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); 801da177e4SLinus Torvalds smp_mb__after_clear_bit(); 811da177e4SLinus Torvalds wake_up_bit(&bh->b_state, BH_Lock); 821da177e4SLinus Torvalds } 831fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(unlock_buffer); 841da177e4SLinus Torvalds 851da177e4SLinus Torvalds /* 861da177e4SLinus Torvalds * Block until a buffer comes unlocked. This doesn't stop it 871da177e4SLinus Torvalds * from becoming locked again - you have to lock it yourself 881da177e4SLinus Torvalds * if you want to preserve its state. 891da177e4SLinus Torvalds */ 901da177e4SLinus Torvalds void __wait_on_buffer(struct buffer_head * bh) 911da177e4SLinus Torvalds { 927eaceaccSJens Axboe wait_on_bit(&bh->b_state, BH_Lock, sleep_on_buffer, TASK_UNINTERRUPTIBLE); 931da177e4SLinus Torvalds } 941fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__wait_on_buffer); 951da177e4SLinus Torvalds 961da177e4SLinus Torvalds static void 971da177e4SLinus Torvalds __clear_page_buffers(struct page *page) 981da177e4SLinus Torvalds { 991da177e4SLinus Torvalds ClearPagePrivate(page); 1004c21e2f2SHugh Dickins set_page_private(page, 0); 1011da177e4SLinus Torvalds page_cache_release(page); 1021da177e4SLinus Torvalds } 1031da177e4SLinus Torvalds 10408bafc03SKeith Mannthey 10508bafc03SKeith Mannthey static int quiet_error(struct buffer_head *bh) 10608bafc03SKeith Mannthey { 10708bafc03SKeith Mannthey if (!test_bit(BH_Quiet, &bh->b_state) && printk_ratelimit()) 10808bafc03SKeith Mannthey return 0; 10908bafc03SKeith Mannthey return 1; 11008bafc03SKeith Mannthey } 11108bafc03SKeith Mannthey 11208bafc03SKeith Mannthey 1131da177e4SLinus Torvalds static void buffer_io_error(struct buffer_head *bh) 1141da177e4SLinus Torvalds { 1151da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 1161da177e4SLinus Torvalds printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n", 1171da177e4SLinus Torvalds bdevname(bh->b_bdev, b), 1181da177e4SLinus Torvalds (unsigned long long)bh->b_blocknr); 1191da177e4SLinus Torvalds } 1201da177e4SLinus Torvalds 1211da177e4SLinus Torvalds /* 12268671f35SDmitry Monakhov * End-of-IO handler helper function which does not touch the bh after 12368671f35SDmitry Monakhov * unlocking it. 12468671f35SDmitry Monakhov * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but 12568671f35SDmitry Monakhov * a race there is benign: unlock_buffer() only use the bh's address for 12668671f35SDmitry Monakhov * hashing after unlocking the buffer, so it doesn't actually touch the bh 12768671f35SDmitry Monakhov * itself. 1281da177e4SLinus Torvalds */ 12968671f35SDmitry Monakhov static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) 1301da177e4SLinus Torvalds { 1311da177e4SLinus Torvalds if (uptodate) { 1321da177e4SLinus Torvalds set_buffer_uptodate(bh); 1331da177e4SLinus Torvalds } else { 1341da177e4SLinus Torvalds /* This happens, due to failed READA attempts. */ 1351da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1361da177e4SLinus Torvalds } 1371da177e4SLinus Torvalds unlock_buffer(bh); 13868671f35SDmitry Monakhov } 13968671f35SDmitry Monakhov 14068671f35SDmitry Monakhov /* 14168671f35SDmitry Monakhov * Default synchronous end-of-IO handler.. Just mark it up-to-date and 14268671f35SDmitry Monakhov * unlock the buffer. This is what ll_rw_block uses too. 14368671f35SDmitry Monakhov */ 14468671f35SDmitry Monakhov void end_buffer_read_sync(struct buffer_head *bh, int uptodate) 14568671f35SDmitry Monakhov { 14668671f35SDmitry Monakhov __end_buffer_read_notouch(bh, uptodate); 1471da177e4SLinus Torvalds put_bh(bh); 1481da177e4SLinus Torvalds } 1491fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_read_sync); 1501da177e4SLinus Torvalds 1511da177e4SLinus Torvalds void end_buffer_write_sync(struct buffer_head *bh, int uptodate) 1521da177e4SLinus Torvalds { 1531da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 1541da177e4SLinus Torvalds 1551da177e4SLinus Torvalds if (uptodate) { 1561da177e4SLinus Torvalds set_buffer_uptodate(bh); 1571da177e4SLinus Torvalds } else { 1580edd55faSChristoph Hellwig if (!quiet_error(bh)) { 1591da177e4SLinus Torvalds buffer_io_error(bh); 1601da177e4SLinus Torvalds printk(KERN_WARNING "lost page write due to " 1611da177e4SLinus Torvalds "I/O error on %s\n", 1621da177e4SLinus Torvalds bdevname(bh->b_bdev, b)); 1631da177e4SLinus Torvalds } 1641da177e4SLinus Torvalds set_buffer_write_io_error(bh); 1651da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1661da177e4SLinus Torvalds } 1671da177e4SLinus Torvalds unlock_buffer(bh); 1681da177e4SLinus Torvalds put_bh(bh); 1691da177e4SLinus Torvalds } 1701fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_write_sync); 1711da177e4SLinus Torvalds 1721da177e4SLinus Torvalds /* 1731da177e4SLinus Torvalds * Various filesystems appear to want __find_get_block to be non-blocking. 1741da177e4SLinus Torvalds * But it's the page lock which protects the buffers. To get around this, 1751da177e4SLinus Torvalds * we get exclusion from try_to_free_buffers with the blockdev mapping's 1761da177e4SLinus Torvalds * private_lock. 1771da177e4SLinus Torvalds * 1781da177e4SLinus Torvalds * Hack idea: for the blockdev mapping, i_bufferlist_lock contention 1791da177e4SLinus Torvalds * may be quite high. This code could TryLock the page, and if that 1801da177e4SLinus Torvalds * succeeds, there is no need to take private_lock. (But if 1811da177e4SLinus Torvalds * private_lock is contended then so is mapping->tree_lock). 1821da177e4SLinus Torvalds */ 1831da177e4SLinus Torvalds static struct buffer_head * 184385fd4c5SCoywolf Qi Hunt __find_get_block_slow(struct block_device *bdev, sector_t block) 1851da177e4SLinus Torvalds { 1861da177e4SLinus Torvalds struct inode *bd_inode = bdev->bd_inode; 1871da177e4SLinus Torvalds struct address_space *bd_mapping = bd_inode->i_mapping; 1881da177e4SLinus Torvalds struct buffer_head *ret = NULL; 1891da177e4SLinus Torvalds pgoff_t index; 1901da177e4SLinus Torvalds struct buffer_head *bh; 1911da177e4SLinus Torvalds struct buffer_head *head; 1921da177e4SLinus Torvalds struct page *page; 1931da177e4SLinus Torvalds int all_mapped = 1; 1941da177e4SLinus Torvalds 1951da177e4SLinus Torvalds index = block >> (PAGE_CACHE_SHIFT - bd_inode->i_blkbits); 1961da177e4SLinus Torvalds page = find_get_page(bd_mapping, index); 1971da177e4SLinus Torvalds if (!page) 1981da177e4SLinus Torvalds goto out; 1991da177e4SLinus Torvalds 2001da177e4SLinus Torvalds spin_lock(&bd_mapping->private_lock); 2011da177e4SLinus Torvalds if (!page_has_buffers(page)) 2021da177e4SLinus Torvalds goto out_unlock; 2031da177e4SLinus Torvalds head = page_buffers(page); 2041da177e4SLinus Torvalds bh = head; 2051da177e4SLinus Torvalds do { 20697f76d3dSNikanth Karthikesan if (!buffer_mapped(bh)) 20797f76d3dSNikanth Karthikesan all_mapped = 0; 20897f76d3dSNikanth Karthikesan else if (bh->b_blocknr == block) { 2091da177e4SLinus Torvalds ret = bh; 2101da177e4SLinus Torvalds get_bh(bh); 2111da177e4SLinus Torvalds goto out_unlock; 2121da177e4SLinus Torvalds } 2131da177e4SLinus Torvalds bh = bh->b_this_page; 2141da177e4SLinus Torvalds } while (bh != head); 2151da177e4SLinus Torvalds 2161da177e4SLinus Torvalds /* we might be here because some of the buffers on this page are 2171da177e4SLinus Torvalds * not mapped. This is due to various races between 2181da177e4SLinus Torvalds * file io on the block device and getblk. It gets dealt with 2191da177e4SLinus Torvalds * elsewhere, don't buffer_error if we had some unmapped buffers 2201da177e4SLinus Torvalds */ 2211da177e4SLinus Torvalds if (all_mapped) { 22272a2ebd8STao Ma char b[BDEVNAME_SIZE]; 22372a2ebd8STao Ma 2241da177e4SLinus Torvalds printk("__find_get_block_slow() failed. " 2251da177e4SLinus Torvalds "block=%llu, b_blocknr=%llu\n", 226205f87f6SBadari Pulavarty (unsigned long long)block, 227205f87f6SBadari Pulavarty (unsigned long long)bh->b_blocknr); 228205f87f6SBadari Pulavarty printk("b_state=0x%08lx, b_size=%zu\n", 229205f87f6SBadari Pulavarty bh->b_state, bh->b_size); 23072a2ebd8STao Ma printk("device %s blocksize: %d\n", bdevname(bdev, b), 23172a2ebd8STao Ma 1 << bd_inode->i_blkbits); 2321da177e4SLinus Torvalds } 2331da177e4SLinus Torvalds out_unlock: 2341da177e4SLinus Torvalds spin_unlock(&bd_mapping->private_lock); 2351da177e4SLinus Torvalds page_cache_release(page); 2361da177e4SLinus Torvalds out: 2371da177e4SLinus Torvalds return ret; 2381da177e4SLinus Torvalds } 2391da177e4SLinus Torvalds 2401da177e4SLinus Torvalds /* 2415b0830cbSJens Axboe * Kick the writeback threads then try to free up some ZONE_NORMAL memory. 2421da177e4SLinus Torvalds */ 2431da177e4SLinus Torvalds static void free_more_memory(void) 2441da177e4SLinus Torvalds { 24519770b32SMel Gorman struct zone *zone; 2460e88460dSMel Gorman int nid; 2471da177e4SLinus Torvalds 2480e175a18SCurt Wohlgemuth wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM); 2491da177e4SLinus Torvalds yield(); 2501da177e4SLinus Torvalds 2510e88460dSMel Gorman for_each_online_node(nid) { 25219770b32SMel Gorman (void)first_zones_zonelist(node_zonelist(nid, GFP_NOFS), 25319770b32SMel Gorman gfp_zone(GFP_NOFS), NULL, 25419770b32SMel Gorman &zone); 25519770b32SMel Gorman if (zone) 25654a6eb5cSMel Gorman try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0, 257327c0e96SKAMEZAWA Hiroyuki GFP_NOFS, NULL); 2581da177e4SLinus Torvalds } 2591da177e4SLinus Torvalds } 2601da177e4SLinus Torvalds 2611da177e4SLinus Torvalds /* 2621da177e4SLinus Torvalds * I/O completion handler for block_read_full_page() - pages 2631da177e4SLinus Torvalds * which come unlocked at the end of I/O. 2641da177e4SLinus Torvalds */ 2651da177e4SLinus Torvalds static void end_buffer_async_read(struct buffer_head *bh, int uptodate) 2661da177e4SLinus Torvalds { 2671da177e4SLinus Torvalds unsigned long flags; 268a3972203SNick Piggin struct buffer_head *first; 2691da177e4SLinus Torvalds struct buffer_head *tmp; 2701da177e4SLinus Torvalds struct page *page; 2711da177e4SLinus Torvalds int page_uptodate = 1; 2721da177e4SLinus Torvalds 2731da177e4SLinus Torvalds BUG_ON(!buffer_async_read(bh)); 2741da177e4SLinus Torvalds 2751da177e4SLinus Torvalds page = bh->b_page; 2761da177e4SLinus Torvalds if (uptodate) { 2771da177e4SLinus Torvalds set_buffer_uptodate(bh); 2781da177e4SLinus Torvalds } else { 2791da177e4SLinus Torvalds clear_buffer_uptodate(bh); 28008bafc03SKeith Mannthey if (!quiet_error(bh)) 2811da177e4SLinus Torvalds buffer_io_error(bh); 2821da177e4SLinus Torvalds SetPageError(page); 2831da177e4SLinus Torvalds } 2841da177e4SLinus Torvalds 2851da177e4SLinus Torvalds /* 2861da177e4SLinus Torvalds * Be _very_ careful from here on. Bad things can happen if 2871da177e4SLinus Torvalds * two buffer heads end IO at almost the same time and both 2881da177e4SLinus Torvalds * decide that the page is now completely done. 2891da177e4SLinus Torvalds */ 290a3972203SNick Piggin first = page_buffers(page); 291a3972203SNick Piggin local_irq_save(flags); 292a3972203SNick Piggin bit_spin_lock(BH_Uptodate_Lock, &first->b_state); 2931da177e4SLinus Torvalds clear_buffer_async_read(bh); 2941da177e4SLinus Torvalds unlock_buffer(bh); 2951da177e4SLinus Torvalds tmp = bh; 2961da177e4SLinus Torvalds do { 2971da177e4SLinus Torvalds if (!buffer_uptodate(tmp)) 2981da177e4SLinus Torvalds page_uptodate = 0; 2991da177e4SLinus Torvalds if (buffer_async_read(tmp)) { 3001da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 3011da177e4SLinus Torvalds goto still_busy; 3021da177e4SLinus Torvalds } 3031da177e4SLinus Torvalds tmp = tmp->b_this_page; 3041da177e4SLinus Torvalds } while (tmp != bh); 305a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 306a3972203SNick Piggin local_irq_restore(flags); 3071da177e4SLinus Torvalds 3081da177e4SLinus Torvalds /* 3091da177e4SLinus Torvalds * If none of the buffers had errors and they are all 3101da177e4SLinus Torvalds * uptodate then we can set the page uptodate. 3111da177e4SLinus Torvalds */ 3121da177e4SLinus Torvalds if (page_uptodate && !PageError(page)) 3131da177e4SLinus Torvalds SetPageUptodate(page); 3141da177e4SLinus Torvalds unlock_page(page); 3151da177e4SLinus Torvalds return; 3161da177e4SLinus Torvalds 3171da177e4SLinus Torvalds still_busy: 318a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 319a3972203SNick Piggin local_irq_restore(flags); 3201da177e4SLinus Torvalds return; 3211da177e4SLinus Torvalds } 3221da177e4SLinus Torvalds 3231da177e4SLinus Torvalds /* 3241da177e4SLinus Torvalds * Completion handler for block_write_full_page() - pages which are unlocked 3251da177e4SLinus Torvalds * during I/O, and which have PageWriteback cleared upon I/O completion. 3261da177e4SLinus Torvalds */ 32735c80d5fSChris Mason void end_buffer_async_write(struct buffer_head *bh, int uptodate) 3281da177e4SLinus Torvalds { 3291da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 3301da177e4SLinus Torvalds unsigned long flags; 331a3972203SNick Piggin struct buffer_head *first; 3321da177e4SLinus Torvalds struct buffer_head *tmp; 3331da177e4SLinus Torvalds struct page *page; 3341da177e4SLinus Torvalds 3351da177e4SLinus Torvalds BUG_ON(!buffer_async_write(bh)); 3361da177e4SLinus Torvalds 3371da177e4SLinus Torvalds page = bh->b_page; 3381da177e4SLinus Torvalds if (uptodate) { 3391da177e4SLinus Torvalds set_buffer_uptodate(bh); 3401da177e4SLinus Torvalds } else { 34108bafc03SKeith Mannthey if (!quiet_error(bh)) { 3421da177e4SLinus Torvalds buffer_io_error(bh); 3431da177e4SLinus Torvalds printk(KERN_WARNING "lost page write due to " 3441da177e4SLinus Torvalds "I/O error on %s\n", 3451da177e4SLinus Torvalds bdevname(bh->b_bdev, b)); 3461da177e4SLinus Torvalds } 3471da177e4SLinus Torvalds set_bit(AS_EIO, &page->mapping->flags); 34858ff407bSJan Kara set_buffer_write_io_error(bh); 3491da177e4SLinus Torvalds clear_buffer_uptodate(bh); 3501da177e4SLinus Torvalds SetPageError(page); 3511da177e4SLinus Torvalds } 3521da177e4SLinus Torvalds 353a3972203SNick Piggin first = page_buffers(page); 354a3972203SNick Piggin local_irq_save(flags); 355a3972203SNick Piggin bit_spin_lock(BH_Uptodate_Lock, &first->b_state); 356a3972203SNick Piggin 3571da177e4SLinus Torvalds clear_buffer_async_write(bh); 3581da177e4SLinus Torvalds unlock_buffer(bh); 3591da177e4SLinus Torvalds tmp = bh->b_this_page; 3601da177e4SLinus Torvalds while (tmp != bh) { 3611da177e4SLinus Torvalds if (buffer_async_write(tmp)) { 3621da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 3631da177e4SLinus Torvalds goto still_busy; 3641da177e4SLinus Torvalds } 3651da177e4SLinus Torvalds tmp = tmp->b_this_page; 3661da177e4SLinus Torvalds } 367a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 368a3972203SNick Piggin local_irq_restore(flags); 3691da177e4SLinus Torvalds end_page_writeback(page); 3701da177e4SLinus Torvalds return; 3711da177e4SLinus Torvalds 3721da177e4SLinus Torvalds still_busy: 373a3972203SNick Piggin bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); 374a3972203SNick Piggin local_irq_restore(flags); 3751da177e4SLinus Torvalds return; 3761da177e4SLinus Torvalds } 3771fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_async_write); 3781da177e4SLinus Torvalds 3791da177e4SLinus Torvalds /* 3801da177e4SLinus Torvalds * If a page's buffers are under async readin (end_buffer_async_read 3811da177e4SLinus Torvalds * completion) then there is a possibility that another thread of 3821da177e4SLinus Torvalds * control could lock one of the buffers after it has completed 3831da177e4SLinus Torvalds * but while some of the other buffers have not completed. This 3841da177e4SLinus Torvalds * locked buffer would confuse end_buffer_async_read() into not unlocking 3851da177e4SLinus Torvalds * the page. So the absence of BH_Async_Read tells end_buffer_async_read() 3861da177e4SLinus Torvalds * that this buffer is not under async I/O. 3871da177e4SLinus Torvalds * 3881da177e4SLinus Torvalds * The page comes unlocked when it has no locked buffer_async buffers 3891da177e4SLinus Torvalds * left. 3901da177e4SLinus Torvalds * 3911da177e4SLinus Torvalds * PageLocked prevents anyone starting new async I/O reads any of 3921da177e4SLinus Torvalds * the buffers. 3931da177e4SLinus Torvalds * 3941da177e4SLinus Torvalds * PageWriteback is used to prevent simultaneous writeout of the same 3951da177e4SLinus Torvalds * page. 3961da177e4SLinus Torvalds * 3971da177e4SLinus Torvalds * PageLocked prevents anyone from starting writeback of a page which is 3981da177e4SLinus Torvalds * under read I/O (PageWriteback is only ever set against a locked page). 3991da177e4SLinus Torvalds */ 4001da177e4SLinus Torvalds static void mark_buffer_async_read(struct buffer_head *bh) 4011da177e4SLinus Torvalds { 4021da177e4SLinus Torvalds bh->b_end_io = end_buffer_async_read; 4031da177e4SLinus Torvalds set_buffer_async_read(bh); 4041da177e4SLinus Torvalds } 4051da177e4SLinus Torvalds 4061fe72eaaSH Hartley Sweeten static void mark_buffer_async_write_endio(struct buffer_head *bh, 40735c80d5fSChris Mason bh_end_io_t *handler) 40835c80d5fSChris Mason { 40935c80d5fSChris Mason bh->b_end_io = handler; 41035c80d5fSChris Mason set_buffer_async_write(bh); 41135c80d5fSChris Mason } 41235c80d5fSChris Mason 4131da177e4SLinus Torvalds void mark_buffer_async_write(struct buffer_head *bh) 4141da177e4SLinus Torvalds { 41535c80d5fSChris Mason mark_buffer_async_write_endio(bh, end_buffer_async_write); 4161da177e4SLinus Torvalds } 4171da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_async_write); 4181da177e4SLinus Torvalds 4191da177e4SLinus Torvalds 4201da177e4SLinus Torvalds /* 4211da177e4SLinus Torvalds * fs/buffer.c contains helper functions for buffer-backed address space's 4221da177e4SLinus Torvalds * fsync functions. A common requirement for buffer-based filesystems is 4231da177e4SLinus Torvalds * that certain data from the backing blockdev needs to be written out for 4241da177e4SLinus Torvalds * a successful fsync(). For example, ext2 indirect blocks need to be 4251da177e4SLinus Torvalds * written back and waited upon before fsync() returns. 4261da177e4SLinus Torvalds * 4271da177e4SLinus Torvalds * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), 4281da177e4SLinus Torvalds * inode_has_buffers() and invalidate_inode_buffers() are provided for the 4291da177e4SLinus Torvalds * management of a list of dependent buffers at ->i_mapping->private_list. 4301da177e4SLinus Torvalds * 4311da177e4SLinus Torvalds * Locking is a little subtle: try_to_free_buffers() will remove buffers 4321da177e4SLinus Torvalds * from their controlling inode's queue when they are being freed. But 4331da177e4SLinus Torvalds * try_to_free_buffers() will be operating against the *blockdev* mapping 4341da177e4SLinus Torvalds * at the time, not against the S_ISREG file which depends on those buffers. 4351da177e4SLinus Torvalds * So the locking for private_list is via the private_lock in the address_space 4361da177e4SLinus Torvalds * which backs the buffers. Which is different from the address_space 4371da177e4SLinus Torvalds * against which the buffers are listed. So for a particular address_space, 4381da177e4SLinus Torvalds * mapping->private_lock does *not* protect mapping->private_list! In fact, 4391da177e4SLinus Torvalds * mapping->private_list will always be protected by the backing blockdev's 4401da177e4SLinus Torvalds * ->private_lock. 4411da177e4SLinus Torvalds * 4421da177e4SLinus Torvalds * Which introduces a requirement: all buffers on an address_space's 4431da177e4SLinus Torvalds * ->private_list must be from the same address_space: the blockdev's. 4441da177e4SLinus Torvalds * 4451da177e4SLinus Torvalds * address_spaces which do not place buffers at ->private_list via these 4461da177e4SLinus Torvalds * utility functions are free to use private_lock and private_list for 4471da177e4SLinus Torvalds * whatever they want. The only requirement is that list_empty(private_list) 4481da177e4SLinus Torvalds * be true at clear_inode() time. 4491da177e4SLinus Torvalds * 4501da177e4SLinus Torvalds * FIXME: clear_inode should not call invalidate_inode_buffers(). The 4511da177e4SLinus Torvalds * filesystems should do that. invalidate_inode_buffers() should just go 4521da177e4SLinus Torvalds * BUG_ON(!list_empty). 4531da177e4SLinus Torvalds * 4541da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should 4551da177e4SLinus Torvalds * take an address_space, not an inode. And it should be called 4561da177e4SLinus Torvalds * mark_buffer_dirty_fsync() to clearly define why those buffers are being 4571da177e4SLinus Torvalds * queued up. 4581da177e4SLinus Torvalds * 4591da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the 4601da177e4SLinus Torvalds * list if it is already on a list. Because if the buffer is on a list, 4611da177e4SLinus Torvalds * it *must* already be on the right one. If not, the filesystem is being 4621da177e4SLinus Torvalds * silly. This will save a ton of locking. But first we have to ensure 4631da177e4SLinus Torvalds * that buffers are taken *off* the old inode's list when they are freed 4641da177e4SLinus Torvalds * (presumably in truncate). That requires careful auditing of all 4651da177e4SLinus Torvalds * filesystems (do it inside bforget()). It could also be done by bringing 4661da177e4SLinus Torvalds * b_inode back. 4671da177e4SLinus Torvalds */ 4681da177e4SLinus Torvalds 4691da177e4SLinus Torvalds /* 4701da177e4SLinus Torvalds * The buffer's backing address_space's private_lock must be held 4711da177e4SLinus Torvalds */ 472dbacefc9SThomas Petazzoni static void __remove_assoc_queue(struct buffer_head *bh) 4731da177e4SLinus Torvalds { 4741da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 47558ff407bSJan Kara WARN_ON(!bh->b_assoc_map); 47658ff407bSJan Kara if (buffer_write_io_error(bh)) 47758ff407bSJan Kara set_bit(AS_EIO, &bh->b_assoc_map->flags); 47858ff407bSJan Kara bh->b_assoc_map = NULL; 4791da177e4SLinus Torvalds } 4801da177e4SLinus Torvalds 4811da177e4SLinus Torvalds int inode_has_buffers(struct inode *inode) 4821da177e4SLinus Torvalds { 4831da177e4SLinus Torvalds return !list_empty(&inode->i_data.private_list); 4841da177e4SLinus Torvalds } 4851da177e4SLinus Torvalds 4861da177e4SLinus Torvalds /* 4871da177e4SLinus Torvalds * osync is designed to support O_SYNC io. It waits synchronously for 4881da177e4SLinus Torvalds * all already-submitted IO to complete, but does not queue any new 4891da177e4SLinus Torvalds * writes to the disk. 4901da177e4SLinus Torvalds * 4911da177e4SLinus Torvalds * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as 4921da177e4SLinus Torvalds * you dirty the buffers, and then use osync_inode_buffers to wait for 4931da177e4SLinus Torvalds * completion. Any other dirty buffers which are not yet queued for 4941da177e4SLinus Torvalds * write will not be flushed to disk by the osync. 4951da177e4SLinus Torvalds */ 4961da177e4SLinus Torvalds static int osync_buffers_list(spinlock_t *lock, struct list_head *list) 4971da177e4SLinus Torvalds { 4981da177e4SLinus Torvalds struct buffer_head *bh; 4991da177e4SLinus Torvalds struct list_head *p; 5001da177e4SLinus Torvalds int err = 0; 5011da177e4SLinus Torvalds 5021da177e4SLinus Torvalds spin_lock(lock); 5031da177e4SLinus Torvalds repeat: 5041da177e4SLinus Torvalds list_for_each_prev(p, list) { 5051da177e4SLinus Torvalds bh = BH_ENTRY(p); 5061da177e4SLinus Torvalds if (buffer_locked(bh)) { 5071da177e4SLinus Torvalds get_bh(bh); 5081da177e4SLinus Torvalds spin_unlock(lock); 5091da177e4SLinus Torvalds wait_on_buffer(bh); 5101da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 5111da177e4SLinus Torvalds err = -EIO; 5121da177e4SLinus Torvalds brelse(bh); 5131da177e4SLinus Torvalds spin_lock(lock); 5141da177e4SLinus Torvalds goto repeat; 5151da177e4SLinus Torvalds } 5161da177e4SLinus Torvalds } 5171da177e4SLinus Torvalds spin_unlock(lock); 5181da177e4SLinus Torvalds return err; 5191da177e4SLinus Torvalds } 5201da177e4SLinus Torvalds 52101a05b33SAl Viro static void do_thaw_one(struct super_block *sb, void *unused) 522c2d75438SEric Sandeen { 523c2d75438SEric Sandeen char b[BDEVNAME_SIZE]; 524c2d75438SEric Sandeen while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb)) 525c2d75438SEric Sandeen printk(KERN_WARNING "Emergency Thaw on %s\n", 526c2d75438SEric Sandeen bdevname(sb->s_bdev, b)); 527c2d75438SEric Sandeen } 52801a05b33SAl Viro 52901a05b33SAl Viro static void do_thaw_all(struct work_struct *work) 53001a05b33SAl Viro { 53101a05b33SAl Viro iterate_supers(do_thaw_one, NULL); 532053c525fSJens Axboe kfree(work); 533c2d75438SEric Sandeen printk(KERN_WARNING "Emergency Thaw complete\n"); 534c2d75438SEric Sandeen } 535c2d75438SEric Sandeen 536c2d75438SEric Sandeen /** 537c2d75438SEric Sandeen * emergency_thaw_all -- forcibly thaw every frozen filesystem 538c2d75438SEric Sandeen * 539c2d75438SEric Sandeen * Used for emergency unfreeze of all filesystems via SysRq 540c2d75438SEric Sandeen */ 541c2d75438SEric Sandeen void emergency_thaw_all(void) 542c2d75438SEric Sandeen { 543053c525fSJens Axboe struct work_struct *work; 544053c525fSJens Axboe 545053c525fSJens Axboe work = kmalloc(sizeof(*work), GFP_ATOMIC); 546053c525fSJens Axboe if (work) { 547053c525fSJens Axboe INIT_WORK(work, do_thaw_all); 548053c525fSJens Axboe schedule_work(work); 549053c525fSJens Axboe } 550c2d75438SEric Sandeen } 551c2d75438SEric Sandeen 5521da177e4SLinus Torvalds /** 55378a4a50aSRandy Dunlap * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers 55467be2dd1SMartin Waitz * @mapping: the mapping which wants those buffers written 5551da177e4SLinus Torvalds * 5561da177e4SLinus Torvalds * Starts I/O against the buffers at mapping->private_list, and waits upon 5571da177e4SLinus Torvalds * that I/O. 5581da177e4SLinus Torvalds * 55967be2dd1SMartin Waitz * Basically, this is a convenience function for fsync(). 56067be2dd1SMartin Waitz * @mapping is a file or directory which needs those buffers to be written for 56167be2dd1SMartin Waitz * a successful fsync(). 5621da177e4SLinus Torvalds */ 5631da177e4SLinus Torvalds int sync_mapping_buffers(struct address_space *mapping) 5641da177e4SLinus Torvalds { 565252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 5661da177e4SLinus Torvalds 5671da177e4SLinus Torvalds if (buffer_mapping == NULL || list_empty(&mapping->private_list)) 5681da177e4SLinus Torvalds return 0; 5691da177e4SLinus Torvalds 5701da177e4SLinus Torvalds return fsync_buffers_list(&buffer_mapping->private_lock, 5711da177e4SLinus Torvalds &mapping->private_list); 5721da177e4SLinus Torvalds } 5731da177e4SLinus Torvalds EXPORT_SYMBOL(sync_mapping_buffers); 5741da177e4SLinus Torvalds 5751da177e4SLinus Torvalds /* 5761da177e4SLinus Torvalds * Called when we've recently written block `bblock', and it is known that 5771da177e4SLinus Torvalds * `bblock' was for a buffer_boundary() buffer. This means that the block at 5781da177e4SLinus Torvalds * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's 5791da177e4SLinus Torvalds * dirty, schedule it for IO. So that indirects merge nicely with their data. 5801da177e4SLinus Torvalds */ 5811da177e4SLinus Torvalds void write_boundary_block(struct block_device *bdev, 5821da177e4SLinus Torvalds sector_t bblock, unsigned blocksize) 5831da177e4SLinus Torvalds { 5841da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); 5851da177e4SLinus Torvalds if (bh) { 5861da177e4SLinus Torvalds if (buffer_dirty(bh)) 5871da177e4SLinus Torvalds ll_rw_block(WRITE, 1, &bh); 5881da177e4SLinus Torvalds put_bh(bh); 5891da177e4SLinus Torvalds } 5901da177e4SLinus Torvalds } 5911da177e4SLinus Torvalds 5921da177e4SLinus Torvalds void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) 5931da177e4SLinus Torvalds { 5941da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 5951da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 5961da177e4SLinus Torvalds 5971da177e4SLinus Torvalds mark_buffer_dirty(bh); 598252aa6f5SRafael Aquini if (!mapping->private_data) { 599252aa6f5SRafael Aquini mapping->private_data = buffer_mapping; 6001da177e4SLinus Torvalds } else { 601252aa6f5SRafael Aquini BUG_ON(mapping->private_data != buffer_mapping); 6021da177e4SLinus Torvalds } 603535ee2fbSJan Kara if (!bh->b_assoc_map) { 6041da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 6051da177e4SLinus Torvalds list_move_tail(&bh->b_assoc_buffers, 6061da177e4SLinus Torvalds &mapping->private_list); 60758ff407bSJan Kara bh->b_assoc_map = mapping; 6081da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 6091da177e4SLinus Torvalds } 6101da177e4SLinus Torvalds } 6111da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_dirty_inode); 6121da177e4SLinus Torvalds 6131da177e4SLinus Torvalds /* 614787d2214SNick Piggin * Mark the page dirty, and set it dirty in the radix tree, and mark the inode 615787d2214SNick Piggin * dirty. 616787d2214SNick Piggin * 617787d2214SNick Piggin * If warn is true, then emit a warning if the page is not uptodate and has 618787d2214SNick Piggin * not been truncated. 619787d2214SNick Piggin */ 620a8e7d49aSLinus Torvalds static void __set_page_dirty(struct page *page, 621787d2214SNick Piggin struct address_space *mapping, int warn) 622787d2214SNick Piggin { 62319fd6231SNick Piggin spin_lock_irq(&mapping->tree_lock); 624787d2214SNick Piggin if (page->mapping) { /* Race with truncate? */ 625787d2214SNick Piggin WARN_ON_ONCE(warn && !PageUptodate(page)); 626e3a7cca1SEdward Shishkin account_page_dirtied(page, mapping); 627787d2214SNick Piggin radix_tree_tag_set(&mapping->page_tree, 628787d2214SNick Piggin page_index(page), PAGECACHE_TAG_DIRTY); 629787d2214SNick Piggin } 63019fd6231SNick Piggin spin_unlock_irq(&mapping->tree_lock); 631787d2214SNick Piggin __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 632787d2214SNick Piggin } 633787d2214SNick Piggin 634787d2214SNick Piggin /* 6351da177e4SLinus Torvalds * Add a page to the dirty page list. 6361da177e4SLinus Torvalds * 6371da177e4SLinus Torvalds * It is a sad fact of life that this function is called from several places 6381da177e4SLinus Torvalds * deeply under spinlocking. It may not sleep. 6391da177e4SLinus Torvalds * 6401da177e4SLinus Torvalds * If the page has buffers, the uptodate buffers are set dirty, to preserve 6411da177e4SLinus Torvalds * dirty-state coherency between the page and the buffers. It the page does 6421da177e4SLinus Torvalds * not have buffers then when they are later attached they will all be set 6431da177e4SLinus Torvalds * dirty. 6441da177e4SLinus Torvalds * 6451da177e4SLinus Torvalds * The buffers are dirtied before the page is dirtied. There's a small race 6461da177e4SLinus Torvalds * window in which a writepage caller may see the page cleanness but not the 6471da177e4SLinus Torvalds * buffer dirtiness. That's fine. If this code were to set the page dirty 6481da177e4SLinus Torvalds * before the buffers, a concurrent writepage caller could clear the page dirty 6491da177e4SLinus Torvalds * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean 6501da177e4SLinus Torvalds * page on the dirty page list. 6511da177e4SLinus Torvalds * 6521da177e4SLinus Torvalds * We use private_lock to lock against try_to_free_buffers while using the 6531da177e4SLinus Torvalds * page's buffer list. Also use this to protect against clean buffers being 6541da177e4SLinus Torvalds * added to the page after it was set dirty. 6551da177e4SLinus Torvalds * 6561da177e4SLinus Torvalds * FIXME: may need to call ->reservepage here as well. That's rather up to the 6571da177e4SLinus Torvalds * address_space though. 6581da177e4SLinus Torvalds */ 6591da177e4SLinus Torvalds int __set_page_dirty_buffers(struct page *page) 6601da177e4SLinus Torvalds { 661a8e7d49aSLinus Torvalds int newly_dirty; 662787d2214SNick Piggin struct address_space *mapping = page_mapping(page); 663ebf7a227SNick Piggin 664ebf7a227SNick Piggin if (unlikely(!mapping)) 665ebf7a227SNick Piggin return !TestSetPageDirty(page); 6661da177e4SLinus Torvalds 6671da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 6681da177e4SLinus Torvalds if (page_has_buffers(page)) { 6691da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 6701da177e4SLinus Torvalds struct buffer_head *bh = head; 6711da177e4SLinus Torvalds 6721da177e4SLinus Torvalds do { 6731da177e4SLinus Torvalds set_buffer_dirty(bh); 6741da177e4SLinus Torvalds bh = bh->b_this_page; 6751da177e4SLinus Torvalds } while (bh != head); 6761da177e4SLinus Torvalds } 677a8e7d49aSLinus Torvalds newly_dirty = !TestSetPageDirty(page); 6781da177e4SLinus Torvalds spin_unlock(&mapping->private_lock); 6791da177e4SLinus Torvalds 680a8e7d49aSLinus Torvalds if (newly_dirty) 681a8e7d49aSLinus Torvalds __set_page_dirty(page, mapping, 1); 682a8e7d49aSLinus Torvalds return newly_dirty; 6831da177e4SLinus Torvalds } 6841da177e4SLinus Torvalds EXPORT_SYMBOL(__set_page_dirty_buffers); 6851da177e4SLinus Torvalds 6861da177e4SLinus Torvalds /* 6871da177e4SLinus Torvalds * Write out and wait upon a list of buffers. 6881da177e4SLinus Torvalds * 6891da177e4SLinus Torvalds * We have conflicting pressures: we want to make sure that all 6901da177e4SLinus Torvalds * initially dirty buffers get waited on, but that any subsequently 6911da177e4SLinus Torvalds * dirtied buffers don't. After all, we don't want fsync to last 6921da177e4SLinus Torvalds * forever if somebody is actively writing to the file. 6931da177e4SLinus Torvalds * 6941da177e4SLinus Torvalds * Do this in two main stages: first we copy dirty buffers to a 6951da177e4SLinus Torvalds * temporary inode list, queueing the writes as we go. Then we clean 6961da177e4SLinus Torvalds * up, waiting for those writes to complete. 6971da177e4SLinus Torvalds * 6981da177e4SLinus Torvalds * During this second stage, any subsequent updates to the file may end 6991da177e4SLinus Torvalds * up refiling the buffer on the original inode's dirty list again, so 7001da177e4SLinus Torvalds * there is a chance we will end up with a buffer queued for write but 7011da177e4SLinus Torvalds * not yet completed on that list. So, as a final cleanup we go through 7021da177e4SLinus Torvalds * the osync code to catch these locked, dirty buffers without requeuing 7031da177e4SLinus Torvalds * any newly dirty buffers for write. 7041da177e4SLinus Torvalds */ 7051da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) 7061da177e4SLinus Torvalds { 7071da177e4SLinus Torvalds struct buffer_head *bh; 7081da177e4SLinus Torvalds struct list_head tmp; 7097eaceaccSJens Axboe struct address_space *mapping; 7101da177e4SLinus Torvalds int err = 0, err2; 7114ee2491eSJens Axboe struct blk_plug plug; 7121da177e4SLinus Torvalds 7131da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 7144ee2491eSJens Axboe blk_start_plug(&plug); 7151da177e4SLinus Torvalds 7161da177e4SLinus Torvalds spin_lock(lock); 7171da177e4SLinus Torvalds while (!list_empty(list)) { 7181da177e4SLinus Torvalds bh = BH_ENTRY(list->next); 719535ee2fbSJan Kara mapping = bh->b_assoc_map; 72058ff407bSJan Kara __remove_assoc_queue(bh); 721535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 722535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 723535ee2fbSJan Kara smp_mb(); 7241da177e4SLinus Torvalds if (buffer_dirty(bh) || buffer_locked(bh)) { 7251da177e4SLinus Torvalds list_add(&bh->b_assoc_buffers, &tmp); 726535ee2fbSJan Kara bh->b_assoc_map = mapping; 7271da177e4SLinus Torvalds if (buffer_dirty(bh)) { 7281da177e4SLinus Torvalds get_bh(bh); 7291da177e4SLinus Torvalds spin_unlock(lock); 7301da177e4SLinus Torvalds /* 7311da177e4SLinus Torvalds * Ensure any pending I/O completes so that 7329cb569d6SChristoph Hellwig * write_dirty_buffer() actually writes the 7339cb569d6SChristoph Hellwig * current contents - it is a noop if I/O is 7349cb569d6SChristoph Hellwig * still in flight on potentially older 7359cb569d6SChristoph Hellwig * contents. 7361da177e4SLinus Torvalds */ 737721a9602SJens Axboe write_dirty_buffer(bh, WRITE_SYNC); 7389cf6b720SJens Axboe 7399cf6b720SJens Axboe /* 7409cf6b720SJens Axboe * Kick off IO for the previous mapping. Note 7419cf6b720SJens Axboe * that we will not run the very last mapping, 7429cf6b720SJens Axboe * wait_on_buffer() will do that for us 7439cf6b720SJens Axboe * through sync_buffer(). 7449cf6b720SJens Axboe */ 7451da177e4SLinus Torvalds brelse(bh); 7461da177e4SLinus Torvalds spin_lock(lock); 7471da177e4SLinus Torvalds } 7481da177e4SLinus Torvalds } 7491da177e4SLinus Torvalds } 7501da177e4SLinus Torvalds 7514ee2491eSJens Axboe spin_unlock(lock); 7524ee2491eSJens Axboe blk_finish_plug(&plug); 7534ee2491eSJens Axboe spin_lock(lock); 7544ee2491eSJens Axboe 7551da177e4SLinus Torvalds while (!list_empty(&tmp)) { 7561da177e4SLinus Torvalds bh = BH_ENTRY(tmp.prev); 7571da177e4SLinus Torvalds get_bh(bh); 758535ee2fbSJan Kara mapping = bh->b_assoc_map; 759535ee2fbSJan Kara __remove_assoc_queue(bh); 760535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 761535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 762535ee2fbSJan Kara smp_mb(); 763535ee2fbSJan Kara if (buffer_dirty(bh)) { 764535ee2fbSJan Kara list_add(&bh->b_assoc_buffers, 765e3892296SJan Kara &mapping->private_list); 766535ee2fbSJan Kara bh->b_assoc_map = mapping; 767535ee2fbSJan Kara } 7681da177e4SLinus Torvalds spin_unlock(lock); 7691da177e4SLinus Torvalds wait_on_buffer(bh); 7701da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 7711da177e4SLinus Torvalds err = -EIO; 7721da177e4SLinus Torvalds brelse(bh); 7731da177e4SLinus Torvalds spin_lock(lock); 7741da177e4SLinus Torvalds } 7751da177e4SLinus Torvalds 7761da177e4SLinus Torvalds spin_unlock(lock); 7771da177e4SLinus Torvalds err2 = osync_buffers_list(lock, list); 7781da177e4SLinus Torvalds if (err) 7791da177e4SLinus Torvalds return err; 7801da177e4SLinus Torvalds else 7811da177e4SLinus Torvalds return err2; 7821da177e4SLinus Torvalds } 7831da177e4SLinus Torvalds 7841da177e4SLinus Torvalds /* 7851da177e4SLinus Torvalds * Invalidate any and all dirty buffers on a given inode. We are 7861da177e4SLinus Torvalds * probably unmounting the fs, but that doesn't mean we have already 7871da177e4SLinus Torvalds * done a sync(). Just drop the buffers from the inode list. 7881da177e4SLinus Torvalds * 7891da177e4SLinus Torvalds * NOTE: we take the inode's blockdev's mapping's private_lock. Which 7901da177e4SLinus Torvalds * assumes that all the buffers are against the blockdev. Not true 7911da177e4SLinus Torvalds * for reiserfs. 7921da177e4SLinus Torvalds */ 7931da177e4SLinus Torvalds void invalidate_inode_buffers(struct inode *inode) 7941da177e4SLinus Torvalds { 7951da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 7961da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 7971da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 798252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 7991da177e4SLinus Torvalds 8001da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8011da177e4SLinus Torvalds while (!list_empty(list)) 8021da177e4SLinus Torvalds __remove_assoc_queue(BH_ENTRY(list->next)); 8031da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8041da177e4SLinus Torvalds } 8051da177e4SLinus Torvalds } 80652b19ac9SJan Kara EXPORT_SYMBOL(invalidate_inode_buffers); 8071da177e4SLinus Torvalds 8081da177e4SLinus Torvalds /* 8091da177e4SLinus Torvalds * Remove any clean buffers from the inode's buffer list. This is called 8101da177e4SLinus Torvalds * when we're trying to free the inode itself. Those buffers can pin it. 8111da177e4SLinus Torvalds * 8121da177e4SLinus Torvalds * Returns true if all buffers were removed. 8131da177e4SLinus Torvalds */ 8141da177e4SLinus Torvalds int remove_inode_buffers(struct inode *inode) 8151da177e4SLinus Torvalds { 8161da177e4SLinus Torvalds int ret = 1; 8171da177e4SLinus Torvalds 8181da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8191da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 8201da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 821252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 8221da177e4SLinus Torvalds 8231da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8241da177e4SLinus Torvalds while (!list_empty(list)) { 8251da177e4SLinus Torvalds struct buffer_head *bh = BH_ENTRY(list->next); 8261da177e4SLinus Torvalds if (buffer_dirty(bh)) { 8271da177e4SLinus Torvalds ret = 0; 8281da177e4SLinus Torvalds break; 8291da177e4SLinus Torvalds } 8301da177e4SLinus Torvalds __remove_assoc_queue(bh); 8311da177e4SLinus Torvalds } 8321da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8331da177e4SLinus Torvalds } 8341da177e4SLinus Torvalds return ret; 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 8371da177e4SLinus Torvalds /* 8381da177e4SLinus Torvalds * Create the appropriate buffers when given a page for data area and 8391da177e4SLinus Torvalds * the size of each buffer.. Use the bh->b_this_page linked list to 8401da177e4SLinus Torvalds * follow the buffers created. Return NULL if unable to create more 8411da177e4SLinus Torvalds * buffers. 8421da177e4SLinus Torvalds * 8431da177e4SLinus Torvalds * The retry flag is used to differentiate async IO (paging, swapping) 8441da177e4SLinus Torvalds * which may not fail from ordinary buffer allocations. 8451da177e4SLinus Torvalds */ 8461da177e4SLinus Torvalds struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, 8471da177e4SLinus Torvalds int retry) 8481da177e4SLinus Torvalds { 8491da177e4SLinus Torvalds struct buffer_head *bh, *head; 8501da177e4SLinus Torvalds long offset; 8511da177e4SLinus Torvalds 8521da177e4SLinus Torvalds try_again: 8531da177e4SLinus Torvalds head = NULL; 8541da177e4SLinus Torvalds offset = PAGE_SIZE; 8551da177e4SLinus Torvalds while ((offset -= size) >= 0) { 8561da177e4SLinus Torvalds bh = alloc_buffer_head(GFP_NOFS); 8571da177e4SLinus Torvalds if (!bh) 8581da177e4SLinus Torvalds goto no_grow; 8591da177e4SLinus Torvalds 8601da177e4SLinus Torvalds bh->b_this_page = head; 8611da177e4SLinus Torvalds bh->b_blocknr = -1; 8621da177e4SLinus Torvalds head = bh; 8631da177e4SLinus Torvalds 8641da177e4SLinus Torvalds bh->b_size = size; 8651da177e4SLinus Torvalds 8661da177e4SLinus Torvalds /* Link the buffer to its page */ 8671da177e4SLinus Torvalds set_bh_page(bh, page, offset); 8681da177e4SLinus Torvalds 86901ffe339SNathan Scott init_buffer(bh, NULL, NULL); 8701da177e4SLinus Torvalds } 8711da177e4SLinus Torvalds return head; 8721da177e4SLinus Torvalds /* 8731da177e4SLinus Torvalds * In case anything failed, we just free everything we got. 8741da177e4SLinus Torvalds */ 8751da177e4SLinus Torvalds no_grow: 8761da177e4SLinus Torvalds if (head) { 8771da177e4SLinus Torvalds do { 8781da177e4SLinus Torvalds bh = head; 8791da177e4SLinus Torvalds head = head->b_this_page; 8801da177e4SLinus Torvalds free_buffer_head(bh); 8811da177e4SLinus Torvalds } while (head); 8821da177e4SLinus Torvalds } 8831da177e4SLinus Torvalds 8841da177e4SLinus Torvalds /* 8851da177e4SLinus Torvalds * Return failure for non-async IO requests. Async IO requests 8861da177e4SLinus Torvalds * are not allowed to fail, so we have to wait until buffer heads 8871da177e4SLinus Torvalds * become available. But we don't want tasks sleeping with 8881da177e4SLinus Torvalds * partially complete buffers, so all were released above. 8891da177e4SLinus Torvalds */ 8901da177e4SLinus Torvalds if (!retry) 8911da177e4SLinus Torvalds return NULL; 8921da177e4SLinus Torvalds 8931da177e4SLinus Torvalds /* We're _really_ low on memory. Now we just 8941da177e4SLinus Torvalds * wait for old buffer heads to become free due to 8951da177e4SLinus Torvalds * finishing IO. Since this is an async request and 8961da177e4SLinus Torvalds * the reserve list is empty, we're sure there are 8971da177e4SLinus Torvalds * async buffer heads in use. 8981da177e4SLinus Torvalds */ 8991da177e4SLinus Torvalds free_more_memory(); 9001da177e4SLinus Torvalds goto try_again; 9011da177e4SLinus Torvalds } 9021da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(alloc_page_buffers); 9031da177e4SLinus Torvalds 9041da177e4SLinus Torvalds static inline void 9051da177e4SLinus Torvalds link_dev_buffers(struct page *page, struct buffer_head *head) 9061da177e4SLinus Torvalds { 9071da177e4SLinus Torvalds struct buffer_head *bh, *tail; 9081da177e4SLinus Torvalds 9091da177e4SLinus Torvalds bh = head; 9101da177e4SLinus Torvalds do { 9111da177e4SLinus Torvalds tail = bh; 9121da177e4SLinus Torvalds bh = bh->b_this_page; 9131da177e4SLinus Torvalds } while (bh); 9141da177e4SLinus Torvalds tail->b_this_page = head; 9151da177e4SLinus Torvalds attach_page_buffers(page, head); 9161da177e4SLinus Torvalds } 9171da177e4SLinus Torvalds 918bbec0270SLinus Torvalds static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) 919bbec0270SLinus Torvalds { 920bbec0270SLinus Torvalds sector_t retval = ~((sector_t)0); 921bbec0270SLinus Torvalds loff_t sz = i_size_read(bdev->bd_inode); 922bbec0270SLinus Torvalds 923bbec0270SLinus Torvalds if (sz) { 924bbec0270SLinus Torvalds unsigned int sizebits = blksize_bits(size); 925bbec0270SLinus Torvalds retval = (sz >> sizebits); 926bbec0270SLinus Torvalds } 927bbec0270SLinus Torvalds return retval; 928bbec0270SLinus Torvalds } 929bbec0270SLinus Torvalds 9301da177e4SLinus Torvalds /* 9311da177e4SLinus Torvalds * Initialise the state of a blockdev page's buffers. 9321da177e4SLinus Torvalds */ 933676ce6d5SHugh Dickins static sector_t 9341da177e4SLinus Torvalds init_page_buffers(struct page *page, struct block_device *bdev, 9351da177e4SLinus Torvalds sector_t block, int size) 9361da177e4SLinus Torvalds { 9371da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 9381da177e4SLinus Torvalds struct buffer_head *bh = head; 9391da177e4SLinus Torvalds int uptodate = PageUptodate(page); 940bbec0270SLinus Torvalds sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size); 9411da177e4SLinus Torvalds 9421da177e4SLinus Torvalds do { 9431da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 9441da177e4SLinus Torvalds init_buffer(bh, NULL, NULL); 9451da177e4SLinus Torvalds bh->b_bdev = bdev; 9461da177e4SLinus Torvalds bh->b_blocknr = block; 9471da177e4SLinus Torvalds if (uptodate) 9481da177e4SLinus Torvalds set_buffer_uptodate(bh); 949080399aaSJeff Moyer if (block < end_block) 9501da177e4SLinus Torvalds set_buffer_mapped(bh); 9511da177e4SLinus Torvalds } 9521da177e4SLinus Torvalds block++; 9531da177e4SLinus Torvalds bh = bh->b_this_page; 9541da177e4SLinus Torvalds } while (bh != head); 955676ce6d5SHugh Dickins 956676ce6d5SHugh Dickins /* 957676ce6d5SHugh Dickins * Caller needs to validate requested block against end of device. 958676ce6d5SHugh Dickins */ 959676ce6d5SHugh Dickins return end_block; 9601da177e4SLinus Torvalds } 9611da177e4SLinus Torvalds 9621da177e4SLinus Torvalds /* 9631da177e4SLinus Torvalds * Create the page-cache page that contains the requested block. 9641da177e4SLinus Torvalds * 965676ce6d5SHugh Dickins * This is used purely for blockdev mappings. 9661da177e4SLinus Torvalds */ 967676ce6d5SHugh Dickins static int 9681da177e4SLinus Torvalds grow_dev_page(struct block_device *bdev, sector_t block, 969676ce6d5SHugh Dickins pgoff_t index, int size, int sizebits) 9701da177e4SLinus Torvalds { 9711da177e4SLinus Torvalds struct inode *inode = bdev->bd_inode; 9721da177e4SLinus Torvalds struct page *page; 9731da177e4SLinus Torvalds struct buffer_head *bh; 974676ce6d5SHugh Dickins sector_t end_block; 975676ce6d5SHugh Dickins int ret = 0; /* Will call free_more_memory() */ 9761da177e4SLinus Torvalds 977ea125892SChristoph Lameter page = find_or_create_page(inode->i_mapping, index, 978769848c0SMel Gorman (mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS)|__GFP_MOVABLE); 9791da177e4SLinus Torvalds if (!page) 980676ce6d5SHugh Dickins return ret; 9811da177e4SLinus Torvalds 982e827f923SEric Sesterhenn BUG_ON(!PageLocked(page)); 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds if (page_has_buffers(page)) { 9851da177e4SLinus Torvalds bh = page_buffers(page); 9861da177e4SLinus Torvalds if (bh->b_size == size) { 987676ce6d5SHugh Dickins end_block = init_page_buffers(page, bdev, 988676ce6d5SHugh Dickins index << sizebits, size); 989676ce6d5SHugh Dickins goto done; 9901da177e4SLinus Torvalds } 9911da177e4SLinus Torvalds if (!try_to_free_buffers(page)) 9921da177e4SLinus Torvalds goto failed; 9931da177e4SLinus Torvalds } 9941da177e4SLinus Torvalds 9951da177e4SLinus Torvalds /* 9961da177e4SLinus Torvalds * Allocate some buffers for this page 9971da177e4SLinus Torvalds */ 9981da177e4SLinus Torvalds bh = alloc_page_buffers(page, size, 0); 9991da177e4SLinus Torvalds if (!bh) 10001da177e4SLinus Torvalds goto failed; 10011da177e4SLinus Torvalds 10021da177e4SLinus Torvalds /* 10031da177e4SLinus Torvalds * Link the page to the buffers and initialise them. Take the 10041da177e4SLinus Torvalds * lock to be atomic wrt __find_get_block(), which does not 10051da177e4SLinus Torvalds * run under the page lock. 10061da177e4SLinus Torvalds */ 10071da177e4SLinus Torvalds spin_lock(&inode->i_mapping->private_lock); 10081da177e4SLinus Torvalds link_dev_buffers(page, bh); 1009676ce6d5SHugh Dickins end_block = init_page_buffers(page, bdev, index << sizebits, size); 10101da177e4SLinus Torvalds spin_unlock(&inode->i_mapping->private_lock); 1011676ce6d5SHugh Dickins done: 1012676ce6d5SHugh Dickins ret = (block < end_block) ? 1 : -ENXIO; 10131da177e4SLinus Torvalds failed: 10141da177e4SLinus Torvalds unlock_page(page); 10151da177e4SLinus Torvalds page_cache_release(page); 1016676ce6d5SHugh Dickins return ret; 10171da177e4SLinus Torvalds } 10181da177e4SLinus Torvalds 10191da177e4SLinus Torvalds /* 10201da177e4SLinus Torvalds * Create buffers for the specified block device block's page. If 10211da177e4SLinus Torvalds * that page was dirty, the buffers are set dirty also. 10221da177e4SLinus Torvalds */ 1023858119e1SArjan van de Ven static int 10241da177e4SLinus Torvalds grow_buffers(struct block_device *bdev, sector_t block, int size) 10251da177e4SLinus Torvalds { 10261da177e4SLinus Torvalds pgoff_t index; 10271da177e4SLinus Torvalds int sizebits; 10281da177e4SLinus Torvalds 10291da177e4SLinus Torvalds sizebits = -1; 10301da177e4SLinus Torvalds do { 10311da177e4SLinus Torvalds sizebits++; 10321da177e4SLinus Torvalds } while ((size << sizebits) < PAGE_SIZE); 10331da177e4SLinus Torvalds 10341da177e4SLinus Torvalds index = block >> sizebits; 10351da177e4SLinus Torvalds 1036e5657933SAndrew Morton /* 1037e5657933SAndrew Morton * Check for a block which wants to lie outside our maximum possible 1038e5657933SAndrew Morton * pagecache index. (this comparison is done using sector_t types). 1039e5657933SAndrew Morton */ 1040e5657933SAndrew Morton if (unlikely(index != block >> sizebits)) { 1041e5657933SAndrew Morton char b[BDEVNAME_SIZE]; 1042e5657933SAndrew Morton 1043e5657933SAndrew Morton printk(KERN_ERR "%s: requested out-of-range block %llu for " 1044e5657933SAndrew Morton "device %s\n", 10458e24eea7SHarvey Harrison __func__, (unsigned long long)block, 1046e5657933SAndrew Morton bdevname(bdev, b)); 1047e5657933SAndrew Morton return -EIO; 1048e5657933SAndrew Morton } 1049676ce6d5SHugh Dickins 10501da177e4SLinus Torvalds /* Create a page with the proper size buffers.. */ 1051676ce6d5SHugh Dickins return grow_dev_page(bdev, block, index, size, sizebits); 10521da177e4SLinus Torvalds } 10531da177e4SLinus Torvalds 105475c96f85SAdrian Bunk static struct buffer_head * 10551da177e4SLinus Torvalds __getblk_slow(struct block_device *bdev, sector_t block, int size) 10561da177e4SLinus Torvalds { 10571da177e4SLinus Torvalds /* Size must be multiple of hard sectorsize */ 1058e1defc4fSMartin K. Petersen if (unlikely(size & (bdev_logical_block_size(bdev)-1) || 10591da177e4SLinus Torvalds (size < 512 || size > PAGE_SIZE))) { 10601da177e4SLinus Torvalds printk(KERN_ERR "getblk(): invalid block size %d requested\n", 10611da177e4SLinus Torvalds size); 1062e1defc4fSMartin K. Petersen printk(KERN_ERR "logical block size: %d\n", 1063e1defc4fSMartin K. Petersen bdev_logical_block_size(bdev)); 10641da177e4SLinus Torvalds 10651da177e4SLinus Torvalds dump_stack(); 10661da177e4SLinus Torvalds return NULL; 10671da177e4SLinus Torvalds } 10681da177e4SLinus Torvalds 1069676ce6d5SHugh Dickins for (;;) { 1070676ce6d5SHugh Dickins struct buffer_head *bh; 1071676ce6d5SHugh Dickins int ret; 1072676ce6d5SHugh Dickins 10731da177e4SLinus Torvalds bh = __find_get_block(bdev, block, size); 10741da177e4SLinus Torvalds if (bh) 10751da177e4SLinus Torvalds return bh; 10761da177e4SLinus Torvalds 1077e5657933SAndrew Morton ret = grow_buffers(bdev, block, size); 1078676ce6d5SHugh Dickins if (ret < 0) 107991f68c89SJeff Moyer return NULL; 1080676ce6d5SHugh Dickins if (ret == 0) 1081676ce6d5SHugh Dickins free_more_memory(); 1082676ce6d5SHugh Dickins } 10831da177e4SLinus Torvalds } 10841da177e4SLinus Torvalds 10851da177e4SLinus Torvalds /* 10861da177e4SLinus Torvalds * The relationship between dirty buffers and dirty pages: 10871da177e4SLinus Torvalds * 10881da177e4SLinus Torvalds * Whenever a page has any dirty buffers, the page's dirty bit is set, and 10891da177e4SLinus Torvalds * the page is tagged dirty in its radix tree. 10901da177e4SLinus Torvalds * 10911da177e4SLinus Torvalds * At all times, the dirtiness of the buffers represents the dirtiness of 10921da177e4SLinus Torvalds * subsections of the page. If the page has buffers, the page dirty bit is 10931da177e4SLinus Torvalds * merely a hint about the true dirty state. 10941da177e4SLinus Torvalds * 10951da177e4SLinus Torvalds * When a page is set dirty in its entirety, all its buffers are marked dirty 10961da177e4SLinus Torvalds * (if the page has buffers). 10971da177e4SLinus Torvalds * 10981da177e4SLinus Torvalds * When a buffer is marked dirty, its page is dirtied, but the page's other 10991da177e4SLinus Torvalds * buffers are not. 11001da177e4SLinus Torvalds * 11011da177e4SLinus Torvalds * Also. When blockdev buffers are explicitly read with bread(), they 11021da177e4SLinus Torvalds * individually become uptodate. But their backing page remains not 11031da177e4SLinus Torvalds * uptodate - even if all of its buffers are uptodate. A subsequent 11041da177e4SLinus Torvalds * block_read_full_page() against that page will discover all the uptodate 11051da177e4SLinus Torvalds * buffers, will set the page uptodate and will perform no I/O. 11061da177e4SLinus Torvalds */ 11071da177e4SLinus Torvalds 11081da177e4SLinus Torvalds /** 11091da177e4SLinus Torvalds * mark_buffer_dirty - mark a buffer_head as needing writeout 111067be2dd1SMartin Waitz * @bh: the buffer_head to mark dirty 11111da177e4SLinus Torvalds * 11121da177e4SLinus Torvalds * mark_buffer_dirty() will set the dirty bit against the buffer, then set its 11131da177e4SLinus Torvalds * backing page dirty, then tag the page as dirty in its address_space's radix 11141da177e4SLinus Torvalds * tree and then attach the address_space's inode to its superblock's dirty 11151da177e4SLinus Torvalds * inode list. 11161da177e4SLinus Torvalds * 11171da177e4SLinus Torvalds * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock, 1118250df6edSDave Chinner * mapping->tree_lock and mapping->host->i_lock. 11191da177e4SLinus Torvalds */ 1120fc9b52cdSHarvey Harrison void mark_buffer_dirty(struct buffer_head *bh) 11211da177e4SLinus Torvalds { 1122787d2214SNick Piggin WARN_ON_ONCE(!buffer_uptodate(bh)); 11231be62dc1SLinus Torvalds 11245305cb83STejun Heo trace_block_dirty_buffer(bh); 11255305cb83STejun Heo 11261be62dc1SLinus Torvalds /* 11271be62dc1SLinus Torvalds * Very *carefully* optimize the it-is-already-dirty case. 11281be62dc1SLinus Torvalds * 11291be62dc1SLinus Torvalds * Don't let the final "is it dirty" escape to before we 11301be62dc1SLinus Torvalds * perhaps modified the buffer. 11311be62dc1SLinus Torvalds */ 11321be62dc1SLinus Torvalds if (buffer_dirty(bh)) { 11331be62dc1SLinus Torvalds smp_mb(); 11341be62dc1SLinus Torvalds if (buffer_dirty(bh)) 11351be62dc1SLinus Torvalds return; 11361be62dc1SLinus Torvalds } 11371be62dc1SLinus Torvalds 1138a8e7d49aSLinus Torvalds if (!test_set_buffer_dirty(bh)) { 1139a8e7d49aSLinus Torvalds struct page *page = bh->b_page; 11408e9d78edSLinus Torvalds if (!TestSetPageDirty(page)) { 11418e9d78edSLinus Torvalds struct address_space *mapping = page_mapping(page); 11428e9d78edSLinus Torvalds if (mapping) 11438e9d78edSLinus Torvalds __set_page_dirty(page, mapping, 0); 11448e9d78edSLinus Torvalds } 1145a8e7d49aSLinus Torvalds } 11461da177e4SLinus Torvalds } 11471fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(mark_buffer_dirty); 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds /* 11501da177e4SLinus Torvalds * Decrement a buffer_head's reference count. If all buffers against a page 11511da177e4SLinus Torvalds * have zero reference count, are clean and unlocked, and if the page is clean 11521da177e4SLinus Torvalds * and unlocked then try_to_free_buffers() may strip the buffers from the page 11531da177e4SLinus Torvalds * in preparation for freeing it (sometimes, rarely, buffers are removed from 11541da177e4SLinus Torvalds * a page but it ends up not being freed, and buffers may later be reattached). 11551da177e4SLinus Torvalds */ 11561da177e4SLinus Torvalds void __brelse(struct buffer_head * buf) 11571da177e4SLinus Torvalds { 11581da177e4SLinus Torvalds if (atomic_read(&buf->b_count)) { 11591da177e4SLinus Torvalds put_bh(buf); 11601da177e4SLinus Torvalds return; 11611da177e4SLinus Torvalds } 11625c752ad9SArjan van de Ven WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); 11631da177e4SLinus Torvalds } 11641fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__brelse); 11651da177e4SLinus Torvalds 11661da177e4SLinus Torvalds /* 11671da177e4SLinus Torvalds * bforget() is like brelse(), except it discards any 11681da177e4SLinus Torvalds * potentially dirty data. 11691da177e4SLinus Torvalds */ 11701da177e4SLinus Torvalds void __bforget(struct buffer_head *bh) 11711da177e4SLinus Torvalds { 11721da177e4SLinus Torvalds clear_buffer_dirty(bh); 1173535ee2fbSJan Kara if (bh->b_assoc_map) { 11741da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 11751da177e4SLinus Torvalds 11761da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 11771da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 117858ff407bSJan Kara bh->b_assoc_map = NULL; 11791da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 11801da177e4SLinus Torvalds } 11811da177e4SLinus Torvalds __brelse(bh); 11821da177e4SLinus Torvalds } 11831fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__bforget); 11841da177e4SLinus Torvalds 11851da177e4SLinus Torvalds static struct buffer_head *__bread_slow(struct buffer_head *bh) 11861da177e4SLinus Torvalds { 11871da177e4SLinus Torvalds lock_buffer(bh); 11881da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 11891da177e4SLinus Torvalds unlock_buffer(bh); 11901da177e4SLinus Torvalds return bh; 11911da177e4SLinus Torvalds } else { 11921da177e4SLinus Torvalds get_bh(bh); 11931da177e4SLinus Torvalds bh->b_end_io = end_buffer_read_sync; 11941da177e4SLinus Torvalds submit_bh(READ, bh); 11951da177e4SLinus Torvalds wait_on_buffer(bh); 11961da177e4SLinus Torvalds if (buffer_uptodate(bh)) 11971da177e4SLinus Torvalds return bh; 11981da177e4SLinus Torvalds } 11991da177e4SLinus Torvalds brelse(bh); 12001da177e4SLinus Torvalds return NULL; 12011da177e4SLinus Torvalds } 12021da177e4SLinus Torvalds 12031da177e4SLinus Torvalds /* 12041da177e4SLinus Torvalds * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). 12051da177e4SLinus Torvalds * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their 12061da177e4SLinus Torvalds * refcount elevated by one when they're in an LRU. A buffer can only appear 12071da177e4SLinus Torvalds * once in a particular CPU's LRU. A single buffer can be present in multiple 12081da177e4SLinus Torvalds * CPU's LRUs at the same time. 12091da177e4SLinus Torvalds * 12101da177e4SLinus Torvalds * This is a transparent caching front-end to sb_bread(), sb_getblk() and 12111da177e4SLinus Torvalds * sb_find_get_block(). 12121da177e4SLinus Torvalds * 12131da177e4SLinus Torvalds * The LRUs themselves only need locking against invalidate_bh_lrus. We use 12141da177e4SLinus Torvalds * a local interrupt disable for that. 12151da177e4SLinus Torvalds */ 12161da177e4SLinus Torvalds 12171da177e4SLinus Torvalds #define BH_LRU_SIZE 8 12181da177e4SLinus Torvalds 12191da177e4SLinus Torvalds struct bh_lru { 12201da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12211da177e4SLinus Torvalds }; 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; 12241da177e4SLinus Torvalds 12251da177e4SLinus Torvalds #ifdef CONFIG_SMP 12261da177e4SLinus Torvalds #define bh_lru_lock() local_irq_disable() 12271da177e4SLinus Torvalds #define bh_lru_unlock() local_irq_enable() 12281da177e4SLinus Torvalds #else 12291da177e4SLinus Torvalds #define bh_lru_lock() preempt_disable() 12301da177e4SLinus Torvalds #define bh_lru_unlock() preempt_enable() 12311da177e4SLinus Torvalds #endif 12321da177e4SLinus Torvalds 12331da177e4SLinus Torvalds static inline void check_irqs_on(void) 12341da177e4SLinus Torvalds { 12351da177e4SLinus Torvalds #ifdef irqs_disabled 12361da177e4SLinus Torvalds BUG_ON(irqs_disabled()); 12371da177e4SLinus Torvalds #endif 12381da177e4SLinus Torvalds } 12391da177e4SLinus Torvalds 12401da177e4SLinus Torvalds /* 12411da177e4SLinus Torvalds * The LRU management algorithm is dopey-but-simple. Sorry. 12421da177e4SLinus Torvalds */ 12431da177e4SLinus Torvalds static void bh_lru_install(struct buffer_head *bh) 12441da177e4SLinus Torvalds { 12451da177e4SLinus Torvalds struct buffer_head *evictee = NULL; 12461da177e4SLinus Torvalds 12471da177e4SLinus Torvalds check_irqs_on(); 12481da177e4SLinus Torvalds bh_lru_lock(); 1249c7b92516SChristoph Lameter if (__this_cpu_read(bh_lrus.bhs[0]) != bh) { 12501da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12511da177e4SLinus Torvalds int in; 12521da177e4SLinus Torvalds int out = 0; 12531da177e4SLinus Torvalds 12541da177e4SLinus Torvalds get_bh(bh); 12551da177e4SLinus Torvalds bhs[out++] = bh; 12561da177e4SLinus Torvalds for (in = 0; in < BH_LRU_SIZE; in++) { 1257c7b92516SChristoph Lameter struct buffer_head *bh2 = 1258c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[in]); 12591da177e4SLinus Torvalds 12601da177e4SLinus Torvalds if (bh2 == bh) { 12611da177e4SLinus Torvalds __brelse(bh2); 12621da177e4SLinus Torvalds } else { 12631da177e4SLinus Torvalds if (out >= BH_LRU_SIZE) { 12641da177e4SLinus Torvalds BUG_ON(evictee != NULL); 12651da177e4SLinus Torvalds evictee = bh2; 12661da177e4SLinus Torvalds } else { 12671da177e4SLinus Torvalds bhs[out++] = bh2; 12681da177e4SLinus Torvalds } 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds } 12711da177e4SLinus Torvalds while (out < BH_LRU_SIZE) 12721da177e4SLinus Torvalds bhs[out++] = NULL; 1273c7b92516SChristoph Lameter memcpy(__this_cpu_ptr(&bh_lrus.bhs), bhs, sizeof(bhs)); 12741da177e4SLinus Torvalds } 12751da177e4SLinus Torvalds bh_lru_unlock(); 12761da177e4SLinus Torvalds 12771da177e4SLinus Torvalds if (evictee) 12781da177e4SLinus Torvalds __brelse(evictee); 12791da177e4SLinus Torvalds } 12801da177e4SLinus Torvalds 12811da177e4SLinus Torvalds /* 12821da177e4SLinus Torvalds * Look up the bh in this cpu's LRU. If it's there, move it to the head. 12831da177e4SLinus Torvalds */ 1284858119e1SArjan van de Ven static struct buffer_head * 12853991d3bdSTomasz Kvarsin lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) 12861da177e4SLinus Torvalds { 12871da177e4SLinus Torvalds struct buffer_head *ret = NULL; 12883991d3bdSTomasz Kvarsin unsigned int i; 12891da177e4SLinus Torvalds 12901da177e4SLinus Torvalds check_irqs_on(); 12911da177e4SLinus Torvalds bh_lru_lock(); 12921da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 1293c7b92516SChristoph Lameter struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); 12941da177e4SLinus Torvalds 12951da177e4SLinus Torvalds if (bh && bh->b_bdev == bdev && 12961da177e4SLinus Torvalds bh->b_blocknr == block && bh->b_size == size) { 12971da177e4SLinus Torvalds if (i) { 12981da177e4SLinus Torvalds while (i) { 1299c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[i], 1300c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[i - 1])); 13011da177e4SLinus Torvalds i--; 13021da177e4SLinus Torvalds } 1303c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[0], bh); 13041da177e4SLinus Torvalds } 13051da177e4SLinus Torvalds get_bh(bh); 13061da177e4SLinus Torvalds ret = bh; 13071da177e4SLinus Torvalds break; 13081da177e4SLinus Torvalds } 13091da177e4SLinus Torvalds } 13101da177e4SLinus Torvalds bh_lru_unlock(); 13111da177e4SLinus Torvalds return ret; 13121da177e4SLinus Torvalds } 13131da177e4SLinus Torvalds 13141da177e4SLinus Torvalds /* 13151da177e4SLinus Torvalds * Perform a pagecache lookup for the matching buffer. If it's there, refresh 13161da177e4SLinus Torvalds * it in the LRU and mark it as accessed. If it is not present then return 13171da177e4SLinus Torvalds * NULL 13181da177e4SLinus Torvalds */ 13191da177e4SLinus Torvalds struct buffer_head * 13203991d3bdSTomasz Kvarsin __find_get_block(struct block_device *bdev, sector_t block, unsigned size) 13211da177e4SLinus Torvalds { 13221da177e4SLinus Torvalds struct buffer_head *bh = lookup_bh_lru(bdev, block, size); 13231da177e4SLinus Torvalds 13241da177e4SLinus Torvalds if (bh == NULL) { 1325385fd4c5SCoywolf Qi Hunt bh = __find_get_block_slow(bdev, block); 13261da177e4SLinus Torvalds if (bh) 13271da177e4SLinus Torvalds bh_lru_install(bh); 13281da177e4SLinus Torvalds } 13291da177e4SLinus Torvalds if (bh) 13301da177e4SLinus Torvalds touch_buffer(bh); 13311da177e4SLinus Torvalds return bh; 13321da177e4SLinus Torvalds } 13331da177e4SLinus Torvalds EXPORT_SYMBOL(__find_get_block); 13341da177e4SLinus Torvalds 13351da177e4SLinus Torvalds /* 13361da177e4SLinus Torvalds * __getblk will locate (and, if necessary, create) the buffer_head 13371da177e4SLinus Torvalds * which corresponds to the passed block_device, block and size. The 13381da177e4SLinus Torvalds * returned buffer has its reference count incremented. 13391da177e4SLinus Torvalds * 13401da177e4SLinus Torvalds * __getblk() will lock up the machine if grow_dev_page's try_to_free_buffers() 13411da177e4SLinus Torvalds * attempt is failing. FIXME, perhaps? 13421da177e4SLinus Torvalds */ 13431da177e4SLinus Torvalds struct buffer_head * 13443991d3bdSTomasz Kvarsin __getblk(struct block_device *bdev, sector_t block, unsigned size) 13451da177e4SLinus Torvalds { 13461da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, block, size); 13471da177e4SLinus Torvalds 13481da177e4SLinus Torvalds might_sleep(); 13491da177e4SLinus Torvalds if (bh == NULL) 13501da177e4SLinus Torvalds bh = __getblk_slow(bdev, block, size); 13511da177e4SLinus Torvalds return bh; 13521da177e4SLinus Torvalds } 13531da177e4SLinus Torvalds EXPORT_SYMBOL(__getblk); 13541da177e4SLinus Torvalds 13551da177e4SLinus Torvalds /* 13561da177e4SLinus Torvalds * Do async read-ahead on a buffer.. 13571da177e4SLinus Torvalds */ 13583991d3bdSTomasz Kvarsin void __breadahead(struct block_device *bdev, sector_t block, unsigned size) 13591da177e4SLinus Torvalds { 13601da177e4SLinus Torvalds struct buffer_head *bh = __getblk(bdev, block, size); 1361a3e713b5SAndrew Morton if (likely(bh)) { 13621da177e4SLinus Torvalds ll_rw_block(READA, 1, &bh); 13631da177e4SLinus Torvalds brelse(bh); 13641da177e4SLinus Torvalds } 1365a3e713b5SAndrew Morton } 13661da177e4SLinus Torvalds EXPORT_SYMBOL(__breadahead); 13671da177e4SLinus Torvalds 13681da177e4SLinus Torvalds /** 13691da177e4SLinus Torvalds * __bread() - reads a specified block and returns the bh 137067be2dd1SMartin Waitz * @bdev: the block_device to read from 13711da177e4SLinus Torvalds * @block: number of block 13721da177e4SLinus Torvalds * @size: size (in bytes) to read 13731da177e4SLinus Torvalds * 13741da177e4SLinus Torvalds * Reads a specified block, and returns buffer head that contains it. 13751da177e4SLinus Torvalds * It returns NULL if the block was unreadable. 13761da177e4SLinus Torvalds */ 13771da177e4SLinus Torvalds struct buffer_head * 13783991d3bdSTomasz Kvarsin __bread(struct block_device *bdev, sector_t block, unsigned size) 13791da177e4SLinus Torvalds { 13801da177e4SLinus Torvalds struct buffer_head *bh = __getblk(bdev, block, size); 13811da177e4SLinus Torvalds 1382a3e713b5SAndrew Morton if (likely(bh) && !buffer_uptodate(bh)) 13831da177e4SLinus Torvalds bh = __bread_slow(bh); 13841da177e4SLinus Torvalds return bh; 13851da177e4SLinus Torvalds } 13861da177e4SLinus Torvalds EXPORT_SYMBOL(__bread); 13871da177e4SLinus Torvalds 13881da177e4SLinus Torvalds /* 13891da177e4SLinus Torvalds * invalidate_bh_lrus() is called rarely - but not only at unmount. 13901da177e4SLinus Torvalds * This doesn't race because it runs in each cpu either in irq 13911da177e4SLinus Torvalds * or with preempt disabled. 13921da177e4SLinus Torvalds */ 13931da177e4SLinus Torvalds static void invalidate_bh_lru(void *arg) 13941da177e4SLinus Torvalds { 13951da177e4SLinus Torvalds struct bh_lru *b = &get_cpu_var(bh_lrus); 13961da177e4SLinus Torvalds int i; 13971da177e4SLinus Torvalds 13981da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 13991da177e4SLinus Torvalds brelse(b->bhs[i]); 14001da177e4SLinus Torvalds b->bhs[i] = NULL; 14011da177e4SLinus Torvalds } 14021da177e4SLinus Torvalds put_cpu_var(bh_lrus); 14031da177e4SLinus Torvalds } 14041da177e4SLinus Torvalds 140542be35d0SGilad Ben-Yossef static bool has_bh_in_lru(int cpu, void *dummy) 140642be35d0SGilad Ben-Yossef { 140742be35d0SGilad Ben-Yossef struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); 140842be35d0SGilad Ben-Yossef int i; 140942be35d0SGilad Ben-Yossef 141042be35d0SGilad Ben-Yossef for (i = 0; i < BH_LRU_SIZE; i++) { 141142be35d0SGilad Ben-Yossef if (b->bhs[i]) 141242be35d0SGilad Ben-Yossef return 1; 141342be35d0SGilad Ben-Yossef } 141442be35d0SGilad Ben-Yossef 141542be35d0SGilad Ben-Yossef return 0; 141642be35d0SGilad Ben-Yossef } 141742be35d0SGilad Ben-Yossef 1418f9a14399SPeter Zijlstra void invalidate_bh_lrus(void) 14191da177e4SLinus Torvalds { 142042be35d0SGilad Ben-Yossef on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL); 14211da177e4SLinus Torvalds } 14229db5579bSNick Piggin EXPORT_SYMBOL_GPL(invalidate_bh_lrus); 14231da177e4SLinus Torvalds 14241da177e4SLinus Torvalds void set_bh_page(struct buffer_head *bh, 14251da177e4SLinus Torvalds struct page *page, unsigned long offset) 14261da177e4SLinus Torvalds { 14271da177e4SLinus Torvalds bh->b_page = page; 1428e827f923SEric Sesterhenn BUG_ON(offset >= PAGE_SIZE); 14291da177e4SLinus Torvalds if (PageHighMem(page)) 14301da177e4SLinus Torvalds /* 14311da177e4SLinus Torvalds * This catches illegal uses and preserves the offset: 14321da177e4SLinus Torvalds */ 14331da177e4SLinus Torvalds bh->b_data = (char *)(0 + offset); 14341da177e4SLinus Torvalds else 14351da177e4SLinus Torvalds bh->b_data = page_address(page) + offset; 14361da177e4SLinus Torvalds } 14371da177e4SLinus Torvalds EXPORT_SYMBOL(set_bh_page); 14381da177e4SLinus Torvalds 14391da177e4SLinus Torvalds /* 14401da177e4SLinus Torvalds * Called when truncating a buffer on a page completely. 14411da177e4SLinus Torvalds */ 1442858119e1SArjan van de Ven static void discard_buffer(struct buffer_head * bh) 14431da177e4SLinus Torvalds { 14441da177e4SLinus Torvalds lock_buffer(bh); 14451da177e4SLinus Torvalds clear_buffer_dirty(bh); 14461da177e4SLinus Torvalds bh->b_bdev = NULL; 14471da177e4SLinus Torvalds clear_buffer_mapped(bh); 14481da177e4SLinus Torvalds clear_buffer_req(bh); 14491da177e4SLinus Torvalds clear_buffer_new(bh); 14501da177e4SLinus Torvalds clear_buffer_delay(bh); 145133a266ddSDavid Chinner clear_buffer_unwritten(bh); 14521da177e4SLinus Torvalds unlock_buffer(bh); 14531da177e4SLinus Torvalds } 14541da177e4SLinus Torvalds 14551da177e4SLinus Torvalds /** 1456814e1d25SWang Sheng-Hui * block_invalidatepage - invalidate part or all of a buffer-backed page 14571da177e4SLinus Torvalds * 14581da177e4SLinus Torvalds * @page: the page which is affected 14591da177e4SLinus Torvalds * @offset: the index of the truncation point 14601da177e4SLinus Torvalds * 14611da177e4SLinus Torvalds * block_invalidatepage() is called when all or part of the page has become 14621da177e4SLinus Torvalds * invalidated by a truncate operation. 14631da177e4SLinus Torvalds * 14641da177e4SLinus Torvalds * block_invalidatepage() does not have to release all buffers, but it must 14651da177e4SLinus Torvalds * ensure that no dirty buffer is left outside @offset and that no I/O 14661da177e4SLinus Torvalds * is underway against any of the blocks which are outside the truncation 14671da177e4SLinus Torvalds * point. Because the caller is about to free (and possibly reuse) those 14681da177e4SLinus Torvalds * blocks on-disk. 14691da177e4SLinus Torvalds */ 14702ff28e22SNeilBrown void block_invalidatepage(struct page *page, unsigned long offset) 14711da177e4SLinus Torvalds { 14721da177e4SLinus Torvalds struct buffer_head *head, *bh, *next; 14731da177e4SLinus Torvalds unsigned int curr_off = 0; 14741da177e4SLinus Torvalds 14751da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 14761da177e4SLinus Torvalds if (!page_has_buffers(page)) 14771da177e4SLinus Torvalds goto out; 14781da177e4SLinus Torvalds 14791da177e4SLinus Torvalds head = page_buffers(page); 14801da177e4SLinus Torvalds bh = head; 14811da177e4SLinus Torvalds do { 14821da177e4SLinus Torvalds unsigned int next_off = curr_off + bh->b_size; 14831da177e4SLinus Torvalds next = bh->b_this_page; 14841da177e4SLinus Torvalds 14851da177e4SLinus Torvalds /* 14861da177e4SLinus Torvalds * is this block fully invalidated? 14871da177e4SLinus Torvalds */ 14881da177e4SLinus Torvalds if (offset <= curr_off) 14891da177e4SLinus Torvalds discard_buffer(bh); 14901da177e4SLinus Torvalds curr_off = next_off; 14911da177e4SLinus Torvalds bh = next; 14921da177e4SLinus Torvalds } while (bh != head); 14931da177e4SLinus Torvalds 14941da177e4SLinus Torvalds /* 14951da177e4SLinus Torvalds * We release buffers only if the entire page is being invalidated. 14961da177e4SLinus Torvalds * The get_block cached value has been unconditionally invalidated, 14971da177e4SLinus Torvalds * so real IO is not possible anymore. 14981da177e4SLinus Torvalds */ 14991da177e4SLinus Torvalds if (offset == 0) 15002ff28e22SNeilBrown try_to_release_page(page, 0); 15011da177e4SLinus Torvalds out: 15022ff28e22SNeilBrown return; 15031da177e4SLinus Torvalds } 15041da177e4SLinus Torvalds EXPORT_SYMBOL(block_invalidatepage); 15051da177e4SLinus Torvalds 15061da177e4SLinus Torvalds /* 15071da177e4SLinus Torvalds * We attach and possibly dirty the buffers atomically wrt 15081da177e4SLinus Torvalds * __set_page_dirty_buffers() via private_lock. try_to_free_buffers 15091da177e4SLinus Torvalds * is already excluded via the page lock. 15101da177e4SLinus Torvalds */ 15111da177e4SLinus Torvalds void create_empty_buffers(struct page *page, 15121da177e4SLinus Torvalds unsigned long blocksize, unsigned long b_state) 15131da177e4SLinus Torvalds { 15141da177e4SLinus Torvalds struct buffer_head *bh, *head, *tail; 15151da177e4SLinus Torvalds 15161da177e4SLinus Torvalds head = alloc_page_buffers(page, blocksize, 1); 15171da177e4SLinus Torvalds bh = head; 15181da177e4SLinus Torvalds do { 15191da177e4SLinus Torvalds bh->b_state |= b_state; 15201da177e4SLinus Torvalds tail = bh; 15211da177e4SLinus Torvalds bh = bh->b_this_page; 15221da177e4SLinus Torvalds } while (bh); 15231da177e4SLinus Torvalds tail->b_this_page = head; 15241da177e4SLinus Torvalds 15251da177e4SLinus Torvalds spin_lock(&page->mapping->private_lock); 15261da177e4SLinus Torvalds if (PageUptodate(page) || PageDirty(page)) { 15271da177e4SLinus Torvalds bh = head; 15281da177e4SLinus Torvalds do { 15291da177e4SLinus Torvalds if (PageDirty(page)) 15301da177e4SLinus Torvalds set_buffer_dirty(bh); 15311da177e4SLinus Torvalds if (PageUptodate(page)) 15321da177e4SLinus Torvalds set_buffer_uptodate(bh); 15331da177e4SLinus Torvalds bh = bh->b_this_page; 15341da177e4SLinus Torvalds } while (bh != head); 15351da177e4SLinus Torvalds } 15361da177e4SLinus Torvalds attach_page_buffers(page, head); 15371da177e4SLinus Torvalds spin_unlock(&page->mapping->private_lock); 15381da177e4SLinus Torvalds } 15391da177e4SLinus Torvalds EXPORT_SYMBOL(create_empty_buffers); 15401da177e4SLinus Torvalds 15411da177e4SLinus Torvalds /* 15421da177e4SLinus Torvalds * We are taking a block for data and we don't want any output from any 15431da177e4SLinus Torvalds * buffer-cache aliases starting from return from that function and 15441da177e4SLinus Torvalds * until the moment when something will explicitly mark the buffer 15451da177e4SLinus Torvalds * dirty (hopefully that will not happen until we will free that block ;-) 15461da177e4SLinus Torvalds * We don't even need to mark it not-uptodate - nobody can expect 15471da177e4SLinus Torvalds * anything from a newly allocated buffer anyway. We used to used 15481da177e4SLinus Torvalds * unmap_buffer() for such invalidation, but that was wrong. We definitely 15491da177e4SLinus Torvalds * don't want to mark the alias unmapped, for example - it would confuse 15501da177e4SLinus Torvalds * anyone who might pick it with bread() afterwards... 15511da177e4SLinus Torvalds * 15521da177e4SLinus Torvalds * Also.. Note that bforget() doesn't lock the buffer. So there can 15531da177e4SLinus Torvalds * be writeout I/O going on against recently-freed buffers. We don't 15541da177e4SLinus Torvalds * wait on that I/O in bforget() - it's more efficient to wait on the I/O 15551da177e4SLinus Torvalds * only if we really need to. That happens here. 15561da177e4SLinus Torvalds */ 15571da177e4SLinus Torvalds void unmap_underlying_metadata(struct block_device *bdev, sector_t block) 15581da177e4SLinus Torvalds { 15591da177e4SLinus Torvalds struct buffer_head *old_bh; 15601da177e4SLinus Torvalds 15611da177e4SLinus Torvalds might_sleep(); 15621da177e4SLinus Torvalds 1563385fd4c5SCoywolf Qi Hunt old_bh = __find_get_block_slow(bdev, block); 15641da177e4SLinus Torvalds if (old_bh) { 15651da177e4SLinus Torvalds clear_buffer_dirty(old_bh); 15661da177e4SLinus Torvalds wait_on_buffer(old_bh); 15671da177e4SLinus Torvalds clear_buffer_req(old_bh); 15681da177e4SLinus Torvalds __brelse(old_bh); 15691da177e4SLinus Torvalds } 15701da177e4SLinus Torvalds } 15711da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_underlying_metadata); 15721da177e4SLinus Torvalds 15731da177e4SLinus Torvalds /* 157445bce8f3SLinus Torvalds * Size is a power-of-two in the range 512..PAGE_SIZE, 157545bce8f3SLinus Torvalds * and the case we care about most is PAGE_SIZE. 157645bce8f3SLinus Torvalds * 157745bce8f3SLinus Torvalds * So this *could* possibly be written with those 157845bce8f3SLinus Torvalds * constraints in mind (relevant mostly if some 157945bce8f3SLinus Torvalds * architecture has a slow bit-scan instruction) 158045bce8f3SLinus Torvalds */ 158145bce8f3SLinus Torvalds static inline int block_size_bits(unsigned int blocksize) 158245bce8f3SLinus Torvalds { 158345bce8f3SLinus Torvalds return ilog2(blocksize); 158445bce8f3SLinus Torvalds } 158545bce8f3SLinus Torvalds 158645bce8f3SLinus Torvalds static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state) 158745bce8f3SLinus Torvalds { 158845bce8f3SLinus Torvalds BUG_ON(!PageLocked(page)); 158945bce8f3SLinus Torvalds 159045bce8f3SLinus Torvalds if (!page_has_buffers(page)) 159145bce8f3SLinus Torvalds create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state); 159245bce8f3SLinus Torvalds return page_buffers(page); 159345bce8f3SLinus Torvalds } 159445bce8f3SLinus Torvalds 159545bce8f3SLinus Torvalds /* 15961da177e4SLinus Torvalds * NOTE! All mapped/uptodate combinations are valid: 15971da177e4SLinus Torvalds * 15981da177e4SLinus Torvalds * Mapped Uptodate Meaning 15991da177e4SLinus Torvalds * 16001da177e4SLinus Torvalds * No No "unknown" - must do get_block() 16011da177e4SLinus Torvalds * No Yes "hole" - zero-filled 16021da177e4SLinus Torvalds * Yes No "allocated" - allocated on disk, not read in 16031da177e4SLinus Torvalds * Yes Yes "valid" - allocated and up-to-date in memory. 16041da177e4SLinus Torvalds * 16051da177e4SLinus Torvalds * "Dirty" is valid only with the last case (mapped+uptodate). 16061da177e4SLinus Torvalds */ 16071da177e4SLinus Torvalds 16081da177e4SLinus Torvalds /* 16091da177e4SLinus Torvalds * While block_write_full_page is writing back the dirty buffers under 16101da177e4SLinus Torvalds * the page lock, whoever dirtied the buffers may decide to clean them 16111da177e4SLinus Torvalds * again at any time. We handle that by only looking at the buffer 16121da177e4SLinus Torvalds * state inside lock_buffer(). 16131da177e4SLinus Torvalds * 16141da177e4SLinus Torvalds * If block_write_full_page() is called for regular writeback 16151da177e4SLinus Torvalds * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a 16161da177e4SLinus Torvalds * locked buffer. This only can happen if someone has written the buffer 16171da177e4SLinus Torvalds * directly, with submit_bh(). At the address_space level PageWriteback 16181da177e4SLinus Torvalds * prevents this contention from occurring. 16196e34eeddSTheodore Ts'o * 16206e34eeddSTheodore Ts'o * If block_write_full_page() is called with wbc->sync_mode == 1621721a9602SJens Axboe * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this 1622721a9602SJens Axboe * causes the writes to be flagged as synchronous writes. 16231da177e4SLinus Torvalds */ 16241da177e4SLinus Torvalds static int __block_write_full_page(struct inode *inode, struct page *page, 162535c80d5fSChris Mason get_block_t *get_block, struct writeback_control *wbc, 162635c80d5fSChris Mason bh_end_io_t *handler) 16271da177e4SLinus Torvalds { 16281da177e4SLinus Torvalds int err; 16291da177e4SLinus Torvalds sector_t block; 16301da177e4SLinus Torvalds sector_t last_block; 1631f0fbd5fcSAndrew Morton struct buffer_head *bh, *head; 163245bce8f3SLinus Torvalds unsigned int blocksize, bbits; 16331da177e4SLinus Torvalds int nr_underway = 0; 16346e34eeddSTheodore Ts'o int write_op = (wbc->sync_mode == WB_SYNC_ALL ? 1635721a9602SJens Axboe WRITE_SYNC : WRITE); 16361da177e4SLinus Torvalds 163745bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 16381da177e4SLinus Torvalds (1 << BH_Dirty)|(1 << BH_Uptodate)); 16391da177e4SLinus Torvalds 16401da177e4SLinus Torvalds /* 16411da177e4SLinus Torvalds * Be very careful. We have no exclusion from __set_page_dirty_buffers 16421da177e4SLinus Torvalds * here, and the (potentially unmapped) buffers may become dirty at 16431da177e4SLinus Torvalds * any time. If a buffer becomes dirty here after we've inspected it 16441da177e4SLinus Torvalds * then we just miss that fact, and the page stays dirty. 16451da177e4SLinus Torvalds * 16461da177e4SLinus Torvalds * Buffers outside i_size may be dirtied by __set_page_dirty_buffers; 16471da177e4SLinus Torvalds * handle that here by just cleaning them. 16481da177e4SLinus Torvalds */ 16491da177e4SLinus Torvalds 16501da177e4SLinus Torvalds bh = head; 165145bce8f3SLinus Torvalds blocksize = bh->b_size; 165245bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 165345bce8f3SLinus Torvalds 165445bce8f3SLinus Torvalds block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); 165545bce8f3SLinus Torvalds last_block = (i_size_read(inode) - 1) >> bbits; 16561da177e4SLinus Torvalds 16571da177e4SLinus Torvalds /* 16581da177e4SLinus Torvalds * Get all the dirty buffers mapped to disk addresses and 16591da177e4SLinus Torvalds * handle any aliases from the underlying blockdev's mapping. 16601da177e4SLinus Torvalds */ 16611da177e4SLinus Torvalds do { 16621da177e4SLinus Torvalds if (block > last_block) { 16631da177e4SLinus Torvalds /* 16641da177e4SLinus Torvalds * mapped buffers outside i_size will occur, because 16651da177e4SLinus Torvalds * this page can be outside i_size when there is a 16661da177e4SLinus Torvalds * truncate in progress. 16671da177e4SLinus Torvalds */ 16681da177e4SLinus Torvalds /* 16691da177e4SLinus Torvalds * The buffer was zeroed by block_write_full_page() 16701da177e4SLinus Torvalds */ 16711da177e4SLinus Torvalds clear_buffer_dirty(bh); 16721da177e4SLinus Torvalds set_buffer_uptodate(bh); 167329a814d2SAlex Tomas } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && 167429a814d2SAlex Tomas buffer_dirty(bh)) { 1675b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 16761da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 16771da177e4SLinus Torvalds if (err) 16781da177e4SLinus Torvalds goto recover; 167929a814d2SAlex Tomas clear_buffer_delay(bh); 16801da177e4SLinus Torvalds if (buffer_new(bh)) { 16811da177e4SLinus Torvalds /* blockdev mappings never come here */ 16821da177e4SLinus Torvalds clear_buffer_new(bh); 16831da177e4SLinus Torvalds unmap_underlying_metadata(bh->b_bdev, 16841da177e4SLinus Torvalds bh->b_blocknr); 16851da177e4SLinus Torvalds } 16861da177e4SLinus Torvalds } 16871da177e4SLinus Torvalds bh = bh->b_this_page; 16881da177e4SLinus Torvalds block++; 16891da177e4SLinus Torvalds } while (bh != head); 16901da177e4SLinus Torvalds 16911da177e4SLinus Torvalds do { 16921da177e4SLinus Torvalds if (!buffer_mapped(bh)) 16931da177e4SLinus Torvalds continue; 16941da177e4SLinus Torvalds /* 16951da177e4SLinus Torvalds * If it's a fully non-blocking write attempt and we cannot 16961da177e4SLinus Torvalds * lock the buffer then redirty the page. Note that this can 16975b0830cbSJens Axboe * potentially cause a busy-wait loop from writeback threads 16985b0830cbSJens Axboe * and kswapd activity, but those code paths have their own 16995b0830cbSJens Axboe * higher-level throttling. 17001da177e4SLinus Torvalds */ 17011b430beeSWu Fengguang if (wbc->sync_mode != WB_SYNC_NONE) { 17021da177e4SLinus Torvalds lock_buffer(bh); 1703ca5de404SNick Piggin } else if (!trylock_buffer(bh)) { 17041da177e4SLinus Torvalds redirty_page_for_writepage(wbc, page); 17051da177e4SLinus Torvalds continue; 17061da177e4SLinus Torvalds } 17071da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 170835c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 17091da177e4SLinus Torvalds } else { 17101da177e4SLinus Torvalds unlock_buffer(bh); 17111da177e4SLinus Torvalds } 17121da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 17131da177e4SLinus Torvalds 17141da177e4SLinus Torvalds /* 17151da177e4SLinus Torvalds * The page and its buffers are protected by PageWriteback(), so we can 17161da177e4SLinus Torvalds * drop the bh refcounts early. 17171da177e4SLinus Torvalds */ 17181da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 17191da177e4SLinus Torvalds set_page_writeback(page); 17201da177e4SLinus Torvalds 17211da177e4SLinus Torvalds do { 17221da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 17231da177e4SLinus Torvalds if (buffer_async_write(bh)) { 1724a64c8610STheodore Ts'o submit_bh(write_op, bh); 17251da177e4SLinus Torvalds nr_underway++; 1726ad576e63SNick Piggin } 17271da177e4SLinus Torvalds bh = next; 17281da177e4SLinus Torvalds } while (bh != head); 172905937baaSAndrew Morton unlock_page(page); 17301da177e4SLinus Torvalds 17311da177e4SLinus Torvalds err = 0; 17321da177e4SLinus Torvalds done: 17331da177e4SLinus Torvalds if (nr_underway == 0) { 17341da177e4SLinus Torvalds /* 17351da177e4SLinus Torvalds * The page was marked dirty, but the buffers were 17361da177e4SLinus Torvalds * clean. Someone wrote them back by hand with 17371da177e4SLinus Torvalds * ll_rw_block/submit_bh. A rare case. 17381da177e4SLinus Torvalds */ 17391da177e4SLinus Torvalds end_page_writeback(page); 17403d67f2d7SNick Piggin 17411da177e4SLinus Torvalds /* 17421da177e4SLinus Torvalds * The page and buffer_heads can be released at any time from 17431da177e4SLinus Torvalds * here on. 17441da177e4SLinus Torvalds */ 17451da177e4SLinus Torvalds } 17461da177e4SLinus Torvalds return err; 17471da177e4SLinus Torvalds 17481da177e4SLinus Torvalds recover: 17491da177e4SLinus Torvalds /* 17501da177e4SLinus Torvalds * ENOSPC, or some other error. We may already have added some 17511da177e4SLinus Torvalds * blocks to the file, so we need to write these out to avoid 17521da177e4SLinus Torvalds * exposing stale data. 17531da177e4SLinus Torvalds * The page is currently locked and not marked for writeback 17541da177e4SLinus Torvalds */ 17551da177e4SLinus Torvalds bh = head; 17561da177e4SLinus Torvalds /* Recovery: lock and submit the mapped buffers */ 17571da177e4SLinus Torvalds do { 175829a814d2SAlex Tomas if (buffer_mapped(bh) && buffer_dirty(bh) && 175929a814d2SAlex Tomas !buffer_delay(bh)) { 17601da177e4SLinus Torvalds lock_buffer(bh); 176135c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 17621da177e4SLinus Torvalds } else { 17631da177e4SLinus Torvalds /* 17641da177e4SLinus Torvalds * The buffer may have been set dirty during 17651da177e4SLinus Torvalds * attachment to a dirty page. 17661da177e4SLinus Torvalds */ 17671da177e4SLinus Torvalds clear_buffer_dirty(bh); 17681da177e4SLinus Torvalds } 17691da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 17701da177e4SLinus Torvalds SetPageError(page); 17711da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 17727e4c3690SAndrew Morton mapping_set_error(page->mapping, err); 17731da177e4SLinus Torvalds set_page_writeback(page); 17741da177e4SLinus Torvalds do { 17751da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 17761da177e4SLinus Torvalds if (buffer_async_write(bh)) { 17771da177e4SLinus Torvalds clear_buffer_dirty(bh); 1778a64c8610STheodore Ts'o submit_bh(write_op, bh); 17791da177e4SLinus Torvalds nr_underway++; 1780ad576e63SNick Piggin } 17811da177e4SLinus Torvalds bh = next; 17821da177e4SLinus Torvalds } while (bh != head); 1783ffda9d30SNick Piggin unlock_page(page); 17841da177e4SLinus Torvalds goto done; 17851da177e4SLinus Torvalds } 17861da177e4SLinus Torvalds 1787afddba49SNick Piggin /* 1788afddba49SNick Piggin * If a page has any new buffers, zero them out here, and mark them uptodate 1789afddba49SNick Piggin * and dirty so they'll be written out (in order to prevent uninitialised 1790afddba49SNick Piggin * block data from leaking). And clear the new bit. 1791afddba49SNick Piggin */ 1792afddba49SNick Piggin void page_zero_new_buffers(struct page *page, unsigned from, unsigned to) 1793afddba49SNick Piggin { 1794afddba49SNick Piggin unsigned int block_start, block_end; 1795afddba49SNick Piggin struct buffer_head *head, *bh; 1796afddba49SNick Piggin 1797afddba49SNick Piggin BUG_ON(!PageLocked(page)); 1798afddba49SNick Piggin if (!page_has_buffers(page)) 1799afddba49SNick Piggin return; 1800afddba49SNick Piggin 1801afddba49SNick Piggin bh = head = page_buffers(page); 1802afddba49SNick Piggin block_start = 0; 1803afddba49SNick Piggin do { 1804afddba49SNick Piggin block_end = block_start + bh->b_size; 1805afddba49SNick Piggin 1806afddba49SNick Piggin if (buffer_new(bh)) { 1807afddba49SNick Piggin if (block_end > from && block_start < to) { 1808afddba49SNick Piggin if (!PageUptodate(page)) { 1809afddba49SNick Piggin unsigned start, size; 1810afddba49SNick Piggin 1811afddba49SNick Piggin start = max(from, block_start); 1812afddba49SNick Piggin size = min(to, block_end) - start; 1813afddba49SNick Piggin 1814eebd2aa3SChristoph Lameter zero_user(page, start, size); 1815afddba49SNick Piggin set_buffer_uptodate(bh); 1816afddba49SNick Piggin } 1817afddba49SNick Piggin 1818afddba49SNick Piggin clear_buffer_new(bh); 1819afddba49SNick Piggin mark_buffer_dirty(bh); 1820afddba49SNick Piggin } 1821afddba49SNick Piggin } 1822afddba49SNick Piggin 1823afddba49SNick Piggin block_start = block_end; 1824afddba49SNick Piggin bh = bh->b_this_page; 1825afddba49SNick Piggin } while (bh != head); 1826afddba49SNick Piggin } 1827afddba49SNick Piggin EXPORT_SYMBOL(page_zero_new_buffers); 1828afddba49SNick Piggin 1829ebdec241SChristoph Hellwig int __block_write_begin(struct page *page, loff_t pos, unsigned len, 18306e1db88dSChristoph Hellwig get_block_t *get_block) 18311da177e4SLinus Torvalds { 1832ebdec241SChristoph Hellwig unsigned from = pos & (PAGE_CACHE_SIZE - 1); 1833ebdec241SChristoph Hellwig unsigned to = from + len; 18346e1db88dSChristoph Hellwig struct inode *inode = page->mapping->host; 18351da177e4SLinus Torvalds unsigned block_start, block_end; 18361da177e4SLinus Torvalds sector_t block; 18371da177e4SLinus Torvalds int err = 0; 18381da177e4SLinus Torvalds unsigned blocksize, bbits; 18391da177e4SLinus Torvalds struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; 18401da177e4SLinus Torvalds 18411da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 18421da177e4SLinus Torvalds BUG_ON(from > PAGE_CACHE_SIZE); 18431da177e4SLinus Torvalds BUG_ON(to > PAGE_CACHE_SIZE); 18441da177e4SLinus Torvalds BUG_ON(from > to); 18451da177e4SLinus Torvalds 184645bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 0); 184745bce8f3SLinus Torvalds blocksize = head->b_size; 184845bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 18491da177e4SLinus Torvalds 18501da177e4SLinus Torvalds block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); 18511da177e4SLinus Torvalds 18521da177e4SLinus Torvalds for(bh = head, block_start = 0; bh != head || !block_start; 18531da177e4SLinus Torvalds block++, block_start=block_end, bh = bh->b_this_page) { 18541da177e4SLinus Torvalds block_end = block_start + blocksize; 18551da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 18561da177e4SLinus Torvalds if (PageUptodate(page)) { 18571da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 18581da177e4SLinus Torvalds set_buffer_uptodate(bh); 18591da177e4SLinus Torvalds } 18601da177e4SLinus Torvalds continue; 18611da177e4SLinus Torvalds } 18621da177e4SLinus Torvalds if (buffer_new(bh)) 18631da177e4SLinus Torvalds clear_buffer_new(bh); 18641da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 1865b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 18661da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 18671da177e4SLinus Torvalds if (err) 1868f3ddbdc6SNick Piggin break; 18691da177e4SLinus Torvalds if (buffer_new(bh)) { 18701da177e4SLinus Torvalds unmap_underlying_metadata(bh->b_bdev, 18711da177e4SLinus Torvalds bh->b_blocknr); 18721da177e4SLinus Torvalds if (PageUptodate(page)) { 1873637aff46SNick Piggin clear_buffer_new(bh); 18741da177e4SLinus Torvalds set_buffer_uptodate(bh); 1875637aff46SNick Piggin mark_buffer_dirty(bh); 18761da177e4SLinus Torvalds continue; 18771da177e4SLinus Torvalds } 1878eebd2aa3SChristoph Lameter if (block_end > to || block_start < from) 1879eebd2aa3SChristoph Lameter zero_user_segments(page, 1880eebd2aa3SChristoph Lameter to, block_end, 1881eebd2aa3SChristoph Lameter block_start, from); 18821da177e4SLinus Torvalds continue; 18831da177e4SLinus Torvalds } 18841da177e4SLinus Torvalds } 18851da177e4SLinus Torvalds if (PageUptodate(page)) { 18861da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 18871da177e4SLinus Torvalds set_buffer_uptodate(bh); 18881da177e4SLinus Torvalds continue; 18891da177e4SLinus Torvalds } 18901da177e4SLinus Torvalds if (!buffer_uptodate(bh) && !buffer_delay(bh) && 189133a266ddSDavid Chinner !buffer_unwritten(bh) && 18921da177e4SLinus Torvalds (block_start < from || block_end > to)) { 18931da177e4SLinus Torvalds ll_rw_block(READ, 1, &bh); 18941da177e4SLinus Torvalds *wait_bh++=bh; 18951da177e4SLinus Torvalds } 18961da177e4SLinus Torvalds } 18971da177e4SLinus Torvalds /* 18981da177e4SLinus Torvalds * If we issued read requests - let them complete. 18991da177e4SLinus Torvalds */ 19001da177e4SLinus Torvalds while(wait_bh > wait) { 19011da177e4SLinus Torvalds wait_on_buffer(*--wait_bh); 19021da177e4SLinus Torvalds if (!buffer_uptodate(*wait_bh)) 1903f3ddbdc6SNick Piggin err = -EIO; 19041da177e4SLinus Torvalds } 1905f9f07b6cSJan Kara if (unlikely(err)) 1906afddba49SNick Piggin page_zero_new_buffers(page, from, to); 19071da177e4SLinus Torvalds return err; 19081da177e4SLinus Torvalds } 1909ebdec241SChristoph Hellwig EXPORT_SYMBOL(__block_write_begin); 19101da177e4SLinus Torvalds 19111da177e4SLinus Torvalds static int __block_commit_write(struct inode *inode, struct page *page, 19121da177e4SLinus Torvalds unsigned from, unsigned to) 19131da177e4SLinus Torvalds { 19141da177e4SLinus Torvalds unsigned block_start, block_end; 19151da177e4SLinus Torvalds int partial = 0; 19161da177e4SLinus Torvalds unsigned blocksize; 19171da177e4SLinus Torvalds struct buffer_head *bh, *head; 19181da177e4SLinus Torvalds 191945bce8f3SLinus Torvalds bh = head = page_buffers(page); 192045bce8f3SLinus Torvalds blocksize = bh->b_size; 19211da177e4SLinus Torvalds 192245bce8f3SLinus Torvalds block_start = 0; 192345bce8f3SLinus Torvalds do { 19241da177e4SLinus Torvalds block_end = block_start + blocksize; 19251da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 19261da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 19271da177e4SLinus Torvalds partial = 1; 19281da177e4SLinus Torvalds } else { 19291da177e4SLinus Torvalds set_buffer_uptodate(bh); 19301da177e4SLinus Torvalds mark_buffer_dirty(bh); 19311da177e4SLinus Torvalds } 1932afddba49SNick Piggin clear_buffer_new(bh); 193345bce8f3SLinus Torvalds 193445bce8f3SLinus Torvalds block_start = block_end; 193545bce8f3SLinus Torvalds bh = bh->b_this_page; 193645bce8f3SLinus Torvalds } while (bh != head); 19371da177e4SLinus Torvalds 19381da177e4SLinus Torvalds /* 19391da177e4SLinus Torvalds * If this is a partial write which happened to make all buffers 19401da177e4SLinus Torvalds * uptodate then we can optimize away a bogus readpage() for 19411da177e4SLinus Torvalds * the next read(). Here we 'discover' whether the page went 19421da177e4SLinus Torvalds * uptodate as a result of this (potentially partial) write. 19431da177e4SLinus Torvalds */ 19441da177e4SLinus Torvalds if (!partial) 19451da177e4SLinus Torvalds SetPageUptodate(page); 19461da177e4SLinus Torvalds return 0; 19471da177e4SLinus Torvalds } 19481da177e4SLinus Torvalds 19491da177e4SLinus Torvalds /* 1950155130a4SChristoph Hellwig * block_write_begin takes care of the basic task of block allocation and 1951155130a4SChristoph Hellwig * bringing partial write blocks uptodate first. 1952155130a4SChristoph Hellwig * 19537bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 1954afddba49SNick Piggin */ 1955155130a4SChristoph Hellwig int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, 1956155130a4SChristoph Hellwig unsigned flags, struct page **pagep, get_block_t *get_block) 1957afddba49SNick Piggin { 19586e1db88dSChristoph Hellwig pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1959afddba49SNick Piggin struct page *page; 19606e1db88dSChristoph Hellwig int status; 1961afddba49SNick Piggin 196254566b2cSNick Piggin page = grab_cache_page_write_begin(mapping, index, flags); 19636e1db88dSChristoph Hellwig if (!page) 19646e1db88dSChristoph Hellwig return -ENOMEM; 1965afddba49SNick Piggin 19666e1db88dSChristoph Hellwig status = __block_write_begin(page, pos, len, get_block); 1967afddba49SNick Piggin if (unlikely(status)) { 1968afddba49SNick Piggin unlock_page(page); 1969afddba49SNick Piggin page_cache_release(page); 19706e1db88dSChristoph Hellwig page = NULL; 1971afddba49SNick Piggin } 1972afddba49SNick Piggin 19736e1db88dSChristoph Hellwig *pagep = page; 1974afddba49SNick Piggin return status; 1975afddba49SNick Piggin } 1976afddba49SNick Piggin EXPORT_SYMBOL(block_write_begin); 1977afddba49SNick Piggin 1978afddba49SNick Piggin int block_write_end(struct file *file, struct address_space *mapping, 1979afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 1980afddba49SNick Piggin struct page *page, void *fsdata) 1981afddba49SNick Piggin { 1982afddba49SNick Piggin struct inode *inode = mapping->host; 1983afddba49SNick Piggin unsigned start; 1984afddba49SNick Piggin 1985afddba49SNick Piggin start = pos & (PAGE_CACHE_SIZE - 1); 1986afddba49SNick Piggin 1987afddba49SNick Piggin if (unlikely(copied < len)) { 1988afddba49SNick Piggin /* 1989afddba49SNick Piggin * The buffers that were written will now be uptodate, so we 1990afddba49SNick Piggin * don't have to worry about a readpage reading them and 1991afddba49SNick Piggin * overwriting a partial write. However if we have encountered 1992afddba49SNick Piggin * a short write and only partially written into a buffer, it 1993afddba49SNick Piggin * will not be marked uptodate, so a readpage might come in and 1994afddba49SNick Piggin * destroy our partial write. 1995afddba49SNick Piggin * 1996afddba49SNick Piggin * Do the simplest thing, and just treat any short write to a 1997afddba49SNick Piggin * non uptodate page as a zero-length write, and force the 1998afddba49SNick Piggin * caller to redo the whole thing. 1999afddba49SNick Piggin */ 2000afddba49SNick Piggin if (!PageUptodate(page)) 2001afddba49SNick Piggin copied = 0; 2002afddba49SNick Piggin 2003afddba49SNick Piggin page_zero_new_buffers(page, start+copied, start+len); 2004afddba49SNick Piggin } 2005afddba49SNick Piggin flush_dcache_page(page); 2006afddba49SNick Piggin 2007afddba49SNick Piggin /* This could be a short (even 0-length) commit */ 2008afddba49SNick Piggin __block_commit_write(inode, page, start, start+copied); 2009afddba49SNick Piggin 2010afddba49SNick Piggin return copied; 2011afddba49SNick Piggin } 2012afddba49SNick Piggin EXPORT_SYMBOL(block_write_end); 2013afddba49SNick Piggin 2014afddba49SNick Piggin int generic_write_end(struct file *file, struct address_space *mapping, 2015afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2016afddba49SNick Piggin struct page *page, void *fsdata) 2017afddba49SNick Piggin { 2018afddba49SNick Piggin struct inode *inode = mapping->host; 2019c7d206b3SJan Kara int i_size_changed = 0; 2020afddba49SNick Piggin 2021afddba49SNick Piggin copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); 2022afddba49SNick Piggin 2023afddba49SNick Piggin /* 2024afddba49SNick Piggin * No need to use i_size_read() here, the i_size 2025afddba49SNick Piggin * cannot change under us because we hold i_mutex. 2026afddba49SNick Piggin * 2027afddba49SNick Piggin * But it's important to update i_size while still holding page lock: 2028afddba49SNick Piggin * page writeout could otherwise come in and zero beyond i_size. 2029afddba49SNick Piggin */ 2030afddba49SNick Piggin if (pos+copied > inode->i_size) { 2031afddba49SNick Piggin i_size_write(inode, pos+copied); 2032c7d206b3SJan Kara i_size_changed = 1; 2033afddba49SNick Piggin } 2034afddba49SNick Piggin 2035afddba49SNick Piggin unlock_page(page); 2036afddba49SNick Piggin page_cache_release(page); 2037afddba49SNick Piggin 2038c7d206b3SJan Kara /* 2039c7d206b3SJan Kara * Don't mark the inode dirty under page lock. First, it unnecessarily 2040c7d206b3SJan Kara * makes the holding time of page lock longer. Second, it forces lock 2041c7d206b3SJan Kara * ordering of page lock and transaction start for journaling 2042c7d206b3SJan Kara * filesystems. 2043c7d206b3SJan Kara */ 2044c7d206b3SJan Kara if (i_size_changed) 2045c7d206b3SJan Kara mark_inode_dirty(inode); 2046c7d206b3SJan Kara 2047afddba49SNick Piggin return copied; 2048afddba49SNick Piggin } 2049afddba49SNick Piggin EXPORT_SYMBOL(generic_write_end); 2050afddba49SNick Piggin 2051afddba49SNick Piggin /* 20528ab22b9aSHisashi Hifumi * block_is_partially_uptodate checks whether buffers within a page are 20538ab22b9aSHisashi Hifumi * uptodate or not. 20548ab22b9aSHisashi Hifumi * 20558ab22b9aSHisashi Hifumi * Returns true if all buffers which correspond to a file portion 20568ab22b9aSHisashi Hifumi * we want to read are uptodate. 20578ab22b9aSHisashi Hifumi */ 20588ab22b9aSHisashi Hifumi int block_is_partially_uptodate(struct page *page, read_descriptor_t *desc, 20598ab22b9aSHisashi Hifumi unsigned long from) 20608ab22b9aSHisashi Hifumi { 20618ab22b9aSHisashi Hifumi unsigned block_start, block_end, blocksize; 20628ab22b9aSHisashi Hifumi unsigned to; 20638ab22b9aSHisashi Hifumi struct buffer_head *bh, *head; 20648ab22b9aSHisashi Hifumi int ret = 1; 20658ab22b9aSHisashi Hifumi 20668ab22b9aSHisashi Hifumi if (!page_has_buffers(page)) 20678ab22b9aSHisashi Hifumi return 0; 20688ab22b9aSHisashi Hifumi 206945bce8f3SLinus Torvalds head = page_buffers(page); 207045bce8f3SLinus Torvalds blocksize = head->b_size; 20718ab22b9aSHisashi Hifumi to = min_t(unsigned, PAGE_CACHE_SIZE - from, desc->count); 20728ab22b9aSHisashi Hifumi to = from + to; 20738ab22b9aSHisashi Hifumi if (from < blocksize && to > PAGE_CACHE_SIZE - blocksize) 20748ab22b9aSHisashi Hifumi return 0; 20758ab22b9aSHisashi Hifumi 20768ab22b9aSHisashi Hifumi bh = head; 20778ab22b9aSHisashi Hifumi block_start = 0; 20788ab22b9aSHisashi Hifumi do { 20798ab22b9aSHisashi Hifumi block_end = block_start + blocksize; 20808ab22b9aSHisashi Hifumi if (block_end > from && block_start < to) { 20818ab22b9aSHisashi Hifumi if (!buffer_uptodate(bh)) { 20828ab22b9aSHisashi Hifumi ret = 0; 20838ab22b9aSHisashi Hifumi break; 20848ab22b9aSHisashi Hifumi } 20858ab22b9aSHisashi Hifumi if (block_end >= to) 20868ab22b9aSHisashi Hifumi break; 20878ab22b9aSHisashi Hifumi } 20888ab22b9aSHisashi Hifumi block_start = block_end; 20898ab22b9aSHisashi Hifumi bh = bh->b_this_page; 20908ab22b9aSHisashi Hifumi } while (bh != head); 20918ab22b9aSHisashi Hifumi 20928ab22b9aSHisashi Hifumi return ret; 20938ab22b9aSHisashi Hifumi } 20948ab22b9aSHisashi Hifumi EXPORT_SYMBOL(block_is_partially_uptodate); 20958ab22b9aSHisashi Hifumi 20968ab22b9aSHisashi Hifumi /* 20971da177e4SLinus Torvalds * Generic "read page" function for block devices that have the normal 20981da177e4SLinus Torvalds * get_block functionality. This is most of the block device filesystems. 20991da177e4SLinus Torvalds * Reads the page asynchronously --- the unlock_buffer() and 21001da177e4SLinus Torvalds * set/clear_buffer_uptodate() functions propagate buffer state into the 21011da177e4SLinus Torvalds * page struct once IO has completed. 21021da177e4SLinus Torvalds */ 21031da177e4SLinus Torvalds int block_read_full_page(struct page *page, get_block_t *get_block) 21041da177e4SLinus Torvalds { 21051da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 21061da177e4SLinus Torvalds sector_t iblock, lblock; 21071da177e4SLinus Torvalds struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 210845bce8f3SLinus Torvalds unsigned int blocksize, bbits; 21091da177e4SLinus Torvalds int nr, i; 21101da177e4SLinus Torvalds int fully_mapped = 1; 21111da177e4SLinus Torvalds 211245bce8f3SLinus Torvalds head = create_page_buffers(page, inode, 0); 211345bce8f3SLinus Torvalds blocksize = head->b_size; 211445bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 21151da177e4SLinus Torvalds 211645bce8f3SLinus Torvalds iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); 211745bce8f3SLinus Torvalds lblock = (i_size_read(inode)+blocksize-1) >> bbits; 21181da177e4SLinus Torvalds bh = head; 21191da177e4SLinus Torvalds nr = 0; 21201da177e4SLinus Torvalds i = 0; 21211da177e4SLinus Torvalds 21221da177e4SLinus Torvalds do { 21231da177e4SLinus Torvalds if (buffer_uptodate(bh)) 21241da177e4SLinus Torvalds continue; 21251da177e4SLinus Torvalds 21261da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2127c64610baSAndrew Morton int err = 0; 2128c64610baSAndrew Morton 21291da177e4SLinus Torvalds fully_mapped = 0; 21301da177e4SLinus Torvalds if (iblock < lblock) { 2131b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2132c64610baSAndrew Morton err = get_block(inode, iblock, bh, 0); 2133c64610baSAndrew Morton if (err) 21341da177e4SLinus Torvalds SetPageError(page); 21351da177e4SLinus Torvalds } 21361da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2137eebd2aa3SChristoph Lameter zero_user(page, i * blocksize, blocksize); 2138c64610baSAndrew Morton if (!err) 21391da177e4SLinus Torvalds set_buffer_uptodate(bh); 21401da177e4SLinus Torvalds continue; 21411da177e4SLinus Torvalds } 21421da177e4SLinus Torvalds /* 21431da177e4SLinus Torvalds * get_block() might have updated the buffer 21441da177e4SLinus Torvalds * synchronously 21451da177e4SLinus Torvalds */ 21461da177e4SLinus Torvalds if (buffer_uptodate(bh)) 21471da177e4SLinus Torvalds continue; 21481da177e4SLinus Torvalds } 21491da177e4SLinus Torvalds arr[nr++] = bh; 21501da177e4SLinus Torvalds } while (i++, iblock++, (bh = bh->b_this_page) != head); 21511da177e4SLinus Torvalds 21521da177e4SLinus Torvalds if (fully_mapped) 21531da177e4SLinus Torvalds SetPageMappedToDisk(page); 21541da177e4SLinus Torvalds 21551da177e4SLinus Torvalds if (!nr) { 21561da177e4SLinus Torvalds /* 21571da177e4SLinus Torvalds * All buffers are uptodate - we can set the page uptodate 21581da177e4SLinus Torvalds * as well. But not if get_block() returned an error. 21591da177e4SLinus Torvalds */ 21601da177e4SLinus Torvalds if (!PageError(page)) 21611da177e4SLinus Torvalds SetPageUptodate(page); 21621da177e4SLinus Torvalds unlock_page(page); 21631da177e4SLinus Torvalds return 0; 21641da177e4SLinus Torvalds } 21651da177e4SLinus Torvalds 21661da177e4SLinus Torvalds /* Stage two: lock the buffers */ 21671da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 21681da177e4SLinus Torvalds bh = arr[i]; 21691da177e4SLinus Torvalds lock_buffer(bh); 21701da177e4SLinus Torvalds mark_buffer_async_read(bh); 21711da177e4SLinus Torvalds } 21721da177e4SLinus Torvalds 21731da177e4SLinus Torvalds /* 21741da177e4SLinus Torvalds * Stage 3: start the IO. Check for uptodateness 21751da177e4SLinus Torvalds * inside the buffer lock in case another process reading 21761da177e4SLinus Torvalds * the underlying blockdev brought it uptodate (the sct fix). 21771da177e4SLinus Torvalds */ 21781da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 21791da177e4SLinus Torvalds bh = arr[i]; 21801da177e4SLinus Torvalds if (buffer_uptodate(bh)) 21811da177e4SLinus Torvalds end_buffer_async_read(bh, 1); 21821da177e4SLinus Torvalds else 21831da177e4SLinus Torvalds submit_bh(READ, bh); 21841da177e4SLinus Torvalds } 21851da177e4SLinus Torvalds return 0; 21861da177e4SLinus Torvalds } 21871fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_read_full_page); 21881da177e4SLinus Torvalds 21891da177e4SLinus Torvalds /* utility function for filesystems that need to do work on expanding 219089e10787SNick Piggin * truncates. Uses filesystem pagecache writes to allow the filesystem to 21911da177e4SLinus Torvalds * deal with the hole. 21921da177e4SLinus Torvalds */ 219389e10787SNick Piggin int generic_cont_expand_simple(struct inode *inode, loff_t size) 21941da177e4SLinus Torvalds { 21951da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 21961da177e4SLinus Torvalds struct page *page; 219789e10787SNick Piggin void *fsdata; 21981da177e4SLinus Torvalds int err; 21991da177e4SLinus Torvalds 2200c08d3b0eSnpiggin@suse.de err = inode_newsize_ok(inode, size); 2201c08d3b0eSnpiggin@suse.de if (err) 22021da177e4SLinus Torvalds goto out; 22031da177e4SLinus Torvalds 220489e10787SNick Piggin err = pagecache_write_begin(NULL, mapping, size, 0, 220589e10787SNick Piggin AOP_FLAG_UNINTERRUPTIBLE|AOP_FLAG_CONT_EXPAND, 220689e10787SNick Piggin &page, &fsdata); 220789e10787SNick Piggin if (err) 220805eb0b51SOGAWA Hirofumi goto out; 220905eb0b51SOGAWA Hirofumi 221089e10787SNick Piggin err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata); 221189e10787SNick Piggin BUG_ON(err > 0); 221205eb0b51SOGAWA Hirofumi 221305eb0b51SOGAWA Hirofumi out: 221405eb0b51SOGAWA Hirofumi return err; 221505eb0b51SOGAWA Hirofumi } 22161fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_cont_expand_simple); 221705eb0b51SOGAWA Hirofumi 2218f1e3af72SAdrian Bunk static int cont_expand_zero(struct file *file, struct address_space *mapping, 221989e10787SNick Piggin loff_t pos, loff_t *bytes) 222005eb0b51SOGAWA Hirofumi { 222189e10787SNick Piggin struct inode *inode = mapping->host; 222289e10787SNick Piggin unsigned blocksize = 1 << inode->i_blkbits; 222389e10787SNick Piggin struct page *page; 222489e10787SNick Piggin void *fsdata; 222589e10787SNick Piggin pgoff_t index, curidx; 222689e10787SNick Piggin loff_t curpos; 222789e10787SNick Piggin unsigned zerofrom, offset, len; 222889e10787SNick Piggin int err = 0; 222905eb0b51SOGAWA Hirofumi 223089e10787SNick Piggin index = pos >> PAGE_CACHE_SHIFT; 223189e10787SNick Piggin offset = pos & ~PAGE_CACHE_MASK; 223289e10787SNick Piggin 223389e10787SNick Piggin while (index > (curidx = (curpos = *bytes)>>PAGE_CACHE_SHIFT)) { 223489e10787SNick Piggin zerofrom = curpos & ~PAGE_CACHE_MASK; 223589e10787SNick Piggin if (zerofrom & (blocksize-1)) { 223689e10787SNick Piggin *bytes |= (blocksize-1); 223789e10787SNick Piggin (*bytes)++; 223889e10787SNick Piggin } 223989e10787SNick Piggin len = PAGE_CACHE_SIZE - zerofrom; 224089e10787SNick Piggin 224189e10787SNick Piggin err = pagecache_write_begin(file, mapping, curpos, len, 224289e10787SNick Piggin AOP_FLAG_UNINTERRUPTIBLE, 224389e10787SNick Piggin &page, &fsdata); 224489e10787SNick Piggin if (err) 224589e10787SNick Piggin goto out; 2246eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 224789e10787SNick Piggin err = pagecache_write_end(file, mapping, curpos, len, len, 224889e10787SNick Piggin page, fsdata); 224989e10787SNick Piggin if (err < 0) 225089e10787SNick Piggin goto out; 225189e10787SNick Piggin BUG_ON(err != len); 225289e10787SNick Piggin err = 0; 2253061e9746SOGAWA Hirofumi 2254061e9746SOGAWA Hirofumi balance_dirty_pages_ratelimited(mapping); 225589e10787SNick Piggin } 225689e10787SNick Piggin 225789e10787SNick Piggin /* page covers the boundary, find the boundary offset */ 225889e10787SNick Piggin if (index == curidx) { 225989e10787SNick Piggin zerofrom = curpos & ~PAGE_CACHE_MASK; 226089e10787SNick Piggin /* if we will expand the thing last block will be filled */ 226189e10787SNick Piggin if (offset <= zerofrom) { 226289e10787SNick Piggin goto out; 226389e10787SNick Piggin } 226489e10787SNick Piggin if (zerofrom & (blocksize-1)) { 226589e10787SNick Piggin *bytes |= (blocksize-1); 226689e10787SNick Piggin (*bytes)++; 226789e10787SNick Piggin } 226889e10787SNick Piggin len = offset - zerofrom; 226989e10787SNick Piggin 227089e10787SNick Piggin err = pagecache_write_begin(file, mapping, curpos, len, 227189e10787SNick Piggin AOP_FLAG_UNINTERRUPTIBLE, 227289e10787SNick Piggin &page, &fsdata); 227389e10787SNick Piggin if (err) 227489e10787SNick Piggin goto out; 2275eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 227689e10787SNick Piggin err = pagecache_write_end(file, mapping, curpos, len, len, 227789e10787SNick Piggin page, fsdata); 227889e10787SNick Piggin if (err < 0) 227989e10787SNick Piggin goto out; 228089e10787SNick Piggin BUG_ON(err != len); 228189e10787SNick Piggin err = 0; 228289e10787SNick Piggin } 228389e10787SNick Piggin out: 228489e10787SNick Piggin return err; 22851da177e4SLinus Torvalds } 22861da177e4SLinus Torvalds 22871da177e4SLinus Torvalds /* 22881da177e4SLinus Torvalds * For moronic filesystems that do not allow holes in file. 22891da177e4SLinus Torvalds * We may have to extend the file. 22901da177e4SLinus Torvalds */ 2291282dc178SChristoph Hellwig int cont_write_begin(struct file *file, struct address_space *mapping, 229289e10787SNick Piggin loff_t pos, unsigned len, unsigned flags, 229389e10787SNick Piggin struct page **pagep, void **fsdata, 229489e10787SNick Piggin get_block_t *get_block, loff_t *bytes) 22951da177e4SLinus Torvalds { 22961da177e4SLinus Torvalds struct inode *inode = mapping->host; 22971da177e4SLinus Torvalds unsigned blocksize = 1 << inode->i_blkbits; 229889e10787SNick Piggin unsigned zerofrom; 229989e10787SNick Piggin int err; 23001da177e4SLinus Torvalds 230189e10787SNick Piggin err = cont_expand_zero(file, mapping, pos, bytes); 230289e10787SNick Piggin if (err) 2303155130a4SChristoph Hellwig return err; 23041da177e4SLinus Torvalds 23051da177e4SLinus Torvalds zerofrom = *bytes & ~PAGE_CACHE_MASK; 230689e10787SNick Piggin if (pos+len > *bytes && zerofrom & (blocksize-1)) { 23071da177e4SLinus Torvalds *bytes |= (blocksize-1); 23081da177e4SLinus Torvalds (*bytes)++; 23091da177e4SLinus Torvalds } 23101da177e4SLinus Torvalds 2311155130a4SChristoph Hellwig return block_write_begin(mapping, pos, len, flags, pagep, get_block); 23121da177e4SLinus Torvalds } 23131fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(cont_write_begin); 23141da177e4SLinus Torvalds 23151da177e4SLinus Torvalds int block_commit_write(struct page *page, unsigned from, unsigned to) 23161da177e4SLinus Torvalds { 23171da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 23181da177e4SLinus Torvalds __block_commit_write(inode,page,from,to); 23191da177e4SLinus Torvalds return 0; 23201da177e4SLinus Torvalds } 23211fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_commit_write); 23221da177e4SLinus Torvalds 232354171690SDavid Chinner /* 232454171690SDavid Chinner * block_page_mkwrite() is not allowed to change the file size as it gets 232554171690SDavid Chinner * called from a page fault handler when a page is first dirtied. Hence we must 232654171690SDavid Chinner * be careful to check for EOF conditions here. We set the page up correctly 232754171690SDavid Chinner * for a written page which means we get ENOSPC checking when writing into 232854171690SDavid Chinner * holes and correct delalloc and unwritten extent mapping on filesystems that 232954171690SDavid Chinner * support these features. 233054171690SDavid Chinner * 233154171690SDavid Chinner * We are not allowed to take the i_mutex here so we have to play games to 233254171690SDavid Chinner * protect against truncate races as the page could now be beyond EOF. Because 23337bb46a67Snpiggin@suse.de * truncate writes the inode size before removing pages, once we have the 233454171690SDavid Chinner * page lock we can determine safely if the page is beyond EOF. If it is not 233554171690SDavid Chinner * beyond EOF, then the page is guaranteed safe against truncation until we 233654171690SDavid Chinner * unlock the page. 2337ea13a864SJan Kara * 233814da9200SJan Kara * Direct callers of this function should protect against filesystem freezing 233914da9200SJan Kara * using sb_start_write() - sb_end_write() functions. 234054171690SDavid Chinner */ 234124da4fabSJan Kara int __block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 234254171690SDavid Chinner get_block_t get_block) 234354171690SDavid Chinner { 2344c2ec175cSNick Piggin struct page *page = vmf->page; 2345496ad9aaSAl Viro struct inode *inode = file_inode(vma->vm_file); 234654171690SDavid Chinner unsigned long end; 234754171690SDavid Chinner loff_t size; 234824da4fabSJan Kara int ret; 234954171690SDavid Chinner 235054171690SDavid Chinner lock_page(page); 235154171690SDavid Chinner size = i_size_read(inode); 235254171690SDavid Chinner if ((page->mapping != inode->i_mapping) || 235318336338SNick Piggin (page_offset(page) > size)) { 235424da4fabSJan Kara /* We overload EFAULT to mean page got truncated */ 235524da4fabSJan Kara ret = -EFAULT; 235624da4fabSJan Kara goto out_unlock; 235754171690SDavid Chinner } 235854171690SDavid Chinner 235954171690SDavid Chinner /* page is wholly or partially inside EOF */ 236054171690SDavid Chinner if (((page->index + 1) << PAGE_CACHE_SHIFT) > size) 236154171690SDavid Chinner end = size & ~PAGE_CACHE_MASK; 236254171690SDavid Chinner else 236354171690SDavid Chinner end = PAGE_CACHE_SIZE; 236454171690SDavid Chinner 2365ebdec241SChristoph Hellwig ret = __block_write_begin(page, 0, end, get_block); 236654171690SDavid Chinner if (!ret) 236754171690SDavid Chinner ret = block_commit_write(page, 0, end); 236854171690SDavid Chinner 236924da4fabSJan Kara if (unlikely(ret < 0)) 237024da4fabSJan Kara goto out_unlock; 2371ea13a864SJan Kara set_page_dirty(page); 23721d1d1a76SDarrick J. Wong wait_for_stable_page(page); 237324da4fabSJan Kara return 0; 237424da4fabSJan Kara out_unlock: 2375b827e496SNick Piggin unlock_page(page); 237654171690SDavid Chinner return ret; 237754171690SDavid Chinner } 237824da4fabSJan Kara EXPORT_SYMBOL(__block_page_mkwrite); 237924da4fabSJan Kara 238024da4fabSJan Kara int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 238124da4fabSJan Kara get_block_t get_block) 238224da4fabSJan Kara { 2383ea13a864SJan Kara int ret; 2384496ad9aaSAl Viro struct super_block *sb = file_inode(vma->vm_file)->i_sb; 238524da4fabSJan Kara 238614da9200SJan Kara sb_start_pagefault(sb); 2387041bbb6dSTheodore Ts'o 2388041bbb6dSTheodore Ts'o /* 2389041bbb6dSTheodore Ts'o * Update file times before taking page lock. We may end up failing the 2390041bbb6dSTheodore Ts'o * fault so this update may be superfluous but who really cares... 2391041bbb6dSTheodore Ts'o */ 2392041bbb6dSTheodore Ts'o file_update_time(vma->vm_file); 2393041bbb6dSTheodore Ts'o 2394ea13a864SJan Kara ret = __block_page_mkwrite(vma, vmf, get_block); 239514da9200SJan Kara sb_end_pagefault(sb); 239624da4fabSJan Kara return block_page_mkwrite_return(ret); 239724da4fabSJan Kara } 23981fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_page_mkwrite); 23991da177e4SLinus Torvalds 24001da177e4SLinus Torvalds /* 240103158cd7SNick Piggin * nobh_write_begin()'s prereads are special: the buffer_heads are freed 24021da177e4SLinus Torvalds * immediately, while under the page lock. So it needs a special end_io 24031da177e4SLinus Torvalds * handler which does not touch the bh after unlocking it. 24041da177e4SLinus Torvalds */ 24051da177e4SLinus Torvalds static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate) 24061da177e4SLinus Torvalds { 240768671f35SDmitry Monakhov __end_buffer_read_notouch(bh, uptodate); 24081da177e4SLinus Torvalds } 24091da177e4SLinus Torvalds 24101da177e4SLinus Torvalds /* 241103158cd7SNick Piggin * Attach the singly-linked list of buffers created by nobh_write_begin, to 241203158cd7SNick Piggin * the page (converting it to circular linked list and taking care of page 241303158cd7SNick Piggin * dirty races). 241403158cd7SNick Piggin */ 241503158cd7SNick Piggin static void attach_nobh_buffers(struct page *page, struct buffer_head *head) 241603158cd7SNick Piggin { 241703158cd7SNick Piggin struct buffer_head *bh; 241803158cd7SNick Piggin 241903158cd7SNick Piggin BUG_ON(!PageLocked(page)); 242003158cd7SNick Piggin 242103158cd7SNick Piggin spin_lock(&page->mapping->private_lock); 242203158cd7SNick Piggin bh = head; 242303158cd7SNick Piggin do { 242403158cd7SNick Piggin if (PageDirty(page)) 242503158cd7SNick Piggin set_buffer_dirty(bh); 242603158cd7SNick Piggin if (!bh->b_this_page) 242703158cd7SNick Piggin bh->b_this_page = head; 242803158cd7SNick Piggin bh = bh->b_this_page; 242903158cd7SNick Piggin } while (bh != head); 243003158cd7SNick Piggin attach_page_buffers(page, head); 243103158cd7SNick Piggin spin_unlock(&page->mapping->private_lock); 243203158cd7SNick Piggin } 243303158cd7SNick Piggin 243403158cd7SNick Piggin /* 2435ea0f04e5SChristoph Hellwig * On entry, the page is fully not uptodate. 2436ea0f04e5SChristoph Hellwig * On exit the page is fully uptodate in the areas outside (from,to) 24377bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 24381da177e4SLinus Torvalds */ 2439ea0f04e5SChristoph Hellwig int nobh_write_begin(struct address_space *mapping, 244003158cd7SNick Piggin loff_t pos, unsigned len, unsigned flags, 244103158cd7SNick Piggin struct page **pagep, void **fsdata, 24421da177e4SLinus Torvalds get_block_t *get_block) 24431da177e4SLinus Torvalds { 244403158cd7SNick Piggin struct inode *inode = mapping->host; 24451da177e4SLinus Torvalds const unsigned blkbits = inode->i_blkbits; 24461da177e4SLinus Torvalds const unsigned blocksize = 1 << blkbits; 2447a4b0672dSNick Piggin struct buffer_head *head, *bh; 244803158cd7SNick Piggin struct page *page; 244903158cd7SNick Piggin pgoff_t index; 245003158cd7SNick Piggin unsigned from, to; 24511da177e4SLinus Torvalds unsigned block_in_page; 2452a4b0672dSNick Piggin unsigned block_start, block_end; 24531da177e4SLinus Torvalds sector_t block_in_file; 24541da177e4SLinus Torvalds int nr_reads = 0; 24551da177e4SLinus Torvalds int ret = 0; 24561da177e4SLinus Torvalds int is_mapped_to_disk = 1; 24571da177e4SLinus Torvalds 245803158cd7SNick Piggin index = pos >> PAGE_CACHE_SHIFT; 245903158cd7SNick Piggin from = pos & (PAGE_CACHE_SIZE - 1); 246003158cd7SNick Piggin to = from + len; 246103158cd7SNick Piggin 246254566b2cSNick Piggin page = grab_cache_page_write_begin(mapping, index, flags); 246303158cd7SNick Piggin if (!page) 246403158cd7SNick Piggin return -ENOMEM; 246503158cd7SNick Piggin *pagep = page; 246603158cd7SNick Piggin *fsdata = NULL; 246703158cd7SNick Piggin 246803158cd7SNick Piggin if (page_has_buffers(page)) { 2469309f77adSNamhyung Kim ret = __block_write_begin(page, pos, len, get_block); 2470309f77adSNamhyung Kim if (unlikely(ret)) 2471309f77adSNamhyung Kim goto out_release; 2472309f77adSNamhyung Kim return ret; 247303158cd7SNick Piggin } 2474a4b0672dSNick Piggin 24751da177e4SLinus Torvalds if (PageMappedToDisk(page)) 24761da177e4SLinus Torvalds return 0; 24771da177e4SLinus Torvalds 2478a4b0672dSNick Piggin /* 2479a4b0672dSNick Piggin * Allocate buffers so that we can keep track of state, and potentially 2480a4b0672dSNick Piggin * attach them to the page if an error occurs. In the common case of 2481a4b0672dSNick Piggin * no error, they will just be freed again without ever being attached 2482a4b0672dSNick Piggin * to the page (which is all OK, because we're under the page lock). 2483a4b0672dSNick Piggin * 2484a4b0672dSNick Piggin * Be careful: the buffer linked list is a NULL terminated one, rather 2485a4b0672dSNick Piggin * than the circular one we're used to. 2486a4b0672dSNick Piggin */ 2487a4b0672dSNick Piggin head = alloc_page_buffers(page, blocksize, 0); 248803158cd7SNick Piggin if (!head) { 248903158cd7SNick Piggin ret = -ENOMEM; 249003158cd7SNick Piggin goto out_release; 249103158cd7SNick Piggin } 2492a4b0672dSNick Piggin 24931da177e4SLinus Torvalds block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits); 24941da177e4SLinus Torvalds 24951da177e4SLinus Torvalds /* 24961da177e4SLinus Torvalds * We loop across all blocks in the page, whether or not they are 24971da177e4SLinus Torvalds * part of the affected region. This is so we can discover if the 24981da177e4SLinus Torvalds * page is fully mapped-to-disk. 24991da177e4SLinus Torvalds */ 2500a4b0672dSNick Piggin for (block_start = 0, block_in_page = 0, bh = head; 25011da177e4SLinus Torvalds block_start < PAGE_CACHE_SIZE; 2502a4b0672dSNick Piggin block_in_page++, block_start += blocksize, bh = bh->b_this_page) { 25031da177e4SLinus Torvalds int create; 25041da177e4SLinus Torvalds 2505a4b0672dSNick Piggin block_end = block_start + blocksize; 2506a4b0672dSNick Piggin bh->b_state = 0; 25071da177e4SLinus Torvalds create = 1; 25081da177e4SLinus Torvalds if (block_start >= to) 25091da177e4SLinus Torvalds create = 0; 25101da177e4SLinus Torvalds ret = get_block(inode, block_in_file + block_in_page, 2511a4b0672dSNick Piggin bh, create); 25121da177e4SLinus Torvalds if (ret) 25131da177e4SLinus Torvalds goto failed; 2514a4b0672dSNick Piggin if (!buffer_mapped(bh)) 25151da177e4SLinus Torvalds is_mapped_to_disk = 0; 2516a4b0672dSNick Piggin if (buffer_new(bh)) 2517a4b0672dSNick Piggin unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); 2518a4b0672dSNick Piggin if (PageUptodate(page)) { 2519a4b0672dSNick Piggin set_buffer_uptodate(bh); 25201da177e4SLinus Torvalds continue; 2521a4b0672dSNick Piggin } 2522a4b0672dSNick Piggin if (buffer_new(bh) || !buffer_mapped(bh)) { 2523eebd2aa3SChristoph Lameter zero_user_segments(page, block_start, from, 2524eebd2aa3SChristoph Lameter to, block_end); 25251da177e4SLinus Torvalds continue; 25261da177e4SLinus Torvalds } 2527a4b0672dSNick Piggin if (buffer_uptodate(bh)) 25281da177e4SLinus Torvalds continue; /* reiserfs does this */ 25291da177e4SLinus Torvalds if (block_start < from || block_end > to) { 2530a4b0672dSNick Piggin lock_buffer(bh); 2531a4b0672dSNick Piggin bh->b_end_io = end_buffer_read_nobh; 2532a4b0672dSNick Piggin submit_bh(READ, bh); 2533a4b0672dSNick Piggin nr_reads++; 25341da177e4SLinus Torvalds } 25351da177e4SLinus Torvalds } 25361da177e4SLinus Torvalds 25371da177e4SLinus Torvalds if (nr_reads) { 25381da177e4SLinus Torvalds /* 25391da177e4SLinus Torvalds * The page is locked, so these buffers are protected from 25401da177e4SLinus Torvalds * any VM or truncate activity. Hence we don't need to care 25411da177e4SLinus Torvalds * for the buffer_head refcounts. 25421da177e4SLinus Torvalds */ 2543a4b0672dSNick Piggin for (bh = head; bh; bh = bh->b_this_page) { 25441da177e4SLinus Torvalds wait_on_buffer(bh); 25451da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 25461da177e4SLinus Torvalds ret = -EIO; 25471da177e4SLinus Torvalds } 25481da177e4SLinus Torvalds if (ret) 25491da177e4SLinus Torvalds goto failed; 25501da177e4SLinus Torvalds } 25511da177e4SLinus Torvalds 25521da177e4SLinus Torvalds if (is_mapped_to_disk) 25531da177e4SLinus Torvalds SetPageMappedToDisk(page); 25541da177e4SLinus Torvalds 255503158cd7SNick Piggin *fsdata = head; /* to be released by nobh_write_end */ 2556a4b0672dSNick Piggin 25571da177e4SLinus Torvalds return 0; 25581da177e4SLinus Torvalds 25591da177e4SLinus Torvalds failed: 256003158cd7SNick Piggin BUG_ON(!ret); 25611da177e4SLinus Torvalds /* 2562a4b0672dSNick Piggin * Error recovery is a bit difficult. We need to zero out blocks that 2563a4b0672dSNick Piggin * were newly allocated, and dirty them to ensure they get written out. 2564a4b0672dSNick Piggin * Buffers need to be attached to the page at this point, otherwise 2565a4b0672dSNick Piggin * the handling of potential IO errors during writeout would be hard 2566a4b0672dSNick Piggin * (could try doing synchronous writeout, but what if that fails too?) 25671da177e4SLinus Torvalds */ 256803158cd7SNick Piggin attach_nobh_buffers(page, head); 256903158cd7SNick Piggin page_zero_new_buffers(page, from, to); 2570a4b0672dSNick Piggin 257103158cd7SNick Piggin out_release: 257203158cd7SNick Piggin unlock_page(page); 257303158cd7SNick Piggin page_cache_release(page); 257403158cd7SNick Piggin *pagep = NULL; 2575a4b0672dSNick Piggin 25767bb46a67Snpiggin@suse.de return ret; 25777bb46a67Snpiggin@suse.de } 257803158cd7SNick Piggin EXPORT_SYMBOL(nobh_write_begin); 25791da177e4SLinus Torvalds 258003158cd7SNick Piggin int nobh_write_end(struct file *file, struct address_space *mapping, 258103158cd7SNick Piggin loff_t pos, unsigned len, unsigned copied, 258203158cd7SNick Piggin struct page *page, void *fsdata) 25831da177e4SLinus Torvalds { 25841da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 2585efdc3131SNick Piggin struct buffer_head *head = fsdata; 258603158cd7SNick Piggin struct buffer_head *bh; 25875b41e74aSDmitri Monakhov BUG_ON(fsdata != NULL && page_has_buffers(page)); 25881da177e4SLinus Torvalds 2589d4cf109fSDave Kleikamp if (unlikely(copied < len) && head) 259003158cd7SNick Piggin attach_nobh_buffers(page, head); 2591a4b0672dSNick Piggin if (page_has_buffers(page)) 259203158cd7SNick Piggin return generic_write_end(file, mapping, pos, len, 259303158cd7SNick Piggin copied, page, fsdata); 2594a4b0672dSNick Piggin 259522c8ca78SNick Piggin SetPageUptodate(page); 25961da177e4SLinus Torvalds set_page_dirty(page); 259703158cd7SNick Piggin if (pos+copied > inode->i_size) { 259803158cd7SNick Piggin i_size_write(inode, pos+copied); 25991da177e4SLinus Torvalds mark_inode_dirty(inode); 26001da177e4SLinus Torvalds } 260103158cd7SNick Piggin 260203158cd7SNick Piggin unlock_page(page); 260303158cd7SNick Piggin page_cache_release(page); 260403158cd7SNick Piggin 260503158cd7SNick Piggin while (head) { 260603158cd7SNick Piggin bh = head; 260703158cd7SNick Piggin head = head->b_this_page; 260803158cd7SNick Piggin free_buffer_head(bh); 26091da177e4SLinus Torvalds } 261003158cd7SNick Piggin 261103158cd7SNick Piggin return copied; 261203158cd7SNick Piggin } 261303158cd7SNick Piggin EXPORT_SYMBOL(nobh_write_end); 26141da177e4SLinus Torvalds 26151da177e4SLinus Torvalds /* 26161da177e4SLinus Torvalds * nobh_writepage() - based on block_full_write_page() except 26171da177e4SLinus Torvalds * that it tries to operate without attaching bufferheads to 26181da177e4SLinus Torvalds * the page. 26191da177e4SLinus Torvalds */ 26201da177e4SLinus Torvalds int nobh_writepage(struct page *page, get_block_t *get_block, 26211da177e4SLinus Torvalds struct writeback_control *wbc) 26221da177e4SLinus Torvalds { 26231da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 26241da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 26251da177e4SLinus Torvalds const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; 26261da177e4SLinus Torvalds unsigned offset; 26271da177e4SLinus Torvalds int ret; 26281da177e4SLinus Torvalds 26291da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 26301da177e4SLinus Torvalds if (page->index < end_index) 26311da177e4SLinus Torvalds goto out; 26321da177e4SLinus Torvalds 26331da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 26341da177e4SLinus Torvalds offset = i_size & (PAGE_CACHE_SIZE-1); 26351da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 26361da177e4SLinus Torvalds /* 26371da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 26381da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 26391da177e4SLinus Torvalds * freeable here, so the page does not leak. 26401da177e4SLinus Torvalds */ 26411da177e4SLinus Torvalds #if 0 26421da177e4SLinus Torvalds /* Not really sure about this - do we need this ? */ 26431da177e4SLinus Torvalds if (page->mapping->a_ops->invalidatepage) 26441da177e4SLinus Torvalds page->mapping->a_ops->invalidatepage(page, offset); 26451da177e4SLinus Torvalds #endif 26461da177e4SLinus Torvalds unlock_page(page); 26471da177e4SLinus Torvalds return 0; /* don't care */ 26481da177e4SLinus Torvalds } 26491da177e4SLinus Torvalds 26501da177e4SLinus Torvalds /* 26511da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 26521da177e4SLinus Torvalds * writepage invocation because it may be mmapped. "A file is mapped 26531da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 26541da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 26551da177e4SLinus Torvalds * writes to that region are not written out to the file." 26561da177e4SLinus Torvalds */ 2657eebd2aa3SChristoph Lameter zero_user_segment(page, offset, PAGE_CACHE_SIZE); 26581da177e4SLinus Torvalds out: 26591da177e4SLinus Torvalds ret = mpage_writepage(page, get_block, wbc); 26601da177e4SLinus Torvalds if (ret == -EAGAIN) 266135c80d5fSChris Mason ret = __block_write_full_page(inode, page, get_block, wbc, 266235c80d5fSChris Mason end_buffer_async_write); 26631da177e4SLinus Torvalds return ret; 26641da177e4SLinus Torvalds } 26651da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_writepage); 26661da177e4SLinus Torvalds 266703158cd7SNick Piggin int nobh_truncate_page(struct address_space *mapping, 266803158cd7SNick Piggin loff_t from, get_block_t *get_block) 26691da177e4SLinus Torvalds { 26701da177e4SLinus Torvalds pgoff_t index = from >> PAGE_CACHE_SHIFT; 26711da177e4SLinus Torvalds unsigned offset = from & (PAGE_CACHE_SIZE-1); 267203158cd7SNick Piggin unsigned blocksize; 267303158cd7SNick Piggin sector_t iblock; 267403158cd7SNick Piggin unsigned length, pos; 267503158cd7SNick Piggin struct inode *inode = mapping->host; 26761da177e4SLinus Torvalds struct page *page; 267703158cd7SNick Piggin struct buffer_head map_bh; 267803158cd7SNick Piggin int err; 26791da177e4SLinus Torvalds 268003158cd7SNick Piggin blocksize = 1 << inode->i_blkbits; 268103158cd7SNick Piggin length = offset & (blocksize - 1); 26821da177e4SLinus Torvalds 268303158cd7SNick Piggin /* Block boundary? Nothing to do */ 268403158cd7SNick Piggin if (!length) 268503158cd7SNick Piggin return 0; 268603158cd7SNick Piggin 268703158cd7SNick Piggin length = blocksize - length; 268803158cd7SNick Piggin iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 268903158cd7SNick Piggin 26901da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 269103158cd7SNick Piggin err = -ENOMEM; 26921da177e4SLinus Torvalds if (!page) 26931da177e4SLinus Torvalds goto out; 26941da177e4SLinus Torvalds 269503158cd7SNick Piggin if (page_has_buffers(page)) { 269603158cd7SNick Piggin has_buffers: 269703158cd7SNick Piggin unlock_page(page); 269803158cd7SNick Piggin page_cache_release(page); 269903158cd7SNick Piggin return block_truncate_page(mapping, from, get_block); 27001da177e4SLinus Torvalds } 270103158cd7SNick Piggin 270203158cd7SNick Piggin /* Find the buffer that contains "offset" */ 270303158cd7SNick Piggin pos = blocksize; 270403158cd7SNick Piggin while (offset >= pos) { 270503158cd7SNick Piggin iblock++; 270603158cd7SNick Piggin pos += blocksize; 270703158cd7SNick Piggin } 270803158cd7SNick Piggin 2709460bcf57STheodore Ts'o map_bh.b_size = blocksize; 2710460bcf57STheodore Ts'o map_bh.b_state = 0; 271103158cd7SNick Piggin err = get_block(inode, iblock, &map_bh, 0); 271203158cd7SNick Piggin if (err) 271303158cd7SNick Piggin goto unlock; 271403158cd7SNick Piggin /* unmapped? It's a hole - nothing to do */ 271503158cd7SNick Piggin if (!buffer_mapped(&map_bh)) 271603158cd7SNick Piggin goto unlock; 271703158cd7SNick Piggin 271803158cd7SNick Piggin /* Ok, it's mapped. Make sure it's up-to-date */ 271903158cd7SNick Piggin if (!PageUptodate(page)) { 272003158cd7SNick Piggin err = mapping->a_ops->readpage(NULL, page); 272103158cd7SNick Piggin if (err) { 272203158cd7SNick Piggin page_cache_release(page); 272303158cd7SNick Piggin goto out; 272403158cd7SNick Piggin } 272503158cd7SNick Piggin lock_page(page); 272603158cd7SNick Piggin if (!PageUptodate(page)) { 272703158cd7SNick Piggin err = -EIO; 272803158cd7SNick Piggin goto unlock; 272903158cd7SNick Piggin } 273003158cd7SNick Piggin if (page_has_buffers(page)) 273103158cd7SNick Piggin goto has_buffers; 273203158cd7SNick Piggin } 2733eebd2aa3SChristoph Lameter zero_user(page, offset, length); 273403158cd7SNick Piggin set_page_dirty(page); 273503158cd7SNick Piggin err = 0; 273603158cd7SNick Piggin 273703158cd7SNick Piggin unlock: 27381da177e4SLinus Torvalds unlock_page(page); 27391da177e4SLinus Torvalds page_cache_release(page); 27401da177e4SLinus Torvalds out: 274103158cd7SNick Piggin return err; 27421da177e4SLinus Torvalds } 27431da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_truncate_page); 27441da177e4SLinus Torvalds 27451da177e4SLinus Torvalds int block_truncate_page(struct address_space *mapping, 27461da177e4SLinus Torvalds loff_t from, get_block_t *get_block) 27471da177e4SLinus Torvalds { 27481da177e4SLinus Torvalds pgoff_t index = from >> PAGE_CACHE_SHIFT; 27491da177e4SLinus Torvalds unsigned offset = from & (PAGE_CACHE_SIZE-1); 27501da177e4SLinus Torvalds unsigned blocksize; 275154b21a79SAndrew Morton sector_t iblock; 27521da177e4SLinus Torvalds unsigned length, pos; 27531da177e4SLinus Torvalds struct inode *inode = mapping->host; 27541da177e4SLinus Torvalds struct page *page; 27551da177e4SLinus Torvalds struct buffer_head *bh; 27561da177e4SLinus Torvalds int err; 27571da177e4SLinus Torvalds 27581da177e4SLinus Torvalds blocksize = 1 << inode->i_blkbits; 27591da177e4SLinus Torvalds length = offset & (blocksize - 1); 27601da177e4SLinus Torvalds 27611da177e4SLinus Torvalds /* Block boundary? Nothing to do */ 27621da177e4SLinus Torvalds if (!length) 27631da177e4SLinus Torvalds return 0; 27641da177e4SLinus Torvalds 27651da177e4SLinus Torvalds length = blocksize - length; 276654b21a79SAndrew Morton iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 27671da177e4SLinus Torvalds 27681da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 27691da177e4SLinus Torvalds err = -ENOMEM; 27701da177e4SLinus Torvalds if (!page) 27711da177e4SLinus Torvalds goto out; 27721da177e4SLinus Torvalds 27731da177e4SLinus Torvalds if (!page_has_buffers(page)) 27741da177e4SLinus Torvalds create_empty_buffers(page, blocksize, 0); 27751da177e4SLinus Torvalds 27761da177e4SLinus Torvalds /* Find the buffer that contains "offset" */ 27771da177e4SLinus Torvalds bh = page_buffers(page); 27781da177e4SLinus Torvalds pos = blocksize; 27791da177e4SLinus Torvalds while (offset >= pos) { 27801da177e4SLinus Torvalds bh = bh->b_this_page; 27811da177e4SLinus Torvalds iblock++; 27821da177e4SLinus Torvalds pos += blocksize; 27831da177e4SLinus Torvalds } 27841da177e4SLinus Torvalds 27851da177e4SLinus Torvalds err = 0; 27861da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2787b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 27881da177e4SLinus Torvalds err = get_block(inode, iblock, bh, 0); 27891da177e4SLinus Torvalds if (err) 27901da177e4SLinus Torvalds goto unlock; 27911da177e4SLinus Torvalds /* unmapped? It's a hole - nothing to do */ 27921da177e4SLinus Torvalds if (!buffer_mapped(bh)) 27931da177e4SLinus Torvalds goto unlock; 27941da177e4SLinus Torvalds } 27951da177e4SLinus Torvalds 27961da177e4SLinus Torvalds /* Ok, it's mapped. Make sure it's up-to-date */ 27971da177e4SLinus Torvalds if (PageUptodate(page)) 27981da177e4SLinus Torvalds set_buffer_uptodate(bh); 27991da177e4SLinus Torvalds 280033a266ddSDavid Chinner if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { 28011da177e4SLinus Torvalds err = -EIO; 28021da177e4SLinus Torvalds ll_rw_block(READ, 1, &bh); 28031da177e4SLinus Torvalds wait_on_buffer(bh); 28041da177e4SLinus Torvalds /* Uhhuh. Read error. Complain and punt. */ 28051da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 28061da177e4SLinus Torvalds goto unlock; 28071da177e4SLinus Torvalds } 28081da177e4SLinus Torvalds 2809eebd2aa3SChristoph Lameter zero_user(page, offset, length); 28101da177e4SLinus Torvalds mark_buffer_dirty(bh); 28111da177e4SLinus Torvalds err = 0; 28121da177e4SLinus Torvalds 28131da177e4SLinus Torvalds unlock: 28141da177e4SLinus Torvalds unlock_page(page); 28151da177e4SLinus Torvalds page_cache_release(page); 28161da177e4SLinus Torvalds out: 28171da177e4SLinus Torvalds return err; 28181da177e4SLinus Torvalds } 28191fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_truncate_page); 28201da177e4SLinus Torvalds 28211da177e4SLinus Torvalds /* 28221da177e4SLinus Torvalds * The generic ->writepage function for buffer-backed address_spaces 282335c80d5fSChris Mason * this form passes in the end_io handler used to finish the IO. 28241da177e4SLinus Torvalds */ 282535c80d5fSChris Mason int block_write_full_page_endio(struct page *page, get_block_t *get_block, 282635c80d5fSChris Mason struct writeback_control *wbc, bh_end_io_t *handler) 28271da177e4SLinus Torvalds { 28281da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 28291da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 28301da177e4SLinus Torvalds const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; 28311da177e4SLinus Torvalds unsigned offset; 28321da177e4SLinus Torvalds 28331da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 28341da177e4SLinus Torvalds if (page->index < end_index) 283535c80d5fSChris Mason return __block_write_full_page(inode, page, get_block, wbc, 283635c80d5fSChris Mason handler); 28371da177e4SLinus Torvalds 28381da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 28391da177e4SLinus Torvalds offset = i_size & (PAGE_CACHE_SIZE-1); 28401da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 28411da177e4SLinus Torvalds /* 28421da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 28431da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 28441da177e4SLinus Torvalds * freeable here, so the page does not leak. 28451da177e4SLinus Torvalds */ 2846aaa4059bSJan Kara do_invalidatepage(page, 0); 28471da177e4SLinus Torvalds unlock_page(page); 28481da177e4SLinus Torvalds return 0; /* don't care */ 28491da177e4SLinus Torvalds } 28501da177e4SLinus Torvalds 28511da177e4SLinus Torvalds /* 28521da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 28532a61aa40SAdam Buchbinder * writepage invocation because it may be mmapped. "A file is mapped 28541da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 28551da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 28561da177e4SLinus Torvalds * writes to that region are not written out to the file." 28571da177e4SLinus Torvalds */ 2858eebd2aa3SChristoph Lameter zero_user_segment(page, offset, PAGE_CACHE_SIZE); 285935c80d5fSChris Mason return __block_write_full_page(inode, page, get_block, wbc, handler); 28601da177e4SLinus Torvalds } 28611fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_write_full_page_endio); 28621da177e4SLinus Torvalds 286335c80d5fSChris Mason /* 286435c80d5fSChris Mason * The generic ->writepage function for buffer-backed address_spaces 286535c80d5fSChris Mason */ 286635c80d5fSChris Mason int block_write_full_page(struct page *page, get_block_t *get_block, 286735c80d5fSChris Mason struct writeback_control *wbc) 286835c80d5fSChris Mason { 286935c80d5fSChris Mason return block_write_full_page_endio(page, get_block, wbc, 287035c80d5fSChris Mason end_buffer_async_write); 287135c80d5fSChris Mason } 28721fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_write_full_page); 287335c80d5fSChris Mason 28741da177e4SLinus Torvalds sector_t generic_block_bmap(struct address_space *mapping, sector_t block, 28751da177e4SLinus Torvalds get_block_t *get_block) 28761da177e4SLinus Torvalds { 28771da177e4SLinus Torvalds struct buffer_head tmp; 28781da177e4SLinus Torvalds struct inode *inode = mapping->host; 28791da177e4SLinus Torvalds tmp.b_state = 0; 28801da177e4SLinus Torvalds tmp.b_blocknr = 0; 2881b0cf2321SBadari Pulavarty tmp.b_size = 1 << inode->i_blkbits; 28821da177e4SLinus Torvalds get_block(inode, block, &tmp, 0); 28831da177e4SLinus Torvalds return tmp.b_blocknr; 28841da177e4SLinus Torvalds } 28851fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_block_bmap); 28861da177e4SLinus Torvalds 28876712ecf8SNeilBrown static void end_bio_bh_io_sync(struct bio *bio, int err) 28881da177e4SLinus Torvalds { 28891da177e4SLinus Torvalds struct buffer_head *bh = bio->bi_private; 28901da177e4SLinus Torvalds 28911da177e4SLinus Torvalds if (err == -EOPNOTSUPP) { 28921da177e4SLinus Torvalds set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); 28931da177e4SLinus Torvalds } 28941da177e4SLinus Torvalds 289508bafc03SKeith Mannthey if (unlikely (test_bit(BIO_QUIET,&bio->bi_flags))) 289608bafc03SKeith Mannthey set_bit(BH_Quiet, &bh->b_state); 289708bafc03SKeith Mannthey 28981da177e4SLinus Torvalds bh->b_end_io(bh, test_bit(BIO_UPTODATE, &bio->bi_flags)); 28991da177e4SLinus Torvalds bio_put(bio); 29001da177e4SLinus Torvalds } 29011da177e4SLinus Torvalds 290257302e0dSLinus Torvalds /* 290357302e0dSLinus Torvalds * This allows us to do IO even on the odd last sectors 290457302e0dSLinus Torvalds * of a device, even if the bh block size is some multiple 290557302e0dSLinus Torvalds * of the physical sector size. 290657302e0dSLinus Torvalds * 290757302e0dSLinus Torvalds * We'll just truncate the bio to the size of the device, 290857302e0dSLinus Torvalds * and clear the end of the buffer head manually. 290957302e0dSLinus Torvalds * 291057302e0dSLinus Torvalds * Truly out-of-range accesses will turn into actual IO 291157302e0dSLinus Torvalds * errors, this only handles the "we need to be able to 291257302e0dSLinus Torvalds * do IO at the final sector" case. 291357302e0dSLinus Torvalds */ 291457302e0dSLinus Torvalds static void guard_bh_eod(int rw, struct bio *bio, struct buffer_head *bh) 291557302e0dSLinus Torvalds { 291657302e0dSLinus Torvalds sector_t maxsector; 291757302e0dSLinus Torvalds unsigned bytes; 291857302e0dSLinus Torvalds 291957302e0dSLinus Torvalds maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; 292057302e0dSLinus Torvalds if (!maxsector) 292157302e0dSLinus Torvalds return; 292257302e0dSLinus Torvalds 292357302e0dSLinus Torvalds /* 292457302e0dSLinus Torvalds * If the *whole* IO is past the end of the device, 292557302e0dSLinus Torvalds * let it through, and the IO layer will turn it into 292657302e0dSLinus Torvalds * an EIO. 292757302e0dSLinus Torvalds */ 292857302e0dSLinus Torvalds if (unlikely(bio->bi_sector >= maxsector)) 292957302e0dSLinus Torvalds return; 293057302e0dSLinus Torvalds 293157302e0dSLinus Torvalds maxsector -= bio->bi_sector; 293257302e0dSLinus Torvalds bytes = bio->bi_size; 293357302e0dSLinus Torvalds if (likely((bytes >> 9) <= maxsector)) 293457302e0dSLinus Torvalds return; 293557302e0dSLinus Torvalds 293657302e0dSLinus Torvalds /* Uhhuh. We've got a bh that straddles the device size! */ 293757302e0dSLinus Torvalds bytes = maxsector << 9; 293857302e0dSLinus Torvalds 293957302e0dSLinus Torvalds /* Truncate the bio.. */ 294057302e0dSLinus Torvalds bio->bi_size = bytes; 294157302e0dSLinus Torvalds bio->bi_io_vec[0].bv_len = bytes; 294257302e0dSLinus Torvalds 294357302e0dSLinus Torvalds /* ..and clear the end of the buffer for reads */ 294427d7c2a0SDan Carpenter if ((rw & RW_MASK) == READ) { 294557302e0dSLinus Torvalds void *kaddr = kmap_atomic(bh->b_page); 294657302e0dSLinus Torvalds memset(kaddr + bh_offset(bh) + bytes, 0, bh->b_size - bytes); 294757302e0dSLinus Torvalds kunmap_atomic(kaddr); 29486d283dbaSLinus Torvalds flush_dcache_page(bh->b_page); 294957302e0dSLinus Torvalds } 295057302e0dSLinus Torvalds } 295157302e0dSLinus Torvalds 29521da177e4SLinus Torvalds int submit_bh(int rw, struct buffer_head * bh) 29531da177e4SLinus Torvalds { 29541da177e4SLinus Torvalds struct bio *bio; 29551da177e4SLinus Torvalds int ret = 0; 29561da177e4SLinus Torvalds 29571da177e4SLinus Torvalds BUG_ON(!buffer_locked(bh)); 29581da177e4SLinus Torvalds BUG_ON(!buffer_mapped(bh)); 29591da177e4SLinus Torvalds BUG_ON(!bh->b_end_io); 29608fb0e342SAneesh Kumar K.V BUG_ON(buffer_delay(bh)); 29618fb0e342SAneesh Kumar K.V BUG_ON(buffer_unwritten(bh)); 29621da177e4SLinus Torvalds 296348fd4f93SJens Axboe /* 296448fd4f93SJens Axboe * Only clear out a write error when rewriting 29651da177e4SLinus Torvalds */ 296648fd4f93SJens Axboe if (test_set_buffer_req(bh) && (rw & WRITE)) 29671da177e4SLinus Torvalds clear_buffer_write_io_error(bh); 29681da177e4SLinus Torvalds 29691da177e4SLinus Torvalds /* 29701da177e4SLinus Torvalds * from here on down, it's all bio -- do the initial mapping, 29711da177e4SLinus Torvalds * submit_bio -> generic_make_request may further map this bio around 29721da177e4SLinus Torvalds */ 29731da177e4SLinus Torvalds bio = bio_alloc(GFP_NOIO, 1); 29741da177e4SLinus Torvalds 29751da177e4SLinus Torvalds bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); 29761da177e4SLinus Torvalds bio->bi_bdev = bh->b_bdev; 29771da177e4SLinus Torvalds bio->bi_io_vec[0].bv_page = bh->b_page; 29781da177e4SLinus Torvalds bio->bi_io_vec[0].bv_len = bh->b_size; 29791da177e4SLinus Torvalds bio->bi_io_vec[0].bv_offset = bh_offset(bh); 29801da177e4SLinus Torvalds 29811da177e4SLinus Torvalds bio->bi_vcnt = 1; 29821da177e4SLinus Torvalds bio->bi_idx = 0; 29831da177e4SLinus Torvalds bio->bi_size = bh->b_size; 29841da177e4SLinus Torvalds 29851da177e4SLinus Torvalds bio->bi_end_io = end_bio_bh_io_sync; 29861da177e4SLinus Torvalds bio->bi_private = bh; 29871da177e4SLinus Torvalds 298857302e0dSLinus Torvalds /* Take care of bh's that straddle the end of the device */ 298957302e0dSLinus Torvalds guard_bh_eod(rw, bio, bh); 299057302e0dSLinus Torvalds 2991*877f962cSTheodore Ts'o if (buffer_meta(bh)) 2992*877f962cSTheodore Ts'o rw |= REQ_META; 2993*877f962cSTheodore Ts'o if (buffer_prio(bh)) 2994*877f962cSTheodore Ts'o rw |= REQ_PRIO; 2995*877f962cSTheodore Ts'o 29961da177e4SLinus Torvalds bio_get(bio); 29971da177e4SLinus Torvalds submit_bio(rw, bio); 29981da177e4SLinus Torvalds 29991da177e4SLinus Torvalds if (bio_flagged(bio, BIO_EOPNOTSUPP)) 30001da177e4SLinus Torvalds ret = -EOPNOTSUPP; 30011da177e4SLinus Torvalds 30021da177e4SLinus Torvalds bio_put(bio); 30031da177e4SLinus Torvalds return ret; 30041da177e4SLinus Torvalds } 30051fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(submit_bh); 30061da177e4SLinus Torvalds 30071da177e4SLinus Torvalds /** 30081da177e4SLinus Torvalds * ll_rw_block: low-level access to block devices (DEPRECATED) 30099cb569d6SChristoph Hellwig * @rw: whether to %READ or %WRITE or maybe %READA (readahead) 30101da177e4SLinus Torvalds * @nr: number of &struct buffer_heads in the array 30111da177e4SLinus Torvalds * @bhs: array of pointers to &struct buffer_head 30121da177e4SLinus Torvalds * 3013a7662236SJan Kara * ll_rw_block() takes an array of pointers to &struct buffer_heads, and 3014a7662236SJan Kara * requests an I/O operation on them, either a %READ or a %WRITE. The third 30159cb569d6SChristoph Hellwig * %READA option is described in the documentation for generic_make_request() 30169cb569d6SChristoph Hellwig * which ll_rw_block() calls. 30171da177e4SLinus Torvalds * 30181da177e4SLinus Torvalds * This function drops any buffer that it cannot get a lock on (with the 30199cb569d6SChristoph Hellwig * BH_Lock state bit), any buffer that appears to be clean when doing a write 30209cb569d6SChristoph Hellwig * request, and any buffer that appears to be up-to-date when doing read 30219cb569d6SChristoph Hellwig * request. Further it marks as clean buffers that are processed for 30229cb569d6SChristoph Hellwig * writing (the buffer cache won't assume that they are actually clean 30239cb569d6SChristoph Hellwig * until the buffer gets unlocked). 30241da177e4SLinus Torvalds * 30251da177e4SLinus Torvalds * ll_rw_block sets b_end_io to simple completion handler that marks 30261da177e4SLinus Torvalds * the buffer up-to-date (if approriate), unlocks the buffer and wakes 30271da177e4SLinus Torvalds * any waiters. 30281da177e4SLinus Torvalds * 30291da177e4SLinus Torvalds * All of the buffers must be for the same device, and must also be a 30301da177e4SLinus Torvalds * multiple of the current approved size for the device. 30311da177e4SLinus Torvalds */ 30321da177e4SLinus Torvalds void ll_rw_block(int rw, int nr, struct buffer_head *bhs[]) 30331da177e4SLinus Torvalds { 30341da177e4SLinus Torvalds int i; 30351da177e4SLinus Torvalds 30361da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 30371da177e4SLinus Torvalds struct buffer_head *bh = bhs[i]; 30381da177e4SLinus Torvalds 30399cb569d6SChristoph Hellwig if (!trylock_buffer(bh)) 30401da177e4SLinus Torvalds continue; 30419cb569d6SChristoph Hellwig if (rw == WRITE) { 30421da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 304376c3073aSakpm@osdl.org bh->b_end_io = end_buffer_write_sync; 3044e60e5c50SOGAWA Hirofumi get_bh(bh); 30451da177e4SLinus Torvalds submit_bh(WRITE, bh); 30461da177e4SLinus Torvalds continue; 30471da177e4SLinus Torvalds } 30481da177e4SLinus Torvalds } else { 30491da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 305076c3073aSakpm@osdl.org bh->b_end_io = end_buffer_read_sync; 3051e60e5c50SOGAWA Hirofumi get_bh(bh); 30521da177e4SLinus Torvalds submit_bh(rw, bh); 30531da177e4SLinus Torvalds continue; 30541da177e4SLinus Torvalds } 30551da177e4SLinus Torvalds } 30561da177e4SLinus Torvalds unlock_buffer(bh); 30571da177e4SLinus Torvalds } 30581da177e4SLinus Torvalds } 30591fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(ll_rw_block); 30601da177e4SLinus Torvalds 30619cb569d6SChristoph Hellwig void write_dirty_buffer(struct buffer_head *bh, int rw) 30629cb569d6SChristoph Hellwig { 30639cb569d6SChristoph Hellwig lock_buffer(bh); 30649cb569d6SChristoph Hellwig if (!test_clear_buffer_dirty(bh)) { 30659cb569d6SChristoph Hellwig unlock_buffer(bh); 30669cb569d6SChristoph Hellwig return; 30679cb569d6SChristoph Hellwig } 30689cb569d6SChristoph Hellwig bh->b_end_io = end_buffer_write_sync; 30699cb569d6SChristoph Hellwig get_bh(bh); 30709cb569d6SChristoph Hellwig submit_bh(rw, bh); 30719cb569d6SChristoph Hellwig } 30729cb569d6SChristoph Hellwig EXPORT_SYMBOL(write_dirty_buffer); 30739cb569d6SChristoph Hellwig 30741da177e4SLinus Torvalds /* 30751da177e4SLinus Torvalds * For a data-integrity writeout, we need to wait upon any in-progress I/O 30761da177e4SLinus Torvalds * and then start new I/O and then wait upon it. The caller must have a ref on 30771da177e4SLinus Torvalds * the buffer_head. 30781da177e4SLinus Torvalds */ 307987e99511SChristoph Hellwig int __sync_dirty_buffer(struct buffer_head *bh, int rw) 30801da177e4SLinus Torvalds { 30811da177e4SLinus Torvalds int ret = 0; 30821da177e4SLinus Torvalds 30831da177e4SLinus Torvalds WARN_ON(atomic_read(&bh->b_count) < 1); 30841da177e4SLinus Torvalds lock_buffer(bh); 30851da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 30861da177e4SLinus Torvalds get_bh(bh); 30871da177e4SLinus Torvalds bh->b_end_io = end_buffer_write_sync; 308887e99511SChristoph Hellwig ret = submit_bh(rw, bh); 30891da177e4SLinus Torvalds wait_on_buffer(bh); 30901da177e4SLinus Torvalds if (!ret && !buffer_uptodate(bh)) 30911da177e4SLinus Torvalds ret = -EIO; 30921da177e4SLinus Torvalds } else { 30931da177e4SLinus Torvalds unlock_buffer(bh); 30941da177e4SLinus Torvalds } 30951da177e4SLinus Torvalds return ret; 30961da177e4SLinus Torvalds } 309787e99511SChristoph Hellwig EXPORT_SYMBOL(__sync_dirty_buffer); 309887e99511SChristoph Hellwig 309987e99511SChristoph Hellwig int sync_dirty_buffer(struct buffer_head *bh) 310087e99511SChristoph Hellwig { 310187e99511SChristoph Hellwig return __sync_dirty_buffer(bh, WRITE_SYNC); 310287e99511SChristoph Hellwig } 31031fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(sync_dirty_buffer); 31041da177e4SLinus Torvalds 31051da177e4SLinus Torvalds /* 31061da177e4SLinus Torvalds * try_to_free_buffers() checks if all the buffers on this particular page 31071da177e4SLinus Torvalds * are unused, and releases them if so. 31081da177e4SLinus Torvalds * 31091da177e4SLinus Torvalds * Exclusion against try_to_free_buffers may be obtained by either 31101da177e4SLinus Torvalds * locking the page or by holding its mapping's private_lock. 31111da177e4SLinus Torvalds * 31121da177e4SLinus Torvalds * If the page is dirty but all the buffers are clean then we need to 31131da177e4SLinus Torvalds * be sure to mark the page clean as well. This is because the page 31141da177e4SLinus Torvalds * may be against a block device, and a later reattachment of buffers 31151da177e4SLinus Torvalds * to a dirty page will set *all* buffers dirty. Which would corrupt 31161da177e4SLinus Torvalds * filesystem data on the same device. 31171da177e4SLinus Torvalds * 31181da177e4SLinus Torvalds * The same applies to regular filesystem pages: if all the buffers are 31191da177e4SLinus Torvalds * clean then we set the page clean and proceed. To do that, we require 31201da177e4SLinus Torvalds * total exclusion from __set_page_dirty_buffers(). That is obtained with 31211da177e4SLinus Torvalds * private_lock. 31221da177e4SLinus Torvalds * 31231da177e4SLinus Torvalds * try_to_free_buffers() is non-blocking. 31241da177e4SLinus Torvalds */ 31251da177e4SLinus Torvalds static inline int buffer_busy(struct buffer_head *bh) 31261da177e4SLinus Torvalds { 31271da177e4SLinus Torvalds return atomic_read(&bh->b_count) | 31281da177e4SLinus Torvalds (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); 31291da177e4SLinus Torvalds } 31301da177e4SLinus Torvalds 31311da177e4SLinus Torvalds static int 31321da177e4SLinus Torvalds drop_buffers(struct page *page, struct buffer_head **buffers_to_free) 31331da177e4SLinus Torvalds { 31341da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 31351da177e4SLinus Torvalds struct buffer_head *bh; 31361da177e4SLinus Torvalds 31371da177e4SLinus Torvalds bh = head; 31381da177e4SLinus Torvalds do { 3139de7d5a3bSakpm@osdl.org if (buffer_write_io_error(bh) && page->mapping) 31401da177e4SLinus Torvalds set_bit(AS_EIO, &page->mapping->flags); 31411da177e4SLinus Torvalds if (buffer_busy(bh)) 31421da177e4SLinus Torvalds goto failed; 31431da177e4SLinus Torvalds bh = bh->b_this_page; 31441da177e4SLinus Torvalds } while (bh != head); 31451da177e4SLinus Torvalds 31461da177e4SLinus Torvalds do { 31471da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 31481da177e4SLinus Torvalds 3149535ee2fbSJan Kara if (bh->b_assoc_map) 31501da177e4SLinus Torvalds __remove_assoc_queue(bh); 31511da177e4SLinus Torvalds bh = next; 31521da177e4SLinus Torvalds } while (bh != head); 31531da177e4SLinus Torvalds *buffers_to_free = head; 31541da177e4SLinus Torvalds __clear_page_buffers(page); 31551da177e4SLinus Torvalds return 1; 31561da177e4SLinus Torvalds failed: 31571da177e4SLinus Torvalds return 0; 31581da177e4SLinus Torvalds } 31591da177e4SLinus Torvalds 31601da177e4SLinus Torvalds int try_to_free_buffers(struct page *page) 31611da177e4SLinus Torvalds { 31621da177e4SLinus Torvalds struct address_space * const mapping = page->mapping; 31631da177e4SLinus Torvalds struct buffer_head *buffers_to_free = NULL; 31641da177e4SLinus Torvalds int ret = 0; 31651da177e4SLinus Torvalds 31661da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 3167ecdfc978SLinus Torvalds if (PageWriteback(page)) 31681da177e4SLinus Torvalds return 0; 31691da177e4SLinus Torvalds 31701da177e4SLinus Torvalds if (mapping == NULL) { /* can this still happen? */ 31711da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 31721da177e4SLinus Torvalds goto out; 31731da177e4SLinus Torvalds } 31741da177e4SLinus Torvalds 31751da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 31761da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 3177ecdfc978SLinus Torvalds 3178ecdfc978SLinus Torvalds /* 3179ecdfc978SLinus Torvalds * If the filesystem writes its buffers by hand (eg ext3) 3180ecdfc978SLinus Torvalds * then we can have clean buffers against a dirty page. We 3181ecdfc978SLinus Torvalds * clean the page here; otherwise the VM will never notice 3182ecdfc978SLinus Torvalds * that the filesystem did any IO at all. 3183ecdfc978SLinus Torvalds * 3184ecdfc978SLinus Torvalds * Also, during truncate, discard_buffer will have marked all 3185ecdfc978SLinus Torvalds * the page's buffers clean. We discover that here and clean 3186ecdfc978SLinus Torvalds * the page also. 318787df7241SNick Piggin * 318887df7241SNick Piggin * private_lock must be held over this entire operation in order 318987df7241SNick Piggin * to synchronise against __set_page_dirty_buffers and prevent the 319087df7241SNick Piggin * dirty bit from being lost. 3191ecdfc978SLinus Torvalds */ 3192ecdfc978SLinus Torvalds if (ret) 3193ecdfc978SLinus Torvalds cancel_dirty_page(page, PAGE_CACHE_SIZE); 319487df7241SNick Piggin spin_unlock(&mapping->private_lock); 31951da177e4SLinus Torvalds out: 31961da177e4SLinus Torvalds if (buffers_to_free) { 31971da177e4SLinus Torvalds struct buffer_head *bh = buffers_to_free; 31981da177e4SLinus Torvalds 31991da177e4SLinus Torvalds do { 32001da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 32011da177e4SLinus Torvalds free_buffer_head(bh); 32021da177e4SLinus Torvalds bh = next; 32031da177e4SLinus Torvalds } while (bh != buffers_to_free); 32041da177e4SLinus Torvalds } 32051da177e4SLinus Torvalds return ret; 32061da177e4SLinus Torvalds } 32071da177e4SLinus Torvalds EXPORT_SYMBOL(try_to_free_buffers); 32081da177e4SLinus Torvalds 32091da177e4SLinus Torvalds /* 32101da177e4SLinus Torvalds * There are no bdflush tunables left. But distributions are 32111da177e4SLinus Torvalds * still running obsolete flush daemons, so we terminate them here. 32121da177e4SLinus Torvalds * 32131da177e4SLinus Torvalds * Use of bdflush() is deprecated and will be removed in a future kernel. 32145b0830cbSJens Axboe * The `flush-X' kernel threads fully replace bdflush daemons and this call. 32151da177e4SLinus Torvalds */ 3216bdc480e3SHeiko Carstens SYSCALL_DEFINE2(bdflush, int, func, long, data) 32171da177e4SLinus Torvalds { 32181da177e4SLinus Torvalds static int msg_count; 32191da177e4SLinus Torvalds 32201da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 32211da177e4SLinus Torvalds return -EPERM; 32221da177e4SLinus Torvalds 32231da177e4SLinus Torvalds if (msg_count < 5) { 32241da177e4SLinus Torvalds msg_count++; 32251da177e4SLinus Torvalds printk(KERN_INFO 32261da177e4SLinus Torvalds "warning: process `%s' used the obsolete bdflush" 32271da177e4SLinus Torvalds " system call\n", current->comm); 32281da177e4SLinus Torvalds printk(KERN_INFO "Fix your initscripts?\n"); 32291da177e4SLinus Torvalds } 32301da177e4SLinus Torvalds 32311da177e4SLinus Torvalds if (func == 1) 32321da177e4SLinus Torvalds do_exit(0); 32331da177e4SLinus Torvalds return 0; 32341da177e4SLinus Torvalds } 32351da177e4SLinus Torvalds 32361da177e4SLinus Torvalds /* 32371da177e4SLinus Torvalds * Buffer-head allocation 32381da177e4SLinus Torvalds */ 3239a0a9b043SShai Fultheim static struct kmem_cache *bh_cachep __read_mostly; 32401da177e4SLinus Torvalds 32411da177e4SLinus Torvalds /* 32421da177e4SLinus Torvalds * Once the number of bh's in the machine exceeds this level, we start 32431da177e4SLinus Torvalds * stripping them in writeback. 32441da177e4SLinus Torvalds */ 324543be594aSZhang Yanfei static unsigned long max_buffer_heads; 32461da177e4SLinus Torvalds 32471da177e4SLinus Torvalds int buffer_heads_over_limit; 32481da177e4SLinus Torvalds 32491da177e4SLinus Torvalds struct bh_accounting { 32501da177e4SLinus Torvalds int nr; /* Number of live bh's */ 32511da177e4SLinus Torvalds int ratelimit; /* Limit cacheline bouncing */ 32521da177e4SLinus Torvalds }; 32531da177e4SLinus Torvalds 32541da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; 32551da177e4SLinus Torvalds 32561da177e4SLinus Torvalds static void recalc_bh_state(void) 32571da177e4SLinus Torvalds { 32581da177e4SLinus Torvalds int i; 32591da177e4SLinus Torvalds int tot = 0; 32601da177e4SLinus Torvalds 3261ee1be862SChristoph Lameter if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) 32621da177e4SLinus Torvalds return; 3263c7b92516SChristoph Lameter __this_cpu_write(bh_accounting.ratelimit, 0); 32648a143426SEric Dumazet for_each_online_cpu(i) 32651da177e4SLinus Torvalds tot += per_cpu(bh_accounting, i).nr; 32661da177e4SLinus Torvalds buffer_heads_over_limit = (tot > max_buffer_heads); 32671da177e4SLinus Torvalds } 32681da177e4SLinus Torvalds 3269dd0fc66fSAl Viro struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) 32701da177e4SLinus Torvalds { 3271019b4d12SRichard Kennedy struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); 32721da177e4SLinus Torvalds if (ret) { 3273a35afb83SChristoph Lameter INIT_LIST_HEAD(&ret->b_assoc_buffers); 3274c7b92516SChristoph Lameter preempt_disable(); 3275c7b92516SChristoph Lameter __this_cpu_inc(bh_accounting.nr); 32761da177e4SLinus Torvalds recalc_bh_state(); 3277c7b92516SChristoph Lameter preempt_enable(); 32781da177e4SLinus Torvalds } 32791da177e4SLinus Torvalds return ret; 32801da177e4SLinus Torvalds } 32811da177e4SLinus Torvalds EXPORT_SYMBOL(alloc_buffer_head); 32821da177e4SLinus Torvalds 32831da177e4SLinus Torvalds void free_buffer_head(struct buffer_head *bh) 32841da177e4SLinus Torvalds { 32851da177e4SLinus Torvalds BUG_ON(!list_empty(&bh->b_assoc_buffers)); 32861da177e4SLinus Torvalds kmem_cache_free(bh_cachep, bh); 3287c7b92516SChristoph Lameter preempt_disable(); 3288c7b92516SChristoph Lameter __this_cpu_dec(bh_accounting.nr); 32891da177e4SLinus Torvalds recalc_bh_state(); 3290c7b92516SChristoph Lameter preempt_enable(); 32911da177e4SLinus Torvalds } 32921da177e4SLinus Torvalds EXPORT_SYMBOL(free_buffer_head); 32931da177e4SLinus Torvalds 32941da177e4SLinus Torvalds static void buffer_exit_cpu(int cpu) 32951da177e4SLinus Torvalds { 32961da177e4SLinus Torvalds int i; 32971da177e4SLinus Torvalds struct bh_lru *b = &per_cpu(bh_lrus, cpu); 32981da177e4SLinus Torvalds 32991da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 33001da177e4SLinus Torvalds brelse(b->bhs[i]); 33011da177e4SLinus Torvalds b->bhs[i] = NULL; 33021da177e4SLinus Torvalds } 3303c7b92516SChristoph Lameter this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); 33048a143426SEric Dumazet per_cpu(bh_accounting, cpu).nr = 0; 33051da177e4SLinus Torvalds } 33061da177e4SLinus Torvalds 33071da177e4SLinus Torvalds static int buffer_cpu_notify(struct notifier_block *self, 33081da177e4SLinus Torvalds unsigned long action, void *hcpu) 33091da177e4SLinus Torvalds { 33108bb78442SRafael J. Wysocki if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) 33111da177e4SLinus Torvalds buffer_exit_cpu((unsigned long)hcpu); 33121da177e4SLinus Torvalds return NOTIFY_OK; 33131da177e4SLinus Torvalds } 33141da177e4SLinus Torvalds 3315389d1b08SAneesh Kumar K.V /** 3316a6b91919SRandy Dunlap * bh_uptodate_or_lock - Test whether the buffer is uptodate 3317389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3318389d1b08SAneesh Kumar K.V * 3319389d1b08SAneesh Kumar K.V * Return true if the buffer is up-to-date and false, 3320389d1b08SAneesh Kumar K.V * with the buffer locked, if not. 3321389d1b08SAneesh Kumar K.V */ 3322389d1b08SAneesh Kumar K.V int bh_uptodate_or_lock(struct buffer_head *bh) 3323389d1b08SAneesh Kumar K.V { 3324389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) { 3325389d1b08SAneesh Kumar K.V lock_buffer(bh); 3326389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) 3327389d1b08SAneesh Kumar K.V return 0; 3328389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3329389d1b08SAneesh Kumar K.V } 3330389d1b08SAneesh Kumar K.V return 1; 3331389d1b08SAneesh Kumar K.V } 3332389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_uptodate_or_lock); 3333389d1b08SAneesh Kumar K.V 3334389d1b08SAneesh Kumar K.V /** 3335a6b91919SRandy Dunlap * bh_submit_read - Submit a locked buffer for reading 3336389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3337389d1b08SAneesh Kumar K.V * 3338389d1b08SAneesh Kumar K.V * Returns zero on success and -EIO on error. 3339389d1b08SAneesh Kumar K.V */ 3340389d1b08SAneesh Kumar K.V int bh_submit_read(struct buffer_head *bh) 3341389d1b08SAneesh Kumar K.V { 3342389d1b08SAneesh Kumar K.V BUG_ON(!buffer_locked(bh)); 3343389d1b08SAneesh Kumar K.V 3344389d1b08SAneesh Kumar K.V if (buffer_uptodate(bh)) { 3345389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3346389d1b08SAneesh Kumar K.V return 0; 3347389d1b08SAneesh Kumar K.V } 3348389d1b08SAneesh Kumar K.V 3349389d1b08SAneesh Kumar K.V get_bh(bh); 3350389d1b08SAneesh Kumar K.V bh->b_end_io = end_buffer_read_sync; 3351389d1b08SAneesh Kumar K.V submit_bh(READ, bh); 3352389d1b08SAneesh Kumar K.V wait_on_buffer(bh); 3353389d1b08SAneesh Kumar K.V if (buffer_uptodate(bh)) 3354389d1b08SAneesh Kumar K.V return 0; 3355389d1b08SAneesh Kumar K.V return -EIO; 3356389d1b08SAneesh Kumar K.V } 3357389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_submit_read); 3358389d1b08SAneesh Kumar K.V 33591da177e4SLinus Torvalds void __init buffer_init(void) 33601da177e4SLinus Torvalds { 336143be594aSZhang Yanfei unsigned long nrpages; 33621da177e4SLinus Torvalds 3363b98938c3SChristoph Lameter bh_cachep = kmem_cache_create("buffer_head", 3364b98938c3SChristoph Lameter sizeof(struct buffer_head), 0, 3365b98938c3SChristoph Lameter (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| 3366b98938c3SChristoph Lameter SLAB_MEM_SPREAD), 3367019b4d12SRichard Kennedy NULL); 33681da177e4SLinus Torvalds 33691da177e4SLinus Torvalds /* 33701da177e4SLinus Torvalds * Limit the bh occupancy to 10% of ZONE_NORMAL 33711da177e4SLinus Torvalds */ 33721da177e4SLinus Torvalds nrpages = (nr_free_buffer_pages() * 10) / 100; 33731da177e4SLinus Torvalds max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); 33741da177e4SLinus Torvalds hotcpu_notifier(buffer_cpu_notify, 0); 33751da177e4SLinus Torvalds } 3376