block, size); 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds if (bh == NULL) { 12942457aec6SMel Gorman /* __find_get_block_slow will mark the page accessed */ 1295385fd4c5SCoywolf Qi Hunt bh = __find_get_block_slow(bdev, block); 12961da177e4SLinus Torvalds if (bh) 12971da177e4SLinus Torvalds bh_lru_install(bh); 12982457aec6SMel Gorman } else 12991da177e4SLinus Torvalds touch_buffer(bh); 13002457aec6SMel Gorman 13011da177e4SLinus Torvalds return bh; 13021da177e4SLinus Torvalds } 13031da177e4SLinus Torvalds EXPORT_SYMBOL(__find_get_block); 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds /* 13063b5e6454SGioh Kim * __getblk_gfp() will locate (and, if necessary, create) the buffer_head 13071da177e4SLinus Torvalds * which corresponds to the passed block_device, block and size. The 13081da177e4SLinus Torvalds * returned buffer has its reference count incremented. 13091da177e4SLinus Torvalds * 13103b5e6454SGioh Kim * __getblk_gfp() will lock up the machine if grow_dev_page's 13113b5e6454SGioh Kim * try_to_free_buffers() attempt is failing. FIXME, perhaps? 13121da177e4SLinus Torvalds */ 13131da177e4SLinus Torvalds struct buffer_head * 13143b5e6454SGioh Kim __getblk_gfp(struct block_device *bdev, sector_t block, 13153b5e6454SGioh Kim unsigned size, gfp_t gfp) 13161da177e4SLinus Torvalds { 13171da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, block, size); 13181da177e4SLinus Torvalds 13191da177e4SLinus Torvalds might_sleep(); 13201da177e4SLinus Torvalds if (bh == NULL) 13213b5e6454SGioh Kim bh = __getblk_slow(bdev, block, size, gfp); 13221da177e4SLinus Torvalds return bh; 13231da177e4SLinus Torvalds } 13243b5e6454SGioh Kim EXPORT_SYMBOL(__getblk_gfp); 13251da177e4SLinus Torvalds 13261da177e4SLinus Torvalds /* 13271da177e4SLinus Torvalds * Do async read-ahead on a buffer.. 13281da177e4SLinus Torvalds */ 13293991d3bdSTomasz Kvarsin void __breadahead(struct block_device *bdev, sector_t block, unsigned size) 13301da177e4SLinus Torvalds { 13311da177e4SLinus Torvalds struct buffer_head *bh = __getblk(bdev, block, size); 1332a3e713b5SAndrew Morton if (likely(bh)) { 133370246286SChristoph Hellwig ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh); 13341da177e4SLinus Torvalds brelse(bh); 13351da177e4SLinus Torvalds } 1336a3e713b5SAndrew Morton } 13371da177e4SLinus Torvalds EXPORT_SYMBOL(__breadahead); 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds /** 13403b5e6454SGioh Kim * __bread_gfp() - reads a specified block and returns the bh 134167be2dd1SMartin Waitz * @bdev: the block_device to read from 13421da177e4SLinus Torvalds * @block: number of block 13431da177e4SLinus Torvalds * @size: size (in bytes) to read 13443b5e6454SGioh Kim * @gfp: page allocation flag 13451da177e4SLinus Torvalds * 13461da177e4SLinus Torvalds * Reads a specified block, and returns buffer head that contains it. 13473b5e6454SGioh Kim * The page cache can be allocated from non-movable area 13483b5e6454SGioh Kim * not to prevent page migration if you set gfp to zero. 13491da177e4SLinus Torvalds * It returns NULL if the block was unreadable. 13501da177e4SLinus Torvalds */ 13511da177e4SLinus Torvalds struct buffer_head * 13523b5e6454SGioh Kim __bread_gfp(struct block_device *bdev, sector_t block, 13533b5e6454SGioh Kim unsigned size, gfp_t gfp) 13541da177e4SLinus Torvalds { 13553b5e6454SGioh Kim struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); 13561da177e4SLinus Torvalds 1357a3e713b5SAndrew Morton if (likely(bh) && !buffer_uptodate(bh)) 13581da177e4SLinus Torvalds bh = __bread_slow(bh); 13591da177e4SLinus Torvalds return bh; 13601da177e4SLinus Torvalds } 13613b5e6454SGioh Kim EXPORT_SYMBOL(__bread_gfp); 13621da177e4SLinus Torvalds 13631da177e4SLinus Torvalds /* 13641da177e4SLinus Torvalds * invalidate_bh_lrus() is called rarely - but not only at unmount. 13651da177e4SLinus Torvalds * This doesn't race because it runs in each cpu either in irq 13661da177e4SLinus Torvalds * or with preempt disabled. 13671da177e4SLinus Torvalds */ 13681da177e4SLinus Torvalds static void invalidate_bh_lru(void *arg) 13691da177e4SLinus Torvalds { 13701da177e4SLinus Torvalds struct bh_lru *b = &get_cpu_var(bh_lrus); 13711da177e4SLinus Torvalds int i; 13721da177e4SLinus Torvalds 13731da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 13741da177e4SLinus Torvalds brelse(b->bhs[i]); 13751da177e4SLinus Torvalds b->bhs[i] = NULL; 13761da177e4SLinus Torvalds } 13771da177e4SLinus Torvalds put_cpu_var(bh_lrus); 13781da177e4SLinus Torvalds } 13791da177e4SLinus Torvalds 138042be35d0SGilad Ben-Yossef static bool has_bh_in_lru(int cpu, void *dummy) 138142be35d0SGilad Ben-Yossef { 138242be35d0SGilad Ben-Yossef struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); 138342be35d0SGilad Ben-Yossef int i; 138442be35d0SGilad Ben-Yossef 138542be35d0SGilad Ben-Yossef for (i = 0; i < BH_LRU_SIZE; i++) { 138642be35d0SGilad Ben-Yossef if (b->bhs[i]) 138742be35d0SGilad Ben-Yossef return 1; 138842be35d0SGilad Ben-Yossef } 138942be35d0SGilad Ben-Yossef 139042be35d0SGilad Ben-Yossef return 0; 139142be35d0SGilad Ben-Yossef } 139242be35d0SGilad Ben-Yossef 1393f9a14399SPeter Zijlstra void invalidate_bh_lrus(void) 13941da177e4SLinus Torvalds { 139542be35d0SGilad Ben-Yossef on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL); 13961da177e4SLinus Torvalds } 13979db5579bSNick Piggin EXPORT_SYMBOL_GPL(invalidate_bh_lrus); 13981da177e4SLinus Torvalds 13991da177e4SLinus Torvalds void set_bh_page(struct buffer_head *bh, 14001da177e4SLinus Torvalds struct page *page, unsigned long offset) 14011da177e4SLinus Torvalds { 14021da177e4SLinus Torvalds bh->b_page = page; 1403e827f923SEric Sesterhenn BUG_ON(offset >= PAGE_SIZE); 14041da177e4SLinus Torvalds if (PageHighMem(page)) 14051da177e4SLinus Torvalds /* 14061da177e4SLinus Torvalds * This catches illegal uses and preserves the offset: 14071da177e4SLinus Torvalds */ 14081da177e4SLinus Torvalds bh->b_data = (char *)(0 + offset); 14091da177e4SLinus Torvalds else 14101da177e4SLinus Torvalds bh->b_data = page_address(page) + offset; 14111da177e4SLinus Torvalds } 14121da177e4SLinus Torvalds EXPORT_SYMBOL(set_bh_page); 14131da177e4SLinus Torvalds 14141da177e4SLinus Torvalds /* 14151da177e4SLinus Torvalds * Called when truncating a buffer on a page completely. 14161da177e4SLinus Torvalds */ 1417e7470ee8SMel Gorman 1418e7470ee8SMel Gorman /* Bits that are cleared during an invalidate */ 1419e7470ee8SMel Gorman #define BUFFER_FLAGS_DISCARD \ 1420e7470ee8SMel Gorman (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ 1421e7470ee8SMel Gorman 1 << BH_Delay | 1 << BH_Unwritten) 1422e7470ee8SMel Gorman 1423858119e1SArjan van de Ven static void discard_buffer(struct buffer_head * bh) 14241da177e4SLinus Torvalds { 1425e7470ee8SMel Gorman unsigned long b_state, b_state_old; 1426e7470ee8SMel Gorman 14271da177e4SLinus Torvalds lock_buffer(bh); 14281da177e4SLinus Torvalds clear_buffer_dirty(bh); 14291da177e4SLinus Torvalds bh->b_bdev = NULL; 1430e7470ee8SMel Gorman b_state = bh->b_state; 1431e7470ee8SMel Gorman for (;;) { 1432e7470ee8SMel Gorman b_state_old = cmpxchg(&bh->b_state, b_state, 1433e7470ee8SMel Gorman (b_state & ~BUFFER_FLAGS_DISCARD)); 1434e7470ee8SMel Gorman if (b_state_old == b_state) 1435e7470ee8SMel Gorman break; 1436e7470ee8SMel Gorman b_state = b_state_old; 1437e7470ee8SMel Gorman } 14381da177e4SLinus Torvalds unlock_buffer(bh); 14391da177e4SLinus Torvalds } 14401da177e4SLinus Torvalds 14411da177e4SLinus Torvalds /** 1442814e1d25SWang Sheng-Hui * block_invalidatepage - invalidate part or all of a buffer-backed page 14431da177e4SLinus Torvalds * 14441da177e4SLinus Torvalds * @page: the page which is affected 1445d47992f8SLukas Czerner * @offset: start of the range to invalidate 1446d47992f8SLukas Czerner * @length: length of the range to invalidate 14471da177e4SLinus Torvalds * 14481da177e4SLinus Torvalds * block_invalidatepage() is called when all or part of the page has become 14491da177e4SLinus Torvalds * invalidated by a truncate operation. 14501da177e4SLinus Torvalds * 14511da177e4SLinus Torvalds * block_invalidatepage() does not have to release all buffers, but it must 14521da177e4SLinus Torvalds * ensure that no dirty buffer is left outside @offset and that no I/O 14531da177e4SLinus Torvalds * is underway against any of the blocks which are outside the truncation 14541da177e4SLinus Torvalds * point. Because the caller is about to free (and possibly reuse) those 14551da177e4SLinus Torvalds * blocks on-disk. 14561da177e4SLinus Torvalds */ 1457d47992f8SLukas Czerner void block_invalidatepage(struct page *page, unsigned int offset, 1458d47992f8SLukas Czerner unsigned int length) 14591da177e4SLinus Torvalds { 14601da177e4SLinus Torvalds struct buffer_head *head, *bh, *next; 14611da177e4SLinus Torvalds unsigned int curr_off = 0; 1462d47992f8SLukas Czerner unsigned int stop = length + offset; 14631da177e4SLinus Torvalds 14641da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 14651da177e4SLinus Torvalds if (!page_has_buffers(page)) 14661da177e4SLinus Torvalds goto out; 14671da177e4SLinus Torvalds 1468d47992f8SLukas Czerner /* 1469d47992f8SLukas Czerner * Check for overflow 1470d47992f8SLukas Czerner */ 147109cbfeafSKirill A. Shutemov BUG_ON(stop > PAGE_SIZE || stop < length); 1472d47992f8SLukas Czerner 14731da177e4SLinus Torvalds head = page_buffers(page); 14741da177e4SLinus Torvalds bh = head; 14751da177e4SLinus Torvalds do { 14761da177e4SLinus Torvalds unsigned int next_off = curr_off + bh->b_size; 14771da177e4SLinus Torvalds next = bh->b_this_page; 14781da177e4SLinus Torvalds 14791da177e4SLinus Torvalds /* 1480d47992f8SLukas Czerner * Are we still fully in range ? 1481d47992f8SLukas Czerner */ 1482d47992f8SLukas Czerner if (next_off > stop) 1483d47992f8SLukas Czerner goto out; 1484d47992f8SLukas Czerner 1485d47992f8SLukas Czerner /* 14861da177e4SLinus Torvalds * is this block fully invalidated? 14871da177e4SLinus Torvalds */ 14881da177e4SLinus Torvalds if (offset <= curr_off) 14891da177e4SLinus Torvalds discard_buffer(bh); 14901da177e4SLinus Torvalds curr_off = next_off; 14911da177e4SLinus Torvalds bh = next; 14921da177e4SLinus Torvalds } while (bh != head); 14931da177e4SLinus Torvalds 14941da177e4SLinus Torvalds /* 14951da177e4SLinus Torvalds * We release buffers only if the entire page is being invalidated. 14961da177e4SLinus Torvalds * The get_block cached value has been unconditionally invalidated, 14971da177e4SLinus Torvalds * so real IO is not possible anymore. 14981da177e4SLinus Torvalds */ 14993172485fSJeff Moyer if (length == PAGE_SIZE) 15002ff28e22SNeilBrown try_to_release_page(page, 0); 15011da177e4SLinus Torvalds out: 15022ff28e22SNeilBrown return; 15031da177e4SLinus Torvalds } 15041da177e4SLinus Torvalds EXPORT_SYMBOL(block_invalidatepage); 15051da177e4SLinus Torvalds 1506d47992f8SLukas Czerner 15071da177e4SLinus Torvalds /* 15081da177e4SLinus Torvalds * We attach and possibly dirty the buffers atomically wrt 15091da177e4SLinus Torvalds * __set_page_dirty_buffers() via private_lock. try_to_free_buffers 15101da177e4SLinus Torvalds * is already excluded via the page lock. 15111da177e4SLinus Torvalds */ 15121da177e4SLinus Torvalds void create_empty_buffers(struct page *page, 15131da177e4SLinus Torvalds unsigned long blocksize, unsigned long b_state) 15141da177e4SLinus Torvalds { 15151da177e4SLinus Torvalds struct buffer_head *bh, *head, *tail; 15161da177e4SLinus Torvalds 1517640ab98fSJens Axboe head = alloc_page_buffers(page, blocksize, true); 15181da177e4SLinus Torvalds bh = head; 15191da177e4SLinus Torvalds do { 15201da177e4SLinus Torvalds bh->b_state |= b_state; 15211da177e4SLinus Torvalds tail = bh; 15221da177e4SLinus Torvalds bh = bh->b_this_page; 15231da177e4SLinus Torvalds } while (bh); 15241da177e4SLinus Torvalds tail->b_this_page = head; 15251da177e4SLinus Torvalds 15261da177e4SLinus Torvalds spin_lock(&page->mapping->private_lock); 15271da177e4SLinus Torvalds if (PageUptodate(page) || PageDirty(page)) { 15281da177e4SLinus Torvalds bh = head; 15291da177e4SLinus Torvalds do { 15301da177e4SLinus Torvalds if (PageDirty(page)) 15311da177e4SLinus Torvalds set_buffer_dirty(bh); 15321da177e4SLinus Torvalds if (PageUptodate(page)) 15331da177e4SLinus Torvalds set_buffer_uptodate(bh); 15341da177e4SLinus Torvalds bh = bh->b_this_page; 15351da177e4SLinus Torvalds } while (bh != head); 15361da177e4SLinus Torvalds } 15371da177e4SLinus Torvalds attach_page_buffers(page, head); 15381da177e4SLinus Torvalds spin_unlock(&page->mapping->private_lock); 15391da177e4SLinus Torvalds } 15401da177e4SLinus Torvalds EXPORT_SYMBOL(create_empty_buffers); 15411da177e4SLinus Torvalds 154229f3ad7dSJan Kara /** 154329f3ad7dSJan Kara * clean_bdev_aliases: clean a range of buffers in block device 154429f3ad7dSJan Kara * @bdev: Block device to clean buffers in 154529f3ad7dSJan Kara * @block: Start of a range of blocks to clean 154629f3ad7dSJan Kara * @len: Number of blocks to clean 15471da177e4SLinus Torvalds * 154829f3ad7dSJan Kara * We are taking a range of blocks for data and we don't want writeback of any 154929f3ad7dSJan Kara * buffer-cache aliases starting from return from this function and until the 155029f3ad7dSJan Kara * moment when something will explicitly mark the buffer dirty (hopefully that 155129f3ad7dSJan Kara * will not happen until we will free that block ;-) We don't even need to mark 155229f3ad7dSJan Kara * it not-uptodate - nobody can expect anything from a newly allocated buffer 155329f3ad7dSJan Kara * anyway. We used to use unmap_buffer() for such invalidation, but that was 155429f3ad7dSJan Kara * wrong. We definitely don't want to mark the alias unmapped, for example - it 155529f3ad7dSJan Kara * would confuse anyone who might pick it with bread() afterwards... 155629f3ad7dSJan Kara * 155729f3ad7dSJan Kara * Also.. Note that bforget() doesn't lock the buffer. So there can be 155829f3ad7dSJan Kara * writeout I/O going on against recently-freed buffers. We don't wait on that 155929f3ad7dSJan Kara * I/O in bforget() - it's more efficient to wait on the I/O only if we really 156029f3ad7dSJan Kara * need to. That happens here. 15611da177e4SLinus Torvalds */ 156229f3ad7dSJan Kara void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) 15631da177e4SLinus Torvalds { 156429f3ad7dSJan Kara struct inode *bd_inode = bdev->bd_inode; 156529f3ad7dSJan Kara struct address_space *bd_mapping = bd_inode->i_mapping; 156629f3ad7dSJan Kara struct pagevec pvec; 156729f3ad7dSJan Kara pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); 156829f3ad7dSJan Kara pgoff_t end; 1569c10f778dSJan Kara int i, count; 157029f3ad7dSJan Kara struct buffer_head *bh; 157129f3ad7dSJan Kara struct buffer_head *head; 15721da177e4SLinus Torvalds 157329f3ad7dSJan Kara end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits); 157486679820SMel Gorman pagevec_init(&pvec); 1575397162ffSJan Kara while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) { 1576c10f778dSJan Kara count = pagevec_count(&pvec); 1577c10f778dSJan Kara for (i = 0; i < count; i++) { 157829f3ad7dSJan Kara struct page *page = pvec.pages[i]; 15791da177e4SLinus Torvalds 158029f3ad7dSJan Kara if (!page_has_buffers(page)) 158129f3ad7dSJan Kara continue; 158229f3ad7dSJan Kara /* 158329f3ad7dSJan Kara * We use page lock instead of bd_mapping->private_lock 158429f3ad7dSJan Kara * to pin buffers here since we can afford to sleep and 158529f3ad7dSJan Kara * it scales better than a global spinlock lock. 158629f3ad7dSJan Kara */ 158729f3ad7dSJan Kara lock_page(page); 158829f3ad7dSJan Kara /* Recheck when the page is locked which pins bhs */ 158929f3ad7dSJan Kara if (!page_has_buffers(page)) 159029f3ad7dSJan Kara goto unlock_page; 159129f3ad7dSJan Kara head = page_buffers(page); 159229f3ad7dSJan Kara bh = head; 159329f3ad7dSJan Kara do { 15946c006a9dSChandan Rajendra if (!buffer_mapped(bh) || (bh->b_blocknr < block)) 159529f3ad7dSJan Kara goto next; 159629f3ad7dSJan Kara if (bh->b_blocknr >= block + len) 159729f3ad7dSJan Kara break; 159829f3ad7dSJan Kara clear_buffer_dirty(bh); 159929f3ad7dSJan Kara wait_on_buffer(bh); 160029f3ad7dSJan Kara clear_buffer_req(bh); 160129f3ad7dSJan Kara next: 160229f3ad7dSJan Kara bh = bh->b_this_page; 160329f3ad7dSJan Kara } while (bh != head); 160429f3ad7dSJan Kara unlock_page: 160529f3ad7dSJan Kara unlock_page(page); 160629f3ad7dSJan Kara } 160729f3ad7dSJan Kara pagevec_release(&pvec); 160829f3ad7dSJan Kara cond_resched(); 1609c10f778dSJan Kara /* End of range already reached? */ 1610c10f778dSJan Kara if (index > end || !index) 1611c10f778dSJan Kara break; 16121da177e4SLinus Torvalds } 16131da177e4SLinus Torvalds } 161429f3ad7dSJan Kara EXPORT_SYMBOL(clean_bdev_aliases); 16151da177e4SLinus Torvalds 16161da177e4SLinus Torvalds /* 161745bce8f3SLinus Torvalds * Size is a power-of-two in the range 512..PAGE_SIZE, 161845bce8f3SLinus Torvalds * and the case we care about most is PAGE_SIZE. 161945bce8f3SLinus Torvalds * 162045bce8f3SLinus Torvalds * So this *could* possibly be written with those 162145bce8f3SLinus Torvalds * constraints in mind (relevant mostly if some 162245bce8f3SLinus Torvalds * architecture has a slow bit-scan instruction) 162345bce8f3SLinus Torvalds */ 162445bce8f3SLinus Torvalds static inline int block_size_bits(unsigned int blocksize) 162545bce8f3SLinus Torvalds { 162645bce8f3SLinus Torvalds return ilog2(blocksize); 162745bce8f3SLinus Torvalds } 162845bce8f3SLinus Torvalds 162945bce8f3SLinus Torvalds static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state) 163045bce8f3SLinus Torvalds { 163145bce8f3SLinus Torvalds BUG_ON(!PageLocked(page)); 163245bce8f3SLinus Torvalds 163345bce8f3SLinus Torvalds if (!page_has_buffers(page)) 16346aa7de05SMark Rutland create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits), 16356aa7de05SMark Rutland b_state); 163645bce8f3SLinus Torvalds return page_buffers(page); 163745bce8f3SLinus Torvalds } 163845bce8f3SLinus Torvalds 163945bce8f3SLinus Torvalds /* 16401da177e4SLinus Torvalds * NOTE! All mapped/uptodate combinations are valid: 16411da177e4SLinus Torvalds * 16421da177e4SLinus Torvalds * Mapped Uptodate Meaning 16431da177e4SLinus Torvalds * 16441da177e4SLinus Torvalds * No No "unknown" - must do get_block() 16451da177e4SLinus Torvalds * No Yes "hole" - zero-filled 16461da177e4SLinus Torvalds * Yes No "allocated" - allocated on disk, not read in 16471da177e4SLinus Torvalds * Yes Yes "valid" - allocated and up-to-date in memory. 16481da177e4SLinus Torvalds * 16491da177e4SLinus Torvalds * "Dirty" is valid only with the last case (mapped+uptodate). 16501da177e4SLinus Torvalds */ 16511da177e4SLinus Torvalds 16521da177e4SLinus Torvalds /* 16531da177e4SLinus Torvalds * While block_write_full_page is writing back the dirty buffers under 16541da177e4SLinus Torvalds * the page lock, whoever dirtied the buffers may decide to clean them 16551da177e4SLinus Torvalds * again at any time. We handle that by only looking at the buffer 16561da177e4SLinus Torvalds * state inside lock_buffer(). 16571da177e4SLinus Torvalds * 16581da177e4SLinus Torvalds * If block_write_full_page() is called for regular writeback 16591da177e4SLinus Torvalds * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a 16601da177e4SLinus Torvalds * locked buffer. This only can happen if someone has written the buffer 16611da177e4SLinus Torvalds * directly, with submit_bh(). At the address_space level PageWriteback 16621da177e4SLinus Torvalds * prevents this contention from occurring. 16636e34eeddSTheodore Ts'o * 16646e34eeddSTheodore Ts'o * If block_write_full_page() is called with wbc->sync_mode == 166570fd7614SChristoph Hellwig * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this 1666721a9602SJens Axboe * causes the writes to be flagged as synchronous writes. 16671da177e4SLinus Torvalds */ 1668b4bba389SBenjamin Marzinski int __block_write_full_page(struct inode *inode, struct page *page, 166935c80d5fSChris Mason get_block_t *get_block, struct writeback_control *wbc, 167035c80d5fSChris Mason bh_end_io_t *handler) 16711da177e4SLinus Torvalds { 16721da177e4SLinus Torvalds int err; 16731da177e4SLinus Torvalds sector_t block; 16741da177e4SLinus Torvalds sector_t last_block; 1675f0fbd5fcSAndrew Morton struct buffer_head *bh, *head; 167645bce8f3SLinus Torvalds unsigned int blocksize, bbits; 16771da177e4SLinus Torvalds int nr_underway = 0; 16787637241eSJens Axboe int write_flags = wbc_to_write_flags(wbc); 16791da177e4SLinus Torvalds 168045bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 16811da177e4SLinus Torvalds (1 << BH_Dirty)|(1 << BH_Uptodate)); 16821da177e4SLinus Torvalds 16831da177e4SLinus Torvalds /* 16841da177e4SLinus Torvalds * Be very careful. We have no exclusion from __set_page_dirty_buffers 16851da177e4SLinus Torvalds * here, and the (potentially unmapped) buffers may become dirty at 16861da177e4SLinus Torvalds * any time. If a buffer becomes dirty here after we've inspected it 16871da177e4SLinus Torvalds * then we just miss that fact, and the page stays dirty. 16881da177e4SLinus Torvalds * 16891da177e4SLinus Torvalds * Buffers outside i_size may be dirtied by __set_page_dirty_buffers; 16901da177e4SLinus Torvalds * handle that here by just cleaning them. 16911da177e4SLinus Torvalds */ 16921da177e4SLinus Torvalds 16931da177e4SLinus Torvalds bh = head; 169445bce8f3SLinus Torvalds blocksize = bh->b_size; 169545bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 169645bce8f3SLinus Torvalds 169709cbfeafSKirill A. Shutemov block = (sector_t)page->index << (PAGE_SHIFT - bbits); 169845bce8f3SLinus Torvalds last_block = (i_size_read(inode) - 1) >> bbits; 16991da177e4SLinus Torvalds 17001da177e4SLinus Torvalds /* 17011da177e4SLinus Torvalds * Get all the dirty buffers mapped to disk addresses and 17021da177e4SLinus Torvalds * handle any aliases from the underlying blockdev's mapping. 17031da177e4SLinus Torvalds */ 17041da177e4SLinus Torvalds do { 17051da177e4SLinus Torvalds if (block > last_block) { 17061da177e4SLinus Torvalds /* 17071da177e4SLinus Torvalds * mapped buffers outside i_size will occur, because 17081da177e4SLinus Torvalds * this page can be outside i_size when there is a 17091da177e4SLinus Torvalds * truncate in progress. 17101da177e4SLinus Torvalds */ 17111da177e4SLinus Torvalds /* 17121da177e4SLinus Torvalds * The buffer was zeroed by block_write_full_page() 17131da177e4SLinus Torvalds */ 17141da177e4SLinus Torvalds clear_buffer_dirty(bh); 17151da177e4SLinus Torvalds set_buffer_uptodate(bh); 171629a814d2SAlex Tomas } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && 171729a814d2SAlex Tomas buffer_dirty(bh)) { 1718b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 17191da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 17201da177e4SLinus Torvalds if (err) 17211da177e4SLinus Torvalds goto recover; 172229a814d2SAlex Tomas clear_buffer_delay(bh); 17231da177e4SLinus Torvalds if (buffer_new(bh)) { 17241da177e4SLinus Torvalds /* blockdev mappings never come here */ 17251da177e4SLinus Torvalds clear_buffer_new(bh); 1726e64855c6SJan Kara clean_bdev_bh_alias(bh); 17271da177e4SLinus Torvalds } 17281da177e4SLinus Torvalds } 17291da177e4SLinus Torvalds bh = bh->b_this_page; 17301da177e4SLinus Torvalds block++; 17311da177e4SLinus Torvalds } while (bh != head); 17321da177e4SLinus Torvalds 17331da177e4SLinus Torvalds do { 17341da177e4SLinus Torvalds if (!buffer_mapped(bh)) 17351da177e4SLinus Torvalds continue; 17361da177e4SLinus Torvalds /* 17371da177e4SLinus Torvalds * If it's a fully non-blocking write attempt and we cannot 17381da177e4SLinus Torvalds * lock the buffer then redirty the page. Note that this can 17395b0830cbSJens Axboe * potentially cause a busy-wait loop from writeback threads 17405b0830cbSJens Axboe * and kswapd activity, but those code paths have their own 17415b0830cbSJens Axboe * higher-level throttling. 17421da177e4SLinus Torvalds */ 17431b430beeSWu Fengguang if (wbc->sync_mode != WB_SYNC_NONE) { 17441da177e4SLinus Torvalds lock_buffer(bh); 1745ca5de404SNick Piggin } else if (!trylock_buffer(bh)) { 17461da177e4SLinus Torvalds redirty_page_for_writepage(wbc, page); 17471da177e4SLinus Torvalds continue; 17481da177e4SLinus Torvalds } 17491da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 175035c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 17511da177e4SLinus Torvalds } else { 17521da177e4SLinus Torvalds unlock_buffer(bh); 17531da177e4SLinus Torvalds } 17541da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 17551da177e4SLinus Torvalds 17561da177e4SLinus Torvalds /* 17571da177e4SLinus Torvalds * The page and its buffers are protected by PageWriteback(), so we can 17581da177e4SLinus Torvalds * drop the bh refcounts early. 17591da177e4SLinus Torvalds */ 17601da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 17611da177e4SLinus Torvalds set_page_writeback(page); 17621da177e4SLinus Torvalds 17631da177e4SLinus Torvalds do { 17641da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 17651da177e4SLinus Torvalds if (buffer_async_write(bh)) { 17668e8f9298SJens Axboe submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, 17678e8f9298SJens Axboe inode->i_write_hint, wbc); 17681da177e4SLinus Torvalds nr_underway++; 1769ad576e63SNick Piggin } 17701da177e4SLinus Torvalds bh = next; 17711da177e4SLinus Torvalds } while (bh != head); 177205937baaSAndrew Morton unlock_page(page); 17731da177e4SLinus Torvalds 17741da177e4SLinus Torvalds err = 0; 17751da177e4SLinus Torvalds done: 17761da177e4SLinus Torvalds if (nr_underway == 0) { 17771da177e4SLinus Torvalds /* 17781da177e4SLinus Torvalds * The page was marked dirty, but the buffers were 17791da177e4SLinus Torvalds * clean. Someone wrote them back by hand with 17801da177e4SLinus Torvalds * ll_rw_block/submit_bh. A rare case. 17811da177e4SLinus Torvalds */ 17821da177e4SLinus Torvalds end_page_writeback(page); 17833d67f2d7SNick Piggin 17841da177e4SLinus Torvalds /* 17851da177e4SLinus Torvalds * The page and buffer_heads can be released at any time from 17861da177e4SLinus Torvalds * here on. 17871da177e4SLinus Torvalds */ 17881da177e4SLinus Torvalds } 17891da177e4SLinus Torvalds return err; 17901da177e4SLinus Torvalds 17911da177e4SLinus Torvalds recover: 17921da177e4SLinus Torvalds /* 17931da177e4SLinus Torvalds * ENOSPC, or some other error. We may already have added some 17941da177e4SLinus Torvalds * blocks to the file, so we need to write these out to avoid 17951da177e4SLinus Torvalds * exposing stale data. 17961da177e4SLinus Torvalds * The page is currently locked and not marked for writeback 17971da177e4SLinus Torvalds */ 17981da177e4SLinus Torvalds bh = head; 17991da177e4SLinus Torvalds /* Recovery: lock and submit the mapped buffers */ 18001da177e4SLinus Torvalds do { 180129a814d2SAlex Tomas if (buffer_mapped(bh) && buffer_dirty(bh) && 180229a814d2SAlex Tomas !buffer_delay(bh)) { 18031da177e4SLinus Torvalds lock_buffer(bh); 180435c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 18051da177e4SLinus Torvalds } else { 18061da177e4SLinus Torvalds /* 18071da177e4SLinus Torvalds * The buffer may have been set dirty during 18081da177e4SLinus Torvalds * attachment to a dirty page. 18091da177e4SLinus Torvalds */ 18101da177e4SLinus Torvalds clear_buffer_dirty(bh); 18111da177e4SLinus Torvalds } 18121da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 18131da177e4SLinus Torvalds SetPageError(page); 18141da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 18157e4c3690SAndrew Morton mapping_set_error(page->mapping, err); 18161da177e4SLinus Torvalds set_page_writeback(page); 18171da177e4SLinus Torvalds do { 18181da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 18191da177e4SLinus Torvalds if (buffer_async_write(bh)) { 18201da177e4SLinus Torvalds clear_buffer_dirty(bh); 18218e8f9298SJens Axboe submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, 18228e8f9298SJens Axboe inode->i_write_hint, wbc); 18231da177e4SLinus Torvalds nr_underway++; 1824ad576e63SNick Piggin } 18251da177e4SLinus Torvalds bh = next; 18261da177e4SLinus Torvalds } while (bh != head); 1827ffda9d30SNick Piggin unlock_page(page); 18281da177e4SLinus Torvalds goto done; 18291da177e4SLinus Torvalds } 1830b4bba389SBenjamin Marzinski EXPORT_SYMBOL(__block_write_full_page); 18311da177e4SLinus Torvalds 1832afddba49SNick Piggin /* 1833afddba49SNick Piggin * If a page has any new buffers, zero them out here, and mark them uptodate 1834afddba49SNick Piggin * and dirty so they'll be written out (in order to prevent uninitialised 1835afddba49SNick Piggin * block data from leaking). And clear the new bit. 1836afddba49SNick Piggin */ 1837afddba49SNick Piggin void page_zero_new_buffers(struct page *page, unsigned from, unsigned to) 1838afddba49SNick Piggin { 1839afddba49SNick Piggin unsigned int block_start, block_end; 1840afddba49SNick Piggin struct buffer_head *head, *bh; 1841afddba49SNick Piggin 1842afddba49SNick Piggin BUG_ON(!PageLocked(page)); 1843afddba49SNick Piggin if (!page_has_buffers(page)) 1844afddba49SNick Piggin return; 1845afddba49SNick Piggin 1846afddba49SNick Piggin bh = head = page_buffers(page); 1847afddba49SNick Piggin block_start = 0; 1848afddba49SNick Piggin do { 1849afddba49SNick Piggin block_end = block_start + bh->b_size; 1850afddba49SNick Piggin 1851afddba49SNick Piggin if (buffer_new(bh)) { 1852afddba49SNick Piggin if (block_end > from && block_start < to) { 1853afddba49SNick Piggin if (!PageUptodate(page)) { 1854afddba49SNick Piggin unsigned start, size; 1855afddba49SNick Piggin 1856afddba49SNick Piggin start = max(from, block_start); 1857afddba49SNick Piggin size = min(to, block_end) - start; 1858afddba49SNick Piggin 1859eebd2aa3SChristoph Lameter zero_user(page, start, size); 1860afddba49SNick Piggin set_buffer_uptodate(bh); 1861afddba49SNick Piggin } 1862afddba49SNick Piggin 1863afddba49SNick Piggin clear_buffer_new(bh); 1864afddba49SNick Piggin mark_buffer_dirty(bh); 1865afddba49SNick Piggin } 1866afddba49SNick Piggin } 1867afddba49SNick Piggin 1868afddba49SNick Piggin block_start = block_end; 1869afddba49SNick Piggin bh = bh->b_this_page; 1870afddba49SNick Piggin } while (bh != head); 1871afddba49SNick Piggin } 1872afddba49SNick Piggin EXPORT_SYMBOL(page_zero_new_buffers); 1873afddba49SNick Piggin 1874ae259a9cSChristoph Hellwig static void 1875ae259a9cSChristoph Hellwig iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, 1876ae259a9cSChristoph Hellwig struct iomap *iomap) 1877ae259a9cSChristoph Hellwig { 1878ae259a9cSChristoph Hellwig loff_t offset = block << inode->i_blkbits; 1879ae259a9cSChristoph Hellwig 1880ae259a9cSChristoph Hellwig bh->b_bdev = iomap->bdev; 1881ae259a9cSChristoph Hellwig 1882ae259a9cSChristoph Hellwig /* 1883ae259a9cSChristoph Hellwig * Block points to offset in file we need to map, iomap contains 1884ae259a9cSChristoph Hellwig * the offset at which the map starts. If the map ends before the 1885ae259a9cSChristoph Hellwig * current block, then do not map the buffer and let the caller 1886ae259a9cSChristoph Hellwig * handle it. 1887ae259a9cSChristoph Hellwig */ 1888ae259a9cSChristoph Hellwig BUG_ON(offset >= iomap->offset + iomap->length); 1889ae259a9cSChristoph Hellwig 1890ae259a9cSChristoph Hellwig switch (iomap->type) { 1891ae259a9cSChristoph Hellwig case IOMAP_HOLE: 1892ae259a9cSChristoph Hellwig /* 1893ae259a9cSChristoph Hellwig * If the buffer is not up to date or beyond the current EOF, 1894ae259a9cSChristoph Hellwig * we need to mark it as new to ensure sub-block zeroing is 1895ae259a9cSChristoph Hellwig * executed if necessary. 1896ae259a9cSChristoph Hellwig */ 1897ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 1898ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 1899ae259a9cSChristoph Hellwig set_buffer_new(bh); 1900ae259a9cSChristoph Hellwig break; 1901ae259a9cSChristoph Hellwig case IOMAP_DELALLOC: 1902ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 1903ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 1904ae259a9cSChristoph Hellwig set_buffer_new(bh); 1905ae259a9cSChristoph Hellwig set_buffer_uptodate(bh); 1906ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 1907ae259a9cSChristoph Hellwig set_buffer_delay(bh); 1908ae259a9cSChristoph Hellwig break; 1909ae259a9cSChristoph Hellwig case IOMAP_UNWRITTEN: 1910ae259a9cSChristoph Hellwig /* 19113d7b6b21SAndreas Gruenbacher * For unwritten regions, we always need to ensure that regions 19123d7b6b21SAndreas Gruenbacher * in the block we are not writing to are zeroed. Mark the 19133d7b6b21SAndreas Gruenbacher * buffer as new to ensure this. 1914ae259a9cSChristoph Hellwig */ 1915ae259a9cSChristoph Hellwig set_buffer_new(bh); 1916ae259a9cSChristoph Hellwig set_buffer_unwritten(bh); 1917ae259a9cSChristoph Hellwig /* FALLTHRU */ 1918ae259a9cSChristoph Hellwig case IOMAP_MAPPED: 19193d7b6b21SAndreas Gruenbacher if ((iomap->flags & IOMAP_F_NEW) || 19203d7b6b21SAndreas Gruenbacher offset >= i_size_read(inode)) 1921ae259a9cSChristoph Hellwig set_buffer_new(bh); 192219fe5f64SAndreas Gruenbacher bh->b_blocknr = (iomap->addr + offset - iomap->offset) >> 192319fe5f64SAndreas Gruenbacher inode->i_blkbits; 1924ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 1925ae259a9cSChristoph Hellwig break; 1926ae259a9cSChristoph Hellwig } 1927ae259a9cSChristoph Hellwig } 1928ae259a9cSChristoph Hellwig 1929ae259a9cSChristoph Hellwig int __block_write_begin_int(struct page *page, loff_t pos, unsigned len, 1930ae259a9cSChristoph Hellwig get_block_t *get_block, struct iomap *iomap) 19311da177e4SLinus Torvalds { 193209cbfeafSKirill A. Shutemov unsigned from = pos & (PAGE_SIZE - 1); 1933ebdec241SChristoph Hellwig unsigned to = from + len; 19346e1db88dSChristoph Hellwig struct inode *inode = page->mapping->host; 19351da177e4SLinus Torvalds unsigned block_start, block_end; 19361da177e4SLinus Torvalds sector_t block; 19371da177e4SLinus Torvalds int err = 0; 19381da177e4SLinus Torvalds unsigned blocksize, bbits; 19391da177e4SLinus Torvalds struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; 19401da177e4SLinus Torvalds 19411da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 194209cbfeafSKirill A. Shutemov BUG_ON(from > PAGE_SIZE); 194309cbfeafSKirill A. Shutemov BUG_ON(to > PAGE_SIZE); 19441da177e4SLinus Torvalds BUG_ON(from > to); 19451da177e4SLinus Torvalds 194645bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 0); 194745bce8f3SLinus Torvalds blocksize = head->b_size; 194845bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 19491da177e4SLinus Torvalds 195009cbfeafSKirill A. Shutemov block = (sector_t)page->index << (PAGE_SHIFT - bbits); 19511da177e4SLinus Torvalds 19521da177e4SLinus Torvalds for(bh = head, block_start = 0; bh != head || !block_start; 19531da177e4SLinus Torvalds block++, block_start=block_end, bh = bh->b_this_page) { 19541da177e4SLinus Torvalds block_end = block_start + blocksize; 19551da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 19561da177e4SLinus Torvalds if (PageUptodate(page)) { 19571da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 19581da177e4SLinus Torvalds set_buffer_uptodate(bh); 19591da177e4SLinus Torvalds } 19601da177e4SLinus Torvalds continue; 19611da177e4SLinus Torvalds } 19621da177e4SLinus Torvalds if (buffer_new(bh)) 19631da177e4SLinus Torvalds clear_buffer_new(bh); 19641da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 1965b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 1966ae259a9cSChristoph Hellwig if (get_block) { 19671da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 19681da177e4SLinus Torvalds if (err) 1969f3ddbdc6SNick Piggin break; 1970ae259a9cSChristoph Hellwig } else { 1971ae259a9cSChristoph Hellwig iomap_to_bh(inode, block, bh, iomap); 1972ae259a9cSChristoph Hellwig } 1973ae259a9cSChristoph Hellwig 19741da177e4SLinus Torvalds if (buffer_new(bh)) { 1975e64855c6SJan Kara clean_bdev_bh_alias(bh); 19761da177e4SLinus Torvalds if (PageUptodate(page)) { 1977637aff46SNick Piggin clear_buffer_new(bh); 19781da177e4SLinus Torvalds set_buffer_uptodate(bh); 1979637aff46SNick Piggin mark_buffer_dirty(bh); 19801da177e4SLinus Torvalds continue; 19811da177e4SLinus Torvalds } 1982eebd2aa3SChristoph Lameter if (block_end > to || block_start < from) 1983eebd2aa3SChristoph Lameter zero_user_segments(page, 1984eebd2aa3SChristoph Lameter to, block_end, 1985eebd2aa3SChristoph Lameter block_start, from); 19861da177e4SLinus Torvalds continue; 19871da177e4SLinus Torvalds } 19881da177e4SLinus Torvalds } 19891da177e4SLinus Torvalds if (PageUptodate(page)) { 19901da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 19911da177e4SLinus Torvalds set_buffer_uptodate(bh); 19921da177e4SLinus Torvalds continue; 19931da177e4SLinus Torvalds } 19941da177e4SLinus Torvalds if (!buffer_uptodate(bh) && !buffer_delay(bh) && 199533a266ddSDavid Chinner !buffer_unwritten(bh) && 19961da177e4SLinus Torvalds (block_start < from || block_end > to)) { 1997dfec8a14SMike Christie ll_rw_block(REQ_OP_READ, 0, 1, &bh); 19981da177e4SLinus Torvalds *wait_bh++=bh; 19991da177e4SLinus Torvalds } 20001da177e4SLinus Torvalds } 20011da177e4SLinus Torvalds /* 20021da177e4SLinus Torvalds * If we issued read requests - let them complete. 20031da177e4SLinus Torvalds */ 20041da177e4SLinus Torvalds while(wait_bh > wait) { 20051da177e4SLinus Torvalds wait_on_buffer(*--wait_bh); 20061da177e4SLinus Torvalds if (!buffer_uptodate(*wait_bh)) 2007f3ddbdc6SNick Piggin err = -EIO; 20081da177e4SLinus Torvalds } 2009f9f07b6cSJan Kara if (unlikely(err)) 2010afddba49SNick Piggin page_zero_new_buffers(page, from, to); 20111da177e4SLinus Torvalds return err; 20121da177e4SLinus Torvalds } 2013ae259a9cSChristoph Hellwig 2014ae259a9cSChristoph Hellwig int __block_write_begin(struct page *page, loff_t pos, unsigned len, 2015ae259a9cSChristoph Hellwig get_block_t *get_block) 2016ae259a9cSChristoph Hellwig { 2017ae259a9cSChristoph Hellwig return __block_write_begin_int(page, pos, len, get_block, NULL); 2018ae259a9cSChristoph Hellwig } 2019ebdec241SChristoph Hellwig EXPORT_SYMBOL(__block_write_begin); 20201da177e4SLinus Torvalds 20211da177e4SLinus Torvalds static int __block_commit_write(struct inode *inode, struct page *page, 20221da177e4SLinus Torvalds unsigned from, unsigned to) 20231da177e4SLinus Torvalds { 20241da177e4SLinus Torvalds unsigned block_start, block_end; 20251da177e4SLinus Torvalds int partial = 0; 20261da177e4SLinus Torvalds unsigned blocksize; 20271da177e4SLinus Torvalds struct buffer_head *bh, *head; 20281da177e4SLinus Torvalds 202945bce8f3SLinus Torvalds bh = head = page_buffers(page); 203045bce8f3SLinus Torvalds blocksize = bh->b_size; 20311da177e4SLinus Torvalds 203245bce8f3SLinus Torvalds block_start = 0; 203345bce8f3SLinus Torvalds do { 20341da177e4SLinus Torvalds block_end = block_start + blocksize; 20351da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 20361da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 20371da177e4SLinus Torvalds partial = 1; 20381da177e4SLinus Torvalds } else { 20391da177e4SLinus Torvalds set_buffer_uptodate(bh); 20401da177e4SLinus Torvalds mark_buffer_dirty(bh); 20411da177e4SLinus Torvalds } 2042afddba49SNick Piggin clear_buffer_new(bh); 204345bce8f3SLinus Torvalds 204445bce8f3SLinus Torvalds block_start = block_end; 204545bce8f3SLinus Torvalds bh = bh->b_this_page; 204645bce8f3SLinus Torvalds } while (bh != head); 20471da177e4SLinus Torvalds 20481da177e4SLinus Torvalds /* 20491da177e4SLinus Torvalds * If this is a partial write which happened to make all buffers 20501da177e4SLinus Torvalds * uptodate then we can optimize away a bogus readpage() for 20511da177e4SLinus Torvalds * the next read(). Here we 'discover' whether the page went 20521da177e4SLinus Torvalds * uptodate as a result of this (potentially partial) write. 20531da177e4SLinus Torvalds */ 20541da177e4SLinus Torvalds if (!partial) 20551da177e4SLinus Torvalds SetPageUptodate(page); 20561da177e4SLinus Torvalds return 0; 20571da177e4SLinus Torvalds } 20581da177e4SLinus Torvalds 20591da177e4SLinus Torvalds /* 2060155130a4SChristoph Hellwig * block_write_begin takes care of the basic task of block allocation and 2061155130a4SChristoph Hellwig * bringing partial write blocks uptodate first. 2062155130a4SChristoph Hellwig * 20637bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 2064afddba49SNick Piggin */ 2065155130a4SChristoph Hellwig int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, 2066155130a4SChristoph Hellwig unsigned flags, struct page **pagep, get_block_t *get_block) 2067afddba49SNick Piggin { 206809cbfeafSKirill A. Shutemov pgoff_t index = pos >> PAGE_SHIFT; 2069afddba49SNick Piggin struct page *page; 20706e1db88dSChristoph Hellwig int status; 2071afddba49SNick Piggin 207254566b2cSNick Piggin page = grab_cache_page_write_begin(mapping, index, flags); 20736e1db88dSChristoph Hellwig if (!page) 20746e1db88dSChristoph Hellwig return -ENOMEM; 2075afddba49SNick Piggin 20766e1db88dSChristoph Hellwig status = __block_write_begin(page, pos, len, get_block); 2077afddba49SNick Piggin if (unlikely(status)) { 2078afddba49SNick Piggin unlock_page(page); 207909cbfeafSKirill A. Shutemov put_page(page); 20806e1db88dSChristoph Hellwig page = NULL; 2081afddba49SNick Piggin } 2082afddba49SNick Piggin 20836e1db88dSChristoph Hellwig *pagep = page; 2084afddba49SNick Piggin return status; 2085afddba49SNick Piggin } 2086afddba49SNick Piggin EXPORT_SYMBOL(block_write_begin); 2087afddba49SNick Piggin 2088*26ddb1f4SAndreas Gruenbacher void __generic_write_end(struct inode *inode, loff_t pos, unsigned copied, 2089a6d639daSChristoph Hellwig struct page *page) 2090a6d639daSChristoph Hellwig { 2091a6d639daSChristoph Hellwig loff_t old_size = inode->i_size; 2092a6d639daSChristoph Hellwig bool i_size_changed = false; 2093a6d639daSChristoph Hellwig 2094a6d639daSChristoph Hellwig /* 2095a6d639daSChristoph Hellwig * No need to use i_size_read() here, the i_size cannot change under us 2096a6d639daSChristoph Hellwig * because we hold i_rwsem. 2097a6d639daSChristoph Hellwig * 2098a6d639daSChristoph Hellwig * But it's important to update i_size while still holding page lock: 2099a6d639daSChristoph Hellwig * page writeout could otherwise come in and zero beyond i_size. 2100a6d639daSChristoph Hellwig */ 2101a6d639daSChristoph Hellwig if (pos + copied > inode->i_size) { 2102a6d639daSChristoph Hellwig i_size_write(inode, pos + copied); 2103a6d639daSChristoph Hellwig i_size_changed = true; 2104a6d639daSChristoph Hellwig } 2105a6d639daSChristoph Hellwig 2106a6d639daSChristoph Hellwig unlock_page(page); 2107a6d639daSChristoph Hellwig put_page(page); 2108a6d639daSChristoph Hellwig 2109a6d639daSChristoph Hellwig if (old_size < pos) 2110a6d639daSChristoph Hellwig pagecache_isize_extended(inode, old_size, pos); 2111a6d639daSChristoph Hellwig /* 2112a6d639daSChristoph Hellwig * Don't mark the inode dirty under page lock. First, it unnecessarily 2113a6d639daSChristoph Hellwig * makes the holding time of page lock longer. Second, it forces lock 2114a6d639daSChristoph Hellwig * ordering of page lock and transaction start for journaling 2115a6d639daSChristoph Hellwig * filesystems. 2116a6d639daSChristoph Hellwig */ 2117a6d639daSChristoph Hellwig if (i_size_changed) 2118a6d639daSChristoph Hellwig mark_inode_dirty(inode); 2119a6d639daSChristoph Hellwig } 2120a6d639daSChristoph Hellwig 2121afddba49SNick Piggin int block_write_end(struct file *file, struct address_space *mapping, 2122afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2123afddba49SNick Piggin struct page *page, void *fsdata) 2124afddba49SNick Piggin { 2125afddba49SNick Piggin struct inode *inode = mapping->host; 2126afddba49SNick Piggin unsigned start; 2127afddba49SNick Piggin 212809cbfeafSKirill A. Shutemov start = pos & (PAGE_SIZE - 1); 2129afddba49SNick Piggin 2130afddba49SNick Piggin if (unlikely(copied < len)) { 2131afddba49SNick Piggin /* 2132afddba49SNick Piggin * The buffers that were written will now be uptodate, so we 2133afddba49SNick Piggin * don't have to worry about a readpage reading them and 2134afddba49SNick Piggin * overwriting a partial write. However if we have encountered 2135afddba49SNick Piggin * a short write and only partially written into a buffer, it 2136afddba49SNick Piggin * will not be marked uptodate, so a readpage might come in and 2137afddba49SNick Piggin * destroy our partial write. 2138afddba49SNick Piggin * 2139afddba49SNick Piggin * Do the simplest thing, and just treat any short write to a 2140afddba49SNick Piggin * non uptodate page as a zero-length write, and force the 2141afddba49SNick Piggin * caller to redo the whole thing. 2142afddba49SNick Piggin */ 2143afddba49SNick Piggin if (!PageUptodate(page)) 2144afddba49SNick Piggin copied = 0; 2145afddba49SNick Piggin 2146afddba49SNick Piggin page_zero_new_buffers(page, start+copied, start+len); 2147afddba49SNick Piggin } 2148afddba49SNick Piggin flush_dcache_page(page); 2149afddba49SNick Piggin 2150afddba49SNick Piggin /* This could be a short (even 0-length) commit */ 2151afddba49SNick Piggin __block_commit_write(inode, page, start, start+copied); 2152afddba49SNick Piggin 2153afddba49SNick Piggin return copied; 2154afddba49SNick Piggin } 2155afddba49SNick Piggin EXPORT_SYMBOL(block_write_end); 2156afddba49SNick Piggin 2157afddba49SNick Piggin int generic_write_end(struct file *file, struct address_space *mapping, 2158afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2159afddba49SNick Piggin struct page *page, void *fsdata) 2160afddba49SNick Piggin { 2161afddba49SNick Piggin copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); 2162*26ddb1f4SAndreas Gruenbacher __generic_write_end(mapping->host, pos, copied, page); 2163*26ddb1f4SAndreas Gruenbacher return copied; 2164afddba49SNick Piggin } 2165afddba49SNick Piggin EXPORT_SYMBOL(generic_write_end); 2166afddba49SNick Piggin 2167afddba49SNick Piggin /* 21688ab22b9aSHisashi Hifumi * block_is_partially_uptodate checks whether buffers within a page are 21698ab22b9aSHisashi Hifumi * uptodate or not. 21708ab22b9aSHisashi Hifumi * 21718ab22b9aSHisashi Hifumi * Returns true if all buffers which correspond to a file portion 21728ab22b9aSHisashi Hifumi * we want to read are uptodate. 21738ab22b9aSHisashi Hifumi */ 2174c186afb4SAl Viro int block_is_partially_uptodate(struct page *page, unsigned long from, 2175c186afb4SAl Viro unsigned long count) 21768ab22b9aSHisashi Hifumi { 21778ab22b9aSHisashi Hifumi unsigned block_start, block_end, blocksize; 21788ab22b9aSHisashi Hifumi unsigned to; 21798ab22b9aSHisashi Hifumi struct buffer_head *bh, *head; 21808ab22b9aSHisashi Hifumi int ret = 1; 21818ab22b9aSHisashi Hifumi 21828ab22b9aSHisashi Hifumi if (!page_has_buffers(page)) 21838ab22b9aSHisashi Hifumi return 0; 21848ab22b9aSHisashi Hifumi 218545bce8f3SLinus Torvalds head = page_buffers(page); 218645bce8f3SLinus Torvalds blocksize = head->b_size; 218709cbfeafSKirill A. Shutemov to = min_t(unsigned, PAGE_SIZE - from, count); 21888ab22b9aSHisashi Hifumi to = from + to; 218909cbfeafSKirill A. Shutemov if (from < blocksize && to > PAGE_SIZE - blocksize) 21908ab22b9aSHisashi Hifumi return 0; 21918ab22b9aSHisashi Hifumi 21928ab22b9aSHisashi Hifumi bh = head; 21938ab22b9aSHisashi Hifumi block_start = 0; 21948ab22b9aSHisashi Hifumi do { 21958ab22b9aSHisashi Hifumi block_end = block_start + blocksize; 21968ab22b9aSHisashi Hifumi if (block_end > from && block_start < to) { 21978ab22b9aSHisashi Hifumi if (!buffer_uptodate(bh)) { 21988ab22b9aSHisashi Hifumi ret = 0; 21998ab22b9aSHisashi Hifumi break; 22008ab22b9aSHisashi Hifumi } 22018ab22b9aSHisashi Hifumi if (block_end >= to) 22028ab22b9aSHisashi Hifumi break; 22038ab22b9aSHisashi Hifumi } 22048ab22b9aSHisashi Hifumi block_start = block_end; 22058ab22b9aSHisashi Hifumi bh = bh->b_this_page; 22068ab22b9aSHisashi Hifumi } while (bh != head); 22078ab22b9aSHisashi Hifumi 22088ab22b9aSHisashi Hifumi return ret; 22098ab22b9aSHisashi Hifumi } 22108ab22b9aSHisashi Hifumi EXPORT_SYMBOL(block_is_partially_uptodate); 22118ab22b9aSHisashi Hifumi 22128ab22b9aSHisashi Hifumi /* 22131da177e4SLinus Torvalds * Generic "read page" function for block devices that have the normal 22141da177e4SLinus Torvalds * get_block functionality. This is most of the block device filesystems. 22151da177e4SLinus Torvalds * Reads the page asynchronously --- the unlock_buffer() and 22161da177e4SLinus Torvalds * set/clear_buffer_uptodate() functions propagate buffer state into the 22171da177e4SLinus Torvalds * page struct once IO has completed. 22181da177e4SLinus Torvalds */ 22191da177e4SLinus Torvalds int block_read_full_page(struct page *page, get_block_t *get_block) 22201da177e4SLinus Torvalds { 22211da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 22221da177e4SLinus Torvalds sector_t iblock, lblock; 22231da177e4SLinus Torvalds struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 222445bce8f3SLinus Torvalds unsigned int blocksize, bbits; 22251da177e4SLinus Torvalds int nr, i; 22261da177e4SLinus Torvalds int fully_mapped = 1; 22271da177e4SLinus Torvalds 222845bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 0); 222945bce8f3SLinus Torvalds blocksize = head->b_size; 223045bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 22311da177e4SLinus Torvalds 223209cbfeafSKirill A. Shutemov iblock = (sector_t)page->index << (PAGE_SHIFT - bbits); 223345bce8f3SLinus Torvalds lblock = (i_size_read(inode)+blocksize-1) >> bbits; 22341da177e4SLinus Torvalds bh = head; 22351da177e4SLinus Torvalds nr = 0; 22361da177e4SLinus Torvalds i = 0; 22371da177e4SLinus Torvalds 22381da177e4SLinus Torvalds do { 22391da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22401da177e4SLinus Torvalds continue; 22411da177e4SLinus Torvalds 22421da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2243c64610baSAndrew Morton int err = 0; 2244c64610baSAndrew Morton 22451da177e4SLinus Torvalds fully_mapped = 0; 22461da177e4SLinus Torvalds if (iblock < lblock) { 2247b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2248c64610baSAndrew Morton err = get_block(inode, iblock, bh, 0); 2249c64610baSAndrew Morton if (err) 22501da177e4SLinus Torvalds SetPageError(page); 22511da177e4SLinus Torvalds } 22521da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2253eebd2aa3SChristoph Lameter zero_user(page, i * blocksize, blocksize); 2254c64610baSAndrew Morton if (!err) 22551da177e4SLinus Torvalds set_buffer_uptodate(bh); 22561da177e4SLinus Torvalds continue; 22571da177e4SLinus Torvalds } 22581da177e4SLinus Torvalds /* 22591da177e4SLinus Torvalds * get_block() might have updated the buffer 22601da177e4SLinus Torvalds * synchronously 22611da177e4SLinus Torvalds */ 22621da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22631da177e4SLinus Torvalds continue; 22641da177e4SLinus Torvalds } 22651da177e4SLinus Torvalds arr[nr++] = bh; 22661da177e4SLinus Torvalds } while (i++, iblock++, (bh = bh->b_this_page) != head); 22671da177e4SLinus Torvalds 22681da177e4SLinus Torvalds if (fully_mapped) 22691da177e4SLinus Torvalds SetPageMappedToDisk(page); 22701da177e4SLinus Torvalds 22711da177e4SLinus Torvalds if (!nr) { 22721da177e4SLinus Torvalds /* 22731da177e4SLinus Torvalds * All buffers are uptodate - we can set the page uptodate 22741da177e4SLinus Torvalds * as well. But not if get_block() returned an error. 22751da177e4SLinus Torvalds */ 22761da177e4SLinus Torvalds if (!PageError(page)) 22771da177e4SLinus Torvalds SetPageUptodate(page); 22781da177e4SLinus Torvalds unlock_page(page); 22791da177e4SLinus Torvalds return 0; 22801da177e4SLinus Torvalds } 22811da177e4SLinus Torvalds 22821da177e4SLinus Torvalds /* Stage two: lock the buffers */ 22831da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 22841da177e4SLinus Torvalds bh = arr[i]; 22851da177e4SLinus Torvalds lock_buffer(bh); 22861da177e4SLinus Torvalds mark_buffer_async_read(bh); 22871da177e4SLinus Torvalds } 22881da177e4SLinus Torvalds 22891da177e4SLinus Torvalds /* 22901da177e4SLinus Torvalds * Stage 3: start the IO. Check for uptodateness 22911da177e4SLinus Torvalds * inside the buffer lock in case another process reading 22921da177e4SLinus Torvalds * the underlying blockdev brought it uptodate (the sct fix). 22931da177e4SLinus Torvalds */ 22941da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 22951da177e4SLinus Torvalds bh = arr[i]; 22961da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22971da177e4SLinus Torvalds end_buffer_async_read(bh, 1); 22981da177e4SLinus Torvalds else 22992a222ca9SMike Christie submit_bh(REQ_OP_READ, 0, bh); 23001da177e4SLinus Torvalds } 23011da177e4SLinus Torvalds return 0; 23021da177e4SLinus Torvalds } 23031fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_read_full_page); 23041da177e4SLinus Torvalds 23051da177e4SLinus Torvalds /* utility function for filesystems that need to do work on expanding 230689e10787SNick Piggin * truncates. Uses filesystem pagecache writes to allow the filesystem to 23071da177e4SLinus Torvalds * deal with the hole. 23081da177e4SLinus Torvalds */ 230989e10787SNick Piggin int generic_cont_expand_simple(struct inode *inode, loff_t size) 23101da177e4SLinus Torvalds { 23111da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 23121da177e4SLinus Torvalds struct page *page; 231389e10787SNick Piggin void *fsdata; 23141da177e4SLinus Torvalds int err; 23151da177e4SLinus Torvalds 2316c08d3b0eSnpiggin@suse.de err = inode_newsize_ok(inode, size); 2317c08d3b0eSnpiggin@suse.de if (err) 23181da177e4SLinus Torvalds goto out; 23191da177e4SLinus Torvalds 232089e10787SNick Piggin err = pagecache_write_begin(NULL, mapping, size, 0, 2321c718a975STetsuo Handa AOP_FLAG_CONT_EXPAND, &page, &fsdata); 232289e10787SNick Piggin if (err) 232305eb0b51SOGAWA Hirofumi goto out; 232405eb0b51SOGAWA Hirofumi 232589e10787SNick Piggin err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata); 232689e10787SNick Piggin BUG_ON(err > 0); 232705eb0b51SOGAWA Hirofumi 232805eb0b51SOGAWA Hirofumi out: 232905eb0b51SOGAWA Hirofumi return err; 233005eb0b51SOGAWA Hirofumi } 23311fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_cont_expand_simple); 233205eb0b51SOGAWA Hirofumi 2333f1e3af72SAdrian Bunk static int cont_expand_zero(struct file *file, struct address_space *mapping, 233489e10787SNick Piggin loff_t pos, loff_t *bytes) 233505eb0b51SOGAWA Hirofumi { 233689e10787SNick Piggin struct inode *inode = mapping->host; 233793407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 233889e10787SNick Piggin struct page *page; 233989e10787SNick Piggin void *fsdata; 234089e10787SNick Piggin pgoff_t index, curidx; 234189e10787SNick Piggin loff_t curpos; 234289e10787SNick Piggin unsigned zerofrom, offset, len; 234389e10787SNick Piggin int err = 0; 234405eb0b51SOGAWA Hirofumi 234509cbfeafSKirill A. Shutemov index = pos >> PAGE_SHIFT; 234609cbfeafSKirill A. Shutemov offset = pos & ~PAGE_MASK; 234789e10787SNick Piggin 234809cbfeafSKirill A. Shutemov while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { 234909cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 235089e10787SNick Piggin if (zerofrom & (blocksize-1)) { 235189e10787SNick Piggin *bytes |= (blocksize-1); 235289e10787SNick Piggin (*bytes)++; 235389e10787SNick Piggin } 235409cbfeafSKirill A. Shutemov len = PAGE_SIZE - zerofrom; 235589e10787SNick Piggin 2356c718a975STetsuo Handa err = pagecache_write_begin(file, mapping, curpos, len, 0, 235789e10787SNick Piggin &page, &fsdata); 235889e10787SNick Piggin if (err) 235989e10787SNick Piggin goto out; 2360eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 236189e10787SNick Piggin err = pagecache_write_end(file, mapping, curpos, len, len, 236289e10787SNick Piggin page, fsdata); 236389e10787SNick Piggin if (err < 0) 236489e10787SNick Piggin goto out; 236589e10787SNick Piggin BUG_ON(err != len); 236689e10787SNick Piggin err = 0; 2367061e9746SOGAWA Hirofumi 2368061e9746SOGAWA Hirofumi balance_dirty_pages_ratelimited(mapping); 2369c2ca0fcdSMikulas Patocka 237008d405c8SDavidlohr Bueso if (fatal_signal_pending(current)) { 2371c2ca0fcdSMikulas Patocka err = -EINTR; 2372c2ca0fcdSMikulas Patocka goto out; 2373c2ca0fcdSMikulas Patocka } 237489e10787SNick Piggin } 237589e10787SNick Piggin 237689e10787SNick Piggin /* page covers the boundary, find the boundary offset */ 237789e10787SNick Piggin if (index == curidx) { 237809cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 237989e10787SNick Piggin /* if we will expand the thing last block will be filled */ 238089e10787SNick Piggin if (offset <= zerofrom) { 238189e10787SNick Piggin goto out; 238289e10787SNick Piggin } 238389e10787SNick Piggin if (zerofrom & (blocksize-1)) { 238489e10787SNick Piggin *bytes |= (blocksize-1); 238589e10787SNick Piggin (*bytes)++; 238689e10787SNick Piggin } 238789e10787SNick Piggin len = offset - zerofrom; 238889e10787SNick Piggin 2389c718a975STetsuo Handa err = pagecache_write_begin(file, mapping, curpos, len, 0, 239089e10787SNick Piggin &page, &fsdata); 239189e10787SNick Piggin if (err) 239289e10787SNick Piggin goto out; 2393eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 239489e10787SNick Piggin err = pagecache_write_end(file, mapping, curpos, len, len, 239589e10787SNick Piggin page, fsdata); 239689e10787SNick Piggin if (err < 0) 239789e10787SNick Piggin goto out; 239889e10787SNick Piggin BUG_ON(err != len); 239989e10787SNick Piggin err = 0; 240089e10787SNick Piggin } 240189e10787SNick Piggin out: 240289e10787SNick Piggin return err; 24031da177e4SLinus Torvalds } 24041da177e4SLinus Torvalds 24051da177e4SLinus Torvalds /* 24061da177e4SLinus Torvalds * For moronic filesystems that do not allow holes in file. 24071da177e4SLinus Torvalds * We may have to extend the file. 24081da177e4SLinus Torvalds */ 2409282dc178SChristoph Hellwig int cont_write_begin(struct file *file, struct address_space *mapping, 241089e10787SNick Piggin loff_t pos, unsigned len, unsigned flags, 241189e10787SNick Piggin struct page **pagep, void **fsdata, 241289e10787SNick Piggin get_block_t *get_block, loff_t *bytes) 24131da177e4SLinus Torvalds { 24141da177e4SLinus Torvalds struct inode *inode = mapping->host; 241593407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 241693407472SFabian Frederick unsigned int zerofrom; 241789e10787SNick Piggin int err; 24181da177e4SLinus Torvalds 241989e10787SNick Piggin err = cont_expand_zero(file, mapping, pos, bytes); 242089e10787SNick Piggin if (err) 2421155130a4SChristoph Hellwig return err; 24221da177e4SLinus Torvalds 242309cbfeafSKirill A. Shutemov zerofrom = *bytes & ~PAGE_MASK; 242489e10787SNick Piggin if (pos+len > *bytes && zerofrom & (blocksize-1)) { 24251da177e4SLinus Torvalds *bytes |= (blocksize-1); 24261da177e4SLinus Torvalds (*bytes)++; 24271da177e4SLinus Torvalds } 24281da177e4SLinus Torvalds 2429155130a4SChristoph Hellwig return block_write_begin(mapping, pos, len, flags, pagep, get_block); 24301da177e4SLinus Torvalds } 24311fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(cont_write_begin); 24321da177e4SLinus Torvalds 24331da177e4SLinus Torvalds int block_commit_write(struct page *page, unsigned from, unsigned to) 24341da177e4SLinus Torvalds { 24351da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 24361da177e4SLinus Torvalds __block_commit_write(inode,page,from,to); 24371da177e4SLinus Torvalds return 0; 24381da177e4SLinus Torvalds } 24391fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_commit_write); 24401da177e4SLinus Torvalds 244154171690SDavid Chinner /* 244254171690SDavid Chinner * block_page_mkwrite() is not allowed to change the file size as it gets 244354171690SDavid Chinner * called from a page fault handler when a page is first dirtied. Hence we must 244454171690SDavid Chinner * be careful to check for EOF conditions here. We set the page up correctly 244554171690SDavid Chinner * for a written page which means we get ENOSPC checking when writing into 244654171690SDavid Chinner * holes and correct delalloc and unwritten extent mapping on filesystems that 244754171690SDavid Chinner * support these features. 244854171690SDavid Chinner * 244954171690SDavid Chinner * We are not allowed to take the i_mutex here so we have to play games to 245054171690SDavid Chinner * protect against truncate races as the page could now be beyond EOF. Because 24517bb46a67Snpiggin@suse.de * truncate writes the inode size before removing pages, once we have the 245254171690SDavid Chinner * page lock we can determine safely if the page is beyond EOF. If it is not 245354171690SDavid Chinner * beyond EOF, then the page is guaranteed safe against truncation until we 245454171690SDavid Chinner * unlock the page. 2455ea13a864SJan Kara * 245614da9200SJan Kara * Direct callers of this function should protect against filesystem freezing 24575c500029SRoss Zwisler * using sb_start_pagefault() - sb_end_pagefault() functions. 245854171690SDavid Chinner */ 24595c500029SRoss Zwisler int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 246054171690SDavid Chinner get_block_t get_block) 246154171690SDavid Chinner { 2462c2ec175cSNick Piggin struct page *page = vmf->page; 2463496ad9aaSAl Viro struct inode *inode = file_inode(vma->vm_file); 246454171690SDavid Chinner unsigned long end; 246554171690SDavid Chinner loff_t size; 246624da4fabSJan Kara int ret; 246754171690SDavid Chinner 246854171690SDavid Chinner lock_page(page); 246954171690SDavid Chinner size = i_size_read(inode); 247054171690SDavid Chinner if ((page->mapping != inode->i_mapping) || 247118336338SNick Piggin (page_offset(page) > size)) { 247224da4fabSJan Kara /* We overload EFAULT to mean page got truncated */ 247324da4fabSJan Kara ret = -EFAULT; 247424da4fabSJan Kara goto out_unlock; 247554171690SDavid Chinner } 247654171690SDavid Chinner 247754171690SDavid Chinner /* page is wholly or partially inside EOF */ 247809cbfeafSKirill A. Shutemov if (((page->index + 1) << PAGE_SHIFT) > size) 247909cbfeafSKirill A. Shutemov end = size & ~PAGE_MASK; 248054171690SDavid Chinner else 248109cbfeafSKirill A. Shutemov end = PAGE_SIZE; 248254171690SDavid Chinner 2483ebdec241SChristoph Hellwig ret = __block_write_begin(page, 0, end, get_block); 248454171690SDavid Chinner if (!ret) 248554171690SDavid Chinner ret = block_commit_write(page, 0, end); 248654171690SDavid Chinner 248724da4fabSJan Kara if (unlikely(ret < 0)) 248824da4fabSJan Kara goto out_unlock; 2489ea13a864SJan Kara set_page_dirty(page); 24901d1d1a76SDarrick J. Wong wait_for_stable_page(page); 249124da4fabSJan Kara return 0; 249224da4fabSJan Kara out_unlock: 2493b827e496SNick Piggin unlock_page(page); 249454171690SDavid Chinner return ret; 249554171690SDavid Chinner } 24961fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_page_mkwrite); 24971da177e4SLinus Torvalds 24981da177e4SLinus Torvalds /* 249903158cd7SNick Piggin * nobh_write_begin()'s prereads are special: the buffer_heads are freed 25001da177e4SLinus Torvalds * immediately, while under the page lock. So it needs a special end_io 25011da177e4SLinus Torvalds * handler which does not touch the bh after unlocking it. 25021da177e4SLinus Torvalds */ 25031da177e4SLinus Torvalds static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate) 25041da177e4SLinus Torvalds { 250568671f35SDmitry Monakhov __end_buffer_read_notouch(bh, uptodate); 25061da177e4SLinus Torvalds } 25071da177e4SLinus Torvalds 25081da177e4SLinus Torvalds /* 250903158cd7SNick Piggin * Attach the singly-linked list of buffers created by nobh_write_begin, to 251003158cd7SNick Piggin * the page (converting it to circular linked list and taking care of page 251103158cd7SNick Piggin * dirty races). 251203158cd7SNick Piggin */ 251303158cd7SNick Piggin static void attach_nobh_buffers(struct page *page, struct buffer_head *head) 251403158cd7SNick Piggin { 251503158cd7SNick Piggin struct buffer_head *bh; 251603158cd7SNick Piggin 251703158cd7SNick Piggin BUG_ON(!PageLocked(page)); 251803158cd7SNick Piggin 251903158cd7SNick Piggin spin_lock(&page->mapping->private_lock); 252003158cd7SNick Piggin bh = head; 252103158cd7SNick Piggin do { 252203158cd7SNick Piggin if (PageDirty(page)) 252303158cd7SNick Piggin set_buffer_dirty(bh); 252403158cd7SNick Piggin if (!bh->b_this_page) 252503158cd7SNick Piggin bh->b_this_page = head; 252603158cd7SNick Piggin bh = bh->b_this_page; 252703158cd7SNick Piggin } while (bh != head); 252803158cd7SNick Piggin attach_page_buffers(page, head); 252903158cd7SNick Piggin spin_unlock(&page->mapping->private_lock); 253003158cd7SNick Piggin } 253103158cd7SNick Piggin 253203158cd7SNick Piggin /* 2533ea0f04e5SChristoph Hellwig * On entry, the page is fully not uptodate. 2534ea0f04e5SChristoph Hellwig * On exit the page is fully uptodate in the areas outside (from,to) 25357bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 25361da177e4SLinus Torvalds */ 2537ea0f04e5SChristoph Hellwig int nobh_write_begin(struct address_space *mapping, 253803158cd7SNick Piggin loff_t pos, unsigned len, unsigned flags, 253903158cd7SNick Piggin struct page **pagep, void **fsdata, 25401da177e4SLinus Torvalds get_block_t *get_block) 25411da177e4SLinus Torvalds { 254203158cd7SNick Piggin struct inode *inode = mapping->host; 25431da177e4SLinus Torvalds const unsigned blkbits = inode->i_blkbits; 25441da177e4SLinus Torvalds const unsigned blocksize = 1 << blkbits; 2545a4b0672dSNick Piggin struct buffer_head *head, *bh; 254603158cd7SNick Piggin struct page *page; 254703158cd7SNick Piggin pgoff_t index; 254803158cd7SNick Piggin unsigned from, to; 25491da177e4SLinus Torvalds unsigned block_in_page; 2550a4b0672dSNick Piggin unsigned block_start, block_end; 25511da177e4SLinus Torvalds sector_t block_in_file; 25521da177e4SLinus Torvalds int nr_reads = 0; 25531da177e4SLinus Torvalds int ret = 0; 25541da177e4SLinus Torvalds int is_mapped_to_disk = 1; 25551da177e4SLinus Torvalds 255609cbfeafSKirill A. Shutemov index = pos >> PAGE_SHIFT; 255709cbfeafSKirill A. Shutemov from = pos & (PAGE_SIZE - 1); 255803158cd7SNick Piggin to = from + len; 255903158cd7SNick Piggin 256054566b2cSNick Piggin page = grab_cache_page_write_begin(mapping, index, flags); 256103158cd7SNick Piggin if (!page) 256203158cd7SNick Piggin return -ENOMEM; 256303158cd7SNick Piggin *pagep = page; 256403158cd7SNick Piggin *fsdata = NULL; 256503158cd7SNick Piggin 256603158cd7SNick Piggin if (page_has_buffers(page)) { 2567309f77adSNamhyung Kim ret = __block_write_begin(page, pos, len, get_block); 2568309f77adSNamhyung Kim if (unlikely(ret)) 2569309f77adSNamhyung Kim goto out_release; 2570309f77adSNamhyung Kim return ret; 257103158cd7SNick Piggin } 2572a4b0672dSNick Piggin 25731da177e4SLinus Torvalds if (PageMappedToDisk(page)) 25741da177e4SLinus Torvalds return 0; 25751da177e4SLinus Torvalds 2576a4b0672dSNick Piggin /* 2577a4b0672dSNick Piggin * Allocate buffers so that we can keep track of state, and potentially 2578a4b0672dSNick Piggin * attach them to the page if an error occurs. In the common case of 2579a4b0672dSNick Piggin * no error, they will just be freed again without ever being attached 2580a4b0672dSNick Piggin * to the page (which is all OK, because we're under the page lock). 2581a4b0672dSNick Piggin * 2582a4b0672dSNick Piggin * Be careful: the buffer linked list is a NULL terminated one, rather 2583a4b0672dSNick Piggin * than the circular one we're used to. 2584a4b0672dSNick Piggin */ 2585640ab98fSJens Axboe head = alloc_page_buffers(page, blocksize, false); 258603158cd7SNick Piggin if (!head) { 258703158cd7SNick Piggin ret = -ENOMEM; 258803158cd7SNick Piggin goto out_release; 258903158cd7SNick Piggin } 2590a4b0672dSNick Piggin 259109cbfeafSKirill A. Shutemov block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits); 25921da177e4SLinus Torvalds 25931da177e4SLinus Torvalds /* 25941da177e4SLinus Torvalds * We loop across all blocks in the page, whether or not they are 25951da177e4SLinus Torvalds * part of the affected region. This is so we can discover if the 25961da177e4SLinus Torvalds * page is fully mapped-to-disk. 25971da177e4SLinus Torvalds */ 2598a4b0672dSNick Piggin for (block_start = 0, block_in_page = 0, bh = head; 259909cbfeafSKirill A. Shutemov block_start < PAGE_SIZE; 2600a4b0672dSNick Piggin block_in_page++, block_start += blocksize, bh = bh->b_this_page) { 26011da177e4SLinus Torvalds int create; 26021da177e4SLinus Torvalds 2603a4b0672dSNick Piggin block_end = block_start + blocksize; 2604a4b0672dSNick Piggin bh->b_state = 0; 26051da177e4SLinus Torvalds create = 1; 26061da177e4SLinus Torvalds if (block_start >= to) 26071da177e4SLinus Torvalds create = 0; 26081da177e4SLinus Torvalds ret = get_block(inode, block_in_file + block_in_page, 2609a4b0672dSNick Piggin bh, create); 26101da177e4SLinus Torvalds if (ret) 26111da177e4SLinus Torvalds goto failed; 2612a4b0672dSNick Piggin if (!buffer_mapped(bh)) 26131da177e4SLinus Torvalds is_mapped_to_disk = 0; 2614a4b0672dSNick Piggin if (buffer_new(bh)) 2615e64855c6SJan Kara clean_bdev_bh_alias(bh); 2616a4b0672dSNick Piggin if (PageUptodate(page)) { 2617a4b0672dSNick Piggin set_buffer_uptodate(bh); 26181da177e4SLinus Torvalds continue; 2619a4b0672dSNick Piggin } 2620a4b0672dSNick Piggin if (buffer_new(bh) || !buffer_mapped(bh)) { 2621eebd2aa3SChristoph Lameter zero_user_segments(page, block_start, from, 2622eebd2aa3SChristoph Lameter to, block_end); 26231da177e4SLinus Torvalds continue; 26241da177e4SLinus Torvalds } 2625a4b0672dSNick Piggin if (buffer_uptodate(bh)) 26261da177e4SLinus Torvalds continue; /* reiserfs does this */ 26271da177e4SLinus Torvalds if (block_start < from || block_end > to) { 2628a4b0672dSNick Piggin lock_buffer(bh); 2629a4b0672dSNick Piggin bh->b_end_io = end_buffer_read_nobh; 26302a222ca9SMike Christie submit_bh(REQ_OP_READ, 0, bh); 2631a4b0672dSNick Piggin nr_reads++; 26321da177e4SLinus Torvalds } 26331da177e4SLinus Torvalds } 26341da177e4SLinus Torvalds 26351da177e4SLinus Torvalds if (nr_reads) { 26361da177e4SLinus Torvalds /* 26371da177e4SLinus Torvalds * The page is locked, so these buffers are protected from 26381da177e4SLinus Torvalds * any VM or truncate activity. Hence we don't need to care 26391da177e4SLinus Torvalds * for the buffer_head refcounts. 26401da177e4SLinus Torvalds */ 2641a4b0672dSNick Piggin for (bh = head; bh; bh = bh->b_this_page) { 26421da177e4SLinus Torvalds wait_on_buffer(bh); 26431da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 26441da177e4SLinus Torvalds ret = -EIO; 26451da177e4SLinus Torvalds } 26461da177e4SLinus Torvalds if (ret) 26471da177e4SLinus Torvalds goto failed; 26481da177e4SLinus Torvalds } 26491da177e4SLinus Torvalds 26501da177e4SLinus Torvalds if (is_mapped_to_disk) 26511da177e4SLinus Torvalds SetPageMappedToDisk(page); 26521da177e4SLinus Torvalds 265303158cd7SNick Piggin *fsdata = head; /* to be released by nobh_write_end */ 2654a4b0672dSNick Piggin 26551da177e4SLinus Torvalds return 0; 26561da177e4SLinus Torvalds 26571da177e4SLinus Torvalds failed: 265803158cd7SNick Piggin BUG_ON(!ret); 26591da177e4SLinus Torvalds /* 2660a4b0672dSNick Piggin * Error recovery is a bit difficult. We need to zero out blocks that 2661a4b0672dSNick Piggin * were newly allocated, and dirty them to ensure they get written out. 2662a4b0672dSNick Piggin * Buffers need to be attached to the page at this point, otherwise 2663a4b0672dSNick Piggin * the handling of potential IO errors during writeout would be hard 2664a4b0672dSNick Piggin * (could try doing synchronous writeout, but what if that fails too?) 26651da177e4SLinus Torvalds */ 266603158cd7SNick Piggin attach_nobh_buffers(page, head); 266703158cd7SNick Piggin page_zero_new_buffers(page, from, to); 2668a4b0672dSNick Piggin 266903158cd7SNick Piggin out_release: 267003158cd7SNick Piggin unlock_page(page); 267109cbfeafSKirill A. Shutemov put_page(page); 267203158cd7SNick Piggin *pagep = NULL; 2673a4b0672dSNick Piggin 26747bb46a67Snpiggin@suse.de return ret; 26757bb46a67Snpiggin@suse.de } 267603158cd7SNick Piggin EXPORT_SYMBOL(nobh_write_begin); 26771da177e4SLinus Torvalds 267803158cd7SNick Piggin int nobh_write_end(struct file *file, struct address_space *mapping, 267903158cd7SNick Piggin loff_t pos, unsigned len, unsigned copied, 268003158cd7SNick Piggin struct page *page, void *fsdata) 26811da177e4SLinus Torvalds { 26821da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 2683efdc3131SNick Piggin struct buffer_head *head = fsdata; 268403158cd7SNick Piggin struct buffer_head *bh; 26855b41e74aSDmitri Monakhov BUG_ON(fsdata != NULL && page_has_buffers(page)); 26861da177e4SLinus Torvalds 2687d4cf109fSDave Kleikamp if (unlikely(copied < len) && head) 268803158cd7SNick Piggin attach_nobh_buffers(page, head); 2689a4b0672dSNick Piggin if (page_has_buffers(page)) 269003158cd7SNick Piggin return generic_write_end(file, mapping, pos, len, 269103158cd7SNick Piggin copied, page, fsdata); 2692a4b0672dSNick Piggin 269322c8ca78SNick Piggin SetPageUptodate(page); 26941da177e4SLinus Torvalds set_page_dirty(page); 269503158cd7SNick Piggin if (pos+copied > inode->i_size) { 269603158cd7SNick Piggin i_size_write(inode, pos+copied); 26971da177e4SLinus Torvalds mark_inode_dirty(inode); 26981da177e4SLinus Torvalds } 269903158cd7SNick Piggin 270003158cd7SNick Piggin unlock_page(page); 270109cbfeafSKirill A. Shutemov put_page(page); 270203158cd7SNick Piggin 270303158cd7SNick Piggin while (head) { 270403158cd7SNick Piggin bh = head; 270503158cd7SNick Piggin head = head->b_this_page; 270603158cd7SNick Piggin free_buffer_head(bh); 27071da177e4SLinus Torvalds } 270803158cd7SNick Piggin 270903158cd7SNick Piggin return copied; 271003158cd7SNick Piggin } 271103158cd7SNick Piggin EXPORT_SYMBOL(nobh_write_end); 27121da177e4SLinus Torvalds 27131da177e4SLinus Torvalds /* 27141da177e4SLinus Torvalds * nobh_writepage() - based on block_full_write_page() except 27151da177e4SLinus Torvalds * that it tries to operate without attaching bufferheads to 27161da177e4SLinus Torvalds * the page. 27171da177e4SLinus Torvalds */ 27181da177e4SLinus Torvalds int nobh_writepage(struct page *page, get_block_t *get_block, 27191da177e4SLinus Torvalds struct writeback_control *wbc) 27201da177e4SLinus Torvalds { 27211da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 27221da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 272309cbfeafSKirill A. Shutemov const pgoff_t end_index = i_size >> PAGE_SHIFT; 27241da177e4SLinus Torvalds unsigned offset; 27251da177e4SLinus Torvalds int ret; 27261da177e4SLinus Torvalds 27271da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 27281da177e4SLinus Torvalds if (page->index < end_index) 27291da177e4SLinus Torvalds goto out; 27301da177e4SLinus Torvalds 27311da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 273209cbfeafSKirill A. Shutemov offset = i_size & (PAGE_SIZE-1); 27331da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 27341da177e4SLinus Torvalds /* 27351da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 27361da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 27371da177e4SLinus Torvalds * freeable here, so the page does not leak. 27381da177e4SLinus Torvalds */ 27391da177e4SLinus Torvalds #if 0 27401da177e4SLinus Torvalds /* Not really sure about this - do we need this ? */ 27411da177e4SLinus Torvalds if (page->mapping->a_ops->invalidatepage) 27421da177e4SLinus Torvalds page->mapping->a_ops->invalidatepage(page, offset); 27431da177e4SLinus Torvalds #endif 27441da177e4SLinus Torvalds unlock_page(page); 27451da177e4SLinus Torvalds return 0; /* don't care */ 27461da177e4SLinus Torvalds } 27471da177e4SLinus Torvalds 27481da177e4SLinus Torvalds /* 27491da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 27501da177e4SLinus Torvalds * writepage invocation because it may be mmapped. "A file is mapped 27511da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 27521da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 27531da177e4SLinus Torvalds * writes to that region are not written out to the file." 27541da177e4SLinus Torvalds */ 275509cbfeafSKirill A. Shutemov zero_user_segment(page, offset, PAGE_SIZE); 27561da177e4SLinus Torvalds out: 27571da177e4SLinus Torvalds ret = mpage_writepage(page, get_block, wbc); 27581da177e4SLinus Torvalds if (ret == -EAGAIN) 275935c80d5fSChris Mason ret = __block_write_full_page(inode, page, get_block, wbc, 276035c80d5fSChris Mason end_buffer_async_write); 27611da177e4SLinus Torvalds return ret; 27621da177e4SLinus Torvalds } 27631da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_writepage); 27641da177e4SLinus Torvalds 276503158cd7SNick Piggin int nobh_truncate_page(struct address_space *mapping, 276603158cd7SNick Piggin loff_t from, get_block_t *get_block) 27671da177e4SLinus Torvalds { 276809cbfeafSKirill A. Shutemov pgoff_t index = from >> PAGE_SHIFT; 276909cbfeafSKirill A. Shutemov unsigned offset = from & (PAGE_SIZE-1); 277003158cd7SNick Piggin unsigned blocksize; 277103158cd7SNick Piggin sector_t iblock; 277203158cd7SNick Piggin unsigned length, pos; 277303158cd7SNick Piggin struct inode *inode = mapping->host; 27741da177e4SLinus Torvalds struct page *page; 277503158cd7SNick Piggin struct buffer_head map_bh; 277603158cd7SNick Piggin int err; 27771da177e4SLinus Torvalds 277893407472SFabian Frederick blocksize = i_blocksize(inode); 277903158cd7SNick Piggin length = offset & (blocksize - 1); 27801da177e4SLinus Torvalds 278103158cd7SNick Piggin /* Block boundary? Nothing to do */ 278203158cd7SNick Piggin if (!length) 278303158cd7SNick Piggin return 0; 278403158cd7SNick Piggin 278503158cd7SNick Piggin length = blocksize - length; 278609cbfeafSKirill A. Shutemov iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); 278703158cd7SNick Piggin 27881da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 278903158cd7SNick Piggin err = -ENOMEM; 27901da177e4SLinus Torvalds if (!page) 27911da177e4SLinus Torvalds goto out; 27921da177e4SLinus Torvalds 279303158cd7SNick Piggin if (page_has_buffers(page)) { 279403158cd7SNick Piggin has_buffers: 279503158cd7SNick Piggin unlock_page(page); 279609cbfeafSKirill A. Shutemov put_page(page); 279703158cd7SNick Piggin return block_truncate_page(mapping, from, get_block); 27981da177e4SLinus Torvalds } 279903158cd7SNick Piggin 280003158cd7SNick Piggin /* Find the buffer that contains "offset" */ 280103158cd7SNick Piggin pos = blocksize; 280203158cd7SNick Piggin while (offset >= pos) { 280303158cd7SNick Piggin iblock++; 280403158cd7SNick Piggin pos += blocksize; 280503158cd7SNick Piggin } 280603158cd7SNick Piggin 2807460bcf57STheodore Ts'o map_bh.b_size = blocksize; 2808460bcf57STheodore Ts'o map_bh.b_state = 0; 280903158cd7SNick Piggin err = get_block(inode, iblock, &map_bh, 0); 281003158cd7SNick Piggin if (err) 281103158cd7SNick Piggin goto unlock; 281203158cd7SNick Piggin /* unmapped? It's a hole - nothing to do */ 281303158cd7SNick Piggin if (!buffer_mapped(&map_bh)) 281403158cd7SNick Piggin goto unlock; 281503158cd7SNick Piggin 281603158cd7SNick Piggin /* Ok, it's mapped. Make sure it's up-to-date */ 281703158cd7SNick Piggin if (!PageUptodate(page)) { 281803158cd7SNick Piggin err = mapping->a_ops->readpage(NULL, page); 281903158cd7SNick Piggin if (err) { 282009cbfeafSKirill A. Shutemov put_page(page); 282103158cd7SNick Piggin goto out; 282203158cd7SNick Piggin } 282303158cd7SNick Piggin lock_page(page); 282403158cd7SNick Piggin if (!PageUptodate(page)) { 282503158cd7SNick Piggin err = -EIO; 282603158cd7SNick Piggin goto unlock; 282703158cd7SNick Piggin } 282803158cd7SNick Piggin if (page_has_buffers(page)) 282903158cd7SNick Piggin goto has_buffers; 283003158cd7SNick Piggin } 2831eebd2aa3SChristoph Lameter zero_user(page, offset, length); 283203158cd7SNick Piggin set_page_dirty(page); 283303158cd7SNick Piggin err = 0; 283403158cd7SNick Piggin 283503158cd7SNick Piggin unlock: 28361da177e4SLinus Torvalds unlock_page(page); 283709cbfeafSKirill A. Shutemov put_page(page); 28381da177e4SLinus Torvalds out: 283903158cd7SNick Piggin return err; 28401da177e4SLinus Torvalds } 28411da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_truncate_page); 28421da177e4SLinus Torvalds 28431da177e4SLinus Torvalds int block_truncate_page(struct address_space *mapping, 28441da177e4SLinus Torvalds loff_t from, get_block_t *get_block) 28451da177e4SLinus Torvalds { 284609cbfeafSKirill A. Shutemov pgoff_t index = from >> PAGE_SHIFT; 284709cbfeafSKirill A. Shutemov unsigned offset = from & (PAGE_SIZE-1); 28481da177e4SLinus Torvalds unsigned blocksize; 284954b21a79SAndrew Morton sector_t iblock; 28501da177e4SLinus Torvalds unsigned length, pos; 28511da177e4SLinus Torvalds struct inode *inode = mapping->host; 28521da177e4SLinus Torvalds struct page *page; 28531da177e4SLinus Torvalds struct buffer_head *bh; 28541da177e4SLinus Torvalds int err; 28551da177e4SLinus Torvalds 285693407472SFabian Frederick blocksize = i_blocksize(inode); 28571da177e4SLinus Torvalds length = offset & (blocksize - 1); 28581da177e4SLinus Torvalds 28591da177e4SLinus Torvalds /* Block boundary? Nothing to do */ 28601da177e4SLinus Torvalds if (!length) 28611da177e4SLinus Torvalds return 0; 28621da177e4SLinus Torvalds 28631da177e4SLinus Torvalds length = blocksize - length; 286409cbfeafSKirill A. Shutemov iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); 28651da177e4SLinus Torvalds 28661da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 28671da177e4SLinus Torvalds err = -ENOMEM; 28681da177e4SLinus Torvalds if (!page) 28691da177e4SLinus Torvalds goto out; 28701da177e4SLinus Torvalds 28711da177e4SLinus Torvalds if (!page_has_buffers(page)) 28721da177e4SLinus Torvalds create_empty_buffers(page, blocksize, 0); 28731da177e4SLinus Torvalds 28741da177e4SLinus Torvalds /* Find the buffer that contains "offset" */ 28751da177e4SLinus Torvalds bh = page_buffers(page); 28761da177e4SLinus Torvalds pos = blocksize; 28771da177e4SLinus Torvalds while (offset >= pos) { 28781da177e4SLinus Torvalds bh = bh->b_this_page; 28791da177e4SLinus Torvalds iblock++; 28801da177e4SLinus Torvalds pos += blocksize; 28811da177e4SLinus Torvalds } 28821da177e4SLinus Torvalds 28831da177e4SLinus Torvalds err = 0; 28841da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2885b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 28861da177e4SLinus Torvalds err = get_block(inode, iblock, bh, 0); 28871da177e4SLinus Torvalds if (err) 28881da177e4SLinus Torvalds goto unlock; 28891da177e4SLinus Torvalds /* unmapped? It's a hole - nothing to do */ 28901da177e4SLinus Torvalds if (!buffer_mapped(bh)) 28911da177e4SLinus Torvalds goto unlock; 28921da177e4SLinus Torvalds } 28931da177e4SLinus Torvalds 28941da177e4SLinus Torvalds /* Ok, it's mapped. Make sure it's up-to-date */ 28951da177e4SLinus Torvalds if (PageUptodate(page)) 28961da177e4SLinus Torvalds set_buffer_uptodate(bh); 28971da177e4SLinus Torvalds 289833a266ddSDavid Chinner if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { 28991da177e4SLinus Torvalds err = -EIO; 2900dfec8a14SMike Christie ll_rw_block(REQ_OP_READ, 0, 1, &bh); 29011da177e4SLinus Torvalds wait_on_buffer(bh); 29021da177e4SLinus Torvalds /* Uhhuh. Read error. Complain and punt. */ 29031da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 29041da177e4SLinus Torvalds goto unlock; 29051da177e4SLinus Torvalds } 29061da177e4SLinus Torvalds 2907eebd2aa3SChristoph Lameter zero_user(page, offset, length); 29081da177e4SLinus Torvalds mark_buffer_dirty(bh); 29091da177e4SLinus Torvalds err = 0; 29101da177e4SLinus Torvalds 29111da177e4SLinus Torvalds unlock: 29121da177e4SLinus Torvalds unlock_page(page); 291309cbfeafSKirill A. Shutemov put_page(page); 29141da177e4SLinus Torvalds out: 29151da177e4SLinus Torvalds return err; 29161da177e4SLinus Torvalds } 29171fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_truncate_page); 29181da177e4SLinus Torvalds 29191da177e4SLinus Torvalds /* 29201da177e4SLinus Torvalds * The generic ->writepage function for buffer-backed address_spaces 29211da177e4SLinus Torvalds */ 29221b938c08SMatthew Wilcox int block_write_full_page(struct page *page, get_block_t *get_block, 29231b938c08SMatthew Wilcox struct writeback_control *wbc) 29241da177e4SLinus Torvalds { 29251da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 29261da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 292709cbfeafSKirill A. Shutemov const pgoff_t end_index = i_size >> PAGE_SHIFT; 29281da177e4SLinus Torvalds unsigned offset; 29291da177e4SLinus Torvalds 29301da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 29311da177e4SLinus Torvalds if (page->index < end_index) 293235c80d5fSChris Mason return __block_write_full_page(inode, page, get_block, wbc, 29331b938c08SMatthew Wilcox end_buffer_async_write); 29341da177e4SLinus Torvalds 29351da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 293609cbfeafSKirill A. Shutemov offset = i_size & (PAGE_SIZE-1); 29371da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 29381da177e4SLinus Torvalds /* 29391da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 29401da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 29411da177e4SLinus Torvalds * freeable here, so the page does not leak. 29421da177e4SLinus Torvalds */ 294309cbfeafSKirill A. Shutemov do_invalidatepage(page, 0, PAGE_SIZE); 29441da177e4SLinus Torvalds unlock_page(page); 29451da177e4SLinus Torvalds return 0; /* don't care */ 29461da177e4SLinus Torvalds } 29471da177e4SLinus Torvalds 29481da177e4SLinus Torvalds /* 29491da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 29502a61aa40SAdam Buchbinder * writepage invocation because it may be mmapped. "A file is mapped 29511da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 29521da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 29531da177e4SLinus Torvalds * writes to that region are not written out to the file." 29541da177e4SLinus Torvalds */ 295509cbfeafSKirill A. Shutemov zero_user_segment(page, offset, PAGE_SIZE); 29561b938c08SMatthew Wilcox return __block_write_full_page(inode, page, get_block, wbc, 295735c80d5fSChris Mason end_buffer_async_write); 295835c80d5fSChris Mason } 29591fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_write_full_page); 296035c80d5fSChris Mason 29611da177e4SLinus Torvalds sector_t generic_block_bmap(struct address_space *mapping, sector_t block, 29621da177e4SLinus Torvalds get_block_t *get_block) 29631da177e4SLinus Torvalds { 29641da177e4SLinus Torvalds struct inode *inode = mapping->host; 29652a527d68SAlexander Potapenko struct buffer_head tmp = { 29662a527d68SAlexander Potapenko .b_size = i_blocksize(inode), 29672a527d68SAlexander Potapenko }; 29682a527d68SAlexander Potapenko 29691da177e4SLinus Torvalds get_block(inode, block, &tmp, 0); 29701da177e4SLinus Torvalds return tmp.b_blocknr; 29711da177e4SLinus Torvalds } 29721fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_block_bmap); 29731da177e4SLinus Torvalds 29744246a0b6SChristoph Hellwig static void end_bio_bh_io_sync(struct bio *bio) 29751da177e4SLinus Torvalds { 29761da177e4SLinus Torvalds struct buffer_head *bh = bio->bi_private; 29771da177e4SLinus Torvalds 2978b7c44ed9SJens Axboe if (unlikely(bio_flagged(bio, BIO_QUIET))) 297908bafc03SKeith Mannthey set_bit(BH_Quiet, &bh->b_state); 298008bafc03SKeith Mannthey 29814e4cbee9SChristoph Hellwig bh->b_end_io(bh, !bio->bi_status); 29821da177e4SLinus Torvalds bio_put(bio); 29831da177e4SLinus Torvalds } 29841da177e4SLinus Torvalds 298557302e0dSLinus Torvalds /* 298657302e0dSLinus Torvalds * This allows us to do IO even on the odd last sectors 298759d43914SAkinobu Mita * of a device, even if the block size is some multiple 298857302e0dSLinus Torvalds * of the physical sector size. 298957302e0dSLinus Torvalds * 299057302e0dSLinus Torvalds * We'll just truncate the bio to the size of the device, 299157302e0dSLinus Torvalds * and clear the end of the buffer head manually. 299257302e0dSLinus Torvalds * 299357302e0dSLinus Torvalds * Truly out-of-range accesses will turn into actual IO 299457302e0dSLinus Torvalds * errors, this only handles the "we need to be able to 299557302e0dSLinus Torvalds * do IO at the final sector" case. 299657302e0dSLinus Torvalds */ 29972a222ca9SMike Christie void guard_bio_eod(int op, struct bio *bio) 299857302e0dSLinus Torvalds { 299957302e0dSLinus Torvalds sector_t maxsector; 3000c45a8f2dSMing Lei struct bio_vec *bvec = bio_last_bvec_all(bio); 300159d43914SAkinobu Mita unsigned truncated_bytes; 300267f2519fSGreg Edwards struct hd_struct *part; 300357302e0dSLinus Torvalds 300467f2519fSGreg Edwards rcu_read_lock(); 300567f2519fSGreg Edwards part = __disk_get_part(bio->bi_disk, bio->bi_partno); 300667f2519fSGreg Edwards if (part) 300767f2519fSGreg Edwards maxsector = part_nr_sects_read(part); 300867f2519fSGreg Edwards else 300974d46992SChristoph Hellwig maxsector = get_capacity(bio->bi_disk); 301067f2519fSGreg Edwards rcu_read_unlock(); 301167f2519fSGreg Edwards 301257302e0dSLinus Torvalds if (!maxsector) 301357302e0dSLinus Torvalds return; 301457302e0dSLinus Torvalds 301557302e0dSLinus Torvalds /* 301657302e0dSLinus Torvalds * If the *whole* IO is past the end of the device, 301757302e0dSLinus Torvalds * let it through, and the IO layer will turn it into 301857302e0dSLinus Torvalds * an EIO. 301957302e0dSLinus Torvalds */ 30204f024f37SKent Overstreet if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 302157302e0dSLinus Torvalds return; 302257302e0dSLinus Torvalds 30234f024f37SKent Overstreet maxsector -= bio->bi_iter.bi_sector; 302459d43914SAkinobu Mita if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 302557302e0dSLinus Torvalds return; 302657302e0dSLinus Torvalds 302759d43914SAkinobu Mita /* Uhhuh. We've got a bio that straddles the device size! */ 302859d43914SAkinobu Mita truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9); 302957302e0dSLinus Torvalds 3030dce30ca9SCarlos Maiolino /* 3031dce30ca9SCarlos Maiolino * The bio contains more than one segment which spans EOD, just return 3032dce30ca9SCarlos Maiolino * and let IO layer turn it into an EIO 3033dce30ca9SCarlos Maiolino */ 3034dce30ca9SCarlos Maiolino if (truncated_bytes > bvec->bv_len) 3035dce30ca9SCarlos Maiolino return; 3036dce30ca9SCarlos Maiolino 303757302e0dSLinus Torvalds /* Truncate the bio.. */ 303859d43914SAkinobu Mita bio->bi_iter.bi_size -= truncated_bytes; 303959d43914SAkinobu Mita bvec->bv_len -= truncated_bytes; 304057302e0dSLinus Torvalds 304157302e0dSLinus Torvalds /* ..and clear the end of the buffer for reads */ 30422a222ca9SMike Christie if (op == REQ_OP_READ) { 3043f70f4464SMing Lei struct bio_vec bv; 3044f70f4464SMing Lei 3045f70f4464SMing Lei mp_bvec_last_segment(bvec, &bv); 3046f70f4464SMing Lei zero_user(bv.bv_page, bv.bv_offset + bv.bv_len, 304759d43914SAkinobu Mita truncated_bytes); 304857302e0dSLinus Torvalds } 304957302e0dSLinus Torvalds } 305057302e0dSLinus Torvalds 30512a222ca9SMike Christie static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh, 30528e8f9298SJens Axboe enum rw_hint write_hint, struct writeback_control *wbc) 30531da177e4SLinus Torvalds { 30541da177e4SLinus Torvalds struct bio *bio; 30551da177e4SLinus Torvalds 30561da177e4SLinus Torvalds BUG_ON(!buffer_locked(bh)); 30571da177e4SLinus Torvalds BUG_ON(!buffer_mapped(bh)); 30581da177e4SLinus Torvalds BUG_ON(!bh->b_end_io); 30598fb0e342SAneesh Kumar K.V BUG_ON(buffer_delay(bh)); 30608fb0e342SAneesh Kumar K.V BUG_ON(buffer_unwritten(bh)); 30611da177e4SLinus Torvalds 306248fd4f93SJens Axboe /* 306348fd4f93SJens Axboe * Only clear out a write error when rewriting 30641da177e4SLinus Torvalds */ 30652a222ca9SMike Christie if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) 30661da177e4SLinus Torvalds clear_buffer_write_io_error(bh); 30671da177e4SLinus Torvalds 30681da177e4SLinus Torvalds /* 30691da177e4SLinus Torvalds * from here on down, it's all bio -- do the initial mapping, 30701da177e4SLinus Torvalds * submit_bio -> generic_make_request may further map this bio around 30711da177e4SLinus Torvalds */ 30721da177e4SLinus Torvalds bio = bio_alloc(GFP_NOIO, 1); 30731da177e4SLinus Torvalds 30744f024f37SKent Overstreet bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); 307574d46992SChristoph Hellwig bio_set_dev(bio, bh->b_bdev); 30768e8f9298SJens Axboe bio->bi_write_hint = write_hint; 30771da177e4SLinus Torvalds 30786cf66b4cSKent Overstreet bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); 30796cf66b4cSKent Overstreet BUG_ON(bio->bi_iter.bi_size != bh->b_size); 30801da177e4SLinus Torvalds 30811da177e4SLinus Torvalds bio->bi_end_io = end_bio_bh_io_sync; 30821da177e4SLinus Torvalds bio->bi_private = bh; 30831da177e4SLinus Torvalds 308457302e0dSLinus Torvalds /* Take care of bh's that straddle the end of the device */ 30852a222ca9SMike Christie guard_bio_eod(op, bio); 308657302e0dSLinus Torvalds 3087877f962cSTheodore Ts'o if (buffer_meta(bh)) 30882a222ca9SMike Christie op_flags |= REQ_META; 3089877f962cSTheodore Ts'o if (buffer_prio(bh)) 30902a222ca9SMike Christie op_flags |= REQ_PRIO; 30912a222ca9SMike Christie bio_set_op_attrs(bio, op, op_flags); 3092877f962cSTheodore Ts'o 3093fd42df30SDennis Zhou if (wbc) { 3094fd42df30SDennis Zhou wbc_init_bio(wbc, bio); 3095fd42df30SDennis Zhou wbc_account_io(wbc, bh->b_page, bh->b_size); 3096fd42df30SDennis Zhou } 3097fd42df30SDennis Zhou 30984e49ea4aSMike Christie submit_bio(bio); 3099f6454b04SJulia Lawall return 0; 31001da177e4SLinus Torvalds } 3101bafc0dbaSTejun Heo 31022a222ca9SMike Christie int submit_bh(int op, int op_flags, struct buffer_head *bh) 310371368511SDarrick J. Wong { 31048e8f9298SJens Axboe return submit_bh_wbc(op, op_flags, bh, 0, NULL); 310571368511SDarrick J. Wong } 31061fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(submit_bh); 31071da177e4SLinus Torvalds 31081da177e4SLinus Torvalds /** 31091da177e4SLinus Torvalds * ll_rw_block: low-level access to block devices (DEPRECATED) 3110dfec8a14SMike Christie * @op: whether to %READ or %WRITE 3111ef295ecfSChristoph Hellwig * @op_flags: req_flag_bits 31121da177e4SLinus Torvalds * @nr: number of &struct buffer_heads in the array 31131da177e4SLinus Torvalds * @bhs: array of pointers to &struct buffer_head 31141da177e4SLinus Torvalds * 3115a7662236SJan Kara * ll_rw_block() takes an array of pointers to &struct buffer_heads, and 311670246286SChristoph Hellwig * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE. 311770246286SChristoph Hellwig * @op_flags contains flags modifying the detailed I/O behavior, most notably 311870246286SChristoph Hellwig * %REQ_RAHEAD. 31191da177e4SLinus Torvalds * 31201da177e4SLinus Torvalds * This function drops any buffer that it cannot get a lock on (with the 31219cb569d6SChristoph Hellwig * BH_Lock state bit), any buffer that appears to be clean when doing a write 31229cb569d6SChristoph Hellwig * request, and any buffer that appears to be up-to-date when doing read 31239cb569d6SChristoph Hellwig * request. Further it marks as clean buffers that are processed for 31249cb569d6SChristoph Hellwig * writing (the buffer cache won't assume that they are actually clean 31259cb569d6SChristoph Hellwig * until the buffer gets unlocked). 31261da177e4SLinus Torvalds * 31271da177e4SLinus Torvalds * ll_rw_block sets b_end_io to simple completion handler that marks 3128e227867fSMasanari Iida * the buffer up-to-date (if appropriate), unlocks the buffer and wakes 31291da177e4SLinus Torvalds * any waiters. 31301da177e4SLinus Torvalds * 31311da177e4SLinus Torvalds * All of the buffers must be for the same device, and must also be a 31321da177e4SLinus Torvalds * multiple of the current approved size for the device. 31331da177e4SLinus Torvalds */ 3134dfec8a14SMike Christie void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[]) 31351da177e4SLinus Torvalds { 31361da177e4SLinus Torvalds int i; 31371da177e4SLinus Torvalds 31381da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 31391da177e4SLinus Torvalds struct buffer_head *bh = bhs[i]; 31401da177e4SLinus Torvalds 31419cb569d6SChristoph Hellwig if (!trylock_buffer(bh)) 31421da177e4SLinus Torvalds continue; 3143dfec8a14SMike Christie if (op == WRITE) { 31441da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 314576c3073aSakpm@osdl.org bh->b_end_io = end_buffer_write_sync; 3146e60e5c50SOGAWA Hirofumi get_bh(bh); 3147dfec8a14SMike Christie submit_bh(op, op_flags, bh); 31481da177e4SLinus Torvalds continue; 31491da177e4SLinus Torvalds } 31501da177e4SLinus Torvalds } else { 31511da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 315276c3073aSakpm@osdl.org bh->b_end_io = end_buffer_read_sync; 3153e60e5c50SOGAWA Hirofumi get_bh(bh); 3154dfec8a14SMike Christie submit_bh(op, op_flags, bh); 31551da177e4SLinus Torvalds continue; 31561da177e4SLinus Torvalds } 31571da177e4SLinus Torvalds } 31581da177e4SLinus Torvalds unlock_buffer(bh); 31591da177e4SLinus Torvalds } 31601da177e4SLinus Torvalds } 31611fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(ll_rw_block); 31621da177e4SLinus Torvalds 31632a222ca9SMike Christie void write_dirty_buffer(struct buffer_head *bh, int op_flags) 31649cb569d6SChristoph Hellwig { 31659cb569d6SChristoph Hellwig lock_buffer(bh); 31669cb569d6SChristoph Hellwig if (!test_clear_buffer_dirty(bh)) { 31679cb569d6SChristoph Hellwig unlock_buffer(bh); 31689cb569d6SChristoph Hellwig return; 31699cb569d6SChristoph Hellwig } 31709cb569d6SChristoph Hellwig bh->b_end_io = end_buffer_write_sync; 31719cb569d6SChristoph Hellwig get_bh(bh); 31722a222ca9SMike Christie submit_bh(REQ_OP_WRITE, op_flags, bh); 31739cb569d6SChristoph Hellwig } 31749cb569d6SChristoph Hellwig EXPORT_SYMBOL(write_dirty_buffer); 31759cb569d6SChristoph Hellwig 31761da177e4SLinus Torvalds /* 31771da177e4SLinus Torvalds * For a data-integrity writeout, we need to wait upon any in-progress I/O 31781da177e4SLinus Torvalds * and then start new I/O and then wait upon it. The caller must have a ref on 31791da177e4SLinus Torvalds * the buffer_head. 31801da177e4SLinus Torvalds */ 31812a222ca9SMike Christie int __sync_dirty_buffer(struct buffer_head *bh, int op_flags) 31821da177e4SLinus Torvalds { 31831da177e4SLinus Torvalds int ret = 0; 31841da177e4SLinus Torvalds 31851da177e4SLinus Torvalds WARN_ON(atomic_read(&bh->b_count) < 1); 31861da177e4SLinus Torvalds lock_buffer(bh); 31871da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 31881da177e4SLinus Torvalds get_bh(bh); 31891da177e4SLinus Torvalds bh->b_end_io = end_buffer_write_sync; 31902a222ca9SMike Christie ret = submit_bh(REQ_OP_WRITE, op_flags, bh); 31911da177e4SLinus Torvalds wait_on_buffer(bh); 31921da177e4SLinus Torvalds if (!ret && !buffer_uptodate(bh)) 31931da177e4SLinus Torvalds ret = -EIO; 31941da177e4SLinus Torvalds } else { 31951da177e4SLinus Torvalds unlock_buffer(bh); 31961da177e4SLinus Torvalds } 31971da177e4SLinus Torvalds return ret; 31981da177e4SLinus Torvalds } 319987e99511SChristoph Hellwig EXPORT_SYMBOL(__sync_dirty_buffer); 320087e99511SChristoph Hellwig 320187e99511SChristoph Hellwig int sync_dirty_buffer(struct buffer_head *bh) 320287e99511SChristoph Hellwig { 320370fd7614SChristoph Hellwig return __sync_dirty_buffer(bh, REQ_SYNC); 320487e99511SChristoph Hellwig } 32051fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(sync_dirty_buffer); 32061da177e4SLinus Torvalds 32071da177e4SLinus Torvalds /* 32081da177e4SLinus Torvalds * try_to_free_buffers() checks if all the buffers on this particular page 32091da177e4SLinus Torvalds * are unused, and releases them if so. 32101da177e4SLinus Torvalds * 32111da177e4SLinus Torvalds * Exclusion against try_to_free_buffers may be obtained by either 32121da177e4SLinus Torvalds * locking the page or by holding its mapping's private_lock. 32131da177e4SLinus Torvalds * 32141da177e4SLinus Torvalds * If the page is dirty but all the buffers are clean then we need to 32151da177e4SLinus Torvalds * be sure to mark the page clean as well. This is because the page 32161da177e4SLinus Torvalds * may be against a block device, and a later reattachment of buffers 32171da177e4SLinus Torvalds * to a dirty page will set *all* buffers dirty. Which would corrupt 32181da177e4SLinus Torvalds * filesystem data on the same device. 32191da177e4SLinus Torvalds * 32201da177e4SLinus Torvalds * The same applies to regular filesystem pages: if all the buffers are 32211da177e4SLinus Torvalds * clean then we set the page clean and proceed. To do that, we require 32221da177e4SLinus Torvalds * total exclusion from __set_page_dirty_buffers(). That is obtained with 32231da177e4SLinus Torvalds * private_lock. 32241da177e4SLinus Torvalds * 32251da177e4SLinus Torvalds * try_to_free_buffers() is non-blocking. 32261da177e4SLinus Torvalds */ 32271da177e4SLinus Torvalds static inline int buffer_busy(struct buffer_head *bh) 32281da177e4SLinus Torvalds { 32291da177e4SLinus Torvalds return atomic_read(&bh->b_count) | 32301da177e4SLinus Torvalds (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); 32311da177e4SLinus Torvalds } 32321da177e4SLinus Torvalds 32331da177e4SLinus Torvalds static int 32341da177e4SLinus Torvalds drop_buffers(struct page *page, struct buffer_head **buffers_to_free) 32351da177e4SLinus Torvalds { 32361da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 32371da177e4SLinus Torvalds struct buffer_head *bh; 32381da177e4SLinus Torvalds 32391da177e4SLinus Torvalds bh = head; 32401da177e4SLinus Torvalds do { 32411da177e4SLinus Torvalds if (buffer_busy(bh)) 32421da177e4SLinus Torvalds goto failed; 32431da177e4SLinus Torvalds bh = bh->b_this_page; 32441da177e4SLinus Torvalds } while (bh != head); 32451da177e4SLinus Torvalds 32461da177e4SLinus Torvalds do { 32471da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 32481da177e4SLinus Torvalds 3249535ee2fbSJan Kara if (bh->b_assoc_map) 32501da177e4SLinus Torvalds __remove_assoc_queue(bh); 32511da177e4SLinus Torvalds bh = next; 32521da177e4SLinus Torvalds } while (bh != head); 32531da177e4SLinus Torvalds *buffers_to_free = head; 32541da177e4SLinus Torvalds __clear_page_buffers(page); 32551da177e4SLinus Torvalds return 1; 32561da177e4SLinus Torvalds failed: 32571da177e4SLinus Torvalds return 0; 32581da177e4SLinus Torvalds } 32591da177e4SLinus Torvalds 32601da177e4SLinus Torvalds int try_to_free_buffers(struct page *page) 32611da177e4SLinus Torvalds { 32621da177e4SLinus Torvalds struct address_space * const mapping = page->mapping; 32631da177e4SLinus Torvalds struct buffer_head *buffers_to_free = NULL; 32641da177e4SLinus Torvalds int ret = 0; 32651da177e4SLinus Torvalds 32661da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 3267ecdfc978SLinus Torvalds if (PageWriteback(page)) 32681da177e4SLinus Torvalds return 0; 32691da177e4SLinus Torvalds 32701da177e4SLinus Torvalds if (mapping == NULL) { /* can this still happen? */ 32711da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 32721da177e4SLinus Torvalds goto out; 32731da177e4SLinus Torvalds } 32741da177e4SLinus Torvalds 32751da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 32761da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 3277ecdfc978SLinus Torvalds 3278ecdfc978SLinus Torvalds /* 3279ecdfc978SLinus Torvalds * If the filesystem writes its buffers by hand (eg ext3) 3280ecdfc978SLinus Torvalds * then we can have clean buffers against a dirty page. We 3281ecdfc978SLinus Torvalds * clean the page here; otherwise the VM will never notice 3282ecdfc978SLinus Torvalds * that the filesystem did any IO at all. 3283ecdfc978SLinus Torvalds * 3284ecdfc978SLinus Torvalds * Also, during truncate, discard_buffer will have marked all 3285ecdfc978SLinus Torvalds * the page's buffers clean. We discover that here and clean 3286ecdfc978SLinus Torvalds * the page also. 328787df7241SNick Piggin * 328887df7241SNick Piggin * private_lock must be held over this entire operation in order 328987df7241SNick Piggin * to synchronise against __set_page_dirty_buffers and prevent the 329087df7241SNick Piggin * dirty bit from being lost. 3291ecdfc978SLinus Torvalds */ 329211f81becSTejun Heo if (ret) 329311f81becSTejun Heo cancel_dirty_page(page); 329487df7241SNick Piggin spin_unlock(&mapping->private_lock); 32951da177e4SLinus Torvalds out: 32961da177e4SLinus Torvalds if (buffers_to_free) { 32971da177e4SLinus Torvalds struct buffer_head *bh = buffers_to_free; 32981da177e4SLinus Torvalds 32991da177e4SLinus Torvalds do { 33001da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 33011da177e4SLinus Torvalds free_buffer_head(bh); 33021da177e4SLinus Torvalds bh = next; 33031da177e4SLinus Torvalds } while (bh != buffers_to_free); 33041da177e4SLinus Torvalds } 33051da177e4SLinus Torvalds return ret; 33061da177e4SLinus Torvalds } 33071da177e4SLinus Torvalds EXPORT_SYMBOL(try_to_free_buffers); 33081da177e4SLinus Torvalds 33091da177e4SLinus Torvalds /* 33101da177e4SLinus Torvalds * There are no bdflush tunables left. But distributions are 33111da177e4SLinus Torvalds * still running obsolete flush daemons, so we terminate them here. 33121da177e4SLinus Torvalds * 33131da177e4SLinus Torvalds * Use of bdflush() is deprecated and will be removed in a future kernel. 33145b0830cbSJens Axboe * The `flush-X' kernel threads fully replace bdflush daemons and this call. 33151da177e4SLinus Torvalds */ 3316bdc480e3SHeiko Carstens SYSCALL_DEFINE2(bdflush, int, func, long, data) 33171da177e4SLinus Torvalds { 33181da177e4SLinus Torvalds static int msg_count; 33191da177e4SLinus Torvalds 33201da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 33211da177e4SLinus Torvalds return -EPERM; 33221da177e4SLinus Torvalds 33231da177e4SLinus Torvalds if (msg_count < 5) { 33241da177e4SLinus Torvalds msg_count++; 33251da177e4SLinus Torvalds printk(KERN_INFO 33261da177e4SLinus Torvalds "warning: process `%s' used the obsolete bdflush" 33271da177e4SLinus Torvalds " system call\n", current->comm); 33281da177e4SLinus Torvalds printk(KERN_INFO "Fix your initscripts?\n"); 33291da177e4SLinus Torvalds } 33301da177e4SLinus Torvalds 33311da177e4SLinus Torvalds if (func == 1) 33321da177e4SLinus Torvalds do_exit(0); 33331da177e4SLinus Torvalds return 0; 33341da177e4SLinus Torvalds } 33351da177e4SLinus Torvalds 33361da177e4SLinus Torvalds /* 33371da177e4SLinus Torvalds * Buffer-head allocation 33381da177e4SLinus Torvalds */ 3339a0a9b043SShai Fultheim static struct kmem_cache *bh_cachep __read_mostly; 33401da177e4SLinus Torvalds 33411da177e4SLinus Torvalds /* 33421da177e4SLinus Torvalds * Once the number of bh's in the machine exceeds this level, we start 33431da177e4SLinus Torvalds * stripping them in writeback. 33441da177e4SLinus Torvalds */ 334543be594aSZhang Yanfei static unsigned long max_buffer_heads; 33461da177e4SLinus Torvalds 33471da177e4SLinus Torvalds int buffer_heads_over_limit; 33481da177e4SLinus Torvalds 33491da177e4SLinus Torvalds struct bh_accounting { 33501da177e4SLinus Torvalds int nr; /* Number of live bh's */ 33511da177e4SLinus Torvalds int ratelimit; /* Limit cacheline bouncing */ 33521da177e4SLinus Torvalds }; 33531da177e4SLinus Torvalds 33541da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; 33551da177e4SLinus Torvalds 33561da177e4SLinus Torvalds static void recalc_bh_state(void) 33571da177e4SLinus Torvalds { 33581da177e4SLinus Torvalds int i; 33591da177e4SLinus Torvalds int tot = 0; 33601da177e4SLinus Torvalds 3361ee1be862SChristoph Lameter if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) 33621da177e4SLinus Torvalds return; 3363c7b92516SChristoph Lameter __this_cpu_write(bh_accounting.ratelimit, 0); 33648a143426SEric Dumazet for_each_online_cpu(i) 33651da177e4SLinus Torvalds tot += per_cpu(bh_accounting, i).nr; 33661da177e4SLinus Torvalds buffer_heads_over_limit = (tot > max_buffer_heads); 33671da177e4SLinus Torvalds } 33681da177e4SLinus Torvalds 3369dd0fc66fSAl Viro struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) 33701da177e4SLinus Torvalds { 3371019b4d12SRichard Kennedy struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); 33721da177e4SLinus Torvalds if (ret) { 3373a35afb83SChristoph Lameter INIT_LIST_HEAD(&ret->b_assoc_buffers); 3374c7b92516SChristoph Lameter preempt_disable(); 3375c7b92516SChristoph Lameter __this_cpu_inc(bh_accounting.nr); 33761da177e4SLinus Torvalds recalc_bh_state(); 3377c7b92516SChristoph Lameter preempt_enable(); 33781da177e4SLinus Torvalds } 33791da177e4SLinus Torvalds return ret; 33801da177e4SLinus Torvalds } 33811da177e4SLinus Torvalds EXPORT_SYMBOL(alloc_buffer_head); 33821da177e4SLinus Torvalds 33831da177e4SLinus Torvalds void free_buffer_head(struct buffer_head *bh) 33841da177e4SLinus Torvalds { 33851da177e4SLinus Torvalds BUG_ON(!list_empty(&bh->b_assoc_buffers)); 33861da177e4SLinus Torvalds kmem_cache_free(bh_cachep, bh); 3387c7b92516SChristoph Lameter preempt_disable(); 3388c7b92516SChristoph Lameter __this_cpu_dec(bh_accounting.nr); 33891da177e4SLinus Torvalds recalc_bh_state(); 3390c7b92516SChristoph Lameter preempt_enable(); 33911da177e4SLinus Torvalds } 33921da177e4SLinus Torvalds EXPORT_SYMBOL(free_buffer_head); 33931da177e4SLinus Torvalds 3394fc4d24c9SSebastian Andrzej Siewior static int buffer_exit_cpu_dead(unsigned int cpu) 33951da177e4SLinus Torvalds { 33961da177e4SLinus Torvalds int i; 33971da177e4SLinus Torvalds struct bh_lru *b = &per_cpu(bh_lrus, cpu); 33981da177e4SLinus Torvalds 33991da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 34001da177e4SLinus Torvalds brelse(b->bhs[i]); 34011da177e4SLinus Torvalds b->bhs[i] = NULL; 34021da177e4SLinus Torvalds } 3403c7b92516SChristoph Lameter this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); 34048a143426SEric Dumazet per_cpu(bh_accounting, cpu).nr = 0; 3405fc4d24c9SSebastian Andrzej Siewior return 0; 34061da177e4SLinus Torvalds } 34071da177e4SLinus Torvalds 3408389d1b08SAneesh Kumar K.V /** 3409a6b91919SRandy Dunlap * bh_uptodate_or_lock - Test whether the buffer is uptodate 3410389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3411389d1b08SAneesh Kumar K.V * 3412389d1b08SAneesh Kumar K.V * Return true if the buffer is up-to-date and false, 3413389d1b08SAneesh Kumar K.V * with the buffer locked, if not. 3414389d1b08SAneesh Kumar K.V */ 3415389d1b08SAneesh Kumar K.V int bh_uptodate_or_lock(struct buffer_head *bh) 3416389d1b08SAneesh Kumar K.V { 3417389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) { 3418389d1b08SAneesh Kumar K.V lock_buffer(bh); 3419389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) 3420389d1b08SAneesh Kumar K.V return 0; 3421389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3422389d1b08SAneesh Kumar K.V } 3423389d1b08SAneesh Kumar K.V return 1; 3424389d1b08SAneesh Kumar K.V } 3425389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_uptodate_or_lock); 3426389d1b08SAneesh Kumar K.V 3427389d1b08SAneesh Kumar K.V /** 3428a6b91919SRandy Dunlap * bh_submit_read - Submit a locked buffer for reading 3429389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3430389d1b08SAneesh Kumar K.V * 3431389d1b08SAneesh Kumar K.V * Returns zero on success and -EIO on error. 3432389d1b08SAneesh Kumar K.V */ 3433389d1b08SAneesh Kumar K.V int bh_submit_read(struct buffer_head *bh) 3434389d1b08SAneesh Kumar K.V { 3435389d1b08SAneesh Kumar K.V BUG_ON(!buffer_locked(bh)); 3436389d1b08SAneesh Kumar K.V 3437389d1b08SAneesh Kumar K.V if (buffer_uptodate(bh)) { 3438389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3439389d1b08SAneesh Kumar K.V return 0; 3440389d1b08SAneesh Kumar K.V } 3441389d1b08SAneesh Kumar K.V 3442389d1b08SAneesh Kumar K.V get_bh(bh); 3443389d1b08SAneesh Kumar K.V bh->b_end_io = end_buffer_read_sync; 34442a222ca9SMike Christie submit_bh(REQ_OP_READ, 0, bh); 3445389d1b08SAneesh Kumar K.V wait_on_buffer(bh); 3446389d1b08SAneesh Kumar K.V if (buffer_uptodate(bh)) 3447389d1b08SAneesh Kumar K.V return 0; 3448389d1b08SAneesh Kumar K.V return -EIO; 3449389d1b08SAneesh Kumar K.V } 3450389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_submit_read); 3451389d1b08SAneesh Kumar K.V 34521da177e4SLinus Torvalds void __init buffer_init(void) 34531da177e4SLinus Torvalds { 345443be594aSZhang Yanfei unsigned long nrpages; 3455fc4d24c9SSebastian Andrzej Siewior int ret; 34561da177e4SLinus Torvalds 3457b98938c3SChristoph Lameter bh_cachep = kmem_cache_create("buffer_head", 3458b98938c3SChristoph Lameter sizeof(struct buffer_head), 0, 3459b98938c3SChristoph Lameter (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| 3460b98938c3SChristoph Lameter SLAB_MEM_SPREAD), 3461019b4d12SRichard Kennedy NULL); 34621da177e4SLinus Torvalds 34631da177e4SLinus Torvalds /* 34641da177e4SLinus Torvalds * Limit the bh occupancy to 10% of ZONE_NORMAL 34651da177e4SLinus Torvalds */ 34661da177e4SLinus Torvalds nrpages = (nr_free_buffer_pages() * 10) / 100; 34671da177e4SLinus Torvalds max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); 3468fc4d24c9SSebastian Andrzej Siewior ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", 3469fc4d24c9SSebastian Andrzej Siewior NULL, buffer_exit_cpu_dead); 3470fc4d24c9SSebastian Andrzej Siewior WARN_ON(ret < 0); 34711da177e4SLinus Torvalds } 3472