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 /* 9371da177e4SLinus Torvalds * Initialise the state of a blockdev page's buffers. 9381da177e4SLinus Torvalds */ 939676ce6d5SHugh Dickins static sector_t 9401da177e4SLinus Torvalds init_page_buffers(struct page *page, struct block_device *bdev, 9411da177e4SLinus Torvalds sector_t block, int size) 9421da177e4SLinus Torvalds { 9431da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 9441da177e4SLinus Torvalds struct buffer_head *bh = head; 9451da177e4SLinus Torvalds int uptodate = PageUptodate(page); 946bcd1d063SChristoph Hellwig sector_t end_block = blkdev_max_block(bdev, size); 9471da177e4SLinus Torvalds 9481da177e4SLinus Torvalds do { 9491da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 95001950a34SEric Biggers bh->b_end_io = NULL; 95101950a34SEric Biggers bh->b_private = NULL; 9521da177e4SLinus Torvalds bh->b_bdev = bdev; 9531da177e4SLinus Torvalds bh->b_blocknr = block; 9541da177e4SLinus Torvalds if (uptodate) 9551da177e4SLinus Torvalds set_buffer_uptodate(bh); 956080399aaSJeff Moyer if (block < end_block) 9571da177e4SLinus Torvalds set_buffer_mapped(bh); 9581da177e4SLinus Torvalds } 9591da177e4SLinus Torvalds block++; 9601da177e4SLinus Torvalds bh = bh->b_this_page; 9611da177e4SLinus Torvalds } while (bh != head); 962676ce6d5SHugh Dickins 963676ce6d5SHugh Dickins /* 964676ce6d5SHugh Dickins * Caller needs to validate requested block against end of device. 965676ce6d5SHugh Dickins */ 966676ce6d5SHugh Dickins return end_block; 9671da177e4SLinus Torvalds } 9681da177e4SLinus Torvalds 9691da177e4SLinus Torvalds /* 9701da177e4SLinus Torvalds * Create the page-cache page that contains the requested block. 9711da177e4SLinus Torvalds * 972676ce6d5SHugh Dickins * This is used purely for blockdev mappings. 9731da177e4SLinus Torvalds */ 974676ce6d5SHugh Dickins static int 9751da177e4SLinus Torvalds grow_dev_page(struct block_device *bdev, sector_t block, 9763b5e6454SGioh Kim pgoff_t index, int size, int sizebits, gfp_t gfp) 9771da177e4SLinus Torvalds { 9781da177e4SLinus Torvalds struct inode *inode = bdev->bd_inode; 9791da177e4SLinus Torvalds struct page *page; 9801da177e4SLinus Torvalds struct buffer_head *bh; 981676ce6d5SHugh Dickins sector_t end_block; 982c4b4c2a7SZhiqiang Liu int ret = 0; 98384235de3SJohannes Weiner gfp_t gfp_mask; 9841da177e4SLinus Torvalds 985c62d2555SMichal Hocko gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp; 9863b5e6454SGioh Kim 98784235de3SJohannes Weiner /* 98884235de3SJohannes Weiner * XXX: __getblk_slow() can not really deal with failure and 98984235de3SJohannes Weiner * will endlessly loop on improvised global reclaim. Prefer 99084235de3SJohannes Weiner * looping in the allocator rather than here, at least that 99184235de3SJohannes Weiner * code knows what it's doing. 99284235de3SJohannes Weiner */ 99384235de3SJohannes Weiner gfp_mask |= __GFP_NOFAIL; 99484235de3SJohannes Weiner 99584235de3SJohannes Weiner page = find_or_create_page(inode->i_mapping, index, gfp_mask); 9961da177e4SLinus Torvalds 997e827f923SEric Sesterhenn BUG_ON(!PageLocked(page)); 9981da177e4SLinus Torvalds 9991da177e4SLinus Torvalds if (page_has_buffers(page)) { 10001da177e4SLinus Torvalds bh = page_buffers(page); 10011da177e4SLinus Torvalds if (bh->b_size == size) { 1002676ce6d5SHugh Dickins end_block = init_page_buffers(page, bdev, 1003f2d5a944SAnton Altaparmakov (sector_t)index << sizebits, 1004f2d5a944SAnton Altaparmakov size); 1005676ce6d5SHugh Dickins goto done; 10061da177e4SLinus Torvalds } 100768189fefSMatthew Wilcox (Oracle) if (!try_to_free_buffers(page_folio(page))) 10081da177e4SLinus Torvalds goto failed; 10091da177e4SLinus Torvalds } 10101da177e4SLinus Torvalds 10111da177e4SLinus Torvalds /* 10121da177e4SLinus Torvalds * Allocate some buffers for this page 10131da177e4SLinus Torvalds */ 101494dc24c0SJens Axboe bh = alloc_page_buffers(page, size, true); 10151da177e4SLinus Torvalds 10161da177e4SLinus Torvalds /* 10171da177e4SLinus Torvalds * Link the page to the buffers and initialise them. Take the 10181da177e4SLinus Torvalds * lock to be atomic wrt __find_get_block(), which does not 10191da177e4SLinus Torvalds * run under the page lock. 10201da177e4SLinus Torvalds */ 10211da177e4SLinus Torvalds spin_lock(&inode->i_mapping->private_lock); 10221da177e4SLinus Torvalds link_dev_buffers(page, bh); 1023f2d5a944SAnton Altaparmakov end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits, 1024f2d5a944SAnton Altaparmakov size); 10251da177e4SLinus Torvalds spin_unlock(&inode->i_mapping->private_lock); 1026676ce6d5SHugh Dickins done: 1027676ce6d5SHugh Dickins ret = (block < end_block) ? 1 : -ENXIO; 10281da177e4SLinus Torvalds failed: 10291da177e4SLinus Torvalds unlock_page(page); 103009cbfeafSKirill A. Shutemov put_page(page); 1031676ce6d5SHugh Dickins return ret; 10321da177e4SLinus Torvalds } 10331da177e4SLinus Torvalds 10341da177e4SLinus Torvalds /* 10351da177e4SLinus Torvalds * Create buffers for the specified block device block's page. If 10361da177e4SLinus Torvalds * that page was dirty, the buffers are set dirty also. 10371da177e4SLinus Torvalds */ 1038858119e1SArjan van de Ven static int 10393b5e6454SGioh Kim grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp) 10401da177e4SLinus Torvalds { 10411da177e4SLinus Torvalds pgoff_t index; 10421da177e4SLinus Torvalds int sizebits; 10431da177e4SLinus Torvalds 104490432e60SMikulas Patocka sizebits = PAGE_SHIFT - __ffs(size); 10451da177e4SLinus Torvalds index = block >> sizebits; 10461da177e4SLinus Torvalds 1047e5657933SAndrew Morton /* 1048e5657933SAndrew Morton * Check for a block which wants to lie outside our maximum possible 1049e5657933SAndrew Morton * pagecache index. (this comparison is done using sector_t types). 1050e5657933SAndrew Morton */ 1051e5657933SAndrew Morton if (unlikely(index != block >> sizebits)) { 1052e5657933SAndrew Morton printk(KERN_ERR "%s: requested out-of-range block %llu for " 1053a1c6f057SDmitry Monakhov "device %pg\n", 10548e24eea7SHarvey Harrison __func__, (unsigned long long)block, 1055a1c6f057SDmitry Monakhov bdev); 1056e5657933SAndrew Morton return -EIO; 1057e5657933SAndrew Morton } 1058676ce6d5SHugh Dickins 10591da177e4SLinus Torvalds /* Create a page with the proper size buffers.. */ 10603b5e6454SGioh Kim return grow_dev_page(bdev, block, index, size, sizebits, gfp); 10611da177e4SLinus Torvalds } 10621da177e4SLinus Torvalds 10630026ba40SEric Biggers static struct buffer_head * 10643b5e6454SGioh Kim __getblk_slow(struct block_device *bdev, sector_t block, 10653b5e6454SGioh Kim unsigned size, gfp_t gfp) 10661da177e4SLinus Torvalds { 10671da177e4SLinus Torvalds /* Size must be multiple of hard sectorsize */ 1068e1defc4fSMartin K. Petersen if (unlikely(size & (bdev_logical_block_size(bdev)-1) || 10691da177e4SLinus Torvalds (size < 512 || size > PAGE_SIZE))) { 10701da177e4SLinus Torvalds printk(KERN_ERR "getblk(): invalid block size %d requested\n", 10711da177e4SLinus Torvalds size); 1072e1defc4fSMartin K. Petersen printk(KERN_ERR "logical block size: %d\n", 1073e1defc4fSMartin K. Petersen bdev_logical_block_size(bdev)); 10741da177e4SLinus Torvalds 10751da177e4SLinus Torvalds dump_stack(); 10761da177e4SLinus Torvalds return NULL; 10771da177e4SLinus Torvalds } 10781da177e4SLinus Torvalds 1079676ce6d5SHugh Dickins for (;;) { 1080676ce6d5SHugh Dickins struct buffer_head *bh; 1081676ce6d5SHugh Dickins int ret; 1082676ce6d5SHugh Dickins 10831da177e4SLinus Torvalds bh = __find_get_block(bdev, block, size); 10841da177e4SLinus Torvalds if (bh) 10851da177e4SLinus Torvalds return bh; 10861da177e4SLinus Torvalds 10873b5e6454SGioh Kim ret = grow_buffers(bdev, block, size, gfp); 1088676ce6d5SHugh Dickins if (ret < 0) 108991f68c89SJeff Moyer return NULL; 1090676ce6d5SHugh Dickins } 10911da177e4SLinus Torvalds } 10921da177e4SLinus Torvalds 10931da177e4SLinus Torvalds /* 10941da177e4SLinus Torvalds * The relationship between dirty buffers and dirty pages: 10951da177e4SLinus Torvalds * 10961da177e4SLinus Torvalds * Whenever a page has any dirty buffers, the page's dirty bit is set, and 1097ec82e1c1SMatthew Wilcox * the page is tagged dirty in the page cache. 10981da177e4SLinus Torvalds * 10991da177e4SLinus Torvalds * At all times, the dirtiness of the buffers represents the dirtiness of 11001da177e4SLinus Torvalds * subsections of the page. If the page has buffers, the page dirty bit is 11011da177e4SLinus Torvalds * merely a hint about the true dirty state. 11021da177e4SLinus Torvalds * 11031da177e4SLinus Torvalds * When a page is set dirty in its entirety, all its buffers are marked dirty 11041da177e4SLinus Torvalds * (if the page has buffers). 11051da177e4SLinus Torvalds * 11061da177e4SLinus Torvalds * When a buffer is marked dirty, its page is dirtied, but the page's other 11071da177e4SLinus Torvalds * buffers are not. 11081da177e4SLinus Torvalds * 11091da177e4SLinus Torvalds * Also. When blockdev buffers are explicitly read with bread(), they 11101da177e4SLinus Torvalds * individually become uptodate. But their backing page remains not 11111da177e4SLinus Torvalds * uptodate - even if all of its buffers are uptodate. A subsequent 11122c69e205SMatthew Wilcox (Oracle) * block_read_full_folio() against that folio will discover all the uptodate 11132c69e205SMatthew Wilcox (Oracle) * buffers, will set the folio uptodate and will perform no I/O. 11141da177e4SLinus Torvalds */ 11151da177e4SLinus Torvalds 11161da177e4SLinus Torvalds /** 11171da177e4SLinus Torvalds * mark_buffer_dirty - mark a buffer_head as needing writeout 111867be2dd1SMartin Waitz * @bh: the buffer_head to mark dirty 11191da177e4SLinus Torvalds * 1120ec82e1c1SMatthew Wilcox * mark_buffer_dirty() will set the dirty bit against the buffer, then set 1121ec82e1c1SMatthew Wilcox * its backing page dirty, then tag the page as dirty in the page cache 1122ec82e1c1SMatthew Wilcox * and then attach the address_space's inode to its superblock's dirty 11231da177e4SLinus Torvalds * inode list. 11241da177e4SLinus Torvalds * 1125abc8a8a2SMatthew Wilcox (Oracle) * mark_buffer_dirty() is atomic. It takes bh->b_folio->mapping->private_lock, 1126b93b0163SMatthew Wilcox * i_pages lock and mapping->host->i_lock. 11271da177e4SLinus Torvalds */ 1128fc9b52cdSHarvey Harrison void mark_buffer_dirty(struct buffer_head *bh) 11291da177e4SLinus Torvalds { 1130787d2214SNick Piggin WARN_ON_ONCE(!buffer_uptodate(bh)); 11311be62dc1SLinus Torvalds 11325305cb83STejun Heo trace_block_dirty_buffer(bh); 11335305cb83STejun Heo 11341be62dc1SLinus Torvalds /* 11351be62dc1SLinus Torvalds * Very *carefully* optimize the it-is-already-dirty case. 11361be62dc1SLinus Torvalds * 11371be62dc1SLinus Torvalds * Don't let the final "is it dirty" escape to before we 11381be62dc1SLinus Torvalds * perhaps modified the buffer. 11391be62dc1SLinus Torvalds */ 11401be62dc1SLinus Torvalds if (buffer_dirty(bh)) { 11411be62dc1SLinus Torvalds smp_mb(); 11421be62dc1SLinus Torvalds if (buffer_dirty(bh)) 11431be62dc1SLinus Torvalds return; 11441be62dc1SLinus Torvalds } 11451be62dc1SLinus Torvalds 1146a8e7d49aSLinus Torvalds if (!test_set_buffer_dirty(bh)) { 1147cf1d3417SMatthew Wilcox (Oracle) struct folio *folio = bh->b_folio; 1148c4843a75SGreg Thelen struct address_space *mapping = NULL; 1149c4843a75SGreg Thelen 1150cf1d3417SMatthew Wilcox (Oracle) folio_memcg_lock(folio); 1151cf1d3417SMatthew Wilcox (Oracle) if (!folio_test_set_dirty(folio)) { 1152cf1d3417SMatthew Wilcox (Oracle) mapping = folio->mapping; 11538e9d78edSLinus Torvalds if (mapping) 1154cf1d3417SMatthew Wilcox (Oracle) __folio_mark_dirty(folio, mapping, 0); 11558e9d78edSLinus Torvalds } 1156cf1d3417SMatthew Wilcox (Oracle) folio_memcg_unlock(folio); 1157c4843a75SGreg Thelen if (mapping) 1158c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 1159a8e7d49aSLinus Torvalds } 11601da177e4SLinus Torvalds } 11611fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(mark_buffer_dirty); 11621da177e4SLinus Torvalds 116387354e5dSJeff Layton void mark_buffer_write_io_error(struct buffer_head *bh) 116487354e5dSJeff Layton { 1165485e9605SJeff Layton struct super_block *sb; 1166485e9605SJeff Layton 116787354e5dSJeff Layton set_buffer_write_io_error(bh); 116887354e5dSJeff Layton /* FIXME: do we need to set this in both places? */ 1169abc8a8a2SMatthew Wilcox (Oracle) if (bh->b_folio && bh->b_folio->mapping) 1170abc8a8a2SMatthew Wilcox (Oracle) mapping_set_error(bh->b_folio->mapping, -EIO); 117187354e5dSJeff Layton if (bh->b_assoc_map) 117287354e5dSJeff Layton mapping_set_error(bh->b_assoc_map, -EIO); 1173485e9605SJeff Layton rcu_read_lock(); 1174485e9605SJeff Layton sb = READ_ONCE(bh->b_bdev->bd_super); 1175485e9605SJeff Layton if (sb) 1176485e9605SJeff Layton errseq_set(&sb->s_wb_err, -EIO); 1177485e9605SJeff Layton rcu_read_unlock(); 117887354e5dSJeff Layton } 117987354e5dSJeff Layton EXPORT_SYMBOL(mark_buffer_write_io_error); 118087354e5dSJeff Layton 11811da177e4SLinus Torvalds /* 11821da177e4SLinus Torvalds * Decrement a buffer_head's reference count. If all buffers against a page 11831da177e4SLinus Torvalds * have zero reference count, are clean and unlocked, and if the page is clean 11841da177e4SLinus Torvalds * and unlocked then try_to_free_buffers() may strip the buffers from the page 11851da177e4SLinus Torvalds * in preparation for freeing it (sometimes, rarely, buffers are removed from 11861da177e4SLinus Torvalds * a page but it ends up not being freed, and buffers may later be reattached). 11871da177e4SLinus Torvalds */ 11881da177e4SLinus Torvalds void __brelse(struct buffer_head * buf) 11891da177e4SLinus Torvalds { 11901da177e4SLinus Torvalds if (atomic_read(&buf->b_count)) { 11911da177e4SLinus Torvalds put_bh(buf); 11921da177e4SLinus Torvalds return; 11931da177e4SLinus Torvalds } 11945c752ad9SArjan van de Ven WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); 11951da177e4SLinus Torvalds } 11961fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__brelse); 11971da177e4SLinus Torvalds 11981da177e4SLinus Torvalds /* 11991da177e4SLinus Torvalds * bforget() is like brelse(), except it discards any 12001da177e4SLinus Torvalds * potentially dirty data. 12011da177e4SLinus Torvalds */ 12021da177e4SLinus Torvalds void __bforget(struct buffer_head *bh) 12031da177e4SLinus Torvalds { 12041da177e4SLinus Torvalds clear_buffer_dirty(bh); 1205535ee2fbSJan Kara if (bh->b_assoc_map) { 1206abc8a8a2SMatthew Wilcox (Oracle) struct address_space *buffer_mapping = bh->b_folio->mapping; 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 12091da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 121058ff407bSJan Kara bh->b_assoc_map = NULL; 12111da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 12121da177e4SLinus Torvalds } 12131da177e4SLinus Torvalds __brelse(bh); 12141da177e4SLinus Torvalds } 12151fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__bforget); 12161da177e4SLinus Torvalds 12171da177e4SLinus Torvalds static struct buffer_head *__bread_slow(struct buffer_head *bh) 12181da177e4SLinus Torvalds { 12191da177e4SLinus Torvalds lock_buffer(bh); 12201da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 12211da177e4SLinus Torvalds unlock_buffer(bh); 12221da177e4SLinus Torvalds return bh; 12231da177e4SLinus Torvalds } else { 12241da177e4SLinus Torvalds get_bh(bh); 12251da177e4SLinus Torvalds bh->b_end_io = end_buffer_read_sync; 12261420c4a5SBart Van Assche submit_bh(REQ_OP_READ, bh); 12271da177e4SLinus Torvalds wait_on_buffer(bh); 12281da177e4SLinus Torvalds if (buffer_uptodate(bh)) 12291da177e4SLinus Torvalds return bh; 12301da177e4SLinus Torvalds } 12311da177e4SLinus Torvalds brelse(bh); 12321da177e4SLinus Torvalds return NULL; 12331da177e4SLinus Torvalds } 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds /* 12361da177e4SLinus Torvalds * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). 12371da177e4SLinus Torvalds * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their 12381da177e4SLinus Torvalds * refcount elevated by one when they're in an LRU. A buffer can only appear 12391da177e4SLinus Torvalds * once in a particular CPU's LRU. A single buffer can be present in multiple 12401da177e4SLinus Torvalds * CPU's LRUs at the same time. 12411da177e4SLinus Torvalds * 12421da177e4SLinus Torvalds * This is a transparent caching front-end to sb_bread(), sb_getblk() and 12431da177e4SLinus Torvalds * sb_find_get_block(). 12441da177e4SLinus Torvalds * 12451da177e4SLinus Torvalds * The LRUs themselves only need locking against invalidate_bh_lrus. We use 12461da177e4SLinus Torvalds * a local interrupt disable for that. 12471da177e4SLinus Torvalds */ 12481da177e4SLinus Torvalds 124986cf78d7SSebastien Buisson #define BH_LRU_SIZE 16 12501da177e4SLinus Torvalds 12511da177e4SLinus Torvalds struct bh_lru { 12521da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12531da177e4SLinus Torvalds }; 12541da177e4SLinus Torvalds 12551da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds #ifdef CONFIG_SMP 12581da177e4SLinus Torvalds #define bh_lru_lock() local_irq_disable() 12591da177e4SLinus Torvalds #define bh_lru_unlock() local_irq_enable() 12601da177e4SLinus Torvalds #else 12611da177e4SLinus Torvalds #define bh_lru_lock() preempt_disable() 12621da177e4SLinus Torvalds #define bh_lru_unlock() preempt_enable() 12631da177e4SLinus Torvalds #endif 12641da177e4SLinus Torvalds 12651da177e4SLinus Torvalds static inline void check_irqs_on(void) 12661da177e4SLinus Torvalds { 12671da177e4SLinus Torvalds #ifdef irqs_disabled 12681da177e4SLinus Torvalds BUG_ON(irqs_disabled()); 12691da177e4SLinus Torvalds #endif 12701da177e4SLinus Torvalds } 12711da177e4SLinus Torvalds 12721da177e4SLinus Torvalds /* 1273241f01fbSEric Biggers * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is 1274241f01fbSEric Biggers * inserted at the front, and the buffer_head at the back if any is evicted. 1275241f01fbSEric Biggers * Or, if already in the LRU it is moved to the front. 12761da177e4SLinus Torvalds */ 12771da177e4SLinus Torvalds static void bh_lru_install(struct buffer_head *bh) 12781da177e4SLinus Torvalds { 1279241f01fbSEric Biggers struct buffer_head *evictee = bh; 1280241f01fbSEric Biggers struct bh_lru *b; 1281241f01fbSEric Biggers int i; 12821da177e4SLinus Torvalds 12831da177e4SLinus Torvalds check_irqs_on(); 1284c0226eb8SMinchan Kim bh_lru_lock(); 1285c0226eb8SMinchan Kim 12868cc621d2SMinchan Kim /* 12878cc621d2SMinchan Kim * the refcount of buffer_head in bh_lru prevents dropping the 12888cc621d2SMinchan Kim * attached page(i.e., try_to_free_buffers) so it could cause 12898cc621d2SMinchan Kim * failing page migration. 12908cc621d2SMinchan Kim * Skip putting upcoming bh into bh_lru until migration is done. 12918cc621d2SMinchan Kim */ 1292c0226eb8SMinchan Kim if (lru_cache_disabled()) { 1293c0226eb8SMinchan Kim bh_lru_unlock(); 12948cc621d2SMinchan Kim return; 1295c0226eb8SMinchan Kim } 1296241f01fbSEric Biggers 1297241f01fbSEric Biggers b = this_cpu_ptr(&bh_lrus); 1298241f01fbSEric Biggers for (i = 0; i < BH_LRU_SIZE; i++) { 1299241f01fbSEric Biggers swap(evictee, b->bhs[i]); 1300241f01fbSEric Biggers if (evictee == bh) { 1301241f01fbSEric Biggers bh_lru_unlock(); 1302241f01fbSEric Biggers return; 1303241f01fbSEric Biggers } 1304241f01fbSEric Biggers } 13051da177e4SLinus Torvalds 13061da177e4SLinus Torvalds get_bh(bh); 13071da177e4SLinus Torvalds bh_lru_unlock(); 1308241f01fbSEric Biggers brelse(evictee); 13091da177e4SLinus Torvalds } 13101da177e4SLinus Torvalds 13111da177e4SLinus Torvalds /* 13121da177e4SLinus Torvalds * Look up the bh in this cpu's LRU. If it's there, move it to the head. 13131da177e4SLinus Torvalds */ 1314858119e1SArjan van de Ven static struct buffer_head * 13153991d3bdSTomasz Kvarsin lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) 13161da177e4SLinus Torvalds { 13171da177e4SLinus Torvalds struct buffer_head *ret = NULL; 13183991d3bdSTomasz Kvarsin unsigned int i; 13191da177e4SLinus Torvalds 13201da177e4SLinus Torvalds check_irqs_on(); 13211da177e4SLinus Torvalds bh_lru_lock(); 13221da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 1323c7b92516SChristoph Lameter struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); 13241da177e4SLinus Torvalds 13259470dd5dSZach Brown if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && 13269470dd5dSZach Brown bh->b_size == size) { 13271da177e4SLinus Torvalds if (i) { 13281da177e4SLinus Torvalds while (i) { 1329c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[i], 1330c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[i - 1])); 13311da177e4SLinus Torvalds i--; 13321da177e4SLinus Torvalds } 1333c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[0], bh); 13341da177e4SLinus Torvalds } 13351da177e4SLinus Torvalds get_bh(bh); 13361da177e4SLinus Torvalds ret = bh; 13371da177e4SLinus Torvalds break; 13381da177e4SLinus Torvalds } 13391da177e4SLinus Torvalds } 13401da177e4SLinus Torvalds bh_lru_unlock(); 13411da177e4SLinus Torvalds return ret; 13421da177e4SLinus Torvalds } 13431da177e4SLinus Torvalds 13441da177e4SLinus Torvalds /* 13451da177e4SLinus Torvalds * Perform a pagecache lookup for the matching buffer. If it's there, refresh 13461da177e4SLinus Torvalds * it in the LRU and mark it as accessed. If it is not present then return 13471da177e4SLinus Torvalds * NULL 13481da177e4SLinus Torvalds */ 13491da177e4SLinus Torvalds struct buffer_head * 13503991d3bdSTomasz Kvarsin __find_get_block(struct block_device *bdev, sector_t block, unsigned size) 13511da177e4SLinus Torvalds { 13521da177e4SLinus Torvalds struct buffer_head *bh = lookup_bh_lru(bdev, block, size); 13531da177e4SLinus Torvalds 13541da177e4SLinus Torvalds if (bh == NULL) { 13552457aec6SMel Gorman /* __find_get_block_slow will mark the page accessed */ 1356385fd4c5SCoywolf Qi Hunt bh = __find_get_block_slow(bdev, block); 13571da177e4SLinus Torvalds if (bh) 13581da177e4SLinus Torvalds bh_lru_install(bh); 13592457aec6SMel Gorman } else 13601da177e4SLinus Torvalds touch_buffer(bh); 13612457aec6SMel Gorman 13621da177e4SLinus Torvalds return bh; 13631da177e4SLinus Torvalds } 13641da177e4SLinus Torvalds EXPORT_SYMBOL(__find_get_block); 13651da177e4SLinus Torvalds 13661da177e4SLinus Torvalds /* 13673b5e6454SGioh Kim * __getblk_gfp() will locate (and, if necessary, create) the buffer_head 13681da177e4SLinus Torvalds * which corresponds to the passed block_device, block and size. The 13691da177e4SLinus Torvalds * returned buffer has its reference count incremented. 13701da177e4SLinus Torvalds * 13713b5e6454SGioh Kim * __getblk_gfp() will lock up the machine if grow_dev_page's 13723b5e6454SGioh Kim * try_to_free_buffers() attempt is failing. FIXME, perhaps? 13731da177e4SLinus Torvalds */ 13741da177e4SLinus Torvalds struct buffer_head * 13753b5e6454SGioh Kim __getblk_gfp(struct block_device *bdev, sector_t block, 13763b5e6454SGioh Kim unsigned size, gfp_t gfp) 13771da177e4SLinus Torvalds { 13781da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, block, size); 13791da177e4SLinus Torvalds 13801da177e4SLinus Torvalds might_sleep(); 13811da177e4SLinus Torvalds if (bh == NULL) 13823b5e6454SGioh Kim bh = __getblk_slow(bdev, block, size, gfp); 13831da177e4SLinus Torvalds return bh; 13841da177e4SLinus Torvalds } 13853b5e6454SGioh Kim EXPORT_SYMBOL(__getblk_gfp); 13861da177e4SLinus Torvalds 13871da177e4SLinus Torvalds /* 13881da177e4SLinus Torvalds * Do async read-ahead on a buffer.. 13891da177e4SLinus Torvalds */ 13903991d3bdSTomasz Kvarsin void __breadahead(struct block_device *bdev, sector_t block, unsigned size) 13911da177e4SLinus Torvalds { 13921da177e4SLinus Torvalds struct buffer_head *bh = __getblk(bdev, block, size); 1393a3e713b5SAndrew Morton if (likely(bh)) { 1394e7ea1129SZhang Yi bh_readahead(bh, REQ_RAHEAD); 13951da177e4SLinus Torvalds brelse(bh); 13961da177e4SLinus Torvalds } 1397a3e713b5SAndrew Morton } 13981da177e4SLinus Torvalds EXPORT_SYMBOL(__breadahead); 13991da177e4SLinus Torvalds 14001da177e4SLinus Torvalds /** 14013b5e6454SGioh Kim * __bread_gfp() - reads a specified block and returns the bh 140267be2dd1SMartin Waitz * @bdev: the block_device to read from 14031da177e4SLinus Torvalds * @block: number of block 14041da177e4SLinus Torvalds * @size: size (in bytes) to read 14053b5e6454SGioh Kim * @gfp: page allocation flag 14061da177e4SLinus Torvalds * 14071da177e4SLinus Torvalds * Reads a specified block, and returns buffer head that contains it. 14083b5e6454SGioh Kim * The page cache can be allocated from non-movable area 14093b5e6454SGioh Kim * not to prevent page migration if you set gfp to zero. 14101da177e4SLinus Torvalds * It returns NULL if the block was unreadable. 14111da177e4SLinus Torvalds */ 14121da177e4SLinus Torvalds struct buffer_head * 14133b5e6454SGioh Kim __bread_gfp(struct block_device *bdev, sector_t block, 14143b5e6454SGioh Kim unsigned size, gfp_t gfp) 14151da177e4SLinus Torvalds { 14163b5e6454SGioh Kim struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); 14171da177e4SLinus Torvalds 1418a3e713b5SAndrew Morton if (likely(bh) && !buffer_uptodate(bh)) 14191da177e4SLinus Torvalds bh = __bread_slow(bh); 14201da177e4SLinus Torvalds return bh; 14211da177e4SLinus Torvalds } 14223b5e6454SGioh Kim EXPORT_SYMBOL(__bread_gfp); 14231da177e4SLinus Torvalds 14248cc621d2SMinchan Kim static void __invalidate_bh_lrus(struct bh_lru *b) 14258cc621d2SMinchan Kim { 14268cc621d2SMinchan Kim int i; 14278cc621d2SMinchan Kim 14288cc621d2SMinchan Kim for (i = 0; i < BH_LRU_SIZE; i++) { 14298cc621d2SMinchan Kim brelse(b->bhs[i]); 14308cc621d2SMinchan Kim b->bhs[i] = NULL; 14318cc621d2SMinchan Kim } 14328cc621d2SMinchan Kim } 14331da177e4SLinus Torvalds /* 14341da177e4SLinus Torvalds * invalidate_bh_lrus() is called rarely - but not only at unmount. 14351da177e4SLinus Torvalds * This doesn't race because it runs in each cpu either in irq 14361da177e4SLinus Torvalds * or with preempt disabled. 14371da177e4SLinus Torvalds */ 14381da177e4SLinus Torvalds static void invalidate_bh_lru(void *arg) 14391da177e4SLinus Torvalds { 14401da177e4SLinus Torvalds struct bh_lru *b = &get_cpu_var(bh_lrus); 14411da177e4SLinus Torvalds 14428cc621d2SMinchan Kim __invalidate_bh_lrus(b); 14431da177e4SLinus Torvalds put_cpu_var(bh_lrus); 14441da177e4SLinus Torvalds } 14451da177e4SLinus Torvalds 14468cc621d2SMinchan Kim bool has_bh_in_lru(int cpu, void *dummy) 144742be35d0SGilad Ben-Yossef { 144842be35d0SGilad Ben-Yossef struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); 144942be35d0SGilad Ben-Yossef int i; 145042be35d0SGilad Ben-Yossef 145142be35d0SGilad Ben-Yossef for (i = 0; i < BH_LRU_SIZE; i++) { 145242be35d0SGilad Ben-Yossef if (b->bhs[i]) 14531d706679SSaurav Girepunje return true; 145442be35d0SGilad Ben-Yossef } 145542be35d0SGilad Ben-Yossef 14561d706679SSaurav Girepunje return false; 145742be35d0SGilad Ben-Yossef } 145842be35d0SGilad Ben-Yossef 1459f9a14399SPeter Zijlstra void invalidate_bh_lrus(void) 14601da177e4SLinus Torvalds { 1461cb923159SSebastian Andrzej Siewior on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1); 14621da177e4SLinus Torvalds } 14639db5579bSNick Piggin EXPORT_SYMBOL_GPL(invalidate_bh_lrus); 14641da177e4SLinus Torvalds 1465243418e3SMinchan Kim /* 1466243418e3SMinchan Kim * It's called from workqueue context so we need a bh_lru_lock to close 1467243418e3SMinchan Kim * the race with preemption/irq. 1468243418e3SMinchan Kim */ 1469243418e3SMinchan Kim void invalidate_bh_lrus_cpu(void) 14708cc621d2SMinchan Kim { 14718cc621d2SMinchan Kim struct bh_lru *b; 14728cc621d2SMinchan Kim 14738cc621d2SMinchan Kim bh_lru_lock(); 1474243418e3SMinchan Kim b = this_cpu_ptr(&bh_lrus); 14758cc621d2SMinchan Kim __invalidate_bh_lrus(b); 14768cc621d2SMinchan Kim bh_lru_unlock(); 14778cc621d2SMinchan Kim } 14788cc621d2SMinchan Kim 14791da177e4SLinus Torvalds void set_bh_page(struct buffer_head *bh, 14801da177e4SLinus Torvalds struct page *page, unsigned long offset) 14811da177e4SLinus Torvalds { 14821da177e4SLinus Torvalds bh->b_page = page; 1483e827f923SEric Sesterhenn BUG_ON(offset >= PAGE_SIZE); 14841da177e4SLinus Torvalds if (PageHighMem(page)) 14851da177e4SLinus Torvalds /* 14861da177e4SLinus Torvalds * This catches illegal uses and preserves the offset: 14871da177e4SLinus Torvalds */ 14881da177e4SLinus Torvalds bh->b_data = (char *)(0 + offset); 14891da177e4SLinus Torvalds else 14901da177e4SLinus Torvalds bh->b_data = page_address(page) + offset; 14911da177e4SLinus Torvalds } 14921da177e4SLinus Torvalds EXPORT_SYMBOL(set_bh_page); 14931da177e4SLinus Torvalds 1494465e5e6aSPankaj Raghav void folio_set_bh(struct buffer_head *bh, struct folio *folio, 1495465e5e6aSPankaj Raghav unsigned long offset) 1496465e5e6aSPankaj Raghav { 1497465e5e6aSPankaj Raghav bh->b_folio = folio; 1498465e5e6aSPankaj Raghav BUG_ON(offset >= folio_size(folio)); 1499465e5e6aSPankaj Raghav if (folio_test_highmem(folio)) 1500465e5e6aSPankaj Raghav /* 1501465e5e6aSPankaj Raghav * This catches illegal uses and preserves the offset: 1502465e5e6aSPankaj Raghav */ 1503465e5e6aSPankaj Raghav bh->b_data = (char *)(0 + offset); 1504465e5e6aSPankaj Raghav else 1505465e5e6aSPankaj Raghav bh->b_data = folio_address(folio) + offset; 1506465e5e6aSPankaj Raghav } 1507465e5e6aSPankaj Raghav EXPORT_SYMBOL(folio_set_bh); 1508465e5e6aSPankaj Raghav 15091da177e4SLinus Torvalds /* 15101da177e4SLinus Torvalds * Called when truncating a buffer on a page completely. 15111da177e4SLinus Torvalds */ 1512e7470ee8SMel Gorman 1513e7470ee8SMel Gorman /* Bits that are cleared during an invalidate */ 1514e7470ee8SMel Gorman #define BUFFER_FLAGS_DISCARD \ 1515e7470ee8SMel Gorman (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ 1516e7470ee8SMel Gorman 1 << BH_Delay | 1 << BH_Unwritten) 1517e7470ee8SMel Gorman 1518858119e1SArjan van de Ven static void discard_buffer(struct buffer_head * bh) 15191da177e4SLinus Torvalds { 1520b0192296SUros Bizjak unsigned long b_state; 1521e7470ee8SMel Gorman 15221da177e4SLinus Torvalds lock_buffer(bh); 15231da177e4SLinus Torvalds clear_buffer_dirty(bh); 15241da177e4SLinus Torvalds bh->b_bdev = NULL; 1525b0192296SUros Bizjak b_state = READ_ONCE(bh->b_state); 1526b0192296SUros Bizjak do { 1527b0192296SUros Bizjak } while (!try_cmpxchg(&bh->b_state, &b_state, 1528b0192296SUros Bizjak b_state & ~BUFFER_FLAGS_DISCARD)); 15291da177e4SLinus Torvalds unlock_buffer(bh); 15301da177e4SLinus Torvalds } 15311da177e4SLinus Torvalds 15321da177e4SLinus Torvalds /** 15337ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio - Invalidate part or all of a buffer-backed folio. 15347ba13abbSMatthew Wilcox (Oracle) * @folio: The folio which is affected. 1535d47992f8SLukas Czerner * @offset: start of the range to invalidate 1536d47992f8SLukas Czerner * @length: length of the range to invalidate 15371da177e4SLinus Torvalds * 15387ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio() is called when all or part of the folio has been 15391da177e4SLinus Torvalds * invalidated by a truncate operation. 15401da177e4SLinus Torvalds * 15417ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio() does not have to release all buffers, but it must 15421da177e4SLinus Torvalds * ensure that no dirty buffer is left outside @offset and that no I/O 15431da177e4SLinus Torvalds * is underway against any of the blocks which are outside the truncation 15441da177e4SLinus Torvalds * point. Because the caller is about to free (and possibly reuse) those 15451da177e4SLinus Torvalds * blocks on-disk. 15461da177e4SLinus Torvalds */ 15477ba13abbSMatthew Wilcox (Oracle) void block_invalidate_folio(struct folio *folio, size_t offset, size_t length) 15481da177e4SLinus Torvalds { 15491da177e4SLinus Torvalds struct buffer_head *head, *bh, *next; 15507ba13abbSMatthew Wilcox (Oracle) size_t curr_off = 0; 15517ba13abbSMatthew Wilcox (Oracle) size_t stop = length + offset; 15521da177e4SLinus Torvalds 15537ba13abbSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 15541da177e4SLinus Torvalds 1555d47992f8SLukas Czerner /* 1556d47992f8SLukas Czerner * Check for overflow 1557d47992f8SLukas Czerner */ 15587ba13abbSMatthew Wilcox (Oracle) BUG_ON(stop > folio_size(folio) || stop < length); 1559d47992f8SLukas Czerner 15607ba13abbSMatthew Wilcox (Oracle) head = folio_buffers(folio); 15617ba13abbSMatthew Wilcox (Oracle) if (!head) 15627ba13abbSMatthew Wilcox (Oracle) return; 15637ba13abbSMatthew Wilcox (Oracle) 15641da177e4SLinus Torvalds bh = head; 15651da177e4SLinus Torvalds do { 15667ba13abbSMatthew Wilcox (Oracle) size_t next_off = curr_off + bh->b_size; 15671da177e4SLinus Torvalds next = bh->b_this_page; 15681da177e4SLinus Torvalds 15691da177e4SLinus Torvalds /* 1570d47992f8SLukas Czerner * Are we still fully in range ? 1571d47992f8SLukas Czerner */ 1572d47992f8SLukas Czerner if (next_off > stop) 1573d47992f8SLukas Czerner goto out; 1574d47992f8SLukas Czerner 1575d47992f8SLukas Czerner /* 15761da177e4SLinus Torvalds * is this block fully invalidated? 15771da177e4SLinus Torvalds */ 15781da177e4SLinus Torvalds if (offset <= curr_off) 15791da177e4SLinus Torvalds discard_buffer(bh); 15801da177e4SLinus Torvalds curr_off = next_off; 15811da177e4SLinus Torvalds bh = next; 15821da177e4SLinus Torvalds } while (bh != head); 15831da177e4SLinus Torvalds 15841da177e4SLinus Torvalds /* 15857ba13abbSMatthew Wilcox (Oracle) * We release buffers only if the entire folio is being invalidated. 15861da177e4SLinus Torvalds * The get_block cached value has been unconditionally invalidated, 15871da177e4SLinus Torvalds * so real IO is not possible anymore. 15881da177e4SLinus Torvalds */ 15897ba13abbSMatthew Wilcox (Oracle) if (length == folio_size(folio)) 15907ba13abbSMatthew Wilcox (Oracle) filemap_release_folio(folio, 0); 15911da177e4SLinus Torvalds out: 15922ff28e22SNeilBrown return; 15931da177e4SLinus Torvalds } 15947ba13abbSMatthew Wilcox (Oracle) EXPORT_SYMBOL(block_invalidate_folio); 15951da177e4SLinus Torvalds 15961da177e4SLinus Torvalds /* 15971da177e4SLinus Torvalds * We attach and possibly dirty the buffers atomically wrt 1598e621900aSMatthew Wilcox (Oracle) * block_dirty_folio() via private_lock. try_to_free_buffers 15998e2e1756SPankaj Raghav * is already excluded via the folio lock. 16001da177e4SLinus Torvalds */ 16018e2e1756SPankaj Raghav void folio_create_empty_buffers(struct folio *folio, unsigned long blocksize, 16028e2e1756SPankaj Raghav unsigned long b_state) 16031da177e4SLinus Torvalds { 16041da177e4SLinus Torvalds struct buffer_head *bh, *head, *tail; 16051da177e4SLinus Torvalds 16068e2e1756SPankaj Raghav head = folio_alloc_buffers(folio, blocksize, true); 16071da177e4SLinus Torvalds bh = head; 16081da177e4SLinus Torvalds do { 16091da177e4SLinus Torvalds bh->b_state |= b_state; 16101da177e4SLinus Torvalds tail = bh; 16111da177e4SLinus Torvalds bh = bh->b_this_page; 16121da177e4SLinus Torvalds } while (bh); 16131da177e4SLinus Torvalds tail->b_this_page = head; 16141da177e4SLinus Torvalds 16158e2e1756SPankaj Raghav spin_lock(&folio->mapping->private_lock); 16168e2e1756SPankaj Raghav if (folio_test_uptodate(folio) || folio_test_dirty(folio)) { 16171da177e4SLinus Torvalds bh = head; 16181da177e4SLinus Torvalds do { 16198e2e1756SPankaj Raghav if (folio_test_dirty(folio)) 16201da177e4SLinus Torvalds set_buffer_dirty(bh); 16218e2e1756SPankaj Raghav if (folio_test_uptodate(folio)) 16221da177e4SLinus Torvalds set_buffer_uptodate(bh); 16231da177e4SLinus Torvalds bh = bh->b_this_page; 16241da177e4SLinus Torvalds } while (bh != head); 16251da177e4SLinus Torvalds } 16268e2e1756SPankaj Raghav folio_attach_private(folio, head); 16278e2e1756SPankaj Raghav spin_unlock(&folio->mapping->private_lock); 16288e2e1756SPankaj Raghav } 16298e2e1756SPankaj Raghav EXPORT_SYMBOL(folio_create_empty_buffers); 16308e2e1756SPankaj Raghav 16318e2e1756SPankaj Raghav void create_empty_buffers(struct page *page, 16328e2e1756SPankaj Raghav unsigned long blocksize, unsigned long b_state) 16338e2e1756SPankaj Raghav { 16348e2e1756SPankaj Raghav folio_create_empty_buffers(page_folio(page), blocksize, b_state); 16351da177e4SLinus Torvalds } 16361da177e4SLinus Torvalds EXPORT_SYMBOL(create_empty_buffers); 16371da177e4SLinus Torvalds 163829f3ad7dSJan Kara /** 163929f3ad7dSJan Kara * clean_bdev_aliases: clean a range of buffers in block device 164029f3ad7dSJan Kara * @bdev: Block device to clean buffers in 164129f3ad7dSJan Kara * @block: Start of a range of blocks to clean 164229f3ad7dSJan Kara * @len: Number of blocks to clean 16431da177e4SLinus Torvalds * 164429f3ad7dSJan Kara * We are taking a range of blocks for data and we don't want writeback of any 164529f3ad7dSJan Kara * buffer-cache aliases starting from return from this function and until the 164629f3ad7dSJan Kara * moment when something will explicitly mark the buffer dirty (hopefully that 164729f3ad7dSJan Kara * will not happen until we will free that block ;-) We don't even need to mark 164829f3ad7dSJan Kara * it not-uptodate - nobody can expect anything from a newly allocated buffer 164929f3ad7dSJan Kara * anyway. We used to use unmap_buffer() for such invalidation, but that was 165029f3ad7dSJan Kara * wrong. We definitely don't want to mark the alias unmapped, for example - it 165129f3ad7dSJan Kara * would confuse anyone who might pick it with bread() afterwards... 165229f3ad7dSJan Kara * 165329f3ad7dSJan Kara * Also.. Note that bforget() doesn't lock the buffer. So there can be 165429f3ad7dSJan Kara * writeout I/O going on against recently-freed buffers. We don't wait on that 165529f3ad7dSJan Kara * I/O in bforget() - it's more efficient to wait on the I/O only if we really 165629f3ad7dSJan Kara * need to. That happens here. 16571da177e4SLinus Torvalds */ 165829f3ad7dSJan Kara void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) 16591da177e4SLinus Torvalds { 166029f3ad7dSJan Kara struct inode *bd_inode = bdev->bd_inode; 166129f3ad7dSJan Kara struct address_space *bd_mapping = bd_inode->i_mapping; 16629e0b6f31SMatthew Wilcox (Oracle) struct folio_batch fbatch; 166329f3ad7dSJan Kara pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); 166429f3ad7dSJan Kara pgoff_t end; 1665c10f778dSJan Kara int i, count; 166629f3ad7dSJan Kara struct buffer_head *bh; 166729f3ad7dSJan Kara struct buffer_head *head; 16681da177e4SLinus Torvalds 166929f3ad7dSJan Kara end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits); 16709e0b6f31SMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 16719e0b6f31SMatthew Wilcox (Oracle) while (filemap_get_folios(bd_mapping, &index, end, &fbatch)) { 16729e0b6f31SMatthew Wilcox (Oracle) count = folio_batch_count(&fbatch); 1673c10f778dSJan Kara for (i = 0; i < count; i++) { 16749e0b6f31SMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 16751da177e4SLinus Torvalds 16769e0b6f31SMatthew Wilcox (Oracle) if (!folio_buffers(folio)) 167729f3ad7dSJan Kara continue; 167829f3ad7dSJan Kara /* 16799e0b6f31SMatthew Wilcox (Oracle) * We use folio lock instead of bd_mapping->private_lock 168029f3ad7dSJan Kara * to pin buffers here since we can afford to sleep and 168129f3ad7dSJan Kara * it scales better than a global spinlock lock. 168229f3ad7dSJan Kara */ 16839e0b6f31SMatthew Wilcox (Oracle) folio_lock(folio); 16849e0b6f31SMatthew Wilcox (Oracle) /* Recheck when the folio is locked which pins bhs */ 16859e0b6f31SMatthew Wilcox (Oracle) head = folio_buffers(folio); 16869e0b6f31SMatthew Wilcox (Oracle) if (!head) 168729f3ad7dSJan Kara goto unlock_page; 168829f3ad7dSJan Kara bh = head; 168929f3ad7dSJan Kara do { 16906c006a9dSChandan Rajendra if (!buffer_mapped(bh) || (bh->b_blocknr < block)) 169129f3ad7dSJan Kara goto next; 169229f3ad7dSJan Kara if (bh->b_blocknr >= block + len) 169329f3ad7dSJan Kara break; 169429f3ad7dSJan Kara clear_buffer_dirty(bh); 169529f3ad7dSJan Kara wait_on_buffer(bh); 169629f3ad7dSJan Kara clear_buffer_req(bh); 169729f3ad7dSJan Kara next: 169829f3ad7dSJan Kara bh = bh->b_this_page; 169929f3ad7dSJan Kara } while (bh != head); 170029f3ad7dSJan Kara unlock_page: 17019e0b6f31SMatthew Wilcox (Oracle) folio_unlock(folio); 170229f3ad7dSJan Kara } 17039e0b6f31SMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 170429f3ad7dSJan Kara cond_resched(); 1705c10f778dSJan Kara /* End of range already reached? */ 1706c10f778dSJan Kara if (index > end || !index) 1707c10f778dSJan Kara break; 17081da177e4SLinus Torvalds } 17091da177e4SLinus Torvalds } 171029f3ad7dSJan Kara EXPORT_SYMBOL(clean_bdev_aliases); 17111da177e4SLinus Torvalds 17121da177e4SLinus Torvalds /* 171345bce8f3SLinus Torvalds * Size is a power-of-two in the range 512..PAGE_SIZE, 171445bce8f3SLinus Torvalds * and the case we care about most is PAGE_SIZE. 171545bce8f3SLinus Torvalds * 171645bce8f3SLinus Torvalds * So this *could* possibly be written with those 171745bce8f3SLinus Torvalds * constraints in mind (relevant mostly if some 171845bce8f3SLinus Torvalds * architecture has a slow bit-scan instruction) 171945bce8f3SLinus Torvalds */ 172045bce8f3SLinus Torvalds static inline int block_size_bits(unsigned int blocksize) 172145bce8f3SLinus Torvalds { 172245bce8f3SLinus Torvalds return ilog2(blocksize); 172345bce8f3SLinus Torvalds } 172445bce8f3SLinus Torvalds 1725c6c8c3e7SPankaj Raghav static struct buffer_head *folio_create_buffers(struct folio *folio, 1726c6c8c3e7SPankaj Raghav struct inode *inode, 1727c6c8c3e7SPankaj Raghav unsigned int b_state) 172845bce8f3SLinus Torvalds { 1729c6c8c3e7SPankaj Raghav BUG_ON(!folio_test_locked(folio)); 173045bce8f3SLinus Torvalds 1731c6c8c3e7SPankaj Raghav if (!folio_buffers(folio)) 1732c6c8c3e7SPankaj Raghav folio_create_empty_buffers(folio, 1733c6c8c3e7SPankaj Raghav 1 << READ_ONCE(inode->i_blkbits), 17346aa7de05SMark Rutland b_state); 1735c6c8c3e7SPankaj Raghav return folio_buffers(folio); 173645bce8f3SLinus Torvalds } 173745bce8f3SLinus Torvalds 173845bce8f3SLinus Torvalds /* 17391da177e4SLinus Torvalds * NOTE! All mapped/uptodate combinations are valid: 17401da177e4SLinus Torvalds * 17411da177e4SLinus Torvalds * Mapped Uptodate Meaning 17421da177e4SLinus Torvalds * 17431da177e4SLinus Torvalds * No No "unknown" - must do get_block() 17441da177e4SLinus Torvalds * No Yes "hole" - zero-filled 17451da177e4SLinus Torvalds * Yes No "allocated" - allocated on disk, not read in 17461da177e4SLinus Torvalds * Yes Yes "valid" - allocated and up-to-date in memory. 17471da177e4SLinus Torvalds * 17481da177e4SLinus Torvalds * "Dirty" is valid only with the last case (mapped+uptodate). 17491da177e4SLinus Torvalds */ 17501da177e4SLinus Torvalds 17511da177e4SLinus Torvalds /* 17521da177e4SLinus Torvalds * While block_write_full_page is writing back the dirty buffers under 17531da177e4SLinus Torvalds * the page lock, whoever dirtied the buffers may decide to clean them 17541da177e4SLinus Torvalds * again at any time. We handle that by only looking at the buffer 17551da177e4SLinus Torvalds * state inside lock_buffer(). 17561da177e4SLinus Torvalds * 17571da177e4SLinus Torvalds * If block_write_full_page() is called for regular writeback 17581da177e4SLinus Torvalds * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a 17591da177e4SLinus Torvalds * locked buffer. This only can happen if someone has written the buffer 17601da177e4SLinus Torvalds * directly, with submit_bh(). At the address_space level PageWriteback 17611da177e4SLinus Torvalds * prevents this contention from occurring. 17626e34eeddSTheodore Ts'o * 17636e34eeddSTheodore Ts'o * If block_write_full_page() is called with wbc->sync_mode == 176470fd7614SChristoph Hellwig * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this 1765721a9602SJens Axboe * causes the writes to be flagged as synchronous writes. 17661da177e4SLinus Torvalds */ 176753418a18SMatthew Wilcox (Oracle) int __block_write_full_folio(struct inode *inode, struct folio *folio, 176835c80d5fSChris Mason get_block_t *get_block, struct writeback_control *wbc, 176935c80d5fSChris Mason bh_end_io_t *handler) 17701da177e4SLinus Torvalds { 17711da177e4SLinus Torvalds int err; 17721da177e4SLinus Torvalds sector_t block; 17731da177e4SLinus Torvalds sector_t last_block; 1774f0fbd5fcSAndrew Morton struct buffer_head *bh, *head; 177545bce8f3SLinus Torvalds unsigned int blocksize, bbits; 17761da177e4SLinus Torvalds int nr_underway = 0; 17773ae72869SBart Van Assche blk_opf_t write_flags = wbc_to_write_flags(wbc); 17781da177e4SLinus Torvalds 177953418a18SMatthew Wilcox (Oracle) head = folio_create_buffers(folio, inode, 17801da177e4SLinus Torvalds (1 << BH_Dirty) | (1 << BH_Uptodate)); 17811da177e4SLinus Torvalds 17821da177e4SLinus Torvalds /* 1783e621900aSMatthew Wilcox (Oracle) * Be very careful. We have no exclusion from block_dirty_folio 17841da177e4SLinus Torvalds * here, and the (potentially unmapped) buffers may become dirty at 17851da177e4SLinus Torvalds * any time. If a buffer becomes dirty here after we've inspected it 178653418a18SMatthew Wilcox (Oracle) * then we just miss that fact, and the folio stays dirty. 17871da177e4SLinus Torvalds * 1788e621900aSMatthew Wilcox (Oracle) * Buffers outside i_size may be dirtied by block_dirty_folio; 17891da177e4SLinus Torvalds * handle that here by just cleaning them. 17901da177e4SLinus Torvalds */ 17911da177e4SLinus Torvalds 17921da177e4SLinus Torvalds bh = head; 179345bce8f3SLinus Torvalds blocksize = bh->b_size; 179445bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 179545bce8f3SLinus Torvalds 179653418a18SMatthew Wilcox (Oracle) block = (sector_t)folio->index << (PAGE_SHIFT - bbits); 179745bce8f3SLinus Torvalds last_block = (i_size_read(inode) - 1) >> bbits; 17981da177e4SLinus Torvalds 17991da177e4SLinus Torvalds /* 18001da177e4SLinus Torvalds * Get all the dirty buffers mapped to disk addresses and 18011da177e4SLinus Torvalds * handle any aliases from the underlying blockdev's mapping. 18021da177e4SLinus Torvalds */ 18031da177e4SLinus Torvalds do { 18041da177e4SLinus Torvalds if (block > last_block) { 18051da177e4SLinus Torvalds /* 18061da177e4SLinus Torvalds * mapped buffers outside i_size will occur, because 180753418a18SMatthew Wilcox (Oracle) * this folio can be outside i_size when there is a 18081da177e4SLinus Torvalds * truncate in progress. 18091da177e4SLinus Torvalds */ 18101da177e4SLinus Torvalds /* 18111da177e4SLinus Torvalds * The buffer was zeroed by block_write_full_page() 18121da177e4SLinus Torvalds */ 18131da177e4SLinus Torvalds clear_buffer_dirty(bh); 18141da177e4SLinus Torvalds set_buffer_uptodate(bh); 181529a814d2SAlex Tomas } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && 181629a814d2SAlex Tomas buffer_dirty(bh)) { 1817b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 18181da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 18191da177e4SLinus Torvalds if (err) 18201da177e4SLinus Torvalds goto recover; 182129a814d2SAlex Tomas clear_buffer_delay(bh); 18221da177e4SLinus Torvalds if (buffer_new(bh)) { 18231da177e4SLinus Torvalds /* blockdev mappings never come here */ 18241da177e4SLinus Torvalds clear_buffer_new(bh); 1825e64855c6SJan Kara clean_bdev_bh_alias(bh); 18261da177e4SLinus Torvalds } 18271da177e4SLinus Torvalds } 18281da177e4SLinus Torvalds bh = bh->b_this_page; 18291da177e4SLinus Torvalds block++; 18301da177e4SLinus Torvalds } while (bh != head); 18311da177e4SLinus Torvalds 18321da177e4SLinus Torvalds do { 18331da177e4SLinus Torvalds if (!buffer_mapped(bh)) 18341da177e4SLinus Torvalds continue; 18351da177e4SLinus Torvalds /* 18361da177e4SLinus Torvalds * If it's a fully non-blocking write attempt and we cannot 183753418a18SMatthew Wilcox (Oracle) * lock the buffer then redirty the folio. Note that this can 18385b0830cbSJens Axboe * potentially cause a busy-wait loop from writeback threads 18395b0830cbSJens Axboe * and kswapd activity, but those code paths have their own 18405b0830cbSJens Axboe * higher-level throttling. 18411da177e4SLinus Torvalds */ 18421b430beeSWu Fengguang if (wbc->sync_mode != WB_SYNC_NONE) { 18431da177e4SLinus Torvalds lock_buffer(bh); 1844ca5de404SNick Piggin } else if (!trylock_buffer(bh)) { 184553418a18SMatthew Wilcox (Oracle) folio_redirty_for_writepage(wbc, folio); 18461da177e4SLinus Torvalds continue; 18471da177e4SLinus Torvalds } 18481da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 184935c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 18501da177e4SLinus Torvalds } else { 18511da177e4SLinus Torvalds unlock_buffer(bh); 18521da177e4SLinus Torvalds } 18531da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 18541da177e4SLinus Torvalds 18551da177e4SLinus Torvalds /* 185653418a18SMatthew Wilcox (Oracle) * The folio and its buffers are protected by the writeback flag, 185753418a18SMatthew Wilcox (Oracle) * so we can drop the bh refcounts early. 18581da177e4SLinus Torvalds */ 185953418a18SMatthew Wilcox (Oracle) BUG_ON(folio_test_writeback(folio)); 186053418a18SMatthew Wilcox (Oracle) folio_start_writeback(folio); 18611da177e4SLinus Torvalds 18621da177e4SLinus Torvalds do { 18631da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 18641da177e4SLinus Torvalds if (buffer_async_write(bh)) { 18651420c4a5SBart Van Assche submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc); 18661da177e4SLinus Torvalds nr_underway++; 1867ad576e63SNick Piggin } 18681da177e4SLinus Torvalds bh = next; 18691da177e4SLinus Torvalds } while (bh != head); 187053418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 18711da177e4SLinus Torvalds 18721da177e4SLinus Torvalds err = 0; 18731da177e4SLinus Torvalds done: 18741da177e4SLinus Torvalds if (nr_underway == 0) { 18751da177e4SLinus Torvalds /* 187653418a18SMatthew Wilcox (Oracle) * The folio was marked dirty, but the buffers were 18771da177e4SLinus Torvalds * clean. Someone wrote them back by hand with 187879f59784SZhang Yi * write_dirty_buffer/submit_bh. A rare case. 18791da177e4SLinus Torvalds */ 188053418a18SMatthew Wilcox (Oracle) folio_end_writeback(folio); 18813d67f2d7SNick Piggin 18821da177e4SLinus Torvalds /* 188353418a18SMatthew Wilcox (Oracle) * The folio and buffer_heads can be released at any time from 18841da177e4SLinus Torvalds * here on. 18851da177e4SLinus Torvalds */ 18861da177e4SLinus Torvalds } 18871da177e4SLinus Torvalds return err; 18881da177e4SLinus Torvalds 18891da177e4SLinus Torvalds recover: 18901da177e4SLinus Torvalds /* 18911da177e4SLinus Torvalds * ENOSPC, or some other error. We may already have added some 18921da177e4SLinus Torvalds * blocks to the file, so we need to write these out to avoid 18931da177e4SLinus Torvalds * exposing stale data. 189453418a18SMatthew Wilcox (Oracle) * The folio is currently locked and not marked for writeback 18951da177e4SLinus Torvalds */ 18961da177e4SLinus Torvalds bh = head; 18971da177e4SLinus Torvalds /* Recovery: lock and submit the mapped buffers */ 18981da177e4SLinus Torvalds do { 189929a814d2SAlex Tomas if (buffer_mapped(bh) && buffer_dirty(bh) && 190029a814d2SAlex Tomas !buffer_delay(bh)) { 19011da177e4SLinus Torvalds lock_buffer(bh); 190235c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 19031da177e4SLinus Torvalds } else { 19041da177e4SLinus Torvalds /* 19051da177e4SLinus Torvalds * The buffer may have been set dirty during 190653418a18SMatthew Wilcox (Oracle) * attachment to a dirty folio. 19071da177e4SLinus Torvalds */ 19081da177e4SLinus Torvalds clear_buffer_dirty(bh); 19091da177e4SLinus Torvalds } 19101da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 191153418a18SMatthew Wilcox (Oracle) folio_set_error(folio); 191253418a18SMatthew Wilcox (Oracle) BUG_ON(folio_test_writeback(folio)); 191353418a18SMatthew Wilcox (Oracle) mapping_set_error(folio->mapping, err); 191453418a18SMatthew Wilcox (Oracle) folio_start_writeback(folio); 19151da177e4SLinus Torvalds do { 19161da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 19171da177e4SLinus Torvalds if (buffer_async_write(bh)) { 19181da177e4SLinus Torvalds clear_buffer_dirty(bh); 19191420c4a5SBart Van Assche submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc); 19201da177e4SLinus Torvalds nr_underway++; 1921ad576e63SNick Piggin } 19221da177e4SLinus Torvalds bh = next; 19231da177e4SLinus Torvalds } while (bh != head); 192453418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 19251da177e4SLinus Torvalds goto done; 19261da177e4SLinus Torvalds } 192753418a18SMatthew Wilcox (Oracle) EXPORT_SYMBOL(__block_write_full_folio); 19281da177e4SLinus Torvalds 1929afddba49SNick Piggin /* 1930*4a9622f2SMatthew Wilcox (Oracle) * If a folio has any new buffers, zero them out here, and mark them uptodate 1931afddba49SNick Piggin * and dirty so they'll be written out (in order to prevent uninitialised 1932afddba49SNick Piggin * block data from leaking). And clear the new bit. 1933afddba49SNick Piggin */ 1934*4a9622f2SMatthew Wilcox (Oracle) void folio_zero_new_buffers(struct folio *folio, size_t from, size_t to) 1935afddba49SNick Piggin { 1936*4a9622f2SMatthew Wilcox (Oracle) size_t block_start, block_end; 1937afddba49SNick Piggin struct buffer_head *head, *bh; 1938afddba49SNick Piggin 1939*4a9622f2SMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 1940*4a9622f2SMatthew Wilcox (Oracle) head = folio_buffers(folio); 1941*4a9622f2SMatthew Wilcox (Oracle) if (!head) 1942afddba49SNick Piggin return; 1943afddba49SNick Piggin 1944*4a9622f2SMatthew Wilcox (Oracle) bh = head; 1945afddba49SNick Piggin block_start = 0; 1946afddba49SNick Piggin do { 1947afddba49SNick Piggin block_end = block_start + bh->b_size; 1948afddba49SNick Piggin 1949afddba49SNick Piggin if (buffer_new(bh)) { 1950afddba49SNick Piggin if (block_end > from && block_start < to) { 1951*4a9622f2SMatthew Wilcox (Oracle) if (!folio_test_uptodate(folio)) { 1952*4a9622f2SMatthew Wilcox (Oracle) size_t start, xend; 1953afddba49SNick Piggin 1954afddba49SNick Piggin start = max(from, block_start); 1955*4a9622f2SMatthew Wilcox (Oracle) xend = min(to, block_end); 1956afddba49SNick Piggin 1957*4a9622f2SMatthew Wilcox (Oracle) folio_zero_segment(folio, start, xend); 1958afddba49SNick Piggin set_buffer_uptodate(bh); 1959afddba49SNick Piggin } 1960afddba49SNick Piggin 1961afddba49SNick Piggin clear_buffer_new(bh); 1962afddba49SNick Piggin mark_buffer_dirty(bh); 1963afddba49SNick Piggin } 1964afddba49SNick Piggin } 1965afddba49SNick Piggin 1966afddba49SNick Piggin block_start = block_end; 1967afddba49SNick Piggin bh = bh->b_this_page; 1968afddba49SNick Piggin } while (bh != head); 1969afddba49SNick Piggin } 1970*4a9622f2SMatthew Wilcox (Oracle) EXPORT_SYMBOL(folio_zero_new_buffers); 1971afddba49SNick Piggin 1972ae259a9cSChristoph Hellwig static void 1973ae259a9cSChristoph Hellwig iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, 19746d49cc85SChristoph Hellwig const struct iomap *iomap) 1975ae259a9cSChristoph Hellwig { 1976ae259a9cSChristoph Hellwig loff_t offset = block << inode->i_blkbits; 1977ae259a9cSChristoph Hellwig 1978ae259a9cSChristoph Hellwig bh->b_bdev = iomap->bdev; 1979ae259a9cSChristoph Hellwig 1980ae259a9cSChristoph Hellwig /* 1981ae259a9cSChristoph Hellwig * Block points to offset in file we need to map, iomap contains 1982ae259a9cSChristoph Hellwig * the offset at which the map starts. If the map ends before the 1983ae259a9cSChristoph Hellwig * current block, then do not map the buffer and let the caller 1984ae259a9cSChristoph Hellwig * handle it. 1985ae259a9cSChristoph Hellwig */ 1986ae259a9cSChristoph Hellwig BUG_ON(offset >= iomap->offset + iomap->length); 1987ae259a9cSChristoph Hellwig 1988ae259a9cSChristoph Hellwig switch (iomap->type) { 1989ae259a9cSChristoph Hellwig case IOMAP_HOLE: 1990ae259a9cSChristoph Hellwig /* 1991ae259a9cSChristoph Hellwig * If the buffer is not up to date or beyond the current EOF, 1992ae259a9cSChristoph Hellwig * we need to mark it as new to ensure sub-block zeroing is 1993ae259a9cSChristoph Hellwig * executed if necessary. 1994ae259a9cSChristoph Hellwig */ 1995ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 1996ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 1997ae259a9cSChristoph Hellwig set_buffer_new(bh); 1998ae259a9cSChristoph Hellwig break; 1999ae259a9cSChristoph Hellwig case IOMAP_DELALLOC: 2000ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 2001ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 2002ae259a9cSChristoph Hellwig set_buffer_new(bh); 2003ae259a9cSChristoph Hellwig set_buffer_uptodate(bh); 2004ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 2005ae259a9cSChristoph Hellwig set_buffer_delay(bh); 2006ae259a9cSChristoph Hellwig break; 2007ae259a9cSChristoph Hellwig case IOMAP_UNWRITTEN: 2008ae259a9cSChristoph Hellwig /* 20093d7b6b21SAndreas Gruenbacher * For unwritten regions, we always need to ensure that regions 20103d7b6b21SAndreas Gruenbacher * in the block we are not writing to are zeroed. Mark the 20113d7b6b21SAndreas Gruenbacher * buffer as new to ensure this. 2012ae259a9cSChristoph Hellwig */ 2013ae259a9cSChristoph Hellwig set_buffer_new(bh); 2014ae259a9cSChristoph Hellwig set_buffer_unwritten(bh); 2015df561f66SGustavo A. R. Silva fallthrough; 2016ae259a9cSChristoph Hellwig case IOMAP_MAPPED: 20173d7b6b21SAndreas Gruenbacher if ((iomap->flags & IOMAP_F_NEW) || 20183d7b6b21SAndreas Gruenbacher offset >= i_size_read(inode)) 2019ae259a9cSChristoph Hellwig set_buffer_new(bh); 202019fe5f64SAndreas Gruenbacher bh->b_blocknr = (iomap->addr + offset - iomap->offset) >> 202119fe5f64SAndreas Gruenbacher inode->i_blkbits; 2022ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 2023ae259a9cSChristoph Hellwig break; 2024ae259a9cSChristoph Hellwig } 2025ae259a9cSChristoph Hellwig } 2026ae259a9cSChristoph Hellwig 2027d1bd0b4eSMatthew Wilcox (Oracle) int __block_write_begin_int(struct folio *folio, loff_t pos, unsigned len, 20286d49cc85SChristoph Hellwig get_block_t *get_block, const struct iomap *iomap) 20291da177e4SLinus Torvalds { 203009cbfeafSKirill A. Shutemov unsigned from = pos & (PAGE_SIZE - 1); 2031ebdec241SChristoph Hellwig unsigned to = from + len; 2032d1bd0b4eSMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 20331da177e4SLinus Torvalds unsigned block_start, block_end; 20341da177e4SLinus Torvalds sector_t block; 20351da177e4SLinus Torvalds int err = 0; 20361da177e4SLinus Torvalds unsigned blocksize, bbits; 20371da177e4SLinus Torvalds struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; 20381da177e4SLinus Torvalds 2039d1bd0b4eSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 204009cbfeafSKirill A. Shutemov BUG_ON(from > PAGE_SIZE); 204109cbfeafSKirill A. Shutemov BUG_ON(to > PAGE_SIZE); 20421da177e4SLinus Torvalds BUG_ON(from > to); 20431da177e4SLinus Torvalds 2044c6c8c3e7SPankaj Raghav head = folio_create_buffers(folio, inode, 0); 204545bce8f3SLinus Torvalds blocksize = head->b_size; 204645bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 20471da177e4SLinus Torvalds 2048d1bd0b4eSMatthew Wilcox (Oracle) block = (sector_t)folio->index << (PAGE_SHIFT - bbits); 20491da177e4SLinus Torvalds 20501da177e4SLinus Torvalds for(bh = head, block_start = 0; bh != head || !block_start; 20511da177e4SLinus Torvalds block++, block_start=block_end, bh = bh->b_this_page) { 20521da177e4SLinus Torvalds block_end = block_start + blocksize; 20531da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 2054d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 20551da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 20561da177e4SLinus Torvalds set_buffer_uptodate(bh); 20571da177e4SLinus Torvalds } 20581da177e4SLinus Torvalds continue; 20591da177e4SLinus Torvalds } 20601da177e4SLinus Torvalds if (buffer_new(bh)) 20611da177e4SLinus Torvalds clear_buffer_new(bh); 20621da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2063b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2064ae259a9cSChristoph Hellwig if (get_block) { 20651da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 20661da177e4SLinus Torvalds if (err) 2067f3ddbdc6SNick Piggin break; 2068ae259a9cSChristoph Hellwig } else { 2069ae259a9cSChristoph Hellwig iomap_to_bh(inode, block, bh, iomap); 2070ae259a9cSChristoph Hellwig } 2071ae259a9cSChristoph Hellwig 20721da177e4SLinus Torvalds if (buffer_new(bh)) { 2073e64855c6SJan Kara clean_bdev_bh_alias(bh); 2074d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 2075637aff46SNick Piggin clear_buffer_new(bh); 20761da177e4SLinus Torvalds set_buffer_uptodate(bh); 2077637aff46SNick Piggin mark_buffer_dirty(bh); 20781da177e4SLinus Torvalds continue; 20791da177e4SLinus Torvalds } 2080eebd2aa3SChristoph Lameter if (block_end > to || block_start < from) 2081d1bd0b4eSMatthew Wilcox (Oracle) folio_zero_segments(folio, 2082eebd2aa3SChristoph Lameter to, block_end, 2083eebd2aa3SChristoph Lameter block_start, from); 20841da177e4SLinus Torvalds continue; 20851da177e4SLinus Torvalds } 20861da177e4SLinus Torvalds } 2087d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 20881da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 20891da177e4SLinus Torvalds set_buffer_uptodate(bh); 20901da177e4SLinus Torvalds continue; 20911da177e4SLinus Torvalds } 20921da177e4SLinus Torvalds if (!buffer_uptodate(bh) && !buffer_delay(bh) && 209333a266ddSDavid Chinner !buffer_unwritten(bh) && 20941da177e4SLinus Torvalds (block_start < from || block_end > to)) { 2095e7ea1129SZhang Yi bh_read_nowait(bh, 0); 20961da177e4SLinus Torvalds *wait_bh++=bh; 20971da177e4SLinus Torvalds } 20981da177e4SLinus Torvalds } 20991da177e4SLinus Torvalds /* 21001da177e4SLinus Torvalds * If we issued read requests - let them complete. 21011da177e4SLinus Torvalds */ 21021da177e4SLinus Torvalds while(wait_bh > wait) { 21031da177e4SLinus Torvalds wait_on_buffer(*--wait_bh); 21041da177e4SLinus Torvalds if (!buffer_uptodate(*wait_bh)) 2105f3ddbdc6SNick Piggin err = -EIO; 21061da177e4SLinus Torvalds } 2107f9f07b6cSJan Kara if (unlikely(err)) 2108*4a9622f2SMatthew Wilcox (Oracle) folio_zero_new_buffers(folio, from, to); 21091da177e4SLinus Torvalds return err; 21101da177e4SLinus Torvalds } 2111ae259a9cSChristoph Hellwig 2112ae259a9cSChristoph Hellwig int __block_write_begin(struct page *page, loff_t pos, unsigned len, 2113ae259a9cSChristoph Hellwig get_block_t *get_block) 2114ae259a9cSChristoph Hellwig { 2115d1bd0b4eSMatthew Wilcox (Oracle) return __block_write_begin_int(page_folio(page), pos, len, get_block, 2116d1bd0b4eSMatthew Wilcox (Oracle) NULL); 2117ae259a9cSChristoph Hellwig } 2118ebdec241SChristoph Hellwig EXPORT_SYMBOL(__block_write_begin); 21191da177e4SLinus Torvalds 21208c6cb3e3SMatthew Wilcox (Oracle) static int __block_commit_write(struct inode *inode, struct folio *folio, 21218c6cb3e3SMatthew Wilcox (Oracle) size_t from, size_t to) 21221da177e4SLinus Torvalds { 21238c6cb3e3SMatthew Wilcox (Oracle) size_t block_start, block_end; 21248c6cb3e3SMatthew Wilcox (Oracle) bool partial = false; 21251da177e4SLinus Torvalds unsigned blocksize; 21261da177e4SLinus Torvalds struct buffer_head *bh, *head; 21271da177e4SLinus Torvalds 21288c6cb3e3SMatthew Wilcox (Oracle) bh = head = folio_buffers(folio); 212945bce8f3SLinus Torvalds blocksize = bh->b_size; 21301da177e4SLinus Torvalds 213145bce8f3SLinus Torvalds block_start = 0; 213245bce8f3SLinus Torvalds do { 21331da177e4SLinus Torvalds block_end = block_start + blocksize; 21341da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 21351da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 21368c6cb3e3SMatthew Wilcox (Oracle) partial = true; 21371da177e4SLinus Torvalds } else { 21381da177e4SLinus Torvalds set_buffer_uptodate(bh); 21391da177e4SLinus Torvalds mark_buffer_dirty(bh); 21401da177e4SLinus Torvalds } 21414ebd3aecSYang Guo if (buffer_new(bh)) 2142afddba49SNick Piggin clear_buffer_new(bh); 214345bce8f3SLinus Torvalds 214445bce8f3SLinus Torvalds block_start = block_end; 214545bce8f3SLinus Torvalds bh = bh->b_this_page; 214645bce8f3SLinus Torvalds } while (bh != head); 21471da177e4SLinus Torvalds 21481da177e4SLinus Torvalds /* 21491da177e4SLinus Torvalds * If this is a partial write which happened to make all buffers 21502c69e205SMatthew Wilcox (Oracle) * uptodate then we can optimize away a bogus read_folio() for 21518c6cb3e3SMatthew Wilcox (Oracle) * the next read(). Here we 'discover' whether the folio went 21521da177e4SLinus Torvalds * uptodate as a result of this (potentially partial) write. 21531da177e4SLinus Torvalds */ 21541da177e4SLinus Torvalds if (!partial) 21558c6cb3e3SMatthew Wilcox (Oracle) folio_mark_uptodate(folio); 21561da177e4SLinus Torvalds return 0; 21571da177e4SLinus Torvalds } 21581da177e4SLinus Torvalds 21591da177e4SLinus Torvalds /* 2160155130a4SChristoph Hellwig * block_write_begin takes care of the basic task of block allocation and 2161155130a4SChristoph Hellwig * bringing partial write blocks uptodate first. 2162155130a4SChristoph Hellwig * 21637bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 2164afddba49SNick Piggin */ 2165155130a4SChristoph Hellwig int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, 2166b3992d1eSMatthew Wilcox (Oracle) struct page **pagep, get_block_t *get_block) 2167afddba49SNick Piggin { 216809cbfeafSKirill A. Shutemov pgoff_t index = pos >> PAGE_SHIFT; 2169afddba49SNick Piggin struct page *page; 21706e1db88dSChristoph Hellwig int status; 2171afddba49SNick Piggin 2172b7446e7cSMatthew Wilcox (Oracle) page = grab_cache_page_write_begin(mapping, index); 21736e1db88dSChristoph Hellwig if (!page) 21746e1db88dSChristoph Hellwig return -ENOMEM; 2175afddba49SNick Piggin 21766e1db88dSChristoph Hellwig status = __block_write_begin(page, pos, len, get_block); 2177afddba49SNick Piggin if (unlikely(status)) { 2178afddba49SNick Piggin unlock_page(page); 217909cbfeafSKirill A. Shutemov put_page(page); 21806e1db88dSChristoph Hellwig page = NULL; 2181afddba49SNick Piggin } 2182afddba49SNick Piggin 21836e1db88dSChristoph Hellwig *pagep = page; 2184afddba49SNick Piggin return status; 2185afddba49SNick Piggin } 2186afddba49SNick Piggin EXPORT_SYMBOL(block_write_begin); 2187afddba49SNick Piggin 2188afddba49SNick Piggin int block_write_end(struct file *file, struct address_space *mapping, 2189afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2190afddba49SNick Piggin struct page *page, void *fsdata) 2191afddba49SNick Piggin { 21928c6cb3e3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 2193afddba49SNick Piggin struct inode *inode = mapping->host; 21948c6cb3e3SMatthew Wilcox (Oracle) size_t start = pos - folio_pos(folio); 2195afddba49SNick Piggin 2196afddba49SNick Piggin if (unlikely(copied < len)) { 2197afddba49SNick Piggin /* 21982c69e205SMatthew Wilcox (Oracle) * The buffers that were written will now be uptodate, so 21992c69e205SMatthew Wilcox (Oracle) * we don't have to worry about a read_folio reading them 22002c69e205SMatthew Wilcox (Oracle) * and overwriting a partial write. However if we have 22012c69e205SMatthew Wilcox (Oracle) * encountered a short write and only partially written 22022c69e205SMatthew Wilcox (Oracle) * into a buffer, it will not be marked uptodate, so a 22032c69e205SMatthew Wilcox (Oracle) * read_folio might come in and destroy our partial write. 2204afddba49SNick Piggin * 2205afddba49SNick Piggin * Do the simplest thing, and just treat any short write to a 22068c6cb3e3SMatthew Wilcox (Oracle) * non uptodate folio as a zero-length write, and force the 2207afddba49SNick Piggin * caller to redo the whole thing. 2208afddba49SNick Piggin */ 22098c6cb3e3SMatthew Wilcox (Oracle) if (!folio_test_uptodate(folio)) 2210afddba49SNick Piggin copied = 0; 2211afddba49SNick Piggin 2212*4a9622f2SMatthew Wilcox (Oracle) folio_zero_new_buffers(folio, start+copied, start+len); 2213afddba49SNick Piggin } 22148c6cb3e3SMatthew Wilcox (Oracle) flush_dcache_folio(folio); 2215afddba49SNick Piggin 2216afddba49SNick Piggin /* This could be a short (even 0-length) commit */ 22178c6cb3e3SMatthew Wilcox (Oracle) __block_commit_write(inode, folio, start, start + copied); 2218afddba49SNick Piggin 2219afddba49SNick Piggin return copied; 2220afddba49SNick Piggin } 2221afddba49SNick Piggin EXPORT_SYMBOL(block_write_end); 2222afddba49SNick Piggin 2223afddba49SNick Piggin int generic_write_end(struct file *file, struct address_space *mapping, 2224afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2225afddba49SNick Piggin struct page *page, void *fsdata) 2226afddba49SNick Piggin { 22278af54f29SChristoph Hellwig struct inode *inode = mapping->host; 22288af54f29SChristoph Hellwig loff_t old_size = inode->i_size; 22298af54f29SChristoph Hellwig bool i_size_changed = false; 22308af54f29SChristoph Hellwig 2231afddba49SNick Piggin copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); 22328af54f29SChristoph Hellwig 22338af54f29SChristoph Hellwig /* 22348af54f29SChristoph Hellwig * No need to use i_size_read() here, the i_size cannot change under us 22358af54f29SChristoph Hellwig * because we hold i_rwsem. 22368af54f29SChristoph Hellwig * 22378af54f29SChristoph Hellwig * But it's important to update i_size while still holding page lock: 22388af54f29SChristoph Hellwig * page writeout could otherwise come in and zero beyond i_size. 22398af54f29SChristoph Hellwig */ 22408af54f29SChristoph Hellwig if (pos + copied > inode->i_size) { 22418af54f29SChristoph Hellwig i_size_write(inode, pos + copied); 22428af54f29SChristoph Hellwig i_size_changed = true; 22438af54f29SChristoph Hellwig } 22448af54f29SChristoph Hellwig 22458af54f29SChristoph Hellwig unlock_page(page); 22467a77dad7SAndreas Gruenbacher put_page(page); 22478af54f29SChristoph Hellwig 22488af54f29SChristoph Hellwig if (old_size < pos) 22498af54f29SChristoph Hellwig pagecache_isize_extended(inode, old_size, pos); 22508af54f29SChristoph Hellwig /* 22518af54f29SChristoph Hellwig * Don't mark the inode dirty under page lock. First, it unnecessarily 22528af54f29SChristoph Hellwig * makes the holding time of page lock longer. Second, it forces lock 22538af54f29SChristoph Hellwig * ordering of page lock and transaction start for journaling 22548af54f29SChristoph Hellwig * filesystems. 22558af54f29SChristoph Hellwig */ 22568af54f29SChristoph Hellwig if (i_size_changed) 22578af54f29SChristoph Hellwig mark_inode_dirty(inode); 225826ddb1f4SAndreas Gruenbacher return copied; 2259afddba49SNick Piggin } 2260afddba49SNick Piggin EXPORT_SYMBOL(generic_write_end); 2261afddba49SNick Piggin 2262afddba49SNick Piggin /* 22632e7e80f7SMatthew Wilcox (Oracle) * block_is_partially_uptodate checks whether buffers within a folio are 22648ab22b9aSHisashi Hifumi * uptodate or not. 22658ab22b9aSHisashi Hifumi * 22662e7e80f7SMatthew Wilcox (Oracle) * Returns true if all buffers which correspond to the specified part 22672e7e80f7SMatthew Wilcox (Oracle) * of the folio are uptodate. 22688ab22b9aSHisashi Hifumi */ 22692e7e80f7SMatthew Wilcox (Oracle) bool block_is_partially_uptodate(struct folio *folio, size_t from, size_t count) 22708ab22b9aSHisashi Hifumi { 22718ab22b9aSHisashi Hifumi unsigned block_start, block_end, blocksize; 22728ab22b9aSHisashi Hifumi unsigned to; 22738ab22b9aSHisashi Hifumi struct buffer_head *bh, *head; 22742e7e80f7SMatthew Wilcox (Oracle) bool ret = true; 22758ab22b9aSHisashi Hifumi 22762e7e80f7SMatthew Wilcox (Oracle) head = folio_buffers(folio); 22772e7e80f7SMatthew Wilcox (Oracle) if (!head) 22782e7e80f7SMatthew Wilcox (Oracle) return false; 227945bce8f3SLinus Torvalds blocksize = head->b_size; 22802e7e80f7SMatthew Wilcox (Oracle) to = min_t(unsigned, folio_size(folio) - from, count); 22818ab22b9aSHisashi Hifumi to = from + to; 22822e7e80f7SMatthew Wilcox (Oracle) if (from < blocksize && to > folio_size(folio) - blocksize) 22832e7e80f7SMatthew Wilcox (Oracle) return false; 22848ab22b9aSHisashi Hifumi 22858ab22b9aSHisashi Hifumi bh = head; 22868ab22b9aSHisashi Hifumi block_start = 0; 22878ab22b9aSHisashi Hifumi do { 22888ab22b9aSHisashi Hifumi block_end = block_start + blocksize; 22898ab22b9aSHisashi Hifumi if (block_end > from && block_start < to) { 22908ab22b9aSHisashi Hifumi if (!buffer_uptodate(bh)) { 22912e7e80f7SMatthew Wilcox (Oracle) ret = false; 22928ab22b9aSHisashi Hifumi break; 22938ab22b9aSHisashi Hifumi } 22948ab22b9aSHisashi Hifumi if (block_end >= to) 22958ab22b9aSHisashi Hifumi break; 22968ab22b9aSHisashi Hifumi } 22978ab22b9aSHisashi Hifumi block_start = block_end; 22988ab22b9aSHisashi Hifumi bh = bh->b_this_page; 22998ab22b9aSHisashi Hifumi } while (bh != head); 23008ab22b9aSHisashi Hifumi 23018ab22b9aSHisashi Hifumi return ret; 23028ab22b9aSHisashi Hifumi } 23038ab22b9aSHisashi Hifumi EXPORT_SYMBOL(block_is_partially_uptodate); 23048ab22b9aSHisashi Hifumi 23058ab22b9aSHisashi Hifumi /* 23062c69e205SMatthew Wilcox (Oracle) * Generic "read_folio" function for block devices that have the normal 23071da177e4SLinus Torvalds * get_block functionality. This is most of the block device filesystems. 23082c69e205SMatthew Wilcox (Oracle) * Reads the folio asynchronously --- the unlock_buffer() and 23091da177e4SLinus Torvalds * set/clear_buffer_uptodate() functions propagate buffer state into the 23102c69e205SMatthew Wilcox (Oracle) * folio once IO has completed. 23111da177e4SLinus Torvalds */ 23122c69e205SMatthew Wilcox (Oracle) int block_read_full_folio(struct folio *folio, get_block_t *get_block) 23131da177e4SLinus Torvalds { 23142c69e205SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 23151da177e4SLinus Torvalds sector_t iblock, lblock; 23161da177e4SLinus Torvalds struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 231745bce8f3SLinus Torvalds unsigned int blocksize, bbits; 23181da177e4SLinus Torvalds int nr, i; 23191da177e4SLinus Torvalds int fully_mapped = 1; 2320b7a6eb22SMatthew Wilcox (Oracle) bool page_error = false; 23214fa512ceSEric Biggers loff_t limit = i_size_read(inode); 23224fa512ceSEric Biggers 23234fa512ceSEric Biggers /* This is needed for ext4. */ 23244fa512ceSEric Biggers if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode)) 23254fa512ceSEric Biggers limit = inode->i_sb->s_maxbytes; 23261da177e4SLinus Torvalds 23272c69e205SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_large(folio), folio); 23282c69e205SMatthew Wilcox (Oracle) 2329c6c8c3e7SPankaj Raghav head = folio_create_buffers(folio, inode, 0); 233045bce8f3SLinus Torvalds blocksize = head->b_size; 233145bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 23321da177e4SLinus Torvalds 23332c69e205SMatthew Wilcox (Oracle) iblock = (sector_t)folio->index << (PAGE_SHIFT - bbits); 23344fa512ceSEric Biggers lblock = (limit+blocksize-1) >> bbits; 23351da177e4SLinus Torvalds bh = head; 23361da177e4SLinus Torvalds nr = 0; 23371da177e4SLinus Torvalds i = 0; 23381da177e4SLinus Torvalds 23391da177e4SLinus Torvalds do { 23401da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23411da177e4SLinus Torvalds continue; 23421da177e4SLinus Torvalds 23431da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2344c64610baSAndrew Morton int err = 0; 2345c64610baSAndrew Morton 23461da177e4SLinus Torvalds fully_mapped = 0; 23471da177e4SLinus Torvalds if (iblock < lblock) { 2348b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2349c64610baSAndrew Morton err = get_block(inode, iblock, bh, 0); 2350b7a6eb22SMatthew Wilcox (Oracle) if (err) { 23512c69e205SMatthew Wilcox (Oracle) folio_set_error(folio); 2352b7a6eb22SMatthew Wilcox (Oracle) page_error = true; 2353b7a6eb22SMatthew Wilcox (Oracle) } 23541da177e4SLinus Torvalds } 23551da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 23562c69e205SMatthew Wilcox (Oracle) folio_zero_range(folio, i * blocksize, 23572c69e205SMatthew Wilcox (Oracle) blocksize); 2358c64610baSAndrew Morton if (!err) 23591da177e4SLinus Torvalds set_buffer_uptodate(bh); 23601da177e4SLinus Torvalds continue; 23611da177e4SLinus Torvalds } 23621da177e4SLinus Torvalds /* 23631da177e4SLinus Torvalds * get_block() might have updated the buffer 23641da177e4SLinus Torvalds * synchronously 23651da177e4SLinus Torvalds */ 23661da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23671da177e4SLinus Torvalds continue; 23681da177e4SLinus Torvalds } 23691da177e4SLinus Torvalds arr[nr++] = bh; 23701da177e4SLinus Torvalds } while (i++, iblock++, (bh = bh->b_this_page) != head); 23711da177e4SLinus Torvalds 23721da177e4SLinus Torvalds if (fully_mapped) 23732c69e205SMatthew Wilcox (Oracle) folio_set_mappedtodisk(folio); 23741da177e4SLinus Torvalds 23751da177e4SLinus Torvalds if (!nr) { 23761da177e4SLinus Torvalds /* 23772c69e205SMatthew Wilcox (Oracle) * All buffers are uptodate - we can set the folio uptodate 23781da177e4SLinus Torvalds * as well. But not if get_block() returned an error. 23791da177e4SLinus Torvalds */ 2380b7a6eb22SMatthew Wilcox (Oracle) if (!page_error) 23812c69e205SMatthew Wilcox (Oracle) folio_mark_uptodate(folio); 23822c69e205SMatthew Wilcox (Oracle) folio_unlock(folio); 23831da177e4SLinus Torvalds return 0; 23841da177e4SLinus Torvalds } 23851da177e4SLinus Torvalds 23861da177e4SLinus Torvalds /* Stage two: lock the buffers */ 23871da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 23881da177e4SLinus Torvalds bh = arr[i]; 23891da177e4SLinus Torvalds lock_buffer(bh); 23901da177e4SLinus Torvalds mark_buffer_async_read(bh); 23911da177e4SLinus Torvalds } 23921da177e4SLinus Torvalds 23931da177e4SLinus Torvalds /* 23941da177e4SLinus Torvalds * Stage 3: start the IO. Check for uptodateness 23951da177e4SLinus Torvalds * inside the buffer lock in case another process reading 23961da177e4SLinus Torvalds * the underlying blockdev brought it uptodate (the sct fix). 23971da177e4SLinus Torvalds */ 23981da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 23991da177e4SLinus Torvalds bh = arr[i]; 24001da177e4SLinus Torvalds if (buffer_uptodate(bh)) 24011da177e4SLinus Torvalds end_buffer_async_read(bh, 1); 24021da177e4SLinus Torvalds else 24031420c4a5SBart Van Assche submit_bh(REQ_OP_READ, bh); 24041da177e4SLinus Torvalds } 24051da177e4SLinus Torvalds return 0; 24061da177e4SLinus Torvalds } 24072c69e205SMatthew Wilcox (Oracle) EXPORT_SYMBOL(block_read_full_folio); 24081da177e4SLinus Torvalds 24091da177e4SLinus Torvalds /* utility function for filesystems that need to do work on expanding 241089e10787SNick Piggin * truncates. Uses filesystem pagecache writes to allow the filesystem to 24111da177e4SLinus Torvalds * deal with the hole. 24121da177e4SLinus Torvalds */ 241389e10787SNick Piggin int generic_cont_expand_simple(struct inode *inode, loff_t size) 24141da177e4SLinus Torvalds { 24151da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 241653b524b8SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 24171da177e4SLinus Torvalds struct page *page; 24181468c6f4SAlexander Potapenko void *fsdata = NULL; 24191da177e4SLinus Torvalds int err; 24201da177e4SLinus Torvalds 2421c08d3b0eSnpiggin@suse.de err = inode_newsize_ok(inode, size); 2422c08d3b0eSnpiggin@suse.de if (err) 24231da177e4SLinus Torvalds goto out; 24241da177e4SLinus Torvalds 242553b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(NULL, mapping, size, 0, &page, &fsdata); 242689e10787SNick Piggin if (err) 242705eb0b51SOGAWA Hirofumi goto out; 242805eb0b51SOGAWA Hirofumi 242953b524b8SMatthew Wilcox (Oracle) err = aops->write_end(NULL, mapping, size, 0, 0, page, fsdata); 243089e10787SNick Piggin BUG_ON(err > 0); 243105eb0b51SOGAWA Hirofumi 243205eb0b51SOGAWA Hirofumi out: 243305eb0b51SOGAWA Hirofumi return err; 243405eb0b51SOGAWA Hirofumi } 24351fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_cont_expand_simple); 243605eb0b51SOGAWA Hirofumi 2437f1e3af72SAdrian Bunk static int cont_expand_zero(struct file *file, struct address_space *mapping, 243889e10787SNick Piggin loff_t pos, loff_t *bytes) 243905eb0b51SOGAWA Hirofumi { 244089e10787SNick Piggin struct inode *inode = mapping->host; 244153b524b8SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 244293407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 244389e10787SNick Piggin struct page *page; 24441468c6f4SAlexander Potapenko void *fsdata = NULL; 244589e10787SNick Piggin pgoff_t index, curidx; 244689e10787SNick Piggin loff_t curpos; 244789e10787SNick Piggin unsigned zerofrom, offset, len; 244889e10787SNick Piggin int err = 0; 244905eb0b51SOGAWA Hirofumi 245009cbfeafSKirill A. Shutemov index = pos >> PAGE_SHIFT; 245109cbfeafSKirill A. Shutemov offset = pos & ~PAGE_MASK; 245289e10787SNick Piggin 245309cbfeafSKirill A. Shutemov while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { 245409cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 245589e10787SNick Piggin if (zerofrom & (blocksize-1)) { 245689e10787SNick Piggin *bytes |= (blocksize-1); 245789e10787SNick Piggin (*bytes)++; 245889e10787SNick Piggin } 245909cbfeafSKirill A. Shutemov len = PAGE_SIZE - zerofrom; 246089e10787SNick Piggin 246153b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(file, mapping, curpos, len, 246289e10787SNick Piggin &page, &fsdata); 246389e10787SNick Piggin if (err) 246489e10787SNick Piggin goto out; 2465eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 246653b524b8SMatthew Wilcox (Oracle) err = aops->write_end(file, mapping, curpos, len, len, 246789e10787SNick Piggin page, fsdata); 246889e10787SNick Piggin if (err < 0) 246989e10787SNick Piggin goto out; 247089e10787SNick Piggin BUG_ON(err != len); 247189e10787SNick Piggin err = 0; 2472061e9746SOGAWA Hirofumi 2473061e9746SOGAWA Hirofumi balance_dirty_pages_ratelimited(mapping); 2474c2ca0fcdSMikulas Patocka 247508d405c8SDavidlohr Bueso if (fatal_signal_pending(current)) { 2476c2ca0fcdSMikulas Patocka err = -EINTR; 2477c2ca0fcdSMikulas Patocka goto out; 2478c2ca0fcdSMikulas Patocka } 247989e10787SNick Piggin } 248089e10787SNick Piggin 248189e10787SNick Piggin /* page covers the boundary, find the boundary offset */ 248289e10787SNick Piggin if (index == curidx) { 248309cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 248489e10787SNick Piggin /* if we will expand the thing last block will be filled */ 248589e10787SNick Piggin if (offset <= zerofrom) { 248689e10787SNick Piggin goto out; 248789e10787SNick Piggin } 248889e10787SNick Piggin if (zerofrom & (blocksize-1)) { 248989e10787SNick Piggin *bytes |= (blocksize-1); 249089e10787SNick Piggin (*bytes)++; 249189e10787SNick Piggin } 249289e10787SNick Piggin len = offset - zerofrom; 249389e10787SNick Piggin 249453b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(file, mapping, curpos, len, 249589e10787SNick Piggin &page, &fsdata); 249689e10787SNick Piggin if (err) 249789e10787SNick Piggin goto out; 2498eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 249953b524b8SMatthew Wilcox (Oracle) err = aops->write_end(file, mapping, curpos, len, len, 250089e10787SNick Piggin page, fsdata); 250189e10787SNick Piggin if (err < 0) 250289e10787SNick Piggin goto out; 250389e10787SNick Piggin BUG_ON(err != len); 250489e10787SNick Piggin err = 0; 250589e10787SNick Piggin } 250689e10787SNick Piggin out: 250789e10787SNick Piggin return err; 25081da177e4SLinus Torvalds } 25091da177e4SLinus Torvalds 25101da177e4SLinus Torvalds /* 25111da177e4SLinus Torvalds * For moronic filesystems that do not allow holes in file. 25121da177e4SLinus Torvalds * We may have to extend the file. 25131da177e4SLinus Torvalds */ 2514282dc178SChristoph Hellwig int cont_write_begin(struct file *file, struct address_space *mapping, 2515be3bbbc5SMatthew Wilcox (Oracle) loff_t pos, unsigned len, 251689e10787SNick Piggin struct page **pagep, void **fsdata, 251789e10787SNick Piggin get_block_t *get_block, loff_t *bytes) 25181da177e4SLinus Torvalds { 25191da177e4SLinus Torvalds struct inode *inode = mapping->host; 252093407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 252193407472SFabian Frederick unsigned int zerofrom; 252289e10787SNick Piggin int err; 25231da177e4SLinus Torvalds 252489e10787SNick Piggin err = cont_expand_zero(file, mapping, pos, bytes); 252589e10787SNick Piggin if (err) 2526155130a4SChristoph Hellwig return err; 25271da177e4SLinus Torvalds 252809cbfeafSKirill A. Shutemov zerofrom = *bytes & ~PAGE_MASK; 252989e10787SNick Piggin if (pos+len > *bytes && zerofrom & (blocksize-1)) { 25301da177e4SLinus Torvalds *bytes |= (blocksize-1); 25311da177e4SLinus Torvalds (*bytes)++; 25321da177e4SLinus Torvalds } 25331da177e4SLinus Torvalds 2534b3992d1eSMatthew Wilcox (Oracle) return block_write_begin(mapping, pos, len, pagep, get_block); 25351da177e4SLinus Torvalds } 25361fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(cont_write_begin); 25371da177e4SLinus Torvalds 25381da177e4SLinus Torvalds int block_commit_write(struct page *page, unsigned from, unsigned to) 25391da177e4SLinus Torvalds { 25408c6cb3e3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 25418c6cb3e3SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 25428c6cb3e3SMatthew Wilcox (Oracle) __block_commit_write(inode, folio, from, to); 25431da177e4SLinus Torvalds return 0; 25441da177e4SLinus Torvalds } 25451fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_commit_write); 25461da177e4SLinus Torvalds 254754171690SDavid Chinner /* 254854171690SDavid Chinner * block_page_mkwrite() is not allowed to change the file size as it gets 254954171690SDavid Chinner * called from a page fault handler when a page is first dirtied. Hence we must 255054171690SDavid Chinner * be careful to check for EOF conditions here. We set the page up correctly 255154171690SDavid Chinner * for a written page which means we get ENOSPC checking when writing into 255254171690SDavid Chinner * holes and correct delalloc and unwritten extent mapping on filesystems that 255354171690SDavid Chinner * support these features. 255454171690SDavid Chinner * 255554171690SDavid Chinner * We are not allowed to take the i_mutex here so we have to play games to 255654171690SDavid Chinner * protect against truncate races as the page could now be beyond EOF. Because 25577bb46a67Snpiggin@suse.de * truncate writes the inode size before removing pages, once we have the 255854171690SDavid Chinner * page lock we can determine safely if the page is beyond EOF. If it is not 255954171690SDavid Chinner * beyond EOF, then the page is guaranteed safe against truncation until we 256054171690SDavid Chinner * unlock the page. 2561ea13a864SJan Kara * 256214da9200SJan Kara * Direct callers of this function should protect against filesystem freezing 25635c500029SRoss Zwisler * using sb_start_pagefault() - sb_end_pagefault() functions. 256454171690SDavid Chinner */ 25655c500029SRoss Zwisler int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 256654171690SDavid Chinner get_block_t get_block) 256754171690SDavid Chinner { 2568fe181377SMatthew Wilcox (Oracle) struct folio *folio = page_folio(vmf->page); 2569496ad9aaSAl Viro struct inode *inode = file_inode(vma->vm_file); 257054171690SDavid Chinner unsigned long end; 257154171690SDavid Chinner loff_t size; 257224da4fabSJan Kara int ret; 257354171690SDavid Chinner 2574fe181377SMatthew Wilcox (Oracle) folio_lock(folio); 257554171690SDavid Chinner size = i_size_read(inode); 2576fe181377SMatthew Wilcox (Oracle) if ((folio->mapping != inode->i_mapping) || 2577fe181377SMatthew Wilcox (Oracle) (folio_pos(folio) >= size)) { 257824da4fabSJan Kara /* We overload EFAULT to mean page got truncated */ 257924da4fabSJan Kara ret = -EFAULT; 258024da4fabSJan Kara goto out_unlock; 258154171690SDavid Chinner } 258254171690SDavid Chinner 2583fe181377SMatthew Wilcox (Oracle) end = folio_size(folio); 2584fe181377SMatthew Wilcox (Oracle) /* folio is wholly or partially inside EOF */ 2585fe181377SMatthew Wilcox (Oracle) if (folio_pos(folio) + end > size) 2586fe181377SMatthew Wilcox (Oracle) end = size - folio_pos(folio); 258754171690SDavid Chinner 2588fe181377SMatthew Wilcox (Oracle) ret = __block_write_begin_int(folio, 0, end, get_block, NULL); 258954171690SDavid Chinner if (!ret) 25908c6cb3e3SMatthew Wilcox (Oracle) ret = __block_commit_write(inode, folio, 0, end); 259154171690SDavid Chinner 259224da4fabSJan Kara if (unlikely(ret < 0)) 259324da4fabSJan Kara goto out_unlock; 2594fe181377SMatthew Wilcox (Oracle) folio_mark_dirty(folio); 2595fe181377SMatthew Wilcox (Oracle) folio_wait_stable(folio); 259624da4fabSJan Kara return 0; 259724da4fabSJan Kara out_unlock: 2598fe181377SMatthew Wilcox (Oracle) folio_unlock(folio); 259954171690SDavid Chinner return ret; 260054171690SDavid Chinner } 26011fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_page_mkwrite); 26021da177e4SLinus Torvalds 26031da177e4SLinus Torvalds int block_truncate_page(struct address_space *mapping, 26041da177e4SLinus Torvalds loff_t from, get_block_t *get_block) 26051da177e4SLinus Torvalds { 260609cbfeafSKirill A. Shutemov pgoff_t index = from >> PAGE_SHIFT; 260709cbfeafSKirill A. Shutemov unsigned offset = from & (PAGE_SIZE-1); 26081da177e4SLinus Torvalds unsigned blocksize; 260954b21a79SAndrew Morton sector_t iblock; 26101da177e4SLinus Torvalds unsigned length, pos; 26111da177e4SLinus Torvalds struct inode *inode = mapping->host; 26121da177e4SLinus Torvalds struct page *page; 26131da177e4SLinus Torvalds struct buffer_head *bh; 2614dc7cb2d2SJiapeng Chong int err = 0; 26151da177e4SLinus Torvalds 261693407472SFabian Frederick blocksize = i_blocksize(inode); 26171da177e4SLinus Torvalds length = offset & (blocksize - 1); 26181da177e4SLinus Torvalds 26191da177e4SLinus Torvalds /* Block boundary? Nothing to do */ 26201da177e4SLinus Torvalds if (!length) 26211da177e4SLinus Torvalds return 0; 26221da177e4SLinus Torvalds 26231da177e4SLinus Torvalds length = blocksize - length; 262409cbfeafSKirill A. Shutemov iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); 26251da177e4SLinus Torvalds 26261da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 26271da177e4SLinus Torvalds if (!page) 2628dc7cb2d2SJiapeng Chong return -ENOMEM; 26291da177e4SLinus Torvalds 26301da177e4SLinus Torvalds if (!page_has_buffers(page)) 26311da177e4SLinus Torvalds create_empty_buffers(page, blocksize, 0); 26321da177e4SLinus Torvalds 26331da177e4SLinus Torvalds /* Find the buffer that contains "offset" */ 26341da177e4SLinus Torvalds bh = page_buffers(page); 26351da177e4SLinus Torvalds pos = blocksize; 26361da177e4SLinus Torvalds while (offset >= pos) { 26371da177e4SLinus Torvalds bh = bh->b_this_page; 26381da177e4SLinus Torvalds iblock++; 26391da177e4SLinus Torvalds pos += blocksize; 26401da177e4SLinus Torvalds } 26411da177e4SLinus Torvalds 26421da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2643b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 26441da177e4SLinus Torvalds err = get_block(inode, iblock, bh, 0); 26451da177e4SLinus Torvalds if (err) 26461da177e4SLinus Torvalds goto unlock; 26471da177e4SLinus Torvalds /* unmapped? It's a hole - nothing to do */ 26481da177e4SLinus Torvalds if (!buffer_mapped(bh)) 26491da177e4SLinus Torvalds goto unlock; 26501da177e4SLinus Torvalds } 26511da177e4SLinus Torvalds 26521da177e4SLinus Torvalds /* Ok, it's mapped. Make sure it's up-to-date */ 26531da177e4SLinus Torvalds if (PageUptodate(page)) 26541da177e4SLinus Torvalds set_buffer_uptodate(bh); 26551da177e4SLinus Torvalds 265633a266ddSDavid Chinner if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { 2657e7ea1129SZhang Yi err = bh_read(bh, 0); 26581da177e4SLinus Torvalds /* Uhhuh. Read error. Complain and punt. */ 2659e7ea1129SZhang Yi if (err < 0) 26601da177e4SLinus Torvalds goto unlock; 26611da177e4SLinus Torvalds } 26621da177e4SLinus Torvalds 2663eebd2aa3SChristoph Lameter zero_user(page, offset, length); 26641da177e4SLinus Torvalds mark_buffer_dirty(bh); 26651da177e4SLinus Torvalds 26661da177e4SLinus Torvalds unlock: 26671da177e4SLinus Torvalds unlock_page(page); 266809cbfeafSKirill A. Shutemov put_page(page); 2669dc7cb2d2SJiapeng Chong 26701da177e4SLinus Torvalds return err; 26711da177e4SLinus Torvalds } 26721fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_truncate_page); 26731da177e4SLinus Torvalds 26741da177e4SLinus Torvalds /* 26751da177e4SLinus Torvalds * The generic ->writepage function for buffer-backed address_spaces 26761da177e4SLinus Torvalds */ 26771b938c08SMatthew Wilcox int block_write_full_page(struct page *page, get_block_t *get_block, 26781b938c08SMatthew Wilcox struct writeback_control *wbc) 26791da177e4SLinus Torvalds { 268053418a18SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 2681bb0ea598SMatthew Wilcox (Oracle) struct inode * const inode = folio->mapping->host; 26821da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 26831da177e4SLinus Torvalds 2684bb0ea598SMatthew Wilcox (Oracle) /* Is the folio fully inside i_size? */ 2685bb0ea598SMatthew Wilcox (Oracle) if (folio_pos(folio) + folio_size(folio) <= i_size) 268653418a18SMatthew Wilcox (Oracle) return __block_write_full_folio(inode, folio, get_block, wbc, 26871b938c08SMatthew Wilcox end_buffer_async_write); 26881da177e4SLinus Torvalds 2689bb0ea598SMatthew Wilcox (Oracle) /* Is the folio fully outside i_size? (truncate in progress) */ 2690bb0ea598SMatthew Wilcox (Oracle) if (folio_pos(folio) >= i_size) { 269153418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 26921da177e4SLinus Torvalds return 0; /* don't care */ 26931da177e4SLinus Torvalds } 26941da177e4SLinus Torvalds 26951da177e4SLinus Torvalds /* 2696bb0ea598SMatthew Wilcox (Oracle) * The folio straddles i_size. It must be zeroed out on each and every 26972a61aa40SAdam Buchbinder * writepage invocation because it may be mmapped. "A file is mapped 26981da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 26991da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 27001da177e4SLinus Torvalds * writes to that region are not written out to the file." 27011da177e4SLinus Torvalds */ 2702bb0ea598SMatthew Wilcox (Oracle) folio_zero_segment(folio, offset_in_folio(folio, i_size), 2703bb0ea598SMatthew Wilcox (Oracle) folio_size(folio)); 270453418a18SMatthew Wilcox (Oracle) return __block_write_full_folio(inode, folio, get_block, wbc, 270535c80d5fSChris Mason end_buffer_async_write); 270635c80d5fSChris Mason } 27071fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_write_full_page); 270835c80d5fSChris Mason 27091da177e4SLinus Torvalds sector_t generic_block_bmap(struct address_space *mapping, sector_t block, 27101da177e4SLinus Torvalds get_block_t *get_block) 27111da177e4SLinus Torvalds { 27121da177e4SLinus Torvalds struct inode *inode = mapping->host; 27132a527d68SAlexander Potapenko struct buffer_head tmp = { 27142a527d68SAlexander Potapenko .b_size = i_blocksize(inode), 27152a527d68SAlexander Potapenko }; 27162a527d68SAlexander Potapenko 27171da177e4SLinus Torvalds get_block(inode, block, &tmp, 0); 27181da177e4SLinus Torvalds return tmp.b_blocknr; 27191da177e4SLinus Torvalds } 27201fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_block_bmap); 27211da177e4SLinus Torvalds 27224246a0b6SChristoph Hellwig static void end_bio_bh_io_sync(struct bio *bio) 27231da177e4SLinus Torvalds { 27241da177e4SLinus Torvalds struct buffer_head *bh = bio->bi_private; 27251da177e4SLinus Torvalds 2726b7c44ed9SJens Axboe if (unlikely(bio_flagged(bio, BIO_QUIET))) 272708bafc03SKeith Mannthey set_bit(BH_Quiet, &bh->b_state); 272808bafc03SKeith Mannthey 27294e4cbee9SChristoph Hellwig bh->b_end_io(bh, !bio->bi_status); 27301da177e4SLinus Torvalds bio_put(bio); 27311da177e4SLinus Torvalds } 27321da177e4SLinus Torvalds 27335bdf402aSRitesh Harjani (IBM) static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh, 27341420c4a5SBart Van Assche struct writeback_control *wbc) 27351da177e4SLinus Torvalds { 27361420c4a5SBart Van Assche const enum req_op op = opf & REQ_OP_MASK; 27371da177e4SLinus Torvalds struct bio *bio; 27381da177e4SLinus Torvalds 27391da177e4SLinus Torvalds BUG_ON(!buffer_locked(bh)); 27401da177e4SLinus Torvalds BUG_ON(!buffer_mapped(bh)); 27411da177e4SLinus Torvalds BUG_ON(!bh->b_end_io); 27428fb0e342SAneesh Kumar K.V BUG_ON(buffer_delay(bh)); 27438fb0e342SAneesh Kumar K.V BUG_ON(buffer_unwritten(bh)); 27441da177e4SLinus Torvalds 274548fd4f93SJens Axboe /* 274648fd4f93SJens Axboe * Only clear out a write error when rewriting 27471da177e4SLinus Torvalds */ 27482a222ca9SMike Christie if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) 27491da177e4SLinus Torvalds clear_buffer_write_io_error(bh); 27501da177e4SLinus Torvalds 275107888c66SChristoph Hellwig if (buffer_meta(bh)) 27521420c4a5SBart Van Assche opf |= REQ_META; 275307888c66SChristoph Hellwig if (buffer_prio(bh)) 27541420c4a5SBart Van Assche opf |= REQ_PRIO; 275507888c66SChristoph Hellwig 27561420c4a5SBart Van Assche bio = bio_alloc(bh->b_bdev, 1, opf, GFP_NOIO); 27571da177e4SLinus Torvalds 27584f74d15fSEric Biggers fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO); 27594f74d15fSEric Biggers 27604f024f37SKent Overstreet bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); 27611da177e4SLinus Torvalds 27626cf66b4cSKent Overstreet bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); 27636cf66b4cSKent Overstreet BUG_ON(bio->bi_iter.bi_size != bh->b_size); 27641da177e4SLinus Torvalds 27651da177e4SLinus Torvalds bio->bi_end_io = end_bio_bh_io_sync; 27661da177e4SLinus Torvalds bio->bi_private = bh; 27671da177e4SLinus Torvalds 276883c9c547SMing Lei /* Take care of bh's that straddle the end of the device */ 276983c9c547SMing Lei guard_bio_eod(bio); 277083c9c547SMing Lei 2771fd42df30SDennis Zhou if (wbc) { 2772fd42df30SDennis Zhou wbc_init_bio(wbc, bio); 277334e51a5eSTejun Heo wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size); 2774fd42df30SDennis Zhou } 2775fd42df30SDennis Zhou 27764e49ea4aSMike Christie submit_bio(bio); 27771da177e4SLinus Torvalds } 2778bafc0dbaSTejun Heo 27795bdf402aSRitesh Harjani (IBM) void submit_bh(blk_opf_t opf, struct buffer_head *bh) 278071368511SDarrick J. Wong { 27815bdf402aSRitesh Harjani (IBM) submit_bh_wbc(opf, bh, NULL); 278271368511SDarrick J. Wong } 27831fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(submit_bh); 27841da177e4SLinus Torvalds 27853ae72869SBart Van Assche void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) 27869cb569d6SChristoph Hellwig { 27879cb569d6SChristoph Hellwig lock_buffer(bh); 27889cb569d6SChristoph Hellwig if (!test_clear_buffer_dirty(bh)) { 27899cb569d6SChristoph Hellwig unlock_buffer(bh); 27909cb569d6SChristoph Hellwig return; 27919cb569d6SChristoph Hellwig } 27929cb569d6SChristoph Hellwig bh->b_end_io = end_buffer_write_sync; 27939cb569d6SChristoph Hellwig get_bh(bh); 27941420c4a5SBart Van Assche submit_bh(REQ_OP_WRITE | op_flags, bh); 27959cb569d6SChristoph Hellwig } 27969cb569d6SChristoph Hellwig EXPORT_SYMBOL(write_dirty_buffer); 27979cb569d6SChristoph Hellwig 27981da177e4SLinus Torvalds /* 27991da177e4SLinus Torvalds * For a data-integrity writeout, we need to wait upon any in-progress I/O 28001da177e4SLinus Torvalds * and then start new I/O and then wait upon it. The caller must have a ref on 28011da177e4SLinus Torvalds * the buffer_head. 28021da177e4SLinus Torvalds */ 28033ae72869SBart Van Assche int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) 28041da177e4SLinus Torvalds { 28051da177e4SLinus Torvalds WARN_ON(atomic_read(&bh->b_count) < 1); 28061da177e4SLinus Torvalds lock_buffer(bh); 28071da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 2808377254b2SXianting Tian /* 2809377254b2SXianting Tian * The bh should be mapped, but it might not be if the 2810377254b2SXianting Tian * device was hot-removed. Not much we can do but fail the I/O. 2811377254b2SXianting Tian */ 2812377254b2SXianting Tian if (!buffer_mapped(bh)) { 2813377254b2SXianting Tian unlock_buffer(bh); 2814377254b2SXianting Tian return -EIO; 2815377254b2SXianting Tian } 2816377254b2SXianting Tian 28171da177e4SLinus Torvalds get_bh(bh); 28181da177e4SLinus Torvalds bh->b_end_io = end_buffer_write_sync; 2819ab620620SRitesh Harjani (IBM) submit_bh(REQ_OP_WRITE | op_flags, bh); 28201da177e4SLinus Torvalds wait_on_buffer(bh); 2821ab620620SRitesh Harjani (IBM) if (!buffer_uptodate(bh)) 2822ab620620SRitesh Harjani (IBM) return -EIO; 28231da177e4SLinus Torvalds } else { 28241da177e4SLinus Torvalds unlock_buffer(bh); 28251da177e4SLinus Torvalds } 2826ab620620SRitesh Harjani (IBM) return 0; 28271da177e4SLinus Torvalds } 282887e99511SChristoph Hellwig EXPORT_SYMBOL(__sync_dirty_buffer); 282987e99511SChristoph Hellwig 283087e99511SChristoph Hellwig int sync_dirty_buffer(struct buffer_head *bh) 283187e99511SChristoph Hellwig { 283270fd7614SChristoph Hellwig return __sync_dirty_buffer(bh, REQ_SYNC); 283387e99511SChristoph Hellwig } 28341fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(sync_dirty_buffer); 28351da177e4SLinus Torvalds 28361da177e4SLinus Torvalds /* 283768189fefSMatthew Wilcox (Oracle) * try_to_free_buffers() checks if all the buffers on this particular folio 28381da177e4SLinus Torvalds * are unused, and releases them if so. 28391da177e4SLinus Torvalds * 28401da177e4SLinus Torvalds * Exclusion against try_to_free_buffers may be obtained by either 284168189fefSMatthew Wilcox (Oracle) * locking the folio or by holding its mapping's private_lock. 28421da177e4SLinus Torvalds * 284368189fefSMatthew Wilcox (Oracle) * If the folio is dirty but all the buffers are clean then we need to 284468189fefSMatthew Wilcox (Oracle) * be sure to mark the folio clean as well. This is because the folio 28451da177e4SLinus Torvalds * may be against a block device, and a later reattachment of buffers 284668189fefSMatthew Wilcox (Oracle) * to a dirty folio will set *all* buffers dirty. Which would corrupt 28471da177e4SLinus Torvalds * filesystem data on the same device. 28481da177e4SLinus Torvalds * 284968189fefSMatthew Wilcox (Oracle) * The same applies to regular filesystem folios: if all the buffers are 285068189fefSMatthew Wilcox (Oracle) * clean then we set the folio clean and proceed. To do that, we require 2851e621900aSMatthew Wilcox (Oracle) * total exclusion from block_dirty_folio(). That is obtained with 28521da177e4SLinus Torvalds * private_lock. 28531da177e4SLinus Torvalds * 28541da177e4SLinus Torvalds * try_to_free_buffers() is non-blocking. 28551da177e4SLinus Torvalds */ 28561da177e4SLinus Torvalds static inline int buffer_busy(struct buffer_head *bh) 28571da177e4SLinus Torvalds { 28581da177e4SLinus Torvalds return atomic_read(&bh->b_count) | 28591da177e4SLinus Torvalds (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); 28601da177e4SLinus Torvalds } 28611da177e4SLinus Torvalds 286264394763SMatthew Wilcox (Oracle) static bool 286364394763SMatthew Wilcox (Oracle) drop_buffers(struct folio *folio, struct buffer_head **buffers_to_free) 28641da177e4SLinus Torvalds { 286564394763SMatthew Wilcox (Oracle) struct buffer_head *head = folio_buffers(folio); 28661da177e4SLinus Torvalds struct buffer_head *bh; 28671da177e4SLinus Torvalds 28681da177e4SLinus Torvalds bh = head; 28691da177e4SLinus Torvalds do { 28701da177e4SLinus Torvalds if (buffer_busy(bh)) 28711da177e4SLinus Torvalds goto failed; 28721da177e4SLinus Torvalds bh = bh->b_this_page; 28731da177e4SLinus Torvalds } while (bh != head); 28741da177e4SLinus Torvalds 28751da177e4SLinus Torvalds do { 28761da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 28771da177e4SLinus Torvalds 2878535ee2fbSJan Kara if (bh->b_assoc_map) 28791da177e4SLinus Torvalds __remove_assoc_queue(bh); 28801da177e4SLinus Torvalds bh = next; 28811da177e4SLinus Torvalds } while (bh != head); 28821da177e4SLinus Torvalds *buffers_to_free = head; 288364394763SMatthew Wilcox (Oracle) folio_detach_private(folio); 288464394763SMatthew Wilcox (Oracle) return true; 28851da177e4SLinus Torvalds failed: 288664394763SMatthew Wilcox (Oracle) return false; 28871da177e4SLinus Torvalds } 28881da177e4SLinus Torvalds 288968189fefSMatthew Wilcox (Oracle) bool try_to_free_buffers(struct folio *folio) 28901da177e4SLinus Torvalds { 289168189fefSMatthew Wilcox (Oracle) struct address_space * const mapping = folio->mapping; 28921da177e4SLinus Torvalds struct buffer_head *buffers_to_free = NULL; 289368189fefSMatthew Wilcox (Oracle) bool ret = 0; 28941da177e4SLinus Torvalds 289568189fefSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 289668189fefSMatthew Wilcox (Oracle) if (folio_test_writeback(folio)) 289768189fefSMatthew Wilcox (Oracle) return false; 28981da177e4SLinus Torvalds 28991da177e4SLinus Torvalds if (mapping == NULL) { /* can this still happen? */ 290064394763SMatthew Wilcox (Oracle) ret = drop_buffers(folio, &buffers_to_free); 29011da177e4SLinus Torvalds goto out; 29021da177e4SLinus Torvalds } 29031da177e4SLinus Torvalds 29041da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 290564394763SMatthew Wilcox (Oracle) ret = drop_buffers(folio, &buffers_to_free); 2906ecdfc978SLinus Torvalds 2907ecdfc978SLinus Torvalds /* 2908ecdfc978SLinus Torvalds * If the filesystem writes its buffers by hand (eg ext3) 290968189fefSMatthew Wilcox (Oracle) * then we can have clean buffers against a dirty folio. We 291068189fefSMatthew Wilcox (Oracle) * clean the folio here; otherwise the VM will never notice 2911ecdfc978SLinus Torvalds * that the filesystem did any IO at all. 2912ecdfc978SLinus Torvalds * 2913ecdfc978SLinus Torvalds * Also, during truncate, discard_buffer will have marked all 291468189fefSMatthew Wilcox (Oracle) * the folio's buffers clean. We discover that here and clean 291568189fefSMatthew Wilcox (Oracle) * the folio also. 291687df7241SNick Piggin * 291787df7241SNick Piggin * private_lock must be held over this entire operation in order 2918e621900aSMatthew Wilcox (Oracle) * to synchronise against block_dirty_folio and prevent the 291987df7241SNick Piggin * dirty bit from being lost. 2920ecdfc978SLinus Torvalds */ 292111f81becSTejun Heo if (ret) 292268189fefSMatthew Wilcox (Oracle) folio_cancel_dirty(folio); 292387df7241SNick Piggin spin_unlock(&mapping->private_lock); 29241da177e4SLinus Torvalds out: 29251da177e4SLinus Torvalds if (buffers_to_free) { 29261da177e4SLinus Torvalds struct buffer_head *bh = buffers_to_free; 29271da177e4SLinus Torvalds 29281da177e4SLinus Torvalds do { 29291da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 29301da177e4SLinus Torvalds free_buffer_head(bh); 29311da177e4SLinus Torvalds bh = next; 29321da177e4SLinus Torvalds } while (bh != buffers_to_free); 29331da177e4SLinus Torvalds } 29341da177e4SLinus Torvalds return ret; 29351da177e4SLinus Torvalds } 29361da177e4SLinus Torvalds EXPORT_SYMBOL(try_to_free_buffers); 29371da177e4SLinus Torvalds 29381da177e4SLinus Torvalds /* 29391da177e4SLinus Torvalds * Buffer-head allocation 29401da177e4SLinus Torvalds */ 2941a0a9b043SShai Fultheim static struct kmem_cache *bh_cachep __read_mostly; 29421da177e4SLinus Torvalds 29431da177e4SLinus Torvalds /* 29441da177e4SLinus Torvalds * Once the number of bh's in the machine exceeds this level, we start 29451da177e4SLinus Torvalds * stripping them in writeback. 29461da177e4SLinus Torvalds */ 294743be594aSZhang Yanfei static unsigned long max_buffer_heads; 29481da177e4SLinus Torvalds 29491da177e4SLinus Torvalds int buffer_heads_over_limit; 29501da177e4SLinus Torvalds 29511da177e4SLinus Torvalds struct bh_accounting { 29521da177e4SLinus Torvalds int nr; /* Number of live bh's */ 29531da177e4SLinus Torvalds int ratelimit; /* Limit cacheline bouncing */ 29541da177e4SLinus Torvalds }; 29551da177e4SLinus Torvalds 29561da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; 29571da177e4SLinus Torvalds 29581da177e4SLinus Torvalds static void recalc_bh_state(void) 29591da177e4SLinus Torvalds { 29601da177e4SLinus Torvalds int i; 29611da177e4SLinus Torvalds int tot = 0; 29621da177e4SLinus Torvalds 2963ee1be862SChristoph Lameter if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) 29641da177e4SLinus Torvalds return; 2965c7b92516SChristoph Lameter __this_cpu_write(bh_accounting.ratelimit, 0); 29668a143426SEric Dumazet for_each_online_cpu(i) 29671da177e4SLinus Torvalds tot += per_cpu(bh_accounting, i).nr; 29681da177e4SLinus Torvalds buffer_heads_over_limit = (tot > max_buffer_heads); 29691da177e4SLinus Torvalds } 29701da177e4SLinus Torvalds 2971dd0fc66fSAl Viro struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) 29721da177e4SLinus Torvalds { 2973019b4d12SRichard Kennedy struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); 29741da177e4SLinus Torvalds if (ret) { 2975a35afb83SChristoph Lameter INIT_LIST_HEAD(&ret->b_assoc_buffers); 2976f1e67e35SThomas Gleixner spin_lock_init(&ret->b_uptodate_lock); 2977c7b92516SChristoph Lameter preempt_disable(); 2978c7b92516SChristoph Lameter __this_cpu_inc(bh_accounting.nr); 29791da177e4SLinus Torvalds recalc_bh_state(); 2980c7b92516SChristoph Lameter preempt_enable(); 29811da177e4SLinus Torvalds } 29821da177e4SLinus Torvalds return ret; 29831da177e4SLinus Torvalds } 29841da177e4SLinus Torvalds EXPORT_SYMBOL(alloc_buffer_head); 29851da177e4SLinus Torvalds 29861da177e4SLinus Torvalds void free_buffer_head(struct buffer_head *bh) 29871da177e4SLinus Torvalds { 29881da177e4SLinus Torvalds BUG_ON(!list_empty(&bh->b_assoc_buffers)); 29891da177e4SLinus Torvalds kmem_cache_free(bh_cachep, bh); 2990c7b92516SChristoph Lameter preempt_disable(); 2991c7b92516SChristoph Lameter __this_cpu_dec(bh_accounting.nr); 29921da177e4SLinus Torvalds recalc_bh_state(); 2993c7b92516SChristoph Lameter preempt_enable(); 29941da177e4SLinus Torvalds } 29951da177e4SLinus Torvalds EXPORT_SYMBOL(free_buffer_head); 29961da177e4SLinus Torvalds 2997fc4d24c9SSebastian Andrzej Siewior static int buffer_exit_cpu_dead(unsigned int cpu) 29981da177e4SLinus Torvalds { 29991da177e4SLinus Torvalds int i; 30001da177e4SLinus Torvalds struct bh_lru *b = &per_cpu(bh_lrus, cpu); 30011da177e4SLinus Torvalds 30021da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 30031da177e4SLinus Torvalds brelse(b->bhs[i]); 30041da177e4SLinus Torvalds b->bhs[i] = NULL; 30051da177e4SLinus Torvalds } 3006c7b92516SChristoph Lameter this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); 30078a143426SEric Dumazet per_cpu(bh_accounting, cpu).nr = 0; 3008fc4d24c9SSebastian Andrzej Siewior return 0; 30091da177e4SLinus Torvalds } 30101da177e4SLinus Torvalds 3011389d1b08SAneesh Kumar K.V /** 3012a6b91919SRandy Dunlap * bh_uptodate_or_lock - Test whether the buffer is uptodate 3013389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3014389d1b08SAneesh Kumar K.V * 3015389d1b08SAneesh Kumar K.V * Return true if the buffer is up-to-date and false, 3016389d1b08SAneesh Kumar K.V * with the buffer locked, if not. 3017389d1b08SAneesh Kumar K.V */ 3018389d1b08SAneesh Kumar K.V int bh_uptodate_or_lock(struct buffer_head *bh) 3019389d1b08SAneesh Kumar K.V { 3020389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) { 3021389d1b08SAneesh Kumar K.V lock_buffer(bh); 3022389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) 3023389d1b08SAneesh Kumar K.V return 0; 3024389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3025389d1b08SAneesh Kumar K.V } 3026389d1b08SAneesh Kumar K.V return 1; 3027389d1b08SAneesh Kumar K.V } 3028389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_uptodate_or_lock); 3029389d1b08SAneesh Kumar K.V 3030389d1b08SAneesh Kumar K.V /** 3031fdee117eSZhang Yi * __bh_read - Submit read for a locked buffer 3032389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3033fdee117eSZhang Yi * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ 3034fdee117eSZhang Yi * @wait: wait until reading finish 3035389d1b08SAneesh Kumar K.V * 3036fdee117eSZhang Yi * Returns zero on success or don't wait, and -EIO on error. 3037389d1b08SAneesh Kumar K.V */ 3038fdee117eSZhang Yi int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait) 3039389d1b08SAneesh Kumar K.V { 3040fdee117eSZhang Yi int ret = 0; 3041389d1b08SAneesh Kumar K.V 3042fdee117eSZhang Yi BUG_ON(!buffer_locked(bh)); 3043389d1b08SAneesh Kumar K.V 3044389d1b08SAneesh Kumar K.V get_bh(bh); 3045389d1b08SAneesh Kumar K.V bh->b_end_io = end_buffer_read_sync; 3046fdee117eSZhang Yi submit_bh(REQ_OP_READ | op_flags, bh); 3047fdee117eSZhang Yi if (wait) { 3048389d1b08SAneesh Kumar K.V wait_on_buffer(bh); 3049fdee117eSZhang Yi if (!buffer_uptodate(bh)) 3050fdee117eSZhang Yi ret = -EIO; 3051389d1b08SAneesh Kumar K.V } 3052fdee117eSZhang Yi return ret; 3053fdee117eSZhang Yi } 3054fdee117eSZhang Yi EXPORT_SYMBOL(__bh_read); 3055fdee117eSZhang Yi 3056fdee117eSZhang Yi /** 3057fdee117eSZhang Yi * __bh_read_batch - Submit read for a batch of unlocked buffers 3058fdee117eSZhang Yi * @nr: entry number of the buffer batch 3059fdee117eSZhang Yi * @bhs: a batch of struct buffer_head 3060fdee117eSZhang Yi * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ 3061fdee117eSZhang Yi * @force_lock: force to get a lock on the buffer if set, otherwise drops any 3062fdee117eSZhang Yi * buffer that cannot lock. 3063fdee117eSZhang Yi * 3064fdee117eSZhang Yi * Returns zero on success or don't wait, and -EIO on error. 3065fdee117eSZhang Yi */ 3066fdee117eSZhang Yi void __bh_read_batch(int nr, struct buffer_head *bhs[], 3067fdee117eSZhang Yi blk_opf_t op_flags, bool force_lock) 3068fdee117eSZhang Yi { 3069fdee117eSZhang Yi int i; 3070fdee117eSZhang Yi 3071fdee117eSZhang Yi for (i = 0; i < nr; i++) { 3072fdee117eSZhang Yi struct buffer_head *bh = bhs[i]; 3073fdee117eSZhang Yi 3074fdee117eSZhang Yi if (buffer_uptodate(bh)) 3075fdee117eSZhang Yi continue; 3076fdee117eSZhang Yi 3077fdee117eSZhang Yi if (force_lock) 3078fdee117eSZhang Yi lock_buffer(bh); 3079fdee117eSZhang Yi else 3080fdee117eSZhang Yi if (!trylock_buffer(bh)) 3081fdee117eSZhang Yi continue; 3082fdee117eSZhang Yi 3083fdee117eSZhang Yi if (buffer_uptodate(bh)) { 3084fdee117eSZhang Yi unlock_buffer(bh); 3085fdee117eSZhang Yi continue; 3086fdee117eSZhang Yi } 3087fdee117eSZhang Yi 3088fdee117eSZhang Yi bh->b_end_io = end_buffer_read_sync; 3089fdee117eSZhang Yi get_bh(bh); 3090fdee117eSZhang Yi submit_bh(REQ_OP_READ | op_flags, bh); 3091fdee117eSZhang Yi } 3092fdee117eSZhang Yi } 3093fdee117eSZhang Yi EXPORT_SYMBOL(__bh_read_batch); 3094389d1b08SAneesh Kumar K.V 30951da177e4SLinus Torvalds void __init buffer_init(void) 30961da177e4SLinus Torvalds { 309743be594aSZhang Yanfei unsigned long nrpages; 3098fc4d24c9SSebastian Andrzej Siewior int ret; 30991da177e4SLinus Torvalds 3100b98938c3SChristoph Lameter bh_cachep = kmem_cache_create("buffer_head", 3101b98938c3SChristoph Lameter sizeof(struct buffer_head), 0, 3102b98938c3SChristoph Lameter (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| 3103b98938c3SChristoph Lameter SLAB_MEM_SPREAD), 3104019b4d12SRichard Kennedy NULL); 31051da177e4SLinus Torvalds 31061da177e4SLinus Torvalds /* 31071da177e4SLinus Torvalds * Limit the bh occupancy to 10% of ZONE_NORMAL 31081da177e4SLinus Torvalds */ 31091da177e4SLinus Torvalds nrpages = (nr_free_buffer_pages() * 10) / 100; 31101da177e4SLinus Torvalds max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); 3111fc4d24c9SSebastian Andrzej Siewior ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", 3112fc4d24c9SSebastian Andrzej Siewior NULL, buffer_exit_cpu_dead); 3113fc4d24c9SSebastian Andrzej Siewior WARN_ON(ret < 0); 31141da177e4SLinus Torvalds } 3115