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> 331da177e4SLinus Torvalds #include <linux/writeback.h> 341da177e4SLinus Torvalds #include <linux/hash.h> 351da177e4SLinus Torvalds #include <linux/suspend.h> 361da177e4SLinus Torvalds #include <linux/buffer_head.h> 3755e829afSAndrew Morton #include <linux/task_io_accounting_ops.h> 381da177e4SLinus Torvalds #include <linux/bio.h> 391da177e4SLinus Torvalds #include <linux/notifier.h> 401da177e4SLinus Torvalds #include <linux/cpu.h> 411da177e4SLinus Torvalds #include <linux/bitops.h> 421da177e4SLinus Torvalds #include <linux/mpage.h> 43fb1c8f93SIngo Molnar #include <linux/bit_spinlock.h> 445305cb83STejun Heo #include <trace/events/block.h> 451da177e4SLinus Torvalds 461da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); 471da177e4SLinus Torvalds 481da177e4SLinus Torvalds #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) 491da177e4SLinus Torvalds 50a3f3c29cSYan Hong void init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private) 511da177e4SLinus Torvalds { 521da177e4SLinus Torvalds bh->b_end_io = handler; 531da177e4SLinus Torvalds bh->b_private = private; 541da177e4SLinus Torvalds } 551fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(init_buffer); 561da177e4SLinus Torvalds 57f0059afdSTejun Heo inline void touch_buffer(struct buffer_head *bh) 58f0059afdSTejun Heo { 595305cb83STejun Heo trace_block_touch_buffer(bh); 60f0059afdSTejun Heo mark_page_accessed(bh->b_page); 61f0059afdSTejun Heo } 62f0059afdSTejun Heo EXPORT_SYMBOL(touch_buffer); 63f0059afdSTejun Heo 647eaceaccSJens Axboe static int sleep_on_buffer(void *word) 651da177e4SLinus Torvalds { 661da177e4SLinus Torvalds io_schedule(); 671da177e4SLinus Torvalds return 0; 681da177e4SLinus Torvalds } 691da177e4SLinus Torvalds 70fc9b52cdSHarvey Harrison void __lock_buffer(struct buffer_head *bh) 711da177e4SLinus Torvalds { 727eaceaccSJens Axboe wait_on_bit_lock(&bh->b_state, BH_Lock, sleep_on_buffer, 731da177e4SLinus Torvalds TASK_UNINTERRUPTIBLE); 741da177e4SLinus Torvalds } 751da177e4SLinus Torvalds EXPORT_SYMBOL(__lock_buffer); 761da177e4SLinus Torvalds 77fc9b52cdSHarvey Harrison void unlock_buffer(struct buffer_head *bh) 781da177e4SLinus Torvalds { 7951b07fc3SNick Piggin clear_bit_unlock(BH_Lock, &bh->b_state); 804e857c58SPeter Zijlstra smp_mb__after_atomic(); 811da177e4SLinus Torvalds wake_up_bit(&bh->b_state, BH_Lock); 821da177e4SLinus Torvalds } 831fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(unlock_buffer); 841da177e4SLinus Torvalds 851da177e4SLinus Torvalds /* 86b4597226SMel Gorman * Returns if the page has dirty or writeback buffers. If all the buffers 87b4597226SMel Gorman * are unlocked and clean then the PageDirty information is stale. If 88b4597226SMel Gorman * any of the pages are locked, it is assumed they are locked for IO. 89b4597226SMel Gorman */ 90b4597226SMel Gorman void buffer_check_dirty_writeback(struct page *page, 91b4597226SMel Gorman bool *dirty, bool *writeback) 92b4597226SMel Gorman { 93b4597226SMel Gorman struct buffer_head *head, *bh; 94b4597226SMel Gorman *dirty = false; 95b4597226SMel Gorman *writeback = false; 96b4597226SMel Gorman 97b4597226SMel Gorman BUG_ON(!PageLocked(page)); 98b4597226SMel Gorman 99b4597226SMel Gorman if (!page_has_buffers(page)) 100b4597226SMel Gorman return; 101b4597226SMel Gorman 102b4597226SMel Gorman if (PageWriteback(page)) 103b4597226SMel Gorman *writeback = true; 104b4597226SMel Gorman 105b4597226SMel Gorman head = page_buffers(page); 106b4597226SMel Gorman bh = head; 107b4597226SMel Gorman do { 108b4597226SMel Gorman if (buffer_locked(bh)) 109b4597226SMel Gorman *writeback = true; 110b4597226SMel Gorman 111b4597226SMel Gorman if (buffer_dirty(bh)) 112b4597226SMel Gorman *dirty = true; 113b4597226SMel Gorman 114b4597226SMel Gorman bh = bh->b_this_page; 115b4597226SMel Gorman } while (bh != head); 116b4597226SMel Gorman } 117b4597226SMel Gorman EXPORT_SYMBOL(buffer_check_dirty_writeback); 118b4597226SMel Gorman 119b4597226SMel Gorman /* 1201da177e4SLinus Torvalds * Block until a buffer comes unlocked. This doesn't stop it 1211da177e4SLinus Torvalds * from becoming locked again - you have to lock it yourself 1221da177e4SLinus Torvalds * if you want to preserve its state. 1231da177e4SLinus Torvalds */ 1241da177e4SLinus Torvalds void __wait_on_buffer(struct buffer_head * bh) 1251da177e4SLinus Torvalds { 1267eaceaccSJens Axboe wait_on_bit(&bh->b_state, BH_Lock, sleep_on_buffer, TASK_UNINTERRUPTIBLE); 1271da177e4SLinus Torvalds } 1281fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__wait_on_buffer); 1291da177e4SLinus Torvalds 1301da177e4SLinus Torvalds static void 1311da177e4SLinus Torvalds __clear_page_buffers(struct page *page) 1321da177e4SLinus Torvalds { 1331da177e4SLinus Torvalds ClearPagePrivate(page); 1344c21e2f2SHugh Dickins set_page_private(page, 0); 1351da177e4SLinus Torvalds page_cache_release(page); 1361da177e4SLinus Torvalds } 1371da177e4SLinus Torvalds 13808bafc03SKeith Mannthey 13908bafc03SKeith Mannthey static int quiet_error(struct buffer_head *bh) 14008bafc03SKeith Mannthey { 14108bafc03SKeith Mannthey if (!test_bit(BH_Quiet, &bh->b_state) && printk_ratelimit()) 14208bafc03SKeith Mannthey return 0; 14308bafc03SKeith Mannthey return 1; 14408bafc03SKeith Mannthey } 14508bafc03SKeith Mannthey 14608bafc03SKeith Mannthey 1471da177e4SLinus Torvalds static void buffer_io_error(struct buffer_head *bh) 1481da177e4SLinus Torvalds { 1491da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 1501da177e4SLinus Torvalds printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n", 1511da177e4SLinus Torvalds bdevname(bh->b_bdev, b), 1521da177e4SLinus Torvalds (unsigned long long)bh->b_blocknr); 1531da177e4SLinus Torvalds } 1541da177e4SLinus Torvalds 1551da177e4SLinus Torvalds /* 15668671f35SDmitry Monakhov * End-of-IO handler helper function which does not touch the bh after 15768671f35SDmitry Monakhov * unlocking it. 15868671f35SDmitry Monakhov * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but 15968671f35SDmitry Monakhov * a race there is benign: unlock_buffer() only use the bh's address for 16068671f35SDmitry Monakhov * hashing after unlocking the buffer, so it doesn't actually touch the bh 16168671f35SDmitry Monakhov * itself. 1621da177e4SLinus Torvalds */ 16368671f35SDmitry Monakhov static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) 1641da177e4SLinus Torvalds { 1651da177e4SLinus Torvalds if (uptodate) { 1661da177e4SLinus Torvalds set_buffer_uptodate(bh); 1671da177e4SLinus Torvalds } else { 1681da177e4SLinus Torvalds /* This happens, due to failed READA attempts. */ 1691da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1701da177e4SLinus Torvalds } 1711da177e4SLinus Torvalds unlock_buffer(bh); 17268671f35SDmitry Monakhov } 17368671f35SDmitry Monakhov 17468671f35SDmitry Monakhov /* 17568671f35SDmitry Monakhov * Default synchronous end-of-IO handler.. Just mark it up-to-date and 17668671f35SDmitry Monakhov * unlock the buffer. This is what ll_rw_block uses too. 17768671f35SDmitry Monakhov */ 17868671f35SDmitry Monakhov void end_buffer_read_sync(struct buffer_head *bh, int uptodate) 17968671f35SDmitry Monakhov { 18068671f35SDmitry Monakhov __end_buffer_read_notouch(bh, uptodate); 1811da177e4SLinus Torvalds put_bh(bh); 1821da177e4SLinus Torvalds } 1831fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_read_sync); 1841da177e4SLinus Torvalds 1851da177e4SLinus Torvalds void end_buffer_write_sync(struct buffer_head *bh, int uptodate) 1861da177e4SLinus Torvalds { 1871da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds if (uptodate) { 1901da177e4SLinus Torvalds set_buffer_uptodate(bh); 1911da177e4SLinus Torvalds } else { 1920edd55faSChristoph Hellwig if (!quiet_error(bh)) { 1931da177e4SLinus Torvalds buffer_io_error(bh); 1941da177e4SLinus Torvalds printk(KERN_WARNING "lost page write due to " 1951da177e4SLinus Torvalds "I/O error on %s\n", 1961da177e4SLinus Torvalds bdevname(bh->b_bdev, b)); 1971da177e4SLinus Torvalds } 1981da177e4SLinus Torvalds set_buffer_write_io_error(bh); 1991da177e4SLinus Torvalds clear_buffer_uptodate(bh); 2001da177e4SLinus Torvalds } 2011da177e4SLinus Torvalds unlock_buffer(bh); 2021da177e4SLinus Torvalds put_bh(bh); 2031da177e4SLinus Torvalds } 2041fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_write_sync); 2051da177e4SLinus Torvalds 2061da177e4SLinus Torvalds /* 2071da177e4SLinus Torvalds * Various filesystems appear to want __find_get_block to be non-blocking. 2081da177e4SLinus Torvalds * But it's the page lock which protects the buffers. To get around this, 2091da177e4SLinus Torvalds * we get exclusion from try_to_free_buffers with the blockdev mapping's 2101da177e4SLinus Torvalds * private_lock. 2111da177e4SLinus Torvalds * 2121da177e4SLinus Torvalds * Hack idea: for the blockdev mapping, i_bufferlist_lock contention 2131da177e4SLinus Torvalds * may be quite high. This code could TryLock the page, and if that 2141da177e4SLinus Torvalds * succeeds, there is no need to take private_lock. (But if 2151da177e4SLinus Torvalds * private_lock is contended then so is mapping->tree_lock). 2161da177e4SLinus Torvalds */ 2171da177e4SLinus Torvalds static struct buffer_head * 218385fd4c5SCoywolf Qi Hunt __find_get_block_slow(struct block_device *bdev, sector_t block) 2191da177e4SLinus Torvalds { 2201da177e4SLinus Torvalds struct inode *bd_inode = bdev->bd_inode; 2211da177e4SLinus Torvalds struct address_space *bd_mapping = bd_inode->i_mapping; 2221da177e4SLinus Torvalds struct buffer_head *ret = NULL; 2231da177e4SLinus Torvalds pgoff_t index; 2241da177e4SLinus Torvalds struct buffer_head *bh; 2251da177e4SLinus Torvalds struct buffer_head *head; 2261da177e4SLinus Torvalds struct page *page; 2271da177e4SLinus Torvalds int all_mapped = 1; 2281da177e4SLinus Torvalds 2291da177e4SLinus Torvalds index = block >> (PAGE_CACHE_SHIFT - bd_inode->i_blkbits); 2301da177e4SLinus Torvalds page = find_get_page(bd_mapping, index); 2311da177e4SLinus Torvalds if (!page) 2321da177e4SLinus Torvalds goto out; 2331da177e4SLinus Torvalds 2341da177e4SLinus Torvalds spin_lock(&bd_mapping->private_lock); 2351da177e4SLinus Torvalds if (!page_has_buffers(page)) 2361da177e4SLinus Torvalds goto out_unlock; 2371da177e4SLinus Torvalds head = page_buffers(page); 2381da177e4SLinus Torvalds bh = head; 2391da177e4SLinus Torvalds do { 24097f76d3dSNikanth Karthikesan if (!buffer_mapped(bh)) 24197f76d3dSNikanth Karthikesan all_mapped = 0; 24297f76d3dSNikanth Karthikesan else if (bh->b_blocknr == block) { 2431da177e4SLinus Torvalds ret = bh; 2441da177e4SLinus Torvalds get_bh(bh); 2451da177e4SLinus Torvalds goto out_unlock; 2461da177e4SLinus Torvalds } 2471da177e4SLinus Torvalds bh = bh->b_this_page; 2481da177e4SLinus Torvalds } while (bh != head); 2491da177e4SLinus Torvalds 2501da177e4SLinus Torvalds /* we might be here because some of the buffers on this page are 2511da177e4SLinus Torvalds * not mapped. This is due to various races between 2521da177e4SLinus Torvalds * file io on the block device and getblk. It gets dealt with 2531da177e4SLinus Torvalds * elsewhere, don't buffer_error if we had some unmapped buffers 2541da177e4SLinus Torvalds */ 2551da177e4SLinus Torvalds if (all_mapped) { 25672a2ebd8STao Ma char b[BDEVNAME_SIZE]; 25772a2ebd8STao Ma 2581da177e4SLinus Torvalds printk("__find_get_block_slow() failed. " 2591da177e4SLinus Torvalds "block=%llu, b_blocknr=%llu\n", 260205f87f6SBadari Pulavarty (unsigned long long)block, 261205f87f6SBadari Pulavarty (unsigned long long)bh->b_blocknr); 262205f87f6SBadari Pulavarty printk("b_state=0x%08lx, b_size=%zu\n", 263205f87f6SBadari Pulavarty bh->b_state, bh->b_size); 26472a2ebd8STao Ma printk("device %s blocksize: %d\n", bdevname(bdev, b), 26572a2ebd8STao Ma 1 << bd_inode->i_blkbits); 2661da177e4SLinus Torvalds } 2671da177e4SLinus Torvalds out_unlock: 2681da177e4SLinus Torvalds spin_unlock(&bd_mapping->private_lock); 2691da177e4SLinus Torvalds page_cache_release(page); 2701da177e4SLinus Torvalds out: 2711da177e4SLinus Torvalds return ret; 2721da177e4SLinus Torvalds } 2731da177e4SLinus Torvalds 2741da177e4SLinus Torvalds /* 2755b0830cbSJens Axboe * Kick the writeback threads then try to free up some ZONE_NORMAL memory. 2761da177e4SLinus Torvalds */ 2771da177e4SLinus Torvalds static void free_more_memory(void) 2781da177e4SLinus Torvalds { 27919770b32SMel Gorman struct zone *zone; 2800e88460dSMel Gorman int nid; 2811da177e4SLinus Torvalds 2820e175a18SCurt Wohlgemuth wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM); 2831da177e4SLinus Torvalds yield(); 2841da177e4SLinus Torvalds 2850e88460dSMel Gorman for_each_online_node(nid) { 28619770b32SMel Gorman (void)first_zones_zonelist(node_zonelist(nid, GFP_NOFS), 28719770b32SMel Gorman gfp_zone(GFP_NOFS), NULL, 28819770b32SMel Gorman &zone); 28919770b32SMel Gorman if (zone) 29054a6eb5cSMel Gorman try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0, 291327c0e96SKAMEZAWA Hiroyuki GFP_NOFS, NULL); 2921da177e4SLinus Torvalds } 2931da177e4SLinus Torvalds } 2941da177e4SLinus Torvalds 2951da177e4SLinus Torvalds /* 2961da177e4SLinus Torvalds * I/O completion handler for block_read_full_page() - pages 2971da177e4SLinus Torvalds * which come unlocked at the end of I/O. 2981da177e4SLinus Torvalds */ 2991da177e4SLinus Torvalds static void end_buffer_async_read(struct buffer_head *bh, int uptodate) 3001da177e4SLinus Torvalds { 3011da177e4SLinus Torvalds unsigned long flags; 302a3972203SNick Piggin struct buffer_head *first; 3031da177e4SLinus Torvalds struct buffer_head *tmp; 3041da177e4SLinus Torvalds struct page *page; 3051da177e4SLinus Torvalds int page_uptodate = 1; 3061da177e4SLinus Torvalds 3071da177e4SLinus Torvalds BUG_ON(!buffer_async_read(bh)); 3081da177e4SLinus Torvalds 3091da177e4SLinus Torvalds page = bh->b_page; 3101da177e4SLinus Torvalds if (uptodate) { 3111da177e4SLinus Torvalds set_buffer_uptodate(bh); 3121da177e4SLinus Torvalds } else { 3131da177e4SLinus Torvalds clear_buffer_uptodate(bh); 31408bafc03SKeith Mannthey if (!quiet_error(bh)) 3151da177e4SLinus Torvalds buffer_io_error(bh); 3161da177e4SLinus Torvalds SetPageError(page); 3171da177e4SLinus Torvalds } 3181da177e4SLinus Torvalds 3191da177e4SLinus Torvalds /* 3201da177e4SLinus Torvalds * Be _very_ careful from here on. Bad things can happen if 3211da177e4SLinus Torvalds * two buffer heads end IO at almost the same time and both 3221da177e4SLinus Torvalds * decide that the page is now completely done. 3231da177e4SLinus Torvalds */ 324a3972203SNick Piggin first = page_buffers(page); 325a3972203SNick Piggin local_irq_save(flags); 326a3972203SNick Piggin bit_spin_lock(BH_Uptodate_Lock, &first->b_state); 3271da177e4SLinus Torvalds clear_buffer_async_read(bh); 3281da177e4SLinus Torvalds unlock_buffer(bh); 3291da177e4SLinus Torvalds tmp = bh; 3301da177e4SLinus Torvalds do { 3311da177e4SLinus Torvalds if (!buffer_uptodate(tmp)) 3321da177e4SLinus Torvalds page_uptodate = 0; 3331da177e4SLinus Torvalds if (buffer_async_read(tmp)) { 3341da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 3351da177e4SLinus Torvalds goto still_busy; 3361da177e4SLinus Torvalds } 3371da177e4SLinus Torvalds tmp = tmp->b_this_page; 3381da177e4SLinus Torvalds } while (tmp != bh); 339a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 340a3972203SNick Piggin local_irq_restore(flags); 3411da177e4SLinus Torvalds 3421da177e4SLinus Torvalds /* 3431da177e4SLinus Torvalds * If none of the buffers had errors and they are all 3441da177e4SLinus Torvalds * uptodate then we can set the page uptodate. 3451da177e4SLinus Torvalds */ 3461da177e4SLinus Torvalds if (page_uptodate && !PageError(page)) 3471da177e4SLinus Torvalds SetPageUptodate(page); 3481da177e4SLinus Torvalds unlock_page(page); 3491da177e4SLinus Torvalds return; 3501da177e4SLinus Torvalds 3511da177e4SLinus Torvalds still_busy: 352a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 353a3972203SNick Piggin local_irq_restore(flags); 3541da177e4SLinus Torvalds return; 3551da177e4SLinus Torvalds } 3561da177e4SLinus Torvalds 3571da177e4SLinus Torvalds /* 3581da177e4SLinus Torvalds * Completion handler for block_write_full_page() - pages which are unlocked 3591da177e4SLinus Torvalds * during I/O, and which have PageWriteback cleared upon I/O completion. 3601da177e4SLinus Torvalds */ 36135c80d5fSChris Mason void end_buffer_async_write(struct buffer_head *bh, int uptodate) 3621da177e4SLinus Torvalds { 3631da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 3641da177e4SLinus Torvalds unsigned long flags; 365a3972203SNick Piggin struct buffer_head *first; 3661da177e4SLinus Torvalds struct buffer_head *tmp; 3671da177e4SLinus Torvalds struct page *page; 3681da177e4SLinus Torvalds 3691da177e4SLinus Torvalds BUG_ON(!buffer_async_write(bh)); 3701da177e4SLinus Torvalds 3711da177e4SLinus Torvalds page = bh->b_page; 3721da177e4SLinus Torvalds if (uptodate) { 3731da177e4SLinus Torvalds set_buffer_uptodate(bh); 3741da177e4SLinus Torvalds } else { 37508bafc03SKeith Mannthey if (!quiet_error(bh)) { 3761da177e4SLinus Torvalds buffer_io_error(bh); 3771da177e4SLinus Torvalds printk(KERN_WARNING "lost page write due to " 3781da177e4SLinus Torvalds "I/O error on %s\n", 3791da177e4SLinus Torvalds bdevname(bh->b_bdev, b)); 3801da177e4SLinus Torvalds } 3811da177e4SLinus Torvalds set_bit(AS_EIO, &page->mapping->flags); 38258ff407bSJan Kara set_buffer_write_io_error(bh); 3831da177e4SLinus Torvalds clear_buffer_uptodate(bh); 3841da177e4SLinus Torvalds SetPageError(page); 3851da177e4SLinus Torvalds } 3861da177e4SLinus Torvalds 387a3972203SNick Piggin first = page_buffers(page); 388a3972203SNick Piggin local_irq_save(flags); 389a3972203SNick Piggin bit_spin_lock(BH_Uptodate_Lock, &first->b_state); 390a3972203SNick Piggin 3911da177e4SLinus Torvalds clear_buffer_async_write(bh); 3921da177e4SLinus Torvalds unlock_buffer(bh); 3931da177e4SLinus Torvalds tmp = bh->b_this_page; 3941da177e4SLinus Torvalds while (tmp != bh) { 3951da177e4SLinus Torvalds if (buffer_async_write(tmp)) { 3961da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 3971da177e4SLinus Torvalds goto still_busy; 3981da177e4SLinus Torvalds } 3991da177e4SLinus Torvalds tmp = tmp->b_this_page; 4001da177e4SLinus Torvalds } 401a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 402a3972203SNick Piggin local_irq_restore(flags); 4031da177e4SLinus Torvalds end_page_writeback(page); 4041da177e4SLinus Torvalds return; 4051da177e4SLinus Torvalds 4061da177e4SLinus Torvalds still_busy: 407a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 408a3972203SNick Piggin local_irq_restore(flags); 4091da177e4SLinus Torvalds return; 4101da177e4SLinus Torvalds } 4111fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_async_write); 4121da177e4SLinus Torvalds 4131da177e4SLinus Torvalds /* 4141da177e4SLinus Torvalds * If a page's buffers are under async readin (end_buffer_async_read 4151da177e4SLinus Torvalds * completion) then there is a possibility that another thread of 4161da177e4SLinus Torvalds * control could lock one of the buffers after it has completed 4171da177e4SLinus Torvalds * but while some of the other buffers have not completed. This 4181da177e4SLinus Torvalds * locked buffer would confuse end_buffer_async_read() into not unlocking 4191da177e4SLinus Torvalds * the page. So the absence of BH_Async_Read tells end_buffer_async_read() 4201da177e4SLinus Torvalds * that this buffer is not under async I/O. 4211da177e4SLinus Torvalds * 4221da177e4SLinus Torvalds * The page comes unlocked when it has no locked buffer_async buffers 4231da177e4SLinus Torvalds * left. 4241da177e4SLinus Torvalds * 4251da177e4SLinus Torvalds * PageLocked prevents anyone starting new async I/O reads any of 4261da177e4SLinus Torvalds * the buffers. 4271da177e4SLinus Torvalds * 4281da177e4SLinus Torvalds * PageWriteback is used to prevent simultaneous writeout of the same 4291da177e4SLinus Torvalds * page. 4301da177e4SLinus Torvalds * 4311da177e4SLinus Torvalds * PageLocked prevents anyone from starting writeback of a page which is 4321da177e4SLinus Torvalds * under read I/O (PageWriteback is only ever set against a locked page). 4331da177e4SLinus Torvalds */ 4341da177e4SLinus Torvalds static void mark_buffer_async_read(struct buffer_head *bh) 4351da177e4SLinus Torvalds { 4361da177e4SLinus Torvalds bh->b_end_io = end_buffer_async_read; 4371da177e4SLinus Torvalds set_buffer_async_read(bh); 4381da177e4SLinus Torvalds } 4391da177e4SLinus Torvalds 4401fe72eaaSH Hartley Sweeten static void mark_buffer_async_write_endio(struct buffer_head *bh, 44135c80d5fSChris Mason bh_end_io_t *handler) 44235c80d5fSChris Mason { 44335c80d5fSChris Mason bh->b_end_io = handler; 44435c80d5fSChris Mason set_buffer_async_write(bh); 44535c80d5fSChris Mason } 44635c80d5fSChris Mason 4471da177e4SLinus Torvalds void mark_buffer_async_write(struct buffer_head *bh) 4481da177e4SLinus Torvalds { 44935c80d5fSChris Mason mark_buffer_async_write_endio(bh, end_buffer_async_write); 4501da177e4SLinus Torvalds } 4511da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_async_write); 4521da177e4SLinus Torvalds 4531da177e4SLinus Torvalds 4541da177e4SLinus Torvalds /* 4551da177e4SLinus Torvalds * fs/buffer.c contains helper functions for buffer-backed address space's 4561da177e4SLinus Torvalds * fsync functions. A common requirement for buffer-based filesystems is 4571da177e4SLinus Torvalds * that certain data from the backing blockdev needs to be written out for 4581da177e4SLinus Torvalds * a successful fsync(). For example, ext2 indirect blocks need to be 4591da177e4SLinus Torvalds * written back and waited upon before fsync() returns. 4601da177e4SLinus Torvalds * 4611da177e4SLinus Torvalds * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), 4621da177e4SLinus Torvalds * inode_has_buffers() and invalidate_inode_buffers() are provided for the 4631da177e4SLinus Torvalds * management of a list of dependent buffers at ->i_mapping->private_list. 4641da177e4SLinus Torvalds * 4651da177e4SLinus Torvalds * Locking is a little subtle: try_to_free_buffers() will remove buffers 4661da177e4SLinus Torvalds * from their controlling inode's queue when they are being freed. But 4671da177e4SLinus Torvalds * try_to_free_buffers() will be operating against the *blockdev* mapping 4681da177e4SLinus Torvalds * at the time, not against the S_ISREG file which depends on those buffers. 4691da177e4SLinus Torvalds * So the locking for private_list is via the private_lock in the address_space 4701da177e4SLinus Torvalds * which backs the buffers. Which is different from the address_space 4711da177e4SLinus Torvalds * against which the buffers are listed. So for a particular address_space, 4721da177e4SLinus Torvalds * mapping->private_lock does *not* protect mapping->private_list! In fact, 4731da177e4SLinus Torvalds * mapping->private_list will always be protected by the backing blockdev's 4741da177e4SLinus Torvalds * ->private_lock. 4751da177e4SLinus Torvalds * 4761da177e4SLinus Torvalds * Which introduces a requirement: all buffers on an address_space's 4771da177e4SLinus Torvalds * ->private_list must be from the same address_space: the blockdev's. 4781da177e4SLinus Torvalds * 4791da177e4SLinus Torvalds * address_spaces which do not place buffers at ->private_list via these 4801da177e4SLinus Torvalds * utility functions are free to use private_lock and private_list for 4811da177e4SLinus Torvalds * whatever they want. The only requirement is that list_empty(private_list) 4821da177e4SLinus Torvalds * be true at clear_inode() time. 4831da177e4SLinus Torvalds * 4841da177e4SLinus Torvalds * FIXME: clear_inode should not call invalidate_inode_buffers(). The 4851da177e4SLinus Torvalds * filesystems should do that. invalidate_inode_buffers() should just go 4861da177e4SLinus Torvalds * BUG_ON(!list_empty). 4871da177e4SLinus Torvalds * 4881da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should 4891da177e4SLinus Torvalds * take an address_space, not an inode. And it should be called 4901da177e4SLinus Torvalds * mark_buffer_dirty_fsync() to clearly define why those buffers are being 4911da177e4SLinus Torvalds * queued up. 4921da177e4SLinus Torvalds * 4931da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the 4941da177e4SLinus Torvalds * list if it is already on a list. Because if the buffer is on a list, 4951da177e4SLinus Torvalds * it *must* already be on the right one. If not, the filesystem is being 4961da177e4SLinus Torvalds * silly. This will save a ton of locking. But first we have to ensure 4971da177e4SLinus Torvalds * that buffers are taken *off* the old inode's list when they are freed 4981da177e4SLinus Torvalds * (presumably in truncate). That requires careful auditing of all 4991da177e4SLinus Torvalds * filesystems (do it inside bforget()). It could also be done by bringing 5001da177e4SLinus Torvalds * b_inode back. 5011da177e4SLinus Torvalds */ 5021da177e4SLinus Torvalds 5031da177e4SLinus Torvalds /* 5041da177e4SLinus Torvalds * The buffer's backing address_space's private_lock must be held 5051da177e4SLinus Torvalds */ 506dbacefc9SThomas Petazzoni static void __remove_assoc_queue(struct buffer_head *bh) 5071da177e4SLinus Torvalds { 5081da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 50958ff407bSJan Kara WARN_ON(!bh->b_assoc_map); 51058ff407bSJan Kara if (buffer_write_io_error(bh)) 51158ff407bSJan Kara set_bit(AS_EIO, &bh->b_assoc_map->flags); 51258ff407bSJan Kara bh->b_assoc_map = NULL; 5131da177e4SLinus Torvalds } 5141da177e4SLinus Torvalds 5151da177e4SLinus Torvalds int inode_has_buffers(struct inode *inode) 5161da177e4SLinus Torvalds { 5171da177e4SLinus Torvalds return !list_empty(&inode->i_data.private_list); 5181da177e4SLinus Torvalds } 5191da177e4SLinus Torvalds 5201da177e4SLinus Torvalds /* 5211da177e4SLinus Torvalds * osync is designed to support O_SYNC io. It waits synchronously for 5221da177e4SLinus Torvalds * all already-submitted IO to complete, but does not queue any new 5231da177e4SLinus Torvalds * writes to the disk. 5241da177e4SLinus Torvalds * 5251da177e4SLinus Torvalds * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as 5261da177e4SLinus Torvalds * you dirty the buffers, and then use osync_inode_buffers to wait for 5271da177e4SLinus Torvalds * completion. Any other dirty buffers which are not yet queued for 5281da177e4SLinus Torvalds * write will not be flushed to disk by the osync. 5291da177e4SLinus Torvalds */ 5301da177e4SLinus Torvalds static int osync_buffers_list(spinlock_t *lock, struct list_head *list) 5311da177e4SLinus Torvalds { 5321da177e4SLinus Torvalds struct buffer_head *bh; 5331da177e4SLinus Torvalds struct list_head *p; 5341da177e4SLinus Torvalds int err = 0; 5351da177e4SLinus Torvalds 5361da177e4SLinus Torvalds spin_lock(lock); 5371da177e4SLinus Torvalds repeat: 5381da177e4SLinus Torvalds list_for_each_prev(p, list) { 5391da177e4SLinus Torvalds bh = BH_ENTRY(p); 5401da177e4SLinus Torvalds if (buffer_locked(bh)) { 5411da177e4SLinus Torvalds get_bh(bh); 5421da177e4SLinus Torvalds spin_unlock(lock); 5431da177e4SLinus Torvalds wait_on_buffer(bh); 5441da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 5451da177e4SLinus Torvalds err = -EIO; 5461da177e4SLinus Torvalds brelse(bh); 5471da177e4SLinus Torvalds spin_lock(lock); 5481da177e4SLinus Torvalds goto repeat; 5491da177e4SLinus Torvalds } 5501da177e4SLinus Torvalds } 5511da177e4SLinus Torvalds spin_unlock(lock); 5521da177e4SLinus Torvalds return err; 5531da177e4SLinus Torvalds } 5541da177e4SLinus Torvalds 55501a05b33SAl Viro static void do_thaw_one(struct super_block *sb, void *unused) 556c2d75438SEric Sandeen { 557c2d75438SEric Sandeen char b[BDEVNAME_SIZE]; 558c2d75438SEric Sandeen while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb)) 559c2d75438SEric Sandeen printk(KERN_WARNING "Emergency Thaw on %s\n", 560c2d75438SEric Sandeen bdevname(sb->s_bdev, b)); 561c2d75438SEric Sandeen } 56201a05b33SAl Viro 56301a05b33SAl Viro static void do_thaw_all(struct work_struct *work) 56401a05b33SAl Viro { 56501a05b33SAl Viro iterate_supers(do_thaw_one, NULL); 566053c525fSJens Axboe kfree(work); 567c2d75438SEric Sandeen printk(KERN_WARNING "Emergency Thaw complete\n"); 568c2d75438SEric Sandeen } 569c2d75438SEric Sandeen 570c2d75438SEric Sandeen /** 571c2d75438SEric Sandeen * emergency_thaw_all -- forcibly thaw every frozen filesystem 572c2d75438SEric Sandeen * 573c2d75438SEric Sandeen * Used for emergency unfreeze of all filesystems via SysRq 574c2d75438SEric Sandeen */ 575c2d75438SEric Sandeen void emergency_thaw_all(void) 576c2d75438SEric Sandeen { 577053c525fSJens Axboe struct work_struct *work; 578053c525fSJens Axboe 579053c525fSJens Axboe work = kmalloc(sizeof(*work), GFP_ATOMIC); 580053c525fSJens Axboe if (work) { 581053c525fSJens Axboe INIT_WORK(work, do_thaw_all); 582053c525fSJens Axboe schedule_work(work); 583053c525fSJens Axboe } 584c2d75438SEric Sandeen } 585c2d75438SEric Sandeen 5861da177e4SLinus Torvalds /** 58778a4a50aSRandy Dunlap * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers 58867be2dd1SMartin Waitz * @mapping: the mapping which wants those buffers written 5891da177e4SLinus Torvalds * 5901da177e4SLinus Torvalds * Starts I/O against the buffers at mapping->private_list, and waits upon 5911da177e4SLinus Torvalds * that I/O. 5921da177e4SLinus Torvalds * 59367be2dd1SMartin Waitz * Basically, this is a convenience function for fsync(). 59467be2dd1SMartin Waitz * @mapping is a file or directory which needs those buffers to be written for 59567be2dd1SMartin Waitz * a successful fsync(). 5961da177e4SLinus Torvalds */ 5971da177e4SLinus Torvalds int sync_mapping_buffers(struct address_space *mapping) 5981da177e4SLinus Torvalds { 599252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 6001da177e4SLinus Torvalds 6011da177e4SLinus Torvalds if (buffer_mapping == NULL || list_empty(&mapping->private_list)) 6021da177e4SLinus Torvalds return 0; 6031da177e4SLinus Torvalds 6041da177e4SLinus Torvalds return fsync_buffers_list(&buffer_mapping->private_lock, 6051da177e4SLinus Torvalds &mapping->private_list); 6061da177e4SLinus Torvalds } 6071da177e4SLinus Torvalds EXPORT_SYMBOL(sync_mapping_buffers); 6081da177e4SLinus Torvalds 6091da177e4SLinus Torvalds /* 6101da177e4SLinus Torvalds * Called when we've recently written block `bblock', and it is known that 6111da177e4SLinus Torvalds * `bblock' was for a buffer_boundary() buffer. This means that the block at 6121da177e4SLinus Torvalds * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's 6131da177e4SLinus Torvalds * dirty, schedule it for IO. So that indirects merge nicely with their data. 6141da177e4SLinus Torvalds */ 6151da177e4SLinus Torvalds void write_boundary_block(struct block_device *bdev, 6161da177e4SLinus Torvalds sector_t bblock, unsigned blocksize) 6171da177e4SLinus Torvalds { 6181da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); 6191da177e4SLinus Torvalds if (bh) { 6201da177e4SLinus Torvalds if (buffer_dirty(bh)) 6211da177e4SLinus Torvalds ll_rw_block(WRITE, 1, &bh); 6221da177e4SLinus Torvalds put_bh(bh); 6231da177e4SLinus Torvalds } 6241da177e4SLinus Torvalds } 6251da177e4SLinus Torvalds 6261da177e4SLinus Torvalds void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) 6271da177e4SLinus Torvalds { 6281da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 6291da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 6301da177e4SLinus Torvalds 6311da177e4SLinus Torvalds mark_buffer_dirty(bh); 632252aa6f5SRafael Aquini if (!mapping->private_data) { 633252aa6f5SRafael Aquini mapping->private_data = buffer_mapping; 6341da177e4SLinus Torvalds } else { 635252aa6f5SRafael Aquini BUG_ON(mapping->private_data != buffer_mapping); 6361da177e4SLinus Torvalds } 637535ee2fbSJan Kara if (!bh->b_assoc_map) { 6381da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 6391da177e4SLinus Torvalds list_move_tail(&bh->b_assoc_buffers, 6401da177e4SLinus Torvalds &mapping->private_list); 64158ff407bSJan Kara bh->b_assoc_map = mapping; 6421da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 6431da177e4SLinus Torvalds } 6441da177e4SLinus Torvalds } 6451da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_dirty_inode); 6461da177e4SLinus Torvalds 6471da177e4SLinus Torvalds /* 648787d2214SNick Piggin * Mark the page dirty, and set it dirty in the radix tree, and mark the inode 649787d2214SNick Piggin * dirty. 650787d2214SNick Piggin * 651787d2214SNick Piggin * If warn is true, then emit a warning if the page is not uptodate and has 652787d2214SNick Piggin * not been truncated. 653787d2214SNick Piggin */ 654a8e7d49aSLinus Torvalds static void __set_page_dirty(struct page *page, 655787d2214SNick Piggin struct address_space *mapping, int warn) 656787d2214SNick Piggin { 657227d53b3SKOSAKI Motohiro unsigned long flags; 658227d53b3SKOSAKI Motohiro 659227d53b3SKOSAKI Motohiro spin_lock_irqsave(&mapping->tree_lock, flags); 660787d2214SNick Piggin if (page->mapping) { /* Race with truncate? */ 661787d2214SNick Piggin WARN_ON_ONCE(warn && !PageUptodate(page)); 662e3a7cca1SEdward Shishkin account_page_dirtied(page, mapping); 663787d2214SNick Piggin radix_tree_tag_set(&mapping->page_tree, 664787d2214SNick Piggin page_index(page), PAGECACHE_TAG_DIRTY); 665787d2214SNick Piggin } 666227d53b3SKOSAKI Motohiro spin_unlock_irqrestore(&mapping->tree_lock, flags); 667787d2214SNick Piggin __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 668787d2214SNick Piggin } 669787d2214SNick Piggin 670787d2214SNick Piggin /* 6711da177e4SLinus Torvalds * Add a page to the dirty page list. 6721da177e4SLinus Torvalds * 6731da177e4SLinus Torvalds * It is a sad fact of life that this function is called from several places 6741da177e4SLinus Torvalds * deeply under spinlocking. It may not sleep. 6751da177e4SLinus Torvalds * 6761da177e4SLinus Torvalds * If the page has buffers, the uptodate buffers are set dirty, to preserve 6771da177e4SLinus Torvalds * dirty-state coherency between the page and the buffers. It the page does 6781da177e4SLinus Torvalds * not have buffers then when they are later attached they will all be set 6791da177e4SLinus Torvalds * dirty. 6801da177e4SLinus Torvalds * 6811da177e4SLinus Torvalds * The buffers are dirtied before the page is dirtied. There's a small race 6821da177e4SLinus Torvalds * window in which a writepage caller may see the page cleanness but not the 6831da177e4SLinus Torvalds * buffer dirtiness. That's fine. If this code were to set the page dirty 6841da177e4SLinus Torvalds * before the buffers, a concurrent writepage caller could clear the page dirty 6851da177e4SLinus Torvalds * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean 6861da177e4SLinus Torvalds * page on the dirty page list. 6871da177e4SLinus Torvalds * 6881da177e4SLinus Torvalds * We use private_lock to lock against try_to_free_buffers while using the 6891da177e4SLinus Torvalds * page's buffer list. Also use this to protect against clean buffers being 6901da177e4SLinus Torvalds * added to the page after it was set dirty. 6911da177e4SLinus Torvalds * 6921da177e4SLinus Torvalds * FIXME: may need to call ->reservepage here as well. That's rather up to the 6931da177e4SLinus Torvalds * address_space though. 6941da177e4SLinus Torvalds */ 6951da177e4SLinus Torvalds int __set_page_dirty_buffers(struct page *page) 6961da177e4SLinus Torvalds { 697a8e7d49aSLinus Torvalds int newly_dirty; 698787d2214SNick Piggin struct address_space *mapping = page_mapping(page); 699ebf7a227SNick Piggin 700ebf7a227SNick Piggin if (unlikely(!mapping)) 701ebf7a227SNick Piggin return !TestSetPageDirty(page); 7021da177e4SLinus Torvalds 7031da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 7041da177e4SLinus Torvalds if (page_has_buffers(page)) { 7051da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 7061da177e4SLinus Torvalds struct buffer_head *bh = head; 7071da177e4SLinus Torvalds 7081da177e4SLinus Torvalds do { 7091da177e4SLinus Torvalds set_buffer_dirty(bh); 7101da177e4SLinus Torvalds bh = bh->b_this_page; 7111da177e4SLinus Torvalds } while (bh != head); 7121da177e4SLinus Torvalds } 713a8e7d49aSLinus Torvalds newly_dirty = !TestSetPageDirty(page); 7141da177e4SLinus Torvalds spin_unlock(&mapping->private_lock); 7151da177e4SLinus Torvalds 716a8e7d49aSLinus Torvalds if (newly_dirty) 717a8e7d49aSLinus Torvalds __set_page_dirty(page, mapping, 1); 718a8e7d49aSLinus Torvalds return newly_dirty; 7191da177e4SLinus Torvalds } 7201da177e4SLinus Torvalds EXPORT_SYMBOL(__set_page_dirty_buffers); 7211da177e4SLinus Torvalds 7221da177e4SLinus Torvalds /* 7231da177e4SLinus Torvalds * Write out and wait upon a list of buffers. 7241da177e4SLinus Torvalds * 7251da177e4SLinus Torvalds * We have conflicting pressures: we want to make sure that all 7261da177e4SLinus Torvalds * initially dirty buffers get waited on, but that any subsequently 7271da177e4SLinus Torvalds * dirtied buffers don't. After all, we don't want fsync to last 7281da177e4SLinus Torvalds * forever if somebody is actively writing to the file. 7291da177e4SLinus Torvalds * 7301da177e4SLinus Torvalds * Do this in two main stages: first we copy dirty buffers to a 7311da177e4SLinus Torvalds * temporary inode list, queueing the writes as we go. Then we clean 7321da177e4SLinus Torvalds * up, waiting for those writes to complete. 7331da177e4SLinus Torvalds * 7341da177e4SLinus Torvalds * During this second stage, any subsequent updates to the file may end 7351da177e4SLinus Torvalds * up refiling the buffer on the original inode's dirty list again, so 7361da177e4SLinus Torvalds * there is a chance we will end up with a buffer queued for write but 7371da177e4SLinus Torvalds * not yet completed on that list. So, as a final cleanup we go through 7381da177e4SLinus Torvalds * the osync code to catch these locked, dirty buffers without requeuing 7391da177e4SLinus Torvalds * any newly dirty buffers for write. 7401da177e4SLinus Torvalds */ 7411da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) 7421da177e4SLinus Torvalds { 7431da177e4SLinus Torvalds struct buffer_head *bh; 7441da177e4SLinus Torvalds struct list_head tmp; 7457eaceaccSJens Axboe struct address_space *mapping; 7461da177e4SLinus Torvalds int err = 0, err2; 7474ee2491eSJens Axboe struct blk_plug plug; 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 7504ee2491eSJens Axboe blk_start_plug(&plug); 7511da177e4SLinus Torvalds 7521da177e4SLinus Torvalds spin_lock(lock); 7531da177e4SLinus Torvalds while (!list_empty(list)) { 7541da177e4SLinus Torvalds bh = BH_ENTRY(list->next); 755535ee2fbSJan Kara mapping = bh->b_assoc_map; 75658ff407bSJan Kara __remove_assoc_queue(bh); 757535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 758535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 759535ee2fbSJan Kara smp_mb(); 7601da177e4SLinus Torvalds if (buffer_dirty(bh) || buffer_locked(bh)) { 7611da177e4SLinus Torvalds list_add(&bh->b_assoc_buffers, &tmp); 762535ee2fbSJan Kara bh->b_assoc_map = mapping; 7631da177e4SLinus Torvalds if (buffer_dirty(bh)) { 7641da177e4SLinus Torvalds get_bh(bh); 7651da177e4SLinus Torvalds spin_unlock(lock); 7661da177e4SLinus Torvalds /* 7671da177e4SLinus Torvalds * Ensure any pending I/O completes so that 7689cb569d6SChristoph Hellwig * write_dirty_buffer() actually writes the 7699cb569d6SChristoph Hellwig * current contents - it is a noop if I/O is 7709cb569d6SChristoph Hellwig * still in flight on potentially older 7719cb569d6SChristoph Hellwig * contents. 7721da177e4SLinus Torvalds */ 773721a9602SJens Axboe write_dirty_buffer(bh, WRITE_SYNC); 7749cf6b720SJens Axboe 7759cf6b720SJens Axboe /* 7769cf6b720SJens Axboe * Kick off IO for the previous mapping. Note 7779cf6b720SJens Axboe * that we will not run the very last mapping, 7789cf6b720SJens Axboe * wait_on_buffer() will do that for us 7799cf6b720SJens Axboe * through sync_buffer(). 7809cf6b720SJens Axboe */ 7811da177e4SLinus Torvalds brelse(bh); 7821da177e4SLinus Torvalds spin_lock(lock); 7831da177e4SLinus Torvalds } 7841da177e4SLinus Torvalds } 7851da177e4SLinus Torvalds } 7861da177e4SLinus Torvalds 7874ee2491eSJens Axboe spin_unlock(lock); 7884ee2491eSJens Axboe blk_finish_plug(&plug); 7894ee2491eSJens Axboe spin_lock(lock); 7904ee2491eSJens Axboe 7911da177e4SLinus Torvalds while (!list_empty(&tmp)) { 7921da177e4SLinus Torvalds bh = BH_ENTRY(tmp.prev); 7931da177e4SLinus Torvalds get_bh(bh); 794535ee2fbSJan Kara mapping = bh->b_assoc_map; 795535ee2fbSJan Kara __remove_assoc_queue(bh); 796535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 797535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 798535ee2fbSJan Kara smp_mb(); 799535ee2fbSJan Kara if (buffer_dirty(bh)) { 800535ee2fbSJan Kara list_add(&bh->b_assoc_buffers, 801e3892296SJan Kara &mapping->private_list); 802535ee2fbSJan Kara bh->b_assoc_map = mapping; 803535ee2fbSJan Kara } 8041da177e4SLinus Torvalds spin_unlock(lock); 8051da177e4SLinus Torvalds wait_on_buffer(bh); 8061da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 8071da177e4SLinus Torvalds err = -EIO; 8081da177e4SLinus Torvalds brelse(bh); 8091da177e4SLinus Torvalds spin_lock(lock); 8101da177e4SLinus Torvalds } 8111da177e4SLinus Torvalds 8121da177e4SLinus Torvalds spin_unlock(lock); 8131da177e4SLinus Torvalds err2 = osync_buffers_list(lock, list); 8141da177e4SLinus Torvalds if (err) 8151da177e4SLinus Torvalds return err; 8161da177e4SLinus Torvalds else 8171da177e4SLinus Torvalds return err2; 8181da177e4SLinus Torvalds } 8191da177e4SLinus Torvalds 8201da177e4SLinus Torvalds /* 8211da177e4SLinus Torvalds * Invalidate any and all dirty buffers on a given inode. We are 8221da177e4SLinus Torvalds * probably unmounting the fs, but that doesn't mean we have already 8231da177e4SLinus Torvalds * done a sync(). Just drop the buffers from the inode list. 8241da177e4SLinus Torvalds * 8251da177e4SLinus Torvalds * NOTE: we take the inode's blockdev's mapping's private_lock. Which 8261da177e4SLinus Torvalds * assumes that all the buffers are against the blockdev. Not true 8271da177e4SLinus Torvalds * for reiserfs. 8281da177e4SLinus Torvalds */ 8291da177e4SLinus Torvalds void invalidate_inode_buffers(struct inode *inode) 8301da177e4SLinus Torvalds { 8311da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8321da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 8331da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 834252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 8351da177e4SLinus Torvalds 8361da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8371da177e4SLinus Torvalds while (!list_empty(list)) 8381da177e4SLinus Torvalds __remove_assoc_queue(BH_ENTRY(list->next)); 8391da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8401da177e4SLinus Torvalds } 8411da177e4SLinus Torvalds } 84252b19ac9SJan Kara EXPORT_SYMBOL(invalidate_inode_buffers); 8431da177e4SLinus Torvalds 8441da177e4SLinus Torvalds /* 8451da177e4SLinus Torvalds * Remove any clean buffers from the inode's buffer list. This is called 8461da177e4SLinus Torvalds * when we're trying to free the inode itself. Those buffers can pin it. 8471da177e4SLinus Torvalds * 8481da177e4SLinus Torvalds * Returns true if all buffers were removed. 8491da177e4SLinus Torvalds */ 8501da177e4SLinus Torvalds int remove_inode_buffers(struct inode *inode) 8511da177e4SLinus Torvalds { 8521da177e4SLinus Torvalds int ret = 1; 8531da177e4SLinus Torvalds 8541da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8551da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 8561da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 857252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8601da177e4SLinus Torvalds while (!list_empty(list)) { 8611da177e4SLinus Torvalds struct buffer_head *bh = BH_ENTRY(list->next); 8621da177e4SLinus Torvalds if (buffer_dirty(bh)) { 8631da177e4SLinus Torvalds ret = 0; 8641da177e4SLinus Torvalds break; 8651da177e4SLinus Torvalds } 8661da177e4SLinus Torvalds __remove_assoc_queue(bh); 8671da177e4SLinus Torvalds } 8681da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8691da177e4SLinus Torvalds } 8701da177e4SLinus Torvalds return ret; 8711da177e4SLinus Torvalds } 8721da177e4SLinus Torvalds 8731da177e4SLinus Torvalds /* 8741da177e4SLinus Torvalds * Create the appropriate buffers when given a page for data area and 8751da177e4SLinus Torvalds * the size of each buffer.. Use the bh->b_this_page linked list to 8761da177e4SLinus Torvalds * follow the buffers created. Return NULL if unable to create more 8771da177e4SLinus Torvalds * buffers. 8781da177e4SLinus Torvalds * 8791da177e4SLinus Torvalds * The retry flag is used to differentiate async IO (paging, swapping) 8801da177e4SLinus Torvalds * which may not fail from ordinary buffer allocations. 8811da177e4SLinus Torvalds */ 8821da177e4SLinus Torvalds struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, 8831da177e4SLinus Torvalds int retry) 8841da177e4SLinus Torvalds { 8851da177e4SLinus Torvalds struct buffer_head *bh, *head; 8861da177e4SLinus Torvalds long offset; 8871da177e4SLinus Torvalds 8881da177e4SLinus Torvalds try_again: 8891da177e4SLinus Torvalds head = NULL; 8901da177e4SLinus Torvalds offset = PAGE_SIZE; 8911da177e4SLinus Torvalds while ((offset -= size) >= 0) { 8921da177e4SLinus Torvalds bh = alloc_buffer_head(GFP_NOFS); 8931da177e4SLinus Torvalds if (!bh) 8941da177e4SLinus Torvalds goto no_grow; 8951da177e4SLinus Torvalds 8961da177e4SLinus Torvalds bh->b_this_page = head; 8971da177e4SLinus Torvalds bh->b_blocknr = -1; 8981da177e4SLinus Torvalds head = bh; 8991da177e4SLinus Torvalds 9001da177e4SLinus Torvalds bh->b_size = size; 9011da177e4SLinus Torvalds 9021da177e4SLinus Torvalds /* Link the buffer to its page */ 9031da177e4SLinus Torvalds set_bh_page(bh, page, offset); 9041da177e4SLinus Torvalds } 9051da177e4SLinus Torvalds return head; 9061da177e4SLinus Torvalds /* 9071da177e4SLinus Torvalds * In case anything failed, we just free everything we got. 9081da177e4SLinus Torvalds */ 9091da177e4SLinus Torvalds no_grow: 9101da177e4SLinus Torvalds if (head) { 9111da177e4SLinus Torvalds do { 9121da177e4SLinus Torvalds bh = head; 9131da177e4SLinus Torvalds head = head->b_this_page; 9141da177e4SLinus Torvalds free_buffer_head(bh); 9151da177e4SLinus Torvalds } while (head); 9161da177e4SLinus Torvalds } 9171da177e4SLinus Torvalds 9181da177e4SLinus Torvalds /* 9191da177e4SLinus Torvalds * Return failure for non-async IO requests. Async IO requests 9201da177e4SLinus Torvalds * are not allowed to fail, so we have to wait until buffer heads 9211da177e4SLinus Torvalds * become available. But we don't want tasks sleeping with 9221da177e4SLinus Torvalds * partially complete buffers, so all were released above. 9231da177e4SLinus Torvalds */ 9241da177e4SLinus Torvalds if (!retry) 9251da177e4SLinus Torvalds return NULL; 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds /* We're _really_ low on memory. Now we just 9281da177e4SLinus Torvalds * wait for old buffer heads to become free due to 9291da177e4SLinus Torvalds * finishing IO. Since this is an async request and 9301da177e4SLinus Torvalds * the reserve list is empty, we're sure there are 9311da177e4SLinus Torvalds * async buffer heads in use. 9321da177e4SLinus Torvalds */ 9331da177e4SLinus Torvalds free_more_memory(); 9341da177e4SLinus Torvalds goto try_again; 9351da177e4SLinus Torvalds } 9361da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(alloc_page_buffers); 9371da177e4SLinus Torvalds 9381da177e4SLinus Torvalds static inline void 9391da177e4SLinus Torvalds link_dev_buffers(struct page *page, struct buffer_head *head) 9401da177e4SLinus Torvalds { 9411da177e4SLinus Torvalds struct buffer_head *bh, *tail; 9421da177e4SLinus Torvalds 9431da177e4SLinus Torvalds bh = head; 9441da177e4SLinus Torvalds do { 9451da177e4SLinus Torvalds tail = bh; 9461da177e4SLinus Torvalds bh = bh->b_this_page; 9471da177e4SLinus Torvalds } while (bh); 9481da177e4SLinus Torvalds tail->b_this_page = head; 9491da177e4SLinus Torvalds attach_page_buffers(page, head); 9501da177e4SLinus Torvalds } 9511da177e4SLinus Torvalds 952bbec0270SLinus Torvalds static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) 953bbec0270SLinus Torvalds { 954bbec0270SLinus Torvalds sector_t retval = ~((sector_t)0); 955bbec0270SLinus Torvalds loff_t sz = i_size_read(bdev->bd_inode); 956bbec0270SLinus Torvalds 957bbec0270SLinus Torvalds if (sz) { 958bbec0270SLinus Torvalds unsigned int sizebits = blksize_bits(size); 959bbec0270SLinus Torvalds retval = (sz >> sizebits); 960bbec0270SLinus Torvalds } 961bbec0270SLinus Torvalds return retval; 962bbec0270SLinus Torvalds } 963bbec0270SLinus Torvalds 9641da177e4SLinus Torvalds /* 9651da177e4SLinus Torvalds * Initialise the state of a blockdev page's buffers. 9661da177e4SLinus Torvalds */ 967676ce6d5SHugh Dickins static sector_t 9681da177e4SLinus Torvalds init_page_buffers(struct page *page, struct block_device *bdev, 9691da177e4SLinus Torvalds sector_t block, int size) 9701da177e4SLinus Torvalds { 9711da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 9721da177e4SLinus Torvalds struct buffer_head *bh = head; 9731da177e4SLinus Torvalds int uptodate = PageUptodate(page); 974bbec0270SLinus Torvalds sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size); 9751da177e4SLinus Torvalds 9761da177e4SLinus Torvalds do { 9771da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 9781da177e4SLinus Torvalds init_buffer(bh, NULL, NULL); 9791da177e4SLinus Torvalds bh->b_bdev = bdev; 9801da177e4SLinus Torvalds bh->b_blocknr = block; 9811da177e4SLinus Torvalds if (uptodate) 9821da177e4SLinus Torvalds set_buffer_uptodate(bh); 983080399aaSJeff Moyer if (block < end_block) 9841da177e4SLinus Torvalds set_buffer_mapped(bh); 9851da177e4SLinus Torvalds } 9861da177e4SLinus Torvalds block++; 9871da177e4SLinus Torvalds bh = bh->b_this_page; 9881da177e4SLinus Torvalds } while (bh != head); 989676ce6d5SHugh Dickins 990676ce6d5SHugh Dickins /* 991676ce6d5SHugh Dickins * Caller needs to validate requested block against end of device. 992676ce6d5SHugh Dickins */ 993676ce6d5SHugh Dickins return end_block; 9941da177e4SLinus Torvalds } 9951da177e4SLinus Torvalds 9961da177e4SLinus Torvalds /* 9971da177e4SLinus Torvalds * Create the page-cache page that contains the requested block. 9981da177e4SLinus Torvalds * 999676ce6d5SHugh Dickins * This is used purely for blockdev mappings. 10001da177e4SLinus Torvalds */ 1001676ce6d5SHugh Dickins static int 10021da177e4SLinus Torvalds grow_dev_page(struct block_device *bdev, sector_t block, 1003676ce6d5SHugh Dickins pgoff_t index, int size, int sizebits) 10041da177e4SLinus Torvalds { 10051da177e4SLinus Torvalds struct inode *inode = bdev->bd_inode; 10061da177e4SLinus Torvalds struct page *page; 10071da177e4SLinus Torvalds struct buffer_head *bh; 1008676ce6d5SHugh Dickins sector_t end_block; 1009676ce6d5SHugh Dickins int ret = 0; /* Will call free_more_memory() */ 101084235de3SJohannes Weiner gfp_t gfp_mask; 10111da177e4SLinus Torvalds 101284235de3SJohannes Weiner gfp_mask = mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS; 101384235de3SJohannes Weiner gfp_mask |= __GFP_MOVABLE; 101484235de3SJohannes Weiner /* 101584235de3SJohannes Weiner * XXX: __getblk_slow() can not really deal with failure and 101684235de3SJohannes Weiner * will endlessly loop on improvised global reclaim. Prefer 101784235de3SJohannes Weiner * looping in the allocator rather than here, at least that 101884235de3SJohannes Weiner * code knows what it's doing. 101984235de3SJohannes Weiner */ 102084235de3SJohannes Weiner gfp_mask |= __GFP_NOFAIL; 102184235de3SJohannes Weiner 102284235de3SJohannes Weiner page = find_or_create_page(inode->i_mapping, index, gfp_mask); 10231da177e4SLinus Torvalds if (!page) 1024676ce6d5SHugh Dickins return ret; 10251da177e4SLinus Torvalds 1026e827f923SEric Sesterhenn BUG_ON(!PageLocked(page)); 10271da177e4SLinus Torvalds 10281da177e4SLinus Torvalds if (page_has_buffers(page)) { 10291da177e4SLinus Torvalds bh = page_buffers(page); 10301da177e4SLinus Torvalds if (bh->b_size == size) { 1031676ce6d5SHugh Dickins end_block = init_page_buffers(page, bdev, 1032676ce6d5SHugh Dickins index << sizebits, size); 1033676ce6d5SHugh Dickins goto done; 10341da177e4SLinus Torvalds } 10351da177e4SLinus Torvalds if (!try_to_free_buffers(page)) 10361da177e4SLinus Torvalds goto failed; 10371da177e4SLinus Torvalds } 10381da177e4SLinus Torvalds 10391da177e4SLinus Torvalds /* 10401da177e4SLinus Torvalds * Allocate some buffers for this page 10411da177e4SLinus Torvalds */ 10421da177e4SLinus Torvalds bh = alloc_page_buffers(page, size, 0); 10431da177e4SLinus Torvalds if (!bh) 10441da177e4SLinus Torvalds goto failed; 10451da177e4SLinus Torvalds 10461da177e4SLinus Torvalds /* 10471da177e4SLinus Torvalds * Link the page to the buffers and initialise them. Take the 10481da177e4SLinus Torvalds * lock to be atomic wrt __find_get_block(), which does not 10491da177e4SLinus Torvalds * run under the page lock. 10501da177e4SLinus Torvalds */ 10511da177e4SLinus Torvalds spin_lock(&inode->i_mapping->private_lock); 10521da177e4SLinus Torvalds link_dev_buffers(page, bh); 1053676ce6d5SHugh Dickins end_block = init_page_buffers(page, bdev, index << sizebits, size); 10541da177e4SLinus Torvalds spin_unlock(&inode->i_mapping->private_lock); 1055676ce6d5SHugh Dickins done: 1056676ce6d5SHugh Dickins ret = (block < end_block) ? 1 : -ENXIO; 10571da177e4SLinus Torvalds failed: 10581da177e4SLinus Torvalds unlock_page(page); 10591da177e4SLinus Torvalds page_cache_release(page); 1060676ce6d5SHugh Dickins return ret; 10611da177e4SLinus Torvalds } 10621da177e4SLinus Torvalds 10631da177e4SLinus Torvalds /* 10641da177e4SLinus Torvalds * Create buffers for the specified block device block's page. If 10651da177e4SLinus Torvalds * that page was dirty, the buffers are set dirty also. 10661da177e4SLinus Torvalds */ 1067858119e1SArjan van de Ven static int 10681da177e4SLinus Torvalds grow_buffers(struct block_device *bdev, sector_t block, int size) 10691da177e4SLinus Torvalds { 10701da177e4SLinus Torvalds pgoff_t index; 10711da177e4SLinus Torvalds int sizebits; 10721da177e4SLinus Torvalds 10731da177e4SLinus Torvalds sizebits = -1; 10741da177e4SLinus Torvalds do { 10751da177e4SLinus Torvalds sizebits++; 10761da177e4SLinus Torvalds } while ((size << sizebits) < PAGE_SIZE); 10771da177e4SLinus Torvalds 10781da177e4SLinus Torvalds index = block >> sizebits; 10791da177e4SLinus Torvalds 1080e5657933SAndrew Morton /* 1081e5657933SAndrew Morton * Check for a block which wants to lie outside our maximum possible 1082e5657933SAndrew Morton * pagecache index. (this comparison is done using sector_t types). 1083e5657933SAndrew Morton */ 1084e5657933SAndrew Morton if (unlikely(index != block >> sizebits)) { 1085e5657933SAndrew Morton char b[BDEVNAME_SIZE]; 1086e5657933SAndrew Morton 1087e5657933SAndrew Morton printk(KERN_ERR "%s: requested out-of-range block %llu for " 1088e5657933SAndrew Morton "device %s\n", 10898e24eea7SHarvey Harrison __func__, (unsigned long long)block, 1090e5657933SAndrew Morton bdevname(bdev, b)); 1091e5657933SAndrew Morton return -EIO; 1092e5657933SAndrew Morton } 1093676ce6d5SHugh Dickins 10941da177e4SLinus Torvalds /* Create a page with the proper size buffers.. */ 1095676ce6d5SHugh Dickins return grow_dev_page(bdev, block, index, size, sizebits); 10961da177e4SLinus Torvalds } 10971da177e4SLinus Torvalds 109875c96f85SAdrian Bunk static struct buffer_head * 10991da177e4SLinus Torvalds __getblk_slow(struct block_device *bdev, sector_t block, int size) 11001da177e4SLinus Torvalds { 11011da177e4SLinus Torvalds /* Size must be multiple of hard sectorsize */ 1102e1defc4fSMartin K. Petersen if (unlikely(size & (bdev_logical_block_size(bdev)-1) || 11031da177e4SLinus Torvalds (size < 512 || size > PAGE_SIZE))) { 11041da177e4SLinus Torvalds printk(KERN_ERR "getblk(): invalid block size %d requested\n", 11051da177e4SLinus Torvalds size); 1106e1defc4fSMartin K. Petersen printk(KERN_ERR "logical block size: %d\n", 1107e1defc4fSMartin K. Petersen bdev_logical_block_size(bdev)); 11081da177e4SLinus Torvalds 11091da177e4SLinus Torvalds dump_stack(); 11101da177e4SLinus Torvalds return NULL; 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds 1113676ce6d5SHugh Dickins for (;;) { 1114676ce6d5SHugh Dickins struct buffer_head *bh; 1115676ce6d5SHugh Dickins int ret; 1116676ce6d5SHugh Dickins 11171da177e4SLinus Torvalds bh = __find_get_block(bdev, block, size); 11181da177e4SLinus Torvalds if (bh) 11191da177e4SLinus Torvalds return bh; 11201da177e4SLinus Torvalds 1121e5657933SAndrew Morton ret = grow_buffers(bdev, block, size); 1122676ce6d5SHugh Dickins if (ret < 0) 112391f68c89SJeff Moyer return NULL; 1124676ce6d5SHugh Dickins if (ret == 0) 1125676ce6d5SHugh Dickins free_more_memory(); 1126676ce6d5SHugh Dickins } 11271da177e4SLinus Torvalds } 11281da177e4SLinus Torvalds 11291da177e4SLinus Torvalds /* 11301da177e4SLinus Torvalds * The relationship between dirty buffers and dirty pages: 11311da177e4SLinus Torvalds * 11321da177e4SLinus Torvalds * Whenever a page has any dirty buffers, the page's dirty bit is set, and 11331da177e4SLinus Torvalds * the page is tagged dirty in its radix tree. 11341da177e4SLinus Torvalds * 11351da177e4SLinus Torvalds * At all times, the dirtiness of the buffers represents the dirtiness of 11361da177e4SLinus Torvalds * subsections of the page. If the page has buffers, the page dirty bit is 11371da177e4SLinus Torvalds * merely a hint about the true dirty state. 11381da177e4SLinus Torvalds * 11391da177e4SLinus Torvalds * When a page is set dirty in its entirety, all its buffers are marked dirty 11401da177e4SLinus Torvalds * (if the page has buffers). 11411da177e4SLinus Torvalds * 11421da177e4SLinus Torvalds * When a buffer is marked dirty, its page is dirtied, but the page's other 11431da177e4SLinus Torvalds * buffers are not. 11441da177e4SLinus Torvalds * 11451da177e4SLinus Torvalds * Also. When blockdev buffers are explicitly read with bread(), they 11461da177e4SLinus Torvalds * individually become uptodate. But their backing page remains not 11471da177e4SLinus Torvalds * uptodate - even if all of its buffers are uptodate. A subsequent 11481da177e4SLinus Torvalds * block_read_full_page() against that page will discover all the uptodate 11491da177e4SLinus Torvalds * buffers, will set the page uptodate and will perform no I/O. 11501da177e4SLinus Torvalds */ 11511da177e4SLinus Torvalds 11521da177e4SLinus Torvalds /** 11531da177e4SLinus Torvalds * mark_buffer_dirty - mark a buffer_head as needing writeout 115467be2dd1SMartin Waitz * @bh: the buffer_head to mark dirty 11551da177e4SLinus Torvalds * 11561da177e4SLinus Torvalds * mark_buffer_dirty() will set the dirty bit against the buffer, then set its 11571da177e4SLinus Torvalds * backing page dirty, then tag the page as dirty in its address_space's radix 11581da177e4SLinus Torvalds * tree and then attach the address_space's inode to its superblock's dirty 11591da177e4SLinus Torvalds * inode list. 11601da177e4SLinus Torvalds * 11611da177e4SLinus Torvalds * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock, 1162250df6edSDave Chinner * mapping->tree_lock and mapping->host->i_lock. 11631da177e4SLinus Torvalds */ 1164fc9b52cdSHarvey Harrison void mark_buffer_dirty(struct buffer_head *bh) 11651da177e4SLinus Torvalds { 1166787d2214SNick Piggin WARN_ON_ONCE(!buffer_uptodate(bh)); 11671be62dc1SLinus Torvalds 11685305cb83STejun Heo trace_block_dirty_buffer(bh); 11695305cb83STejun Heo 11701be62dc1SLinus Torvalds /* 11711be62dc1SLinus Torvalds * Very *carefully* optimize the it-is-already-dirty case. 11721be62dc1SLinus Torvalds * 11731be62dc1SLinus Torvalds * Don't let the final "is it dirty" escape to before we 11741be62dc1SLinus Torvalds * perhaps modified the buffer. 11751be62dc1SLinus Torvalds */ 11761be62dc1SLinus Torvalds if (buffer_dirty(bh)) { 11771be62dc1SLinus Torvalds smp_mb(); 11781be62dc1SLinus Torvalds if (buffer_dirty(bh)) 11791be62dc1SLinus Torvalds return; 11801be62dc1SLinus Torvalds } 11811be62dc1SLinus Torvalds 1182a8e7d49aSLinus Torvalds if (!test_set_buffer_dirty(bh)) { 1183a8e7d49aSLinus Torvalds struct page *page = bh->b_page; 11848e9d78edSLinus Torvalds if (!TestSetPageDirty(page)) { 11858e9d78edSLinus Torvalds struct address_space *mapping = page_mapping(page); 11868e9d78edSLinus Torvalds if (mapping) 11878e9d78edSLinus Torvalds __set_page_dirty(page, mapping, 0); 11888e9d78edSLinus Torvalds } 1189a8e7d49aSLinus Torvalds } 11901da177e4SLinus Torvalds } 11911fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(mark_buffer_dirty); 11921da177e4SLinus Torvalds 11931da177e4SLinus Torvalds /* 11941da177e4SLinus Torvalds * Decrement a buffer_head's reference count. If all buffers against a page 11951da177e4SLinus Torvalds * have zero reference count, are clean and unlocked, and if the page is clean 11961da177e4SLinus Torvalds * and unlocked then try_to_free_buffers() may strip the buffers from the page 11971da177e4SLinus Torvalds * in preparation for freeing it (sometimes, rarely, buffers are removed from 11981da177e4SLinus Torvalds * a page but it ends up not being freed, and buffers may later be reattached). 11991da177e4SLinus Torvalds */ 12001da177e4SLinus Torvalds void __brelse(struct buffer_head * buf) 12011da177e4SLinus Torvalds { 12021da177e4SLinus Torvalds if (atomic_read(&buf->b_count)) { 12031da177e4SLinus Torvalds put_bh(buf); 12041da177e4SLinus Torvalds return; 12051da177e4SLinus Torvalds } 12065c752ad9SArjan van de Ven WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); 12071da177e4SLinus Torvalds } 12081fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__brelse); 12091da177e4SLinus Torvalds 12101da177e4SLinus Torvalds /* 12111da177e4SLinus Torvalds * bforget() is like brelse(), except it discards any 12121da177e4SLinus Torvalds * potentially dirty data. 12131da177e4SLinus Torvalds */ 12141da177e4SLinus Torvalds void __bforget(struct buffer_head *bh) 12151da177e4SLinus Torvalds { 12161da177e4SLinus Torvalds clear_buffer_dirty(bh); 1217535ee2fbSJan Kara if (bh->b_assoc_map) { 12181da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 12211da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 122258ff407bSJan Kara bh->b_assoc_map = NULL; 12231da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 12241da177e4SLinus Torvalds } 12251da177e4SLinus Torvalds __brelse(bh); 12261da177e4SLinus Torvalds } 12271fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__bforget); 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds static struct buffer_head *__bread_slow(struct buffer_head *bh) 12301da177e4SLinus Torvalds { 12311da177e4SLinus Torvalds lock_buffer(bh); 12321da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 12331da177e4SLinus Torvalds unlock_buffer(bh); 12341da177e4SLinus Torvalds return bh; 12351da177e4SLinus Torvalds } else { 12361da177e4SLinus Torvalds get_bh(bh); 12371da177e4SLinus Torvalds bh->b_end_io = end_buffer_read_sync; 12381da177e4SLinus Torvalds submit_bh(READ, bh); 12391da177e4SLinus Torvalds wait_on_buffer(bh); 12401da177e4SLinus Torvalds if (buffer_uptodate(bh)) 12411da177e4SLinus Torvalds return bh; 12421da177e4SLinus Torvalds } 12431da177e4SLinus Torvalds brelse(bh); 12441da177e4SLinus Torvalds return NULL; 12451da177e4SLinus Torvalds } 12461da177e4SLinus Torvalds 12471da177e4SLinus Torvalds /* 12481da177e4SLinus Torvalds * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). 12491da177e4SLinus Torvalds * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their 12501da177e4SLinus Torvalds * refcount elevated by one when they're in an LRU. A buffer can only appear 12511da177e4SLinus Torvalds * once in a particular CPU's LRU. A single buffer can be present in multiple 12521da177e4SLinus Torvalds * CPU's LRUs at the same time. 12531da177e4SLinus Torvalds * 12541da177e4SLinus Torvalds * This is a transparent caching front-end to sb_bread(), sb_getblk() and 12551da177e4SLinus Torvalds * sb_find_get_block(). 12561da177e4SLinus Torvalds * 12571da177e4SLinus Torvalds * The LRUs themselves only need locking against invalidate_bh_lrus. We use 12581da177e4SLinus Torvalds * a local interrupt disable for that. 12591da177e4SLinus Torvalds */ 12601da177e4SLinus Torvalds 12611da177e4SLinus Torvalds #define BH_LRU_SIZE 8 12621da177e4SLinus Torvalds 12631da177e4SLinus Torvalds struct bh_lru { 12641da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12651da177e4SLinus Torvalds }; 12661da177e4SLinus Torvalds 12671da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; 12681da177e4SLinus Torvalds 12691da177e4SLinus Torvalds #ifdef CONFIG_SMP 12701da177e4SLinus Torvalds #define bh_lru_lock() local_irq_disable() 12711da177e4SLinus Torvalds #define bh_lru_unlock() local_irq_enable() 12721da177e4SLinus Torvalds #else 12731da177e4SLinus Torvalds #define bh_lru_lock() preempt_disable() 12741da177e4SLinus Torvalds #define bh_lru_unlock() preempt_enable() 12751da177e4SLinus Torvalds #endif 12761da177e4SLinus Torvalds 12771da177e4SLinus Torvalds static inline void check_irqs_on(void) 12781da177e4SLinus Torvalds { 12791da177e4SLinus Torvalds #ifdef irqs_disabled 12801da177e4SLinus Torvalds BUG_ON(irqs_disabled()); 12811da177e4SLinus Torvalds #endif 12821da177e4SLinus Torvalds } 12831da177e4SLinus Torvalds 12841da177e4SLinus Torvalds /* 12851da177e4SLinus Torvalds * The LRU management algorithm is dopey-but-simple. Sorry. 12861da177e4SLinus Torvalds */ 12871da177e4SLinus Torvalds static void bh_lru_install(struct buffer_head *bh) 12881da177e4SLinus Torvalds { 12891da177e4SLinus Torvalds struct buffer_head *evictee = NULL; 12901da177e4SLinus Torvalds 12911da177e4SLinus Torvalds check_irqs_on(); 12921da177e4SLinus Torvalds bh_lru_lock(); 1293c7b92516SChristoph Lameter if (__this_cpu_read(bh_lrus.bhs[0]) != bh) { 12941da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12951da177e4SLinus Torvalds int in; 12961da177e4SLinus Torvalds int out = 0; 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds get_bh(bh); 12991da177e4SLinus Torvalds bhs[out++] = bh; 13001da177e4SLinus Torvalds for (in = 0; in < BH_LRU_SIZE; in++) { 1301c7b92516SChristoph Lameter struct buffer_head *bh2 = 1302c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[in]); 13031da177e4SLinus Torvalds 13041da177e4SLinus Torvalds if (bh2 == bh) { 13051da177e4SLinus Torvalds __brelse(bh2); 13061da177e4SLinus Torvalds } else { 13071da177e4SLinus Torvalds if (out >= BH_LRU_SIZE) { 13081da177e4SLinus Torvalds BUG_ON(evictee != NULL); 13091da177e4SLinus Torvalds evictee = bh2; 13101da177e4SLinus Torvalds } else { 13111da177e4SLinus Torvalds bhs[out++] = bh2; 13121da177e4SLinus Torvalds } 13131da177e4SLinus Torvalds } 13141da177e4SLinus Torvalds } 13151da177e4SLinus Torvalds while (out < BH_LRU_SIZE) 13161da177e4SLinus Torvalds bhs[out++] = NULL; 1317ca6673b0SChristoph Lameter memcpy(this_cpu_ptr(&bh_lrus.bhs), bhs, sizeof(bhs)); 13181da177e4SLinus Torvalds } 13191da177e4SLinus Torvalds bh_lru_unlock(); 13201da177e4SLinus Torvalds 13211da177e4SLinus Torvalds if (evictee) 13221da177e4SLinus Torvalds __brelse(evictee); 13231da177e4SLinus Torvalds } 13241da177e4SLinus Torvalds 13251da177e4SLinus Torvalds /* 13261da177e4SLinus Torvalds * Look up the bh in this cpu's LRU. If it's there, move it to the head. 13271da177e4SLinus Torvalds */ 1328858119e1SArjan van de Ven static struct buffer_head * 13293991d3bdSTomasz Kvarsin lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) 13301da177e4SLinus Torvalds { 13311da177e4SLinus Torvalds struct buffer_head *ret = NULL; 13323991d3bdSTomasz Kvarsin unsigned int i; 13331da177e4SLinus Torvalds 13341da177e4SLinus Torvalds check_irqs_on(); 13351da177e4SLinus Torvalds bh_lru_lock(); 13361da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 1337c7b92516SChristoph Lameter struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds if (bh && bh->b_bdev == bdev && 13401da177e4SLinus Torvalds bh->b_blocknr == block && bh->b_size == size) { 13411da177e4SLinus Torvalds if (i) { 13421da177e4SLinus Torvalds while (i) { 1343c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[i], 1344c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[i - 1])); 13451da177e4SLinus Torvalds i--; 13461da177e4SLinus Torvalds } 1347c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[0], bh); 13481da177e4SLinus Torvalds } 13491da177e4SLinus Torvalds get_bh(bh); 13501da177e4SLinus Torvalds ret = bh; 13511da177e4SLinus Torvalds break; 13521da177e4SLinus Torvalds } 13531da177e4SLinus Torvalds } 13541da177e4SLinus Torvalds bh_lru_unlock(); 13551da177e4SLinus Torvalds return ret; 13561da177e4SLinus Torvalds } 13571da177e4SLinus Torvalds 13581da177e4SLinus Torvalds /* 13591da177e4SLinus Torvalds * Perform a pagecache lookup for the matching buffer. If it's there, refresh 13601da177e4SLinus Torvalds * it in the LRU and mark it as accessed. If it is not present then return 13611da177e4SLinus Torvalds * NULL 13621da177e4SLinus Torvalds */ 13631da177e4SLinus Torvalds struct buffer_head * 13643991d3bdSTomasz Kvarsin __find_get_block(struct block_device *bdev, sector_t block, unsigned size) 13651da177e4SLinus Torvalds { 13661da177e4SLinus Torvalds struct buffer_head *bh = lookup_bh_lru(bdev, block, size); 13671da177e4SLinus Torvalds 13681da177e4SLinus Torvalds if (bh == NULL) { 1369385fd4c5SCoywolf Qi Hunt bh = __find_get_block_slow(bdev, block); 13701da177e4SLinus Torvalds if (bh) 13711da177e4SLinus Torvalds bh_lru_install(bh); 13721da177e4SLinus Torvalds } 13731da177e4SLinus Torvalds if (bh) 13741da177e4SLinus Torvalds touch_buffer(bh); 13751da177e4SLinus Torvalds return bh; 13761da177e4SLinus Torvalds } 13771da177e4SLinus Torvalds EXPORT_SYMBOL(__find_get_block); 13781da177e4SLinus Torvalds 13791da177e4SLinus Torvalds /* 13801da177e4SLinus Torvalds * __getblk will locate (and, if necessary, create) the buffer_head 13811da177e4SLinus Torvalds * which corresponds to the passed block_device, block and size. The 13821da177e4SLinus Torvalds * returned buffer has its reference count incremented. 13831da177e4SLinus Torvalds * 13841da177e4SLinus Torvalds * __getblk() will lock up the machine if grow_dev_page's try_to_free_buffers() 13851da177e4SLinus Torvalds * attempt is failing. FIXME, perhaps? 13861da177e4SLinus Torvalds */ 13871da177e4SLinus Torvalds struct buffer_head * 13883991d3bdSTomasz Kvarsin __getblk(struct block_device *bdev, sector_t block, unsigned size) 13891da177e4SLinus Torvalds { 13901da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, block, size); 13911da177e4SLinus Torvalds 13921da177e4SLinus Torvalds might_sleep(); 13931da177e4SLinus Torvalds if (bh == NULL) 13941da177e4SLinus Torvalds bh = __getblk_slow(bdev, block, size); 13951da177e4SLinus Torvalds return bh; 13961da177e4SLinus Torvalds } 13971da177e4SLinus Torvalds EXPORT_SYMBOL(__getblk); 13981da177e4SLinus Torvalds 13991da177e4SLinus Torvalds /* 14001da177e4SLinus Torvalds * Do async read-ahead on a buffer.. 14011da177e4SLinus Torvalds */ 14023991d3bdSTomasz Kvarsin void __breadahead(struct block_device *bdev, sector_t block, unsigned size) 14031da177e4SLinus Torvalds { 14041da177e4SLinus Torvalds struct buffer_head *bh = __getblk(bdev, block, size); 1405a3e713b5SAndrew Morton if (likely(bh)) { 14061da177e4SLinus Torvalds ll_rw_block(READA, 1, &bh); 14071da177e4SLinus Torvalds brelse(bh); 14081da177e4SLinus Torvalds } 1409a3e713b5SAndrew Morton } 14101da177e4SLinus Torvalds EXPORT_SYMBOL(__breadahead); 14111da177e4SLinus Torvalds 14121da177e4SLinus Torvalds /** 14131da177e4SLinus Torvalds * __bread() - reads a specified block and returns the bh 141467be2dd1SMartin Waitz * @bdev: the block_device to read from 14151da177e4SLinus Torvalds * @block: number of block 14161da177e4SLinus Torvalds * @size: size (in bytes) to read 14171da177e4SLinus Torvalds * 14181da177e4SLinus Torvalds * Reads a specified block, and returns buffer head that contains it. 14191da177e4SLinus Torvalds * It returns NULL if the block was unreadable. 14201da177e4SLinus Torvalds */ 14211da177e4SLinus Torvalds struct buffer_head * 14223991d3bdSTomasz Kvarsin __bread(struct block_device *bdev, sector_t block, unsigned size) 14231da177e4SLinus Torvalds { 14241da177e4SLinus Torvalds struct buffer_head *bh = __getblk(bdev, block, size); 14251da177e4SLinus Torvalds 1426a3e713b5SAndrew Morton if (likely(bh) && !buffer_uptodate(bh)) 14271da177e4SLinus Torvalds bh = __bread_slow(bh); 14281da177e4SLinus Torvalds return bh; 14291da177e4SLinus Torvalds } 14301da177e4SLinus Torvalds EXPORT_SYMBOL(__bread); 14311da177e4SLinus Torvalds 14321da177e4SLinus Torvalds /* 14331da177e4SLinus Torvalds * invalidate_bh_lrus() is called rarely - but not only at unmount. 14341da177e4SLinus Torvalds * This doesn't race because it runs in each cpu either in irq 14351da177e4SLinus Torvalds * or with preempt disabled. 14361da177e4SLinus Torvalds */ 14371da177e4SLinus Torvalds static void invalidate_bh_lru(void *arg) 14381da177e4SLinus Torvalds { 14391da177e4SLinus Torvalds struct bh_lru *b = &get_cpu_var(bh_lrus); 14401da177e4SLinus Torvalds int i; 14411da177e4SLinus Torvalds 14421da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 14431da177e4SLinus Torvalds brelse(b->bhs[i]); 14441da177e4SLinus Torvalds b->bhs[i] = NULL; 14451da177e4SLinus Torvalds } 14461da177e4SLinus Torvalds put_cpu_var(bh_lrus); 14471da177e4SLinus Torvalds } 14481da177e4SLinus Torvalds 144942be35d0SGilad Ben-Yossef static bool has_bh_in_lru(int cpu, void *dummy) 145042be35d0SGilad Ben-Yossef { 145142be35d0SGilad Ben-Yossef struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); 145242be35d0SGilad Ben-Yossef int i; 145342be35d0SGilad Ben-Yossef 145442be35d0SGilad Ben-Yossef for (i = 0; i < BH_LRU_SIZE; i++) { 145542be35d0SGilad Ben-Yossef if (b->bhs[i]) 145642be35d0SGilad Ben-Yossef return 1; 145742be35d0SGilad Ben-Yossef } 145842be35d0SGilad Ben-Yossef 145942be35d0SGilad Ben-Yossef return 0; 146042be35d0SGilad Ben-Yossef } 146142be35d0SGilad Ben-Yossef 1462f9a14399SPeter Zijlstra void invalidate_bh_lrus(void) 14631da177e4SLinus Torvalds { 146442be35d0SGilad Ben-Yossef on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL); 14651da177e4SLinus Torvalds } 14669db5579bSNick Piggin EXPORT_SYMBOL_GPL(invalidate_bh_lrus); 14671da177e4SLinus Torvalds 14681da177e4SLinus Torvalds void set_bh_page(struct buffer_head *bh, 14691da177e4SLinus Torvalds struct page *page, unsigned long offset) 14701da177e4SLinus Torvalds { 14711da177e4SLinus Torvalds bh->b_page = page; 1472e827f923SEric Sesterhenn BUG_ON(offset >= PAGE_SIZE); 14731da177e4SLinus Torvalds if (PageHighMem(page)) 14741da177e4SLinus Torvalds /* 14751da177e4SLinus Torvalds * This catches illegal uses and preserves the offset: 14761da177e4SLinus Torvalds */ 14771da177e4SLinus Torvalds bh->b_data = (char *)(0 + offset); 14781da177e4SLinus Torvalds else 14791da177e4SLinus Torvalds bh->b_data = page_address(page) + offset; 14801da177e4SLinus Torvalds } 14811da177e4SLinus Torvalds EXPORT_SYMBOL(set_bh_page); 14821da177e4SLinus Torvalds 14831da177e4SLinus Torvalds /* 14841da177e4SLinus Torvalds * Called when truncating a buffer on a page completely. 14851da177e4SLinus Torvalds */ 1486858119e1SArjan van de Ven static void discard_buffer(struct buffer_head * bh) 14871da177e4SLinus Torvalds { 14881da177e4SLinus Torvalds lock_buffer(bh); 14891da177e4SLinus Torvalds clear_buffer_dirty(bh); 14901da177e4SLinus Torvalds bh->b_bdev = NULL; 14911da177e4SLinus Torvalds clear_buffer_mapped(bh); 14921da177e4SLinus Torvalds clear_buffer_req(bh); 14931da177e4SLinus Torvalds clear_buffer_new(bh); 14941da177e4SLinus Torvalds clear_buffer_delay(bh); 149533a266ddSDavid Chinner clear_buffer_unwritten(bh); 14961da177e4SLinus Torvalds unlock_buffer(bh); 14971da177e4SLinus Torvalds } 14981da177e4SLinus Torvalds 14991da177e4SLinus Torvalds /** 1500814e1d25SWang Sheng-Hui * block_invalidatepage - invalidate part or all of a buffer-backed page 15011da177e4SLinus Torvalds * 15021da177e4SLinus Torvalds * @page: the page which is affected 1503d47992f8SLukas Czerner * @offset: start of the range to invalidate 1504d47992f8SLukas Czerner * @length: length of the range to invalidate 15051da177e4SLinus Torvalds * 15061da177e4SLinus Torvalds * block_invalidatepage() is called when all or part of the page has become 15071da177e4SLinus Torvalds * invalidated by a truncate operation. 15081da177e4SLinus Torvalds * 15091da177e4SLinus Torvalds * block_invalidatepage() does not have to release all buffers, but it must 15101da177e4SLinus Torvalds * ensure that no dirty buffer is left outside @offset and that no I/O 15111da177e4SLinus Torvalds * is underway against any of the blocks which are outside the truncation 15121da177e4SLinus Torvalds * point. Because the caller is about to free (and possibly reuse) those 15131da177e4SLinus Torvalds * blocks on-disk. 15141da177e4SLinus Torvalds */ 1515d47992f8SLukas Czerner void block_invalidatepage(struct page *page, unsigned int offset, 1516d47992f8SLukas Czerner unsigned int length) 15171da177e4SLinus Torvalds { 15181da177e4SLinus Torvalds struct buffer_head *head, *bh, *next; 15191da177e4SLinus Torvalds unsigned int curr_off = 0; 1520d47992f8SLukas Czerner unsigned int stop = length + offset; 15211da177e4SLinus Torvalds 15221da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 15231da177e4SLinus Torvalds if (!page_has_buffers(page)) 15241da177e4SLinus Torvalds goto out; 15251da177e4SLinus Torvalds 1526d47992f8SLukas Czerner /* 1527d47992f8SLukas Czerner * Check for overflow 1528d47992f8SLukas Czerner */ 1529d47992f8SLukas Czerner BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); 1530d47992f8SLukas Czerner 15311da177e4SLinus Torvalds head = page_buffers(page); 15321da177e4SLinus Torvalds bh = head; 15331da177e4SLinus Torvalds do { 15341da177e4SLinus Torvalds unsigned int next_off = curr_off + bh->b_size; 15351da177e4SLinus Torvalds next = bh->b_this_page; 15361da177e4SLinus Torvalds 15371da177e4SLinus Torvalds /* 1538d47992f8SLukas Czerner * Are we still fully in range ? 1539d47992f8SLukas Czerner */ 1540d47992f8SLukas Czerner if (next_off > stop) 1541d47992f8SLukas Czerner goto out; 1542d47992f8SLukas Czerner 1543d47992f8SLukas Czerner /* 15441da177e4SLinus Torvalds * is this block fully invalidated? 15451da177e4SLinus Torvalds */ 15461da177e4SLinus Torvalds if (offset <= curr_off) 15471da177e4SLinus Torvalds discard_buffer(bh); 15481da177e4SLinus Torvalds curr_off = next_off; 15491da177e4SLinus Torvalds bh = next; 15501da177e4SLinus Torvalds } while (bh != head); 15511da177e4SLinus Torvalds 15521da177e4SLinus Torvalds /* 15531da177e4SLinus Torvalds * We release buffers only if the entire page is being invalidated. 15541da177e4SLinus Torvalds * The get_block cached value has been unconditionally invalidated, 15551da177e4SLinus Torvalds * so real IO is not possible anymore. 15561da177e4SLinus Torvalds */ 15571da177e4SLinus Torvalds if (offset == 0) 15582ff28e22SNeilBrown try_to_release_page(page, 0); 15591da177e4SLinus Torvalds out: 15602ff28e22SNeilBrown return; 15611da177e4SLinus Torvalds } 15621da177e4SLinus Torvalds EXPORT_SYMBOL(block_invalidatepage); 15631da177e4SLinus Torvalds 1564d47992f8SLukas Czerner 15651da177e4SLinus Torvalds /* 15661da177e4SLinus Torvalds * We attach and possibly dirty the buffers atomically wrt 15671da177e4SLinus Torvalds * __set_page_dirty_buffers() via private_lock. try_to_free_buffers 15681da177e4SLinus Torvalds * is already excluded via the page lock. 15691da177e4SLinus Torvalds */ 15701da177e4SLinus Torvalds void create_empty_buffers(struct page *page, 15711da177e4SLinus Torvalds unsigned long blocksize, unsigned long b_state) 15721da177e4SLinus Torvalds { 15731da177e4SLinus Torvalds struct buffer_head *bh, *head, *tail; 15741da177e4SLinus Torvalds 15751da177e4SLinus Torvalds head = alloc_page_buffers(page, blocksize, 1); 15761da177e4SLinus Torvalds bh = head; 15771da177e4SLinus Torvalds do { 15781da177e4SLinus Torvalds bh->b_state |= b_state; 15791da177e4SLinus Torvalds tail = bh; 15801da177e4SLinus Torvalds bh = bh->b_this_page; 15811da177e4SLinus Torvalds } while (bh); 15821da177e4SLinus Torvalds tail->b_this_page = head; 15831da177e4SLinus Torvalds 15841da177e4SLinus Torvalds spin_lock(&page->mapping->private_lock); 15851da177e4SLinus Torvalds if (PageUptodate(page) || PageDirty(page)) { 15861da177e4SLinus Torvalds bh = head; 15871da177e4SLinus Torvalds do { 15881da177e4SLinus Torvalds if (PageDirty(page)) 15891da177e4SLinus Torvalds set_buffer_dirty(bh); 15901da177e4SLinus Torvalds if (PageUptodate(page)) 15911da177e4SLinus Torvalds set_buffer_uptodate(bh); 15921da177e4SLinus Torvalds bh = bh->b_this_page; 15931da177e4SLinus Torvalds } while (bh != head); 15941da177e4SLinus Torvalds } 15951da177e4SLinus Torvalds attach_page_buffers(page, head); 15961da177e4SLinus Torvalds spin_unlock(&page->mapping->private_lock); 15971da177e4SLinus Torvalds } 15981da177e4SLinus Torvalds EXPORT_SYMBOL(create_empty_buffers); 15991da177e4SLinus Torvalds 16001da177e4SLinus Torvalds /* 16011da177e4SLinus Torvalds * We are taking a block for data and we don't want any output from any 16021da177e4SLinus Torvalds * buffer-cache aliases starting from return from that function and 16031da177e4SLinus Torvalds * until the moment when something will explicitly mark the buffer 16041da177e4SLinus Torvalds * dirty (hopefully that will not happen until we will free that block ;-) 16051da177e4SLinus Torvalds * We don't even need to mark it not-uptodate - nobody can expect 16061da177e4SLinus Torvalds * anything from a newly allocated buffer anyway. We used to used 16071da177e4SLinus Torvalds * unmap_buffer() for such invalidation, but that was wrong. We definitely 16081da177e4SLinus Torvalds * don't want to mark the alias unmapped, for example - it would confuse 16091da177e4SLinus Torvalds * anyone who might pick it with bread() afterwards... 16101da177e4SLinus Torvalds * 16111da177e4SLinus Torvalds * Also.. Note that bforget() doesn't lock the buffer. So there can 16121da177e4SLinus Torvalds * be writeout I/O going on against recently-freed buffers. We don't 16131da177e4SLinus Torvalds * wait on that I/O in bforget() - it's more efficient to wait on the I/O 16141da177e4SLinus Torvalds * only if we really need to. That happens here. 16151da177e4SLinus Torvalds */ 16161da177e4SLinus Torvalds void unmap_underlying_metadata(struct block_device *bdev, sector_t block) 16171da177e4SLinus Torvalds { 16181da177e4SLinus Torvalds struct buffer_head *old_bh; 16191da177e4SLinus Torvalds 16201da177e4SLinus Torvalds might_sleep(); 16211da177e4SLinus Torvalds 1622385fd4c5SCoywolf Qi Hunt old_bh = __find_get_block_slow(bdev, block); 16231da177e4SLinus Torvalds if (old_bh) { 16241da177e4SLinus Torvalds clear_buffer_dirty(old_bh); 16251da177e4SLinus Torvalds wait_on_buffer(old_bh); 16261da177e4SLinus Torvalds clear_buffer_req(old_bh); 16271da177e4SLinus Torvalds __brelse(old_bh); 16281da177e4SLinus Torvalds } 16291da177e4SLinus Torvalds } 16301da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_underlying_metadata); 16311da177e4SLinus Torvalds 16321da177e4SLinus Torvalds /* 163345bce8f3SLinus Torvalds * Size is a power-of-two in the range 512..PAGE_SIZE, 163445bce8f3SLinus Torvalds * and the case we care about most is PAGE_SIZE. 163545bce8f3SLinus Torvalds * 163645bce8f3SLinus Torvalds * So this *could* possibly be written with those 163745bce8f3SLinus Torvalds * constraints in mind (relevant mostly if some 163845bce8f3SLinus Torvalds * architecture has a slow bit-scan instruction) 163945bce8f3SLinus Torvalds */ 164045bce8f3SLinus Torvalds static inline int block_size_bits(unsigned int blocksize) 164145bce8f3SLinus Torvalds { 164245bce8f3SLinus Torvalds return ilog2(blocksize); 164345bce8f3SLinus Torvalds } 164445bce8f3SLinus Torvalds 164545bce8f3SLinus Torvalds static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state) 164645bce8f3SLinus Torvalds { 164745bce8f3SLinus Torvalds BUG_ON(!PageLocked(page)); 164845bce8f3SLinus Torvalds 164945bce8f3SLinus Torvalds if (!page_has_buffers(page)) 165045bce8f3SLinus Torvalds create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state); 165145bce8f3SLinus Torvalds return page_buffers(page); 165245bce8f3SLinus Torvalds } 165345bce8f3SLinus Torvalds 165445bce8f3SLinus Torvalds /* 16551da177e4SLinus Torvalds * NOTE! All mapped/uptodate combinations are valid: 16561da177e4SLinus Torvalds * 16571da177e4SLinus Torvalds * Mapped Uptodate Meaning 16581da177e4SLinus Torvalds * 16591da177e4SLinus Torvalds * No No "unknown" - must do get_block() 16601da177e4SLinus Torvalds * No Yes "hole" - zero-filled 16611da177e4SLinus Torvalds * Yes No "allocated" - allocated on disk, not read in 16621da177e4SLinus Torvalds * Yes Yes "valid" - allocated and up-to-date in memory. 16631da177e4SLinus Torvalds * 16641da177e4SLinus Torvalds * "Dirty" is valid only with the last case (mapped+uptodate). 16651da177e4SLinus Torvalds */ 16661da177e4SLinus Torvalds 16671da177e4SLinus Torvalds /* 16681da177e4SLinus Torvalds * While block_write_full_page is writing back the dirty buffers under 16691da177e4SLinus Torvalds * the page lock, whoever dirtied the buffers may decide to clean them 16701da177e4SLinus Torvalds * again at any time. We handle that by only looking at the buffer 16711da177e4SLinus Torvalds * state inside lock_buffer(). 16721da177e4SLinus Torvalds * 16731da177e4SLinus Torvalds * If block_write_full_page() is called for regular writeback 16741da177e4SLinus Torvalds * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a 16751da177e4SLinus Torvalds * locked buffer. This only can happen if someone has written the buffer 16761da177e4SLinus Torvalds * directly, with submit_bh(). At the address_space level PageWriteback 16771da177e4SLinus Torvalds * prevents this contention from occurring. 16786e34eeddSTheodore Ts'o * 16796e34eeddSTheodore Ts'o * If block_write_full_page() is called with wbc->sync_mode == 1680721a9602SJens Axboe * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this 1681721a9602SJens Axboe * causes the writes to be flagged as synchronous writes. 16821da177e4SLinus Torvalds */ 16831da177e4SLinus Torvalds static int __block_write_full_page(struct inode *inode, struct page *page, 168435c80d5fSChris Mason get_block_t *get_block, struct writeback_control *wbc, 168535c80d5fSChris Mason bh_end_io_t *handler) 16861da177e4SLinus Torvalds { 16871da177e4SLinus Torvalds int err; 16881da177e4SLinus Torvalds sector_t block; 16891da177e4SLinus Torvalds sector_t last_block; 1690f0fbd5fcSAndrew Morton struct buffer_head *bh, *head; 169145bce8f3SLinus Torvalds unsigned int blocksize, bbits; 16921da177e4SLinus Torvalds int nr_underway = 0; 16936e34eeddSTheodore Ts'o int write_op = (wbc->sync_mode == WB_SYNC_ALL ? 1694721a9602SJens Axboe WRITE_SYNC : WRITE); 16951da177e4SLinus Torvalds 169645bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 16971da177e4SLinus Torvalds (1 << BH_Dirty)|(1 << BH_Uptodate)); 16981da177e4SLinus Torvalds 16991da177e4SLinus Torvalds /* 17001da177e4SLinus Torvalds * Be very careful. We have no exclusion from __set_page_dirty_buffers 17011da177e4SLinus Torvalds * here, and the (potentially unmapped) buffers may become dirty at 17021da177e4SLinus Torvalds * any time. If a buffer becomes dirty here after we've inspected it 17031da177e4SLinus Torvalds * then we just miss that fact, and the page stays dirty. 17041da177e4SLinus Torvalds * 17051da177e4SLinus Torvalds * Buffers outside i_size may be dirtied by __set_page_dirty_buffers; 17061da177e4SLinus Torvalds * handle that here by just cleaning them. 17071da177e4SLinus Torvalds */ 17081da177e4SLinus Torvalds 17091da177e4SLinus Torvalds bh = head; 171045bce8f3SLinus Torvalds blocksize = bh->b_size; 171145bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 171245bce8f3SLinus Torvalds 171345bce8f3SLinus Torvalds block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); 171445bce8f3SLinus Torvalds last_block = (i_size_read(inode) - 1) >> bbits; 17151da177e4SLinus Torvalds 17161da177e4SLinus Torvalds /* 17171da177e4SLinus Torvalds * Get all the dirty buffers mapped to disk addresses and 17181da177e4SLinus Torvalds * handle any aliases from the underlying blockdev's mapping. 17191da177e4SLinus Torvalds */ 17201da177e4SLinus Torvalds do { 17211da177e4SLinus Torvalds if (block > last_block) { 17221da177e4SLinus Torvalds /* 17231da177e4SLinus Torvalds * mapped buffers outside i_size will occur, because 17241da177e4SLinus Torvalds * this page can be outside i_size when there is a 17251da177e4SLinus Torvalds * truncate in progress. 17261da177e4SLinus Torvalds */ 17271da177e4SLinus Torvalds /* 17281da177e4SLinus Torvalds * The buffer was zeroed by block_write_full_page() 17291da177e4SLinus Torvalds */ 17301da177e4SLinus Torvalds clear_buffer_dirty(bh); 17311da177e4SLinus Torvalds set_buffer_uptodate(bh); 173229a814d2SAlex Tomas } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && 173329a814d2SAlex Tomas buffer_dirty(bh)) { 1734b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 17351da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 17361da177e4SLinus Torvalds if (err) 17371da177e4SLinus Torvalds goto recover; 173829a814d2SAlex Tomas clear_buffer_delay(bh); 17391da177e4SLinus Torvalds if (buffer_new(bh)) { 17401da177e4SLinus Torvalds /* blockdev mappings never come here */ 17411da177e4SLinus Torvalds clear_buffer_new(bh); 17421da177e4SLinus Torvalds unmap_underlying_metadata(bh->b_bdev, 17431da177e4SLinus Torvalds bh->b_blocknr); 17441da177e4SLinus Torvalds } 17451da177e4SLinus Torvalds } 17461da177e4SLinus Torvalds bh = bh->b_this_page; 17471da177e4SLinus Torvalds block++; 17481da177e4SLinus Torvalds } while (bh != head); 17491da177e4SLinus Torvalds 17501da177e4SLinus Torvalds do { 17511da177e4SLinus Torvalds if (!buffer_mapped(bh)) 17521da177e4SLinus Torvalds continue; 17531da177e4SLinus Torvalds /* 17541da177e4SLinus Torvalds * If it's a fully non-blocking write attempt and we cannot 17551da177e4SLinus Torvalds * lock the buffer then redirty the page. Note that this can 17565b0830cbSJens Axboe * potentially cause a busy-wait loop from writeback threads 17575b0830cbSJens Axboe * and kswapd activity, but those code paths have their own 17585b0830cbSJens Axboe * higher-level throttling. 17591da177e4SLinus Torvalds */ 17601b430beeSWu Fengguang if (wbc->sync_mode != WB_SYNC_NONE) { 17611da177e4SLinus Torvalds lock_buffer(bh); 1762ca5de404SNick Piggin } else if (!trylock_buffer(bh)) { 17631da177e4SLinus Torvalds redirty_page_for_writepage(wbc, page); 17641da177e4SLinus Torvalds continue; 17651da177e4SLinus Torvalds } 17661da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 176735c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 17681da177e4SLinus Torvalds } else { 17691da177e4SLinus Torvalds unlock_buffer(bh); 17701da177e4SLinus Torvalds } 17711da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 17721da177e4SLinus Torvalds 17731da177e4SLinus Torvalds /* 17741da177e4SLinus Torvalds * The page and its buffers are protected by PageWriteback(), so we can 17751da177e4SLinus Torvalds * drop the bh refcounts early. 17761da177e4SLinus Torvalds */ 17771da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 17781da177e4SLinus Torvalds set_page_writeback(page); 17791da177e4SLinus Torvalds 17801da177e4SLinus Torvalds do { 17811da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 17821da177e4SLinus Torvalds if (buffer_async_write(bh)) { 1783a64c8610STheodore Ts'o submit_bh(write_op, bh); 17841da177e4SLinus Torvalds nr_underway++; 1785ad576e63SNick Piggin } 17861da177e4SLinus Torvalds bh = next; 17871da177e4SLinus Torvalds } while (bh != head); 178805937baaSAndrew Morton unlock_page(page); 17891da177e4SLinus Torvalds 17901da177e4SLinus Torvalds err = 0; 17911da177e4SLinus Torvalds done: 17921da177e4SLinus Torvalds if (nr_underway == 0) { 17931da177e4SLinus Torvalds /* 17941da177e4SLinus Torvalds * The page was marked dirty, but the buffers were 17951da177e4SLinus Torvalds * clean. Someone wrote them back by hand with 17961da177e4SLinus Torvalds * ll_rw_block/submit_bh. A rare case. 17971da177e4SLinus Torvalds */ 17981da177e4SLinus Torvalds end_page_writeback(page); 17993d67f2d7SNick Piggin 18001da177e4SLinus Torvalds /* 18011da177e4SLinus Torvalds * The page and buffer_heads can be released at any time from 18021da177e4SLinus Torvalds * here on. 18031da177e4SLinus Torvalds */ 18041da177e4SLinus Torvalds } 18051da177e4SLinus Torvalds return err; 18061da177e4SLinus Torvalds 18071da177e4SLinus Torvalds recover: 18081da177e4SLinus Torvalds /* 18091da177e4SLinus Torvalds * ENOSPC, or some other error. We may already have added some 18101da177e4SLinus Torvalds * blocks to the file, so we need to write these out to avoid 18111da177e4SLinus Torvalds * exposing stale data. 18121da177e4SLinus Torvalds * The page is currently locked and not marked for writeback 18131da177e4SLinus Torvalds */ 18141da177e4SLinus Torvalds bh = head; 18151da177e4SLinus Torvalds /* Recovery: lock and submit the mapped buffers */ 18161da177e4SLinus Torvalds do { 181729a814d2SAlex Tomas if (buffer_mapped(bh) && buffer_dirty(bh) && 181829a814d2SAlex Tomas !buffer_delay(bh)) { 18191da177e4SLinus Torvalds lock_buffer(bh); 182035c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 18211da177e4SLinus Torvalds } else { 18221da177e4SLinus Torvalds /* 18231da177e4SLinus Torvalds * The buffer may have been set dirty during 18241da177e4SLinus Torvalds * attachment to a dirty page. 18251da177e4SLinus Torvalds */ 18261da177e4SLinus Torvalds clear_buffer_dirty(bh); 18271da177e4SLinus Torvalds } 18281da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 18291da177e4SLinus Torvalds SetPageError(page); 18301da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 18317e4c3690SAndrew Morton mapping_set_error(page->mapping, err); 18321da177e4SLinus Torvalds set_page_writeback(page); 18331da177e4SLinus Torvalds do { 18341da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 18351da177e4SLinus Torvalds if (buffer_async_write(bh)) { 18361da177e4SLinus Torvalds clear_buffer_dirty(bh); 1837a64c8610STheodore Ts'o submit_bh(write_op, bh); 18381da177e4SLinus Torvalds nr_underway++; 1839ad576e63SNick Piggin } 18401da177e4SLinus Torvalds bh = next; 18411da177e4SLinus Torvalds } while (bh != head); 1842ffda9d30SNick Piggin unlock_page(page); 18431da177e4SLinus Torvalds goto done; 18441da177e4SLinus Torvalds } 18451da177e4SLinus Torvalds 1846afddba49SNick Piggin /* 1847afddba49SNick Piggin * If a page has any new buffers, zero them out here, and mark them uptodate 1848afddba49SNick Piggin * and dirty so they'll be written out (in order to prevent uninitialised 1849afddba49SNick Piggin * block data from leaking). And clear the new bit. 1850afddba49SNick Piggin */ 1851afddba49SNick Piggin void page_zero_new_buffers(struct page *page, unsigned from, unsigned to) 1852afddba49SNick Piggin { 1853afddba49SNick Piggin unsigned int block_start, block_end; 1854afddba49SNick Piggin struct buffer_head *head, *bh; 1855afddba49SNick Piggin 1856afddba49SNick Piggin BUG_ON(!PageLocked(page)); 1857afddba49SNick Piggin if (!page_has_buffers(page)) 1858afddba49SNick Piggin return; 1859afddba49SNick Piggin 1860afddba49SNick Piggin bh = head = page_buffers(page); 1861afddba49SNick Piggin block_start = 0; 1862afddba49SNick Piggin do { 1863afddba49SNick Piggin block_end = block_start + bh->b_size; 1864afddba49SNick Piggin 1865afddba49SNick Piggin if (buffer_new(bh)) { 1866afddba49SNick Piggin if (block_end > from && block_start < to) { 1867afddba49SNick Piggin if (!PageUptodate(page)) { 1868afddba49SNick Piggin unsigned start, size; 1869afddba49SNick Piggin 1870afddba49SNick Piggin start = max(from, block_start); 1871afddba49SNick Piggin size = min(to, block_end) - start; 1872afddba49SNick Piggin 1873eebd2aa3SChristoph Lameter zero_user(page, start, size); 1874afddba49SNick Piggin set_buffer_uptodate(bh); 1875afddba49SNick Piggin } 1876afddba49SNick Piggin 1877afddba49SNick Piggin clear_buffer_new(bh); 1878afddba49SNick Piggin mark_buffer_dirty(bh); 1879afddba49SNick Piggin } 1880afddba49SNick Piggin } 1881afddba49SNick Piggin 1882afddba49SNick Piggin block_start = block_end; 1883afddba49SNick Piggin bh = bh->b_this_page; 1884afddba49SNick Piggin } while (bh != head); 1885afddba49SNick Piggin } 1886afddba49SNick Piggin EXPORT_SYMBOL(page_zero_new_buffers); 1887afddba49SNick Piggin 1888ebdec241SChristoph Hellwig int __block_write_begin(struct page *page, loff_t pos, unsigned len, 18896e1db88dSChristoph Hellwig get_block_t *get_block) 18901da177e4SLinus Torvalds { 1891ebdec241SChristoph Hellwig unsigned from = pos & (PAGE_CACHE_SIZE - 1); 1892ebdec241SChristoph Hellwig unsigned to = from + len; 18936e1db88dSChristoph Hellwig struct inode *inode = page->mapping->host; 18941da177e4SLinus Torvalds unsigned block_start, block_end; 18951da177e4SLinus Torvalds sector_t block; 18961da177e4SLinus Torvalds int err = 0; 18971da177e4SLinus Torvalds unsigned blocksize, bbits; 18981da177e4SLinus Torvalds struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; 18991da177e4SLinus Torvalds 19001da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 19011da177e4SLinus Torvalds BUG_ON(from > PAGE_CACHE_SIZE); 19021da177e4SLinus Torvalds BUG_ON(to > PAGE_CACHE_SIZE); 19031da177e4SLinus Torvalds BUG_ON(from > to); 19041da177e4SLinus Torvalds 190545bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 0); 190645bce8f3SLinus Torvalds blocksize = head->b_size; 190745bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 19081da177e4SLinus Torvalds 19091da177e4SLinus Torvalds block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); 19101da177e4SLinus Torvalds 19111da177e4SLinus Torvalds for(bh = head, block_start = 0; bh != head || !block_start; 19121da177e4SLinus Torvalds block++, block_start=block_end, bh = bh->b_this_page) { 19131da177e4SLinus Torvalds block_end = block_start + blocksize; 19141da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 19151da177e4SLinus Torvalds if (PageUptodate(page)) { 19161da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 19171da177e4SLinus Torvalds set_buffer_uptodate(bh); 19181da177e4SLinus Torvalds } 19191da177e4SLinus Torvalds continue; 19201da177e4SLinus Torvalds } 19211da177e4SLinus Torvalds if (buffer_new(bh)) 19221da177e4SLinus Torvalds clear_buffer_new(bh); 19231da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 1924b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 19251da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 19261da177e4SLinus Torvalds if (err) 1927f3ddbdc6SNick Piggin break; 19281da177e4SLinus Torvalds if (buffer_new(bh)) { 19291da177e4SLinus Torvalds unmap_underlying_metadata(bh->b_bdev, 19301da177e4SLinus Torvalds bh->b_blocknr); 19311da177e4SLinus Torvalds if (PageUptodate(page)) { 1932637aff46SNick Piggin clear_buffer_new(bh); 19331da177e4SLinus Torvalds set_buffer_uptodate(bh); 1934637aff46SNick Piggin mark_buffer_dirty(bh); 19351da177e4SLinus Torvalds continue; 19361da177e4SLinus Torvalds } 1937eebd2aa3SChristoph Lameter if (block_end > to || block_start < from) 1938eebd2aa3SChristoph Lameter zero_user_segments(page, 1939eebd2aa3SChristoph Lameter to, block_end, 1940eebd2aa3SChristoph Lameter block_start, from); 19411da177e4SLinus Torvalds continue; 19421da177e4SLinus Torvalds } 19431da177e4SLinus Torvalds } 19441da177e4SLinus Torvalds if (PageUptodate(page)) { 19451da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 19461da177e4SLinus Torvalds set_buffer_uptodate(bh); 19471da177e4SLinus Torvalds continue; 19481da177e4SLinus Torvalds } 19491da177e4SLinus Torvalds if (!buffer_uptodate(bh) && !buffer_delay(bh) && 195033a266ddSDavid Chinner !buffer_unwritten(bh) && 19511da177e4SLinus Torvalds (block_start < from || block_end > to)) { 19521da177e4SLinus Torvalds ll_rw_block(READ, 1, &bh); 19531da177e4SLinus Torvalds *wait_bh++=bh; 19541da177e4SLinus Torvalds } 19551da177e4SLinus Torvalds } 19561da177e4SLinus Torvalds /* 19571da177e4SLinus Torvalds * If we issued read requests - let them complete. 19581da177e4SLinus Torvalds */ 19591da177e4SLinus Torvalds while(wait_bh > wait) { 19601da177e4SLinus Torvalds wait_on_buffer(*--wait_bh); 19611da177e4SLinus Torvalds if (!buffer_uptodate(*wait_bh)) 1962f3ddbdc6SNick Piggin err = -EIO; 19631da177e4SLinus Torvalds } 1964f9f07b6cSJan Kara if (unlikely(err)) 1965afddba49SNick Piggin page_zero_new_buffers(page, from, to); 19661da177e4SLinus Torvalds return err; 19671da177e4SLinus Torvalds } 1968ebdec241SChristoph Hellwig EXPORT_SYMBOL(__block_write_begin); 19691da177e4SLinus Torvalds 19701da177e4SLinus Torvalds static int __block_commit_write(struct inode *inode, struct page *page, 19711da177e4SLinus Torvalds unsigned from, unsigned to) 19721da177e4SLinus Torvalds { 19731da177e4SLinus Torvalds unsigned block_start, block_end; 19741da177e4SLinus Torvalds int partial = 0; 19751da177e4SLinus Torvalds unsigned blocksize; 19761da177e4SLinus Torvalds struct buffer_head *bh, *head; 19771da177e4SLinus Torvalds 197845bce8f3SLinus Torvalds bh = head = page_buffers(page); 197945bce8f3SLinus Torvalds blocksize = bh->b_size; 19801da177e4SLinus Torvalds 198145bce8f3SLinus Torvalds block_start = 0; 198245bce8f3SLinus Torvalds do { 19831da177e4SLinus Torvalds block_end = block_start + blocksize; 19841da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 19851da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 19861da177e4SLinus Torvalds partial = 1; 19871da177e4SLinus Torvalds } else { 19881da177e4SLinus Torvalds set_buffer_uptodate(bh); 19891da177e4SLinus Torvalds mark_buffer_dirty(bh); 19901da177e4SLinus Torvalds } 1991afddba49SNick Piggin clear_buffer_new(bh); 199245bce8f3SLinus Torvalds 199345bce8f3SLinus Torvalds block_start = block_end; 199445bce8f3SLinus Torvalds bh = bh->b_this_page; 199545bce8f3SLinus Torvalds } while (bh != head); 19961da177e4SLinus Torvalds 19971da177e4SLinus Torvalds /* 19981da177e4SLinus Torvalds * If this is a partial write which happened to make all buffers 19991da177e4SLinus Torvalds * uptodate then we can optimize away a bogus readpage() for 20001da177e4SLinus Torvalds * the next read(). Here we 'discover' whether the page went 20011da177e4SLinus Torvalds * uptodate as a result of this (potentially partial) write. 20021da177e4SLinus Torvalds */ 20031da177e4SLinus Torvalds if (!partial) 20041da177e4SLinus Torvalds SetPageUptodate(page); 20051da177e4SLinus Torvalds return 0; 20061da177e4SLinus Torvalds } 20071da177e4SLinus Torvalds 20081da177e4SLinus Torvalds /* 2009155130a4SChristoph Hellwig * block_write_begin takes care of the basic task of block allocation and 2010155130a4SChristoph Hellwig * bringing partial write blocks uptodate first. 2011155130a4SChristoph Hellwig * 20127bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 2013afddba49SNick Piggin */ 2014155130a4SChristoph Hellwig int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, 2015155130a4SChristoph Hellwig unsigned flags, struct page **pagep, get_block_t *get_block) 2016afddba49SNick Piggin { 20176e1db88dSChristoph Hellwig pgoff_t index = pos >> PAGE_CACHE_SHIFT; 2018afddba49SNick Piggin struct page *page; 20196e1db88dSChristoph Hellwig int status; 2020afddba49SNick Piggin 202154566b2cSNick Piggin page = grab_cache_page_write_begin(mapping, index, flags); 20226e1db88dSChristoph Hellwig if (!page) 20236e1db88dSChristoph Hellwig return -ENOMEM; 2024afddba49SNick Piggin 20256e1db88dSChristoph Hellwig status = __block_write_begin(page, pos, len, get_block); 2026afddba49SNick Piggin if (unlikely(status)) { 2027afddba49SNick Piggin unlock_page(page); 2028afddba49SNick Piggin page_cache_release(page); 20296e1db88dSChristoph Hellwig page = NULL; 2030afddba49SNick Piggin } 2031afddba49SNick Piggin 20326e1db88dSChristoph Hellwig *pagep = page; 2033afddba49SNick Piggin return status; 2034afddba49SNick Piggin } 2035afddba49SNick Piggin EXPORT_SYMBOL(block_write_begin); 2036afddba49SNick Piggin 2037afddba49SNick Piggin int block_write_end(struct file *file, struct address_space *mapping, 2038afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2039afddba49SNick Piggin struct page *page, void *fsdata) 2040afddba49SNick Piggin { 2041afddba49SNick Piggin struct inode *inode = mapping->host; 2042afddba49SNick Piggin unsigned start; 2043afddba49SNick Piggin 2044afddba49SNick Piggin start = pos & (PAGE_CACHE_SIZE - 1); 2045afddba49SNick Piggin 2046afddba49SNick Piggin if (unlikely(copied < len)) { 2047afddba49SNick Piggin /* 2048afddba49SNick Piggin * The buffers that were written will now be uptodate, so we 2049afddba49SNick Piggin * don't have to worry about a readpage reading them and 2050afddba49SNick Piggin * overwriting a partial write. However if we have encountered 2051afddba49SNick Piggin * a short write and only partially written into a buffer, it 2052afddba49SNick Piggin * will not be marked uptodate, so a readpage might come in and 2053afddba49SNick Piggin * destroy our partial write. 2054afddba49SNick Piggin * 2055afddba49SNick Piggin * Do the simplest thing, and just treat any short write to a 2056afddba49SNick Piggin * non uptodate page as a zero-length write, and force the 2057afddba49SNick Piggin * caller to redo the whole thing. 2058afddba49SNick Piggin */ 2059afddba49SNick Piggin if (!PageUptodate(page)) 2060afddba49SNick Piggin copied = 0; 2061afddba49SNick Piggin 2062afddba49SNick Piggin page_zero_new_buffers(page, start+copied, start+len); 2063afddba49SNick Piggin } 2064afddba49SNick Piggin flush_dcache_page(page); 2065afddba49SNick Piggin 2066afddba49SNick Piggin /* This could be a short (even 0-length) commit */ 2067afddba49SNick Piggin __block_commit_write(inode, page, start, start+copied); 2068afddba49SNick Piggin 2069afddba49SNick Piggin return copied; 2070afddba49SNick Piggin } 2071afddba49SNick Piggin EXPORT_SYMBOL(block_write_end); 2072afddba49SNick Piggin 2073afddba49SNick Piggin int generic_write_end(struct file *file, struct address_space *mapping, 2074afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2075afddba49SNick Piggin struct page *page, void *fsdata) 2076afddba49SNick Piggin { 2077afddba49SNick Piggin struct inode *inode = mapping->host; 2078c7d206b3SJan Kara int i_size_changed = 0; 2079afddba49SNick Piggin 2080afddba49SNick Piggin copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); 2081afddba49SNick Piggin 2082afddba49SNick Piggin /* 2083afddba49SNick Piggin * No need to use i_size_read() here, the i_size 2084afddba49SNick Piggin * cannot change under us because we hold i_mutex. 2085afddba49SNick Piggin * 2086afddba49SNick Piggin * But it's important to update i_size while still holding page lock: 2087afddba49SNick Piggin * page writeout could otherwise come in and zero beyond i_size. 2088afddba49SNick Piggin */ 2089afddba49SNick Piggin if (pos+copied > inode->i_size) { 2090afddba49SNick Piggin i_size_write(inode, pos+copied); 2091c7d206b3SJan Kara i_size_changed = 1; 2092afddba49SNick Piggin } 2093afddba49SNick Piggin 2094afddba49SNick Piggin unlock_page(page); 2095afddba49SNick Piggin page_cache_release(page); 2096afddba49SNick Piggin 2097c7d206b3SJan Kara /* 2098c7d206b3SJan Kara * Don't mark the inode dirty under page lock. First, it unnecessarily 2099c7d206b3SJan Kara * makes the holding time of page lock longer. Second, it forces lock 2100c7d206b3SJan Kara * ordering of page lock and transaction start for journaling 2101c7d206b3SJan Kara * filesystems. 2102c7d206b3SJan Kara */ 2103c7d206b3SJan Kara if (i_size_changed) 2104c7d206b3SJan Kara mark_inode_dirty(inode); 2105c7d206b3SJan Kara 2106afddba49SNick Piggin return copied; 2107afddba49SNick Piggin } 2108afddba49SNick Piggin EXPORT_SYMBOL(generic_write_end); 2109afddba49SNick Piggin 2110afddba49SNick Piggin /* 21118ab22b9aSHisashi Hifumi * block_is_partially_uptodate checks whether buffers within a page are 21128ab22b9aSHisashi Hifumi * uptodate or not. 21138ab22b9aSHisashi Hifumi * 21148ab22b9aSHisashi Hifumi * Returns true if all buffers which correspond to a file portion 21158ab22b9aSHisashi Hifumi * we want to read are uptodate. 21168ab22b9aSHisashi Hifumi */ 2117c186afb4SAl Viro int block_is_partially_uptodate(struct page *page, unsigned long from, 2118c186afb4SAl Viro unsigned long count) 21198ab22b9aSHisashi Hifumi { 21208ab22b9aSHisashi Hifumi unsigned block_start, block_end, blocksize; 21218ab22b9aSHisashi Hifumi unsigned to; 21228ab22b9aSHisashi Hifumi struct buffer_head *bh, *head; 21238ab22b9aSHisashi Hifumi int ret = 1; 21248ab22b9aSHisashi Hifumi 21258ab22b9aSHisashi Hifumi if (!page_has_buffers(page)) 21268ab22b9aSHisashi Hifumi return 0; 21278ab22b9aSHisashi Hifumi 212845bce8f3SLinus Torvalds head = page_buffers(page); 212945bce8f3SLinus Torvalds blocksize = head->b_size; 2130c186afb4SAl Viro to = min_t(unsigned, PAGE_CACHE_SIZE - from, count); 21318ab22b9aSHisashi Hifumi to = from + to; 21328ab22b9aSHisashi Hifumi if (from < blocksize && to > PAGE_CACHE_SIZE - blocksize) 21338ab22b9aSHisashi Hifumi return 0; 21348ab22b9aSHisashi Hifumi 21358ab22b9aSHisashi Hifumi bh = head; 21368ab22b9aSHisashi Hifumi block_start = 0; 21378ab22b9aSHisashi Hifumi do { 21388ab22b9aSHisashi Hifumi block_end = block_start + blocksize; 21398ab22b9aSHisashi Hifumi if (block_end > from && block_start < to) { 21408ab22b9aSHisashi Hifumi if (!buffer_uptodate(bh)) { 21418ab22b9aSHisashi Hifumi ret = 0; 21428ab22b9aSHisashi Hifumi break; 21438ab22b9aSHisashi Hifumi } 21448ab22b9aSHisashi Hifumi if (block_end >= to) 21458ab22b9aSHisashi Hifumi break; 21468ab22b9aSHisashi Hifumi } 21478ab22b9aSHisashi Hifumi block_start = block_end; 21488ab22b9aSHisashi Hifumi bh = bh->b_this_page; 21498ab22b9aSHisashi Hifumi } while (bh != head); 21508ab22b9aSHisashi Hifumi 21518ab22b9aSHisashi Hifumi return ret; 21528ab22b9aSHisashi Hifumi } 21538ab22b9aSHisashi Hifumi EXPORT_SYMBOL(block_is_partially_uptodate); 21548ab22b9aSHisashi Hifumi 21558ab22b9aSHisashi Hifumi /* 21561da177e4SLinus Torvalds * Generic "read page" function for block devices that have the normal 21571da177e4SLinus Torvalds * get_block functionality. This is most of the block device filesystems. 21581da177e4SLinus Torvalds * Reads the page asynchronously --- the unlock_buffer() and 21591da177e4SLinus Torvalds * set/clear_buffer_uptodate() functions propagate buffer state into the 21601da177e4SLinus Torvalds * page struct once IO has completed. 21611da177e4SLinus Torvalds */ 21621da177e4SLinus Torvalds int block_read_full_page(struct page *page, get_block_t *get_block) 21631da177e4SLinus Torvalds { 21641da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 21651da177e4SLinus Torvalds sector_t iblock, lblock; 21661da177e4SLinus Torvalds struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 216745bce8f3SLinus Torvalds unsigned int blocksize, bbits; 21681da177e4SLinus Torvalds int nr, i; 21691da177e4SLinus Torvalds int fully_mapped = 1; 21701da177e4SLinus Torvalds 217145bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 0); 217245bce8f3SLinus Torvalds blocksize = head->b_size; 217345bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 21741da177e4SLinus Torvalds 217545bce8f3SLinus Torvalds iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); 217645bce8f3SLinus Torvalds lblock = (i_size_read(inode)+blocksize-1) >> bbits; 21771da177e4SLinus Torvalds bh = head; 21781da177e4SLinus Torvalds nr = 0; 21791da177e4SLinus Torvalds i = 0; 21801da177e4SLinus Torvalds 21811da177e4SLinus Torvalds do { 21821da177e4SLinus Torvalds if (buffer_uptodate(bh)) 21831da177e4SLinus Torvalds continue; 21841da177e4SLinus Torvalds 21851da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2186c64610baSAndrew Morton int err = 0; 2187c64610baSAndrew Morton 21881da177e4SLinus Torvalds fully_mapped = 0; 21891da177e4SLinus Torvalds if (iblock < lblock) { 2190b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2191c64610baSAndrew Morton err = get_block(inode, iblock, bh, 0); 2192c64610baSAndrew Morton if (err) 21931da177e4SLinus Torvalds SetPageError(page); 21941da177e4SLinus Torvalds } 21951da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2196eebd2aa3SChristoph Lameter zero_user(page, i * blocksize, blocksize); 2197c64610baSAndrew Morton if (!err) 21981da177e4SLinus Torvalds set_buffer_uptodate(bh); 21991da177e4SLinus Torvalds continue; 22001da177e4SLinus Torvalds } 22011da177e4SLinus Torvalds /* 22021da177e4SLinus Torvalds * get_block() might have updated the buffer 22031da177e4SLinus Torvalds * synchronously 22041da177e4SLinus Torvalds */ 22051da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22061da177e4SLinus Torvalds continue; 22071da177e4SLinus Torvalds } 22081da177e4SLinus Torvalds arr[nr++] = bh; 22091da177e4SLinus Torvalds } while (i++, iblock++, (bh = bh->b_this_page) != head); 22101da177e4SLinus Torvalds 22111da177e4SLinus Torvalds if (fully_mapped) 22121da177e4SLinus Torvalds SetPageMappedToDisk(page); 22131da177e4SLinus Torvalds 22141da177e4SLinus Torvalds if (!nr) { 22151da177e4SLinus Torvalds /* 22161da177e4SLinus Torvalds * All buffers are uptodate - we can set the page uptodate 22171da177e4SLinus Torvalds * as well. But not if get_block() returned an error. 22181da177e4SLinus Torvalds */ 22191da177e4SLinus Torvalds if (!PageError(page)) 22201da177e4SLinus Torvalds SetPageUptodate(page); 22211da177e4SLinus Torvalds unlock_page(page); 22221da177e4SLinus Torvalds return 0; 22231da177e4SLinus Torvalds } 22241da177e4SLinus Torvalds 22251da177e4SLinus Torvalds /* Stage two: lock the buffers */ 22261da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 22271da177e4SLinus Torvalds bh = arr[i]; 22281da177e4SLinus Torvalds lock_buffer(bh); 22291da177e4SLinus Torvalds mark_buffer_async_read(bh); 22301da177e4SLinus Torvalds } 22311da177e4SLinus Torvalds 22321da177e4SLinus Torvalds /* 22331da177e4SLinus Torvalds * Stage 3: start the IO. Check for uptodateness 22341da177e4SLinus Torvalds * inside the buffer lock in case another process reading 22351da177e4SLinus Torvalds * the underlying blockdev brought it uptodate (the sct fix). 22361da177e4SLinus Torvalds */ 22371da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 22381da177e4SLinus Torvalds bh = arr[i]; 22391da177e4SLinus Torvalds if (buffer_uptodate(bh)) 22401da177e4SLinus Torvalds end_buffer_async_read(bh, 1); 22411da177e4SLinus Torvalds else 22421da177e4SLinus Torvalds submit_bh(READ, bh); 22431da177e4SLinus Torvalds } 22441da177e4SLinus Torvalds return 0; 22451da177e4SLinus Torvalds } 22461fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_read_full_page); 22471da177e4SLinus Torvalds 22481da177e4SLinus Torvalds /* utility function for filesystems that need to do work on expanding 224989e10787SNick Piggin * truncates. Uses filesystem pagecache writes to allow the filesystem to 22501da177e4SLinus Torvalds * deal with the hole. 22511da177e4SLinus Torvalds */ 225289e10787SNick Piggin int generic_cont_expand_simple(struct inode *inode, loff_t size) 22531da177e4SLinus Torvalds { 22541da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 22551da177e4SLinus Torvalds struct page *page; 225689e10787SNick Piggin void *fsdata; 22571da177e4SLinus Torvalds int err; 22581da177e4SLinus Torvalds 2259c08d3b0eSnpiggin@suse.de err = inode_newsize_ok(inode, size); 2260c08d3b0eSnpiggin@suse.de if (err) 22611da177e4SLinus Torvalds goto out; 22621da177e4SLinus Torvalds 226389e10787SNick Piggin err = pagecache_write_begin(NULL, mapping, size, 0, 226489e10787SNick Piggin AOP_FLAG_UNINTERRUPTIBLE|AOP_FLAG_CONT_EXPAND, 226589e10787SNick Piggin &page, &fsdata); 226689e10787SNick Piggin if (err) 226705eb0b51SOGAWA Hirofumi goto out; 226805eb0b51SOGAWA Hirofumi 226989e10787SNick Piggin err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata); 227089e10787SNick Piggin BUG_ON(err > 0); 227105eb0b51SOGAWA Hirofumi 227205eb0b51SOGAWA Hirofumi out: 227305eb0b51SOGAWA Hirofumi return err; 227405eb0b51SOGAWA Hirofumi } 22751fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_cont_expand_simple); 227605eb0b51SOGAWA Hirofumi 2277f1e3af72SAdrian Bunk static int cont_expand_zero(struct file *file, struct address_space *mapping, 227889e10787SNick Piggin loff_t pos, loff_t *bytes) 227905eb0b51SOGAWA Hirofumi { 228089e10787SNick Piggin struct inode *inode = mapping->host; 228189e10787SNick Piggin unsigned blocksize = 1 << inode->i_blkbits; 228289e10787SNick Piggin struct page *page; 228389e10787SNick Piggin void *fsdata; 228489e10787SNick Piggin pgoff_t index, curidx; 228589e10787SNick Piggin loff_t curpos; 228689e10787SNick Piggin unsigned zerofrom, offset, len; 228789e10787SNick Piggin int err = 0; 228805eb0b51SOGAWA Hirofumi 228989e10787SNick Piggin index = pos >> PAGE_CACHE_SHIFT; 229089e10787SNick Piggin offset = pos & ~PAGE_CACHE_MASK; 229189e10787SNick Piggin 229289e10787SNick Piggin while (index > (curidx = (curpos = *bytes)>>PAGE_CACHE_SHIFT)) { 229389e10787SNick Piggin zerofrom = curpos & ~PAGE_CACHE_MASK; 229489e10787SNick Piggin if (zerofrom & (blocksize-1)) { 229589e10787SNick Piggin *bytes |= (blocksize-1); 229689e10787SNick Piggin (*bytes)++; 229789e10787SNick Piggin } 229889e10787SNick Piggin len = PAGE_CACHE_SIZE - zerofrom; 229989e10787SNick Piggin 230089e10787SNick Piggin err = pagecache_write_begin(file, mapping, curpos, len, 230189e10787SNick Piggin AOP_FLAG_UNINTERRUPTIBLE, 230289e10787SNick Piggin &page, &fsdata); 230389e10787SNick Piggin if (err) 230489e10787SNick Piggin goto out; 2305eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 230689e10787SNick Piggin err = pagecache_write_end(file, mapping, curpos, len, len, 230789e10787SNick Piggin page, fsdata); 230889e10787SNick Piggin if (err < 0) 230989e10787SNick Piggin goto out; 231089e10787SNick Piggin BUG_ON(err != len); 231189e10787SNick Piggin err = 0; 2312061e9746SOGAWA Hirofumi 2313061e9746SOGAWA Hirofumi balance_dirty_pages_ratelimited(mapping); 231489e10787SNick Piggin } 231589e10787SNick Piggin 231689e10787SNick Piggin /* page covers the boundary, find the boundary offset */ 231789e10787SNick Piggin if (index == curidx) { 231889e10787SNick Piggin zerofrom = curpos & ~PAGE_CACHE_MASK; 231989e10787SNick Piggin /* if we will expand the thing last block will be filled */ 232089e10787SNick Piggin if (offset <= zerofrom) { 232189e10787SNick Piggin goto out; 232289e10787SNick Piggin } 232389e10787SNick Piggin if (zerofrom & (blocksize-1)) { 232489e10787SNick Piggin *bytes |= (blocksize-1); 232589e10787SNick Piggin (*bytes)++; 232689e10787SNick Piggin } 232789e10787SNick Piggin len = offset - zerofrom; 232889e10787SNick Piggin 232989e10787SNick Piggin err = pagecache_write_begin(file, mapping, curpos, len, 233089e10787SNick Piggin AOP_FLAG_UNINTERRUPTIBLE, 233189e10787SNick Piggin &page, &fsdata); 233289e10787SNick Piggin if (err) 233389e10787SNick Piggin goto out; 2334eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 233589e10787SNick Piggin err = pagecache_write_end(file, mapping, curpos, len, len, 233689e10787SNick Piggin page, fsdata); 233789e10787SNick Piggin if (err < 0) 233889e10787SNick Piggin goto out; 233989e10787SNick Piggin BUG_ON(err != len); 234089e10787SNick Piggin err = 0; 234189e10787SNick Piggin } 234289e10787SNick Piggin out: 234389e10787SNick Piggin return err; 23441da177e4SLinus Torvalds } 23451da177e4SLinus Torvalds 23461da177e4SLinus Torvalds /* 23471da177e4SLinus Torvalds * For moronic filesystems that do not allow holes in file. 23481da177e4SLinus Torvalds * We may have to extend the file. 23491da177e4SLinus Torvalds */ 2350282dc178SChristoph Hellwig int cont_write_begin(struct file *file, struct address_space *mapping, 235189e10787SNick Piggin loff_t pos, unsigned len, unsigned flags, 235289e10787SNick Piggin struct page **pagep, void **fsdata, 235389e10787SNick Piggin get_block_t *get_block, loff_t *bytes) 23541da177e4SLinus Torvalds { 23551da177e4SLinus Torvalds struct inode *inode = mapping->host; 23561da177e4SLinus Torvalds unsigned blocksize = 1 << inode->i_blkbits; 235789e10787SNick Piggin unsigned zerofrom; 235889e10787SNick Piggin int err; 23591da177e4SLinus Torvalds 236089e10787SNick Piggin err = cont_expand_zero(file, mapping, pos, bytes); 236189e10787SNick Piggin if (err) 2362155130a4SChristoph Hellwig return err; 23631da177e4SLinus Torvalds 23641da177e4SLinus Torvalds zerofrom = *bytes & ~PAGE_CACHE_MASK; 236589e10787SNick Piggin if (pos+len > *bytes && zerofrom & (blocksize-1)) { 23661da177e4SLinus Torvalds *bytes |= (blocksize-1); 23671da177e4SLinus Torvalds (*bytes)++; 23681da177e4SLinus Torvalds } 23691da177e4SLinus Torvalds 2370155130a4SChristoph Hellwig return block_write_begin(mapping, pos, len, flags, pagep, get_block); 23711da177e4SLinus Torvalds } 23721fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(cont_write_begin); 23731da177e4SLinus Torvalds 23741da177e4SLinus Torvalds int block_commit_write(struct page *page, unsigned from, unsigned to) 23751da177e4SLinus Torvalds { 23761da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 23771da177e4SLinus Torvalds __block_commit_write(inode,page,from,to); 23781da177e4SLinus Torvalds return 0; 23791da177e4SLinus Torvalds } 23801fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_commit_write); 23811da177e4SLinus Torvalds 238254171690SDavid Chinner /* 238354171690SDavid Chinner * block_page_mkwrite() is not allowed to change the file size as it gets 238454171690SDavid Chinner * called from a page fault handler when a page is first dirtied. Hence we must 238554171690SDavid Chinner * be careful to check for EOF conditions here. We set the page up correctly 238654171690SDavid Chinner * for a written page which means we get ENOSPC checking when writing into 238754171690SDavid Chinner * holes and correct delalloc and unwritten extent mapping on filesystems that 238854171690SDavid Chinner * support these features. 238954171690SDavid Chinner * 239054171690SDavid Chinner * We are not allowed to take the i_mutex here so we have to play games to 239154171690SDavid Chinner * protect against truncate races as the page could now be beyond EOF. Because 23927bb46a67Snpiggin@suse.de * truncate writes the inode size before removing pages, once we have the 239354171690SDavid Chinner * page lock we can determine safely if the page is beyond EOF. If it is not 239454171690SDavid Chinner * beyond EOF, then the page is guaranteed safe against truncation until we 239554171690SDavid Chinner * unlock the page. 2396ea13a864SJan Kara * 239714da9200SJan Kara * Direct callers of this function should protect against filesystem freezing 239814da9200SJan Kara * using sb_start_write() - sb_end_write() functions. 239954171690SDavid Chinner */ 240024da4fabSJan Kara int __block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 240154171690SDavid Chinner get_block_t get_block) 240254171690SDavid Chinner { 2403c2ec175cSNick Piggin struct page *page = vmf->page; 2404496ad9aaSAl Viro struct inode *inode = file_inode(vma->vm_file); 240554171690SDavid Chinner unsigned long end; 240654171690SDavid Chinner loff_t size; 240724da4fabSJan Kara int ret; 240854171690SDavid Chinner 240954171690SDavid Chinner lock_page(page); 241054171690SDavid Chinner size = i_size_read(inode); 241154171690SDavid Chinner if ((page->mapping != inode->i_mapping) || 241218336338SNick Piggin (page_offset(page) > size)) { 241324da4fabSJan Kara /* We overload EFAULT to mean page got truncated */ 241424da4fabSJan Kara ret = -EFAULT; 241524da4fabSJan Kara goto out_unlock; 241654171690SDavid Chinner } 241754171690SDavid Chinner 241854171690SDavid Chinner /* page is wholly or partially inside EOF */ 241954171690SDavid Chinner if (((page->index + 1) << PAGE_CACHE_SHIFT) > size) 242054171690SDavid Chinner end = size & ~PAGE_CACHE_MASK; 242154171690SDavid Chinner else 242254171690SDavid Chinner end = PAGE_CACHE_SIZE; 242354171690SDavid Chinner 2424ebdec241SChristoph Hellwig ret = __block_write_begin(page, 0, end, get_block); 242554171690SDavid Chinner if (!ret) 242654171690SDavid Chinner ret = block_commit_write(page, 0, end); 242754171690SDavid Chinner 242824da4fabSJan Kara if (unlikely(ret < 0)) 242924da4fabSJan Kara goto out_unlock; 2430ea13a864SJan Kara set_page_dirty(page); 24311d1d1a76SDarrick J. Wong wait_for_stable_page(page); 243224da4fabSJan Kara return 0; 243324da4fabSJan Kara out_unlock: 2434b827e496SNick Piggin unlock_page(page); 243554171690SDavid Chinner return ret; 243654171690SDavid Chinner } 243724da4fabSJan Kara EXPORT_SYMBOL(__block_page_mkwrite); 243824da4fabSJan Kara 243924da4fabSJan Kara int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 244024da4fabSJan Kara get_block_t get_block) 244124da4fabSJan Kara { 2442ea13a864SJan Kara int ret; 2443496ad9aaSAl Viro struct super_block *sb = file_inode(vma->vm_file)->i_sb; 244424da4fabSJan Kara 244514da9200SJan Kara sb_start_pagefault(sb); 2446041bbb6dSTheodore Ts'o 2447041bbb6dSTheodore Ts'o /* 2448041bbb6dSTheodore Ts'o * Update file times before taking page lock. We may end up failing the 2449041bbb6dSTheodore Ts'o * fault so this update may be superfluous but who really cares... 2450041bbb6dSTheodore Ts'o */ 2451041bbb6dSTheodore Ts'o file_update_time(vma->vm_file); 2452041bbb6dSTheodore Ts'o 2453ea13a864SJan Kara ret = __block_page_mkwrite(vma, vmf, get_block); 245414da9200SJan Kara sb_end_pagefault(sb); 245524da4fabSJan Kara return block_page_mkwrite_return(ret); 245624da4fabSJan Kara } 24571fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_page_mkwrite); 24581da177e4SLinus Torvalds 24591da177e4SLinus Torvalds /* 246003158cd7SNick Piggin * nobh_write_begin()'s prereads are special: the buffer_heads are freed 24611da177e4SLinus Torvalds * immediately, while under the page lock. So it needs a special end_io 24621da177e4SLinus Torvalds * handler which does not touch the bh after unlocking it. 24631da177e4SLinus Torvalds */ 24641da177e4SLinus Torvalds static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate) 24651da177e4SLinus Torvalds { 246668671f35SDmitry Monakhov __end_buffer_read_notouch(bh, uptodate); 24671da177e4SLinus Torvalds } 24681da177e4SLinus Torvalds 24691da177e4SLinus Torvalds /* 247003158cd7SNick Piggin * Attach the singly-linked list of buffers created by nobh_write_begin, to 247103158cd7SNick Piggin * the page (converting it to circular linked list and taking care of page 247203158cd7SNick Piggin * dirty races). 247303158cd7SNick Piggin */ 247403158cd7SNick Piggin static void attach_nobh_buffers(struct page *page, struct buffer_head *head) 247503158cd7SNick Piggin { 247603158cd7SNick Piggin struct buffer_head *bh; 247703158cd7SNick Piggin 247803158cd7SNick Piggin BUG_ON(!PageLocked(page)); 247903158cd7SNick Piggin 248003158cd7SNick Piggin spin_lock(&page->mapping->private_lock); 248103158cd7SNick Piggin bh = head; 248203158cd7SNick Piggin do { 248303158cd7SNick Piggin if (PageDirty(page)) 248403158cd7SNick Piggin set_buffer_dirty(bh); 248503158cd7SNick Piggin if (!bh->b_this_page) 248603158cd7SNick Piggin bh->b_this_page = head; 248703158cd7SNick Piggin bh = bh->b_this_page; 248803158cd7SNick Piggin } while (bh != head); 248903158cd7SNick Piggin attach_page_buffers(page, head); 249003158cd7SNick Piggin spin_unlock(&page->mapping->private_lock); 249103158cd7SNick Piggin } 249203158cd7SNick Piggin 249303158cd7SNick Piggin /* 2494ea0f04e5SChristoph Hellwig * On entry, the page is fully not uptodate. 2495ea0f04e5SChristoph Hellwig * On exit the page is fully uptodate in the areas outside (from,to) 24967bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 24971da177e4SLinus Torvalds */ 2498ea0f04e5SChristoph Hellwig int nobh_write_begin(struct address_space *mapping, 249903158cd7SNick Piggin loff_t pos, unsigned len, unsigned flags, 250003158cd7SNick Piggin struct page **pagep, void **fsdata, 25011da177e4SLinus Torvalds get_block_t *get_block) 25021da177e4SLinus Torvalds { 250303158cd7SNick Piggin struct inode *inode = mapping->host; 25041da177e4SLinus Torvalds const unsigned blkbits = inode->i_blkbits; 25051da177e4SLinus Torvalds const unsigned blocksize = 1 << blkbits; 2506a4b0672dSNick Piggin struct buffer_head *head, *bh; 250703158cd7SNick Piggin struct page *page; 250803158cd7SNick Piggin pgoff_t index; 250903158cd7SNick Piggin unsigned from, to; 25101da177e4SLinus Torvalds unsigned block_in_page; 2511a4b0672dSNick Piggin unsigned block_start, block_end; 25121da177e4SLinus Torvalds sector_t block_in_file; 25131da177e4SLinus Torvalds int nr_reads = 0; 25141da177e4SLinus Torvalds int ret = 0; 25151da177e4SLinus Torvalds int is_mapped_to_disk = 1; 25161da177e4SLinus Torvalds 251703158cd7SNick Piggin index = pos >> PAGE_CACHE_SHIFT; 251803158cd7SNick Piggin from = pos & (PAGE_CACHE_SIZE - 1); 251903158cd7SNick Piggin to = from + len; 252003158cd7SNick Piggin 252154566b2cSNick Piggin page = grab_cache_page_write_begin(mapping, index, flags); 252203158cd7SNick Piggin if (!page) 252303158cd7SNick Piggin return -ENOMEM; 252403158cd7SNick Piggin *pagep = page; 252503158cd7SNick Piggin *fsdata = NULL; 252603158cd7SNick Piggin 252703158cd7SNick Piggin if (page_has_buffers(page)) { 2528309f77adSNamhyung Kim ret = __block_write_begin(page, pos, len, get_block); 2529309f77adSNamhyung Kim if (unlikely(ret)) 2530309f77adSNamhyung Kim goto out_release; 2531309f77adSNamhyung Kim return ret; 253203158cd7SNick Piggin } 2533a4b0672dSNick Piggin 25341da177e4SLinus Torvalds if (PageMappedToDisk(page)) 25351da177e4SLinus Torvalds return 0; 25361da177e4SLinus Torvalds 2537a4b0672dSNick Piggin /* 2538a4b0672dSNick Piggin * Allocate buffers so that we can keep track of state, and potentially 2539a4b0672dSNick Piggin * attach them to the page if an error occurs. In the common case of 2540a4b0672dSNick Piggin * no error, they will just be freed again without ever being attached 2541a4b0672dSNick Piggin * to the page (which is all OK, because we're under the page lock). 2542a4b0672dSNick Piggin * 2543a4b0672dSNick Piggin * Be careful: the buffer linked list is a NULL terminated one, rather 2544a4b0672dSNick Piggin * than the circular one we're used to. 2545a4b0672dSNick Piggin */ 2546a4b0672dSNick Piggin head = alloc_page_buffers(page, blocksize, 0); 254703158cd7SNick Piggin if (!head) { 254803158cd7SNick Piggin ret = -ENOMEM; 254903158cd7SNick Piggin goto out_release; 255003158cd7SNick Piggin } 2551a4b0672dSNick Piggin 25521da177e4SLinus Torvalds block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits); 25531da177e4SLinus Torvalds 25541da177e4SLinus Torvalds /* 25551da177e4SLinus Torvalds * We loop across all blocks in the page, whether or not they are 25561da177e4SLinus Torvalds * part of the affected region. This is so we can discover if the 25571da177e4SLinus Torvalds * page is fully mapped-to-disk. 25581da177e4SLinus Torvalds */ 2559a4b0672dSNick Piggin for (block_start = 0, block_in_page = 0, bh = head; 25601da177e4SLinus Torvalds block_start < PAGE_CACHE_SIZE; 2561a4b0672dSNick Piggin block_in_page++, block_start += blocksize, bh = bh->b_this_page) { 25621da177e4SLinus Torvalds int create; 25631da177e4SLinus Torvalds 2564a4b0672dSNick Piggin block_end = block_start + blocksize; 2565a4b0672dSNick Piggin bh->b_state = 0; 25661da177e4SLinus Torvalds create = 1; 25671da177e4SLinus Torvalds if (block_start >= to) 25681da177e4SLinus Torvalds create = 0; 25691da177e4SLinus Torvalds ret = get_block(inode, block_in_file + block_in_page, 2570a4b0672dSNick Piggin bh, create); 25711da177e4SLinus Torvalds if (ret) 25721da177e4SLinus Torvalds goto failed; 2573a4b0672dSNick Piggin if (!buffer_mapped(bh)) 25741da177e4SLinus Torvalds is_mapped_to_disk = 0; 2575a4b0672dSNick Piggin if (buffer_new(bh)) 2576a4b0672dSNick Piggin unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); 2577a4b0672dSNick Piggin if (PageUptodate(page)) { 2578a4b0672dSNick Piggin set_buffer_uptodate(bh); 25791da177e4SLinus Torvalds continue; 2580a4b0672dSNick Piggin } 2581a4b0672dSNick Piggin if (buffer_new(bh) || !buffer_mapped(bh)) { 2582eebd2aa3SChristoph Lameter zero_user_segments(page, block_start, from, 2583eebd2aa3SChristoph Lameter to, block_end); 25841da177e4SLinus Torvalds continue; 25851da177e4SLinus Torvalds } 2586a4b0672dSNick Piggin if (buffer_uptodate(bh)) 25871da177e4SLinus Torvalds continue; /* reiserfs does this */ 25881da177e4SLinus Torvalds if (block_start < from || block_end > to) { 2589a4b0672dSNick Piggin lock_buffer(bh); 2590a4b0672dSNick Piggin bh->b_end_io = end_buffer_read_nobh; 2591a4b0672dSNick Piggin submit_bh(READ, bh); 2592a4b0672dSNick Piggin nr_reads++; 25931da177e4SLinus Torvalds } 25941da177e4SLinus Torvalds } 25951da177e4SLinus Torvalds 25961da177e4SLinus Torvalds if (nr_reads) { 25971da177e4SLinus Torvalds /* 25981da177e4SLinus Torvalds * The page is locked, so these buffers are protected from 25991da177e4SLinus Torvalds * any VM or truncate activity. Hence we don't need to care 26001da177e4SLinus Torvalds * for the buffer_head refcounts. 26011da177e4SLinus Torvalds */ 2602a4b0672dSNick Piggin for (bh = head; bh; bh = bh->b_this_page) { 26031da177e4SLinus Torvalds wait_on_buffer(bh); 26041da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 26051da177e4SLinus Torvalds ret = -EIO; 26061da177e4SLinus Torvalds } 26071da177e4SLinus Torvalds if (ret) 26081da177e4SLinus Torvalds goto failed; 26091da177e4SLinus Torvalds } 26101da177e4SLinus Torvalds 26111da177e4SLinus Torvalds if (is_mapped_to_disk) 26121da177e4SLinus Torvalds SetPageMappedToDisk(page); 26131da177e4SLinus Torvalds 261403158cd7SNick Piggin *fsdata = head; /* to be released by nobh_write_end */ 2615a4b0672dSNick Piggin 26161da177e4SLinus Torvalds return 0; 26171da177e4SLinus Torvalds 26181da177e4SLinus Torvalds failed: 261903158cd7SNick Piggin BUG_ON(!ret); 26201da177e4SLinus Torvalds /* 2621a4b0672dSNick Piggin * Error recovery is a bit difficult. We need to zero out blocks that 2622a4b0672dSNick Piggin * were newly allocated, and dirty them to ensure they get written out. 2623a4b0672dSNick Piggin * Buffers need to be attached to the page at this point, otherwise 2624a4b0672dSNick Piggin * the handling of potential IO errors during writeout would be hard 2625a4b0672dSNick Piggin * (could try doing synchronous writeout, but what if that fails too?) 26261da177e4SLinus Torvalds */ 262703158cd7SNick Piggin attach_nobh_buffers(page, head); 262803158cd7SNick Piggin page_zero_new_buffers(page, from, to); 2629a4b0672dSNick Piggin 263003158cd7SNick Piggin out_release: 263103158cd7SNick Piggin unlock_page(page); 263203158cd7SNick Piggin page_cache_release(page); 263303158cd7SNick Piggin *pagep = NULL; 2634a4b0672dSNick Piggin 26357bb46a67Snpiggin@suse.de return ret; 26367bb46a67Snpiggin@suse.de } 263703158cd7SNick Piggin EXPORT_SYMBOL(nobh_write_begin); 26381da177e4SLinus Torvalds 263903158cd7SNick Piggin int nobh_write_end(struct file *file, struct address_space *mapping, 264003158cd7SNick Piggin loff_t pos, unsigned len, unsigned copied, 264103158cd7SNick Piggin struct page *page, void *fsdata) 26421da177e4SLinus Torvalds { 26431da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 2644efdc3131SNick Piggin struct buffer_head *head = fsdata; 264503158cd7SNick Piggin struct buffer_head *bh; 26465b41e74aSDmitri Monakhov BUG_ON(fsdata != NULL && page_has_buffers(page)); 26471da177e4SLinus Torvalds 2648d4cf109fSDave Kleikamp if (unlikely(copied < len) && head) 264903158cd7SNick Piggin attach_nobh_buffers(page, head); 2650a4b0672dSNick Piggin if (page_has_buffers(page)) 265103158cd7SNick Piggin return generic_write_end(file, mapping, pos, len, 265203158cd7SNick Piggin copied, page, fsdata); 2653a4b0672dSNick Piggin 265422c8ca78SNick Piggin SetPageUptodate(page); 26551da177e4SLinus Torvalds set_page_dirty(page); 265603158cd7SNick Piggin if (pos+copied > inode->i_size) { 265703158cd7SNick Piggin i_size_write(inode, pos+copied); 26581da177e4SLinus Torvalds mark_inode_dirty(inode); 26591da177e4SLinus Torvalds } 266003158cd7SNick Piggin 266103158cd7SNick Piggin unlock_page(page); 266203158cd7SNick Piggin page_cache_release(page); 266303158cd7SNick Piggin 266403158cd7SNick Piggin while (head) { 266503158cd7SNick Piggin bh = head; 266603158cd7SNick Piggin head = head->b_this_page; 266703158cd7SNick Piggin free_buffer_head(bh); 26681da177e4SLinus Torvalds } 266903158cd7SNick Piggin 267003158cd7SNick Piggin return copied; 267103158cd7SNick Piggin } 267203158cd7SNick Piggin EXPORT_SYMBOL(nobh_write_end); 26731da177e4SLinus Torvalds 26741da177e4SLinus Torvalds /* 26751da177e4SLinus Torvalds * nobh_writepage() - based on block_full_write_page() except 26761da177e4SLinus Torvalds * that it tries to operate without attaching bufferheads to 26771da177e4SLinus Torvalds * the page. 26781da177e4SLinus Torvalds */ 26791da177e4SLinus Torvalds int nobh_writepage(struct page *page, get_block_t *get_block, 26801da177e4SLinus Torvalds struct writeback_control *wbc) 26811da177e4SLinus Torvalds { 26821da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 26831da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 26841da177e4SLinus Torvalds const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; 26851da177e4SLinus Torvalds unsigned offset; 26861da177e4SLinus Torvalds int ret; 26871da177e4SLinus Torvalds 26881da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 26891da177e4SLinus Torvalds if (page->index < end_index) 26901da177e4SLinus Torvalds goto out; 26911da177e4SLinus Torvalds 26921da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 26931da177e4SLinus Torvalds offset = i_size & (PAGE_CACHE_SIZE-1); 26941da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 26951da177e4SLinus Torvalds /* 26961da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 26971da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 26981da177e4SLinus Torvalds * freeable here, so the page does not leak. 26991da177e4SLinus Torvalds */ 27001da177e4SLinus Torvalds #if 0 27011da177e4SLinus Torvalds /* Not really sure about this - do we need this ? */ 27021da177e4SLinus Torvalds if (page->mapping->a_ops->invalidatepage) 27031da177e4SLinus Torvalds page->mapping->a_ops->invalidatepage(page, offset); 27041da177e4SLinus Torvalds #endif 27051da177e4SLinus Torvalds unlock_page(page); 27061da177e4SLinus Torvalds return 0; /* don't care */ 27071da177e4SLinus Torvalds } 27081da177e4SLinus Torvalds 27091da177e4SLinus Torvalds /* 27101da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 27111da177e4SLinus Torvalds * writepage invocation because it may be mmapped. "A file is mapped 27121da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 27131da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 27141da177e4SLinus Torvalds * writes to that region are not written out to the file." 27151da177e4SLinus Torvalds */ 2716eebd2aa3SChristoph Lameter zero_user_segment(page, offset, PAGE_CACHE_SIZE); 27171da177e4SLinus Torvalds out: 27181da177e4SLinus Torvalds ret = mpage_writepage(page, get_block, wbc); 27191da177e4SLinus Torvalds if (ret == -EAGAIN) 272035c80d5fSChris Mason ret = __block_write_full_page(inode, page, get_block, wbc, 272135c80d5fSChris Mason end_buffer_async_write); 27221da177e4SLinus Torvalds return ret; 27231da177e4SLinus Torvalds } 27241da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_writepage); 27251da177e4SLinus Torvalds 272603158cd7SNick Piggin int nobh_truncate_page(struct address_space *mapping, 272703158cd7SNick Piggin loff_t from, get_block_t *get_block) 27281da177e4SLinus Torvalds { 27291da177e4SLinus Torvalds pgoff_t index = from >> PAGE_CACHE_SHIFT; 27301da177e4SLinus Torvalds unsigned offset = from & (PAGE_CACHE_SIZE-1); 273103158cd7SNick Piggin unsigned blocksize; 273203158cd7SNick Piggin sector_t iblock; 273303158cd7SNick Piggin unsigned length, pos; 273403158cd7SNick Piggin struct inode *inode = mapping->host; 27351da177e4SLinus Torvalds struct page *page; 273603158cd7SNick Piggin struct buffer_head map_bh; 273703158cd7SNick Piggin int err; 27381da177e4SLinus Torvalds 273903158cd7SNick Piggin blocksize = 1 << inode->i_blkbits; 274003158cd7SNick Piggin length = offset & (blocksize - 1); 27411da177e4SLinus Torvalds 274203158cd7SNick Piggin /* Block boundary? Nothing to do */ 274303158cd7SNick Piggin if (!length) 274403158cd7SNick Piggin return 0; 274503158cd7SNick Piggin 274603158cd7SNick Piggin length = blocksize - length; 274703158cd7SNick Piggin iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 274803158cd7SNick Piggin 27491da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 275003158cd7SNick Piggin err = -ENOMEM; 27511da177e4SLinus Torvalds if (!page) 27521da177e4SLinus Torvalds goto out; 27531da177e4SLinus Torvalds 275403158cd7SNick Piggin if (page_has_buffers(page)) { 275503158cd7SNick Piggin has_buffers: 275603158cd7SNick Piggin unlock_page(page); 275703158cd7SNick Piggin page_cache_release(page); 275803158cd7SNick Piggin return block_truncate_page(mapping, from, get_block); 27591da177e4SLinus Torvalds } 276003158cd7SNick Piggin 276103158cd7SNick Piggin /* Find the buffer that contains "offset" */ 276203158cd7SNick Piggin pos = blocksize; 276303158cd7SNick Piggin while (offset >= pos) { 276403158cd7SNick Piggin iblock++; 276503158cd7SNick Piggin pos += blocksize; 276603158cd7SNick Piggin } 276703158cd7SNick Piggin 2768460bcf57STheodore Ts'o map_bh.b_size = blocksize; 2769460bcf57STheodore Ts'o map_bh.b_state = 0; 277003158cd7SNick Piggin err = get_block(inode, iblock, &map_bh, 0); 277103158cd7SNick Piggin if (err) 277203158cd7SNick Piggin goto unlock; 277303158cd7SNick Piggin /* unmapped? It's a hole - nothing to do */ 277403158cd7SNick Piggin if (!buffer_mapped(&map_bh)) 277503158cd7SNick Piggin goto unlock; 277603158cd7SNick Piggin 277703158cd7SNick Piggin /* Ok, it's mapped. Make sure it's up-to-date */ 277803158cd7SNick Piggin if (!PageUptodate(page)) { 277903158cd7SNick Piggin err = mapping->a_ops->readpage(NULL, page); 278003158cd7SNick Piggin if (err) { 278103158cd7SNick Piggin page_cache_release(page); 278203158cd7SNick Piggin goto out; 278303158cd7SNick Piggin } 278403158cd7SNick Piggin lock_page(page); 278503158cd7SNick Piggin if (!PageUptodate(page)) { 278603158cd7SNick Piggin err = -EIO; 278703158cd7SNick Piggin goto unlock; 278803158cd7SNick Piggin } 278903158cd7SNick Piggin if (page_has_buffers(page)) 279003158cd7SNick Piggin goto has_buffers; 279103158cd7SNick Piggin } 2792eebd2aa3SChristoph Lameter zero_user(page, offset, length); 279303158cd7SNick Piggin set_page_dirty(page); 279403158cd7SNick Piggin err = 0; 279503158cd7SNick Piggin 279603158cd7SNick Piggin unlock: 27971da177e4SLinus Torvalds unlock_page(page); 27981da177e4SLinus Torvalds page_cache_release(page); 27991da177e4SLinus Torvalds out: 280003158cd7SNick Piggin return err; 28011da177e4SLinus Torvalds } 28021da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_truncate_page); 28031da177e4SLinus Torvalds 28041da177e4SLinus Torvalds int block_truncate_page(struct address_space *mapping, 28051da177e4SLinus Torvalds loff_t from, get_block_t *get_block) 28061da177e4SLinus Torvalds { 28071da177e4SLinus Torvalds pgoff_t index = from >> PAGE_CACHE_SHIFT; 28081da177e4SLinus Torvalds unsigned offset = from & (PAGE_CACHE_SIZE-1); 28091da177e4SLinus Torvalds unsigned blocksize; 281054b21a79SAndrew Morton sector_t iblock; 28111da177e4SLinus Torvalds unsigned length, pos; 28121da177e4SLinus Torvalds struct inode *inode = mapping->host; 28131da177e4SLinus Torvalds struct page *page; 28141da177e4SLinus Torvalds struct buffer_head *bh; 28151da177e4SLinus Torvalds int err; 28161da177e4SLinus Torvalds 28171da177e4SLinus Torvalds blocksize = 1 << inode->i_blkbits; 28181da177e4SLinus Torvalds length = offset & (blocksize - 1); 28191da177e4SLinus Torvalds 28201da177e4SLinus Torvalds /* Block boundary? Nothing to do */ 28211da177e4SLinus Torvalds if (!length) 28221da177e4SLinus Torvalds return 0; 28231da177e4SLinus Torvalds 28241da177e4SLinus Torvalds length = blocksize - length; 282554b21a79SAndrew Morton iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 28261da177e4SLinus Torvalds 28271da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 28281da177e4SLinus Torvalds err = -ENOMEM; 28291da177e4SLinus Torvalds if (!page) 28301da177e4SLinus Torvalds goto out; 28311da177e4SLinus Torvalds 28321da177e4SLinus Torvalds if (!page_has_buffers(page)) 28331da177e4SLinus Torvalds create_empty_buffers(page, blocksize, 0); 28341da177e4SLinus Torvalds 28351da177e4SLinus Torvalds /* Find the buffer that contains "offset" */ 28361da177e4SLinus Torvalds bh = page_buffers(page); 28371da177e4SLinus Torvalds pos = blocksize; 28381da177e4SLinus Torvalds while (offset >= pos) { 28391da177e4SLinus Torvalds bh = bh->b_this_page; 28401da177e4SLinus Torvalds iblock++; 28411da177e4SLinus Torvalds pos += blocksize; 28421da177e4SLinus Torvalds } 28431da177e4SLinus Torvalds 28441da177e4SLinus Torvalds err = 0; 28451da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2846b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 28471da177e4SLinus Torvalds err = get_block(inode, iblock, bh, 0); 28481da177e4SLinus Torvalds if (err) 28491da177e4SLinus Torvalds goto unlock; 28501da177e4SLinus Torvalds /* unmapped? It's a hole - nothing to do */ 28511da177e4SLinus Torvalds if (!buffer_mapped(bh)) 28521da177e4SLinus Torvalds goto unlock; 28531da177e4SLinus Torvalds } 28541da177e4SLinus Torvalds 28551da177e4SLinus Torvalds /* Ok, it's mapped. Make sure it's up-to-date */ 28561da177e4SLinus Torvalds if (PageUptodate(page)) 28571da177e4SLinus Torvalds set_buffer_uptodate(bh); 28581da177e4SLinus Torvalds 285933a266ddSDavid Chinner if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { 28601da177e4SLinus Torvalds err = -EIO; 28611da177e4SLinus Torvalds ll_rw_block(READ, 1, &bh); 28621da177e4SLinus Torvalds wait_on_buffer(bh); 28631da177e4SLinus Torvalds /* Uhhuh. Read error. Complain and punt. */ 28641da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 28651da177e4SLinus Torvalds goto unlock; 28661da177e4SLinus Torvalds } 28671da177e4SLinus Torvalds 2868eebd2aa3SChristoph Lameter zero_user(page, offset, length); 28691da177e4SLinus Torvalds mark_buffer_dirty(bh); 28701da177e4SLinus Torvalds err = 0; 28711da177e4SLinus Torvalds 28721da177e4SLinus Torvalds unlock: 28731da177e4SLinus Torvalds unlock_page(page); 28741da177e4SLinus Torvalds page_cache_release(page); 28751da177e4SLinus Torvalds out: 28761da177e4SLinus Torvalds return err; 28771da177e4SLinus Torvalds } 28781fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_truncate_page); 28791da177e4SLinus Torvalds 28801da177e4SLinus Torvalds /* 28811da177e4SLinus Torvalds * The generic ->writepage function for buffer-backed address_spaces 28821da177e4SLinus Torvalds */ 2883*1b938c08SMatthew Wilcox int block_write_full_page(struct page *page, get_block_t *get_block, 2884*1b938c08SMatthew Wilcox struct writeback_control *wbc) 28851da177e4SLinus Torvalds { 28861da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 28871da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 28881da177e4SLinus Torvalds const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; 28891da177e4SLinus Torvalds unsigned offset; 28901da177e4SLinus Torvalds 28911da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 28921da177e4SLinus Torvalds if (page->index < end_index) 289335c80d5fSChris Mason return __block_write_full_page(inode, page, get_block, wbc, 2894*1b938c08SMatthew Wilcox end_buffer_async_write); 28951da177e4SLinus Torvalds 28961da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 28971da177e4SLinus Torvalds offset = i_size & (PAGE_CACHE_SIZE-1); 28981da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 28991da177e4SLinus Torvalds /* 29001da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 29011da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 29021da177e4SLinus Torvalds * freeable here, so the page does not leak. 29031da177e4SLinus Torvalds */ 2904d47992f8SLukas Czerner do_invalidatepage(page, 0, PAGE_CACHE_SIZE); 29051da177e4SLinus Torvalds unlock_page(page); 29061da177e4SLinus Torvalds return 0; /* don't care */ 29071da177e4SLinus Torvalds } 29081da177e4SLinus Torvalds 29091da177e4SLinus Torvalds /* 29101da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 29112a61aa40SAdam Buchbinder * writepage invocation because it may be mmapped. "A file is mapped 29121da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 29131da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 29141da177e4SLinus Torvalds * writes to that region are not written out to the file." 29151da177e4SLinus Torvalds */ 2916eebd2aa3SChristoph Lameter zero_user_segment(page, offset, PAGE_CACHE_SIZE); 2917*1b938c08SMatthew Wilcox return __block_write_full_page(inode, page, get_block, wbc, 291835c80d5fSChris Mason end_buffer_async_write); 291935c80d5fSChris Mason } 29201fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_write_full_page); 292135c80d5fSChris Mason 29221da177e4SLinus Torvalds sector_t generic_block_bmap(struct address_space *mapping, sector_t block, 29231da177e4SLinus Torvalds get_block_t *get_block) 29241da177e4SLinus Torvalds { 29251da177e4SLinus Torvalds struct buffer_head tmp; 29261da177e4SLinus Torvalds struct inode *inode = mapping->host; 29271da177e4SLinus Torvalds tmp.b_state = 0; 29281da177e4SLinus Torvalds tmp.b_blocknr = 0; 2929b0cf2321SBadari Pulavarty tmp.b_size = 1 << inode->i_blkbits; 29301da177e4SLinus Torvalds get_block(inode, block, &tmp, 0); 29311da177e4SLinus Torvalds return tmp.b_blocknr; 29321da177e4SLinus Torvalds } 29331fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_block_bmap); 29341da177e4SLinus Torvalds 29356712ecf8SNeilBrown static void end_bio_bh_io_sync(struct bio *bio, int err) 29361da177e4SLinus Torvalds { 29371da177e4SLinus Torvalds struct buffer_head *bh = bio->bi_private; 29381da177e4SLinus Torvalds 29391da177e4SLinus Torvalds if (err == -EOPNOTSUPP) { 29401da177e4SLinus Torvalds set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); 29411da177e4SLinus Torvalds } 29421da177e4SLinus Torvalds 294308bafc03SKeith Mannthey if (unlikely (test_bit(BIO_QUIET,&bio->bi_flags))) 294408bafc03SKeith Mannthey set_bit(BH_Quiet, &bh->b_state); 294508bafc03SKeith Mannthey 29461da177e4SLinus Torvalds bh->b_end_io(bh, test_bit(BIO_UPTODATE, &bio->bi_flags)); 29471da177e4SLinus Torvalds bio_put(bio); 29481da177e4SLinus Torvalds } 29491da177e4SLinus Torvalds 295057302e0dSLinus Torvalds /* 295157302e0dSLinus Torvalds * This allows us to do IO even on the odd last sectors 295257302e0dSLinus Torvalds * of a device, even if the bh block size is some multiple 295357302e0dSLinus Torvalds * of the physical sector size. 295457302e0dSLinus Torvalds * 295557302e0dSLinus Torvalds * We'll just truncate the bio to the size of the device, 295657302e0dSLinus Torvalds * and clear the end of the buffer head manually. 295757302e0dSLinus Torvalds * 295857302e0dSLinus Torvalds * Truly out-of-range accesses will turn into actual IO 295957302e0dSLinus Torvalds * errors, this only handles the "we need to be able to 296057302e0dSLinus Torvalds * do IO at the final sector" case. 296157302e0dSLinus Torvalds */ 296257302e0dSLinus Torvalds static void guard_bh_eod(int rw, struct bio *bio, struct buffer_head *bh) 296357302e0dSLinus Torvalds { 296457302e0dSLinus Torvalds sector_t maxsector; 296557302e0dSLinus Torvalds unsigned bytes; 296657302e0dSLinus Torvalds 296757302e0dSLinus Torvalds maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; 296857302e0dSLinus Torvalds if (!maxsector) 296957302e0dSLinus Torvalds return; 297057302e0dSLinus Torvalds 297157302e0dSLinus Torvalds /* 297257302e0dSLinus Torvalds * If the *whole* IO is past the end of the device, 297357302e0dSLinus Torvalds * let it through, and the IO layer will turn it into 297457302e0dSLinus Torvalds * an EIO. 297557302e0dSLinus Torvalds */ 29764f024f37SKent Overstreet if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 297757302e0dSLinus Torvalds return; 297857302e0dSLinus Torvalds 29794f024f37SKent Overstreet maxsector -= bio->bi_iter.bi_sector; 29804f024f37SKent Overstreet bytes = bio->bi_iter.bi_size; 298157302e0dSLinus Torvalds if (likely((bytes >> 9) <= maxsector)) 298257302e0dSLinus Torvalds return; 298357302e0dSLinus Torvalds 298457302e0dSLinus Torvalds /* Uhhuh. We've got a bh that straddles the device size! */ 298557302e0dSLinus Torvalds bytes = maxsector << 9; 298657302e0dSLinus Torvalds 298757302e0dSLinus Torvalds /* Truncate the bio.. */ 29884f024f37SKent Overstreet bio->bi_iter.bi_size = bytes; 298957302e0dSLinus Torvalds bio->bi_io_vec[0].bv_len = bytes; 299057302e0dSLinus Torvalds 299157302e0dSLinus Torvalds /* ..and clear the end of the buffer for reads */ 299227d7c2a0SDan Carpenter if ((rw & RW_MASK) == READ) { 299357302e0dSLinus Torvalds void *kaddr = kmap_atomic(bh->b_page); 299457302e0dSLinus Torvalds memset(kaddr + bh_offset(bh) + bytes, 0, bh->b_size - bytes); 299557302e0dSLinus Torvalds kunmap_atomic(kaddr); 29966d283dbaSLinus Torvalds flush_dcache_page(bh->b_page); 299757302e0dSLinus Torvalds } 299857302e0dSLinus Torvalds } 299957302e0dSLinus Torvalds 300071368511SDarrick J. Wong int _submit_bh(int rw, struct buffer_head *bh, unsigned long bio_flags) 30011da177e4SLinus Torvalds { 30021da177e4SLinus Torvalds struct bio *bio; 30031da177e4SLinus Torvalds int ret = 0; 30041da177e4SLinus Torvalds 30051da177e4SLinus Torvalds BUG_ON(!buffer_locked(bh)); 30061da177e4SLinus Torvalds BUG_ON(!buffer_mapped(bh)); 30071da177e4SLinus Torvalds BUG_ON(!bh->b_end_io); 30088fb0e342SAneesh Kumar K.V BUG_ON(buffer_delay(bh)); 30098fb0e342SAneesh Kumar K.V BUG_ON(buffer_unwritten(bh)); 30101da177e4SLinus Torvalds 301148fd4f93SJens Axboe /* 301248fd4f93SJens Axboe * Only clear out a write error when rewriting 30131da177e4SLinus Torvalds */ 301448fd4f93SJens Axboe if (test_set_buffer_req(bh) && (rw & WRITE)) 30151da177e4SLinus Torvalds clear_buffer_write_io_error(bh); 30161da177e4SLinus Torvalds 30171da177e4SLinus Torvalds /* 30181da177e4SLinus Torvalds * from here on down, it's all bio -- do the initial mapping, 30191da177e4SLinus Torvalds * submit_bio -> generic_make_request may further map this bio around 30201da177e4SLinus Torvalds */ 30211da177e4SLinus Torvalds bio = bio_alloc(GFP_NOIO, 1); 30221da177e4SLinus Torvalds 30234f024f37SKent Overstreet bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); 30241da177e4SLinus Torvalds bio->bi_bdev = bh->b_bdev; 30251da177e4SLinus Torvalds bio->bi_io_vec[0].bv_page = bh->b_page; 30261da177e4SLinus Torvalds bio->bi_io_vec[0].bv_len = bh->b_size; 30271da177e4SLinus Torvalds bio->bi_io_vec[0].bv_offset = bh_offset(bh); 30281da177e4SLinus Torvalds 30291da177e4SLinus Torvalds bio->bi_vcnt = 1; 30304f024f37SKent Overstreet bio->bi_iter.bi_size = bh->b_size; 30311da177e4SLinus Torvalds 30321da177e4SLinus Torvalds bio->bi_end_io = end_bio_bh_io_sync; 30331da177e4SLinus Torvalds bio->bi_private = bh; 303471368511SDarrick J. Wong bio->bi_flags |= bio_flags; 30351da177e4SLinus Torvalds 303657302e0dSLinus Torvalds /* Take care of bh's that straddle the end of the device */ 303757302e0dSLinus Torvalds guard_bh_eod(rw, bio, bh); 303857302e0dSLinus Torvalds 3039877f962cSTheodore Ts'o if (buffer_meta(bh)) 3040877f962cSTheodore Ts'o rw |= REQ_META; 3041877f962cSTheodore Ts'o if (buffer_prio(bh)) 3042877f962cSTheodore Ts'o rw |= REQ_PRIO; 3043877f962cSTheodore Ts'o 30441da177e4SLinus Torvalds bio_get(bio); 30451da177e4SLinus Torvalds submit_bio(rw, bio); 30461da177e4SLinus Torvalds 30471da177e4SLinus Torvalds if (bio_flagged(bio, BIO_EOPNOTSUPP)) 30481da177e4SLinus Torvalds ret = -EOPNOTSUPP; 30491da177e4SLinus Torvalds 30501da177e4SLinus Torvalds bio_put(bio); 30511da177e4SLinus Torvalds return ret; 30521da177e4SLinus Torvalds } 305371368511SDarrick J. Wong EXPORT_SYMBOL_GPL(_submit_bh); 305471368511SDarrick J. Wong 305571368511SDarrick J. Wong int submit_bh(int rw, struct buffer_head *bh) 305671368511SDarrick J. Wong { 305771368511SDarrick J. Wong return _submit_bh(rw, bh, 0); 305871368511SDarrick J. Wong } 30591fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(submit_bh); 30601da177e4SLinus Torvalds 30611da177e4SLinus Torvalds /** 30621da177e4SLinus Torvalds * ll_rw_block: low-level access to block devices (DEPRECATED) 30639cb569d6SChristoph Hellwig * @rw: whether to %READ or %WRITE or maybe %READA (readahead) 30641da177e4SLinus Torvalds * @nr: number of &struct buffer_heads in the array 30651da177e4SLinus Torvalds * @bhs: array of pointers to &struct buffer_head 30661da177e4SLinus Torvalds * 3067a7662236SJan Kara * ll_rw_block() takes an array of pointers to &struct buffer_heads, and 3068a7662236SJan Kara * requests an I/O operation on them, either a %READ or a %WRITE. The third 30699cb569d6SChristoph Hellwig * %READA option is described in the documentation for generic_make_request() 30709cb569d6SChristoph Hellwig * which ll_rw_block() calls. 30711da177e4SLinus Torvalds * 30721da177e4SLinus Torvalds * This function drops any buffer that it cannot get a lock on (with the 30739cb569d6SChristoph Hellwig * BH_Lock state bit), any buffer that appears to be clean when doing a write 30749cb569d6SChristoph Hellwig * request, and any buffer that appears to be up-to-date when doing read 30759cb569d6SChristoph Hellwig * request. Further it marks as clean buffers that are processed for 30769cb569d6SChristoph Hellwig * writing (the buffer cache won't assume that they are actually clean 30779cb569d6SChristoph Hellwig * until the buffer gets unlocked). 30781da177e4SLinus Torvalds * 30791da177e4SLinus Torvalds * ll_rw_block sets b_end_io to simple completion handler that marks 3080e227867fSMasanari Iida * the buffer up-to-date (if appropriate), unlocks the buffer and wakes 30811da177e4SLinus Torvalds * any waiters. 30821da177e4SLinus Torvalds * 30831da177e4SLinus Torvalds * All of the buffers must be for the same device, and must also be a 30841da177e4SLinus Torvalds * multiple of the current approved size for the device. 30851da177e4SLinus Torvalds */ 30861da177e4SLinus Torvalds void ll_rw_block(int rw, int nr, struct buffer_head *bhs[]) 30871da177e4SLinus Torvalds { 30881da177e4SLinus Torvalds int i; 30891da177e4SLinus Torvalds 30901da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 30911da177e4SLinus Torvalds struct buffer_head *bh = bhs[i]; 30921da177e4SLinus Torvalds 30939cb569d6SChristoph Hellwig if (!trylock_buffer(bh)) 30941da177e4SLinus Torvalds continue; 30959cb569d6SChristoph Hellwig if (rw == WRITE) { 30961da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 309776c3073aSakpm@osdl.org bh->b_end_io = end_buffer_write_sync; 3098e60e5c50SOGAWA Hirofumi get_bh(bh); 30991da177e4SLinus Torvalds submit_bh(WRITE, bh); 31001da177e4SLinus Torvalds continue; 31011da177e4SLinus Torvalds } 31021da177e4SLinus Torvalds } else { 31031da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 310476c3073aSakpm@osdl.org bh->b_end_io = end_buffer_read_sync; 3105e60e5c50SOGAWA Hirofumi get_bh(bh); 31061da177e4SLinus Torvalds submit_bh(rw, bh); 31071da177e4SLinus Torvalds continue; 31081da177e4SLinus Torvalds } 31091da177e4SLinus Torvalds } 31101da177e4SLinus Torvalds unlock_buffer(bh); 31111da177e4SLinus Torvalds } 31121da177e4SLinus Torvalds } 31131fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(ll_rw_block); 31141da177e4SLinus Torvalds 31159cb569d6SChristoph Hellwig void write_dirty_buffer(struct buffer_head *bh, int rw) 31169cb569d6SChristoph Hellwig { 31179cb569d6SChristoph Hellwig lock_buffer(bh); 31189cb569d6SChristoph Hellwig if (!test_clear_buffer_dirty(bh)) { 31199cb569d6SChristoph Hellwig unlock_buffer(bh); 31209cb569d6SChristoph Hellwig return; 31219cb569d6SChristoph Hellwig } 31229cb569d6SChristoph Hellwig bh->b_end_io = end_buffer_write_sync; 31239cb569d6SChristoph Hellwig get_bh(bh); 31249cb569d6SChristoph Hellwig submit_bh(rw, bh); 31259cb569d6SChristoph Hellwig } 31269cb569d6SChristoph Hellwig EXPORT_SYMBOL(write_dirty_buffer); 31279cb569d6SChristoph Hellwig 31281da177e4SLinus Torvalds /* 31291da177e4SLinus Torvalds * For a data-integrity writeout, we need to wait upon any in-progress I/O 31301da177e4SLinus Torvalds * and then start new I/O and then wait upon it. The caller must have a ref on 31311da177e4SLinus Torvalds * the buffer_head. 31321da177e4SLinus Torvalds */ 313387e99511SChristoph Hellwig int __sync_dirty_buffer(struct buffer_head *bh, int rw) 31341da177e4SLinus Torvalds { 31351da177e4SLinus Torvalds int ret = 0; 31361da177e4SLinus Torvalds 31371da177e4SLinus Torvalds WARN_ON(atomic_read(&bh->b_count) < 1); 31381da177e4SLinus Torvalds lock_buffer(bh); 31391da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 31401da177e4SLinus Torvalds get_bh(bh); 31411da177e4SLinus Torvalds bh->b_end_io = end_buffer_write_sync; 314287e99511SChristoph Hellwig ret = submit_bh(rw, bh); 31431da177e4SLinus Torvalds wait_on_buffer(bh); 31441da177e4SLinus Torvalds if (!ret && !buffer_uptodate(bh)) 31451da177e4SLinus Torvalds ret = -EIO; 31461da177e4SLinus Torvalds } else { 31471da177e4SLinus Torvalds unlock_buffer(bh); 31481da177e4SLinus Torvalds } 31491da177e4SLinus Torvalds return ret; 31501da177e4SLinus Torvalds } 315187e99511SChristoph Hellwig EXPORT_SYMBOL(__sync_dirty_buffer); 315287e99511SChristoph Hellwig 315387e99511SChristoph Hellwig int sync_dirty_buffer(struct buffer_head *bh) 315487e99511SChristoph Hellwig { 315587e99511SChristoph Hellwig return __sync_dirty_buffer(bh, WRITE_SYNC); 315687e99511SChristoph Hellwig } 31571fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(sync_dirty_buffer); 31581da177e4SLinus Torvalds 31591da177e4SLinus Torvalds /* 31601da177e4SLinus Torvalds * try_to_free_buffers() checks if all the buffers on this particular page 31611da177e4SLinus Torvalds * are unused, and releases them if so. 31621da177e4SLinus Torvalds * 31631da177e4SLinus Torvalds * Exclusion against try_to_free_buffers may be obtained by either 31641da177e4SLinus Torvalds * locking the page or by holding its mapping's private_lock. 31651da177e4SLinus Torvalds * 31661da177e4SLinus Torvalds * If the page is dirty but all the buffers are clean then we need to 31671da177e4SLinus Torvalds * be sure to mark the page clean as well. This is because the page 31681da177e4SLinus Torvalds * may be against a block device, and a later reattachment of buffers 31691da177e4SLinus Torvalds * to a dirty page will set *all* buffers dirty. Which would corrupt 31701da177e4SLinus Torvalds * filesystem data on the same device. 31711da177e4SLinus Torvalds * 31721da177e4SLinus Torvalds * The same applies to regular filesystem pages: if all the buffers are 31731da177e4SLinus Torvalds * clean then we set the page clean and proceed. To do that, we require 31741da177e4SLinus Torvalds * total exclusion from __set_page_dirty_buffers(). That is obtained with 31751da177e4SLinus Torvalds * private_lock. 31761da177e4SLinus Torvalds * 31771da177e4SLinus Torvalds * try_to_free_buffers() is non-blocking. 31781da177e4SLinus Torvalds */ 31791da177e4SLinus Torvalds static inline int buffer_busy(struct buffer_head *bh) 31801da177e4SLinus Torvalds { 31811da177e4SLinus Torvalds return atomic_read(&bh->b_count) | 31821da177e4SLinus Torvalds (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); 31831da177e4SLinus Torvalds } 31841da177e4SLinus Torvalds 31851da177e4SLinus Torvalds static int 31861da177e4SLinus Torvalds drop_buffers(struct page *page, struct buffer_head **buffers_to_free) 31871da177e4SLinus Torvalds { 31881da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 31891da177e4SLinus Torvalds struct buffer_head *bh; 31901da177e4SLinus Torvalds 31911da177e4SLinus Torvalds bh = head; 31921da177e4SLinus Torvalds do { 3193de7d5a3bSakpm@osdl.org if (buffer_write_io_error(bh) && page->mapping) 31941da177e4SLinus Torvalds set_bit(AS_EIO, &page->mapping->flags); 31951da177e4SLinus Torvalds if (buffer_busy(bh)) 31961da177e4SLinus Torvalds goto failed; 31971da177e4SLinus Torvalds bh = bh->b_this_page; 31981da177e4SLinus Torvalds } while (bh != head); 31991da177e4SLinus Torvalds 32001da177e4SLinus Torvalds do { 32011da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 32021da177e4SLinus Torvalds 3203535ee2fbSJan Kara if (bh->b_assoc_map) 32041da177e4SLinus Torvalds __remove_assoc_queue(bh); 32051da177e4SLinus Torvalds bh = next; 32061da177e4SLinus Torvalds } while (bh != head); 32071da177e4SLinus Torvalds *buffers_to_free = head; 32081da177e4SLinus Torvalds __clear_page_buffers(page); 32091da177e4SLinus Torvalds return 1; 32101da177e4SLinus Torvalds failed: 32111da177e4SLinus Torvalds return 0; 32121da177e4SLinus Torvalds } 32131da177e4SLinus Torvalds 32141da177e4SLinus Torvalds int try_to_free_buffers(struct page *page) 32151da177e4SLinus Torvalds { 32161da177e4SLinus Torvalds struct address_space * const mapping = page->mapping; 32171da177e4SLinus Torvalds struct buffer_head *buffers_to_free = NULL; 32181da177e4SLinus Torvalds int ret = 0; 32191da177e4SLinus Torvalds 32201da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 3221ecdfc978SLinus Torvalds if (PageWriteback(page)) 32221da177e4SLinus Torvalds return 0; 32231da177e4SLinus Torvalds 32241da177e4SLinus Torvalds if (mapping == NULL) { /* can this still happen? */ 32251da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 32261da177e4SLinus Torvalds goto out; 32271da177e4SLinus Torvalds } 32281da177e4SLinus Torvalds 32291da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 32301da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 3231ecdfc978SLinus Torvalds 3232ecdfc978SLinus Torvalds /* 3233ecdfc978SLinus Torvalds * If the filesystem writes its buffers by hand (eg ext3) 3234ecdfc978SLinus Torvalds * then we can have clean buffers against a dirty page. We 3235ecdfc978SLinus Torvalds * clean the page here; otherwise the VM will never notice 3236ecdfc978SLinus Torvalds * that the filesystem did any IO at all. 3237ecdfc978SLinus Torvalds * 3238ecdfc978SLinus Torvalds * Also, during truncate, discard_buffer will have marked all 3239ecdfc978SLinus Torvalds * the page's buffers clean. We discover that here and clean 3240ecdfc978SLinus Torvalds * the page also. 324187df7241SNick Piggin * 324287df7241SNick Piggin * private_lock must be held over this entire operation in order 324387df7241SNick Piggin * to synchronise against __set_page_dirty_buffers and prevent the 324487df7241SNick Piggin * dirty bit from being lost. 3245ecdfc978SLinus Torvalds */ 3246ecdfc978SLinus Torvalds if (ret) 3247ecdfc978SLinus Torvalds cancel_dirty_page(page, PAGE_CACHE_SIZE); 324887df7241SNick Piggin spin_unlock(&mapping->private_lock); 32491da177e4SLinus Torvalds out: 32501da177e4SLinus Torvalds if (buffers_to_free) { 32511da177e4SLinus Torvalds struct buffer_head *bh = buffers_to_free; 32521da177e4SLinus Torvalds 32531da177e4SLinus Torvalds do { 32541da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 32551da177e4SLinus Torvalds free_buffer_head(bh); 32561da177e4SLinus Torvalds bh = next; 32571da177e4SLinus Torvalds } while (bh != buffers_to_free); 32581da177e4SLinus Torvalds } 32591da177e4SLinus Torvalds return ret; 32601da177e4SLinus Torvalds } 32611da177e4SLinus Torvalds EXPORT_SYMBOL(try_to_free_buffers); 32621da177e4SLinus Torvalds 32631da177e4SLinus Torvalds /* 32641da177e4SLinus Torvalds * There are no bdflush tunables left. But distributions are 32651da177e4SLinus Torvalds * still running obsolete flush daemons, so we terminate them here. 32661da177e4SLinus Torvalds * 32671da177e4SLinus Torvalds * Use of bdflush() is deprecated and will be removed in a future kernel. 32685b0830cbSJens Axboe * The `flush-X' kernel threads fully replace bdflush daemons and this call. 32691da177e4SLinus Torvalds */ 3270bdc480e3SHeiko Carstens SYSCALL_DEFINE2(bdflush, int, func, long, data) 32711da177e4SLinus Torvalds { 32721da177e4SLinus Torvalds static int msg_count; 32731da177e4SLinus Torvalds 32741da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 32751da177e4SLinus Torvalds return -EPERM; 32761da177e4SLinus Torvalds 32771da177e4SLinus Torvalds if (msg_count < 5) { 32781da177e4SLinus Torvalds msg_count++; 32791da177e4SLinus Torvalds printk(KERN_INFO 32801da177e4SLinus Torvalds "warning: process `%s' used the obsolete bdflush" 32811da177e4SLinus Torvalds " system call\n", current->comm); 32821da177e4SLinus Torvalds printk(KERN_INFO "Fix your initscripts?\n"); 32831da177e4SLinus Torvalds } 32841da177e4SLinus Torvalds 32851da177e4SLinus Torvalds if (func == 1) 32861da177e4SLinus Torvalds do_exit(0); 32871da177e4SLinus Torvalds return 0; 32881da177e4SLinus Torvalds } 32891da177e4SLinus Torvalds 32901da177e4SLinus Torvalds /* 32911da177e4SLinus Torvalds * Buffer-head allocation 32921da177e4SLinus Torvalds */ 3293a0a9b043SShai Fultheim static struct kmem_cache *bh_cachep __read_mostly; 32941da177e4SLinus Torvalds 32951da177e4SLinus Torvalds /* 32961da177e4SLinus Torvalds * Once the number of bh's in the machine exceeds this level, we start 32971da177e4SLinus Torvalds * stripping them in writeback. 32981da177e4SLinus Torvalds */ 329943be594aSZhang Yanfei static unsigned long max_buffer_heads; 33001da177e4SLinus Torvalds 33011da177e4SLinus Torvalds int buffer_heads_over_limit; 33021da177e4SLinus Torvalds 33031da177e4SLinus Torvalds struct bh_accounting { 33041da177e4SLinus Torvalds int nr; /* Number of live bh's */ 33051da177e4SLinus Torvalds int ratelimit; /* Limit cacheline bouncing */ 33061da177e4SLinus Torvalds }; 33071da177e4SLinus Torvalds 33081da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; 33091da177e4SLinus Torvalds 33101da177e4SLinus Torvalds static void recalc_bh_state(void) 33111da177e4SLinus Torvalds { 33121da177e4SLinus Torvalds int i; 33131da177e4SLinus Torvalds int tot = 0; 33141da177e4SLinus Torvalds 3315ee1be862SChristoph Lameter if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) 33161da177e4SLinus Torvalds return; 3317c7b92516SChristoph Lameter __this_cpu_write(bh_accounting.ratelimit, 0); 33188a143426SEric Dumazet for_each_online_cpu(i) 33191da177e4SLinus Torvalds tot += per_cpu(bh_accounting, i).nr; 33201da177e4SLinus Torvalds buffer_heads_over_limit = (tot > max_buffer_heads); 33211da177e4SLinus Torvalds } 33221da177e4SLinus Torvalds 3323dd0fc66fSAl Viro struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) 33241da177e4SLinus Torvalds { 3325019b4d12SRichard Kennedy struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); 33261da177e4SLinus Torvalds if (ret) { 3327a35afb83SChristoph Lameter INIT_LIST_HEAD(&ret->b_assoc_buffers); 3328c7b92516SChristoph Lameter preempt_disable(); 3329c7b92516SChristoph Lameter __this_cpu_inc(bh_accounting.nr); 33301da177e4SLinus Torvalds recalc_bh_state(); 3331c7b92516SChristoph Lameter preempt_enable(); 33321da177e4SLinus Torvalds } 33331da177e4SLinus Torvalds return ret; 33341da177e4SLinus Torvalds } 33351da177e4SLinus Torvalds EXPORT_SYMBOL(alloc_buffer_head); 33361da177e4SLinus Torvalds 33371da177e4SLinus Torvalds void free_buffer_head(struct buffer_head *bh) 33381da177e4SLinus Torvalds { 33391da177e4SLinus Torvalds BUG_ON(!list_empty(&bh->b_assoc_buffers)); 33401da177e4SLinus Torvalds kmem_cache_free(bh_cachep, bh); 3341c7b92516SChristoph Lameter preempt_disable(); 3342c7b92516SChristoph Lameter __this_cpu_dec(bh_accounting.nr); 33431da177e4SLinus Torvalds recalc_bh_state(); 3344c7b92516SChristoph Lameter preempt_enable(); 33451da177e4SLinus Torvalds } 33461da177e4SLinus Torvalds EXPORT_SYMBOL(free_buffer_head); 33471da177e4SLinus Torvalds 33481da177e4SLinus Torvalds static void buffer_exit_cpu(int cpu) 33491da177e4SLinus Torvalds { 33501da177e4SLinus Torvalds int i; 33511da177e4SLinus Torvalds struct bh_lru *b = &per_cpu(bh_lrus, cpu); 33521da177e4SLinus Torvalds 33531da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 33541da177e4SLinus Torvalds brelse(b->bhs[i]); 33551da177e4SLinus Torvalds b->bhs[i] = NULL; 33561da177e4SLinus Torvalds } 3357c7b92516SChristoph Lameter this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); 33588a143426SEric Dumazet per_cpu(bh_accounting, cpu).nr = 0; 33591da177e4SLinus Torvalds } 33601da177e4SLinus Torvalds 33611da177e4SLinus Torvalds static int buffer_cpu_notify(struct notifier_block *self, 33621da177e4SLinus Torvalds unsigned long action, void *hcpu) 33631da177e4SLinus Torvalds { 33648bb78442SRafael J. Wysocki if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) 33651da177e4SLinus Torvalds buffer_exit_cpu((unsigned long)hcpu); 33661da177e4SLinus Torvalds return NOTIFY_OK; 33671da177e4SLinus Torvalds } 33681da177e4SLinus Torvalds 3369389d1b08SAneesh Kumar K.V /** 3370a6b91919SRandy Dunlap * bh_uptodate_or_lock - Test whether the buffer is uptodate 3371389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3372389d1b08SAneesh Kumar K.V * 3373389d1b08SAneesh Kumar K.V * Return true if the buffer is up-to-date and false, 3374389d1b08SAneesh Kumar K.V * with the buffer locked, if not. 3375389d1b08SAneesh Kumar K.V */ 3376389d1b08SAneesh Kumar K.V int bh_uptodate_or_lock(struct buffer_head *bh) 3377389d1b08SAneesh Kumar K.V { 3378389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) { 3379389d1b08SAneesh Kumar K.V lock_buffer(bh); 3380389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) 3381389d1b08SAneesh Kumar K.V return 0; 3382389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3383389d1b08SAneesh Kumar K.V } 3384389d1b08SAneesh Kumar K.V return 1; 3385389d1b08SAneesh Kumar K.V } 3386389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_uptodate_or_lock); 3387389d1b08SAneesh Kumar K.V 3388389d1b08SAneesh Kumar K.V /** 3389a6b91919SRandy Dunlap * bh_submit_read - Submit a locked buffer for reading 3390389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3391389d1b08SAneesh Kumar K.V * 3392389d1b08SAneesh Kumar K.V * Returns zero on success and -EIO on error. 3393389d1b08SAneesh Kumar K.V */ 3394389d1b08SAneesh Kumar K.V int bh_submit_read(struct buffer_head *bh) 3395389d1b08SAneesh Kumar K.V { 3396389d1b08SAneesh Kumar K.V BUG_ON(!buffer_locked(bh)); 3397389d1b08SAneesh Kumar K.V 3398389d1b08SAneesh Kumar K.V if (buffer_uptodate(bh)) { 3399389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3400389d1b08SAneesh Kumar K.V return 0; 3401389d1b08SAneesh Kumar K.V } 3402389d1b08SAneesh Kumar K.V 3403389d1b08SAneesh Kumar K.V get_bh(bh); 3404389d1b08SAneesh Kumar K.V bh->b_end_io = end_buffer_read_sync; 3405389d1b08SAneesh Kumar K.V submit_bh(READ, bh); 3406389d1b08SAneesh Kumar K.V wait_on_buffer(bh); 3407389d1b08SAneesh Kumar K.V if (buffer_uptodate(bh)) 3408389d1b08SAneesh Kumar K.V return 0; 3409389d1b08SAneesh Kumar K.V return -EIO; 3410389d1b08SAneesh Kumar K.V } 3411389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_submit_read); 3412389d1b08SAneesh Kumar K.V 34131da177e4SLinus Torvalds void __init buffer_init(void) 34141da177e4SLinus Torvalds { 341543be594aSZhang Yanfei unsigned long nrpages; 34161da177e4SLinus Torvalds 3417b98938c3SChristoph Lameter bh_cachep = kmem_cache_create("buffer_head", 3418b98938c3SChristoph Lameter sizeof(struct buffer_head), 0, 3419b98938c3SChristoph Lameter (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| 3420b98938c3SChristoph Lameter SLAB_MEM_SPREAD), 3421019b4d12SRichard Kennedy NULL); 34221da177e4SLinus Torvalds 34231da177e4SLinus Torvalds /* 34241da177e4SLinus Torvalds * Limit the bh occupancy to 10% of ZONE_NORMAL 34251da177e4SLinus Torvalds */ 34261da177e4SLinus Torvalds nrpages = (nr_free_buffer_pages() * 10) / 100; 34271da177e4SLinus Torvalds max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); 34281da177e4SLinus Torvalds hotcpu_notifier(buffer_cpu_notify, 0); 34291da177e4SLinus Torvalds } 3430