1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/buffer.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 2002 Linus Torvalds 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 91da177e4SLinus Torvalds * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95 101da177e4SLinus Torvalds * 111da177e4SLinus Torvalds * Removed a lot of unnecessary code and simplified things now that 121da177e4SLinus Torvalds * the buffer cache isn't our primary cache - Andrew Tridgell 12/96 131da177e4SLinus Torvalds * 141da177e4SLinus Torvalds * Speed up hash, lru, and free list operations. Use gfp() for allocating 151da177e4SLinus Torvalds * hash table, use SLAB cache for buffer heads. SMP threading. -DaveM 161da177e4SLinus Torvalds * 171da177e4SLinus Torvalds * Added 32k buffer block sizes - these are required older ARM systems. - RMK 181da177e4SLinus Torvalds * 191da177e4SLinus Torvalds * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de> 201da177e4SLinus Torvalds */ 211da177e4SLinus Torvalds 221da177e4SLinus Torvalds #include <linux/kernel.h> 23f361bf4aSIngo Molnar #include <linux/sched/signal.h> 241da177e4SLinus Torvalds #include <linux/syscalls.h> 251da177e4SLinus Torvalds #include <linux/fs.h> 26ae259a9cSChristoph Hellwig #include <linux/iomap.h> 271da177e4SLinus Torvalds #include <linux/mm.h> 281da177e4SLinus Torvalds #include <linux/percpu.h> 291da177e4SLinus Torvalds #include <linux/slab.h> 3016f7e0feSRandy Dunlap #include <linux/capability.h> 311da177e4SLinus Torvalds #include <linux/blkdev.h> 321da177e4SLinus Torvalds #include <linux/file.h> 331da177e4SLinus Torvalds #include <linux/quotaops.h> 341da177e4SLinus Torvalds #include <linux/highmem.h> 35630d9c47SPaul Gortmaker #include <linux/export.h> 36bafc0dbaSTejun Heo #include <linux/backing-dev.h> 371da177e4SLinus Torvalds #include <linux/writeback.h> 381da177e4SLinus Torvalds #include <linux/hash.h> 391da177e4SLinus Torvalds #include <linux/suspend.h> 401da177e4SLinus Torvalds #include <linux/buffer_head.h> 4155e829afSAndrew Morton #include <linux/task_io_accounting_ops.h> 421da177e4SLinus Torvalds #include <linux/bio.h> 431da177e4SLinus Torvalds #include <linux/cpu.h> 441da177e4SLinus Torvalds #include <linux/bitops.h> 451da177e4SLinus Torvalds #include <linux/mpage.h> 46fb1c8f93SIngo Molnar #include <linux/bit_spinlock.h> 4729f3ad7dSJan Kara #include <linux/pagevec.h> 48f745c6f5SShakeel Butt #include <linux/sched/mm.h> 495305cb83STejun Heo #include <trace/events/block.h> 5031fb992cSEric Biggers #include <linux/fscrypt.h> 514fa512ceSEric Biggers #include <linux/fsverity.h> 521da177e4SLinus Torvalds 532b211dc0SBen Dooks #include "internal.h" 542b211dc0SBen Dooks 551da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); 565bdf402aSRitesh Harjani (IBM) static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh, 571420c4a5SBart Van Assche struct writeback_control *wbc); 581da177e4SLinus Torvalds 591da177e4SLinus Torvalds #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) 601da177e4SLinus Torvalds 61f0059afdSTejun Heo inline void touch_buffer(struct buffer_head *bh) 62f0059afdSTejun Heo { 635305cb83STejun Heo trace_block_touch_buffer(bh); 6403c5f331SMatthew Wilcox (Oracle) folio_mark_accessed(bh->b_folio); 65f0059afdSTejun Heo } 66f0059afdSTejun Heo EXPORT_SYMBOL(touch_buffer); 67f0059afdSTejun Heo 68fc9b52cdSHarvey Harrison void __lock_buffer(struct buffer_head *bh) 691da177e4SLinus Torvalds { 7074316201SNeilBrown wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); 711da177e4SLinus Torvalds } 721da177e4SLinus Torvalds EXPORT_SYMBOL(__lock_buffer); 731da177e4SLinus Torvalds 74fc9b52cdSHarvey Harrison void unlock_buffer(struct buffer_head *bh) 751da177e4SLinus Torvalds { 7651b07fc3SNick Piggin clear_bit_unlock(BH_Lock, &bh->b_state); 774e857c58SPeter Zijlstra smp_mb__after_atomic(); 781da177e4SLinus Torvalds wake_up_bit(&bh->b_state, BH_Lock); 791da177e4SLinus Torvalds } 801fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(unlock_buffer); 811da177e4SLinus Torvalds 821da177e4SLinus Torvalds /* 83520f301cSMatthew Wilcox (Oracle) * Returns if the folio has dirty or writeback buffers. If all the buffers 84520f301cSMatthew Wilcox (Oracle) * are unlocked and clean then the folio_test_dirty information is stale. If 85520f301cSMatthew Wilcox (Oracle) * any of the buffers are locked, it is assumed they are locked for IO. 86b4597226SMel Gorman */ 87520f301cSMatthew Wilcox (Oracle) void buffer_check_dirty_writeback(struct folio *folio, 88b4597226SMel Gorman bool *dirty, bool *writeback) 89b4597226SMel Gorman { 90b4597226SMel Gorman struct buffer_head *head, *bh; 91b4597226SMel Gorman *dirty = false; 92b4597226SMel Gorman *writeback = false; 93b4597226SMel Gorman 94520f301cSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 95b4597226SMel Gorman 96520f301cSMatthew Wilcox (Oracle) head = folio_buffers(folio); 97520f301cSMatthew Wilcox (Oracle) if (!head) 98b4597226SMel Gorman return; 99b4597226SMel Gorman 100520f301cSMatthew Wilcox (Oracle) if (folio_test_writeback(folio)) 101b4597226SMel Gorman *writeback = true; 102b4597226SMel Gorman 103b4597226SMel Gorman bh = head; 104b4597226SMel Gorman do { 105b4597226SMel Gorman if (buffer_locked(bh)) 106b4597226SMel Gorman *writeback = true; 107b4597226SMel Gorman 108b4597226SMel Gorman if (buffer_dirty(bh)) 109b4597226SMel Gorman *dirty = true; 110b4597226SMel Gorman 111b4597226SMel Gorman bh = bh->b_this_page; 112b4597226SMel Gorman } while (bh != head); 113b4597226SMel Gorman } 114b4597226SMel Gorman EXPORT_SYMBOL(buffer_check_dirty_writeback); 115b4597226SMel Gorman 116b4597226SMel Gorman /* 1171da177e4SLinus Torvalds * Block until a buffer comes unlocked. This doesn't stop it 1181da177e4SLinus Torvalds * from becoming locked again - you have to lock it yourself 1191da177e4SLinus Torvalds * if you want to preserve its state. 1201da177e4SLinus Torvalds */ 1211da177e4SLinus Torvalds void __wait_on_buffer(struct buffer_head * bh) 1221da177e4SLinus Torvalds { 12374316201SNeilBrown wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); 1241da177e4SLinus Torvalds } 1251fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__wait_on_buffer); 1261da177e4SLinus Torvalds 127b744c2acSRobert Elliott static void buffer_io_error(struct buffer_head *bh, char *msg) 1281da177e4SLinus Torvalds { 129432f16e6SRobert Elliott if (!test_bit(BH_Quiet, &bh->b_state)) 130432f16e6SRobert Elliott printk_ratelimited(KERN_ERR 131a1c6f057SDmitry Monakhov "Buffer I/O error on dev %pg, logical block %llu%s\n", 132a1c6f057SDmitry Monakhov bh->b_bdev, (unsigned long long)bh->b_blocknr, msg); 1331da177e4SLinus Torvalds } 1341da177e4SLinus Torvalds 1351da177e4SLinus Torvalds /* 13668671f35SDmitry Monakhov * End-of-IO handler helper function which does not touch the bh after 13768671f35SDmitry Monakhov * unlocking it. 13868671f35SDmitry Monakhov * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but 13968671f35SDmitry Monakhov * a race there is benign: unlock_buffer() only use the bh's address for 14068671f35SDmitry Monakhov * hashing after unlocking the buffer, so it doesn't actually touch the bh 14168671f35SDmitry Monakhov * itself. 1421da177e4SLinus Torvalds */ 14368671f35SDmitry Monakhov static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) 1441da177e4SLinus Torvalds { 1451da177e4SLinus Torvalds if (uptodate) { 1461da177e4SLinus Torvalds set_buffer_uptodate(bh); 1471da177e4SLinus Torvalds } else { 14870246286SChristoph Hellwig /* This happens, due to failed read-ahead attempts. */ 1491da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1501da177e4SLinus Torvalds } 1511da177e4SLinus Torvalds unlock_buffer(bh); 15268671f35SDmitry Monakhov } 15368671f35SDmitry Monakhov 15468671f35SDmitry Monakhov /* 15568671f35SDmitry Monakhov * Default synchronous end-of-IO handler.. Just mark it up-to-date and 15679f59784SZhang Yi * unlock the buffer. 15768671f35SDmitry Monakhov */ 15868671f35SDmitry Monakhov void end_buffer_read_sync(struct buffer_head *bh, int uptodate) 15968671f35SDmitry Monakhov { 16068671f35SDmitry Monakhov __end_buffer_read_notouch(bh, uptodate); 1611da177e4SLinus Torvalds put_bh(bh); 1621da177e4SLinus Torvalds } 1631fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_read_sync); 1641da177e4SLinus Torvalds 1651da177e4SLinus Torvalds void end_buffer_write_sync(struct buffer_head *bh, int uptodate) 1661da177e4SLinus Torvalds { 1671da177e4SLinus Torvalds if (uptodate) { 1681da177e4SLinus Torvalds set_buffer_uptodate(bh); 1691da177e4SLinus Torvalds } else { 170b744c2acSRobert Elliott buffer_io_error(bh, ", lost sync page write"); 17187354e5dSJeff Layton mark_buffer_write_io_error(bh); 1721da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1731da177e4SLinus Torvalds } 1741da177e4SLinus Torvalds unlock_buffer(bh); 1751da177e4SLinus Torvalds put_bh(bh); 1761da177e4SLinus Torvalds } 1771fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_write_sync); 1781da177e4SLinus Torvalds 1791da177e4SLinus Torvalds /* 1801da177e4SLinus Torvalds * Various filesystems appear to want __find_get_block to be non-blocking. 1811da177e4SLinus Torvalds * But it's the page lock which protects the buffers. To get around this, 1821da177e4SLinus Torvalds * we get exclusion from try_to_free_buffers with the blockdev mapping's 1831da177e4SLinus Torvalds * private_lock. 1841da177e4SLinus Torvalds * 185b93b0163SMatthew Wilcox * Hack idea: for the blockdev mapping, private_lock contention 1861da177e4SLinus Torvalds * may be quite high. This code could TryLock the page, and if that 187b93b0163SMatthew Wilcox * succeeds, there is no need to take private_lock. 1881da177e4SLinus Torvalds */ 1891da177e4SLinus Torvalds static struct buffer_head * 190385fd4c5SCoywolf Qi Hunt __find_get_block_slow(struct block_device *bdev, sector_t block) 1911da177e4SLinus Torvalds { 1921da177e4SLinus Torvalds struct inode *bd_inode = bdev->bd_inode; 1931da177e4SLinus Torvalds struct address_space *bd_mapping = bd_inode->i_mapping; 1941da177e4SLinus Torvalds struct buffer_head *ret = NULL; 1951da177e4SLinus Torvalds pgoff_t index; 1961da177e4SLinus Torvalds struct buffer_head *bh; 1971da177e4SLinus Torvalds struct buffer_head *head; 1981da177e4SLinus Torvalds struct page *page; 1991da177e4SLinus Torvalds int all_mapped = 1; 20043636c80STetsuo Handa static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1); 2011da177e4SLinus Torvalds 20209cbfeafSKirill A. Shutemov index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); 2032457aec6SMel Gorman page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED); 2041da177e4SLinus Torvalds if (!page) 2051da177e4SLinus Torvalds goto out; 2061da177e4SLinus Torvalds 2071da177e4SLinus Torvalds spin_lock(&bd_mapping->private_lock); 2081da177e4SLinus Torvalds if (!page_has_buffers(page)) 2091da177e4SLinus Torvalds goto out_unlock; 2101da177e4SLinus Torvalds head = page_buffers(page); 2111da177e4SLinus Torvalds bh = head; 2121da177e4SLinus Torvalds do { 21397f76d3dSNikanth Karthikesan if (!buffer_mapped(bh)) 21497f76d3dSNikanth Karthikesan all_mapped = 0; 21597f76d3dSNikanth Karthikesan else if (bh->b_blocknr == block) { 2161da177e4SLinus Torvalds ret = bh; 2171da177e4SLinus Torvalds get_bh(bh); 2181da177e4SLinus Torvalds goto out_unlock; 2191da177e4SLinus Torvalds } 2201da177e4SLinus Torvalds bh = bh->b_this_page; 2211da177e4SLinus Torvalds } while (bh != head); 2221da177e4SLinus Torvalds 2231da177e4SLinus Torvalds /* we might be here because some of the buffers on this page are 2241da177e4SLinus Torvalds * not mapped. This is due to various races between 2251da177e4SLinus Torvalds * file io on the block device and getblk. It gets dealt with 2261da177e4SLinus Torvalds * elsewhere, don't buffer_error if we had some unmapped buffers 2271da177e4SLinus Torvalds */ 22843636c80STetsuo Handa ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE); 22943636c80STetsuo Handa if (all_mapped && __ratelimit(&last_warned)) { 23043636c80STetsuo Handa printk("__find_get_block_slow() failed. block=%llu, " 23143636c80STetsuo Handa "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, " 23243636c80STetsuo Handa "device %pg blocksize: %d\n", 233205f87f6SBadari Pulavarty (unsigned long long)block, 23443636c80STetsuo Handa (unsigned long long)bh->b_blocknr, 23543636c80STetsuo Handa bh->b_state, bh->b_size, bdev, 23672a2ebd8STao Ma 1 << bd_inode->i_blkbits); 2371da177e4SLinus Torvalds } 2381da177e4SLinus Torvalds out_unlock: 2391da177e4SLinus Torvalds spin_unlock(&bd_mapping->private_lock); 24009cbfeafSKirill A. Shutemov put_page(page); 2411da177e4SLinus Torvalds out: 2421da177e4SLinus Torvalds return ret; 2431da177e4SLinus Torvalds } 2441da177e4SLinus Torvalds 2451da177e4SLinus Torvalds static void end_buffer_async_read(struct buffer_head *bh, int uptodate) 2461da177e4SLinus Torvalds { 2471da177e4SLinus Torvalds unsigned long flags; 248a3972203SNick Piggin struct buffer_head *first; 2491da177e4SLinus Torvalds struct buffer_head *tmp; 2502e2dba15SMatthew Wilcox (Oracle) struct folio *folio; 2512e2dba15SMatthew Wilcox (Oracle) int folio_uptodate = 1; 2521da177e4SLinus Torvalds 2531da177e4SLinus Torvalds BUG_ON(!buffer_async_read(bh)); 2541da177e4SLinus Torvalds 2552e2dba15SMatthew Wilcox (Oracle) folio = bh->b_folio; 2561da177e4SLinus Torvalds if (uptodate) { 2571da177e4SLinus Torvalds set_buffer_uptodate(bh); 2581da177e4SLinus Torvalds } else { 2591da177e4SLinus Torvalds clear_buffer_uptodate(bh); 260b744c2acSRobert Elliott buffer_io_error(bh, ", async page read"); 2612e2dba15SMatthew Wilcox (Oracle) folio_set_error(folio); 2621da177e4SLinus Torvalds } 2631da177e4SLinus Torvalds 2641da177e4SLinus Torvalds /* 2651da177e4SLinus Torvalds * Be _very_ careful from here on. Bad things can happen if 2661da177e4SLinus Torvalds * two buffer heads end IO at almost the same time and both 2671da177e4SLinus Torvalds * decide that the page is now completely done. 2681da177e4SLinus Torvalds */ 2692e2dba15SMatthew Wilcox (Oracle) first = folio_buffers(folio); 270f1e67e35SThomas Gleixner spin_lock_irqsave(&first->b_uptodate_lock, flags); 2711da177e4SLinus Torvalds clear_buffer_async_read(bh); 2721da177e4SLinus Torvalds unlock_buffer(bh); 2731da177e4SLinus Torvalds tmp = bh; 2741da177e4SLinus Torvalds do { 2751da177e4SLinus Torvalds if (!buffer_uptodate(tmp)) 2762e2dba15SMatthew Wilcox (Oracle) folio_uptodate = 0; 2771da177e4SLinus Torvalds if (buffer_async_read(tmp)) { 2781da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 2791da177e4SLinus Torvalds goto still_busy; 2801da177e4SLinus Torvalds } 2811da177e4SLinus Torvalds tmp = tmp->b_this_page; 2821da177e4SLinus Torvalds } while (tmp != bh); 283f1e67e35SThomas Gleixner spin_unlock_irqrestore(&first->b_uptodate_lock, flags); 2841da177e4SLinus Torvalds 2851da177e4SLinus Torvalds /* 2866e8e79fcSMatthew Wilcox (Oracle) * If all of the buffers are uptodate then we can set the page 2876e8e79fcSMatthew Wilcox (Oracle) * uptodate. 2881da177e4SLinus Torvalds */ 2892e2dba15SMatthew Wilcox (Oracle) if (folio_uptodate) 2902e2dba15SMatthew Wilcox (Oracle) folio_mark_uptodate(folio); 2912e2dba15SMatthew Wilcox (Oracle) folio_unlock(folio); 2921da177e4SLinus Torvalds return; 2931da177e4SLinus Torvalds 2941da177e4SLinus Torvalds still_busy: 295f1e67e35SThomas Gleixner spin_unlock_irqrestore(&first->b_uptodate_lock, flags); 2961da177e4SLinus Torvalds return; 2971da177e4SLinus Torvalds } 2981da177e4SLinus Torvalds 2994fa512ceSEric Biggers struct postprocess_bh_ctx { 30031fb992cSEric Biggers struct work_struct work; 30131fb992cSEric Biggers struct buffer_head *bh; 30231fb992cSEric Biggers }; 30331fb992cSEric Biggers 3044fa512ceSEric Biggers static void verify_bh(struct work_struct *work) 3054fa512ceSEric Biggers { 3064fa512ceSEric Biggers struct postprocess_bh_ctx *ctx = 3074fa512ceSEric Biggers container_of(work, struct postprocess_bh_ctx, work); 3084fa512ceSEric Biggers struct buffer_head *bh = ctx->bh; 3094fa512ceSEric Biggers bool valid; 3104fa512ceSEric Biggers 3118b7d3fe9SEric Biggers valid = fsverity_verify_blocks(bh->b_folio, bh->b_size, bh_offset(bh)); 3124fa512ceSEric Biggers end_buffer_async_read(bh, valid); 3134fa512ceSEric Biggers kfree(ctx); 3144fa512ceSEric Biggers } 3154fa512ceSEric Biggers 3164fa512ceSEric Biggers static bool need_fsverity(struct buffer_head *bh) 3174fa512ceSEric Biggers { 3188b7d3fe9SEric Biggers struct folio *folio = bh->b_folio; 3198b7d3fe9SEric Biggers struct inode *inode = folio->mapping->host; 3204fa512ceSEric Biggers 3214fa512ceSEric Biggers return fsverity_active(inode) && 3224fa512ceSEric Biggers /* needed by ext4 */ 3238b7d3fe9SEric Biggers folio->index < DIV_ROUND_UP(inode->i_size, PAGE_SIZE); 3244fa512ceSEric Biggers } 3254fa512ceSEric Biggers 32631fb992cSEric Biggers static void decrypt_bh(struct work_struct *work) 32731fb992cSEric Biggers { 3284fa512ceSEric Biggers struct postprocess_bh_ctx *ctx = 3294fa512ceSEric Biggers container_of(work, struct postprocess_bh_ctx, work); 33031fb992cSEric Biggers struct buffer_head *bh = ctx->bh; 33131fb992cSEric Biggers int err; 33231fb992cSEric Biggers 3339c7fb7f7SEric Biggers err = fscrypt_decrypt_pagecache_blocks(bh->b_folio, bh->b_size, 3349c7fb7f7SEric Biggers bh_offset(bh)); 3354fa512ceSEric Biggers if (err == 0 && need_fsverity(bh)) { 3364fa512ceSEric Biggers /* 3374fa512ceSEric Biggers * We use different work queues for decryption and for verity 3384fa512ceSEric Biggers * because verity may require reading metadata pages that need 3394fa512ceSEric Biggers * decryption, and we shouldn't recurse to the same workqueue. 3404fa512ceSEric Biggers */ 3414fa512ceSEric Biggers INIT_WORK(&ctx->work, verify_bh); 3424fa512ceSEric Biggers fsverity_enqueue_verify_work(&ctx->work); 3434fa512ceSEric Biggers return; 3444fa512ceSEric Biggers } 34531fb992cSEric Biggers end_buffer_async_read(bh, err == 0); 34631fb992cSEric Biggers kfree(ctx); 34731fb992cSEric Biggers } 34831fb992cSEric Biggers 34931fb992cSEric Biggers /* 3502c69e205SMatthew Wilcox (Oracle) * I/O completion handler for block_read_full_folio() - pages 35131fb992cSEric Biggers * which come unlocked at the end of I/O. 35231fb992cSEric Biggers */ 35331fb992cSEric Biggers static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate) 35431fb992cSEric Biggers { 3553822a7c4SLinus Torvalds struct inode *inode = bh->b_folio->mapping->host; 3564fa512ceSEric Biggers bool decrypt = fscrypt_inode_uses_fs_layer_crypto(inode); 3574fa512ceSEric Biggers bool verify = need_fsverity(bh); 3584fa512ceSEric Biggers 3594fa512ceSEric Biggers /* Decrypt (with fscrypt) and/or verify (with fsverity) if needed. */ 3604fa512ceSEric Biggers if (uptodate && (decrypt || verify)) { 3614fa512ceSEric Biggers struct postprocess_bh_ctx *ctx = 3624fa512ceSEric Biggers kmalloc(sizeof(*ctx), GFP_ATOMIC); 36331fb992cSEric Biggers 36431fb992cSEric Biggers if (ctx) { 36531fb992cSEric Biggers ctx->bh = bh; 3664fa512ceSEric Biggers if (decrypt) { 3674fa512ceSEric Biggers INIT_WORK(&ctx->work, decrypt_bh); 36831fb992cSEric Biggers fscrypt_enqueue_decrypt_work(&ctx->work); 3694fa512ceSEric Biggers } else { 3704fa512ceSEric Biggers INIT_WORK(&ctx->work, verify_bh); 3714fa512ceSEric Biggers fsverity_enqueue_verify_work(&ctx->work); 3724fa512ceSEric Biggers } 37331fb992cSEric Biggers return; 37431fb992cSEric Biggers } 37531fb992cSEric Biggers uptodate = 0; 37631fb992cSEric Biggers } 37731fb992cSEric Biggers end_buffer_async_read(bh, uptodate); 37831fb992cSEric Biggers } 37931fb992cSEric Biggers 3801da177e4SLinus Torvalds /* 3811da177e4SLinus Torvalds * Completion handler for block_write_full_page() - pages which are unlocked 3821da177e4SLinus Torvalds * during I/O, and which have PageWriteback cleared upon I/O completion. 3831da177e4SLinus Torvalds */ 38435c80d5fSChris Mason void end_buffer_async_write(struct buffer_head *bh, int uptodate) 3851da177e4SLinus Torvalds { 3861da177e4SLinus Torvalds unsigned long flags; 387a3972203SNick Piggin struct buffer_head *first; 3881da177e4SLinus Torvalds struct buffer_head *tmp; 389743ed81eSMatthew Wilcox (Oracle) struct folio *folio; 3901da177e4SLinus Torvalds 3911da177e4SLinus Torvalds BUG_ON(!buffer_async_write(bh)); 3921da177e4SLinus Torvalds 393743ed81eSMatthew Wilcox (Oracle) folio = bh->b_folio; 3941da177e4SLinus Torvalds if (uptodate) { 3951da177e4SLinus Torvalds set_buffer_uptodate(bh); 3961da177e4SLinus Torvalds } else { 397b744c2acSRobert Elliott buffer_io_error(bh, ", lost async page write"); 39887354e5dSJeff Layton mark_buffer_write_io_error(bh); 3991da177e4SLinus Torvalds clear_buffer_uptodate(bh); 400743ed81eSMatthew Wilcox (Oracle) folio_set_error(folio); 4011da177e4SLinus Torvalds } 4021da177e4SLinus Torvalds 403743ed81eSMatthew Wilcox (Oracle) first = folio_buffers(folio); 404f1e67e35SThomas Gleixner spin_lock_irqsave(&first->b_uptodate_lock, flags); 405a3972203SNick Piggin 4061da177e4SLinus Torvalds clear_buffer_async_write(bh); 4071da177e4SLinus Torvalds unlock_buffer(bh); 4081da177e4SLinus Torvalds tmp = bh->b_this_page; 4091da177e4SLinus Torvalds while (tmp != bh) { 4101da177e4SLinus Torvalds if (buffer_async_write(tmp)) { 4111da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 4121da177e4SLinus Torvalds goto still_busy; 4131da177e4SLinus Torvalds } 4141da177e4SLinus Torvalds tmp = tmp->b_this_page; 4151da177e4SLinus Torvalds } 416f1e67e35SThomas Gleixner spin_unlock_irqrestore(&first->b_uptodate_lock, flags); 417743ed81eSMatthew Wilcox (Oracle) folio_end_writeback(folio); 4181da177e4SLinus Torvalds return; 4191da177e4SLinus Torvalds 4201da177e4SLinus Torvalds still_busy: 421f1e67e35SThomas Gleixner spin_unlock_irqrestore(&first->b_uptodate_lock, flags); 4221da177e4SLinus Torvalds return; 4231da177e4SLinus Torvalds } 4241fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_async_write); 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds /* 4271da177e4SLinus Torvalds * If a page's buffers are under async readin (end_buffer_async_read 4281da177e4SLinus Torvalds * completion) then there is a possibility that another thread of 4291da177e4SLinus Torvalds * control could lock one of the buffers after it has completed 4301da177e4SLinus Torvalds * but while some of the other buffers have not completed. This 4311da177e4SLinus Torvalds * locked buffer would confuse end_buffer_async_read() into not unlocking 4321da177e4SLinus Torvalds * the page. So the absence of BH_Async_Read tells end_buffer_async_read() 4331da177e4SLinus Torvalds * that this buffer is not under async I/O. 4341da177e4SLinus Torvalds * 4351da177e4SLinus Torvalds * The page comes unlocked when it has no locked buffer_async buffers 4361da177e4SLinus Torvalds * left. 4371da177e4SLinus Torvalds * 4381da177e4SLinus Torvalds * PageLocked prevents anyone starting new async I/O reads any of 4391da177e4SLinus Torvalds * the buffers. 4401da177e4SLinus Torvalds * 4411da177e4SLinus Torvalds * PageWriteback is used to prevent simultaneous writeout of the same 4421da177e4SLinus Torvalds * page. 4431da177e4SLinus Torvalds * 4441da177e4SLinus Torvalds * PageLocked prevents anyone from starting writeback of a page which is 4451da177e4SLinus Torvalds * under read I/O (PageWriteback is only ever set against a locked page). 4461da177e4SLinus Torvalds */ 4471da177e4SLinus Torvalds static void mark_buffer_async_read(struct buffer_head *bh) 4481da177e4SLinus Torvalds { 44931fb992cSEric Biggers bh->b_end_io = end_buffer_async_read_io; 4501da177e4SLinus Torvalds set_buffer_async_read(bh); 4511da177e4SLinus Torvalds } 4521da177e4SLinus Torvalds 4531fe72eaaSH Hartley Sweeten static void mark_buffer_async_write_endio(struct buffer_head *bh, 45435c80d5fSChris Mason bh_end_io_t *handler) 45535c80d5fSChris Mason { 45635c80d5fSChris Mason bh->b_end_io = handler; 45735c80d5fSChris Mason set_buffer_async_write(bh); 45835c80d5fSChris Mason } 45935c80d5fSChris Mason 4601da177e4SLinus Torvalds void mark_buffer_async_write(struct buffer_head *bh) 4611da177e4SLinus Torvalds { 46235c80d5fSChris Mason mark_buffer_async_write_endio(bh, end_buffer_async_write); 4631da177e4SLinus Torvalds } 4641da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_async_write); 4651da177e4SLinus Torvalds 4661da177e4SLinus Torvalds 4671da177e4SLinus Torvalds /* 4681da177e4SLinus Torvalds * fs/buffer.c contains helper functions for buffer-backed address space's 4691da177e4SLinus Torvalds * fsync functions. A common requirement for buffer-based filesystems is 4701da177e4SLinus Torvalds * that certain data from the backing blockdev needs to be written out for 4711da177e4SLinus Torvalds * a successful fsync(). For example, ext2 indirect blocks need to be 4721da177e4SLinus Torvalds * written back and waited upon before fsync() returns. 4731da177e4SLinus Torvalds * 4741da177e4SLinus Torvalds * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), 4751da177e4SLinus Torvalds * inode_has_buffers() and invalidate_inode_buffers() are provided for the 4761da177e4SLinus Torvalds * management of a list of dependent buffers at ->i_mapping->private_list. 4771da177e4SLinus Torvalds * 4781da177e4SLinus Torvalds * Locking is a little subtle: try_to_free_buffers() will remove buffers 4791da177e4SLinus Torvalds * from their controlling inode's queue when they are being freed. But 4801da177e4SLinus Torvalds * try_to_free_buffers() will be operating against the *blockdev* mapping 4811da177e4SLinus Torvalds * at the time, not against the S_ISREG file which depends on those buffers. 4821da177e4SLinus Torvalds * So the locking for private_list is via the private_lock in the address_space 4831da177e4SLinus Torvalds * which backs the buffers. Which is different from the address_space 4841da177e4SLinus Torvalds * against which the buffers are listed. So for a particular address_space, 4851da177e4SLinus Torvalds * mapping->private_lock does *not* protect mapping->private_list! In fact, 4861da177e4SLinus Torvalds * mapping->private_list will always be protected by the backing blockdev's 4871da177e4SLinus Torvalds * ->private_lock. 4881da177e4SLinus Torvalds * 4891da177e4SLinus Torvalds * Which introduces a requirement: all buffers on an address_space's 4901da177e4SLinus Torvalds * ->private_list must be from the same address_space: the blockdev's. 4911da177e4SLinus Torvalds * 4921da177e4SLinus Torvalds * address_spaces which do not place buffers at ->private_list via these 4931da177e4SLinus Torvalds * utility functions are free to use private_lock and private_list for 4941da177e4SLinus Torvalds * whatever they want. The only requirement is that list_empty(private_list) 4951da177e4SLinus Torvalds * be true at clear_inode() time. 4961da177e4SLinus Torvalds * 4971da177e4SLinus Torvalds * FIXME: clear_inode should not call invalidate_inode_buffers(). The 4981da177e4SLinus Torvalds * filesystems should do that. invalidate_inode_buffers() should just go 4991da177e4SLinus Torvalds * BUG_ON(!list_empty). 5001da177e4SLinus Torvalds * 5011da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should 5021da177e4SLinus Torvalds * take an address_space, not an inode. And it should be called 5031da177e4SLinus Torvalds * mark_buffer_dirty_fsync() to clearly define why those buffers are being 5041da177e4SLinus Torvalds * queued up. 5051da177e4SLinus Torvalds * 5061da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the 5071da177e4SLinus Torvalds * list if it is already on a list. Because if the buffer is on a list, 5081da177e4SLinus Torvalds * it *must* already be on the right one. If not, the filesystem is being 5091da177e4SLinus Torvalds * silly. This will save a ton of locking. But first we have to ensure 5101da177e4SLinus Torvalds * that buffers are taken *off* the old inode's list when they are freed 5111da177e4SLinus Torvalds * (presumably in truncate). That requires careful auditing of all 5121da177e4SLinus Torvalds * filesystems (do it inside bforget()). It could also be done by bringing 5131da177e4SLinus Torvalds * b_inode back. 5141da177e4SLinus Torvalds */ 5151da177e4SLinus Torvalds 5161da177e4SLinus Torvalds /* 5171da177e4SLinus Torvalds * The buffer's backing address_space's private_lock must be held 5181da177e4SLinus Torvalds */ 519dbacefc9SThomas Petazzoni static void __remove_assoc_queue(struct buffer_head *bh) 5201da177e4SLinus Torvalds { 5211da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 52258ff407bSJan Kara WARN_ON(!bh->b_assoc_map); 52358ff407bSJan Kara bh->b_assoc_map = NULL; 5241da177e4SLinus Torvalds } 5251da177e4SLinus Torvalds 5261da177e4SLinus Torvalds int inode_has_buffers(struct inode *inode) 5271da177e4SLinus Torvalds { 5281da177e4SLinus Torvalds return !list_empty(&inode->i_data.private_list); 5291da177e4SLinus Torvalds } 5301da177e4SLinus Torvalds 5311da177e4SLinus Torvalds /* 5321da177e4SLinus Torvalds * osync is designed to support O_SYNC io. It waits synchronously for 5331da177e4SLinus Torvalds * all already-submitted IO to complete, but does not queue any new 5341da177e4SLinus Torvalds * writes to the disk. 5351da177e4SLinus Torvalds * 53679f59784SZhang Yi * To do O_SYNC writes, just queue the buffer writes with write_dirty_buffer 53779f59784SZhang Yi * as you dirty the buffers, and then use osync_inode_buffers to wait for 5381da177e4SLinus Torvalds * completion. Any other dirty buffers which are not yet queued for 5391da177e4SLinus Torvalds * write will not be flushed to disk by the osync. 5401da177e4SLinus Torvalds */ 5411da177e4SLinus Torvalds static int osync_buffers_list(spinlock_t *lock, struct list_head *list) 5421da177e4SLinus Torvalds { 5431da177e4SLinus Torvalds struct buffer_head *bh; 5441da177e4SLinus Torvalds struct list_head *p; 5451da177e4SLinus Torvalds int err = 0; 5461da177e4SLinus Torvalds 5471da177e4SLinus Torvalds spin_lock(lock); 5481da177e4SLinus Torvalds repeat: 5491da177e4SLinus Torvalds list_for_each_prev(p, list) { 5501da177e4SLinus Torvalds bh = BH_ENTRY(p); 5511da177e4SLinus Torvalds if (buffer_locked(bh)) { 5521da177e4SLinus Torvalds get_bh(bh); 5531da177e4SLinus Torvalds spin_unlock(lock); 5541da177e4SLinus Torvalds wait_on_buffer(bh); 5551da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 5561da177e4SLinus Torvalds err = -EIO; 5571da177e4SLinus Torvalds brelse(bh); 5581da177e4SLinus Torvalds spin_lock(lock); 5591da177e4SLinus Torvalds goto repeat; 5601da177e4SLinus Torvalds } 5611da177e4SLinus Torvalds } 5621da177e4SLinus Torvalds spin_unlock(lock); 5631da177e4SLinus Torvalds return err; 5641da177e4SLinus Torvalds } 5651da177e4SLinus Torvalds 56608fdc8a0SMateusz Guzik void emergency_thaw_bdev(struct super_block *sb) 567c2d75438SEric Sandeen { 568040f04bdSChristoph Hellwig while (sb->s_bdev && !thaw_bdev(sb->s_bdev)) 569a1c6f057SDmitry Monakhov printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev); 570c2d75438SEric Sandeen } 57101a05b33SAl Viro 5721da177e4SLinus Torvalds /** 57378a4a50aSRandy Dunlap * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers 57467be2dd1SMartin Waitz * @mapping: the mapping which wants those buffers written 5751da177e4SLinus Torvalds * 5761da177e4SLinus Torvalds * Starts I/O against the buffers at mapping->private_list, and waits upon 5771da177e4SLinus Torvalds * that I/O. 5781da177e4SLinus Torvalds * 57967be2dd1SMartin Waitz * Basically, this is a convenience function for fsync(). 58067be2dd1SMartin Waitz * @mapping is a file or directory which needs those buffers to be written for 58167be2dd1SMartin Waitz * a successful fsync(). 5821da177e4SLinus Torvalds */ 5831da177e4SLinus Torvalds int sync_mapping_buffers(struct address_space *mapping) 5841da177e4SLinus Torvalds { 585252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 5861da177e4SLinus Torvalds 5871da177e4SLinus Torvalds if (buffer_mapping == NULL || list_empty(&mapping->private_list)) 5881da177e4SLinus Torvalds return 0; 5891da177e4SLinus Torvalds 5901da177e4SLinus Torvalds return fsync_buffers_list(&buffer_mapping->private_lock, 5911da177e4SLinus Torvalds &mapping->private_list); 5921da177e4SLinus Torvalds } 5931da177e4SLinus Torvalds EXPORT_SYMBOL(sync_mapping_buffers); 5941da177e4SLinus Torvalds 5951da177e4SLinus Torvalds /* 5961da177e4SLinus Torvalds * Called when we've recently written block `bblock', and it is known that 5971da177e4SLinus Torvalds * `bblock' was for a buffer_boundary() buffer. This means that the block at 5981da177e4SLinus Torvalds * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's 5991da177e4SLinus Torvalds * dirty, schedule it for IO. So that indirects merge nicely with their data. 6001da177e4SLinus Torvalds */ 6011da177e4SLinus Torvalds void write_boundary_block(struct block_device *bdev, 6021da177e4SLinus Torvalds sector_t bblock, unsigned blocksize) 6031da177e4SLinus Torvalds { 6041da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); 6051da177e4SLinus Torvalds if (bh) { 6061da177e4SLinus Torvalds if (buffer_dirty(bh)) 607e7ea1129SZhang Yi write_dirty_buffer(bh, 0); 6081da177e4SLinus Torvalds put_bh(bh); 6091da177e4SLinus Torvalds } 6101da177e4SLinus Torvalds } 6111da177e4SLinus Torvalds 6121da177e4SLinus Torvalds void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) 6131da177e4SLinus Torvalds { 6141da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 615abc8a8a2SMatthew Wilcox (Oracle) struct address_space *buffer_mapping = bh->b_folio->mapping; 6161da177e4SLinus Torvalds 6171da177e4SLinus Torvalds mark_buffer_dirty(bh); 618252aa6f5SRafael Aquini if (!mapping->private_data) { 619252aa6f5SRafael Aquini mapping->private_data = buffer_mapping; 6201da177e4SLinus Torvalds } else { 621252aa6f5SRafael Aquini BUG_ON(mapping->private_data != buffer_mapping); 6221da177e4SLinus Torvalds } 623535ee2fbSJan Kara if (!bh->b_assoc_map) { 6241da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 6251da177e4SLinus Torvalds list_move_tail(&bh->b_assoc_buffers, 6261da177e4SLinus Torvalds &mapping->private_list); 62758ff407bSJan Kara bh->b_assoc_map = mapping; 6281da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 6291da177e4SLinus Torvalds } 6301da177e4SLinus Torvalds } 6311da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_dirty_inode); 6321da177e4SLinus Torvalds 6331da177e4SLinus Torvalds /* 6341da177e4SLinus Torvalds * Add a page to the dirty page list. 6351da177e4SLinus Torvalds * 6361da177e4SLinus Torvalds * It is a sad fact of life that this function is called from several places 6371da177e4SLinus Torvalds * deeply under spinlocking. It may not sleep. 6381da177e4SLinus Torvalds * 6391da177e4SLinus Torvalds * If the page has buffers, the uptodate buffers are set dirty, to preserve 6401da177e4SLinus Torvalds * dirty-state coherency between the page and the buffers. It the page does 6411da177e4SLinus Torvalds * not have buffers then when they are later attached they will all be set 6421da177e4SLinus Torvalds * dirty. 6431da177e4SLinus Torvalds * 6441da177e4SLinus Torvalds * The buffers are dirtied before the page is dirtied. There's a small race 6451da177e4SLinus Torvalds * window in which a writepage caller may see the page cleanness but not the 6461da177e4SLinus Torvalds * buffer dirtiness. That's fine. If this code were to set the page dirty 6471da177e4SLinus Torvalds * before the buffers, a concurrent writepage caller could clear the page dirty 6481da177e4SLinus Torvalds * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean 6491da177e4SLinus Torvalds * page on the dirty page list. 6501da177e4SLinus Torvalds * 6511da177e4SLinus Torvalds * We use private_lock to lock against try_to_free_buffers while using the 6521da177e4SLinus Torvalds * page's buffer list. Also use this to protect against clean buffers being 6531da177e4SLinus Torvalds * added to the page after it was set dirty. 6541da177e4SLinus Torvalds * 6551da177e4SLinus Torvalds * FIXME: may need to call ->reservepage here as well. That's rather up to the 6561da177e4SLinus Torvalds * address_space though. 6571da177e4SLinus Torvalds */ 658e621900aSMatthew Wilcox (Oracle) bool block_dirty_folio(struct address_space *mapping, struct folio *folio) 6591da177e4SLinus Torvalds { 660e621900aSMatthew Wilcox (Oracle) struct buffer_head *head; 661e621900aSMatthew Wilcox (Oracle) bool newly_dirty; 6621da177e4SLinus Torvalds 6631da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 664e621900aSMatthew Wilcox (Oracle) head = folio_buffers(folio); 665e621900aSMatthew Wilcox (Oracle) if (head) { 6661da177e4SLinus Torvalds struct buffer_head *bh = head; 6671da177e4SLinus Torvalds 6681da177e4SLinus Torvalds do { 6691da177e4SLinus Torvalds set_buffer_dirty(bh); 6701da177e4SLinus Torvalds bh = bh->b_this_page; 6711da177e4SLinus Torvalds } while (bh != head); 6721da177e4SLinus Torvalds } 673c4843a75SGreg Thelen /* 674bcfe06bfSRoman Gushchin * Lock out page's memcg migration to keep PageDirty 67581f8c3a4SJohannes Weiner * synchronized with per-memcg dirty page counters. 676c4843a75SGreg Thelen */ 677e621900aSMatthew Wilcox (Oracle) folio_memcg_lock(folio); 678e621900aSMatthew Wilcox (Oracle) newly_dirty = !folio_test_set_dirty(folio); 6791da177e4SLinus Torvalds spin_unlock(&mapping->private_lock); 6801da177e4SLinus Torvalds 681a8e7d49aSLinus Torvalds if (newly_dirty) 682e621900aSMatthew Wilcox (Oracle) __folio_mark_dirty(folio, mapping, 1); 683c4843a75SGreg Thelen 684e621900aSMatthew Wilcox (Oracle) folio_memcg_unlock(folio); 685c4843a75SGreg Thelen 686c4843a75SGreg Thelen if (newly_dirty) 687c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 688c4843a75SGreg Thelen 689a8e7d49aSLinus Torvalds return newly_dirty; 6901da177e4SLinus Torvalds } 691e621900aSMatthew Wilcox (Oracle) EXPORT_SYMBOL(block_dirty_folio); 6921da177e4SLinus Torvalds 6931da177e4SLinus Torvalds /* 6941da177e4SLinus Torvalds * Write out and wait upon a list of buffers. 6951da177e4SLinus Torvalds * 6961da177e4SLinus Torvalds * We have conflicting pressures: we want to make sure that all 6971da177e4SLinus Torvalds * initially dirty buffers get waited on, but that any subsequently 6981da177e4SLinus Torvalds * dirtied buffers don't. After all, we don't want fsync to last 6991da177e4SLinus Torvalds * forever if somebody is actively writing to the file. 7001da177e4SLinus Torvalds * 7011da177e4SLinus Torvalds * Do this in two main stages: first we copy dirty buffers to a 7021da177e4SLinus Torvalds * temporary inode list, queueing the writes as we go. Then we clean 7031da177e4SLinus Torvalds * up, waiting for those writes to complete. 7041da177e4SLinus Torvalds * 7051da177e4SLinus Torvalds * During this second stage, any subsequent updates to the file may end 7061da177e4SLinus Torvalds * up refiling the buffer on the original inode's dirty list again, so 7071da177e4SLinus Torvalds * there is a chance we will end up with a buffer queued for write but 7081da177e4SLinus Torvalds * not yet completed on that list. So, as a final cleanup we go through 7091da177e4SLinus Torvalds * the osync code to catch these locked, dirty buffers without requeuing 7101da177e4SLinus Torvalds * any newly dirty buffers for write. 7111da177e4SLinus Torvalds */ 7121da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) 7131da177e4SLinus Torvalds { 7141da177e4SLinus Torvalds struct buffer_head *bh; 7151da177e4SLinus Torvalds struct list_head tmp; 7167eaceaccSJens Axboe struct address_space *mapping; 7171da177e4SLinus Torvalds int err = 0, err2; 7184ee2491eSJens Axboe struct blk_plug plug; 7191da177e4SLinus Torvalds 7201da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 7214ee2491eSJens Axboe blk_start_plug(&plug); 7221da177e4SLinus Torvalds 7231da177e4SLinus Torvalds spin_lock(lock); 7241da177e4SLinus Torvalds while (!list_empty(list)) { 7251da177e4SLinus Torvalds bh = BH_ENTRY(list->next); 726535ee2fbSJan Kara mapping = bh->b_assoc_map; 72758ff407bSJan Kara __remove_assoc_queue(bh); 728535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 729535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 730535ee2fbSJan Kara smp_mb(); 7311da177e4SLinus Torvalds if (buffer_dirty(bh) || buffer_locked(bh)) { 7321da177e4SLinus Torvalds list_add(&bh->b_assoc_buffers, &tmp); 733535ee2fbSJan Kara bh->b_assoc_map = mapping; 7341da177e4SLinus Torvalds if (buffer_dirty(bh)) { 7351da177e4SLinus Torvalds get_bh(bh); 7361da177e4SLinus Torvalds spin_unlock(lock); 7371da177e4SLinus Torvalds /* 7381da177e4SLinus Torvalds * Ensure any pending I/O completes so that 7399cb569d6SChristoph Hellwig * write_dirty_buffer() actually writes the 7409cb569d6SChristoph Hellwig * current contents - it is a noop if I/O is 7419cb569d6SChristoph Hellwig * still in flight on potentially older 7429cb569d6SChristoph Hellwig * contents. 7431da177e4SLinus Torvalds */ 74470fd7614SChristoph Hellwig write_dirty_buffer(bh, REQ_SYNC); 7459cf6b720SJens Axboe 7469cf6b720SJens Axboe /* 7479cf6b720SJens Axboe * Kick off IO for the previous mapping. Note 7489cf6b720SJens Axboe * that we will not run the very last mapping, 7499cf6b720SJens Axboe * wait_on_buffer() will do that for us 7509cf6b720SJens Axboe * through sync_buffer(). 7519cf6b720SJens Axboe */ 7521da177e4SLinus Torvalds brelse(bh); 7531da177e4SLinus Torvalds spin_lock(lock); 7541da177e4SLinus Torvalds } 7551da177e4SLinus Torvalds } 7561da177e4SLinus Torvalds } 7571da177e4SLinus Torvalds 7584ee2491eSJens Axboe spin_unlock(lock); 7594ee2491eSJens Axboe blk_finish_plug(&plug); 7604ee2491eSJens Axboe spin_lock(lock); 7614ee2491eSJens Axboe 7621da177e4SLinus Torvalds while (!list_empty(&tmp)) { 7631da177e4SLinus Torvalds bh = BH_ENTRY(tmp.prev); 7641da177e4SLinus Torvalds get_bh(bh); 765535ee2fbSJan Kara mapping = bh->b_assoc_map; 766535ee2fbSJan Kara __remove_assoc_queue(bh); 767535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 768535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 769535ee2fbSJan Kara smp_mb(); 770535ee2fbSJan Kara if (buffer_dirty(bh)) { 771535ee2fbSJan Kara list_add(&bh->b_assoc_buffers, 772e3892296SJan Kara &mapping->private_list); 773535ee2fbSJan Kara bh->b_assoc_map = mapping; 774535ee2fbSJan Kara } 7751da177e4SLinus Torvalds spin_unlock(lock); 7761da177e4SLinus Torvalds wait_on_buffer(bh); 7771da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 7781da177e4SLinus Torvalds err = -EIO; 7791da177e4SLinus Torvalds brelse(bh); 7801da177e4SLinus Torvalds spin_lock(lock); 7811da177e4SLinus Torvalds } 7821da177e4SLinus Torvalds 7831da177e4SLinus Torvalds spin_unlock(lock); 7841da177e4SLinus Torvalds err2 = osync_buffers_list(lock, list); 7851da177e4SLinus Torvalds if (err) 7861da177e4SLinus Torvalds return err; 7871da177e4SLinus Torvalds else 7881da177e4SLinus Torvalds return err2; 7891da177e4SLinus Torvalds } 7901da177e4SLinus Torvalds 7911da177e4SLinus Torvalds /* 7921da177e4SLinus Torvalds * Invalidate any and all dirty buffers on a given inode. We are 7931da177e4SLinus Torvalds * probably unmounting the fs, but that doesn't mean we have already 7941da177e4SLinus Torvalds * done a sync(). Just drop the buffers from the inode list. 7951da177e4SLinus Torvalds * 7961da177e4SLinus Torvalds * NOTE: we take the inode's blockdev's mapping's private_lock. Which 7971da177e4SLinus Torvalds * assumes that all the buffers are against the blockdev. Not true 7981da177e4SLinus Torvalds * for reiserfs. 7991da177e4SLinus Torvalds */ 8001da177e4SLinus Torvalds void invalidate_inode_buffers(struct inode *inode) 8011da177e4SLinus Torvalds { 8021da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8031da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 8041da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 805252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 8061da177e4SLinus Torvalds 8071da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8081da177e4SLinus Torvalds while (!list_empty(list)) 8091da177e4SLinus Torvalds __remove_assoc_queue(BH_ENTRY(list->next)); 8101da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8111da177e4SLinus Torvalds } 8121da177e4SLinus Torvalds } 81352b19ac9SJan Kara EXPORT_SYMBOL(invalidate_inode_buffers); 8141da177e4SLinus Torvalds 8151da177e4SLinus Torvalds /* 8161da177e4SLinus Torvalds * Remove any clean buffers from the inode's buffer list. This is called 8171da177e4SLinus Torvalds * when we're trying to free the inode itself. Those buffers can pin it. 8181da177e4SLinus Torvalds * 8191da177e4SLinus Torvalds * Returns true if all buffers were removed. 8201da177e4SLinus Torvalds */ 8211da177e4SLinus Torvalds int remove_inode_buffers(struct inode *inode) 8221da177e4SLinus Torvalds { 8231da177e4SLinus Torvalds int ret = 1; 8241da177e4SLinus Torvalds 8251da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8261da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 8271da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 828252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 8291da177e4SLinus Torvalds 8301da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8311da177e4SLinus Torvalds while (!list_empty(list)) { 8321da177e4SLinus Torvalds struct buffer_head *bh = BH_ENTRY(list->next); 8331da177e4SLinus Torvalds if (buffer_dirty(bh)) { 8341da177e4SLinus Torvalds ret = 0; 8351da177e4SLinus Torvalds break; 8361da177e4SLinus Torvalds } 8371da177e4SLinus Torvalds __remove_assoc_queue(bh); 8381da177e4SLinus Torvalds } 8391da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8401da177e4SLinus Torvalds } 8411da177e4SLinus Torvalds return ret; 8421da177e4SLinus Torvalds } 8431da177e4SLinus Torvalds 8441da177e4SLinus Torvalds /* 845c71124a8SPankaj Raghav * Create the appropriate buffers when given a folio for data area and 8461da177e4SLinus Torvalds * the size of each buffer.. Use the bh->b_this_page linked list to 8471da177e4SLinus Torvalds * follow the buffers created. Return NULL if unable to create more 8481da177e4SLinus Torvalds * buffers. 8491da177e4SLinus Torvalds * 8501da177e4SLinus Torvalds * The retry flag is used to differentiate async IO (paging, swapping) 8511da177e4SLinus Torvalds * which may not fail from ordinary buffer allocations. 8521da177e4SLinus Torvalds */ 853c71124a8SPankaj Raghav struct buffer_head *folio_alloc_buffers(struct folio *folio, unsigned long size, 854640ab98fSJens Axboe bool retry) 8551da177e4SLinus Torvalds { 8561da177e4SLinus Torvalds struct buffer_head *bh, *head; 857f745c6f5SShakeel Butt gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT; 8581da177e4SLinus Torvalds long offset; 859b87d8cefSRoman Gushchin struct mem_cgroup *memcg, *old_memcg; 8601da177e4SLinus Torvalds 861640ab98fSJens Axboe if (retry) 862640ab98fSJens Axboe gfp |= __GFP_NOFAIL; 863640ab98fSJens Axboe 864c71124a8SPankaj Raghav /* The folio lock pins the memcg */ 865c71124a8SPankaj Raghav memcg = folio_memcg(folio); 866b87d8cefSRoman Gushchin old_memcg = set_active_memcg(memcg); 867f745c6f5SShakeel Butt 8681da177e4SLinus Torvalds head = NULL; 869c71124a8SPankaj Raghav offset = folio_size(folio); 8701da177e4SLinus Torvalds while ((offset -= size) >= 0) { 871640ab98fSJens Axboe bh = alloc_buffer_head(gfp); 8721da177e4SLinus Torvalds if (!bh) 8731da177e4SLinus Torvalds goto no_grow; 8741da177e4SLinus Torvalds 8751da177e4SLinus Torvalds bh->b_this_page = head; 8761da177e4SLinus Torvalds bh->b_blocknr = -1; 8771da177e4SLinus Torvalds head = bh; 8781da177e4SLinus Torvalds 8791da177e4SLinus Torvalds bh->b_size = size; 8801da177e4SLinus Torvalds 881c71124a8SPankaj Raghav /* Link the buffer to its folio */ 882c71124a8SPankaj Raghav folio_set_bh(bh, folio, offset); 8831da177e4SLinus Torvalds } 884f745c6f5SShakeel Butt out: 885b87d8cefSRoman Gushchin set_active_memcg(old_memcg); 8861da177e4SLinus Torvalds return head; 8871da177e4SLinus Torvalds /* 8881da177e4SLinus Torvalds * In case anything failed, we just free everything we got. 8891da177e4SLinus Torvalds */ 8901da177e4SLinus Torvalds no_grow: 8911da177e4SLinus Torvalds if (head) { 8921da177e4SLinus Torvalds do { 8931da177e4SLinus Torvalds bh = head; 8941da177e4SLinus Torvalds head = head->b_this_page; 8951da177e4SLinus Torvalds free_buffer_head(bh); 8961da177e4SLinus Torvalds } while (head); 8971da177e4SLinus Torvalds } 8981da177e4SLinus Torvalds 899f745c6f5SShakeel Butt goto out; 9001da177e4SLinus Torvalds } 901c71124a8SPankaj Raghav EXPORT_SYMBOL_GPL(folio_alloc_buffers); 902c71124a8SPankaj Raghav 903c71124a8SPankaj Raghav struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, 904c71124a8SPankaj Raghav bool retry) 905c71124a8SPankaj Raghav { 906c71124a8SPankaj Raghav return folio_alloc_buffers(page_folio(page), size, retry); 907c71124a8SPankaj Raghav } 9081da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(alloc_page_buffers); 9091da177e4SLinus Torvalds 9101da177e4SLinus Torvalds static inline void 9111da177e4SLinus Torvalds link_dev_buffers(struct page *page, struct buffer_head *head) 9121da177e4SLinus Torvalds { 9131da177e4SLinus Torvalds struct buffer_head *bh, *tail; 9141da177e4SLinus Torvalds 9151da177e4SLinus Torvalds bh = head; 9161da177e4SLinus Torvalds do { 9171da177e4SLinus Torvalds tail = bh; 9181da177e4SLinus Torvalds bh = bh->b_this_page; 9191da177e4SLinus Torvalds } while (bh); 9201da177e4SLinus Torvalds tail->b_this_page = head; 92145dcfc27SGuoqing Jiang attach_page_private(page, head); 9221da177e4SLinus Torvalds } 9231da177e4SLinus Torvalds 924bbec0270SLinus Torvalds static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) 925bbec0270SLinus Torvalds { 926bbec0270SLinus Torvalds sector_t retval = ~((sector_t)0); 927b86058f9SChristoph Hellwig loff_t sz = bdev_nr_bytes(bdev); 928bbec0270SLinus Torvalds 929bbec0270SLinus Torvalds if (sz) { 930bbec0270SLinus Torvalds unsigned int sizebits = blksize_bits(size); 931bbec0270SLinus Torvalds retval = (sz >> sizebits); 932bbec0270SLinus Torvalds } 933bbec0270SLinus Torvalds return retval; 934bbec0270SLinus Torvalds } 935bbec0270SLinus Torvalds 9361da177e4SLinus Torvalds /* 937*6f24ce6bSMatthew Wilcox (Oracle) * Initialise the state of a blockdev folio's buffers. 9381da177e4SLinus Torvalds */ 939*6f24ce6bSMatthew Wilcox (Oracle) static sector_t folio_init_buffers(struct folio *folio, 940*6f24ce6bSMatthew Wilcox (Oracle) struct block_device *bdev, sector_t block, int size) 9411da177e4SLinus Torvalds { 942*6f24ce6bSMatthew Wilcox (Oracle) struct buffer_head *head = folio_buffers(folio); 9431da177e4SLinus Torvalds struct buffer_head *bh = head; 944*6f24ce6bSMatthew Wilcox (Oracle) bool uptodate = folio_test_uptodate(folio); 945bcd1d063SChristoph Hellwig sector_t end_block = blkdev_max_block(bdev, size); 9461da177e4SLinus Torvalds 9471da177e4SLinus Torvalds do { 9481da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 94901950a34SEric Biggers bh->b_end_io = NULL; 95001950a34SEric Biggers bh->b_private = NULL; 9511da177e4SLinus Torvalds bh->b_bdev = bdev; 9521da177e4SLinus Torvalds bh->b_blocknr = block; 9531da177e4SLinus Torvalds if (uptodate) 9541da177e4SLinus Torvalds set_buffer_uptodate(bh); 955080399aaSJeff Moyer if (block < end_block) 9561da177e4SLinus Torvalds set_buffer_mapped(bh); 9571da177e4SLinus Torvalds } 9581da177e4SLinus Torvalds block++; 9591da177e4SLinus Torvalds bh = bh->b_this_page; 9601da177e4SLinus Torvalds } while (bh != head); 961676ce6d5SHugh Dickins 962676ce6d5SHugh Dickins /* 963676ce6d5SHugh Dickins * Caller needs to validate requested block against end of device. 964676ce6d5SHugh Dickins */ 965676ce6d5SHugh Dickins return end_block; 9661da177e4SLinus Torvalds } 9671da177e4SLinus Torvalds 9681da177e4SLinus Torvalds /* 9691da177e4SLinus Torvalds * Create the page-cache page that contains the requested block. 9701da177e4SLinus Torvalds * 971676ce6d5SHugh Dickins * This is used purely for blockdev mappings. 9721da177e4SLinus Torvalds */ 973676ce6d5SHugh Dickins static int 9741da177e4SLinus Torvalds grow_dev_page(struct block_device *bdev, sector_t block, 9753b5e6454SGioh Kim pgoff_t index, int size, int sizebits, gfp_t gfp) 9761da177e4SLinus Torvalds { 9771da177e4SLinus Torvalds struct inode *inode = bdev->bd_inode; 9783c98a41cSMatthew Wilcox (Oracle) struct folio *folio; 9791da177e4SLinus Torvalds struct buffer_head *bh; 980676ce6d5SHugh Dickins sector_t end_block; 981c4b4c2a7SZhiqiang Liu int ret = 0; 98284235de3SJohannes Weiner gfp_t gfp_mask; 9831da177e4SLinus Torvalds 984c62d2555SMichal Hocko gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp; 9853b5e6454SGioh Kim 98684235de3SJohannes Weiner /* 98784235de3SJohannes Weiner * XXX: __getblk_slow() can not really deal with failure and 98884235de3SJohannes Weiner * will endlessly loop on improvised global reclaim. Prefer 98984235de3SJohannes Weiner * looping in the allocator rather than here, at least that 99084235de3SJohannes Weiner * code knows what it's doing. 99184235de3SJohannes Weiner */ 99284235de3SJohannes Weiner gfp_mask |= __GFP_NOFAIL; 99384235de3SJohannes Weiner 9943c98a41cSMatthew Wilcox (Oracle) folio = __filemap_get_folio(inode->i_mapping, index, 9953c98a41cSMatthew Wilcox (Oracle) FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp_mask); 9961da177e4SLinus Torvalds 9973c98a41cSMatthew Wilcox (Oracle) bh = folio_buffers(folio); 9983c98a41cSMatthew Wilcox (Oracle) if (bh) { 9991da177e4SLinus Torvalds if (bh->b_size == size) { 1000*6f24ce6bSMatthew Wilcox (Oracle) end_block = folio_init_buffers(folio, bdev, 1001*6f24ce6bSMatthew Wilcox (Oracle) (sector_t)index << sizebits, size); 1002676ce6d5SHugh Dickins goto done; 10031da177e4SLinus Torvalds } 10043c98a41cSMatthew Wilcox (Oracle) if (!try_to_free_buffers(folio)) 10051da177e4SLinus Torvalds goto failed; 10061da177e4SLinus Torvalds } 10071da177e4SLinus Torvalds 10083c98a41cSMatthew Wilcox (Oracle) bh = folio_alloc_buffers(folio, size, true); 10091da177e4SLinus Torvalds 10101da177e4SLinus Torvalds /* 10113c98a41cSMatthew Wilcox (Oracle) * Link the folio to the buffers and initialise them. Take the 10121da177e4SLinus Torvalds * lock to be atomic wrt __find_get_block(), which does not 10133c98a41cSMatthew Wilcox (Oracle) * run under the folio lock. 10141da177e4SLinus Torvalds */ 10151da177e4SLinus Torvalds spin_lock(&inode->i_mapping->private_lock); 10163c98a41cSMatthew Wilcox (Oracle) link_dev_buffers(&folio->page, bh); 1017*6f24ce6bSMatthew Wilcox (Oracle) end_block = folio_init_buffers(folio, bdev, 10183c98a41cSMatthew Wilcox (Oracle) (sector_t)index << sizebits, size); 10191da177e4SLinus Torvalds spin_unlock(&inode->i_mapping->private_lock); 1020676ce6d5SHugh Dickins done: 1021676ce6d5SHugh Dickins ret = (block < end_block) ? 1 : -ENXIO; 10221da177e4SLinus Torvalds failed: 10233c98a41cSMatthew Wilcox (Oracle) folio_unlock(folio); 10243c98a41cSMatthew Wilcox (Oracle) folio_put(folio); 1025676ce6d5SHugh Dickins return ret; 10261da177e4SLinus Torvalds } 10271da177e4SLinus Torvalds 10281da177e4SLinus Torvalds /* 10291da177e4SLinus Torvalds * Create buffers for the specified block device block's page. If 10301da177e4SLinus Torvalds * that page was dirty, the buffers are set dirty also. 10311da177e4SLinus Torvalds */ 1032858119e1SArjan van de Ven static int 10333b5e6454SGioh Kim grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp) 10341da177e4SLinus Torvalds { 10351da177e4SLinus Torvalds pgoff_t index; 10361da177e4SLinus Torvalds int sizebits; 10371da177e4SLinus Torvalds 103890432e60SMikulas Patocka sizebits = PAGE_SHIFT - __ffs(size); 10391da177e4SLinus Torvalds index = block >> sizebits; 10401da177e4SLinus Torvalds 1041e5657933SAndrew Morton /* 1042e5657933SAndrew Morton * Check for a block which wants to lie outside our maximum possible 1043e5657933SAndrew Morton * pagecache index. (this comparison is done using sector_t types). 1044e5657933SAndrew Morton */ 1045e5657933SAndrew Morton if (unlikely(index != block >> sizebits)) { 1046e5657933SAndrew Morton printk(KERN_ERR "%s: requested out-of-range block %llu for " 1047a1c6f057SDmitry Monakhov "device %pg\n", 10488e24eea7SHarvey Harrison __func__, (unsigned long long)block, 1049a1c6f057SDmitry Monakhov bdev); 1050e5657933SAndrew Morton return -EIO; 1051e5657933SAndrew Morton } 1052676ce6d5SHugh Dickins 10531da177e4SLinus Torvalds /* Create a page with the proper size buffers.. */ 10543b5e6454SGioh Kim return grow_dev_page(bdev, block, index, size, sizebits, gfp); 10551da177e4SLinus Torvalds } 10561da177e4SLinus Torvalds 10570026ba40SEric Biggers static struct buffer_head * 10583b5e6454SGioh Kim __getblk_slow(struct block_device *bdev, sector_t block, 10593b5e6454SGioh Kim unsigned size, gfp_t gfp) 10601da177e4SLinus Torvalds { 10611da177e4SLinus Torvalds /* Size must be multiple of hard sectorsize */ 1062e1defc4fSMartin K. Petersen if (unlikely(size & (bdev_logical_block_size(bdev)-1) || 10631da177e4SLinus Torvalds (size < 512 || size > PAGE_SIZE))) { 10641da177e4SLinus Torvalds printk(KERN_ERR "getblk(): invalid block size %d requested\n", 10651da177e4SLinus Torvalds size); 1066e1defc4fSMartin K. Petersen printk(KERN_ERR "logical block size: %d\n", 1067e1defc4fSMartin K. Petersen bdev_logical_block_size(bdev)); 10681da177e4SLinus Torvalds 10691da177e4SLinus Torvalds dump_stack(); 10701da177e4SLinus Torvalds return NULL; 10711da177e4SLinus Torvalds } 10721da177e4SLinus Torvalds 1073676ce6d5SHugh Dickins for (;;) { 1074676ce6d5SHugh Dickins struct buffer_head *bh; 1075676ce6d5SHugh Dickins int ret; 1076676ce6d5SHugh Dickins 10771da177e4SLinus Torvalds bh = __find_get_block(bdev, block, size); 10781da177e4SLinus Torvalds if (bh) 10791da177e4SLinus Torvalds return bh; 10801da177e4SLinus Torvalds 10813b5e6454SGioh Kim ret = grow_buffers(bdev, block, size, gfp); 1082676ce6d5SHugh Dickins if (ret < 0) 108391f68c89SJeff Moyer return NULL; 1084676ce6d5SHugh Dickins } 10851da177e4SLinus Torvalds } 10861da177e4SLinus Torvalds 10871da177e4SLinus Torvalds /* 10881da177e4SLinus Torvalds * The relationship between dirty buffers and dirty pages: 10891da177e4SLinus Torvalds * 10901da177e4SLinus Torvalds * Whenever a page has any dirty buffers, the page's dirty bit is set, and 1091ec82e1c1SMatthew Wilcox * the page is tagged dirty in the page cache. 10921da177e4SLinus Torvalds * 10931da177e4SLinus Torvalds * At all times, the dirtiness of the buffers represents the dirtiness of 10941da177e4SLinus Torvalds * subsections of the page. If the page has buffers, the page dirty bit is 10951da177e4SLinus Torvalds * merely a hint about the true dirty state. 10961da177e4SLinus Torvalds * 10971da177e4SLinus Torvalds * When a page is set dirty in its entirety, all its buffers are marked dirty 10981da177e4SLinus Torvalds * (if the page has buffers). 10991da177e4SLinus Torvalds * 11001da177e4SLinus Torvalds * When a buffer is marked dirty, its page is dirtied, but the page's other 11011da177e4SLinus Torvalds * buffers are not. 11021da177e4SLinus Torvalds * 11031da177e4SLinus Torvalds * Also. When blockdev buffers are explicitly read with bread(), they 11041da177e4SLinus Torvalds * individually become uptodate. But their backing page remains not 11051da177e4SLinus Torvalds * uptodate - even if all of its buffers are uptodate. A subsequent 11062c69e205SMatthew Wilcox (Oracle) * block_read_full_folio() against that folio will discover all the uptodate 11072c69e205SMatthew Wilcox (Oracle) * buffers, will set the folio uptodate and will perform no I/O. 11081da177e4SLinus Torvalds */ 11091da177e4SLinus Torvalds 11101da177e4SLinus Torvalds /** 11111da177e4SLinus Torvalds * mark_buffer_dirty - mark a buffer_head as needing writeout 111267be2dd1SMartin Waitz * @bh: the buffer_head to mark dirty 11131da177e4SLinus Torvalds * 1114ec82e1c1SMatthew Wilcox * mark_buffer_dirty() will set the dirty bit against the buffer, then set 1115ec82e1c1SMatthew Wilcox * its backing page dirty, then tag the page as dirty in the page cache 1116ec82e1c1SMatthew Wilcox * and then attach the address_space's inode to its superblock's dirty 11171da177e4SLinus Torvalds * inode list. 11181da177e4SLinus Torvalds * 1119abc8a8a2SMatthew Wilcox (Oracle) * mark_buffer_dirty() is atomic. It takes bh->b_folio->mapping->private_lock, 1120b93b0163SMatthew Wilcox * i_pages lock and mapping->host->i_lock. 11211da177e4SLinus Torvalds */ 1122fc9b52cdSHarvey Harrison void mark_buffer_dirty(struct buffer_head *bh) 11231da177e4SLinus Torvalds { 1124787d2214SNick Piggin WARN_ON_ONCE(!buffer_uptodate(bh)); 11251be62dc1SLinus Torvalds 11265305cb83STejun Heo trace_block_dirty_buffer(bh); 11275305cb83STejun Heo 11281be62dc1SLinus Torvalds /* 11291be62dc1SLinus Torvalds * Very *carefully* optimize the it-is-already-dirty case. 11301be62dc1SLinus Torvalds * 11311be62dc1SLinus Torvalds * Don't let the final "is it dirty" escape to before we 11321be62dc1SLinus Torvalds * perhaps modified the buffer. 11331be62dc1SLinus Torvalds */ 11341be62dc1SLinus Torvalds if (buffer_dirty(bh)) { 11351be62dc1SLinus Torvalds smp_mb(); 11361be62dc1SLinus Torvalds if (buffer_dirty(bh)) 11371be62dc1SLinus Torvalds return; 11381be62dc1SLinus Torvalds } 11391be62dc1SLinus Torvalds 1140a8e7d49aSLinus Torvalds if (!test_set_buffer_dirty(bh)) { 1141cf1d3417SMatthew Wilcox (Oracle) struct folio *folio = bh->b_folio; 1142c4843a75SGreg Thelen struct address_space *mapping = NULL; 1143c4843a75SGreg Thelen 1144cf1d3417SMatthew Wilcox (Oracle) folio_memcg_lock(folio); 1145cf1d3417SMatthew Wilcox (Oracle) if (!folio_test_set_dirty(folio)) { 1146cf1d3417SMatthew Wilcox (Oracle) mapping = folio->mapping; 11478e9d78edSLinus Torvalds if (mapping) 1148cf1d3417SMatthew Wilcox (Oracle) __folio_mark_dirty(folio, mapping, 0); 11498e9d78edSLinus Torvalds } 1150cf1d3417SMatthew Wilcox (Oracle) folio_memcg_unlock(folio); 1151c4843a75SGreg Thelen if (mapping) 1152c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 1153a8e7d49aSLinus Torvalds } 11541da177e4SLinus Torvalds } 11551fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(mark_buffer_dirty); 11561da177e4SLinus Torvalds 115787354e5dSJeff Layton void mark_buffer_write_io_error(struct buffer_head *bh) 115887354e5dSJeff Layton { 1159485e9605SJeff Layton struct super_block *sb; 1160485e9605SJeff Layton 116187354e5dSJeff Layton set_buffer_write_io_error(bh); 116287354e5dSJeff Layton /* FIXME: do we need to set this in both places? */ 1163abc8a8a2SMatthew Wilcox (Oracle) if (bh->b_folio && bh->b_folio->mapping) 1164abc8a8a2SMatthew Wilcox (Oracle) mapping_set_error(bh->b_folio->mapping, -EIO); 116587354e5dSJeff Layton if (bh->b_assoc_map) 116687354e5dSJeff Layton mapping_set_error(bh->b_assoc_map, -EIO); 1167485e9605SJeff Layton rcu_read_lock(); 1168485e9605SJeff Layton sb = READ_ONCE(bh->b_bdev->bd_super); 1169485e9605SJeff Layton if (sb) 1170485e9605SJeff Layton errseq_set(&sb->s_wb_err, -EIO); 1171485e9605SJeff Layton rcu_read_unlock(); 117287354e5dSJeff Layton } 117387354e5dSJeff Layton EXPORT_SYMBOL(mark_buffer_write_io_error); 117487354e5dSJeff Layton 11751da177e4SLinus Torvalds /* 11761da177e4SLinus Torvalds * Decrement a buffer_head's reference count. If all buffers against a page 11771da177e4SLinus Torvalds * have zero reference count, are clean and unlocked, and if the page is clean 11781da177e4SLinus Torvalds * and unlocked then try_to_free_buffers() may strip the buffers from the page 11791da177e4SLinus Torvalds * in preparation for freeing it (sometimes, rarely, buffers are removed from 11801da177e4SLinus Torvalds * a page but it ends up not being freed, and buffers may later be reattached). 11811da177e4SLinus Torvalds */ 11821da177e4SLinus Torvalds void __brelse(struct buffer_head * buf) 11831da177e4SLinus Torvalds { 11841da177e4SLinus Torvalds if (atomic_read(&buf->b_count)) { 11851da177e4SLinus Torvalds put_bh(buf); 11861da177e4SLinus Torvalds return; 11871da177e4SLinus Torvalds } 11885c752ad9SArjan van de Ven WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); 11891da177e4SLinus Torvalds } 11901fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__brelse); 11911da177e4SLinus Torvalds 11921da177e4SLinus Torvalds /* 11931da177e4SLinus Torvalds * bforget() is like brelse(), except it discards any 11941da177e4SLinus Torvalds * potentially dirty data. 11951da177e4SLinus Torvalds */ 11961da177e4SLinus Torvalds void __bforget(struct buffer_head *bh) 11971da177e4SLinus Torvalds { 11981da177e4SLinus Torvalds clear_buffer_dirty(bh); 1199535ee2fbSJan Kara if (bh->b_assoc_map) { 1200abc8a8a2SMatthew Wilcox (Oracle) struct address_space *buffer_mapping = bh->b_folio->mapping; 12011da177e4SLinus Torvalds 12021da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 12031da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 120458ff407bSJan Kara bh->b_assoc_map = NULL; 12051da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 12061da177e4SLinus Torvalds } 12071da177e4SLinus Torvalds __brelse(bh); 12081da177e4SLinus Torvalds } 12091fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__bforget); 12101da177e4SLinus Torvalds 12111da177e4SLinus Torvalds static struct buffer_head *__bread_slow(struct buffer_head *bh) 12121da177e4SLinus Torvalds { 12131da177e4SLinus Torvalds lock_buffer(bh); 12141da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 12151da177e4SLinus Torvalds unlock_buffer(bh); 12161da177e4SLinus Torvalds return bh; 12171da177e4SLinus Torvalds } else { 12181da177e4SLinus Torvalds get_bh(bh); 12191da177e4SLinus Torvalds bh->b_end_io = end_buffer_read_sync; 12201420c4a5SBart Van Assche submit_bh(REQ_OP_READ, bh); 12211da177e4SLinus Torvalds wait_on_buffer(bh); 12221da177e4SLinus Torvalds if (buffer_uptodate(bh)) 12231da177e4SLinus Torvalds return bh; 12241da177e4SLinus Torvalds } 12251da177e4SLinus Torvalds brelse(bh); 12261da177e4SLinus Torvalds return NULL; 12271da177e4SLinus Torvalds } 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds /* 12301da177e4SLinus Torvalds * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). 12311da177e4SLinus Torvalds * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their 12321da177e4SLinus Torvalds * refcount elevated by one when they're in an LRU. A buffer can only appear 12331da177e4SLinus Torvalds * once in a particular CPU's LRU. A single buffer can be present in multiple 12341da177e4SLinus Torvalds * CPU's LRUs at the same time. 12351da177e4SLinus Torvalds * 12361da177e4SLinus Torvalds * This is a transparent caching front-end to sb_bread(), sb_getblk() and 12371da177e4SLinus Torvalds * sb_find_get_block(). 12381da177e4SLinus Torvalds * 12391da177e4SLinus Torvalds * The LRUs themselves only need locking against invalidate_bh_lrus. We use 12401da177e4SLinus Torvalds * a local interrupt disable for that. 12411da177e4SLinus Torvalds */ 12421da177e4SLinus Torvalds 124386cf78d7SSebastien Buisson #define BH_LRU_SIZE 16 12441da177e4SLinus Torvalds 12451da177e4SLinus Torvalds struct bh_lru { 12461da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12471da177e4SLinus Torvalds }; 12481da177e4SLinus Torvalds 12491da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; 12501da177e4SLinus Torvalds 12511da177e4SLinus Torvalds #ifdef CONFIG_SMP 12521da177e4SLinus Torvalds #define bh_lru_lock() local_irq_disable() 12531da177e4SLinus Torvalds #define bh_lru_unlock() local_irq_enable() 12541da177e4SLinus Torvalds #else 12551da177e4SLinus Torvalds #define bh_lru_lock() preempt_disable() 12561da177e4SLinus Torvalds #define bh_lru_unlock() preempt_enable() 12571da177e4SLinus Torvalds #endif 12581da177e4SLinus Torvalds 12591da177e4SLinus Torvalds static inline void check_irqs_on(void) 12601da177e4SLinus Torvalds { 12611da177e4SLinus Torvalds #ifdef irqs_disabled 12621da177e4SLinus Torvalds BUG_ON(irqs_disabled()); 12631da177e4SLinus Torvalds #endif 12641da177e4SLinus Torvalds } 12651da177e4SLinus Torvalds 12661da177e4SLinus Torvalds /* 1267241f01fbSEric Biggers * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is 1268241f01fbSEric Biggers * inserted at the front, and the buffer_head at the back if any is evicted. 1269241f01fbSEric Biggers * Or, if already in the LRU it is moved to the front. 12701da177e4SLinus Torvalds */ 12711da177e4SLinus Torvalds static void bh_lru_install(struct buffer_head *bh) 12721da177e4SLinus Torvalds { 1273241f01fbSEric Biggers struct buffer_head *evictee = bh; 1274241f01fbSEric Biggers struct bh_lru *b; 1275241f01fbSEric Biggers int i; 12761da177e4SLinus Torvalds 12771da177e4SLinus Torvalds check_irqs_on(); 1278c0226eb8SMinchan Kim bh_lru_lock(); 1279c0226eb8SMinchan Kim 12808cc621d2SMinchan Kim /* 12818cc621d2SMinchan Kim * the refcount of buffer_head in bh_lru prevents dropping the 12828cc621d2SMinchan Kim * attached page(i.e., try_to_free_buffers) so it could cause 12838cc621d2SMinchan Kim * failing page migration. 12848cc621d2SMinchan Kim * Skip putting upcoming bh into bh_lru until migration is done. 12858cc621d2SMinchan Kim */ 1286c0226eb8SMinchan Kim if (lru_cache_disabled()) { 1287c0226eb8SMinchan Kim bh_lru_unlock(); 12888cc621d2SMinchan Kim return; 1289c0226eb8SMinchan Kim } 1290241f01fbSEric Biggers 1291241f01fbSEric Biggers b = this_cpu_ptr(&bh_lrus); 1292241f01fbSEric Biggers for (i = 0; i < BH_LRU_SIZE; i++) { 1293241f01fbSEric Biggers swap(evictee, b->bhs[i]); 1294241f01fbSEric Biggers if (evictee == bh) { 1295241f01fbSEric Biggers bh_lru_unlock(); 1296241f01fbSEric Biggers return; 1297241f01fbSEric Biggers } 1298241f01fbSEric Biggers } 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds get_bh(bh); 13011da177e4SLinus Torvalds bh_lru_unlock(); 1302241f01fbSEric Biggers brelse(evictee); 13031da177e4SLinus Torvalds } 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds /* 13061da177e4SLinus Torvalds * Look up the bh in this cpu's LRU. If it's there, move it to the head. 13071da177e4SLinus Torvalds */ 1308858119e1SArjan van de Ven static struct buffer_head * 13093991d3bdSTomasz Kvarsin lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) 13101da177e4SLinus Torvalds { 13111da177e4SLinus Torvalds struct buffer_head *ret = NULL; 13123991d3bdSTomasz Kvarsin unsigned int i; 13131da177e4SLinus Torvalds 13141da177e4SLinus Torvalds check_irqs_on(); 13151da177e4SLinus Torvalds bh_lru_lock(); 13161da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 1317c7b92516SChristoph Lameter struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); 13181da177e4SLinus Torvalds 13199470dd5dSZach Brown if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && 13209470dd5dSZach Brown bh->b_size == size) { 13211da177e4SLinus Torvalds if (i) { 13221da177e4SLinus Torvalds while (i) { 1323c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[i], 1324c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[i - 1])); 13251da177e4SLinus Torvalds i--; 13261da177e4SLinus Torvalds } 1327c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[0], bh); 13281da177e4SLinus Torvalds } 13291da177e4SLinus Torvalds get_bh(bh); 13301da177e4SLinus Torvalds ret = bh; 13311da177e4SLinus Torvalds break; 13321da177e4SLinus Torvalds } 13331da177e4SLinus Torvalds } 13341da177e4SLinus Torvalds bh_lru_unlock(); 13351da177e4SLinus Torvalds return ret; 13361da177e4SLinus Torvalds } 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds /* 13391da177e4SLinus Torvalds * Perform a pagecache lookup for the matching buffer. If it's there, refresh 13401da177e4SLinus Torvalds * it in the LRU and mark it as accessed. If it is not present then return 13411da177e4SLinus Torvalds * NULL 13421da177e4SLinus Torvalds */ 13431da177e4SLinus Torvalds struct buffer_head * 13443991d3bdSTomasz Kvarsin __find_get_block(struct block_device *bdev, sector_t block, unsigned size) 13451da177e4SLinus Torvalds { 13461da177e4SLinus Torvalds struct buffer_head *bh = lookup_bh_lru(bdev, block, size); 13471da177e4SLinus Torvalds 13481da177e4SLinus Torvalds if (bh == NULL) { 13492457aec6SMel Gorman /* __find_get_block_slow will mark the page accessed */ 1350385fd4c5SCoywolf Qi Hunt bh = __find_get_block_slow(bdev, block); 13511da177e4SLinus Torvalds if (bh) 13521da177e4SLinus Torvalds bh_lru_install(bh); 13532457aec6SMel Gorman } else 13541da177e4SLinus Torvalds touch_buffer(bh); 13552457aec6SMel Gorman 13561da177e4SLinus Torvalds return bh; 13571da177e4SLinus Torvalds } 13581da177e4SLinus Torvalds EXPORT_SYMBOL(__find_get_block); 13591da177e4SLinus Torvalds 13601da177e4SLinus Torvalds /* 13613b5e6454SGioh Kim * __getblk_gfp() will locate (and, if necessary, create) the buffer_head 13621da177e4SLinus Torvalds * which corresponds to the passed block_device, block and size. The 13631da177e4SLinus Torvalds * returned buffer has its reference count incremented. 13641da177e4SLinus Torvalds * 13653b5e6454SGioh Kim * __getblk_gfp() will lock up the machine if grow_dev_page's 13663b5e6454SGioh Kim * try_to_free_buffers() attempt is failing. FIXME, perhaps? 13671da177e4SLinus Torvalds */ 13681da177e4SLinus Torvalds struct buffer_head * 13693b5e6454SGioh Kim __getblk_gfp(struct block_device *bdev, sector_t block, 13703b5e6454SGioh Kim unsigned size, gfp_t gfp) 13711da177e4SLinus Torvalds { 13721da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, block, size); 13731da177e4SLinus Torvalds 13741da177e4SLinus Torvalds might_sleep(); 13751da177e4SLinus Torvalds if (bh == NULL) 13763b5e6454SGioh Kim bh = __getblk_slow(bdev, block, size, gfp); 13771da177e4SLinus Torvalds return bh; 13781da177e4SLinus Torvalds } 13793b5e6454SGioh Kim EXPORT_SYMBOL(__getblk_gfp); 13801da177e4SLinus Torvalds 13811da177e4SLinus Torvalds /* 13821da177e4SLinus Torvalds * Do async read-ahead on a buffer.. 13831da177e4SLinus Torvalds */ 13843991d3bdSTomasz Kvarsin void __breadahead(struct block_device *bdev, sector_t block, unsigned size) 13851da177e4SLinus Torvalds { 13861da177e4SLinus Torvalds struct buffer_head *bh = __getblk(bdev, block, size); 1387a3e713b5SAndrew Morton if (likely(bh)) { 1388e7ea1129SZhang Yi bh_readahead(bh, REQ_RAHEAD); 13891da177e4SLinus Torvalds brelse(bh); 13901da177e4SLinus Torvalds } 1391a3e713b5SAndrew Morton } 13921da177e4SLinus Torvalds EXPORT_SYMBOL(__breadahead); 13931da177e4SLinus Torvalds 13941da177e4SLinus Torvalds /** 13953b5e6454SGioh Kim * __bread_gfp() - reads a specified block and returns the bh 139667be2dd1SMartin Waitz * @bdev: the block_device to read from 13971da177e4SLinus Torvalds * @block: number of block 13981da177e4SLinus Torvalds * @size: size (in bytes) to read 13993b5e6454SGioh Kim * @gfp: page allocation flag 14001da177e4SLinus Torvalds * 14011da177e4SLinus Torvalds * Reads a specified block, and returns buffer head that contains it. 14023b5e6454SGioh Kim * The page cache can be allocated from non-movable area 14033b5e6454SGioh Kim * not to prevent page migration if you set gfp to zero. 14041da177e4SLinus Torvalds * It returns NULL if the block was unreadable. 14051da177e4SLinus Torvalds */ 14061da177e4SLinus Torvalds struct buffer_head * 14073b5e6454SGioh Kim __bread_gfp(struct block_device *bdev, sector_t block, 14083b5e6454SGioh Kim unsigned size, gfp_t gfp) 14091da177e4SLinus Torvalds { 14103b5e6454SGioh Kim struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); 14111da177e4SLinus Torvalds 1412a3e713b5SAndrew Morton if (likely(bh) && !buffer_uptodate(bh)) 14131da177e4SLinus Torvalds bh = __bread_slow(bh); 14141da177e4SLinus Torvalds return bh; 14151da177e4SLinus Torvalds } 14163b5e6454SGioh Kim EXPORT_SYMBOL(__bread_gfp); 14171da177e4SLinus Torvalds 14188cc621d2SMinchan Kim static void __invalidate_bh_lrus(struct bh_lru *b) 14198cc621d2SMinchan Kim { 14208cc621d2SMinchan Kim int i; 14218cc621d2SMinchan Kim 14228cc621d2SMinchan Kim for (i = 0; i < BH_LRU_SIZE; i++) { 14238cc621d2SMinchan Kim brelse(b->bhs[i]); 14248cc621d2SMinchan Kim b->bhs[i] = NULL; 14258cc621d2SMinchan Kim } 14268cc621d2SMinchan Kim } 14271da177e4SLinus Torvalds /* 14281da177e4SLinus Torvalds * invalidate_bh_lrus() is called rarely - but not only at unmount. 14291da177e4SLinus Torvalds * This doesn't race because it runs in each cpu either in irq 14301da177e4SLinus Torvalds * or with preempt disabled. 14311da177e4SLinus Torvalds */ 14321da177e4SLinus Torvalds static void invalidate_bh_lru(void *arg) 14331da177e4SLinus Torvalds { 14341da177e4SLinus Torvalds struct bh_lru *b = &get_cpu_var(bh_lrus); 14351da177e4SLinus Torvalds 14368cc621d2SMinchan Kim __invalidate_bh_lrus(b); 14371da177e4SLinus Torvalds put_cpu_var(bh_lrus); 14381da177e4SLinus Torvalds } 14391da177e4SLinus Torvalds 14408cc621d2SMinchan Kim bool has_bh_in_lru(int cpu, void *dummy) 144142be35d0SGilad Ben-Yossef { 144242be35d0SGilad Ben-Yossef struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); 144342be35d0SGilad Ben-Yossef int i; 144442be35d0SGilad Ben-Yossef 144542be35d0SGilad Ben-Yossef for (i = 0; i < BH_LRU_SIZE; i++) { 144642be35d0SGilad Ben-Yossef if (b->bhs[i]) 14471d706679SSaurav Girepunje return true; 144842be35d0SGilad Ben-Yossef } 144942be35d0SGilad Ben-Yossef 14501d706679SSaurav Girepunje return false; 145142be35d0SGilad Ben-Yossef } 145242be35d0SGilad Ben-Yossef 1453f9a14399SPeter Zijlstra void invalidate_bh_lrus(void) 14541da177e4SLinus Torvalds { 1455cb923159SSebastian Andrzej Siewior on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1); 14561da177e4SLinus Torvalds } 14579db5579bSNick Piggin EXPORT_SYMBOL_GPL(invalidate_bh_lrus); 14581da177e4SLinus Torvalds 1459243418e3SMinchan Kim /* 1460243418e3SMinchan Kim * It's called from workqueue context so we need a bh_lru_lock to close 1461243418e3SMinchan Kim * the race with preemption/irq. 1462243418e3SMinchan Kim */ 1463243418e3SMinchan Kim void invalidate_bh_lrus_cpu(void) 14648cc621d2SMinchan Kim { 14658cc621d2SMinchan Kim struct bh_lru *b; 14668cc621d2SMinchan Kim 14678cc621d2SMinchan Kim bh_lru_lock(); 1468243418e3SMinchan Kim b = this_cpu_ptr(&bh_lrus); 14698cc621d2SMinchan Kim __invalidate_bh_lrus(b); 14708cc621d2SMinchan Kim bh_lru_unlock(); 14718cc621d2SMinchan Kim } 14728cc621d2SMinchan Kim 14731da177e4SLinus Torvalds void set_bh_page(struct buffer_head *bh, 14741da177e4SLinus Torvalds struct page *page, unsigned long offset) 14751da177e4SLinus Torvalds { 14761da177e4SLinus Torvalds bh->b_page = page; 1477e827f923SEric Sesterhenn BUG_ON(offset >= PAGE_SIZE); 14781da177e4SLinus Torvalds if (PageHighMem(page)) 14791da177e4SLinus Torvalds /* 14801da177e4SLinus Torvalds * This catches illegal uses and preserves the offset: 14811da177e4SLinus Torvalds */ 14821da177e4SLinus Torvalds bh->b_data = (char *)(0 + offset); 14831da177e4SLinus Torvalds else 14841da177e4SLinus Torvalds bh->b_data = page_address(page) + offset; 14851da177e4SLinus Torvalds } 14861da177e4SLinus Torvalds EXPORT_SYMBOL(set_bh_page); 14871da177e4SLinus Torvalds 1488465e5e6aSPankaj Raghav void folio_set_bh(struct buffer_head *bh, struct folio *folio, 1489465e5e6aSPankaj Raghav unsigned long offset) 1490465e5e6aSPankaj Raghav { 1491465e5e6aSPankaj Raghav bh->b_folio = folio; 1492465e5e6aSPankaj Raghav BUG_ON(offset >= folio_size(folio)); 1493465e5e6aSPankaj Raghav if (folio_test_highmem(folio)) 1494465e5e6aSPankaj Raghav /* 1495465e5e6aSPankaj Raghav * This catches illegal uses and preserves the offset: 1496465e5e6aSPankaj Raghav */ 1497465e5e6aSPankaj Raghav bh->b_data = (char *)(0 + offset); 1498465e5e6aSPankaj Raghav else 1499465e5e6aSPankaj Raghav bh->b_data = folio_address(folio) + offset; 1500465e5e6aSPankaj Raghav } 1501465e5e6aSPankaj Raghav EXPORT_SYMBOL(folio_set_bh); 1502465e5e6aSPankaj Raghav 15031da177e4SLinus Torvalds /* 15041da177e4SLinus Torvalds * Called when truncating a buffer on a page completely. 15051da177e4SLinus Torvalds */ 1506e7470ee8SMel Gorman 1507e7470ee8SMel Gorman /* Bits that are cleared during an invalidate */ 1508e7470ee8SMel Gorman #define BUFFER_FLAGS_DISCARD \ 1509e7470ee8SMel Gorman (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ 1510e7470ee8SMel Gorman 1 << BH_Delay | 1 << BH_Unwritten) 1511e7470ee8SMel Gorman 1512858119e1SArjan van de Ven static void discard_buffer(struct buffer_head * bh) 15131da177e4SLinus Torvalds { 1514b0192296SUros Bizjak unsigned long b_state; 1515e7470ee8SMel Gorman 15161da177e4SLinus Torvalds lock_buffer(bh); 15171da177e4SLinus Torvalds clear_buffer_dirty(bh); 15181da177e4SLinus Torvalds bh->b_bdev = NULL; 1519b0192296SUros Bizjak b_state = READ_ONCE(bh->b_state); 1520b0192296SUros Bizjak do { 1521b0192296SUros Bizjak } while (!try_cmpxchg(&bh->b_state, &b_state, 1522b0192296SUros Bizjak b_state & ~BUFFER_FLAGS_DISCARD)); 15231da177e4SLinus Torvalds unlock_buffer(bh); 15241da177e4SLinus Torvalds } 15251da177e4SLinus Torvalds 15261da177e4SLinus Torvalds /** 15277ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio - Invalidate part or all of a buffer-backed folio. 15287ba13abbSMatthew Wilcox (Oracle) * @folio: The folio which is affected. 1529d47992f8SLukas Czerner * @offset: start of the range to invalidate 1530d47992f8SLukas Czerner * @length: length of the range to invalidate 15311da177e4SLinus Torvalds * 15327ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio() is called when all or part of the folio has been 15331da177e4SLinus Torvalds * invalidated by a truncate operation. 15341da177e4SLinus Torvalds * 15357ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio() does not have to release all buffers, but it must 15361da177e4SLinus Torvalds * ensure that no dirty buffer is left outside @offset and that no I/O 15371da177e4SLinus Torvalds * is underway against any of the blocks which are outside the truncation 15381da177e4SLinus Torvalds * point. Because the caller is about to free (and possibly reuse) those 15391da177e4SLinus Torvalds * blocks on-disk. 15401da177e4SLinus Torvalds */ 15417ba13abbSMatthew Wilcox (Oracle) void block_invalidate_folio(struct folio *folio, size_t offset, size_t length) 15421da177e4SLinus Torvalds { 15431da177e4SLinus Torvalds struct buffer_head *head, *bh, *next; 15447ba13abbSMatthew Wilcox (Oracle) size_t curr_off = 0; 15457ba13abbSMatthew Wilcox (Oracle) size_t stop = length + offset; 15461da177e4SLinus Torvalds 15477ba13abbSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 15481da177e4SLinus Torvalds 1549d47992f8SLukas Czerner /* 1550d47992f8SLukas Czerner * Check for overflow 1551d47992f8SLukas Czerner */ 15527ba13abbSMatthew Wilcox (Oracle) BUG_ON(stop > folio_size(folio) || stop < length); 1553d47992f8SLukas Czerner 15547ba13abbSMatthew Wilcox (Oracle) head = folio_buffers(folio); 15557ba13abbSMatthew Wilcox (Oracle) if (!head) 15567ba13abbSMatthew Wilcox (Oracle) return; 15577ba13abbSMatthew Wilcox (Oracle) 15581da177e4SLinus Torvalds bh = head; 15591da177e4SLinus Torvalds do { 15607ba13abbSMatthew Wilcox (Oracle) size_t next_off = curr_off + bh->b_size; 15611da177e4SLinus Torvalds next = bh->b_this_page; 15621da177e4SLinus Torvalds 15631da177e4SLinus Torvalds /* 1564d47992f8SLukas Czerner * Are we still fully in range ? 1565d47992f8SLukas Czerner */ 1566d47992f8SLukas Czerner if (next_off > stop) 1567d47992f8SLukas Czerner goto out; 1568d47992f8SLukas Czerner 1569d47992f8SLukas Czerner /* 15701da177e4SLinus Torvalds * is this block fully invalidated? 15711da177e4SLinus Torvalds */ 15721da177e4SLinus Torvalds if (offset <= curr_off) 15731da177e4SLinus Torvalds discard_buffer(bh); 15741da177e4SLinus Torvalds curr_off = next_off; 15751da177e4SLinus Torvalds bh = next; 15761da177e4SLinus Torvalds } while (bh != head); 15771da177e4SLinus Torvalds 15781da177e4SLinus Torvalds /* 15797ba13abbSMatthew Wilcox (Oracle) * We release buffers only if the entire folio is being invalidated. 15801da177e4SLinus Torvalds * The get_block cached value has been unconditionally invalidated, 15811da177e4SLinus Torvalds * so real IO is not possible anymore. 15821da177e4SLinus Torvalds */ 15837ba13abbSMatthew Wilcox (Oracle) if (length == folio_size(folio)) 15847ba13abbSMatthew Wilcox (Oracle) filemap_release_folio(folio, 0); 15851da177e4SLinus Torvalds out: 15862ff28e22SNeilBrown return; 15871da177e4SLinus Torvalds } 15887ba13abbSMatthew Wilcox (Oracle) EXPORT_SYMBOL(block_invalidate_folio); 15891da177e4SLinus Torvalds 15901da177e4SLinus Torvalds /* 15911da177e4SLinus Torvalds * We attach and possibly dirty the buffers atomically wrt 1592e621900aSMatthew Wilcox (Oracle) * block_dirty_folio() via private_lock. try_to_free_buffers 15938e2e1756SPankaj Raghav * is already excluded via the folio lock. 15941da177e4SLinus Torvalds */ 15958e2e1756SPankaj Raghav void folio_create_empty_buffers(struct folio *folio, unsigned long blocksize, 15968e2e1756SPankaj Raghav unsigned long b_state) 15971da177e4SLinus Torvalds { 15981da177e4SLinus Torvalds struct buffer_head *bh, *head, *tail; 15991da177e4SLinus Torvalds 16008e2e1756SPankaj Raghav head = folio_alloc_buffers(folio, blocksize, true); 16011da177e4SLinus Torvalds bh = head; 16021da177e4SLinus Torvalds do { 16031da177e4SLinus Torvalds bh->b_state |= b_state; 16041da177e4SLinus Torvalds tail = bh; 16051da177e4SLinus Torvalds bh = bh->b_this_page; 16061da177e4SLinus Torvalds } while (bh); 16071da177e4SLinus Torvalds tail->b_this_page = head; 16081da177e4SLinus Torvalds 16098e2e1756SPankaj Raghav spin_lock(&folio->mapping->private_lock); 16108e2e1756SPankaj Raghav if (folio_test_uptodate(folio) || folio_test_dirty(folio)) { 16111da177e4SLinus Torvalds bh = head; 16121da177e4SLinus Torvalds do { 16138e2e1756SPankaj Raghav if (folio_test_dirty(folio)) 16141da177e4SLinus Torvalds set_buffer_dirty(bh); 16158e2e1756SPankaj Raghav if (folio_test_uptodate(folio)) 16161da177e4SLinus Torvalds set_buffer_uptodate(bh); 16171da177e4SLinus Torvalds bh = bh->b_this_page; 16181da177e4SLinus Torvalds } while (bh != head); 16191da177e4SLinus Torvalds } 16208e2e1756SPankaj Raghav folio_attach_private(folio, head); 16218e2e1756SPankaj Raghav spin_unlock(&folio->mapping->private_lock); 16228e2e1756SPankaj Raghav } 16238e2e1756SPankaj Raghav EXPORT_SYMBOL(folio_create_empty_buffers); 16248e2e1756SPankaj Raghav 16258e2e1756SPankaj Raghav void create_empty_buffers(struct page *page, 16268e2e1756SPankaj Raghav unsigned long blocksize, unsigned long b_state) 16278e2e1756SPankaj Raghav { 16288e2e1756SPankaj Raghav folio_create_empty_buffers(page_folio(page), blocksize, b_state); 16291da177e4SLinus Torvalds } 16301da177e4SLinus Torvalds EXPORT_SYMBOL(create_empty_buffers); 16311da177e4SLinus Torvalds 163229f3ad7dSJan Kara /** 163329f3ad7dSJan Kara * clean_bdev_aliases: clean a range of buffers in block device 163429f3ad7dSJan Kara * @bdev: Block device to clean buffers in 163529f3ad7dSJan Kara * @block: Start of a range of blocks to clean 163629f3ad7dSJan Kara * @len: Number of blocks to clean 16371da177e4SLinus Torvalds * 163829f3ad7dSJan Kara * We are taking a range of blocks for data and we don't want writeback of any 163929f3ad7dSJan Kara * buffer-cache aliases starting from return from this function and until the 164029f3ad7dSJan Kara * moment when something will explicitly mark the buffer dirty (hopefully that 164129f3ad7dSJan Kara * will not happen until we will free that block ;-) We don't even need to mark 164229f3ad7dSJan Kara * it not-uptodate - nobody can expect anything from a newly allocated buffer 164329f3ad7dSJan Kara * anyway. We used to use unmap_buffer() for such invalidation, but that was 164429f3ad7dSJan Kara * wrong. We definitely don't want to mark the alias unmapped, for example - it 164529f3ad7dSJan Kara * would confuse anyone who might pick it with bread() afterwards... 164629f3ad7dSJan Kara * 164729f3ad7dSJan Kara * Also.. Note that bforget() doesn't lock the buffer. So there can be 164829f3ad7dSJan Kara * writeout I/O going on against recently-freed buffers. We don't wait on that 164929f3ad7dSJan Kara * I/O in bforget() - it's more efficient to wait on the I/O only if we really 165029f3ad7dSJan Kara * need to. That happens here. 16511da177e4SLinus Torvalds */ 165229f3ad7dSJan Kara void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) 16531da177e4SLinus Torvalds { 165429f3ad7dSJan Kara struct inode *bd_inode = bdev->bd_inode; 165529f3ad7dSJan Kara struct address_space *bd_mapping = bd_inode->i_mapping; 16569e0b6f31SMatthew Wilcox (Oracle) struct folio_batch fbatch; 165729f3ad7dSJan Kara pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); 165829f3ad7dSJan Kara pgoff_t end; 1659c10f778dSJan Kara int i, count; 166029f3ad7dSJan Kara struct buffer_head *bh; 166129f3ad7dSJan Kara struct buffer_head *head; 16621da177e4SLinus Torvalds 166329f3ad7dSJan Kara end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits); 16649e0b6f31SMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 16659e0b6f31SMatthew Wilcox (Oracle) while (filemap_get_folios(bd_mapping, &index, end, &fbatch)) { 16669e0b6f31SMatthew Wilcox (Oracle) count = folio_batch_count(&fbatch); 1667c10f778dSJan Kara for (i = 0; i < count; i++) { 16689e0b6f31SMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 16691da177e4SLinus Torvalds 16709e0b6f31SMatthew Wilcox (Oracle) if (!folio_buffers(folio)) 167129f3ad7dSJan Kara continue; 167229f3ad7dSJan Kara /* 16739e0b6f31SMatthew Wilcox (Oracle) * We use folio lock instead of bd_mapping->private_lock 167429f3ad7dSJan Kara * to pin buffers here since we can afford to sleep and 167529f3ad7dSJan Kara * it scales better than a global spinlock lock. 167629f3ad7dSJan Kara */ 16779e0b6f31SMatthew Wilcox (Oracle) folio_lock(folio); 16789e0b6f31SMatthew Wilcox (Oracle) /* Recheck when the folio is locked which pins bhs */ 16799e0b6f31SMatthew Wilcox (Oracle) head = folio_buffers(folio); 16809e0b6f31SMatthew Wilcox (Oracle) if (!head) 168129f3ad7dSJan Kara goto unlock_page; 168229f3ad7dSJan Kara bh = head; 168329f3ad7dSJan Kara do { 16846c006a9dSChandan Rajendra if (!buffer_mapped(bh) || (bh->b_blocknr < block)) 168529f3ad7dSJan Kara goto next; 168629f3ad7dSJan Kara if (bh->b_blocknr >= block + len) 168729f3ad7dSJan Kara break; 168829f3ad7dSJan Kara clear_buffer_dirty(bh); 168929f3ad7dSJan Kara wait_on_buffer(bh); 169029f3ad7dSJan Kara clear_buffer_req(bh); 169129f3ad7dSJan Kara next: 169229f3ad7dSJan Kara bh = bh->b_this_page; 169329f3ad7dSJan Kara } while (bh != head); 169429f3ad7dSJan Kara unlock_page: 16959e0b6f31SMatthew Wilcox (Oracle) folio_unlock(folio); 169629f3ad7dSJan Kara } 16979e0b6f31SMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 169829f3ad7dSJan Kara cond_resched(); 1699c10f778dSJan Kara /* End of range already reached? */ 1700c10f778dSJan Kara if (index > end || !index) 1701c10f778dSJan Kara break; 17021da177e4SLinus Torvalds } 17031da177e4SLinus Torvalds } 170429f3ad7dSJan Kara EXPORT_SYMBOL(clean_bdev_aliases); 17051da177e4SLinus Torvalds 17061da177e4SLinus Torvalds /* 170745bce8f3SLinus Torvalds * Size is a power-of-two in the range 512..PAGE_SIZE, 170845bce8f3SLinus Torvalds * and the case we care about most is PAGE_SIZE. 170945bce8f3SLinus Torvalds * 171045bce8f3SLinus Torvalds * So this *could* possibly be written with those 171145bce8f3SLinus Torvalds * constraints in mind (relevant mostly if some 171245bce8f3SLinus Torvalds * architecture has a slow bit-scan instruction) 171345bce8f3SLinus Torvalds */ 171445bce8f3SLinus Torvalds static inline int block_size_bits(unsigned int blocksize) 171545bce8f3SLinus Torvalds { 171645bce8f3SLinus Torvalds return ilog2(blocksize); 171745bce8f3SLinus Torvalds } 171845bce8f3SLinus Torvalds 1719c6c8c3e7SPankaj Raghav static struct buffer_head *folio_create_buffers(struct folio *folio, 1720c6c8c3e7SPankaj Raghav struct inode *inode, 1721c6c8c3e7SPankaj Raghav unsigned int b_state) 172245bce8f3SLinus Torvalds { 1723c6c8c3e7SPankaj Raghav BUG_ON(!folio_test_locked(folio)); 172445bce8f3SLinus Torvalds 1725c6c8c3e7SPankaj Raghav if (!folio_buffers(folio)) 1726c6c8c3e7SPankaj Raghav folio_create_empty_buffers(folio, 1727c6c8c3e7SPankaj Raghav 1 << READ_ONCE(inode->i_blkbits), 17286aa7de05SMark Rutland b_state); 1729c6c8c3e7SPankaj Raghav return folio_buffers(folio); 173045bce8f3SLinus Torvalds } 173145bce8f3SLinus Torvalds 173245bce8f3SLinus Torvalds /* 17331da177e4SLinus Torvalds * NOTE! All mapped/uptodate combinations are valid: 17341da177e4SLinus Torvalds * 17351da177e4SLinus Torvalds * Mapped Uptodate Meaning 17361da177e4SLinus Torvalds * 17371da177e4SLinus Torvalds * No No "unknown" - must do get_block() 17381da177e4SLinus Torvalds * No Yes "hole" - zero-filled 17391da177e4SLinus Torvalds * Yes No "allocated" - allocated on disk, not read in 17401da177e4SLinus Torvalds * Yes Yes "valid" - allocated and up-to-date in memory. 17411da177e4SLinus Torvalds * 17421da177e4SLinus Torvalds * "Dirty" is valid only with the last case (mapped+uptodate). 17431da177e4SLinus Torvalds */ 17441da177e4SLinus Torvalds 17451da177e4SLinus Torvalds /* 17461da177e4SLinus Torvalds * While block_write_full_page is writing back the dirty buffers under 17471da177e4SLinus Torvalds * the page lock, whoever dirtied the buffers may decide to clean them 17481da177e4SLinus Torvalds * again at any time. We handle that by only looking at the buffer 17491da177e4SLinus Torvalds * state inside lock_buffer(). 17501da177e4SLinus Torvalds * 17511da177e4SLinus Torvalds * If block_write_full_page() is called for regular writeback 17521da177e4SLinus Torvalds * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a 17531da177e4SLinus Torvalds * locked buffer. This only can happen if someone has written the buffer 17541da177e4SLinus Torvalds * directly, with submit_bh(). At the address_space level PageWriteback 17551da177e4SLinus Torvalds * prevents this contention from occurring. 17566e34eeddSTheodore Ts'o * 17576e34eeddSTheodore Ts'o * If block_write_full_page() is called with wbc->sync_mode == 175870fd7614SChristoph Hellwig * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this 1759721a9602SJens Axboe * causes the writes to be flagged as synchronous writes. 17601da177e4SLinus Torvalds */ 176153418a18SMatthew Wilcox (Oracle) int __block_write_full_folio(struct inode *inode, struct folio *folio, 176235c80d5fSChris Mason get_block_t *get_block, struct writeback_control *wbc, 176335c80d5fSChris Mason bh_end_io_t *handler) 17641da177e4SLinus Torvalds { 17651da177e4SLinus Torvalds int err; 17661da177e4SLinus Torvalds sector_t block; 17671da177e4SLinus Torvalds sector_t last_block; 1768f0fbd5fcSAndrew Morton struct buffer_head *bh, *head; 176945bce8f3SLinus Torvalds unsigned int blocksize, bbits; 17701da177e4SLinus Torvalds int nr_underway = 0; 17713ae72869SBart Van Assche blk_opf_t write_flags = wbc_to_write_flags(wbc); 17721da177e4SLinus Torvalds 177353418a18SMatthew Wilcox (Oracle) head = folio_create_buffers(folio, inode, 17741da177e4SLinus Torvalds (1 << BH_Dirty) | (1 << BH_Uptodate)); 17751da177e4SLinus Torvalds 17761da177e4SLinus Torvalds /* 1777e621900aSMatthew Wilcox (Oracle) * Be very careful. We have no exclusion from block_dirty_folio 17781da177e4SLinus Torvalds * here, and the (potentially unmapped) buffers may become dirty at 17791da177e4SLinus Torvalds * any time. If a buffer becomes dirty here after we've inspected it 178053418a18SMatthew Wilcox (Oracle) * then we just miss that fact, and the folio stays dirty. 17811da177e4SLinus Torvalds * 1782e621900aSMatthew Wilcox (Oracle) * Buffers outside i_size may be dirtied by block_dirty_folio; 17831da177e4SLinus Torvalds * handle that here by just cleaning them. 17841da177e4SLinus Torvalds */ 17851da177e4SLinus Torvalds 17861da177e4SLinus Torvalds bh = head; 178745bce8f3SLinus Torvalds blocksize = bh->b_size; 178845bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 178945bce8f3SLinus Torvalds 179053418a18SMatthew Wilcox (Oracle) block = (sector_t)folio->index << (PAGE_SHIFT - bbits); 179145bce8f3SLinus Torvalds last_block = (i_size_read(inode) - 1) >> bbits; 17921da177e4SLinus Torvalds 17931da177e4SLinus Torvalds /* 17941da177e4SLinus Torvalds * Get all the dirty buffers mapped to disk addresses and 17951da177e4SLinus Torvalds * handle any aliases from the underlying blockdev's mapping. 17961da177e4SLinus Torvalds */ 17971da177e4SLinus Torvalds do { 17981da177e4SLinus Torvalds if (block > last_block) { 17991da177e4SLinus Torvalds /* 18001da177e4SLinus Torvalds * mapped buffers outside i_size will occur, because 180153418a18SMatthew Wilcox (Oracle) * this folio can be outside i_size when there is a 18021da177e4SLinus Torvalds * truncate in progress. 18031da177e4SLinus Torvalds */ 18041da177e4SLinus Torvalds /* 18051da177e4SLinus Torvalds * The buffer was zeroed by block_write_full_page() 18061da177e4SLinus Torvalds */ 18071da177e4SLinus Torvalds clear_buffer_dirty(bh); 18081da177e4SLinus Torvalds set_buffer_uptodate(bh); 180929a814d2SAlex Tomas } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && 181029a814d2SAlex Tomas buffer_dirty(bh)) { 1811b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 18121da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 18131da177e4SLinus Torvalds if (err) 18141da177e4SLinus Torvalds goto recover; 181529a814d2SAlex Tomas clear_buffer_delay(bh); 18161da177e4SLinus Torvalds if (buffer_new(bh)) { 18171da177e4SLinus Torvalds /* blockdev mappings never come here */ 18181da177e4SLinus Torvalds clear_buffer_new(bh); 1819e64855c6SJan Kara clean_bdev_bh_alias(bh); 18201da177e4SLinus Torvalds } 18211da177e4SLinus Torvalds } 18221da177e4SLinus Torvalds bh = bh->b_this_page; 18231da177e4SLinus Torvalds block++; 18241da177e4SLinus Torvalds } while (bh != head); 18251da177e4SLinus Torvalds 18261da177e4SLinus Torvalds do { 18271da177e4SLinus Torvalds if (!buffer_mapped(bh)) 18281da177e4SLinus Torvalds continue; 18291da177e4SLinus Torvalds /* 18301da177e4SLinus Torvalds * If it's a fully non-blocking write attempt and we cannot 183153418a18SMatthew Wilcox (Oracle) * lock the buffer then redirty the folio. Note that this can 18325b0830cbSJens Axboe * potentially cause a busy-wait loop from writeback threads 18335b0830cbSJens Axboe * and kswapd activity, but those code paths have their own 18345b0830cbSJens Axboe * higher-level throttling. 18351da177e4SLinus Torvalds */ 18361b430beeSWu Fengguang if (wbc->sync_mode != WB_SYNC_NONE) { 18371da177e4SLinus Torvalds lock_buffer(bh); 1838ca5de404SNick Piggin } else if (!trylock_buffer(bh)) { 183953418a18SMatthew Wilcox (Oracle) folio_redirty_for_writepage(wbc, folio); 18401da177e4SLinus Torvalds continue; 18411da177e4SLinus Torvalds } 18421da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 184335c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 18441da177e4SLinus Torvalds } else { 18451da177e4SLinus Torvalds unlock_buffer(bh); 18461da177e4SLinus Torvalds } 18471da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 18481da177e4SLinus Torvalds 18491da177e4SLinus Torvalds /* 185053418a18SMatthew Wilcox (Oracle) * The folio and its buffers are protected by the writeback flag, 185153418a18SMatthew Wilcox (Oracle) * so we can drop the bh refcounts early. 18521da177e4SLinus Torvalds */ 185353418a18SMatthew Wilcox (Oracle) BUG_ON(folio_test_writeback(folio)); 185453418a18SMatthew Wilcox (Oracle) folio_start_writeback(folio); 18551da177e4SLinus Torvalds 18561da177e4SLinus Torvalds do { 18571da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 18581da177e4SLinus Torvalds if (buffer_async_write(bh)) { 18591420c4a5SBart Van Assche submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc); 18601da177e4SLinus Torvalds nr_underway++; 1861ad576e63SNick Piggin } 18621da177e4SLinus Torvalds bh = next; 18631da177e4SLinus Torvalds } while (bh != head); 186453418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 18651da177e4SLinus Torvalds 18661da177e4SLinus Torvalds err = 0; 18671da177e4SLinus Torvalds done: 18681da177e4SLinus Torvalds if (nr_underway == 0) { 18691da177e4SLinus Torvalds /* 187053418a18SMatthew Wilcox (Oracle) * The folio was marked dirty, but the buffers were 18711da177e4SLinus Torvalds * clean. Someone wrote them back by hand with 187279f59784SZhang Yi * write_dirty_buffer/submit_bh. A rare case. 18731da177e4SLinus Torvalds */ 187453418a18SMatthew Wilcox (Oracle) folio_end_writeback(folio); 18753d67f2d7SNick Piggin 18761da177e4SLinus Torvalds /* 187753418a18SMatthew Wilcox (Oracle) * The folio and buffer_heads can be released at any time from 18781da177e4SLinus Torvalds * here on. 18791da177e4SLinus Torvalds */ 18801da177e4SLinus Torvalds } 18811da177e4SLinus Torvalds return err; 18821da177e4SLinus Torvalds 18831da177e4SLinus Torvalds recover: 18841da177e4SLinus Torvalds /* 18851da177e4SLinus Torvalds * ENOSPC, or some other error. We may already have added some 18861da177e4SLinus Torvalds * blocks to the file, so we need to write these out to avoid 18871da177e4SLinus Torvalds * exposing stale data. 188853418a18SMatthew Wilcox (Oracle) * The folio is currently locked and not marked for writeback 18891da177e4SLinus Torvalds */ 18901da177e4SLinus Torvalds bh = head; 18911da177e4SLinus Torvalds /* Recovery: lock and submit the mapped buffers */ 18921da177e4SLinus Torvalds do { 189329a814d2SAlex Tomas if (buffer_mapped(bh) && buffer_dirty(bh) && 189429a814d2SAlex Tomas !buffer_delay(bh)) { 18951da177e4SLinus Torvalds lock_buffer(bh); 189635c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 18971da177e4SLinus Torvalds } else { 18981da177e4SLinus Torvalds /* 18991da177e4SLinus Torvalds * The buffer may have been set dirty during 190053418a18SMatthew Wilcox (Oracle) * attachment to a dirty folio. 19011da177e4SLinus Torvalds */ 19021da177e4SLinus Torvalds clear_buffer_dirty(bh); 19031da177e4SLinus Torvalds } 19041da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 190553418a18SMatthew Wilcox (Oracle) folio_set_error(folio); 190653418a18SMatthew Wilcox (Oracle) BUG_ON(folio_test_writeback(folio)); 190753418a18SMatthew Wilcox (Oracle) mapping_set_error(folio->mapping, err); 190853418a18SMatthew Wilcox (Oracle) folio_start_writeback(folio); 19091da177e4SLinus Torvalds do { 19101da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 19111da177e4SLinus Torvalds if (buffer_async_write(bh)) { 19121da177e4SLinus Torvalds clear_buffer_dirty(bh); 19131420c4a5SBart Van Assche submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc); 19141da177e4SLinus Torvalds nr_underway++; 1915ad576e63SNick Piggin } 19161da177e4SLinus Torvalds bh = next; 19171da177e4SLinus Torvalds } while (bh != head); 191853418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 19191da177e4SLinus Torvalds goto done; 19201da177e4SLinus Torvalds } 192153418a18SMatthew Wilcox (Oracle) EXPORT_SYMBOL(__block_write_full_folio); 19221da177e4SLinus Torvalds 1923afddba49SNick Piggin /* 19244a9622f2SMatthew Wilcox (Oracle) * If a folio has any new buffers, zero them out here, and mark them uptodate 1925afddba49SNick Piggin * and dirty so they'll be written out (in order to prevent uninitialised 1926afddba49SNick Piggin * block data from leaking). And clear the new bit. 1927afddba49SNick Piggin */ 19284a9622f2SMatthew Wilcox (Oracle) void folio_zero_new_buffers(struct folio *folio, size_t from, size_t to) 1929afddba49SNick Piggin { 19304a9622f2SMatthew Wilcox (Oracle) size_t block_start, block_end; 1931afddba49SNick Piggin struct buffer_head *head, *bh; 1932afddba49SNick Piggin 19334a9622f2SMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 19344a9622f2SMatthew Wilcox (Oracle) head = folio_buffers(folio); 19354a9622f2SMatthew Wilcox (Oracle) if (!head) 1936afddba49SNick Piggin return; 1937afddba49SNick Piggin 19384a9622f2SMatthew Wilcox (Oracle) bh = head; 1939afddba49SNick Piggin block_start = 0; 1940afddba49SNick Piggin do { 1941afddba49SNick Piggin block_end = block_start + bh->b_size; 1942afddba49SNick Piggin 1943afddba49SNick Piggin if (buffer_new(bh)) { 1944afddba49SNick Piggin if (block_end > from && block_start < to) { 19454a9622f2SMatthew Wilcox (Oracle) if (!folio_test_uptodate(folio)) { 19464a9622f2SMatthew Wilcox (Oracle) size_t start, xend; 1947afddba49SNick Piggin 1948afddba49SNick Piggin start = max(from, block_start); 19494a9622f2SMatthew Wilcox (Oracle) xend = min(to, block_end); 1950afddba49SNick Piggin 19514a9622f2SMatthew Wilcox (Oracle) folio_zero_segment(folio, start, xend); 1952afddba49SNick Piggin set_buffer_uptodate(bh); 1953afddba49SNick Piggin } 1954afddba49SNick Piggin 1955afddba49SNick Piggin clear_buffer_new(bh); 1956afddba49SNick Piggin mark_buffer_dirty(bh); 1957afddba49SNick Piggin } 1958afddba49SNick Piggin } 1959afddba49SNick Piggin 1960afddba49SNick Piggin block_start = block_end; 1961afddba49SNick Piggin bh = bh->b_this_page; 1962afddba49SNick Piggin } while (bh != head); 1963afddba49SNick Piggin } 19644a9622f2SMatthew Wilcox (Oracle) EXPORT_SYMBOL(folio_zero_new_buffers); 1965afddba49SNick Piggin 1966ae259a9cSChristoph Hellwig static void 1967ae259a9cSChristoph Hellwig iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, 19686d49cc85SChristoph Hellwig const struct iomap *iomap) 1969ae259a9cSChristoph Hellwig { 1970ae259a9cSChristoph Hellwig loff_t offset = block << inode->i_blkbits; 1971ae259a9cSChristoph Hellwig 1972ae259a9cSChristoph Hellwig bh->b_bdev = iomap->bdev; 1973ae259a9cSChristoph Hellwig 1974ae259a9cSChristoph Hellwig /* 1975ae259a9cSChristoph Hellwig * Block points to offset in file we need to map, iomap contains 1976ae259a9cSChristoph Hellwig * the offset at which the map starts. If the map ends before the 1977ae259a9cSChristoph Hellwig * current block, then do not map the buffer and let the caller 1978ae259a9cSChristoph Hellwig * handle it. 1979ae259a9cSChristoph Hellwig */ 1980ae259a9cSChristoph Hellwig BUG_ON(offset >= iomap->offset + iomap->length); 1981ae259a9cSChristoph Hellwig 1982ae259a9cSChristoph Hellwig switch (iomap->type) { 1983ae259a9cSChristoph Hellwig case IOMAP_HOLE: 1984ae259a9cSChristoph Hellwig /* 1985ae259a9cSChristoph Hellwig * If the buffer is not up to date or beyond the current EOF, 1986ae259a9cSChristoph Hellwig * we need to mark it as new to ensure sub-block zeroing is 1987ae259a9cSChristoph Hellwig * executed if necessary. 1988ae259a9cSChristoph Hellwig */ 1989ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 1990ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 1991ae259a9cSChristoph Hellwig set_buffer_new(bh); 1992ae259a9cSChristoph Hellwig break; 1993ae259a9cSChristoph Hellwig case IOMAP_DELALLOC: 1994ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 1995ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 1996ae259a9cSChristoph Hellwig set_buffer_new(bh); 1997ae259a9cSChristoph Hellwig set_buffer_uptodate(bh); 1998ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 1999ae259a9cSChristoph Hellwig set_buffer_delay(bh); 2000ae259a9cSChristoph Hellwig break; 2001ae259a9cSChristoph Hellwig case IOMAP_UNWRITTEN: 2002ae259a9cSChristoph Hellwig /* 20033d7b6b21SAndreas Gruenbacher * For unwritten regions, we always need to ensure that regions 20043d7b6b21SAndreas Gruenbacher * in the block we are not writing to are zeroed. Mark the 20053d7b6b21SAndreas Gruenbacher * buffer as new to ensure this. 2006ae259a9cSChristoph Hellwig */ 2007ae259a9cSChristoph Hellwig set_buffer_new(bh); 2008ae259a9cSChristoph Hellwig set_buffer_unwritten(bh); 2009df561f66SGustavo A. R. Silva fallthrough; 2010ae259a9cSChristoph Hellwig case IOMAP_MAPPED: 20113d7b6b21SAndreas Gruenbacher if ((iomap->flags & IOMAP_F_NEW) || 20123d7b6b21SAndreas Gruenbacher offset >= i_size_read(inode)) 2013ae259a9cSChristoph Hellwig set_buffer_new(bh); 201419fe5f64SAndreas Gruenbacher bh->b_blocknr = (iomap->addr + offset - iomap->offset) >> 201519fe5f64SAndreas Gruenbacher inode->i_blkbits; 2016ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 2017ae259a9cSChristoph Hellwig break; 2018ae259a9cSChristoph Hellwig } 2019ae259a9cSChristoph Hellwig } 2020ae259a9cSChristoph Hellwig 2021d1bd0b4eSMatthew Wilcox (Oracle) int __block_write_begin_int(struct folio *folio, loff_t pos, unsigned len, 20226d49cc85SChristoph Hellwig get_block_t *get_block, const struct iomap *iomap) 20231da177e4SLinus Torvalds { 202409cbfeafSKirill A. Shutemov unsigned from = pos & (PAGE_SIZE - 1); 2025ebdec241SChristoph Hellwig unsigned to = from + len; 2026d1bd0b4eSMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 20271da177e4SLinus Torvalds unsigned block_start, block_end; 20281da177e4SLinus Torvalds sector_t block; 20291da177e4SLinus Torvalds int err = 0; 20301da177e4SLinus Torvalds unsigned blocksize, bbits; 20311da177e4SLinus Torvalds struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; 20321da177e4SLinus Torvalds 2033d1bd0b4eSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 203409cbfeafSKirill A. Shutemov BUG_ON(from > PAGE_SIZE); 203509cbfeafSKirill A. Shutemov BUG_ON(to > PAGE_SIZE); 20361da177e4SLinus Torvalds BUG_ON(from > to); 20371da177e4SLinus Torvalds 2038c6c8c3e7SPankaj Raghav head = folio_create_buffers(folio, inode, 0); 203945bce8f3SLinus Torvalds blocksize = head->b_size; 204045bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 20411da177e4SLinus Torvalds 2042d1bd0b4eSMatthew Wilcox (Oracle) block = (sector_t)folio->index << (PAGE_SHIFT - bbits); 20431da177e4SLinus Torvalds 20441da177e4SLinus Torvalds for(bh = head, block_start = 0; bh != head || !block_start; 20451da177e4SLinus Torvalds block++, block_start=block_end, bh = bh->b_this_page) { 20461da177e4SLinus Torvalds block_end = block_start + blocksize; 20471da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 2048d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 20491da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 20501da177e4SLinus Torvalds set_buffer_uptodate(bh); 20511da177e4SLinus Torvalds } 20521da177e4SLinus Torvalds continue; 20531da177e4SLinus Torvalds } 20541da177e4SLinus Torvalds if (buffer_new(bh)) 20551da177e4SLinus Torvalds clear_buffer_new(bh); 20561da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2057b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2058ae259a9cSChristoph Hellwig if (get_block) { 20591da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 20601da177e4SLinus Torvalds if (err) 2061f3ddbdc6SNick Piggin break; 2062ae259a9cSChristoph Hellwig } else { 2063ae259a9cSChristoph Hellwig iomap_to_bh(inode, block, bh, iomap); 2064ae259a9cSChristoph Hellwig } 2065ae259a9cSChristoph Hellwig 20661da177e4SLinus Torvalds if (buffer_new(bh)) { 2067e64855c6SJan Kara clean_bdev_bh_alias(bh); 2068d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 2069637aff46SNick Piggin clear_buffer_new(bh); 20701da177e4SLinus Torvalds set_buffer_uptodate(bh); 2071637aff46SNick Piggin mark_buffer_dirty(bh); 20721da177e4SLinus Torvalds continue; 20731da177e4SLinus Torvalds } 2074eebd2aa3SChristoph Lameter if (block_end > to || block_start < from) 2075d1bd0b4eSMatthew Wilcox (Oracle) folio_zero_segments(folio, 2076eebd2aa3SChristoph Lameter to, block_end, 2077eebd2aa3SChristoph Lameter block_start, from); 20781da177e4SLinus Torvalds continue; 20791da177e4SLinus Torvalds } 20801da177e4SLinus Torvalds } 2081d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 20821da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 20831da177e4SLinus Torvalds set_buffer_uptodate(bh); 20841da177e4SLinus Torvalds continue; 20851da177e4SLinus Torvalds } 20861da177e4SLinus Torvalds if (!buffer_uptodate(bh) && !buffer_delay(bh) && 208733a266ddSDavid Chinner !buffer_unwritten(bh) && 20881da177e4SLinus Torvalds (block_start < from || block_end > to)) { 2089e7ea1129SZhang Yi bh_read_nowait(bh, 0); 20901da177e4SLinus Torvalds *wait_bh++=bh; 20911da177e4SLinus Torvalds } 20921da177e4SLinus Torvalds } 20931da177e4SLinus Torvalds /* 20941da177e4SLinus Torvalds * If we issued read requests - let them complete. 20951da177e4SLinus Torvalds */ 20961da177e4SLinus Torvalds while(wait_bh > wait) { 20971da177e4SLinus Torvalds wait_on_buffer(*--wait_bh); 20981da177e4SLinus Torvalds if (!buffer_uptodate(*wait_bh)) 2099f3ddbdc6SNick Piggin err = -EIO; 21001da177e4SLinus Torvalds } 2101f9f07b6cSJan Kara if (unlikely(err)) 21024a9622f2SMatthew Wilcox (Oracle) folio_zero_new_buffers(folio, from, to); 21031da177e4SLinus Torvalds return err; 21041da177e4SLinus Torvalds } 2105ae259a9cSChristoph Hellwig 2106ae259a9cSChristoph Hellwig int __block_write_begin(struct page *page, loff_t pos, unsigned len, 2107ae259a9cSChristoph Hellwig get_block_t *get_block) 2108ae259a9cSChristoph Hellwig { 2109d1bd0b4eSMatthew Wilcox (Oracle) return __block_write_begin_int(page_folio(page), pos, len, get_block, 2110d1bd0b4eSMatthew Wilcox (Oracle) NULL); 2111ae259a9cSChristoph Hellwig } 2112ebdec241SChristoph Hellwig EXPORT_SYMBOL(__block_write_begin); 21131da177e4SLinus Torvalds 21148c6cb3e3SMatthew Wilcox (Oracle) static int __block_commit_write(struct inode *inode, struct folio *folio, 21158c6cb3e3SMatthew Wilcox (Oracle) size_t from, size_t to) 21161da177e4SLinus Torvalds { 21178c6cb3e3SMatthew Wilcox (Oracle) size_t block_start, block_end; 21188c6cb3e3SMatthew Wilcox (Oracle) bool partial = false; 21191da177e4SLinus Torvalds unsigned blocksize; 21201da177e4SLinus Torvalds struct buffer_head *bh, *head; 21211da177e4SLinus Torvalds 21228c6cb3e3SMatthew Wilcox (Oracle) bh = head = folio_buffers(folio); 212345bce8f3SLinus Torvalds blocksize = bh->b_size; 21241da177e4SLinus Torvalds 212545bce8f3SLinus Torvalds block_start = 0; 212645bce8f3SLinus Torvalds do { 21271da177e4SLinus Torvalds block_end = block_start + blocksize; 21281da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 21291da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 21308c6cb3e3SMatthew Wilcox (Oracle) partial = true; 21311da177e4SLinus Torvalds } else { 21321da177e4SLinus Torvalds set_buffer_uptodate(bh); 21331da177e4SLinus Torvalds mark_buffer_dirty(bh); 21341da177e4SLinus Torvalds } 21354ebd3aecSYang Guo if (buffer_new(bh)) 2136afddba49SNick Piggin clear_buffer_new(bh); 213745bce8f3SLinus Torvalds 213845bce8f3SLinus Torvalds block_start = block_end; 213945bce8f3SLinus Torvalds bh = bh->b_this_page; 214045bce8f3SLinus Torvalds } while (bh != head); 21411da177e4SLinus Torvalds 21421da177e4SLinus Torvalds /* 21431da177e4SLinus Torvalds * If this is a partial write which happened to make all buffers 21442c69e205SMatthew Wilcox (Oracle) * uptodate then we can optimize away a bogus read_folio() for 21458c6cb3e3SMatthew Wilcox (Oracle) * the next read(). Here we 'discover' whether the folio went 21461da177e4SLinus Torvalds * uptodate as a result of this (potentially partial) write. 21471da177e4SLinus Torvalds */ 21481da177e4SLinus Torvalds if (!partial) 21498c6cb3e3SMatthew Wilcox (Oracle) folio_mark_uptodate(folio); 21501da177e4SLinus Torvalds return 0; 21511da177e4SLinus Torvalds } 21521da177e4SLinus Torvalds 21531da177e4SLinus Torvalds /* 2154155130a4SChristoph Hellwig * block_write_begin takes care of the basic task of block allocation and 2155155130a4SChristoph Hellwig * bringing partial write blocks uptodate first. 2156155130a4SChristoph Hellwig * 21577bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 2158afddba49SNick Piggin */ 2159155130a4SChristoph Hellwig int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, 2160b3992d1eSMatthew Wilcox (Oracle) struct page **pagep, get_block_t *get_block) 2161afddba49SNick Piggin { 216209cbfeafSKirill A. Shutemov pgoff_t index = pos >> PAGE_SHIFT; 2163afddba49SNick Piggin struct page *page; 21646e1db88dSChristoph Hellwig int status; 2165afddba49SNick Piggin 2166b7446e7cSMatthew Wilcox (Oracle) page = grab_cache_page_write_begin(mapping, index); 21676e1db88dSChristoph Hellwig if (!page) 21686e1db88dSChristoph Hellwig return -ENOMEM; 2169afddba49SNick Piggin 21706e1db88dSChristoph Hellwig status = __block_write_begin(page, pos, len, get_block); 2171afddba49SNick Piggin if (unlikely(status)) { 2172afddba49SNick Piggin unlock_page(page); 217309cbfeafSKirill A. Shutemov put_page(page); 21746e1db88dSChristoph Hellwig page = NULL; 2175afddba49SNick Piggin } 2176afddba49SNick Piggin 21776e1db88dSChristoph Hellwig *pagep = page; 2178afddba49SNick Piggin return status; 2179afddba49SNick Piggin } 2180afddba49SNick Piggin EXPORT_SYMBOL(block_write_begin); 2181afddba49SNick Piggin 2182afddba49SNick Piggin int block_write_end(struct file *file, struct address_space *mapping, 2183afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2184afddba49SNick Piggin struct page *page, void *fsdata) 2185afddba49SNick Piggin { 21868c6cb3e3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 2187afddba49SNick Piggin struct inode *inode = mapping->host; 21888c6cb3e3SMatthew Wilcox (Oracle) size_t start = pos - folio_pos(folio); 2189afddba49SNick Piggin 2190afddba49SNick Piggin if (unlikely(copied < len)) { 2191afddba49SNick Piggin /* 21922c69e205SMatthew Wilcox (Oracle) * The buffers that were written will now be uptodate, so 21932c69e205SMatthew Wilcox (Oracle) * we don't have to worry about a read_folio reading them 21942c69e205SMatthew Wilcox (Oracle) * and overwriting a partial write. However if we have 21952c69e205SMatthew Wilcox (Oracle) * encountered a short write and only partially written 21962c69e205SMatthew Wilcox (Oracle) * into a buffer, it will not be marked uptodate, so a 21972c69e205SMatthew Wilcox (Oracle) * read_folio might come in and destroy our partial write. 2198afddba49SNick Piggin * 2199afddba49SNick Piggin * Do the simplest thing, and just treat any short write to a 22008c6cb3e3SMatthew Wilcox (Oracle) * non uptodate folio as a zero-length write, and force the 2201afddba49SNick Piggin * caller to redo the whole thing. 2202afddba49SNick Piggin */ 22038c6cb3e3SMatthew Wilcox (Oracle) if (!folio_test_uptodate(folio)) 2204afddba49SNick Piggin copied = 0; 2205afddba49SNick Piggin 22064a9622f2SMatthew Wilcox (Oracle) folio_zero_new_buffers(folio, start+copied, start+len); 2207afddba49SNick Piggin } 22088c6cb3e3SMatthew Wilcox (Oracle) flush_dcache_folio(folio); 2209afddba49SNick Piggin 2210afddba49SNick Piggin /* This could be a short (even 0-length) commit */ 22118c6cb3e3SMatthew Wilcox (Oracle) __block_commit_write(inode, folio, start, start + copied); 2212afddba49SNick Piggin 2213afddba49SNick Piggin return copied; 2214afddba49SNick Piggin } 2215afddba49SNick Piggin EXPORT_SYMBOL(block_write_end); 2216afddba49SNick Piggin 2217afddba49SNick Piggin int generic_write_end(struct file *file, struct address_space *mapping, 2218afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2219afddba49SNick Piggin struct page *page, void *fsdata) 2220afddba49SNick Piggin { 22218af54f29SChristoph Hellwig struct inode *inode = mapping->host; 22228af54f29SChristoph Hellwig loff_t old_size = inode->i_size; 22238af54f29SChristoph Hellwig bool i_size_changed = false; 22248af54f29SChristoph Hellwig 2225afddba49SNick Piggin copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); 22268af54f29SChristoph Hellwig 22278af54f29SChristoph Hellwig /* 22288af54f29SChristoph Hellwig * No need to use i_size_read() here, the i_size cannot change under us 22298af54f29SChristoph Hellwig * because we hold i_rwsem. 22308af54f29SChristoph Hellwig * 22318af54f29SChristoph Hellwig * But it's important to update i_size while still holding page lock: 22328af54f29SChristoph Hellwig * page writeout could otherwise come in and zero beyond i_size. 22338af54f29SChristoph Hellwig */ 22348af54f29SChristoph Hellwig if (pos + copied > inode->i_size) { 22358af54f29SChristoph Hellwig i_size_write(inode, pos + copied); 22368af54f29SChristoph Hellwig i_size_changed = true; 22378af54f29SChristoph Hellwig } 22388af54f29SChristoph Hellwig 22398af54f29SChristoph Hellwig unlock_page(page); 22407a77dad7SAndreas Gruenbacher put_page(page); 22418af54f29SChristoph Hellwig 22428af54f29SChristoph Hellwig if (old_size < pos) 22438af54f29SChristoph Hellwig pagecache_isize_extended(inode, old_size, pos); 22448af54f29SChristoph Hellwig /* 22458af54f29SChristoph Hellwig * Don't mark the inode dirty under page lock. First, it unnecessarily 22468af54f29SChristoph Hellwig * makes the holding time of page lock longer. Second, it forces lock 22478af54f29SChristoph Hellwig * ordering of page lock and transaction start for journaling 22488af54f29SChristoph Hellwig * filesystems. 22498af54f29SChristoph Hellwig */ 22508af54f29SChristoph Hellwig if (i_size_changed) 22518af54f29SChristoph Hellwig mark_inode_dirty(inode); 225226ddb1f4SAndreas Gruenbacher return copied; 2253afddba49SNick Piggin } 2254afddba49SNick Piggin EXPORT_SYMBOL(generic_write_end); 2255afddba49SNick Piggin 2256afddba49SNick Piggin /* 22572e7e80f7SMatthew Wilcox (Oracle) * block_is_partially_uptodate checks whether buffers within a folio are 22588ab22b9aSHisashi Hifumi * uptodate or not. 22598ab22b9aSHisashi Hifumi * 22602e7e80f7SMatthew Wilcox (Oracle) * Returns true if all buffers which correspond to the specified part 22612e7e80f7SMatthew Wilcox (Oracle) * of the folio are uptodate. 22628ab22b9aSHisashi Hifumi */ 22632e7e80f7SMatthew Wilcox (Oracle) bool block_is_partially_uptodate(struct folio *folio, size_t from, size_t count) 22648ab22b9aSHisashi Hifumi { 22658ab22b9aSHisashi Hifumi unsigned block_start, block_end, blocksize; 22668ab22b9aSHisashi Hifumi unsigned to; 22678ab22b9aSHisashi Hifumi struct buffer_head *bh, *head; 22682e7e80f7SMatthew Wilcox (Oracle) bool ret = true; 22698ab22b9aSHisashi Hifumi 22702e7e80f7SMatthew Wilcox (Oracle) head = folio_buffers(folio); 22712e7e80f7SMatthew Wilcox (Oracle) if (!head) 22722e7e80f7SMatthew Wilcox (Oracle) return false; 227345bce8f3SLinus Torvalds blocksize = head->b_size; 22742e7e80f7SMatthew Wilcox (Oracle) to = min_t(unsigned, folio_size(folio) - from, count); 22758ab22b9aSHisashi Hifumi to = from + to; 22762e7e80f7SMatthew Wilcox (Oracle) if (from < blocksize && to > folio_size(folio) - blocksize) 22772e7e80f7SMatthew Wilcox (Oracle) return false; 22788ab22b9aSHisashi Hifumi 22798ab22b9aSHisashi Hifumi bh = head; 22808ab22b9aSHisashi Hifumi block_start = 0; 22818ab22b9aSHisashi Hifumi do { 22828ab22b9aSHisashi Hifumi block_end = block_start + blocksize; 22838ab22b9aSHisashi Hifumi if (block_end > from && block_start < to) { 22848ab22b9aSHisashi Hifumi if (!buffer_uptodate(bh)) { 22852e7e80f7SMatthew Wilcox (Oracle) ret = false; 22868ab22b9aSHisashi Hifumi break; 22878ab22b9aSHisashi Hifumi } 22888ab22b9aSHisashi Hifumi if (block_end >= to) 22898ab22b9aSHisashi Hifumi break; 22908ab22b9aSHisashi Hifumi } 22918ab22b9aSHisashi Hifumi block_start = block_end; 22928ab22b9aSHisashi Hifumi bh = bh->b_this_page; 22938ab22b9aSHisashi Hifumi } while (bh != head); 22948ab22b9aSHisashi Hifumi 22958ab22b9aSHisashi Hifumi return ret; 22968ab22b9aSHisashi Hifumi } 22978ab22b9aSHisashi Hifumi EXPORT_SYMBOL(block_is_partially_uptodate); 22988ab22b9aSHisashi Hifumi 22998ab22b9aSHisashi Hifumi /* 23002c69e205SMatthew Wilcox (Oracle) * Generic "read_folio" function for block devices that have the normal 23011da177e4SLinus Torvalds * get_block functionality. This is most of the block device filesystems. 23022c69e205SMatthew Wilcox (Oracle) * Reads the folio asynchronously --- the unlock_buffer() and 23031da177e4SLinus Torvalds * set/clear_buffer_uptodate() functions propagate buffer state into the 23042c69e205SMatthew Wilcox (Oracle) * folio once IO has completed. 23051da177e4SLinus Torvalds */ 23062c69e205SMatthew Wilcox (Oracle) int block_read_full_folio(struct folio *folio, get_block_t *get_block) 23071da177e4SLinus Torvalds { 23082c69e205SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 23091da177e4SLinus Torvalds sector_t iblock, lblock; 23101da177e4SLinus Torvalds struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 231145bce8f3SLinus Torvalds unsigned int blocksize, bbits; 23121da177e4SLinus Torvalds int nr, i; 23131da177e4SLinus Torvalds int fully_mapped = 1; 2314b7a6eb22SMatthew Wilcox (Oracle) bool page_error = false; 23154fa512ceSEric Biggers loff_t limit = i_size_read(inode); 23164fa512ceSEric Biggers 23174fa512ceSEric Biggers /* This is needed for ext4. */ 23184fa512ceSEric Biggers if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode)) 23194fa512ceSEric Biggers limit = inode->i_sb->s_maxbytes; 23201da177e4SLinus Torvalds 23212c69e205SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_large(folio), folio); 23222c69e205SMatthew Wilcox (Oracle) 2323c6c8c3e7SPankaj Raghav head = folio_create_buffers(folio, inode, 0); 232445bce8f3SLinus Torvalds blocksize = head->b_size; 232545bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 23261da177e4SLinus Torvalds 23272c69e205SMatthew Wilcox (Oracle) iblock = (sector_t)folio->index << (PAGE_SHIFT - bbits); 23284fa512ceSEric Biggers lblock = (limit+blocksize-1) >> bbits; 23291da177e4SLinus Torvalds bh = head; 23301da177e4SLinus Torvalds nr = 0; 23311da177e4SLinus Torvalds i = 0; 23321da177e4SLinus Torvalds 23331da177e4SLinus Torvalds do { 23341da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23351da177e4SLinus Torvalds continue; 23361da177e4SLinus Torvalds 23371da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2338c64610baSAndrew Morton int err = 0; 2339c64610baSAndrew Morton 23401da177e4SLinus Torvalds fully_mapped = 0; 23411da177e4SLinus Torvalds if (iblock < lblock) { 2342b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2343c64610baSAndrew Morton err = get_block(inode, iblock, bh, 0); 2344b7a6eb22SMatthew Wilcox (Oracle) if (err) { 23452c69e205SMatthew Wilcox (Oracle) folio_set_error(folio); 2346b7a6eb22SMatthew Wilcox (Oracle) page_error = true; 2347b7a6eb22SMatthew Wilcox (Oracle) } 23481da177e4SLinus Torvalds } 23491da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 23502c69e205SMatthew Wilcox (Oracle) folio_zero_range(folio, i * blocksize, 23512c69e205SMatthew Wilcox (Oracle) blocksize); 2352c64610baSAndrew Morton if (!err) 23531da177e4SLinus Torvalds set_buffer_uptodate(bh); 23541da177e4SLinus Torvalds continue; 23551da177e4SLinus Torvalds } 23561da177e4SLinus Torvalds /* 23571da177e4SLinus Torvalds * get_block() might have updated the buffer 23581da177e4SLinus Torvalds * synchronously 23591da177e4SLinus Torvalds */ 23601da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23611da177e4SLinus Torvalds continue; 23621da177e4SLinus Torvalds } 23631da177e4SLinus Torvalds arr[nr++] = bh; 23641da177e4SLinus Torvalds } while (i++, iblock++, (bh = bh->b_this_page) != head); 23651da177e4SLinus Torvalds 23661da177e4SLinus Torvalds if (fully_mapped) 23672c69e205SMatthew Wilcox (Oracle) folio_set_mappedtodisk(folio); 23681da177e4SLinus Torvalds 23691da177e4SLinus Torvalds if (!nr) { 23701da177e4SLinus Torvalds /* 23712c69e205SMatthew Wilcox (Oracle) * All buffers are uptodate - we can set the folio uptodate 23721da177e4SLinus Torvalds * as well. But not if get_block() returned an error. 23731da177e4SLinus Torvalds */ 2374b7a6eb22SMatthew Wilcox (Oracle) if (!page_error) 23752c69e205SMatthew Wilcox (Oracle) folio_mark_uptodate(folio); 23762c69e205SMatthew Wilcox (Oracle) folio_unlock(folio); 23771da177e4SLinus Torvalds return 0; 23781da177e4SLinus Torvalds } 23791da177e4SLinus Torvalds 23801da177e4SLinus Torvalds /* Stage two: lock the buffers */ 23811da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 23821da177e4SLinus Torvalds bh = arr[i]; 23831da177e4SLinus Torvalds lock_buffer(bh); 23841da177e4SLinus Torvalds mark_buffer_async_read(bh); 23851da177e4SLinus Torvalds } 23861da177e4SLinus Torvalds 23871da177e4SLinus Torvalds /* 23881da177e4SLinus Torvalds * Stage 3: start the IO. Check for uptodateness 23891da177e4SLinus Torvalds * inside the buffer lock in case another process reading 23901da177e4SLinus Torvalds * the underlying blockdev brought it uptodate (the sct fix). 23911da177e4SLinus Torvalds */ 23921da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 23931da177e4SLinus Torvalds bh = arr[i]; 23941da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23951da177e4SLinus Torvalds end_buffer_async_read(bh, 1); 23961da177e4SLinus Torvalds else 23971420c4a5SBart Van Assche submit_bh(REQ_OP_READ, bh); 23981da177e4SLinus Torvalds } 23991da177e4SLinus Torvalds return 0; 24001da177e4SLinus Torvalds } 24012c69e205SMatthew Wilcox (Oracle) EXPORT_SYMBOL(block_read_full_folio); 24021da177e4SLinus Torvalds 24031da177e4SLinus Torvalds /* utility function for filesystems that need to do work on expanding 240489e10787SNick Piggin * truncates. Uses filesystem pagecache writes to allow the filesystem to 24051da177e4SLinus Torvalds * deal with the hole. 24061da177e4SLinus Torvalds */ 240789e10787SNick Piggin int generic_cont_expand_simple(struct inode *inode, loff_t size) 24081da177e4SLinus Torvalds { 24091da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 241053b524b8SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 24111da177e4SLinus Torvalds struct page *page; 24121468c6f4SAlexander Potapenko void *fsdata = NULL; 24131da177e4SLinus Torvalds int err; 24141da177e4SLinus Torvalds 2415c08d3b0eSnpiggin@suse.de err = inode_newsize_ok(inode, size); 2416c08d3b0eSnpiggin@suse.de if (err) 24171da177e4SLinus Torvalds goto out; 24181da177e4SLinus Torvalds 241953b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(NULL, mapping, size, 0, &page, &fsdata); 242089e10787SNick Piggin if (err) 242105eb0b51SOGAWA Hirofumi goto out; 242205eb0b51SOGAWA Hirofumi 242353b524b8SMatthew Wilcox (Oracle) err = aops->write_end(NULL, mapping, size, 0, 0, page, fsdata); 242489e10787SNick Piggin BUG_ON(err > 0); 242505eb0b51SOGAWA Hirofumi 242605eb0b51SOGAWA Hirofumi out: 242705eb0b51SOGAWA Hirofumi return err; 242805eb0b51SOGAWA Hirofumi } 24291fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_cont_expand_simple); 243005eb0b51SOGAWA Hirofumi 2431f1e3af72SAdrian Bunk static int cont_expand_zero(struct file *file, struct address_space *mapping, 243289e10787SNick Piggin loff_t pos, loff_t *bytes) 243305eb0b51SOGAWA Hirofumi { 243489e10787SNick Piggin struct inode *inode = mapping->host; 243553b524b8SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 243693407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 243789e10787SNick Piggin struct page *page; 24381468c6f4SAlexander Potapenko void *fsdata = NULL; 243989e10787SNick Piggin pgoff_t index, curidx; 244089e10787SNick Piggin loff_t curpos; 244189e10787SNick Piggin unsigned zerofrom, offset, len; 244289e10787SNick Piggin int err = 0; 244305eb0b51SOGAWA Hirofumi 244409cbfeafSKirill A. Shutemov index = pos >> PAGE_SHIFT; 244509cbfeafSKirill A. Shutemov offset = pos & ~PAGE_MASK; 244689e10787SNick Piggin 244709cbfeafSKirill A. Shutemov while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { 244809cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 244989e10787SNick Piggin if (zerofrom & (blocksize-1)) { 245089e10787SNick Piggin *bytes |= (blocksize-1); 245189e10787SNick Piggin (*bytes)++; 245289e10787SNick Piggin } 245309cbfeafSKirill A. Shutemov len = PAGE_SIZE - zerofrom; 245489e10787SNick Piggin 245553b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(file, mapping, curpos, len, 245689e10787SNick Piggin &page, &fsdata); 245789e10787SNick Piggin if (err) 245889e10787SNick Piggin goto out; 2459eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 246053b524b8SMatthew Wilcox (Oracle) err = aops->write_end(file, mapping, curpos, len, len, 246189e10787SNick Piggin page, fsdata); 246289e10787SNick Piggin if (err < 0) 246389e10787SNick Piggin goto out; 246489e10787SNick Piggin BUG_ON(err != len); 246589e10787SNick Piggin err = 0; 2466061e9746SOGAWA Hirofumi 2467061e9746SOGAWA Hirofumi balance_dirty_pages_ratelimited(mapping); 2468c2ca0fcdSMikulas Patocka 246908d405c8SDavidlohr Bueso if (fatal_signal_pending(current)) { 2470c2ca0fcdSMikulas Patocka err = -EINTR; 2471c2ca0fcdSMikulas Patocka goto out; 2472c2ca0fcdSMikulas Patocka } 247389e10787SNick Piggin } 247489e10787SNick Piggin 247589e10787SNick Piggin /* page covers the boundary, find the boundary offset */ 247689e10787SNick Piggin if (index == curidx) { 247709cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 247889e10787SNick Piggin /* if we will expand the thing last block will be filled */ 247989e10787SNick Piggin if (offset <= zerofrom) { 248089e10787SNick Piggin goto out; 248189e10787SNick Piggin } 248289e10787SNick Piggin if (zerofrom & (blocksize-1)) { 248389e10787SNick Piggin *bytes |= (blocksize-1); 248489e10787SNick Piggin (*bytes)++; 248589e10787SNick Piggin } 248689e10787SNick Piggin len = offset - zerofrom; 248789e10787SNick Piggin 248853b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(file, mapping, curpos, len, 248989e10787SNick Piggin &page, &fsdata); 249089e10787SNick Piggin if (err) 249189e10787SNick Piggin goto out; 2492eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 249353b524b8SMatthew Wilcox (Oracle) err = aops->write_end(file, mapping, curpos, len, len, 249489e10787SNick Piggin page, fsdata); 249589e10787SNick Piggin if (err < 0) 249689e10787SNick Piggin goto out; 249789e10787SNick Piggin BUG_ON(err != len); 249889e10787SNick Piggin err = 0; 249989e10787SNick Piggin } 250089e10787SNick Piggin out: 250189e10787SNick Piggin return err; 25021da177e4SLinus Torvalds } 25031da177e4SLinus Torvalds 25041da177e4SLinus Torvalds /* 25051da177e4SLinus Torvalds * For moronic filesystems that do not allow holes in file. 25061da177e4SLinus Torvalds * We may have to extend the file. 25071da177e4SLinus Torvalds */ 2508282dc178SChristoph Hellwig int cont_write_begin(struct file *file, struct address_space *mapping, 2509be3bbbc5SMatthew Wilcox (Oracle) loff_t pos, unsigned len, 251089e10787SNick Piggin struct page **pagep, void **fsdata, 251189e10787SNick Piggin get_block_t *get_block, loff_t *bytes) 25121da177e4SLinus Torvalds { 25131da177e4SLinus Torvalds struct inode *inode = mapping->host; 251493407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 251593407472SFabian Frederick unsigned int zerofrom; 251689e10787SNick Piggin int err; 25171da177e4SLinus Torvalds 251889e10787SNick Piggin err = cont_expand_zero(file, mapping, pos, bytes); 251989e10787SNick Piggin if (err) 2520155130a4SChristoph Hellwig return err; 25211da177e4SLinus Torvalds 252209cbfeafSKirill A. Shutemov zerofrom = *bytes & ~PAGE_MASK; 252389e10787SNick Piggin if (pos+len > *bytes && zerofrom & (blocksize-1)) { 25241da177e4SLinus Torvalds *bytes |= (blocksize-1); 25251da177e4SLinus Torvalds (*bytes)++; 25261da177e4SLinus Torvalds } 25271da177e4SLinus Torvalds 2528b3992d1eSMatthew Wilcox (Oracle) return block_write_begin(mapping, pos, len, pagep, get_block); 25291da177e4SLinus Torvalds } 25301fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(cont_write_begin); 25311da177e4SLinus Torvalds 25321da177e4SLinus Torvalds int block_commit_write(struct page *page, unsigned from, unsigned to) 25331da177e4SLinus Torvalds { 25348c6cb3e3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 25358c6cb3e3SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 25368c6cb3e3SMatthew Wilcox (Oracle) __block_commit_write(inode, folio, from, to); 25371da177e4SLinus Torvalds return 0; 25381da177e4SLinus Torvalds } 25391fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_commit_write); 25401da177e4SLinus Torvalds 254154171690SDavid Chinner /* 254254171690SDavid Chinner * block_page_mkwrite() is not allowed to change the file size as it gets 254354171690SDavid Chinner * called from a page fault handler when a page is first dirtied. Hence we must 254454171690SDavid Chinner * be careful to check for EOF conditions here. We set the page up correctly 254554171690SDavid Chinner * for a written page which means we get ENOSPC checking when writing into 254654171690SDavid Chinner * holes and correct delalloc and unwritten extent mapping on filesystems that 254754171690SDavid Chinner * support these features. 254854171690SDavid Chinner * 254954171690SDavid Chinner * We are not allowed to take the i_mutex here so we have to play games to 255054171690SDavid Chinner * protect against truncate races as the page could now be beyond EOF. Because 25517bb46a67Snpiggin@suse.de * truncate writes the inode size before removing pages, once we have the 255254171690SDavid Chinner * page lock we can determine safely if the page is beyond EOF. If it is not 255354171690SDavid Chinner * beyond EOF, then the page is guaranteed safe against truncation until we 255454171690SDavid Chinner * unlock the page. 2555ea13a864SJan Kara * 255614da9200SJan Kara * Direct callers of this function should protect against filesystem freezing 25575c500029SRoss Zwisler * using sb_start_pagefault() - sb_end_pagefault() functions. 255854171690SDavid Chinner */ 25595c500029SRoss Zwisler int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 256054171690SDavid Chinner get_block_t get_block) 256154171690SDavid Chinner { 2562fe181377SMatthew Wilcox (Oracle) struct folio *folio = page_folio(vmf->page); 2563496ad9aaSAl Viro struct inode *inode = file_inode(vma->vm_file); 256454171690SDavid Chinner unsigned long end; 256554171690SDavid Chinner loff_t size; 256624da4fabSJan Kara int ret; 256754171690SDavid Chinner 2568fe181377SMatthew Wilcox (Oracle) folio_lock(folio); 256954171690SDavid Chinner size = i_size_read(inode); 2570fe181377SMatthew Wilcox (Oracle) if ((folio->mapping != inode->i_mapping) || 2571fe181377SMatthew Wilcox (Oracle) (folio_pos(folio) >= size)) { 257224da4fabSJan Kara /* We overload EFAULT to mean page got truncated */ 257324da4fabSJan Kara ret = -EFAULT; 257424da4fabSJan Kara goto out_unlock; 257554171690SDavid Chinner } 257654171690SDavid Chinner 2577fe181377SMatthew Wilcox (Oracle) end = folio_size(folio); 2578fe181377SMatthew Wilcox (Oracle) /* folio is wholly or partially inside EOF */ 2579fe181377SMatthew Wilcox (Oracle) if (folio_pos(folio) + end > size) 2580fe181377SMatthew Wilcox (Oracle) end = size - folio_pos(folio); 258154171690SDavid Chinner 2582fe181377SMatthew Wilcox (Oracle) ret = __block_write_begin_int(folio, 0, end, get_block, NULL); 258354171690SDavid Chinner if (!ret) 25848c6cb3e3SMatthew Wilcox (Oracle) ret = __block_commit_write(inode, folio, 0, end); 258554171690SDavid Chinner 258624da4fabSJan Kara if (unlikely(ret < 0)) 258724da4fabSJan Kara goto out_unlock; 2588fe181377SMatthew Wilcox (Oracle) folio_mark_dirty(folio); 2589fe181377SMatthew Wilcox (Oracle) folio_wait_stable(folio); 259024da4fabSJan Kara return 0; 259124da4fabSJan Kara out_unlock: 2592fe181377SMatthew Wilcox (Oracle) folio_unlock(folio); 259354171690SDavid Chinner return ret; 259454171690SDavid Chinner } 25951fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_page_mkwrite); 25961da177e4SLinus Torvalds 25971da177e4SLinus Torvalds int block_truncate_page(struct address_space *mapping, 25981da177e4SLinus Torvalds loff_t from, get_block_t *get_block) 25991da177e4SLinus Torvalds { 260009cbfeafSKirill A. Shutemov pgoff_t index = from >> PAGE_SHIFT; 260109cbfeafSKirill A. Shutemov unsigned offset = from & (PAGE_SIZE-1); 26021da177e4SLinus Torvalds unsigned blocksize; 260354b21a79SAndrew Morton sector_t iblock; 26041da177e4SLinus Torvalds unsigned length, pos; 26051da177e4SLinus Torvalds struct inode *inode = mapping->host; 26061da177e4SLinus Torvalds struct page *page; 26071da177e4SLinus Torvalds struct buffer_head *bh; 2608dc7cb2d2SJiapeng Chong int err = 0; 26091da177e4SLinus Torvalds 261093407472SFabian Frederick blocksize = i_blocksize(inode); 26111da177e4SLinus Torvalds length = offset & (blocksize - 1); 26121da177e4SLinus Torvalds 26131da177e4SLinus Torvalds /* Block boundary? Nothing to do */ 26141da177e4SLinus Torvalds if (!length) 26151da177e4SLinus Torvalds return 0; 26161da177e4SLinus Torvalds 26171da177e4SLinus Torvalds length = blocksize - length; 261809cbfeafSKirill A. Shutemov iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); 26191da177e4SLinus Torvalds 26201da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 26211da177e4SLinus Torvalds if (!page) 2622dc7cb2d2SJiapeng Chong return -ENOMEM; 26231da177e4SLinus Torvalds 26241da177e4SLinus Torvalds if (!page_has_buffers(page)) 26251da177e4SLinus Torvalds create_empty_buffers(page, blocksize, 0); 26261da177e4SLinus Torvalds 26271da177e4SLinus Torvalds /* Find the buffer that contains "offset" */ 26281da177e4SLinus Torvalds bh = page_buffers(page); 26291da177e4SLinus Torvalds pos = blocksize; 26301da177e4SLinus Torvalds while (offset >= pos) { 26311da177e4SLinus Torvalds bh = bh->b_this_page; 26321da177e4SLinus Torvalds iblock++; 26331da177e4SLinus Torvalds pos += blocksize; 26341da177e4SLinus Torvalds } 26351da177e4SLinus Torvalds 26361da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2637b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 26381da177e4SLinus Torvalds err = get_block(inode, iblock, bh, 0); 26391da177e4SLinus Torvalds if (err) 26401da177e4SLinus Torvalds goto unlock; 26411da177e4SLinus Torvalds /* unmapped? It's a hole - nothing to do */ 26421da177e4SLinus Torvalds if (!buffer_mapped(bh)) 26431da177e4SLinus Torvalds goto unlock; 26441da177e4SLinus Torvalds } 26451da177e4SLinus Torvalds 26461da177e4SLinus Torvalds /* Ok, it's mapped. Make sure it's up-to-date */ 26471da177e4SLinus Torvalds if (PageUptodate(page)) 26481da177e4SLinus Torvalds set_buffer_uptodate(bh); 26491da177e4SLinus Torvalds 265033a266ddSDavid Chinner if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { 2651e7ea1129SZhang Yi err = bh_read(bh, 0); 26521da177e4SLinus Torvalds /* Uhhuh. Read error. Complain and punt. */ 2653e7ea1129SZhang Yi if (err < 0) 26541da177e4SLinus Torvalds goto unlock; 26551da177e4SLinus Torvalds } 26561da177e4SLinus Torvalds 2657eebd2aa3SChristoph Lameter zero_user(page, offset, length); 26581da177e4SLinus Torvalds mark_buffer_dirty(bh); 26591da177e4SLinus Torvalds 26601da177e4SLinus Torvalds unlock: 26611da177e4SLinus Torvalds unlock_page(page); 266209cbfeafSKirill A. Shutemov put_page(page); 2663dc7cb2d2SJiapeng Chong 26641da177e4SLinus Torvalds return err; 26651da177e4SLinus Torvalds } 26661fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_truncate_page); 26671da177e4SLinus Torvalds 26681da177e4SLinus Torvalds /* 26691da177e4SLinus Torvalds * The generic ->writepage function for buffer-backed address_spaces 26701da177e4SLinus Torvalds */ 26711b938c08SMatthew Wilcox int block_write_full_page(struct page *page, get_block_t *get_block, 26721b938c08SMatthew Wilcox struct writeback_control *wbc) 26731da177e4SLinus Torvalds { 267453418a18SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 2675bb0ea598SMatthew Wilcox (Oracle) struct inode * const inode = folio->mapping->host; 26761da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 26771da177e4SLinus Torvalds 2678bb0ea598SMatthew Wilcox (Oracle) /* Is the folio fully inside i_size? */ 2679bb0ea598SMatthew Wilcox (Oracle) if (folio_pos(folio) + folio_size(folio) <= i_size) 268053418a18SMatthew Wilcox (Oracle) return __block_write_full_folio(inode, folio, get_block, wbc, 26811b938c08SMatthew Wilcox end_buffer_async_write); 26821da177e4SLinus Torvalds 2683bb0ea598SMatthew Wilcox (Oracle) /* Is the folio fully outside i_size? (truncate in progress) */ 2684bb0ea598SMatthew Wilcox (Oracle) if (folio_pos(folio) >= i_size) { 268553418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 26861da177e4SLinus Torvalds return 0; /* don't care */ 26871da177e4SLinus Torvalds } 26881da177e4SLinus Torvalds 26891da177e4SLinus Torvalds /* 2690bb0ea598SMatthew Wilcox (Oracle) * The folio straddles i_size. It must be zeroed out on each and every 26912a61aa40SAdam Buchbinder * writepage invocation because it may be mmapped. "A file is mapped 26921da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 26931da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 26941da177e4SLinus Torvalds * writes to that region are not written out to the file." 26951da177e4SLinus Torvalds */ 2696bb0ea598SMatthew Wilcox (Oracle) folio_zero_segment(folio, offset_in_folio(folio, i_size), 2697bb0ea598SMatthew Wilcox (Oracle) folio_size(folio)); 269853418a18SMatthew Wilcox (Oracle) return __block_write_full_folio(inode, folio, get_block, wbc, 269935c80d5fSChris Mason end_buffer_async_write); 270035c80d5fSChris Mason } 27011fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_write_full_page); 270235c80d5fSChris Mason 27031da177e4SLinus Torvalds sector_t generic_block_bmap(struct address_space *mapping, sector_t block, 27041da177e4SLinus Torvalds get_block_t *get_block) 27051da177e4SLinus Torvalds { 27061da177e4SLinus Torvalds struct inode *inode = mapping->host; 27072a527d68SAlexander Potapenko struct buffer_head tmp = { 27082a527d68SAlexander Potapenko .b_size = i_blocksize(inode), 27092a527d68SAlexander Potapenko }; 27102a527d68SAlexander Potapenko 27111da177e4SLinus Torvalds get_block(inode, block, &tmp, 0); 27121da177e4SLinus Torvalds return tmp.b_blocknr; 27131da177e4SLinus Torvalds } 27141fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_block_bmap); 27151da177e4SLinus Torvalds 27164246a0b6SChristoph Hellwig static void end_bio_bh_io_sync(struct bio *bio) 27171da177e4SLinus Torvalds { 27181da177e4SLinus Torvalds struct buffer_head *bh = bio->bi_private; 27191da177e4SLinus Torvalds 2720b7c44ed9SJens Axboe if (unlikely(bio_flagged(bio, BIO_QUIET))) 272108bafc03SKeith Mannthey set_bit(BH_Quiet, &bh->b_state); 272208bafc03SKeith Mannthey 27234e4cbee9SChristoph Hellwig bh->b_end_io(bh, !bio->bi_status); 27241da177e4SLinus Torvalds bio_put(bio); 27251da177e4SLinus Torvalds } 27261da177e4SLinus Torvalds 27275bdf402aSRitesh Harjani (IBM) static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh, 27281420c4a5SBart Van Assche struct writeback_control *wbc) 27291da177e4SLinus Torvalds { 27301420c4a5SBart Van Assche const enum req_op op = opf & REQ_OP_MASK; 27311da177e4SLinus Torvalds struct bio *bio; 27321da177e4SLinus Torvalds 27331da177e4SLinus Torvalds BUG_ON(!buffer_locked(bh)); 27341da177e4SLinus Torvalds BUG_ON(!buffer_mapped(bh)); 27351da177e4SLinus Torvalds BUG_ON(!bh->b_end_io); 27368fb0e342SAneesh Kumar K.V BUG_ON(buffer_delay(bh)); 27378fb0e342SAneesh Kumar K.V BUG_ON(buffer_unwritten(bh)); 27381da177e4SLinus Torvalds 273948fd4f93SJens Axboe /* 274048fd4f93SJens Axboe * Only clear out a write error when rewriting 27411da177e4SLinus Torvalds */ 27422a222ca9SMike Christie if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) 27431da177e4SLinus Torvalds clear_buffer_write_io_error(bh); 27441da177e4SLinus Torvalds 274507888c66SChristoph Hellwig if (buffer_meta(bh)) 27461420c4a5SBart Van Assche opf |= REQ_META; 274707888c66SChristoph Hellwig if (buffer_prio(bh)) 27481420c4a5SBart Van Assche opf |= REQ_PRIO; 274907888c66SChristoph Hellwig 27501420c4a5SBart Van Assche bio = bio_alloc(bh->b_bdev, 1, opf, GFP_NOIO); 27511da177e4SLinus Torvalds 27524f74d15fSEric Biggers fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO); 27534f74d15fSEric Biggers 27544f024f37SKent Overstreet bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); 27551da177e4SLinus Torvalds 27566cf66b4cSKent Overstreet bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); 27576cf66b4cSKent Overstreet BUG_ON(bio->bi_iter.bi_size != bh->b_size); 27581da177e4SLinus Torvalds 27591da177e4SLinus Torvalds bio->bi_end_io = end_bio_bh_io_sync; 27601da177e4SLinus Torvalds bio->bi_private = bh; 27611da177e4SLinus Torvalds 276283c9c547SMing Lei /* Take care of bh's that straddle the end of the device */ 276383c9c547SMing Lei guard_bio_eod(bio); 276483c9c547SMing Lei 2765fd42df30SDennis Zhou if (wbc) { 2766fd42df30SDennis Zhou wbc_init_bio(wbc, bio); 276734e51a5eSTejun Heo wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size); 2768fd42df30SDennis Zhou } 2769fd42df30SDennis Zhou 27704e49ea4aSMike Christie submit_bio(bio); 27711da177e4SLinus Torvalds } 2772bafc0dbaSTejun Heo 27735bdf402aSRitesh Harjani (IBM) void submit_bh(blk_opf_t opf, struct buffer_head *bh) 277471368511SDarrick J. Wong { 27755bdf402aSRitesh Harjani (IBM) submit_bh_wbc(opf, bh, NULL); 277671368511SDarrick J. Wong } 27771fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(submit_bh); 27781da177e4SLinus Torvalds 27793ae72869SBart Van Assche void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) 27809cb569d6SChristoph Hellwig { 27819cb569d6SChristoph Hellwig lock_buffer(bh); 27829cb569d6SChristoph Hellwig if (!test_clear_buffer_dirty(bh)) { 27839cb569d6SChristoph Hellwig unlock_buffer(bh); 27849cb569d6SChristoph Hellwig return; 27859cb569d6SChristoph Hellwig } 27869cb569d6SChristoph Hellwig bh->b_end_io = end_buffer_write_sync; 27879cb569d6SChristoph Hellwig get_bh(bh); 27881420c4a5SBart Van Assche submit_bh(REQ_OP_WRITE | op_flags, bh); 27899cb569d6SChristoph Hellwig } 27909cb569d6SChristoph Hellwig EXPORT_SYMBOL(write_dirty_buffer); 27919cb569d6SChristoph Hellwig 27921da177e4SLinus Torvalds /* 27931da177e4SLinus Torvalds * For a data-integrity writeout, we need to wait upon any in-progress I/O 27941da177e4SLinus Torvalds * and then start new I/O and then wait upon it. The caller must have a ref on 27951da177e4SLinus Torvalds * the buffer_head. 27961da177e4SLinus Torvalds */ 27973ae72869SBart Van Assche int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) 27981da177e4SLinus Torvalds { 27991da177e4SLinus Torvalds WARN_ON(atomic_read(&bh->b_count) < 1); 28001da177e4SLinus Torvalds lock_buffer(bh); 28011da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 2802377254b2SXianting Tian /* 2803377254b2SXianting Tian * The bh should be mapped, but it might not be if the 2804377254b2SXianting Tian * device was hot-removed. Not much we can do but fail the I/O. 2805377254b2SXianting Tian */ 2806377254b2SXianting Tian if (!buffer_mapped(bh)) { 2807377254b2SXianting Tian unlock_buffer(bh); 2808377254b2SXianting Tian return -EIO; 2809377254b2SXianting Tian } 2810377254b2SXianting Tian 28111da177e4SLinus Torvalds get_bh(bh); 28121da177e4SLinus Torvalds bh->b_end_io = end_buffer_write_sync; 2813ab620620SRitesh Harjani (IBM) submit_bh(REQ_OP_WRITE | op_flags, bh); 28141da177e4SLinus Torvalds wait_on_buffer(bh); 2815ab620620SRitesh Harjani (IBM) if (!buffer_uptodate(bh)) 2816ab620620SRitesh Harjani (IBM) return -EIO; 28171da177e4SLinus Torvalds } else { 28181da177e4SLinus Torvalds unlock_buffer(bh); 28191da177e4SLinus Torvalds } 2820ab620620SRitesh Harjani (IBM) return 0; 28211da177e4SLinus Torvalds } 282287e99511SChristoph Hellwig EXPORT_SYMBOL(__sync_dirty_buffer); 282387e99511SChristoph Hellwig 282487e99511SChristoph Hellwig int sync_dirty_buffer(struct buffer_head *bh) 282587e99511SChristoph Hellwig { 282670fd7614SChristoph Hellwig return __sync_dirty_buffer(bh, REQ_SYNC); 282787e99511SChristoph Hellwig } 28281fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(sync_dirty_buffer); 28291da177e4SLinus Torvalds 28301da177e4SLinus Torvalds /* 283168189fefSMatthew Wilcox (Oracle) * try_to_free_buffers() checks if all the buffers on this particular folio 28321da177e4SLinus Torvalds * are unused, and releases them if so. 28331da177e4SLinus Torvalds * 28341da177e4SLinus Torvalds * Exclusion against try_to_free_buffers may be obtained by either 283568189fefSMatthew Wilcox (Oracle) * locking the folio or by holding its mapping's private_lock. 28361da177e4SLinus Torvalds * 283768189fefSMatthew Wilcox (Oracle) * If the folio is dirty but all the buffers are clean then we need to 283868189fefSMatthew Wilcox (Oracle) * be sure to mark the folio clean as well. This is because the folio 28391da177e4SLinus Torvalds * may be against a block device, and a later reattachment of buffers 284068189fefSMatthew Wilcox (Oracle) * to a dirty folio will set *all* buffers dirty. Which would corrupt 28411da177e4SLinus Torvalds * filesystem data on the same device. 28421da177e4SLinus Torvalds * 284368189fefSMatthew Wilcox (Oracle) * The same applies to regular filesystem folios: if all the buffers are 284468189fefSMatthew Wilcox (Oracle) * clean then we set the folio clean and proceed. To do that, we require 2845e621900aSMatthew Wilcox (Oracle) * total exclusion from block_dirty_folio(). That is obtained with 28461da177e4SLinus Torvalds * private_lock. 28471da177e4SLinus Torvalds * 28481da177e4SLinus Torvalds * try_to_free_buffers() is non-blocking. 28491da177e4SLinus Torvalds */ 28501da177e4SLinus Torvalds static inline int buffer_busy(struct buffer_head *bh) 28511da177e4SLinus Torvalds { 28521da177e4SLinus Torvalds return atomic_read(&bh->b_count) | 28531da177e4SLinus Torvalds (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); 28541da177e4SLinus Torvalds } 28551da177e4SLinus Torvalds 285664394763SMatthew Wilcox (Oracle) static bool 285764394763SMatthew Wilcox (Oracle) drop_buffers(struct folio *folio, struct buffer_head **buffers_to_free) 28581da177e4SLinus Torvalds { 285964394763SMatthew Wilcox (Oracle) struct buffer_head *head = folio_buffers(folio); 28601da177e4SLinus Torvalds struct buffer_head *bh; 28611da177e4SLinus Torvalds 28621da177e4SLinus Torvalds bh = head; 28631da177e4SLinus Torvalds do { 28641da177e4SLinus Torvalds if (buffer_busy(bh)) 28651da177e4SLinus Torvalds goto failed; 28661da177e4SLinus Torvalds bh = bh->b_this_page; 28671da177e4SLinus Torvalds } while (bh != head); 28681da177e4SLinus Torvalds 28691da177e4SLinus Torvalds do { 28701da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 28711da177e4SLinus Torvalds 2872535ee2fbSJan Kara if (bh->b_assoc_map) 28731da177e4SLinus Torvalds __remove_assoc_queue(bh); 28741da177e4SLinus Torvalds bh = next; 28751da177e4SLinus Torvalds } while (bh != head); 28761da177e4SLinus Torvalds *buffers_to_free = head; 287764394763SMatthew Wilcox (Oracle) folio_detach_private(folio); 287864394763SMatthew Wilcox (Oracle) return true; 28791da177e4SLinus Torvalds failed: 288064394763SMatthew Wilcox (Oracle) return false; 28811da177e4SLinus Torvalds } 28821da177e4SLinus Torvalds 288368189fefSMatthew Wilcox (Oracle) bool try_to_free_buffers(struct folio *folio) 28841da177e4SLinus Torvalds { 288568189fefSMatthew Wilcox (Oracle) struct address_space * const mapping = folio->mapping; 28861da177e4SLinus Torvalds struct buffer_head *buffers_to_free = NULL; 288768189fefSMatthew Wilcox (Oracle) bool ret = 0; 28881da177e4SLinus Torvalds 288968189fefSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 289068189fefSMatthew Wilcox (Oracle) if (folio_test_writeback(folio)) 289168189fefSMatthew Wilcox (Oracle) return false; 28921da177e4SLinus Torvalds 28931da177e4SLinus Torvalds if (mapping == NULL) { /* can this still happen? */ 289464394763SMatthew Wilcox (Oracle) ret = drop_buffers(folio, &buffers_to_free); 28951da177e4SLinus Torvalds goto out; 28961da177e4SLinus Torvalds } 28971da177e4SLinus Torvalds 28981da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 289964394763SMatthew Wilcox (Oracle) ret = drop_buffers(folio, &buffers_to_free); 2900ecdfc978SLinus Torvalds 2901ecdfc978SLinus Torvalds /* 2902ecdfc978SLinus Torvalds * If the filesystem writes its buffers by hand (eg ext3) 290368189fefSMatthew Wilcox (Oracle) * then we can have clean buffers against a dirty folio. We 290468189fefSMatthew Wilcox (Oracle) * clean the folio here; otherwise the VM will never notice 2905ecdfc978SLinus Torvalds * that the filesystem did any IO at all. 2906ecdfc978SLinus Torvalds * 2907ecdfc978SLinus Torvalds * Also, during truncate, discard_buffer will have marked all 290868189fefSMatthew Wilcox (Oracle) * the folio's buffers clean. We discover that here and clean 290968189fefSMatthew Wilcox (Oracle) * the folio also. 291087df7241SNick Piggin * 291187df7241SNick Piggin * private_lock must be held over this entire operation in order 2912e621900aSMatthew Wilcox (Oracle) * to synchronise against block_dirty_folio and prevent the 291387df7241SNick Piggin * dirty bit from being lost. 2914ecdfc978SLinus Torvalds */ 291511f81becSTejun Heo if (ret) 291668189fefSMatthew Wilcox (Oracle) folio_cancel_dirty(folio); 291787df7241SNick Piggin spin_unlock(&mapping->private_lock); 29181da177e4SLinus Torvalds out: 29191da177e4SLinus Torvalds if (buffers_to_free) { 29201da177e4SLinus Torvalds struct buffer_head *bh = buffers_to_free; 29211da177e4SLinus Torvalds 29221da177e4SLinus Torvalds do { 29231da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 29241da177e4SLinus Torvalds free_buffer_head(bh); 29251da177e4SLinus Torvalds bh = next; 29261da177e4SLinus Torvalds } while (bh != buffers_to_free); 29271da177e4SLinus Torvalds } 29281da177e4SLinus Torvalds return ret; 29291da177e4SLinus Torvalds } 29301da177e4SLinus Torvalds EXPORT_SYMBOL(try_to_free_buffers); 29311da177e4SLinus Torvalds 29321da177e4SLinus Torvalds /* 29331da177e4SLinus Torvalds * Buffer-head allocation 29341da177e4SLinus Torvalds */ 2935a0a9b043SShai Fultheim static struct kmem_cache *bh_cachep __read_mostly; 29361da177e4SLinus Torvalds 29371da177e4SLinus Torvalds /* 29381da177e4SLinus Torvalds * Once the number of bh's in the machine exceeds this level, we start 29391da177e4SLinus Torvalds * stripping them in writeback. 29401da177e4SLinus Torvalds */ 294143be594aSZhang Yanfei static unsigned long max_buffer_heads; 29421da177e4SLinus Torvalds 29431da177e4SLinus Torvalds int buffer_heads_over_limit; 29441da177e4SLinus Torvalds 29451da177e4SLinus Torvalds struct bh_accounting { 29461da177e4SLinus Torvalds int nr; /* Number of live bh's */ 29471da177e4SLinus Torvalds int ratelimit; /* Limit cacheline bouncing */ 29481da177e4SLinus Torvalds }; 29491da177e4SLinus Torvalds 29501da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; 29511da177e4SLinus Torvalds 29521da177e4SLinus Torvalds static void recalc_bh_state(void) 29531da177e4SLinus Torvalds { 29541da177e4SLinus Torvalds int i; 29551da177e4SLinus Torvalds int tot = 0; 29561da177e4SLinus Torvalds 2957ee1be862SChristoph Lameter if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) 29581da177e4SLinus Torvalds return; 2959c7b92516SChristoph Lameter __this_cpu_write(bh_accounting.ratelimit, 0); 29608a143426SEric Dumazet for_each_online_cpu(i) 29611da177e4SLinus Torvalds tot += per_cpu(bh_accounting, i).nr; 29621da177e4SLinus Torvalds buffer_heads_over_limit = (tot > max_buffer_heads); 29631da177e4SLinus Torvalds } 29641da177e4SLinus Torvalds 2965dd0fc66fSAl Viro struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) 29661da177e4SLinus Torvalds { 2967019b4d12SRichard Kennedy struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); 29681da177e4SLinus Torvalds if (ret) { 2969a35afb83SChristoph Lameter INIT_LIST_HEAD(&ret->b_assoc_buffers); 2970f1e67e35SThomas Gleixner spin_lock_init(&ret->b_uptodate_lock); 2971c7b92516SChristoph Lameter preempt_disable(); 2972c7b92516SChristoph Lameter __this_cpu_inc(bh_accounting.nr); 29731da177e4SLinus Torvalds recalc_bh_state(); 2974c7b92516SChristoph Lameter preempt_enable(); 29751da177e4SLinus Torvalds } 29761da177e4SLinus Torvalds return ret; 29771da177e4SLinus Torvalds } 29781da177e4SLinus Torvalds EXPORT_SYMBOL(alloc_buffer_head); 29791da177e4SLinus Torvalds 29801da177e4SLinus Torvalds void free_buffer_head(struct buffer_head *bh) 29811da177e4SLinus Torvalds { 29821da177e4SLinus Torvalds BUG_ON(!list_empty(&bh->b_assoc_buffers)); 29831da177e4SLinus Torvalds kmem_cache_free(bh_cachep, bh); 2984c7b92516SChristoph Lameter preempt_disable(); 2985c7b92516SChristoph Lameter __this_cpu_dec(bh_accounting.nr); 29861da177e4SLinus Torvalds recalc_bh_state(); 2987c7b92516SChristoph Lameter preempt_enable(); 29881da177e4SLinus Torvalds } 29891da177e4SLinus Torvalds EXPORT_SYMBOL(free_buffer_head); 29901da177e4SLinus Torvalds 2991fc4d24c9SSebastian Andrzej Siewior static int buffer_exit_cpu_dead(unsigned int cpu) 29921da177e4SLinus Torvalds { 29931da177e4SLinus Torvalds int i; 29941da177e4SLinus Torvalds struct bh_lru *b = &per_cpu(bh_lrus, cpu); 29951da177e4SLinus Torvalds 29961da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 29971da177e4SLinus Torvalds brelse(b->bhs[i]); 29981da177e4SLinus Torvalds b->bhs[i] = NULL; 29991da177e4SLinus Torvalds } 3000c7b92516SChristoph Lameter this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); 30018a143426SEric Dumazet per_cpu(bh_accounting, cpu).nr = 0; 3002fc4d24c9SSebastian Andrzej Siewior return 0; 30031da177e4SLinus Torvalds } 30041da177e4SLinus Torvalds 3005389d1b08SAneesh Kumar K.V /** 3006a6b91919SRandy Dunlap * bh_uptodate_or_lock - Test whether the buffer is uptodate 3007389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3008389d1b08SAneesh Kumar K.V * 3009389d1b08SAneesh Kumar K.V * Return true if the buffer is up-to-date and false, 3010389d1b08SAneesh Kumar K.V * with the buffer locked, if not. 3011389d1b08SAneesh Kumar K.V */ 3012389d1b08SAneesh Kumar K.V int bh_uptodate_or_lock(struct buffer_head *bh) 3013389d1b08SAneesh Kumar K.V { 3014389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) { 3015389d1b08SAneesh Kumar K.V lock_buffer(bh); 3016389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) 3017389d1b08SAneesh Kumar K.V return 0; 3018389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3019389d1b08SAneesh Kumar K.V } 3020389d1b08SAneesh Kumar K.V return 1; 3021389d1b08SAneesh Kumar K.V } 3022389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_uptodate_or_lock); 3023389d1b08SAneesh Kumar K.V 3024389d1b08SAneesh Kumar K.V /** 3025fdee117eSZhang Yi * __bh_read - Submit read for a locked buffer 3026389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3027fdee117eSZhang Yi * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ 3028fdee117eSZhang Yi * @wait: wait until reading finish 3029389d1b08SAneesh Kumar K.V * 3030fdee117eSZhang Yi * Returns zero on success or don't wait, and -EIO on error. 3031389d1b08SAneesh Kumar K.V */ 3032fdee117eSZhang Yi int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait) 3033389d1b08SAneesh Kumar K.V { 3034fdee117eSZhang Yi int ret = 0; 3035389d1b08SAneesh Kumar K.V 3036fdee117eSZhang Yi BUG_ON(!buffer_locked(bh)); 3037389d1b08SAneesh Kumar K.V 3038389d1b08SAneesh Kumar K.V get_bh(bh); 3039389d1b08SAneesh Kumar K.V bh->b_end_io = end_buffer_read_sync; 3040fdee117eSZhang Yi submit_bh(REQ_OP_READ | op_flags, bh); 3041fdee117eSZhang Yi if (wait) { 3042389d1b08SAneesh Kumar K.V wait_on_buffer(bh); 3043fdee117eSZhang Yi if (!buffer_uptodate(bh)) 3044fdee117eSZhang Yi ret = -EIO; 3045389d1b08SAneesh Kumar K.V } 3046fdee117eSZhang Yi return ret; 3047fdee117eSZhang Yi } 3048fdee117eSZhang Yi EXPORT_SYMBOL(__bh_read); 3049fdee117eSZhang Yi 3050fdee117eSZhang Yi /** 3051fdee117eSZhang Yi * __bh_read_batch - Submit read for a batch of unlocked buffers 3052fdee117eSZhang Yi * @nr: entry number of the buffer batch 3053fdee117eSZhang Yi * @bhs: a batch of struct buffer_head 3054fdee117eSZhang Yi * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ 3055fdee117eSZhang Yi * @force_lock: force to get a lock on the buffer if set, otherwise drops any 3056fdee117eSZhang Yi * buffer that cannot lock. 3057fdee117eSZhang Yi * 3058fdee117eSZhang Yi * Returns zero on success or don't wait, and -EIO on error. 3059fdee117eSZhang Yi */ 3060fdee117eSZhang Yi void __bh_read_batch(int nr, struct buffer_head *bhs[], 3061fdee117eSZhang Yi blk_opf_t op_flags, bool force_lock) 3062fdee117eSZhang Yi { 3063fdee117eSZhang Yi int i; 3064fdee117eSZhang Yi 3065fdee117eSZhang Yi for (i = 0; i < nr; i++) { 3066fdee117eSZhang Yi struct buffer_head *bh = bhs[i]; 3067fdee117eSZhang Yi 3068fdee117eSZhang Yi if (buffer_uptodate(bh)) 3069fdee117eSZhang Yi continue; 3070fdee117eSZhang Yi 3071fdee117eSZhang Yi if (force_lock) 3072fdee117eSZhang Yi lock_buffer(bh); 3073fdee117eSZhang Yi else 3074fdee117eSZhang Yi if (!trylock_buffer(bh)) 3075fdee117eSZhang Yi continue; 3076fdee117eSZhang Yi 3077fdee117eSZhang Yi if (buffer_uptodate(bh)) { 3078fdee117eSZhang Yi unlock_buffer(bh); 3079fdee117eSZhang Yi continue; 3080fdee117eSZhang Yi } 3081fdee117eSZhang Yi 3082fdee117eSZhang Yi bh->b_end_io = end_buffer_read_sync; 3083fdee117eSZhang Yi get_bh(bh); 3084fdee117eSZhang Yi submit_bh(REQ_OP_READ | op_flags, bh); 3085fdee117eSZhang Yi } 3086fdee117eSZhang Yi } 3087fdee117eSZhang Yi EXPORT_SYMBOL(__bh_read_batch); 3088389d1b08SAneesh Kumar K.V 30891da177e4SLinus Torvalds void __init buffer_init(void) 30901da177e4SLinus Torvalds { 309143be594aSZhang Yanfei unsigned long nrpages; 3092fc4d24c9SSebastian Andrzej Siewior int ret; 30931da177e4SLinus Torvalds 3094b98938c3SChristoph Lameter bh_cachep = kmem_cache_create("buffer_head", 3095b98938c3SChristoph Lameter sizeof(struct buffer_head), 0, 3096b98938c3SChristoph Lameter (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| 3097b98938c3SChristoph Lameter SLAB_MEM_SPREAD), 3098019b4d12SRichard Kennedy NULL); 30991da177e4SLinus Torvalds 31001da177e4SLinus Torvalds /* 31011da177e4SLinus Torvalds * Limit the bh occupancy to 10% of ZONE_NORMAL 31021da177e4SLinus Torvalds */ 31031da177e4SLinus Torvalds nrpages = (nr_free_buffer_pages() * 10) / 100; 31041da177e4SLinus Torvalds max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); 3105fc4d24c9SSebastian Andrzej Siewior ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", 3106fc4d24c9SSebastian Andrzej Siewior NULL, buffer_exit_cpu_dead); 3107fc4d24c9SSebastian Andrzej Siewior WARN_ON(ret < 0); 31081da177e4SLinus Torvalds } 3109