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/notifier.h> 431da177e4SLinus Torvalds #include <linux/cpu.h> 441da177e4SLinus Torvalds #include <linux/bitops.h> 451da177e4SLinus Torvalds #include <linux/mpage.h> 46fb1c8f93SIngo Molnar #include <linux/bit_spinlock.h> 4729f3ad7dSJan Kara #include <linux/pagevec.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 56a3f3c29cSYan Hong void init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private) 571da177e4SLinus Torvalds { 581da177e4SLinus Torvalds bh->b_end_io = handler; 591da177e4SLinus Torvalds bh->b_private = private; 601da177e4SLinus Torvalds } 611fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(init_buffer); 621da177e4SLinus Torvalds 63f0059afdSTejun Heo inline void touch_buffer(struct buffer_head *bh) 64f0059afdSTejun Heo { 655305cb83STejun Heo trace_block_touch_buffer(bh); 66f0059afdSTejun Heo mark_page_accessed(bh->b_page); 67f0059afdSTejun Heo } 68f0059afdSTejun Heo EXPORT_SYMBOL(touch_buffer); 69f0059afdSTejun Heo 70fc9b52cdSHarvey Harrison void __lock_buffer(struct buffer_head *bh) 711da177e4SLinus Torvalds { 7274316201SNeilBrown wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); 731da177e4SLinus Torvalds } 741da177e4SLinus Torvalds EXPORT_SYMBOL(__lock_buffer); 751da177e4SLinus Torvalds 76fc9b52cdSHarvey Harrison void unlock_buffer(struct buffer_head *bh) 771da177e4SLinus Torvalds { 7851b07fc3SNick Piggin clear_bit_unlock(BH_Lock, &bh->b_state); 794e857c58SPeter Zijlstra smp_mb__after_atomic(); 801da177e4SLinus Torvalds wake_up_bit(&bh->b_state, BH_Lock); 811da177e4SLinus Torvalds } 821fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(unlock_buffer); 831da177e4SLinus Torvalds 841da177e4SLinus Torvalds /* 85b4597226SMel Gorman * Returns if the page has dirty or writeback buffers. If all the buffers 86b4597226SMel Gorman * are unlocked and clean then the PageDirty information is stale. If 87b4597226SMel Gorman * any of the pages are locked, it is assumed they are locked for IO. 88b4597226SMel Gorman */ 89b4597226SMel Gorman void buffer_check_dirty_writeback(struct page *page, 90b4597226SMel Gorman bool *dirty, bool *writeback) 91b4597226SMel Gorman { 92b4597226SMel Gorman struct buffer_head *head, *bh; 93b4597226SMel Gorman *dirty = false; 94b4597226SMel Gorman *writeback = false; 95b4597226SMel Gorman 96b4597226SMel Gorman BUG_ON(!PageLocked(page)); 97b4597226SMel Gorman 98b4597226SMel Gorman if (!page_has_buffers(page)) 99b4597226SMel Gorman return; 100b4597226SMel Gorman 101b4597226SMel Gorman if (PageWriteback(page)) 102b4597226SMel Gorman *writeback = true; 103b4597226SMel Gorman 104b4597226SMel Gorman head = page_buffers(page); 105b4597226SMel Gorman bh = head; 106b4597226SMel Gorman do { 107b4597226SMel Gorman if (buffer_locked(bh)) 108b4597226SMel Gorman *writeback = true; 109b4597226SMel Gorman 110b4597226SMel Gorman if (buffer_dirty(bh)) 111b4597226SMel Gorman *dirty = true; 112b4597226SMel Gorman 113b4597226SMel Gorman bh = bh->b_this_page; 114b4597226SMel Gorman } while (bh != head); 115b4597226SMel Gorman } 116b4597226SMel Gorman EXPORT_SYMBOL(buffer_check_dirty_writeback); 117b4597226SMel Gorman 118b4597226SMel Gorman /* 1191da177e4SLinus Torvalds * Block until a buffer comes unlocked. This doesn't stop it 1201da177e4SLinus Torvalds * from becoming locked again - you have to lock it yourself 1211da177e4SLinus Torvalds * if you want to preserve its state. 1221da177e4SLinus Torvalds */ 1231da177e4SLinus Torvalds void __wait_on_buffer(struct buffer_head * bh) 1241da177e4SLinus Torvalds { 12574316201SNeilBrown wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); 1261da177e4SLinus Torvalds } 1271fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__wait_on_buffer); 1281da177e4SLinus Torvalds 1291da177e4SLinus Torvalds static void 1301da177e4SLinus Torvalds __clear_page_buffers(struct page *page) 1311da177e4SLinus Torvalds { 1321da177e4SLinus Torvalds ClearPagePrivate(page); 1334c21e2f2SHugh Dickins set_page_private(page, 0); 13409cbfeafSKirill A. Shutemov put_page(page); 1351da177e4SLinus Torvalds } 1361da177e4SLinus Torvalds 137b744c2acSRobert Elliott static void buffer_io_error(struct buffer_head *bh, char *msg) 1381da177e4SLinus Torvalds { 139432f16e6SRobert Elliott if (!test_bit(BH_Quiet, &bh->b_state)) 140432f16e6SRobert Elliott printk_ratelimited(KERN_ERR 141a1c6f057SDmitry Monakhov "Buffer I/O error on dev %pg, logical block %llu%s\n", 142a1c6f057SDmitry Monakhov bh->b_bdev, (unsigned long long)bh->b_blocknr, msg); 1431da177e4SLinus Torvalds } 1441da177e4SLinus Torvalds 1451da177e4SLinus Torvalds /* 14668671f35SDmitry Monakhov * End-of-IO handler helper function which does not touch the bh after 14768671f35SDmitry Monakhov * unlocking it. 14868671f35SDmitry Monakhov * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but 14968671f35SDmitry Monakhov * a race there is benign: unlock_buffer() only use the bh's address for 15068671f35SDmitry Monakhov * hashing after unlocking the buffer, so it doesn't actually touch the bh 15168671f35SDmitry Monakhov * itself. 1521da177e4SLinus Torvalds */ 15368671f35SDmitry Monakhov static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) 1541da177e4SLinus Torvalds { 1551da177e4SLinus Torvalds if (uptodate) { 1561da177e4SLinus Torvalds set_buffer_uptodate(bh); 1571da177e4SLinus Torvalds } else { 15870246286SChristoph Hellwig /* This happens, due to failed read-ahead attempts. */ 1591da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1601da177e4SLinus Torvalds } 1611da177e4SLinus Torvalds unlock_buffer(bh); 16268671f35SDmitry Monakhov } 16368671f35SDmitry Monakhov 16468671f35SDmitry Monakhov /* 16568671f35SDmitry Monakhov * Default synchronous end-of-IO handler.. Just mark it up-to-date and 16668671f35SDmitry Monakhov * unlock the buffer. This is what ll_rw_block uses too. 16768671f35SDmitry Monakhov */ 16868671f35SDmitry Monakhov void end_buffer_read_sync(struct buffer_head *bh, int uptodate) 16968671f35SDmitry Monakhov { 17068671f35SDmitry Monakhov __end_buffer_read_notouch(bh, uptodate); 1711da177e4SLinus Torvalds put_bh(bh); 1721da177e4SLinus Torvalds } 1731fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_read_sync); 1741da177e4SLinus Torvalds 1751da177e4SLinus Torvalds void end_buffer_write_sync(struct buffer_head *bh, int uptodate) 1761da177e4SLinus Torvalds { 1771da177e4SLinus Torvalds if (uptodate) { 1781da177e4SLinus Torvalds set_buffer_uptodate(bh); 1791da177e4SLinus Torvalds } else { 180b744c2acSRobert Elliott buffer_io_error(bh, ", lost sync page write"); 18187354e5dSJeff Layton mark_buffer_write_io_error(bh); 1821da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1831da177e4SLinus Torvalds } 1841da177e4SLinus Torvalds unlock_buffer(bh); 1851da177e4SLinus Torvalds put_bh(bh); 1861da177e4SLinus Torvalds } 1871fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_write_sync); 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds /* 1901da177e4SLinus Torvalds * Various filesystems appear to want __find_get_block to be non-blocking. 1911da177e4SLinus Torvalds * But it's the page lock which protects the buffers. To get around this, 1921da177e4SLinus Torvalds * we get exclusion from try_to_free_buffers with the blockdev mapping's 1931da177e4SLinus Torvalds * private_lock. 1941da177e4SLinus Torvalds * 1951da177e4SLinus Torvalds * Hack idea: for the blockdev mapping, i_bufferlist_lock contention 1961da177e4SLinus Torvalds * may be quite high. This code could TryLock the page, and if that 1971da177e4SLinus Torvalds * succeeds, there is no need to take private_lock. (But if 1981da177e4SLinus Torvalds * private_lock is contended then so is mapping->tree_lock). 1991da177e4SLinus Torvalds */ 2001da177e4SLinus Torvalds static struct buffer_head * 201385fd4c5SCoywolf Qi Hunt __find_get_block_slow(struct block_device *bdev, sector_t block) 2021da177e4SLinus Torvalds { 2031da177e4SLinus Torvalds struct inode *bd_inode = bdev->bd_inode; 2041da177e4SLinus Torvalds struct address_space *bd_mapping = bd_inode->i_mapping; 2051da177e4SLinus Torvalds struct buffer_head *ret = NULL; 2061da177e4SLinus Torvalds pgoff_t index; 2071da177e4SLinus Torvalds struct buffer_head *bh; 2081da177e4SLinus Torvalds struct buffer_head *head; 2091da177e4SLinus Torvalds struct page *page; 2101da177e4SLinus Torvalds int all_mapped = 1; 2111da177e4SLinus Torvalds 21209cbfeafSKirill A. Shutemov index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); 2132457aec6SMel Gorman page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED); 2141da177e4SLinus Torvalds if (!page) 2151da177e4SLinus Torvalds goto out; 2161da177e4SLinus Torvalds 2171da177e4SLinus Torvalds spin_lock(&bd_mapping->private_lock); 2181da177e4SLinus Torvalds if (!page_has_buffers(page)) 2191da177e4SLinus Torvalds goto out_unlock; 2201da177e4SLinus Torvalds head = page_buffers(page); 2211da177e4SLinus Torvalds bh = head; 2221da177e4SLinus Torvalds do { 22397f76d3dSNikanth Karthikesan if (!buffer_mapped(bh)) 22497f76d3dSNikanth Karthikesan all_mapped = 0; 22597f76d3dSNikanth Karthikesan else if (bh->b_blocknr == block) { 2261da177e4SLinus Torvalds ret = bh; 2271da177e4SLinus Torvalds get_bh(bh); 2281da177e4SLinus Torvalds goto out_unlock; 2291da177e4SLinus Torvalds } 2301da177e4SLinus Torvalds bh = bh->b_this_page; 2311da177e4SLinus Torvalds } while (bh != head); 2321da177e4SLinus Torvalds 2331da177e4SLinus Torvalds /* we might be here because some of the buffers on this page are 2341da177e4SLinus Torvalds * not mapped. This is due to various races between 2351da177e4SLinus Torvalds * file io on the block device and getblk. It gets dealt with 2361da177e4SLinus Torvalds * elsewhere, don't buffer_error if we had some unmapped buffers 2371da177e4SLinus Torvalds */ 2381da177e4SLinus Torvalds if (all_mapped) { 2391da177e4SLinus Torvalds printk("__find_get_block_slow() failed. " 2401da177e4SLinus Torvalds "block=%llu, b_blocknr=%llu\n", 241205f87f6SBadari Pulavarty (unsigned long long)block, 242205f87f6SBadari Pulavarty (unsigned long long)bh->b_blocknr); 243205f87f6SBadari Pulavarty printk("b_state=0x%08lx, b_size=%zu\n", 244205f87f6SBadari Pulavarty bh->b_state, bh->b_size); 245a1c6f057SDmitry Monakhov printk("device %pg blocksize: %d\n", bdev, 24672a2ebd8STao Ma 1 << bd_inode->i_blkbits); 2471da177e4SLinus Torvalds } 2481da177e4SLinus Torvalds out_unlock: 2491da177e4SLinus Torvalds spin_unlock(&bd_mapping->private_lock); 25009cbfeafSKirill A. Shutemov put_page(page); 2511da177e4SLinus Torvalds out: 2521da177e4SLinus Torvalds return ret; 2531da177e4SLinus Torvalds } 2541da177e4SLinus Torvalds 2551da177e4SLinus Torvalds /* 2565b0830cbSJens Axboe * Kick the writeback threads then try to free up some ZONE_NORMAL memory. 2571da177e4SLinus Torvalds */ 2581da177e4SLinus Torvalds static void free_more_memory(void) 2591da177e4SLinus Torvalds { 260c33d6c06SMel Gorman struct zoneref *z; 2610e88460dSMel Gorman int nid; 2621da177e4SLinus Torvalds 2630e175a18SCurt Wohlgemuth wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM); 2641da177e4SLinus Torvalds yield(); 2651da177e4SLinus Torvalds 2660e88460dSMel Gorman for_each_online_node(nid) { 267c33d6c06SMel Gorman 268c33d6c06SMel Gorman z = first_zones_zonelist(node_zonelist(nid, GFP_NOFS), 269c33d6c06SMel Gorman gfp_zone(GFP_NOFS), NULL); 270c33d6c06SMel Gorman if (z->zone) 27154a6eb5cSMel Gorman try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0, 272327c0e96SKAMEZAWA Hiroyuki GFP_NOFS, NULL); 2731da177e4SLinus Torvalds } 2741da177e4SLinus Torvalds } 2751da177e4SLinus Torvalds 2761da177e4SLinus Torvalds /* 2771da177e4SLinus Torvalds * I/O completion handler for block_read_full_page() - pages 2781da177e4SLinus Torvalds * which come unlocked at the end of I/O. 2791da177e4SLinus Torvalds */ 2801da177e4SLinus Torvalds static void end_buffer_async_read(struct buffer_head *bh, int uptodate) 2811da177e4SLinus Torvalds { 2821da177e4SLinus Torvalds unsigned long flags; 283a3972203SNick Piggin struct buffer_head *first; 2841da177e4SLinus Torvalds struct buffer_head *tmp; 2851da177e4SLinus Torvalds struct page *page; 2861da177e4SLinus Torvalds int page_uptodate = 1; 2871da177e4SLinus Torvalds 2881da177e4SLinus Torvalds BUG_ON(!buffer_async_read(bh)); 2891da177e4SLinus Torvalds 2901da177e4SLinus Torvalds page = bh->b_page; 2911da177e4SLinus Torvalds if (uptodate) { 2921da177e4SLinus Torvalds set_buffer_uptodate(bh); 2931da177e4SLinus Torvalds } else { 2941da177e4SLinus Torvalds clear_buffer_uptodate(bh); 295b744c2acSRobert Elliott buffer_io_error(bh, ", async page read"); 2961da177e4SLinus Torvalds SetPageError(page); 2971da177e4SLinus Torvalds } 2981da177e4SLinus Torvalds 2991da177e4SLinus Torvalds /* 3001da177e4SLinus Torvalds * Be _very_ careful from here on. Bad things can happen if 3011da177e4SLinus Torvalds * two buffer heads end IO at almost the same time and both 3021da177e4SLinus Torvalds * decide that the page is now completely done. 3031da177e4SLinus Torvalds */ 304a3972203SNick Piggin first = page_buffers(page); 305a3972203SNick Piggin local_irq_save(flags); 306a3972203SNick Piggin bit_spin_lock(BH_Uptodate_Lock, &first->b_state); 3071da177e4SLinus Torvalds clear_buffer_async_read(bh); 3081da177e4SLinus Torvalds unlock_buffer(bh); 3091da177e4SLinus Torvalds tmp = bh; 3101da177e4SLinus Torvalds do { 3111da177e4SLinus Torvalds if (!buffer_uptodate(tmp)) 3121da177e4SLinus Torvalds page_uptodate = 0; 3131da177e4SLinus Torvalds if (buffer_async_read(tmp)) { 3141da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 3151da177e4SLinus Torvalds goto still_busy; 3161da177e4SLinus Torvalds } 3171da177e4SLinus Torvalds tmp = tmp->b_this_page; 3181da177e4SLinus Torvalds } while (tmp != bh); 319a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 320a3972203SNick Piggin local_irq_restore(flags); 3211da177e4SLinus Torvalds 3221da177e4SLinus Torvalds /* 3231da177e4SLinus Torvalds * If none of the buffers had errors and they are all 3241da177e4SLinus Torvalds * uptodate then we can set the page uptodate. 3251da177e4SLinus Torvalds */ 3261da177e4SLinus Torvalds if (page_uptodate && !PageError(page)) 3271da177e4SLinus Torvalds SetPageUptodate(page); 3281da177e4SLinus Torvalds unlock_page(page); 3291da177e4SLinus Torvalds return; 3301da177e4SLinus Torvalds 3311da177e4SLinus Torvalds still_busy: 332a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 333a3972203SNick Piggin local_irq_restore(flags); 3341da177e4SLinus Torvalds return; 3351da177e4SLinus Torvalds } 3361da177e4SLinus Torvalds 3371da177e4SLinus Torvalds /* 3381da177e4SLinus Torvalds * Completion handler for block_write_full_page() - pages which are unlocked 3391da177e4SLinus Torvalds * during I/O, and which have PageWriteback cleared upon I/O completion. 3401da177e4SLinus Torvalds */ 34135c80d5fSChris Mason void end_buffer_async_write(struct buffer_head *bh, int uptodate) 3421da177e4SLinus Torvalds { 3431da177e4SLinus Torvalds unsigned long flags; 344a3972203SNick Piggin struct buffer_head *first; 3451da177e4SLinus Torvalds struct buffer_head *tmp; 3461da177e4SLinus Torvalds struct page *page; 3471da177e4SLinus Torvalds 3481da177e4SLinus Torvalds BUG_ON(!buffer_async_write(bh)); 3491da177e4SLinus Torvalds 3501da177e4SLinus Torvalds page = bh->b_page; 3511da177e4SLinus Torvalds if (uptodate) { 3521da177e4SLinus Torvalds set_buffer_uptodate(bh); 3531da177e4SLinus Torvalds } else { 354b744c2acSRobert Elliott buffer_io_error(bh, ", lost async page write"); 35587354e5dSJeff Layton mark_buffer_write_io_error(bh); 3561da177e4SLinus Torvalds clear_buffer_uptodate(bh); 3571da177e4SLinus Torvalds SetPageError(page); 3581da177e4SLinus Torvalds } 3591da177e4SLinus Torvalds 360a3972203SNick Piggin first = page_buffers(page); 361a3972203SNick Piggin local_irq_save(flags); 362a3972203SNick Piggin bit_spin_lock(BH_Uptodate_Lock, &first->b_state); 363a3972203SNick Piggin 3641da177e4SLinus Torvalds clear_buffer_async_write(bh); 3651da177e4SLinus Torvalds unlock_buffer(bh); 3661da177e4SLinus Torvalds tmp = bh->b_this_page; 3671da177e4SLinus Torvalds while (tmp != bh) { 3681da177e4SLinus Torvalds if (buffer_async_write(tmp)) { 3691da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 3701da177e4SLinus Torvalds goto still_busy; 3711da177e4SLinus Torvalds } 3721da177e4SLinus Torvalds tmp = tmp->b_this_page; 3731da177e4SLinus Torvalds } 374a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 375a3972203SNick Piggin local_irq_restore(flags); 3761da177e4SLinus Torvalds end_page_writeback(page); 3771da177e4SLinus Torvalds return; 3781da177e4SLinus Torvalds 3791da177e4SLinus Torvalds still_busy: 380a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 381a3972203SNick Piggin local_irq_restore(flags); 3821da177e4SLinus Torvalds return; 3831da177e4SLinus Torvalds } 3841fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_async_write); 3851da177e4SLinus Torvalds 3861da177e4SLinus Torvalds /* 3871da177e4SLinus Torvalds * If a page's buffers are under async readin (end_buffer_async_read 3881da177e4SLinus Torvalds * completion) then there is a possibility that another thread of 3891da177e4SLinus Torvalds * control could lock one of the buffers after it has completed 3901da177e4SLinus Torvalds * but while some of the other buffers have not completed. This 3911da177e4SLinus Torvalds * locked buffer would confuse end_buffer_async_read() into not unlocking 3921da177e4SLinus Torvalds * the page. So the absence of BH_Async_Read tells end_buffer_async_read() 3931da177e4SLinus Torvalds * that this buffer is not under async I/O. 3941da177e4SLinus Torvalds * 3951da177e4SLinus Torvalds * The page comes unlocked when it has no locked buffer_async buffers 3961da177e4SLinus Torvalds * left. 3971da177e4SLinus Torvalds * 3981da177e4SLinus Torvalds * PageLocked prevents anyone starting new async I/O reads any of 3991da177e4SLinus Torvalds * the buffers. 4001da177e4SLinus Torvalds * 4011da177e4SLinus Torvalds * PageWriteback is used to prevent simultaneous writeout of the same 4021da177e4SLinus Torvalds * page. 4031da177e4SLinus Torvalds * 4041da177e4SLinus Torvalds * PageLocked prevents anyone from starting writeback of a page which is 4051da177e4SLinus Torvalds * under read I/O (PageWriteback is only ever set against a locked page). 4061da177e4SLinus Torvalds */ 4071da177e4SLinus Torvalds static void mark_buffer_async_read(struct buffer_head *bh) 4081da177e4SLinus Torvalds { 4091da177e4SLinus Torvalds bh->b_end_io = end_buffer_async_read; 4101da177e4SLinus Torvalds set_buffer_async_read(bh); 4111da177e4SLinus Torvalds } 4121da177e4SLinus Torvalds 4131fe72eaaSH Hartley Sweeten static void mark_buffer_async_write_endio(struct buffer_head *bh, 41435c80d5fSChris Mason bh_end_io_t *handler) 41535c80d5fSChris Mason { 41635c80d5fSChris Mason bh->b_end_io = handler; 41735c80d5fSChris Mason set_buffer_async_write(bh); 41835c80d5fSChris Mason } 41935c80d5fSChris Mason 4201da177e4SLinus Torvalds void mark_buffer_async_write(struct buffer_head *bh) 4211da177e4SLinus Torvalds { 42235c80d5fSChris Mason mark_buffer_async_write_endio(bh, end_buffer_async_write); 4231da177e4SLinus Torvalds } 4241da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_async_write); 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds 4271da177e4SLinus Torvalds /* 4281da177e4SLinus Torvalds * fs/buffer.c contains helper functions for buffer-backed address space's 4291da177e4SLinus Torvalds * fsync functions. A common requirement for buffer-based filesystems is 4301da177e4SLinus Torvalds * that certain data from the backing blockdev needs to be written out for 4311da177e4SLinus Torvalds * a successful fsync(). For example, ext2 indirect blocks need to be 4321da177e4SLinus Torvalds * written back and waited upon before fsync() returns. 4331da177e4SLinus Torvalds * 4341da177e4SLinus Torvalds * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), 4351da177e4SLinus Torvalds * inode_has_buffers() and invalidate_inode_buffers() are provided for the 4361da177e4SLinus Torvalds * management of a list of dependent buffers at ->i_mapping->private_list. 4371da177e4SLinus Torvalds * 4381da177e4SLinus Torvalds * Locking is a little subtle: try_to_free_buffers() will remove buffers 4391da177e4SLinus Torvalds * from their controlling inode's queue when they are being freed. But 4401da177e4SLinus Torvalds * try_to_free_buffers() will be operating against the *blockdev* mapping 4411da177e4SLinus Torvalds * at the time, not against the S_ISREG file which depends on those buffers. 4421da177e4SLinus Torvalds * So the locking for private_list is via the private_lock in the address_space 4431da177e4SLinus Torvalds * which backs the buffers. Which is different from the address_space 4441da177e4SLinus Torvalds * against which the buffers are listed. So for a particular address_space, 4451da177e4SLinus Torvalds * mapping->private_lock does *not* protect mapping->private_list! In fact, 4461da177e4SLinus Torvalds * mapping->private_list will always be protected by the backing blockdev's 4471da177e4SLinus Torvalds * ->private_lock. 4481da177e4SLinus Torvalds * 4491da177e4SLinus Torvalds * Which introduces a requirement: all buffers on an address_space's 4501da177e4SLinus Torvalds * ->private_list must be from the same address_space: the blockdev's. 4511da177e4SLinus Torvalds * 4521da177e4SLinus Torvalds * address_spaces which do not place buffers at ->private_list via these 4531da177e4SLinus Torvalds * utility functions are free to use private_lock and private_list for 4541da177e4SLinus Torvalds * whatever they want. The only requirement is that list_empty(private_list) 4551da177e4SLinus Torvalds * be true at clear_inode() time. 4561da177e4SLinus Torvalds * 4571da177e4SLinus Torvalds * FIXME: clear_inode should not call invalidate_inode_buffers(). The 4581da177e4SLinus Torvalds * filesystems should do that. invalidate_inode_buffers() should just go 4591da177e4SLinus Torvalds * BUG_ON(!list_empty). 4601da177e4SLinus Torvalds * 4611da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should 4621da177e4SLinus Torvalds * take an address_space, not an inode. And it should be called 4631da177e4SLinus Torvalds * mark_buffer_dirty_fsync() to clearly define why those buffers are being 4641da177e4SLinus Torvalds * queued up. 4651da177e4SLinus Torvalds * 4661da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the 4671da177e4SLinus Torvalds * list if it is already on a list. Because if the buffer is on a list, 4681da177e4SLinus Torvalds * it *must* already be on the right one. If not, the filesystem is being 4691da177e4SLinus Torvalds * silly. This will save a ton of locking. But first we have to ensure 4701da177e4SLinus Torvalds * that buffers are taken *off* the old inode's list when they are freed 4711da177e4SLinus Torvalds * (presumably in truncate). That requires careful auditing of all 4721da177e4SLinus Torvalds * filesystems (do it inside bforget()). It could also be done by bringing 4731da177e4SLinus Torvalds * b_inode back. 4741da177e4SLinus Torvalds */ 4751da177e4SLinus Torvalds 4761da177e4SLinus Torvalds /* 4771da177e4SLinus Torvalds * The buffer's backing address_space's private_lock must be held 4781da177e4SLinus Torvalds */ 479dbacefc9SThomas Petazzoni static void __remove_assoc_queue(struct buffer_head *bh) 4801da177e4SLinus Torvalds { 4811da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 48258ff407bSJan Kara WARN_ON(!bh->b_assoc_map); 48358ff407bSJan Kara bh->b_assoc_map = NULL; 4841da177e4SLinus Torvalds } 4851da177e4SLinus Torvalds 4861da177e4SLinus Torvalds int inode_has_buffers(struct inode *inode) 4871da177e4SLinus Torvalds { 4881da177e4SLinus Torvalds return !list_empty(&inode->i_data.private_list); 4891da177e4SLinus Torvalds } 4901da177e4SLinus Torvalds 4911da177e4SLinus Torvalds /* 4921da177e4SLinus Torvalds * osync is designed to support O_SYNC io. It waits synchronously for 4931da177e4SLinus Torvalds * all already-submitted IO to complete, but does not queue any new 4941da177e4SLinus Torvalds * writes to the disk. 4951da177e4SLinus Torvalds * 4961da177e4SLinus Torvalds * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as 4971da177e4SLinus Torvalds * you dirty the buffers, and then use osync_inode_buffers to wait for 4981da177e4SLinus Torvalds * completion. Any other dirty buffers which are not yet queued for 4991da177e4SLinus Torvalds * write will not be flushed to disk by the osync. 5001da177e4SLinus Torvalds */ 5011da177e4SLinus Torvalds static int osync_buffers_list(spinlock_t *lock, struct list_head *list) 5021da177e4SLinus Torvalds { 5031da177e4SLinus Torvalds struct buffer_head *bh; 5041da177e4SLinus Torvalds struct list_head *p; 5051da177e4SLinus Torvalds int err = 0; 5061da177e4SLinus Torvalds 5071da177e4SLinus Torvalds spin_lock(lock); 5081da177e4SLinus Torvalds repeat: 5091da177e4SLinus Torvalds list_for_each_prev(p, list) { 5101da177e4SLinus Torvalds bh = BH_ENTRY(p); 5111da177e4SLinus Torvalds if (buffer_locked(bh)) { 5121da177e4SLinus Torvalds get_bh(bh); 5131da177e4SLinus Torvalds spin_unlock(lock); 5141da177e4SLinus Torvalds wait_on_buffer(bh); 5151da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 5161da177e4SLinus Torvalds err = -EIO; 5171da177e4SLinus Torvalds brelse(bh); 5181da177e4SLinus Torvalds spin_lock(lock); 5191da177e4SLinus Torvalds goto repeat; 5201da177e4SLinus Torvalds } 5211da177e4SLinus Torvalds } 5221da177e4SLinus Torvalds spin_unlock(lock); 5231da177e4SLinus Torvalds return err; 5241da177e4SLinus Torvalds } 5251da177e4SLinus Torvalds 52601a05b33SAl Viro static void do_thaw_one(struct super_block *sb, void *unused) 527c2d75438SEric Sandeen { 528c2d75438SEric Sandeen while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb)) 529a1c6f057SDmitry Monakhov printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev); 530c2d75438SEric Sandeen } 53101a05b33SAl Viro 53201a05b33SAl Viro static void do_thaw_all(struct work_struct *work) 53301a05b33SAl Viro { 53401a05b33SAl Viro iterate_supers(do_thaw_one, NULL); 535053c525fSJens Axboe kfree(work); 536c2d75438SEric Sandeen printk(KERN_WARNING "Emergency Thaw complete\n"); 537c2d75438SEric Sandeen } 538c2d75438SEric Sandeen 539c2d75438SEric Sandeen /** 540c2d75438SEric Sandeen * emergency_thaw_all -- forcibly thaw every frozen filesystem 541c2d75438SEric Sandeen * 542c2d75438SEric Sandeen * Used for emergency unfreeze of all filesystems via SysRq 543c2d75438SEric Sandeen */ 544c2d75438SEric Sandeen void emergency_thaw_all(void) 545c2d75438SEric Sandeen { 546053c525fSJens Axboe struct work_struct *work; 547053c525fSJens Axboe 548053c525fSJens Axboe work = kmalloc(sizeof(*work), GFP_ATOMIC); 549053c525fSJens Axboe if (work) { 550053c525fSJens Axboe INIT_WORK(work, do_thaw_all); 551053c525fSJens Axboe schedule_work(work); 552053c525fSJens Axboe } 553c2d75438SEric Sandeen } 554c2d75438SEric Sandeen 5551da177e4SLinus Torvalds /** 55678a4a50aSRandy Dunlap * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers 55767be2dd1SMartin Waitz * @mapping: the mapping which wants those buffers written 5581da177e4SLinus Torvalds * 5591da177e4SLinus Torvalds * Starts I/O against the buffers at mapping->private_list, and waits upon 5601da177e4SLinus Torvalds * that I/O. 5611da177e4SLinus Torvalds * 56267be2dd1SMartin Waitz * Basically, this is a convenience function for fsync(). 56367be2dd1SMartin Waitz * @mapping is a file or directory which needs those buffers to be written for 56467be2dd1SMartin Waitz * a successful fsync(). 5651da177e4SLinus Torvalds */ 5661da177e4SLinus Torvalds int sync_mapping_buffers(struct address_space *mapping) 5671da177e4SLinus Torvalds { 568252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 5691da177e4SLinus Torvalds 5701da177e4SLinus Torvalds if (buffer_mapping == NULL || list_empty(&mapping->private_list)) 5711da177e4SLinus Torvalds return 0; 5721da177e4SLinus Torvalds 5731da177e4SLinus Torvalds return fsync_buffers_list(&buffer_mapping->private_lock, 5741da177e4SLinus Torvalds &mapping->private_list); 5751da177e4SLinus Torvalds } 5761da177e4SLinus Torvalds EXPORT_SYMBOL(sync_mapping_buffers); 5771da177e4SLinus Torvalds 5781da177e4SLinus Torvalds /* 5791da177e4SLinus Torvalds * Called when we've recently written block `bblock', and it is known that 5801da177e4SLinus Torvalds * `bblock' was for a buffer_boundary() buffer. This means that the block at 5811da177e4SLinus Torvalds * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's 5821da177e4SLinus Torvalds * dirty, schedule it for IO. So that indirects merge nicely with their data. 5831da177e4SLinus Torvalds */ 5841da177e4SLinus Torvalds void write_boundary_block(struct block_device *bdev, 5851da177e4SLinus Torvalds sector_t bblock, unsigned blocksize) 5861da177e4SLinus Torvalds { 5871da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); 5881da177e4SLinus Torvalds if (bh) { 5891da177e4SLinus Torvalds if (buffer_dirty(bh)) 590dfec8a14SMike Christie ll_rw_block(REQ_OP_WRITE, 0, 1, &bh); 5911da177e4SLinus Torvalds put_bh(bh); 5921da177e4SLinus Torvalds } 5931da177e4SLinus Torvalds } 5941da177e4SLinus Torvalds 5951da177e4SLinus Torvalds void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) 5961da177e4SLinus Torvalds { 5971da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 5981da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 5991da177e4SLinus Torvalds 6001da177e4SLinus Torvalds mark_buffer_dirty(bh); 601252aa6f5SRafael Aquini if (!mapping->private_data) { 602252aa6f5SRafael Aquini mapping->private_data = buffer_mapping; 6031da177e4SLinus Torvalds } else { 604252aa6f5SRafael Aquini BUG_ON(mapping->private_data != buffer_mapping); 6051da177e4SLinus Torvalds } 606535ee2fbSJan Kara if (!bh->b_assoc_map) { 6071da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 6081da177e4SLinus Torvalds list_move_tail(&bh->b_assoc_buffers, 6091da177e4SLinus Torvalds &mapping->private_list); 61058ff407bSJan Kara bh->b_assoc_map = mapping; 6111da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 6121da177e4SLinus Torvalds } 6131da177e4SLinus Torvalds } 6141da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_dirty_inode); 6151da177e4SLinus Torvalds 6161da177e4SLinus Torvalds /* 617787d2214SNick Piggin * Mark the page dirty, and set it dirty in the radix tree, and mark the inode 618787d2214SNick Piggin * dirty. 619787d2214SNick Piggin * 620787d2214SNick Piggin * If warn is true, then emit a warning if the page is not uptodate and has 621787d2214SNick Piggin * not been truncated. 622c4843a75SGreg Thelen * 62381f8c3a4SJohannes Weiner * The caller must hold lock_page_memcg(). 624787d2214SNick Piggin */ 625c4843a75SGreg Thelen static void __set_page_dirty(struct page *page, struct address_space *mapping, 62662cccb8cSJohannes Weiner int warn) 627787d2214SNick Piggin { 628227d53b3SKOSAKI Motohiro unsigned long flags; 629227d53b3SKOSAKI Motohiro 630227d53b3SKOSAKI Motohiro spin_lock_irqsave(&mapping->tree_lock, flags); 631787d2214SNick Piggin if (page->mapping) { /* Race with truncate? */ 632787d2214SNick Piggin WARN_ON_ONCE(warn && !PageUptodate(page)); 63362cccb8cSJohannes Weiner account_page_dirtied(page, mapping); 634787d2214SNick Piggin radix_tree_tag_set(&mapping->page_tree, 635787d2214SNick Piggin page_index(page), PAGECACHE_TAG_DIRTY); 636787d2214SNick Piggin } 637227d53b3SKOSAKI Motohiro spin_unlock_irqrestore(&mapping->tree_lock, flags); 638787d2214SNick Piggin } 639787d2214SNick Piggin 640787d2214SNick Piggin /* 6411da177e4SLinus Torvalds * Add a page to the dirty page list. 6421da177e4SLinus Torvalds * 6431da177e4SLinus Torvalds * It is a sad fact of life that this function is called from several places 6441da177e4SLinus Torvalds * deeply under spinlocking. It may not sleep. 6451da177e4SLinus Torvalds * 6461da177e4SLinus Torvalds * If the page has buffers, the uptodate buffers are set dirty, to preserve 6471da177e4SLinus Torvalds * dirty-state coherency between the page and the buffers. It the page does 6481da177e4SLinus Torvalds * not have buffers then when they are later attached they will all be set 6491da177e4SLinus Torvalds * dirty. 6501da177e4SLinus Torvalds * 6511da177e4SLinus Torvalds * The buffers are dirtied before the page is dirtied. There's a small race 6521da177e4SLinus Torvalds * window in which a writepage caller may see the page cleanness but not the 6531da177e4SLinus Torvalds * buffer dirtiness. That's fine. If this code were to set the page dirty 6541da177e4SLinus Torvalds * before the buffers, a concurrent writepage caller could clear the page dirty 6551da177e4SLinus Torvalds * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean 6561da177e4SLinus Torvalds * page on the dirty page list. 6571da177e4SLinus Torvalds * 6581da177e4SLinus Torvalds * We use private_lock to lock against try_to_free_buffers while using the 6591da177e4SLinus Torvalds * page's buffer list. Also use this to protect against clean buffers being 6601da177e4SLinus Torvalds * added to the page after it was set dirty. 6611da177e4SLinus Torvalds * 6621da177e4SLinus Torvalds * FIXME: may need to call ->reservepage here as well. That's rather up to the 6631da177e4SLinus Torvalds * address_space though. 6641da177e4SLinus Torvalds */ 6651da177e4SLinus Torvalds int __set_page_dirty_buffers(struct page *page) 6661da177e4SLinus Torvalds { 667a8e7d49aSLinus Torvalds int newly_dirty; 668787d2214SNick Piggin struct address_space *mapping = page_mapping(page); 669ebf7a227SNick Piggin 670ebf7a227SNick Piggin if (unlikely(!mapping)) 671ebf7a227SNick Piggin return !TestSetPageDirty(page); 6721da177e4SLinus Torvalds 6731da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 6741da177e4SLinus Torvalds if (page_has_buffers(page)) { 6751da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 6761da177e4SLinus Torvalds struct buffer_head *bh = head; 6771da177e4SLinus Torvalds 6781da177e4SLinus Torvalds do { 6791da177e4SLinus Torvalds set_buffer_dirty(bh); 6801da177e4SLinus Torvalds bh = bh->b_this_page; 6811da177e4SLinus Torvalds } while (bh != head); 6821da177e4SLinus Torvalds } 683c4843a75SGreg Thelen /* 68481f8c3a4SJohannes Weiner * Lock out page->mem_cgroup migration to keep PageDirty 68581f8c3a4SJohannes Weiner * synchronized with per-memcg dirty page counters. 686c4843a75SGreg Thelen */ 68762cccb8cSJohannes Weiner lock_page_memcg(page); 688a8e7d49aSLinus Torvalds newly_dirty = !TestSetPageDirty(page); 6891da177e4SLinus Torvalds spin_unlock(&mapping->private_lock); 6901da177e4SLinus Torvalds 691a8e7d49aSLinus Torvalds if (newly_dirty) 69262cccb8cSJohannes Weiner __set_page_dirty(page, mapping, 1); 693c4843a75SGreg Thelen 69462cccb8cSJohannes Weiner unlock_page_memcg(page); 695c4843a75SGreg Thelen 696c4843a75SGreg Thelen if (newly_dirty) 697c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 698c4843a75SGreg Thelen 699a8e7d49aSLinus Torvalds return newly_dirty; 7001da177e4SLinus Torvalds } 7011da177e4SLinus Torvalds EXPORT_SYMBOL(__set_page_dirty_buffers); 7021da177e4SLinus Torvalds 7031da177e4SLinus Torvalds /* 7041da177e4SLinus Torvalds * Write out and wait upon a list of buffers. 7051da177e4SLinus Torvalds * 7061da177e4SLinus Torvalds * We have conflicting pressures: we want to make sure that all 7071da177e4SLinus Torvalds * initially dirty buffers get waited on, but that any subsequently 7081da177e4SLinus Torvalds * dirtied buffers don't. After all, we don't want fsync to last 7091da177e4SLinus Torvalds * forever if somebody is actively writing to the file. 7101da177e4SLinus Torvalds * 7111da177e4SLinus Torvalds * Do this in two main stages: first we copy dirty buffers to a 7121da177e4SLinus Torvalds * temporary inode list, queueing the writes as we go. Then we clean 7131da177e4SLinus Torvalds * up, waiting for those writes to complete. 7141da177e4SLinus Torvalds * 7151da177e4SLinus Torvalds * During this second stage, any subsequent updates to the file may end 7161da177e4SLinus Torvalds * up refiling the buffer on the original inode's dirty list again, so 7171da177e4SLinus Torvalds * there is a chance we will end up with a buffer queued for write but 7181da177e4SLinus Torvalds * not yet completed on that list. So, as a final cleanup we go through 7191da177e4SLinus Torvalds * the osync code to catch these locked, dirty buffers without requeuing 7201da177e4SLinus Torvalds * any newly dirty buffers for write. 7211da177e4SLinus Torvalds */ 7221da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) 7231da177e4SLinus Torvalds { 7241da177e4SLinus Torvalds struct buffer_head *bh; 7251da177e4SLinus Torvalds struct list_head tmp; 7267eaceaccSJens Axboe struct address_space *mapping; 7271da177e4SLinus Torvalds int err = 0, err2; 7284ee2491eSJens Axboe struct blk_plug plug; 7291da177e4SLinus Torvalds 7301da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 7314ee2491eSJens Axboe blk_start_plug(&plug); 7321da177e4SLinus Torvalds 7331da177e4SLinus Torvalds spin_lock(lock); 7341da177e4SLinus Torvalds while (!list_empty(list)) { 7351da177e4SLinus Torvalds bh = BH_ENTRY(list->next); 736535ee2fbSJan Kara mapping = bh->b_assoc_map; 73758ff407bSJan Kara __remove_assoc_queue(bh); 738535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 739535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 740535ee2fbSJan Kara smp_mb(); 7411da177e4SLinus Torvalds if (buffer_dirty(bh) || buffer_locked(bh)) { 7421da177e4SLinus Torvalds list_add(&bh->b_assoc_buffers, &tmp); 743535ee2fbSJan Kara bh->b_assoc_map = mapping; 7441da177e4SLinus Torvalds if (buffer_dirty(bh)) { 7451da177e4SLinus Torvalds get_bh(bh); 7461da177e4SLinus Torvalds spin_unlock(lock); 7471da177e4SLinus Torvalds /* 7481da177e4SLinus Torvalds * Ensure any pending I/O completes so that 7499cb569d6SChristoph Hellwig * write_dirty_buffer() actually writes the 7509cb569d6SChristoph Hellwig * current contents - it is a noop if I/O is 7519cb569d6SChristoph Hellwig * still in flight on potentially older 7529cb569d6SChristoph Hellwig * contents. 7531da177e4SLinus Torvalds */ 75470fd7614SChristoph Hellwig write_dirty_buffer(bh, REQ_SYNC); 7559cf6b720SJens Axboe 7569cf6b720SJens Axboe /* 7579cf6b720SJens Axboe * Kick off IO for the previous mapping. Note 7589cf6b720SJens Axboe * that we will not run the very last mapping, 7599cf6b720SJens Axboe * wait_on_buffer() will do that for us 7609cf6b720SJens Axboe * through sync_buffer(). 7619cf6b720SJens Axboe */ 7621da177e4SLinus Torvalds brelse(bh); 7631da177e4SLinus Torvalds spin_lock(lock); 7641da177e4SLinus Torvalds } 7651da177e4SLinus Torvalds } 7661da177e4SLinus Torvalds } 7671da177e4SLinus Torvalds 7684ee2491eSJens Axboe spin_unlock(lock); 7694ee2491eSJens Axboe blk_finish_plug(&plug); 7704ee2491eSJens Axboe spin_lock(lock); 7714ee2491eSJens Axboe 7721da177e4SLinus Torvalds while (!list_empty(&tmp)) { 7731da177e4SLinus Torvalds bh = BH_ENTRY(tmp.prev); 7741da177e4SLinus Torvalds get_bh(bh); 775535ee2fbSJan Kara mapping = bh->b_assoc_map; 776535ee2fbSJan Kara __remove_assoc_queue(bh); 777535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 778535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 779535ee2fbSJan Kara smp_mb(); 780535ee2fbSJan Kara if (buffer_dirty(bh)) { 781535ee2fbSJan Kara list_add(&bh->b_assoc_buffers, 782e3892296SJan Kara &mapping->private_list); 783535ee2fbSJan Kara bh->b_assoc_map = mapping; 784535ee2fbSJan Kara } 7851da177e4SLinus Torvalds spin_unlock(lock); 7861da177e4SLinus Torvalds wait_on_buffer(bh); 7871da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 7881da177e4SLinus Torvalds err = -EIO; 7891da177e4SLinus Torvalds brelse(bh); 7901da177e4SLinus Torvalds spin_lock(lock); 7911da177e4SLinus Torvalds } 7921da177e4SLinus Torvalds 7931da177e4SLinus Torvalds spin_unlock(lock); 7941da177e4SLinus Torvalds err2 = osync_buffers_list(lock, list); 7951da177e4SLinus Torvalds if (err) 7961da177e4SLinus Torvalds return err; 7971da177e4SLinus Torvalds else 7981da177e4SLinus Torvalds return err2; 7991da177e4SLinus Torvalds } 8001da177e4SLinus Torvalds 8011da177e4SLinus Torvalds /* 8021da177e4SLinus Torvalds * Invalidate any and all dirty buffers on a given inode. We are 8031da177e4SLinus Torvalds * probably unmounting the fs, but that doesn't mean we have already 8041da177e4SLinus Torvalds * done a sync(). Just drop the buffers from the inode list. 8051da177e4SLinus Torvalds * 8061da177e4SLinus Torvalds * NOTE: we take the inode's blockdev's mapping's private_lock. Which 8071da177e4SLinus Torvalds * assumes that all the buffers are against the blockdev. Not true 8081da177e4SLinus Torvalds * for reiserfs. 8091da177e4SLinus Torvalds */ 8101da177e4SLinus Torvalds void invalidate_inode_buffers(struct inode *inode) 8111da177e4SLinus Torvalds { 8121da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8131da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 8141da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 815252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 8161da177e4SLinus Torvalds 8171da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8181da177e4SLinus Torvalds while (!list_empty(list)) 8191da177e4SLinus Torvalds __remove_assoc_queue(BH_ENTRY(list->next)); 8201da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8211da177e4SLinus Torvalds } 8221da177e4SLinus Torvalds } 82352b19ac9SJan Kara EXPORT_SYMBOL(invalidate_inode_buffers); 8241da177e4SLinus Torvalds 8251da177e4SLinus Torvalds /* 8261da177e4SLinus Torvalds * Remove any clean buffers from the inode's buffer list. This is called 8271da177e4SLinus Torvalds * when we're trying to free the inode itself. Those buffers can pin it. 8281da177e4SLinus Torvalds * 8291da177e4SLinus Torvalds * Returns true if all buffers were removed. 8301da177e4SLinus Torvalds */ 8311da177e4SLinus Torvalds int remove_inode_buffers(struct inode *inode) 8321da177e4SLinus Torvalds { 8331da177e4SLinus Torvalds int ret = 1; 8341da177e4SLinus Torvalds 8351da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8361da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 8371da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 838252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 8391da177e4SLinus Torvalds 8401da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8411da177e4SLinus Torvalds while (!list_empty(list)) { 8421da177e4SLinus Torvalds struct buffer_head *bh = BH_ENTRY(list->next); 8431da177e4SLinus Torvalds if (buffer_dirty(bh)) { 8441da177e4SLinus Torvalds ret = 0; 8451da177e4SLinus Torvalds break; 8461da177e4SLinus Torvalds } 8471da177e4SLinus Torvalds __remove_assoc_queue(bh); 8481da177e4SLinus Torvalds } 8491da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8501da177e4SLinus Torvalds } 8511da177e4SLinus Torvalds return ret; 8521da177e4SLinus Torvalds } 8531da177e4SLinus Torvalds 8541da177e4SLinus Torvalds /* 8551da177e4SLinus Torvalds * Create the appropriate buffers when given a page for data area and 8561da177e4SLinus Torvalds * the size of each buffer.. Use the bh->b_this_page linked list to 8571da177e4SLinus Torvalds * follow the buffers created. Return NULL if unable to create more 8581da177e4SLinus Torvalds * buffers. 8591da177e4SLinus Torvalds * 8601da177e4SLinus Torvalds * The retry flag is used to differentiate async IO (paging, swapping) 8611da177e4SLinus Torvalds * which may not fail from ordinary buffer allocations. 8621da177e4SLinus Torvalds */ 8631da177e4SLinus Torvalds struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, 864640ab98fSJens Axboe bool retry) 8651da177e4SLinus Torvalds { 8661da177e4SLinus Torvalds struct buffer_head *bh, *head; 867640ab98fSJens Axboe gfp_t gfp = GFP_NOFS; 8681da177e4SLinus Torvalds long offset; 8691da177e4SLinus Torvalds 870640ab98fSJens Axboe if (retry) 871640ab98fSJens Axboe gfp |= __GFP_NOFAIL; 872640ab98fSJens Axboe 8731da177e4SLinus Torvalds head = NULL; 8741da177e4SLinus Torvalds offset = PAGE_SIZE; 8751da177e4SLinus Torvalds while ((offset -= size) >= 0) { 876640ab98fSJens Axboe bh = alloc_buffer_head(gfp); 8771da177e4SLinus Torvalds if (!bh) 8781da177e4SLinus Torvalds goto no_grow; 8791da177e4SLinus Torvalds 8801da177e4SLinus Torvalds bh->b_this_page = head; 8811da177e4SLinus Torvalds bh->b_blocknr = -1; 8821da177e4SLinus Torvalds head = bh; 8831da177e4SLinus Torvalds 8841da177e4SLinus Torvalds bh->b_size = size; 8851da177e4SLinus Torvalds 8861da177e4SLinus Torvalds /* Link the buffer to its page */ 8871da177e4SLinus Torvalds set_bh_page(bh, page, offset); 8881da177e4SLinus Torvalds } 8891da177e4SLinus Torvalds return head; 8901da177e4SLinus Torvalds /* 8911da177e4SLinus Torvalds * In case anything failed, we just free everything we got. 8921da177e4SLinus Torvalds */ 8931da177e4SLinus Torvalds no_grow: 8941da177e4SLinus Torvalds if (head) { 8951da177e4SLinus Torvalds do { 8961da177e4SLinus Torvalds bh = head; 8971da177e4SLinus Torvalds head = head->b_this_page; 8981da177e4SLinus Torvalds free_buffer_head(bh); 8991da177e4SLinus Torvalds } while (head); 9001da177e4SLinus Torvalds } 9011da177e4SLinus Torvalds 9021da177e4SLinus Torvalds return NULL; 9031da177e4SLinus Torvalds } 9041da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(alloc_page_buffers); 9051da177e4SLinus Torvalds 9061da177e4SLinus Torvalds static inline void 9071da177e4SLinus Torvalds link_dev_buffers(struct page *page, struct buffer_head *head) 9081da177e4SLinus Torvalds { 9091da177e4SLinus Torvalds struct buffer_head *bh, *tail; 9101da177e4SLinus Torvalds 9111da177e4SLinus Torvalds bh = head; 9121da177e4SLinus Torvalds do { 9131da177e4SLinus Torvalds tail = bh; 9141da177e4SLinus Torvalds bh = bh->b_this_page; 9151da177e4SLinus Torvalds } while (bh); 9161da177e4SLinus Torvalds tail->b_this_page = head; 9171da177e4SLinus Torvalds attach_page_buffers(page, head); 9181da177e4SLinus Torvalds } 9191da177e4SLinus Torvalds 920bbec0270SLinus Torvalds static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) 921bbec0270SLinus Torvalds { 922bbec0270SLinus Torvalds sector_t retval = ~((sector_t)0); 923bbec0270SLinus Torvalds loff_t sz = i_size_read(bdev->bd_inode); 924bbec0270SLinus Torvalds 925bbec0270SLinus Torvalds if (sz) { 926bbec0270SLinus Torvalds unsigned int sizebits = blksize_bits(size); 927bbec0270SLinus Torvalds retval = (sz >> sizebits); 928bbec0270SLinus Torvalds } 929bbec0270SLinus Torvalds return retval; 930bbec0270SLinus Torvalds } 931bbec0270SLinus Torvalds 9321da177e4SLinus Torvalds /* 9331da177e4SLinus Torvalds * Initialise the state of a blockdev page's buffers. 9341da177e4SLinus Torvalds */ 935676ce6d5SHugh Dickins static sector_t 9361da177e4SLinus Torvalds init_page_buffers(struct page *page, struct block_device *bdev, 9371da177e4SLinus Torvalds sector_t block, int size) 9381da177e4SLinus Torvalds { 9391da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 9401da177e4SLinus Torvalds struct buffer_head *bh = head; 9411da177e4SLinus Torvalds int uptodate = PageUptodate(page); 942bbec0270SLinus Torvalds sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size); 9431da177e4SLinus Torvalds 9441da177e4SLinus Torvalds do { 9451da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 9461da177e4SLinus Torvalds init_buffer(bh, NULL, NULL); 9471da177e4SLinus Torvalds bh->b_bdev = bdev; 9481da177e4SLinus Torvalds bh->b_blocknr = block; 9491da177e4SLinus Torvalds if (uptodate) 9501da177e4SLinus Torvalds set_buffer_uptodate(bh); 951080399aaSJeff Moyer if (block < end_block) 9521da177e4SLinus Torvalds set_buffer_mapped(bh); 9531da177e4SLinus Torvalds } 9541da177e4SLinus Torvalds block++; 9551da177e4SLinus Torvalds bh = bh->b_this_page; 9561da177e4SLinus Torvalds } while (bh != head); 957676ce6d5SHugh Dickins 958676ce6d5SHugh Dickins /* 959676ce6d5SHugh Dickins * Caller needs to validate requested block against end of device. 960676ce6d5SHugh Dickins */ 961676ce6d5SHugh Dickins return end_block; 9621da177e4SLinus Torvalds } 9631da177e4SLinus Torvalds 9641da177e4SLinus Torvalds /* 9651da177e4SLinus Torvalds * Create the page-cache page that contains the requested block. 9661da177e4SLinus Torvalds * 967676ce6d5SHugh Dickins * This is used purely for blockdev mappings. 9681da177e4SLinus Torvalds */ 969676ce6d5SHugh Dickins static int 9701da177e4SLinus Torvalds grow_dev_page(struct block_device *bdev, sector_t block, 9713b5e6454SGioh Kim pgoff_t index, int size, int sizebits, gfp_t gfp) 9721da177e4SLinus Torvalds { 9731da177e4SLinus Torvalds struct inode *inode = bdev->bd_inode; 9741da177e4SLinus Torvalds struct page *page; 9751da177e4SLinus Torvalds struct buffer_head *bh; 976676ce6d5SHugh Dickins sector_t end_block; 977676ce6d5SHugh Dickins int ret = 0; /* Will call free_more_memory() */ 97884235de3SJohannes Weiner gfp_t gfp_mask; 9791da177e4SLinus Torvalds 980c62d2555SMichal Hocko gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp; 9813b5e6454SGioh Kim 98284235de3SJohannes Weiner /* 98384235de3SJohannes Weiner * XXX: __getblk_slow() can not really deal with failure and 98484235de3SJohannes Weiner * will endlessly loop on improvised global reclaim. Prefer 98584235de3SJohannes Weiner * looping in the allocator rather than here, at least that 98684235de3SJohannes Weiner * code knows what it's doing. 98784235de3SJohannes Weiner */ 98884235de3SJohannes Weiner gfp_mask |= __GFP_NOFAIL; 98984235de3SJohannes Weiner 99084235de3SJohannes Weiner page = find_or_create_page(inode->i_mapping, index, gfp_mask); 9911da177e4SLinus Torvalds 992e827f923SEric Sesterhenn BUG_ON(!PageLocked(page)); 9931da177e4SLinus Torvalds 9941da177e4SLinus Torvalds if (page_has_buffers(page)) { 9951da177e4SLinus Torvalds bh = page_buffers(page); 9961da177e4SLinus Torvalds if (bh->b_size == size) { 997676ce6d5SHugh Dickins end_block = init_page_buffers(page, bdev, 998f2d5a944SAnton Altaparmakov (sector_t)index << sizebits, 999f2d5a944SAnton Altaparmakov size); 1000676ce6d5SHugh Dickins goto done; 10011da177e4SLinus Torvalds } 10021da177e4SLinus Torvalds if (!try_to_free_buffers(page)) 10031da177e4SLinus Torvalds goto failed; 10041da177e4SLinus Torvalds } 10051da177e4SLinus Torvalds 10061da177e4SLinus Torvalds /* 10071da177e4SLinus Torvalds * Allocate some buffers for this page 10081da177e4SLinus Torvalds */ 1009*94dc24c0SJens Axboe bh = alloc_page_buffers(page, size, true); 10101da177e4SLinus Torvalds 10111da177e4SLinus Torvalds /* 10121da177e4SLinus Torvalds * Link the page to the buffers and initialise them. Take the 10131da177e4SLinus Torvalds * lock to be atomic wrt __find_get_block(), which does not 10141da177e4SLinus Torvalds * run under the page lock. 10151da177e4SLinus Torvalds */ 10161da177e4SLinus Torvalds spin_lock(&inode->i_mapping->private_lock); 10171da177e4SLinus Torvalds link_dev_buffers(page, bh); 1018f2d5a944SAnton Altaparmakov end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits, 1019f2d5a944SAnton Altaparmakov size); 10201da177e4SLinus Torvalds spin_unlock(&inode->i_mapping->private_lock); 1021676ce6d5SHugh Dickins done: 1022676ce6d5SHugh Dickins ret = (block < end_block) ? 1 : -ENXIO; 10231da177e4SLinus Torvalds failed: 10241da177e4SLinus Torvalds unlock_page(page); 102509cbfeafSKirill A. Shutemov put_page(page); 1026676ce6d5SHugh Dickins return ret; 10271da177e4SLinus Torvalds } 10281da177e4SLinus Torvalds 10291da177e4SLinus Torvalds /* 10301da177e4SLinus Torvalds * Create buffers for the specified block device block's page. If 10311da177e4SLinus Torvalds * that page was dirty, the buffers are set dirty also. 10321da177e4SLinus Torvalds */ 1033858119e1SArjan van de Ven static int 10343b5e6454SGioh Kim grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp) 10351da177e4SLinus Torvalds { 10361da177e4SLinus Torvalds pgoff_t index; 10371da177e4SLinus Torvalds int sizebits; 10381da177e4SLinus Torvalds 10391da177e4SLinus Torvalds sizebits = -1; 10401da177e4SLinus Torvalds do { 10411da177e4SLinus Torvalds sizebits++; 10421da177e4SLinus Torvalds } while ((size << sizebits) < PAGE_SIZE); 10431da177e4SLinus Torvalds 10441da177e4SLinus Torvalds index = block >> sizebits; 10451da177e4SLinus Torvalds 1046e5657933SAndrew Morton /* 1047e5657933SAndrew Morton * Check for a block which wants to lie outside our maximum possible 1048e5657933SAndrew Morton * pagecache index. (this comparison is done using sector_t types). 1049e5657933SAndrew Morton */ 1050e5657933SAndrew Morton if (unlikely(index != block >> sizebits)) { 1051e5657933SAndrew Morton printk(KERN_ERR "%s: requested out-of-range block %llu for " 1052a1c6f057SDmitry Monakhov "device %pg\n", 10538e24eea7SHarvey Harrison __func__, (unsigned long long)block, 1054a1c6f057SDmitry Monakhov bdev); 1055e5657933SAndrew Morton return -EIO; 1056e5657933SAndrew Morton } 1057676ce6d5SHugh Dickins 10581da177e4SLinus Torvalds /* Create a page with the proper size buffers.. */ 10593b5e6454SGioh Kim return grow_dev_page(bdev, block, index, size, sizebits, gfp); 10601da177e4SLinus Torvalds } 10611da177e4SLinus Torvalds 10620026ba40SEric Biggers static struct buffer_head * 10633b5e6454SGioh Kim __getblk_slow(struct block_device *bdev, sector_t block, 10643b5e6454SGioh Kim unsigned size, gfp_t gfp) 10651da177e4SLinus Torvalds { 10661da177e4SLinus Torvalds /* Size must be multiple of hard sectorsize */ 1067e1defc4fSMartin K. Petersen if (unlikely(size & (bdev_logical_block_size(bdev)-1) || 10681da177e4SLinus Torvalds (size < 512 || size > PAGE_SIZE))) { 10691da177e4SLinus Torvalds printk(KERN_ERR "getblk(): invalid block size %d requested\n", 10701da177e4SLinus Torvalds size); 1071e1defc4fSMartin K. Petersen printk(KERN_ERR "logical block size: %d\n", 1072e1defc4fSMartin K. Petersen bdev_logical_block_size(bdev)); 10731da177e4SLinus Torvalds 10741da177e4SLinus Torvalds dump_stack(); 10751da177e4SLinus Torvalds return NULL; 10761da177e4SLinus Torvalds } 10771da177e4SLinus Torvalds 1078676ce6d5SHugh Dickins for (;;) { 1079676ce6d5SHugh Dickins struct buffer_head *bh; 1080676ce6d5SHugh Dickins int ret; 1081676ce6d5SHugh Dickins 10821da177e4SLinus Torvalds bh = __find_get_block(bdev, block, size); 10831da177e4SLinus Torvalds if (bh) 10841da177e4SLinus Torvalds return bh; 10851da177e4SLinus Torvalds 10863b5e6454SGioh Kim ret = grow_buffers(bdev, block, size, gfp); 1087676ce6d5SHugh Dickins if (ret < 0) 108891f68c89SJeff Moyer return NULL; 1089676ce6d5SHugh Dickins if (ret == 0) 1090676ce6d5SHugh Dickins free_more_memory(); 1091676ce6d5SHugh Dickins } 10921da177e4SLinus Torvalds } 10931da177e4SLinus Torvalds 10941da177e4SLinus Torvalds /* 10951da177e4SLinus Torvalds * The relationship between dirty buffers and dirty pages: 10961da177e4SLinus Torvalds * 10971da177e4SLinus Torvalds * Whenever a page has any dirty buffers, the page's dirty bit is set, and 10981da177e4SLinus Torvalds * the page is tagged dirty in its radix tree. 10991da177e4SLinus Torvalds * 11001da177e4SLinus Torvalds * At all times, the dirtiness of the buffers represents the dirtiness of 11011da177e4SLinus Torvalds * subsections of the page. If the page has buffers, the page dirty bit is 11021da177e4SLinus Torvalds * merely a hint about the true dirty state. 11031da177e4SLinus Torvalds * 11041da177e4SLinus Torvalds * When a page is set dirty in its entirety, all its buffers are marked dirty 11051da177e4SLinus Torvalds * (if the page has buffers). 11061da177e4SLinus Torvalds * 11071da177e4SLinus Torvalds * When a buffer is marked dirty, its page is dirtied, but the page's other 11081da177e4SLinus Torvalds * buffers are not. 11091da177e4SLinus Torvalds * 11101da177e4SLinus Torvalds * Also. When blockdev buffers are explicitly read with bread(), they 11111da177e4SLinus Torvalds * individually become uptodate. But their backing page remains not 11121da177e4SLinus Torvalds * uptodate - even if all of its buffers are uptodate. A subsequent 11131da177e4SLinus Torvalds * block_read_full_page() against that page will discover all the uptodate 11141da177e4SLinus Torvalds * buffers, will set the page uptodate and will perform no I/O. 11151da177e4SLinus Torvalds */ 11161da177e4SLinus Torvalds 11171da177e4SLinus Torvalds /** 11181da177e4SLinus Torvalds * mark_buffer_dirty - mark a buffer_head as needing writeout 111967be2dd1SMartin Waitz * @bh: the buffer_head to mark dirty 11201da177e4SLinus Torvalds * 11211da177e4SLinus Torvalds * mark_buffer_dirty() will set the dirty bit against the buffer, then set its 11221da177e4SLinus Torvalds * backing page dirty, then tag the page as dirty in its address_space's radix 11231da177e4SLinus Torvalds * tree and then attach the address_space's inode to its superblock's dirty 11241da177e4SLinus Torvalds * inode list. 11251da177e4SLinus Torvalds * 11261da177e4SLinus Torvalds * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock, 1127250df6edSDave Chinner * mapping->tree_lock and mapping->host->i_lock. 11281da177e4SLinus Torvalds */ 1129fc9b52cdSHarvey Harrison void mark_buffer_dirty(struct buffer_head *bh) 11301da177e4SLinus Torvalds { 1131787d2214SNick Piggin WARN_ON_ONCE(!buffer_uptodate(bh)); 11321be62dc1SLinus Torvalds 11335305cb83STejun Heo trace_block_dirty_buffer(bh); 11345305cb83STejun Heo 11351be62dc1SLinus Torvalds /* 11361be62dc1SLinus Torvalds * Very *carefully* optimize the it-is-already-dirty case. 11371be62dc1SLinus Torvalds * 11381be62dc1SLinus Torvalds * Don't let the final "is it dirty" escape to before we 11391be62dc1SLinus Torvalds * perhaps modified the buffer. 11401be62dc1SLinus Torvalds */ 11411be62dc1SLinus Torvalds if (buffer_dirty(bh)) { 11421be62dc1SLinus Torvalds smp_mb(); 11431be62dc1SLinus Torvalds if (buffer_dirty(bh)) 11441be62dc1SLinus Torvalds return; 11451be62dc1SLinus Torvalds } 11461be62dc1SLinus Torvalds 1147a8e7d49aSLinus Torvalds if (!test_set_buffer_dirty(bh)) { 1148a8e7d49aSLinus Torvalds struct page *page = bh->b_page; 1149c4843a75SGreg Thelen struct address_space *mapping = NULL; 1150c4843a75SGreg Thelen 115162cccb8cSJohannes Weiner lock_page_memcg(page); 11528e9d78edSLinus Torvalds if (!TestSetPageDirty(page)) { 1153c4843a75SGreg Thelen mapping = page_mapping(page); 11548e9d78edSLinus Torvalds if (mapping) 115562cccb8cSJohannes Weiner __set_page_dirty(page, mapping, 0); 11568e9d78edSLinus Torvalds } 115762cccb8cSJohannes Weiner unlock_page_memcg(page); 1158c4843a75SGreg Thelen if (mapping) 1159c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 1160a8e7d49aSLinus Torvalds } 11611da177e4SLinus Torvalds } 11621fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(mark_buffer_dirty); 11631da177e4SLinus Torvalds 116487354e5dSJeff Layton void mark_buffer_write_io_error(struct buffer_head *bh) 116587354e5dSJeff Layton { 116687354e5dSJeff Layton set_buffer_write_io_error(bh); 116787354e5dSJeff Layton /* FIXME: do we need to set this in both places? */ 116887354e5dSJeff Layton if (bh->b_page && bh->b_page->mapping) 116987354e5dSJeff Layton mapping_set_error(bh->b_page->mapping, -EIO); 117087354e5dSJeff Layton if (bh->b_assoc_map) 117187354e5dSJeff Layton mapping_set_error(bh->b_assoc_map, -EIO); 117287354e5dSJeff Layton } 117387354e5dSJeff Layton EXPORT_SYMBOL(mark_buffer_write_io_error); 117487354e5dSJeff Layton 11751da177e4SLinus Torvalds /* 11761da177e4SLinus Torvalds * Decrement a buffer_head's reference count. If all buffers against a page 11771da177e4SLinus Torvalds * have zero reference count, are clean and unlocked, and if the page is clean 11781da177e4SLinus Torvalds * and unlocked then try_to_free_buffers() may strip the buffers from the page 11791da177e4SLinus Torvalds * in preparation for freeing it (sometimes, rarely, buffers are removed from 11801da177e4SLinus Torvalds * a page but it ends up not being freed, and buffers may later be reattached). 11811da177e4SLinus Torvalds */ 11821da177e4SLinus Torvalds void __brelse(struct buffer_head * buf) 11831da177e4SLinus Torvalds { 11841da177e4SLinus Torvalds if (atomic_read(&buf->b_count)) { 11851da177e4SLinus Torvalds put_bh(buf); 11861da177e4SLinus Torvalds return; 11871da177e4SLinus Torvalds } 11885c752ad9SArjan van de Ven WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); 11891da177e4SLinus Torvalds } 11901fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__brelse); 11911da177e4SLinus Torvalds 11921da177e4SLinus Torvalds /* 11931da177e4SLinus Torvalds * bforget() is like brelse(), except it discards any 11941da177e4SLinus Torvalds * potentially dirty data. 11951da177e4SLinus Torvalds */ 11961da177e4SLinus Torvalds void __bforget(struct buffer_head *bh) 11971da177e4SLinus Torvalds { 11981da177e4SLinus Torvalds clear_buffer_dirty(bh); 1199535ee2fbSJan Kara if (bh->b_assoc_map) { 12001da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 12011da177e4SLinus Torvalds 12021da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 12031da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 120458ff407bSJan Kara bh->b_assoc_map = NULL; 12051da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 12061da177e4SLinus Torvalds } 12071da177e4SLinus Torvalds __brelse(bh); 12081da177e4SLinus Torvalds } 12091fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__bforget); 12101da177e4SLinus Torvalds 12111da177e4SLinus Torvalds static struct buffer_head *__bread_slow(struct buffer_head *bh) 12121da177e4SLinus Torvalds { 12131da177e4SLinus Torvalds lock_buffer(bh); 12141da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 12151da177e4SLinus Torvalds unlock_buffer(bh); 12161da177e4SLinus Torvalds return bh; 12171da177e4SLinus Torvalds } else { 12181da177e4SLinus Torvalds get_bh(bh); 12191da177e4SLinus Torvalds bh->b_end_io = end_buffer_read_sync; 12202a222ca9SMike Christie submit_bh(REQ_OP_READ, 0, bh); 12211da177e4SLinus Torvalds wait_on_buffer(bh); 12221da177e4SLinus Torvalds if (buffer_uptodate(bh)) 12231da177e4SLinus Torvalds return bh; 12241da177e4SLinus Torvalds } 12251da177e4SLinus Torvalds brelse(bh); 12261da177e4SLinus Torvalds return NULL; 12271da177e4SLinus Torvalds } 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds /* 12301da177e4SLinus Torvalds * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). 12311da177e4SLinus Torvalds * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their 12321da177e4SLinus Torvalds * refcount elevated by one when they're in an LRU. A buffer can only appear 12331da177e4SLinus Torvalds * once in a particular CPU's LRU. A single buffer can be present in multiple 12341da177e4SLinus Torvalds * CPU's LRUs at the same time. 12351da177e4SLinus Torvalds * 12361da177e4SLinus Torvalds * This is a transparent caching front-end to sb_bread(), sb_getblk() and 12371da177e4SLinus Torvalds * sb_find_get_block(). 12381da177e4SLinus Torvalds * 12391da177e4SLinus Torvalds * The LRUs themselves only need locking against invalidate_bh_lrus. We use 12401da177e4SLinus Torvalds * a local interrupt disable for that. 12411da177e4SLinus Torvalds */ 12421da177e4SLinus Torvalds 124386cf78d7SSebastien Buisson #define BH_LRU_SIZE 16 12441da177e4SLinus Torvalds 12451da177e4SLinus Torvalds struct bh_lru { 12461da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12471da177e4SLinus Torvalds }; 12481da177e4SLinus Torvalds 12491da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; 12501da177e4SLinus Torvalds 12511da177e4SLinus Torvalds #ifdef CONFIG_SMP 12521da177e4SLinus Torvalds #define bh_lru_lock() local_irq_disable() 12531da177e4SLinus Torvalds #define bh_lru_unlock() local_irq_enable() 12541da177e4SLinus Torvalds #else 12551da177e4SLinus Torvalds #define bh_lru_lock() preempt_disable() 12561da177e4SLinus Torvalds #define bh_lru_unlock() preempt_enable() 12571da177e4SLinus Torvalds #endif 12581da177e4SLinus Torvalds 12591da177e4SLinus Torvalds static inline void check_irqs_on(void) 12601da177e4SLinus Torvalds { 12611da177e4SLinus Torvalds #ifdef irqs_disabled 12621da177e4SLinus Torvalds BUG_ON(irqs_disabled()); 12631da177e4SLinus Torvalds #endif 12641da177e4SLinus Torvalds } 12651da177e4SLinus Torvalds 12661da177e4SLinus Torvalds /* 1267241f01fbSEric Biggers * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is 1268241f01fbSEric Biggers * inserted at the front, and the buffer_head at the back if any is evicted. 1269241f01fbSEric Biggers * Or, if already in the LRU it is moved to the front. 12701da177e4SLinus Torvalds */ 12711da177e4SLinus Torvalds static void bh_lru_install(struct buffer_head *bh) 12721da177e4SLinus Torvalds { 1273241f01fbSEric Biggers struct buffer_head *evictee = bh; 1274241f01fbSEric Biggers struct bh_lru *b; 1275241f01fbSEric Biggers int i; 12761da177e4SLinus Torvalds 12771da177e4SLinus Torvalds check_irqs_on(); 12781da177e4SLinus Torvalds bh_lru_lock(); 1279241f01fbSEric Biggers 1280241f01fbSEric Biggers b = this_cpu_ptr(&bh_lrus); 1281241f01fbSEric Biggers for (i = 0; i < BH_LRU_SIZE; i++) { 1282241f01fbSEric Biggers swap(evictee, b->bhs[i]); 1283241f01fbSEric Biggers if (evictee == bh) { 1284241f01fbSEric Biggers bh_lru_unlock(); 1285241f01fbSEric Biggers return; 1286241f01fbSEric Biggers } 1287241f01fbSEric Biggers } 12881da177e4SLinus Torvalds 12891da177e4SLinus Torvalds get_bh(bh); 12901da177e4SLinus Torvalds bh_lru_unlock(); 1291241f01fbSEric Biggers brelse(evictee); 12921da177e4SLinus Torvalds } 12931da177e4SLinus Torvalds 12941da177e4SLinus Torvalds /* 12951da177e4SLinus Torvalds * Look up the bh in this cpu's LRU. If it's there, move it to the head. 12961da177e4SLinus Torvalds */ 1297858119e1SArjan van de Ven static struct buffer_head * 12983991d3bdSTomasz Kvarsin lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) 12991da177e4SLinus Torvalds { 13001da177e4SLinus Torvalds struct buffer_head *ret = NULL; 13013991d3bdSTomasz Kvarsin unsigned int i; 13021da177e4SLinus Torvalds 13031da177e4SLinus Torvalds check_irqs_on(); 13041da177e4SLinus Torvalds bh_lru_lock(); 13051da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 1306c7b92516SChristoph Lameter struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); 13071da177e4SLinus Torvalds 13089470dd5dSZach Brown if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && 13099470dd5dSZach Brown bh->b_size == size) { 13101da177e4SLinus Torvalds if (i) { 13111da177e4SLinus Torvalds while (i) { 1312c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[i], 1313c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[i - 1])); 13141da177e4SLinus Torvalds i--; 13151da177e4SLinus Torvalds } 1316c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[0], bh); 13171da177e4SLinus Torvalds } 13181da177e4SLinus Torvalds get_bh(bh); 13191da177e4SLinus Torvalds ret = bh; 13201da177e4SLinus Torvalds break; 13211da177e4SLinus Torvalds } 13221da177e4SLinus Torvalds } 13231da177e4SLinus Torvalds bh_lru_unlock(); 13241da177e4SLinus Torvalds return ret; 13251da177e4SLinus Torvalds } 13261da177e4SLinus Torvalds 13271da177e4SLinus Torvalds /* 13281da177e4SLinus Torvalds * Perform a pagecache lookup for the matching buffer. If it's there, refresh 13291da177e4SLinus Torvalds * it in the LRU and mark it as accessed. If it is not present then return 13301da177e4SLinus Torvalds * NULL 13311da177e4SLinus Torvalds */ 13321da177e4SLinus Torvalds struct buffer_head * 13333991d3bdSTomasz Kvarsin __find_get_block(struct block_device *bdev, sector_t block, unsigned size) 13341da177e4SLinus Torvalds { 13351da177e4SLinus Torvalds struct buffer_head *bh = lookup_bh_lru(bdev, block, size); 13361da177e4SLinus Torvalds 13371da177e4SLinus Torvalds if (bh == NULL) { 13382457aec6SMel Gorman /* __find_get_block_slow will mark the page accessed */ 1339385fd4c5SCoywolf Qi Hunt bh = __find_get_block_slow(bdev, block); 13401da177e4SLinus Torvalds if (bh) 13411da177e4SLinus Torvalds bh_lru_install(bh); 13422457aec6SMel Gorman } else 13431da177e4SLinus Torvalds touch_buffer(bh); 13442457aec6SMel Gorman 13451da177e4SLinus Torvalds return bh; 13461da177e4SLinus Torvalds } 13471da177e4SLinus Torvalds EXPORT_SYMBOL(__find_get_block); 13481da177e4SLinus Torvalds 13491da177e4SLinus Torvalds /* 13503b5e6454SGioh Kim * __getblk_gfp() will locate (and, if necessary, create) the buffer_head 13511da177e4SLinus Torvalds * which corresponds to the passed block_device, block and size. The 13521da177e4SLinus Torvalds * returned buffer has its reference count incremented. 13531da177e4SLinus Torvalds * 13543b5e6454SGioh Kim * __getblk_gfp() will lock up the machine if grow_dev_page's 13553b5e6454SGioh Kim * try_to_free_buffers() attempt is failing. FIXME, perhaps? 13561da177e4SLinus Torvalds */ 13571da177e4SLinus Torvalds struct buffer_head * 13583b5e6454SGioh Kim __getblk_gfp(struct block_device *bdev, sector_t block, 13593b5e6454SGioh Kim unsigned size, gfp_t gfp) 13601da177e4SLinus Torvalds { 13611da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, block, size); 13621da177e4SLinus Torvalds 13631da177e4SLinus Torvalds might_sleep(); 13641da177e4SLinus Torvalds if (bh == NULL) 13653b5e6454SGioh Kim bh = __getblk_slow(bdev, block, size, gfp); 13661da177e4SLinus Torvalds return bh; 13671da177e4SLinus Torvalds } 13683b5e6454SGioh Kim EXPORT_SYMBOL(__getblk_gfp); 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds /* 13711da177e4SLinus Torvalds * Do async read-ahead on a buffer.. 13721da177e4SLinus Torvalds */ 13733991d3bdSTomasz Kvarsin void __breadahead(struct block_device *bdev, sector_t block, unsigned size) 13741da177e4SLinus Torvalds { 13751da177e4SLinus Torvalds struct buffer_head *bh = __getblk(bdev, block, size); 1376a3e713b5SAndrew Morton if (likely(bh)) { 137770246286SChristoph Hellwig ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh); 13781da177e4SLinus Torvalds brelse(bh); 13791da177e4SLinus Torvalds } 1380a3e713b5SAndrew Morton } 13811da177e4SLinus Torvalds EXPORT_SYMBOL(__breadahead); 13821da177e4SLinus Torvalds 13831da177e4SLinus Torvalds /** 13843b5e6454SGioh Kim * __bread_gfp() - reads a specified block and returns the bh 138567be2dd1SMartin Waitz * @bdev: the block_device to read from 13861da177e4SLinus Torvalds * @block: number of block 13871da177e4SLinus Torvalds * @size: size (in bytes) to read 13883b5e6454SGioh Kim * @gfp: page allocation flag 13891da177e4SLinus Torvalds * 13901da177e4SLinus Torvalds * Reads a specified block, and returns buffer head that contains it. 13913b5e6454SGioh Kim * The page cache can be allocated from non-movable area 13923b5e6454SGioh Kim * not to prevent page migration if you set gfp to zero. 13931da177e4SLinus Torvalds * It returns NULL if the block was unreadable. 13941da177e4SLinus Torvalds */ 13951da177e4SLinus Torvalds struct buffer_head * 13963b5e6454SGioh Kim __bread_gfp(struct block_device *bdev, sector_t block, 13973b5e6454SGioh Kim unsigned size, gfp_t gfp) 13981da177e4SLinus Torvalds { 13993b5e6454SGioh Kim struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); 14001da177e4SLinus Torvalds 1401a3e713b5SAndrew Morton if (likely(bh) && !buffer_uptodate(bh)) 14021da177e4SLinus Torvalds bh = __bread_slow(bh); 14031da177e4SLinus Torvalds return bh; 14041da177e4SLinus Torvalds } 14053b5e6454SGioh Kim EXPORT_SYMBOL(__bread_gfp); 14061da177e4SLinus Torvalds 14071da177e4SLinus Torvalds /* 14081da177e4SLinus Torvalds * invalidate_bh_lrus() is called rarely - but not only at unmount. 14091da177e4SLinus Torvalds * This doesn't race because it runs in each cpu either in irq 14101da177e4SLinus Torvalds * or with preempt disabled. 14111da177e4SLinus Torvalds */ 14121da177e4SLinus Torvalds static void invalidate_bh_lru(void *arg) 14131da177e4SLinus Torvalds { 14141da177e4SLinus Torvalds struct bh_lru *b = &get_cpu_var(bh_lrus); 14151da177e4SLinus Torvalds int i; 14161da177e4SLinus Torvalds 14171da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 14181da177e4SLinus Torvalds brelse(b->bhs[i]); 14191da177e4SLinus Torvalds b->bhs[i] = NULL; 14201da177e4SLinus Torvalds } 14211da177e4SLinus Torvalds put_cpu_var(bh_lrus); 14221da177e4SLinus Torvalds } 14231da177e4SLinus Torvalds 142442be35d0SGilad Ben-Yossef static bool has_bh_in_lru(int cpu, void *dummy) 142542be35d0SGilad Ben-Yossef { 142642be35d0SGilad Ben-Yossef struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); 142742be35d0SGilad Ben-Yossef int i; 142842be35d0SGilad Ben-Yossef 142942be35d0SGilad Ben-Yossef for (i = 0; i < BH_LRU_SIZE; i++) { 143042be35d0SGilad Ben-Yossef if (b->bhs[i]) 143142be35d0SGilad Ben-Yossef return 1; 143242be35d0SGilad Ben-Yossef } 143342be35d0SGilad Ben-Yossef 143442be35d0SGilad Ben-Yossef return 0; 143542be35d0SGilad Ben-Yossef } 143642be35d0SGilad Ben-Yossef 1437f9a14399SPeter Zijlstra void invalidate_bh_lrus(void) 14381da177e4SLinus Torvalds { 143942be35d0SGilad Ben-Yossef on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL); 14401da177e4SLinus Torvalds } 14419db5579bSNick Piggin EXPORT_SYMBOL_GPL(invalidate_bh_lrus); 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds void set_bh_page(struct buffer_head *bh, 14441da177e4SLinus Torvalds struct page *page, unsigned long offset) 14451da177e4SLinus Torvalds { 14461da177e4SLinus Torvalds bh->b_page = page; 1447e827f923SEric Sesterhenn BUG_ON(offset >= PAGE_SIZE); 14481da177e4SLinus Torvalds if (PageHighMem(page)) 14491da177e4SLinus Torvalds /* 14501da177e4SLinus Torvalds * This catches illegal uses and preserves the offset: 14511da177e4SLinus Torvalds */ 14521da177e4SLinus Torvalds bh->b_data = (char *)(0 + offset); 14531da177e4SLinus Torvalds else 14541da177e4SLinus Torvalds bh->b_data = page_address(page) + offset; 14551da177e4SLinus Torvalds } 14561da177e4SLinus Torvalds EXPORT_SYMBOL(set_bh_page); 14571da177e4SLinus Torvalds 14581da177e4SLinus Torvalds /* 14591da177e4SLinus Torvalds * Called when truncating a buffer on a page completely. 14601da177e4SLinus Torvalds */ 1461e7470ee8SMel Gorman 1462e7470ee8SMel Gorman /* Bits that are cleared during an invalidate */ 1463e7470ee8SMel Gorman #define BUFFER_FLAGS_DISCARD \ 1464e7470ee8SMel Gorman (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ 1465e7470ee8SMel Gorman 1 << BH_Delay | 1 << BH_Unwritten) 1466e7470ee8SMel Gorman 1467858119e1SArjan van de Ven static void discard_buffer(struct buffer_head * bh) 14681da177e4SLinus Torvalds { 1469e7470ee8SMel Gorman unsigned long b_state, b_state_old; 1470e7470ee8SMel Gorman 14711da177e4SLinus Torvalds lock_buffer(bh); 14721da177e4SLinus Torvalds clear_buffer_dirty(bh); 14731da177e4SLinus Torvalds bh->b_bdev = NULL; 1474e7470ee8SMel Gorman b_state = bh->b_state; 1475e7470ee8SMel Gorman for (;;) { 1476e7470ee8SMel Gorman b_state_old = cmpxchg(&bh->b_state, b_state, 1477e7470ee8SMel Gorman (b_state & ~BUFFER_FLAGS_DISCARD)); 1478e7470ee8SMel Gorman if (b_state_old == b_state) 1479e7470ee8SMel Gorman break; 1480e7470ee8SMel Gorman b_state = b_state_old; 1481e7470ee8SMel Gorman } 14821da177e4SLinus Torvalds unlock_buffer(bh); 14831da177e4SLinus Torvalds } 14841da177e4SLinus Torvalds 14851da177e4SLinus Torvalds /** 1486814e1d25SWang Sheng-Hui * block_invalidatepage - invalidate part or all of a buffer-backed page 14871da177e4SLinus Torvalds * 14881da177e4SLinus Torvalds * @page: the page which is affected 1489d47992f8SLukas Czerner * @offset: start of the range to invalidate 1490d47992f8SLukas Czerner * @length: length of the range to invalidate 14911da177e4SLinus Torvalds * 14921da177e4SLinus Torvalds * block_invalidatepage() is called when all or part of the page has become 14931da177e4SLinus Torvalds * invalidated by a truncate operation. 14941da177e4SLinus Torvalds * 14951da177e4SLinus Torvalds * block_invalidatepage() does not have to release all buffers, but it must 14961da177e4SLinus Torvalds * ensure that no dirty buffer is left outside @offset and that no I/O 14971da177e4SLinus Torvalds * is underway against any of the blocks which are outside the truncation 14981da177e4SLinus Torvalds * point. Because the caller is about to free (and possibly reuse) those 14991da177e4SLinus Torvalds * blocks on-disk. 15001da177e4SLinus Torvalds */ 1501d47992f8SLukas Czerner void block_invalidatepage(struct page *page, unsigned int offset, 1502d47992f8SLukas Czerner unsigned int length) 15031da177e4SLinus Torvalds { 15041da177e4SLinus Torvalds struct buffer_head *head, *bh, *next; 15051da177e4SLinus Torvalds unsigned int curr_off = 0; 1506d47992f8SLukas Czerner unsigned int stop = length + offset; 15071da177e4SLinus Torvalds 15081da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 15091da177e4SLinus Torvalds if (!page_has_buffers(page)) 15101da177e4SLinus Torvalds goto out; 15111da177e4SLinus Torvalds 1512d47992f8SLukas Czerner /* 1513d47992f8SLukas Czerner * Check for overflow 1514d47992f8SLukas Czerner */ 151509cbfeafSKirill A. Shutemov BUG_ON(stop > PAGE_SIZE || stop < length); 1516d47992f8SLukas Czerner 15171da177e4SLinus Torvalds head = page_buffers(page); 15181da177e4SLinus Torvalds bh = head; 15191da177e4SLinus Torvalds do { 15201da177e4SLinus Torvalds unsigned int next_off = curr_off + bh->b_size; 15211da177e4SLinus Torvalds next = bh->b_this_page; 15221da177e4SLinus Torvalds 15231da177e4SLinus Torvalds /* 1524d47992f8SLukas Czerner * Are we still fully in range ? 1525d47992f8SLukas Czerner */ 1526d47992f8SLukas Czerner if (next_off > stop) 1527d47992f8SLukas Czerner goto out; 1528d47992f8SLukas Czerner 1529d47992f8SLukas Czerner /* 15301da177e4SLinus Torvalds * is this block fully invalidated? 15311da177e4SLinus Torvalds */ 15321da177e4SLinus Torvalds if (offset <= curr_off) 15331da177e4SLinus Torvalds discard_buffer(bh); 15341da177e4SLinus Torvalds curr_off = next_off; 15351da177e4SLinus Torvalds bh = next; 15361da177e4SLinus Torvalds } while (bh != head); 15371da177e4SLinus Torvalds 15381da177e4SLinus Torvalds /* 15391da177e4SLinus Torvalds * We release buffers only if the entire page is being invalidated. 15401da177e4SLinus Torvalds * The get_block cached value has been unconditionally invalidated, 15411da177e4SLinus Torvalds * so real IO is not possible anymore. 15421da177e4SLinus Torvalds */ 15431da177e4SLinus Torvalds if (offset == 0) 15442ff28e22SNeilBrown try_to_release_page(page, 0); 15451da177e4SLinus Torvalds out: 15462ff28e22SNeilBrown return; 15471da177e4SLinus Torvalds } 15481da177e4SLinus Torvalds EXPORT_SYMBOL(block_invalidatepage); 15491da177e4SLinus Torvalds 1550d47992f8SLukas Czerner 15511da177e4SLinus Torvalds /* 15521da177e4SLinus Torvalds * We attach and possibly dirty the buffers atomically wrt 15531da177e4SLinus Torvalds * __set_page_dirty_buffers() via private_lock. try_to_free_buffers 15541da177e4SLinus Torvalds * is already excluded via the page lock. 15551da177e4SLinus Torvalds */ 15561da177e4SLinus Torvalds void create_empty_buffers(struct page *page, 15571da177e4SLinus Torvalds unsigned long blocksize, unsigned long b_state) 15581da177e4SLinus Torvalds { 15591da177e4SLinus Torvalds struct buffer_head *bh, *head, *tail; 15601da177e4SLinus Torvalds 1561640ab98fSJens Axboe head = alloc_page_buffers(page, blocksize, true); 15621da177e4SLinus Torvalds bh = head; 15631da177e4SLinus Torvalds do { 15641da177e4SLinus Torvalds bh->b_state |= b_state; 15651da177e4SLinus Torvalds tail = bh; 15661da177e4SLinus Torvalds bh = bh->b_this_page; 15671da177e4SLinus Torvalds } while (bh); 15681da177e4SLinus Torvalds tail->b_this_page = head; 15691da177e4SLinus Torvalds 15701da177e4SLinus Torvalds spin_lock(&page->mapping->private_lock); 15711da177e4SLinus Torvalds if (PageUptodate(page) || PageDirty(page)) { 15721da177e4SLinus Torvalds bh = head; 15731da177e4SLinus Torvalds do { 15741da177e4SLinus Torvalds if (PageDirty(page)) 15751da177e4SLinus Torvalds set_buffer_dirty(bh); 15761da177e4SLinus Torvalds if (PageUptodate(page)) 15771da177e4SLinus Torvalds set_buffer_uptodate(bh); 15781da177e4SLinus Torvalds bh = bh->b_this_page; 15791da177e4SLinus Torvalds } while (bh != head); 15801da177e4SLinus Torvalds } 15811da177e4SLinus Torvalds attach_page_buffers(page, head); 15821da177e4SLinus Torvalds spin_unlock(&page->mapping->private_lock); 15831da177e4SLinus Torvalds } 15841da177e4SLinus Torvalds EXPORT_SYMBOL(create_empty_buffers); 15851da177e4SLinus Torvalds 158629f3ad7dSJan Kara /** 158729f3ad7dSJan Kara * clean_bdev_aliases: clean a range of buffers in block device 158829f3ad7dSJan Kara * @bdev: Block device to clean buffers in 158929f3ad7dSJan Kara * @block: Start of a range of blocks to clean 159029f3ad7dSJan Kara * @len: Number of blocks to clean 15911da177e4SLinus Torvalds * 159229f3ad7dSJan Kara * We are taking a range of blocks for data and we don't want writeback of any 159329f3ad7dSJan Kara * buffer-cache aliases starting from return from this function and until the 159429f3ad7dSJan Kara * moment when something will explicitly mark the buffer dirty (hopefully that 159529f3ad7dSJan Kara * will not happen until we will free that block ;-) We don't even need to mark 159629f3ad7dSJan Kara * it not-uptodate - nobody can expect anything from a newly allocated buffer 159729f3ad7dSJan Kara * anyway. We used to use unmap_buffer() for such invalidation, but that was 159829f3ad7dSJan Kara * wrong. We definitely don't want to mark the alias unmapped, for example - it 159929f3ad7dSJan Kara * would confuse anyone who might pick it with bread() afterwards... 160029f3ad7dSJan Kara * 160129f3ad7dSJan Kara * Also.. Note that bforget() doesn't lock the buffer. So there can be 160229f3ad7dSJan Kara * writeout I/O going on against recently-freed buffers. We don't wait on that 160329f3ad7dSJan Kara * I/O in bforget() - it's more efficient to wait on the I/O only if we really 160429f3ad7dSJan Kara * need to. That happens here. 16051da177e4SLinus Torvalds */ 160629f3ad7dSJan Kara void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) 16071da177e4SLinus Torvalds { 160829f3ad7dSJan Kara struct inode *bd_inode = bdev->bd_inode; 160929f3ad7dSJan Kara struct address_space *bd_mapping = bd_inode->i_mapping; 161029f3ad7dSJan Kara struct pagevec pvec; 161129f3ad7dSJan Kara pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); 161229f3ad7dSJan Kara pgoff_t end; 1613c10f778dSJan Kara int i, count; 161429f3ad7dSJan Kara struct buffer_head *bh; 161529f3ad7dSJan Kara struct buffer_head *head; 16161da177e4SLinus Torvalds 161729f3ad7dSJan Kara end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits); 161829f3ad7dSJan Kara pagevec_init(&pvec, 0); 1619397162ffSJan Kara while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) { 1620c10f778dSJan Kara count = pagevec_count(&pvec); 1621c10f778dSJan Kara for (i = 0; i < count; i++) { 162229f3ad7dSJan Kara struct page *page = pvec.pages[i]; 16231da177e4SLinus Torvalds 162429f3ad7dSJan Kara if (!page_has_buffers(page)) 162529f3ad7dSJan Kara continue; 162629f3ad7dSJan Kara /* 162729f3ad7dSJan Kara * We use page lock instead of bd_mapping->private_lock 162829f3ad7dSJan Kara * to pin buffers here since we can afford to sleep and 162929f3ad7dSJan Kara * it scales better than a global spinlock lock. 163029f3ad7dSJan Kara */ 163129f3ad7dSJan Kara lock_page(page); 163229f3ad7dSJan Kara /* Recheck when the page is locked which pins bhs */ 163329f3ad7dSJan Kara if (!page_has_buffers(page)) 163429f3ad7dSJan Kara goto unlock_page; 163529f3ad7dSJan Kara head = page_buffers(page); 163629f3ad7dSJan Kara bh = head; 163729f3ad7dSJan Kara do { 16386c006a9dSChandan Rajendra if (!buffer_mapped(bh) || (bh->b_blocknr < block)) 163929f3ad7dSJan Kara goto next; 164029f3ad7dSJan Kara if (bh->b_blocknr >= block + len) 164129f3ad7dSJan Kara break; 164229f3ad7dSJan Kara clear_buffer_dirty(bh); 164329f3ad7dSJan Kara wait_on_buffer(bh); 164429f3ad7dSJan Kara clear_buffer_req(bh); 164529f3ad7dSJan Kara next: 164629f3ad7dSJan Kara bh = bh->b_this_page; 164729f3ad7dSJan Kara } while (bh != head); 164829f3ad7dSJan Kara unlock_page: 164929f3ad7dSJan Kara unlock_page(page); 165029f3ad7dSJan Kara } 165129f3ad7dSJan Kara pagevec_release(&pvec); 165229f3ad7dSJan Kara cond_resched(); 1653c10f778dSJan Kara /* End of range already reached? */ 1654c10f778dSJan Kara if (index > end || !index) 1655c10f778dSJan Kara break; 16561da177e4SLinus Torvalds } 16571da177e4SLinus Torvalds } 165829f3ad7dSJan Kara EXPORT_SYMBOL(clean_bdev_aliases); 16591da177e4SLinus Torvalds 16601da177e4SLinus Torvalds /* 166145bce8f3SLinus Torvalds * Size is a power-of-two in the range 512..PAGE_SIZE, 166245bce8f3SLinus Torvalds * and the case we care about most is PAGE_SIZE. 166345bce8f3SLinus Torvalds * 166445bce8f3SLinus Torvalds * So this *could* possibly be written with those 166545bce8f3SLinus Torvalds * constraints in mind (relevant mostly if some 166645bce8f3SLinus Torvalds * architecture has a slow bit-scan instruction) 166745bce8f3SLinus Torvalds */ 166845bce8f3SLinus Torvalds static inline int block_size_bits(unsigned int blocksize) 166945bce8f3SLinus Torvalds { 167045bce8f3SLinus Torvalds return ilog2(blocksize); 167145bce8f3SLinus Torvalds } 167245bce8f3SLinus Torvalds 167345bce8f3SLinus Torvalds static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state) 167445bce8f3SLinus Torvalds { 167545bce8f3SLinus Torvalds BUG_ON(!PageLocked(page)); 167645bce8f3SLinus Torvalds 167745bce8f3SLinus Torvalds if (!page_has_buffers(page)) 167845bce8f3SLinus Torvalds create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state); 167945bce8f3SLinus Torvalds return page_buffers(page); 168045bce8f3SLinus Torvalds } 168145bce8f3SLinus Torvalds 168245bce8f3SLinus Torvalds /* 16831da177e4SLinus Torvalds * NOTE! All mapped/uptodate combinations are valid: 16841da177e4SLinus Torvalds * 16851da177e4SLinus Torvalds * Mapped Uptodate Meaning 16861da177e4SLinus Torvalds * 16871da177e4SLinus Torvalds * No No "unknown" - must do get_block() 16881da177e4SLinus Torvalds * No Yes "hole" - zero-filled 16891da177e4SLinus Torvalds * Yes No "allocated" - allocated on disk, not read in 16901da177e4SLinus Torvalds * Yes Yes "valid" - allocated and up-to-date in memory. 16911da177e4SLinus Torvalds * 16921da177e4SLinus Torvalds * "Dirty" is valid only with the last case (mapped+uptodate). 16931da177e4SLinus Torvalds */ 16941da177e4SLinus Torvalds 16951da177e4SLinus Torvalds /* 16961da177e4SLinus Torvalds * While block_write_full_page is writing back the dirty buffers under 16971da177e4SLinus Torvalds * the page lock, whoever dirtied the buffers may decide to clean them 16981da177e4SLinus Torvalds * again at any time. We handle that by only looking at the buffer 16991da177e4SLinus Torvalds * state inside lock_buffer(). 17001da177e4SLinus Torvalds * 17011da177e4SLinus Torvalds * If block_write_full_page() is called for regular writeback 17021da177e4SLinus Torvalds * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a 17031da177e4SLinus Torvalds * locked buffer. This only can happen if someone has written the buffer 17041da177e4SLinus Torvalds * directly, with submit_bh(). At the address_space level PageWriteback 17051da177e4SLinus Torvalds * prevents this contention from occurring. 17066e34eeddSTheodore Ts'o * 17076e34eeddSTheodore Ts'o * If block_write_full_page() is called with wbc->sync_mode == 170870fd7614SChristoph Hellwig * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this 1709721a9602SJens Axboe * causes the writes to be flagged as synchronous writes. 17101da177e4SLinus Torvalds */ 1711b4bba389SBenjamin Marzinski int __block_write_full_page(struct inode *inode, struct page *page, 171235c80d5fSChris Mason get_block_t *get_block, struct writeback_control *wbc, 171335c80d5fSChris Mason bh_end_io_t *handler) 17141da177e4SLinus Torvalds { 17151da177e4SLinus Torvalds int err; 17161da177e4SLinus Torvalds sector_t block; 17171da177e4SLinus Torvalds sector_t last_block; 1718f0fbd5fcSAndrew Morton struct buffer_head *bh, *head; 171945bce8f3SLinus Torvalds unsigned int blocksize, bbits; 17201da177e4SLinus Torvalds int nr_underway = 0; 17217637241eSJens Axboe int write_flags = wbc_to_write_flags(wbc); 17221da177e4SLinus Torvalds 172345bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 17241da177e4SLinus Torvalds (1 << BH_Dirty)|(1 << BH_Uptodate)); 17251da177e4SLinus Torvalds 17261da177e4SLinus Torvalds /* 17271da177e4SLinus Torvalds * Be very careful. We have no exclusion from __set_page_dirty_buffers 17281da177e4SLinus Torvalds * here, and the (potentially unmapped) buffers may become dirty at 17291da177e4SLinus Torvalds * any time. If a buffer becomes dirty here after we've inspected it 17301da177e4SLinus Torvalds * then we just miss that fact, and the page stays dirty. 17311da177e4SLinus Torvalds * 17321da177e4SLinus Torvalds * Buffers outside i_size may be dirtied by __set_page_dirty_buffers; 17331da177e4SLinus Torvalds * handle that here by just cleaning them. 17341da177e4SLinus Torvalds */ 17351da177e4SLinus Torvalds 17361da177e4SLinus Torvalds bh = head; 173745bce8f3SLinus Torvalds blocksize = bh->b_size; 173845bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 173945bce8f3SLinus Torvalds 174009cbfeafSKirill A. Shutemov block = (sector_t)page->index << (PAGE_SHIFT - bbits); 174145bce8f3SLinus Torvalds last_block = (i_size_read(inode) - 1) >> bbits; 17421da177e4SLinus Torvalds 17431da177e4SLinus Torvalds /* 17441da177e4SLinus Torvalds * Get all the dirty buffers mapped to disk addresses and 17451da177e4SLinus Torvalds * handle any aliases from the underlying blockdev's mapping. 17461da177e4SLinus Torvalds */ 17471da177e4SLinus Torvalds do { 17481da177e4SLinus Torvalds if (block > last_block) { 17491da177e4SLinus Torvalds /* 17501da177e4SLinus Torvalds * mapped buffers outside i_size will occur, because 17511da177e4SLinus Torvalds * this page can be outside i_size when there is a 17521da177e4SLinus Torvalds * truncate in progress. 17531da177e4SLinus Torvalds */ 17541da177e4SLinus Torvalds /* 17551da177e4SLinus Torvalds * The buffer was zeroed by block_write_full_page() 17561da177e4SLinus Torvalds */ 17571da177e4SLinus Torvalds clear_buffer_dirty(bh); 17581da177e4SLinus Torvalds set_buffer_uptodate(bh); 175929a814d2SAlex Tomas } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && 176029a814d2SAlex Tomas buffer_dirty(bh)) { 1761b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 17621da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 17631da177e4SLinus Torvalds if (err) 17641da177e4SLinus Torvalds goto recover; 176529a814d2SAlex Tomas clear_buffer_delay(bh); 17661da177e4SLinus Torvalds if (buffer_new(bh)) { 17671da177e4SLinus Torvalds /* blockdev mappings never come here */ 17681da177e4SLinus Torvalds clear_buffer_new(bh); 1769e64855c6SJan Kara clean_bdev_bh_alias(bh); 17701da177e4SLinus Torvalds } 17711da177e4SLinus Torvalds } 17721da177e4SLinus Torvalds bh = bh->b_this_page; 17731da177e4SLinus Torvalds block++; 17741da177e4SLinus Torvalds } while (bh != head); 17751da177e4SLinus Torvalds 17761da177e4SLinus Torvalds do { 17771da177e4SLinus Torvalds if (!buffer_mapped(bh)) 17781da177e4SLinus Torvalds continue; 17791da177e4SLinus Torvalds /* 17801da177e4SLinus Torvalds * If it's a fully non-blocking write attempt and we cannot 17811da177e4SLinus Torvalds * lock the buffer then redirty the page. Note that this can 17825b0830cbSJens Axboe * potentially cause a busy-wait loop from writeback threads 17835b0830cbSJens Axboe * and kswapd activity, but those code paths have their own 17845b0830cbSJens Axboe * higher-level throttling. 17851da177e4SLinus Torvalds */ 17861b430beeSWu Fengguang if (wbc->sync_mode != WB_SYNC_NONE) { 17871da177e4SLinus Torvalds lock_buffer(bh); 1788ca5de404SNick Piggin } else if (!trylock_buffer(bh)) { 17891da177e4SLinus Torvalds redirty_page_for_writepage(wbc, page); 17901da177e4SLinus Torvalds continue; 17911da177e4SLinus Torvalds } 17921da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 179335c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 17941da177e4SLinus Torvalds } else { 17951da177e4SLinus Torvalds unlock_buffer(bh); 17961da177e4SLinus Torvalds } 17971da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 17981da177e4SLinus Torvalds 17991da177e4SLinus Torvalds /* 18001da177e4SLinus Torvalds * The page and its buffers are protected by PageWriteback(), so we can 18011da177e4SLinus Torvalds * drop the bh refcounts early. 18021da177e4SLinus Torvalds */ 18031da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 18041da177e4SLinus Torvalds set_page_writeback(page); 18051da177e4SLinus Torvalds 18061da177e4SLinus Torvalds do { 18071da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 18081da177e4SLinus Torvalds if (buffer_async_write(bh)) { 18098e8f9298SJens Axboe submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, 18108e8f9298SJens Axboe inode->i_write_hint, wbc); 18111da177e4SLinus Torvalds nr_underway++; 1812ad576e63SNick Piggin } 18131da177e4SLinus Torvalds bh = next; 18141da177e4SLinus Torvalds } while (bh != head); 181505937baaSAndrew Morton unlock_page(page); 18161da177e4SLinus Torvalds 18171da177e4SLinus Torvalds err = 0; 18181da177e4SLinus Torvalds done: 18191da177e4SLinus Torvalds if (nr_underway == 0) { 18201da177e4SLinus Torvalds /* 18211da177e4SLinus Torvalds * The page was marked dirty, but the buffers were 18221da177e4SLinus Torvalds * clean. Someone wrote them back by hand with 18231da177e4SLinus Torvalds * ll_rw_block/submit_bh. A rare case. 18241da177e4SLinus Torvalds */ 18251da177e4SLinus Torvalds end_page_writeback(page); 18263d67f2d7SNick Piggin 18271da177e4SLinus Torvalds /* 18281da177e4SLinus Torvalds * The page and buffer_heads can be released at any time from 18291da177e4SLinus Torvalds * here on. 18301da177e4SLinus Torvalds */ 18311da177e4SLinus Torvalds } 18321da177e4SLinus Torvalds return err; 18331da177e4SLinus Torvalds 18341da177e4SLinus Torvalds recover: 18351da177e4SLinus Torvalds /* 18361da177e4SLinus Torvalds * ENOSPC, or some other error. We may already have added some 18371da177e4SLinus Torvalds * blocks to the file, so we need to write these out to avoid 18381da177e4SLinus Torvalds * exposing stale data. 18391da177e4SLinus Torvalds * The page is currently locked and not marked for writeback 18401da177e4SLinus Torvalds */ 18411da177e4SLinus Torvalds bh = head; 18421da177e4SLinus Torvalds /* Recovery: lock and submit the mapped buffers */ 18431da177e4SLinus Torvalds do { 184429a814d2SAlex Tomas if (buffer_mapped(bh) && buffer_dirty(bh) && 184529a814d2SAlex Tomas !buffer_delay(bh)) { 18461da177e4SLinus Torvalds lock_buffer(bh); 184735c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 18481da177e4SLinus Torvalds } else { 18491da177e4SLinus Torvalds /* 18501da177e4SLinus Torvalds * The buffer may have been set dirty during 18511da177e4SLinus Torvalds * attachment to a dirty page. 18521da177e4SLinus Torvalds */ 18531da177e4SLinus Torvalds clear_buffer_dirty(bh); 18541da177e4SLinus Torvalds } 18551da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 18561da177e4SLinus Torvalds SetPageError(page); 18571da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 18587e4c3690SAndrew Morton mapping_set_error(page->mapping, err); 18591da177e4SLinus Torvalds set_page_writeback(page); 18601da177e4SLinus Torvalds do { 18611da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 18621da177e4SLinus Torvalds if (buffer_async_write(bh)) { 18631da177e4SLinus Torvalds clear_buffer_dirty(bh); 18648e8f9298SJens Axboe submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, 18658e8f9298SJens Axboe inode->i_write_hint, wbc); 18661da177e4SLinus Torvalds nr_underway++; 1867ad576e63SNick Piggin } 18681da177e4SLinus Torvalds bh = next; 18691da177e4SLinus Torvalds } while (bh != head); 1870ffda9d30SNick Piggin unlock_page(page); 18711da177e4SLinus Torvalds goto done; 18721da177e4SLinus Torvalds } 1873b4bba389SBenjamin Marzinski EXPORT_SYMBOL(__block_write_full_page); 18741da177e4SLinus Torvalds 1875afddba49SNick Piggin /* 1876afddba49SNick Piggin * If a page has any new buffers, zero them out here, and mark them uptodate 1877afddba49SNick Piggin * and dirty so they'll be written out (in order to prevent uninitialised 1878afddba49SNick Piggin * block data from leaking). And clear the new bit. 1879afddba49SNick Piggin */ 1880afddba49SNick Piggin void page_zero_new_buffers(struct page *page, unsigned from, unsigned to) 1881afddba49SNick Piggin { 1882afddba49SNick Piggin unsigned int block_start, block_end; 1883afddba49SNick Piggin struct buffer_head *head, *bh; 1884afddba49SNick Piggin 1885afddba49SNick Piggin BUG_ON(!PageLocked(page)); 1886afddba49SNick Piggin if (!page_has_buffers(page)) 1887afddba49SNick Piggin return; 1888afddba49SNick Piggin 1889afddba49SNick Piggin bh = head = page_buffers(page); 1890afddba49SNick Piggin block_start = 0; 1891afddba49SNick Piggin do { 1892afddba49SNick Piggin block_end = block_start + bh->b_size; 1893afddba49SNick Piggin 1894afddba49SNick Piggin if (buffer_new(bh)) { 1895afddba49SNick Piggin if (block_end > from && block_start < to) { 1896afddba49SNick Piggin if (!PageUptodate(page)) { 1897afddba49SNick Piggin unsigned start, size; 1898afddba49SNick Piggin 1899afddba49SNick Piggin start = max(from, block_start); 1900afddba49SNick Piggin size = min(to, block_end) - start; 1901afddba49SNick Piggin 1902eebd2aa3SChristoph Lameter zero_user(page, start, size); 1903afddba49SNick Piggin set_buffer_uptodate(bh); 1904afddba49SNick Piggin } 1905afddba49SNick Piggin 1906afddba49SNick Piggin clear_buffer_new(bh); 1907afddba49SNick Piggin mark_buffer_dirty(bh); 1908afddba49SNick Piggin } 1909afddba49SNick Piggin } 1910afddba49SNick Piggin 1911afddba49SNick Piggin block_start = block_end; 1912afddba49SNick Piggin bh = bh->b_this_page; 1913afddba49SNick Piggin } while (bh != head); 1914afddba49SNick Piggin } 1915afddba49SNick Piggin EXPORT_SYMBOL(page_zero_new_buffers); 1916afddba49SNick Piggin 1917ae259a9cSChristoph Hellwig static void 1918ae259a9cSChristoph Hellwig iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, 1919ae259a9cSChristoph Hellwig struct iomap *iomap) 1920ae259a9cSChristoph Hellwig { 1921ae259a9cSChristoph Hellwig loff_t offset = block << inode->i_blkbits; 1922ae259a9cSChristoph Hellwig 1923ae259a9cSChristoph Hellwig bh->b_bdev = iomap->bdev; 1924ae259a9cSChristoph Hellwig 1925ae259a9cSChristoph Hellwig /* 1926ae259a9cSChristoph Hellwig * Block points to offset in file we need to map, iomap contains 1927ae259a9cSChristoph Hellwig * the offset at which the map starts. If the map ends before the 1928ae259a9cSChristoph Hellwig * current block, then do not map the buffer and let the caller 1929ae259a9cSChristoph Hellwig * handle it. 1930ae259a9cSChristoph Hellwig */ 1931ae259a9cSChristoph Hellwig BUG_ON(offset >= iomap->offset + iomap->length); 1932ae259a9cSChristoph Hellwig 1933ae259a9cSChristoph Hellwig switch (iomap->type) { 1934ae259a9cSChristoph Hellwig case IOMAP_HOLE: 1935ae259a9cSChristoph Hellwig /* 1936ae259a9cSChristoph Hellwig * If the buffer is not up to date or beyond the current EOF, 1937ae259a9cSChristoph Hellwig * we need to mark it as new to ensure sub-block zeroing is 1938ae259a9cSChristoph Hellwig * executed if necessary. 1939ae259a9cSChristoph Hellwig */ 1940ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 1941ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 1942ae259a9cSChristoph Hellwig set_buffer_new(bh); 1943ae259a9cSChristoph Hellwig break; 1944ae259a9cSChristoph Hellwig case IOMAP_DELALLOC: 1945ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 1946ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 1947ae259a9cSChristoph Hellwig set_buffer_new(bh); 1948ae259a9cSChristoph Hellwig set_buffer_uptodate(bh); 1949ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 1950ae259a9cSChristoph Hellwig set_buffer_delay(bh); 1951ae259a9cSChristoph Hellwig break; 1952ae259a9cSChristoph Hellwig case IOMAP_UNWRITTEN: 1953ae259a9cSChristoph Hellwig /* 1954ae259a9cSChristoph Hellwig * For unwritten regions, we always need to ensure that 1955ae259a9cSChristoph Hellwig * sub-block writes cause the regions in the block we are not 1956ae259a9cSChristoph Hellwig * writing to are zeroed. Set the buffer as new to ensure this. 1957ae259a9cSChristoph Hellwig */ 1958ae259a9cSChristoph Hellwig set_buffer_new(bh); 1959ae259a9cSChristoph Hellwig set_buffer_unwritten(bh); 1960ae259a9cSChristoph Hellwig /* FALLTHRU */ 1961ae259a9cSChristoph Hellwig case IOMAP_MAPPED: 1962ae259a9cSChristoph Hellwig if (offset >= i_size_read(inode)) 1963ae259a9cSChristoph Hellwig set_buffer_new(bh); 1964ae259a9cSChristoph Hellwig bh->b_blocknr = (iomap->blkno >> (inode->i_blkbits - 9)) + 1965ae259a9cSChristoph Hellwig ((offset - iomap->offset) >> inode->i_blkbits); 1966ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 1967ae259a9cSChristoph Hellwig break; 1968ae259a9cSChristoph Hellwig } 1969ae259a9cSChristoph Hellwig } 1970ae259a9cSChristoph Hellwig 1971ae259a9cSChristoph Hellwig int __block_write_begin_int(struct page *page, loff_t pos, unsigned len, 1972ae259a9cSChristoph Hellwig get_block_t *get_block, struct iomap *iomap) 19731da177e4SLinus Torvalds { 197409cbfeafSKirill A. Shutemov unsigned from = pos & (PAGE_SIZE - 1); 1975ebdec241SChristoph Hellwig unsigned to = from + len; 19766e1db88dSChristoph Hellwig struct inode *inode = page->mapping->host; 19771da177e4SLinus Torvalds unsigned block_start, block_end; 19781da177e4SLinus Torvalds sector_t block; 19791da177e4SLinus Torvalds int err = 0; 19801da177e4SLinus Torvalds unsigned blocksize, bbits; 19811da177e4SLinus Torvalds struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; 19821da177e4SLinus Torvalds 19831da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 198409cbfeafSKirill A. Shutemov BUG_ON(from > PAGE_SIZE); 198509cbfeafSKirill A. Shutemov BUG_ON(to > PAGE_SIZE); 19861da177e4SLinus Torvalds BUG_ON(from > to); 19871da177e4SLinus Torvalds 198845bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 0); 198945bce8f3SLinus Torvalds blocksize = head->b_size; 199045bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 19911da177e4SLinus Torvalds 199209cbfeafSKirill A. Shutemov block = (sector_t)page->index << (PAGE_SHIFT - bbits); 19931da177e4SLinus Torvalds 19941da177e4SLinus Torvalds for(bh = head, block_start = 0; bh != head || !block_start; 19951da177e4SLinus Torvalds block++, block_start=block_end, bh = bh->b_this_page) { 19961da177e4SLinus Torvalds block_end = block_start + blocksize; 19971da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 19981da177e4SLinus Torvalds if (PageUptodate(page)) { 19991da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 20001da177e4SLinus Torvalds set_buffer_uptodate(bh); 20011da177e4SLinus Torvalds } 20021da177e4SLinus Torvalds continue; 20031da177e4SLinus Torvalds } 20041da177e4SLinus Torvalds if (buffer_new(bh)) 20051da177e4SLinus Torvalds clear_buffer_new(bh); 20061da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2007b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2008ae259a9cSChristoph Hellwig if (get_block) { 20091da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 20101da177e4SLinus Torvalds if (err) 2011f3ddbdc6SNick Piggin break; 2012ae259a9cSChristoph Hellwig } else { 2013ae259a9cSChristoph Hellwig iomap_to_bh(inode, block, bh, iomap); 2014ae259a9cSChristoph Hellwig } 2015ae259a9cSChristoph Hellwig 20161da177e4SLinus Torvalds if (buffer_new(bh)) { 2017e64855c6SJan Kara clean_bdev_bh_alias(bh); 20181da177e4SLinus Torvalds if (PageUptodate(page)) { 2019637aff46SNick Piggin clear_buffer_new(bh); 20201da177e4SLinus Torvalds set_buffer_uptodate(bh); 2021637aff46SNick Piggin mark_buffer_dirty(bh); 20221da177e4SLinus Torvalds continue; 20231da177e4SLinus Torvalds } 2024eebd2aa3SChristoph Lameter if (block_end > to || block_start < from) 2025eebd2aa3SChristoph Lameter zero_user_segments(page, 2026eebd2aa3SChristoph Lameter to, block_end, 2027eebd2aa3SChristoph Lameter block_start, from); 20281da177e4SLinus Torvalds continue; 20291da177e4SLinus Torvalds } 20301da177e4SLinus Torvalds } 20311da177e4SLinus Torvalds if (PageUptodate(page)) { 20321da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 20331da177e4SLinus Torvalds set_buffer_uptodate(bh); 20341da177e4SLinus Torvalds continue; 20351da177e4SLinus Torvalds } 20361da177e4SLinus Torvalds if (!buffer_uptodate(bh) && !buffer_delay(bh) && 203733a266ddSDavid Chinner !buffer_unwritten(bh) && 20381da177e4SLinus Torvalds (block_start < from || block_end > to)) { 2039dfec8a14SMike Christie ll_rw_block(REQ_OP_READ, 0, 1, &bh); 20401da177e4SLinus Torvalds *wait_bh++=bh; 20411da177e4SLinus Torvalds } 20421da177e4SLinus Torvalds } 20431da177e4SLinus Torvalds /* 20441da177e4SLinus Torvalds * If we issued read requests - let them complete. 20451da177e4SLinus Torvalds */ 20461da177e4SLinus Torvalds while(wait_bh > wait) { 20471da177e4SLinus Torvalds wait_on_buffer(*--wait_bh); 20481da177e4SLinus Torvalds if (!buffer_uptodate(*wait_bh)) 2049f3ddbdc6SNick Piggin err = -EIO; 20501da177e4SLinus Torvalds } 2051f9f07b6cSJan Kara if (unlikely(err)) 2052afddba49SNick Piggin page_zero_new_buffers(page, from, to); 20531da177e4SLinus Torvalds return err; 20541da177e4SLinus Torvalds } 2055ae259a9cSChristoph Hellwig 2056ae259a9cSChristoph Hellwig int __block_write_begin(struct page *page, loff_t pos, unsigned len, 2057ae259a9cSChristoph Hellwig get_block_t *get_block) 2058ae259a9cSChristoph Hellwig { 2059ae259a9cSChristoph Hellwig return __block_write_begin_int(page, pos, len, get_block, NULL); 2060ae259a9cSChristoph Hellwig } 2061ebdec241SChristoph Hellwig EXPORT_SYMBOL(__block_write_begin); 20621da177e4SLinus Torvalds 20631da177e4SLinus Torvalds static int __block_commit_write(struct inode *inode, struct page *page, 20641da177e4SLinus Torvalds unsigned from, unsigned to) 20651da177e4SLinus Torvalds { 20661da177e4SLinus Torvalds unsigned block_start, block_end; 20671da177e4SLinus Torvalds int partial = 0; 20681da177e4SLinus Torvalds unsigned blocksize; 20691da177e4SLinus Torvalds struct buffer_head *bh, *head; 20701da177e4SLinus Torvalds 207145bce8f3SLinus Torvalds bh = head = page_buffers(page); 207245bce8f3SLinus Torvalds blocksize = bh->b_size; 20731da177e4SLinus Torvalds 207445bce8f3SLinus Torvalds block_start = 0; 207545bce8f3SLinus Torvalds do { 20761da177e4SLinus Torvalds block_end = block_start + blocksize; 20771da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 20781da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 20791da177e4SLinus Torvalds partial = 1; 20801da177e4SLinus Torvalds } else { 20811da177e4SLinus Torvalds set_buffer_uptodate(bh); 20821da177e4SLinus Torvalds mark_buffer_dirty(bh); 20831da177e4SLinus Torvalds } 2084afddba49SNick Piggin clear_buffer_new(bh); 208545bce8f3SLinus Torvalds 208645bce8f3SLinus Torvalds block_start = block_end; 208745bce8f3SLinus Torvalds bh = bh->b_this_page; 208845bce8f3SLinus Torvalds } while (bh != head); 20891da177e4SLinus Torvalds 20901da177e4SLinus Torvalds /* 20911da177e4SLinus Torvalds * If this is a partial write which happened to make all buffers 20921da177e4SLinus Torvalds * uptodate then we can optimize away a bogus readpage() for 20931da177e4SLinus Torvalds * the next read(). Here we 'discover' whether the page went 20941da177e4SLinus Torvalds * uptodate as a result of this (potentially partial) write. 20951da177e4SLinus Torvalds */ 20961da177e4SLinus Torvalds if (!partial) 20971da177e4SLinus Torvalds SetPageUptodate(page); 20981da177e4SLinus Torvalds return 0; 20991da177e4SLinus Torvalds } 21001da177e4SLinus Torvalds 21011da177e4SLinus Torvalds /* 2102155130a4SChristoph Hellwig * block_write_begin takes care of the basic task of block allocation and 2103155130a4SChristoph Hellwig * bringing partial write blocks uptodate first. 2104155130a4SChristoph Hellwig * 21057bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 2106afddba49SNick Piggin */ 2107155130a4SChristoph Hellwig int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, 2108155130a4SChristoph Hellwig unsigned flags, struct page **pagep, get_block_t *get_block) 2109afddba49SNick Piggin { 211009cbfeafSKirill A. Shutemov pgoff_t index = pos >> PAGE_SHIFT; 2111afddba49SNick Piggin struct page *page; 21126e1db88dSChristoph Hellwig int status; 2113afddba49SNick Piggin 211454566b2cSNick Piggin page = grab_cache_page_write_begin(mapping, index, flags); 21156e1db88dSChristoph Hellwig if (!page) 21166e1db88dSChristoph Hellwig return -ENOMEM; 2117afddba49SNick Piggin 21186e1db88dSChristoph Hellwig status = __block_write_begin(page, pos, len, get_block); 2119afddba49SNick Piggin if (unlikely(status)) { 2120afddba49SNick Piggin unlock_page(page); 212109cbfeafSKirill A. Shutemov put_page(page); 21226e1db88dSChristoph Hellwig page = NULL; 2123afddba49SNick Piggin } 2124afddba49SNick Piggin 21256e1db88dSChristoph Hellwig *pagep = page; 2126afddba49SNick Piggin return status; 2127afddba49SNick Piggin } 2128afddba49SNick Piggin EXPORT_SYMBOL(block_write_begin); 2129afddba49SNick Piggin 2130afddba49SNick Piggin int block_write_end(struct file *file, struct address_space *mapping, 2131afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2132afddba49SNick Piggin struct page *page, void *fsdata) 2133afddba49SNick Piggin { 2134afddba49SNick Piggin struct inode *inode = mapping->host; 2135afddba49SNick Piggin unsigned start; 2136afddba49SNick Piggin 213709cbfeafSKirill A. Shutemov start = pos & (PAGE_SIZE - 1); 2138afddba49SNick Piggin 2139afddba49SNick Piggin if (unlikely(copied < len)) { 2140afddba49SNick Piggin /* 2141afddba49SNick Piggin * The buffers that were written will now be uptodate, so we 2142afddba49SNick Piggin * don't have to worry about a readpage reading them and 2143afddba49SNick Piggin * overwriting a partial write. However if we have encountered 2144afddba49SNick Piggin * a short write and only partially written into a buffer, it 2145afddba49SNick Piggin * will not be marked uptodate, so a readpage might come in and 2146afddba49SNick Piggin * destroy our partial write. 2147afddba49SNick Piggin * 2148afddba49SNick Piggin * Do the simplest thing, and just treat any short write to a 2149afddba49SNick Piggin * non uptodate page as a zero-length write, and force the 2150afddba49SNick Piggin * caller to redo the whole thing. 2151afddba49SNick Piggin */ 2152afddba49SNick Piggin if (!PageUptodate(page)) 2153afddba49SNick Piggin copied = 0; 2154afddba49SNick Piggin 2155afddba49SNick Piggin page_zero_new_buffers(page, start+copied, start+len); 2156afddba49SNick Piggin } 2157afddba49SNick Piggin flush_dcache_page(page); 2158afddba49SNick Piggin 2159afddba49SNick Piggin /* This could be a short (even 0-length) commit */ 2160afddba49SNick Piggin __block_commit_write(inode, page, start, start+copied); 2161afddba49SNick Piggin 2162afddba49SNick Piggin return copied; 2163afddba49SNick Piggin } 2164afddba49SNick Piggin EXPORT_SYMBOL(block_write_end); 2165afddba49SNick Piggin 2166afddba49SNick Piggin int generic_write_end(struct file *file, struct address_space *mapping, 2167afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2168afddba49SNick Piggin struct page *page, void *fsdata) 2169afddba49SNick Piggin { 2170afddba49SNick Piggin struct inode *inode = mapping->host; 217190a80202SJan Kara loff_t old_size = inode->i_size; 2172c7d206b3SJan Kara int i_size_changed = 0; 2173afddba49SNick Piggin 2174afddba49SNick Piggin copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); 2175afddba49SNick Piggin 2176afddba49SNick Piggin /* 2177afddba49SNick Piggin * No need to use i_size_read() here, the i_size 2178afddba49SNick Piggin * cannot change under us because we hold i_mutex. 2179afddba49SNick Piggin * 2180afddba49SNick Piggin * But it's important to update i_size while still holding page lock: 2181afddba49SNick Piggin * page writeout could otherwise come in and zero beyond i_size. 2182afddba49SNick Piggin */ 2183afddba49SNick Piggin if (pos+copied > inode->i_size) { 2184afddba49SNick Piggin i_size_write(inode, pos+copied); 2185c7d206b3SJan Kara i_size_changed = 1; 2186afddba49SNick Piggin } 2187afddba49SNick Piggin 2188afddba49SNick Piggin unlock_page(page); 218909cbfeafSKirill A. Shutemov put_page(page); 2190afddba49SNick Piggin 219190a80202SJan Kara if (old_size < pos) 219290a80202SJan Kara pagecache_isize_extended(inode, old_size, pos); 2193c7d206b3SJan Kara /* 2194c7d206b3SJan Kara * Don't mark the inode dirty under page lock. First, it unnecessarily 2195c7d206b3SJan Kara * makes the holding time of page lock longer. Second, it forces lock 2196c7d206b3SJan Kara * ordering of page lock and transaction start for journaling 2197c7d206b3SJan Kara * filesystems. 2198c7d206b3SJan Kara */ 2199c7d206b3SJan Kara if (i_size_changed) 2200c7d206b3SJan Kara mark_inode_dirty(inode); 2201c7d206b3SJan Kara 2202afddba49SNick Piggin return copied; 2203afddba49SNick Piggin } 2204afddba49SNick Piggin EXPORT_SYMBOL(generic_write_end); 2205afddba49SNick Piggin 2206afddba49SNick Piggin /* 22078ab22b9aSHisashi Hifumi * block_is_partially_uptodate checks whether buffers within a page are 22088ab22b9aSHisashi Hifumi * uptodate or not. 22098ab22b9aSHisashi Hifumi * 22108ab22b9aSHisashi Hifumi * Returns true if all buffers which correspond to a file portion 22118ab22b9aSHisashi Hifumi * we want to read are uptodate. 22128ab22b9aSHisashi Hifumi */ 2213c186afb4SAl Viro int block_is_partially_uptodate(struct page *page, unsigned long from, 2214c186afb4SAl Viro unsigned long count) 22158ab22b9aSHisashi Hifumi { 22168ab22b9aSHisashi Hifumi unsigned block_start, block_end, blocksize; 22178ab22b9aSHisashi Hifumi unsigned to; 22188ab22b9aSHisashi Hifumi struct buffer_head *bh, *head; 22198ab22b9aSHisashi Hifumi int ret = 1; 22208ab22b9aSHisashi Hifumi 22218ab22b9aSHisashi Hifumi if (!page_has_buffers(page)) 22228ab22b9aSHisashi Hifumi return 0; 22238ab22b9aSHisashi Hifumi 222445bce8f3SLinus Torvalds head = page_buffers(page); 222545bce8f3SLinus Torvalds blocksize = head->b_size; 222609cbfeafSKirill A. Shutemov to = min_t(unsigned, PAGE_SIZE - from, count); 22278ab22b9aSHisashi Hifumi to = from + to; 222809cbfeafSKirill A. Shutemov if (from < blocksize && to > PAGE_SIZE - blocksize) 22298ab22b9aSHisashi Hifumi return 0; 22308ab22b9aSHisashi Hifumi 22318ab22b9aSHisashi Hifumi bh = head; 22328ab22b9aSHisashi Hifumi block_start = 0; 22338ab22b9aSHisashi Hifumi do { 22348ab22b9aSHisashi Hifumi block_end = block_start + blocksize; 22358ab22b9aSHisashi Hifumi if (block_end > from && block_start < to) { 22368ab22b9aSHisashi Hifumi if (!buffer_uptodate(bh)) { 22378ab22b9aSHisashi Hifumi ret = 0; 22388ab22b9aSHisashi Hifumi break; 22398ab22b9aSHisashi Hifumi } 22408ab22b9aSHisashi Hifumi if (block_end >= to) 22418ab22b9aSHisashi Hifumi break; 22428ab22b9aSHisashi Hifumi } 22438ab22b9aSHisashi Hifumi block_start = block_end; 22448ab22b9aSHisashi Hifumi bh = bh->b_this_page; 22458ab22b9aSHisashi Hifumi } while (bh != head); 22468ab22b9aSHisashi Hifumi 22478ab22b9aSHisashi Hifumi return ret; 22488ab22b9aSHisashi Hifumi } 22498ab22b9aSHisashi Hifumi EXPORT_SYMBOL(block_is_partially_uptodate); 22508ab22b9aSHisashi Hifumi 22518ab22b9aSHisashi Hifumi /* 22521da177e4SLinus Torvalds * Generic "read page" function for block devices that have the normal 22531da177e4SLinus Torvalds * get_block functionality. This is most of the block device filesystems. 22541da177e4SLinus Torvalds * Reads the page asynchronously --- the unlock_buffer() and 22551da177e4SLinus Torvalds * set/clear_buffer_uptodate() functions propagate buffer state into the 22561da177e4SLinus Torvalds * page struct once IO has completed. 22571da177e4SLinus Torvalds */ 22581da177e4SLinus Torvalds int block_read_full_page(struct page *page, get_block_t *get_block) 22591da177e4SLinus Torvalds { 22601da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 22611da177e4SLinus Torvalds sector_t iblock, lblock; 22621da177e4SLinus Torvalds struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 226345bce8f3SLinus Torvalds unsigned int blocksize, bbits; 22641da177e4SLinus Torvalds int nr, i; 22651da177e4SLinus Torvalds int fully_mapped = 1; 22661da177e4SLinus Torvalds 226745bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 0); 226845bce8f3SLinus Torvalds blocksize = head->b_size; 226945bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 22701da177e4SLinus Torvalds 227109cbfeafSKirill A. Shutemov iblock = (sector_t)page->index << (PAGE_SHIFT - bbits); 227245bce8f3SLinus Torvalds lblock = (i_size_read(inode)+blocksize-1) >> bbits; 22731da177e4SLinus Torvalds bh = head; 22741da177e4SLinus Torvalds nr = 0; 22751da177e4SLinus Torvalds i = 0; 22761da177e4SLinus Torvalds 22771da177e4SLinus Torvalds do { 22781da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22791da177e4SLinus Torvalds continue; 22801da177e4SLinus Torvalds 22811da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2282c64610baSAndrew Morton int err = 0; 2283c64610baSAndrew Morton 22841da177e4SLinus Torvalds fully_mapped = 0; 22851da177e4SLinus Torvalds if (iblock < lblock) { 2286b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2287c64610baSAndrew Morton err = get_block(inode, iblock, bh, 0); 2288c64610baSAndrew Morton if (err) 22891da177e4SLinus Torvalds SetPageError(page); 22901da177e4SLinus Torvalds } 22911da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2292eebd2aa3SChristoph Lameter zero_user(page, i * blocksize, blocksize); 2293c64610baSAndrew Morton if (!err) 22941da177e4SLinus Torvalds set_buffer_uptodate(bh); 22951da177e4SLinus Torvalds continue; 22961da177e4SLinus Torvalds } 22971da177e4SLinus Torvalds /* 22981da177e4SLinus Torvalds * get_block() might have updated the buffer 22991da177e4SLinus Torvalds * synchronously 23001da177e4SLinus Torvalds */ 23011da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23021da177e4SLinus Torvalds continue; 23031da177e4SLinus Torvalds } 23041da177e4SLinus Torvalds arr[nr++] = bh; 23051da177e4SLinus Torvalds } while (i++, iblock++, (bh = bh->b_this_page) != head); 23061da177e4SLinus Torvalds 23071da177e4SLinus Torvalds if (fully_mapped) 23081da177e4SLinus Torvalds SetPageMappedToDisk(page); 23091da177e4SLinus Torvalds 23101da177e4SLinus Torvalds if (!nr) { 23111da177e4SLinus Torvalds /* 23121da177e4SLinus Torvalds * All buffers are uptodate - we can set the page uptodate 23131da177e4SLinus Torvalds * as well. But not if get_block() returned an error. 23141da177e4SLinus Torvalds */ 23151da177e4SLinus Torvalds if (!PageError(page)) 23161da177e4SLinus Torvalds SetPageUptodate(page); 23171da177e4SLinus Torvalds unlock_page(page); 23181da177e4SLinus Torvalds return 0; 23191da177e4SLinus Torvalds } 23201da177e4SLinus Torvalds 23211da177e4SLinus Torvalds /* Stage two: lock the buffers */ 23221da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 23231da177e4SLinus Torvalds bh = arr[i]; 23241da177e4SLinus Torvalds lock_buffer(bh); 23251da177e4SLinus Torvalds mark_buffer_async_read(bh); 23261da177e4SLinus Torvalds } 23271da177e4SLinus Torvalds 23281da177e4SLinus Torvalds /* 23291da177e4SLinus Torvalds * Stage 3: start the IO. Check for uptodateness 23301da177e4SLinus Torvalds * inside the buffer lock in case another process reading 23311da177e4SLinus Torvalds * the underlying blockdev brought it uptodate (the sct fix). 23321da177e4SLinus Torvalds */ 23331da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 23341da177e4SLinus Torvalds bh = arr[i]; 23351da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23361da177e4SLinus Torvalds end_buffer_async_read(bh, 1); 23371da177e4SLinus Torvalds else 23382a222ca9SMike Christie submit_bh(REQ_OP_READ, 0, bh); 23391da177e4SLinus Torvalds } 23401da177e4SLinus Torvalds return 0; 23411da177e4SLinus Torvalds } 23421fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_read_full_page); 23431da177e4SLinus Torvalds 23441da177e4SLinus Torvalds /* utility function for filesystems that need to do work on expanding 234589e10787SNick Piggin * truncates. Uses filesystem pagecache writes to allow the filesystem to 23461da177e4SLinus Torvalds * deal with the hole. 23471da177e4SLinus Torvalds */ 234889e10787SNick Piggin int generic_cont_expand_simple(struct inode *inode, loff_t size) 23491da177e4SLinus Torvalds { 23501da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 23511da177e4SLinus Torvalds struct page *page; 235289e10787SNick Piggin void *fsdata; 23531da177e4SLinus Torvalds int err; 23541da177e4SLinus Torvalds 2355c08d3b0eSnpiggin@suse.de err = inode_newsize_ok(inode, size); 2356c08d3b0eSnpiggin@suse.de if (err) 23571da177e4SLinus Torvalds goto out; 23581da177e4SLinus Torvalds 235989e10787SNick Piggin err = pagecache_write_begin(NULL, mapping, size, 0, 2360c718a975STetsuo Handa AOP_FLAG_CONT_EXPAND, &page, &fsdata); 236189e10787SNick Piggin if (err) 236205eb0b51SOGAWA Hirofumi goto out; 236305eb0b51SOGAWA Hirofumi 236489e10787SNick Piggin err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata); 236589e10787SNick Piggin BUG_ON(err > 0); 236605eb0b51SOGAWA Hirofumi 236705eb0b51SOGAWA Hirofumi out: 236805eb0b51SOGAWA Hirofumi return err; 236905eb0b51SOGAWA Hirofumi } 23701fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_cont_expand_simple); 237105eb0b51SOGAWA Hirofumi 2372f1e3af72SAdrian Bunk static int cont_expand_zero(struct file *file, struct address_space *mapping, 237389e10787SNick Piggin loff_t pos, loff_t *bytes) 237405eb0b51SOGAWA Hirofumi { 237589e10787SNick Piggin struct inode *inode = mapping->host; 237693407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 237789e10787SNick Piggin struct page *page; 237889e10787SNick Piggin void *fsdata; 237989e10787SNick Piggin pgoff_t index, curidx; 238089e10787SNick Piggin loff_t curpos; 238189e10787SNick Piggin unsigned zerofrom, offset, len; 238289e10787SNick Piggin int err = 0; 238305eb0b51SOGAWA Hirofumi 238409cbfeafSKirill A. Shutemov index = pos >> PAGE_SHIFT; 238509cbfeafSKirill A. Shutemov offset = pos & ~PAGE_MASK; 238689e10787SNick Piggin 238709cbfeafSKirill A. Shutemov while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { 238809cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 238989e10787SNick Piggin if (zerofrom & (blocksize-1)) { 239089e10787SNick Piggin *bytes |= (blocksize-1); 239189e10787SNick Piggin (*bytes)++; 239289e10787SNick Piggin } 239309cbfeafSKirill A. Shutemov len = PAGE_SIZE - zerofrom; 239489e10787SNick Piggin 2395c718a975STetsuo Handa err = pagecache_write_begin(file, mapping, curpos, len, 0, 239689e10787SNick Piggin &page, &fsdata); 239789e10787SNick Piggin if (err) 239889e10787SNick Piggin goto out; 2399eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 240089e10787SNick Piggin err = pagecache_write_end(file, mapping, curpos, len, len, 240189e10787SNick Piggin page, fsdata); 240289e10787SNick Piggin if (err < 0) 240389e10787SNick Piggin goto out; 240489e10787SNick Piggin BUG_ON(err != len); 240589e10787SNick Piggin err = 0; 2406061e9746SOGAWA Hirofumi 2407061e9746SOGAWA Hirofumi balance_dirty_pages_ratelimited(mapping); 2408c2ca0fcdSMikulas Patocka 2409c2ca0fcdSMikulas Patocka if (unlikely(fatal_signal_pending(current))) { 2410c2ca0fcdSMikulas Patocka err = -EINTR; 2411c2ca0fcdSMikulas Patocka goto out; 2412c2ca0fcdSMikulas Patocka } 241389e10787SNick Piggin } 241489e10787SNick Piggin 241589e10787SNick Piggin /* page covers the boundary, find the boundary offset */ 241689e10787SNick Piggin if (index == curidx) { 241709cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 241889e10787SNick Piggin /* if we will expand the thing last block will be filled */ 241989e10787SNick Piggin if (offset <= zerofrom) { 242089e10787SNick Piggin goto out; 242189e10787SNick Piggin } 242289e10787SNick Piggin if (zerofrom & (blocksize-1)) { 242389e10787SNick Piggin *bytes |= (blocksize-1); 242489e10787SNick Piggin (*bytes)++; 242589e10787SNick Piggin } 242689e10787SNick Piggin len = offset - zerofrom; 242789e10787SNick Piggin 2428c718a975STetsuo Handa err = pagecache_write_begin(file, mapping, curpos, len, 0, 242989e10787SNick Piggin &page, &fsdata); 243089e10787SNick Piggin if (err) 243189e10787SNick Piggin goto out; 2432eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 243389e10787SNick Piggin err = pagecache_write_end(file, mapping, curpos, len, len, 243489e10787SNick Piggin page, fsdata); 243589e10787SNick Piggin if (err < 0) 243689e10787SNick Piggin goto out; 243789e10787SNick Piggin BUG_ON(err != len); 243889e10787SNick Piggin err = 0; 243989e10787SNick Piggin } 244089e10787SNick Piggin out: 244189e10787SNick Piggin return err; 24421da177e4SLinus Torvalds } 24431da177e4SLinus Torvalds 24441da177e4SLinus Torvalds /* 24451da177e4SLinus Torvalds * For moronic filesystems that do not allow holes in file. 24461da177e4SLinus Torvalds * We may have to extend the file. 24471da177e4SLinus Torvalds */ 2448282dc178SChristoph Hellwig int cont_write_begin(struct file *file, struct address_space *mapping, 244989e10787SNick Piggin loff_t pos, unsigned len, unsigned flags, 245089e10787SNick Piggin struct page **pagep, void **fsdata, 245189e10787SNick Piggin get_block_t *get_block, loff_t *bytes) 24521da177e4SLinus Torvalds { 24531da177e4SLinus Torvalds struct inode *inode = mapping->host; 245493407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 245593407472SFabian Frederick unsigned int zerofrom; 245689e10787SNick Piggin int err; 24571da177e4SLinus Torvalds 245889e10787SNick Piggin err = cont_expand_zero(file, mapping, pos, bytes); 245989e10787SNick Piggin if (err) 2460155130a4SChristoph Hellwig return err; 24611da177e4SLinus Torvalds 246209cbfeafSKirill A. Shutemov zerofrom = *bytes & ~PAGE_MASK; 246389e10787SNick Piggin if (pos+len > *bytes && zerofrom & (blocksize-1)) { 24641da177e4SLinus Torvalds *bytes |= (blocksize-1); 24651da177e4SLinus Torvalds (*bytes)++; 24661da177e4SLinus Torvalds } 24671da177e4SLinus Torvalds 2468155130a4SChristoph Hellwig return block_write_begin(mapping, pos, len, flags, pagep, get_block); 24691da177e4SLinus Torvalds } 24701fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(cont_write_begin); 24711da177e4SLinus Torvalds 24721da177e4SLinus Torvalds int block_commit_write(struct page *page, unsigned from, unsigned to) 24731da177e4SLinus Torvalds { 24741da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 24751da177e4SLinus Torvalds __block_commit_write(inode,page,from,to); 24761da177e4SLinus Torvalds return 0; 24771da177e4SLinus Torvalds } 24781fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_commit_write); 24791da177e4SLinus Torvalds 248054171690SDavid Chinner /* 248154171690SDavid Chinner * block_page_mkwrite() is not allowed to change the file size as it gets 248254171690SDavid Chinner * called from a page fault handler when a page is first dirtied. Hence we must 248354171690SDavid Chinner * be careful to check for EOF conditions here. We set the page up correctly 248454171690SDavid Chinner * for a written page which means we get ENOSPC checking when writing into 248554171690SDavid Chinner * holes and correct delalloc and unwritten extent mapping on filesystems that 248654171690SDavid Chinner * support these features. 248754171690SDavid Chinner * 248854171690SDavid Chinner * We are not allowed to take the i_mutex here so we have to play games to 248954171690SDavid Chinner * protect against truncate races as the page could now be beyond EOF. Because 24907bb46a67Snpiggin@suse.de * truncate writes the inode size before removing pages, once we have the 249154171690SDavid Chinner * page lock we can determine safely if the page is beyond EOF. If it is not 249254171690SDavid Chinner * beyond EOF, then the page is guaranteed safe against truncation until we 249354171690SDavid Chinner * unlock the page. 2494ea13a864SJan Kara * 249514da9200SJan Kara * Direct callers of this function should protect against filesystem freezing 24965c500029SRoss Zwisler * using sb_start_pagefault() - sb_end_pagefault() functions. 249754171690SDavid Chinner */ 24985c500029SRoss Zwisler int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 249954171690SDavid Chinner get_block_t get_block) 250054171690SDavid Chinner { 2501c2ec175cSNick Piggin struct page *page = vmf->page; 2502496ad9aaSAl Viro struct inode *inode = file_inode(vma->vm_file); 250354171690SDavid Chinner unsigned long end; 250454171690SDavid Chinner loff_t size; 250524da4fabSJan Kara int ret; 250654171690SDavid Chinner 250754171690SDavid Chinner lock_page(page); 250854171690SDavid Chinner size = i_size_read(inode); 250954171690SDavid Chinner if ((page->mapping != inode->i_mapping) || 251018336338SNick Piggin (page_offset(page) > size)) { 251124da4fabSJan Kara /* We overload EFAULT to mean page got truncated */ 251224da4fabSJan Kara ret = -EFAULT; 251324da4fabSJan Kara goto out_unlock; 251454171690SDavid Chinner } 251554171690SDavid Chinner 251654171690SDavid Chinner /* page is wholly or partially inside EOF */ 251709cbfeafSKirill A. Shutemov if (((page->index + 1) << PAGE_SHIFT) > size) 251809cbfeafSKirill A. Shutemov end = size & ~PAGE_MASK; 251954171690SDavid Chinner else 252009cbfeafSKirill A. Shutemov end = PAGE_SIZE; 252154171690SDavid Chinner 2522ebdec241SChristoph Hellwig ret = __block_write_begin(page, 0, end, get_block); 252354171690SDavid Chinner if (!ret) 252454171690SDavid Chinner ret = block_commit_write(page, 0, end); 252554171690SDavid Chinner 252624da4fabSJan Kara if (unlikely(ret < 0)) 252724da4fabSJan Kara goto out_unlock; 2528ea13a864SJan Kara set_page_dirty(page); 25291d1d1a76SDarrick J. Wong wait_for_stable_page(page); 253024da4fabSJan Kara return 0; 253124da4fabSJan Kara out_unlock: 2532b827e496SNick Piggin unlock_page(page); 253354171690SDavid Chinner return ret; 253454171690SDavid Chinner } 25351fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_page_mkwrite); 25361da177e4SLinus Torvalds 25371da177e4SLinus Torvalds /* 253803158cd7SNick Piggin * nobh_write_begin()'s prereads are special: the buffer_heads are freed 25391da177e4SLinus Torvalds * immediately, while under the page lock. So it needs a special end_io 25401da177e4SLinus Torvalds * handler which does not touch the bh after unlocking it. 25411da177e4SLinus Torvalds */ 25421da177e4SLinus Torvalds static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate) 25431da177e4SLinus Torvalds { 254468671f35SDmitry Monakhov __end_buffer_read_notouch(bh, uptodate); 25451da177e4SLinus Torvalds } 25461da177e4SLinus Torvalds 25471da177e4SLinus Torvalds /* 254803158cd7SNick Piggin * Attach the singly-linked list of buffers created by nobh_write_begin, to 254903158cd7SNick Piggin * the page (converting it to circular linked list and taking care of page 255003158cd7SNick Piggin * dirty races). 255103158cd7SNick Piggin */ 255203158cd7SNick Piggin static void attach_nobh_buffers(struct page *page, struct buffer_head *head) 255303158cd7SNick Piggin { 255403158cd7SNick Piggin struct buffer_head *bh; 255503158cd7SNick Piggin 255603158cd7SNick Piggin BUG_ON(!PageLocked(page)); 255703158cd7SNick Piggin 255803158cd7SNick Piggin spin_lock(&page->mapping->private_lock); 255903158cd7SNick Piggin bh = head; 256003158cd7SNick Piggin do { 256103158cd7SNick Piggin if (PageDirty(page)) 256203158cd7SNick Piggin set_buffer_dirty(bh); 256303158cd7SNick Piggin if (!bh->b_this_page) 256403158cd7SNick Piggin bh->b_this_page = head; 256503158cd7SNick Piggin bh = bh->b_this_page; 256603158cd7SNick Piggin } while (bh != head); 256703158cd7SNick Piggin attach_page_buffers(page, head); 256803158cd7SNick Piggin spin_unlock(&page->mapping->private_lock); 256903158cd7SNick Piggin } 257003158cd7SNick Piggin 257103158cd7SNick Piggin /* 2572ea0f04e5SChristoph Hellwig * On entry, the page is fully not uptodate. 2573ea0f04e5SChristoph Hellwig * On exit the page is fully uptodate in the areas outside (from,to) 25747bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 25751da177e4SLinus Torvalds */ 2576ea0f04e5SChristoph Hellwig int nobh_write_begin(struct address_space *mapping, 257703158cd7SNick Piggin loff_t pos, unsigned len, unsigned flags, 257803158cd7SNick Piggin struct page **pagep, void **fsdata, 25791da177e4SLinus Torvalds get_block_t *get_block) 25801da177e4SLinus Torvalds { 258103158cd7SNick Piggin struct inode *inode = mapping->host; 25821da177e4SLinus Torvalds const unsigned blkbits = inode->i_blkbits; 25831da177e4SLinus Torvalds const unsigned blocksize = 1 << blkbits; 2584a4b0672dSNick Piggin struct buffer_head *head, *bh; 258503158cd7SNick Piggin struct page *page; 258603158cd7SNick Piggin pgoff_t index; 258703158cd7SNick Piggin unsigned from, to; 25881da177e4SLinus Torvalds unsigned block_in_page; 2589a4b0672dSNick Piggin unsigned block_start, block_end; 25901da177e4SLinus Torvalds sector_t block_in_file; 25911da177e4SLinus Torvalds int nr_reads = 0; 25921da177e4SLinus Torvalds int ret = 0; 25931da177e4SLinus Torvalds int is_mapped_to_disk = 1; 25941da177e4SLinus Torvalds 259509cbfeafSKirill A. Shutemov index = pos >> PAGE_SHIFT; 259609cbfeafSKirill A. Shutemov from = pos & (PAGE_SIZE - 1); 259703158cd7SNick Piggin to = from + len; 259803158cd7SNick Piggin 259954566b2cSNick Piggin page = grab_cache_page_write_begin(mapping, index, flags); 260003158cd7SNick Piggin if (!page) 260103158cd7SNick Piggin return -ENOMEM; 260203158cd7SNick Piggin *pagep = page; 260303158cd7SNick Piggin *fsdata = NULL; 260403158cd7SNick Piggin 260503158cd7SNick Piggin if (page_has_buffers(page)) { 2606309f77adSNamhyung Kim ret = __block_write_begin(page, pos, len, get_block); 2607309f77adSNamhyung Kim if (unlikely(ret)) 2608309f77adSNamhyung Kim goto out_release; 2609309f77adSNamhyung Kim return ret; 261003158cd7SNick Piggin } 2611a4b0672dSNick Piggin 26121da177e4SLinus Torvalds if (PageMappedToDisk(page)) 26131da177e4SLinus Torvalds return 0; 26141da177e4SLinus Torvalds 2615a4b0672dSNick Piggin /* 2616a4b0672dSNick Piggin * Allocate buffers so that we can keep track of state, and potentially 2617a4b0672dSNick Piggin * attach them to the page if an error occurs. In the common case of 2618a4b0672dSNick Piggin * no error, they will just be freed again without ever being attached 2619a4b0672dSNick Piggin * to the page (which is all OK, because we're under the page lock). 2620a4b0672dSNick Piggin * 2621a4b0672dSNick Piggin * Be careful: the buffer linked list is a NULL terminated one, rather 2622a4b0672dSNick Piggin * than the circular one we're used to. 2623a4b0672dSNick Piggin */ 2624640ab98fSJens Axboe head = alloc_page_buffers(page, blocksize, false); 262503158cd7SNick Piggin if (!head) { 262603158cd7SNick Piggin ret = -ENOMEM; 262703158cd7SNick Piggin goto out_release; 262803158cd7SNick Piggin } 2629a4b0672dSNick Piggin 263009cbfeafSKirill A. Shutemov block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits); 26311da177e4SLinus Torvalds 26321da177e4SLinus Torvalds /* 26331da177e4SLinus Torvalds * We loop across all blocks in the page, whether or not they are 26341da177e4SLinus Torvalds * part of the affected region. This is so we can discover if the 26351da177e4SLinus Torvalds * page is fully mapped-to-disk. 26361da177e4SLinus Torvalds */ 2637a4b0672dSNick Piggin for (block_start = 0, block_in_page = 0, bh = head; 263809cbfeafSKirill A. Shutemov block_start < PAGE_SIZE; 2639a4b0672dSNick Piggin block_in_page++, block_start += blocksize, bh = bh->b_this_page) { 26401da177e4SLinus Torvalds int create; 26411da177e4SLinus Torvalds 2642a4b0672dSNick Piggin block_end = block_start + blocksize; 2643a4b0672dSNick Piggin bh->b_state = 0; 26441da177e4SLinus Torvalds create = 1; 26451da177e4SLinus Torvalds if (block_start >= to) 26461da177e4SLinus Torvalds create = 0; 26471da177e4SLinus Torvalds ret = get_block(inode, block_in_file + block_in_page, 2648a4b0672dSNick Piggin bh, create); 26491da177e4SLinus Torvalds if (ret) 26501da177e4SLinus Torvalds goto failed; 2651a4b0672dSNick Piggin if (!buffer_mapped(bh)) 26521da177e4SLinus Torvalds is_mapped_to_disk = 0; 2653a4b0672dSNick Piggin if (buffer_new(bh)) 2654e64855c6SJan Kara clean_bdev_bh_alias(bh); 2655a4b0672dSNick Piggin if (PageUptodate(page)) { 2656a4b0672dSNick Piggin set_buffer_uptodate(bh); 26571da177e4SLinus Torvalds continue; 2658a4b0672dSNick Piggin } 2659a4b0672dSNick Piggin if (buffer_new(bh) || !buffer_mapped(bh)) { 2660eebd2aa3SChristoph Lameter zero_user_segments(page, block_start, from, 2661eebd2aa3SChristoph Lameter to, block_end); 26621da177e4SLinus Torvalds continue; 26631da177e4SLinus Torvalds } 2664a4b0672dSNick Piggin if (buffer_uptodate(bh)) 26651da177e4SLinus Torvalds continue; /* reiserfs does this */ 26661da177e4SLinus Torvalds if (block_start < from || block_end > to) { 2667a4b0672dSNick Piggin lock_buffer(bh); 2668a4b0672dSNick Piggin bh->b_end_io = end_buffer_read_nobh; 26692a222ca9SMike Christie submit_bh(REQ_OP_READ, 0, bh); 2670a4b0672dSNick Piggin nr_reads++; 26711da177e4SLinus Torvalds } 26721da177e4SLinus Torvalds } 26731da177e4SLinus Torvalds 26741da177e4SLinus Torvalds if (nr_reads) { 26751da177e4SLinus Torvalds /* 26761da177e4SLinus Torvalds * The page is locked, so these buffers are protected from 26771da177e4SLinus Torvalds * any VM or truncate activity. Hence we don't need to care 26781da177e4SLinus Torvalds * for the buffer_head refcounts. 26791da177e4SLinus Torvalds */ 2680a4b0672dSNick Piggin for (bh = head; bh; bh = bh->b_this_page) { 26811da177e4SLinus Torvalds wait_on_buffer(bh); 26821da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 26831da177e4SLinus Torvalds ret = -EIO; 26841da177e4SLinus Torvalds } 26851da177e4SLinus Torvalds if (ret) 26861da177e4SLinus Torvalds goto failed; 26871da177e4SLinus Torvalds } 26881da177e4SLinus Torvalds 26891da177e4SLinus Torvalds if (is_mapped_to_disk) 26901da177e4SLinus Torvalds SetPageMappedToDisk(page); 26911da177e4SLinus Torvalds 269203158cd7SNick Piggin *fsdata = head; /* to be released by nobh_write_end */ 2693a4b0672dSNick Piggin 26941da177e4SLinus Torvalds return 0; 26951da177e4SLinus Torvalds 26961da177e4SLinus Torvalds failed: 269703158cd7SNick Piggin BUG_ON(!ret); 26981da177e4SLinus Torvalds /* 2699a4b0672dSNick Piggin * Error recovery is a bit difficult. We need to zero out blocks that 2700a4b0672dSNick Piggin * were newly allocated, and dirty them to ensure they get written out. 2701a4b0672dSNick Piggin * Buffers need to be attached to the page at this point, otherwise 2702a4b0672dSNick Piggin * the handling of potential IO errors during writeout would be hard 2703a4b0672dSNick Piggin * (could try doing synchronous writeout, but what if that fails too?) 27041da177e4SLinus Torvalds */ 270503158cd7SNick Piggin attach_nobh_buffers(page, head); 270603158cd7SNick Piggin page_zero_new_buffers(page, from, to); 2707a4b0672dSNick Piggin 270803158cd7SNick Piggin out_release: 270903158cd7SNick Piggin unlock_page(page); 271009cbfeafSKirill A. Shutemov put_page(page); 271103158cd7SNick Piggin *pagep = NULL; 2712a4b0672dSNick Piggin 27137bb46a67Snpiggin@suse.de return ret; 27147bb46a67Snpiggin@suse.de } 271503158cd7SNick Piggin EXPORT_SYMBOL(nobh_write_begin); 27161da177e4SLinus Torvalds 271703158cd7SNick Piggin int nobh_write_end(struct file *file, struct address_space *mapping, 271803158cd7SNick Piggin loff_t pos, unsigned len, unsigned copied, 271903158cd7SNick Piggin struct page *page, void *fsdata) 27201da177e4SLinus Torvalds { 27211da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 2722efdc3131SNick Piggin struct buffer_head *head = fsdata; 272303158cd7SNick Piggin struct buffer_head *bh; 27245b41e74aSDmitri Monakhov BUG_ON(fsdata != NULL && page_has_buffers(page)); 27251da177e4SLinus Torvalds 2726d4cf109fSDave Kleikamp if (unlikely(copied < len) && head) 272703158cd7SNick Piggin attach_nobh_buffers(page, head); 2728a4b0672dSNick Piggin if (page_has_buffers(page)) 272903158cd7SNick Piggin return generic_write_end(file, mapping, pos, len, 273003158cd7SNick Piggin copied, page, fsdata); 2731a4b0672dSNick Piggin 273222c8ca78SNick Piggin SetPageUptodate(page); 27331da177e4SLinus Torvalds set_page_dirty(page); 273403158cd7SNick Piggin if (pos+copied > inode->i_size) { 273503158cd7SNick Piggin i_size_write(inode, pos+copied); 27361da177e4SLinus Torvalds mark_inode_dirty(inode); 27371da177e4SLinus Torvalds } 273803158cd7SNick Piggin 273903158cd7SNick Piggin unlock_page(page); 274009cbfeafSKirill A. Shutemov put_page(page); 274103158cd7SNick Piggin 274203158cd7SNick Piggin while (head) { 274303158cd7SNick Piggin bh = head; 274403158cd7SNick Piggin head = head->b_this_page; 274503158cd7SNick Piggin free_buffer_head(bh); 27461da177e4SLinus Torvalds } 274703158cd7SNick Piggin 274803158cd7SNick Piggin return copied; 274903158cd7SNick Piggin } 275003158cd7SNick Piggin EXPORT_SYMBOL(nobh_write_end); 27511da177e4SLinus Torvalds 27521da177e4SLinus Torvalds /* 27531da177e4SLinus Torvalds * nobh_writepage() - based on block_full_write_page() except 27541da177e4SLinus Torvalds * that it tries to operate without attaching bufferheads to 27551da177e4SLinus Torvalds * the page. 27561da177e4SLinus Torvalds */ 27571da177e4SLinus Torvalds int nobh_writepage(struct page *page, get_block_t *get_block, 27581da177e4SLinus Torvalds struct writeback_control *wbc) 27591da177e4SLinus Torvalds { 27601da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 27611da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 276209cbfeafSKirill A. Shutemov const pgoff_t end_index = i_size >> PAGE_SHIFT; 27631da177e4SLinus Torvalds unsigned offset; 27641da177e4SLinus Torvalds int ret; 27651da177e4SLinus Torvalds 27661da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 27671da177e4SLinus Torvalds if (page->index < end_index) 27681da177e4SLinus Torvalds goto out; 27691da177e4SLinus Torvalds 27701da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 277109cbfeafSKirill A. Shutemov offset = i_size & (PAGE_SIZE-1); 27721da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 27731da177e4SLinus Torvalds /* 27741da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 27751da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 27761da177e4SLinus Torvalds * freeable here, so the page does not leak. 27771da177e4SLinus Torvalds */ 27781da177e4SLinus Torvalds #if 0 27791da177e4SLinus Torvalds /* Not really sure about this - do we need this ? */ 27801da177e4SLinus Torvalds if (page->mapping->a_ops->invalidatepage) 27811da177e4SLinus Torvalds page->mapping->a_ops->invalidatepage(page, offset); 27821da177e4SLinus Torvalds #endif 27831da177e4SLinus Torvalds unlock_page(page); 27841da177e4SLinus Torvalds return 0; /* don't care */ 27851da177e4SLinus Torvalds } 27861da177e4SLinus Torvalds 27871da177e4SLinus Torvalds /* 27881da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 27891da177e4SLinus Torvalds * writepage invocation because it may be mmapped. "A file is mapped 27901da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 27911da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 27921da177e4SLinus Torvalds * writes to that region are not written out to the file." 27931da177e4SLinus Torvalds */ 279409cbfeafSKirill A. Shutemov zero_user_segment(page, offset, PAGE_SIZE); 27951da177e4SLinus Torvalds out: 27961da177e4SLinus Torvalds ret = mpage_writepage(page, get_block, wbc); 27971da177e4SLinus Torvalds if (ret == -EAGAIN) 279835c80d5fSChris Mason ret = __block_write_full_page(inode, page, get_block, wbc, 279935c80d5fSChris Mason end_buffer_async_write); 28001da177e4SLinus Torvalds return ret; 28011da177e4SLinus Torvalds } 28021da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_writepage); 28031da177e4SLinus Torvalds 280403158cd7SNick Piggin int nobh_truncate_page(struct address_space *mapping, 280503158cd7SNick Piggin loff_t from, get_block_t *get_block) 28061da177e4SLinus Torvalds { 280709cbfeafSKirill A. Shutemov pgoff_t index = from >> PAGE_SHIFT; 280809cbfeafSKirill A. Shutemov unsigned offset = from & (PAGE_SIZE-1); 280903158cd7SNick Piggin unsigned blocksize; 281003158cd7SNick Piggin sector_t iblock; 281103158cd7SNick Piggin unsigned length, pos; 281203158cd7SNick Piggin struct inode *inode = mapping->host; 28131da177e4SLinus Torvalds struct page *page; 281403158cd7SNick Piggin struct buffer_head map_bh; 281503158cd7SNick Piggin int err; 28161da177e4SLinus Torvalds 281793407472SFabian Frederick blocksize = i_blocksize(inode); 281803158cd7SNick Piggin length = offset & (blocksize - 1); 28191da177e4SLinus Torvalds 282003158cd7SNick Piggin /* Block boundary? Nothing to do */ 282103158cd7SNick Piggin if (!length) 282203158cd7SNick Piggin return 0; 282303158cd7SNick Piggin 282403158cd7SNick Piggin length = blocksize - length; 282509cbfeafSKirill A. Shutemov iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); 282603158cd7SNick Piggin 28271da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 282803158cd7SNick Piggin err = -ENOMEM; 28291da177e4SLinus Torvalds if (!page) 28301da177e4SLinus Torvalds goto out; 28311da177e4SLinus Torvalds 283203158cd7SNick Piggin if (page_has_buffers(page)) { 283303158cd7SNick Piggin has_buffers: 283403158cd7SNick Piggin unlock_page(page); 283509cbfeafSKirill A. Shutemov put_page(page); 283603158cd7SNick Piggin return block_truncate_page(mapping, from, get_block); 28371da177e4SLinus Torvalds } 283803158cd7SNick Piggin 283903158cd7SNick Piggin /* Find the buffer that contains "offset" */ 284003158cd7SNick Piggin pos = blocksize; 284103158cd7SNick Piggin while (offset >= pos) { 284203158cd7SNick Piggin iblock++; 284303158cd7SNick Piggin pos += blocksize; 284403158cd7SNick Piggin } 284503158cd7SNick Piggin 2846460bcf57STheodore Ts'o map_bh.b_size = blocksize; 2847460bcf57STheodore Ts'o map_bh.b_state = 0; 284803158cd7SNick Piggin err = get_block(inode, iblock, &map_bh, 0); 284903158cd7SNick Piggin if (err) 285003158cd7SNick Piggin goto unlock; 285103158cd7SNick Piggin /* unmapped? It's a hole - nothing to do */ 285203158cd7SNick Piggin if (!buffer_mapped(&map_bh)) 285303158cd7SNick Piggin goto unlock; 285403158cd7SNick Piggin 285503158cd7SNick Piggin /* Ok, it's mapped. Make sure it's up-to-date */ 285603158cd7SNick Piggin if (!PageUptodate(page)) { 285703158cd7SNick Piggin err = mapping->a_ops->readpage(NULL, page); 285803158cd7SNick Piggin if (err) { 285909cbfeafSKirill A. Shutemov put_page(page); 286003158cd7SNick Piggin goto out; 286103158cd7SNick Piggin } 286203158cd7SNick Piggin lock_page(page); 286303158cd7SNick Piggin if (!PageUptodate(page)) { 286403158cd7SNick Piggin err = -EIO; 286503158cd7SNick Piggin goto unlock; 286603158cd7SNick Piggin } 286703158cd7SNick Piggin if (page_has_buffers(page)) 286803158cd7SNick Piggin goto has_buffers; 286903158cd7SNick Piggin } 2870eebd2aa3SChristoph Lameter zero_user(page, offset, length); 287103158cd7SNick Piggin set_page_dirty(page); 287203158cd7SNick Piggin err = 0; 287303158cd7SNick Piggin 287403158cd7SNick Piggin unlock: 28751da177e4SLinus Torvalds unlock_page(page); 287609cbfeafSKirill A. Shutemov put_page(page); 28771da177e4SLinus Torvalds out: 287803158cd7SNick Piggin return err; 28791da177e4SLinus Torvalds } 28801da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_truncate_page); 28811da177e4SLinus Torvalds 28821da177e4SLinus Torvalds int block_truncate_page(struct address_space *mapping, 28831da177e4SLinus Torvalds loff_t from, get_block_t *get_block) 28841da177e4SLinus Torvalds { 288509cbfeafSKirill A. Shutemov pgoff_t index = from >> PAGE_SHIFT; 288609cbfeafSKirill A. Shutemov unsigned offset = from & (PAGE_SIZE-1); 28871da177e4SLinus Torvalds unsigned blocksize; 288854b21a79SAndrew Morton sector_t iblock; 28891da177e4SLinus Torvalds unsigned length, pos; 28901da177e4SLinus Torvalds struct inode *inode = mapping->host; 28911da177e4SLinus Torvalds struct page *page; 28921da177e4SLinus Torvalds struct buffer_head *bh; 28931da177e4SLinus Torvalds int err; 28941da177e4SLinus Torvalds 289593407472SFabian Frederick blocksize = i_blocksize(inode); 28961da177e4SLinus Torvalds length = offset & (blocksize - 1); 28971da177e4SLinus Torvalds 28981da177e4SLinus Torvalds /* Block boundary? Nothing to do */ 28991da177e4SLinus Torvalds if (!length) 29001da177e4SLinus Torvalds return 0; 29011da177e4SLinus Torvalds 29021da177e4SLinus Torvalds length = blocksize - length; 290309cbfeafSKirill A. Shutemov iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); 29041da177e4SLinus Torvalds 29051da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 29061da177e4SLinus Torvalds err = -ENOMEM; 29071da177e4SLinus Torvalds if (!page) 29081da177e4SLinus Torvalds goto out; 29091da177e4SLinus Torvalds 29101da177e4SLinus Torvalds if (!page_has_buffers(page)) 29111da177e4SLinus Torvalds create_empty_buffers(page, blocksize, 0); 29121da177e4SLinus Torvalds 29131da177e4SLinus Torvalds /* Find the buffer that contains "offset" */ 29141da177e4SLinus Torvalds bh = page_buffers(page); 29151da177e4SLinus Torvalds pos = blocksize; 29161da177e4SLinus Torvalds while (offset >= pos) { 29171da177e4SLinus Torvalds bh = bh->b_this_page; 29181da177e4SLinus Torvalds iblock++; 29191da177e4SLinus Torvalds pos += blocksize; 29201da177e4SLinus Torvalds } 29211da177e4SLinus Torvalds 29221da177e4SLinus Torvalds err = 0; 29231da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2924b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 29251da177e4SLinus Torvalds err = get_block(inode, iblock, bh, 0); 29261da177e4SLinus Torvalds if (err) 29271da177e4SLinus Torvalds goto unlock; 29281da177e4SLinus Torvalds /* unmapped? It's a hole - nothing to do */ 29291da177e4SLinus Torvalds if (!buffer_mapped(bh)) 29301da177e4SLinus Torvalds goto unlock; 29311da177e4SLinus Torvalds } 29321da177e4SLinus Torvalds 29331da177e4SLinus Torvalds /* Ok, it's mapped. Make sure it's up-to-date */ 29341da177e4SLinus Torvalds if (PageUptodate(page)) 29351da177e4SLinus Torvalds set_buffer_uptodate(bh); 29361da177e4SLinus Torvalds 293733a266ddSDavid Chinner if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { 29381da177e4SLinus Torvalds err = -EIO; 2939dfec8a14SMike Christie ll_rw_block(REQ_OP_READ, 0, 1, &bh); 29401da177e4SLinus Torvalds wait_on_buffer(bh); 29411da177e4SLinus Torvalds /* Uhhuh. Read error. Complain and punt. */ 29421da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 29431da177e4SLinus Torvalds goto unlock; 29441da177e4SLinus Torvalds } 29451da177e4SLinus Torvalds 2946eebd2aa3SChristoph Lameter zero_user(page, offset, length); 29471da177e4SLinus Torvalds mark_buffer_dirty(bh); 29481da177e4SLinus Torvalds err = 0; 29491da177e4SLinus Torvalds 29501da177e4SLinus Torvalds unlock: 29511da177e4SLinus Torvalds unlock_page(page); 295209cbfeafSKirill A. Shutemov put_page(page); 29531da177e4SLinus Torvalds out: 29541da177e4SLinus Torvalds return err; 29551da177e4SLinus Torvalds } 29561fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_truncate_page); 29571da177e4SLinus Torvalds 29581da177e4SLinus Torvalds /* 29591da177e4SLinus Torvalds * The generic ->writepage function for buffer-backed address_spaces 29601da177e4SLinus Torvalds */ 29611b938c08SMatthew Wilcox int block_write_full_page(struct page *page, get_block_t *get_block, 29621b938c08SMatthew Wilcox struct writeback_control *wbc) 29631da177e4SLinus Torvalds { 29641da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 29651da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 296609cbfeafSKirill A. Shutemov const pgoff_t end_index = i_size >> PAGE_SHIFT; 29671da177e4SLinus Torvalds unsigned offset; 29681da177e4SLinus Torvalds 29691da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 29701da177e4SLinus Torvalds if (page->index < end_index) 297135c80d5fSChris Mason return __block_write_full_page(inode, page, get_block, wbc, 29721b938c08SMatthew Wilcox end_buffer_async_write); 29731da177e4SLinus Torvalds 29741da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 297509cbfeafSKirill A. Shutemov offset = i_size & (PAGE_SIZE-1); 29761da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 29771da177e4SLinus Torvalds /* 29781da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 29791da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 29801da177e4SLinus Torvalds * freeable here, so the page does not leak. 29811da177e4SLinus Torvalds */ 298209cbfeafSKirill A. Shutemov do_invalidatepage(page, 0, PAGE_SIZE); 29831da177e4SLinus Torvalds unlock_page(page); 29841da177e4SLinus Torvalds return 0; /* don't care */ 29851da177e4SLinus Torvalds } 29861da177e4SLinus Torvalds 29871da177e4SLinus Torvalds /* 29881da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 29892a61aa40SAdam Buchbinder * writepage invocation because it may be mmapped. "A file is mapped 29901da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 29911da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 29921da177e4SLinus Torvalds * writes to that region are not written out to the file." 29931da177e4SLinus Torvalds */ 299409cbfeafSKirill A. Shutemov zero_user_segment(page, offset, PAGE_SIZE); 29951b938c08SMatthew Wilcox return __block_write_full_page(inode, page, get_block, wbc, 299635c80d5fSChris Mason end_buffer_async_write); 299735c80d5fSChris Mason } 29981fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_write_full_page); 299935c80d5fSChris Mason 30001da177e4SLinus Torvalds sector_t generic_block_bmap(struct address_space *mapping, sector_t block, 30011da177e4SLinus Torvalds get_block_t *get_block) 30021da177e4SLinus Torvalds { 30031da177e4SLinus Torvalds struct inode *inode = mapping->host; 30042a527d68SAlexander Potapenko struct buffer_head tmp = { 30052a527d68SAlexander Potapenko .b_size = i_blocksize(inode), 30062a527d68SAlexander Potapenko }; 30072a527d68SAlexander Potapenko 30081da177e4SLinus Torvalds get_block(inode, block, &tmp, 0); 30091da177e4SLinus Torvalds return tmp.b_blocknr; 30101da177e4SLinus Torvalds } 30111fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_block_bmap); 30121da177e4SLinus Torvalds 30134246a0b6SChristoph Hellwig static void end_bio_bh_io_sync(struct bio *bio) 30141da177e4SLinus Torvalds { 30151da177e4SLinus Torvalds struct buffer_head *bh = bio->bi_private; 30161da177e4SLinus Torvalds 3017b7c44ed9SJens Axboe if (unlikely(bio_flagged(bio, BIO_QUIET))) 301808bafc03SKeith Mannthey set_bit(BH_Quiet, &bh->b_state); 301908bafc03SKeith Mannthey 30204e4cbee9SChristoph Hellwig bh->b_end_io(bh, !bio->bi_status); 30211da177e4SLinus Torvalds bio_put(bio); 30221da177e4SLinus Torvalds } 30231da177e4SLinus Torvalds 302457302e0dSLinus Torvalds /* 302557302e0dSLinus Torvalds * This allows us to do IO even on the odd last sectors 302659d43914SAkinobu Mita * of a device, even if the block size is some multiple 302757302e0dSLinus Torvalds * of the physical sector size. 302857302e0dSLinus Torvalds * 302957302e0dSLinus Torvalds * We'll just truncate the bio to the size of the device, 303057302e0dSLinus Torvalds * and clear the end of the buffer head manually. 303157302e0dSLinus Torvalds * 303257302e0dSLinus Torvalds * Truly out-of-range accesses will turn into actual IO 303357302e0dSLinus Torvalds * errors, this only handles the "we need to be able to 303457302e0dSLinus Torvalds * do IO at the final sector" case. 303557302e0dSLinus Torvalds */ 30362a222ca9SMike Christie void guard_bio_eod(int op, struct bio *bio) 303757302e0dSLinus Torvalds { 303857302e0dSLinus Torvalds sector_t maxsector; 303959d43914SAkinobu Mita struct bio_vec *bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 304059d43914SAkinobu Mita unsigned truncated_bytes; 304157302e0dSLinus Torvalds 304274d46992SChristoph Hellwig maxsector = get_capacity(bio->bi_disk); 304357302e0dSLinus Torvalds if (!maxsector) 304457302e0dSLinus Torvalds return; 304557302e0dSLinus Torvalds 304657302e0dSLinus Torvalds /* 304757302e0dSLinus Torvalds * If the *whole* IO is past the end of the device, 304857302e0dSLinus Torvalds * let it through, and the IO layer will turn it into 304957302e0dSLinus Torvalds * an EIO. 305057302e0dSLinus Torvalds */ 30514f024f37SKent Overstreet if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 305257302e0dSLinus Torvalds return; 305357302e0dSLinus Torvalds 30544f024f37SKent Overstreet maxsector -= bio->bi_iter.bi_sector; 305559d43914SAkinobu Mita if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 305657302e0dSLinus Torvalds return; 305757302e0dSLinus Torvalds 305859d43914SAkinobu Mita /* Uhhuh. We've got a bio that straddles the device size! */ 305959d43914SAkinobu Mita truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9); 306057302e0dSLinus Torvalds 306157302e0dSLinus Torvalds /* Truncate the bio.. */ 306259d43914SAkinobu Mita bio->bi_iter.bi_size -= truncated_bytes; 306359d43914SAkinobu Mita bvec->bv_len -= truncated_bytes; 306457302e0dSLinus Torvalds 306557302e0dSLinus Torvalds /* ..and clear the end of the buffer for reads */ 30662a222ca9SMike Christie if (op == REQ_OP_READ) { 306759d43914SAkinobu Mita zero_user(bvec->bv_page, bvec->bv_offset + bvec->bv_len, 306859d43914SAkinobu Mita truncated_bytes); 306957302e0dSLinus Torvalds } 307057302e0dSLinus Torvalds } 307157302e0dSLinus Torvalds 30722a222ca9SMike Christie static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh, 30738e8f9298SJens Axboe enum rw_hint write_hint, struct writeback_control *wbc) 30741da177e4SLinus Torvalds { 30751da177e4SLinus Torvalds struct bio *bio; 30761da177e4SLinus Torvalds 30771da177e4SLinus Torvalds BUG_ON(!buffer_locked(bh)); 30781da177e4SLinus Torvalds BUG_ON(!buffer_mapped(bh)); 30791da177e4SLinus Torvalds BUG_ON(!bh->b_end_io); 30808fb0e342SAneesh Kumar K.V BUG_ON(buffer_delay(bh)); 30818fb0e342SAneesh Kumar K.V BUG_ON(buffer_unwritten(bh)); 30821da177e4SLinus Torvalds 308348fd4f93SJens Axboe /* 308448fd4f93SJens Axboe * Only clear out a write error when rewriting 30851da177e4SLinus Torvalds */ 30862a222ca9SMike Christie if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) 30871da177e4SLinus Torvalds clear_buffer_write_io_error(bh); 30881da177e4SLinus Torvalds 30891da177e4SLinus Torvalds /* 30901da177e4SLinus Torvalds * from here on down, it's all bio -- do the initial mapping, 30911da177e4SLinus Torvalds * submit_bio -> generic_make_request may further map this bio around 30921da177e4SLinus Torvalds */ 30931da177e4SLinus Torvalds bio = bio_alloc(GFP_NOIO, 1); 30941da177e4SLinus Torvalds 30952a814908STejun Heo if (wbc) { 3096b16b1debSTejun Heo wbc_init_bio(wbc, bio); 30972a814908STejun Heo wbc_account_io(wbc, bh->b_page, bh->b_size); 30982a814908STejun Heo } 3099bafc0dbaSTejun Heo 31004f024f37SKent Overstreet bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); 310174d46992SChristoph Hellwig bio_set_dev(bio, bh->b_bdev); 31028e8f9298SJens Axboe bio->bi_write_hint = write_hint; 31031da177e4SLinus Torvalds 31046cf66b4cSKent Overstreet bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); 31056cf66b4cSKent Overstreet BUG_ON(bio->bi_iter.bi_size != bh->b_size); 31061da177e4SLinus Torvalds 31071da177e4SLinus Torvalds bio->bi_end_io = end_bio_bh_io_sync; 31081da177e4SLinus Torvalds bio->bi_private = bh; 31091da177e4SLinus Torvalds 311057302e0dSLinus Torvalds /* Take care of bh's that straddle the end of the device */ 31112a222ca9SMike Christie guard_bio_eod(op, bio); 311257302e0dSLinus Torvalds 3113877f962cSTheodore Ts'o if (buffer_meta(bh)) 31142a222ca9SMike Christie op_flags |= REQ_META; 3115877f962cSTheodore Ts'o if (buffer_prio(bh)) 31162a222ca9SMike Christie op_flags |= REQ_PRIO; 31172a222ca9SMike Christie bio_set_op_attrs(bio, op, op_flags); 3118877f962cSTheodore Ts'o 31194e49ea4aSMike Christie submit_bio(bio); 3120f6454b04SJulia Lawall return 0; 31211da177e4SLinus Torvalds } 3122bafc0dbaSTejun Heo 31232a222ca9SMike Christie int submit_bh(int op, int op_flags, struct buffer_head *bh) 312471368511SDarrick J. Wong { 31258e8f9298SJens Axboe return submit_bh_wbc(op, op_flags, bh, 0, NULL); 312671368511SDarrick J. Wong } 31271fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(submit_bh); 31281da177e4SLinus Torvalds 31291da177e4SLinus Torvalds /** 31301da177e4SLinus Torvalds * ll_rw_block: low-level access to block devices (DEPRECATED) 3131dfec8a14SMike Christie * @op: whether to %READ or %WRITE 3132ef295ecfSChristoph Hellwig * @op_flags: req_flag_bits 31331da177e4SLinus Torvalds * @nr: number of &struct buffer_heads in the array 31341da177e4SLinus Torvalds * @bhs: array of pointers to &struct buffer_head 31351da177e4SLinus Torvalds * 3136a7662236SJan Kara * ll_rw_block() takes an array of pointers to &struct buffer_heads, and 313770246286SChristoph Hellwig * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE. 313870246286SChristoph Hellwig * @op_flags contains flags modifying the detailed I/O behavior, most notably 313970246286SChristoph Hellwig * %REQ_RAHEAD. 31401da177e4SLinus Torvalds * 31411da177e4SLinus Torvalds * This function drops any buffer that it cannot get a lock on (with the 31429cb569d6SChristoph Hellwig * BH_Lock state bit), any buffer that appears to be clean when doing a write 31439cb569d6SChristoph Hellwig * request, and any buffer that appears to be up-to-date when doing read 31449cb569d6SChristoph Hellwig * request. Further it marks as clean buffers that are processed for 31459cb569d6SChristoph Hellwig * writing (the buffer cache won't assume that they are actually clean 31469cb569d6SChristoph Hellwig * until the buffer gets unlocked). 31471da177e4SLinus Torvalds * 31481da177e4SLinus Torvalds * ll_rw_block sets b_end_io to simple completion handler that marks 3149e227867fSMasanari Iida * the buffer up-to-date (if appropriate), unlocks the buffer and wakes 31501da177e4SLinus Torvalds * any waiters. 31511da177e4SLinus Torvalds * 31521da177e4SLinus Torvalds * All of the buffers must be for the same device, and must also be a 31531da177e4SLinus Torvalds * multiple of the current approved size for the device. 31541da177e4SLinus Torvalds */ 3155dfec8a14SMike Christie void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[]) 31561da177e4SLinus Torvalds { 31571da177e4SLinus Torvalds int i; 31581da177e4SLinus Torvalds 31591da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 31601da177e4SLinus Torvalds struct buffer_head *bh = bhs[i]; 31611da177e4SLinus Torvalds 31629cb569d6SChristoph Hellwig if (!trylock_buffer(bh)) 31631da177e4SLinus Torvalds continue; 3164dfec8a14SMike Christie if (op == WRITE) { 31651da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 316676c3073aSakpm@osdl.org bh->b_end_io = end_buffer_write_sync; 3167e60e5c50SOGAWA Hirofumi get_bh(bh); 3168dfec8a14SMike Christie submit_bh(op, op_flags, bh); 31691da177e4SLinus Torvalds continue; 31701da177e4SLinus Torvalds } 31711da177e4SLinus Torvalds } else { 31721da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 317376c3073aSakpm@osdl.org bh->b_end_io = end_buffer_read_sync; 3174e60e5c50SOGAWA Hirofumi get_bh(bh); 3175dfec8a14SMike Christie submit_bh(op, op_flags, bh); 31761da177e4SLinus Torvalds continue; 31771da177e4SLinus Torvalds } 31781da177e4SLinus Torvalds } 31791da177e4SLinus Torvalds unlock_buffer(bh); 31801da177e4SLinus Torvalds } 31811da177e4SLinus Torvalds } 31821fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(ll_rw_block); 31831da177e4SLinus Torvalds 31842a222ca9SMike Christie void write_dirty_buffer(struct buffer_head *bh, int op_flags) 31859cb569d6SChristoph Hellwig { 31869cb569d6SChristoph Hellwig lock_buffer(bh); 31879cb569d6SChristoph Hellwig if (!test_clear_buffer_dirty(bh)) { 31889cb569d6SChristoph Hellwig unlock_buffer(bh); 31899cb569d6SChristoph Hellwig return; 31909cb569d6SChristoph Hellwig } 31919cb569d6SChristoph Hellwig bh->b_end_io = end_buffer_write_sync; 31929cb569d6SChristoph Hellwig get_bh(bh); 31932a222ca9SMike Christie submit_bh(REQ_OP_WRITE, op_flags, bh); 31949cb569d6SChristoph Hellwig } 31959cb569d6SChristoph Hellwig EXPORT_SYMBOL(write_dirty_buffer); 31969cb569d6SChristoph Hellwig 31971da177e4SLinus Torvalds /* 31981da177e4SLinus Torvalds * For a data-integrity writeout, we need to wait upon any in-progress I/O 31991da177e4SLinus Torvalds * and then start new I/O and then wait upon it. The caller must have a ref on 32001da177e4SLinus Torvalds * the buffer_head. 32011da177e4SLinus Torvalds */ 32022a222ca9SMike Christie int __sync_dirty_buffer(struct buffer_head *bh, int op_flags) 32031da177e4SLinus Torvalds { 32041da177e4SLinus Torvalds int ret = 0; 32051da177e4SLinus Torvalds 32061da177e4SLinus Torvalds WARN_ON(atomic_read(&bh->b_count) < 1); 32071da177e4SLinus Torvalds lock_buffer(bh); 32081da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 32091da177e4SLinus Torvalds get_bh(bh); 32101da177e4SLinus Torvalds bh->b_end_io = end_buffer_write_sync; 32112a222ca9SMike Christie ret = submit_bh(REQ_OP_WRITE, op_flags, bh); 32121da177e4SLinus Torvalds wait_on_buffer(bh); 32131da177e4SLinus Torvalds if (!ret && !buffer_uptodate(bh)) 32141da177e4SLinus Torvalds ret = -EIO; 32151da177e4SLinus Torvalds } else { 32161da177e4SLinus Torvalds unlock_buffer(bh); 32171da177e4SLinus Torvalds } 32181da177e4SLinus Torvalds return ret; 32191da177e4SLinus Torvalds } 322087e99511SChristoph Hellwig EXPORT_SYMBOL(__sync_dirty_buffer); 322187e99511SChristoph Hellwig 322287e99511SChristoph Hellwig int sync_dirty_buffer(struct buffer_head *bh) 322387e99511SChristoph Hellwig { 322470fd7614SChristoph Hellwig return __sync_dirty_buffer(bh, REQ_SYNC); 322587e99511SChristoph Hellwig } 32261fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(sync_dirty_buffer); 32271da177e4SLinus Torvalds 32281da177e4SLinus Torvalds /* 32291da177e4SLinus Torvalds * try_to_free_buffers() checks if all the buffers on this particular page 32301da177e4SLinus Torvalds * are unused, and releases them if so. 32311da177e4SLinus Torvalds * 32321da177e4SLinus Torvalds * Exclusion against try_to_free_buffers may be obtained by either 32331da177e4SLinus Torvalds * locking the page or by holding its mapping's private_lock. 32341da177e4SLinus Torvalds * 32351da177e4SLinus Torvalds * If the page is dirty but all the buffers are clean then we need to 32361da177e4SLinus Torvalds * be sure to mark the page clean as well. This is because the page 32371da177e4SLinus Torvalds * may be against a block device, and a later reattachment of buffers 32381da177e4SLinus Torvalds * to a dirty page will set *all* buffers dirty. Which would corrupt 32391da177e4SLinus Torvalds * filesystem data on the same device. 32401da177e4SLinus Torvalds * 32411da177e4SLinus Torvalds * The same applies to regular filesystem pages: if all the buffers are 32421da177e4SLinus Torvalds * clean then we set the page clean and proceed. To do that, we require 32431da177e4SLinus Torvalds * total exclusion from __set_page_dirty_buffers(). That is obtained with 32441da177e4SLinus Torvalds * private_lock. 32451da177e4SLinus Torvalds * 32461da177e4SLinus Torvalds * try_to_free_buffers() is non-blocking. 32471da177e4SLinus Torvalds */ 32481da177e4SLinus Torvalds static inline int buffer_busy(struct buffer_head *bh) 32491da177e4SLinus Torvalds { 32501da177e4SLinus Torvalds return atomic_read(&bh->b_count) | 32511da177e4SLinus Torvalds (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); 32521da177e4SLinus Torvalds } 32531da177e4SLinus Torvalds 32541da177e4SLinus Torvalds static int 32551da177e4SLinus Torvalds drop_buffers(struct page *page, struct buffer_head **buffers_to_free) 32561da177e4SLinus Torvalds { 32571da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 32581da177e4SLinus Torvalds struct buffer_head *bh; 32591da177e4SLinus Torvalds 32601da177e4SLinus Torvalds bh = head; 32611da177e4SLinus Torvalds do { 32621da177e4SLinus Torvalds if (buffer_busy(bh)) 32631da177e4SLinus Torvalds goto failed; 32641da177e4SLinus Torvalds bh = bh->b_this_page; 32651da177e4SLinus Torvalds } while (bh != head); 32661da177e4SLinus Torvalds 32671da177e4SLinus Torvalds do { 32681da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 32691da177e4SLinus Torvalds 3270535ee2fbSJan Kara if (bh->b_assoc_map) 32711da177e4SLinus Torvalds __remove_assoc_queue(bh); 32721da177e4SLinus Torvalds bh = next; 32731da177e4SLinus Torvalds } while (bh != head); 32741da177e4SLinus Torvalds *buffers_to_free = head; 32751da177e4SLinus Torvalds __clear_page_buffers(page); 32761da177e4SLinus Torvalds return 1; 32771da177e4SLinus Torvalds failed: 32781da177e4SLinus Torvalds return 0; 32791da177e4SLinus Torvalds } 32801da177e4SLinus Torvalds 32811da177e4SLinus Torvalds int try_to_free_buffers(struct page *page) 32821da177e4SLinus Torvalds { 32831da177e4SLinus Torvalds struct address_space * const mapping = page->mapping; 32841da177e4SLinus Torvalds struct buffer_head *buffers_to_free = NULL; 32851da177e4SLinus Torvalds int ret = 0; 32861da177e4SLinus Torvalds 32871da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 3288ecdfc978SLinus Torvalds if (PageWriteback(page)) 32891da177e4SLinus Torvalds return 0; 32901da177e4SLinus Torvalds 32911da177e4SLinus Torvalds if (mapping == NULL) { /* can this still happen? */ 32921da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 32931da177e4SLinus Torvalds goto out; 32941da177e4SLinus Torvalds } 32951da177e4SLinus Torvalds 32961da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 32971da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 3298ecdfc978SLinus Torvalds 3299ecdfc978SLinus Torvalds /* 3300ecdfc978SLinus Torvalds * If the filesystem writes its buffers by hand (eg ext3) 3301ecdfc978SLinus Torvalds * then we can have clean buffers against a dirty page. We 3302ecdfc978SLinus Torvalds * clean the page here; otherwise the VM will never notice 3303ecdfc978SLinus Torvalds * that the filesystem did any IO at all. 3304ecdfc978SLinus Torvalds * 3305ecdfc978SLinus Torvalds * Also, during truncate, discard_buffer will have marked all 3306ecdfc978SLinus Torvalds * the page's buffers clean. We discover that here and clean 3307ecdfc978SLinus Torvalds * the page also. 330887df7241SNick Piggin * 330987df7241SNick Piggin * private_lock must be held over this entire operation in order 331087df7241SNick Piggin * to synchronise against __set_page_dirty_buffers and prevent the 331187df7241SNick Piggin * dirty bit from being lost. 3312ecdfc978SLinus Torvalds */ 331311f81becSTejun Heo if (ret) 331411f81becSTejun Heo cancel_dirty_page(page); 331587df7241SNick Piggin spin_unlock(&mapping->private_lock); 33161da177e4SLinus Torvalds out: 33171da177e4SLinus Torvalds if (buffers_to_free) { 33181da177e4SLinus Torvalds struct buffer_head *bh = buffers_to_free; 33191da177e4SLinus Torvalds 33201da177e4SLinus Torvalds do { 33211da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 33221da177e4SLinus Torvalds free_buffer_head(bh); 33231da177e4SLinus Torvalds bh = next; 33241da177e4SLinus Torvalds } while (bh != buffers_to_free); 33251da177e4SLinus Torvalds } 33261da177e4SLinus Torvalds return ret; 33271da177e4SLinus Torvalds } 33281da177e4SLinus Torvalds EXPORT_SYMBOL(try_to_free_buffers); 33291da177e4SLinus Torvalds 33301da177e4SLinus Torvalds /* 33311da177e4SLinus Torvalds * There are no bdflush tunables left. But distributions are 33321da177e4SLinus Torvalds * still running obsolete flush daemons, so we terminate them here. 33331da177e4SLinus Torvalds * 33341da177e4SLinus Torvalds * Use of bdflush() is deprecated and will be removed in a future kernel. 33355b0830cbSJens Axboe * The `flush-X' kernel threads fully replace bdflush daemons and this call. 33361da177e4SLinus Torvalds */ 3337bdc480e3SHeiko Carstens SYSCALL_DEFINE2(bdflush, int, func, long, data) 33381da177e4SLinus Torvalds { 33391da177e4SLinus Torvalds static int msg_count; 33401da177e4SLinus Torvalds 33411da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 33421da177e4SLinus Torvalds return -EPERM; 33431da177e4SLinus Torvalds 33441da177e4SLinus Torvalds if (msg_count < 5) { 33451da177e4SLinus Torvalds msg_count++; 33461da177e4SLinus Torvalds printk(KERN_INFO 33471da177e4SLinus Torvalds "warning: process `%s' used the obsolete bdflush" 33481da177e4SLinus Torvalds " system call\n", current->comm); 33491da177e4SLinus Torvalds printk(KERN_INFO "Fix your initscripts?\n"); 33501da177e4SLinus Torvalds } 33511da177e4SLinus Torvalds 33521da177e4SLinus Torvalds if (func == 1) 33531da177e4SLinus Torvalds do_exit(0); 33541da177e4SLinus Torvalds return 0; 33551da177e4SLinus Torvalds } 33561da177e4SLinus Torvalds 33571da177e4SLinus Torvalds /* 33581da177e4SLinus Torvalds * Buffer-head allocation 33591da177e4SLinus Torvalds */ 3360a0a9b043SShai Fultheim static struct kmem_cache *bh_cachep __read_mostly; 33611da177e4SLinus Torvalds 33621da177e4SLinus Torvalds /* 33631da177e4SLinus Torvalds * Once the number of bh's in the machine exceeds this level, we start 33641da177e4SLinus Torvalds * stripping them in writeback. 33651da177e4SLinus Torvalds */ 336643be594aSZhang Yanfei static unsigned long max_buffer_heads; 33671da177e4SLinus Torvalds 33681da177e4SLinus Torvalds int buffer_heads_over_limit; 33691da177e4SLinus Torvalds 33701da177e4SLinus Torvalds struct bh_accounting { 33711da177e4SLinus Torvalds int nr; /* Number of live bh's */ 33721da177e4SLinus Torvalds int ratelimit; /* Limit cacheline bouncing */ 33731da177e4SLinus Torvalds }; 33741da177e4SLinus Torvalds 33751da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; 33761da177e4SLinus Torvalds 33771da177e4SLinus Torvalds static void recalc_bh_state(void) 33781da177e4SLinus Torvalds { 33791da177e4SLinus Torvalds int i; 33801da177e4SLinus Torvalds int tot = 0; 33811da177e4SLinus Torvalds 3382ee1be862SChristoph Lameter if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) 33831da177e4SLinus Torvalds return; 3384c7b92516SChristoph Lameter __this_cpu_write(bh_accounting.ratelimit, 0); 33858a143426SEric Dumazet for_each_online_cpu(i) 33861da177e4SLinus Torvalds tot += per_cpu(bh_accounting, i).nr; 33871da177e4SLinus Torvalds buffer_heads_over_limit = (tot > max_buffer_heads); 33881da177e4SLinus Torvalds } 33891da177e4SLinus Torvalds 3390dd0fc66fSAl Viro struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) 33911da177e4SLinus Torvalds { 3392019b4d12SRichard Kennedy struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); 33931da177e4SLinus Torvalds if (ret) { 3394a35afb83SChristoph Lameter INIT_LIST_HEAD(&ret->b_assoc_buffers); 3395c7b92516SChristoph Lameter preempt_disable(); 3396c7b92516SChristoph Lameter __this_cpu_inc(bh_accounting.nr); 33971da177e4SLinus Torvalds recalc_bh_state(); 3398c7b92516SChristoph Lameter preempt_enable(); 33991da177e4SLinus Torvalds } 34001da177e4SLinus Torvalds return ret; 34011da177e4SLinus Torvalds } 34021da177e4SLinus Torvalds EXPORT_SYMBOL(alloc_buffer_head); 34031da177e4SLinus Torvalds 34041da177e4SLinus Torvalds void free_buffer_head(struct buffer_head *bh) 34051da177e4SLinus Torvalds { 34061da177e4SLinus Torvalds BUG_ON(!list_empty(&bh->b_assoc_buffers)); 34071da177e4SLinus Torvalds kmem_cache_free(bh_cachep, bh); 3408c7b92516SChristoph Lameter preempt_disable(); 3409c7b92516SChristoph Lameter __this_cpu_dec(bh_accounting.nr); 34101da177e4SLinus Torvalds recalc_bh_state(); 3411c7b92516SChristoph Lameter preempt_enable(); 34121da177e4SLinus Torvalds } 34131da177e4SLinus Torvalds EXPORT_SYMBOL(free_buffer_head); 34141da177e4SLinus Torvalds 3415fc4d24c9SSebastian Andrzej Siewior static int buffer_exit_cpu_dead(unsigned int cpu) 34161da177e4SLinus Torvalds { 34171da177e4SLinus Torvalds int i; 34181da177e4SLinus Torvalds struct bh_lru *b = &per_cpu(bh_lrus, cpu); 34191da177e4SLinus Torvalds 34201da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 34211da177e4SLinus Torvalds brelse(b->bhs[i]); 34221da177e4SLinus Torvalds b->bhs[i] = NULL; 34231da177e4SLinus Torvalds } 3424c7b92516SChristoph Lameter this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); 34258a143426SEric Dumazet per_cpu(bh_accounting, cpu).nr = 0; 3426fc4d24c9SSebastian Andrzej Siewior return 0; 34271da177e4SLinus Torvalds } 34281da177e4SLinus Torvalds 3429389d1b08SAneesh Kumar K.V /** 3430a6b91919SRandy Dunlap * bh_uptodate_or_lock - Test whether the buffer is uptodate 3431389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3432389d1b08SAneesh Kumar K.V * 3433389d1b08SAneesh Kumar K.V * Return true if the buffer is up-to-date and false, 3434389d1b08SAneesh Kumar K.V * with the buffer locked, if not. 3435389d1b08SAneesh Kumar K.V */ 3436389d1b08SAneesh Kumar K.V int bh_uptodate_or_lock(struct buffer_head *bh) 3437389d1b08SAneesh Kumar K.V { 3438389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) { 3439389d1b08SAneesh Kumar K.V lock_buffer(bh); 3440389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) 3441389d1b08SAneesh Kumar K.V return 0; 3442389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3443389d1b08SAneesh Kumar K.V } 3444389d1b08SAneesh Kumar K.V return 1; 3445389d1b08SAneesh Kumar K.V } 3446389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_uptodate_or_lock); 3447389d1b08SAneesh Kumar K.V 3448389d1b08SAneesh Kumar K.V /** 3449a6b91919SRandy Dunlap * bh_submit_read - Submit a locked buffer for reading 3450389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3451389d1b08SAneesh Kumar K.V * 3452389d1b08SAneesh Kumar K.V * Returns zero on success and -EIO on error. 3453389d1b08SAneesh Kumar K.V */ 3454389d1b08SAneesh Kumar K.V int bh_submit_read(struct buffer_head *bh) 3455389d1b08SAneesh Kumar K.V { 3456389d1b08SAneesh Kumar K.V BUG_ON(!buffer_locked(bh)); 3457389d1b08SAneesh Kumar K.V 3458389d1b08SAneesh Kumar K.V if (buffer_uptodate(bh)) { 3459389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3460389d1b08SAneesh Kumar K.V return 0; 3461389d1b08SAneesh Kumar K.V } 3462389d1b08SAneesh Kumar K.V 3463389d1b08SAneesh Kumar K.V get_bh(bh); 3464389d1b08SAneesh Kumar K.V bh->b_end_io = end_buffer_read_sync; 34652a222ca9SMike Christie submit_bh(REQ_OP_READ, 0, bh); 3466389d1b08SAneesh Kumar K.V wait_on_buffer(bh); 3467389d1b08SAneesh Kumar K.V if (buffer_uptodate(bh)) 3468389d1b08SAneesh Kumar K.V return 0; 3469389d1b08SAneesh Kumar K.V return -EIO; 3470389d1b08SAneesh Kumar K.V } 3471389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_submit_read); 3472389d1b08SAneesh Kumar K.V 3473334fd34dSAndreas Gruenbacher /* 3474334fd34dSAndreas Gruenbacher * Seek for SEEK_DATA / SEEK_HOLE within @page, starting at @lastoff. 3475334fd34dSAndreas Gruenbacher * 3476334fd34dSAndreas Gruenbacher * Returns the offset within the file on success, and -ENOENT otherwise. 3477334fd34dSAndreas Gruenbacher */ 3478334fd34dSAndreas Gruenbacher static loff_t 3479334fd34dSAndreas Gruenbacher page_seek_hole_data(struct page *page, loff_t lastoff, int whence) 3480334fd34dSAndreas Gruenbacher { 3481334fd34dSAndreas Gruenbacher loff_t offset = page_offset(page); 3482334fd34dSAndreas Gruenbacher struct buffer_head *bh, *head; 3483334fd34dSAndreas Gruenbacher bool seek_data = whence == SEEK_DATA; 3484334fd34dSAndreas Gruenbacher 3485334fd34dSAndreas Gruenbacher if (lastoff < offset) 3486334fd34dSAndreas Gruenbacher lastoff = offset; 3487334fd34dSAndreas Gruenbacher 3488334fd34dSAndreas Gruenbacher bh = head = page_buffers(page); 3489334fd34dSAndreas Gruenbacher do { 3490334fd34dSAndreas Gruenbacher offset += bh->b_size; 3491334fd34dSAndreas Gruenbacher if (lastoff >= offset) 3492334fd34dSAndreas Gruenbacher continue; 3493334fd34dSAndreas Gruenbacher 3494334fd34dSAndreas Gruenbacher /* 3495334fd34dSAndreas Gruenbacher * Unwritten extents that have data in the page cache covering 3496334fd34dSAndreas Gruenbacher * them can be identified by the BH_Unwritten state flag. 3497334fd34dSAndreas Gruenbacher * Pages with multiple buffers might have a mix of holes, data 3498334fd34dSAndreas Gruenbacher * and unwritten extents - any buffer with valid data in it 3499334fd34dSAndreas Gruenbacher * should have BH_Uptodate flag set on it. 3500334fd34dSAndreas Gruenbacher */ 3501334fd34dSAndreas Gruenbacher 3502334fd34dSAndreas Gruenbacher if ((buffer_unwritten(bh) || buffer_uptodate(bh)) == seek_data) 3503334fd34dSAndreas Gruenbacher return lastoff; 3504334fd34dSAndreas Gruenbacher 3505334fd34dSAndreas Gruenbacher lastoff = offset; 3506334fd34dSAndreas Gruenbacher } while ((bh = bh->b_this_page) != head); 3507334fd34dSAndreas Gruenbacher return -ENOENT; 3508334fd34dSAndreas Gruenbacher } 3509334fd34dSAndreas Gruenbacher 3510334fd34dSAndreas Gruenbacher /* 3511334fd34dSAndreas Gruenbacher * Seek for SEEK_DATA / SEEK_HOLE in the page cache. 3512334fd34dSAndreas Gruenbacher * 3513334fd34dSAndreas Gruenbacher * Within unwritten extents, the page cache determines which parts are holes 3514334fd34dSAndreas Gruenbacher * and which are data: unwritten and uptodate buffer heads count as data; 3515334fd34dSAndreas Gruenbacher * everything else counts as a hole. 3516334fd34dSAndreas Gruenbacher * 3517334fd34dSAndreas Gruenbacher * Returns the resulting offset on successs, and -ENOENT otherwise. 3518334fd34dSAndreas Gruenbacher */ 3519334fd34dSAndreas Gruenbacher loff_t 3520334fd34dSAndreas Gruenbacher page_cache_seek_hole_data(struct inode *inode, loff_t offset, loff_t length, 3521334fd34dSAndreas Gruenbacher int whence) 3522334fd34dSAndreas Gruenbacher { 3523334fd34dSAndreas Gruenbacher pgoff_t index = offset >> PAGE_SHIFT; 3524334fd34dSAndreas Gruenbacher pgoff_t end = DIV_ROUND_UP(offset + length, PAGE_SIZE); 3525334fd34dSAndreas Gruenbacher loff_t lastoff = offset; 3526334fd34dSAndreas Gruenbacher struct pagevec pvec; 3527334fd34dSAndreas Gruenbacher 3528334fd34dSAndreas Gruenbacher if (length <= 0) 3529334fd34dSAndreas Gruenbacher return -ENOENT; 3530334fd34dSAndreas Gruenbacher 3531334fd34dSAndreas Gruenbacher pagevec_init(&pvec, 0); 3532334fd34dSAndreas Gruenbacher 3533334fd34dSAndreas Gruenbacher do { 35348338141fSJan Kara unsigned nr_pages, i; 3535334fd34dSAndreas Gruenbacher 35368338141fSJan Kara nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping, &index, 3537397162ffSJan Kara end - 1); 3538334fd34dSAndreas Gruenbacher if (nr_pages == 0) 3539334fd34dSAndreas Gruenbacher break; 3540334fd34dSAndreas Gruenbacher 3541334fd34dSAndreas Gruenbacher for (i = 0; i < nr_pages; i++) { 3542334fd34dSAndreas Gruenbacher struct page *page = pvec.pages[i]; 3543334fd34dSAndreas Gruenbacher 3544334fd34dSAndreas Gruenbacher /* 3545334fd34dSAndreas Gruenbacher * At this point, the page may be truncated or 3546334fd34dSAndreas Gruenbacher * invalidated (changing page->mapping to NULL), or 3547334fd34dSAndreas Gruenbacher * even swizzled back from swapper_space to tmpfs file 3548334fd34dSAndreas Gruenbacher * mapping. However, page->index will not change 3549334fd34dSAndreas Gruenbacher * because we have a reference on the page. 3550334fd34dSAndreas Gruenbacher * 3551334fd34dSAndreas Gruenbacher * If current page offset is beyond where we've ended, 3552334fd34dSAndreas Gruenbacher * we've found a hole. 3553334fd34dSAndreas Gruenbacher */ 3554334fd34dSAndreas Gruenbacher if (whence == SEEK_HOLE && 3555334fd34dSAndreas Gruenbacher lastoff < page_offset(page)) 3556334fd34dSAndreas Gruenbacher goto check_range; 3557334fd34dSAndreas Gruenbacher 3558334fd34dSAndreas Gruenbacher lock_page(page); 3559334fd34dSAndreas Gruenbacher if (likely(page->mapping == inode->i_mapping) && 3560334fd34dSAndreas Gruenbacher page_has_buffers(page)) { 3561334fd34dSAndreas Gruenbacher lastoff = page_seek_hole_data(page, lastoff, whence); 3562334fd34dSAndreas Gruenbacher if (lastoff >= 0) { 3563334fd34dSAndreas Gruenbacher unlock_page(page); 3564334fd34dSAndreas Gruenbacher goto check_range; 3565334fd34dSAndreas Gruenbacher } 3566334fd34dSAndreas Gruenbacher } 3567334fd34dSAndreas Gruenbacher unlock_page(page); 3568334fd34dSAndreas Gruenbacher lastoff = page_offset(page) + PAGE_SIZE; 3569334fd34dSAndreas Gruenbacher } 3570334fd34dSAndreas Gruenbacher pagevec_release(&pvec); 3571334fd34dSAndreas Gruenbacher } while (index < end); 3572334fd34dSAndreas Gruenbacher 3573334fd34dSAndreas Gruenbacher /* When no page at lastoff and we are not done, we found a hole. */ 3574334fd34dSAndreas Gruenbacher if (whence != SEEK_HOLE) 3575334fd34dSAndreas Gruenbacher goto not_found; 3576334fd34dSAndreas Gruenbacher 3577334fd34dSAndreas Gruenbacher check_range: 3578334fd34dSAndreas Gruenbacher if (lastoff < offset + length) 3579334fd34dSAndreas Gruenbacher goto out; 3580334fd34dSAndreas Gruenbacher not_found: 3581334fd34dSAndreas Gruenbacher lastoff = -ENOENT; 3582334fd34dSAndreas Gruenbacher out: 3583334fd34dSAndreas Gruenbacher pagevec_release(&pvec); 3584334fd34dSAndreas Gruenbacher return lastoff; 3585334fd34dSAndreas Gruenbacher } 3586334fd34dSAndreas Gruenbacher 35871da177e4SLinus Torvalds void __init buffer_init(void) 35881da177e4SLinus Torvalds { 358943be594aSZhang Yanfei unsigned long nrpages; 3590fc4d24c9SSebastian Andrzej Siewior int ret; 35911da177e4SLinus Torvalds 3592b98938c3SChristoph Lameter bh_cachep = kmem_cache_create("buffer_head", 3593b98938c3SChristoph Lameter sizeof(struct buffer_head), 0, 3594b98938c3SChristoph Lameter (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| 3595b98938c3SChristoph Lameter SLAB_MEM_SPREAD), 3596019b4d12SRichard Kennedy NULL); 35971da177e4SLinus Torvalds 35981da177e4SLinus Torvalds /* 35991da177e4SLinus Torvalds * Limit the bh occupancy to 10% of ZONE_NORMAL 36001da177e4SLinus Torvalds */ 36011da177e4SLinus Torvalds nrpages = (nr_free_buffer_pages() * 10) / 100; 36021da177e4SLinus Torvalds max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); 3603fc4d24c9SSebastian Andrzej Siewior ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", 3604fc4d24c9SSebastian Andrzej Siewior NULL, buffer_exit_cpu_dead); 3605fc4d24c9SSebastian Andrzej Siewior WARN_ON(ret < 0); 36061da177e4SLinus Torvalds } 3607