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> 221da177e4SLinus Torvalds #include <linux/syscalls.h> 231da177e4SLinus Torvalds #include <linux/fs.h> 241da177e4SLinus Torvalds #include <linux/mm.h> 251da177e4SLinus Torvalds #include <linux/percpu.h> 261da177e4SLinus Torvalds #include <linux/slab.h> 2716f7e0feSRandy Dunlap #include <linux/capability.h> 281da177e4SLinus Torvalds #include <linux/blkdev.h> 291da177e4SLinus Torvalds #include <linux/file.h> 301da177e4SLinus Torvalds #include <linux/quotaops.h> 311da177e4SLinus Torvalds #include <linux/highmem.h> 32630d9c47SPaul Gortmaker #include <linux/export.h> 33bafc0dbaSTejun Heo #include <linux/backing-dev.h> 341da177e4SLinus Torvalds #include <linux/writeback.h> 351da177e4SLinus Torvalds #include <linux/hash.h> 361da177e4SLinus Torvalds #include <linux/suspend.h> 371da177e4SLinus Torvalds #include <linux/buffer_head.h> 3855e829afSAndrew Morton #include <linux/task_io_accounting_ops.h> 391da177e4SLinus Torvalds #include <linux/bio.h> 401da177e4SLinus Torvalds #include <linux/notifier.h> 411da177e4SLinus Torvalds #include <linux/cpu.h> 421da177e4SLinus Torvalds #include <linux/bitops.h> 431da177e4SLinus Torvalds #include <linux/mpage.h> 44fb1c8f93SIngo Molnar #include <linux/bit_spinlock.h> 455305cb83STejun Heo #include <trace/events/block.h> 461da177e4SLinus Torvalds 471da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); 48b16b1debSTejun Heo static int submit_bh_wbc(int rw, struct buffer_head *bh, 49bafc0dbaSTejun Heo unsigned long bio_flags, 50b16b1debSTejun Heo struct writeback_control *wbc); 511da177e4SLinus Torvalds 521da177e4SLinus Torvalds #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) 531da177e4SLinus Torvalds 54a3f3c29cSYan Hong void init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private) 551da177e4SLinus Torvalds { 561da177e4SLinus Torvalds bh->b_end_io = handler; 571da177e4SLinus Torvalds bh->b_private = private; 581da177e4SLinus Torvalds } 591fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(init_buffer); 601da177e4SLinus Torvalds 61f0059afdSTejun Heo inline void touch_buffer(struct buffer_head *bh) 62f0059afdSTejun Heo { 635305cb83STejun Heo trace_block_touch_buffer(bh); 64f0059afdSTejun Heo mark_page_accessed(bh->b_page); 65f0059afdSTejun Heo } 66f0059afdSTejun Heo EXPORT_SYMBOL(touch_buffer); 67f0059afdSTejun Heo 68fc9b52cdSHarvey Harrison void __lock_buffer(struct buffer_head *bh) 691da177e4SLinus Torvalds { 7074316201SNeilBrown wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); 711da177e4SLinus Torvalds } 721da177e4SLinus Torvalds EXPORT_SYMBOL(__lock_buffer); 731da177e4SLinus Torvalds 74fc9b52cdSHarvey Harrison void unlock_buffer(struct buffer_head *bh) 751da177e4SLinus Torvalds { 7651b07fc3SNick Piggin clear_bit_unlock(BH_Lock, &bh->b_state); 774e857c58SPeter Zijlstra smp_mb__after_atomic(); 781da177e4SLinus Torvalds wake_up_bit(&bh->b_state, BH_Lock); 791da177e4SLinus Torvalds } 801fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(unlock_buffer); 811da177e4SLinus Torvalds 821da177e4SLinus Torvalds /* 83b4597226SMel Gorman * Returns if the page has dirty or writeback buffers. If all the buffers 84b4597226SMel Gorman * are unlocked and clean then the PageDirty information is stale. If 85b4597226SMel Gorman * any of the pages are locked, it is assumed they are locked for IO. 86b4597226SMel Gorman */ 87b4597226SMel Gorman void buffer_check_dirty_writeback(struct page *page, 88b4597226SMel Gorman bool *dirty, bool *writeback) 89b4597226SMel Gorman { 90b4597226SMel Gorman struct buffer_head *head, *bh; 91b4597226SMel Gorman *dirty = false; 92b4597226SMel Gorman *writeback = false; 93b4597226SMel Gorman 94b4597226SMel Gorman BUG_ON(!PageLocked(page)); 95b4597226SMel Gorman 96b4597226SMel Gorman if (!page_has_buffers(page)) 97b4597226SMel Gorman return; 98b4597226SMel Gorman 99b4597226SMel Gorman if (PageWriteback(page)) 100b4597226SMel Gorman *writeback = true; 101b4597226SMel Gorman 102b4597226SMel Gorman head = page_buffers(page); 103b4597226SMel Gorman bh = head; 104b4597226SMel Gorman do { 105b4597226SMel Gorman if (buffer_locked(bh)) 106b4597226SMel Gorman *writeback = true; 107b4597226SMel Gorman 108b4597226SMel Gorman if (buffer_dirty(bh)) 109b4597226SMel Gorman *dirty = true; 110b4597226SMel Gorman 111b4597226SMel Gorman bh = bh->b_this_page; 112b4597226SMel Gorman } while (bh != head); 113b4597226SMel Gorman } 114b4597226SMel Gorman EXPORT_SYMBOL(buffer_check_dirty_writeback); 115b4597226SMel Gorman 116b4597226SMel Gorman /* 1171da177e4SLinus Torvalds * Block until a buffer comes unlocked. This doesn't stop it 1181da177e4SLinus Torvalds * from becoming locked again - you have to lock it yourself 1191da177e4SLinus Torvalds * if you want to preserve its state. 1201da177e4SLinus Torvalds */ 1211da177e4SLinus Torvalds void __wait_on_buffer(struct buffer_head * bh) 1221da177e4SLinus Torvalds { 12374316201SNeilBrown wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); 1241da177e4SLinus Torvalds } 1251fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__wait_on_buffer); 1261da177e4SLinus Torvalds 1271da177e4SLinus Torvalds static void 1281da177e4SLinus Torvalds __clear_page_buffers(struct page *page) 1291da177e4SLinus Torvalds { 1301da177e4SLinus Torvalds ClearPagePrivate(page); 1314c21e2f2SHugh Dickins set_page_private(page, 0); 1321da177e4SLinus Torvalds page_cache_release(page); 1331da177e4SLinus Torvalds } 1341da177e4SLinus Torvalds 135b744c2acSRobert Elliott static void buffer_io_error(struct buffer_head *bh, char *msg) 1361da177e4SLinus Torvalds { 1371da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 138432f16e6SRobert Elliott 139432f16e6SRobert Elliott if (!test_bit(BH_Quiet, &bh->b_state)) 140432f16e6SRobert Elliott printk_ratelimited(KERN_ERR 141432f16e6SRobert Elliott "Buffer I/O error on dev %s, logical block %llu%s\n", 1421da177e4SLinus Torvalds bdevname(bh->b_bdev, b), 143b744c2acSRobert Elliott (unsigned long long)bh->b_blocknr, msg); 1441da177e4SLinus Torvalds } 1451da177e4SLinus Torvalds 1461da177e4SLinus Torvalds /* 14768671f35SDmitry Monakhov * End-of-IO handler helper function which does not touch the bh after 14868671f35SDmitry Monakhov * unlocking it. 14968671f35SDmitry Monakhov * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but 15068671f35SDmitry Monakhov * a race there is benign: unlock_buffer() only use the bh's address for 15168671f35SDmitry Monakhov * hashing after unlocking the buffer, so it doesn't actually touch the bh 15268671f35SDmitry Monakhov * itself. 1531da177e4SLinus Torvalds */ 15468671f35SDmitry Monakhov static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) 1551da177e4SLinus Torvalds { 1561da177e4SLinus Torvalds if (uptodate) { 1571da177e4SLinus Torvalds set_buffer_uptodate(bh); 1581da177e4SLinus Torvalds } else { 1591da177e4SLinus Torvalds /* This happens, due to failed READA attempts. */ 1601da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1611da177e4SLinus Torvalds } 1621da177e4SLinus Torvalds unlock_buffer(bh); 16368671f35SDmitry Monakhov } 16468671f35SDmitry Monakhov 16568671f35SDmitry Monakhov /* 16668671f35SDmitry Monakhov * Default synchronous end-of-IO handler.. Just mark it up-to-date and 16768671f35SDmitry Monakhov * unlock the buffer. This is what ll_rw_block uses too. 16868671f35SDmitry Monakhov */ 16968671f35SDmitry Monakhov void end_buffer_read_sync(struct buffer_head *bh, int uptodate) 17068671f35SDmitry Monakhov { 17168671f35SDmitry Monakhov __end_buffer_read_notouch(bh, uptodate); 1721da177e4SLinus Torvalds put_bh(bh); 1731da177e4SLinus Torvalds } 1741fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_read_sync); 1751da177e4SLinus Torvalds 1761da177e4SLinus Torvalds void end_buffer_write_sync(struct buffer_head *bh, int uptodate) 1771da177e4SLinus Torvalds { 1781da177e4SLinus Torvalds if (uptodate) { 1791da177e4SLinus Torvalds set_buffer_uptodate(bh); 1801da177e4SLinus Torvalds } else { 181b744c2acSRobert Elliott buffer_io_error(bh, ", lost sync page write"); 1821da177e4SLinus Torvalds set_buffer_write_io_error(bh); 1831da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1841da177e4SLinus Torvalds } 1851da177e4SLinus Torvalds unlock_buffer(bh); 1861da177e4SLinus Torvalds put_bh(bh); 1871da177e4SLinus Torvalds } 1881fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_write_sync); 1891da177e4SLinus Torvalds 1901da177e4SLinus Torvalds /* 1911da177e4SLinus Torvalds * Various filesystems appear to want __find_get_block to be non-blocking. 1921da177e4SLinus Torvalds * But it's the page lock which protects the buffers. To get around this, 1931da177e4SLinus Torvalds * we get exclusion from try_to_free_buffers with the blockdev mapping's 1941da177e4SLinus Torvalds * private_lock. 1951da177e4SLinus Torvalds * 1961da177e4SLinus Torvalds * Hack idea: for the blockdev mapping, i_bufferlist_lock contention 1971da177e4SLinus Torvalds * may be quite high. This code could TryLock the page, and if that 1981da177e4SLinus Torvalds * succeeds, there is no need to take private_lock. (But if 1991da177e4SLinus Torvalds * private_lock is contended then so is mapping->tree_lock). 2001da177e4SLinus Torvalds */ 2011da177e4SLinus Torvalds static struct buffer_head * 202385fd4c5SCoywolf Qi Hunt __find_get_block_slow(struct block_device *bdev, sector_t block) 2031da177e4SLinus Torvalds { 2041da177e4SLinus Torvalds struct inode *bd_inode = bdev->bd_inode; 2051da177e4SLinus Torvalds struct address_space *bd_mapping = bd_inode->i_mapping; 2061da177e4SLinus Torvalds struct buffer_head *ret = NULL; 2071da177e4SLinus Torvalds pgoff_t index; 2081da177e4SLinus Torvalds struct buffer_head *bh; 2091da177e4SLinus Torvalds struct buffer_head *head; 2101da177e4SLinus Torvalds struct page *page; 2111da177e4SLinus Torvalds int all_mapped = 1; 2121da177e4SLinus Torvalds 2131da177e4SLinus Torvalds index = block >> (PAGE_CACHE_SHIFT - bd_inode->i_blkbits); 2142457aec6SMel Gorman page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED); 2151da177e4SLinus Torvalds if (!page) 2161da177e4SLinus Torvalds goto out; 2171da177e4SLinus Torvalds 2181da177e4SLinus Torvalds spin_lock(&bd_mapping->private_lock); 2191da177e4SLinus Torvalds if (!page_has_buffers(page)) 2201da177e4SLinus Torvalds goto out_unlock; 2211da177e4SLinus Torvalds head = page_buffers(page); 2221da177e4SLinus Torvalds bh = head; 2231da177e4SLinus Torvalds do { 22497f76d3dSNikanth Karthikesan if (!buffer_mapped(bh)) 22597f76d3dSNikanth Karthikesan all_mapped = 0; 22697f76d3dSNikanth Karthikesan else if (bh->b_blocknr == block) { 2271da177e4SLinus Torvalds ret = bh; 2281da177e4SLinus Torvalds get_bh(bh); 2291da177e4SLinus Torvalds goto out_unlock; 2301da177e4SLinus Torvalds } 2311da177e4SLinus Torvalds bh = bh->b_this_page; 2321da177e4SLinus Torvalds } while (bh != head); 2331da177e4SLinus Torvalds 2341da177e4SLinus Torvalds /* we might be here because some of the buffers on this page are 2351da177e4SLinus Torvalds * not mapped. This is due to various races between 2361da177e4SLinus Torvalds * file io on the block device and getblk. It gets dealt with 2371da177e4SLinus Torvalds * elsewhere, don't buffer_error if we had some unmapped buffers 2381da177e4SLinus Torvalds */ 2391da177e4SLinus Torvalds if (all_mapped) { 24072a2ebd8STao Ma char b[BDEVNAME_SIZE]; 24172a2ebd8STao Ma 2421da177e4SLinus Torvalds printk("__find_get_block_slow() failed. " 2431da177e4SLinus Torvalds "block=%llu, b_blocknr=%llu\n", 244205f87f6SBadari Pulavarty (unsigned long long)block, 245205f87f6SBadari Pulavarty (unsigned long long)bh->b_blocknr); 246205f87f6SBadari Pulavarty printk("b_state=0x%08lx, b_size=%zu\n", 247205f87f6SBadari Pulavarty bh->b_state, bh->b_size); 24872a2ebd8STao Ma printk("device %s blocksize: %d\n", bdevname(bdev, b), 24972a2ebd8STao Ma 1 << bd_inode->i_blkbits); 2501da177e4SLinus Torvalds } 2511da177e4SLinus Torvalds out_unlock: 2521da177e4SLinus Torvalds spin_unlock(&bd_mapping->private_lock); 2531da177e4SLinus Torvalds page_cache_release(page); 2541da177e4SLinus Torvalds out: 2551da177e4SLinus Torvalds return ret; 2561da177e4SLinus Torvalds } 2571da177e4SLinus Torvalds 2581da177e4SLinus Torvalds /* 2595b0830cbSJens Axboe * Kick the writeback threads then try to free up some ZONE_NORMAL memory. 2601da177e4SLinus Torvalds */ 2611da177e4SLinus Torvalds static void free_more_memory(void) 2621da177e4SLinus Torvalds { 26319770b32SMel Gorman struct zone *zone; 2640e88460dSMel Gorman int nid; 2651da177e4SLinus Torvalds 2660e175a18SCurt Wohlgemuth wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM); 2671da177e4SLinus Torvalds yield(); 2681da177e4SLinus Torvalds 2690e88460dSMel Gorman for_each_online_node(nid) { 27019770b32SMel Gorman (void)first_zones_zonelist(node_zonelist(nid, GFP_NOFS), 27119770b32SMel Gorman gfp_zone(GFP_NOFS), NULL, 27219770b32SMel Gorman &zone); 27319770b32SMel Gorman if (zone) 27454a6eb5cSMel Gorman try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0, 275327c0e96SKAMEZAWA Hiroyuki GFP_NOFS, NULL); 2761da177e4SLinus Torvalds } 2771da177e4SLinus Torvalds } 2781da177e4SLinus Torvalds 2791da177e4SLinus Torvalds /* 2801da177e4SLinus Torvalds * I/O completion handler for block_read_full_page() - pages 2811da177e4SLinus Torvalds * which come unlocked at the end of I/O. 2821da177e4SLinus Torvalds */ 2831da177e4SLinus Torvalds static void end_buffer_async_read(struct buffer_head *bh, int uptodate) 2841da177e4SLinus Torvalds { 2851da177e4SLinus Torvalds unsigned long flags; 286a3972203SNick Piggin struct buffer_head *first; 2871da177e4SLinus Torvalds struct buffer_head *tmp; 2881da177e4SLinus Torvalds struct page *page; 2891da177e4SLinus Torvalds int page_uptodate = 1; 2901da177e4SLinus Torvalds 2911da177e4SLinus Torvalds BUG_ON(!buffer_async_read(bh)); 2921da177e4SLinus Torvalds 2931da177e4SLinus Torvalds page = bh->b_page; 2941da177e4SLinus Torvalds if (uptodate) { 2951da177e4SLinus Torvalds set_buffer_uptodate(bh); 2961da177e4SLinus Torvalds } else { 2971da177e4SLinus Torvalds clear_buffer_uptodate(bh); 298b744c2acSRobert Elliott buffer_io_error(bh, ", async page read"); 2991da177e4SLinus Torvalds SetPageError(page); 3001da177e4SLinus Torvalds } 3011da177e4SLinus Torvalds 3021da177e4SLinus Torvalds /* 3031da177e4SLinus Torvalds * Be _very_ careful from here on. Bad things can happen if 3041da177e4SLinus Torvalds * two buffer heads end IO at almost the same time and both 3051da177e4SLinus Torvalds * decide that the page is now completely done. 3061da177e4SLinus Torvalds */ 307a3972203SNick Piggin first = page_buffers(page); 308a3972203SNick Piggin local_irq_save(flags); 309a3972203SNick Piggin bit_spin_lock(BH_Uptodate_Lock, &first->b_state); 3101da177e4SLinus Torvalds clear_buffer_async_read(bh); 3111da177e4SLinus Torvalds unlock_buffer(bh); 3121da177e4SLinus Torvalds tmp = bh; 3131da177e4SLinus Torvalds do { 3141da177e4SLinus Torvalds if (!buffer_uptodate(tmp)) 3151da177e4SLinus Torvalds page_uptodate = 0; 3161da177e4SLinus Torvalds if (buffer_async_read(tmp)) { 3171da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 3181da177e4SLinus Torvalds goto still_busy; 3191da177e4SLinus Torvalds } 3201da177e4SLinus Torvalds tmp = tmp->b_this_page; 3211da177e4SLinus Torvalds } while (tmp != bh); 322a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 323a3972203SNick Piggin local_irq_restore(flags); 3241da177e4SLinus Torvalds 3251da177e4SLinus Torvalds /* 3261da177e4SLinus Torvalds * If none of the buffers had errors and they are all 3271da177e4SLinus Torvalds * uptodate then we can set the page uptodate. 3281da177e4SLinus Torvalds */ 3291da177e4SLinus Torvalds if (page_uptodate && !PageError(page)) 3301da177e4SLinus Torvalds SetPageUptodate(page); 3311da177e4SLinus Torvalds unlock_page(page); 3321da177e4SLinus Torvalds return; 3331da177e4SLinus Torvalds 3341da177e4SLinus Torvalds still_busy: 335a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 336a3972203SNick Piggin local_irq_restore(flags); 3371da177e4SLinus Torvalds return; 3381da177e4SLinus Torvalds } 3391da177e4SLinus Torvalds 3401da177e4SLinus Torvalds /* 3411da177e4SLinus Torvalds * Completion handler for block_write_full_page() - pages which are unlocked 3421da177e4SLinus Torvalds * during I/O, and which have PageWriteback cleared upon I/O completion. 3431da177e4SLinus Torvalds */ 34435c80d5fSChris Mason void end_buffer_async_write(struct buffer_head *bh, int uptodate) 3451da177e4SLinus Torvalds { 3461da177e4SLinus Torvalds unsigned long flags; 347a3972203SNick Piggin struct buffer_head *first; 3481da177e4SLinus Torvalds struct buffer_head *tmp; 3491da177e4SLinus Torvalds struct page *page; 3501da177e4SLinus Torvalds 3511da177e4SLinus Torvalds BUG_ON(!buffer_async_write(bh)); 3521da177e4SLinus Torvalds 3531da177e4SLinus Torvalds page = bh->b_page; 3541da177e4SLinus Torvalds if (uptodate) { 3551da177e4SLinus Torvalds set_buffer_uptodate(bh); 3561da177e4SLinus Torvalds } else { 357b744c2acSRobert Elliott buffer_io_error(bh, ", lost async page write"); 3581da177e4SLinus Torvalds set_bit(AS_EIO, &page->mapping->flags); 35958ff407bSJan Kara set_buffer_write_io_error(bh); 3601da177e4SLinus Torvalds clear_buffer_uptodate(bh); 3611da177e4SLinus Torvalds SetPageError(page); 3621da177e4SLinus Torvalds } 3631da177e4SLinus Torvalds 364a3972203SNick Piggin first = page_buffers(page); 365a3972203SNick Piggin local_irq_save(flags); 366a3972203SNick Piggin bit_spin_lock(BH_Uptodate_Lock, &first->b_state); 367a3972203SNick Piggin 3681da177e4SLinus Torvalds clear_buffer_async_write(bh); 3691da177e4SLinus Torvalds unlock_buffer(bh); 3701da177e4SLinus Torvalds tmp = bh->b_this_page; 3711da177e4SLinus Torvalds while (tmp != bh) { 3721da177e4SLinus Torvalds if (buffer_async_write(tmp)) { 3731da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 3741da177e4SLinus Torvalds goto still_busy; 3751da177e4SLinus Torvalds } 3761da177e4SLinus Torvalds tmp = tmp->b_this_page; 3771da177e4SLinus Torvalds } 378a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 379a3972203SNick Piggin local_irq_restore(flags); 3801da177e4SLinus Torvalds end_page_writeback(page); 3811da177e4SLinus Torvalds return; 3821da177e4SLinus Torvalds 3831da177e4SLinus Torvalds still_busy: 384a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 385a3972203SNick Piggin local_irq_restore(flags); 3861da177e4SLinus Torvalds return; 3871da177e4SLinus Torvalds } 3881fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_async_write); 3891da177e4SLinus Torvalds 3901da177e4SLinus Torvalds /* 3911da177e4SLinus Torvalds * If a page's buffers are under async readin (end_buffer_async_read 3921da177e4SLinus Torvalds * completion) then there is a possibility that another thread of 3931da177e4SLinus Torvalds * control could lock one of the buffers after it has completed 3941da177e4SLinus Torvalds * but while some of the other buffers have not completed. This 3951da177e4SLinus Torvalds * locked buffer would confuse end_buffer_async_read() into not unlocking 3961da177e4SLinus Torvalds * the page. So the absence of BH_Async_Read tells end_buffer_async_read() 3971da177e4SLinus Torvalds * that this buffer is not under async I/O. 3981da177e4SLinus Torvalds * 3991da177e4SLinus Torvalds * The page comes unlocked when it has no locked buffer_async buffers 4001da177e4SLinus Torvalds * left. 4011da177e4SLinus Torvalds * 4021da177e4SLinus Torvalds * PageLocked prevents anyone starting new async I/O reads any of 4031da177e4SLinus Torvalds * the buffers. 4041da177e4SLinus Torvalds * 4051da177e4SLinus Torvalds * PageWriteback is used to prevent simultaneous writeout of the same 4061da177e4SLinus Torvalds * page. 4071da177e4SLinus Torvalds * 4081da177e4SLinus Torvalds * PageLocked prevents anyone from starting writeback of a page which is 4091da177e4SLinus Torvalds * under read I/O (PageWriteback is only ever set against a locked page). 4101da177e4SLinus Torvalds */ 4111da177e4SLinus Torvalds static void mark_buffer_async_read(struct buffer_head *bh) 4121da177e4SLinus Torvalds { 4131da177e4SLinus Torvalds bh->b_end_io = end_buffer_async_read; 4141da177e4SLinus Torvalds set_buffer_async_read(bh); 4151da177e4SLinus Torvalds } 4161da177e4SLinus Torvalds 4171fe72eaaSH Hartley Sweeten static void mark_buffer_async_write_endio(struct buffer_head *bh, 41835c80d5fSChris Mason bh_end_io_t *handler) 41935c80d5fSChris Mason { 42035c80d5fSChris Mason bh->b_end_io = handler; 42135c80d5fSChris Mason set_buffer_async_write(bh); 42235c80d5fSChris Mason } 42335c80d5fSChris Mason 4241da177e4SLinus Torvalds void mark_buffer_async_write(struct buffer_head *bh) 4251da177e4SLinus Torvalds { 42635c80d5fSChris Mason mark_buffer_async_write_endio(bh, end_buffer_async_write); 4271da177e4SLinus Torvalds } 4281da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_async_write); 4291da177e4SLinus Torvalds 4301da177e4SLinus Torvalds 4311da177e4SLinus Torvalds /* 4321da177e4SLinus Torvalds * fs/buffer.c contains helper functions for buffer-backed address space's 4331da177e4SLinus Torvalds * fsync functions. A common requirement for buffer-based filesystems is 4341da177e4SLinus Torvalds * that certain data from the backing blockdev needs to be written out for 4351da177e4SLinus Torvalds * a successful fsync(). For example, ext2 indirect blocks need to be 4361da177e4SLinus Torvalds * written back and waited upon before fsync() returns. 4371da177e4SLinus Torvalds * 4381da177e4SLinus Torvalds * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), 4391da177e4SLinus Torvalds * inode_has_buffers() and invalidate_inode_buffers() are provided for the 4401da177e4SLinus Torvalds * management of a list of dependent buffers at ->i_mapping->private_list. 4411da177e4SLinus Torvalds * 4421da177e4SLinus Torvalds * Locking is a little subtle: try_to_free_buffers() will remove buffers 4431da177e4SLinus Torvalds * from their controlling inode's queue when they are being freed. But 4441da177e4SLinus Torvalds * try_to_free_buffers() will be operating against the *blockdev* mapping 4451da177e4SLinus Torvalds * at the time, not against the S_ISREG file which depends on those buffers. 4461da177e4SLinus Torvalds * So the locking for private_list is via the private_lock in the address_space 4471da177e4SLinus Torvalds * which backs the buffers. Which is different from the address_space 4481da177e4SLinus Torvalds * against which the buffers are listed. So for a particular address_space, 4491da177e4SLinus Torvalds * mapping->private_lock does *not* protect mapping->private_list! In fact, 4501da177e4SLinus Torvalds * mapping->private_list will always be protected by the backing blockdev's 4511da177e4SLinus Torvalds * ->private_lock. 4521da177e4SLinus Torvalds * 4531da177e4SLinus Torvalds * Which introduces a requirement: all buffers on an address_space's 4541da177e4SLinus Torvalds * ->private_list must be from the same address_space: the blockdev's. 4551da177e4SLinus Torvalds * 4561da177e4SLinus Torvalds * address_spaces which do not place buffers at ->private_list via these 4571da177e4SLinus Torvalds * utility functions are free to use private_lock and private_list for 4581da177e4SLinus Torvalds * whatever they want. The only requirement is that list_empty(private_list) 4591da177e4SLinus Torvalds * be true at clear_inode() time. 4601da177e4SLinus Torvalds * 4611da177e4SLinus Torvalds * FIXME: clear_inode should not call invalidate_inode_buffers(). The 4621da177e4SLinus Torvalds * filesystems should do that. invalidate_inode_buffers() should just go 4631da177e4SLinus Torvalds * BUG_ON(!list_empty). 4641da177e4SLinus Torvalds * 4651da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should 4661da177e4SLinus Torvalds * take an address_space, not an inode. And it should be called 4671da177e4SLinus Torvalds * mark_buffer_dirty_fsync() to clearly define why those buffers are being 4681da177e4SLinus Torvalds * queued up. 4691da177e4SLinus Torvalds * 4701da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the 4711da177e4SLinus Torvalds * list if it is already on a list. Because if the buffer is on a list, 4721da177e4SLinus Torvalds * it *must* already be on the right one. If not, the filesystem is being 4731da177e4SLinus Torvalds * silly. This will save a ton of locking. But first we have to ensure 4741da177e4SLinus Torvalds * that buffers are taken *off* the old inode's list when they are freed 4751da177e4SLinus Torvalds * (presumably in truncate). That requires careful auditing of all 4761da177e4SLinus Torvalds * filesystems (do it inside bforget()). It could also be done by bringing 4771da177e4SLinus Torvalds * b_inode back. 4781da177e4SLinus Torvalds */ 4791da177e4SLinus Torvalds 4801da177e4SLinus Torvalds /* 4811da177e4SLinus Torvalds * The buffer's backing address_space's private_lock must be held 4821da177e4SLinus Torvalds */ 483dbacefc9SThomas Petazzoni static void __remove_assoc_queue(struct buffer_head *bh) 4841da177e4SLinus Torvalds { 4851da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 48658ff407bSJan Kara WARN_ON(!bh->b_assoc_map); 48758ff407bSJan Kara if (buffer_write_io_error(bh)) 48858ff407bSJan Kara set_bit(AS_EIO, &bh->b_assoc_map->flags); 48958ff407bSJan Kara bh->b_assoc_map = NULL; 4901da177e4SLinus Torvalds } 4911da177e4SLinus Torvalds 4921da177e4SLinus Torvalds int inode_has_buffers(struct inode *inode) 4931da177e4SLinus Torvalds { 4941da177e4SLinus Torvalds return !list_empty(&inode->i_data.private_list); 4951da177e4SLinus Torvalds } 4961da177e4SLinus Torvalds 4971da177e4SLinus Torvalds /* 4981da177e4SLinus Torvalds * osync is designed to support O_SYNC io. It waits synchronously for 4991da177e4SLinus Torvalds * all already-submitted IO to complete, but does not queue any new 5001da177e4SLinus Torvalds * writes to the disk. 5011da177e4SLinus Torvalds * 5021da177e4SLinus Torvalds * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as 5031da177e4SLinus Torvalds * you dirty the buffers, and then use osync_inode_buffers to wait for 5041da177e4SLinus Torvalds * completion. Any other dirty buffers which are not yet queued for 5051da177e4SLinus Torvalds * write will not be flushed to disk by the osync. 5061da177e4SLinus Torvalds */ 5071da177e4SLinus Torvalds static int osync_buffers_list(spinlock_t *lock, struct list_head *list) 5081da177e4SLinus Torvalds { 5091da177e4SLinus Torvalds struct buffer_head *bh; 5101da177e4SLinus Torvalds struct list_head *p; 5111da177e4SLinus Torvalds int err = 0; 5121da177e4SLinus Torvalds 5131da177e4SLinus Torvalds spin_lock(lock); 5141da177e4SLinus Torvalds repeat: 5151da177e4SLinus Torvalds list_for_each_prev(p, list) { 5161da177e4SLinus Torvalds bh = BH_ENTRY(p); 5171da177e4SLinus Torvalds if (buffer_locked(bh)) { 5181da177e4SLinus Torvalds get_bh(bh); 5191da177e4SLinus Torvalds spin_unlock(lock); 5201da177e4SLinus Torvalds wait_on_buffer(bh); 5211da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 5221da177e4SLinus Torvalds err = -EIO; 5231da177e4SLinus Torvalds brelse(bh); 5241da177e4SLinus Torvalds spin_lock(lock); 5251da177e4SLinus Torvalds goto repeat; 5261da177e4SLinus Torvalds } 5271da177e4SLinus Torvalds } 5281da177e4SLinus Torvalds spin_unlock(lock); 5291da177e4SLinus Torvalds return err; 5301da177e4SLinus Torvalds } 5311da177e4SLinus Torvalds 53201a05b33SAl Viro static void do_thaw_one(struct super_block *sb, void *unused) 533c2d75438SEric Sandeen { 534c2d75438SEric Sandeen char b[BDEVNAME_SIZE]; 535c2d75438SEric Sandeen while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb)) 536c2d75438SEric Sandeen printk(KERN_WARNING "Emergency Thaw on %s\n", 537c2d75438SEric Sandeen bdevname(sb->s_bdev, b)); 538c2d75438SEric Sandeen } 53901a05b33SAl Viro 54001a05b33SAl Viro static void do_thaw_all(struct work_struct *work) 54101a05b33SAl Viro { 54201a05b33SAl Viro iterate_supers(do_thaw_one, NULL); 543053c525fSJens Axboe kfree(work); 544c2d75438SEric Sandeen printk(KERN_WARNING "Emergency Thaw complete\n"); 545c2d75438SEric Sandeen } 546c2d75438SEric Sandeen 547c2d75438SEric Sandeen /** 548c2d75438SEric Sandeen * emergency_thaw_all -- forcibly thaw every frozen filesystem 549c2d75438SEric Sandeen * 550c2d75438SEric Sandeen * Used for emergency unfreeze of all filesystems via SysRq 551c2d75438SEric Sandeen */ 552c2d75438SEric Sandeen void emergency_thaw_all(void) 553c2d75438SEric Sandeen { 554053c525fSJens Axboe struct work_struct *work; 555053c525fSJens Axboe 556053c525fSJens Axboe work = kmalloc(sizeof(*work), GFP_ATOMIC); 557053c525fSJens Axboe if (work) { 558053c525fSJens Axboe INIT_WORK(work, do_thaw_all); 559053c525fSJens Axboe schedule_work(work); 560053c525fSJens Axboe } 561c2d75438SEric Sandeen } 562c2d75438SEric Sandeen 5631da177e4SLinus Torvalds /** 56478a4a50aSRandy Dunlap * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers 56567be2dd1SMartin Waitz * @mapping: the mapping which wants those buffers written 5661da177e4SLinus Torvalds * 5671da177e4SLinus Torvalds * Starts I/O against the buffers at mapping->private_list, and waits upon 5681da177e4SLinus Torvalds * that I/O. 5691da177e4SLinus Torvalds * 57067be2dd1SMartin Waitz * Basically, this is a convenience function for fsync(). 57167be2dd1SMartin Waitz * @mapping is a file or directory which needs those buffers to be written for 57267be2dd1SMartin Waitz * a successful fsync(). 5731da177e4SLinus Torvalds */ 5741da177e4SLinus Torvalds int sync_mapping_buffers(struct address_space *mapping) 5751da177e4SLinus Torvalds { 576252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 5771da177e4SLinus Torvalds 5781da177e4SLinus Torvalds if (buffer_mapping == NULL || list_empty(&mapping->private_list)) 5791da177e4SLinus Torvalds return 0; 5801da177e4SLinus Torvalds 5811da177e4SLinus Torvalds return fsync_buffers_list(&buffer_mapping->private_lock, 5821da177e4SLinus Torvalds &mapping->private_list); 5831da177e4SLinus Torvalds } 5841da177e4SLinus Torvalds EXPORT_SYMBOL(sync_mapping_buffers); 5851da177e4SLinus Torvalds 5861da177e4SLinus Torvalds /* 5871da177e4SLinus Torvalds * Called when we've recently written block `bblock', and it is known that 5881da177e4SLinus Torvalds * `bblock' was for a buffer_boundary() buffer. This means that the block at 5891da177e4SLinus Torvalds * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's 5901da177e4SLinus Torvalds * dirty, schedule it for IO. So that indirects merge nicely with their data. 5911da177e4SLinus Torvalds */ 5921da177e4SLinus Torvalds void write_boundary_block(struct block_device *bdev, 5931da177e4SLinus Torvalds sector_t bblock, unsigned blocksize) 5941da177e4SLinus Torvalds { 5951da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); 5961da177e4SLinus Torvalds if (bh) { 5971da177e4SLinus Torvalds if (buffer_dirty(bh)) 5981da177e4SLinus Torvalds ll_rw_block(WRITE, 1, &bh); 5991da177e4SLinus Torvalds put_bh(bh); 6001da177e4SLinus Torvalds } 6011da177e4SLinus Torvalds } 6021da177e4SLinus Torvalds 6031da177e4SLinus Torvalds void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) 6041da177e4SLinus Torvalds { 6051da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 6061da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 6071da177e4SLinus Torvalds 6081da177e4SLinus Torvalds mark_buffer_dirty(bh); 609252aa6f5SRafael Aquini if (!mapping->private_data) { 610252aa6f5SRafael Aquini mapping->private_data = buffer_mapping; 6111da177e4SLinus Torvalds } else { 612252aa6f5SRafael Aquini BUG_ON(mapping->private_data != buffer_mapping); 6131da177e4SLinus Torvalds } 614535ee2fbSJan Kara if (!bh->b_assoc_map) { 6151da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 6161da177e4SLinus Torvalds list_move_tail(&bh->b_assoc_buffers, 6171da177e4SLinus Torvalds &mapping->private_list); 61858ff407bSJan Kara bh->b_assoc_map = mapping; 6191da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 6201da177e4SLinus Torvalds } 6211da177e4SLinus Torvalds } 6221da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_dirty_inode); 6231da177e4SLinus Torvalds 6241da177e4SLinus Torvalds /* 625787d2214SNick Piggin * Mark the page dirty, and set it dirty in the radix tree, and mark the inode 626787d2214SNick Piggin * dirty. 627787d2214SNick Piggin * 628787d2214SNick Piggin * If warn is true, then emit a warning if the page is not uptodate and has 629787d2214SNick Piggin * not been truncated. 630c4843a75SGreg Thelen * 631c4843a75SGreg Thelen * The caller must hold mem_cgroup_begin_page_stat() lock. 632787d2214SNick Piggin */ 633c4843a75SGreg Thelen static void __set_page_dirty(struct page *page, struct address_space *mapping, 634c4843a75SGreg Thelen struct mem_cgroup *memcg, int warn) 635787d2214SNick Piggin { 636227d53b3SKOSAKI Motohiro unsigned long flags; 637227d53b3SKOSAKI Motohiro 638227d53b3SKOSAKI Motohiro spin_lock_irqsave(&mapping->tree_lock, flags); 639787d2214SNick Piggin if (page->mapping) { /* Race with truncate? */ 640787d2214SNick Piggin WARN_ON_ONCE(warn && !PageUptodate(page)); 641c4843a75SGreg Thelen account_page_dirtied(page, mapping, memcg); 642787d2214SNick Piggin radix_tree_tag_set(&mapping->page_tree, 643787d2214SNick Piggin page_index(page), PAGECACHE_TAG_DIRTY); 644787d2214SNick Piggin } 645227d53b3SKOSAKI Motohiro spin_unlock_irqrestore(&mapping->tree_lock, flags); 646787d2214SNick Piggin } 647787d2214SNick Piggin 648787d2214SNick Piggin /* 6491da177e4SLinus Torvalds * Add a page to the dirty page list. 6501da177e4SLinus Torvalds * 6511da177e4SLinus Torvalds * It is a sad fact of life that this function is called from several places 6521da177e4SLinus Torvalds * deeply under spinlocking. It may not sleep. 6531da177e4SLinus Torvalds * 6541da177e4SLinus Torvalds * If the page has buffers, the uptodate buffers are set dirty, to preserve 6551da177e4SLinus Torvalds * dirty-state coherency between the page and the buffers. It the page does 6561da177e4SLinus Torvalds * not have buffers then when they are later attached they will all be set 6571da177e4SLinus Torvalds * dirty. 6581da177e4SLinus Torvalds * 6591da177e4SLinus Torvalds * The buffers are dirtied before the page is dirtied. There's a small race 6601da177e4SLinus Torvalds * window in which a writepage caller may see the page cleanness but not the 6611da177e4SLinus Torvalds * buffer dirtiness. That's fine. If this code were to set the page dirty 6621da177e4SLinus Torvalds * before the buffers, a concurrent writepage caller could clear the page dirty 6631da177e4SLinus Torvalds * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean 6641da177e4SLinus Torvalds * page on the dirty page list. 6651da177e4SLinus Torvalds * 6661da177e4SLinus Torvalds * We use private_lock to lock against try_to_free_buffers while using the 6671da177e4SLinus Torvalds * page's buffer list. Also use this to protect against clean buffers being 6681da177e4SLinus Torvalds * added to the page after it was set dirty. 6691da177e4SLinus Torvalds * 6701da177e4SLinus Torvalds * FIXME: may need to call ->reservepage here as well. That's rather up to the 6711da177e4SLinus Torvalds * address_space though. 6721da177e4SLinus Torvalds */ 6731da177e4SLinus Torvalds int __set_page_dirty_buffers(struct page *page) 6741da177e4SLinus Torvalds { 675a8e7d49aSLinus Torvalds int newly_dirty; 676c4843a75SGreg Thelen struct mem_cgroup *memcg; 677787d2214SNick Piggin struct address_space *mapping = page_mapping(page); 678ebf7a227SNick Piggin 679ebf7a227SNick Piggin if (unlikely(!mapping)) 680ebf7a227SNick Piggin return !TestSetPageDirty(page); 6811da177e4SLinus Torvalds 6821da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 6831da177e4SLinus Torvalds if (page_has_buffers(page)) { 6841da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 6851da177e4SLinus Torvalds struct buffer_head *bh = head; 6861da177e4SLinus Torvalds 6871da177e4SLinus Torvalds do { 6881da177e4SLinus Torvalds set_buffer_dirty(bh); 6891da177e4SLinus Torvalds bh = bh->b_this_page; 6901da177e4SLinus Torvalds } while (bh != head); 6911da177e4SLinus Torvalds } 692c4843a75SGreg Thelen /* 693c4843a75SGreg Thelen * Use mem_group_begin_page_stat() to keep PageDirty synchronized with 694c4843a75SGreg Thelen * per-memcg dirty page counters. 695c4843a75SGreg Thelen */ 696c4843a75SGreg Thelen memcg = mem_cgroup_begin_page_stat(page); 697a8e7d49aSLinus Torvalds newly_dirty = !TestSetPageDirty(page); 6981da177e4SLinus Torvalds spin_unlock(&mapping->private_lock); 6991da177e4SLinus Torvalds 700a8e7d49aSLinus Torvalds if (newly_dirty) 701c4843a75SGreg Thelen __set_page_dirty(page, mapping, memcg, 1); 702c4843a75SGreg Thelen 703c4843a75SGreg Thelen mem_cgroup_end_page_stat(memcg); 704c4843a75SGreg Thelen 705c4843a75SGreg Thelen if (newly_dirty) 706c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 707c4843a75SGreg Thelen 708a8e7d49aSLinus Torvalds return newly_dirty; 7091da177e4SLinus Torvalds } 7101da177e4SLinus Torvalds EXPORT_SYMBOL(__set_page_dirty_buffers); 7111da177e4SLinus Torvalds 7121da177e4SLinus Torvalds /* 7131da177e4SLinus Torvalds * Write out and wait upon a list of buffers. 7141da177e4SLinus Torvalds * 7151da177e4SLinus Torvalds * We have conflicting pressures: we want to make sure that all 7161da177e4SLinus Torvalds * initially dirty buffers get waited on, but that any subsequently 7171da177e4SLinus Torvalds * dirtied buffers don't. After all, we don't want fsync to last 7181da177e4SLinus Torvalds * forever if somebody is actively writing to the file. 7191da177e4SLinus Torvalds * 7201da177e4SLinus Torvalds * Do this in two main stages: first we copy dirty buffers to a 7211da177e4SLinus Torvalds * temporary inode list, queueing the writes as we go. Then we clean 7221da177e4SLinus Torvalds * up, waiting for those writes to complete. 7231da177e4SLinus Torvalds * 7241da177e4SLinus Torvalds * During this second stage, any subsequent updates to the file may end 7251da177e4SLinus Torvalds * up refiling the buffer on the original inode's dirty list again, so 7261da177e4SLinus Torvalds * there is a chance we will end up with a buffer queued for write but 7271da177e4SLinus Torvalds * not yet completed on that list. So, as a final cleanup we go through 7281da177e4SLinus Torvalds * the osync code to catch these locked, dirty buffers without requeuing 7291da177e4SLinus Torvalds * any newly dirty buffers for write. 7301da177e4SLinus Torvalds */ 7311da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) 7321da177e4SLinus Torvalds { 7331da177e4SLinus Torvalds struct buffer_head *bh; 7341da177e4SLinus Torvalds struct list_head tmp; 7357eaceaccSJens Axboe struct address_space *mapping; 7361da177e4SLinus Torvalds int err = 0, err2; 7374ee2491eSJens Axboe struct blk_plug plug; 7381da177e4SLinus Torvalds 7391da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 7404ee2491eSJens Axboe blk_start_plug(&plug); 7411da177e4SLinus Torvalds 7421da177e4SLinus Torvalds spin_lock(lock); 7431da177e4SLinus Torvalds while (!list_empty(list)) { 7441da177e4SLinus Torvalds bh = BH_ENTRY(list->next); 745535ee2fbSJan Kara mapping = bh->b_assoc_map; 74658ff407bSJan Kara __remove_assoc_queue(bh); 747535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 748535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 749535ee2fbSJan Kara smp_mb(); 7501da177e4SLinus Torvalds if (buffer_dirty(bh) || buffer_locked(bh)) { 7511da177e4SLinus Torvalds list_add(&bh->b_assoc_buffers, &tmp); 752535ee2fbSJan Kara bh->b_assoc_map = mapping; 7531da177e4SLinus Torvalds if (buffer_dirty(bh)) { 7541da177e4SLinus Torvalds get_bh(bh); 7551da177e4SLinus Torvalds spin_unlock(lock); 7561da177e4SLinus Torvalds /* 7571da177e4SLinus Torvalds * Ensure any pending I/O completes so that 7589cb569d6SChristoph Hellwig * write_dirty_buffer() actually writes the 7599cb569d6SChristoph Hellwig * current contents - it is a noop if I/O is 7609cb569d6SChristoph Hellwig * still in flight on potentially older 7619cb569d6SChristoph Hellwig * contents. 7621da177e4SLinus Torvalds */ 763721a9602SJens Axboe write_dirty_buffer(bh, WRITE_SYNC); 7649cf6b720SJens Axboe 7659cf6b720SJens Axboe /* 7669cf6b720SJens Axboe * Kick off IO for the previous mapping. Note 7679cf6b720SJens Axboe * that we will not run the very last mapping, 7689cf6b720SJens Axboe * wait_on_buffer() will do that for us 7699cf6b720SJens Axboe * through sync_buffer(). 7709cf6b720SJens Axboe */ 7711da177e4SLinus Torvalds brelse(bh); 7721da177e4SLinus Torvalds spin_lock(lock); 7731da177e4SLinus Torvalds } 7741da177e4SLinus Torvalds } 7751da177e4SLinus Torvalds } 7761da177e4SLinus Torvalds 7774ee2491eSJens Axboe spin_unlock(lock); 7784ee2491eSJens Axboe blk_finish_plug(&plug); 7794ee2491eSJens Axboe spin_lock(lock); 7804ee2491eSJens Axboe 7811da177e4SLinus Torvalds while (!list_empty(&tmp)) { 7821da177e4SLinus Torvalds bh = BH_ENTRY(tmp.prev); 7831da177e4SLinus Torvalds get_bh(bh); 784535ee2fbSJan Kara mapping = bh->b_assoc_map; 785535ee2fbSJan Kara __remove_assoc_queue(bh); 786535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 787535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 788535ee2fbSJan Kara smp_mb(); 789535ee2fbSJan Kara if (buffer_dirty(bh)) { 790535ee2fbSJan Kara list_add(&bh->b_assoc_buffers, 791e3892296SJan Kara &mapping->private_list); 792535ee2fbSJan Kara bh->b_assoc_map = mapping; 793535ee2fbSJan Kara } 7941da177e4SLinus Torvalds spin_unlock(lock); 7951da177e4SLinus Torvalds wait_on_buffer(bh); 7961da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 7971da177e4SLinus Torvalds err = -EIO; 7981da177e4SLinus Torvalds brelse(bh); 7991da177e4SLinus Torvalds spin_lock(lock); 8001da177e4SLinus Torvalds } 8011da177e4SLinus Torvalds 8021da177e4SLinus Torvalds spin_unlock(lock); 8031da177e4SLinus Torvalds err2 = osync_buffers_list(lock, list); 8041da177e4SLinus Torvalds if (err) 8051da177e4SLinus Torvalds return err; 8061da177e4SLinus Torvalds else 8071da177e4SLinus Torvalds return err2; 8081da177e4SLinus Torvalds } 8091da177e4SLinus Torvalds 8101da177e4SLinus Torvalds /* 8111da177e4SLinus Torvalds * Invalidate any and all dirty buffers on a given inode. We are 8121da177e4SLinus Torvalds * probably unmounting the fs, but that doesn't mean we have already 8131da177e4SLinus Torvalds * done a sync(). Just drop the buffers from the inode list. 8141da177e4SLinus Torvalds * 8151da177e4SLinus Torvalds * NOTE: we take the inode's blockdev's mapping's private_lock. Which 8161da177e4SLinus Torvalds * assumes that all the buffers are against the blockdev. Not true 8171da177e4SLinus Torvalds * for reiserfs. 8181da177e4SLinus Torvalds */ 8191da177e4SLinus Torvalds void invalidate_inode_buffers(struct inode *inode) 8201da177e4SLinus Torvalds { 8211da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8221da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 8231da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 824252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 8251da177e4SLinus Torvalds 8261da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8271da177e4SLinus Torvalds while (!list_empty(list)) 8281da177e4SLinus Torvalds __remove_assoc_queue(BH_ENTRY(list->next)); 8291da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8301da177e4SLinus Torvalds } 8311da177e4SLinus Torvalds } 83252b19ac9SJan Kara EXPORT_SYMBOL(invalidate_inode_buffers); 8331da177e4SLinus Torvalds 8341da177e4SLinus Torvalds /* 8351da177e4SLinus Torvalds * Remove any clean buffers from the inode's buffer list. This is called 8361da177e4SLinus Torvalds * when we're trying to free the inode itself. Those buffers can pin it. 8371da177e4SLinus Torvalds * 8381da177e4SLinus Torvalds * Returns true if all buffers were removed. 8391da177e4SLinus Torvalds */ 8401da177e4SLinus Torvalds int remove_inode_buffers(struct inode *inode) 8411da177e4SLinus Torvalds { 8421da177e4SLinus Torvalds int ret = 1; 8431da177e4SLinus Torvalds 8441da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8451da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 8461da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 847252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 8481da177e4SLinus Torvalds 8491da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8501da177e4SLinus Torvalds while (!list_empty(list)) { 8511da177e4SLinus Torvalds struct buffer_head *bh = BH_ENTRY(list->next); 8521da177e4SLinus Torvalds if (buffer_dirty(bh)) { 8531da177e4SLinus Torvalds ret = 0; 8541da177e4SLinus Torvalds break; 8551da177e4SLinus Torvalds } 8561da177e4SLinus Torvalds __remove_assoc_queue(bh); 8571da177e4SLinus Torvalds } 8581da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8591da177e4SLinus Torvalds } 8601da177e4SLinus Torvalds return ret; 8611da177e4SLinus Torvalds } 8621da177e4SLinus Torvalds 8631da177e4SLinus Torvalds /* 8641da177e4SLinus Torvalds * Create the appropriate buffers when given a page for data area and 8651da177e4SLinus Torvalds * the size of each buffer.. Use the bh->b_this_page linked list to 8661da177e4SLinus Torvalds * follow the buffers created. Return NULL if unable to create more 8671da177e4SLinus Torvalds * buffers. 8681da177e4SLinus Torvalds * 8691da177e4SLinus Torvalds * The retry flag is used to differentiate async IO (paging, swapping) 8701da177e4SLinus Torvalds * which may not fail from ordinary buffer allocations. 8711da177e4SLinus Torvalds */ 8721da177e4SLinus Torvalds struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, 8731da177e4SLinus Torvalds int retry) 8741da177e4SLinus Torvalds { 8751da177e4SLinus Torvalds struct buffer_head *bh, *head; 8761da177e4SLinus Torvalds long offset; 8771da177e4SLinus Torvalds 8781da177e4SLinus Torvalds try_again: 8791da177e4SLinus Torvalds head = NULL; 8801da177e4SLinus Torvalds offset = PAGE_SIZE; 8811da177e4SLinus Torvalds while ((offset -= size) >= 0) { 8821da177e4SLinus Torvalds bh = alloc_buffer_head(GFP_NOFS); 8831da177e4SLinus Torvalds if (!bh) 8841da177e4SLinus Torvalds goto no_grow; 8851da177e4SLinus Torvalds 8861da177e4SLinus Torvalds bh->b_this_page = head; 8871da177e4SLinus Torvalds bh->b_blocknr = -1; 8881da177e4SLinus Torvalds head = bh; 8891da177e4SLinus Torvalds 8901da177e4SLinus Torvalds bh->b_size = size; 8911da177e4SLinus Torvalds 8921da177e4SLinus Torvalds /* Link the buffer to its page */ 8931da177e4SLinus Torvalds set_bh_page(bh, page, offset); 8941da177e4SLinus Torvalds } 8951da177e4SLinus Torvalds return head; 8961da177e4SLinus Torvalds /* 8971da177e4SLinus Torvalds * In case anything failed, we just free everything we got. 8981da177e4SLinus Torvalds */ 8991da177e4SLinus Torvalds no_grow: 9001da177e4SLinus Torvalds if (head) { 9011da177e4SLinus Torvalds do { 9021da177e4SLinus Torvalds bh = head; 9031da177e4SLinus Torvalds head = head->b_this_page; 9041da177e4SLinus Torvalds free_buffer_head(bh); 9051da177e4SLinus Torvalds } while (head); 9061da177e4SLinus Torvalds } 9071da177e4SLinus Torvalds 9081da177e4SLinus Torvalds /* 9091da177e4SLinus Torvalds * Return failure for non-async IO requests. Async IO requests 9101da177e4SLinus Torvalds * are not allowed to fail, so we have to wait until buffer heads 9111da177e4SLinus Torvalds * become available. But we don't want tasks sleeping with 9121da177e4SLinus Torvalds * partially complete buffers, so all were released above. 9131da177e4SLinus Torvalds */ 9141da177e4SLinus Torvalds if (!retry) 9151da177e4SLinus Torvalds return NULL; 9161da177e4SLinus Torvalds 9171da177e4SLinus Torvalds /* We're _really_ low on memory. Now we just 9181da177e4SLinus Torvalds * wait for old buffer heads to become free due to 9191da177e4SLinus Torvalds * finishing IO. Since this is an async request and 9201da177e4SLinus Torvalds * the reserve list is empty, we're sure there are 9211da177e4SLinus Torvalds * async buffer heads in use. 9221da177e4SLinus Torvalds */ 9231da177e4SLinus Torvalds free_more_memory(); 9241da177e4SLinus Torvalds goto try_again; 9251da177e4SLinus Torvalds } 9261da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(alloc_page_buffers); 9271da177e4SLinus Torvalds 9281da177e4SLinus Torvalds static inline void 9291da177e4SLinus Torvalds link_dev_buffers(struct page *page, struct buffer_head *head) 9301da177e4SLinus Torvalds { 9311da177e4SLinus Torvalds struct buffer_head *bh, *tail; 9321da177e4SLinus Torvalds 9331da177e4SLinus Torvalds bh = head; 9341da177e4SLinus Torvalds do { 9351da177e4SLinus Torvalds tail = bh; 9361da177e4SLinus Torvalds bh = bh->b_this_page; 9371da177e4SLinus Torvalds } while (bh); 9381da177e4SLinus Torvalds tail->b_this_page = head; 9391da177e4SLinus Torvalds attach_page_buffers(page, head); 9401da177e4SLinus Torvalds } 9411da177e4SLinus Torvalds 942bbec0270SLinus Torvalds static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) 943bbec0270SLinus Torvalds { 944bbec0270SLinus Torvalds sector_t retval = ~((sector_t)0); 945bbec0270SLinus Torvalds loff_t sz = i_size_read(bdev->bd_inode); 946bbec0270SLinus Torvalds 947bbec0270SLinus Torvalds if (sz) { 948bbec0270SLinus Torvalds unsigned int sizebits = blksize_bits(size); 949bbec0270SLinus Torvalds retval = (sz >> sizebits); 950bbec0270SLinus Torvalds } 951bbec0270SLinus Torvalds return retval; 952bbec0270SLinus Torvalds } 953bbec0270SLinus Torvalds 9541da177e4SLinus Torvalds /* 9551da177e4SLinus Torvalds * Initialise the state of a blockdev page's buffers. 9561da177e4SLinus Torvalds */ 957676ce6d5SHugh Dickins static sector_t 9581da177e4SLinus Torvalds init_page_buffers(struct page *page, struct block_device *bdev, 9591da177e4SLinus Torvalds sector_t block, int size) 9601da177e4SLinus Torvalds { 9611da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 9621da177e4SLinus Torvalds struct buffer_head *bh = head; 9631da177e4SLinus Torvalds int uptodate = PageUptodate(page); 964bbec0270SLinus Torvalds sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size); 9651da177e4SLinus Torvalds 9661da177e4SLinus Torvalds do { 9671da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 9681da177e4SLinus Torvalds init_buffer(bh, NULL, NULL); 9691da177e4SLinus Torvalds bh->b_bdev = bdev; 9701da177e4SLinus Torvalds bh->b_blocknr = block; 9711da177e4SLinus Torvalds if (uptodate) 9721da177e4SLinus Torvalds set_buffer_uptodate(bh); 973080399aaSJeff Moyer if (block < end_block) 9741da177e4SLinus Torvalds set_buffer_mapped(bh); 9751da177e4SLinus Torvalds } 9761da177e4SLinus Torvalds block++; 9771da177e4SLinus Torvalds bh = bh->b_this_page; 9781da177e4SLinus Torvalds } while (bh != head); 979676ce6d5SHugh Dickins 980676ce6d5SHugh Dickins /* 981676ce6d5SHugh Dickins * Caller needs to validate requested block against end of device. 982676ce6d5SHugh Dickins */ 983676ce6d5SHugh Dickins return end_block; 9841da177e4SLinus Torvalds } 9851da177e4SLinus Torvalds 9861da177e4SLinus Torvalds /* 9871da177e4SLinus Torvalds * Create the page-cache page that contains the requested block. 9881da177e4SLinus Torvalds * 989676ce6d5SHugh Dickins * This is used purely for blockdev mappings. 9901da177e4SLinus Torvalds */ 991676ce6d5SHugh Dickins static int 9921da177e4SLinus Torvalds grow_dev_page(struct block_device *bdev, sector_t block, 9933b5e6454SGioh Kim pgoff_t index, int size, int sizebits, gfp_t gfp) 9941da177e4SLinus Torvalds { 9951da177e4SLinus Torvalds struct inode *inode = bdev->bd_inode; 9961da177e4SLinus Torvalds struct page *page; 9971da177e4SLinus Torvalds struct buffer_head *bh; 998676ce6d5SHugh Dickins sector_t end_block; 999676ce6d5SHugh Dickins int ret = 0; /* Will call free_more_memory() */ 100084235de3SJohannes Weiner gfp_t gfp_mask; 10011da177e4SLinus Torvalds 10023b5e6454SGioh Kim gfp_mask = (mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS) | gfp; 10033b5e6454SGioh Kim 100484235de3SJohannes Weiner /* 100584235de3SJohannes Weiner * XXX: __getblk_slow() can not really deal with failure and 100684235de3SJohannes Weiner * will endlessly loop on improvised global reclaim. Prefer 100784235de3SJohannes Weiner * looping in the allocator rather than here, at least that 100884235de3SJohannes Weiner * code knows what it's doing. 100984235de3SJohannes Weiner */ 101084235de3SJohannes Weiner gfp_mask |= __GFP_NOFAIL; 101184235de3SJohannes Weiner 101284235de3SJohannes Weiner page = find_or_create_page(inode->i_mapping, index, gfp_mask); 10131da177e4SLinus Torvalds if (!page) 1014676ce6d5SHugh Dickins return ret; 10151da177e4SLinus Torvalds 1016e827f923SEric Sesterhenn BUG_ON(!PageLocked(page)); 10171da177e4SLinus Torvalds 10181da177e4SLinus Torvalds if (page_has_buffers(page)) { 10191da177e4SLinus Torvalds bh = page_buffers(page); 10201da177e4SLinus Torvalds if (bh->b_size == size) { 1021676ce6d5SHugh Dickins end_block = init_page_buffers(page, bdev, 1022f2d5a944SAnton Altaparmakov (sector_t)index << sizebits, 1023f2d5a944SAnton Altaparmakov size); 1024676ce6d5SHugh Dickins goto done; 10251da177e4SLinus Torvalds } 10261da177e4SLinus Torvalds if (!try_to_free_buffers(page)) 10271da177e4SLinus Torvalds goto failed; 10281da177e4SLinus Torvalds } 10291da177e4SLinus Torvalds 10301da177e4SLinus Torvalds /* 10311da177e4SLinus Torvalds * Allocate some buffers for this page 10321da177e4SLinus Torvalds */ 10331da177e4SLinus Torvalds bh = alloc_page_buffers(page, size, 0); 10341da177e4SLinus Torvalds if (!bh) 10351da177e4SLinus Torvalds goto failed; 10361da177e4SLinus Torvalds 10371da177e4SLinus Torvalds /* 10381da177e4SLinus Torvalds * Link the page to the buffers and initialise them. Take the 10391da177e4SLinus Torvalds * lock to be atomic wrt __find_get_block(), which does not 10401da177e4SLinus Torvalds * run under the page lock. 10411da177e4SLinus Torvalds */ 10421da177e4SLinus Torvalds spin_lock(&inode->i_mapping->private_lock); 10431da177e4SLinus Torvalds link_dev_buffers(page, bh); 1044f2d5a944SAnton Altaparmakov end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits, 1045f2d5a944SAnton Altaparmakov size); 10461da177e4SLinus Torvalds spin_unlock(&inode->i_mapping->private_lock); 1047676ce6d5SHugh Dickins done: 1048676ce6d5SHugh Dickins ret = (block < end_block) ? 1 : -ENXIO; 10491da177e4SLinus Torvalds failed: 10501da177e4SLinus Torvalds unlock_page(page); 10511da177e4SLinus Torvalds page_cache_release(page); 1052676ce6d5SHugh Dickins return ret; 10531da177e4SLinus Torvalds } 10541da177e4SLinus Torvalds 10551da177e4SLinus Torvalds /* 10561da177e4SLinus Torvalds * Create buffers for the specified block device block's page. If 10571da177e4SLinus Torvalds * that page was dirty, the buffers are set dirty also. 10581da177e4SLinus Torvalds */ 1059858119e1SArjan van de Ven static int 10603b5e6454SGioh Kim grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp) 10611da177e4SLinus Torvalds { 10621da177e4SLinus Torvalds pgoff_t index; 10631da177e4SLinus Torvalds int sizebits; 10641da177e4SLinus Torvalds 10651da177e4SLinus Torvalds sizebits = -1; 10661da177e4SLinus Torvalds do { 10671da177e4SLinus Torvalds sizebits++; 10681da177e4SLinus Torvalds } while ((size << sizebits) < PAGE_SIZE); 10691da177e4SLinus Torvalds 10701da177e4SLinus Torvalds index = block >> sizebits; 10711da177e4SLinus Torvalds 1072e5657933SAndrew Morton /* 1073e5657933SAndrew Morton * Check for a block which wants to lie outside our maximum possible 1074e5657933SAndrew Morton * pagecache index. (this comparison is done using sector_t types). 1075e5657933SAndrew Morton */ 1076e5657933SAndrew Morton if (unlikely(index != block >> sizebits)) { 1077e5657933SAndrew Morton char b[BDEVNAME_SIZE]; 1078e5657933SAndrew Morton 1079e5657933SAndrew Morton printk(KERN_ERR "%s: requested out-of-range block %llu for " 1080e5657933SAndrew Morton "device %s\n", 10818e24eea7SHarvey Harrison __func__, (unsigned long long)block, 1082e5657933SAndrew Morton bdevname(bdev, b)); 1083e5657933SAndrew Morton return -EIO; 1084e5657933SAndrew Morton } 1085676ce6d5SHugh Dickins 10861da177e4SLinus Torvalds /* Create a page with the proper size buffers.. */ 10873b5e6454SGioh Kim return grow_dev_page(bdev, block, index, size, sizebits, gfp); 10881da177e4SLinus Torvalds } 10891da177e4SLinus Torvalds 10903b5e6454SGioh Kim struct buffer_head * 10913b5e6454SGioh Kim __getblk_slow(struct block_device *bdev, sector_t block, 10923b5e6454SGioh Kim unsigned size, gfp_t gfp) 10931da177e4SLinus Torvalds { 10941da177e4SLinus Torvalds /* Size must be multiple of hard sectorsize */ 1095e1defc4fSMartin K. Petersen if (unlikely(size & (bdev_logical_block_size(bdev)-1) || 10961da177e4SLinus Torvalds (size < 512 || size > PAGE_SIZE))) { 10971da177e4SLinus Torvalds printk(KERN_ERR "getblk(): invalid block size %d requested\n", 10981da177e4SLinus Torvalds size); 1099e1defc4fSMartin K. Petersen printk(KERN_ERR "logical block size: %d\n", 1100e1defc4fSMartin K. Petersen bdev_logical_block_size(bdev)); 11011da177e4SLinus Torvalds 11021da177e4SLinus Torvalds dump_stack(); 11031da177e4SLinus Torvalds return NULL; 11041da177e4SLinus Torvalds } 11051da177e4SLinus Torvalds 1106676ce6d5SHugh Dickins for (;;) { 1107676ce6d5SHugh Dickins struct buffer_head *bh; 1108676ce6d5SHugh Dickins int ret; 1109676ce6d5SHugh Dickins 11101da177e4SLinus Torvalds bh = __find_get_block(bdev, block, size); 11111da177e4SLinus Torvalds if (bh) 11121da177e4SLinus Torvalds return bh; 11131da177e4SLinus Torvalds 11143b5e6454SGioh Kim ret = grow_buffers(bdev, block, size, gfp); 1115676ce6d5SHugh Dickins if (ret < 0) 111691f68c89SJeff Moyer return NULL; 1117676ce6d5SHugh Dickins if (ret == 0) 1118676ce6d5SHugh Dickins free_more_memory(); 1119676ce6d5SHugh Dickins } 11201da177e4SLinus Torvalds } 11213b5e6454SGioh Kim EXPORT_SYMBOL(__getblk_slow); 11221da177e4SLinus Torvalds 11231da177e4SLinus Torvalds /* 11241da177e4SLinus Torvalds * The relationship between dirty buffers and dirty pages: 11251da177e4SLinus Torvalds * 11261da177e4SLinus Torvalds * Whenever a page has any dirty buffers, the page's dirty bit is set, and 11271da177e4SLinus Torvalds * the page is tagged dirty in its radix tree. 11281da177e4SLinus Torvalds * 11291da177e4SLinus Torvalds * At all times, the dirtiness of the buffers represents the dirtiness of 11301da177e4SLinus Torvalds * subsections of the page. If the page has buffers, the page dirty bit is 11311da177e4SLinus Torvalds * merely a hint about the true dirty state. 11321da177e4SLinus Torvalds * 11331da177e4SLinus Torvalds * When a page is set dirty in its entirety, all its buffers are marked dirty 11341da177e4SLinus Torvalds * (if the page has buffers). 11351da177e4SLinus Torvalds * 11361da177e4SLinus Torvalds * When a buffer is marked dirty, its page is dirtied, but the page's other 11371da177e4SLinus Torvalds * buffers are not. 11381da177e4SLinus Torvalds * 11391da177e4SLinus Torvalds * Also. When blockdev buffers are explicitly read with bread(), they 11401da177e4SLinus Torvalds * individually become uptodate. But their backing page remains not 11411da177e4SLinus Torvalds * uptodate - even if all of its buffers are uptodate. A subsequent 11421da177e4SLinus Torvalds * block_read_full_page() against that page will discover all the uptodate 11431da177e4SLinus Torvalds * buffers, will set the page uptodate and will perform no I/O. 11441da177e4SLinus Torvalds */ 11451da177e4SLinus Torvalds 11461da177e4SLinus Torvalds /** 11471da177e4SLinus Torvalds * mark_buffer_dirty - mark a buffer_head as needing writeout 114867be2dd1SMartin Waitz * @bh: the buffer_head to mark dirty 11491da177e4SLinus Torvalds * 11501da177e4SLinus Torvalds * mark_buffer_dirty() will set the dirty bit against the buffer, then set its 11511da177e4SLinus Torvalds * backing page dirty, then tag the page as dirty in its address_space's radix 11521da177e4SLinus Torvalds * tree and then attach the address_space's inode to its superblock's dirty 11531da177e4SLinus Torvalds * inode list. 11541da177e4SLinus Torvalds * 11551da177e4SLinus Torvalds * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock, 1156250df6edSDave Chinner * mapping->tree_lock and mapping->host->i_lock. 11571da177e4SLinus Torvalds */ 1158fc9b52cdSHarvey Harrison void mark_buffer_dirty(struct buffer_head *bh) 11591da177e4SLinus Torvalds { 1160787d2214SNick Piggin WARN_ON_ONCE(!buffer_uptodate(bh)); 11611be62dc1SLinus Torvalds 11625305cb83STejun Heo trace_block_dirty_buffer(bh); 11635305cb83STejun Heo 11641be62dc1SLinus Torvalds /* 11651be62dc1SLinus Torvalds * Very *carefully* optimize the it-is-already-dirty case. 11661be62dc1SLinus Torvalds * 11671be62dc1SLinus Torvalds * Don't let the final "is it dirty" escape to before we 11681be62dc1SLinus Torvalds * perhaps modified the buffer. 11691be62dc1SLinus Torvalds */ 11701be62dc1SLinus Torvalds if (buffer_dirty(bh)) { 11711be62dc1SLinus Torvalds smp_mb(); 11721be62dc1SLinus Torvalds if (buffer_dirty(bh)) 11731be62dc1SLinus Torvalds return; 11741be62dc1SLinus Torvalds } 11751be62dc1SLinus Torvalds 1176a8e7d49aSLinus Torvalds if (!test_set_buffer_dirty(bh)) { 1177a8e7d49aSLinus Torvalds struct page *page = bh->b_page; 1178c4843a75SGreg Thelen struct address_space *mapping = NULL; 1179c4843a75SGreg Thelen struct mem_cgroup *memcg; 1180c4843a75SGreg Thelen 1181c4843a75SGreg Thelen memcg = mem_cgroup_begin_page_stat(page); 11828e9d78edSLinus Torvalds if (!TestSetPageDirty(page)) { 1183c4843a75SGreg Thelen mapping = page_mapping(page); 11848e9d78edSLinus Torvalds if (mapping) 1185c4843a75SGreg Thelen __set_page_dirty(page, mapping, memcg, 0); 11868e9d78edSLinus Torvalds } 1187c4843a75SGreg Thelen mem_cgroup_end_page_stat(memcg); 1188c4843a75SGreg Thelen if (mapping) 1189c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 1190a8e7d49aSLinus Torvalds } 11911da177e4SLinus Torvalds } 11921fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(mark_buffer_dirty); 11931da177e4SLinus Torvalds 11941da177e4SLinus Torvalds /* 11951da177e4SLinus Torvalds * Decrement a buffer_head's reference count. If all buffers against a page 11961da177e4SLinus Torvalds * have zero reference count, are clean and unlocked, and if the page is clean 11971da177e4SLinus Torvalds * and unlocked then try_to_free_buffers() may strip the buffers from the page 11981da177e4SLinus Torvalds * in preparation for freeing it (sometimes, rarely, buffers are removed from 11991da177e4SLinus Torvalds * a page but it ends up not being freed, and buffers may later be reattached). 12001da177e4SLinus Torvalds */ 12011da177e4SLinus Torvalds void __brelse(struct buffer_head * buf) 12021da177e4SLinus Torvalds { 12031da177e4SLinus Torvalds if (atomic_read(&buf->b_count)) { 12041da177e4SLinus Torvalds put_bh(buf); 12051da177e4SLinus Torvalds return; 12061da177e4SLinus Torvalds } 12075c752ad9SArjan van de Ven WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); 12081da177e4SLinus Torvalds } 12091fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__brelse); 12101da177e4SLinus Torvalds 12111da177e4SLinus Torvalds /* 12121da177e4SLinus Torvalds * bforget() is like brelse(), except it discards any 12131da177e4SLinus Torvalds * potentially dirty data. 12141da177e4SLinus Torvalds */ 12151da177e4SLinus Torvalds void __bforget(struct buffer_head *bh) 12161da177e4SLinus Torvalds { 12171da177e4SLinus Torvalds clear_buffer_dirty(bh); 1218535ee2fbSJan Kara if (bh->b_assoc_map) { 12191da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 12201da177e4SLinus Torvalds 12211da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 12221da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 122358ff407bSJan Kara bh->b_assoc_map = NULL; 12241da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 12251da177e4SLinus Torvalds } 12261da177e4SLinus Torvalds __brelse(bh); 12271da177e4SLinus Torvalds } 12281fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__bforget); 12291da177e4SLinus Torvalds 12301da177e4SLinus Torvalds static struct buffer_head *__bread_slow(struct buffer_head *bh) 12311da177e4SLinus Torvalds { 12321da177e4SLinus Torvalds lock_buffer(bh); 12331da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 12341da177e4SLinus Torvalds unlock_buffer(bh); 12351da177e4SLinus Torvalds return bh; 12361da177e4SLinus Torvalds } else { 12371da177e4SLinus Torvalds get_bh(bh); 12381da177e4SLinus Torvalds bh->b_end_io = end_buffer_read_sync; 12391da177e4SLinus Torvalds submit_bh(READ, bh); 12401da177e4SLinus Torvalds wait_on_buffer(bh); 12411da177e4SLinus Torvalds if (buffer_uptodate(bh)) 12421da177e4SLinus Torvalds return bh; 12431da177e4SLinus Torvalds } 12441da177e4SLinus Torvalds brelse(bh); 12451da177e4SLinus Torvalds return NULL; 12461da177e4SLinus Torvalds } 12471da177e4SLinus Torvalds 12481da177e4SLinus Torvalds /* 12491da177e4SLinus Torvalds * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). 12501da177e4SLinus Torvalds * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their 12511da177e4SLinus Torvalds * refcount elevated by one when they're in an LRU. A buffer can only appear 12521da177e4SLinus Torvalds * once in a particular CPU's LRU. A single buffer can be present in multiple 12531da177e4SLinus Torvalds * CPU's LRUs at the same time. 12541da177e4SLinus Torvalds * 12551da177e4SLinus Torvalds * This is a transparent caching front-end to sb_bread(), sb_getblk() and 12561da177e4SLinus Torvalds * sb_find_get_block(). 12571da177e4SLinus Torvalds * 12581da177e4SLinus Torvalds * The LRUs themselves only need locking against invalidate_bh_lrus. We use 12591da177e4SLinus Torvalds * a local interrupt disable for that. 12601da177e4SLinus Torvalds */ 12611da177e4SLinus Torvalds 126286cf78d7SSebastien Buisson #define BH_LRU_SIZE 16 12631da177e4SLinus Torvalds 12641da177e4SLinus Torvalds struct bh_lru { 12651da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12661da177e4SLinus Torvalds }; 12671da177e4SLinus Torvalds 12681da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; 12691da177e4SLinus Torvalds 12701da177e4SLinus Torvalds #ifdef CONFIG_SMP 12711da177e4SLinus Torvalds #define bh_lru_lock() local_irq_disable() 12721da177e4SLinus Torvalds #define bh_lru_unlock() local_irq_enable() 12731da177e4SLinus Torvalds #else 12741da177e4SLinus Torvalds #define bh_lru_lock() preempt_disable() 12751da177e4SLinus Torvalds #define bh_lru_unlock() preempt_enable() 12761da177e4SLinus Torvalds #endif 12771da177e4SLinus Torvalds 12781da177e4SLinus Torvalds static inline void check_irqs_on(void) 12791da177e4SLinus Torvalds { 12801da177e4SLinus Torvalds #ifdef irqs_disabled 12811da177e4SLinus Torvalds BUG_ON(irqs_disabled()); 12821da177e4SLinus Torvalds #endif 12831da177e4SLinus Torvalds } 12841da177e4SLinus Torvalds 12851da177e4SLinus Torvalds /* 12861da177e4SLinus Torvalds * The LRU management algorithm is dopey-but-simple. Sorry. 12871da177e4SLinus Torvalds */ 12881da177e4SLinus Torvalds static void bh_lru_install(struct buffer_head *bh) 12891da177e4SLinus Torvalds { 12901da177e4SLinus Torvalds struct buffer_head *evictee = NULL; 12911da177e4SLinus Torvalds 12921da177e4SLinus Torvalds check_irqs_on(); 12931da177e4SLinus Torvalds bh_lru_lock(); 1294c7b92516SChristoph Lameter if (__this_cpu_read(bh_lrus.bhs[0]) != bh) { 12951da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12961da177e4SLinus Torvalds int in; 12971da177e4SLinus Torvalds int out = 0; 12981da177e4SLinus Torvalds 12991da177e4SLinus Torvalds get_bh(bh); 13001da177e4SLinus Torvalds bhs[out++] = bh; 13011da177e4SLinus Torvalds for (in = 0; in < BH_LRU_SIZE; in++) { 1302c7b92516SChristoph Lameter struct buffer_head *bh2 = 1303c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[in]); 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds if (bh2 == bh) { 13061da177e4SLinus Torvalds __brelse(bh2); 13071da177e4SLinus Torvalds } else { 13081da177e4SLinus Torvalds if (out >= BH_LRU_SIZE) { 13091da177e4SLinus Torvalds BUG_ON(evictee != NULL); 13101da177e4SLinus Torvalds evictee = bh2; 13111da177e4SLinus Torvalds } else { 13121da177e4SLinus Torvalds bhs[out++] = bh2; 13131da177e4SLinus Torvalds } 13141da177e4SLinus Torvalds } 13151da177e4SLinus Torvalds } 13161da177e4SLinus Torvalds while (out < BH_LRU_SIZE) 13171da177e4SLinus Torvalds bhs[out++] = NULL; 1318ca6673b0SChristoph Lameter memcpy(this_cpu_ptr(&bh_lrus.bhs), bhs, sizeof(bhs)); 13191da177e4SLinus Torvalds } 13201da177e4SLinus Torvalds bh_lru_unlock(); 13211da177e4SLinus Torvalds 13221da177e4SLinus Torvalds if (evictee) 13231da177e4SLinus Torvalds __brelse(evictee); 13241da177e4SLinus Torvalds } 13251da177e4SLinus Torvalds 13261da177e4SLinus Torvalds /* 13271da177e4SLinus Torvalds * Look up the bh in this cpu's LRU. If it's there, move it to the head. 13281da177e4SLinus Torvalds */ 1329858119e1SArjan van de Ven static struct buffer_head * 13303991d3bdSTomasz Kvarsin lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) 13311da177e4SLinus Torvalds { 13321da177e4SLinus Torvalds struct buffer_head *ret = NULL; 13333991d3bdSTomasz Kvarsin unsigned int i; 13341da177e4SLinus Torvalds 13351da177e4SLinus Torvalds check_irqs_on(); 13361da177e4SLinus Torvalds bh_lru_lock(); 13371da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 1338c7b92516SChristoph Lameter struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); 13391da177e4SLinus Torvalds 13409470dd5dSZach Brown if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && 13419470dd5dSZach Brown bh->b_size == size) { 13421da177e4SLinus Torvalds if (i) { 13431da177e4SLinus Torvalds while (i) { 1344c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[i], 1345c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[i - 1])); 13461da177e4SLinus Torvalds i--; 13471da177e4SLinus Torvalds } 1348c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[0], bh); 13491da177e4SLinus Torvalds } 13501da177e4SLinus Torvalds get_bh(bh); 13511da177e4SLinus Torvalds ret = bh; 13521da177e4SLinus Torvalds break; 13531da177e4SLinus Torvalds } 13541da177e4SLinus Torvalds } 13551da177e4SLinus Torvalds bh_lru_unlock(); 13561da177e4SLinus Torvalds return ret; 13571da177e4SLinus Torvalds } 13581da177e4SLinus Torvalds 13591da177e4SLinus Torvalds /* 13601da177e4SLinus Torvalds * Perform a pagecache lookup for the matching buffer. If it's there, refresh 13611da177e4SLinus Torvalds * it in the LRU and mark it as accessed. If it is not present then return 13621da177e4SLinus Torvalds * NULL 13631da177e4SLinus Torvalds */ 13641da177e4SLinus Torvalds struct buffer_head * 13653991d3bdSTomasz Kvarsin __find_get_block(struct block_device *bdev, sector_t block, unsigned size) 13661da177e4SLinus Torvalds { 13671da177e4SLinus Torvalds struct buffer_head *bh = lookup_bh_lru(bdev, block, size); 13681da177e4SLinus Torvalds 13691da177e4SLinus Torvalds if (bh == NULL) { 13702457aec6SMel Gorman /* __find_get_block_slow will mark the page accessed */ 1371385fd4c5SCoywolf Qi Hunt bh = __find_get_block_slow(bdev, block); 13721da177e4SLinus Torvalds if (bh) 13731da177e4SLinus Torvalds bh_lru_install(bh); 13742457aec6SMel Gorman } else 13751da177e4SLinus Torvalds touch_buffer(bh); 13762457aec6SMel Gorman 13771da177e4SLinus Torvalds return bh; 13781da177e4SLinus Torvalds } 13791da177e4SLinus Torvalds EXPORT_SYMBOL(__find_get_block); 13801da177e4SLinus Torvalds 13811da177e4SLinus Torvalds /* 13823b5e6454SGioh Kim * __getblk_gfp() will locate (and, if necessary, create) the buffer_head 13831da177e4SLinus Torvalds * which corresponds to the passed block_device, block and size. The 13841da177e4SLinus Torvalds * returned buffer has its reference count incremented. 13851da177e4SLinus Torvalds * 13863b5e6454SGioh Kim * __getblk_gfp() will lock up the machine if grow_dev_page's 13873b5e6454SGioh Kim * try_to_free_buffers() attempt is failing. FIXME, perhaps? 13881da177e4SLinus Torvalds */ 13891da177e4SLinus Torvalds struct buffer_head * 13903b5e6454SGioh Kim __getblk_gfp(struct block_device *bdev, sector_t block, 13913b5e6454SGioh Kim unsigned size, gfp_t gfp) 13921da177e4SLinus Torvalds { 13931da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, block, size); 13941da177e4SLinus Torvalds 13951da177e4SLinus Torvalds might_sleep(); 13961da177e4SLinus Torvalds if (bh == NULL) 13973b5e6454SGioh Kim bh = __getblk_slow(bdev, block, size, gfp); 13981da177e4SLinus Torvalds return bh; 13991da177e4SLinus Torvalds } 14003b5e6454SGioh Kim EXPORT_SYMBOL(__getblk_gfp); 14011da177e4SLinus Torvalds 14021da177e4SLinus Torvalds /* 14031da177e4SLinus Torvalds * Do async read-ahead on a buffer.. 14041da177e4SLinus Torvalds */ 14053991d3bdSTomasz Kvarsin void __breadahead(struct block_device *bdev, sector_t block, unsigned size) 14061da177e4SLinus Torvalds { 14071da177e4SLinus Torvalds struct buffer_head *bh = __getblk(bdev, block, size); 1408a3e713b5SAndrew Morton if (likely(bh)) { 14091da177e4SLinus Torvalds ll_rw_block(READA, 1, &bh); 14101da177e4SLinus Torvalds brelse(bh); 14111da177e4SLinus Torvalds } 1412a3e713b5SAndrew Morton } 14131da177e4SLinus Torvalds EXPORT_SYMBOL(__breadahead); 14141da177e4SLinus Torvalds 14151da177e4SLinus Torvalds /** 14163b5e6454SGioh Kim * __bread_gfp() - reads a specified block and returns the bh 141767be2dd1SMartin Waitz * @bdev: the block_device to read from 14181da177e4SLinus Torvalds * @block: number of block 14191da177e4SLinus Torvalds * @size: size (in bytes) to read 14203b5e6454SGioh Kim * @gfp: page allocation flag 14211da177e4SLinus Torvalds * 14221da177e4SLinus Torvalds * Reads a specified block, and returns buffer head that contains it. 14233b5e6454SGioh Kim * The page cache can be allocated from non-movable area 14243b5e6454SGioh Kim * not to prevent page migration if you set gfp to zero. 14251da177e4SLinus Torvalds * It returns NULL if the block was unreadable. 14261da177e4SLinus Torvalds */ 14271da177e4SLinus Torvalds struct buffer_head * 14283b5e6454SGioh Kim __bread_gfp(struct block_device *bdev, sector_t block, 14293b5e6454SGioh Kim unsigned size, gfp_t gfp) 14301da177e4SLinus Torvalds { 14313b5e6454SGioh Kim struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); 14321da177e4SLinus Torvalds 1433a3e713b5SAndrew Morton if (likely(bh) && !buffer_uptodate(bh)) 14341da177e4SLinus Torvalds bh = __bread_slow(bh); 14351da177e4SLinus Torvalds return bh; 14361da177e4SLinus Torvalds } 14373b5e6454SGioh Kim EXPORT_SYMBOL(__bread_gfp); 14381da177e4SLinus Torvalds 14391da177e4SLinus Torvalds /* 14401da177e4SLinus Torvalds * invalidate_bh_lrus() is called rarely - but not only at unmount. 14411da177e4SLinus Torvalds * This doesn't race because it runs in each cpu either in irq 14421da177e4SLinus Torvalds * or with preempt disabled. 14431da177e4SLinus Torvalds */ 14441da177e4SLinus Torvalds static void invalidate_bh_lru(void *arg) 14451da177e4SLinus Torvalds { 14461da177e4SLinus Torvalds struct bh_lru *b = &get_cpu_var(bh_lrus); 14471da177e4SLinus Torvalds int i; 14481da177e4SLinus Torvalds 14491da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 14501da177e4SLinus Torvalds brelse(b->bhs[i]); 14511da177e4SLinus Torvalds b->bhs[i] = NULL; 14521da177e4SLinus Torvalds } 14531da177e4SLinus Torvalds put_cpu_var(bh_lrus); 14541da177e4SLinus Torvalds } 14551da177e4SLinus Torvalds 145642be35d0SGilad Ben-Yossef static bool has_bh_in_lru(int cpu, void *dummy) 145742be35d0SGilad Ben-Yossef { 145842be35d0SGilad Ben-Yossef struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); 145942be35d0SGilad Ben-Yossef int i; 146042be35d0SGilad Ben-Yossef 146142be35d0SGilad Ben-Yossef for (i = 0; i < BH_LRU_SIZE; i++) { 146242be35d0SGilad Ben-Yossef if (b->bhs[i]) 146342be35d0SGilad Ben-Yossef return 1; 146442be35d0SGilad Ben-Yossef } 146542be35d0SGilad Ben-Yossef 146642be35d0SGilad Ben-Yossef return 0; 146742be35d0SGilad Ben-Yossef } 146842be35d0SGilad Ben-Yossef 1469f9a14399SPeter Zijlstra void invalidate_bh_lrus(void) 14701da177e4SLinus Torvalds { 147142be35d0SGilad Ben-Yossef on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL); 14721da177e4SLinus Torvalds } 14739db5579bSNick Piggin EXPORT_SYMBOL_GPL(invalidate_bh_lrus); 14741da177e4SLinus Torvalds 14751da177e4SLinus Torvalds void set_bh_page(struct buffer_head *bh, 14761da177e4SLinus Torvalds struct page *page, unsigned long offset) 14771da177e4SLinus Torvalds { 14781da177e4SLinus Torvalds bh->b_page = page; 1479e827f923SEric Sesterhenn BUG_ON(offset >= PAGE_SIZE); 14801da177e4SLinus Torvalds if (PageHighMem(page)) 14811da177e4SLinus Torvalds /* 14821da177e4SLinus Torvalds * This catches illegal uses and preserves the offset: 14831da177e4SLinus Torvalds */ 14841da177e4SLinus Torvalds bh->b_data = (char *)(0 + offset); 14851da177e4SLinus Torvalds else 14861da177e4SLinus Torvalds bh->b_data = page_address(page) + offset; 14871da177e4SLinus Torvalds } 14881da177e4SLinus Torvalds EXPORT_SYMBOL(set_bh_page); 14891da177e4SLinus Torvalds 14901da177e4SLinus Torvalds /* 14911da177e4SLinus Torvalds * Called when truncating a buffer on a page completely. 14921da177e4SLinus Torvalds */ 1493e7470ee8SMel Gorman 1494e7470ee8SMel Gorman /* Bits that are cleared during an invalidate */ 1495e7470ee8SMel Gorman #define BUFFER_FLAGS_DISCARD \ 1496e7470ee8SMel Gorman (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ 1497e7470ee8SMel Gorman 1 << BH_Delay | 1 << BH_Unwritten) 1498e7470ee8SMel Gorman 1499858119e1SArjan van de Ven static void discard_buffer(struct buffer_head * bh) 15001da177e4SLinus Torvalds { 1501e7470ee8SMel Gorman unsigned long b_state, b_state_old; 1502e7470ee8SMel Gorman 15031da177e4SLinus Torvalds lock_buffer(bh); 15041da177e4SLinus Torvalds clear_buffer_dirty(bh); 15051da177e4SLinus Torvalds bh->b_bdev = NULL; 1506e7470ee8SMel Gorman b_state = bh->b_state; 1507e7470ee8SMel Gorman for (;;) { 1508e7470ee8SMel Gorman b_state_old = cmpxchg(&bh->b_state, b_state, 1509e7470ee8SMel Gorman (b_state & ~BUFFER_FLAGS_DISCARD)); 1510e7470ee8SMel Gorman if (b_state_old == b_state) 1511e7470ee8SMel Gorman break; 1512e7470ee8SMel Gorman b_state = b_state_old; 1513e7470ee8SMel Gorman } 15141da177e4SLinus Torvalds unlock_buffer(bh); 15151da177e4SLinus Torvalds } 15161da177e4SLinus Torvalds 15171da177e4SLinus Torvalds /** 1518814e1d25SWang Sheng-Hui * block_invalidatepage - invalidate part or all of a buffer-backed page 15191da177e4SLinus Torvalds * 15201da177e4SLinus Torvalds * @page: the page which is affected 1521d47992f8SLukas Czerner * @offset: start of the range to invalidate 1522d47992f8SLukas Czerner * @length: length of the range to invalidate 15231da177e4SLinus Torvalds * 15241da177e4SLinus Torvalds * block_invalidatepage() is called when all or part of the page has become 15251da177e4SLinus Torvalds * invalidated by a truncate operation. 15261da177e4SLinus Torvalds * 15271da177e4SLinus Torvalds * block_invalidatepage() does not have to release all buffers, but it must 15281da177e4SLinus Torvalds * ensure that no dirty buffer is left outside @offset and that no I/O 15291da177e4SLinus Torvalds * is underway against any of the blocks which are outside the truncation 15301da177e4SLinus Torvalds * point. Because the caller is about to free (and possibly reuse) those 15311da177e4SLinus Torvalds * blocks on-disk. 15321da177e4SLinus Torvalds */ 1533d47992f8SLukas Czerner void block_invalidatepage(struct page *page, unsigned int offset, 1534d47992f8SLukas Czerner unsigned int length) 15351da177e4SLinus Torvalds { 15361da177e4SLinus Torvalds struct buffer_head *head, *bh, *next; 15371da177e4SLinus Torvalds unsigned int curr_off = 0; 1538d47992f8SLukas Czerner unsigned int stop = length + offset; 15391da177e4SLinus Torvalds 15401da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 15411da177e4SLinus Torvalds if (!page_has_buffers(page)) 15421da177e4SLinus Torvalds goto out; 15431da177e4SLinus Torvalds 1544d47992f8SLukas Czerner /* 1545d47992f8SLukas Czerner * Check for overflow 1546d47992f8SLukas Czerner */ 1547d47992f8SLukas Czerner BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); 1548d47992f8SLukas Czerner 15491da177e4SLinus Torvalds head = page_buffers(page); 15501da177e4SLinus Torvalds bh = head; 15511da177e4SLinus Torvalds do { 15521da177e4SLinus Torvalds unsigned int next_off = curr_off + bh->b_size; 15531da177e4SLinus Torvalds next = bh->b_this_page; 15541da177e4SLinus Torvalds 15551da177e4SLinus Torvalds /* 1556d47992f8SLukas Czerner * Are we still fully in range ? 1557d47992f8SLukas Czerner */ 1558d47992f8SLukas Czerner if (next_off > stop) 1559d47992f8SLukas Czerner goto out; 1560d47992f8SLukas Czerner 1561d47992f8SLukas Czerner /* 15621da177e4SLinus Torvalds * is this block fully invalidated? 15631da177e4SLinus Torvalds */ 15641da177e4SLinus Torvalds if (offset <= curr_off) 15651da177e4SLinus Torvalds discard_buffer(bh); 15661da177e4SLinus Torvalds curr_off = next_off; 15671da177e4SLinus Torvalds bh = next; 15681da177e4SLinus Torvalds } while (bh != head); 15691da177e4SLinus Torvalds 15701da177e4SLinus Torvalds /* 15711da177e4SLinus Torvalds * We release buffers only if the entire page is being invalidated. 15721da177e4SLinus Torvalds * The get_block cached value has been unconditionally invalidated, 15731da177e4SLinus Torvalds * so real IO is not possible anymore. 15741da177e4SLinus Torvalds */ 15751da177e4SLinus Torvalds if (offset == 0) 15762ff28e22SNeilBrown try_to_release_page(page, 0); 15771da177e4SLinus Torvalds out: 15782ff28e22SNeilBrown return; 15791da177e4SLinus Torvalds } 15801da177e4SLinus Torvalds EXPORT_SYMBOL(block_invalidatepage); 15811da177e4SLinus Torvalds 1582d47992f8SLukas Czerner 15831da177e4SLinus Torvalds /* 15841da177e4SLinus Torvalds * We attach and possibly dirty the buffers atomically wrt 15851da177e4SLinus Torvalds * __set_page_dirty_buffers() via private_lock. try_to_free_buffers 15861da177e4SLinus Torvalds * is already excluded via the page lock. 15871da177e4SLinus Torvalds */ 15881da177e4SLinus Torvalds void create_empty_buffers(struct page *page, 15891da177e4SLinus Torvalds unsigned long blocksize, unsigned long b_state) 15901da177e4SLinus Torvalds { 15911da177e4SLinus Torvalds struct buffer_head *bh, *head, *tail; 15921da177e4SLinus Torvalds 15931da177e4SLinus Torvalds head = alloc_page_buffers(page, blocksize, 1); 15941da177e4SLinus Torvalds bh = head; 15951da177e4SLinus Torvalds do { 15961da177e4SLinus Torvalds bh->b_state |= b_state; 15971da177e4SLinus Torvalds tail = bh; 15981da177e4SLinus Torvalds bh = bh->b_this_page; 15991da177e4SLinus Torvalds } while (bh); 16001da177e4SLinus Torvalds tail->b_this_page = head; 16011da177e4SLinus Torvalds 16021da177e4SLinus Torvalds spin_lock(&page->mapping->private_lock); 16031da177e4SLinus Torvalds if (PageUptodate(page) || PageDirty(page)) { 16041da177e4SLinus Torvalds bh = head; 16051da177e4SLinus Torvalds do { 16061da177e4SLinus Torvalds if (PageDirty(page)) 16071da177e4SLinus Torvalds set_buffer_dirty(bh); 16081da177e4SLinus Torvalds if (PageUptodate(page)) 16091da177e4SLinus Torvalds set_buffer_uptodate(bh); 16101da177e4SLinus Torvalds bh = bh->b_this_page; 16111da177e4SLinus Torvalds } while (bh != head); 16121da177e4SLinus Torvalds } 16131da177e4SLinus Torvalds attach_page_buffers(page, head); 16141da177e4SLinus Torvalds spin_unlock(&page->mapping->private_lock); 16151da177e4SLinus Torvalds } 16161da177e4SLinus Torvalds EXPORT_SYMBOL(create_empty_buffers); 16171da177e4SLinus Torvalds 16181da177e4SLinus Torvalds /* 16191da177e4SLinus Torvalds * We are taking a block for data and we don't want any output from any 16201da177e4SLinus Torvalds * buffer-cache aliases starting from return from that function and 16211da177e4SLinus Torvalds * until the moment when something will explicitly mark the buffer 16221da177e4SLinus Torvalds * dirty (hopefully that will not happen until we will free that block ;-) 16231da177e4SLinus Torvalds * We don't even need to mark it not-uptodate - nobody can expect 16241da177e4SLinus Torvalds * anything from a newly allocated buffer anyway. We used to used 16251da177e4SLinus Torvalds * unmap_buffer() for such invalidation, but that was wrong. We definitely 16261da177e4SLinus Torvalds * don't want to mark the alias unmapped, for example - it would confuse 16271da177e4SLinus Torvalds * anyone who might pick it with bread() afterwards... 16281da177e4SLinus Torvalds * 16291da177e4SLinus Torvalds * Also.. Note that bforget() doesn't lock the buffer. So there can 16301da177e4SLinus Torvalds * be writeout I/O going on against recently-freed buffers. We don't 16311da177e4SLinus Torvalds * wait on that I/O in bforget() - it's more efficient to wait on the I/O 16321da177e4SLinus Torvalds * only if we really need to. That happens here. 16331da177e4SLinus Torvalds */ 16341da177e4SLinus Torvalds void unmap_underlying_metadata(struct block_device *bdev, sector_t block) 16351da177e4SLinus Torvalds { 16361da177e4SLinus Torvalds struct buffer_head *old_bh; 16371da177e4SLinus Torvalds 16381da177e4SLinus Torvalds might_sleep(); 16391da177e4SLinus Torvalds 1640385fd4c5SCoywolf Qi Hunt old_bh = __find_get_block_slow(bdev, block); 16411da177e4SLinus Torvalds if (old_bh) { 16421da177e4SLinus Torvalds clear_buffer_dirty(old_bh); 16431da177e4SLinus Torvalds wait_on_buffer(old_bh); 16441da177e4SLinus Torvalds clear_buffer_req(old_bh); 16451da177e4SLinus Torvalds __brelse(old_bh); 16461da177e4SLinus Torvalds } 16471da177e4SLinus Torvalds } 16481da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_underlying_metadata); 16491da177e4SLinus Torvalds 16501da177e4SLinus Torvalds /* 165145bce8f3SLinus Torvalds * Size is a power-of-two in the range 512..PAGE_SIZE, 165245bce8f3SLinus Torvalds * and the case we care about most is PAGE_SIZE. 165345bce8f3SLinus Torvalds * 165445bce8f3SLinus Torvalds * So this *could* possibly be written with those 165545bce8f3SLinus Torvalds * constraints in mind (relevant mostly if some 165645bce8f3SLinus Torvalds * architecture has a slow bit-scan instruction) 165745bce8f3SLinus Torvalds */ 165845bce8f3SLinus Torvalds static inline int block_size_bits(unsigned int blocksize) 165945bce8f3SLinus Torvalds { 166045bce8f3SLinus Torvalds return ilog2(blocksize); 166145bce8f3SLinus Torvalds } 166245bce8f3SLinus Torvalds 166345bce8f3SLinus Torvalds static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state) 166445bce8f3SLinus Torvalds { 166545bce8f3SLinus Torvalds BUG_ON(!PageLocked(page)); 166645bce8f3SLinus Torvalds 166745bce8f3SLinus Torvalds if (!page_has_buffers(page)) 166845bce8f3SLinus Torvalds create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state); 166945bce8f3SLinus Torvalds return page_buffers(page); 167045bce8f3SLinus Torvalds } 167145bce8f3SLinus Torvalds 167245bce8f3SLinus Torvalds /* 16731da177e4SLinus Torvalds * NOTE! All mapped/uptodate combinations are valid: 16741da177e4SLinus Torvalds * 16751da177e4SLinus Torvalds * Mapped Uptodate Meaning 16761da177e4SLinus Torvalds * 16771da177e4SLinus Torvalds * No No "unknown" - must do get_block() 16781da177e4SLinus Torvalds * No Yes "hole" - zero-filled 16791da177e4SLinus Torvalds * Yes No "allocated" - allocated on disk, not read in 16801da177e4SLinus Torvalds * Yes Yes "valid" - allocated and up-to-date in memory. 16811da177e4SLinus Torvalds * 16821da177e4SLinus Torvalds * "Dirty" is valid only with the last case (mapped+uptodate). 16831da177e4SLinus Torvalds */ 16841da177e4SLinus Torvalds 16851da177e4SLinus Torvalds /* 16861da177e4SLinus Torvalds * While block_write_full_page is writing back the dirty buffers under 16871da177e4SLinus Torvalds * the page lock, whoever dirtied the buffers may decide to clean them 16881da177e4SLinus Torvalds * again at any time. We handle that by only looking at the buffer 16891da177e4SLinus Torvalds * state inside lock_buffer(). 16901da177e4SLinus Torvalds * 16911da177e4SLinus Torvalds * If block_write_full_page() is called for regular writeback 16921da177e4SLinus Torvalds * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a 16931da177e4SLinus Torvalds * locked buffer. This only can happen if someone has written the buffer 16941da177e4SLinus Torvalds * directly, with submit_bh(). At the address_space level PageWriteback 16951da177e4SLinus Torvalds * prevents this contention from occurring. 16966e34eeddSTheodore Ts'o * 16976e34eeddSTheodore Ts'o * If block_write_full_page() is called with wbc->sync_mode == 1698721a9602SJens Axboe * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this 1699721a9602SJens Axboe * causes the writes to be flagged as synchronous writes. 17001da177e4SLinus Torvalds */ 17011da177e4SLinus Torvalds static int __block_write_full_page(struct inode *inode, struct page *page, 170235c80d5fSChris Mason get_block_t *get_block, struct writeback_control *wbc, 170335c80d5fSChris Mason bh_end_io_t *handler) 17041da177e4SLinus Torvalds { 17051da177e4SLinus Torvalds int err; 17061da177e4SLinus Torvalds sector_t block; 17071da177e4SLinus Torvalds sector_t last_block; 1708f0fbd5fcSAndrew Morton struct buffer_head *bh, *head; 170945bce8f3SLinus Torvalds unsigned int blocksize, bbits; 17101da177e4SLinus Torvalds int nr_underway = 0; 1711bafc0dbaSTejun Heo int write_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE); 17121da177e4SLinus Torvalds 171345bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 17141da177e4SLinus Torvalds (1 << BH_Dirty)|(1 << BH_Uptodate)); 17151da177e4SLinus Torvalds 17161da177e4SLinus Torvalds /* 17171da177e4SLinus Torvalds * Be very careful. We have no exclusion from __set_page_dirty_buffers 17181da177e4SLinus Torvalds * here, and the (potentially unmapped) buffers may become dirty at 17191da177e4SLinus Torvalds * any time. If a buffer becomes dirty here after we've inspected it 17201da177e4SLinus Torvalds * then we just miss that fact, and the page stays dirty. 17211da177e4SLinus Torvalds * 17221da177e4SLinus Torvalds * Buffers outside i_size may be dirtied by __set_page_dirty_buffers; 17231da177e4SLinus Torvalds * handle that here by just cleaning them. 17241da177e4SLinus Torvalds */ 17251da177e4SLinus Torvalds 17261da177e4SLinus Torvalds bh = head; 172745bce8f3SLinus Torvalds blocksize = bh->b_size; 172845bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 172945bce8f3SLinus Torvalds 173045bce8f3SLinus Torvalds block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); 173145bce8f3SLinus Torvalds last_block = (i_size_read(inode) - 1) >> bbits; 17321da177e4SLinus Torvalds 17331da177e4SLinus Torvalds /* 17341da177e4SLinus Torvalds * Get all the dirty buffers mapped to disk addresses and 17351da177e4SLinus Torvalds * handle any aliases from the underlying blockdev's mapping. 17361da177e4SLinus Torvalds */ 17371da177e4SLinus Torvalds do { 17381da177e4SLinus Torvalds if (block > last_block) { 17391da177e4SLinus Torvalds /* 17401da177e4SLinus Torvalds * mapped buffers outside i_size will occur, because 17411da177e4SLinus Torvalds * this page can be outside i_size when there is a 17421da177e4SLinus Torvalds * truncate in progress. 17431da177e4SLinus Torvalds */ 17441da177e4SLinus Torvalds /* 17451da177e4SLinus Torvalds * The buffer was zeroed by block_write_full_page() 17461da177e4SLinus Torvalds */ 17471da177e4SLinus Torvalds clear_buffer_dirty(bh); 17481da177e4SLinus Torvalds set_buffer_uptodate(bh); 174929a814d2SAlex Tomas } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && 175029a814d2SAlex Tomas buffer_dirty(bh)) { 1751b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 17521da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 17531da177e4SLinus Torvalds if (err) 17541da177e4SLinus Torvalds goto recover; 175529a814d2SAlex Tomas clear_buffer_delay(bh); 17561da177e4SLinus Torvalds if (buffer_new(bh)) { 17571da177e4SLinus Torvalds /* blockdev mappings never come here */ 17581da177e4SLinus Torvalds clear_buffer_new(bh); 17591da177e4SLinus Torvalds unmap_underlying_metadata(bh->b_bdev, 17601da177e4SLinus Torvalds bh->b_blocknr); 17611da177e4SLinus Torvalds } 17621da177e4SLinus Torvalds } 17631da177e4SLinus Torvalds bh = bh->b_this_page; 17641da177e4SLinus Torvalds block++; 17651da177e4SLinus Torvalds } while (bh != head); 17661da177e4SLinus Torvalds 17671da177e4SLinus Torvalds do { 17681da177e4SLinus Torvalds if (!buffer_mapped(bh)) 17691da177e4SLinus Torvalds continue; 17701da177e4SLinus Torvalds /* 17711da177e4SLinus Torvalds * If it's a fully non-blocking write attempt and we cannot 17721da177e4SLinus Torvalds * lock the buffer then redirty the page. Note that this can 17735b0830cbSJens Axboe * potentially cause a busy-wait loop from writeback threads 17745b0830cbSJens Axboe * and kswapd activity, but those code paths have their own 17755b0830cbSJens Axboe * higher-level throttling. 17761da177e4SLinus Torvalds */ 17771b430beeSWu Fengguang if (wbc->sync_mode != WB_SYNC_NONE) { 17781da177e4SLinus Torvalds lock_buffer(bh); 1779ca5de404SNick Piggin } else if (!trylock_buffer(bh)) { 17801da177e4SLinus Torvalds redirty_page_for_writepage(wbc, page); 17811da177e4SLinus Torvalds continue; 17821da177e4SLinus Torvalds } 17831da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 178435c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 17851da177e4SLinus Torvalds } else { 17861da177e4SLinus Torvalds unlock_buffer(bh); 17871da177e4SLinus Torvalds } 17881da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 17891da177e4SLinus Torvalds 17901da177e4SLinus Torvalds /* 17911da177e4SLinus Torvalds * The page and its buffers are protected by PageWriteback(), so we can 17921da177e4SLinus Torvalds * drop the bh refcounts early. 17931da177e4SLinus Torvalds */ 17941da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 17951da177e4SLinus Torvalds set_page_writeback(page); 17961da177e4SLinus Torvalds 17971da177e4SLinus Torvalds do { 17981da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 17991da177e4SLinus Torvalds if (buffer_async_write(bh)) { 1800b16b1debSTejun Heo submit_bh_wbc(write_op, bh, 0, wbc); 18011da177e4SLinus Torvalds nr_underway++; 1802ad576e63SNick Piggin } 18031da177e4SLinus Torvalds bh = next; 18041da177e4SLinus Torvalds } while (bh != head); 180505937baaSAndrew Morton unlock_page(page); 18061da177e4SLinus Torvalds 18071da177e4SLinus Torvalds err = 0; 18081da177e4SLinus Torvalds done: 18091da177e4SLinus Torvalds if (nr_underway == 0) { 18101da177e4SLinus Torvalds /* 18111da177e4SLinus Torvalds * The page was marked dirty, but the buffers were 18121da177e4SLinus Torvalds * clean. Someone wrote them back by hand with 18131da177e4SLinus Torvalds * ll_rw_block/submit_bh. A rare case. 18141da177e4SLinus Torvalds */ 18151da177e4SLinus Torvalds end_page_writeback(page); 18163d67f2d7SNick Piggin 18171da177e4SLinus Torvalds /* 18181da177e4SLinus Torvalds * The page and buffer_heads can be released at any time from 18191da177e4SLinus Torvalds * here on. 18201da177e4SLinus Torvalds */ 18211da177e4SLinus Torvalds } 18221da177e4SLinus Torvalds return err; 18231da177e4SLinus Torvalds 18241da177e4SLinus Torvalds recover: 18251da177e4SLinus Torvalds /* 18261da177e4SLinus Torvalds * ENOSPC, or some other error. We may already have added some 18271da177e4SLinus Torvalds * blocks to the file, so we need to write these out to avoid 18281da177e4SLinus Torvalds * exposing stale data. 18291da177e4SLinus Torvalds * The page is currently locked and not marked for writeback 18301da177e4SLinus Torvalds */ 18311da177e4SLinus Torvalds bh = head; 18321da177e4SLinus Torvalds /* Recovery: lock and submit the mapped buffers */ 18331da177e4SLinus Torvalds do { 183429a814d2SAlex Tomas if (buffer_mapped(bh) && buffer_dirty(bh) && 183529a814d2SAlex Tomas !buffer_delay(bh)) { 18361da177e4SLinus Torvalds lock_buffer(bh); 183735c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 18381da177e4SLinus Torvalds } else { 18391da177e4SLinus Torvalds /* 18401da177e4SLinus Torvalds * The buffer may have been set dirty during 18411da177e4SLinus Torvalds * attachment to a dirty page. 18421da177e4SLinus Torvalds */ 18431da177e4SLinus Torvalds clear_buffer_dirty(bh); 18441da177e4SLinus Torvalds } 18451da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 18461da177e4SLinus Torvalds SetPageError(page); 18471da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 18487e4c3690SAndrew Morton mapping_set_error(page->mapping, err); 18491da177e4SLinus Torvalds set_page_writeback(page); 18501da177e4SLinus Torvalds do { 18511da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 18521da177e4SLinus Torvalds if (buffer_async_write(bh)) { 18531da177e4SLinus Torvalds clear_buffer_dirty(bh); 1854b16b1debSTejun Heo submit_bh_wbc(write_op, bh, 0, wbc); 18551da177e4SLinus Torvalds nr_underway++; 1856ad576e63SNick Piggin } 18571da177e4SLinus Torvalds bh = next; 18581da177e4SLinus Torvalds } while (bh != head); 1859ffda9d30SNick Piggin unlock_page(page); 18601da177e4SLinus Torvalds goto done; 18611da177e4SLinus Torvalds } 18621da177e4SLinus Torvalds 1863afddba49SNick Piggin /* 1864afddba49SNick Piggin * If a page has any new buffers, zero them out here, and mark them uptodate 1865afddba49SNick Piggin * and dirty so they'll be written out (in order to prevent uninitialised 1866afddba49SNick Piggin * block data from leaking). And clear the new bit. 1867afddba49SNick Piggin */ 1868afddba49SNick Piggin void page_zero_new_buffers(struct page *page, unsigned from, unsigned to) 1869afddba49SNick Piggin { 1870afddba49SNick Piggin unsigned int block_start, block_end; 1871afddba49SNick Piggin struct buffer_head *head, *bh; 1872afddba49SNick Piggin 1873afddba49SNick Piggin BUG_ON(!PageLocked(page)); 1874afddba49SNick Piggin if (!page_has_buffers(page)) 1875afddba49SNick Piggin return; 1876afddba49SNick Piggin 1877afddba49SNick Piggin bh = head = page_buffers(page); 1878afddba49SNick Piggin block_start = 0; 1879afddba49SNick Piggin do { 1880afddba49SNick Piggin block_end = block_start + bh->b_size; 1881afddba49SNick Piggin 1882afddba49SNick Piggin if (buffer_new(bh)) { 1883afddba49SNick Piggin if (block_end > from && block_start < to) { 1884afddba49SNick Piggin if (!PageUptodate(page)) { 1885afddba49SNick Piggin unsigned start, size; 1886afddba49SNick Piggin 1887afddba49SNick Piggin start = max(from, block_start); 1888afddba49SNick Piggin size = min(to, block_end) - start; 1889afddba49SNick Piggin 1890eebd2aa3SChristoph Lameter zero_user(page, start, size); 1891afddba49SNick Piggin set_buffer_uptodate(bh); 1892afddba49SNick Piggin } 1893afddba49SNick Piggin 1894afddba49SNick Piggin clear_buffer_new(bh); 1895afddba49SNick Piggin mark_buffer_dirty(bh); 1896afddba49SNick Piggin } 1897afddba49SNick Piggin } 1898afddba49SNick Piggin 1899afddba49SNick Piggin block_start = block_end; 1900afddba49SNick Piggin bh = bh->b_this_page; 1901afddba49SNick Piggin } while (bh != head); 1902afddba49SNick Piggin } 1903afddba49SNick Piggin EXPORT_SYMBOL(page_zero_new_buffers); 1904afddba49SNick Piggin 1905ebdec241SChristoph Hellwig int __block_write_begin(struct page *page, loff_t pos, unsigned len, 19066e1db88dSChristoph Hellwig get_block_t *get_block) 19071da177e4SLinus Torvalds { 1908ebdec241SChristoph Hellwig unsigned from = pos & (PAGE_CACHE_SIZE - 1); 1909ebdec241SChristoph Hellwig unsigned to = from + len; 19106e1db88dSChristoph Hellwig struct inode *inode = page->mapping->host; 19111da177e4SLinus Torvalds unsigned block_start, block_end; 19121da177e4SLinus Torvalds sector_t block; 19131da177e4SLinus Torvalds int err = 0; 19141da177e4SLinus Torvalds unsigned blocksize, bbits; 19151da177e4SLinus Torvalds struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; 19161da177e4SLinus Torvalds 19171da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 19181da177e4SLinus Torvalds BUG_ON(from > PAGE_CACHE_SIZE); 19191da177e4SLinus Torvalds BUG_ON(to > PAGE_CACHE_SIZE); 19201da177e4SLinus Torvalds BUG_ON(from > to); 19211da177e4SLinus Torvalds 192245bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 0); 192345bce8f3SLinus Torvalds blocksize = head->b_size; 192445bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 19251da177e4SLinus Torvalds 19261da177e4SLinus Torvalds block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); 19271da177e4SLinus Torvalds 19281da177e4SLinus Torvalds for(bh = head, block_start = 0; bh != head || !block_start; 19291da177e4SLinus Torvalds block++, block_start=block_end, bh = bh->b_this_page) { 19301da177e4SLinus Torvalds block_end = block_start + blocksize; 19311da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 19321da177e4SLinus Torvalds if (PageUptodate(page)) { 19331da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 19341da177e4SLinus Torvalds set_buffer_uptodate(bh); 19351da177e4SLinus Torvalds } 19361da177e4SLinus Torvalds continue; 19371da177e4SLinus Torvalds } 19381da177e4SLinus Torvalds if (buffer_new(bh)) 19391da177e4SLinus Torvalds clear_buffer_new(bh); 19401da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 1941b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 19421da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 19431da177e4SLinus Torvalds if (err) 1944f3ddbdc6SNick Piggin break; 19451da177e4SLinus Torvalds if (buffer_new(bh)) { 19461da177e4SLinus Torvalds unmap_underlying_metadata(bh->b_bdev, 19471da177e4SLinus Torvalds bh->b_blocknr); 19481da177e4SLinus Torvalds if (PageUptodate(page)) { 1949637aff46SNick Piggin clear_buffer_new(bh); 19501da177e4SLinus Torvalds set_buffer_uptodate(bh); 1951637aff46SNick Piggin mark_buffer_dirty(bh); 19521da177e4SLinus Torvalds continue; 19531da177e4SLinus Torvalds } 1954eebd2aa3SChristoph Lameter if (block_end > to || block_start < from) 1955eebd2aa3SChristoph Lameter zero_user_segments(page, 1956eebd2aa3SChristoph Lameter to, block_end, 1957eebd2aa3SChristoph Lameter block_start, from); 19581da177e4SLinus Torvalds continue; 19591da177e4SLinus Torvalds } 19601da177e4SLinus Torvalds } 19611da177e4SLinus Torvalds if (PageUptodate(page)) { 19621da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 19631da177e4SLinus Torvalds set_buffer_uptodate(bh); 19641da177e4SLinus Torvalds continue; 19651da177e4SLinus Torvalds } 19661da177e4SLinus Torvalds if (!buffer_uptodate(bh) && !buffer_delay(bh) && 196733a266ddSDavid Chinner !buffer_unwritten(bh) && 19681da177e4SLinus Torvalds (block_start < from || block_end > to)) { 19691da177e4SLinus Torvalds ll_rw_block(READ, 1, &bh); 19701da177e4SLinus Torvalds *wait_bh++=bh; 19711da177e4SLinus Torvalds } 19721da177e4SLinus Torvalds } 19731da177e4SLinus Torvalds /* 19741da177e4SLinus Torvalds * If we issued read requests - let them complete. 19751da177e4SLinus Torvalds */ 19761da177e4SLinus Torvalds while(wait_bh > wait) { 19771da177e4SLinus Torvalds wait_on_buffer(*--wait_bh); 19781da177e4SLinus Torvalds if (!buffer_uptodate(*wait_bh)) 1979f3ddbdc6SNick Piggin err = -EIO; 19801da177e4SLinus Torvalds } 1981f9f07b6cSJan Kara if (unlikely(err)) 1982afddba49SNick Piggin page_zero_new_buffers(page, from, to); 19831da177e4SLinus Torvalds return err; 19841da177e4SLinus Torvalds } 1985ebdec241SChristoph Hellwig EXPORT_SYMBOL(__block_write_begin); 19861da177e4SLinus Torvalds 19871da177e4SLinus Torvalds static int __block_commit_write(struct inode *inode, struct page *page, 19881da177e4SLinus Torvalds unsigned from, unsigned to) 19891da177e4SLinus Torvalds { 19901da177e4SLinus Torvalds unsigned block_start, block_end; 19911da177e4SLinus Torvalds int partial = 0; 19921da177e4SLinus Torvalds unsigned blocksize; 19931da177e4SLinus Torvalds struct buffer_head *bh, *head; 19941da177e4SLinus Torvalds 199545bce8f3SLinus Torvalds bh = head = page_buffers(page); 199645bce8f3SLinus Torvalds blocksize = bh->b_size; 19971da177e4SLinus Torvalds 199845bce8f3SLinus Torvalds block_start = 0; 199945bce8f3SLinus Torvalds do { 20001da177e4SLinus Torvalds block_end = block_start + blocksize; 20011da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 20021da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 20031da177e4SLinus Torvalds partial = 1; 20041da177e4SLinus Torvalds } else { 20051da177e4SLinus Torvalds set_buffer_uptodate(bh); 20061da177e4SLinus Torvalds mark_buffer_dirty(bh); 20071da177e4SLinus Torvalds } 2008afddba49SNick Piggin clear_buffer_new(bh); 200945bce8f3SLinus Torvalds 201045bce8f3SLinus Torvalds block_start = block_end; 201145bce8f3SLinus Torvalds bh = bh->b_this_page; 201245bce8f3SLinus Torvalds } while (bh != head); 20131da177e4SLinus Torvalds 20141da177e4SLinus Torvalds /* 20151da177e4SLinus Torvalds * If this is a partial write which happened to make all buffers 20161da177e4SLinus Torvalds * uptodate then we can optimize away a bogus readpage() for 20171da177e4SLinus Torvalds * the next read(). Here we 'discover' whether the page went 20181da177e4SLinus Torvalds * uptodate as a result of this (potentially partial) write. 20191da177e4SLinus Torvalds */ 20201da177e4SLinus Torvalds if (!partial) 20211da177e4SLinus Torvalds SetPageUptodate(page); 20221da177e4SLinus Torvalds return 0; 20231da177e4SLinus Torvalds } 20241da177e4SLinus Torvalds 20251da177e4SLinus Torvalds /* 2026155130a4SChristoph Hellwig * block_write_begin takes care of the basic task of block allocation and 2027155130a4SChristoph Hellwig * bringing partial write blocks uptodate first. 2028155130a4SChristoph Hellwig * 20297bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 2030afddba49SNick Piggin */ 2031155130a4SChristoph Hellwig int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, 2032155130a4SChristoph Hellwig unsigned flags, struct page **pagep, get_block_t *get_block) 2033afddba49SNick Piggin { 20346e1db88dSChristoph Hellwig pgoff_t index = pos >> PAGE_CACHE_SHIFT; 2035afddba49SNick Piggin struct page *page; 20366e1db88dSChristoph Hellwig int status; 2037afddba49SNick Piggin 203854566b2cSNick Piggin page = grab_cache_page_write_begin(mapping, index, flags); 20396e1db88dSChristoph Hellwig if (!page) 20406e1db88dSChristoph Hellwig return -ENOMEM; 2041afddba49SNick Piggin 20426e1db88dSChristoph Hellwig status = __block_write_begin(page, pos, len, get_block); 2043afddba49SNick Piggin if (unlikely(status)) { 2044afddba49SNick Piggin unlock_page(page); 2045afddba49SNick Piggin page_cache_release(page); 20466e1db88dSChristoph Hellwig page = NULL; 2047afddba49SNick Piggin } 2048afddba49SNick Piggin 20496e1db88dSChristoph Hellwig *pagep = page; 2050afddba49SNick Piggin return status; 2051afddba49SNick Piggin } 2052afddba49SNick Piggin EXPORT_SYMBOL(block_write_begin); 2053afddba49SNick Piggin 2054afddba49SNick Piggin int block_write_end(struct file *file, struct address_space *mapping, 2055afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2056afddba49SNick Piggin struct page *page, void *fsdata) 2057afddba49SNick Piggin { 2058afddba49SNick Piggin struct inode *inode = mapping->host; 2059afddba49SNick Piggin unsigned start; 2060afddba49SNick Piggin 2061afddba49SNick Piggin start = pos & (PAGE_CACHE_SIZE - 1); 2062afddba49SNick Piggin 2063afddba49SNick Piggin if (unlikely(copied < len)) { 2064afddba49SNick Piggin /* 2065afddba49SNick Piggin * The buffers that were written will now be uptodate, so we 2066afddba49SNick Piggin * don't have to worry about a readpage reading them and 2067afddba49SNick Piggin * overwriting a partial write. However if we have encountered 2068afddba49SNick Piggin * a short write and only partially written into a buffer, it 2069afddba49SNick Piggin * will not be marked uptodate, so a readpage might come in and 2070afddba49SNick Piggin * destroy our partial write. 2071afddba49SNick Piggin * 2072afddba49SNick Piggin * Do the simplest thing, and just treat any short write to a 2073afddba49SNick Piggin * non uptodate page as a zero-length write, and force the 2074afddba49SNick Piggin * caller to redo the whole thing. 2075afddba49SNick Piggin */ 2076afddba49SNick Piggin if (!PageUptodate(page)) 2077afddba49SNick Piggin copied = 0; 2078afddba49SNick Piggin 2079afddba49SNick Piggin page_zero_new_buffers(page, start+copied, start+len); 2080afddba49SNick Piggin } 2081afddba49SNick Piggin flush_dcache_page(page); 2082afddba49SNick Piggin 2083afddba49SNick Piggin /* This could be a short (even 0-length) commit */ 2084afddba49SNick Piggin __block_commit_write(inode, page, start, start+copied); 2085afddba49SNick Piggin 2086afddba49SNick Piggin return copied; 2087afddba49SNick Piggin } 2088afddba49SNick Piggin EXPORT_SYMBOL(block_write_end); 2089afddba49SNick Piggin 2090afddba49SNick Piggin int generic_write_end(struct file *file, struct address_space *mapping, 2091afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2092afddba49SNick Piggin struct page *page, void *fsdata) 2093afddba49SNick Piggin { 2094afddba49SNick Piggin struct inode *inode = mapping->host; 209590a80202SJan Kara loff_t old_size = inode->i_size; 2096c7d206b3SJan Kara int i_size_changed = 0; 2097afddba49SNick Piggin 2098afddba49SNick Piggin copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); 2099afddba49SNick Piggin 2100afddba49SNick Piggin /* 2101afddba49SNick Piggin * No need to use i_size_read() here, the i_size 2102afddba49SNick Piggin * cannot change under us because we hold i_mutex. 2103afddba49SNick Piggin * 2104afddba49SNick Piggin * But it's important to update i_size while still holding page lock: 2105afddba49SNick Piggin * page writeout could otherwise come in and zero beyond i_size. 2106afddba49SNick Piggin */ 2107afddba49SNick Piggin if (pos+copied > inode->i_size) { 2108afddba49SNick Piggin i_size_write(inode, pos+copied); 2109c7d206b3SJan Kara i_size_changed = 1; 2110afddba49SNick Piggin } 2111afddba49SNick Piggin 2112afddba49SNick Piggin unlock_page(page); 2113afddba49SNick Piggin page_cache_release(page); 2114afddba49SNick Piggin 211590a80202SJan Kara if (old_size < pos) 211690a80202SJan Kara pagecache_isize_extended(inode, old_size, pos); 2117c7d206b3SJan Kara /* 2118c7d206b3SJan Kara * Don't mark the inode dirty under page lock. First, it unnecessarily 2119c7d206b3SJan Kara * makes the holding time of page lock longer. Second, it forces lock 2120c7d206b3SJan Kara * ordering of page lock and transaction start for journaling 2121c7d206b3SJan Kara * filesystems. 2122c7d206b3SJan Kara */ 2123c7d206b3SJan Kara if (i_size_changed) 2124c7d206b3SJan Kara mark_inode_dirty(inode); 2125c7d206b3SJan Kara 2126afddba49SNick Piggin return copied; 2127afddba49SNick Piggin } 2128afddba49SNick Piggin EXPORT_SYMBOL(generic_write_end); 2129afddba49SNick Piggin 2130afddba49SNick Piggin /* 21318ab22b9aSHisashi Hifumi * block_is_partially_uptodate checks whether buffers within a page are 21328ab22b9aSHisashi Hifumi * uptodate or not. 21338ab22b9aSHisashi Hifumi * 21348ab22b9aSHisashi Hifumi * Returns true if all buffers which correspond to a file portion 21358ab22b9aSHisashi Hifumi * we want to read are uptodate. 21368ab22b9aSHisashi Hifumi */ 2137c186afb4SAl Viro int block_is_partially_uptodate(struct page *page, unsigned long from, 2138c186afb4SAl Viro unsigned long count) 21398ab22b9aSHisashi Hifumi { 21408ab22b9aSHisashi Hifumi unsigned block_start, block_end, blocksize; 21418ab22b9aSHisashi Hifumi unsigned to; 21428ab22b9aSHisashi Hifumi struct buffer_head *bh, *head; 21438ab22b9aSHisashi Hifumi int ret = 1; 21448ab22b9aSHisashi Hifumi 21458ab22b9aSHisashi Hifumi if (!page_has_buffers(page)) 21468ab22b9aSHisashi Hifumi return 0; 21478ab22b9aSHisashi Hifumi 214845bce8f3SLinus Torvalds head = page_buffers(page); 214945bce8f3SLinus Torvalds blocksize = head->b_size; 2150c186afb4SAl Viro to = min_t(unsigned, PAGE_CACHE_SIZE - from, count); 21518ab22b9aSHisashi Hifumi to = from + to; 21528ab22b9aSHisashi Hifumi if (from < blocksize && to > PAGE_CACHE_SIZE - blocksize) 21538ab22b9aSHisashi Hifumi return 0; 21548ab22b9aSHisashi Hifumi 21558ab22b9aSHisashi Hifumi bh = head; 21568ab22b9aSHisashi Hifumi block_start = 0; 21578ab22b9aSHisashi Hifumi do { 21588ab22b9aSHisashi Hifumi block_end = block_start + blocksize; 21598ab22b9aSHisashi Hifumi if (block_end > from && block_start < to) { 21608ab22b9aSHisashi Hifumi if (!buffer_uptodate(bh)) { 21618ab22b9aSHisashi Hifumi ret = 0; 21628ab22b9aSHisashi Hifumi break; 21638ab22b9aSHisashi Hifumi } 21648ab22b9aSHisashi Hifumi if (block_end >= to) 21658ab22b9aSHisashi Hifumi break; 21668ab22b9aSHisashi Hifumi } 21678ab22b9aSHisashi Hifumi block_start = block_end; 21688ab22b9aSHisashi Hifumi bh = bh->b_this_page; 21698ab22b9aSHisashi Hifumi } while (bh != head); 21708ab22b9aSHisashi Hifumi 21718ab22b9aSHisashi Hifumi return ret; 21728ab22b9aSHisashi Hifumi } 21738ab22b9aSHisashi Hifumi EXPORT_SYMBOL(block_is_partially_uptodate); 21748ab22b9aSHisashi Hifumi 21758ab22b9aSHisashi Hifumi /* 21761da177e4SLinus Torvalds * Generic "read page" function for block devices that have the normal 21771da177e4SLinus Torvalds * get_block functionality. This is most of the block device filesystems. 21781da177e4SLinus Torvalds * Reads the page asynchronously --- the unlock_buffer() and 21791da177e4SLinus Torvalds * set/clear_buffer_uptodate() functions propagate buffer state into the 21801da177e4SLinus Torvalds * page struct once IO has completed. 21811da177e4SLinus Torvalds */ 21821da177e4SLinus Torvalds int block_read_full_page(struct page *page, get_block_t *get_block) 21831da177e4SLinus Torvalds { 21841da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 21851da177e4SLinus Torvalds sector_t iblock, lblock; 21861da177e4SLinus Torvalds struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 218745bce8f3SLinus Torvalds unsigned int blocksize, bbits; 21881da177e4SLinus Torvalds int nr, i; 21891da177e4SLinus Torvalds int fully_mapped = 1; 21901da177e4SLinus Torvalds 219145bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 0); 219245bce8f3SLinus Torvalds blocksize = head->b_size; 219345bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 21941da177e4SLinus Torvalds 219545bce8f3SLinus Torvalds iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); 219645bce8f3SLinus Torvalds lblock = (i_size_read(inode)+blocksize-1) >> bbits; 21971da177e4SLinus Torvalds bh = head; 21981da177e4SLinus Torvalds nr = 0; 21991da177e4SLinus Torvalds i = 0; 22001da177e4SLinus Torvalds 22011da177e4SLinus Torvalds do { 22021da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22031da177e4SLinus Torvalds continue; 22041da177e4SLinus Torvalds 22051da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2206c64610baSAndrew Morton int err = 0; 2207c64610baSAndrew Morton 22081da177e4SLinus Torvalds fully_mapped = 0; 22091da177e4SLinus Torvalds if (iblock < lblock) { 2210b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2211c64610baSAndrew Morton err = get_block(inode, iblock, bh, 0); 2212c64610baSAndrew Morton if (err) 22131da177e4SLinus Torvalds SetPageError(page); 22141da177e4SLinus Torvalds } 22151da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2216eebd2aa3SChristoph Lameter zero_user(page, i * blocksize, blocksize); 2217c64610baSAndrew Morton if (!err) 22181da177e4SLinus Torvalds set_buffer_uptodate(bh); 22191da177e4SLinus Torvalds continue; 22201da177e4SLinus Torvalds } 22211da177e4SLinus Torvalds /* 22221da177e4SLinus Torvalds * get_block() might have updated the buffer 22231da177e4SLinus Torvalds * synchronously 22241da177e4SLinus Torvalds */ 22251da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22261da177e4SLinus Torvalds continue; 22271da177e4SLinus Torvalds } 22281da177e4SLinus Torvalds arr[nr++] = bh; 22291da177e4SLinus Torvalds } while (i++, iblock++, (bh = bh->b_this_page) != head); 22301da177e4SLinus Torvalds 22311da177e4SLinus Torvalds if (fully_mapped) 22321da177e4SLinus Torvalds SetPageMappedToDisk(page); 22331da177e4SLinus Torvalds 22341da177e4SLinus Torvalds if (!nr) { 22351da177e4SLinus Torvalds /* 22361da177e4SLinus Torvalds * All buffers are uptodate - we can set the page uptodate 22371da177e4SLinus Torvalds * as well. But not if get_block() returned an error. 22381da177e4SLinus Torvalds */ 22391da177e4SLinus Torvalds if (!PageError(page)) 22401da177e4SLinus Torvalds SetPageUptodate(page); 22411da177e4SLinus Torvalds unlock_page(page); 22421da177e4SLinus Torvalds return 0; 22431da177e4SLinus Torvalds } 22441da177e4SLinus Torvalds 22451da177e4SLinus Torvalds /* Stage two: lock the buffers */ 22461da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 22471da177e4SLinus Torvalds bh = arr[i]; 22481da177e4SLinus Torvalds lock_buffer(bh); 22491da177e4SLinus Torvalds mark_buffer_async_read(bh); 22501da177e4SLinus Torvalds } 22511da177e4SLinus Torvalds 22521da177e4SLinus Torvalds /* 22531da177e4SLinus Torvalds * Stage 3: start the IO. Check for uptodateness 22541da177e4SLinus Torvalds * inside the buffer lock in case another process reading 22551da177e4SLinus Torvalds * the underlying blockdev brought it uptodate (the sct fix). 22561da177e4SLinus Torvalds */ 22571da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 22581da177e4SLinus Torvalds bh = arr[i]; 22591da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22601da177e4SLinus Torvalds end_buffer_async_read(bh, 1); 22611da177e4SLinus Torvalds else 22621da177e4SLinus Torvalds submit_bh(READ, bh); 22631da177e4SLinus Torvalds } 22641da177e4SLinus Torvalds return 0; 22651da177e4SLinus Torvalds } 22661fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_read_full_page); 22671da177e4SLinus Torvalds 22681da177e4SLinus Torvalds /* utility function for filesystems that need to do work on expanding 226989e10787SNick Piggin * truncates. Uses filesystem pagecache writes to allow the filesystem to 22701da177e4SLinus Torvalds * deal with the hole. 22711da177e4SLinus Torvalds */ 227289e10787SNick Piggin int generic_cont_expand_simple(struct inode *inode, loff_t size) 22731da177e4SLinus Torvalds { 22741da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 22751da177e4SLinus Torvalds struct page *page; 227689e10787SNick Piggin void *fsdata; 22771da177e4SLinus Torvalds int err; 22781da177e4SLinus Torvalds 2279c08d3b0eSnpiggin@suse.de err = inode_newsize_ok(inode, size); 2280c08d3b0eSnpiggin@suse.de if (err) 22811da177e4SLinus Torvalds goto out; 22821da177e4SLinus Torvalds 228389e10787SNick Piggin err = pagecache_write_begin(NULL, mapping, size, 0, 228489e10787SNick Piggin AOP_FLAG_UNINTERRUPTIBLE|AOP_FLAG_CONT_EXPAND, 228589e10787SNick Piggin &page, &fsdata); 228689e10787SNick Piggin if (err) 228705eb0b51SOGAWA Hirofumi goto out; 228805eb0b51SOGAWA Hirofumi 228989e10787SNick Piggin err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata); 229089e10787SNick Piggin BUG_ON(err > 0); 229105eb0b51SOGAWA Hirofumi 229205eb0b51SOGAWA Hirofumi out: 229305eb0b51SOGAWA Hirofumi return err; 229405eb0b51SOGAWA Hirofumi } 22951fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_cont_expand_simple); 229605eb0b51SOGAWA Hirofumi 2297f1e3af72SAdrian Bunk static int cont_expand_zero(struct file *file, struct address_space *mapping, 229889e10787SNick Piggin loff_t pos, loff_t *bytes) 229905eb0b51SOGAWA Hirofumi { 230089e10787SNick Piggin struct inode *inode = mapping->host; 230189e10787SNick Piggin unsigned blocksize = 1 << inode->i_blkbits; 230289e10787SNick Piggin struct page *page; 230389e10787SNick Piggin void *fsdata; 230489e10787SNick Piggin pgoff_t index, curidx; 230589e10787SNick Piggin loff_t curpos; 230689e10787SNick Piggin unsigned zerofrom, offset, len; 230789e10787SNick Piggin int err = 0; 230805eb0b51SOGAWA Hirofumi 230989e10787SNick Piggin index = pos >> PAGE_CACHE_SHIFT; 231089e10787SNick Piggin offset = pos & ~PAGE_CACHE_MASK; 231189e10787SNick Piggin 231289e10787SNick Piggin while (index > (curidx = (curpos = *bytes)>>PAGE_CACHE_SHIFT)) { 231389e10787SNick Piggin zerofrom = curpos & ~PAGE_CACHE_MASK; 231489e10787SNick Piggin if (zerofrom & (blocksize-1)) { 231589e10787SNick Piggin *bytes |= (blocksize-1); 231689e10787SNick Piggin (*bytes)++; 231789e10787SNick Piggin } 231889e10787SNick Piggin len = PAGE_CACHE_SIZE - zerofrom; 231989e10787SNick Piggin 232089e10787SNick Piggin err = pagecache_write_begin(file, mapping, curpos, len, 232189e10787SNick Piggin AOP_FLAG_UNINTERRUPTIBLE, 232289e10787SNick Piggin &page, &fsdata); 232389e10787SNick Piggin if (err) 232489e10787SNick Piggin goto out; 2325eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 232689e10787SNick Piggin err = pagecache_write_end(file, mapping, curpos, len, len, 232789e10787SNick Piggin page, fsdata); 232889e10787SNick Piggin if (err < 0) 232989e10787SNick Piggin goto out; 233089e10787SNick Piggin BUG_ON(err != len); 233189e10787SNick Piggin err = 0; 2332061e9746SOGAWA Hirofumi 2333061e9746SOGAWA Hirofumi balance_dirty_pages_ratelimited(mapping); 2334c2ca0fcdSMikulas Patocka 2335c2ca0fcdSMikulas Patocka if (unlikely(fatal_signal_pending(current))) { 2336c2ca0fcdSMikulas Patocka err = -EINTR; 2337c2ca0fcdSMikulas Patocka goto out; 2338c2ca0fcdSMikulas Patocka } 233989e10787SNick Piggin } 234089e10787SNick Piggin 234189e10787SNick Piggin /* page covers the boundary, find the boundary offset */ 234289e10787SNick Piggin if (index == curidx) { 234389e10787SNick Piggin zerofrom = curpos & ~PAGE_CACHE_MASK; 234489e10787SNick Piggin /* if we will expand the thing last block will be filled */ 234589e10787SNick Piggin if (offset <= zerofrom) { 234689e10787SNick Piggin goto out; 234789e10787SNick Piggin } 234889e10787SNick Piggin if (zerofrom & (blocksize-1)) { 234989e10787SNick Piggin *bytes |= (blocksize-1); 235089e10787SNick Piggin (*bytes)++; 235189e10787SNick Piggin } 235289e10787SNick Piggin len = offset - zerofrom; 235389e10787SNick Piggin 235489e10787SNick Piggin err = pagecache_write_begin(file, mapping, curpos, len, 235589e10787SNick Piggin AOP_FLAG_UNINTERRUPTIBLE, 235689e10787SNick Piggin &page, &fsdata); 235789e10787SNick Piggin if (err) 235889e10787SNick Piggin goto out; 2359eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 236089e10787SNick Piggin err = pagecache_write_end(file, mapping, curpos, len, len, 236189e10787SNick Piggin page, fsdata); 236289e10787SNick Piggin if (err < 0) 236389e10787SNick Piggin goto out; 236489e10787SNick Piggin BUG_ON(err != len); 236589e10787SNick Piggin err = 0; 236689e10787SNick Piggin } 236789e10787SNick Piggin out: 236889e10787SNick Piggin return err; 23691da177e4SLinus Torvalds } 23701da177e4SLinus Torvalds 23711da177e4SLinus Torvalds /* 23721da177e4SLinus Torvalds * For moronic filesystems that do not allow holes in file. 23731da177e4SLinus Torvalds * We may have to extend the file. 23741da177e4SLinus Torvalds */ 2375282dc178SChristoph Hellwig int cont_write_begin(struct file *file, struct address_space *mapping, 237689e10787SNick Piggin loff_t pos, unsigned len, unsigned flags, 237789e10787SNick Piggin struct page **pagep, void **fsdata, 237889e10787SNick Piggin get_block_t *get_block, loff_t *bytes) 23791da177e4SLinus Torvalds { 23801da177e4SLinus Torvalds struct inode *inode = mapping->host; 23811da177e4SLinus Torvalds unsigned blocksize = 1 << inode->i_blkbits; 238289e10787SNick Piggin unsigned zerofrom; 238389e10787SNick Piggin int err; 23841da177e4SLinus Torvalds 238589e10787SNick Piggin err = cont_expand_zero(file, mapping, pos, bytes); 238689e10787SNick Piggin if (err) 2387155130a4SChristoph Hellwig return err; 23881da177e4SLinus Torvalds 23891da177e4SLinus Torvalds zerofrom = *bytes & ~PAGE_CACHE_MASK; 239089e10787SNick Piggin if (pos+len > *bytes && zerofrom & (blocksize-1)) { 23911da177e4SLinus Torvalds *bytes |= (blocksize-1); 23921da177e4SLinus Torvalds (*bytes)++; 23931da177e4SLinus Torvalds } 23941da177e4SLinus Torvalds 2395155130a4SChristoph Hellwig return block_write_begin(mapping, pos, len, flags, pagep, get_block); 23961da177e4SLinus Torvalds } 23971fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(cont_write_begin); 23981da177e4SLinus Torvalds 23991da177e4SLinus Torvalds int block_commit_write(struct page *page, unsigned from, unsigned to) 24001da177e4SLinus Torvalds { 24011da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 24021da177e4SLinus Torvalds __block_commit_write(inode,page,from,to); 24031da177e4SLinus Torvalds return 0; 24041da177e4SLinus Torvalds } 24051fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_commit_write); 24061da177e4SLinus Torvalds 240754171690SDavid Chinner /* 240854171690SDavid Chinner * block_page_mkwrite() is not allowed to change the file size as it gets 240954171690SDavid Chinner * called from a page fault handler when a page is first dirtied. Hence we must 241054171690SDavid Chinner * be careful to check for EOF conditions here. We set the page up correctly 241154171690SDavid Chinner * for a written page which means we get ENOSPC checking when writing into 241254171690SDavid Chinner * holes and correct delalloc and unwritten extent mapping on filesystems that 241354171690SDavid Chinner * support these features. 241454171690SDavid Chinner * 241554171690SDavid Chinner * We are not allowed to take the i_mutex here so we have to play games to 241654171690SDavid Chinner * protect against truncate races as the page could now be beyond EOF. Because 24177bb46a67Snpiggin@suse.de * truncate writes the inode size before removing pages, once we have the 241854171690SDavid Chinner * page lock we can determine safely if the page is beyond EOF. If it is not 241954171690SDavid Chinner * beyond EOF, then the page is guaranteed safe against truncation until we 242054171690SDavid Chinner * unlock the page. 2421ea13a864SJan Kara * 242214da9200SJan Kara * Direct callers of this function should protect against filesystem freezing 242314da9200SJan Kara * using sb_start_write() - sb_end_write() functions. 242454171690SDavid Chinner */ 242524da4fabSJan Kara int __block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 242654171690SDavid Chinner get_block_t get_block) 242754171690SDavid Chinner { 2428c2ec175cSNick Piggin struct page *page = vmf->page; 2429496ad9aaSAl Viro struct inode *inode = file_inode(vma->vm_file); 243054171690SDavid Chinner unsigned long end; 243154171690SDavid Chinner loff_t size; 243224da4fabSJan Kara int ret; 243354171690SDavid Chinner 243454171690SDavid Chinner lock_page(page); 243554171690SDavid Chinner size = i_size_read(inode); 243654171690SDavid Chinner if ((page->mapping != inode->i_mapping) || 243718336338SNick Piggin (page_offset(page) > size)) { 243824da4fabSJan Kara /* We overload EFAULT to mean page got truncated */ 243924da4fabSJan Kara ret = -EFAULT; 244024da4fabSJan Kara goto out_unlock; 244154171690SDavid Chinner } 244254171690SDavid Chinner 244354171690SDavid Chinner /* page is wholly or partially inside EOF */ 244454171690SDavid Chinner if (((page->index + 1) << PAGE_CACHE_SHIFT) > size) 244554171690SDavid Chinner end = size & ~PAGE_CACHE_MASK; 244654171690SDavid Chinner else 244754171690SDavid Chinner end = PAGE_CACHE_SIZE; 244854171690SDavid Chinner 2449ebdec241SChristoph Hellwig ret = __block_write_begin(page, 0, end, get_block); 245054171690SDavid Chinner if (!ret) 245154171690SDavid Chinner ret = block_commit_write(page, 0, end); 245254171690SDavid Chinner 245324da4fabSJan Kara if (unlikely(ret < 0)) 245424da4fabSJan Kara goto out_unlock; 2455ea13a864SJan Kara set_page_dirty(page); 24561d1d1a76SDarrick J. Wong wait_for_stable_page(page); 245724da4fabSJan Kara return 0; 245824da4fabSJan Kara out_unlock: 2459b827e496SNick Piggin unlock_page(page); 246054171690SDavid Chinner return ret; 246154171690SDavid Chinner } 246224da4fabSJan Kara EXPORT_SYMBOL(__block_page_mkwrite); 246324da4fabSJan Kara 246424da4fabSJan Kara int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 246524da4fabSJan Kara get_block_t get_block) 246624da4fabSJan Kara { 2467ea13a864SJan Kara int ret; 2468496ad9aaSAl Viro struct super_block *sb = file_inode(vma->vm_file)->i_sb; 246924da4fabSJan Kara 247014da9200SJan Kara sb_start_pagefault(sb); 2471041bbb6dSTheodore Ts'o 2472041bbb6dSTheodore Ts'o /* 2473041bbb6dSTheodore Ts'o * Update file times before taking page lock. We may end up failing the 2474041bbb6dSTheodore Ts'o * fault so this update may be superfluous but who really cares... 2475041bbb6dSTheodore Ts'o */ 2476041bbb6dSTheodore Ts'o file_update_time(vma->vm_file); 2477041bbb6dSTheodore Ts'o 2478ea13a864SJan Kara ret = __block_page_mkwrite(vma, vmf, get_block); 247914da9200SJan Kara sb_end_pagefault(sb); 248024da4fabSJan Kara return block_page_mkwrite_return(ret); 248124da4fabSJan Kara } 24821fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_page_mkwrite); 24831da177e4SLinus Torvalds 24841da177e4SLinus Torvalds /* 248503158cd7SNick Piggin * nobh_write_begin()'s prereads are special: the buffer_heads are freed 24861da177e4SLinus Torvalds * immediately, while under the page lock. So it needs a special end_io 24871da177e4SLinus Torvalds * handler which does not touch the bh after unlocking it. 24881da177e4SLinus Torvalds */ 24891da177e4SLinus Torvalds static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate) 24901da177e4SLinus Torvalds { 249168671f35SDmitry Monakhov __end_buffer_read_notouch(bh, uptodate); 24921da177e4SLinus Torvalds } 24931da177e4SLinus Torvalds 24941da177e4SLinus Torvalds /* 249503158cd7SNick Piggin * Attach the singly-linked list of buffers created by nobh_write_begin, to 249603158cd7SNick Piggin * the page (converting it to circular linked list and taking care of page 249703158cd7SNick Piggin * dirty races). 249803158cd7SNick Piggin */ 249903158cd7SNick Piggin static void attach_nobh_buffers(struct page *page, struct buffer_head *head) 250003158cd7SNick Piggin { 250103158cd7SNick Piggin struct buffer_head *bh; 250203158cd7SNick Piggin 250303158cd7SNick Piggin BUG_ON(!PageLocked(page)); 250403158cd7SNick Piggin 250503158cd7SNick Piggin spin_lock(&page->mapping->private_lock); 250603158cd7SNick Piggin bh = head; 250703158cd7SNick Piggin do { 250803158cd7SNick Piggin if (PageDirty(page)) 250903158cd7SNick Piggin set_buffer_dirty(bh); 251003158cd7SNick Piggin if (!bh->b_this_page) 251103158cd7SNick Piggin bh->b_this_page = head; 251203158cd7SNick Piggin bh = bh->b_this_page; 251303158cd7SNick Piggin } while (bh != head); 251403158cd7SNick Piggin attach_page_buffers(page, head); 251503158cd7SNick Piggin spin_unlock(&page->mapping->private_lock); 251603158cd7SNick Piggin } 251703158cd7SNick Piggin 251803158cd7SNick Piggin /* 2519ea0f04e5SChristoph Hellwig * On entry, the page is fully not uptodate. 2520ea0f04e5SChristoph Hellwig * On exit the page is fully uptodate in the areas outside (from,to) 25217bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 25221da177e4SLinus Torvalds */ 2523ea0f04e5SChristoph Hellwig int nobh_write_begin(struct address_space *mapping, 252403158cd7SNick Piggin loff_t pos, unsigned len, unsigned flags, 252503158cd7SNick Piggin struct page **pagep, void **fsdata, 25261da177e4SLinus Torvalds get_block_t *get_block) 25271da177e4SLinus Torvalds { 252803158cd7SNick Piggin struct inode *inode = mapping->host; 25291da177e4SLinus Torvalds const unsigned blkbits = inode->i_blkbits; 25301da177e4SLinus Torvalds const unsigned blocksize = 1 << blkbits; 2531a4b0672dSNick Piggin struct buffer_head *head, *bh; 253203158cd7SNick Piggin struct page *page; 253303158cd7SNick Piggin pgoff_t index; 253403158cd7SNick Piggin unsigned from, to; 25351da177e4SLinus Torvalds unsigned block_in_page; 2536a4b0672dSNick Piggin unsigned block_start, block_end; 25371da177e4SLinus Torvalds sector_t block_in_file; 25381da177e4SLinus Torvalds int nr_reads = 0; 25391da177e4SLinus Torvalds int ret = 0; 25401da177e4SLinus Torvalds int is_mapped_to_disk = 1; 25411da177e4SLinus Torvalds 254203158cd7SNick Piggin index = pos >> PAGE_CACHE_SHIFT; 254303158cd7SNick Piggin from = pos & (PAGE_CACHE_SIZE - 1); 254403158cd7SNick Piggin to = from + len; 254503158cd7SNick Piggin 254654566b2cSNick Piggin page = grab_cache_page_write_begin(mapping, index, flags); 254703158cd7SNick Piggin if (!page) 254803158cd7SNick Piggin return -ENOMEM; 254903158cd7SNick Piggin *pagep = page; 255003158cd7SNick Piggin *fsdata = NULL; 255103158cd7SNick Piggin 255203158cd7SNick Piggin if (page_has_buffers(page)) { 2553309f77adSNamhyung Kim ret = __block_write_begin(page, pos, len, get_block); 2554309f77adSNamhyung Kim if (unlikely(ret)) 2555309f77adSNamhyung Kim goto out_release; 2556309f77adSNamhyung Kim return ret; 255703158cd7SNick Piggin } 2558a4b0672dSNick Piggin 25591da177e4SLinus Torvalds if (PageMappedToDisk(page)) 25601da177e4SLinus Torvalds return 0; 25611da177e4SLinus Torvalds 2562a4b0672dSNick Piggin /* 2563a4b0672dSNick Piggin * Allocate buffers so that we can keep track of state, and potentially 2564a4b0672dSNick Piggin * attach them to the page if an error occurs. In the common case of 2565a4b0672dSNick Piggin * no error, they will just be freed again without ever being attached 2566a4b0672dSNick Piggin * to the page (which is all OK, because we're under the page lock). 2567a4b0672dSNick Piggin * 2568a4b0672dSNick Piggin * Be careful: the buffer linked list is a NULL terminated one, rather 2569a4b0672dSNick Piggin * than the circular one we're used to. 2570a4b0672dSNick Piggin */ 2571a4b0672dSNick Piggin head = alloc_page_buffers(page, blocksize, 0); 257203158cd7SNick Piggin if (!head) { 257303158cd7SNick Piggin ret = -ENOMEM; 257403158cd7SNick Piggin goto out_release; 257503158cd7SNick Piggin } 2576a4b0672dSNick Piggin 25771da177e4SLinus Torvalds block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits); 25781da177e4SLinus Torvalds 25791da177e4SLinus Torvalds /* 25801da177e4SLinus Torvalds * We loop across all blocks in the page, whether or not they are 25811da177e4SLinus Torvalds * part of the affected region. This is so we can discover if the 25821da177e4SLinus Torvalds * page is fully mapped-to-disk. 25831da177e4SLinus Torvalds */ 2584a4b0672dSNick Piggin for (block_start = 0, block_in_page = 0, bh = head; 25851da177e4SLinus Torvalds block_start < PAGE_CACHE_SIZE; 2586a4b0672dSNick Piggin block_in_page++, block_start += blocksize, bh = bh->b_this_page) { 25871da177e4SLinus Torvalds int create; 25881da177e4SLinus Torvalds 2589a4b0672dSNick Piggin block_end = block_start + blocksize; 2590a4b0672dSNick Piggin bh->b_state = 0; 25911da177e4SLinus Torvalds create = 1; 25921da177e4SLinus Torvalds if (block_start >= to) 25931da177e4SLinus Torvalds create = 0; 25941da177e4SLinus Torvalds ret = get_block(inode, block_in_file + block_in_page, 2595a4b0672dSNick Piggin bh, create); 25961da177e4SLinus Torvalds if (ret) 25971da177e4SLinus Torvalds goto failed; 2598a4b0672dSNick Piggin if (!buffer_mapped(bh)) 25991da177e4SLinus Torvalds is_mapped_to_disk = 0; 2600a4b0672dSNick Piggin if (buffer_new(bh)) 2601a4b0672dSNick Piggin unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); 2602a4b0672dSNick Piggin if (PageUptodate(page)) { 2603a4b0672dSNick Piggin set_buffer_uptodate(bh); 26041da177e4SLinus Torvalds continue; 2605a4b0672dSNick Piggin } 2606a4b0672dSNick Piggin if (buffer_new(bh) || !buffer_mapped(bh)) { 2607eebd2aa3SChristoph Lameter zero_user_segments(page, block_start, from, 2608eebd2aa3SChristoph Lameter to, block_end); 26091da177e4SLinus Torvalds continue; 26101da177e4SLinus Torvalds } 2611a4b0672dSNick Piggin if (buffer_uptodate(bh)) 26121da177e4SLinus Torvalds continue; /* reiserfs does this */ 26131da177e4SLinus Torvalds if (block_start < from || block_end > to) { 2614a4b0672dSNick Piggin lock_buffer(bh); 2615a4b0672dSNick Piggin bh->b_end_io = end_buffer_read_nobh; 2616a4b0672dSNick Piggin submit_bh(READ, bh); 2617a4b0672dSNick Piggin nr_reads++; 26181da177e4SLinus Torvalds } 26191da177e4SLinus Torvalds } 26201da177e4SLinus Torvalds 26211da177e4SLinus Torvalds if (nr_reads) { 26221da177e4SLinus Torvalds /* 26231da177e4SLinus Torvalds * The page is locked, so these buffers are protected from 26241da177e4SLinus Torvalds * any VM or truncate activity. Hence we don't need to care 26251da177e4SLinus Torvalds * for the buffer_head refcounts. 26261da177e4SLinus Torvalds */ 2627a4b0672dSNick Piggin for (bh = head; bh; bh = bh->b_this_page) { 26281da177e4SLinus Torvalds wait_on_buffer(bh); 26291da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 26301da177e4SLinus Torvalds ret = -EIO; 26311da177e4SLinus Torvalds } 26321da177e4SLinus Torvalds if (ret) 26331da177e4SLinus Torvalds goto failed; 26341da177e4SLinus Torvalds } 26351da177e4SLinus Torvalds 26361da177e4SLinus Torvalds if (is_mapped_to_disk) 26371da177e4SLinus Torvalds SetPageMappedToDisk(page); 26381da177e4SLinus Torvalds 263903158cd7SNick Piggin *fsdata = head; /* to be released by nobh_write_end */ 2640a4b0672dSNick Piggin 26411da177e4SLinus Torvalds return 0; 26421da177e4SLinus Torvalds 26431da177e4SLinus Torvalds failed: 264403158cd7SNick Piggin BUG_ON(!ret); 26451da177e4SLinus Torvalds /* 2646a4b0672dSNick Piggin * Error recovery is a bit difficult. We need to zero out blocks that 2647a4b0672dSNick Piggin * were newly allocated, and dirty them to ensure they get written out. 2648a4b0672dSNick Piggin * Buffers need to be attached to the page at this point, otherwise 2649a4b0672dSNick Piggin * the handling of potential IO errors during writeout would be hard 2650a4b0672dSNick Piggin * (could try doing synchronous writeout, but what if that fails too?) 26511da177e4SLinus Torvalds */ 265203158cd7SNick Piggin attach_nobh_buffers(page, head); 265303158cd7SNick Piggin page_zero_new_buffers(page, from, to); 2654a4b0672dSNick Piggin 265503158cd7SNick Piggin out_release: 265603158cd7SNick Piggin unlock_page(page); 265703158cd7SNick Piggin page_cache_release(page); 265803158cd7SNick Piggin *pagep = NULL; 2659a4b0672dSNick Piggin 26607bb46a67Snpiggin@suse.de return ret; 26617bb46a67Snpiggin@suse.de } 266203158cd7SNick Piggin EXPORT_SYMBOL(nobh_write_begin); 26631da177e4SLinus Torvalds 266403158cd7SNick Piggin int nobh_write_end(struct file *file, struct address_space *mapping, 266503158cd7SNick Piggin loff_t pos, unsigned len, unsigned copied, 266603158cd7SNick Piggin struct page *page, void *fsdata) 26671da177e4SLinus Torvalds { 26681da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 2669efdc3131SNick Piggin struct buffer_head *head = fsdata; 267003158cd7SNick Piggin struct buffer_head *bh; 26715b41e74aSDmitri Monakhov BUG_ON(fsdata != NULL && page_has_buffers(page)); 26721da177e4SLinus Torvalds 2673d4cf109fSDave Kleikamp if (unlikely(copied < len) && head) 267403158cd7SNick Piggin attach_nobh_buffers(page, head); 2675a4b0672dSNick Piggin if (page_has_buffers(page)) 267603158cd7SNick Piggin return generic_write_end(file, mapping, pos, len, 267703158cd7SNick Piggin copied, page, fsdata); 2678a4b0672dSNick Piggin 267922c8ca78SNick Piggin SetPageUptodate(page); 26801da177e4SLinus Torvalds set_page_dirty(page); 268103158cd7SNick Piggin if (pos+copied > inode->i_size) { 268203158cd7SNick Piggin i_size_write(inode, pos+copied); 26831da177e4SLinus Torvalds mark_inode_dirty(inode); 26841da177e4SLinus Torvalds } 268503158cd7SNick Piggin 268603158cd7SNick Piggin unlock_page(page); 268703158cd7SNick Piggin page_cache_release(page); 268803158cd7SNick Piggin 268903158cd7SNick Piggin while (head) { 269003158cd7SNick Piggin bh = head; 269103158cd7SNick Piggin head = head->b_this_page; 269203158cd7SNick Piggin free_buffer_head(bh); 26931da177e4SLinus Torvalds } 269403158cd7SNick Piggin 269503158cd7SNick Piggin return copied; 269603158cd7SNick Piggin } 269703158cd7SNick Piggin EXPORT_SYMBOL(nobh_write_end); 26981da177e4SLinus Torvalds 26991da177e4SLinus Torvalds /* 27001da177e4SLinus Torvalds * nobh_writepage() - based on block_full_write_page() except 27011da177e4SLinus Torvalds * that it tries to operate without attaching bufferheads to 27021da177e4SLinus Torvalds * the page. 27031da177e4SLinus Torvalds */ 27041da177e4SLinus Torvalds int nobh_writepage(struct page *page, get_block_t *get_block, 27051da177e4SLinus Torvalds struct writeback_control *wbc) 27061da177e4SLinus Torvalds { 27071da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 27081da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 27091da177e4SLinus Torvalds const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; 27101da177e4SLinus Torvalds unsigned offset; 27111da177e4SLinus Torvalds int ret; 27121da177e4SLinus Torvalds 27131da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 27141da177e4SLinus Torvalds if (page->index < end_index) 27151da177e4SLinus Torvalds goto out; 27161da177e4SLinus Torvalds 27171da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 27181da177e4SLinus Torvalds offset = i_size & (PAGE_CACHE_SIZE-1); 27191da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 27201da177e4SLinus Torvalds /* 27211da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 27221da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 27231da177e4SLinus Torvalds * freeable here, so the page does not leak. 27241da177e4SLinus Torvalds */ 27251da177e4SLinus Torvalds #if 0 27261da177e4SLinus Torvalds /* Not really sure about this - do we need this ? */ 27271da177e4SLinus Torvalds if (page->mapping->a_ops->invalidatepage) 27281da177e4SLinus Torvalds page->mapping->a_ops->invalidatepage(page, offset); 27291da177e4SLinus Torvalds #endif 27301da177e4SLinus Torvalds unlock_page(page); 27311da177e4SLinus Torvalds return 0; /* don't care */ 27321da177e4SLinus Torvalds } 27331da177e4SLinus Torvalds 27341da177e4SLinus Torvalds /* 27351da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 27361da177e4SLinus Torvalds * writepage invocation because it may be mmapped. "A file is mapped 27371da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 27381da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 27391da177e4SLinus Torvalds * writes to that region are not written out to the file." 27401da177e4SLinus Torvalds */ 2741eebd2aa3SChristoph Lameter zero_user_segment(page, offset, PAGE_CACHE_SIZE); 27421da177e4SLinus Torvalds out: 27431da177e4SLinus Torvalds ret = mpage_writepage(page, get_block, wbc); 27441da177e4SLinus Torvalds if (ret == -EAGAIN) 274535c80d5fSChris Mason ret = __block_write_full_page(inode, page, get_block, wbc, 274635c80d5fSChris Mason end_buffer_async_write); 27471da177e4SLinus Torvalds return ret; 27481da177e4SLinus Torvalds } 27491da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_writepage); 27501da177e4SLinus Torvalds 275103158cd7SNick Piggin int nobh_truncate_page(struct address_space *mapping, 275203158cd7SNick Piggin loff_t from, get_block_t *get_block) 27531da177e4SLinus Torvalds { 27541da177e4SLinus Torvalds pgoff_t index = from >> PAGE_CACHE_SHIFT; 27551da177e4SLinus Torvalds unsigned offset = from & (PAGE_CACHE_SIZE-1); 275603158cd7SNick Piggin unsigned blocksize; 275703158cd7SNick Piggin sector_t iblock; 275803158cd7SNick Piggin unsigned length, pos; 275903158cd7SNick Piggin struct inode *inode = mapping->host; 27601da177e4SLinus Torvalds struct page *page; 276103158cd7SNick Piggin struct buffer_head map_bh; 276203158cd7SNick Piggin int err; 27631da177e4SLinus Torvalds 276403158cd7SNick Piggin blocksize = 1 << inode->i_blkbits; 276503158cd7SNick Piggin length = offset & (blocksize - 1); 27661da177e4SLinus Torvalds 276703158cd7SNick Piggin /* Block boundary? Nothing to do */ 276803158cd7SNick Piggin if (!length) 276903158cd7SNick Piggin return 0; 277003158cd7SNick Piggin 277103158cd7SNick Piggin length = blocksize - length; 277203158cd7SNick Piggin iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 277303158cd7SNick Piggin 27741da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 277503158cd7SNick Piggin err = -ENOMEM; 27761da177e4SLinus Torvalds if (!page) 27771da177e4SLinus Torvalds goto out; 27781da177e4SLinus Torvalds 277903158cd7SNick Piggin if (page_has_buffers(page)) { 278003158cd7SNick Piggin has_buffers: 278103158cd7SNick Piggin unlock_page(page); 278203158cd7SNick Piggin page_cache_release(page); 278303158cd7SNick Piggin return block_truncate_page(mapping, from, get_block); 27841da177e4SLinus Torvalds } 278503158cd7SNick Piggin 278603158cd7SNick Piggin /* Find the buffer that contains "offset" */ 278703158cd7SNick Piggin pos = blocksize; 278803158cd7SNick Piggin while (offset >= pos) { 278903158cd7SNick Piggin iblock++; 279003158cd7SNick Piggin pos += blocksize; 279103158cd7SNick Piggin } 279203158cd7SNick Piggin 2793460bcf57STheodore Ts'o map_bh.b_size = blocksize; 2794460bcf57STheodore Ts'o map_bh.b_state = 0; 279503158cd7SNick Piggin err = get_block(inode, iblock, &map_bh, 0); 279603158cd7SNick Piggin if (err) 279703158cd7SNick Piggin goto unlock; 279803158cd7SNick Piggin /* unmapped? It's a hole - nothing to do */ 279903158cd7SNick Piggin if (!buffer_mapped(&map_bh)) 280003158cd7SNick Piggin goto unlock; 280103158cd7SNick Piggin 280203158cd7SNick Piggin /* Ok, it's mapped. Make sure it's up-to-date */ 280303158cd7SNick Piggin if (!PageUptodate(page)) { 280403158cd7SNick Piggin err = mapping->a_ops->readpage(NULL, page); 280503158cd7SNick Piggin if (err) { 280603158cd7SNick Piggin page_cache_release(page); 280703158cd7SNick Piggin goto out; 280803158cd7SNick Piggin } 280903158cd7SNick Piggin lock_page(page); 281003158cd7SNick Piggin if (!PageUptodate(page)) { 281103158cd7SNick Piggin err = -EIO; 281203158cd7SNick Piggin goto unlock; 281303158cd7SNick Piggin } 281403158cd7SNick Piggin if (page_has_buffers(page)) 281503158cd7SNick Piggin goto has_buffers; 281603158cd7SNick Piggin } 2817eebd2aa3SChristoph Lameter zero_user(page, offset, length); 281803158cd7SNick Piggin set_page_dirty(page); 281903158cd7SNick Piggin err = 0; 282003158cd7SNick Piggin 282103158cd7SNick Piggin unlock: 28221da177e4SLinus Torvalds unlock_page(page); 28231da177e4SLinus Torvalds page_cache_release(page); 28241da177e4SLinus Torvalds out: 282503158cd7SNick Piggin return err; 28261da177e4SLinus Torvalds } 28271da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_truncate_page); 28281da177e4SLinus Torvalds 28291da177e4SLinus Torvalds int block_truncate_page(struct address_space *mapping, 28301da177e4SLinus Torvalds loff_t from, get_block_t *get_block) 28311da177e4SLinus Torvalds { 28321da177e4SLinus Torvalds pgoff_t index = from >> PAGE_CACHE_SHIFT; 28331da177e4SLinus Torvalds unsigned offset = from & (PAGE_CACHE_SIZE-1); 28341da177e4SLinus Torvalds unsigned blocksize; 283554b21a79SAndrew Morton sector_t iblock; 28361da177e4SLinus Torvalds unsigned length, pos; 28371da177e4SLinus Torvalds struct inode *inode = mapping->host; 28381da177e4SLinus Torvalds struct page *page; 28391da177e4SLinus Torvalds struct buffer_head *bh; 28401da177e4SLinus Torvalds int err; 28411da177e4SLinus Torvalds 28421da177e4SLinus Torvalds blocksize = 1 << inode->i_blkbits; 28431da177e4SLinus Torvalds length = offset & (blocksize - 1); 28441da177e4SLinus Torvalds 28451da177e4SLinus Torvalds /* Block boundary? Nothing to do */ 28461da177e4SLinus Torvalds if (!length) 28471da177e4SLinus Torvalds return 0; 28481da177e4SLinus Torvalds 28491da177e4SLinus Torvalds length = blocksize - length; 285054b21a79SAndrew Morton iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 28511da177e4SLinus Torvalds 28521da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 28531da177e4SLinus Torvalds err = -ENOMEM; 28541da177e4SLinus Torvalds if (!page) 28551da177e4SLinus Torvalds goto out; 28561da177e4SLinus Torvalds 28571da177e4SLinus Torvalds if (!page_has_buffers(page)) 28581da177e4SLinus Torvalds create_empty_buffers(page, blocksize, 0); 28591da177e4SLinus Torvalds 28601da177e4SLinus Torvalds /* Find the buffer that contains "offset" */ 28611da177e4SLinus Torvalds bh = page_buffers(page); 28621da177e4SLinus Torvalds pos = blocksize; 28631da177e4SLinus Torvalds while (offset >= pos) { 28641da177e4SLinus Torvalds bh = bh->b_this_page; 28651da177e4SLinus Torvalds iblock++; 28661da177e4SLinus Torvalds pos += blocksize; 28671da177e4SLinus Torvalds } 28681da177e4SLinus Torvalds 28691da177e4SLinus Torvalds err = 0; 28701da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2871b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 28721da177e4SLinus Torvalds err = get_block(inode, iblock, bh, 0); 28731da177e4SLinus Torvalds if (err) 28741da177e4SLinus Torvalds goto unlock; 28751da177e4SLinus Torvalds /* unmapped? It's a hole - nothing to do */ 28761da177e4SLinus Torvalds if (!buffer_mapped(bh)) 28771da177e4SLinus Torvalds goto unlock; 28781da177e4SLinus Torvalds } 28791da177e4SLinus Torvalds 28801da177e4SLinus Torvalds /* Ok, it's mapped. Make sure it's up-to-date */ 28811da177e4SLinus Torvalds if (PageUptodate(page)) 28821da177e4SLinus Torvalds set_buffer_uptodate(bh); 28831da177e4SLinus Torvalds 288433a266ddSDavid Chinner if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { 28851da177e4SLinus Torvalds err = -EIO; 28861da177e4SLinus Torvalds ll_rw_block(READ, 1, &bh); 28871da177e4SLinus Torvalds wait_on_buffer(bh); 28881da177e4SLinus Torvalds /* Uhhuh. Read error. Complain and punt. */ 28891da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 28901da177e4SLinus Torvalds goto unlock; 28911da177e4SLinus Torvalds } 28921da177e4SLinus Torvalds 2893eebd2aa3SChristoph Lameter zero_user(page, offset, length); 28941da177e4SLinus Torvalds mark_buffer_dirty(bh); 28951da177e4SLinus Torvalds err = 0; 28961da177e4SLinus Torvalds 28971da177e4SLinus Torvalds unlock: 28981da177e4SLinus Torvalds unlock_page(page); 28991da177e4SLinus Torvalds page_cache_release(page); 29001da177e4SLinus Torvalds out: 29011da177e4SLinus Torvalds return err; 29021da177e4SLinus Torvalds } 29031fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_truncate_page); 29041da177e4SLinus Torvalds 29051da177e4SLinus Torvalds /* 29061da177e4SLinus Torvalds * The generic ->writepage function for buffer-backed address_spaces 29071da177e4SLinus Torvalds */ 29081b938c08SMatthew Wilcox int block_write_full_page(struct page *page, get_block_t *get_block, 29091b938c08SMatthew Wilcox struct writeback_control *wbc) 29101da177e4SLinus Torvalds { 29111da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 29121da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 29131da177e4SLinus Torvalds const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; 29141da177e4SLinus Torvalds unsigned offset; 29151da177e4SLinus Torvalds 29161da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 29171da177e4SLinus Torvalds if (page->index < end_index) 291835c80d5fSChris Mason return __block_write_full_page(inode, page, get_block, wbc, 29191b938c08SMatthew Wilcox end_buffer_async_write); 29201da177e4SLinus Torvalds 29211da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 29221da177e4SLinus Torvalds offset = i_size & (PAGE_CACHE_SIZE-1); 29231da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 29241da177e4SLinus Torvalds /* 29251da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 29261da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 29271da177e4SLinus Torvalds * freeable here, so the page does not leak. 29281da177e4SLinus Torvalds */ 2929d47992f8SLukas Czerner do_invalidatepage(page, 0, PAGE_CACHE_SIZE); 29301da177e4SLinus Torvalds unlock_page(page); 29311da177e4SLinus Torvalds return 0; /* don't care */ 29321da177e4SLinus Torvalds } 29331da177e4SLinus Torvalds 29341da177e4SLinus Torvalds /* 29351da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 29362a61aa40SAdam Buchbinder * writepage invocation because it may be mmapped. "A file is mapped 29371da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 29381da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 29391da177e4SLinus Torvalds * writes to that region are not written out to the file." 29401da177e4SLinus Torvalds */ 2941eebd2aa3SChristoph Lameter zero_user_segment(page, offset, PAGE_CACHE_SIZE); 29421b938c08SMatthew Wilcox return __block_write_full_page(inode, page, get_block, wbc, 294335c80d5fSChris Mason end_buffer_async_write); 294435c80d5fSChris Mason } 29451fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_write_full_page); 294635c80d5fSChris Mason 29471da177e4SLinus Torvalds sector_t generic_block_bmap(struct address_space *mapping, sector_t block, 29481da177e4SLinus Torvalds get_block_t *get_block) 29491da177e4SLinus Torvalds { 29501da177e4SLinus Torvalds struct buffer_head tmp; 29511da177e4SLinus Torvalds struct inode *inode = mapping->host; 29521da177e4SLinus Torvalds tmp.b_state = 0; 29531da177e4SLinus Torvalds tmp.b_blocknr = 0; 2954b0cf2321SBadari Pulavarty tmp.b_size = 1 << inode->i_blkbits; 29551da177e4SLinus Torvalds get_block(inode, block, &tmp, 0); 29561da177e4SLinus Torvalds return tmp.b_blocknr; 29571da177e4SLinus Torvalds } 29581fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_block_bmap); 29591da177e4SLinus Torvalds 29606712ecf8SNeilBrown static void end_bio_bh_io_sync(struct bio *bio, int err) 29611da177e4SLinus Torvalds { 29621da177e4SLinus Torvalds struct buffer_head *bh = bio->bi_private; 29631da177e4SLinus Torvalds 296408bafc03SKeith Mannthey if (unlikely (test_bit(BIO_QUIET,&bio->bi_flags))) 296508bafc03SKeith Mannthey set_bit(BH_Quiet, &bh->b_state); 296608bafc03SKeith Mannthey 29671da177e4SLinus Torvalds bh->b_end_io(bh, test_bit(BIO_UPTODATE, &bio->bi_flags)); 29681da177e4SLinus Torvalds bio_put(bio); 29691da177e4SLinus Torvalds } 29701da177e4SLinus Torvalds 297157302e0dSLinus Torvalds /* 297257302e0dSLinus Torvalds * This allows us to do IO even on the odd last sectors 297359d43914SAkinobu Mita * of a device, even if the block size is some multiple 297457302e0dSLinus Torvalds * of the physical sector size. 297557302e0dSLinus Torvalds * 297657302e0dSLinus Torvalds * We'll just truncate the bio to the size of the device, 297757302e0dSLinus Torvalds * and clear the end of the buffer head manually. 297857302e0dSLinus Torvalds * 297957302e0dSLinus Torvalds * Truly out-of-range accesses will turn into actual IO 298057302e0dSLinus Torvalds * errors, this only handles the "we need to be able to 298157302e0dSLinus Torvalds * do IO at the final sector" case. 298257302e0dSLinus Torvalds */ 29834db96b71SAkinobu Mita void guard_bio_eod(int rw, struct bio *bio) 298457302e0dSLinus Torvalds { 298557302e0dSLinus Torvalds sector_t maxsector; 298659d43914SAkinobu Mita struct bio_vec *bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 298759d43914SAkinobu Mita unsigned truncated_bytes; 298857302e0dSLinus Torvalds 298957302e0dSLinus Torvalds maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; 299057302e0dSLinus Torvalds if (!maxsector) 299157302e0dSLinus Torvalds return; 299257302e0dSLinus Torvalds 299357302e0dSLinus Torvalds /* 299457302e0dSLinus Torvalds * If the *whole* IO is past the end of the device, 299557302e0dSLinus Torvalds * let it through, and the IO layer will turn it into 299657302e0dSLinus Torvalds * an EIO. 299757302e0dSLinus Torvalds */ 29984f024f37SKent Overstreet if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 299957302e0dSLinus Torvalds return; 300057302e0dSLinus Torvalds 30014f024f37SKent Overstreet maxsector -= bio->bi_iter.bi_sector; 300259d43914SAkinobu Mita if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 300357302e0dSLinus Torvalds return; 300457302e0dSLinus Torvalds 300559d43914SAkinobu Mita /* Uhhuh. We've got a bio that straddles the device size! */ 300659d43914SAkinobu Mita truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9); 300757302e0dSLinus Torvalds 300857302e0dSLinus Torvalds /* Truncate the bio.. */ 300959d43914SAkinobu Mita bio->bi_iter.bi_size -= truncated_bytes; 301059d43914SAkinobu Mita bvec->bv_len -= truncated_bytes; 301157302e0dSLinus Torvalds 301257302e0dSLinus Torvalds /* ..and clear the end of the buffer for reads */ 301327d7c2a0SDan Carpenter if ((rw & RW_MASK) == READ) { 301459d43914SAkinobu Mita zero_user(bvec->bv_page, bvec->bv_offset + bvec->bv_len, 301559d43914SAkinobu Mita truncated_bytes); 301657302e0dSLinus Torvalds } 301757302e0dSLinus Torvalds } 301857302e0dSLinus Torvalds 3019b16b1debSTejun Heo static int submit_bh_wbc(int rw, struct buffer_head *bh, 3020b16b1debSTejun Heo unsigned long bio_flags, struct writeback_control *wbc) 30211da177e4SLinus Torvalds { 30221da177e4SLinus Torvalds struct bio *bio; 3023b16b1debSTejun Heo int ret = 0; 30241da177e4SLinus Torvalds 30251da177e4SLinus Torvalds BUG_ON(!buffer_locked(bh)); 30261da177e4SLinus Torvalds BUG_ON(!buffer_mapped(bh)); 30271da177e4SLinus Torvalds BUG_ON(!bh->b_end_io); 30288fb0e342SAneesh Kumar K.V BUG_ON(buffer_delay(bh)); 30298fb0e342SAneesh Kumar K.V BUG_ON(buffer_unwritten(bh)); 30301da177e4SLinus Torvalds 303148fd4f93SJens Axboe /* 303248fd4f93SJens Axboe * Only clear out a write error when rewriting 30331da177e4SLinus Torvalds */ 303448fd4f93SJens Axboe if (test_set_buffer_req(bh) && (rw & WRITE)) 30351da177e4SLinus Torvalds clear_buffer_write_io_error(bh); 30361da177e4SLinus Torvalds 30371da177e4SLinus Torvalds /* 30381da177e4SLinus Torvalds * from here on down, it's all bio -- do the initial mapping, 30391da177e4SLinus Torvalds * submit_bio -> generic_make_request may further map this bio around 30401da177e4SLinus Torvalds */ 30411da177e4SLinus Torvalds bio = bio_alloc(GFP_NOIO, 1); 30421da177e4SLinus Torvalds 3043*2a814908STejun Heo if (wbc) { 3044b16b1debSTejun Heo wbc_init_bio(wbc, bio); 3045*2a814908STejun Heo wbc_account_io(wbc, bh->b_page, bh->b_size); 3046*2a814908STejun Heo } 3047bafc0dbaSTejun Heo 30484f024f37SKent Overstreet bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); 30491da177e4SLinus Torvalds bio->bi_bdev = bh->b_bdev; 30501da177e4SLinus Torvalds bio->bi_io_vec[0].bv_page = bh->b_page; 30511da177e4SLinus Torvalds bio->bi_io_vec[0].bv_len = bh->b_size; 30521da177e4SLinus Torvalds bio->bi_io_vec[0].bv_offset = bh_offset(bh); 30531da177e4SLinus Torvalds 30541da177e4SLinus Torvalds bio->bi_vcnt = 1; 30554f024f37SKent Overstreet bio->bi_iter.bi_size = bh->b_size; 30561da177e4SLinus Torvalds 30571da177e4SLinus Torvalds bio->bi_end_io = end_bio_bh_io_sync; 30581da177e4SLinus Torvalds bio->bi_private = bh; 305971368511SDarrick J. Wong bio->bi_flags |= bio_flags; 30601da177e4SLinus Torvalds 306157302e0dSLinus Torvalds /* Take care of bh's that straddle the end of the device */ 306259d43914SAkinobu Mita guard_bio_eod(rw, bio); 306357302e0dSLinus Torvalds 3064877f962cSTheodore Ts'o if (buffer_meta(bh)) 3065877f962cSTheodore Ts'o rw |= REQ_META; 3066877f962cSTheodore Ts'o if (buffer_prio(bh)) 3067877f962cSTheodore Ts'o rw |= REQ_PRIO; 3068877f962cSTheodore Ts'o 30691da177e4SLinus Torvalds submit_bio(rw, bio); 3070f6454b04SJulia Lawall return 0; 30711da177e4SLinus Torvalds } 3072bafc0dbaSTejun Heo 3073bafc0dbaSTejun Heo int _submit_bh(int rw, struct buffer_head *bh, unsigned long bio_flags) 3074bafc0dbaSTejun Heo { 3075b16b1debSTejun Heo return submit_bh_wbc(rw, bh, bio_flags, NULL); 3076bafc0dbaSTejun Heo } 307771368511SDarrick J. Wong EXPORT_SYMBOL_GPL(_submit_bh); 307871368511SDarrick J. Wong 307971368511SDarrick J. Wong int submit_bh(int rw, struct buffer_head *bh) 308071368511SDarrick J. Wong { 3081b16b1debSTejun Heo return submit_bh_wbc(rw, bh, 0, NULL); 308271368511SDarrick J. Wong } 30831fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(submit_bh); 30841da177e4SLinus Torvalds 30851da177e4SLinus Torvalds /** 30861da177e4SLinus Torvalds * ll_rw_block: low-level access to block devices (DEPRECATED) 30879cb569d6SChristoph Hellwig * @rw: whether to %READ or %WRITE or maybe %READA (readahead) 30881da177e4SLinus Torvalds * @nr: number of &struct buffer_heads in the array 30891da177e4SLinus Torvalds * @bhs: array of pointers to &struct buffer_head 30901da177e4SLinus Torvalds * 3091a7662236SJan Kara * ll_rw_block() takes an array of pointers to &struct buffer_heads, and 3092a7662236SJan Kara * requests an I/O operation on them, either a %READ or a %WRITE. The third 30939cb569d6SChristoph Hellwig * %READA option is described in the documentation for generic_make_request() 30949cb569d6SChristoph Hellwig * which ll_rw_block() calls. 30951da177e4SLinus Torvalds * 30961da177e4SLinus Torvalds * This function drops any buffer that it cannot get a lock on (with the 30979cb569d6SChristoph Hellwig * BH_Lock state bit), any buffer that appears to be clean when doing a write 30989cb569d6SChristoph Hellwig * request, and any buffer that appears to be up-to-date when doing read 30999cb569d6SChristoph Hellwig * request. Further it marks as clean buffers that are processed for 31009cb569d6SChristoph Hellwig * writing (the buffer cache won't assume that they are actually clean 31019cb569d6SChristoph Hellwig * until the buffer gets unlocked). 31021da177e4SLinus Torvalds * 31031da177e4SLinus Torvalds * ll_rw_block sets b_end_io to simple completion handler that marks 3104e227867fSMasanari Iida * the buffer up-to-date (if appropriate), unlocks the buffer and wakes 31051da177e4SLinus Torvalds * any waiters. 31061da177e4SLinus Torvalds * 31071da177e4SLinus Torvalds * All of the buffers must be for the same device, and must also be a 31081da177e4SLinus Torvalds * multiple of the current approved size for the device. 31091da177e4SLinus Torvalds */ 31101da177e4SLinus Torvalds void ll_rw_block(int rw, int nr, struct buffer_head *bhs[]) 31111da177e4SLinus Torvalds { 31121da177e4SLinus Torvalds int i; 31131da177e4SLinus Torvalds 31141da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 31151da177e4SLinus Torvalds struct buffer_head *bh = bhs[i]; 31161da177e4SLinus Torvalds 31179cb569d6SChristoph Hellwig if (!trylock_buffer(bh)) 31181da177e4SLinus Torvalds continue; 31199cb569d6SChristoph Hellwig if (rw == WRITE) { 31201da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 312176c3073aSakpm@osdl.org bh->b_end_io = end_buffer_write_sync; 3122e60e5c50SOGAWA Hirofumi get_bh(bh); 31231da177e4SLinus Torvalds submit_bh(WRITE, bh); 31241da177e4SLinus Torvalds continue; 31251da177e4SLinus Torvalds } 31261da177e4SLinus Torvalds } else { 31271da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 312876c3073aSakpm@osdl.org bh->b_end_io = end_buffer_read_sync; 3129e60e5c50SOGAWA Hirofumi get_bh(bh); 31301da177e4SLinus Torvalds submit_bh(rw, bh); 31311da177e4SLinus Torvalds continue; 31321da177e4SLinus Torvalds } 31331da177e4SLinus Torvalds } 31341da177e4SLinus Torvalds unlock_buffer(bh); 31351da177e4SLinus Torvalds } 31361da177e4SLinus Torvalds } 31371fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(ll_rw_block); 31381da177e4SLinus Torvalds 31399cb569d6SChristoph Hellwig void write_dirty_buffer(struct buffer_head *bh, int rw) 31409cb569d6SChristoph Hellwig { 31419cb569d6SChristoph Hellwig lock_buffer(bh); 31429cb569d6SChristoph Hellwig if (!test_clear_buffer_dirty(bh)) { 31439cb569d6SChristoph Hellwig unlock_buffer(bh); 31449cb569d6SChristoph Hellwig return; 31459cb569d6SChristoph Hellwig } 31469cb569d6SChristoph Hellwig bh->b_end_io = end_buffer_write_sync; 31479cb569d6SChristoph Hellwig get_bh(bh); 31489cb569d6SChristoph Hellwig submit_bh(rw, bh); 31499cb569d6SChristoph Hellwig } 31509cb569d6SChristoph Hellwig EXPORT_SYMBOL(write_dirty_buffer); 31519cb569d6SChristoph Hellwig 31521da177e4SLinus Torvalds /* 31531da177e4SLinus Torvalds * For a data-integrity writeout, we need to wait upon any in-progress I/O 31541da177e4SLinus Torvalds * and then start new I/O and then wait upon it. The caller must have a ref on 31551da177e4SLinus Torvalds * the buffer_head. 31561da177e4SLinus Torvalds */ 315787e99511SChristoph Hellwig int __sync_dirty_buffer(struct buffer_head *bh, int rw) 31581da177e4SLinus Torvalds { 31591da177e4SLinus Torvalds int ret = 0; 31601da177e4SLinus Torvalds 31611da177e4SLinus Torvalds WARN_ON(atomic_read(&bh->b_count) < 1); 31621da177e4SLinus Torvalds lock_buffer(bh); 31631da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 31641da177e4SLinus Torvalds get_bh(bh); 31651da177e4SLinus Torvalds bh->b_end_io = end_buffer_write_sync; 316687e99511SChristoph Hellwig ret = submit_bh(rw, bh); 31671da177e4SLinus Torvalds wait_on_buffer(bh); 31681da177e4SLinus Torvalds if (!ret && !buffer_uptodate(bh)) 31691da177e4SLinus Torvalds ret = -EIO; 31701da177e4SLinus Torvalds } else { 31711da177e4SLinus Torvalds unlock_buffer(bh); 31721da177e4SLinus Torvalds } 31731da177e4SLinus Torvalds return ret; 31741da177e4SLinus Torvalds } 317587e99511SChristoph Hellwig EXPORT_SYMBOL(__sync_dirty_buffer); 317687e99511SChristoph Hellwig 317787e99511SChristoph Hellwig int sync_dirty_buffer(struct buffer_head *bh) 317887e99511SChristoph Hellwig { 317987e99511SChristoph Hellwig return __sync_dirty_buffer(bh, WRITE_SYNC); 318087e99511SChristoph Hellwig } 31811fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(sync_dirty_buffer); 31821da177e4SLinus Torvalds 31831da177e4SLinus Torvalds /* 31841da177e4SLinus Torvalds * try_to_free_buffers() checks if all the buffers on this particular page 31851da177e4SLinus Torvalds * are unused, and releases them if so. 31861da177e4SLinus Torvalds * 31871da177e4SLinus Torvalds * Exclusion against try_to_free_buffers may be obtained by either 31881da177e4SLinus Torvalds * locking the page or by holding its mapping's private_lock. 31891da177e4SLinus Torvalds * 31901da177e4SLinus Torvalds * If the page is dirty but all the buffers are clean then we need to 31911da177e4SLinus Torvalds * be sure to mark the page clean as well. This is because the page 31921da177e4SLinus Torvalds * may be against a block device, and a later reattachment of buffers 31931da177e4SLinus Torvalds * to a dirty page will set *all* buffers dirty. Which would corrupt 31941da177e4SLinus Torvalds * filesystem data on the same device. 31951da177e4SLinus Torvalds * 31961da177e4SLinus Torvalds * The same applies to regular filesystem pages: if all the buffers are 31971da177e4SLinus Torvalds * clean then we set the page clean and proceed. To do that, we require 31981da177e4SLinus Torvalds * total exclusion from __set_page_dirty_buffers(). That is obtained with 31991da177e4SLinus Torvalds * private_lock. 32001da177e4SLinus Torvalds * 32011da177e4SLinus Torvalds * try_to_free_buffers() is non-blocking. 32021da177e4SLinus Torvalds */ 32031da177e4SLinus Torvalds static inline int buffer_busy(struct buffer_head *bh) 32041da177e4SLinus Torvalds { 32051da177e4SLinus Torvalds return atomic_read(&bh->b_count) | 32061da177e4SLinus Torvalds (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); 32071da177e4SLinus Torvalds } 32081da177e4SLinus Torvalds 32091da177e4SLinus Torvalds static int 32101da177e4SLinus Torvalds drop_buffers(struct page *page, struct buffer_head **buffers_to_free) 32111da177e4SLinus Torvalds { 32121da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 32131da177e4SLinus Torvalds struct buffer_head *bh; 32141da177e4SLinus Torvalds 32151da177e4SLinus Torvalds bh = head; 32161da177e4SLinus Torvalds do { 3217de7d5a3bSakpm@osdl.org if (buffer_write_io_error(bh) && page->mapping) 32181da177e4SLinus Torvalds set_bit(AS_EIO, &page->mapping->flags); 32191da177e4SLinus Torvalds if (buffer_busy(bh)) 32201da177e4SLinus Torvalds goto failed; 32211da177e4SLinus Torvalds bh = bh->b_this_page; 32221da177e4SLinus Torvalds } while (bh != head); 32231da177e4SLinus Torvalds 32241da177e4SLinus Torvalds do { 32251da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 32261da177e4SLinus Torvalds 3227535ee2fbSJan Kara if (bh->b_assoc_map) 32281da177e4SLinus Torvalds __remove_assoc_queue(bh); 32291da177e4SLinus Torvalds bh = next; 32301da177e4SLinus Torvalds } while (bh != head); 32311da177e4SLinus Torvalds *buffers_to_free = head; 32321da177e4SLinus Torvalds __clear_page_buffers(page); 32331da177e4SLinus Torvalds return 1; 32341da177e4SLinus Torvalds failed: 32351da177e4SLinus Torvalds return 0; 32361da177e4SLinus Torvalds } 32371da177e4SLinus Torvalds 32381da177e4SLinus Torvalds int try_to_free_buffers(struct page *page) 32391da177e4SLinus Torvalds { 32401da177e4SLinus Torvalds struct address_space * const mapping = page->mapping; 32411da177e4SLinus Torvalds struct buffer_head *buffers_to_free = NULL; 32421da177e4SLinus Torvalds int ret = 0; 32431da177e4SLinus Torvalds 32441da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 3245ecdfc978SLinus Torvalds if (PageWriteback(page)) 32461da177e4SLinus Torvalds return 0; 32471da177e4SLinus Torvalds 32481da177e4SLinus Torvalds if (mapping == NULL) { /* can this still happen? */ 32491da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 32501da177e4SLinus Torvalds goto out; 32511da177e4SLinus Torvalds } 32521da177e4SLinus Torvalds 32531da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 32541da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 3255ecdfc978SLinus Torvalds 3256ecdfc978SLinus Torvalds /* 3257ecdfc978SLinus Torvalds * If the filesystem writes its buffers by hand (eg ext3) 3258ecdfc978SLinus Torvalds * then we can have clean buffers against a dirty page. We 3259ecdfc978SLinus Torvalds * clean the page here; otherwise the VM will never notice 3260ecdfc978SLinus Torvalds * that the filesystem did any IO at all. 3261ecdfc978SLinus Torvalds * 3262ecdfc978SLinus Torvalds * Also, during truncate, discard_buffer will have marked all 3263ecdfc978SLinus Torvalds * the page's buffers clean. We discover that here and clean 3264ecdfc978SLinus Torvalds * the page also. 326587df7241SNick Piggin * 326687df7241SNick Piggin * private_lock must be held over this entire operation in order 326787df7241SNick Piggin * to synchronise against __set_page_dirty_buffers and prevent the 326887df7241SNick Piggin * dirty bit from being lost. 3269ecdfc978SLinus Torvalds */ 327011f81becSTejun Heo if (ret) 327111f81becSTejun Heo cancel_dirty_page(page); 327287df7241SNick Piggin spin_unlock(&mapping->private_lock); 32731da177e4SLinus Torvalds out: 32741da177e4SLinus Torvalds if (buffers_to_free) { 32751da177e4SLinus Torvalds struct buffer_head *bh = buffers_to_free; 32761da177e4SLinus Torvalds 32771da177e4SLinus Torvalds do { 32781da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 32791da177e4SLinus Torvalds free_buffer_head(bh); 32801da177e4SLinus Torvalds bh = next; 32811da177e4SLinus Torvalds } while (bh != buffers_to_free); 32821da177e4SLinus Torvalds } 32831da177e4SLinus Torvalds return ret; 32841da177e4SLinus Torvalds } 32851da177e4SLinus Torvalds EXPORT_SYMBOL(try_to_free_buffers); 32861da177e4SLinus Torvalds 32871da177e4SLinus Torvalds /* 32881da177e4SLinus Torvalds * There are no bdflush tunables left. But distributions are 32891da177e4SLinus Torvalds * still running obsolete flush daemons, so we terminate them here. 32901da177e4SLinus Torvalds * 32911da177e4SLinus Torvalds * Use of bdflush() is deprecated and will be removed in a future kernel. 32925b0830cbSJens Axboe * The `flush-X' kernel threads fully replace bdflush daemons and this call. 32931da177e4SLinus Torvalds */ 3294bdc480e3SHeiko Carstens SYSCALL_DEFINE2(bdflush, int, func, long, data) 32951da177e4SLinus Torvalds { 32961da177e4SLinus Torvalds static int msg_count; 32971da177e4SLinus Torvalds 32981da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 32991da177e4SLinus Torvalds return -EPERM; 33001da177e4SLinus Torvalds 33011da177e4SLinus Torvalds if (msg_count < 5) { 33021da177e4SLinus Torvalds msg_count++; 33031da177e4SLinus Torvalds printk(KERN_INFO 33041da177e4SLinus Torvalds "warning: process `%s' used the obsolete bdflush" 33051da177e4SLinus Torvalds " system call\n", current->comm); 33061da177e4SLinus Torvalds printk(KERN_INFO "Fix your initscripts?\n"); 33071da177e4SLinus Torvalds } 33081da177e4SLinus Torvalds 33091da177e4SLinus Torvalds if (func == 1) 33101da177e4SLinus Torvalds do_exit(0); 33111da177e4SLinus Torvalds return 0; 33121da177e4SLinus Torvalds } 33131da177e4SLinus Torvalds 33141da177e4SLinus Torvalds /* 33151da177e4SLinus Torvalds * Buffer-head allocation 33161da177e4SLinus Torvalds */ 3317a0a9b043SShai Fultheim static struct kmem_cache *bh_cachep __read_mostly; 33181da177e4SLinus Torvalds 33191da177e4SLinus Torvalds /* 33201da177e4SLinus Torvalds * Once the number of bh's in the machine exceeds this level, we start 33211da177e4SLinus Torvalds * stripping them in writeback. 33221da177e4SLinus Torvalds */ 332343be594aSZhang Yanfei static unsigned long max_buffer_heads; 33241da177e4SLinus Torvalds 33251da177e4SLinus Torvalds int buffer_heads_over_limit; 33261da177e4SLinus Torvalds 33271da177e4SLinus Torvalds struct bh_accounting { 33281da177e4SLinus Torvalds int nr; /* Number of live bh's */ 33291da177e4SLinus Torvalds int ratelimit; /* Limit cacheline bouncing */ 33301da177e4SLinus Torvalds }; 33311da177e4SLinus Torvalds 33321da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; 33331da177e4SLinus Torvalds 33341da177e4SLinus Torvalds static void recalc_bh_state(void) 33351da177e4SLinus Torvalds { 33361da177e4SLinus Torvalds int i; 33371da177e4SLinus Torvalds int tot = 0; 33381da177e4SLinus Torvalds 3339ee1be862SChristoph Lameter if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) 33401da177e4SLinus Torvalds return; 3341c7b92516SChristoph Lameter __this_cpu_write(bh_accounting.ratelimit, 0); 33428a143426SEric Dumazet for_each_online_cpu(i) 33431da177e4SLinus Torvalds tot += per_cpu(bh_accounting, i).nr; 33441da177e4SLinus Torvalds buffer_heads_over_limit = (tot > max_buffer_heads); 33451da177e4SLinus Torvalds } 33461da177e4SLinus Torvalds 3347dd0fc66fSAl Viro struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) 33481da177e4SLinus Torvalds { 3349019b4d12SRichard Kennedy struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); 33501da177e4SLinus Torvalds if (ret) { 3351a35afb83SChristoph Lameter INIT_LIST_HEAD(&ret->b_assoc_buffers); 3352c7b92516SChristoph Lameter preempt_disable(); 3353c7b92516SChristoph Lameter __this_cpu_inc(bh_accounting.nr); 33541da177e4SLinus Torvalds recalc_bh_state(); 3355c7b92516SChristoph Lameter preempt_enable(); 33561da177e4SLinus Torvalds } 33571da177e4SLinus Torvalds return ret; 33581da177e4SLinus Torvalds } 33591da177e4SLinus Torvalds EXPORT_SYMBOL(alloc_buffer_head); 33601da177e4SLinus Torvalds 33611da177e4SLinus Torvalds void free_buffer_head(struct buffer_head *bh) 33621da177e4SLinus Torvalds { 33631da177e4SLinus Torvalds BUG_ON(!list_empty(&bh->b_assoc_buffers)); 33641da177e4SLinus Torvalds kmem_cache_free(bh_cachep, bh); 3365c7b92516SChristoph Lameter preempt_disable(); 3366c7b92516SChristoph Lameter __this_cpu_dec(bh_accounting.nr); 33671da177e4SLinus Torvalds recalc_bh_state(); 3368c7b92516SChristoph Lameter preempt_enable(); 33691da177e4SLinus Torvalds } 33701da177e4SLinus Torvalds EXPORT_SYMBOL(free_buffer_head); 33711da177e4SLinus Torvalds 33721da177e4SLinus Torvalds static void buffer_exit_cpu(int cpu) 33731da177e4SLinus Torvalds { 33741da177e4SLinus Torvalds int i; 33751da177e4SLinus Torvalds struct bh_lru *b = &per_cpu(bh_lrus, cpu); 33761da177e4SLinus Torvalds 33771da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 33781da177e4SLinus Torvalds brelse(b->bhs[i]); 33791da177e4SLinus Torvalds b->bhs[i] = NULL; 33801da177e4SLinus Torvalds } 3381c7b92516SChristoph Lameter this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); 33828a143426SEric Dumazet per_cpu(bh_accounting, cpu).nr = 0; 33831da177e4SLinus Torvalds } 33841da177e4SLinus Torvalds 33851da177e4SLinus Torvalds static int buffer_cpu_notify(struct notifier_block *self, 33861da177e4SLinus Torvalds unsigned long action, void *hcpu) 33871da177e4SLinus Torvalds { 33888bb78442SRafael J. Wysocki if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) 33891da177e4SLinus Torvalds buffer_exit_cpu((unsigned long)hcpu); 33901da177e4SLinus Torvalds return NOTIFY_OK; 33911da177e4SLinus Torvalds } 33921da177e4SLinus Torvalds 3393389d1b08SAneesh Kumar K.V /** 3394a6b91919SRandy Dunlap * bh_uptodate_or_lock - Test whether the buffer is uptodate 3395389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3396389d1b08SAneesh Kumar K.V * 3397389d1b08SAneesh Kumar K.V * Return true if the buffer is up-to-date and false, 3398389d1b08SAneesh Kumar K.V * with the buffer locked, if not. 3399389d1b08SAneesh Kumar K.V */ 3400389d1b08SAneesh Kumar K.V int bh_uptodate_or_lock(struct buffer_head *bh) 3401389d1b08SAneesh Kumar K.V { 3402389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) { 3403389d1b08SAneesh Kumar K.V lock_buffer(bh); 3404389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) 3405389d1b08SAneesh Kumar K.V return 0; 3406389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3407389d1b08SAneesh Kumar K.V } 3408389d1b08SAneesh Kumar K.V return 1; 3409389d1b08SAneesh Kumar K.V } 3410389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_uptodate_or_lock); 3411389d1b08SAneesh Kumar K.V 3412389d1b08SAneesh Kumar K.V /** 3413a6b91919SRandy Dunlap * bh_submit_read - Submit a locked buffer for reading 3414389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3415389d1b08SAneesh Kumar K.V * 3416389d1b08SAneesh Kumar K.V * Returns zero on success and -EIO on error. 3417389d1b08SAneesh Kumar K.V */ 3418389d1b08SAneesh Kumar K.V int bh_submit_read(struct buffer_head *bh) 3419389d1b08SAneesh Kumar K.V { 3420389d1b08SAneesh Kumar K.V BUG_ON(!buffer_locked(bh)); 3421389d1b08SAneesh Kumar K.V 3422389d1b08SAneesh Kumar K.V if (buffer_uptodate(bh)) { 3423389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3424389d1b08SAneesh Kumar K.V return 0; 3425389d1b08SAneesh Kumar K.V } 3426389d1b08SAneesh Kumar K.V 3427389d1b08SAneesh Kumar K.V get_bh(bh); 3428389d1b08SAneesh Kumar K.V bh->b_end_io = end_buffer_read_sync; 3429389d1b08SAneesh Kumar K.V submit_bh(READ, bh); 3430389d1b08SAneesh Kumar K.V wait_on_buffer(bh); 3431389d1b08SAneesh Kumar K.V if (buffer_uptodate(bh)) 3432389d1b08SAneesh Kumar K.V return 0; 3433389d1b08SAneesh Kumar K.V return -EIO; 3434389d1b08SAneesh Kumar K.V } 3435389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_submit_read); 3436389d1b08SAneesh Kumar K.V 34371da177e4SLinus Torvalds void __init buffer_init(void) 34381da177e4SLinus Torvalds { 343943be594aSZhang Yanfei unsigned long nrpages; 34401da177e4SLinus Torvalds 3441b98938c3SChristoph Lameter bh_cachep = kmem_cache_create("buffer_head", 3442b98938c3SChristoph Lameter sizeof(struct buffer_head), 0, 3443b98938c3SChristoph Lameter (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| 3444b98938c3SChristoph Lameter SLAB_MEM_SPREAD), 3445019b4d12SRichard Kennedy NULL); 34461da177e4SLinus Torvalds 34471da177e4SLinus Torvalds /* 34481da177e4SLinus Torvalds * Limit the bh occupancy to 10% of ZONE_NORMAL 34491da177e4SLinus Torvalds */ 34501da177e4SLinus Torvalds nrpages = (nr_free_buffer_pages() * 10) / 100; 34511da177e4SLinus Torvalds max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); 34521da177e4SLinus Torvalds hotcpu_notifier(buffer_cpu_notify, 0); 34531da177e4SLinus Torvalds } 3454