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> 528a237adfSMarcelo Tosatti #include <linux/sched/isolation.h> 531da177e4SLinus Torvalds 542b211dc0SBen Dooks #include "internal.h" 552b211dc0SBen Dooks 561da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); 575bdf402aSRitesh Harjani (IBM) static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh, 581420c4a5SBart Van Assche struct writeback_control *wbc); 591da177e4SLinus Torvalds 601da177e4SLinus Torvalds #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) 611da177e4SLinus Torvalds 62f0059afdSTejun Heo inline void touch_buffer(struct buffer_head *bh) 63f0059afdSTejun Heo { 645305cb83STejun Heo trace_block_touch_buffer(bh); 6503c5f331SMatthew Wilcox (Oracle) folio_mark_accessed(bh->b_folio); 66f0059afdSTejun Heo } 67f0059afdSTejun Heo EXPORT_SYMBOL(touch_buffer); 68f0059afdSTejun Heo 69fc9b52cdSHarvey Harrison void __lock_buffer(struct buffer_head *bh) 701da177e4SLinus Torvalds { 7174316201SNeilBrown wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); 721da177e4SLinus Torvalds } 731da177e4SLinus Torvalds EXPORT_SYMBOL(__lock_buffer); 741da177e4SLinus Torvalds 75fc9b52cdSHarvey Harrison void unlock_buffer(struct buffer_head *bh) 761da177e4SLinus Torvalds { 7751b07fc3SNick Piggin clear_bit_unlock(BH_Lock, &bh->b_state); 784e857c58SPeter Zijlstra smp_mb__after_atomic(); 791da177e4SLinus Torvalds wake_up_bit(&bh->b_state, BH_Lock); 801da177e4SLinus Torvalds } 811fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(unlock_buffer); 821da177e4SLinus Torvalds 831da177e4SLinus Torvalds /* 84520f301cSMatthew Wilcox (Oracle) * Returns if the folio has dirty or writeback buffers. If all the buffers 85520f301cSMatthew Wilcox (Oracle) * are unlocked and clean then the folio_test_dirty information is stale. If 86520f301cSMatthew Wilcox (Oracle) * any of the buffers are locked, it is assumed they are locked for IO. 87b4597226SMel Gorman */ 88520f301cSMatthew Wilcox (Oracle) void buffer_check_dirty_writeback(struct folio *folio, 89b4597226SMel Gorman bool *dirty, bool *writeback) 90b4597226SMel Gorman { 91b4597226SMel Gorman struct buffer_head *head, *bh; 92b4597226SMel Gorman *dirty = false; 93b4597226SMel Gorman *writeback = false; 94b4597226SMel Gorman 95520f301cSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 96b4597226SMel Gorman 97520f301cSMatthew Wilcox (Oracle) head = folio_buffers(folio); 98520f301cSMatthew Wilcox (Oracle) if (!head) 99b4597226SMel Gorman return; 100b4597226SMel Gorman 101520f301cSMatthew Wilcox (Oracle) if (folio_test_writeback(folio)) 102b4597226SMel Gorman *writeback = true; 103b4597226SMel Gorman 104b4597226SMel Gorman bh = head; 105b4597226SMel Gorman do { 106b4597226SMel Gorman if (buffer_locked(bh)) 107b4597226SMel Gorman *writeback = true; 108b4597226SMel Gorman 109b4597226SMel Gorman if (buffer_dirty(bh)) 110b4597226SMel Gorman *dirty = true; 111b4597226SMel Gorman 112b4597226SMel Gorman bh = bh->b_this_page; 113b4597226SMel Gorman } while (bh != head); 114b4597226SMel Gorman } 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; 198eee25182SMatthew Wilcox (Oracle) struct folio *folio; 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); 203eee25182SMatthew Wilcox (Oracle) folio = __filemap_get_folio(bd_mapping, index, FGP_ACCESSED, 0); 204eee25182SMatthew Wilcox (Oracle) if (IS_ERR(folio)) 2051da177e4SLinus Torvalds goto out; 2061da177e4SLinus Torvalds 2071da177e4SLinus Torvalds spin_lock(&bd_mapping->private_lock); 208eee25182SMatthew Wilcox (Oracle) head = folio_buffers(folio); 209eee25182SMatthew Wilcox (Oracle) if (!head) 2101da177e4SLinus Torvalds goto out_unlock; 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); 240eee25182SMatthew Wilcox (Oracle) folio_put(folio); 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 2856ba924d3SMatthew Wilcox (Oracle) folio_end_read(folio, folio_uptodate); 2861da177e4SLinus Torvalds return; 2871da177e4SLinus Torvalds 2881da177e4SLinus Torvalds still_busy: 289f1e67e35SThomas Gleixner spin_unlock_irqrestore(&first->b_uptodate_lock, flags); 2901da177e4SLinus Torvalds return; 2911da177e4SLinus Torvalds } 2921da177e4SLinus Torvalds 2934fa512ceSEric Biggers struct postprocess_bh_ctx { 29431fb992cSEric Biggers struct work_struct work; 29531fb992cSEric Biggers struct buffer_head *bh; 29631fb992cSEric Biggers }; 29731fb992cSEric Biggers 2984fa512ceSEric Biggers static void verify_bh(struct work_struct *work) 2994fa512ceSEric Biggers { 3004fa512ceSEric Biggers struct postprocess_bh_ctx *ctx = 3014fa512ceSEric Biggers container_of(work, struct postprocess_bh_ctx, work); 3024fa512ceSEric Biggers struct buffer_head *bh = ctx->bh; 3034fa512ceSEric Biggers bool valid; 3044fa512ceSEric Biggers 3058b7d3fe9SEric Biggers valid = fsverity_verify_blocks(bh->b_folio, bh->b_size, bh_offset(bh)); 3064fa512ceSEric Biggers end_buffer_async_read(bh, valid); 3074fa512ceSEric Biggers kfree(ctx); 3084fa512ceSEric Biggers } 3094fa512ceSEric Biggers 3104fa512ceSEric Biggers static bool need_fsverity(struct buffer_head *bh) 3114fa512ceSEric Biggers { 3128b7d3fe9SEric Biggers struct folio *folio = bh->b_folio; 3138b7d3fe9SEric Biggers struct inode *inode = folio->mapping->host; 3144fa512ceSEric Biggers 3154fa512ceSEric Biggers return fsverity_active(inode) && 3164fa512ceSEric Biggers /* needed by ext4 */ 3178b7d3fe9SEric Biggers folio->index < DIV_ROUND_UP(inode->i_size, PAGE_SIZE); 3184fa512ceSEric Biggers } 3194fa512ceSEric Biggers 32031fb992cSEric Biggers static void decrypt_bh(struct work_struct *work) 32131fb992cSEric Biggers { 3224fa512ceSEric Biggers struct postprocess_bh_ctx *ctx = 3234fa512ceSEric Biggers container_of(work, struct postprocess_bh_ctx, work); 32431fb992cSEric Biggers struct buffer_head *bh = ctx->bh; 32531fb992cSEric Biggers int err; 32631fb992cSEric Biggers 3279c7fb7f7SEric Biggers err = fscrypt_decrypt_pagecache_blocks(bh->b_folio, bh->b_size, 3289c7fb7f7SEric Biggers bh_offset(bh)); 3294fa512ceSEric Biggers if (err == 0 && need_fsverity(bh)) { 3304fa512ceSEric Biggers /* 3314fa512ceSEric Biggers * We use different work queues for decryption and for verity 3324fa512ceSEric Biggers * because verity may require reading metadata pages that need 3334fa512ceSEric Biggers * decryption, and we shouldn't recurse to the same workqueue. 3344fa512ceSEric Biggers */ 3354fa512ceSEric Biggers INIT_WORK(&ctx->work, verify_bh); 3364fa512ceSEric Biggers fsverity_enqueue_verify_work(&ctx->work); 3374fa512ceSEric Biggers return; 3384fa512ceSEric Biggers } 33931fb992cSEric Biggers end_buffer_async_read(bh, err == 0); 34031fb992cSEric Biggers kfree(ctx); 34131fb992cSEric Biggers } 34231fb992cSEric Biggers 34331fb992cSEric Biggers /* 3442c69e205SMatthew Wilcox (Oracle) * I/O completion handler for block_read_full_folio() - pages 34531fb992cSEric Biggers * which come unlocked at the end of I/O. 34631fb992cSEric Biggers */ 34731fb992cSEric Biggers static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate) 34831fb992cSEric Biggers { 3493822a7c4SLinus Torvalds struct inode *inode = bh->b_folio->mapping->host; 3504fa512ceSEric Biggers bool decrypt = fscrypt_inode_uses_fs_layer_crypto(inode); 3514fa512ceSEric Biggers bool verify = need_fsverity(bh); 3524fa512ceSEric Biggers 3534fa512ceSEric Biggers /* Decrypt (with fscrypt) and/or verify (with fsverity) if needed. */ 3544fa512ceSEric Biggers if (uptodate && (decrypt || verify)) { 3554fa512ceSEric Biggers struct postprocess_bh_ctx *ctx = 3564fa512ceSEric Biggers kmalloc(sizeof(*ctx), GFP_ATOMIC); 35731fb992cSEric Biggers 35831fb992cSEric Biggers if (ctx) { 35931fb992cSEric Biggers ctx->bh = bh; 3604fa512ceSEric Biggers if (decrypt) { 3614fa512ceSEric Biggers INIT_WORK(&ctx->work, decrypt_bh); 36231fb992cSEric Biggers fscrypt_enqueue_decrypt_work(&ctx->work); 3634fa512ceSEric Biggers } else { 3644fa512ceSEric Biggers INIT_WORK(&ctx->work, verify_bh); 3654fa512ceSEric Biggers fsverity_enqueue_verify_work(&ctx->work); 3664fa512ceSEric Biggers } 36731fb992cSEric Biggers return; 36831fb992cSEric Biggers } 36931fb992cSEric Biggers uptodate = 0; 37031fb992cSEric Biggers } 37131fb992cSEric Biggers end_buffer_async_read(bh, uptodate); 37231fb992cSEric Biggers } 37331fb992cSEric Biggers 3741da177e4SLinus Torvalds /* 3751da177e4SLinus Torvalds * Completion handler for block_write_full_page() - pages which are unlocked 3761da177e4SLinus Torvalds * during I/O, and which have PageWriteback cleared upon I/O completion. 3771da177e4SLinus Torvalds */ 37835c80d5fSChris Mason void end_buffer_async_write(struct buffer_head *bh, int uptodate) 3791da177e4SLinus Torvalds { 3801da177e4SLinus Torvalds unsigned long flags; 381a3972203SNick Piggin struct buffer_head *first; 3821da177e4SLinus Torvalds struct buffer_head *tmp; 383743ed81eSMatthew Wilcox (Oracle) struct folio *folio; 3841da177e4SLinus Torvalds 3851da177e4SLinus Torvalds BUG_ON(!buffer_async_write(bh)); 3861da177e4SLinus Torvalds 387743ed81eSMatthew Wilcox (Oracle) folio = bh->b_folio; 3881da177e4SLinus Torvalds if (uptodate) { 3891da177e4SLinus Torvalds set_buffer_uptodate(bh); 3901da177e4SLinus Torvalds } else { 391b744c2acSRobert Elliott buffer_io_error(bh, ", lost async page write"); 39287354e5dSJeff Layton mark_buffer_write_io_error(bh); 3931da177e4SLinus Torvalds clear_buffer_uptodate(bh); 394743ed81eSMatthew Wilcox (Oracle) folio_set_error(folio); 3951da177e4SLinus Torvalds } 3961da177e4SLinus Torvalds 397743ed81eSMatthew Wilcox (Oracle) first = folio_buffers(folio); 398f1e67e35SThomas Gleixner spin_lock_irqsave(&first->b_uptodate_lock, flags); 399a3972203SNick Piggin 4001da177e4SLinus Torvalds clear_buffer_async_write(bh); 4011da177e4SLinus Torvalds unlock_buffer(bh); 4021da177e4SLinus Torvalds tmp = bh->b_this_page; 4031da177e4SLinus Torvalds while (tmp != bh) { 4041da177e4SLinus Torvalds if (buffer_async_write(tmp)) { 4051da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 4061da177e4SLinus Torvalds goto still_busy; 4071da177e4SLinus Torvalds } 4081da177e4SLinus Torvalds tmp = tmp->b_this_page; 4091da177e4SLinus Torvalds } 410f1e67e35SThomas Gleixner spin_unlock_irqrestore(&first->b_uptodate_lock, flags); 411743ed81eSMatthew Wilcox (Oracle) folio_end_writeback(folio); 4121da177e4SLinus Torvalds return; 4131da177e4SLinus Torvalds 4141da177e4SLinus Torvalds still_busy: 415f1e67e35SThomas Gleixner spin_unlock_irqrestore(&first->b_uptodate_lock, flags); 4161da177e4SLinus Torvalds return; 4171da177e4SLinus Torvalds } 4181fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(end_buffer_async_write); 4191da177e4SLinus Torvalds 4201da177e4SLinus Torvalds /* 4211da177e4SLinus Torvalds * If a page's buffers are under async readin (end_buffer_async_read 4221da177e4SLinus Torvalds * completion) then there is a possibility that another thread of 4231da177e4SLinus Torvalds * control could lock one of the buffers after it has completed 4241da177e4SLinus Torvalds * but while some of the other buffers have not completed. This 4251da177e4SLinus Torvalds * locked buffer would confuse end_buffer_async_read() into not unlocking 4261da177e4SLinus Torvalds * the page. So the absence of BH_Async_Read tells end_buffer_async_read() 4271da177e4SLinus Torvalds * that this buffer is not under async I/O. 4281da177e4SLinus Torvalds * 4291da177e4SLinus Torvalds * The page comes unlocked when it has no locked buffer_async buffers 4301da177e4SLinus Torvalds * left. 4311da177e4SLinus Torvalds * 4321da177e4SLinus Torvalds * PageLocked prevents anyone starting new async I/O reads any of 4331da177e4SLinus Torvalds * the buffers. 4341da177e4SLinus Torvalds * 4351da177e4SLinus Torvalds * PageWriteback is used to prevent simultaneous writeout of the same 4361da177e4SLinus Torvalds * page. 4371da177e4SLinus Torvalds * 4381da177e4SLinus Torvalds * PageLocked prevents anyone from starting writeback of a page which is 4391da177e4SLinus Torvalds * under read I/O (PageWriteback is only ever set against a locked page). 4401da177e4SLinus Torvalds */ 4411da177e4SLinus Torvalds static void mark_buffer_async_read(struct buffer_head *bh) 4421da177e4SLinus Torvalds { 44331fb992cSEric Biggers bh->b_end_io = end_buffer_async_read_io; 4441da177e4SLinus Torvalds set_buffer_async_read(bh); 4451da177e4SLinus Torvalds } 4461da177e4SLinus Torvalds 4471fe72eaaSH Hartley Sweeten static void mark_buffer_async_write_endio(struct buffer_head *bh, 44835c80d5fSChris Mason bh_end_io_t *handler) 44935c80d5fSChris Mason { 45035c80d5fSChris Mason bh->b_end_io = handler; 45135c80d5fSChris Mason set_buffer_async_write(bh); 45235c80d5fSChris Mason } 45335c80d5fSChris Mason 4541da177e4SLinus Torvalds void mark_buffer_async_write(struct buffer_head *bh) 4551da177e4SLinus Torvalds { 45635c80d5fSChris Mason mark_buffer_async_write_endio(bh, end_buffer_async_write); 4571da177e4SLinus Torvalds } 4581da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_async_write); 4591da177e4SLinus Torvalds 4601da177e4SLinus Torvalds 4611da177e4SLinus Torvalds /* 4621da177e4SLinus Torvalds * fs/buffer.c contains helper functions for buffer-backed address space's 4631da177e4SLinus Torvalds * fsync functions. A common requirement for buffer-based filesystems is 4641da177e4SLinus Torvalds * that certain data from the backing blockdev needs to be written out for 4651da177e4SLinus Torvalds * a successful fsync(). For example, ext2 indirect blocks need to be 4661da177e4SLinus Torvalds * written back and waited upon before fsync() returns. 4671da177e4SLinus Torvalds * 4681da177e4SLinus Torvalds * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), 4691da177e4SLinus Torvalds * inode_has_buffers() and invalidate_inode_buffers() are provided for the 4701da177e4SLinus Torvalds * management of a list of dependent buffers at ->i_mapping->private_list. 4711da177e4SLinus Torvalds * 4721da177e4SLinus Torvalds * Locking is a little subtle: try_to_free_buffers() will remove buffers 4731da177e4SLinus Torvalds * from their controlling inode's queue when they are being freed. But 4741da177e4SLinus Torvalds * try_to_free_buffers() will be operating against the *blockdev* mapping 4751da177e4SLinus Torvalds * at the time, not against the S_ISREG file which depends on those buffers. 4761da177e4SLinus Torvalds * So the locking for private_list is via the private_lock in the address_space 4771da177e4SLinus Torvalds * which backs the buffers. Which is different from the address_space 4781da177e4SLinus Torvalds * against which the buffers are listed. So for a particular address_space, 4791da177e4SLinus Torvalds * mapping->private_lock does *not* protect mapping->private_list! In fact, 4801da177e4SLinus Torvalds * mapping->private_list will always be protected by the backing blockdev's 4811da177e4SLinus Torvalds * ->private_lock. 4821da177e4SLinus Torvalds * 4831da177e4SLinus Torvalds * Which introduces a requirement: all buffers on an address_space's 4841da177e4SLinus Torvalds * ->private_list must be from the same address_space: the blockdev's. 4851da177e4SLinus Torvalds * 4861da177e4SLinus Torvalds * address_spaces which do not place buffers at ->private_list via these 4871da177e4SLinus Torvalds * utility functions are free to use private_lock and private_list for 4881da177e4SLinus Torvalds * whatever they want. The only requirement is that list_empty(private_list) 4891da177e4SLinus Torvalds * be true at clear_inode() time. 4901da177e4SLinus Torvalds * 4911da177e4SLinus Torvalds * FIXME: clear_inode should not call invalidate_inode_buffers(). The 4921da177e4SLinus Torvalds * filesystems should do that. invalidate_inode_buffers() should just go 4931da177e4SLinus Torvalds * BUG_ON(!list_empty). 4941da177e4SLinus Torvalds * 4951da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should 4961da177e4SLinus Torvalds * take an address_space, not an inode. And it should be called 4971da177e4SLinus Torvalds * mark_buffer_dirty_fsync() to clearly define why those buffers are being 4981da177e4SLinus Torvalds * queued up. 4991da177e4SLinus Torvalds * 5001da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the 5011da177e4SLinus Torvalds * list if it is already on a list. Because if the buffer is on a list, 5021da177e4SLinus Torvalds * it *must* already be on the right one. If not, the filesystem is being 5031da177e4SLinus Torvalds * silly. This will save a ton of locking. But first we have to ensure 5041da177e4SLinus Torvalds * that buffers are taken *off* the old inode's list when they are freed 5051da177e4SLinus Torvalds * (presumably in truncate). That requires careful auditing of all 5061da177e4SLinus Torvalds * filesystems (do it inside bforget()). It could also be done by bringing 5071da177e4SLinus Torvalds * b_inode back. 5081da177e4SLinus Torvalds */ 5091da177e4SLinus Torvalds 5101da177e4SLinus Torvalds /* 5111da177e4SLinus Torvalds * The buffer's backing address_space's private_lock must be held 5121da177e4SLinus Torvalds */ 513dbacefc9SThomas Petazzoni static void __remove_assoc_queue(struct buffer_head *bh) 5141da177e4SLinus Torvalds { 5151da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 51658ff407bSJan Kara WARN_ON(!bh->b_assoc_map); 51758ff407bSJan Kara bh->b_assoc_map = NULL; 5181da177e4SLinus Torvalds } 5191da177e4SLinus Torvalds 5201da177e4SLinus Torvalds int inode_has_buffers(struct inode *inode) 5211da177e4SLinus Torvalds { 5221da177e4SLinus Torvalds return !list_empty(&inode->i_data.private_list); 5231da177e4SLinus Torvalds } 5241da177e4SLinus Torvalds 5251da177e4SLinus Torvalds /* 5261da177e4SLinus Torvalds * osync is designed to support O_SYNC io. It waits synchronously for 5271da177e4SLinus Torvalds * all already-submitted IO to complete, but does not queue any new 5281da177e4SLinus Torvalds * writes to the disk. 5291da177e4SLinus Torvalds * 53079f59784SZhang Yi * To do O_SYNC writes, just queue the buffer writes with write_dirty_buffer 53179f59784SZhang Yi * as you dirty the buffers, and then use osync_inode_buffers to wait for 5321da177e4SLinus Torvalds * completion. Any other dirty buffers which are not yet queued for 5331da177e4SLinus Torvalds * write will not be flushed to disk by the osync. 5341da177e4SLinus Torvalds */ 5351da177e4SLinus Torvalds static int osync_buffers_list(spinlock_t *lock, struct list_head *list) 5361da177e4SLinus Torvalds { 5371da177e4SLinus Torvalds struct buffer_head *bh; 5381da177e4SLinus Torvalds struct list_head *p; 5391da177e4SLinus Torvalds int err = 0; 5401da177e4SLinus Torvalds 5411da177e4SLinus Torvalds spin_lock(lock); 5421da177e4SLinus Torvalds repeat: 5431da177e4SLinus Torvalds list_for_each_prev(p, list) { 5441da177e4SLinus Torvalds bh = BH_ENTRY(p); 5451da177e4SLinus Torvalds if (buffer_locked(bh)) { 5461da177e4SLinus Torvalds get_bh(bh); 5471da177e4SLinus Torvalds spin_unlock(lock); 5481da177e4SLinus Torvalds wait_on_buffer(bh); 5491da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 5501da177e4SLinus Torvalds err = -EIO; 5511da177e4SLinus Torvalds brelse(bh); 5521da177e4SLinus Torvalds spin_lock(lock); 5531da177e4SLinus Torvalds goto repeat; 5541da177e4SLinus Torvalds } 5551da177e4SLinus Torvalds } 5561da177e4SLinus Torvalds spin_unlock(lock); 5571da177e4SLinus Torvalds return err; 5581da177e4SLinus Torvalds } 5591da177e4SLinus Torvalds 5601da177e4SLinus Torvalds /** 56178a4a50aSRandy Dunlap * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers 56267be2dd1SMartin Waitz * @mapping: the mapping which wants those buffers written 5631da177e4SLinus Torvalds * 5641da177e4SLinus Torvalds * Starts I/O against the buffers at mapping->private_list, and waits upon 5651da177e4SLinus Torvalds * that I/O. 5661da177e4SLinus Torvalds * 56767be2dd1SMartin Waitz * Basically, this is a convenience function for fsync(). 56867be2dd1SMartin Waitz * @mapping is a file or directory which needs those buffers to be written for 56967be2dd1SMartin Waitz * a successful fsync(). 5701da177e4SLinus Torvalds */ 5711da177e4SLinus Torvalds int sync_mapping_buffers(struct address_space *mapping) 5721da177e4SLinus Torvalds { 573252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 5741da177e4SLinus Torvalds 5751da177e4SLinus Torvalds if (buffer_mapping == NULL || list_empty(&mapping->private_list)) 5761da177e4SLinus Torvalds return 0; 5771da177e4SLinus Torvalds 5781da177e4SLinus Torvalds return fsync_buffers_list(&buffer_mapping->private_lock, 5791da177e4SLinus Torvalds &mapping->private_list); 5801da177e4SLinus Torvalds } 5811da177e4SLinus Torvalds EXPORT_SYMBOL(sync_mapping_buffers); 5821da177e4SLinus Torvalds 58331b2ebc0SRitesh Harjani (IBM) /** 58431b2ebc0SRitesh Harjani (IBM) * generic_buffers_fsync_noflush - generic buffer fsync implementation 58531b2ebc0SRitesh Harjani (IBM) * for simple filesystems with no inode lock 58631b2ebc0SRitesh Harjani (IBM) * 58731b2ebc0SRitesh Harjani (IBM) * @file: file to synchronize 58831b2ebc0SRitesh Harjani (IBM) * @start: start offset in bytes 58931b2ebc0SRitesh Harjani (IBM) * @end: end offset in bytes (inclusive) 59031b2ebc0SRitesh Harjani (IBM) * @datasync: only synchronize essential metadata if true 59131b2ebc0SRitesh Harjani (IBM) * 59231b2ebc0SRitesh Harjani (IBM) * This is a generic implementation of the fsync method for simple 59331b2ebc0SRitesh Harjani (IBM) * filesystems which track all non-inode metadata in the buffers list 59431b2ebc0SRitesh Harjani (IBM) * hanging off the address_space structure. 59531b2ebc0SRitesh Harjani (IBM) */ 59631b2ebc0SRitesh Harjani (IBM) int generic_buffers_fsync_noflush(struct file *file, loff_t start, loff_t end, 59731b2ebc0SRitesh Harjani (IBM) bool datasync) 59831b2ebc0SRitesh Harjani (IBM) { 59931b2ebc0SRitesh Harjani (IBM) struct inode *inode = file->f_mapping->host; 60031b2ebc0SRitesh Harjani (IBM) int err; 60131b2ebc0SRitesh Harjani (IBM) int ret; 60231b2ebc0SRitesh Harjani (IBM) 60331b2ebc0SRitesh Harjani (IBM) err = file_write_and_wait_range(file, start, end); 60431b2ebc0SRitesh Harjani (IBM) if (err) 60531b2ebc0SRitesh Harjani (IBM) return err; 60631b2ebc0SRitesh Harjani (IBM) 60731b2ebc0SRitesh Harjani (IBM) ret = sync_mapping_buffers(inode->i_mapping); 60831b2ebc0SRitesh Harjani (IBM) if (!(inode->i_state & I_DIRTY_ALL)) 60931b2ebc0SRitesh Harjani (IBM) goto out; 61031b2ebc0SRitesh Harjani (IBM) if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) 61131b2ebc0SRitesh Harjani (IBM) goto out; 61231b2ebc0SRitesh Harjani (IBM) 61331b2ebc0SRitesh Harjani (IBM) err = sync_inode_metadata(inode, 1); 61431b2ebc0SRitesh Harjani (IBM) if (ret == 0) 61531b2ebc0SRitesh Harjani (IBM) ret = err; 61631b2ebc0SRitesh Harjani (IBM) 61731b2ebc0SRitesh Harjani (IBM) out: 61831b2ebc0SRitesh Harjani (IBM) /* check and advance again to catch errors after syncing out buffers */ 61931b2ebc0SRitesh Harjani (IBM) err = file_check_and_advance_wb_err(file); 62031b2ebc0SRitesh Harjani (IBM) if (ret == 0) 62131b2ebc0SRitesh Harjani (IBM) ret = err; 62231b2ebc0SRitesh Harjani (IBM) return ret; 62331b2ebc0SRitesh Harjani (IBM) } 62431b2ebc0SRitesh Harjani (IBM) EXPORT_SYMBOL(generic_buffers_fsync_noflush); 62531b2ebc0SRitesh Harjani (IBM) 62631b2ebc0SRitesh Harjani (IBM) /** 62731b2ebc0SRitesh Harjani (IBM) * generic_buffers_fsync - generic buffer fsync implementation 62831b2ebc0SRitesh Harjani (IBM) * for simple filesystems with no inode lock 62931b2ebc0SRitesh Harjani (IBM) * 63031b2ebc0SRitesh Harjani (IBM) * @file: file to synchronize 63131b2ebc0SRitesh Harjani (IBM) * @start: start offset in bytes 63231b2ebc0SRitesh Harjani (IBM) * @end: end offset in bytes (inclusive) 63331b2ebc0SRitesh Harjani (IBM) * @datasync: only synchronize essential metadata if true 63431b2ebc0SRitesh Harjani (IBM) * 63531b2ebc0SRitesh Harjani (IBM) * This is a generic implementation of the fsync method for simple 63631b2ebc0SRitesh Harjani (IBM) * filesystems which track all non-inode metadata in the buffers list 63731b2ebc0SRitesh Harjani (IBM) * hanging off the address_space structure. This also makes sure that 63831b2ebc0SRitesh Harjani (IBM) * a device cache flush operation is called at the end. 63931b2ebc0SRitesh Harjani (IBM) */ 64031b2ebc0SRitesh Harjani (IBM) int generic_buffers_fsync(struct file *file, loff_t start, loff_t end, 64131b2ebc0SRitesh Harjani (IBM) bool datasync) 64231b2ebc0SRitesh Harjani (IBM) { 64331b2ebc0SRitesh Harjani (IBM) struct inode *inode = file->f_mapping->host; 64431b2ebc0SRitesh Harjani (IBM) int ret; 64531b2ebc0SRitesh Harjani (IBM) 64631b2ebc0SRitesh Harjani (IBM) ret = generic_buffers_fsync_noflush(file, start, end, datasync); 64731b2ebc0SRitesh Harjani (IBM) if (!ret) 64831b2ebc0SRitesh Harjani (IBM) ret = blkdev_issue_flush(inode->i_sb->s_bdev); 64931b2ebc0SRitesh Harjani (IBM) return ret; 65031b2ebc0SRitesh Harjani (IBM) } 65131b2ebc0SRitesh Harjani (IBM) EXPORT_SYMBOL(generic_buffers_fsync); 65231b2ebc0SRitesh Harjani (IBM) 6531da177e4SLinus Torvalds /* 6541da177e4SLinus Torvalds * Called when we've recently written block `bblock', and it is known that 6551da177e4SLinus Torvalds * `bblock' was for a buffer_boundary() buffer. This means that the block at 6561da177e4SLinus Torvalds * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's 6571da177e4SLinus Torvalds * dirty, schedule it for IO. So that indirects merge nicely with their data. 6581da177e4SLinus Torvalds */ 6591da177e4SLinus Torvalds void write_boundary_block(struct block_device *bdev, 6601da177e4SLinus Torvalds sector_t bblock, unsigned blocksize) 6611da177e4SLinus Torvalds { 6621da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); 6631da177e4SLinus Torvalds if (bh) { 6641da177e4SLinus Torvalds if (buffer_dirty(bh)) 665e7ea1129SZhang Yi write_dirty_buffer(bh, 0); 6661da177e4SLinus Torvalds put_bh(bh); 6671da177e4SLinus Torvalds } 6681da177e4SLinus Torvalds } 6691da177e4SLinus Torvalds 6701da177e4SLinus Torvalds void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) 6711da177e4SLinus Torvalds { 6721da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 673abc8a8a2SMatthew Wilcox (Oracle) struct address_space *buffer_mapping = bh->b_folio->mapping; 6741da177e4SLinus Torvalds 6751da177e4SLinus Torvalds mark_buffer_dirty(bh); 676252aa6f5SRafael Aquini if (!mapping->private_data) { 677252aa6f5SRafael Aquini mapping->private_data = buffer_mapping; 6781da177e4SLinus Torvalds } else { 679252aa6f5SRafael Aquini BUG_ON(mapping->private_data != buffer_mapping); 6801da177e4SLinus Torvalds } 681535ee2fbSJan Kara if (!bh->b_assoc_map) { 6821da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 6831da177e4SLinus Torvalds list_move_tail(&bh->b_assoc_buffers, 6841da177e4SLinus Torvalds &mapping->private_list); 68558ff407bSJan Kara bh->b_assoc_map = mapping; 6861da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 6871da177e4SLinus Torvalds } 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_dirty_inode); 6901da177e4SLinus Torvalds 6911da177e4SLinus Torvalds /* 6921da177e4SLinus Torvalds * Add a page to the dirty page list. 6931da177e4SLinus Torvalds * 6941da177e4SLinus Torvalds * It is a sad fact of life that this function is called from several places 6951da177e4SLinus Torvalds * deeply under spinlocking. It may not sleep. 6961da177e4SLinus Torvalds * 6971da177e4SLinus Torvalds * If the page has buffers, the uptodate buffers are set dirty, to preserve 6981da177e4SLinus Torvalds * dirty-state coherency between the page and the buffers. It the page does 6991da177e4SLinus Torvalds * not have buffers then when they are later attached they will all be set 7001da177e4SLinus Torvalds * dirty. 7011da177e4SLinus Torvalds * 7021da177e4SLinus Torvalds * The buffers are dirtied before the page is dirtied. There's a small race 7031da177e4SLinus Torvalds * window in which a writepage caller may see the page cleanness but not the 7041da177e4SLinus Torvalds * buffer dirtiness. That's fine. If this code were to set the page dirty 7051da177e4SLinus Torvalds * before the buffers, a concurrent writepage caller could clear the page dirty 7061da177e4SLinus Torvalds * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean 7071da177e4SLinus Torvalds * page on the dirty page list. 7081da177e4SLinus Torvalds * 7091da177e4SLinus Torvalds * We use private_lock to lock against try_to_free_buffers while using the 7101da177e4SLinus Torvalds * page's buffer list. Also use this to protect against clean buffers being 7111da177e4SLinus Torvalds * added to the page after it was set dirty. 7121da177e4SLinus Torvalds * 7131da177e4SLinus Torvalds * FIXME: may need to call ->reservepage here as well. That's rather up to the 7141da177e4SLinus Torvalds * address_space though. 7151da177e4SLinus Torvalds */ 716e621900aSMatthew Wilcox (Oracle) bool block_dirty_folio(struct address_space *mapping, struct folio *folio) 7171da177e4SLinus Torvalds { 718e621900aSMatthew Wilcox (Oracle) struct buffer_head *head; 719e621900aSMatthew Wilcox (Oracle) bool newly_dirty; 7201da177e4SLinus Torvalds 7211da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 722e621900aSMatthew Wilcox (Oracle) head = folio_buffers(folio); 723e621900aSMatthew Wilcox (Oracle) if (head) { 7241da177e4SLinus Torvalds struct buffer_head *bh = head; 7251da177e4SLinus Torvalds 7261da177e4SLinus Torvalds do { 7271da177e4SLinus Torvalds set_buffer_dirty(bh); 7281da177e4SLinus Torvalds bh = bh->b_this_page; 7291da177e4SLinus Torvalds } while (bh != head); 7301da177e4SLinus Torvalds } 731c4843a75SGreg Thelen /* 732bcfe06bfSRoman Gushchin * Lock out page's memcg migration to keep PageDirty 73381f8c3a4SJohannes Weiner * synchronized with per-memcg dirty page counters. 734c4843a75SGreg Thelen */ 735e621900aSMatthew Wilcox (Oracle) folio_memcg_lock(folio); 736e621900aSMatthew Wilcox (Oracle) newly_dirty = !folio_test_set_dirty(folio); 7371da177e4SLinus Torvalds spin_unlock(&mapping->private_lock); 7381da177e4SLinus Torvalds 739a8e7d49aSLinus Torvalds if (newly_dirty) 740e621900aSMatthew Wilcox (Oracle) __folio_mark_dirty(folio, mapping, 1); 741c4843a75SGreg Thelen 742e621900aSMatthew Wilcox (Oracle) folio_memcg_unlock(folio); 743c4843a75SGreg Thelen 744c4843a75SGreg Thelen if (newly_dirty) 745c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 746c4843a75SGreg Thelen 747a8e7d49aSLinus Torvalds return newly_dirty; 7481da177e4SLinus Torvalds } 749e621900aSMatthew Wilcox (Oracle) EXPORT_SYMBOL(block_dirty_folio); 7501da177e4SLinus Torvalds 7511da177e4SLinus Torvalds /* 7521da177e4SLinus Torvalds * Write out and wait upon a list of buffers. 7531da177e4SLinus Torvalds * 7541da177e4SLinus Torvalds * We have conflicting pressures: we want to make sure that all 7551da177e4SLinus Torvalds * initially dirty buffers get waited on, but that any subsequently 7561da177e4SLinus Torvalds * dirtied buffers don't. After all, we don't want fsync to last 7571da177e4SLinus Torvalds * forever if somebody is actively writing to the file. 7581da177e4SLinus Torvalds * 7591da177e4SLinus Torvalds * Do this in two main stages: first we copy dirty buffers to a 7601da177e4SLinus Torvalds * temporary inode list, queueing the writes as we go. Then we clean 7611da177e4SLinus Torvalds * up, waiting for those writes to complete. 7621da177e4SLinus Torvalds * 7631da177e4SLinus Torvalds * During this second stage, any subsequent updates to the file may end 7641da177e4SLinus Torvalds * up refiling the buffer on the original inode's dirty list again, so 7651da177e4SLinus Torvalds * there is a chance we will end up with a buffer queued for write but 7661da177e4SLinus Torvalds * not yet completed on that list. So, as a final cleanup we go through 7671da177e4SLinus Torvalds * the osync code to catch these locked, dirty buffers without requeuing 7681da177e4SLinus Torvalds * any newly dirty buffers for write. 7691da177e4SLinus Torvalds */ 7701da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) 7711da177e4SLinus Torvalds { 7721da177e4SLinus Torvalds struct buffer_head *bh; 7731da177e4SLinus Torvalds struct list_head tmp; 7747eaceaccSJens Axboe struct address_space *mapping; 7751da177e4SLinus Torvalds int err = 0, err2; 7764ee2491eSJens Axboe struct blk_plug plug; 7771da177e4SLinus Torvalds 7781da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 7794ee2491eSJens Axboe blk_start_plug(&plug); 7801da177e4SLinus Torvalds 7811da177e4SLinus Torvalds spin_lock(lock); 7821da177e4SLinus Torvalds while (!list_empty(list)) { 7831da177e4SLinus Torvalds bh = BH_ENTRY(list->next); 784535ee2fbSJan Kara mapping = bh->b_assoc_map; 78558ff407bSJan Kara __remove_assoc_queue(bh); 786535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 787535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 788535ee2fbSJan Kara smp_mb(); 7891da177e4SLinus Torvalds if (buffer_dirty(bh) || buffer_locked(bh)) { 7901da177e4SLinus Torvalds list_add(&bh->b_assoc_buffers, &tmp); 791535ee2fbSJan Kara bh->b_assoc_map = mapping; 7921da177e4SLinus Torvalds if (buffer_dirty(bh)) { 7931da177e4SLinus Torvalds get_bh(bh); 7941da177e4SLinus Torvalds spin_unlock(lock); 7951da177e4SLinus Torvalds /* 7961da177e4SLinus Torvalds * Ensure any pending I/O completes so that 7979cb569d6SChristoph Hellwig * write_dirty_buffer() actually writes the 7989cb569d6SChristoph Hellwig * current contents - it is a noop if I/O is 7999cb569d6SChristoph Hellwig * still in flight on potentially older 8009cb569d6SChristoph Hellwig * contents. 8011da177e4SLinus Torvalds */ 80270fd7614SChristoph Hellwig write_dirty_buffer(bh, REQ_SYNC); 8039cf6b720SJens Axboe 8049cf6b720SJens Axboe /* 8059cf6b720SJens Axboe * Kick off IO for the previous mapping. Note 8069cf6b720SJens Axboe * that we will not run the very last mapping, 8079cf6b720SJens Axboe * wait_on_buffer() will do that for us 8089cf6b720SJens Axboe * through sync_buffer(). 8099cf6b720SJens Axboe */ 8101da177e4SLinus Torvalds brelse(bh); 8111da177e4SLinus Torvalds spin_lock(lock); 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds } 8141da177e4SLinus Torvalds } 8151da177e4SLinus Torvalds 8164ee2491eSJens Axboe spin_unlock(lock); 8174ee2491eSJens Axboe blk_finish_plug(&plug); 8184ee2491eSJens Axboe spin_lock(lock); 8194ee2491eSJens Axboe 8201da177e4SLinus Torvalds while (!list_empty(&tmp)) { 8211da177e4SLinus Torvalds bh = BH_ENTRY(tmp.prev); 8221da177e4SLinus Torvalds get_bh(bh); 823535ee2fbSJan Kara mapping = bh->b_assoc_map; 824535ee2fbSJan Kara __remove_assoc_queue(bh); 825535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 826535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 827535ee2fbSJan Kara smp_mb(); 828535ee2fbSJan Kara if (buffer_dirty(bh)) { 829535ee2fbSJan Kara list_add(&bh->b_assoc_buffers, 830e3892296SJan Kara &mapping->private_list); 831535ee2fbSJan Kara bh->b_assoc_map = mapping; 832535ee2fbSJan Kara } 8331da177e4SLinus Torvalds spin_unlock(lock); 8341da177e4SLinus Torvalds wait_on_buffer(bh); 8351da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 8361da177e4SLinus Torvalds err = -EIO; 8371da177e4SLinus Torvalds brelse(bh); 8381da177e4SLinus Torvalds spin_lock(lock); 8391da177e4SLinus Torvalds } 8401da177e4SLinus Torvalds 8411da177e4SLinus Torvalds spin_unlock(lock); 8421da177e4SLinus Torvalds err2 = osync_buffers_list(lock, list); 8431da177e4SLinus Torvalds if (err) 8441da177e4SLinus Torvalds return err; 8451da177e4SLinus Torvalds else 8461da177e4SLinus Torvalds return err2; 8471da177e4SLinus Torvalds } 8481da177e4SLinus Torvalds 8491da177e4SLinus Torvalds /* 8501da177e4SLinus Torvalds * Invalidate any and all dirty buffers on a given inode. We are 8511da177e4SLinus Torvalds * probably unmounting the fs, but that doesn't mean we have already 8521da177e4SLinus Torvalds * done a sync(). Just drop the buffers from the inode list. 8531da177e4SLinus Torvalds * 8541da177e4SLinus Torvalds * NOTE: we take the inode's blockdev's mapping's private_lock. Which 8551da177e4SLinus Torvalds * assumes that all the buffers are against the blockdev. Not true 8561da177e4SLinus Torvalds * for reiserfs. 8571da177e4SLinus Torvalds */ 8581da177e4SLinus Torvalds void invalidate_inode_buffers(struct inode *inode) 8591da177e4SLinus Torvalds { 8601da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8611da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 8621da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 863252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 8641da177e4SLinus Torvalds 8651da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8661da177e4SLinus Torvalds while (!list_empty(list)) 8671da177e4SLinus Torvalds __remove_assoc_queue(BH_ENTRY(list->next)); 8681da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8691da177e4SLinus Torvalds } 8701da177e4SLinus Torvalds } 87152b19ac9SJan Kara EXPORT_SYMBOL(invalidate_inode_buffers); 8721da177e4SLinus Torvalds 8731da177e4SLinus Torvalds /* 8741da177e4SLinus Torvalds * Remove any clean buffers from the inode's buffer list. This is called 8751da177e4SLinus Torvalds * when we're trying to free the inode itself. Those buffers can pin it. 8761da177e4SLinus Torvalds * 8771da177e4SLinus Torvalds * Returns true if all buffers were removed. 8781da177e4SLinus Torvalds */ 8791da177e4SLinus Torvalds int remove_inode_buffers(struct inode *inode) 8801da177e4SLinus Torvalds { 8811da177e4SLinus Torvalds int ret = 1; 8821da177e4SLinus Torvalds 8831da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8841da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 8851da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 886252aa6f5SRafael Aquini struct address_space *buffer_mapping = mapping->private_data; 8871da177e4SLinus Torvalds 8881da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8891da177e4SLinus Torvalds while (!list_empty(list)) { 8901da177e4SLinus Torvalds struct buffer_head *bh = BH_ENTRY(list->next); 8911da177e4SLinus Torvalds if (buffer_dirty(bh)) { 8921da177e4SLinus Torvalds ret = 0; 8931da177e4SLinus Torvalds break; 8941da177e4SLinus Torvalds } 8951da177e4SLinus Torvalds __remove_assoc_queue(bh); 8961da177e4SLinus Torvalds } 8971da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8981da177e4SLinus Torvalds } 8991da177e4SLinus Torvalds return ret; 9001da177e4SLinus Torvalds } 9011da177e4SLinus Torvalds 9021da177e4SLinus Torvalds /* 903c71124a8SPankaj Raghav * Create the appropriate buffers when given a folio for data area and 9041da177e4SLinus Torvalds * the size of each buffer.. Use the bh->b_this_page linked list to 9051da177e4SLinus Torvalds * follow the buffers created. Return NULL if unable to create more 9061da177e4SLinus Torvalds * buffers. 9071da177e4SLinus Torvalds * 9081da177e4SLinus Torvalds * The retry flag is used to differentiate async IO (paging, swapping) 9091da177e4SLinus Torvalds * which may not fail from ordinary buffer allocations. 9101da177e4SLinus Torvalds */ 911c71124a8SPankaj Raghav struct buffer_head *folio_alloc_buffers(struct folio *folio, unsigned long size, 9122a418157SMatthew Wilcox (Oracle) gfp_t gfp) 9131da177e4SLinus Torvalds { 9141da177e4SLinus Torvalds struct buffer_head *bh, *head; 9151da177e4SLinus Torvalds long offset; 916b87d8cefSRoman Gushchin struct mem_cgroup *memcg, *old_memcg; 9171da177e4SLinus Torvalds 918c71124a8SPankaj Raghav /* The folio lock pins the memcg */ 919c71124a8SPankaj Raghav memcg = folio_memcg(folio); 920b87d8cefSRoman Gushchin old_memcg = set_active_memcg(memcg); 921f745c6f5SShakeel Butt 9221da177e4SLinus Torvalds head = NULL; 923c71124a8SPankaj Raghav offset = folio_size(folio); 9241da177e4SLinus Torvalds while ((offset -= size) >= 0) { 925640ab98fSJens Axboe bh = alloc_buffer_head(gfp); 9261da177e4SLinus Torvalds if (!bh) 9271da177e4SLinus Torvalds goto no_grow; 9281da177e4SLinus Torvalds 9291da177e4SLinus Torvalds bh->b_this_page = head; 9301da177e4SLinus Torvalds bh->b_blocknr = -1; 9311da177e4SLinus Torvalds head = bh; 9321da177e4SLinus Torvalds 9331da177e4SLinus Torvalds bh->b_size = size; 9341da177e4SLinus Torvalds 935c71124a8SPankaj Raghav /* Link the buffer to its folio */ 936c71124a8SPankaj Raghav folio_set_bh(bh, folio, offset); 9371da177e4SLinus Torvalds } 938f745c6f5SShakeel Butt out: 939b87d8cefSRoman Gushchin set_active_memcg(old_memcg); 9401da177e4SLinus Torvalds return head; 9411da177e4SLinus Torvalds /* 9421da177e4SLinus Torvalds * In case anything failed, we just free everything we got. 9431da177e4SLinus Torvalds */ 9441da177e4SLinus Torvalds no_grow: 9451da177e4SLinus Torvalds if (head) { 9461da177e4SLinus Torvalds do { 9471da177e4SLinus Torvalds bh = head; 9481da177e4SLinus Torvalds head = head->b_this_page; 9491da177e4SLinus Torvalds free_buffer_head(bh); 9501da177e4SLinus Torvalds } while (head); 9511da177e4SLinus Torvalds } 9521da177e4SLinus Torvalds 953f745c6f5SShakeel Butt goto out; 9541da177e4SLinus Torvalds } 955c71124a8SPankaj Raghav EXPORT_SYMBOL_GPL(folio_alloc_buffers); 956c71124a8SPankaj Raghav 957c71124a8SPankaj Raghav struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, 958c71124a8SPankaj Raghav bool retry) 959c71124a8SPankaj Raghav { 9602a418157SMatthew Wilcox (Oracle) gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT; 9612a418157SMatthew Wilcox (Oracle) if (retry) 9622a418157SMatthew Wilcox (Oracle) gfp |= __GFP_NOFAIL; 9632a418157SMatthew Wilcox (Oracle) 9642a418157SMatthew Wilcox (Oracle) return folio_alloc_buffers(page_folio(page), size, gfp); 965c71124a8SPankaj Raghav } 9661da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(alloc_page_buffers); 9671da177e4SLinus Torvalds 96808d84addSMatthew Wilcox (Oracle) static inline void link_dev_buffers(struct folio *folio, 96908d84addSMatthew Wilcox (Oracle) struct buffer_head *head) 9701da177e4SLinus Torvalds { 9711da177e4SLinus Torvalds struct buffer_head *bh, *tail; 9721da177e4SLinus Torvalds 9731da177e4SLinus Torvalds bh = head; 9741da177e4SLinus Torvalds do { 9751da177e4SLinus Torvalds tail = bh; 9761da177e4SLinus Torvalds bh = bh->b_this_page; 9771da177e4SLinus Torvalds } while (bh); 9781da177e4SLinus Torvalds tail->b_this_page = head; 97908d84addSMatthew Wilcox (Oracle) folio_attach_private(folio, head); 9801da177e4SLinus Torvalds } 9811da177e4SLinus Torvalds 982bbec0270SLinus Torvalds static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) 983bbec0270SLinus Torvalds { 984bbec0270SLinus Torvalds sector_t retval = ~((sector_t)0); 985b86058f9SChristoph Hellwig loff_t sz = bdev_nr_bytes(bdev); 986bbec0270SLinus Torvalds 987bbec0270SLinus Torvalds if (sz) { 988bbec0270SLinus Torvalds unsigned int sizebits = blksize_bits(size); 989bbec0270SLinus Torvalds retval = (sz >> sizebits); 990bbec0270SLinus Torvalds } 991bbec0270SLinus Torvalds return retval; 992bbec0270SLinus Torvalds } 993bbec0270SLinus Torvalds 9941da177e4SLinus Torvalds /* 9956f24ce6bSMatthew Wilcox (Oracle) * Initialise the state of a blockdev folio's buffers. 9961da177e4SLinus Torvalds */ 9976f24ce6bSMatthew Wilcox (Oracle) static sector_t folio_init_buffers(struct folio *folio, 998382497adSMatthew Wilcox (Oracle) struct block_device *bdev, unsigned size) 9991da177e4SLinus Torvalds { 10006f24ce6bSMatthew Wilcox (Oracle) struct buffer_head *head = folio_buffers(folio); 10011da177e4SLinus Torvalds struct buffer_head *bh = head; 10026f24ce6bSMatthew Wilcox (Oracle) bool uptodate = folio_test_uptodate(folio); 1003382497adSMatthew Wilcox (Oracle) sector_t block = div_u64(folio_pos(folio), size); 1004bcd1d063SChristoph Hellwig sector_t end_block = blkdev_max_block(bdev, size); 10051da177e4SLinus Torvalds 10061da177e4SLinus Torvalds do { 10071da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 100801950a34SEric Biggers bh->b_end_io = NULL; 100901950a34SEric Biggers bh->b_private = NULL; 10101da177e4SLinus Torvalds bh->b_bdev = bdev; 10111da177e4SLinus Torvalds bh->b_blocknr = block; 10121da177e4SLinus Torvalds if (uptodate) 10131da177e4SLinus Torvalds set_buffer_uptodate(bh); 1014080399aaSJeff Moyer if (block < end_block) 10151da177e4SLinus Torvalds set_buffer_mapped(bh); 10161da177e4SLinus Torvalds } 10171da177e4SLinus Torvalds block++; 10181da177e4SLinus Torvalds bh = bh->b_this_page; 10191da177e4SLinus Torvalds } while (bh != head); 1020676ce6d5SHugh Dickins 1021676ce6d5SHugh Dickins /* 1022676ce6d5SHugh Dickins * Caller needs to validate requested block against end of device. 1023676ce6d5SHugh Dickins */ 1024676ce6d5SHugh Dickins return end_block; 10251da177e4SLinus Torvalds } 10261da177e4SLinus Torvalds 10271da177e4SLinus Torvalds /* 10286d840a18SMatthew Wilcox (Oracle) * Create the page-cache folio that contains the requested block. 10291da177e4SLinus Torvalds * 1030676ce6d5SHugh Dickins * This is used purely for blockdev mappings. 10316d840a18SMatthew Wilcox (Oracle) * 10326d840a18SMatthew Wilcox (Oracle) * Returns false if we have a 'permanent' failure. Returns true if 10336d840a18SMatthew Wilcox (Oracle) * we succeeded, or the caller should retry. 10341da177e4SLinus Torvalds */ 10356d840a18SMatthew Wilcox (Oracle) static bool grow_dev_folio(struct block_device *bdev, sector_t block, 1036382497adSMatthew Wilcox (Oracle) pgoff_t index, unsigned size, gfp_t gfp) 10371da177e4SLinus Torvalds { 10381da177e4SLinus Torvalds struct inode *inode = bdev->bd_inode; 10393c98a41cSMatthew Wilcox (Oracle) struct folio *folio; 10401da177e4SLinus Torvalds struct buffer_head *bh; 10416d840a18SMatthew Wilcox (Oracle) sector_t end_block = 0; 104284235de3SJohannes Weiner 10433c98a41cSMatthew Wilcox (Oracle) folio = __filemap_get_folio(inode->i_mapping, index, 10443ed65f04SMatthew Wilcox (Oracle) FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp); 10453ed65f04SMatthew Wilcox (Oracle) if (IS_ERR(folio)) 10466d840a18SMatthew Wilcox (Oracle) return false; 10471da177e4SLinus Torvalds 10483c98a41cSMatthew Wilcox (Oracle) bh = folio_buffers(folio); 10493c98a41cSMatthew Wilcox (Oracle) if (bh) { 10501da177e4SLinus Torvalds if (bh->b_size == size) { 1051382497adSMatthew Wilcox (Oracle) end_block = folio_init_buffers(folio, bdev, size); 10526d840a18SMatthew Wilcox (Oracle) goto unlock; 10531da177e4SLinus Torvalds } 10541da177e4SLinus Torvalds 10556d840a18SMatthew Wilcox (Oracle) /* Caller should retry if this call fails */ 10566d840a18SMatthew Wilcox (Oracle) end_block = ~0ULL; 10576d840a18SMatthew Wilcox (Oracle) if (!try_to_free_buffers(folio)) 10586d840a18SMatthew Wilcox (Oracle) goto unlock; 10596d840a18SMatthew Wilcox (Oracle) } 10606d840a18SMatthew Wilcox (Oracle) 10613ed65f04SMatthew Wilcox (Oracle) bh = folio_alloc_buffers(folio, size, gfp | __GFP_ACCOUNT); 10623ed65f04SMatthew Wilcox (Oracle) if (!bh) 10636d840a18SMatthew Wilcox (Oracle) goto unlock; 10641da177e4SLinus Torvalds 10651da177e4SLinus Torvalds /* 10663c98a41cSMatthew Wilcox (Oracle) * Link the folio to the buffers and initialise them. Take the 10671da177e4SLinus Torvalds * lock to be atomic wrt __find_get_block(), which does not 10683c98a41cSMatthew Wilcox (Oracle) * run under the folio lock. 10691da177e4SLinus Torvalds */ 10701da177e4SLinus Torvalds spin_lock(&inode->i_mapping->private_lock); 107108d84addSMatthew Wilcox (Oracle) link_dev_buffers(folio, bh); 1072382497adSMatthew Wilcox (Oracle) end_block = folio_init_buffers(folio, bdev, size); 10731da177e4SLinus Torvalds spin_unlock(&inode->i_mapping->private_lock); 10746d840a18SMatthew Wilcox (Oracle) unlock: 10753c98a41cSMatthew Wilcox (Oracle) folio_unlock(folio); 10763c98a41cSMatthew Wilcox (Oracle) folio_put(folio); 10776d840a18SMatthew Wilcox (Oracle) return block < end_block; 10781da177e4SLinus Torvalds } 10791da177e4SLinus Torvalds 10801da177e4SLinus Torvalds /* 10816d840a18SMatthew Wilcox (Oracle) * Create buffers for the specified block device block's folio. If 10826d840a18SMatthew Wilcox (Oracle) * that folio was dirty, the buffers are set dirty also. Returns false 10836d840a18SMatthew Wilcox (Oracle) * if we've hit a permanent error. 10841da177e4SLinus Torvalds */ 10856d840a18SMatthew Wilcox (Oracle) static bool grow_buffers(struct block_device *bdev, sector_t block, 10866d840a18SMatthew Wilcox (Oracle) unsigned size, gfp_t gfp) 10871da177e4SLinus Torvalds { 10885f3bd90dSMatthew Wilcox (Oracle) loff_t pos; 10891da177e4SLinus Torvalds 1090e5657933SAndrew Morton /* 10915f3bd90dSMatthew Wilcox (Oracle) * Check for a block which lies outside our maximum possible 10925f3bd90dSMatthew Wilcox (Oracle) * pagecache index. 1093e5657933SAndrew Morton */ 10945f3bd90dSMatthew Wilcox (Oracle) if (check_mul_overflow(block, (sector_t)size, &pos) || pos > MAX_LFS_FILESIZE) { 10955f3bd90dSMatthew Wilcox (Oracle) printk(KERN_ERR "%s: requested out-of-range block %llu for device %pg\n", 10968e24eea7SHarvey Harrison __func__, (unsigned long long)block, 1097a1c6f057SDmitry Monakhov bdev); 10986d840a18SMatthew Wilcox (Oracle) return false; 1099e5657933SAndrew Morton } 1100676ce6d5SHugh Dickins 11016d840a18SMatthew Wilcox (Oracle) /* Create a folio with the proper size buffers */ 11025f3bd90dSMatthew Wilcox (Oracle) return grow_dev_folio(bdev, block, pos / PAGE_SIZE, size, gfp); 11031da177e4SLinus Torvalds } 11041da177e4SLinus Torvalds 11050026ba40SEric Biggers static struct buffer_head * 11063b5e6454SGioh Kim __getblk_slow(struct block_device *bdev, sector_t block, 11073b5e6454SGioh Kim unsigned size, gfp_t gfp) 11081da177e4SLinus Torvalds { 11091da177e4SLinus Torvalds /* Size must be multiple of hard sectorsize */ 1110e1defc4fSMartin K. Petersen if (unlikely(size & (bdev_logical_block_size(bdev)-1) || 11111da177e4SLinus Torvalds (size < 512 || size > PAGE_SIZE))) { 11121da177e4SLinus Torvalds printk(KERN_ERR "getblk(): invalid block size %d requested\n", 11131da177e4SLinus Torvalds size); 1114e1defc4fSMartin K. Petersen printk(KERN_ERR "logical block size: %d\n", 1115e1defc4fSMartin K. Petersen bdev_logical_block_size(bdev)); 11161da177e4SLinus Torvalds 11171da177e4SLinus Torvalds dump_stack(); 11181da177e4SLinus Torvalds return NULL; 11191da177e4SLinus Torvalds } 11201da177e4SLinus Torvalds 1121676ce6d5SHugh Dickins for (;;) { 1122676ce6d5SHugh Dickins struct buffer_head *bh; 1123676ce6d5SHugh Dickins 11241da177e4SLinus Torvalds bh = __find_get_block(bdev, block, size); 11251da177e4SLinus Torvalds if (bh) 11261da177e4SLinus Torvalds return bh; 11271da177e4SLinus Torvalds 11286d840a18SMatthew Wilcox (Oracle) if (!grow_buffers(bdev, block, size, gfp)) 112991f68c89SJeff Moyer return NULL; 1130676ce6d5SHugh Dickins } 11311da177e4SLinus Torvalds } 11321da177e4SLinus Torvalds 11331da177e4SLinus Torvalds /* 11341da177e4SLinus Torvalds * The relationship between dirty buffers and dirty pages: 11351da177e4SLinus Torvalds * 11361da177e4SLinus Torvalds * Whenever a page has any dirty buffers, the page's dirty bit is set, and 1137ec82e1c1SMatthew Wilcox * the page is tagged dirty in the page cache. 11381da177e4SLinus Torvalds * 11391da177e4SLinus Torvalds * At all times, the dirtiness of the buffers represents the dirtiness of 11401da177e4SLinus Torvalds * subsections of the page. If the page has buffers, the page dirty bit is 11411da177e4SLinus Torvalds * merely a hint about the true dirty state. 11421da177e4SLinus Torvalds * 11431da177e4SLinus Torvalds * When a page is set dirty in its entirety, all its buffers are marked dirty 11441da177e4SLinus Torvalds * (if the page has buffers). 11451da177e4SLinus Torvalds * 11461da177e4SLinus Torvalds * When a buffer is marked dirty, its page is dirtied, but the page's other 11471da177e4SLinus Torvalds * buffers are not. 11481da177e4SLinus Torvalds * 11491da177e4SLinus Torvalds * Also. When blockdev buffers are explicitly read with bread(), they 11501da177e4SLinus Torvalds * individually become uptodate. But their backing page remains not 11511da177e4SLinus Torvalds * uptodate - even if all of its buffers are uptodate. A subsequent 11522c69e205SMatthew Wilcox (Oracle) * block_read_full_folio() against that folio will discover all the uptodate 11532c69e205SMatthew Wilcox (Oracle) * buffers, will set the folio uptodate and will perform no I/O. 11541da177e4SLinus Torvalds */ 11551da177e4SLinus Torvalds 11561da177e4SLinus Torvalds /** 11571da177e4SLinus Torvalds * mark_buffer_dirty - mark a buffer_head as needing writeout 115867be2dd1SMartin Waitz * @bh: the buffer_head to mark dirty 11591da177e4SLinus Torvalds * 1160ec82e1c1SMatthew Wilcox * mark_buffer_dirty() will set the dirty bit against the buffer, then set 1161ec82e1c1SMatthew Wilcox * its backing page dirty, then tag the page as dirty in the page cache 1162ec82e1c1SMatthew Wilcox * and then attach the address_space's inode to its superblock's dirty 11631da177e4SLinus Torvalds * inode list. 11641da177e4SLinus Torvalds * 1165abc8a8a2SMatthew Wilcox (Oracle) * mark_buffer_dirty() is atomic. It takes bh->b_folio->mapping->private_lock, 1166b93b0163SMatthew Wilcox * i_pages lock and mapping->host->i_lock. 11671da177e4SLinus Torvalds */ 1168fc9b52cdSHarvey Harrison void mark_buffer_dirty(struct buffer_head *bh) 11691da177e4SLinus Torvalds { 1170787d2214SNick Piggin WARN_ON_ONCE(!buffer_uptodate(bh)); 11711be62dc1SLinus Torvalds 11725305cb83STejun Heo trace_block_dirty_buffer(bh); 11735305cb83STejun Heo 11741be62dc1SLinus Torvalds /* 11751be62dc1SLinus Torvalds * Very *carefully* optimize the it-is-already-dirty case. 11761be62dc1SLinus Torvalds * 11771be62dc1SLinus Torvalds * Don't let the final "is it dirty" escape to before we 11781be62dc1SLinus Torvalds * perhaps modified the buffer. 11791be62dc1SLinus Torvalds */ 11801be62dc1SLinus Torvalds if (buffer_dirty(bh)) { 11811be62dc1SLinus Torvalds smp_mb(); 11821be62dc1SLinus Torvalds if (buffer_dirty(bh)) 11831be62dc1SLinus Torvalds return; 11841be62dc1SLinus Torvalds } 11851be62dc1SLinus Torvalds 1186a8e7d49aSLinus Torvalds if (!test_set_buffer_dirty(bh)) { 1187cf1d3417SMatthew Wilcox (Oracle) struct folio *folio = bh->b_folio; 1188c4843a75SGreg Thelen struct address_space *mapping = NULL; 1189c4843a75SGreg Thelen 1190cf1d3417SMatthew Wilcox (Oracle) folio_memcg_lock(folio); 1191cf1d3417SMatthew Wilcox (Oracle) if (!folio_test_set_dirty(folio)) { 1192cf1d3417SMatthew Wilcox (Oracle) mapping = folio->mapping; 11938e9d78edSLinus Torvalds if (mapping) 1194cf1d3417SMatthew Wilcox (Oracle) __folio_mark_dirty(folio, mapping, 0); 11958e9d78edSLinus Torvalds } 1196cf1d3417SMatthew Wilcox (Oracle) folio_memcg_unlock(folio); 1197c4843a75SGreg Thelen if (mapping) 1198c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 1199a8e7d49aSLinus Torvalds } 12001da177e4SLinus Torvalds } 12011fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(mark_buffer_dirty); 12021da177e4SLinus Torvalds 120387354e5dSJeff Layton void mark_buffer_write_io_error(struct buffer_head *bh) 120487354e5dSJeff Layton { 120587354e5dSJeff Layton set_buffer_write_io_error(bh); 120687354e5dSJeff Layton /* FIXME: do we need to set this in both places? */ 1207abc8a8a2SMatthew Wilcox (Oracle) if (bh->b_folio && bh->b_folio->mapping) 1208abc8a8a2SMatthew Wilcox (Oracle) mapping_set_error(bh->b_folio->mapping, -EIO); 12094b2201daSChristoph Hellwig if (bh->b_assoc_map) { 121087354e5dSJeff Layton mapping_set_error(bh->b_assoc_map, -EIO); 12114b2201daSChristoph Hellwig errseq_set(&bh->b_assoc_map->host->i_sb->s_wb_err, -EIO); 12124b2201daSChristoph Hellwig } 121387354e5dSJeff Layton } 121487354e5dSJeff Layton EXPORT_SYMBOL(mark_buffer_write_io_error); 121587354e5dSJeff Layton 12161da177e4SLinus Torvalds /* 12171da177e4SLinus Torvalds * Decrement a buffer_head's reference count. If all buffers against a page 12181da177e4SLinus Torvalds * have zero reference count, are clean and unlocked, and if the page is clean 12191da177e4SLinus Torvalds * and unlocked then try_to_free_buffers() may strip the buffers from the page 12201da177e4SLinus Torvalds * in preparation for freeing it (sometimes, rarely, buffers are removed from 12211da177e4SLinus Torvalds * a page but it ends up not being freed, and buffers may later be reattached). 12221da177e4SLinus Torvalds */ 12231da177e4SLinus Torvalds void __brelse(struct buffer_head * buf) 12241da177e4SLinus Torvalds { 12251da177e4SLinus Torvalds if (atomic_read(&buf->b_count)) { 12261da177e4SLinus Torvalds put_bh(buf); 12271da177e4SLinus Torvalds return; 12281da177e4SLinus Torvalds } 12295c752ad9SArjan van de Ven WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); 12301da177e4SLinus Torvalds } 12311fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__brelse); 12321da177e4SLinus Torvalds 12331da177e4SLinus Torvalds /* 12341da177e4SLinus Torvalds * bforget() is like brelse(), except it discards any 12351da177e4SLinus Torvalds * potentially dirty data. 12361da177e4SLinus Torvalds */ 12371da177e4SLinus Torvalds void __bforget(struct buffer_head *bh) 12381da177e4SLinus Torvalds { 12391da177e4SLinus Torvalds clear_buffer_dirty(bh); 1240535ee2fbSJan Kara if (bh->b_assoc_map) { 1241abc8a8a2SMatthew Wilcox (Oracle) struct address_space *buffer_mapping = bh->b_folio->mapping; 12421da177e4SLinus Torvalds 12431da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 12441da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 124558ff407bSJan Kara bh->b_assoc_map = NULL; 12461da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 12471da177e4SLinus Torvalds } 12481da177e4SLinus Torvalds __brelse(bh); 12491da177e4SLinus Torvalds } 12501fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__bforget); 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds static struct buffer_head *__bread_slow(struct buffer_head *bh) 12531da177e4SLinus Torvalds { 12541da177e4SLinus Torvalds lock_buffer(bh); 12551da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 12561da177e4SLinus Torvalds unlock_buffer(bh); 12571da177e4SLinus Torvalds return bh; 12581da177e4SLinus Torvalds } else { 12591da177e4SLinus Torvalds get_bh(bh); 12601da177e4SLinus Torvalds bh->b_end_io = end_buffer_read_sync; 12611420c4a5SBart Van Assche submit_bh(REQ_OP_READ, bh); 12621da177e4SLinus Torvalds wait_on_buffer(bh); 12631da177e4SLinus Torvalds if (buffer_uptodate(bh)) 12641da177e4SLinus Torvalds return bh; 12651da177e4SLinus Torvalds } 12661da177e4SLinus Torvalds brelse(bh); 12671da177e4SLinus Torvalds return NULL; 12681da177e4SLinus Torvalds } 12691da177e4SLinus Torvalds 12701da177e4SLinus Torvalds /* 12711da177e4SLinus Torvalds * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). 12721da177e4SLinus Torvalds * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their 12731da177e4SLinus Torvalds * refcount elevated by one when they're in an LRU. A buffer can only appear 12741da177e4SLinus Torvalds * once in a particular CPU's LRU. A single buffer can be present in multiple 12751da177e4SLinus Torvalds * CPU's LRUs at the same time. 12761da177e4SLinus Torvalds * 12771da177e4SLinus Torvalds * This is a transparent caching front-end to sb_bread(), sb_getblk() and 12781da177e4SLinus Torvalds * sb_find_get_block(). 12791da177e4SLinus Torvalds * 12801da177e4SLinus Torvalds * The LRUs themselves only need locking against invalidate_bh_lrus. We use 12811da177e4SLinus Torvalds * a local interrupt disable for that. 12821da177e4SLinus Torvalds */ 12831da177e4SLinus Torvalds 128486cf78d7SSebastien Buisson #define BH_LRU_SIZE 16 12851da177e4SLinus Torvalds 12861da177e4SLinus Torvalds struct bh_lru { 12871da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12881da177e4SLinus Torvalds }; 12891da177e4SLinus Torvalds 12901da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; 12911da177e4SLinus Torvalds 12921da177e4SLinus Torvalds #ifdef CONFIG_SMP 12931da177e4SLinus Torvalds #define bh_lru_lock() local_irq_disable() 12941da177e4SLinus Torvalds #define bh_lru_unlock() local_irq_enable() 12951da177e4SLinus Torvalds #else 12961da177e4SLinus Torvalds #define bh_lru_lock() preempt_disable() 12971da177e4SLinus Torvalds #define bh_lru_unlock() preempt_enable() 12981da177e4SLinus Torvalds #endif 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds static inline void check_irqs_on(void) 13011da177e4SLinus Torvalds { 13021da177e4SLinus Torvalds #ifdef irqs_disabled 13031da177e4SLinus Torvalds BUG_ON(irqs_disabled()); 13041da177e4SLinus Torvalds #endif 13051da177e4SLinus Torvalds } 13061da177e4SLinus Torvalds 13071da177e4SLinus Torvalds /* 1308241f01fbSEric Biggers * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is 1309241f01fbSEric Biggers * inserted at the front, and the buffer_head at the back if any is evicted. 1310241f01fbSEric Biggers * Or, if already in the LRU it is moved to the front. 13111da177e4SLinus Torvalds */ 13121da177e4SLinus Torvalds static void bh_lru_install(struct buffer_head *bh) 13131da177e4SLinus Torvalds { 1314241f01fbSEric Biggers struct buffer_head *evictee = bh; 1315241f01fbSEric Biggers struct bh_lru *b; 1316241f01fbSEric Biggers int i; 13171da177e4SLinus Torvalds 13181da177e4SLinus Torvalds check_irqs_on(); 1319c0226eb8SMinchan Kim bh_lru_lock(); 1320c0226eb8SMinchan Kim 13218cc621d2SMinchan Kim /* 13228cc621d2SMinchan Kim * the refcount of buffer_head in bh_lru prevents dropping the 13238cc621d2SMinchan Kim * attached page(i.e., try_to_free_buffers) so it could cause 13248cc621d2SMinchan Kim * failing page migration. 13258cc621d2SMinchan Kim * Skip putting upcoming bh into bh_lru until migration is done. 13268cc621d2SMinchan Kim */ 13278a237adfSMarcelo Tosatti if (lru_cache_disabled() || cpu_is_isolated(smp_processor_id())) { 1328c0226eb8SMinchan Kim bh_lru_unlock(); 13298cc621d2SMinchan Kim return; 1330c0226eb8SMinchan Kim } 1331241f01fbSEric Biggers 1332241f01fbSEric Biggers b = this_cpu_ptr(&bh_lrus); 1333241f01fbSEric Biggers for (i = 0; i < BH_LRU_SIZE; i++) { 1334241f01fbSEric Biggers swap(evictee, b->bhs[i]); 1335241f01fbSEric Biggers if (evictee == bh) { 1336241f01fbSEric Biggers bh_lru_unlock(); 1337241f01fbSEric Biggers return; 1338241f01fbSEric Biggers } 1339241f01fbSEric Biggers } 13401da177e4SLinus Torvalds 13411da177e4SLinus Torvalds get_bh(bh); 13421da177e4SLinus Torvalds bh_lru_unlock(); 1343241f01fbSEric Biggers brelse(evictee); 13441da177e4SLinus Torvalds } 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds /* 13471da177e4SLinus Torvalds * Look up the bh in this cpu's LRU. If it's there, move it to the head. 13481da177e4SLinus Torvalds */ 1349858119e1SArjan van de Ven static struct buffer_head * 13503991d3bdSTomasz Kvarsin lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) 13511da177e4SLinus Torvalds { 13521da177e4SLinus Torvalds struct buffer_head *ret = NULL; 13533991d3bdSTomasz Kvarsin unsigned int i; 13541da177e4SLinus Torvalds 13551da177e4SLinus Torvalds check_irqs_on(); 13561da177e4SLinus Torvalds bh_lru_lock(); 13578a237adfSMarcelo Tosatti if (cpu_is_isolated(smp_processor_id())) { 13588a237adfSMarcelo Tosatti bh_lru_unlock(); 13598a237adfSMarcelo Tosatti return NULL; 13608a237adfSMarcelo Tosatti } 13611da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 1362c7b92516SChristoph Lameter struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); 13631da177e4SLinus Torvalds 13649470dd5dSZach Brown if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && 13659470dd5dSZach Brown bh->b_size == size) { 13661da177e4SLinus Torvalds if (i) { 13671da177e4SLinus Torvalds while (i) { 1368c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[i], 1369c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[i - 1])); 13701da177e4SLinus Torvalds i--; 13711da177e4SLinus Torvalds } 1372c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[0], bh); 13731da177e4SLinus Torvalds } 13741da177e4SLinus Torvalds get_bh(bh); 13751da177e4SLinus Torvalds ret = bh; 13761da177e4SLinus Torvalds break; 13771da177e4SLinus Torvalds } 13781da177e4SLinus Torvalds } 13791da177e4SLinus Torvalds bh_lru_unlock(); 13801da177e4SLinus Torvalds return ret; 13811da177e4SLinus Torvalds } 13821da177e4SLinus Torvalds 13831da177e4SLinus Torvalds /* 13841da177e4SLinus Torvalds * Perform a pagecache lookup for the matching buffer. If it's there, refresh 13851da177e4SLinus Torvalds * it in the LRU and mark it as accessed. If it is not present then return 13861da177e4SLinus Torvalds * NULL 13871da177e4SLinus Torvalds */ 13881da177e4SLinus Torvalds struct buffer_head * 13893991d3bdSTomasz Kvarsin __find_get_block(struct block_device *bdev, sector_t block, unsigned size) 13901da177e4SLinus Torvalds { 13911da177e4SLinus Torvalds struct buffer_head *bh = lookup_bh_lru(bdev, block, size); 13921da177e4SLinus Torvalds 13931da177e4SLinus Torvalds if (bh == NULL) { 13942457aec6SMel Gorman /* __find_get_block_slow will mark the page accessed */ 1395385fd4c5SCoywolf Qi Hunt bh = __find_get_block_slow(bdev, block); 13961da177e4SLinus Torvalds if (bh) 13971da177e4SLinus Torvalds bh_lru_install(bh); 13982457aec6SMel Gorman } else 13991da177e4SLinus Torvalds touch_buffer(bh); 14002457aec6SMel Gorman 14011da177e4SLinus Torvalds return bh; 14021da177e4SLinus Torvalds } 14031da177e4SLinus Torvalds EXPORT_SYMBOL(__find_get_block); 14041da177e4SLinus Torvalds 14053ed65f04SMatthew Wilcox (Oracle) /** 14063ed65f04SMatthew Wilcox (Oracle) * bdev_getblk - Get a buffer_head in a block device's buffer cache. 14073ed65f04SMatthew Wilcox (Oracle) * @bdev: The block device. 14083ed65f04SMatthew Wilcox (Oracle) * @block: The block number. 14093ed65f04SMatthew Wilcox (Oracle) * @size: The size of buffer_heads for this @bdev. 14103ed65f04SMatthew Wilcox (Oracle) * @gfp: The memory allocation flags to use. 14113ed65f04SMatthew Wilcox (Oracle) * 14123ed65f04SMatthew Wilcox (Oracle) * Return: The buffer head, or NULL if memory could not be allocated. 14133ed65f04SMatthew Wilcox (Oracle) */ 14143ed65f04SMatthew Wilcox (Oracle) struct buffer_head *bdev_getblk(struct block_device *bdev, sector_t block, 14153ed65f04SMatthew Wilcox (Oracle) unsigned size, gfp_t gfp) 14163ed65f04SMatthew Wilcox (Oracle) { 14173ed65f04SMatthew Wilcox (Oracle) struct buffer_head *bh = __find_get_block(bdev, block, size); 14183ed65f04SMatthew Wilcox (Oracle) 14193ed65f04SMatthew Wilcox (Oracle) might_alloc(gfp); 14203ed65f04SMatthew Wilcox (Oracle) if (bh) 14213ed65f04SMatthew Wilcox (Oracle) return bh; 14223ed65f04SMatthew Wilcox (Oracle) 14233ed65f04SMatthew Wilcox (Oracle) return __getblk_slow(bdev, block, size, gfp); 14243ed65f04SMatthew Wilcox (Oracle) } 14253ed65f04SMatthew Wilcox (Oracle) EXPORT_SYMBOL(bdev_getblk); 14263ed65f04SMatthew Wilcox (Oracle) 14271da177e4SLinus Torvalds /* 14281da177e4SLinus Torvalds * Do async read-ahead on a buffer.. 14291da177e4SLinus Torvalds */ 14303991d3bdSTomasz Kvarsin void __breadahead(struct block_device *bdev, sector_t block, unsigned size) 14311da177e4SLinus Torvalds { 1432775d9b10SMatthew Wilcox (Oracle) struct buffer_head *bh = bdev_getblk(bdev, block, size, 1433775d9b10SMatthew Wilcox (Oracle) GFP_NOWAIT | __GFP_MOVABLE); 1434775d9b10SMatthew Wilcox (Oracle) 1435a3e713b5SAndrew Morton if (likely(bh)) { 1436e7ea1129SZhang Yi bh_readahead(bh, REQ_RAHEAD); 14371da177e4SLinus Torvalds brelse(bh); 14381da177e4SLinus Torvalds } 1439a3e713b5SAndrew Morton } 14401da177e4SLinus Torvalds EXPORT_SYMBOL(__breadahead); 14411da177e4SLinus Torvalds 14421da177e4SLinus Torvalds /** 14433b5e6454SGioh Kim * __bread_gfp() - reads a specified block and returns the bh 144467be2dd1SMartin Waitz * @bdev: the block_device to read from 14451da177e4SLinus Torvalds * @block: number of block 14461da177e4SLinus Torvalds * @size: size (in bytes) to read 14473b5e6454SGioh Kim * @gfp: page allocation flag 14481da177e4SLinus Torvalds * 14491da177e4SLinus Torvalds * Reads a specified block, and returns buffer head that contains it. 14503b5e6454SGioh Kim * The page cache can be allocated from non-movable area 14513b5e6454SGioh Kim * not to prevent page migration if you set gfp to zero. 14521da177e4SLinus Torvalds * It returns NULL if the block was unreadable. 14531da177e4SLinus Torvalds */ 14541da177e4SLinus Torvalds struct buffer_head * 14553b5e6454SGioh Kim __bread_gfp(struct block_device *bdev, sector_t block, 14563b5e6454SGioh Kim unsigned size, gfp_t gfp) 14571da177e4SLinus Torvalds { 145893b13ecaSMatthew Wilcox (Oracle) struct buffer_head *bh; 145993b13ecaSMatthew Wilcox (Oracle) 146093b13ecaSMatthew Wilcox (Oracle) gfp |= mapping_gfp_constraint(bdev->bd_inode->i_mapping, ~__GFP_FS); 146193b13ecaSMatthew Wilcox (Oracle) 146293b13ecaSMatthew Wilcox (Oracle) /* 146393b13ecaSMatthew Wilcox (Oracle) * Prefer looping in the allocator rather than here, at least that 146493b13ecaSMatthew Wilcox (Oracle) * code knows what it's doing. 146593b13ecaSMatthew Wilcox (Oracle) */ 146693b13ecaSMatthew Wilcox (Oracle) gfp |= __GFP_NOFAIL; 146793b13ecaSMatthew Wilcox (Oracle) 146893b13ecaSMatthew Wilcox (Oracle) bh = bdev_getblk(bdev, block, size, gfp); 14691da177e4SLinus Torvalds 1470a3e713b5SAndrew Morton if (likely(bh) && !buffer_uptodate(bh)) 14711da177e4SLinus Torvalds bh = __bread_slow(bh); 14721da177e4SLinus Torvalds return bh; 14731da177e4SLinus Torvalds } 14743b5e6454SGioh Kim EXPORT_SYMBOL(__bread_gfp); 14751da177e4SLinus Torvalds 14768cc621d2SMinchan Kim static void __invalidate_bh_lrus(struct bh_lru *b) 14778cc621d2SMinchan Kim { 14788cc621d2SMinchan Kim int i; 14798cc621d2SMinchan Kim 14808cc621d2SMinchan Kim for (i = 0; i < BH_LRU_SIZE; i++) { 14818cc621d2SMinchan Kim brelse(b->bhs[i]); 14828cc621d2SMinchan Kim b->bhs[i] = NULL; 14838cc621d2SMinchan Kim } 14848cc621d2SMinchan Kim } 14851da177e4SLinus Torvalds /* 14861da177e4SLinus Torvalds * invalidate_bh_lrus() is called rarely - but not only at unmount. 14871da177e4SLinus Torvalds * This doesn't race because it runs in each cpu either in irq 14881da177e4SLinus Torvalds * or with preempt disabled. 14891da177e4SLinus Torvalds */ 14901da177e4SLinus Torvalds static void invalidate_bh_lru(void *arg) 14911da177e4SLinus Torvalds { 14921da177e4SLinus Torvalds struct bh_lru *b = &get_cpu_var(bh_lrus); 14931da177e4SLinus Torvalds 14948cc621d2SMinchan Kim __invalidate_bh_lrus(b); 14951da177e4SLinus Torvalds put_cpu_var(bh_lrus); 14961da177e4SLinus Torvalds } 14971da177e4SLinus Torvalds 14988cc621d2SMinchan Kim bool has_bh_in_lru(int cpu, void *dummy) 149942be35d0SGilad Ben-Yossef { 150042be35d0SGilad Ben-Yossef struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); 150142be35d0SGilad Ben-Yossef int i; 150242be35d0SGilad Ben-Yossef 150342be35d0SGilad Ben-Yossef for (i = 0; i < BH_LRU_SIZE; i++) { 150442be35d0SGilad Ben-Yossef if (b->bhs[i]) 15051d706679SSaurav Girepunje return true; 150642be35d0SGilad Ben-Yossef } 150742be35d0SGilad Ben-Yossef 15081d706679SSaurav Girepunje return false; 150942be35d0SGilad Ben-Yossef } 151042be35d0SGilad Ben-Yossef 1511f9a14399SPeter Zijlstra void invalidate_bh_lrus(void) 15121da177e4SLinus Torvalds { 1513cb923159SSebastian Andrzej Siewior on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1); 15141da177e4SLinus Torvalds } 15159db5579bSNick Piggin EXPORT_SYMBOL_GPL(invalidate_bh_lrus); 15161da177e4SLinus Torvalds 1517243418e3SMinchan Kim /* 1518243418e3SMinchan Kim * It's called from workqueue context so we need a bh_lru_lock to close 1519243418e3SMinchan Kim * the race with preemption/irq. 1520243418e3SMinchan Kim */ 1521243418e3SMinchan Kim void invalidate_bh_lrus_cpu(void) 15228cc621d2SMinchan Kim { 15238cc621d2SMinchan Kim struct bh_lru *b; 15248cc621d2SMinchan Kim 15258cc621d2SMinchan Kim bh_lru_lock(); 1526243418e3SMinchan Kim b = this_cpu_ptr(&bh_lrus); 15278cc621d2SMinchan Kim __invalidate_bh_lrus(b); 15288cc621d2SMinchan Kim bh_lru_unlock(); 15298cc621d2SMinchan Kim } 15308cc621d2SMinchan Kim 1531465e5e6aSPankaj Raghav void folio_set_bh(struct buffer_head *bh, struct folio *folio, 1532465e5e6aSPankaj Raghav unsigned long offset) 1533465e5e6aSPankaj Raghav { 1534465e5e6aSPankaj Raghav bh->b_folio = folio; 1535465e5e6aSPankaj Raghav BUG_ON(offset >= folio_size(folio)); 1536465e5e6aSPankaj Raghav if (folio_test_highmem(folio)) 1537465e5e6aSPankaj Raghav /* 1538465e5e6aSPankaj Raghav * This catches illegal uses and preserves the offset: 1539465e5e6aSPankaj Raghav */ 1540465e5e6aSPankaj Raghav bh->b_data = (char *)(0 + offset); 1541465e5e6aSPankaj Raghav else 1542465e5e6aSPankaj Raghav bh->b_data = folio_address(folio) + offset; 1543465e5e6aSPankaj Raghav } 1544465e5e6aSPankaj Raghav EXPORT_SYMBOL(folio_set_bh); 1545465e5e6aSPankaj Raghav 15461da177e4SLinus Torvalds /* 15471da177e4SLinus Torvalds * Called when truncating a buffer on a page completely. 15481da177e4SLinus Torvalds */ 1549e7470ee8SMel Gorman 1550e7470ee8SMel Gorman /* Bits that are cleared during an invalidate */ 1551e7470ee8SMel Gorman #define BUFFER_FLAGS_DISCARD \ 1552e7470ee8SMel Gorman (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ 1553e7470ee8SMel Gorman 1 << BH_Delay | 1 << BH_Unwritten) 1554e7470ee8SMel Gorman 1555858119e1SArjan van de Ven static void discard_buffer(struct buffer_head * bh) 15561da177e4SLinus Torvalds { 1557b0192296SUros Bizjak unsigned long b_state; 1558e7470ee8SMel Gorman 15591da177e4SLinus Torvalds lock_buffer(bh); 15601da177e4SLinus Torvalds clear_buffer_dirty(bh); 15611da177e4SLinus Torvalds bh->b_bdev = NULL; 1562b0192296SUros Bizjak b_state = READ_ONCE(bh->b_state); 1563b0192296SUros Bizjak do { 1564b0192296SUros Bizjak } while (!try_cmpxchg(&bh->b_state, &b_state, 1565b0192296SUros Bizjak b_state & ~BUFFER_FLAGS_DISCARD)); 15661da177e4SLinus Torvalds unlock_buffer(bh); 15671da177e4SLinus Torvalds } 15681da177e4SLinus Torvalds 15691da177e4SLinus Torvalds /** 15707ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio - Invalidate part or all of a buffer-backed folio. 15717ba13abbSMatthew Wilcox (Oracle) * @folio: The folio which is affected. 1572d47992f8SLukas Czerner * @offset: start of the range to invalidate 1573d47992f8SLukas Czerner * @length: length of the range to invalidate 15741da177e4SLinus Torvalds * 15757ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio() is called when all or part of the folio has been 15761da177e4SLinus Torvalds * invalidated by a truncate operation. 15771da177e4SLinus Torvalds * 15787ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio() does not have to release all buffers, but it must 15791da177e4SLinus Torvalds * ensure that no dirty buffer is left outside @offset and that no I/O 15801da177e4SLinus Torvalds * is underway against any of the blocks which are outside the truncation 15811da177e4SLinus Torvalds * point. Because the caller is about to free (and possibly reuse) those 15821da177e4SLinus Torvalds * blocks on-disk. 15831da177e4SLinus Torvalds */ 15847ba13abbSMatthew Wilcox (Oracle) void block_invalidate_folio(struct folio *folio, size_t offset, size_t length) 15851da177e4SLinus Torvalds { 15861da177e4SLinus Torvalds struct buffer_head *head, *bh, *next; 15877ba13abbSMatthew Wilcox (Oracle) size_t curr_off = 0; 15887ba13abbSMatthew Wilcox (Oracle) size_t stop = length + offset; 15891da177e4SLinus Torvalds 15907ba13abbSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 15911da177e4SLinus Torvalds 1592d47992f8SLukas Czerner /* 1593d47992f8SLukas Czerner * Check for overflow 1594d47992f8SLukas Czerner */ 15957ba13abbSMatthew Wilcox (Oracle) BUG_ON(stop > folio_size(folio) || stop < length); 1596d47992f8SLukas Czerner 15977ba13abbSMatthew Wilcox (Oracle) head = folio_buffers(folio); 15987ba13abbSMatthew Wilcox (Oracle) if (!head) 15997ba13abbSMatthew Wilcox (Oracle) return; 16007ba13abbSMatthew Wilcox (Oracle) 16011da177e4SLinus Torvalds bh = head; 16021da177e4SLinus Torvalds do { 16037ba13abbSMatthew Wilcox (Oracle) size_t next_off = curr_off + bh->b_size; 16041da177e4SLinus Torvalds next = bh->b_this_page; 16051da177e4SLinus Torvalds 16061da177e4SLinus Torvalds /* 1607d47992f8SLukas Czerner * Are we still fully in range ? 1608d47992f8SLukas Czerner */ 1609d47992f8SLukas Czerner if (next_off > stop) 1610d47992f8SLukas Czerner goto out; 1611d47992f8SLukas Czerner 1612d47992f8SLukas Czerner /* 16131da177e4SLinus Torvalds * is this block fully invalidated? 16141da177e4SLinus Torvalds */ 16151da177e4SLinus Torvalds if (offset <= curr_off) 16161da177e4SLinus Torvalds discard_buffer(bh); 16171da177e4SLinus Torvalds curr_off = next_off; 16181da177e4SLinus Torvalds bh = next; 16191da177e4SLinus Torvalds } while (bh != head); 16201da177e4SLinus Torvalds 16211da177e4SLinus Torvalds /* 16227ba13abbSMatthew Wilcox (Oracle) * We release buffers only if the entire folio is being invalidated. 16231da177e4SLinus Torvalds * The get_block cached value has been unconditionally invalidated, 16241da177e4SLinus Torvalds * so real IO is not possible anymore. 16251da177e4SLinus Torvalds */ 16267ba13abbSMatthew Wilcox (Oracle) if (length == folio_size(folio)) 16277ba13abbSMatthew Wilcox (Oracle) filemap_release_folio(folio, 0); 16281da177e4SLinus Torvalds out: 16292ff28e22SNeilBrown return; 16301da177e4SLinus Torvalds } 16317ba13abbSMatthew Wilcox (Oracle) EXPORT_SYMBOL(block_invalidate_folio); 16321da177e4SLinus Torvalds 16331da177e4SLinus Torvalds /* 16341da177e4SLinus Torvalds * We attach and possibly dirty the buffers atomically wrt 1635e621900aSMatthew Wilcox (Oracle) * block_dirty_folio() via private_lock. try_to_free_buffers 16368e2e1756SPankaj Raghav * is already excluded via the folio lock. 16371da177e4SLinus Torvalds */ 16380a88810dSMatthew Wilcox (Oracle) struct buffer_head *create_empty_buffers(struct folio *folio, 16393decb856SMatthew Wilcox (Oracle) unsigned long blocksize, unsigned long b_state) 16401da177e4SLinus Torvalds { 16411da177e4SLinus Torvalds struct buffer_head *bh, *head, *tail; 16422a418157SMatthew Wilcox (Oracle) gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT | __GFP_NOFAIL; 16431da177e4SLinus Torvalds 16442a418157SMatthew Wilcox (Oracle) head = folio_alloc_buffers(folio, blocksize, gfp); 16451da177e4SLinus Torvalds bh = head; 16461da177e4SLinus Torvalds do { 16471da177e4SLinus Torvalds bh->b_state |= b_state; 16481da177e4SLinus Torvalds tail = bh; 16491da177e4SLinus Torvalds bh = bh->b_this_page; 16501da177e4SLinus Torvalds } while (bh); 16511da177e4SLinus Torvalds tail->b_this_page = head; 16521da177e4SLinus Torvalds 16538e2e1756SPankaj Raghav spin_lock(&folio->mapping->private_lock); 16548e2e1756SPankaj Raghav if (folio_test_uptodate(folio) || folio_test_dirty(folio)) { 16551da177e4SLinus Torvalds bh = head; 16561da177e4SLinus Torvalds do { 16578e2e1756SPankaj Raghav if (folio_test_dirty(folio)) 16581da177e4SLinus Torvalds set_buffer_dirty(bh); 16598e2e1756SPankaj Raghav if (folio_test_uptodate(folio)) 16601da177e4SLinus Torvalds set_buffer_uptodate(bh); 16611da177e4SLinus Torvalds bh = bh->b_this_page; 16621da177e4SLinus Torvalds } while (bh != head); 16631da177e4SLinus Torvalds } 16648e2e1756SPankaj Raghav folio_attach_private(folio, head); 16658e2e1756SPankaj Raghav spin_unlock(&folio->mapping->private_lock); 16663decb856SMatthew Wilcox (Oracle) 16673decb856SMatthew Wilcox (Oracle) return head; 16688e2e1756SPankaj Raghav } 16691da177e4SLinus Torvalds EXPORT_SYMBOL(create_empty_buffers); 16701da177e4SLinus Torvalds 167129f3ad7dSJan Kara /** 167229f3ad7dSJan Kara * clean_bdev_aliases: clean a range of buffers in block device 167329f3ad7dSJan Kara * @bdev: Block device to clean buffers in 167429f3ad7dSJan Kara * @block: Start of a range of blocks to clean 167529f3ad7dSJan Kara * @len: Number of blocks to clean 16761da177e4SLinus Torvalds * 167729f3ad7dSJan Kara * We are taking a range of blocks for data and we don't want writeback of any 167829f3ad7dSJan Kara * buffer-cache aliases starting from return from this function and until the 167929f3ad7dSJan Kara * moment when something will explicitly mark the buffer dirty (hopefully that 168029f3ad7dSJan Kara * will not happen until we will free that block ;-) We don't even need to mark 168129f3ad7dSJan Kara * it not-uptodate - nobody can expect anything from a newly allocated buffer 168229f3ad7dSJan Kara * anyway. We used to use unmap_buffer() for such invalidation, but that was 168329f3ad7dSJan Kara * wrong. We definitely don't want to mark the alias unmapped, for example - it 168429f3ad7dSJan Kara * would confuse anyone who might pick it with bread() afterwards... 168529f3ad7dSJan Kara * 168629f3ad7dSJan Kara * Also.. Note that bforget() doesn't lock the buffer. So there can be 168729f3ad7dSJan Kara * writeout I/O going on against recently-freed buffers. We don't wait on that 168829f3ad7dSJan Kara * I/O in bforget() - it's more efficient to wait on the I/O only if we really 168929f3ad7dSJan Kara * need to. That happens here. 16901da177e4SLinus Torvalds */ 169129f3ad7dSJan Kara void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) 16921da177e4SLinus Torvalds { 169329f3ad7dSJan Kara struct inode *bd_inode = bdev->bd_inode; 169429f3ad7dSJan Kara struct address_space *bd_mapping = bd_inode->i_mapping; 16959e0b6f31SMatthew Wilcox (Oracle) struct folio_batch fbatch; 169629f3ad7dSJan Kara pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); 169729f3ad7dSJan Kara pgoff_t end; 1698c10f778dSJan Kara int i, count; 169929f3ad7dSJan Kara struct buffer_head *bh; 170029f3ad7dSJan Kara struct buffer_head *head; 17011da177e4SLinus Torvalds 170229f3ad7dSJan Kara end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits); 17039e0b6f31SMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 17049e0b6f31SMatthew Wilcox (Oracle) while (filemap_get_folios(bd_mapping, &index, end, &fbatch)) { 17059e0b6f31SMatthew Wilcox (Oracle) count = folio_batch_count(&fbatch); 1706c10f778dSJan Kara for (i = 0; i < count; i++) { 17079e0b6f31SMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 17081da177e4SLinus Torvalds 17099e0b6f31SMatthew Wilcox (Oracle) if (!folio_buffers(folio)) 171029f3ad7dSJan Kara continue; 171129f3ad7dSJan Kara /* 17129e0b6f31SMatthew Wilcox (Oracle) * We use folio lock instead of bd_mapping->private_lock 171329f3ad7dSJan Kara * to pin buffers here since we can afford to sleep and 171429f3ad7dSJan Kara * it scales better than a global spinlock lock. 171529f3ad7dSJan Kara */ 17169e0b6f31SMatthew Wilcox (Oracle) folio_lock(folio); 17179e0b6f31SMatthew Wilcox (Oracle) /* Recheck when the folio is locked which pins bhs */ 17189e0b6f31SMatthew Wilcox (Oracle) head = folio_buffers(folio); 17199e0b6f31SMatthew Wilcox (Oracle) if (!head) 172029f3ad7dSJan Kara goto unlock_page; 172129f3ad7dSJan Kara bh = head; 172229f3ad7dSJan Kara do { 17236c006a9dSChandan Rajendra if (!buffer_mapped(bh) || (bh->b_blocknr < block)) 172429f3ad7dSJan Kara goto next; 172529f3ad7dSJan Kara if (bh->b_blocknr >= block + len) 172629f3ad7dSJan Kara break; 172729f3ad7dSJan Kara clear_buffer_dirty(bh); 172829f3ad7dSJan Kara wait_on_buffer(bh); 172929f3ad7dSJan Kara clear_buffer_req(bh); 173029f3ad7dSJan Kara next: 173129f3ad7dSJan Kara bh = bh->b_this_page; 173229f3ad7dSJan Kara } while (bh != head); 173329f3ad7dSJan Kara unlock_page: 17349e0b6f31SMatthew Wilcox (Oracle) folio_unlock(folio); 173529f3ad7dSJan Kara } 17369e0b6f31SMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 173729f3ad7dSJan Kara cond_resched(); 1738c10f778dSJan Kara /* End of range already reached? */ 1739c10f778dSJan Kara if (index > end || !index) 1740c10f778dSJan Kara break; 17411da177e4SLinus Torvalds } 17421da177e4SLinus Torvalds } 174329f3ad7dSJan Kara EXPORT_SYMBOL(clean_bdev_aliases); 17441da177e4SLinus Torvalds 17451da177e4SLinus Torvalds /* 174645bce8f3SLinus Torvalds * Size is a power-of-two in the range 512..PAGE_SIZE, 174745bce8f3SLinus Torvalds * and the case we care about most is PAGE_SIZE. 174845bce8f3SLinus Torvalds * 174945bce8f3SLinus Torvalds * So this *could* possibly be written with those 175045bce8f3SLinus Torvalds * constraints in mind (relevant mostly if some 175145bce8f3SLinus Torvalds * architecture has a slow bit-scan instruction) 175245bce8f3SLinus Torvalds */ 175345bce8f3SLinus Torvalds static inline int block_size_bits(unsigned int blocksize) 175445bce8f3SLinus Torvalds { 175545bce8f3SLinus Torvalds return ilog2(blocksize); 175645bce8f3SLinus Torvalds } 175745bce8f3SLinus Torvalds 1758c6c8c3e7SPankaj Raghav static struct buffer_head *folio_create_buffers(struct folio *folio, 1759c6c8c3e7SPankaj Raghav struct inode *inode, 1760c6c8c3e7SPankaj Raghav unsigned int b_state) 176145bce8f3SLinus Torvalds { 17623decb856SMatthew Wilcox (Oracle) struct buffer_head *bh; 17633decb856SMatthew Wilcox (Oracle) 1764c6c8c3e7SPankaj Raghav BUG_ON(!folio_test_locked(folio)); 176545bce8f3SLinus Torvalds 17663decb856SMatthew Wilcox (Oracle) bh = folio_buffers(folio); 17673decb856SMatthew Wilcox (Oracle) if (!bh) 17680a88810dSMatthew Wilcox (Oracle) bh = create_empty_buffers(folio, 17693decb856SMatthew Wilcox (Oracle) 1 << READ_ONCE(inode->i_blkbits), b_state); 17703decb856SMatthew Wilcox (Oracle) return bh; 177145bce8f3SLinus Torvalds } 177245bce8f3SLinus Torvalds 177345bce8f3SLinus Torvalds /* 17741da177e4SLinus Torvalds * NOTE! All mapped/uptodate combinations are valid: 17751da177e4SLinus Torvalds * 17761da177e4SLinus Torvalds * Mapped Uptodate Meaning 17771da177e4SLinus Torvalds * 17781da177e4SLinus Torvalds * No No "unknown" - must do get_block() 17791da177e4SLinus Torvalds * No Yes "hole" - zero-filled 17801da177e4SLinus Torvalds * Yes No "allocated" - allocated on disk, not read in 17811da177e4SLinus Torvalds * Yes Yes "valid" - allocated and up-to-date in memory. 17821da177e4SLinus Torvalds * 17831da177e4SLinus Torvalds * "Dirty" is valid only with the last case (mapped+uptodate). 17841da177e4SLinus Torvalds */ 17851da177e4SLinus Torvalds 17861da177e4SLinus Torvalds /* 17871da177e4SLinus Torvalds * While block_write_full_page is writing back the dirty buffers under 17881da177e4SLinus Torvalds * the page lock, whoever dirtied the buffers may decide to clean them 17891da177e4SLinus Torvalds * again at any time. We handle that by only looking at the buffer 17901da177e4SLinus Torvalds * state inside lock_buffer(). 17911da177e4SLinus Torvalds * 17921da177e4SLinus Torvalds * If block_write_full_page() is called for regular writeback 17931da177e4SLinus Torvalds * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a 17941da177e4SLinus Torvalds * locked buffer. This only can happen if someone has written the buffer 17951da177e4SLinus Torvalds * directly, with submit_bh(). At the address_space level PageWriteback 17961da177e4SLinus Torvalds * prevents this contention from occurring. 17976e34eeddSTheodore Ts'o * 17986e34eeddSTheodore Ts'o * If block_write_full_page() is called with wbc->sync_mode == 179970fd7614SChristoph Hellwig * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this 1800721a9602SJens Axboe * causes the writes to be flagged as synchronous writes. 18011da177e4SLinus Torvalds */ 180253418a18SMatthew Wilcox (Oracle) int __block_write_full_folio(struct inode *inode, struct folio *folio, 180335c80d5fSChris Mason get_block_t *get_block, struct writeback_control *wbc, 180435c80d5fSChris Mason bh_end_io_t *handler) 18051da177e4SLinus Torvalds { 18061da177e4SLinus Torvalds int err; 18071da177e4SLinus Torvalds sector_t block; 18081da177e4SLinus Torvalds sector_t last_block; 1809f0fbd5fcSAndrew Morton struct buffer_head *bh, *head; 181045bce8f3SLinus Torvalds unsigned int blocksize, bbits; 18111da177e4SLinus Torvalds int nr_underway = 0; 18123ae72869SBart Van Assche blk_opf_t write_flags = wbc_to_write_flags(wbc); 18131da177e4SLinus Torvalds 181453418a18SMatthew Wilcox (Oracle) head = folio_create_buffers(folio, inode, 18151da177e4SLinus Torvalds (1 << BH_Dirty) | (1 << BH_Uptodate)); 18161da177e4SLinus Torvalds 18171da177e4SLinus Torvalds /* 1818e621900aSMatthew Wilcox (Oracle) * Be very careful. We have no exclusion from block_dirty_folio 18191da177e4SLinus Torvalds * here, and the (potentially unmapped) buffers may become dirty at 18201da177e4SLinus Torvalds * any time. If a buffer becomes dirty here after we've inspected it 182153418a18SMatthew Wilcox (Oracle) * then we just miss that fact, and the folio stays dirty. 18221da177e4SLinus Torvalds * 1823e621900aSMatthew Wilcox (Oracle) * Buffers outside i_size may be dirtied by block_dirty_folio; 18241da177e4SLinus Torvalds * handle that here by just cleaning them. 18251da177e4SLinus Torvalds */ 18261da177e4SLinus Torvalds 18271da177e4SLinus Torvalds bh = head; 182845bce8f3SLinus Torvalds blocksize = bh->b_size; 182945bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 183045bce8f3SLinus Torvalds 183153418a18SMatthew Wilcox (Oracle) block = (sector_t)folio->index << (PAGE_SHIFT - bbits); 183245bce8f3SLinus Torvalds last_block = (i_size_read(inode) - 1) >> bbits; 18331da177e4SLinus Torvalds 18341da177e4SLinus Torvalds /* 18351da177e4SLinus Torvalds * Get all the dirty buffers mapped to disk addresses and 18361da177e4SLinus Torvalds * handle any aliases from the underlying blockdev's mapping. 18371da177e4SLinus Torvalds */ 18381da177e4SLinus Torvalds do { 18391da177e4SLinus Torvalds if (block > last_block) { 18401da177e4SLinus Torvalds /* 18411da177e4SLinus Torvalds * mapped buffers outside i_size will occur, because 184253418a18SMatthew Wilcox (Oracle) * this folio can be outside i_size when there is a 18431da177e4SLinus Torvalds * truncate in progress. 18441da177e4SLinus Torvalds */ 18451da177e4SLinus Torvalds /* 18461da177e4SLinus Torvalds * The buffer was zeroed by block_write_full_page() 18471da177e4SLinus Torvalds */ 18481da177e4SLinus Torvalds clear_buffer_dirty(bh); 18491da177e4SLinus Torvalds set_buffer_uptodate(bh); 185029a814d2SAlex Tomas } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && 185129a814d2SAlex Tomas buffer_dirty(bh)) { 1852b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 18531da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 18541da177e4SLinus Torvalds if (err) 18551da177e4SLinus Torvalds goto recover; 185629a814d2SAlex Tomas clear_buffer_delay(bh); 18571da177e4SLinus Torvalds if (buffer_new(bh)) { 18581da177e4SLinus Torvalds /* blockdev mappings never come here */ 18591da177e4SLinus Torvalds clear_buffer_new(bh); 1860e64855c6SJan Kara clean_bdev_bh_alias(bh); 18611da177e4SLinus Torvalds } 18621da177e4SLinus Torvalds } 18631da177e4SLinus Torvalds bh = bh->b_this_page; 18641da177e4SLinus Torvalds block++; 18651da177e4SLinus Torvalds } while (bh != head); 18661da177e4SLinus Torvalds 18671da177e4SLinus Torvalds do { 18681da177e4SLinus Torvalds if (!buffer_mapped(bh)) 18691da177e4SLinus Torvalds continue; 18701da177e4SLinus Torvalds /* 18711da177e4SLinus Torvalds * If it's a fully non-blocking write attempt and we cannot 187253418a18SMatthew Wilcox (Oracle) * lock the buffer then redirty the folio. Note that this can 18735b0830cbSJens Axboe * potentially cause a busy-wait loop from writeback threads 18745b0830cbSJens Axboe * and kswapd activity, but those code paths have their own 18755b0830cbSJens Axboe * higher-level throttling. 18761da177e4SLinus Torvalds */ 18771b430beeSWu Fengguang if (wbc->sync_mode != WB_SYNC_NONE) { 18781da177e4SLinus Torvalds lock_buffer(bh); 1879ca5de404SNick Piggin } else if (!trylock_buffer(bh)) { 188053418a18SMatthew Wilcox (Oracle) folio_redirty_for_writepage(wbc, folio); 18811da177e4SLinus Torvalds continue; 18821da177e4SLinus Torvalds } 18831da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 188435c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 18851da177e4SLinus Torvalds } else { 18861da177e4SLinus Torvalds unlock_buffer(bh); 18871da177e4SLinus Torvalds } 18881da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 18891da177e4SLinus Torvalds 18901da177e4SLinus Torvalds /* 189153418a18SMatthew Wilcox (Oracle) * The folio and its buffers are protected by the writeback flag, 189253418a18SMatthew Wilcox (Oracle) * so we can drop the bh refcounts early. 18931da177e4SLinus Torvalds */ 189453418a18SMatthew Wilcox (Oracle) BUG_ON(folio_test_writeback(folio)); 189553418a18SMatthew Wilcox (Oracle) folio_start_writeback(folio); 18961da177e4SLinus Torvalds 18971da177e4SLinus Torvalds do { 18981da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 18991da177e4SLinus Torvalds if (buffer_async_write(bh)) { 19001420c4a5SBart Van Assche submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc); 19011da177e4SLinus Torvalds nr_underway++; 1902ad576e63SNick Piggin } 19031da177e4SLinus Torvalds bh = next; 19041da177e4SLinus Torvalds } while (bh != head); 190553418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 19061da177e4SLinus Torvalds 19071da177e4SLinus Torvalds err = 0; 19081da177e4SLinus Torvalds done: 19091da177e4SLinus Torvalds if (nr_underway == 0) { 19101da177e4SLinus Torvalds /* 191153418a18SMatthew Wilcox (Oracle) * The folio was marked dirty, but the buffers were 19121da177e4SLinus Torvalds * clean. Someone wrote them back by hand with 191379f59784SZhang Yi * write_dirty_buffer/submit_bh. A rare case. 19141da177e4SLinus Torvalds */ 191553418a18SMatthew Wilcox (Oracle) folio_end_writeback(folio); 19163d67f2d7SNick Piggin 19171da177e4SLinus Torvalds /* 191853418a18SMatthew Wilcox (Oracle) * The folio and buffer_heads can be released at any time from 19191da177e4SLinus Torvalds * here on. 19201da177e4SLinus Torvalds */ 19211da177e4SLinus Torvalds } 19221da177e4SLinus Torvalds return err; 19231da177e4SLinus Torvalds 19241da177e4SLinus Torvalds recover: 19251da177e4SLinus Torvalds /* 19261da177e4SLinus Torvalds * ENOSPC, or some other error. We may already have added some 19271da177e4SLinus Torvalds * blocks to the file, so we need to write these out to avoid 19281da177e4SLinus Torvalds * exposing stale data. 192953418a18SMatthew Wilcox (Oracle) * The folio is currently locked and not marked for writeback 19301da177e4SLinus Torvalds */ 19311da177e4SLinus Torvalds bh = head; 19321da177e4SLinus Torvalds /* Recovery: lock and submit the mapped buffers */ 19331da177e4SLinus Torvalds do { 193429a814d2SAlex Tomas if (buffer_mapped(bh) && buffer_dirty(bh) && 193529a814d2SAlex Tomas !buffer_delay(bh)) { 19361da177e4SLinus Torvalds lock_buffer(bh); 193735c80d5fSChris Mason mark_buffer_async_write_endio(bh, handler); 19381da177e4SLinus Torvalds } else { 19391da177e4SLinus Torvalds /* 19401da177e4SLinus Torvalds * The buffer may have been set dirty during 194153418a18SMatthew Wilcox (Oracle) * attachment to a dirty folio. 19421da177e4SLinus Torvalds */ 19431da177e4SLinus Torvalds clear_buffer_dirty(bh); 19441da177e4SLinus Torvalds } 19451da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 194653418a18SMatthew Wilcox (Oracle) folio_set_error(folio); 194753418a18SMatthew Wilcox (Oracle) BUG_ON(folio_test_writeback(folio)); 194853418a18SMatthew Wilcox (Oracle) mapping_set_error(folio->mapping, err); 194953418a18SMatthew Wilcox (Oracle) folio_start_writeback(folio); 19501da177e4SLinus Torvalds do { 19511da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 19521da177e4SLinus Torvalds if (buffer_async_write(bh)) { 19531da177e4SLinus Torvalds clear_buffer_dirty(bh); 19541420c4a5SBart Van Assche submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc); 19551da177e4SLinus Torvalds nr_underway++; 1956ad576e63SNick Piggin } 19571da177e4SLinus Torvalds bh = next; 19581da177e4SLinus Torvalds } while (bh != head); 195953418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 19601da177e4SLinus Torvalds goto done; 19611da177e4SLinus Torvalds } 196253418a18SMatthew Wilcox (Oracle) EXPORT_SYMBOL(__block_write_full_folio); 19631da177e4SLinus Torvalds 1964afddba49SNick Piggin /* 19654a9622f2SMatthew Wilcox (Oracle) * If a folio has any new buffers, zero them out here, and mark them uptodate 1966afddba49SNick Piggin * and dirty so they'll be written out (in order to prevent uninitialised 1967afddba49SNick Piggin * block data from leaking). And clear the new bit. 1968afddba49SNick Piggin */ 19694a9622f2SMatthew Wilcox (Oracle) void folio_zero_new_buffers(struct folio *folio, size_t from, size_t to) 1970afddba49SNick Piggin { 19714a9622f2SMatthew Wilcox (Oracle) size_t block_start, block_end; 1972afddba49SNick Piggin struct buffer_head *head, *bh; 1973afddba49SNick Piggin 19744a9622f2SMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 19754a9622f2SMatthew Wilcox (Oracle) head = folio_buffers(folio); 19764a9622f2SMatthew Wilcox (Oracle) if (!head) 1977afddba49SNick Piggin return; 1978afddba49SNick Piggin 19794a9622f2SMatthew Wilcox (Oracle) bh = head; 1980afddba49SNick Piggin block_start = 0; 1981afddba49SNick Piggin do { 1982afddba49SNick Piggin block_end = block_start + bh->b_size; 1983afddba49SNick Piggin 1984afddba49SNick Piggin if (buffer_new(bh)) { 1985afddba49SNick Piggin if (block_end > from && block_start < to) { 19864a9622f2SMatthew Wilcox (Oracle) if (!folio_test_uptodate(folio)) { 19874a9622f2SMatthew Wilcox (Oracle) size_t start, xend; 1988afddba49SNick Piggin 1989afddba49SNick Piggin start = max(from, block_start); 19904a9622f2SMatthew Wilcox (Oracle) xend = min(to, block_end); 1991afddba49SNick Piggin 19924a9622f2SMatthew Wilcox (Oracle) folio_zero_segment(folio, start, xend); 1993afddba49SNick Piggin set_buffer_uptodate(bh); 1994afddba49SNick Piggin } 1995afddba49SNick Piggin 1996afddba49SNick Piggin clear_buffer_new(bh); 1997afddba49SNick Piggin mark_buffer_dirty(bh); 1998afddba49SNick Piggin } 1999afddba49SNick Piggin } 2000afddba49SNick Piggin 2001afddba49SNick Piggin block_start = block_end; 2002afddba49SNick Piggin bh = bh->b_this_page; 2003afddba49SNick Piggin } while (bh != head); 2004afddba49SNick Piggin } 20054a9622f2SMatthew Wilcox (Oracle) EXPORT_SYMBOL(folio_zero_new_buffers); 2006afddba49SNick Piggin 20074aa8cdd5SChristoph Hellwig static int 2008ae259a9cSChristoph Hellwig iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, 20096d49cc85SChristoph Hellwig const struct iomap *iomap) 2010ae259a9cSChristoph Hellwig { 2011*80844194SMatthew Wilcox (Oracle) loff_t offset = (loff_t)block << inode->i_blkbits; 2012ae259a9cSChristoph Hellwig 2013ae259a9cSChristoph Hellwig bh->b_bdev = iomap->bdev; 2014ae259a9cSChristoph Hellwig 2015ae259a9cSChristoph Hellwig /* 2016ae259a9cSChristoph Hellwig * Block points to offset in file we need to map, iomap contains 2017ae259a9cSChristoph Hellwig * the offset at which the map starts. If the map ends before the 2018ae259a9cSChristoph Hellwig * current block, then do not map the buffer and let the caller 2019ae259a9cSChristoph Hellwig * handle it. 2020ae259a9cSChristoph Hellwig */ 20214aa8cdd5SChristoph Hellwig if (offset >= iomap->offset + iomap->length) 20224aa8cdd5SChristoph Hellwig return -EIO; 2023ae259a9cSChristoph Hellwig 2024ae259a9cSChristoph Hellwig switch (iomap->type) { 2025ae259a9cSChristoph Hellwig case IOMAP_HOLE: 2026ae259a9cSChristoph Hellwig /* 2027ae259a9cSChristoph Hellwig * If the buffer is not up to date or beyond the current EOF, 2028ae259a9cSChristoph Hellwig * we need to mark it as new to ensure sub-block zeroing is 2029ae259a9cSChristoph Hellwig * executed if necessary. 2030ae259a9cSChristoph Hellwig */ 2031ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 2032ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 2033ae259a9cSChristoph Hellwig set_buffer_new(bh); 20344aa8cdd5SChristoph Hellwig return 0; 2035ae259a9cSChristoph Hellwig case IOMAP_DELALLOC: 2036ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 2037ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 2038ae259a9cSChristoph Hellwig set_buffer_new(bh); 2039ae259a9cSChristoph Hellwig set_buffer_uptodate(bh); 2040ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 2041ae259a9cSChristoph Hellwig set_buffer_delay(bh); 20424aa8cdd5SChristoph Hellwig return 0; 2043ae259a9cSChristoph Hellwig case IOMAP_UNWRITTEN: 2044ae259a9cSChristoph Hellwig /* 20453d7b6b21SAndreas Gruenbacher * For unwritten regions, we always need to ensure that regions 20463d7b6b21SAndreas Gruenbacher * in the block we are not writing to are zeroed. Mark the 20473d7b6b21SAndreas Gruenbacher * buffer as new to ensure this. 2048ae259a9cSChristoph Hellwig */ 2049ae259a9cSChristoph Hellwig set_buffer_new(bh); 2050ae259a9cSChristoph Hellwig set_buffer_unwritten(bh); 2051df561f66SGustavo A. R. Silva fallthrough; 2052ae259a9cSChristoph Hellwig case IOMAP_MAPPED: 20533d7b6b21SAndreas Gruenbacher if ((iomap->flags & IOMAP_F_NEW) || 2054381c0432SChristoph Hellwig offset >= i_size_read(inode)) { 2055381c0432SChristoph Hellwig /* 2056381c0432SChristoph Hellwig * This can happen if truncating the block device races 2057381c0432SChristoph Hellwig * with the check in the caller as i_size updates on 2058381c0432SChristoph Hellwig * block devices aren't synchronized by i_rwsem for 2059381c0432SChristoph Hellwig * block devices. 2060381c0432SChristoph Hellwig */ 2061381c0432SChristoph Hellwig if (S_ISBLK(inode->i_mode)) 2062381c0432SChristoph Hellwig return -EIO; 2063ae259a9cSChristoph Hellwig set_buffer_new(bh); 2064381c0432SChristoph Hellwig } 206519fe5f64SAndreas Gruenbacher bh->b_blocknr = (iomap->addr + offset - iomap->offset) >> 206619fe5f64SAndreas Gruenbacher inode->i_blkbits; 2067ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 20684aa8cdd5SChristoph Hellwig return 0; 20694aa8cdd5SChristoph Hellwig default: 20704aa8cdd5SChristoph Hellwig WARN_ON_ONCE(1); 20714aa8cdd5SChristoph Hellwig return -EIO; 2072ae259a9cSChristoph Hellwig } 2073ae259a9cSChristoph Hellwig } 2074ae259a9cSChristoph Hellwig 2075d1bd0b4eSMatthew Wilcox (Oracle) int __block_write_begin_int(struct folio *folio, loff_t pos, unsigned len, 20766d49cc85SChristoph Hellwig get_block_t *get_block, const struct iomap *iomap) 20771da177e4SLinus Torvalds { 207809cbfeafSKirill A. Shutemov unsigned from = pos & (PAGE_SIZE - 1); 2079ebdec241SChristoph Hellwig unsigned to = from + len; 2080d1bd0b4eSMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 20811da177e4SLinus Torvalds unsigned block_start, block_end; 20821da177e4SLinus Torvalds sector_t block; 20831da177e4SLinus Torvalds int err = 0; 20841da177e4SLinus Torvalds unsigned blocksize, bbits; 20851da177e4SLinus Torvalds struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; 20861da177e4SLinus Torvalds 2087d1bd0b4eSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 208809cbfeafSKirill A. Shutemov BUG_ON(from > PAGE_SIZE); 208909cbfeafSKirill A. Shutemov BUG_ON(to > PAGE_SIZE); 20901da177e4SLinus Torvalds BUG_ON(from > to); 20911da177e4SLinus Torvalds 2092c6c8c3e7SPankaj Raghav head = folio_create_buffers(folio, inode, 0); 209345bce8f3SLinus Torvalds blocksize = head->b_size; 209445bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 20951da177e4SLinus Torvalds 2096d1bd0b4eSMatthew Wilcox (Oracle) block = (sector_t)folio->index << (PAGE_SHIFT - bbits); 20971da177e4SLinus Torvalds 20981da177e4SLinus Torvalds for(bh = head, block_start = 0; bh != head || !block_start; 20991da177e4SLinus Torvalds block++, block_start=block_end, bh = bh->b_this_page) { 21001da177e4SLinus Torvalds block_end = block_start + blocksize; 21011da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 2102d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 21031da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 21041da177e4SLinus Torvalds set_buffer_uptodate(bh); 21051da177e4SLinus Torvalds } 21061da177e4SLinus Torvalds continue; 21071da177e4SLinus Torvalds } 21081da177e4SLinus Torvalds if (buffer_new(bh)) 21091da177e4SLinus Torvalds clear_buffer_new(bh); 21101da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2111b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 21124aa8cdd5SChristoph Hellwig if (get_block) 21131da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 21144aa8cdd5SChristoph Hellwig else 21154aa8cdd5SChristoph Hellwig err = iomap_to_bh(inode, block, bh, iomap); 21161da177e4SLinus Torvalds if (err) 2117f3ddbdc6SNick Piggin break; 2118ae259a9cSChristoph Hellwig 21191da177e4SLinus Torvalds if (buffer_new(bh)) { 2120e64855c6SJan Kara clean_bdev_bh_alias(bh); 2121d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 2122637aff46SNick Piggin clear_buffer_new(bh); 21231da177e4SLinus Torvalds set_buffer_uptodate(bh); 2124637aff46SNick Piggin mark_buffer_dirty(bh); 21251da177e4SLinus Torvalds continue; 21261da177e4SLinus Torvalds } 2127eebd2aa3SChristoph Lameter if (block_end > to || block_start < from) 2128d1bd0b4eSMatthew Wilcox (Oracle) folio_zero_segments(folio, 2129eebd2aa3SChristoph Lameter to, block_end, 2130eebd2aa3SChristoph Lameter block_start, from); 21311da177e4SLinus Torvalds continue; 21321da177e4SLinus Torvalds } 21331da177e4SLinus Torvalds } 2134d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 21351da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 21361da177e4SLinus Torvalds set_buffer_uptodate(bh); 21371da177e4SLinus Torvalds continue; 21381da177e4SLinus Torvalds } 21391da177e4SLinus Torvalds if (!buffer_uptodate(bh) && !buffer_delay(bh) && 214033a266ddSDavid Chinner !buffer_unwritten(bh) && 21411da177e4SLinus Torvalds (block_start < from || block_end > to)) { 2142e7ea1129SZhang Yi bh_read_nowait(bh, 0); 21431da177e4SLinus Torvalds *wait_bh++=bh; 21441da177e4SLinus Torvalds } 21451da177e4SLinus Torvalds } 21461da177e4SLinus Torvalds /* 21471da177e4SLinus Torvalds * If we issued read requests - let them complete. 21481da177e4SLinus Torvalds */ 21491da177e4SLinus Torvalds while(wait_bh > wait) { 21501da177e4SLinus Torvalds wait_on_buffer(*--wait_bh); 21511da177e4SLinus Torvalds if (!buffer_uptodate(*wait_bh)) 2152f3ddbdc6SNick Piggin err = -EIO; 21531da177e4SLinus Torvalds } 2154f9f07b6cSJan Kara if (unlikely(err)) 21554a9622f2SMatthew Wilcox (Oracle) folio_zero_new_buffers(folio, from, to); 21561da177e4SLinus Torvalds return err; 21571da177e4SLinus Torvalds } 2158ae259a9cSChristoph Hellwig 2159ae259a9cSChristoph Hellwig int __block_write_begin(struct page *page, loff_t pos, unsigned len, 2160ae259a9cSChristoph Hellwig get_block_t *get_block) 2161ae259a9cSChristoph Hellwig { 2162d1bd0b4eSMatthew Wilcox (Oracle) return __block_write_begin_int(page_folio(page), pos, len, get_block, 2163d1bd0b4eSMatthew Wilcox (Oracle) NULL); 2164ae259a9cSChristoph Hellwig } 2165ebdec241SChristoph Hellwig EXPORT_SYMBOL(__block_write_begin); 21661da177e4SLinus Torvalds 2167a524fcfeSBean Huo static void __block_commit_write(struct folio *folio, size_t from, size_t to) 21681da177e4SLinus Torvalds { 21698c6cb3e3SMatthew Wilcox (Oracle) size_t block_start, block_end; 21708c6cb3e3SMatthew Wilcox (Oracle) bool partial = false; 21711da177e4SLinus Torvalds unsigned blocksize; 21721da177e4SLinus Torvalds struct buffer_head *bh, *head; 21731da177e4SLinus Torvalds 21748c6cb3e3SMatthew Wilcox (Oracle) bh = head = folio_buffers(folio); 217545bce8f3SLinus Torvalds blocksize = bh->b_size; 21761da177e4SLinus Torvalds 217745bce8f3SLinus Torvalds block_start = 0; 217845bce8f3SLinus Torvalds do { 21791da177e4SLinus Torvalds block_end = block_start + blocksize; 21801da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 21811da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 21828c6cb3e3SMatthew Wilcox (Oracle) partial = true; 21831da177e4SLinus Torvalds } else { 21841da177e4SLinus Torvalds set_buffer_uptodate(bh); 21851da177e4SLinus Torvalds mark_buffer_dirty(bh); 21861da177e4SLinus Torvalds } 21874ebd3aecSYang Guo if (buffer_new(bh)) 2188afddba49SNick Piggin clear_buffer_new(bh); 218945bce8f3SLinus Torvalds 219045bce8f3SLinus Torvalds block_start = block_end; 219145bce8f3SLinus Torvalds bh = bh->b_this_page; 219245bce8f3SLinus Torvalds } while (bh != head); 21931da177e4SLinus Torvalds 21941da177e4SLinus Torvalds /* 21951da177e4SLinus Torvalds * If this is a partial write which happened to make all buffers 21962c69e205SMatthew Wilcox (Oracle) * uptodate then we can optimize away a bogus read_folio() for 21978c6cb3e3SMatthew Wilcox (Oracle) * the next read(). Here we 'discover' whether the folio went 21981da177e4SLinus Torvalds * uptodate as a result of this (potentially partial) write. 21991da177e4SLinus Torvalds */ 22001da177e4SLinus Torvalds if (!partial) 22018c6cb3e3SMatthew Wilcox (Oracle) folio_mark_uptodate(folio); 22021da177e4SLinus Torvalds } 22031da177e4SLinus Torvalds 22041da177e4SLinus Torvalds /* 2205155130a4SChristoph Hellwig * block_write_begin takes care of the basic task of block allocation and 2206155130a4SChristoph Hellwig * bringing partial write blocks uptodate first. 2207155130a4SChristoph Hellwig * 22087bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 2209afddba49SNick Piggin */ 2210155130a4SChristoph Hellwig int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, 2211b3992d1eSMatthew Wilcox (Oracle) struct page **pagep, get_block_t *get_block) 2212afddba49SNick Piggin { 221309cbfeafSKirill A. Shutemov pgoff_t index = pos >> PAGE_SHIFT; 2214afddba49SNick Piggin struct page *page; 22156e1db88dSChristoph Hellwig int status; 2216afddba49SNick Piggin 2217b7446e7cSMatthew Wilcox (Oracle) page = grab_cache_page_write_begin(mapping, index); 22186e1db88dSChristoph Hellwig if (!page) 22196e1db88dSChristoph Hellwig return -ENOMEM; 2220afddba49SNick Piggin 22216e1db88dSChristoph Hellwig status = __block_write_begin(page, pos, len, get_block); 2222afddba49SNick Piggin if (unlikely(status)) { 2223afddba49SNick Piggin unlock_page(page); 222409cbfeafSKirill A. Shutemov put_page(page); 22256e1db88dSChristoph Hellwig page = NULL; 2226afddba49SNick Piggin } 2227afddba49SNick Piggin 22286e1db88dSChristoph Hellwig *pagep = page; 2229afddba49SNick Piggin return status; 2230afddba49SNick Piggin } 2231afddba49SNick Piggin EXPORT_SYMBOL(block_write_begin); 2232afddba49SNick Piggin 2233afddba49SNick Piggin int block_write_end(struct file *file, struct address_space *mapping, 2234afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2235afddba49SNick Piggin struct page *page, void *fsdata) 2236afddba49SNick Piggin { 22378c6cb3e3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 22388c6cb3e3SMatthew Wilcox (Oracle) size_t start = pos - folio_pos(folio); 2239afddba49SNick Piggin 2240afddba49SNick Piggin if (unlikely(copied < len)) { 2241afddba49SNick Piggin /* 22422c69e205SMatthew Wilcox (Oracle) * The buffers that were written will now be uptodate, so 22432c69e205SMatthew Wilcox (Oracle) * we don't have to worry about a read_folio reading them 22442c69e205SMatthew Wilcox (Oracle) * and overwriting a partial write. However if we have 22452c69e205SMatthew Wilcox (Oracle) * encountered a short write and only partially written 22462c69e205SMatthew Wilcox (Oracle) * into a buffer, it will not be marked uptodate, so a 22472c69e205SMatthew Wilcox (Oracle) * read_folio might come in and destroy our partial write. 2248afddba49SNick Piggin * 2249afddba49SNick Piggin * Do the simplest thing, and just treat any short write to a 22508c6cb3e3SMatthew Wilcox (Oracle) * non uptodate folio as a zero-length write, and force the 2251afddba49SNick Piggin * caller to redo the whole thing. 2252afddba49SNick Piggin */ 22538c6cb3e3SMatthew Wilcox (Oracle) if (!folio_test_uptodate(folio)) 2254afddba49SNick Piggin copied = 0; 2255afddba49SNick Piggin 22564a9622f2SMatthew Wilcox (Oracle) folio_zero_new_buffers(folio, start+copied, start+len); 2257afddba49SNick Piggin } 22588c6cb3e3SMatthew Wilcox (Oracle) flush_dcache_folio(folio); 2259afddba49SNick Piggin 2260afddba49SNick Piggin /* This could be a short (even 0-length) commit */ 2261489b7e72SBean Huo __block_commit_write(folio, start, start + copied); 2262afddba49SNick Piggin 2263afddba49SNick Piggin return copied; 2264afddba49SNick Piggin } 2265afddba49SNick Piggin EXPORT_SYMBOL(block_write_end); 2266afddba49SNick Piggin 2267afddba49SNick Piggin int generic_write_end(struct file *file, struct address_space *mapping, 2268afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2269afddba49SNick Piggin struct page *page, void *fsdata) 2270afddba49SNick Piggin { 22718af54f29SChristoph Hellwig struct inode *inode = mapping->host; 22728af54f29SChristoph Hellwig loff_t old_size = inode->i_size; 22738af54f29SChristoph Hellwig bool i_size_changed = false; 22748af54f29SChristoph Hellwig 2275afddba49SNick Piggin copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); 22768af54f29SChristoph Hellwig 22778af54f29SChristoph Hellwig /* 22788af54f29SChristoph Hellwig * No need to use i_size_read() here, the i_size cannot change under us 22798af54f29SChristoph Hellwig * because we hold i_rwsem. 22808af54f29SChristoph Hellwig * 22818af54f29SChristoph Hellwig * But it's important to update i_size while still holding page lock: 22828af54f29SChristoph Hellwig * page writeout could otherwise come in and zero beyond i_size. 22838af54f29SChristoph Hellwig */ 22848af54f29SChristoph Hellwig if (pos + copied > inode->i_size) { 22858af54f29SChristoph Hellwig i_size_write(inode, pos + copied); 22868af54f29SChristoph Hellwig i_size_changed = true; 22878af54f29SChristoph Hellwig } 22888af54f29SChristoph Hellwig 22898af54f29SChristoph Hellwig unlock_page(page); 22907a77dad7SAndreas Gruenbacher put_page(page); 22918af54f29SChristoph Hellwig 22928af54f29SChristoph Hellwig if (old_size < pos) 22938af54f29SChristoph Hellwig pagecache_isize_extended(inode, old_size, pos); 22948af54f29SChristoph Hellwig /* 22958af54f29SChristoph Hellwig * Don't mark the inode dirty under page lock. First, it unnecessarily 22968af54f29SChristoph Hellwig * makes the holding time of page lock longer. Second, it forces lock 22978af54f29SChristoph Hellwig * ordering of page lock and transaction start for journaling 22988af54f29SChristoph Hellwig * filesystems. 22998af54f29SChristoph Hellwig */ 23008af54f29SChristoph Hellwig if (i_size_changed) 23018af54f29SChristoph Hellwig mark_inode_dirty(inode); 230226ddb1f4SAndreas Gruenbacher return copied; 2303afddba49SNick Piggin } 2304afddba49SNick Piggin EXPORT_SYMBOL(generic_write_end); 2305afddba49SNick Piggin 2306afddba49SNick Piggin /* 23072e7e80f7SMatthew Wilcox (Oracle) * block_is_partially_uptodate checks whether buffers within a folio are 23088ab22b9aSHisashi Hifumi * uptodate or not. 23098ab22b9aSHisashi Hifumi * 23102e7e80f7SMatthew Wilcox (Oracle) * Returns true if all buffers which correspond to the specified part 23112e7e80f7SMatthew Wilcox (Oracle) * of the folio are uptodate. 23128ab22b9aSHisashi Hifumi */ 23132e7e80f7SMatthew Wilcox (Oracle) bool block_is_partially_uptodate(struct folio *folio, size_t from, size_t count) 23148ab22b9aSHisashi Hifumi { 23158ab22b9aSHisashi Hifumi unsigned block_start, block_end, blocksize; 23168ab22b9aSHisashi Hifumi unsigned to; 23178ab22b9aSHisashi Hifumi struct buffer_head *bh, *head; 23182e7e80f7SMatthew Wilcox (Oracle) bool ret = true; 23198ab22b9aSHisashi Hifumi 23202e7e80f7SMatthew Wilcox (Oracle) head = folio_buffers(folio); 23212e7e80f7SMatthew Wilcox (Oracle) if (!head) 23222e7e80f7SMatthew Wilcox (Oracle) return false; 232345bce8f3SLinus Torvalds blocksize = head->b_size; 23242e7e80f7SMatthew Wilcox (Oracle) to = min_t(unsigned, folio_size(folio) - from, count); 23258ab22b9aSHisashi Hifumi to = from + to; 23262e7e80f7SMatthew Wilcox (Oracle) if (from < blocksize && to > folio_size(folio) - blocksize) 23272e7e80f7SMatthew Wilcox (Oracle) return false; 23288ab22b9aSHisashi Hifumi 23298ab22b9aSHisashi Hifumi bh = head; 23308ab22b9aSHisashi Hifumi block_start = 0; 23318ab22b9aSHisashi Hifumi do { 23328ab22b9aSHisashi Hifumi block_end = block_start + blocksize; 23338ab22b9aSHisashi Hifumi if (block_end > from && block_start < to) { 23348ab22b9aSHisashi Hifumi if (!buffer_uptodate(bh)) { 23352e7e80f7SMatthew Wilcox (Oracle) ret = false; 23368ab22b9aSHisashi Hifumi break; 23378ab22b9aSHisashi Hifumi } 23388ab22b9aSHisashi Hifumi if (block_end >= to) 23398ab22b9aSHisashi Hifumi break; 23408ab22b9aSHisashi Hifumi } 23418ab22b9aSHisashi Hifumi block_start = block_end; 23428ab22b9aSHisashi Hifumi bh = bh->b_this_page; 23438ab22b9aSHisashi Hifumi } while (bh != head); 23448ab22b9aSHisashi Hifumi 23458ab22b9aSHisashi Hifumi return ret; 23468ab22b9aSHisashi Hifumi } 23478ab22b9aSHisashi Hifumi EXPORT_SYMBOL(block_is_partially_uptodate); 23488ab22b9aSHisashi Hifumi 23498ab22b9aSHisashi Hifumi /* 23502c69e205SMatthew Wilcox (Oracle) * Generic "read_folio" function for block devices that have the normal 23511da177e4SLinus Torvalds * get_block functionality. This is most of the block device filesystems. 23522c69e205SMatthew Wilcox (Oracle) * Reads the folio asynchronously --- the unlock_buffer() and 23531da177e4SLinus Torvalds * set/clear_buffer_uptodate() functions propagate buffer state into the 23542c69e205SMatthew Wilcox (Oracle) * folio once IO has completed. 23551da177e4SLinus Torvalds */ 23562c69e205SMatthew Wilcox (Oracle) int block_read_full_folio(struct folio *folio, get_block_t *get_block) 23571da177e4SLinus Torvalds { 23582c69e205SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 23591da177e4SLinus Torvalds sector_t iblock, lblock; 23601da177e4SLinus Torvalds struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 236145bce8f3SLinus Torvalds unsigned int blocksize, bbits; 23621da177e4SLinus Torvalds int nr, i; 23631da177e4SLinus Torvalds int fully_mapped = 1; 2364b7a6eb22SMatthew Wilcox (Oracle) bool page_error = false; 23654fa512ceSEric Biggers loff_t limit = i_size_read(inode); 23664fa512ceSEric Biggers 23674fa512ceSEric Biggers /* This is needed for ext4. */ 23684fa512ceSEric Biggers if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode)) 23694fa512ceSEric Biggers limit = inode->i_sb->s_maxbytes; 23701da177e4SLinus Torvalds 23712c69e205SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_large(folio), folio); 23722c69e205SMatthew Wilcox (Oracle) 2373c6c8c3e7SPankaj Raghav head = folio_create_buffers(folio, inode, 0); 237445bce8f3SLinus Torvalds blocksize = head->b_size; 237545bce8f3SLinus Torvalds bbits = block_size_bits(blocksize); 23761da177e4SLinus Torvalds 23772c69e205SMatthew Wilcox (Oracle) iblock = (sector_t)folio->index << (PAGE_SHIFT - bbits); 23784fa512ceSEric Biggers lblock = (limit+blocksize-1) >> bbits; 23791da177e4SLinus Torvalds bh = head; 23801da177e4SLinus Torvalds nr = 0; 23811da177e4SLinus Torvalds i = 0; 23821da177e4SLinus Torvalds 23831da177e4SLinus Torvalds do { 23841da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23851da177e4SLinus Torvalds continue; 23861da177e4SLinus Torvalds 23871da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2388c64610baSAndrew Morton int err = 0; 2389c64610baSAndrew Morton 23901da177e4SLinus Torvalds fully_mapped = 0; 23911da177e4SLinus Torvalds if (iblock < lblock) { 2392b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2393c64610baSAndrew Morton err = get_block(inode, iblock, bh, 0); 2394b7a6eb22SMatthew Wilcox (Oracle) if (err) { 23952c69e205SMatthew Wilcox (Oracle) folio_set_error(folio); 2396b7a6eb22SMatthew Wilcox (Oracle) page_error = true; 2397b7a6eb22SMatthew Wilcox (Oracle) } 23981da177e4SLinus Torvalds } 23991da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 24002c69e205SMatthew Wilcox (Oracle) folio_zero_range(folio, i * blocksize, 24012c69e205SMatthew Wilcox (Oracle) blocksize); 2402c64610baSAndrew Morton if (!err) 24031da177e4SLinus Torvalds set_buffer_uptodate(bh); 24041da177e4SLinus Torvalds continue; 24051da177e4SLinus Torvalds } 24061da177e4SLinus Torvalds /* 24071da177e4SLinus Torvalds * get_block() might have updated the buffer 24081da177e4SLinus Torvalds * synchronously 24091da177e4SLinus Torvalds */ 24101da177e4SLinus Torvalds if (buffer_uptodate(bh)) 24111da177e4SLinus Torvalds continue; 24121da177e4SLinus Torvalds } 24131da177e4SLinus Torvalds arr[nr++] = bh; 24141da177e4SLinus Torvalds } while (i++, iblock++, (bh = bh->b_this_page) != head); 24151da177e4SLinus Torvalds 24161da177e4SLinus Torvalds if (fully_mapped) 24172c69e205SMatthew Wilcox (Oracle) folio_set_mappedtodisk(folio); 24181da177e4SLinus Torvalds 24191da177e4SLinus Torvalds if (!nr) { 24201da177e4SLinus Torvalds /* 24216ba924d3SMatthew Wilcox (Oracle) * All buffers are uptodate or get_block() returned an 24226ba924d3SMatthew Wilcox (Oracle) * error when trying to map them - we can finish the read. 24231da177e4SLinus Torvalds */ 24246ba924d3SMatthew Wilcox (Oracle) folio_end_read(folio, !page_error); 24251da177e4SLinus Torvalds return 0; 24261da177e4SLinus Torvalds } 24271da177e4SLinus Torvalds 24281da177e4SLinus Torvalds /* Stage two: lock the buffers */ 24291da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 24301da177e4SLinus Torvalds bh = arr[i]; 24311da177e4SLinus Torvalds lock_buffer(bh); 24321da177e4SLinus Torvalds mark_buffer_async_read(bh); 24331da177e4SLinus Torvalds } 24341da177e4SLinus Torvalds 24351da177e4SLinus Torvalds /* 24361da177e4SLinus Torvalds * Stage 3: start the IO. Check for uptodateness 24371da177e4SLinus Torvalds * inside the buffer lock in case another process reading 24381da177e4SLinus Torvalds * the underlying blockdev brought it uptodate (the sct fix). 24391da177e4SLinus Torvalds */ 24401da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 24411da177e4SLinus Torvalds bh = arr[i]; 24421da177e4SLinus Torvalds if (buffer_uptodate(bh)) 24431da177e4SLinus Torvalds end_buffer_async_read(bh, 1); 24441da177e4SLinus Torvalds else 24451420c4a5SBart Van Assche submit_bh(REQ_OP_READ, bh); 24461da177e4SLinus Torvalds } 24471da177e4SLinus Torvalds return 0; 24481da177e4SLinus Torvalds } 24492c69e205SMatthew Wilcox (Oracle) EXPORT_SYMBOL(block_read_full_folio); 24501da177e4SLinus Torvalds 24511da177e4SLinus Torvalds /* utility function for filesystems that need to do work on expanding 245289e10787SNick Piggin * truncates. Uses filesystem pagecache writes to allow the filesystem to 24531da177e4SLinus Torvalds * deal with the hole. 24541da177e4SLinus Torvalds */ 245589e10787SNick Piggin int generic_cont_expand_simple(struct inode *inode, loff_t size) 24561da177e4SLinus Torvalds { 24571da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 245853b524b8SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 24591da177e4SLinus Torvalds struct page *page; 24601468c6f4SAlexander Potapenko void *fsdata = NULL; 24611da177e4SLinus Torvalds int err; 24621da177e4SLinus Torvalds 2463c08d3b0eSnpiggin@suse.de err = inode_newsize_ok(inode, size); 2464c08d3b0eSnpiggin@suse.de if (err) 24651da177e4SLinus Torvalds goto out; 24661da177e4SLinus Torvalds 246753b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(NULL, mapping, size, 0, &page, &fsdata); 246889e10787SNick Piggin if (err) 246905eb0b51SOGAWA Hirofumi goto out; 247005eb0b51SOGAWA Hirofumi 247153b524b8SMatthew Wilcox (Oracle) err = aops->write_end(NULL, mapping, size, 0, 0, page, fsdata); 247289e10787SNick Piggin BUG_ON(err > 0); 247305eb0b51SOGAWA Hirofumi 247405eb0b51SOGAWA Hirofumi out: 247505eb0b51SOGAWA Hirofumi return err; 247605eb0b51SOGAWA Hirofumi } 24771fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_cont_expand_simple); 247805eb0b51SOGAWA Hirofumi 2479f1e3af72SAdrian Bunk static int cont_expand_zero(struct file *file, struct address_space *mapping, 248089e10787SNick Piggin loff_t pos, loff_t *bytes) 248105eb0b51SOGAWA Hirofumi { 248289e10787SNick Piggin struct inode *inode = mapping->host; 248353b524b8SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 248493407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 248589e10787SNick Piggin struct page *page; 24861468c6f4SAlexander Potapenko void *fsdata = NULL; 248789e10787SNick Piggin pgoff_t index, curidx; 248889e10787SNick Piggin loff_t curpos; 248989e10787SNick Piggin unsigned zerofrom, offset, len; 249089e10787SNick Piggin int err = 0; 249105eb0b51SOGAWA Hirofumi 249209cbfeafSKirill A. Shutemov index = pos >> PAGE_SHIFT; 249309cbfeafSKirill A. Shutemov offset = pos & ~PAGE_MASK; 249489e10787SNick Piggin 249509cbfeafSKirill A. Shutemov while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { 249609cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 249789e10787SNick Piggin if (zerofrom & (blocksize-1)) { 249889e10787SNick Piggin *bytes |= (blocksize-1); 249989e10787SNick Piggin (*bytes)++; 250089e10787SNick Piggin } 250109cbfeafSKirill A. Shutemov len = PAGE_SIZE - zerofrom; 250289e10787SNick Piggin 250353b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(file, mapping, curpos, len, 250489e10787SNick Piggin &page, &fsdata); 250589e10787SNick Piggin if (err) 250689e10787SNick Piggin goto out; 2507eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 250853b524b8SMatthew Wilcox (Oracle) err = aops->write_end(file, mapping, curpos, len, len, 250989e10787SNick Piggin page, fsdata); 251089e10787SNick Piggin if (err < 0) 251189e10787SNick Piggin goto out; 251289e10787SNick Piggin BUG_ON(err != len); 251389e10787SNick Piggin err = 0; 2514061e9746SOGAWA Hirofumi 2515061e9746SOGAWA Hirofumi balance_dirty_pages_ratelimited(mapping); 2516c2ca0fcdSMikulas Patocka 251708d405c8SDavidlohr Bueso if (fatal_signal_pending(current)) { 2518c2ca0fcdSMikulas Patocka err = -EINTR; 2519c2ca0fcdSMikulas Patocka goto out; 2520c2ca0fcdSMikulas Patocka } 252189e10787SNick Piggin } 252289e10787SNick Piggin 252389e10787SNick Piggin /* page covers the boundary, find the boundary offset */ 252489e10787SNick Piggin if (index == curidx) { 252509cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 252689e10787SNick Piggin /* if we will expand the thing last block will be filled */ 252789e10787SNick Piggin if (offset <= zerofrom) { 252889e10787SNick Piggin goto out; 252989e10787SNick Piggin } 253089e10787SNick Piggin if (zerofrom & (blocksize-1)) { 253189e10787SNick Piggin *bytes |= (blocksize-1); 253289e10787SNick Piggin (*bytes)++; 253389e10787SNick Piggin } 253489e10787SNick Piggin len = offset - zerofrom; 253589e10787SNick Piggin 253653b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(file, mapping, curpos, len, 253789e10787SNick Piggin &page, &fsdata); 253889e10787SNick Piggin if (err) 253989e10787SNick Piggin goto out; 2540eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 254153b524b8SMatthew Wilcox (Oracle) err = aops->write_end(file, mapping, curpos, len, len, 254289e10787SNick Piggin page, fsdata); 254389e10787SNick Piggin if (err < 0) 254489e10787SNick Piggin goto out; 254589e10787SNick Piggin BUG_ON(err != len); 254689e10787SNick Piggin err = 0; 254789e10787SNick Piggin } 254889e10787SNick Piggin out: 254989e10787SNick Piggin return err; 25501da177e4SLinus Torvalds } 25511da177e4SLinus Torvalds 25521da177e4SLinus Torvalds /* 25531da177e4SLinus Torvalds * For moronic filesystems that do not allow holes in file. 25541da177e4SLinus Torvalds * We may have to extend the file. 25551da177e4SLinus Torvalds */ 2556282dc178SChristoph Hellwig int cont_write_begin(struct file *file, struct address_space *mapping, 2557be3bbbc5SMatthew Wilcox (Oracle) loff_t pos, unsigned len, 255889e10787SNick Piggin struct page **pagep, void **fsdata, 255989e10787SNick Piggin get_block_t *get_block, loff_t *bytes) 25601da177e4SLinus Torvalds { 25611da177e4SLinus Torvalds struct inode *inode = mapping->host; 256293407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 256393407472SFabian Frederick unsigned int zerofrom; 256489e10787SNick Piggin int err; 25651da177e4SLinus Torvalds 256689e10787SNick Piggin err = cont_expand_zero(file, mapping, pos, bytes); 256789e10787SNick Piggin if (err) 2568155130a4SChristoph Hellwig return err; 25691da177e4SLinus Torvalds 257009cbfeafSKirill A. Shutemov zerofrom = *bytes & ~PAGE_MASK; 257189e10787SNick Piggin if (pos+len > *bytes && zerofrom & (blocksize-1)) { 25721da177e4SLinus Torvalds *bytes |= (blocksize-1); 25731da177e4SLinus Torvalds (*bytes)++; 25741da177e4SLinus Torvalds } 25751da177e4SLinus Torvalds 2576b3992d1eSMatthew Wilcox (Oracle) return block_write_begin(mapping, pos, len, pagep, get_block); 25771da177e4SLinus Torvalds } 25781fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(cont_write_begin); 25791da177e4SLinus Torvalds 2580a524fcfeSBean Huo void block_commit_write(struct page *page, unsigned from, unsigned to) 25811da177e4SLinus Torvalds { 25828c6cb3e3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 2583489b7e72SBean Huo __block_commit_write(folio, from, to); 25841da177e4SLinus Torvalds } 25851fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_commit_write); 25861da177e4SLinus Torvalds 258754171690SDavid Chinner /* 258854171690SDavid Chinner * block_page_mkwrite() is not allowed to change the file size as it gets 258954171690SDavid Chinner * called from a page fault handler when a page is first dirtied. Hence we must 259054171690SDavid Chinner * be careful to check for EOF conditions here. We set the page up correctly 259154171690SDavid Chinner * for a written page which means we get ENOSPC checking when writing into 259254171690SDavid Chinner * holes and correct delalloc and unwritten extent mapping on filesystems that 259354171690SDavid Chinner * support these features. 259454171690SDavid Chinner * 259554171690SDavid Chinner * We are not allowed to take the i_mutex here so we have to play games to 259654171690SDavid Chinner * protect against truncate races as the page could now be beyond EOF. Because 25977bb46a67Snpiggin@suse.de * truncate writes the inode size before removing pages, once we have the 259854171690SDavid Chinner * page lock we can determine safely if the page is beyond EOF. If it is not 259954171690SDavid Chinner * beyond EOF, then the page is guaranteed safe against truncation until we 260054171690SDavid Chinner * unlock the page. 2601ea13a864SJan Kara * 260214da9200SJan Kara * Direct callers of this function should protect against filesystem freezing 26035c500029SRoss Zwisler * using sb_start_pagefault() - sb_end_pagefault() functions. 260454171690SDavid Chinner */ 26055c500029SRoss Zwisler int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 260654171690SDavid Chinner get_block_t get_block) 260754171690SDavid Chinner { 2608fe181377SMatthew Wilcox (Oracle) struct folio *folio = page_folio(vmf->page); 2609496ad9aaSAl Viro struct inode *inode = file_inode(vma->vm_file); 261054171690SDavid Chinner unsigned long end; 261154171690SDavid Chinner loff_t size; 261224da4fabSJan Kara int ret; 261354171690SDavid Chinner 2614fe181377SMatthew Wilcox (Oracle) folio_lock(folio); 261554171690SDavid Chinner size = i_size_read(inode); 2616fe181377SMatthew Wilcox (Oracle) if ((folio->mapping != inode->i_mapping) || 2617fe181377SMatthew Wilcox (Oracle) (folio_pos(folio) >= size)) { 261824da4fabSJan Kara /* We overload EFAULT to mean page got truncated */ 261924da4fabSJan Kara ret = -EFAULT; 262024da4fabSJan Kara goto out_unlock; 262154171690SDavid Chinner } 262254171690SDavid Chinner 2623fe181377SMatthew Wilcox (Oracle) end = folio_size(folio); 2624fe181377SMatthew Wilcox (Oracle) /* folio is wholly or partially inside EOF */ 2625fe181377SMatthew Wilcox (Oracle) if (folio_pos(folio) + end > size) 2626fe181377SMatthew Wilcox (Oracle) end = size - folio_pos(folio); 262754171690SDavid Chinner 2628fe181377SMatthew Wilcox (Oracle) ret = __block_write_begin_int(folio, 0, end, get_block, NULL); 2629a524fcfeSBean Huo if (unlikely(ret)) 263024da4fabSJan Kara goto out_unlock; 2631a524fcfeSBean Huo 2632a524fcfeSBean Huo __block_commit_write(folio, 0, end); 2633a524fcfeSBean Huo 2634fe181377SMatthew Wilcox (Oracle) folio_mark_dirty(folio); 2635fe181377SMatthew Wilcox (Oracle) folio_wait_stable(folio); 263624da4fabSJan Kara return 0; 263724da4fabSJan Kara out_unlock: 2638fe181377SMatthew Wilcox (Oracle) folio_unlock(folio); 263954171690SDavid Chinner return ret; 264054171690SDavid Chinner } 26411fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_page_mkwrite); 26421da177e4SLinus Torvalds 26431da177e4SLinus Torvalds int block_truncate_page(struct address_space *mapping, 26441da177e4SLinus Torvalds loff_t from, get_block_t *get_block) 26451da177e4SLinus Torvalds { 264609cbfeafSKirill A. Shutemov pgoff_t index = from >> PAGE_SHIFT; 26471da177e4SLinus Torvalds unsigned blocksize; 264854b21a79SAndrew Morton sector_t iblock; 26496d68f644SMatthew Wilcox (Oracle) size_t offset, length, pos; 26501da177e4SLinus Torvalds struct inode *inode = mapping->host; 26516d68f644SMatthew Wilcox (Oracle) struct folio *folio; 26521da177e4SLinus Torvalds struct buffer_head *bh; 2653dc7cb2d2SJiapeng Chong int err = 0; 26541da177e4SLinus Torvalds 265593407472SFabian Frederick blocksize = i_blocksize(inode); 26566d68f644SMatthew Wilcox (Oracle) length = from & (blocksize - 1); 26571da177e4SLinus Torvalds 26581da177e4SLinus Torvalds /* Block boundary? Nothing to do */ 26591da177e4SLinus Torvalds if (!length) 26601da177e4SLinus Torvalds return 0; 26611da177e4SLinus Torvalds 26621da177e4SLinus Torvalds length = blocksize - length; 266309cbfeafSKirill A. Shutemov iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); 26641da177e4SLinus Torvalds 26656d68f644SMatthew Wilcox (Oracle) folio = filemap_grab_folio(mapping, index); 26666d68f644SMatthew Wilcox (Oracle) if (IS_ERR(folio)) 26676d68f644SMatthew Wilcox (Oracle) return PTR_ERR(folio); 26681da177e4SLinus Torvalds 26696d68f644SMatthew Wilcox (Oracle) bh = folio_buffers(folio); 26703decb856SMatthew Wilcox (Oracle) if (!bh) 26710a88810dSMatthew Wilcox (Oracle) bh = create_empty_buffers(folio, blocksize, 0); 26721da177e4SLinus Torvalds 26731da177e4SLinus Torvalds /* Find the buffer that contains "offset" */ 26746d68f644SMatthew Wilcox (Oracle) offset = offset_in_folio(folio, from); 26751da177e4SLinus Torvalds pos = blocksize; 26761da177e4SLinus Torvalds while (offset >= pos) { 26771da177e4SLinus Torvalds bh = bh->b_this_page; 26781da177e4SLinus Torvalds iblock++; 26791da177e4SLinus Torvalds pos += blocksize; 26801da177e4SLinus Torvalds } 26811da177e4SLinus Torvalds 26821da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2683b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 26841da177e4SLinus Torvalds err = get_block(inode, iblock, bh, 0); 26851da177e4SLinus Torvalds if (err) 26861da177e4SLinus Torvalds goto unlock; 26871da177e4SLinus Torvalds /* unmapped? It's a hole - nothing to do */ 26881da177e4SLinus Torvalds if (!buffer_mapped(bh)) 26891da177e4SLinus Torvalds goto unlock; 26901da177e4SLinus Torvalds } 26911da177e4SLinus Torvalds 26921da177e4SLinus Torvalds /* Ok, it's mapped. Make sure it's up-to-date */ 26936d68f644SMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) 26941da177e4SLinus Torvalds set_buffer_uptodate(bh); 26951da177e4SLinus Torvalds 269633a266ddSDavid Chinner if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { 2697e7ea1129SZhang Yi err = bh_read(bh, 0); 26981da177e4SLinus Torvalds /* Uhhuh. Read error. Complain and punt. */ 2699e7ea1129SZhang Yi if (err < 0) 27001da177e4SLinus Torvalds goto unlock; 27011da177e4SLinus Torvalds } 27021da177e4SLinus Torvalds 27036d68f644SMatthew Wilcox (Oracle) folio_zero_range(folio, offset, length); 27041da177e4SLinus Torvalds mark_buffer_dirty(bh); 27051da177e4SLinus Torvalds 27061da177e4SLinus Torvalds unlock: 27076d68f644SMatthew Wilcox (Oracle) folio_unlock(folio); 27086d68f644SMatthew Wilcox (Oracle) folio_put(folio); 2709dc7cb2d2SJiapeng Chong 27101da177e4SLinus Torvalds return err; 27111da177e4SLinus Torvalds } 27121fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_truncate_page); 27131da177e4SLinus Torvalds 27141da177e4SLinus Torvalds /* 27151da177e4SLinus Torvalds * The generic ->writepage function for buffer-backed address_spaces 27161da177e4SLinus Torvalds */ 27171b938c08SMatthew Wilcox int block_write_full_page(struct page *page, get_block_t *get_block, 27181b938c08SMatthew Wilcox struct writeback_control *wbc) 27191da177e4SLinus Torvalds { 272053418a18SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 2721bb0ea598SMatthew Wilcox (Oracle) struct inode * const inode = folio->mapping->host; 27221da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 27231da177e4SLinus Torvalds 2724bb0ea598SMatthew Wilcox (Oracle) /* Is the folio fully inside i_size? */ 2725bb0ea598SMatthew Wilcox (Oracle) if (folio_pos(folio) + folio_size(folio) <= i_size) 272653418a18SMatthew Wilcox (Oracle) return __block_write_full_folio(inode, folio, get_block, wbc, 27271b938c08SMatthew Wilcox end_buffer_async_write); 27281da177e4SLinus Torvalds 2729bb0ea598SMatthew Wilcox (Oracle) /* Is the folio fully outside i_size? (truncate in progress) */ 2730bb0ea598SMatthew Wilcox (Oracle) if (folio_pos(folio) >= i_size) { 273153418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 27321da177e4SLinus Torvalds return 0; /* don't care */ 27331da177e4SLinus Torvalds } 27341da177e4SLinus Torvalds 27351da177e4SLinus Torvalds /* 2736bb0ea598SMatthew Wilcox (Oracle) * The folio straddles i_size. It must be zeroed out on each and every 27372a61aa40SAdam Buchbinder * writepage invocation because it may be mmapped. "A file is mapped 27381da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 27391da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 27401da177e4SLinus Torvalds * writes to that region are not written out to the file." 27411da177e4SLinus Torvalds */ 2742bb0ea598SMatthew Wilcox (Oracle) folio_zero_segment(folio, offset_in_folio(folio, i_size), 2743bb0ea598SMatthew Wilcox (Oracle) folio_size(folio)); 274453418a18SMatthew Wilcox (Oracle) return __block_write_full_folio(inode, folio, get_block, wbc, 274535c80d5fSChris Mason end_buffer_async_write); 274635c80d5fSChris Mason } 27471fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_write_full_page); 274835c80d5fSChris Mason 27491da177e4SLinus Torvalds sector_t generic_block_bmap(struct address_space *mapping, sector_t block, 27501da177e4SLinus Torvalds get_block_t *get_block) 27511da177e4SLinus Torvalds { 27521da177e4SLinus Torvalds struct inode *inode = mapping->host; 27532a527d68SAlexander Potapenko struct buffer_head tmp = { 27542a527d68SAlexander Potapenko .b_size = i_blocksize(inode), 27552a527d68SAlexander Potapenko }; 27562a527d68SAlexander Potapenko 27571da177e4SLinus Torvalds get_block(inode, block, &tmp, 0); 27581da177e4SLinus Torvalds return tmp.b_blocknr; 27591da177e4SLinus Torvalds } 27601fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_block_bmap); 27611da177e4SLinus Torvalds 27624246a0b6SChristoph Hellwig static void end_bio_bh_io_sync(struct bio *bio) 27631da177e4SLinus Torvalds { 27641da177e4SLinus Torvalds struct buffer_head *bh = bio->bi_private; 27651da177e4SLinus Torvalds 2766b7c44ed9SJens Axboe if (unlikely(bio_flagged(bio, BIO_QUIET))) 276708bafc03SKeith Mannthey set_bit(BH_Quiet, &bh->b_state); 276808bafc03SKeith Mannthey 27694e4cbee9SChristoph Hellwig bh->b_end_io(bh, !bio->bi_status); 27701da177e4SLinus Torvalds bio_put(bio); 27711da177e4SLinus Torvalds } 27721da177e4SLinus Torvalds 27735bdf402aSRitesh Harjani (IBM) static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh, 27741420c4a5SBart Van Assche struct writeback_control *wbc) 27751da177e4SLinus Torvalds { 27761420c4a5SBart Van Assche const enum req_op op = opf & REQ_OP_MASK; 27771da177e4SLinus Torvalds struct bio *bio; 27781da177e4SLinus Torvalds 27791da177e4SLinus Torvalds BUG_ON(!buffer_locked(bh)); 27801da177e4SLinus Torvalds BUG_ON(!buffer_mapped(bh)); 27811da177e4SLinus Torvalds BUG_ON(!bh->b_end_io); 27828fb0e342SAneesh Kumar K.V BUG_ON(buffer_delay(bh)); 27838fb0e342SAneesh Kumar K.V BUG_ON(buffer_unwritten(bh)); 27841da177e4SLinus Torvalds 278548fd4f93SJens Axboe /* 278648fd4f93SJens Axboe * Only clear out a write error when rewriting 27871da177e4SLinus Torvalds */ 27882a222ca9SMike Christie if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) 27891da177e4SLinus Torvalds clear_buffer_write_io_error(bh); 27901da177e4SLinus Torvalds 279107888c66SChristoph Hellwig if (buffer_meta(bh)) 27921420c4a5SBart Van Assche opf |= REQ_META; 279307888c66SChristoph Hellwig if (buffer_prio(bh)) 27941420c4a5SBart Van Assche opf |= REQ_PRIO; 279507888c66SChristoph Hellwig 27961420c4a5SBart Van Assche bio = bio_alloc(bh->b_bdev, 1, opf, GFP_NOIO); 27971da177e4SLinus Torvalds 27984f74d15fSEric Biggers fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO); 27994f74d15fSEric Biggers 28004f024f37SKent Overstreet bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); 28011da177e4SLinus Torvalds 2802741af75dSJohannes Thumshirn __bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); 28031da177e4SLinus Torvalds 28041da177e4SLinus Torvalds bio->bi_end_io = end_bio_bh_io_sync; 28051da177e4SLinus Torvalds bio->bi_private = bh; 28061da177e4SLinus Torvalds 280783c9c547SMing Lei /* Take care of bh's that straddle the end of the device */ 280883c9c547SMing Lei guard_bio_eod(bio); 280983c9c547SMing Lei 2810fd42df30SDennis Zhou if (wbc) { 2811fd42df30SDennis Zhou wbc_init_bio(wbc, bio); 281234e51a5eSTejun Heo wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size); 2813fd42df30SDennis Zhou } 2814fd42df30SDennis Zhou 28154e49ea4aSMike Christie submit_bio(bio); 28161da177e4SLinus Torvalds } 2817bafc0dbaSTejun Heo 28185bdf402aSRitesh Harjani (IBM) void submit_bh(blk_opf_t opf, struct buffer_head *bh) 281971368511SDarrick J. Wong { 28205bdf402aSRitesh Harjani (IBM) submit_bh_wbc(opf, bh, NULL); 282171368511SDarrick J. Wong } 28221fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(submit_bh); 28231da177e4SLinus Torvalds 28243ae72869SBart Van Assche void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) 28259cb569d6SChristoph Hellwig { 28269cb569d6SChristoph Hellwig lock_buffer(bh); 28279cb569d6SChristoph Hellwig if (!test_clear_buffer_dirty(bh)) { 28289cb569d6SChristoph Hellwig unlock_buffer(bh); 28299cb569d6SChristoph Hellwig return; 28309cb569d6SChristoph Hellwig } 28319cb569d6SChristoph Hellwig bh->b_end_io = end_buffer_write_sync; 28329cb569d6SChristoph Hellwig get_bh(bh); 28331420c4a5SBart Van Assche submit_bh(REQ_OP_WRITE | op_flags, bh); 28349cb569d6SChristoph Hellwig } 28359cb569d6SChristoph Hellwig EXPORT_SYMBOL(write_dirty_buffer); 28369cb569d6SChristoph Hellwig 28371da177e4SLinus Torvalds /* 28381da177e4SLinus Torvalds * For a data-integrity writeout, we need to wait upon any in-progress I/O 28391da177e4SLinus Torvalds * and then start new I/O and then wait upon it. The caller must have a ref on 28401da177e4SLinus Torvalds * the buffer_head. 28411da177e4SLinus Torvalds */ 28423ae72869SBart Van Assche int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) 28431da177e4SLinus Torvalds { 28441da177e4SLinus Torvalds WARN_ON(atomic_read(&bh->b_count) < 1); 28451da177e4SLinus Torvalds lock_buffer(bh); 28461da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 2847377254b2SXianting Tian /* 2848377254b2SXianting Tian * The bh should be mapped, but it might not be if the 2849377254b2SXianting Tian * device was hot-removed. Not much we can do but fail the I/O. 2850377254b2SXianting Tian */ 2851377254b2SXianting Tian if (!buffer_mapped(bh)) { 2852377254b2SXianting Tian unlock_buffer(bh); 2853377254b2SXianting Tian return -EIO; 2854377254b2SXianting Tian } 2855377254b2SXianting Tian 28561da177e4SLinus Torvalds get_bh(bh); 28571da177e4SLinus Torvalds bh->b_end_io = end_buffer_write_sync; 2858ab620620SRitesh Harjani (IBM) submit_bh(REQ_OP_WRITE | op_flags, bh); 28591da177e4SLinus Torvalds wait_on_buffer(bh); 2860ab620620SRitesh Harjani (IBM) if (!buffer_uptodate(bh)) 2861ab620620SRitesh Harjani (IBM) return -EIO; 28621da177e4SLinus Torvalds } else { 28631da177e4SLinus Torvalds unlock_buffer(bh); 28641da177e4SLinus Torvalds } 2865ab620620SRitesh Harjani (IBM) return 0; 28661da177e4SLinus Torvalds } 286787e99511SChristoph Hellwig EXPORT_SYMBOL(__sync_dirty_buffer); 286887e99511SChristoph Hellwig 286987e99511SChristoph Hellwig int sync_dirty_buffer(struct buffer_head *bh) 287087e99511SChristoph Hellwig { 287170fd7614SChristoph Hellwig return __sync_dirty_buffer(bh, REQ_SYNC); 287287e99511SChristoph Hellwig } 28731fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(sync_dirty_buffer); 28741da177e4SLinus Torvalds 28751da177e4SLinus Torvalds /* 287668189fefSMatthew Wilcox (Oracle) * try_to_free_buffers() checks if all the buffers on this particular folio 28771da177e4SLinus Torvalds * are unused, and releases them if so. 28781da177e4SLinus Torvalds * 28791da177e4SLinus Torvalds * Exclusion against try_to_free_buffers may be obtained by either 288068189fefSMatthew Wilcox (Oracle) * locking the folio or by holding its mapping's private_lock. 28811da177e4SLinus Torvalds * 288268189fefSMatthew Wilcox (Oracle) * If the folio is dirty but all the buffers are clean then we need to 288368189fefSMatthew Wilcox (Oracle) * be sure to mark the folio clean as well. This is because the folio 28841da177e4SLinus Torvalds * may be against a block device, and a later reattachment of buffers 288568189fefSMatthew Wilcox (Oracle) * to a dirty folio will set *all* buffers dirty. Which would corrupt 28861da177e4SLinus Torvalds * filesystem data on the same device. 28871da177e4SLinus Torvalds * 288868189fefSMatthew Wilcox (Oracle) * The same applies to regular filesystem folios: if all the buffers are 288968189fefSMatthew Wilcox (Oracle) * clean then we set the folio clean and proceed. To do that, we require 2890e621900aSMatthew Wilcox (Oracle) * total exclusion from block_dirty_folio(). That is obtained with 28911da177e4SLinus Torvalds * private_lock. 28921da177e4SLinus Torvalds * 28931da177e4SLinus Torvalds * try_to_free_buffers() is non-blocking. 28941da177e4SLinus Torvalds */ 28951da177e4SLinus Torvalds static inline int buffer_busy(struct buffer_head *bh) 28961da177e4SLinus Torvalds { 28971da177e4SLinus Torvalds return atomic_read(&bh->b_count) | 28981da177e4SLinus Torvalds (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); 28991da177e4SLinus Torvalds } 29001da177e4SLinus Torvalds 290164394763SMatthew Wilcox (Oracle) static bool 290264394763SMatthew Wilcox (Oracle) drop_buffers(struct folio *folio, struct buffer_head **buffers_to_free) 29031da177e4SLinus Torvalds { 290464394763SMatthew Wilcox (Oracle) struct buffer_head *head = folio_buffers(folio); 29051da177e4SLinus Torvalds struct buffer_head *bh; 29061da177e4SLinus Torvalds 29071da177e4SLinus Torvalds bh = head; 29081da177e4SLinus Torvalds do { 29091da177e4SLinus Torvalds if (buffer_busy(bh)) 29101da177e4SLinus Torvalds goto failed; 29111da177e4SLinus Torvalds bh = bh->b_this_page; 29121da177e4SLinus Torvalds } while (bh != head); 29131da177e4SLinus Torvalds 29141da177e4SLinus Torvalds do { 29151da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 29161da177e4SLinus Torvalds 2917535ee2fbSJan Kara if (bh->b_assoc_map) 29181da177e4SLinus Torvalds __remove_assoc_queue(bh); 29191da177e4SLinus Torvalds bh = next; 29201da177e4SLinus Torvalds } while (bh != head); 29211da177e4SLinus Torvalds *buffers_to_free = head; 292264394763SMatthew Wilcox (Oracle) folio_detach_private(folio); 292364394763SMatthew Wilcox (Oracle) return true; 29241da177e4SLinus Torvalds failed: 292564394763SMatthew Wilcox (Oracle) return false; 29261da177e4SLinus Torvalds } 29271da177e4SLinus Torvalds 292868189fefSMatthew Wilcox (Oracle) bool try_to_free_buffers(struct folio *folio) 29291da177e4SLinus Torvalds { 293068189fefSMatthew Wilcox (Oracle) struct address_space * const mapping = folio->mapping; 29311da177e4SLinus Torvalds struct buffer_head *buffers_to_free = NULL; 293268189fefSMatthew Wilcox (Oracle) bool ret = 0; 29331da177e4SLinus Torvalds 293468189fefSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 293568189fefSMatthew Wilcox (Oracle) if (folio_test_writeback(folio)) 293668189fefSMatthew Wilcox (Oracle) return false; 29371da177e4SLinus Torvalds 29381da177e4SLinus Torvalds if (mapping == NULL) { /* can this still happen? */ 293964394763SMatthew Wilcox (Oracle) ret = drop_buffers(folio, &buffers_to_free); 29401da177e4SLinus Torvalds goto out; 29411da177e4SLinus Torvalds } 29421da177e4SLinus Torvalds 29431da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 294464394763SMatthew Wilcox (Oracle) ret = drop_buffers(folio, &buffers_to_free); 2945ecdfc978SLinus Torvalds 2946ecdfc978SLinus Torvalds /* 2947ecdfc978SLinus Torvalds * If the filesystem writes its buffers by hand (eg ext3) 294868189fefSMatthew Wilcox (Oracle) * then we can have clean buffers against a dirty folio. We 294968189fefSMatthew Wilcox (Oracle) * clean the folio here; otherwise the VM will never notice 2950ecdfc978SLinus Torvalds * that the filesystem did any IO at all. 2951ecdfc978SLinus Torvalds * 2952ecdfc978SLinus Torvalds * Also, during truncate, discard_buffer will have marked all 295368189fefSMatthew Wilcox (Oracle) * the folio's buffers clean. We discover that here and clean 295468189fefSMatthew Wilcox (Oracle) * the folio also. 295587df7241SNick Piggin * 295687df7241SNick Piggin * private_lock must be held over this entire operation in order 2957e621900aSMatthew Wilcox (Oracle) * to synchronise against block_dirty_folio and prevent the 295887df7241SNick Piggin * dirty bit from being lost. 2959ecdfc978SLinus Torvalds */ 296011f81becSTejun Heo if (ret) 296168189fefSMatthew Wilcox (Oracle) folio_cancel_dirty(folio); 296287df7241SNick Piggin spin_unlock(&mapping->private_lock); 29631da177e4SLinus Torvalds out: 29641da177e4SLinus Torvalds if (buffers_to_free) { 29651da177e4SLinus Torvalds struct buffer_head *bh = buffers_to_free; 29661da177e4SLinus Torvalds 29671da177e4SLinus Torvalds do { 29681da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 29691da177e4SLinus Torvalds free_buffer_head(bh); 29701da177e4SLinus Torvalds bh = next; 29711da177e4SLinus Torvalds } while (bh != buffers_to_free); 29721da177e4SLinus Torvalds } 29731da177e4SLinus Torvalds return ret; 29741da177e4SLinus Torvalds } 29751da177e4SLinus Torvalds EXPORT_SYMBOL(try_to_free_buffers); 29761da177e4SLinus Torvalds 29771da177e4SLinus Torvalds /* 29781da177e4SLinus Torvalds * Buffer-head allocation 29791da177e4SLinus Torvalds */ 298068279f9cSAlexey Dobriyan static struct kmem_cache *bh_cachep __ro_after_init; 29811da177e4SLinus Torvalds 29821da177e4SLinus Torvalds /* 29831da177e4SLinus Torvalds * Once the number of bh's in the machine exceeds this level, we start 29841da177e4SLinus Torvalds * stripping them in writeback. 29851da177e4SLinus Torvalds */ 298668279f9cSAlexey Dobriyan static unsigned long max_buffer_heads __ro_after_init; 29871da177e4SLinus Torvalds 29881da177e4SLinus Torvalds int buffer_heads_over_limit; 29891da177e4SLinus Torvalds 29901da177e4SLinus Torvalds struct bh_accounting { 29911da177e4SLinus Torvalds int nr; /* Number of live bh's */ 29921da177e4SLinus Torvalds int ratelimit; /* Limit cacheline bouncing */ 29931da177e4SLinus Torvalds }; 29941da177e4SLinus Torvalds 29951da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; 29961da177e4SLinus Torvalds 29971da177e4SLinus Torvalds static void recalc_bh_state(void) 29981da177e4SLinus Torvalds { 29991da177e4SLinus Torvalds int i; 30001da177e4SLinus Torvalds int tot = 0; 30011da177e4SLinus Torvalds 3002ee1be862SChristoph Lameter if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) 30031da177e4SLinus Torvalds return; 3004c7b92516SChristoph Lameter __this_cpu_write(bh_accounting.ratelimit, 0); 30058a143426SEric Dumazet for_each_online_cpu(i) 30061da177e4SLinus Torvalds tot += per_cpu(bh_accounting, i).nr; 30071da177e4SLinus Torvalds buffer_heads_over_limit = (tot > max_buffer_heads); 30081da177e4SLinus Torvalds } 30091da177e4SLinus Torvalds 3010dd0fc66fSAl Viro struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) 30111da177e4SLinus Torvalds { 3012019b4d12SRichard Kennedy struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); 30131da177e4SLinus Torvalds if (ret) { 3014a35afb83SChristoph Lameter INIT_LIST_HEAD(&ret->b_assoc_buffers); 3015f1e67e35SThomas Gleixner spin_lock_init(&ret->b_uptodate_lock); 3016c7b92516SChristoph Lameter preempt_disable(); 3017c7b92516SChristoph Lameter __this_cpu_inc(bh_accounting.nr); 30181da177e4SLinus Torvalds recalc_bh_state(); 3019c7b92516SChristoph Lameter preempt_enable(); 30201da177e4SLinus Torvalds } 30211da177e4SLinus Torvalds return ret; 30221da177e4SLinus Torvalds } 30231da177e4SLinus Torvalds EXPORT_SYMBOL(alloc_buffer_head); 30241da177e4SLinus Torvalds 30251da177e4SLinus Torvalds void free_buffer_head(struct buffer_head *bh) 30261da177e4SLinus Torvalds { 30271da177e4SLinus Torvalds BUG_ON(!list_empty(&bh->b_assoc_buffers)); 30281da177e4SLinus Torvalds kmem_cache_free(bh_cachep, bh); 3029c7b92516SChristoph Lameter preempt_disable(); 3030c7b92516SChristoph Lameter __this_cpu_dec(bh_accounting.nr); 30311da177e4SLinus Torvalds recalc_bh_state(); 3032c7b92516SChristoph Lameter preempt_enable(); 30331da177e4SLinus Torvalds } 30341da177e4SLinus Torvalds EXPORT_SYMBOL(free_buffer_head); 30351da177e4SLinus Torvalds 3036fc4d24c9SSebastian Andrzej Siewior static int buffer_exit_cpu_dead(unsigned int cpu) 30371da177e4SLinus Torvalds { 30381da177e4SLinus Torvalds int i; 30391da177e4SLinus Torvalds struct bh_lru *b = &per_cpu(bh_lrus, cpu); 30401da177e4SLinus Torvalds 30411da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 30421da177e4SLinus Torvalds brelse(b->bhs[i]); 30431da177e4SLinus Torvalds b->bhs[i] = NULL; 30441da177e4SLinus Torvalds } 3045c7b92516SChristoph Lameter this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); 30468a143426SEric Dumazet per_cpu(bh_accounting, cpu).nr = 0; 3047fc4d24c9SSebastian Andrzej Siewior return 0; 30481da177e4SLinus Torvalds } 30491da177e4SLinus Torvalds 3050389d1b08SAneesh Kumar K.V /** 3051a6b91919SRandy Dunlap * bh_uptodate_or_lock - Test whether the buffer is uptodate 3052389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3053389d1b08SAneesh Kumar K.V * 3054389d1b08SAneesh Kumar K.V * Return true if the buffer is up-to-date and false, 3055389d1b08SAneesh Kumar K.V * with the buffer locked, if not. 3056389d1b08SAneesh Kumar K.V */ 3057389d1b08SAneesh Kumar K.V int bh_uptodate_or_lock(struct buffer_head *bh) 3058389d1b08SAneesh Kumar K.V { 3059389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) { 3060389d1b08SAneesh Kumar K.V lock_buffer(bh); 3061389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) 3062389d1b08SAneesh Kumar K.V return 0; 3063389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3064389d1b08SAneesh Kumar K.V } 3065389d1b08SAneesh Kumar K.V return 1; 3066389d1b08SAneesh Kumar K.V } 3067389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_uptodate_or_lock); 3068389d1b08SAneesh Kumar K.V 3069389d1b08SAneesh Kumar K.V /** 3070fdee117eSZhang Yi * __bh_read - Submit read for a locked buffer 3071389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3072fdee117eSZhang Yi * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ 3073fdee117eSZhang Yi * @wait: wait until reading finish 3074389d1b08SAneesh Kumar K.V * 3075fdee117eSZhang Yi * Returns zero on success or don't wait, and -EIO on error. 3076389d1b08SAneesh Kumar K.V */ 3077fdee117eSZhang Yi int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait) 3078389d1b08SAneesh Kumar K.V { 3079fdee117eSZhang Yi int ret = 0; 3080389d1b08SAneesh Kumar K.V 3081fdee117eSZhang Yi BUG_ON(!buffer_locked(bh)); 3082389d1b08SAneesh Kumar K.V 3083389d1b08SAneesh Kumar K.V get_bh(bh); 3084389d1b08SAneesh Kumar K.V bh->b_end_io = end_buffer_read_sync; 3085fdee117eSZhang Yi submit_bh(REQ_OP_READ | op_flags, bh); 3086fdee117eSZhang Yi if (wait) { 3087389d1b08SAneesh Kumar K.V wait_on_buffer(bh); 3088fdee117eSZhang Yi if (!buffer_uptodate(bh)) 3089fdee117eSZhang Yi ret = -EIO; 3090389d1b08SAneesh Kumar K.V } 3091fdee117eSZhang Yi return ret; 3092fdee117eSZhang Yi } 3093fdee117eSZhang Yi EXPORT_SYMBOL(__bh_read); 3094fdee117eSZhang Yi 3095fdee117eSZhang Yi /** 3096fdee117eSZhang Yi * __bh_read_batch - Submit read for a batch of unlocked buffers 3097fdee117eSZhang Yi * @nr: entry number of the buffer batch 3098fdee117eSZhang Yi * @bhs: a batch of struct buffer_head 3099fdee117eSZhang Yi * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ 3100fdee117eSZhang Yi * @force_lock: force to get a lock on the buffer if set, otherwise drops any 3101fdee117eSZhang Yi * buffer that cannot lock. 3102fdee117eSZhang Yi * 3103fdee117eSZhang Yi * Returns zero on success or don't wait, and -EIO on error. 3104fdee117eSZhang Yi */ 3105fdee117eSZhang Yi void __bh_read_batch(int nr, struct buffer_head *bhs[], 3106fdee117eSZhang Yi blk_opf_t op_flags, bool force_lock) 3107fdee117eSZhang Yi { 3108fdee117eSZhang Yi int i; 3109fdee117eSZhang Yi 3110fdee117eSZhang Yi for (i = 0; i < nr; i++) { 3111fdee117eSZhang Yi struct buffer_head *bh = bhs[i]; 3112fdee117eSZhang Yi 3113fdee117eSZhang Yi if (buffer_uptodate(bh)) 3114fdee117eSZhang Yi continue; 3115fdee117eSZhang Yi 3116fdee117eSZhang Yi if (force_lock) 3117fdee117eSZhang Yi lock_buffer(bh); 3118fdee117eSZhang Yi else 3119fdee117eSZhang Yi if (!trylock_buffer(bh)) 3120fdee117eSZhang Yi continue; 3121fdee117eSZhang Yi 3122fdee117eSZhang Yi if (buffer_uptodate(bh)) { 3123fdee117eSZhang Yi unlock_buffer(bh); 3124fdee117eSZhang Yi continue; 3125fdee117eSZhang Yi } 3126fdee117eSZhang Yi 3127fdee117eSZhang Yi bh->b_end_io = end_buffer_read_sync; 3128fdee117eSZhang Yi get_bh(bh); 3129fdee117eSZhang Yi submit_bh(REQ_OP_READ | op_flags, bh); 3130fdee117eSZhang Yi } 3131fdee117eSZhang Yi } 3132fdee117eSZhang Yi EXPORT_SYMBOL(__bh_read_batch); 3133389d1b08SAneesh Kumar K.V 31341da177e4SLinus Torvalds void __init buffer_init(void) 31351da177e4SLinus Torvalds { 313643be594aSZhang Yanfei unsigned long nrpages; 3137fc4d24c9SSebastian Andrzej Siewior int ret; 31381da177e4SLinus Torvalds 3139b98938c3SChristoph Lameter bh_cachep = kmem_cache_create("buffer_head", 3140b98938c3SChristoph Lameter sizeof(struct buffer_head), 0, 3141b98938c3SChristoph Lameter (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| 3142b98938c3SChristoph Lameter SLAB_MEM_SPREAD), 3143019b4d12SRichard Kennedy NULL); 31441da177e4SLinus Torvalds 31451da177e4SLinus Torvalds /* 31461da177e4SLinus Torvalds * Limit the bh occupancy to 10% of ZONE_NORMAL 31471da177e4SLinus Torvalds */ 31481da177e4SLinus Torvalds nrpages = (nr_free_buffer_pages() * 10) / 100; 31491da177e4SLinus Torvalds max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); 3150fc4d24c9SSebastian Andrzej Siewior ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", 3151fc4d24c9SSebastian Andrzej Siewior NULL, buffer_exit_cpu_dead); 3152fc4d24c9SSebastian Andrzej Siewior WARN_ON(ret < 0); 31531da177e4SLinus Torvalds } 3154