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 183600f111eSMatthew Wilcox (Oracle) * i_private_lock. 1841da177e4SLinus Torvalds * 185600f111eSMatthew Wilcox (Oracle) * Hack idea: for the blockdev mapping, i_private_lock contention 1861da177e4SLinus Torvalds * may be quite high. This code could TryLock the page, and if that 187600f111eSMatthew Wilcox (Oracle) * succeeds, there is no need to take i_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 2024b04646cSMatthew Wilcox (Oracle) index = ((loff_t)block << bd_inode->i_blkbits) / PAGE_SIZE; 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 207600f111eSMatthew Wilcox (Oracle) spin_lock(&bd_mapping->i_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: 239600f111eSMatthew Wilcox (Oracle) spin_unlock(&bd_mapping->i_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 /* 37514059f66SMatthew Wilcox (Oracle) * Completion handler for block_write_full_folio() - folios which are unlocked 37614059f66SMatthew Wilcox (Oracle) * during I/O, and which have the writeback flag cleared upon I/O completion. 3771da177e4SLinus Torvalds */ 37814059f66SMatthew Wilcox (Oracle) static 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 } 4181da177e4SLinus Torvalds 4191da177e4SLinus Torvalds /* 4201da177e4SLinus Torvalds * If a page's buffers are under async readin (end_buffer_async_read 4211da177e4SLinus Torvalds * completion) then there is a possibility that another thread of 4221da177e4SLinus Torvalds * control could lock one of the buffers after it has completed 4231da177e4SLinus Torvalds * but while some of the other buffers have not completed. This 4241da177e4SLinus Torvalds * locked buffer would confuse end_buffer_async_read() into not unlocking 4251da177e4SLinus Torvalds * the page. So the absence of BH_Async_Read tells end_buffer_async_read() 4261da177e4SLinus Torvalds * that this buffer is not under async I/O. 4271da177e4SLinus Torvalds * 4281da177e4SLinus Torvalds * The page comes unlocked when it has no locked buffer_async buffers 4291da177e4SLinus Torvalds * left. 4301da177e4SLinus Torvalds * 4311da177e4SLinus Torvalds * PageLocked prevents anyone starting new async I/O reads any of 4321da177e4SLinus Torvalds * the buffers. 4331da177e4SLinus Torvalds * 4341da177e4SLinus Torvalds * PageWriteback is used to prevent simultaneous writeout of the same 4351da177e4SLinus Torvalds * page. 4361da177e4SLinus Torvalds * 4371da177e4SLinus Torvalds * PageLocked prevents anyone from starting writeback of a page which is 4381da177e4SLinus Torvalds * under read I/O (PageWriteback is only ever set against a locked page). 4391da177e4SLinus Torvalds */ 4401da177e4SLinus Torvalds static void mark_buffer_async_read(struct buffer_head *bh) 4411da177e4SLinus Torvalds { 44231fb992cSEric Biggers bh->b_end_io = end_buffer_async_read_io; 4431da177e4SLinus Torvalds set_buffer_async_read(bh); 4441da177e4SLinus Torvalds } 4451da177e4SLinus Torvalds 4461fe72eaaSH Hartley Sweeten static void mark_buffer_async_write_endio(struct buffer_head *bh, 44735c80d5fSChris Mason bh_end_io_t *handler) 44835c80d5fSChris Mason { 44935c80d5fSChris Mason bh->b_end_io = handler; 45035c80d5fSChris Mason set_buffer_async_write(bh); 45135c80d5fSChris Mason } 45235c80d5fSChris Mason 4531da177e4SLinus Torvalds void mark_buffer_async_write(struct buffer_head *bh) 4541da177e4SLinus Torvalds { 45535c80d5fSChris Mason mark_buffer_async_write_endio(bh, end_buffer_async_write); 4561da177e4SLinus Torvalds } 4571da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_async_write); 4581da177e4SLinus Torvalds 4591da177e4SLinus Torvalds 4601da177e4SLinus Torvalds /* 4611da177e4SLinus Torvalds * fs/buffer.c contains helper functions for buffer-backed address space's 4621da177e4SLinus Torvalds * fsync functions. A common requirement for buffer-based filesystems is 4631da177e4SLinus Torvalds * that certain data from the backing blockdev needs to be written out for 4641da177e4SLinus Torvalds * a successful fsync(). For example, ext2 indirect blocks need to be 4651da177e4SLinus Torvalds * written back and waited upon before fsync() returns. 4661da177e4SLinus Torvalds * 467*73f65b8bSAndreas Gruenbacher * The functions mark_buffer_dirty_inode(), fsync_inode_buffers(), 4681da177e4SLinus Torvalds * inode_has_buffers() and invalidate_inode_buffers() are provided for the 469600f111eSMatthew Wilcox (Oracle) * management of a list of dependent buffers at ->i_mapping->i_private_list. 4701da177e4SLinus Torvalds * 4711da177e4SLinus Torvalds * Locking is a little subtle: try_to_free_buffers() will remove buffers 4721da177e4SLinus Torvalds * from their controlling inode's queue when they are being freed. But 4731da177e4SLinus Torvalds * try_to_free_buffers() will be operating against the *blockdev* mapping 4741da177e4SLinus Torvalds * at the time, not against the S_ISREG file which depends on those buffers. 475600f111eSMatthew Wilcox (Oracle) * So the locking for i_private_list is via the i_private_lock in the address_space 4761da177e4SLinus Torvalds * which backs the buffers. Which is different from the address_space 4771da177e4SLinus Torvalds * against which the buffers are listed. So for a particular address_space, 478600f111eSMatthew Wilcox (Oracle) * mapping->i_private_lock does *not* protect mapping->i_private_list! In fact, 479600f111eSMatthew Wilcox (Oracle) * mapping->i_private_list will always be protected by the backing blockdev's 480600f111eSMatthew Wilcox (Oracle) * ->i_private_lock. 4811da177e4SLinus Torvalds * 4821da177e4SLinus Torvalds * Which introduces a requirement: all buffers on an address_space's 483600f111eSMatthew Wilcox (Oracle) * ->i_private_list must be from the same address_space: the blockdev's. 4841da177e4SLinus Torvalds * 485600f111eSMatthew Wilcox (Oracle) * address_spaces which do not place buffers at ->i_private_list via these 486600f111eSMatthew Wilcox (Oracle) * utility functions are free to use i_private_lock and i_private_list for 487600f111eSMatthew Wilcox (Oracle) * whatever they want. The only requirement is that list_empty(i_private_list) 4881da177e4SLinus Torvalds * be true at clear_inode() time. 4891da177e4SLinus Torvalds * 4901da177e4SLinus Torvalds * FIXME: clear_inode should not call invalidate_inode_buffers(). The 4911da177e4SLinus Torvalds * filesystems should do that. invalidate_inode_buffers() should just go 4921da177e4SLinus Torvalds * BUG_ON(!list_empty). 4931da177e4SLinus Torvalds * 4941da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should 4951da177e4SLinus Torvalds * take an address_space, not an inode. And it should be called 4961da177e4SLinus Torvalds * mark_buffer_dirty_fsync() to clearly define why those buffers are being 4971da177e4SLinus Torvalds * queued up. 4981da177e4SLinus Torvalds * 4991da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the 5001da177e4SLinus Torvalds * list if it is already on a list. Because if the buffer is on a list, 5011da177e4SLinus Torvalds * it *must* already be on the right one. If not, the filesystem is being 5021da177e4SLinus Torvalds * silly. This will save a ton of locking. But first we have to ensure 5031da177e4SLinus Torvalds * that buffers are taken *off* the old inode's list when they are freed 5041da177e4SLinus Torvalds * (presumably in truncate). That requires careful auditing of all 5051da177e4SLinus Torvalds * filesystems (do it inside bforget()). It could also be done by bringing 5061da177e4SLinus Torvalds * b_inode back. 5071da177e4SLinus Torvalds */ 5081da177e4SLinus Torvalds 5091da177e4SLinus Torvalds /* 510600f111eSMatthew Wilcox (Oracle) * The buffer's backing address_space's i_private_lock must be held 5111da177e4SLinus Torvalds */ 512dbacefc9SThomas Petazzoni static void __remove_assoc_queue(struct buffer_head *bh) 5131da177e4SLinus Torvalds { 5141da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 51558ff407bSJan Kara WARN_ON(!bh->b_assoc_map); 51658ff407bSJan Kara bh->b_assoc_map = NULL; 5171da177e4SLinus Torvalds } 5181da177e4SLinus Torvalds 5191da177e4SLinus Torvalds int inode_has_buffers(struct inode *inode) 5201da177e4SLinus Torvalds { 521600f111eSMatthew Wilcox (Oracle) return !list_empty(&inode->i_data.i_private_list); 5221da177e4SLinus Torvalds } 5231da177e4SLinus Torvalds 5241da177e4SLinus Torvalds /* 5251da177e4SLinus Torvalds * osync is designed to support O_SYNC io. It waits synchronously for 5261da177e4SLinus Torvalds * all already-submitted IO to complete, but does not queue any new 5271da177e4SLinus Torvalds * writes to the disk. 5281da177e4SLinus Torvalds * 52979f59784SZhang Yi * To do O_SYNC writes, just queue the buffer writes with write_dirty_buffer 53079f59784SZhang Yi * as you dirty the buffers, and then use osync_inode_buffers to wait for 5311da177e4SLinus Torvalds * completion. Any other dirty buffers which are not yet queued for 5321da177e4SLinus Torvalds * write will not be flushed to disk by the osync. 5331da177e4SLinus Torvalds */ 5341da177e4SLinus Torvalds static int osync_buffers_list(spinlock_t *lock, struct list_head *list) 5351da177e4SLinus Torvalds { 5361da177e4SLinus Torvalds struct buffer_head *bh; 5371da177e4SLinus Torvalds struct list_head *p; 5381da177e4SLinus Torvalds int err = 0; 5391da177e4SLinus Torvalds 5401da177e4SLinus Torvalds spin_lock(lock); 5411da177e4SLinus Torvalds repeat: 5421da177e4SLinus Torvalds list_for_each_prev(p, list) { 5431da177e4SLinus Torvalds bh = BH_ENTRY(p); 5441da177e4SLinus Torvalds if (buffer_locked(bh)) { 5451da177e4SLinus Torvalds get_bh(bh); 5461da177e4SLinus Torvalds spin_unlock(lock); 5471da177e4SLinus Torvalds wait_on_buffer(bh); 5481da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 5491da177e4SLinus Torvalds err = -EIO; 5501da177e4SLinus Torvalds brelse(bh); 5511da177e4SLinus Torvalds spin_lock(lock); 5521da177e4SLinus Torvalds goto repeat; 5531da177e4SLinus Torvalds } 5541da177e4SLinus Torvalds } 5551da177e4SLinus Torvalds spin_unlock(lock); 5561da177e4SLinus Torvalds return err; 5571da177e4SLinus Torvalds } 5581da177e4SLinus Torvalds 5591da177e4SLinus Torvalds /** 56078a4a50aSRandy Dunlap * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers 56167be2dd1SMartin Waitz * @mapping: the mapping which wants those buffers written 5621da177e4SLinus Torvalds * 563600f111eSMatthew Wilcox (Oracle) * Starts I/O against the buffers at mapping->i_private_list, and waits upon 5641da177e4SLinus Torvalds * that I/O. 5651da177e4SLinus Torvalds * 56667be2dd1SMartin Waitz * Basically, this is a convenience function for fsync(). 56767be2dd1SMartin Waitz * @mapping is a file or directory which needs those buffers to be written for 56867be2dd1SMartin Waitz * a successful fsync(). 5691da177e4SLinus Torvalds */ 5701da177e4SLinus Torvalds int sync_mapping_buffers(struct address_space *mapping) 5711da177e4SLinus Torvalds { 572600f111eSMatthew Wilcox (Oracle) struct address_space *buffer_mapping = mapping->i_private_data; 5731da177e4SLinus Torvalds 574600f111eSMatthew Wilcox (Oracle) if (buffer_mapping == NULL || list_empty(&mapping->i_private_list)) 5751da177e4SLinus Torvalds return 0; 5761da177e4SLinus Torvalds 577600f111eSMatthew Wilcox (Oracle) return fsync_buffers_list(&buffer_mapping->i_private_lock, 578600f111eSMatthew Wilcox (Oracle) &mapping->i_private_list); 5791da177e4SLinus Torvalds } 5801da177e4SLinus Torvalds EXPORT_SYMBOL(sync_mapping_buffers); 5811da177e4SLinus Torvalds 58231b2ebc0SRitesh Harjani (IBM) /** 58331b2ebc0SRitesh Harjani (IBM) * generic_buffers_fsync_noflush - generic buffer fsync implementation 58431b2ebc0SRitesh Harjani (IBM) * for simple filesystems with no inode lock 58531b2ebc0SRitesh Harjani (IBM) * 58631b2ebc0SRitesh Harjani (IBM) * @file: file to synchronize 58731b2ebc0SRitesh Harjani (IBM) * @start: start offset in bytes 58831b2ebc0SRitesh Harjani (IBM) * @end: end offset in bytes (inclusive) 58931b2ebc0SRitesh Harjani (IBM) * @datasync: only synchronize essential metadata if true 59031b2ebc0SRitesh Harjani (IBM) * 59131b2ebc0SRitesh Harjani (IBM) * This is a generic implementation of the fsync method for simple 59231b2ebc0SRitesh Harjani (IBM) * filesystems which track all non-inode metadata in the buffers list 59331b2ebc0SRitesh Harjani (IBM) * hanging off the address_space structure. 59431b2ebc0SRitesh Harjani (IBM) */ 59531b2ebc0SRitesh Harjani (IBM) int generic_buffers_fsync_noflush(struct file *file, loff_t start, loff_t end, 59631b2ebc0SRitesh Harjani (IBM) bool datasync) 59731b2ebc0SRitesh Harjani (IBM) { 59831b2ebc0SRitesh Harjani (IBM) struct inode *inode = file->f_mapping->host; 59931b2ebc0SRitesh Harjani (IBM) int err; 60031b2ebc0SRitesh Harjani (IBM) int ret; 60131b2ebc0SRitesh Harjani (IBM) 60231b2ebc0SRitesh Harjani (IBM) err = file_write_and_wait_range(file, start, end); 60331b2ebc0SRitesh Harjani (IBM) if (err) 60431b2ebc0SRitesh Harjani (IBM) return err; 60531b2ebc0SRitesh Harjani (IBM) 60631b2ebc0SRitesh Harjani (IBM) ret = sync_mapping_buffers(inode->i_mapping); 60731b2ebc0SRitesh Harjani (IBM) if (!(inode->i_state & I_DIRTY_ALL)) 60831b2ebc0SRitesh Harjani (IBM) goto out; 60931b2ebc0SRitesh Harjani (IBM) if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) 61031b2ebc0SRitesh Harjani (IBM) goto out; 61131b2ebc0SRitesh Harjani (IBM) 61231b2ebc0SRitesh Harjani (IBM) err = sync_inode_metadata(inode, 1); 61331b2ebc0SRitesh Harjani (IBM) if (ret == 0) 61431b2ebc0SRitesh Harjani (IBM) ret = err; 61531b2ebc0SRitesh Harjani (IBM) 61631b2ebc0SRitesh Harjani (IBM) out: 61731b2ebc0SRitesh Harjani (IBM) /* check and advance again to catch errors after syncing out buffers */ 61831b2ebc0SRitesh Harjani (IBM) err = file_check_and_advance_wb_err(file); 61931b2ebc0SRitesh Harjani (IBM) if (ret == 0) 62031b2ebc0SRitesh Harjani (IBM) ret = err; 62131b2ebc0SRitesh Harjani (IBM) return ret; 62231b2ebc0SRitesh Harjani (IBM) } 62331b2ebc0SRitesh Harjani (IBM) EXPORT_SYMBOL(generic_buffers_fsync_noflush); 62431b2ebc0SRitesh Harjani (IBM) 62531b2ebc0SRitesh Harjani (IBM) /** 62631b2ebc0SRitesh Harjani (IBM) * generic_buffers_fsync - generic buffer fsync implementation 62731b2ebc0SRitesh Harjani (IBM) * for simple filesystems with no inode lock 62831b2ebc0SRitesh Harjani (IBM) * 62931b2ebc0SRitesh Harjani (IBM) * @file: file to synchronize 63031b2ebc0SRitesh Harjani (IBM) * @start: start offset in bytes 63131b2ebc0SRitesh Harjani (IBM) * @end: end offset in bytes (inclusive) 63231b2ebc0SRitesh Harjani (IBM) * @datasync: only synchronize essential metadata if true 63331b2ebc0SRitesh Harjani (IBM) * 63431b2ebc0SRitesh Harjani (IBM) * This is a generic implementation of the fsync method for simple 63531b2ebc0SRitesh Harjani (IBM) * filesystems which track all non-inode metadata in the buffers list 63631b2ebc0SRitesh Harjani (IBM) * hanging off the address_space structure. This also makes sure that 63731b2ebc0SRitesh Harjani (IBM) * a device cache flush operation is called at the end. 63831b2ebc0SRitesh Harjani (IBM) */ 63931b2ebc0SRitesh Harjani (IBM) int generic_buffers_fsync(struct file *file, loff_t start, loff_t end, 64031b2ebc0SRitesh Harjani (IBM) bool datasync) 64131b2ebc0SRitesh Harjani (IBM) { 64231b2ebc0SRitesh Harjani (IBM) struct inode *inode = file->f_mapping->host; 64331b2ebc0SRitesh Harjani (IBM) int ret; 64431b2ebc0SRitesh Harjani (IBM) 64531b2ebc0SRitesh Harjani (IBM) ret = generic_buffers_fsync_noflush(file, start, end, datasync); 64631b2ebc0SRitesh Harjani (IBM) if (!ret) 64731b2ebc0SRitesh Harjani (IBM) ret = blkdev_issue_flush(inode->i_sb->s_bdev); 64831b2ebc0SRitesh Harjani (IBM) return ret; 64931b2ebc0SRitesh Harjani (IBM) } 65031b2ebc0SRitesh Harjani (IBM) EXPORT_SYMBOL(generic_buffers_fsync); 65131b2ebc0SRitesh Harjani (IBM) 6521da177e4SLinus Torvalds /* 6531da177e4SLinus Torvalds * Called when we've recently written block `bblock', and it is known that 6541da177e4SLinus Torvalds * `bblock' was for a buffer_boundary() buffer. This means that the block at 6551da177e4SLinus Torvalds * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's 6561da177e4SLinus Torvalds * dirty, schedule it for IO. So that indirects merge nicely with their data. 6571da177e4SLinus Torvalds */ 6581da177e4SLinus Torvalds void write_boundary_block(struct block_device *bdev, 6591da177e4SLinus Torvalds sector_t bblock, unsigned blocksize) 6601da177e4SLinus Torvalds { 6611da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); 6621da177e4SLinus Torvalds if (bh) { 6631da177e4SLinus Torvalds if (buffer_dirty(bh)) 664e7ea1129SZhang Yi write_dirty_buffer(bh, 0); 6651da177e4SLinus Torvalds put_bh(bh); 6661da177e4SLinus Torvalds } 6671da177e4SLinus Torvalds } 6681da177e4SLinus Torvalds 6691da177e4SLinus Torvalds void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) 6701da177e4SLinus Torvalds { 6711da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 672abc8a8a2SMatthew Wilcox (Oracle) struct address_space *buffer_mapping = bh->b_folio->mapping; 6731da177e4SLinus Torvalds 6741da177e4SLinus Torvalds mark_buffer_dirty(bh); 675600f111eSMatthew Wilcox (Oracle) if (!mapping->i_private_data) { 676600f111eSMatthew Wilcox (Oracle) mapping->i_private_data = buffer_mapping; 6771da177e4SLinus Torvalds } else { 678600f111eSMatthew Wilcox (Oracle) BUG_ON(mapping->i_private_data != buffer_mapping); 6791da177e4SLinus Torvalds } 680535ee2fbSJan Kara if (!bh->b_assoc_map) { 681600f111eSMatthew Wilcox (Oracle) spin_lock(&buffer_mapping->i_private_lock); 6821da177e4SLinus Torvalds list_move_tail(&bh->b_assoc_buffers, 683600f111eSMatthew Wilcox (Oracle) &mapping->i_private_list); 68458ff407bSJan Kara bh->b_assoc_map = mapping; 685600f111eSMatthew Wilcox (Oracle) spin_unlock(&buffer_mapping->i_private_lock); 6861da177e4SLinus Torvalds } 6871da177e4SLinus Torvalds } 6881da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_dirty_inode); 6891da177e4SLinus Torvalds 6901da177e4SLinus Torvalds /* 6911da177e4SLinus Torvalds * Add a page to the dirty page list. 6921da177e4SLinus Torvalds * 6931da177e4SLinus Torvalds * It is a sad fact of life that this function is called from several places 6941da177e4SLinus Torvalds * deeply under spinlocking. It may not sleep. 6951da177e4SLinus Torvalds * 6961da177e4SLinus Torvalds * If the page has buffers, the uptodate buffers are set dirty, to preserve 6971da177e4SLinus Torvalds * dirty-state coherency between the page and the buffers. It the page does 6981da177e4SLinus Torvalds * not have buffers then when they are later attached they will all be set 6991da177e4SLinus Torvalds * dirty. 7001da177e4SLinus Torvalds * 7011da177e4SLinus Torvalds * The buffers are dirtied before the page is dirtied. There's a small race 7021da177e4SLinus Torvalds * window in which a writepage caller may see the page cleanness but not the 7031da177e4SLinus Torvalds * buffer dirtiness. That's fine. If this code were to set the page dirty 7041da177e4SLinus Torvalds * before the buffers, a concurrent writepage caller could clear the page dirty 7051da177e4SLinus Torvalds * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean 7061da177e4SLinus Torvalds * page on the dirty page list. 7071da177e4SLinus Torvalds * 708600f111eSMatthew Wilcox (Oracle) * We use i_private_lock to lock against try_to_free_buffers while using the 7091da177e4SLinus Torvalds * page's buffer list. Also use this to protect against clean buffers being 7101da177e4SLinus Torvalds * added to the page after it was set dirty. 7111da177e4SLinus Torvalds * 7121da177e4SLinus Torvalds * FIXME: may need to call ->reservepage here as well. That's rather up to the 7131da177e4SLinus Torvalds * address_space though. 7141da177e4SLinus Torvalds */ 715e621900aSMatthew Wilcox (Oracle) bool block_dirty_folio(struct address_space *mapping, struct folio *folio) 7161da177e4SLinus Torvalds { 717e621900aSMatthew Wilcox (Oracle) struct buffer_head *head; 718e621900aSMatthew Wilcox (Oracle) bool newly_dirty; 7191da177e4SLinus Torvalds 720600f111eSMatthew Wilcox (Oracle) spin_lock(&mapping->i_private_lock); 721e621900aSMatthew Wilcox (Oracle) head = folio_buffers(folio); 722e621900aSMatthew Wilcox (Oracle) if (head) { 7231da177e4SLinus Torvalds struct buffer_head *bh = head; 7241da177e4SLinus Torvalds 7251da177e4SLinus Torvalds do { 7261da177e4SLinus Torvalds set_buffer_dirty(bh); 7271da177e4SLinus Torvalds bh = bh->b_this_page; 7281da177e4SLinus Torvalds } while (bh != head); 7291da177e4SLinus Torvalds } 730c4843a75SGreg Thelen /* 731bcfe06bfSRoman Gushchin * Lock out page's memcg migration to keep PageDirty 73281f8c3a4SJohannes Weiner * synchronized with per-memcg dirty page counters. 733c4843a75SGreg Thelen */ 734e621900aSMatthew Wilcox (Oracle) folio_memcg_lock(folio); 735e621900aSMatthew Wilcox (Oracle) newly_dirty = !folio_test_set_dirty(folio); 736600f111eSMatthew Wilcox (Oracle) spin_unlock(&mapping->i_private_lock); 7371da177e4SLinus Torvalds 738a8e7d49aSLinus Torvalds if (newly_dirty) 739e621900aSMatthew Wilcox (Oracle) __folio_mark_dirty(folio, mapping, 1); 740c4843a75SGreg Thelen 741e621900aSMatthew Wilcox (Oracle) folio_memcg_unlock(folio); 742c4843a75SGreg Thelen 743c4843a75SGreg Thelen if (newly_dirty) 744c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 745c4843a75SGreg Thelen 746a8e7d49aSLinus Torvalds return newly_dirty; 7471da177e4SLinus Torvalds } 748e621900aSMatthew Wilcox (Oracle) EXPORT_SYMBOL(block_dirty_folio); 7491da177e4SLinus Torvalds 7501da177e4SLinus Torvalds /* 7511da177e4SLinus Torvalds * Write out and wait upon a list of buffers. 7521da177e4SLinus Torvalds * 7531da177e4SLinus Torvalds * We have conflicting pressures: we want to make sure that all 7541da177e4SLinus Torvalds * initially dirty buffers get waited on, but that any subsequently 7551da177e4SLinus Torvalds * dirtied buffers don't. After all, we don't want fsync to last 7561da177e4SLinus Torvalds * forever if somebody is actively writing to the file. 7571da177e4SLinus Torvalds * 7581da177e4SLinus Torvalds * Do this in two main stages: first we copy dirty buffers to a 7591da177e4SLinus Torvalds * temporary inode list, queueing the writes as we go. Then we clean 7601da177e4SLinus Torvalds * up, waiting for those writes to complete. 7611da177e4SLinus Torvalds * 7621da177e4SLinus Torvalds * During this second stage, any subsequent updates to the file may end 7631da177e4SLinus Torvalds * up refiling the buffer on the original inode's dirty list again, so 7641da177e4SLinus Torvalds * there is a chance we will end up with a buffer queued for write but 7651da177e4SLinus Torvalds * not yet completed on that list. So, as a final cleanup we go through 7661da177e4SLinus Torvalds * the osync code to catch these locked, dirty buffers without requeuing 7671da177e4SLinus Torvalds * any newly dirty buffers for write. 7681da177e4SLinus Torvalds */ 7691da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) 7701da177e4SLinus Torvalds { 7711da177e4SLinus Torvalds struct buffer_head *bh; 7721da177e4SLinus Torvalds struct list_head tmp; 7737eaceaccSJens Axboe struct address_space *mapping; 7741da177e4SLinus Torvalds int err = 0, err2; 7754ee2491eSJens Axboe struct blk_plug plug; 7761da177e4SLinus Torvalds 7771da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 7784ee2491eSJens Axboe blk_start_plug(&plug); 7791da177e4SLinus Torvalds 7801da177e4SLinus Torvalds spin_lock(lock); 7811da177e4SLinus Torvalds while (!list_empty(list)) { 7821da177e4SLinus Torvalds bh = BH_ENTRY(list->next); 783535ee2fbSJan Kara mapping = bh->b_assoc_map; 78458ff407bSJan Kara __remove_assoc_queue(bh); 785535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 786535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 787535ee2fbSJan Kara smp_mb(); 7881da177e4SLinus Torvalds if (buffer_dirty(bh) || buffer_locked(bh)) { 7891da177e4SLinus Torvalds list_add(&bh->b_assoc_buffers, &tmp); 790535ee2fbSJan Kara bh->b_assoc_map = mapping; 7911da177e4SLinus Torvalds if (buffer_dirty(bh)) { 7921da177e4SLinus Torvalds get_bh(bh); 7931da177e4SLinus Torvalds spin_unlock(lock); 7941da177e4SLinus Torvalds /* 7951da177e4SLinus Torvalds * Ensure any pending I/O completes so that 7969cb569d6SChristoph Hellwig * write_dirty_buffer() actually writes the 7979cb569d6SChristoph Hellwig * current contents - it is a noop if I/O is 7989cb569d6SChristoph Hellwig * still in flight on potentially older 7999cb569d6SChristoph Hellwig * contents. 8001da177e4SLinus Torvalds */ 80170fd7614SChristoph Hellwig write_dirty_buffer(bh, REQ_SYNC); 8029cf6b720SJens Axboe 8039cf6b720SJens Axboe /* 8049cf6b720SJens Axboe * Kick off IO for the previous mapping. Note 8059cf6b720SJens Axboe * that we will not run the very last mapping, 8069cf6b720SJens Axboe * wait_on_buffer() will do that for us 8079cf6b720SJens Axboe * through sync_buffer(). 8089cf6b720SJens Axboe */ 8091da177e4SLinus Torvalds brelse(bh); 8101da177e4SLinus Torvalds spin_lock(lock); 8111da177e4SLinus Torvalds } 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds } 8141da177e4SLinus Torvalds 8154ee2491eSJens Axboe spin_unlock(lock); 8164ee2491eSJens Axboe blk_finish_plug(&plug); 8174ee2491eSJens Axboe spin_lock(lock); 8184ee2491eSJens Axboe 8191da177e4SLinus Torvalds while (!list_empty(&tmp)) { 8201da177e4SLinus Torvalds bh = BH_ENTRY(tmp.prev); 8211da177e4SLinus Torvalds get_bh(bh); 822535ee2fbSJan Kara mapping = bh->b_assoc_map; 823535ee2fbSJan Kara __remove_assoc_queue(bh); 824535ee2fbSJan Kara /* Avoid race with mark_buffer_dirty_inode() which does 825535ee2fbSJan Kara * a lockless check and we rely on seeing the dirty bit */ 826535ee2fbSJan Kara smp_mb(); 827535ee2fbSJan Kara if (buffer_dirty(bh)) { 828535ee2fbSJan Kara list_add(&bh->b_assoc_buffers, 829600f111eSMatthew Wilcox (Oracle) &mapping->i_private_list); 830535ee2fbSJan Kara bh->b_assoc_map = mapping; 831535ee2fbSJan Kara } 8321da177e4SLinus Torvalds spin_unlock(lock); 8331da177e4SLinus Torvalds wait_on_buffer(bh); 8341da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 8351da177e4SLinus Torvalds err = -EIO; 8361da177e4SLinus Torvalds brelse(bh); 8371da177e4SLinus Torvalds spin_lock(lock); 8381da177e4SLinus Torvalds } 8391da177e4SLinus Torvalds 8401da177e4SLinus Torvalds spin_unlock(lock); 8411da177e4SLinus Torvalds err2 = osync_buffers_list(lock, list); 8421da177e4SLinus Torvalds if (err) 8431da177e4SLinus Torvalds return err; 8441da177e4SLinus Torvalds else 8451da177e4SLinus Torvalds return err2; 8461da177e4SLinus Torvalds } 8471da177e4SLinus Torvalds 8481da177e4SLinus Torvalds /* 8491da177e4SLinus Torvalds * Invalidate any and all dirty buffers on a given inode. We are 8501da177e4SLinus Torvalds * probably unmounting the fs, but that doesn't mean we have already 8511da177e4SLinus Torvalds * done a sync(). Just drop the buffers from the inode list. 8521da177e4SLinus Torvalds * 853600f111eSMatthew Wilcox (Oracle) * NOTE: we take the inode's blockdev's mapping's i_private_lock. Which 8541da177e4SLinus Torvalds * assumes that all the buffers are against the blockdev. Not true 8551da177e4SLinus Torvalds * for reiserfs. 8561da177e4SLinus Torvalds */ 8571da177e4SLinus Torvalds void invalidate_inode_buffers(struct inode *inode) 8581da177e4SLinus Torvalds { 8591da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8601da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 861600f111eSMatthew Wilcox (Oracle) struct list_head *list = &mapping->i_private_list; 862600f111eSMatthew Wilcox (Oracle) struct address_space *buffer_mapping = mapping->i_private_data; 8631da177e4SLinus Torvalds 864600f111eSMatthew Wilcox (Oracle) spin_lock(&buffer_mapping->i_private_lock); 8651da177e4SLinus Torvalds while (!list_empty(list)) 8661da177e4SLinus Torvalds __remove_assoc_queue(BH_ENTRY(list->next)); 867600f111eSMatthew Wilcox (Oracle) spin_unlock(&buffer_mapping->i_private_lock); 8681da177e4SLinus Torvalds } 8691da177e4SLinus Torvalds } 87052b19ac9SJan Kara EXPORT_SYMBOL(invalidate_inode_buffers); 8711da177e4SLinus Torvalds 8721da177e4SLinus Torvalds /* 8731da177e4SLinus Torvalds * Remove any clean buffers from the inode's buffer list. This is called 8741da177e4SLinus Torvalds * when we're trying to free the inode itself. Those buffers can pin it. 8751da177e4SLinus Torvalds * 8761da177e4SLinus Torvalds * Returns true if all buffers were removed. 8771da177e4SLinus Torvalds */ 8781da177e4SLinus Torvalds int remove_inode_buffers(struct inode *inode) 8791da177e4SLinus Torvalds { 8801da177e4SLinus Torvalds int ret = 1; 8811da177e4SLinus Torvalds 8821da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 8831da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 884600f111eSMatthew Wilcox (Oracle) struct list_head *list = &mapping->i_private_list; 885600f111eSMatthew Wilcox (Oracle) struct address_space *buffer_mapping = mapping->i_private_data; 8861da177e4SLinus Torvalds 887600f111eSMatthew Wilcox (Oracle) spin_lock(&buffer_mapping->i_private_lock); 8881da177e4SLinus Torvalds while (!list_empty(list)) { 8891da177e4SLinus Torvalds struct buffer_head *bh = BH_ENTRY(list->next); 8901da177e4SLinus Torvalds if (buffer_dirty(bh)) { 8911da177e4SLinus Torvalds ret = 0; 8921da177e4SLinus Torvalds break; 8931da177e4SLinus Torvalds } 8941da177e4SLinus Torvalds __remove_assoc_queue(bh); 8951da177e4SLinus Torvalds } 896600f111eSMatthew Wilcox (Oracle) spin_unlock(&buffer_mapping->i_private_lock); 8971da177e4SLinus Torvalds } 8981da177e4SLinus Torvalds return ret; 8991da177e4SLinus Torvalds } 9001da177e4SLinus Torvalds 9011da177e4SLinus Torvalds /* 902c71124a8SPankaj Raghav * Create the appropriate buffers when given a folio for data area and 9031da177e4SLinus Torvalds * the size of each buffer.. Use the bh->b_this_page linked list to 9041da177e4SLinus Torvalds * follow the buffers created. Return NULL if unable to create more 9051da177e4SLinus Torvalds * buffers. 9061da177e4SLinus Torvalds * 9071da177e4SLinus Torvalds * The retry flag is used to differentiate async IO (paging, swapping) 9081da177e4SLinus Torvalds * which may not fail from ordinary buffer allocations. 9091da177e4SLinus Torvalds */ 910c71124a8SPankaj Raghav struct buffer_head *folio_alloc_buffers(struct folio *folio, unsigned long size, 9112a418157SMatthew Wilcox (Oracle) gfp_t gfp) 9121da177e4SLinus Torvalds { 9131da177e4SLinus Torvalds struct buffer_head *bh, *head; 9141da177e4SLinus Torvalds long offset; 915b87d8cefSRoman Gushchin struct mem_cgroup *memcg, *old_memcg; 9161da177e4SLinus Torvalds 917c71124a8SPankaj Raghav /* The folio lock pins the memcg */ 918c71124a8SPankaj Raghav memcg = folio_memcg(folio); 919b87d8cefSRoman Gushchin old_memcg = set_active_memcg(memcg); 920f745c6f5SShakeel Butt 9211da177e4SLinus Torvalds head = NULL; 922c71124a8SPankaj Raghav offset = folio_size(folio); 9231da177e4SLinus Torvalds while ((offset -= size) >= 0) { 924640ab98fSJens Axboe bh = alloc_buffer_head(gfp); 9251da177e4SLinus Torvalds if (!bh) 9261da177e4SLinus Torvalds goto no_grow; 9271da177e4SLinus Torvalds 9281da177e4SLinus Torvalds bh->b_this_page = head; 9291da177e4SLinus Torvalds bh->b_blocknr = -1; 9301da177e4SLinus Torvalds head = bh; 9311da177e4SLinus Torvalds 9321da177e4SLinus Torvalds bh->b_size = size; 9331da177e4SLinus Torvalds 934c71124a8SPankaj Raghav /* Link the buffer to its folio */ 935c71124a8SPankaj Raghav folio_set_bh(bh, folio, offset); 9361da177e4SLinus Torvalds } 937f745c6f5SShakeel Butt out: 938b87d8cefSRoman Gushchin set_active_memcg(old_memcg); 9391da177e4SLinus Torvalds return head; 9401da177e4SLinus Torvalds /* 9411da177e4SLinus Torvalds * In case anything failed, we just free everything we got. 9421da177e4SLinus Torvalds */ 9431da177e4SLinus Torvalds no_grow: 9441da177e4SLinus Torvalds if (head) { 9451da177e4SLinus Torvalds do { 9461da177e4SLinus Torvalds bh = head; 9471da177e4SLinus Torvalds head = head->b_this_page; 9481da177e4SLinus Torvalds free_buffer_head(bh); 9491da177e4SLinus Torvalds } while (head); 9501da177e4SLinus Torvalds } 9511da177e4SLinus Torvalds 952f745c6f5SShakeel Butt goto out; 9531da177e4SLinus Torvalds } 954c71124a8SPankaj Raghav EXPORT_SYMBOL_GPL(folio_alloc_buffers); 955c71124a8SPankaj Raghav 956c71124a8SPankaj Raghav struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, 957c71124a8SPankaj Raghav bool retry) 958c71124a8SPankaj Raghav { 9592a418157SMatthew Wilcox (Oracle) gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT; 9602a418157SMatthew Wilcox (Oracle) if (retry) 9612a418157SMatthew Wilcox (Oracle) gfp |= __GFP_NOFAIL; 9622a418157SMatthew Wilcox (Oracle) 9632a418157SMatthew Wilcox (Oracle) return folio_alloc_buffers(page_folio(page), size, gfp); 964c71124a8SPankaj Raghav } 9651da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(alloc_page_buffers); 9661da177e4SLinus Torvalds 96708d84addSMatthew Wilcox (Oracle) static inline void link_dev_buffers(struct folio *folio, 96808d84addSMatthew Wilcox (Oracle) struct buffer_head *head) 9691da177e4SLinus Torvalds { 9701da177e4SLinus Torvalds struct buffer_head *bh, *tail; 9711da177e4SLinus Torvalds 9721da177e4SLinus Torvalds bh = head; 9731da177e4SLinus Torvalds do { 9741da177e4SLinus Torvalds tail = bh; 9751da177e4SLinus Torvalds bh = bh->b_this_page; 9761da177e4SLinus Torvalds } while (bh); 9771da177e4SLinus Torvalds tail->b_this_page = head; 97808d84addSMatthew Wilcox (Oracle) folio_attach_private(folio, head); 9791da177e4SLinus Torvalds } 9801da177e4SLinus Torvalds 981bbec0270SLinus Torvalds static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) 982bbec0270SLinus Torvalds { 983bbec0270SLinus Torvalds sector_t retval = ~((sector_t)0); 984b86058f9SChristoph Hellwig loff_t sz = bdev_nr_bytes(bdev); 985bbec0270SLinus Torvalds 986bbec0270SLinus Torvalds if (sz) { 987bbec0270SLinus Torvalds unsigned int sizebits = blksize_bits(size); 988bbec0270SLinus Torvalds retval = (sz >> sizebits); 989bbec0270SLinus Torvalds } 990bbec0270SLinus Torvalds return retval; 991bbec0270SLinus Torvalds } 992bbec0270SLinus Torvalds 9931da177e4SLinus Torvalds /* 9946f24ce6bSMatthew Wilcox (Oracle) * Initialise the state of a blockdev folio's buffers. 9951da177e4SLinus Torvalds */ 9966f24ce6bSMatthew Wilcox (Oracle) static sector_t folio_init_buffers(struct folio *folio, 997382497adSMatthew Wilcox (Oracle) struct block_device *bdev, unsigned size) 9981da177e4SLinus Torvalds { 9996f24ce6bSMatthew Wilcox (Oracle) struct buffer_head *head = folio_buffers(folio); 10001da177e4SLinus Torvalds struct buffer_head *bh = head; 10016f24ce6bSMatthew Wilcox (Oracle) bool uptodate = folio_test_uptodate(folio); 1002382497adSMatthew Wilcox (Oracle) sector_t block = div_u64(folio_pos(folio), size); 1003bcd1d063SChristoph Hellwig sector_t end_block = blkdev_max_block(bdev, size); 10041da177e4SLinus Torvalds 10051da177e4SLinus Torvalds do { 10061da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 100701950a34SEric Biggers bh->b_end_io = NULL; 100801950a34SEric Biggers bh->b_private = NULL; 10091da177e4SLinus Torvalds bh->b_bdev = bdev; 10101da177e4SLinus Torvalds bh->b_blocknr = block; 10111da177e4SLinus Torvalds if (uptodate) 10121da177e4SLinus Torvalds set_buffer_uptodate(bh); 1013080399aaSJeff Moyer if (block < end_block) 10141da177e4SLinus Torvalds set_buffer_mapped(bh); 10151da177e4SLinus Torvalds } 10161da177e4SLinus Torvalds block++; 10171da177e4SLinus Torvalds bh = bh->b_this_page; 10181da177e4SLinus Torvalds } while (bh != head); 1019676ce6d5SHugh Dickins 1020676ce6d5SHugh Dickins /* 1021676ce6d5SHugh Dickins * Caller needs to validate requested block against end of device. 1022676ce6d5SHugh Dickins */ 1023676ce6d5SHugh Dickins return end_block; 10241da177e4SLinus Torvalds } 10251da177e4SLinus Torvalds 10261da177e4SLinus Torvalds /* 10276d840a18SMatthew Wilcox (Oracle) * Create the page-cache folio that contains the requested block. 10281da177e4SLinus Torvalds * 1029676ce6d5SHugh Dickins * This is used purely for blockdev mappings. 10306d840a18SMatthew Wilcox (Oracle) * 1031bcd30d4cSMatthew Wilcox (Oracle) * Returns false if we have a failure which cannot be cured by retrying 1032bcd30d4cSMatthew Wilcox (Oracle) * without sleeping. Returns true if we succeeded, or the caller should retry. 10331da177e4SLinus Torvalds */ 10346d840a18SMatthew Wilcox (Oracle) static bool grow_dev_folio(struct block_device *bdev, sector_t block, 1035382497adSMatthew Wilcox (Oracle) pgoff_t index, unsigned size, gfp_t gfp) 10361da177e4SLinus Torvalds { 10371da177e4SLinus Torvalds struct inode *inode = bdev->bd_inode; 10383c98a41cSMatthew Wilcox (Oracle) struct folio *folio; 10391da177e4SLinus Torvalds struct buffer_head *bh; 10406d840a18SMatthew Wilcox (Oracle) sector_t end_block = 0; 104184235de3SJohannes Weiner 10423c98a41cSMatthew Wilcox (Oracle) folio = __filemap_get_folio(inode->i_mapping, index, 10433ed65f04SMatthew Wilcox (Oracle) FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp); 10443ed65f04SMatthew Wilcox (Oracle) if (IS_ERR(folio)) 10456d840a18SMatthew Wilcox (Oracle) return false; 10461da177e4SLinus Torvalds 10473c98a41cSMatthew Wilcox (Oracle) bh = folio_buffers(folio); 10483c98a41cSMatthew Wilcox (Oracle) if (bh) { 10491da177e4SLinus Torvalds if (bh->b_size == size) { 1050382497adSMatthew Wilcox (Oracle) end_block = folio_init_buffers(folio, bdev, size); 10516d840a18SMatthew Wilcox (Oracle) goto unlock; 10521da177e4SLinus Torvalds } 10531da177e4SLinus Torvalds 1054bcd30d4cSMatthew Wilcox (Oracle) /* 1055bcd30d4cSMatthew Wilcox (Oracle) * Retrying may succeed; for example the folio may finish 1056bcd30d4cSMatthew Wilcox (Oracle) * writeback, or buffers may be cleaned. This should not 1057bcd30d4cSMatthew Wilcox (Oracle) * happen very often; maybe we have old buffers attached to 1058bcd30d4cSMatthew Wilcox (Oracle) * this blockdev's page cache and we're trying to change 1059bcd30d4cSMatthew Wilcox (Oracle) * the block size? 1060bcd30d4cSMatthew Wilcox (Oracle) */ 1061bcd30d4cSMatthew Wilcox (Oracle) if (!try_to_free_buffers(folio)) { 10626d840a18SMatthew Wilcox (Oracle) end_block = ~0ULL; 10636d840a18SMatthew Wilcox (Oracle) goto unlock; 10646d840a18SMatthew Wilcox (Oracle) } 1065bcd30d4cSMatthew Wilcox (Oracle) } 10666d840a18SMatthew Wilcox (Oracle) 10673ed65f04SMatthew Wilcox (Oracle) bh = folio_alloc_buffers(folio, size, gfp | __GFP_ACCOUNT); 10683ed65f04SMatthew Wilcox (Oracle) if (!bh) 10696d840a18SMatthew Wilcox (Oracle) goto unlock; 10701da177e4SLinus Torvalds 10711da177e4SLinus Torvalds /* 10723c98a41cSMatthew Wilcox (Oracle) * Link the folio to the buffers and initialise them. Take the 10731da177e4SLinus Torvalds * lock to be atomic wrt __find_get_block(), which does not 10743c98a41cSMatthew Wilcox (Oracle) * run under the folio lock. 10751da177e4SLinus Torvalds */ 1076600f111eSMatthew Wilcox (Oracle) spin_lock(&inode->i_mapping->i_private_lock); 107708d84addSMatthew Wilcox (Oracle) link_dev_buffers(folio, bh); 1078382497adSMatthew Wilcox (Oracle) end_block = folio_init_buffers(folio, bdev, size); 1079600f111eSMatthew Wilcox (Oracle) spin_unlock(&inode->i_mapping->i_private_lock); 10806d840a18SMatthew Wilcox (Oracle) unlock: 10813c98a41cSMatthew Wilcox (Oracle) folio_unlock(folio); 10823c98a41cSMatthew Wilcox (Oracle) folio_put(folio); 10836d840a18SMatthew Wilcox (Oracle) return block < end_block; 10841da177e4SLinus Torvalds } 10851da177e4SLinus Torvalds 10861da177e4SLinus Torvalds /* 10876d840a18SMatthew Wilcox (Oracle) * Create buffers for the specified block device block's folio. If 10886d840a18SMatthew Wilcox (Oracle) * that folio was dirty, the buffers are set dirty also. Returns false 10896d840a18SMatthew Wilcox (Oracle) * if we've hit a permanent error. 10901da177e4SLinus Torvalds */ 10916d840a18SMatthew Wilcox (Oracle) static bool grow_buffers(struct block_device *bdev, sector_t block, 10926d840a18SMatthew Wilcox (Oracle) unsigned size, gfp_t gfp) 10931da177e4SLinus Torvalds { 10945f3bd90dSMatthew Wilcox (Oracle) loff_t pos; 10951da177e4SLinus Torvalds 1096e5657933SAndrew Morton /* 10975f3bd90dSMatthew Wilcox (Oracle) * Check for a block which lies outside our maximum possible 10985f3bd90dSMatthew Wilcox (Oracle) * pagecache index. 1099e5657933SAndrew Morton */ 11005f3bd90dSMatthew Wilcox (Oracle) if (check_mul_overflow(block, (sector_t)size, &pos) || pos > MAX_LFS_FILESIZE) { 11015f3bd90dSMatthew Wilcox (Oracle) printk(KERN_ERR "%s: requested out-of-range block %llu for device %pg\n", 11028e24eea7SHarvey Harrison __func__, (unsigned long long)block, 1103a1c6f057SDmitry Monakhov bdev); 11046d840a18SMatthew Wilcox (Oracle) return false; 1105e5657933SAndrew Morton } 1106676ce6d5SHugh Dickins 11076d840a18SMatthew Wilcox (Oracle) /* Create a folio with the proper size buffers */ 11085f3bd90dSMatthew Wilcox (Oracle) return grow_dev_folio(bdev, block, pos / PAGE_SIZE, size, gfp); 11091da177e4SLinus Torvalds } 11101da177e4SLinus Torvalds 11110026ba40SEric Biggers static struct buffer_head * 11123b5e6454SGioh Kim __getblk_slow(struct block_device *bdev, sector_t block, 11133b5e6454SGioh Kim unsigned size, gfp_t gfp) 11141da177e4SLinus Torvalds { 11151da177e4SLinus Torvalds /* Size must be multiple of hard sectorsize */ 1116e1defc4fSMartin K. Petersen if (unlikely(size & (bdev_logical_block_size(bdev)-1) || 11171da177e4SLinus Torvalds (size < 512 || size > PAGE_SIZE))) { 11181da177e4SLinus Torvalds printk(KERN_ERR "getblk(): invalid block size %d requested\n", 11191da177e4SLinus Torvalds size); 1120e1defc4fSMartin K. Petersen printk(KERN_ERR "logical block size: %d\n", 1121e1defc4fSMartin K. Petersen bdev_logical_block_size(bdev)); 11221da177e4SLinus Torvalds 11231da177e4SLinus Torvalds dump_stack(); 11241da177e4SLinus Torvalds return NULL; 11251da177e4SLinus Torvalds } 11261da177e4SLinus Torvalds 1127676ce6d5SHugh Dickins for (;;) { 1128676ce6d5SHugh Dickins struct buffer_head *bh; 1129676ce6d5SHugh Dickins 11301da177e4SLinus Torvalds bh = __find_get_block(bdev, block, size); 11311da177e4SLinus Torvalds if (bh) 11321da177e4SLinus Torvalds return bh; 11331da177e4SLinus Torvalds 11346d840a18SMatthew Wilcox (Oracle) if (!grow_buffers(bdev, block, size, gfp)) 113591f68c89SJeff Moyer return NULL; 1136676ce6d5SHugh Dickins } 11371da177e4SLinus Torvalds } 11381da177e4SLinus Torvalds 11391da177e4SLinus Torvalds /* 11401da177e4SLinus Torvalds * The relationship between dirty buffers and dirty pages: 11411da177e4SLinus Torvalds * 11421da177e4SLinus Torvalds * Whenever a page has any dirty buffers, the page's dirty bit is set, and 1143ec82e1c1SMatthew Wilcox * the page is tagged dirty in the page cache. 11441da177e4SLinus Torvalds * 11451da177e4SLinus Torvalds * At all times, the dirtiness of the buffers represents the dirtiness of 11461da177e4SLinus Torvalds * subsections of the page. If the page has buffers, the page dirty bit is 11471da177e4SLinus Torvalds * merely a hint about the true dirty state. 11481da177e4SLinus Torvalds * 11491da177e4SLinus Torvalds * When a page is set dirty in its entirety, all its buffers are marked dirty 11501da177e4SLinus Torvalds * (if the page has buffers). 11511da177e4SLinus Torvalds * 11521da177e4SLinus Torvalds * When a buffer is marked dirty, its page is dirtied, but the page's other 11531da177e4SLinus Torvalds * buffers are not. 11541da177e4SLinus Torvalds * 11551da177e4SLinus Torvalds * Also. When blockdev buffers are explicitly read with bread(), they 11561da177e4SLinus Torvalds * individually become uptodate. But their backing page remains not 11571da177e4SLinus Torvalds * uptodate - even if all of its buffers are uptodate. A subsequent 11582c69e205SMatthew Wilcox (Oracle) * block_read_full_folio() against that folio will discover all the uptodate 11592c69e205SMatthew Wilcox (Oracle) * buffers, will set the folio uptodate and will perform no I/O. 11601da177e4SLinus Torvalds */ 11611da177e4SLinus Torvalds 11621da177e4SLinus Torvalds /** 11631da177e4SLinus Torvalds * mark_buffer_dirty - mark a buffer_head as needing writeout 116467be2dd1SMartin Waitz * @bh: the buffer_head to mark dirty 11651da177e4SLinus Torvalds * 1166ec82e1c1SMatthew Wilcox * mark_buffer_dirty() will set the dirty bit against the buffer, then set 1167ec82e1c1SMatthew Wilcox * its backing page dirty, then tag the page as dirty in the page cache 1168ec82e1c1SMatthew Wilcox * and then attach the address_space's inode to its superblock's dirty 11691da177e4SLinus Torvalds * inode list. 11701da177e4SLinus Torvalds * 1171600f111eSMatthew Wilcox (Oracle) * mark_buffer_dirty() is atomic. It takes bh->b_folio->mapping->i_private_lock, 1172b93b0163SMatthew Wilcox * i_pages lock and mapping->host->i_lock. 11731da177e4SLinus Torvalds */ 1174fc9b52cdSHarvey Harrison void mark_buffer_dirty(struct buffer_head *bh) 11751da177e4SLinus Torvalds { 1176787d2214SNick Piggin WARN_ON_ONCE(!buffer_uptodate(bh)); 11771be62dc1SLinus Torvalds 11785305cb83STejun Heo trace_block_dirty_buffer(bh); 11795305cb83STejun Heo 11801be62dc1SLinus Torvalds /* 11811be62dc1SLinus Torvalds * Very *carefully* optimize the it-is-already-dirty case. 11821be62dc1SLinus Torvalds * 11831be62dc1SLinus Torvalds * Don't let the final "is it dirty" escape to before we 11841be62dc1SLinus Torvalds * perhaps modified the buffer. 11851be62dc1SLinus Torvalds */ 11861be62dc1SLinus Torvalds if (buffer_dirty(bh)) { 11871be62dc1SLinus Torvalds smp_mb(); 11881be62dc1SLinus Torvalds if (buffer_dirty(bh)) 11891be62dc1SLinus Torvalds return; 11901be62dc1SLinus Torvalds } 11911be62dc1SLinus Torvalds 1192a8e7d49aSLinus Torvalds if (!test_set_buffer_dirty(bh)) { 1193cf1d3417SMatthew Wilcox (Oracle) struct folio *folio = bh->b_folio; 1194c4843a75SGreg Thelen struct address_space *mapping = NULL; 1195c4843a75SGreg Thelen 1196cf1d3417SMatthew Wilcox (Oracle) folio_memcg_lock(folio); 1197cf1d3417SMatthew Wilcox (Oracle) if (!folio_test_set_dirty(folio)) { 1198cf1d3417SMatthew Wilcox (Oracle) mapping = folio->mapping; 11998e9d78edSLinus Torvalds if (mapping) 1200cf1d3417SMatthew Wilcox (Oracle) __folio_mark_dirty(folio, mapping, 0); 12018e9d78edSLinus Torvalds } 1202cf1d3417SMatthew Wilcox (Oracle) folio_memcg_unlock(folio); 1203c4843a75SGreg Thelen if (mapping) 1204c4843a75SGreg Thelen __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 1205a8e7d49aSLinus Torvalds } 12061da177e4SLinus Torvalds } 12071fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(mark_buffer_dirty); 12081da177e4SLinus Torvalds 120987354e5dSJeff Layton void mark_buffer_write_io_error(struct buffer_head *bh) 121087354e5dSJeff Layton { 121187354e5dSJeff Layton set_buffer_write_io_error(bh); 121287354e5dSJeff Layton /* FIXME: do we need to set this in both places? */ 1213abc8a8a2SMatthew Wilcox (Oracle) if (bh->b_folio && bh->b_folio->mapping) 1214abc8a8a2SMatthew Wilcox (Oracle) mapping_set_error(bh->b_folio->mapping, -EIO); 12154b2201daSChristoph Hellwig if (bh->b_assoc_map) { 121687354e5dSJeff Layton mapping_set_error(bh->b_assoc_map, -EIO); 12174b2201daSChristoph Hellwig errseq_set(&bh->b_assoc_map->host->i_sb->s_wb_err, -EIO); 12184b2201daSChristoph Hellwig } 121987354e5dSJeff Layton } 122087354e5dSJeff Layton EXPORT_SYMBOL(mark_buffer_write_io_error); 122187354e5dSJeff Layton 12221da177e4SLinus Torvalds /* 12231da177e4SLinus Torvalds * Decrement a buffer_head's reference count. If all buffers against a page 12241da177e4SLinus Torvalds * have zero reference count, are clean and unlocked, and if the page is clean 12251da177e4SLinus Torvalds * and unlocked then try_to_free_buffers() may strip the buffers from the page 12261da177e4SLinus Torvalds * in preparation for freeing it (sometimes, rarely, buffers are removed from 12271da177e4SLinus Torvalds * a page but it ends up not being freed, and buffers may later be reattached). 12281da177e4SLinus Torvalds */ 12291da177e4SLinus Torvalds void __brelse(struct buffer_head * buf) 12301da177e4SLinus Torvalds { 12311da177e4SLinus Torvalds if (atomic_read(&buf->b_count)) { 12321da177e4SLinus Torvalds put_bh(buf); 12331da177e4SLinus Torvalds return; 12341da177e4SLinus Torvalds } 12355c752ad9SArjan van de Ven WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); 12361da177e4SLinus Torvalds } 12371fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__brelse); 12381da177e4SLinus Torvalds 12391da177e4SLinus Torvalds /* 12401da177e4SLinus Torvalds * bforget() is like brelse(), except it discards any 12411da177e4SLinus Torvalds * potentially dirty data. 12421da177e4SLinus Torvalds */ 12431da177e4SLinus Torvalds void __bforget(struct buffer_head *bh) 12441da177e4SLinus Torvalds { 12451da177e4SLinus Torvalds clear_buffer_dirty(bh); 1246535ee2fbSJan Kara if (bh->b_assoc_map) { 1247abc8a8a2SMatthew Wilcox (Oracle) struct address_space *buffer_mapping = bh->b_folio->mapping; 12481da177e4SLinus Torvalds 1249600f111eSMatthew Wilcox (Oracle) spin_lock(&buffer_mapping->i_private_lock); 12501da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 125158ff407bSJan Kara bh->b_assoc_map = NULL; 1252600f111eSMatthew Wilcox (Oracle) spin_unlock(&buffer_mapping->i_private_lock); 12531da177e4SLinus Torvalds } 12541da177e4SLinus Torvalds __brelse(bh); 12551da177e4SLinus Torvalds } 12561fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(__bforget); 12571da177e4SLinus Torvalds 12581da177e4SLinus Torvalds static struct buffer_head *__bread_slow(struct buffer_head *bh) 12591da177e4SLinus Torvalds { 12601da177e4SLinus Torvalds lock_buffer(bh); 12611da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 12621da177e4SLinus Torvalds unlock_buffer(bh); 12631da177e4SLinus Torvalds return bh; 12641da177e4SLinus Torvalds } else { 12651da177e4SLinus Torvalds get_bh(bh); 12661da177e4SLinus Torvalds bh->b_end_io = end_buffer_read_sync; 12671420c4a5SBart Van Assche submit_bh(REQ_OP_READ, bh); 12681da177e4SLinus Torvalds wait_on_buffer(bh); 12691da177e4SLinus Torvalds if (buffer_uptodate(bh)) 12701da177e4SLinus Torvalds return bh; 12711da177e4SLinus Torvalds } 12721da177e4SLinus Torvalds brelse(bh); 12731da177e4SLinus Torvalds return NULL; 12741da177e4SLinus Torvalds } 12751da177e4SLinus Torvalds 12761da177e4SLinus Torvalds /* 12771da177e4SLinus Torvalds * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). 12781da177e4SLinus Torvalds * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their 12791da177e4SLinus Torvalds * refcount elevated by one when they're in an LRU. A buffer can only appear 12801da177e4SLinus Torvalds * once in a particular CPU's LRU. A single buffer can be present in multiple 12811da177e4SLinus Torvalds * CPU's LRUs at the same time. 12821da177e4SLinus Torvalds * 12831da177e4SLinus Torvalds * This is a transparent caching front-end to sb_bread(), sb_getblk() and 12841da177e4SLinus Torvalds * sb_find_get_block(). 12851da177e4SLinus Torvalds * 12861da177e4SLinus Torvalds * The LRUs themselves only need locking against invalidate_bh_lrus. We use 12871da177e4SLinus Torvalds * a local interrupt disable for that. 12881da177e4SLinus Torvalds */ 12891da177e4SLinus Torvalds 129086cf78d7SSebastien Buisson #define BH_LRU_SIZE 16 12911da177e4SLinus Torvalds 12921da177e4SLinus Torvalds struct bh_lru { 12931da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 12941da177e4SLinus Torvalds }; 12951da177e4SLinus Torvalds 12961da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds #ifdef CONFIG_SMP 12991da177e4SLinus Torvalds #define bh_lru_lock() local_irq_disable() 13001da177e4SLinus Torvalds #define bh_lru_unlock() local_irq_enable() 13011da177e4SLinus Torvalds #else 13021da177e4SLinus Torvalds #define bh_lru_lock() preempt_disable() 13031da177e4SLinus Torvalds #define bh_lru_unlock() preempt_enable() 13041da177e4SLinus Torvalds #endif 13051da177e4SLinus Torvalds 13061da177e4SLinus Torvalds static inline void check_irqs_on(void) 13071da177e4SLinus Torvalds { 13081da177e4SLinus Torvalds #ifdef irqs_disabled 13091da177e4SLinus Torvalds BUG_ON(irqs_disabled()); 13101da177e4SLinus Torvalds #endif 13111da177e4SLinus Torvalds } 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds /* 1314241f01fbSEric Biggers * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is 1315241f01fbSEric Biggers * inserted at the front, and the buffer_head at the back if any is evicted. 1316241f01fbSEric Biggers * Or, if already in the LRU it is moved to the front. 13171da177e4SLinus Torvalds */ 13181da177e4SLinus Torvalds static void bh_lru_install(struct buffer_head *bh) 13191da177e4SLinus Torvalds { 1320241f01fbSEric Biggers struct buffer_head *evictee = bh; 1321241f01fbSEric Biggers struct bh_lru *b; 1322241f01fbSEric Biggers int i; 13231da177e4SLinus Torvalds 13241da177e4SLinus Torvalds check_irqs_on(); 1325c0226eb8SMinchan Kim bh_lru_lock(); 1326c0226eb8SMinchan Kim 13278cc621d2SMinchan Kim /* 13288cc621d2SMinchan Kim * the refcount of buffer_head in bh_lru prevents dropping the 13298cc621d2SMinchan Kim * attached page(i.e., try_to_free_buffers) so it could cause 13308cc621d2SMinchan Kim * failing page migration. 13318cc621d2SMinchan Kim * Skip putting upcoming bh into bh_lru until migration is done. 13328cc621d2SMinchan Kim */ 13338a237adfSMarcelo Tosatti if (lru_cache_disabled() || cpu_is_isolated(smp_processor_id())) { 1334c0226eb8SMinchan Kim bh_lru_unlock(); 13358cc621d2SMinchan Kim return; 1336c0226eb8SMinchan Kim } 1337241f01fbSEric Biggers 1338241f01fbSEric Biggers b = this_cpu_ptr(&bh_lrus); 1339241f01fbSEric Biggers for (i = 0; i < BH_LRU_SIZE; i++) { 1340241f01fbSEric Biggers swap(evictee, b->bhs[i]); 1341241f01fbSEric Biggers if (evictee == bh) { 1342241f01fbSEric Biggers bh_lru_unlock(); 1343241f01fbSEric Biggers return; 1344241f01fbSEric Biggers } 1345241f01fbSEric Biggers } 13461da177e4SLinus Torvalds 13471da177e4SLinus Torvalds get_bh(bh); 13481da177e4SLinus Torvalds bh_lru_unlock(); 1349241f01fbSEric Biggers brelse(evictee); 13501da177e4SLinus Torvalds } 13511da177e4SLinus Torvalds 13521da177e4SLinus Torvalds /* 13531da177e4SLinus Torvalds * Look up the bh in this cpu's LRU. If it's there, move it to the head. 13541da177e4SLinus Torvalds */ 1355858119e1SArjan van de Ven static struct buffer_head * 13563991d3bdSTomasz Kvarsin lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) 13571da177e4SLinus Torvalds { 13581da177e4SLinus Torvalds struct buffer_head *ret = NULL; 13593991d3bdSTomasz Kvarsin unsigned int i; 13601da177e4SLinus Torvalds 13611da177e4SLinus Torvalds check_irqs_on(); 13621da177e4SLinus Torvalds bh_lru_lock(); 13638a237adfSMarcelo Tosatti if (cpu_is_isolated(smp_processor_id())) { 13648a237adfSMarcelo Tosatti bh_lru_unlock(); 13658a237adfSMarcelo Tosatti return NULL; 13668a237adfSMarcelo Tosatti } 13671da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 1368c7b92516SChristoph Lameter struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); 13691da177e4SLinus Torvalds 13709470dd5dSZach Brown if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && 13719470dd5dSZach Brown bh->b_size == size) { 13721da177e4SLinus Torvalds if (i) { 13731da177e4SLinus Torvalds while (i) { 1374c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[i], 1375c7b92516SChristoph Lameter __this_cpu_read(bh_lrus.bhs[i - 1])); 13761da177e4SLinus Torvalds i--; 13771da177e4SLinus Torvalds } 1378c7b92516SChristoph Lameter __this_cpu_write(bh_lrus.bhs[0], bh); 13791da177e4SLinus Torvalds } 13801da177e4SLinus Torvalds get_bh(bh); 13811da177e4SLinus Torvalds ret = bh; 13821da177e4SLinus Torvalds break; 13831da177e4SLinus Torvalds } 13841da177e4SLinus Torvalds } 13851da177e4SLinus Torvalds bh_lru_unlock(); 13861da177e4SLinus Torvalds return ret; 13871da177e4SLinus Torvalds } 13881da177e4SLinus Torvalds 13891da177e4SLinus Torvalds /* 13901da177e4SLinus Torvalds * Perform a pagecache lookup for the matching buffer. If it's there, refresh 13911da177e4SLinus Torvalds * it in the LRU and mark it as accessed. If it is not present then return 13921da177e4SLinus Torvalds * NULL 13931da177e4SLinus Torvalds */ 13941da177e4SLinus Torvalds struct buffer_head * 13953991d3bdSTomasz Kvarsin __find_get_block(struct block_device *bdev, sector_t block, unsigned size) 13961da177e4SLinus Torvalds { 13971da177e4SLinus Torvalds struct buffer_head *bh = lookup_bh_lru(bdev, block, size); 13981da177e4SLinus Torvalds 13991da177e4SLinus Torvalds if (bh == NULL) { 14002457aec6SMel Gorman /* __find_get_block_slow will mark the page accessed */ 1401385fd4c5SCoywolf Qi Hunt bh = __find_get_block_slow(bdev, block); 14021da177e4SLinus Torvalds if (bh) 14031da177e4SLinus Torvalds bh_lru_install(bh); 14042457aec6SMel Gorman } else 14051da177e4SLinus Torvalds touch_buffer(bh); 14062457aec6SMel Gorman 14071da177e4SLinus Torvalds return bh; 14081da177e4SLinus Torvalds } 14091da177e4SLinus Torvalds EXPORT_SYMBOL(__find_get_block); 14101da177e4SLinus Torvalds 14113ed65f04SMatthew Wilcox (Oracle) /** 14123ed65f04SMatthew Wilcox (Oracle) * bdev_getblk - Get a buffer_head in a block device's buffer cache. 14133ed65f04SMatthew Wilcox (Oracle) * @bdev: The block device. 14143ed65f04SMatthew Wilcox (Oracle) * @block: The block number. 14153ed65f04SMatthew Wilcox (Oracle) * @size: The size of buffer_heads for this @bdev. 14163ed65f04SMatthew Wilcox (Oracle) * @gfp: The memory allocation flags to use. 14173ed65f04SMatthew Wilcox (Oracle) * 14183ed65f04SMatthew Wilcox (Oracle) * Return: The buffer head, or NULL if memory could not be allocated. 14193ed65f04SMatthew Wilcox (Oracle) */ 14203ed65f04SMatthew Wilcox (Oracle) struct buffer_head *bdev_getblk(struct block_device *bdev, sector_t block, 14213ed65f04SMatthew Wilcox (Oracle) unsigned size, gfp_t gfp) 14223ed65f04SMatthew Wilcox (Oracle) { 14233ed65f04SMatthew Wilcox (Oracle) struct buffer_head *bh = __find_get_block(bdev, block, size); 14243ed65f04SMatthew Wilcox (Oracle) 14253ed65f04SMatthew Wilcox (Oracle) might_alloc(gfp); 14263ed65f04SMatthew Wilcox (Oracle) if (bh) 14273ed65f04SMatthew Wilcox (Oracle) return bh; 14283ed65f04SMatthew Wilcox (Oracle) 14293ed65f04SMatthew Wilcox (Oracle) return __getblk_slow(bdev, block, size, gfp); 14303ed65f04SMatthew Wilcox (Oracle) } 14313ed65f04SMatthew Wilcox (Oracle) EXPORT_SYMBOL(bdev_getblk); 14323ed65f04SMatthew Wilcox (Oracle) 14331da177e4SLinus Torvalds /* 14341da177e4SLinus Torvalds * Do async read-ahead on a buffer.. 14351da177e4SLinus Torvalds */ 14363991d3bdSTomasz Kvarsin void __breadahead(struct block_device *bdev, sector_t block, unsigned size) 14371da177e4SLinus Torvalds { 1438775d9b10SMatthew Wilcox (Oracle) struct buffer_head *bh = bdev_getblk(bdev, block, size, 1439775d9b10SMatthew Wilcox (Oracle) GFP_NOWAIT | __GFP_MOVABLE); 1440775d9b10SMatthew Wilcox (Oracle) 1441a3e713b5SAndrew Morton if (likely(bh)) { 1442e7ea1129SZhang Yi bh_readahead(bh, REQ_RAHEAD); 14431da177e4SLinus Torvalds brelse(bh); 14441da177e4SLinus Torvalds } 1445a3e713b5SAndrew Morton } 14461da177e4SLinus Torvalds EXPORT_SYMBOL(__breadahead); 14471da177e4SLinus Torvalds 14481da177e4SLinus Torvalds /** 14493b5e6454SGioh Kim * __bread_gfp() - reads a specified block and returns the bh 145067be2dd1SMartin Waitz * @bdev: the block_device to read from 14511da177e4SLinus Torvalds * @block: number of block 14521da177e4SLinus Torvalds * @size: size (in bytes) to read 14533b5e6454SGioh Kim * @gfp: page allocation flag 14541da177e4SLinus Torvalds * 14551da177e4SLinus Torvalds * Reads a specified block, and returns buffer head that contains it. 14563b5e6454SGioh Kim * The page cache can be allocated from non-movable area 14573b5e6454SGioh Kim * not to prevent page migration if you set gfp to zero. 14581da177e4SLinus Torvalds * It returns NULL if the block was unreadable. 14591da177e4SLinus Torvalds */ 14601da177e4SLinus Torvalds struct buffer_head * 14613b5e6454SGioh Kim __bread_gfp(struct block_device *bdev, sector_t block, 14623b5e6454SGioh Kim unsigned size, gfp_t gfp) 14631da177e4SLinus Torvalds { 146493b13ecaSMatthew Wilcox (Oracle) struct buffer_head *bh; 146593b13ecaSMatthew Wilcox (Oracle) 146693b13ecaSMatthew Wilcox (Oracle) gfp |= mapping_gfp_constraint(bdev->bd_inode->i_mapping, ~__GFP_FS); 146793b13ecaSMatthew Wilcox (Oracle) 146893b13ecaSMatthew Wilcox (Oracle) /* 146993b13ecaSMatthew Wilcox (Oracle) * Prefer looping in the allocator rather than here, at least that 147093b13ecaSMatthew Wilcox (Oracle) * code knows what it's doing. 147193b13ecaSMatthew Wilcox (Oracle) */ 147293b13ecaSMatthew Wilcox (Oracle) gfp |= __GFP_NOFAIL; 147393b13ecaSMatthew Wilcox (Oracle) 147493b13ecaSMatthew Wilcox (Oracle) bh = bdev_getblk(bdev, block, size, gfp); 14751da177e4SLinus Torvalds 1476a3e713b5SAndrew Morton if (likely(bh) && !buffer_uptodate(bh)) 14771da177e4SLinus Torvalds bh = __bread_slow(bh); 14781da177e4SLinus Torvalds return bh; 14791da177e4SLinus Torvalds } 14803b5e6454SGioh Kim EXPORT_SYMBOL(__bread_gfp); 14811da177e4SLinus Torvalds 14828cc621d2SMinchan Kim static void __invalidate_bh_lrus(struct bh_lru *b) 14838cc621d2SMinchan Kim { 14848cc621d2SMinchan Kim int i; 14858cc621d2SMinchan Kim 14868cc621d2SMinchan Kim for (i = 0; i < BH_LRU_SIZE; i++) { 14878cc621d2SMinchan Kim brelse(b->bhs[i]); 14888cc621d2SMinchan Kim b->bhs[i] = NULL; 14898cc621d2SMinchan Kim } 14908cc621d2SMinchan Kim } 14911da177e4SLinus Torvalds /* 14921da177e4SLinus Torvalds * invalidate_bh_lrus() is called rarely - but not only at unmount. 14931da177e4SLinus Torvalds * This doesn't race because it runs in each cpu either in irq 14941da177e4SLinus Torvalds * or with preempt disabled. 14951da177e4SLinus Torvalds */ 14961da177e4SLinus Torvalds static void invalidate_bh_lru(void *arg) 14971da177e4SLinus Torvalds { 14981da177e4SLinus Torvalds struct bh_lru *b = &get_cpu_var(bh_lrus); 14991da177e4SLinus Torvalds 15008cc621d2SMinchan Kim __invalidate_bh_lrus(b); 15011da177e4SLinus Torvalds put_cpu_var(bh_lrus); 15021da177e4SLinus Torvalds } 15031da177e4SLinus Torvalds 15048cc621d2SMinchan Kim bool has_bh_in_lru(int cpu, void *dummy) 150542be35d0SGilad Ben-Yossef { 150642be35d0SGilad Ben-Yossef struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); 150742be35d0SGilad Ben-Yossef int i; 150842be35d0SGilad Ben-Yossef 150942be35d0SGilad Ben-Yossef for (i = 0; i < BH_LRU_SIZE; i++) { 151042be35d0SGilad Ben-Yossef if (b->bhs[i]) 15111d706679SSaurav Girepunje return true; 151242be35d0SGilad Ben-Yossef } 151342be35d0SGilad Ben-Yossef 15141d706679SSaurav Girepunje return false; 151542be35d0SGilad Ben-Yossef } 151642be35d0SGilad Ben-Yossef 1517f9a14399SPeter Zijlstra void invalidate_bh_lrus(void) 15181da177e4SLinus Torvalds { 1519cb923159SSebastian Andrzej Siewior on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1); 15201da177e4SLinus Torvalds } 15219db5579bSNick Piggin EXPORT_SYMBOL_GPL(invalidate_bh_lrus); 15221da177e4SLinus Torvalds 1523243418e3SMinchan Kim /* 1524243418e3SMinchan Kim * It's called from workqueue context so we need a bh_lru_lock to close 1525243418e3SMinchan Kim * the race with preemption/irq. 1526243418e3SMinchan Kim */ 1527243418e3SMinchan Kim void invalidate_bh_lrus_cpu(void) 15288cc621d2SMinchan Kim { 15298cc621d2SMinchan Kim struct bh_lru *b; 15308cc621d2SMinchan Kim 15318cc621d2SMinchan Kim bh_lru_lock(); 1532243418e3SMinchan Kim b = this_cpu_ptr(&bh_lrus); 15338cc621d2SMinchan Kim __invalidate_bh_lrus(b); 15348cc621d2SMinchan Kim bh_lru_unlock(); 15358cc621d2SMinchan Kim } 15368cc621d2SMinchan Kim 1537465e5e6aSPankaj Raghav void folio_set_bh(struct buffer_head *bh, struct folio *folio, 1538465e5e6aSPankaj Raghav unsigned long offset) 1539465e5e6aSPankaj Raghav { 1540465e5e6aSPankaj Raghav bh->b_folio = folio; 1541465e5e6aSPankaj Raghav BUG_ON(offset >= folio_size(folio)); 1542465e5e6aSPankaj Raghav if (folio_test_highmem(folio)) 1543465e5e6aSPankaj Raghav /* 1544465e5e6aSPankaj Raghav * This catches illegal uses and preserves the offset: 1545465e5e6aSPankaj Raghav */ 1546465e5e6aSPankaj Raghav bh->b_data = (char *)(0 + offset); 1547465e5e6aSPankaj Raghav else 1548465e5e6aSPankaj Raghav bh->b_data = folio_address(folio) + offset; 1549465e5e6aSPankaj Raghav } 1550465e5e6aSPankaj Raghav EXPORT_SYMBOL(folio_set_bh); 1551465e5e6aSPankaj Raghav 15521da177e4SLinus Torvalds /* 15531da177e4SLinus Torvalds * Called when truncating a buffer on a page completely. 15541da177e4SLinus Torvalds */ 1555e7470ee8SMel Gorman 1556e7470ee8SMel Gorman /* Bits that are cleared during an invalidate */ 1557e7470ee8SMel Gorman #define BUFFER_FLAGS_DISCARD \ 1558e7470ee8SMel Gorman (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ 1559e7470ee8SMel Gorman 1 << BH_Delay | 1 << BH_Unwritten) 1560e7470ee8SMel Gorman 1561858119e1SArjan van de Ven static void discard_buffer(struct buffer_head * bh) 15621da177e4SLinus Torvalds { 1563b0192296SUros Bizjak unsigned long b_state; 1564e7470ee8SMel Gorman 15651da177e4SLinus Torvalds lock_buffer(bh); 15661da177e4SLinus Torvalds clear_buffer_dirty(bh); 15671da177e4SLinus Torvalds bh->b_bdev = NULL; 1568b0192296SUros Bizjak b_state = READ_ONCE(bh->b_state); 1569b0192296SUros Bizjak do { 1570b0192296SUros Bizjak } while (!try_cmpxchg(&bh->b_state, &b_state, 1571b0192296SUros Bizjak b_state & ~BUFFER_FLAGS_DISCARD)); 15721da177e4SLinus Torvalds unlock_buffer(bh); 15731da177e4SLinus Torvalds } 15741da177e4SLinus Torvalds 15751da177e4SLinus Torvalds /** 15767ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio - Invalidate part or all of a buffer-backed folio. 15777ba13abbSMatthew Wilcox (Oracle) * @folio: The folio which is affected. 1578d47992f8SLukas Czerner * @offset: start of the range to invalidate 1579d47992f8SLukas Czerner * @length: length of the range to invalidate 15801da177e4SLinus Torvalds * 15817ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio() is called when all or part of the folio has been 15821da177e4SLinus Torvalds * invalidated by a truncate operation. 15831da177e4SLinus Torvalds * 15847ba13abbSMatthew Wilcox (Oracle) * block_invalidate_folio() does not have to release all buffers, but it must 15851da177e4SLinus Torvalds * ensure that no dirty buffer is left outside @offset and that no I/O 15861da177e4SLinus Torvalds * is underway against any of the blocks which are outside the truncation 15871da177e4SLinus Torvalds * point. Because the caller is about to free (and possibly reuse) those 15881da177e4SLinus Torvalds * blocks on-disk. 15891da177e4SLinus Torvalds */ 15907ba13abbSMatthew Wilcox (Oracle) void block_invalidate_folio(struct folio *folio, size_t offset, size_t length) 15911da177e4SLinus Torvalds { 15921da177e4SLinus Torvalds struct buffer_head *head, *bh, *next; 15937ba13abbSMatthew Wilcox (Oracle) size_t curr_off = 0; 15947ba13abbSMatthew Wilcox (Oracle) size_t stop = length + offset; 15951da177e4SLinus Torvalds 15967ba13abbSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 15971da177e4SLinus Torvalds 1598d47992f8SLukas Czerner /* 1599d47992f8SLukas Czerner * Check for overflow 1600d47992f8SLukas Czerner */ 16017ba13abbSMatthew Wilcox (Oracle) BUG_ON(stop > folio_size(folio) || stop < length); 1602d47992f8SLukas Czerner 16037ba13abbSMatthew Wilcox (Oracle) head = folio_buffers(folio); 16047ba13abbSMatthew Wilcox (Oracle) if (!head) 16057ba13abbSMatthew Wilcox (Oracle) return; 16067ba13abbSMatthew Wilcox (Oracle) 16071da177e4SLinus Torvalds bh = head; 16081da177e4SLinus Torvalds do { 16097ba13abbSMatthew Wilcox (Oracle) size_t next_off = curr_off + bh->b_size; 16101da177e4SLinus Torvalds next = bh->b_this_page; 16111da177e4SLinus Torvalds 16121da177e4SLinus Torvalds /* 1613d47992f8SLukas Czerner * Are we still fully in range ? 1614d47992f8SLukas Czerner */ 1615d47992f8SLukas Czerner if (next_off > stop) 1616d47992f8SLukas Czerner goto out; 1617d47992f8SLukas Czerner 1618d47992f8SLukas Czerner /* 16191da177e4SLinus Torvalds * is this block fully invalidated? 16201da177e4SLinus Torvalds */ 16211da177e4SLinus Torvalds if (offset <= curr_off) 16221da177e4SLinus Torvalds discard_buffer(bh); 16231da177e4SLinus Torvalds curr_off = next_off; 16241da177e4SLinus Torvalds bh = next; 16251da177e4SLinus Torvalds } while (bh != head); 16261da177e4SLinus Torvalds 16271da177e4SLinus Torvalds /* 16287ba13abbSMatthew Wilcox (Oracle) * We release buffers only if the entire folio is being invalidated. 16291da177e4SLinus Torvalds * The get_block cached value has been unconditionally invalidated, 16301da177e4SLinus Torvalds * so real IO is not possible anymore. 16311da177e4SLinus Torvalds */ 16327ba13abbSMatthew Wilcox (Oracle) if (length == folio_size(folio)) 16337ba13abbSMatthew Wilcox (Oracle) filemap_release_folio(folio, 0); 16341da177e4SLinus Torvalds out: 16352ff28e22SNeilBrown return; 16361da177e4SLinus Torvalds } 16377ba13abbSMatthew Wilcox (Oracle) EXPORT_SYMBOL(block_invalidate_folio); 16381da177e4SLinus Torvalds 16391da177e4SLinus Torvalds /* 16401da177e4SLinus Torvalds * We attach and possibly dirty the buffers atomically wrt 1641600f111eSMatthew Wilcox (Oracle) * block_dirty_folio() via i_private_lock. try_to_free_buffers 16428e2e1756SPankaj Raghav * is already excluded via the folio lock. 16431da177e4SLinus Torvalds */ 16440a88810dSMatthew Wilcox (Oracle) struct buffer_head *create_empty_buffers(struct folio *folio, 16453decb856SMatthew Wilcox (Oracle) unsigned long blocksize, unsigned long b_state) 16461da177e4SLinus Torvalds { 16471da177e4SLinus Torvalds struct buffer_head *bh, *head, *tail; 16482a418157SMatthew Wilcox (Oracle) gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT | __GFP_NOFAIL; 16491da177e4SLinus Torvalds 16502a418157SMatthew Wilcox (Oracle) head = folio_alloc_buffers(folio, blocksize, gfp); 16511da177e4SLinus Torvalds bh = head; 16521da177e4SLinus Torvalds do { 16531da177e4SLinus Torvalds bh->b_state |= b_state; 16541da177e4SLinus Torvalds tail = bh; 16551da177e4SLinus Torvalds bh = bh->b_this_page; 16561da177e4SLinus Torvalds } while (bh); 16571da177e4SLinus Torvalds tail->b_this_page = head; 16581da177e4SLinus Torvalds 1659600f111eSMatthew Wilcox (Oracle) spin_lock(&folio->mapping->i_private_lock); 16608e2e1756SPankaj Raghav if (folio_test_uptodate(folio) || folio_test_dirty(folio)) { 16611da177e4SLinus Torvalds bh = head; 16621da177e4SLinus Torvalds do { 16638e2e1756SPankaj Raghav if (folio_test_dirty(folio)) 16641da177e4SLinus Torvalds set_buffer_dirty(bh); 16658e2e1756SPankaj Raghav if (folio_test_uptodate(folio)) 16661da177e4SLinus Torvalds set_buffer_uptodate(bh); 16671da177e4SLinus Torvalds bh = bh->b_this_page; 16681da177e4SLinus Torvalds } while (bh != head); 16691da177e4SLinus Torvalds } 16708e2e1756SPankaj Raghav folio_attach_private(folio, head); 1671600f111eSMatthew Wilcox (Oracle) spin_unlock(&folio->mapping->i_private_lock); 16723decb856SMatthew Wilcox (Oracle) 16733decb856SMatthew Wilcox (Oracle) return head; 16748e2e1756SPankaj Raghav } 16751da177e4SLinus Torvalds EXPORT_SYMBOL(create_empty_buffers); 16761da177e4SLinus Torvalds 167729f3ad7dSJan Kara /** 167829f3ad7dSJan Kara * clean_bdev_aliases: clean a range of buffers in block device 167929f3ad7dSJan Kara * @bdev: Block device to clean buffers in 168029f3ad7dSJan Kara * @block: Start of a range of blocks to clean 168129f3ad7dSJan Kara * @len: Number of blocks to clean 16821da177e4SLinus Torvalds * 168329f3ad7dSJan Kara * We are taking a range of blocks for data and we don't want writeback of any 168429f3ad7dSJan Kara * buffer-cache aliases starting from return from this function and until the 168529f3ad7dSJan Kara * moment when something will explicitly mark the buffer dirty (hopefully that 168629f3ad7dSJan Kara * will not happen until we will free that block ;-) We don't even need to mark 168729f3ad7dSJan Kara * it not-uptodate - nobody can expect anything from a newly allocated buffer 168829f3ad7dSJan Kara * anyway. We used to use unmap_buffer() for such invalidation, but that was 168929f3ad7dSJan Kara * wrong. We definitely don't want to mark the alias unmapped, for example - it 169029f3ad7dSJan Kara * would confuse anyone who might pick it with bread() afterwards... 169129f3ad7dSJan Kara * 169229f3ad7dSJan Kara * Also.. Note that bforget() doesn't lock the buffer. So there can be 169329f3ad7dSJan Kara * writeout I/O going on against recently-freed buffers. We don't wait on that 169429f3ad7dSJan Kara * I/O in bforget() - it's more efficient to wait on the I/O only if we really 169529f3ad7dSJan Kara * need to. That happens here. 16961da177e4SLinus Torvalds */ 169729f3ad7dSJan Kara void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) 16981da177e4SLinus Torvalds { 169929f3ad7dSJan Kara struct inode *bd_inode = bdev->bd_inode; 170029f3ad7dSJan Kara struct address_space *bd_mapping = bd_inode->i_mapping; 17019e0b6f31SMatthew Wilcox (Oracle) struct folio_batch fbatch; 17024b04646cSMatthew Wilcox (Oracle) pgoff_t index = ((loff_t)block << bd_inode->i_blkbits) / PAGE_SIZE; 170329f3ad7dSJan Kara pgoff_t end; 1704c10f778dSJan Kara int i, count; 170529f3ad7dSJan Kara struct buffer_head *bh; 170629f3ad7dSJan Kara struct buffer_head *head; 17071da177e4SLinus Torvalds 17084b04646cSMatthew Wilcox (Oracle) end = ((loff_t)(block + len - 1) << bd_inode->i_blkbits) / PAGE_SIZE; 17099e0b6f31SMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 17109e0b6f31SMatthew Wilcox (Oracle) while (filemap_get_folios(bd_mapping, &index, end, &fbatch)) { 17119e0b6f31SMatthew Wilcox (Oracle) count = folio_batch_count(&fbatch); 1712c10f778dSJan Kara for (i = 0; i < count; i++) { 17139e0b6f31SMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 17141da177e4SLinus Torvalds 17159e0b6f31SMatthew Wilcox (Oracle) if (!folio_buffers(folio)) 171629f3ad7dSJan Kara continue; 171729f3ad7dSJan Kara /* 1718600f111eSMatthew Wilcox (Oracle) * We use folio lock instead of bd_mapping->i_private_lock 171929f3ad7dSJan Kara * to pin buffers here since we can afford to sleep and 172029f3ad7dSJan Kara * it scales better than a global spinlock lock. 172129f3ad7dSJan Kara */ 17229e0b6f31SMatthew Wilcox (Oracle) folio_lock(folio); 17239e0b6f31SMatthew Wilcox (Oracle) /* Recheck when the folio is locked which pins bhs */ 17249e0b6f31SMatthew Wilcox (Oracle) head = folio_buffers(folio); 17259e0b6f31SMatthew Wilcox (Oracle) if (!head) 172629f3ad7dSJan Kara goto unlock_page; 172729f3ad7dSJan Kara bh = head; 172829f3ad7dSJan Kara do { 17296c006a9dSChandan Rajendra if (!buffer_mapped(bh) || (bh->b_blocknr < block)) 173029f3ad7dSJan Kara goto next; 173129f3ad7dSJan Kara if (bh->b_blocknr >= block + len) 173229f3ad7dSJan Kara break; 173329f3ad7dSJan Kara clear_buffer_dirty(bh); 173429f3ad7dSJan Kara wait_on_buffer(bh); 173529f3ad7dSJan Kara clear_buffer_req(bh); 173629f3ad7dSJan Kara next: 173729f3ad7dSJan Kara bh = bh->b_this_page; 173829f3ad7dSJan Kara } while (bh != head); 173929f3ad7dSJan Kara unlock_page: 17409e0b6f31SMatthew Wilcox (Oracle) folio_unlock(folio); 174129f3ad7dSJan Kara } 17429e0b6f31SMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 174329f3ad7dSJan Kara cond_resched(); 1744c10f778dSJan Kara /* End of range already reached? */ 1745c10f778dSJan Kara if (index > end || !index) 1746c10f778dSJan Kara break; 17471da177e4SLinus Torvalds } 17481da177e4SLinus Torvalds } 174929f3ad7dSJan Kara EXPORT_SYMBOL(clean_bdev_aliases); 17501da177e4SLinus Torvalds 1751c6c8c3e7SPankaj Raghav static struct buffer_head *folio_create_buffers(struct folio *folio, 1752c6c8c3e7SPankaj Raghav struct inode *inode, 1753c6c8c3e7SPankaj Raghav unsigned int b_state) 175445bce8f3SLinus Torvalds { 17553decb856SMatthew Wilcox (Oracle) struct buffer_head *bh; 17563decb856SMatthew Wilcox (Oracle) 1757c6c8c3e7SPankaj Raghav BUG_ON(!folio_test_locked(folio)); 175845bce8f3SLinus Torvalds 17593decb856SMatthew Wilcox (Oracle) bh = folio_buffers(folio); 17603decb856SMatthew Wilcox (Oracle) if (!bh) 17610a88810dSMatthew Wilcox (Oracle) bh = create_empty_buffers(folio, 17623decb856SMatthew Wilcox (Oracle) 1 << READ_ONCE(inode->i_blkbits), b_state); 17633decb856SMatthew Wilcox (Oracle) return bh; 176445bce8f3SLinus Torvalds } 176545bce8f3SLinus Torvalds 176645bce8f3SLinus Torvalds /* 17671da177e4SLinus Torvalds * NOTE! All mapped/uptodate combinations are valid: 17681da177e4SLinus Torvalds * 17691da177e4SLinus Torvalds * Mapped Uptodate Meaning 17701da177e4SLinus Torvalds * 17711da177e4SLinus Torvalds * No No "unknown" - must do get_block() 17721da177e4SLinus Torvalds * No Yes "hole" - zero-filled 17731da177e4SLinus Torvalds * Yes No "allocated" - allocated on disk, not read in 17741da177e4SLinus Torvalds * Yes Yes "valid" - allocated and up-to-date in memory. 17751da177e4SLinus Torvalds * 17761da177e4SLinus Torvalds * "Dirty" is valid only with the last case (mapped+uptodate). 17771da177e4SLinus Torvalds */ 17781da177e4SLinus Torvalds 17791da177e4SLinus Torvalds /* 178017bf23a9SMatthew Wilcox (Oracle) * While block_write_full_folio is writing back the dirty buffers under 17811da177e4SLinus Torvalds * the page lock, whoever dirtied the buffers may decide to clean them 17821da177e4SLinus Torvalds * again at any time. We handle that by only looking at the buffer 17831da177e4SLinus Torvalds * state inside lock_buffer(). 17841da177e4SLinus Torvalds * 178517bf23a9SMatthew Wilcox (Oracle) * If block_write_full_folio() is called for regular writeback 17861da177e4SLinus Torvalds * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a 17871da177e4SLinus Torvalds * locked buffer. This only can happen if someone has written the buffer 17881da177e4SLinus Torvalds * directly, with submit_bh(). At the address_space level PageWriteback 17891da177e4SLinus Torvalds * prevents this contention from occurring. 17906e34eeddSTheodore Ts'o * 179117bf23a9SMatthew Wilcox (Oracle) * If block_write_full_folio() is called with wbc->sync_mode == 179270fd7614SChristoph Hellwig * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this 1793721a9602SJens Axboe * causes the writes to be flagged as synchronous writes. 17941da177e4SLinus Torvalds */ 179553418a18SMatthew Wilcox (Oracle) int __block_write_full_folio(struct inode *inode, struct folio *folio, 179614059f66SMatthew Wilcox (Oracle) get_block_t *get_block, struct writeback_control *wbc) 17971da177e4SLinus Torvalds { 17981da177e4SLinus Torvalds int err; 17991da177e4SLinus Torvalds sector_t block; 18001da177e4SLinus Torvalds sector_t last_block; 1801f0fbd5fcSAndrew Morton struct buffer_head *bh, *head; 1802fa399c31SMatthew Wilcox (Oracle) size_t blocksize; 18031da177e4SLinus Torvalds int nr_underway = 0; 18043ae72869SBart Van Assche blk_opf_t write_flags = wbc_to_write_flags(wbc); 18051da177e4SLinus Torvalds 180653418a18SMatthew Wilcox (Oracle) head = folio_create_buffers(folio, inode, 18071da177e4SLinus Torvalds (1 << BH_Dirty) | (1 << BH_Uptodate)); 18081da177e4SLinus Torvalds 18091da177e4SLinus Torvalds /* 1810e621900aSMatthew Wilcox (Oracle) * Be very careful. We have no exclusion from block_dirty_folio 18111da177e4SLinus Torvalds * here, and the (potentially unmapped) buffers may become dirty at 18121da177e4SLinus Torvalds * any time. If a buffer becomes dirty here after we've inspected it 181353418a18SMatthew Wilcox (Oracle) * then we just miss that fact, and the folio stays dirty. 18141da177e4SLinus Torvalds * 1815e621900aSMatthew Wilcox (Oracle) * Buffers outside i_size may be dirtied by block_dirty_folio; 18161da177e4SLinus Torvalds * handle that here by just cleaning them. 18171da177e4SLinus Torvalds */ 18181da177e4SLinus Torvalds 18191da177e4SLinus Torvalds bh = head; 182045bce8f3SLinus Torvalds blocksize = bh->b_size; 182145bce8f3SLinus Torvalds 1822fa399c31SMatthew Wilcox (Oracle) block = div_u64(folio_pos(folio), blocksize); 1823fa399c31SMatthew Wilcox (Oracle) last_block = div_u64(i_size_read(inode) - 1, blocksize); 18241da177e4SLinus Torvalds 18251da177e4SLinus Torvalds /* 18261da177e4SLinus Torvalds * Get all the dirty buffers mapped to disk addresses and 18271da177e4SLinus Torvalds * handle any aliases from the underlying blockdev's mapping. 18281da177e4SLinus Torvalds */ 18291da177e4SLinus Torvalds do { 18301da177e4SLinus Torvalds if (block > last_block) { 18311da177e4SLinus Torvalds /* 18321da177e4SLinus Torvalds * mapped buffers outside i_size will occur, because 183353418a18SMatthew Wilcox (Oracle) * this folio can be outside i_size when there is a 18341da177e4SLinus Torvalds * truncate in progress. 18351da177e4SLinus Torvalds */ 18361da177e4SLinus Torvalds /* 183717bf23a9SMatthew Wilcox (Oracle) * The buffer was zeroed by block_write_full_folio() 18381da177e4SLinus Torvalds */ 18391da177e4SLinus Torvalds clear_buffer_dirty(bh); 18401da177e4SLinus Torvalds set_buffer_uptodate(bh); 184129a814d2SAlex Tomas } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && 184229a814d2SAlex Tomas buffer_dirty(bh)) { 1843b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 18441da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 18451da177e4SLinus Torvalds if (err) 18461da177e4SLinus Torvalds goto recover; 184729a814d2SAlex Tomas clear_buffer_delay(bh); 18481da177e4SLinus Torvalds if (buffer_new(bh)) { 18491da177e4SLinus Torvalds /* blockdev mappings never come here */ 18501da177e4SLinus Torvalds clear_buffer_new(bh); 1851e64855c6SJan Kara clean_bdev_bh_alias(bh); 18521da177e4SLinus Torvalds } 18531da177e4SLinus Torvalds } 18541da177e4SLinus Torvalds bh = bh->b_this_page; 18551da177e4SLinus Torvalds block++; 18561da177e4SLinus Torvalds } while (bh != head); 18571da177e4SLinus Torvalds 18581da177e4SLinus Torvalds do { 18591da177e4SLinus Torvalds if (!buffer_mapped(bh)) 18601da177e4SLinus Torvalds continue; 18611da177e4SLinus Torvalds /* 18621da177e4SLinus Torvalds * If it's a fully non-blocking write attempt and we cannot 186353418a18SMatthew Wilcox (Oracle) * lock the buffer then redirty the folio. Note that this can 18645b0830cbSJens Axboe * potentially cause a busy-wait loop from writeback threads 18655b0830cbSJens Axboe * and kswapd activity, but those code paths have their own 18665b0830cbSJens Axboe * higher-level throttling. 18671da177e4SLinus Torvalds */ 18681b430beeSWu Fengguang if (wbc->sync_mode != WB_SYNC_NONE) { 18691da177e4SLinus Torvalds lock_buffer(bh); 1870ca5de404SNick Piggin } else if (!trylock_buffer(bh)) { 187153418a18SMatthew Wilcox (Oracle) folio_redirty_for_writepage(wbc, folio); 18721da177e4SLinus Torvalds continue; 18731da177e4SLinus Torvalds } 18741da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 187514059f66SMatthew Wilcox (Oracle) mark_buffer_async_write_endio(bh, 187614059f66SMatthew Wilcox (Oracle) end_buffer_async_write); 18771da177e4SLinus Torvalds } else { 18781da177e4SLinus Torvalds unlock_buffer(bh); 18791da177e4SLinus Torvalds } 18801da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 18811da177e4SLinus Torvalds 18821da177e4SLinus Torvalds /* 188353418a18SMatthew Wilcox (Oracle) * The folio and its buffers are protected by the writeback flag, 188453418a18SMatthew Wilcox (Oracle) * so we can drop the bh refcounts early. 18851da177e4SLinus Torvalds */ 188653418a18SMatthew Wilcox (Oracle) BUG_ON(folio_test_writeback(folio)); 188753418a18SMatthew Wilcox (Oracle) folio_start_writeback(folio); 18881da177e4SLinus Torvalds 18891da177e4SLinus Torvalds do { 18901da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 18911da177e4SLinus Torvalds if (buffer_async_write(bh)) { 18921420c4a5SBart Van Assche submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc); 18931da177e4SLinus Torvalds nr_underway++; 1894ad576e63SNick Piggin } 18951da177e4SLinus Torvalds bh = next; 18961da177e4SLinus Torvalds } while (bh != head); 189753418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 18981da177e4SLinus Torvalds 18991da177e4SLinus Torvalds err = 0; 19001da177e4SLinus Torvalds done: 19011da177e4SLinus Torvalds if (nr_underway == 0) { 19021da177e4SLinus Torvalds /* 190353418a18SMatthew Wilcox (Oracle) * The folio was marked dirty, but the buffers were 19041da177e4SLinus Torvalds * clean. Someone wrote them back by hand with 190579f59784SZhang Yi * write_dirty_buffer/submit_bh. A rare case. 19061da177e4SLinus Torvalds */ 190753418a18SMatthew Wilcox (Oracle) folio_end_writeback(folio); 19083d67f2d7SNick Piggin 19091da177e4SLinus Torvalds /* 191053418a18SMatthew Wilcox (Oracle) * The folio and buffer_heads can be released at any time from 19111da177e4SLinus Torvalds * here on. 19121da177e4SLinus Torvalds */ 19131da177e4SLinus Torvalds } 19141da177e4SLinus Torvalds return err; 19151da177e4SLinus Torvalds 19161da177e4SLinus Torvalds recover: 19171da177e4SLinus Torvalds /* 19181da177e4SLinus Torvalds * ENOSPC, or some other error. We may already have added some 19191da177e4SLinus Torvalds * blocks to the file, so we need to write these out to avoid 19201da177e4SLinus Torvalds * exposing stale data. 192153418a18SMatthew Wilcox (Oracle) * The folio is currently locked and not marked for writeback 19221da177e4SLinus Torvalds */ 19231da177e4SLinus Torvalds bh = head; 19241da177e4SLinus Torvalds /* Recovery: lock and submit the mapped buffers */ 19251da177e4SLinus Torvalds do { 192629a814d2SAlex Tomas if (buffer_mapped(bh) && buffer_dirty(bh) && 192729a814d2SAlex Tomas !buffer_delay(bh)) { 19281da177e4SLinus Torvalds lock_buffer(bh); 192914059f66SMatthew Wilcox (Oracle) mark_buffer_async_write_endio(bh, 193014059f66SMatthew Wilcox (Oracle) end_buffer_async_write); 19311da177e4SLinus Torvalds } else { 19321da177e4SLinus Torvalds /* 19331da177e4SLinus Torvalds * The buffer may have been set dirty during 193453418a18SMatthew Wilcox (Oracle) * attachment to a dirty folio. 19351da177e4SLinus Torvalds */ 19361da177e4SLinus Torvalds clear_buffer_dirty(bh); 19371da177e4SLinus Torvalds } 19381da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 193953418a18SMatthew Wilcox (Oracle) folio_set_error(folio); 194053418a18SMatthew Wilcox (Oracle) BUG_ON(folio_test_writeback(folio)); 194153418a18SMatthew Wilcox (Oracle) mapping_set_error(folio->mapping, err); 194253418a18SMatthew Wilcox (Oracle) folio_start_writeback(folio); 19431da177e4SLinus Torvalds do { 19441da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 19451da177e4SLinus Torvalds if (buffer_async_write(bh)) { 19461da177e4SLinus Torvalds clear_buffer_dirty(bh); 19471420c4a5SBart Van Assche submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc); 19481da177e4SLinus Torvalds nr_underway++; 1949ad576e63SNick Piggin } 19501da177e4SLinus Torvalds bh = next; 19511da177e4SLinus Torvalds } while (bh != head); 195253418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 19531da177e4SLinus Torvalds goto done; 19541da177e4SLinus Torvalds } 195553418a18SMatthew Wilcox (Oracle) EXPORT_SYMBOL(__block_write_full_folio); 19561da177e4SLinus Torvalds 1957afddba49SNick Piggin /* 19584a9622f2SMatthew Wilcox (Oracle) * If a folio has any new buffers, zero them out here, and mark them uptodate 1959afddba49SNick Piggin * and dirty so they'll be written out (in order to prevent uninitialised 1960afddba49SNick Piggin * block data from leaking). And clear the new bit. 1961afddba49SNick Piggin */ 19624a9622f2SMatthew Wilcox (Oracle) void folio_zero_new_buffers(struct folio *folio, size_t from, size_t to) 1963afddba49SNick Piggin { 19644a9622f2SMatthew Wilcox (Oracle) size_t block_start, block_end; 1965afddba49SNick Piggin struct buffer_head *head, *bh; 1966afddba49SNick Piggin 19674a9622f2SMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 19684a9622f2SMatthew Wilcox (Oracle) head = folio_buffers(folio); 19694a9622f2SMatthew Wilcox (Oracle) if (!head) 1970afddba49SNick Piggin return; 1971afddba49SNick Piggin 19724a9622f2SMatthew Wilcox (Oracle) bh = head; 1973afddba49SNick Piggin block_start = 0; 1974afddba49SNick Piggin do { 1975afddba49SNick Piggin block_end = block_start + bh->b_size; 1976afddba49SNick Piggin 1977afddba49SNick Piggin if (buffer_new(bh)) { 1978afddba49SNick Piggin if (block_end > from && block_start < to) { 19794a9622f2SMatthew Wilcox (Oracle) if (!folio_test_uptodate(folio)) { 19804a9622f2SMatthew Wilcox (Oracle) size_t start, xend; 1981afddba49SNick Piggin 1982afddba49SNick Piggin start = max(from, block_start); 19834a9622f2SMatthew Wilcox (Oracle) xend = min(to, block_end); 1984afddba49SNick Piggin 19854a9622f2SMatthew Wilcox (Oracle) folio_zero_segment(folio, start, xend); 1986afddba49SNick Piggin set_buffer_uptodate(bh); 1987afddba49SNick Piggin } 1988afddba49SNick Piggin 1989afddba49SNick Piggin clear_buffer_new(bh); 1990afddba49SNick Piggin mark_buffer_dirty(bh); 1991afddba49SNick Piggin } 1992afddba49SNick Piggin } 1993afddba49SNick Piggin 1994afddba49SNick Piggin block_start = block_end; 1995afddba49SNick Piggin bh = bh->b_this_page; 1996afddba49SNick Piggin } while (bh != head); 1997afddba49SNick Piggin } 19984a9622f2SMatthew Wilcox (Oracle) EXPORT_SYMBOL(folio_zero_new_buffers); 1999afddba49SNick Piggin 20004aa8cdd5SChristoph Hellwig static int 2001ae259a9cSChristoph Hellwig iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, 20026d49cc85SChristoph Hellwig const struct iomap *iomap) 2003ae259a9cSChristoph Hellwig { 200480844194SMatthew Wilcox (Oracle) loff_t offset = (loff_t)block << inode->i_blkbits; 2005ae259a9cSChristoph Hellwig 2006ae259a9cSChristoph Hellwig bh->b_bdev = iomap->bdev; 2007ae259a9cSChristoph Hellwig 2008ae259a9cSChristoph Hellwig /* 2009ae259a9cSChristoph Hellwig * Block points to offset in file we need to map, iomap contains 2010ae259a9cSChristoph Hellwig * the offset at which the map starts. If the map ends before the 2011ae259a9cSChristoph Hellwig * current block, then do not map the buffer and let the caller 2012ae259a9cSChristoph Hellwig * handle it. 2013ae259a9cSChristoph Hellwig */ 20144aa8cdd5SChristoph Hellwig if (offset >= iomap->offset + iomap->length) 20154aa8cdd5SChristoph Hellwig return -EIO; 2016ae259a9cSChristoph Hellwig 2017ae259a9cSChristoph Hellwig switch (iomap->type) { 2018ae259a9cSChristoph Hellwig case IOMAP_HOLE: 2019ae259a9cSChristoph Hellwig /* 2020ae259a9cSChristoph Hellwig * If the buffer is not up to date or beyond the current EOF, 2021ae259a9cSChristoph Hellwig * we need to mark it as new to ensure sub-block zeroing is 2022ae259a9cSChristoph Hellwig * executed if necessary. 2023ae259a9cSChristoph Hellwig */ 2024ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 2025ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 2026ae259a9cSChristoph Hellwig set_buffer_new(bh); 20274aa8cdd5SChristoph Hellwig return 0; 2028ae259a9cSChristoph Hellwig case IOMAP_DELALLOC: 2029ae259a9cSChristoph Hellwig if (!buffer_uptodate(bh) || 2030ae259a9cSChristoph Hellwig (offset >= i_size_read(inode))) 2031ae259a9cSChristoph Hellwig set_buffer_new(bh); 2032ae259a9cSChristoph Hellwig set_buffer_uptodate(bh); 2033ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 2034ae259a9cSChristoph Hellwig set_buffer_delay(bh); 20354aa8cdd5SChristoph Hellwig return 0; 2036ae259a9cSChristoph Hellwig case IOMAP_UNWRITTEN: 2037ae259a9cSChristoph Hellwig /* 20383d7b6b21SAndreas Gruenbacher * For unwritten regions, we always need to ensure that regions 20393d7b6b21SAndreas Gruenbacher * in the block we are not writing to are zeroed. Mark the 20403d7b6b21SAndreas Gruenbacher * buffer as new to ensure this. 2041ae259a9cSChristoph Hellwig */ 2042ae259a9cSChristoph Hellwig set_buffer_new(bh); 2043ae259a9cSChristoph Hellwig set_buffer_unwritten(bh); 2044df561f66SGustavo A. R. Silva fallthrough; 2045ae259a9cSChristoph Hellwig case IOMAP_MAPPED: 20463d7b6b21SAndreas Gruenbacher if ((iomap->flags & IOMAP_F_NEW) || 2047381c0432SChristoph Hellwig offset >= i_size_read(inode)) { 2048381c0432SChristoph Hellwig /* 2049381c0432SChristoph Hellwig * This can happen if truncating the block device races 2050381c0432SChristoph Hellwig * with the check in the caller as i_size updates on 2051381c0432SChristoph Hellwig * block devices aren't synchronized by i_rwsem for 2052381c0432SChristoph Hellwig * block devices. 2053381c0432SChristoph Hellwig */ 2054381c0432SChristoph Hellwig if (S_ISBLK(inode->i_mode)) 2055381c0432SChristoph Hellwig return -EIO; 2056ae259a9cSChristoph Hellwig set_buffer_new(bh); 2057381c0432SChristoph Hellwig } 205819fe5f64SAndreas Gruenbacher bh->b_blocknr = (iomap->addr + offset - iomap->offset) >> 205919fe5f64SAndreas Gruenbacher inode->i_blkbits; 2060ae259a9cSChristoph Hellwig set_buffer_mapped(bh); 20614aa8cdd5SChristoph Hellwig return 0; 20624aa8cdd5SChristoph Hellwig default: 20634aa8cdd5SChristoph Hellwig WARN_ON_ONCE(1); 20644aa8cdd5SChristoph Hellwig return -EIO; 2065ae259a9cSChristoph Hellwig } 2066ae259a9cSChristoph Hellwig } 2067ae259a9cSChristoph Hellwig 2068d1bd0b4eSMatthew Wilcox (Oracle) int __block_write_begin_int(struct folio *folio, loff_t pos, unsigned len, 20696d49cc85SChristoph Hellwig get_block_t *get_block, const struct iomap *iomap) 20701da177e4SLinus Torvalds { 2071b0619401SMatthew Wilcox (Oracle) size_t from = offset_in_folio(folio, pos); 2072b0619401SMatthew Wilcox (Oracle) size_t to = from + len; 2073d1bd0b4eSMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 2074b0619401SMatthew Wilcox (Oracle) size_t block_start, block_end; 20751da177e4SLinus Torvalds sector_t block; 20761da177e4SLinus Torvalds int err = 0; 2077b0619401SMatthew Wilcox (Oracle) size_t blocksize; 20781da177e4SLinus Torvalds struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; 20791da177e4SLinus Torvalds 2080d1bd0b4eSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 2081b0619401SMatthew Wilcox (Oracle) BUG_ON(to > folio_size(folio)); 20821da177e4SLinus Torvalds BUG_ON(from > to); 20831da177e4SLinus Torvalds 2084c6c8c3e7SPankaj Raghav head = folio_create_buffers(folio, inode, 0); 208545bce8f3SLinus Torvalds blocksize = head->b_size; 2086b0619401SMatthew Wilcox (Oracle) block = div_u64(folio_pos(folio), blocksize); 20871da177e4SLinus Torvalds 20881da177e4SLinus Torvalds for (bh = head, block_start = 0; bh != head || !block_start; 20891da177e4SLinus Torvalds block++, block_start=block_end, bh = bh->b_this_page) { 20901da177e4SLinus Torvalds block_end = block_start + blocksize; 20911da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 2092d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 20931da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 20941da177e4SLinus Torvalds set_buffer_uptodate(bh); 20951da177e4SLinus Torvalds } 20961da177e4SLinus Torvalds continue; 20971da177e4SLinus Torvalds } 20981da177e4SLinus Torvalds if (buffer_new(bh)) 20991da177e4SLinus Torvalds clear_buffer_new(bh); 21001da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2101b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 21024aa8cdd5SChristoph Hellwig if (get_block) 21031da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 21044aa8cdd5SChristoph Hellwig else 21054aa8cdd5SChristoph Hellwig err = iomap_to_bh(inode, block, bh, iomap); 21061da177e4SLinus Torvalds if (err) 2107f3ddbdc6SNick Piggin break; 2108ae259a9cSChristoph Hellwig 21091da177e4SLinus Torvalds if (buffer_new(bh)) { 2110e64855c6SJan Kara clean_bdev_bh_alias(bh); 2111d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 2112637aff46SNick Piggin clear_buffer_new(bh); 21131da177e4SLinus Torvalds set_buffer_uptodate(bh); 2114637aff46SNick Piggin mark_buffer_dirty(bh); 21151da177e4SLinus Torvalds continue; 21161da177e4SLinus Torvalds } 2117eebd2aa3SChristoph Lameter if (block_end > to || block_start < from) 2118d1bd0b4eSMatthew Wilcox (Oracle) folio_zero_segments(folio, 2119eebd2aa3SChristoph Lameter to, block_end, 2120eebd2aa3SChristoph Lameter block_start, from); 21211da177e4SLinus Torvalds continue; 21221da177e4SLinus Torvalds } 21231da177e4SLinus Torvalds } 2124d1bd0b4eSMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) { 21251da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 21261da177e4SLinus Torvalds set_buffer_uptodate(bh); 21271da177e4SLinus Torvalds continue; 21281da177e4SLinus Torvalds } 21291da177e4SLinus Torvalds if (!buffer_uptodate(bh) && !buffer_delay(bh) && 213033a266ddSDavid Chinner !buffer_unwritten(bh) && 21311da177e4SLinus Torvalds (block_start < from || block_end > to)) { 2132e7ea1129SZhang Yi bh_read_nowait(bh, 0); 21331da177e4SLinus Torvalds *wait_bh++=bh; 21341da177e4SLinus Torvalds } 21351da177e4SLinus Torvalds } 21361da177e4SLinus Torvalds /* 21371da177e4SLinus Torvalds * If we issued read requests - let them complete. 21381da177e4SLinus Torvalds */ 21391da177e4SLinus Torvalds while(wait_bh > wait) { 21401da177e4SLinus Torvalds wait_on_buffer(*--wait_bh); 21411da177e4SLinus Torvalds if (!buffer_uptodate(*wait_bh)) 2142f3ddbdc6SNick Piggin err = -EIO; 21431da177e4SLinus Torvalds } 2144f9f07b6cSJan Kara if (unlikely(err)) 21454a9622f2SMatthew Wilcox (Oracle) folio_zero_new_buffers(folio, from, to); 21461da177e4SLinus Torvalds return err; 21471da177e4SLinus Torvalds } 2148ae259a9cSChristoph Hellwig 2149ae259a9cSChristoph Hellwig int __block_write_begin(struct page *page, loff_t pos, unsigned len, 2150ae259a9cSChristoph Hellwig get_block_t *get_block) 2151ae259a9cSChristoph Hellwig { 2152d1bd0b4eSMatthew Wilcox (Oracle) return __block_write_begin_int(page_folio(page), pos, len, get_block, 2153d1bd0b4eSMatthew Wilcox (Oracle) NULL); 2154ae259a9cSChristoph Hellwig } 2155ebdec241SChristoph Hellwig EXPORT_SYMBOL(__block_write_begin); 21561da177e4SLinus Torvalds 2157a524fcfeSBean Huo static void __block_commit_write(struct folio *folio, size_t from, size_t to) 21581da177e4SLinus Torvalds { 21598c6cb3e3SMatthew Wilcox (Oracle) size_t block_start, block_end; 21608c6cb3e3SMatthew Wilcox (Oracle) bool partial = false; 21611da177e4SLinus Torvalds unsigned blocksize; 21621da177e4SLinus Torvalds struct buffer_head *bh, *head; 21631da177e4SLinus Torvalds 21648c6cb3e3SMatthew Wilcox (Oracle) bh = head = folio_buffers(folio); 216545bce8f3SLinus Torvalds blocksize = bh->b_size; 21661da177e4SLinus Torvalds 216745bce8f3SLinus Torvalds block_start = 0; 216845bce8f3SLinus Torvalds do { 21691da177e4SLinus Torvalds block_end = block_start + blocksize; 21701da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 21711da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 21728c6cb3e3SMatthew Wilcox (Oracle) partial = true; 21731da177e4SLinus Torvalds } else { 21741da177e4SLinus Torvalds set_buffer_uptodate(bh); 21751da177e4SLinus Torvalds mark_buffer_dirty(bh); 21761da177e4SLinus Torvalds } 21774ebd3aecSYang Guo if (buffer_new(bh)) 2178afddba49SNick Piggin clear_buffer_new(bh); 217945bce8f3SLinus Torvalds 218045bce8f3SLinus Torvalds block_start = block_end; 218145bce8f3SLinus Torvalds bh = bh->b_this_page; 218245bce8f3SLinus Torvalds } while (bh != head); 21831da177e4SLinus Torvalds 21841da177e4SLinus Torvalds /* 21851da177e4SLinus Torvalds * If this is a partial write which happened to make all buffers 21862c69e205SMatthew Wilcox (Oracle) * uptodate then we can optimize away a bogus read_folio() for 21878c6cb3e3SMatthew Wilcox (Oracle) * the next read(). Here we 'discover' whether the folio went 21881da177e4SLinus Torvalds * uptodate as a result of this (potentially partial) write. 21891da177e4SLinus Torvalds */ 21901da177e4SLinus Torvalds if (!partial) 21918c6cb3e3SMatthew Wilcox (Oracle) folio_mark_uptodate(folio); 21921da177e4SLinus Torvalds } 21931da177e4SLinus Torvalds 21941da177e4SLinus Torvalds /* 2195155130a4SChristoph Hellwig * block_write_begin takes care of the basic task of block allocation and 2196155130a4SChristoph Hellwig * bringing partial write blocks uptodate first. 2197155130a4SChristoph Hellwig * 21987bb46a67Snpiggin@suse.de * The filesystem needs to handle block truncation upon failure. 2199afddba49SNick Piggin */ 2200155130a4SChristoph Hellwig int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, 2201b3992d1eSMatthew Wilcox (Oracle) struct page **pagep, get_block_t *get_block) 2202afddba49SNick Piggin { 220309cbfeafSKirill A. Shutemov pgoff_t index = pos >> PAGE_SHIFT; 2204afddba49SNick Piggin struct page *page; 22056e1db88dSChristoph Hellwig int status; 2206afddba49SNick Piggin 2207b7446e7cSMatthew Wilcox (Oracle) page = grab_cache_page_write_begin(mapping, index); 22086e1db88dSChristoph Hellwig if (!page) 22096e1db88dSChristoph Hellwig return -ENOMEM; 2210afddba49SNick Piggin 22116e1db88dSChristoph Hellwig status = __block_write_begin(page, pos, len, get_block); 2212afddba49SNick Piggin if (unlikely(status)) { 2213afddba49SNick Piggin unlock_page(page); 221409cbfeafSKirill A. Shutemov put_page(page); 22156e1db88dSChristoph Hellwig page = NULL; 2216afddba49SNick Piggin } 2217afddba49SNick Piggin 22186e1db88dSChristoph Hellwig *pagep = page; 2219afddba49SNick Piggin return status; 2220afddba49SNick Piggin } 2221afddba49SNick Piggin EXPORT_SYMBOL(block_write_begin); 2222afddba49SNick Piggin 2223afddba49SNick Piggin int block_write_end(struct file *file, struct address_space *mapping, 2224afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2225afddba49SNick Piggin struct page *page, void *fsdata) 2226afddba49SNick Piggin { 22278c6cb3e3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 22288c6cb3e3SMatthew Wilcox (Oracle) size_t start = pos - folio_pos(folio); 2229afddba49SNick Piggin 2230afddba49SNick Piggin if (unlikely(copied < len)) { 2231afddba49SNick Piggin /* 22322c69e205SMatthew Wilcox (Oracle) * The buffers that were written will now be uptodate, so 22332c69e205SMatthew Wilcox (Oracle) * we don't have to worry about a read_folio reading them 22342c69e205SMatthew Wilcox (Oracle) * and overwriting a partial write. However if we have 22352c69e205SMatthew Wilcox (Oracle) * encountered a short write and only partially written 22362c69e205SMatthew Wilcox (Oracle) * into a buffer, it will not be marked uptodate, so a 22372c69e205SMatthew Wilcox (Oracle) * read_folio might come in and destroy our partial write. 2238afddba49SNick Piggin * 2239afddba49SNick Piggin * Do the simplest thing, and just treat any short write to a 22408c6cb3e3SMatthew Wilcox (Oracle) * non uptodate folio as a zero-length write, and force the 2241afddba49SNick Piggin * caller to redo the whole thing. 2242afddba49SNick Piggin */ 22438c6cb3e3SMatthew Wilcox (Oracle) if (!folio_test_uptodate(folio)) 2244afddba49SNick Piggin copied = 0; 2245afddba49SNick Piggin 22464a9622f2SMatthew Wilcox (Oracle) folio_zero_new_buffers(folio, start+copied, start+len); 2247afddba49SNick Piggin } 22488c6cb3e3SMatthew Wilcox (Oracle) flush_dcache_folio(folio); 2249afddba49SNick Piggin 2250afddba49SNick Piggin /* This could be a short (even 0-length) commit */ 2251489b7e72SBean Huo __block_commit_write(folio, start, start + copied); 2252afddba49SNick Piggin 2253afddba49SNick Piggin return copied; 2254afddba49SNick Piggin } 2255afddba49SNick Piggin EXPORT_SYMBOL(block_write_end); 2256afddba49SNick Piggin 2257afddba49SNick Piggin int generic_write_end(struct file *file, struct address_space *mapping, 2258afddba49SNick Piggin loff_t pos, unsigned len, unsigned copied, 2259afddba49SNick Piggin struct page *page, void *fsdata) 2260afddba49SNick Piggin { 22618af54f29SChristoph Hellwig struct inode *inode = mapping->host; 22628af54f29SChristoph Hellwig loff_t old_size = inode->i_size; 22638af54f29SChristoph Hellwig bool i_size_changed = false; 22648af54f29SChristoph Hellwig 2265afddba49SNick Piggin copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); 22668af54f29SChristoph Hellwig 22678af54f29SChristoph Hellwig /* 22688af54f29SChristoph Hellwig * No need to use i_size_read() here, the i_size cannot change under us 22698af54f29SChristoph Hellwig * because we hold i_rwsem. 22708af54f29SChristoph Hellwig * 22718af54f29SChristoph Hellwig * But it's important to update i_size while still holding page lock: 22728af54f29SChristoph Hellwig * page writeout could otherwise come in and zero beyond i_size. 22738af54f29SChristoph Hellwig */ 22748af54f29SChristoph Hellwig if (pos + copied > inode->i_size) { 22758af54f29SChristoph Hellwig i_size_write(inode, pos + copied); 22768af54f29SChristoph Hellwig i_size_changed = true; 22778af54f29SChristoph Hellwig } 22788af54f29SChristoph Hellwig 22798af54f29SChristoph Hellwig unlock_page(page); 22807a77dad7SAndreas Gruenbacher put_page(page); 22818af54f29SChristoph Hellwig 22828af54f29SChristoph Hellwig if (old_size < pos) 22838af54f29SChristoph Hellwig pagecache_isize_extended(inode, old_size, pos); 22848af54f29SChristoph Hellwig /* 22858af54f29SChristoph Hellwig * Don't mark the inode dirty under page lock. First, it unnecessarily 22868af54f29SChristoph Hellwig * makes the holding time of page lock longer. Second, it forces lock 22878af54f29SChristoph Hellwig * ordering of page lock and transaction start for journaling 22888af54f29SChristoph Hellwig * filesystems. 22898af54f29SChristoph Hellwig */ 22908af54f29SChristoph Hellwig if (i_size_changed) 22918af54f29SChristoph Hellwig mark_inode_dirty(inode); 229226ddb1f4SAndreas Gruenbacher return copied; 2293afddba49SNick Piggin } 2294afddba49SNick Piggin EXPORT_SYMBOL(generic_write_end); 2295afddba49SNick Piggin 2296afddba49SNick Piggin /* 22972e7e80f7SMatthew Wilcox (Oracle) * block_is_partially_uptodate checks whether buffers within a folio are 22988ab22b9aSHisashi Hifumi * uptodate or not. 22998ab22b9aSHisashi Hifumi * 23002e7e80f7SMatthew Wilcox (Oracle) * Returns true if all buffers which correspond to the specified part 23012e7e80f7SMatthew Wilcox (Oracle) * of the folio are uptodate. 23028ab22b9aSHisashi Hifumi */ 23032e7e80f7SMatthew Wilcox (Oracle) bool block_is_partially_uptodate(struct folio *folio, size_t from, size_t count) 23048ab22b9aSHisashi Hifumi { 23058ab22b9aSHisashi Hifumi unsigned block_start, block_end, blocksize; 23068ab22b9aSHisashi Hifumi unsigned to; 23078ab22b9aSHisashi Hifumi struct buffer_head *bh, *head; 23082e7e80f7SMatthew Wilcox (Oracle) bool ret = true; 23098ab22b9aSHisashi Hifumi 23102e7e80f7SMatthew Wilcox (Oracle) head = folio_buffers(folio); 23112e7e80f7SMatthew Wilcox (Oracle) if (!head) 23122e7e80f7SMatthew Wilcox (Oracle) return false; 231345bce8f3SLinus Torvalds blocksize = head->b_size; 23142e7e80f7SMatthew Wilcox (Oracle) to = min_t(unsigned, folio_size(folio) - from, count); 23158ab22b9aSHisashi Hifumi to = from + to; 23162e7e80f7SMatthew Wilcox (Oracle) if (from < blocksize && to > folio_size(folio) - blocksize) 23172e7e80f7SMatthew Wilcox (Oracle) return false; 23188ab22b9aSHisashi Hifumi 23198ab22b9aSHisashi Hifumi bh = head; 23208ab22b9aSHisashi Hifumi block_start = 0; 23218ab22b9aSHisashi Hifumi do { 23228ab22b9aSHisashi Hifumi block_end = block_start + blocksize; 23238ab22b9aSHisashi Hifumi if (block_end > from && block_start < to) { 23248ab22b9aSHisashi Hifumi if (!buffer_uptodate(bh)) { 23252e7e80f7SMatthew Wilcox (Oracle) ret = false; 23268ab22b9aSHisashi Hifumi break; 23278ab22b9aSHisashi Hifumi } 23288ab22b9aSHisashi Hifumi if (block_end >= to) 23298ab22b9aSHisashi Hifumi break; 23308ab22b9aSHisashi Hifumi } 23318ab22b9aSHisashi Hifumi block_start = block_end; 23328ab22b9aSHisashi Hifumi bh = bh->b_this_page; 23338ab22b9aSHisashi Hifumi } while (bh != head); 23348ab22b9aSHisashi Hifumi 23358ab22b9aSHisashi Hifumi return ret; 23368ab22b9aSHisashi Hifumi } 23378ab22b9aSHisashi Hifumi EXPORT_SYMBOL(block_is_partially_uptodate); 23388ab22b9aSHisashi Hifumi 23398ab22b9aSHisashi Hifumi /* 23402c69e205SMatthew Wilcox (Oracle) * Generic "read_folio" function for block devices that have the normal 23411da177e4SLinus Torvalds * get_block functionality. This is most of the block device filesystems. 23422c69e205SMatthew Wilcox (Oracle) * Reads the folio asynchronously --- the unlock_buffer() and 23431da177e4SLinus Torvalds * set/clear_buffer_uptodate() functions propagate buffer state into the 23442c69e205SMatthew Wilcox (Oracle) * folio once IO has completed. 23451da177e4SLinus Torvalds */ 23462c69e205SMatthew Wilcox (Oracle) int block_read_full_folio(struct folio *folio, get_block_t *get_block) 23471da177e4SLinus Torvalds { 23482c69e205SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 23491da177e4SLinus Torvalds sector_t iblock, lblock; 23501da177e4SLinus Torvalds struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 2351fa399c31SMatthew Wilcox (Oracle) size_t blocksize; 23521da177e4SLinus Torvalds int nr, i; 23531da177e4SLinus Torvalds int fully_mapped = 1; 2354b7a6eb22SMatthew Wilcox (Oracle) bool page_error = false; 23554fa512ceSEric Biggers loff_t limit = i_size_read(inode); 23564fa512ceSEric Biggers 23574fa512ceSEric Biggers /* This is needed for ext4. */ 23584fa512ceSEric Biggers if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode)) 23594fa512ceSEric Biggers limit = inode->i_sb->s_maxbytes; 23601da177e4SLinus Torvalds 23612c69e205SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(folio_test_large(folio), folio); 23622c69e205SMatthew Wilcox (Oracle) 2363c6c8c3e7SPankaj Raghav head = folio_create_buffers(folio, inode, 0); 236445bce8f3SLinus Torvalds blocksize = head->b_size; 23651da177e4SLinus Torvalds 2366fa399c31SMatthew Wilcox (Oracle) iblock = div_u64(folio_pos(folio), blocksize); 2367fa399c31SMatthew Wilcox (Oracle) lblock = div_u64(limit + blocksize - 1, blocksize); 23681da177e4SLinus Torvalds bh = head; 23691da177e4SLinus Torvalds nr = 0; 23701da177e4SLinus Torvalds i = 0; 23711da177e4SLinus Torvalds 23721da177e4SLinus Torvalds do { 23731da177e4SLinus Torvalds if (buffer_uptodate(bh)) 23741da177e4SLinus Torvalds continue; 23751da177e4SLinus Torvalds 23761da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2377c64610baSAndrew Morton int err = 0; 2378c64610baSAndrew Morton 23791da177e4SLinus Torvalds fully_mapped = 0; 23801da177e4SLinus Torvalds if (iblock < lblock) { 2381b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 2382c64610baSAndrew Morton err = get_block(inode, iblock, bh, 0); 2383b7a6eb22SMatthew Wilcox (Oracle) if (err) { 23842c69e205SMatthew Wilcox (Oracle) folio_set_error(folio); 2385b7a6eb22SMatthew Wilcox (Oracle) page_error = true; 2386b7a6eb22SMatthew Wilcox (Oracle) } 23871da177e4SLinus Torvalds } 23881da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 23892c69e205SMatthew Wilcox (Oracle) folio_zero_range(folio, i * blocksize, 23902c69e205SMatthew Wilcox (Oracle) blocksize); 2391c64610baSAndrew Morton if (!err) 23921da177e4SLinus Torvalds set_buffer_uptodate(bh); 23931da177e4SLinus Torvalds continue; 23941da177e4SLinus Torvalds } 23951da177e4SLinus Torvalds /* 23961da177e4SLinus Torvalds * get_block() might have updated the buffer 23971da177e4SLinus Torvalds * synchronously 23981da177e4SLinus Torvalds */ 23991da177e4SLinus Torvalds if (buffer_uptodate(bh)) 24001da177e4SLinus Torvalds continue; 24011da177e4SLinus Torvalds } 24021da177e4SLinus Torvalds arr[nr++] = bh; 24031da177e4SLinus Torvalds } while (i++, iblock++, (bh = bh->b_this_page) != head); 24041da177e4SLinus Torvalds 24051da177e4SLinus Torvalds if (fully_mapped) 24062c69e205SMatthew Wilcox (Oracle) folio_set_mappedtodisk(folio); 24071da177e4SLinus Torvalds 24081da177e4SLinus Torvalds if (!nr) { 24091da177e4SLinus Torvalds /* 24106ba924d3SMatthew Wilcox (Oracle) * All buffers are uptodate or get_block() returned an 24116ba924d3SMatthew Wilcox (Oracle) * error when trying to map them - we can finish the read. 24121da177e4SLinus Torvalds */ 24136ba924d3SMatthew Wilcox (Oracle) folio_end_read(folio, !page_error); 24141da177e4SLinus Torvalds return 0; 24151da177e4SLinus Torvalds } 24161da177e4SLinus Torvalds 24171da177e4SLinus Torvalds /* Stage two: lock the buffers */ 24181da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 24191da177e4SLinus Torvalds bh = arr[i]; 24201da177e4SLinus Torvalds lock_buffer(bh); 24211da177e4SLinus Torvalds mark_buffer_async_read(bh); 24221da177e4SLinus Torvalds } 24231da177e4SLinus Torvalds 24241da177e4SLinus Torvalds /* 24251da177e4SLinus Torvalds * Stage 3: start the IO. Check for uptodateness 24261da177e4SLinus Torvalds * inside the buffer lock in case another process reading 24271da177e4SLinus Torvalds * the underlying blockdev brought it uptodate (the sct fix). 24281da177e4SLinus Torvalds */ 24291da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 24301da177e4SLinus Torvalds bh = arr[i]; 24311da177e4SLinus Torvalds if (buffer_uptodate(bh)) 24321da177e4SLinus Torvalds end_buffer_async_read(bh, 1); 24331da177e4SLinus Torvalds else 24341420c4a5SBart Van Assche submit_bh(REQ_OP_READ, bh); 24351da177e4SLinus Torvalds } 24361da177e4SLinus Torvalds return 0; 24371da177e4SLinus Torvalds } 24382c69e205SMatthew Wilcox (Oracle) EXPORT_SYMBOL(block_read_full_folio); 24391da177e4SLinus Torvalds 24401da177e4SLinus Torvalds /* utility function for filesystems that need to do work on expanding 244189e10787SNick Piggin * truncates. Uses filesystem pagecache writes to allow the filesystem to 24421da177e4SLinus Torvalds * deal with the hole. 24431da177e4SLinus Torvalds */ 244489e10787SNick Piggin int generic_cont_expand_simple(struct inode *inode, loff_t size) 24451da177e4SLinus Torvalds { 24461da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 244753b524b8SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 24481da177e4SLinus Torvalds struct page *page; 24491468c6f4SAlexander Potapenko void *fsdata = NULL; 24501da177e4SLinus Torvalds int err; 24511da177e4SLinus Torvalds 2452c08d3b0eSnpiggin@suse.de err = inode_newsize_ok(inode, size); 2453c08d3b0eSnpiggin@suse.de if (err) 24541da177e4SLinus Torvalds goto out; 24551da177e4SLinus Torvalds 245653b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(NULL, mapping, size, 0, &page, &fsdata); 245789e10787SNick Piggin if (err) 245805eb0b51SOGAWA Hirofumi goto out; 245905eb0b51SOGAWA Hirofumi 246053b524b8SMatthew Wilcox (Oracle) err = aops->write_end(NULL, mapping, size, 0, 0, page, fsdata); 246189e10787SNick Piggin BUG_ON(err > 0); 246205eb0b51SOGAWA Hirofumi 246305eb0b51SOGAWA Hirofumi out: 246405eb0b51SOGAWA Hirofumi return err; 246505eb0b51SOGAWA Hirofumi } 24661fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_cont_expand_simple); 246705eb0b51SOGAWA Hirofumi 2468f1e3af72SAdrian Bunk static int cont_expand_zero(struct file *file, struct address_space *mapping, 246989e10787SNick Piggin loff_t pos, loff_t *bytes) 247005eb0b51SOGAWA Hirofumi { 247189e10787SNick Piggin struct inode *inode = mapping->host; 247253b524b8SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 247393407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 247489e10787SNick Piggin struct page *page; 24751468c6f4SAlexander Potapenko void *fsdata = NULL; 247689e10787SNick Piggin pgoff_t index, curidx; 247789e10787SNick Piggin loff_t curpos; 247889e10787SNick Piggin unsigned zerofrom, offset, len; 247989e10787SNick Piggin int err = 0; 248005eb0b51SOGAWA Hirofumi 248109cbfeafSKirill A. Shutemov index = pos >> PAGE_SHIFT; 248209cbfeafSKirill A. Shutemov offset = pos & ~PAGE_MASK; 248389e10787SNick Piggin 248409cbfeafSKirill A. Shutemov while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { 248509cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 248689e10787SNick Piggin if (zerofrom & (blocksize-1)) { 248789e10787SNick Piggin *bytes |= (blocksize-1); 248889e10787SNick Piggin (*bytes)++; 248989e10787SNick Piggin } 249009cbfeafSKirill A. Shutemov len = PAGE_SIZE - zerofrom; 249189e10787SNick Piggin 249253b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(file, mapping, curpos, len, 249389e10787SNick Piggin &page, &fsdata); 249489e10787SNick Piggin if (err) 249589e10787SNick Piggin goto out; 2496eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 249753b524b8SMatthew Wilcox (Oracle) err = aops->write_end(file, mapping, curpos, len, len, 249889e10787SNick Piggin page, fsdata); 249989e10787SNick Piggin if (err < 0) 250089e10787SNick Piggin goto out; 250189e10787SNick Piggin BUG_ON(err != len); 250289e10787SNick Piggin err = 0; 2503061e9746SOGAWA Hirofumi 2504061e9746SOGAWA Hirofumi balance_dirty_pages_ratelimited(mapping); 2505c2ca0fcdSMikulas Patocka 250608d405c8SDavidlohr Bueso if (fatal_signal_pending(current)) { 2507c2ca0fcdSMikulas Patocka err = -EINTR; 2508c2ca0fcdSMikulas Patocka goto out; 2509c2ca0fcdSMikulas Patocka } 251089e10787SNick Piggin } 251189e10787SNick Piggin 251289e10787SNick Piggin /* page covers the boundary, find the boundary offset */ 251389e10787SNick Piggin if (index == curidx) { 251409cbfeafSKirill A. Shutemov zerofrom = curpos & ~PAGE_MASK; 251589e10787SNick Piggin /* if we will expand the thing last block will be filled */ 251689e10787SNick Piggin if (offset <= zerofrom) { 251789e10787SNick Piggin goto out; 251889e10787SNick Piggin } 251989e10787SNick Piggin if (zerofrom & (blocksize-1)) { 252089e10787SNick Piggin *bytes |= (blocksize-1); 252189e10787SNick Piggin (*bytes)++; 252289e10787SNick Piggin } 252389e10787SNick Piggin len = offset - zerofrom; 252489e10787SNick Piggin 252553b524b8SMatthew Wilcox (Oracle) err = aops->write_begin(file, mapping, curpos, len, 252689e10787SNick Piggin &page, &fsdata); 252789e10787SNick Piggin if (err) 252889e10787SNick Piggin goto out; 2529eebd2aa3SChristoph Lameter zero_user(page, zerofrom, len); 253053b524b8SMatthew Wilcox (Oracle) err = aops->write_end(file, mapping, curpos, len, len, 253189e10787SNick Piggin page, fsdata); 253289e10787SNick Piggin if (err < 0) 253389e10787SNick Piggin goto out; 253489e10787SNick Piggin BUG_ON(err != len); 253589e10787SNick Piggin err = 0; 253689e10787SNick Piggin } 253789e10787SNick Piggin out: 253889e10787SNick Piggin return err; 25391da177e4SLinus Torvalds } 25401da177e4SLinus Torvalds 25411da177e4SLinus Torvalds /* 25421da177e4SLinus Torvalds * For moronic filesystems that do not allow holes in file. 25431da177e4SLinus Torvalds * We may have to extend the file. 25441da177e4SLinus Torvalds */ 2545282dc178SChristoph Hellwig int cont_write_begin(struct file *file, struct address_space *mapping, 2546be3bbbc5SMatthew Wilcox (Oracle) loff_t pos, unsigned len, 254789e10787SNick Piggin struct page **pagep, void **fsdata, 254889e10787SNick Piggin get_block_t *get_block, loff_t *bytes) 25491da177e4SLinus Torvalds { 25501da177e4SLinus Torvalds struct inode *inode = mapping->host; 255193407472SFabian Frederick unsigned int blocksize = i_blocksize(inode); 255293407472SFabian Frederick unsigned int zerofrom; 255389e10787SNick Piggin int err; 25541da177e4SLinus Torvalds 255589e10787SNick Piggin err = cont_expand_zero(file, mapping, pos, bytes); 255689e10787SNick Piggin if (err) 2557155130a4SChristoph Hellwig return err; 25581da177e4SLinus Torvalds 255909cbfeafSKirill A. Shutemov zerofrom = *bytes & ~PAGE_MASK; 256089e10787SNick Piggin if (pos+len > *bytes && zerofrom & (blocksize-1)) { 25611da177e4SLinus Torvalds *bytes |= (blocksize-1); 25621da177e4SLinus Torvalds (*bytes)++; 25631da177e4SLinus Torvalds } 25641da177e4SLinus Torvalds 2565b3992d1eSMatthew Wilcox (Oracle) return block_write_begin(mapping, pos, len, pagep, get_block); 25661da177e4SLinus Torvalds } 25671fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(cont_write_begin); 25681da177e4SLinus Torvalds 2569a524fcfeSBean Huo void block_commit_write(struct page *page, unsigned from, unsigned to) 25701da177e4SLinus Torvalds { 25718c6cb3e3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 2572489b7e72SBean Huo __block_commit_write(folio, from, to); 25731da177e4SLinus Torvalds } 25741fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_commit_write); 25751da177e4SLinus Torvalds 257654171690SDavid Chinner /* 257754171690SDavid Chinner * block_page_mkwrite() is not allowed to change the file size as it gets 257854171690SDavid Chinner * called from a page fault handler when a page is first dirtied. Hence we must 257954171690SDavid Chinner * be careful to check for EOF conditions here. We set the page up correctly 258054171690SDavid Chinner * for a written page which means we get ENOSPC checking when writing into 258154171690SDavid Chinner * holes and correct delalloc and unwritten extent mapping on filesystems that 258254171690SDavid Chinner * support these features. 258354171690SDavid Chinner * 258454171690SDavid Chinner * We are not allowed to take the i_mutex here so we have to play games to 258554171690SDavid Chinner * protect against truncate races as the page could now be beyond EOF. Because 25867bb46a67Snpiggin@suse.de * truncate writes the inode size before removing pages, once we have the 258754171690SDavid Chinner * page lock we can determine safely if the page is beyond EOF. If it is not 258854171690SDavid Chinner * beyond EOF, then the page is guaranteed safe against truncation until we 258954171690SDavid Chinner * unlock the page. 2590ea13a864SJan Kara * 259114da9200SJan Kara * Direct callers of this function should protect against filesystem freezing 25925c500029SRoss Zwisler * using sb_start_pagefault() - sb_end_pagefault() functions. 259354171690SDavid Chinner */ 25945c500029SRoss Zwisler int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, 259554171690SDavid Chinner get_block_t get_block) 259654171690SDavid Chinner { 2597fe181377SMatthew Wilcox (Oracle) struct folio *folio = page_folio(vmf->page); 2598496ad9aaSAl Viro struct inode *inode = file_inode(vma->vm_file); 259954171690SDavid Chinner unsigned long end; 260054171690SDavid Chinner loff_t size; 260124da4fabSJan Kara int ret; 260254171690SDavid Chinner 2603fe181377SMatthew Wilcox (Oracle) folio_lock(folio); 260454171690SDavid Chinner size = i_size_read(inode); 2605fe181377SMatthew Wilcox (Oracle) if ((folio->mapping != inode->i_mapping) || 2606fe181377SMatthew Wilcox (Oracle) (folio_pos(folio) >= size)) { 260724da4fabSJan Kara /* We overload EFAULT to mean page got truncated */ 260824da4fabSJan Kara ret = -EFAULT; 260924da4fabSJan Kara goto out_unlock; 261054171690SDavid Chinner } 261154171690SDavid Chinner 2612fe181377SMatthew Wilcox (Oracle) end = folio_size(folio); 2613fe181377SMatthew Wilcox (Oracle) /* folio is wholly or partially inside EOF */ 2614fe181377SMatthew Wilcox (Oracle) if (folio_pos(folio) + end > size) 2615fe181377SMatthew Wilcox (Oracle) end = size - folio_pos(folio); 261654171690SDavid Chinner 2617fe181377SMatthew Wilcox (Oracle) ret = __block_write_begin_int(folio, 0, end, get_block, NULL); 2618a524fcfeSBean Huo if (unlikely(ret)) 261924da4fabSJan Kara goto out_unlock; 2620a524fcfeSBean Huo 2621a524fcfeSBean Huo __block_commit_write(folio, 0, end); 2622a524fcfeSBean Huo 2623fe181377SMatthew Wilcox (Oracle) folio_mark_dirty(folio); 2624fe181377SMatthew Wilcox (Oracle) folio_wait_stable(folio); 262524da4fabSJan Kara return 0; 262624da4fabSJan Kara out_unlock: 2627fe181377SMatthew Wilcox (Oracle) folio_unlock(folio); 262854171690SDavid Chinner return ret; 262954171690SDavid Chinner } 26301fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_page_mkwrite); 26311da177e4SLinus Torvalds 26321da177e4SLinus Torvalds int block_truncate_page(struct address_space *mapping, 26331da177e4SLinus Torvalds loff_t from, get_block_t *get_block) 26341da177e4SLinus Torvalds { 263509cbfeafSKirill A. Shutemov pgoff_t index = from >> PAGE_SHIFT; 26361da177e4SLinus Torvalds unsigned blocksize; 263754b21a79SAndrew Morton sector_t iblock; 26386d68f644SMatthew Wilcox (Oracle) size_t offset, length, pos; 26391da177e4SLinus Torvalds struct inode *inode = mapping->host; 26406d68f644SMatthew Wilcox (Oracle) struct folio *folio; 26411da177e4SLinus Torvalds struct buffer_head *bh; 2642dc7cb2d2SJiapeng Chong int err = 0; 26431da177e4SLinus Torvalds 264493407472SFabian Frederick blocksize = i_blocksize(inode); 26456d68f644SMatthew Wilcox (Oracle) length = from & (blocksize - 1); 26461da177e4SLinus Torvalds 26471da177e4SLinus Torvalds /* Block boundary? Nothing to do */ 26481da177e4SLinus Torvalds if (!length) 26491da177e4SLinus Torvalds return 0; 26501da177e4SLinus Torvalds 26511da177e4SLinus Torvalds length = blocksize - length; 26524b04646cSMatthew Wilcox (Oracle) iblock = ((loff_t)index * PAGE_SIZE) >> inode->i_blkbits; 26531da177e4SLinus Torvalds 26546d68f644SMatthew Wilcox (Oracle) folio = filemap_grab_folio(mapping, index); 26556d68f644SMatthew Wilcox (Oracle) if (IS_ERR(folio)) 26566d68f644SMatthew Wilcox (Oracle) return PTR_ERR(folio); 26571da177e4SLinus Torvalds 26586d68f644SMatthew Wilcox (Oracle) bh = folio_buffers(folio); 26593decb856SMatthew Wilcox (Oracle) if (!bh) 26600a88810dSMatthew Wilcox (Oracle) bh = create_empty_buffers(folio, blocksize, 0); 26611da177e4SLinus Torvalds 26621da177e4SLinus Torvalds /* Find the buffer that contains "offset" */ 26636d68f644SMatthew Wilcox (Oracle) offset = offset_in_folio(folio, from); 26641da177e4SLinus Torvalds pos = blocksize; 26651da177e4SLinus Torvalds while (offset >= pos) { 26661da177e4SLinus Torvalds bh = bh->b_this_page; 26671da177e4SLinus Torvalds iblock++; 26681da177e4SLinus Torvalds pos += blocksize; 26691da177e4SLinus Torvalds } 26701da177e4SLinus Torvalds 26711da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 2672b0cf2321SBadari Pulavarty WARN_ON(bh->b_size != blocksize); 26731da177e4SLinus Torvalds err = get_block(inode, iblock, bh, 0); 26741da177e4SLinus Torvalds if (err) 26751da177e4SLinus Torvalds goto unlock; 26761da177e4SLinus Torvalds /* unmapped? It's a hole - nothing to do */ 26771da177e4SLinus Torvalds if (!buffer_mapped(bh)) 26781da177e4SLinus Torvalds goto unlock; 26791da177e4SLinus Torvalds } 26801da177e4SLinus Torvalds 26811da177e4SLinus Torvalds /* Ok, it's mapped. Make sure it's up-to-date */ 26826d68f644SMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) 26831da177e4SLinus Torvalds set_buffer_uptodate(bh); 26841da177e4SLinus Torvalds 268533a266ddSDavid Chinner if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { 2686e7ea1129SZhang Yi err = bh_read(bh, 0); 26871da177e4SLinus Torvalds /* Uhhuh. Read error. Complain and punt. */ 2688e7ea1129SZhang Yi if (err < 0) 26891da177e4SLinus Torvalds goto unlock; 26901da177e4SLinus Torvalds } 26911da177e4SLinus Torvalds 26926d68f644SMatthew Wilcox (Oracle) folio_zero_range(folio, offset, length); 26931da177e4SLinus Torvalds mark_buffer_dirty(bh); 26941da177e4SLinus Torvalds 26951da177e4SLinus Torvalds unlock: 26966d68f644SMatthew Wilcox (Oracle) folio_unlock(folio); 26976d68f644SMatthew Wilcox (Oracle) folio_put(folio); 2698dc7cb2d2SJiapeng Chong 26991da177e4SLinus Torvalds return err; 27001da177e4SLinus Torvalds } 27011fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(block_truncate_page); 27021da177e4SLinus Torvalds 27031da177e4SLinus Torvalds /* 27041da177e4SLinus Torvalds * The generic ->writepage function for buffer-backed address_spaces 27051da177e4SLinus Torvalds */ 270617bf23a9SMatthew Wilcox (Oracle) int block_write_full_folio(struct folio *folio, struct writeback_control *wbc, 270717bf23a9SMatthew Wilcox (Oracle) void *get_block) 27081da177e4SLinus Torvalds { 2709bb0ea598SMatthew Wilcox (Oracle) struct inode * const inode = folio->mapping->host; 27101da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 27111da177e4SLinus Torvalds 2712bb0ea598SMatthew Wilcox (Oracle) /* Is the folio fully inside i_size? */ 2713bb0ea598SMatthew Wilcox (Oracle) if (folio_pos(folio) + folio_size(folio) <= i_size) 271414059f66SMatthew Wilcox (Oracle) return __block_write_full_folio(inode, folio, get_block, wbc); 27151da177e4SLinus Torvalds 2716bb0ea598SMatthew Wilcox (Oracle) /* Is the folio fully outside i_size? (truncate in progress) */ 2717bb0ea598SMatthew Wilcox (Oracle) if (folio_pos(folio) >= i_size) { 271853418a18SMatthew Wilcox (Oracle) folio_unlock(folio); 27191da177e4SLinus Torvalds return 0; /* don't care */ 27201da177e4SLinus Torvalds } 27211da177e4SLinus Torvalds 27221da177e4SLinus Torvalds /* 2723bb0ea598SMatthew Wilcox (Oracle) * The folio straddles i_size. It must be zeroed out on each and every 27242a61aa40SAdam Buchbinder * writepage invocation because it may be mmapped. "A file is mapped 27251da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 27261da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 27271da177e4SLinus Torvalds * writes to that region are not written out to the file." 27281da177e4SLinus Torvalds */ 2729bb0ea598SMatthew Wilcox (Oracle) folio_zero_segment(folio, offset_in_folio(folio, i_size), 2730bb0ea598SMatthew Wilcox (Oracle) folio_size(folio)); 273114059f66SMatthew Wilcox (Oracle) return __block_write_full_folio(inode, folio, get_block, wbc); 273235c80d5fSChris Mason } 273335c80d5fSChris Mason 27341da177e4SLinus Torvalds sector_t generic_block_bmap(struct address_space *mapping, sector_t block, 27351da177e4SLinus Torvalds get_block_t *get_block) 27361da177e4SLinus Torvalds { 27371da177e4SLinus Torvalds struct inode *inode = mapping->host; 27382a527d68SAlexander Potapenko struct buffer_head tmp = { 27392a527d68SAlexander Potapenko .b_size = i_blocksize(inode), 27402a527d68SAlexander Potapenko }; 27412a527d68SAlexander Potapenko 27421da177e4SLinus Torvalds get_block(inode, block, &tmp, 0); 27431da177e4SLinus Torvalds return tmp.b_blocknr; 27441da177e4SLinus Torvalds } 27451fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(generic_block_bmap); 27461da177e4SLinus Torvalds 27474246a0b6SChristoph Hellwig static void end_bio_bh_io_sync(struct bio *bio) 27481da177e4SLinus Torvalds { 27491da177e4SLinus Torvalds struct buffer_head *bh = bio->bi_private; 27501da177e4SLinus Torvalds 2751b7c44ed9SJens Axboe if (unlikely(bio_flagged(bio, BIO_QUIET))) 275208bafc03SKeith Mannthey set_bit(BH_Quiet, &bh->b_state); 275308bafc03SKeith Mannthey 27544e4cbee9SChristoph Hellwig bh->b_end_io(bh, !bio->bi_status); 27551da177e4SLinus Torvalds bio_put(bio); 27561da177e4SLinus Torvalds } 27571da177e4SLinus Torvalds 27585bdf402aSRitesh Harjani (IBM) static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh, 27591420c4a5SBart Van Assche struct writeback_control *wbc) 27601da177e4SLinus Torvalds { 27611420c4a5SBart Van Assche const enum req_op op = opf & REQ_OP_MASK; 27621da177e4SLinus Torvalds struct bio *bio; 27631da177e4SLinus Torvalds 27641da177e4SLinus Torvalds BUG_ON(!buffer_locked(bh)); 27651da177e4SLinus Torvalds BUG_ON(!buffer_mapped(bh)); 27661da177e4SLinus Torvalds BUG_ON(!bh->b_end_io); 27678fb0e342SAneesh Kumar K.V BUG_ON(buffer_delay(bh)); 27688fb0e342SAneesh Kumar K.V BUG_ON(buffer_unwritten(bh)); 27691da177e4SLinus Torvalds 277048fd4f93SJens Axboe /* 277148fd4f93SJens Axboe * Only clear out a write error when rewriting 27721da177e4SLinus Torvalds */ 27732a222ca9SMike Christie if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) 27741da177e4SLinus Torvalds clear_buffer_write_io_error(bh); 27751da177e4SLinus Torvalds 277607888c66SChristoph Hellwig if (buffer_meta(bh)) 27771420c4a5SBart Van Assche opf |= REQ_META; 277807888c66SChristoph Hellwig if (buffer_prio(bh)) 27791420c4a5SBart Van Assche opf |= REQ_PRIO; 278007888c66SChristoph Hellwig 27811420c4a5SBart Van Assche bio = bio_alloc(bh->b_bdev, 1, opf, GFP_NOIO); 27821da177e4SLinus Torvalds 27834f74d15fSEric Biggers fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO); 27844f74d15fSEric Biggers 27854f024f37SKent Overstreet bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); 27861da177e4SLinus Torvalds 2787741af75dSJohannes Thumshirn __bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); 27881da177e4SLinus Torvalds 27891da177e4SLinus Torvalds bio->bi_end_io = end_bio_bh_io_sync; 27901da177e4SLinus Torvalds bio->bi_private = bh; 27911da177e4SLinus Torvalds 279283c9c547SMing Lei /* Take care of bh's that straddle the end of the device */ 279383c9c547SMing Lei guard_bio_eod(bio); 279483c9c547SMing Lei 2795fd42df30SDennis Zhou if (wbc) { 2796fd42df30SDennis Zhou wbc_init_bio(wbc, bio); 279734e51a5eSTejun Heo wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size); 2798fd42df30SDennis Zhou } 2799fd42df30SDennis Zhou 28004e49ea4aSMike Christie submit_bio(bio); 28011da177e4SLinus Torvalds } 2802bafc0dbaSTejun Heo 28035bdf402aSRitesh Harjani (IBM) void submit_bh(blk_opf_t opf, struct buffer_head *bh) 280471368511SDarrick J. Wong { 28055bdf402aSRitesh Harjani (IBM) submit_bh_wbc(opf, bh, NULL); 280671368511SDarrick J. Wong } 28071fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(submit_bh); 28081da177e4SLinus Torvalds 28093ae72869SBart Van Assche void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) 28109cb569d6SChristoph Hellwig { 28119cb569d6SChristoph Hellwig lock_buffer(bh); 28129cb569d6SChristoph Hellwig if (!test_clear_buffer_dirty(bh)) { 28139cb569d6SChristoph Hellwig unlock_buffer(bh); 28149cb569d6SChristoph Hellwig return; 28159cb569d6SChristoph Hellwig } 28169cb569d6SChristoph Hellwig bh->b_end_io = end_buffer_write_sync; 28179cb569d6SChristoph Hellwig get_bh(bh); 28181420c4a5SBart Van Assche submit_bh(REQ_OP_WRITE | op_flags, bh); 28199cb569d6SChristoph Hellwig } 28209cb569d6SChristoph Hellwig EXPORT_SYMBOL(write_dirty_buffer); 28219cb569d6SChristoph Hellwig 28221da177e4SLinus Torvalds /* 28231da177e4SLinus Torvalds * For a data-integrity writeout, we need to wait upon any in-progress I/O 28241da177e4SLinus Torvalds * and then start new I/O and then wait upon it. The caller must have a ref on 28251da177e4SLinus Torvalds * the buffer_head. 28261da177e4SLinus Torvalds */ 28273ae72869SBart Van Assche int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) 28281da177e4SLinus Torvalds { 28291da177e4SLinus Torvalds WARN_ON(atomic_read(&bh->b_count) < 1); 28301da177e4SLinus Torvalds lock_buffer(bh); 28311da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 2832377254b2SXianting Tian /* 2833377254b2SXianting Tian * The bh should be mapped, but it might not be if the 2834377254b2SXianting Tian * device was hot-removed. Not much we can do but fail the I/O. 2835377254b2SXianting Tian */ 2836377254b2SXianting Tian if (!buffer_mapped(bh)) { 2837377254b2SXianting Tian unlock_buffer(bh); 2838377254b2SXianting Tian return -EIO; 2839377254b2SXianting Tian } 2840377254b2SXianting Tian 28411da177e4SLinus Torvalds get_bh(bh); 28421da177e4SLinus Torvalds bh->b_end_io = end_buffer_write_sync; 2843ab620620SRitesh Harjani (IBM) submit_bh(REQ_OP_WRITE | op_flags, bh); 28441da177e4SLinus Torvalds wait_on_buffer(bh); 2845ab620620SRitesh Harjani (IBM) if (!buffer_uptodate(bh)) 2846ab620620SRitesh Harjani (IBM) return -EIO; 28471da177e4SLinus Torvalds } else { 28481da177e4SLinus Torvalds unlock_buffer(bh); 28491da177e4SLinus Torvalds } 2850ab620620SRitesh Harjani (IBM) return 0; 28511da177e4SLinus Torvalds } 285287e99511SChristoph Hellwig EXPORT_SYMBOL(__sync_dirty_buffer); 285387e99511SChristoph Hellwig 285487e99511SChristoph Hellwig int sync_dirty_buffer(struct buffer_head *bh) 285587e99511SChristoph Hellwig { 285670fd7614SChristoph Hellwig return __sync_dirty_buffer(bh, REQ_SYNC); 285787e99511SChristoph Hellwig } 28581fe72eaaSH Hartley Sweeten EXPORT_SYMBOL(sync_dirty_buffer); 28591da177e4SLinus Torvalds 28601da177e4SLinus Torvalds /* 286168189fefSMatthew Wilcox (Oracle) * try_to_free_buffers() checks if all the buffers on this particular folio 28621da177e4SLinus Torvalds * are unused, and releases them if so. 28631da177e4SLinus Torvalds * 28641da177e4SLinus Torvalds * Exclusion against try_to_free_buffers may be obtained by either 2865600f111eSMatthew Wilcox (Oracle) * locking the folio or by holding its mapping's i_private_lock. 28661da177e4SLinus Torvalds * 286768189fefSMatthew Wilcox (Oracle) * If the folio is dirty but all the buffers are clean then we need to 286868189fefSMatthew Wilcox (Oracle) * be sure to mark the folio clean as well. This is because the folio 28691da177e4SLinus Torvalds * may be against a block device, and a later reattachment of buffers 287068189fefSMatthew Wilcox (Oracle) * to a dirty folio will set *all* buffers dirty. Which would corrupt 28711da177e4SLinus Torvalds * filesystem data on the same device. 28721da177e4SLinus Torvalds * 287368189fefSMatthew Wilcox (Oracle) * The same applies to regular filesystem folios: if all the buffers are 287468189fefSMatthew Wilcox (Oracle) * clean then we set the folio clean and proceed. To do that, we require 2875e621900aSMatthew Wilcox (Oracle) * total exclusion from block_dirty_folio(). That is obtained with 2876600f111eSMatthew Wilcox (Oracle) * i_private_lock. 28771da177e4SLinus Torvalds * 28781da177e4SLinus Torvalds * try_to_free_buffers() is non-blocking. 28791da177e4SLinus Torvalds */ 28801da177e4SLinus Torvalds static inline int buffer_busy(struct buffer_head *bh) 28811da177e4SLinus Torvalds { 28821da177e4SLinus Torvalds return atomic_read(&bh->b_count) | 28831da177e4SLinus Torvalds (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); 28841da177e4SLinus Torvalds } 28851da177e4SLinus Torvalds 288664394763SMatthew Wilcox (Oracle) static bool 288764394763SMatthew Wilcox (Oracle) drop_buffers(struct folio *folio, struct buffer_head **buffers_to_free) 28881da177e4SLinus Torvalds { 288964394763SMatthew Wilcox (Oracle) struct buffer_head *head = folio_buffers(folio); 28901da177e4SLinus Torvalds struct buffer_head *bh; 28911da177e4SLinus Torvalds 28921da177e4SLinus Torvalds bh = head; 28931da177e4SLinus Torvalds do { 28941da177e4SLinus Torvalds if (buffer_busy(bh)) 28951da177e4SLinus Torvalds goto failed; 28961da177e4SLinus Torvalds bh = bh->b_this_page; 28971da177e4SLinus Torvalds } while (bh != head); 28981da177e4SLinus Torvalds 28991da177e4SLinus Torvalds do { 29001da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 29011da177e4SLinus Torvalds 2902535ee2fbSJan Kara if (bh->b_assoc_map) 29031da177e4SLinus Torvalds __remove_assoc_queue(bh); 29041da177e4SLinus Torvalds bh = next; 29051da177e4SLinus Torvalds } while (bh != head); 29061da177e4SLinus Torvalds *buffers_to_free = head; 290764394763SMatthew Wilcox (Oracle) folio_detach_private(folio); 290864394763SMatthew Wilcox (Oracle) return true; 29091da177e4SLinus Torvalds failed: 291064394763SMatthew Wilcox (Oracle) return false; 29111da177e4SLinus Torvalds } 29121da177e4SLinus Torvalds 291368189fefSMatthew Wilcox (Oracle) bool try_to_free_buffers(struct folio *folio) 29141da177e4SLinus Torvalds { 291568189fefSMatthew Wilcox (Oracle) struct address_space * const mapping = folio->mapping; 29161da177e4SLinus Torvalds struct buffer_head *buffers_to_free = NULL; 291768189fefSMatthew Wilcox (Oracle) bool ret = 0; 29181da177e4SLinus Torvalds 291968189fefSMatthew Wilcox (Oracle) BUG_ON(!folio_test_locked(folio)); 292068189fefSMatthew Wilcox (Oracle) if (folio_test_writeback(folio)) 292168189fefSMatthew Wilcox (Oracle) return false; 29221da177e4SLinus Torvalds 29231da177e4SLinus Torvalds if (mapping == NULL) { /* can this still happen? */ 292464394763SMatthew Wilcox (Oracle) ret = drop_buffers(folio, &buffers_to_free); 29251da177e4SLinus Torvalds goto out; 29261da177e4SLinus Torvalds } 29271da177e4SLinus Torvalds 2928600f111eSMatthew Wilcox (Oracle) spin_lock(&mapping->i_private_lock); 292964394763SMatthew Wilcox (Oracle) ret = drop_buffers(folio, &buffers_to_free); 2930ecdfc978SLinus Torvalds 2931ecdfc978SLinus Torvalds /* 2932ecdfc978SLinus Torvalds * If the filesystem writes its buffers by hand (eg ext3) 293368189fefSMatthew Wilcox (Oracle) * then we can have clean buffers against a dirty folio. We 293468189fefSMatthew Wilcox (Oracle) * clean the folio here; otherwise the VM will never notice 2935ecdfc978SLinus Torvalds * that the filesystem did any IO at all. 2936ecdfc978SLinus Torvalds * 2937ecdfc978SLinus Torvalds * Also, during truncate, discard_buffer will have marked all 293868189fefSMatthew Wilcox (Oracle) * the folio's buffers clean. We discover that here and clean 293968189fefSMatthew Wilcox (Oracle) * the folio also. 294087df7241SNick Piggin * 2941600f111eSMatthew Wilcox (Oracle) * i_private_lock must be held over this entire operation in order 2942e621900aSMatthew Wilcox (Oracle) * to synchronise against block_dirty_folio and prevent the 294387df7241SNick Piggin * dirty bit from being lost. 2944ecdfc978SLinus Torvalds */ 294511f81becSTejun Heo if (ret) 294668189fefSMatthew Wilcox (Oracle) folio_cancel_dirty(folio); 2947600f111eSMatthew Wilcox (Oracle) spin_unlock(&mapping->i_private_lock); 29481da177e4SLinus Torvalds out: 29491da177e4SLinus Torvalds if (buffers_to_free) { 29501da177e4SLinus Torvalds struct buffer_head *bh = buffers_to_free; 29511da177e4SLinus Torvalds 29521da177e4SLinus Torvalds do { 29531da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 29541da177e4SLinus Torvalds free_buffer_head(bh); 29551da177e4SLinus Torvalds bh = next; 29561da177e4SLinus Torvalds } while (bh != buffers_to_free); 29571da177e4SLinus Torvalds } 29581da177e4SLinus Torvalds return ret; 29591da177e4SLinus Torvalds } 29601da177e4SLinus Torvalds EXPORT_SYMBOL(try_to_free_buffers); 29611da177e4SLinus Torvalds 29621da177e4SLinus Torvalds /* 29631da177e4SLinus Torvalds * Buffer-head allocation 29641da177e4SLinus Torvalds */ 296568279f9cSAlexey Dobriyan static struct kmem_cache *bh_cachep __ro_after_init; 29661da177e4SLinus Torvalds 29671da177e4SLinus Torvalds /* 29681da177e4SLinus Torvalds * Once the number of bh's in the machine exceeds this level, we start 29691da177e4SLinus Torvalds * stripping them in writeback. 29701da177e4SLinus Torvalds */ 297168279f9cSAlexey Dobriyan static unsigned long max_buffer_heads __ro_after_init; 29721da177e4SLinus Torvalds 29731da177e4SLinus Torvalds int buffer_heads_over_limit; 29741da177e4SLinus Torvalds 29751da177e4SLinus Torvalds struct bh_accounting { 29761da177e4SLinus Torvalds int nr; /* Number of live bh's */ 29771da177e4SLinus Torvalds int ratelimit; /* Limit cacheline bouncing */ 29781da177e4SLinus Torvalds }; 29791da177e4SLinus Torvalds 29801da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; 29811da177e4SLinus Torvalds 29821da177e4SLinus Torvalds static void recalc_bh_state(void) 29831da177e4SLinus Torvalds { 29841da177e4SLinus Torvalds int i; 29851da177e4SLinus Torvalds int tot = 0; 29861da177e4SLinus Torvalds 2987ee1be862SChristoph Lameter if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) 29881da177e4SLinus Torvalds return; 2989c7b92516SChristoph Lameter __this_cpu_write(bh_accounting.ratelimit, 0); 29908a143426SEric Dumazet for_each_online_cpu(i) 29911da177e4SLinus Torvalds tot += per_cpu(bh_accounting, i).nr; 29921da177e4SLinus Torvalds buffer_heads_over_limit = (tot > max_buffer_heads); 29931da177e4SLinus Torvalds } 29941da177e4SLinus Torvalds 2995dd0fc66fSAl Viro struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) 29961da177e4SLinus Torvalds { 2997019b4d12SRichard Kennedy struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); 29981da177e4SLinus Torvalds if (ret) { 2999a35afb83SChristoph Lameter INIT_LIST_HEAD(&ret->b_assoc_buffers); 3000f1e67e35SThomas Gleixner spin_lock_init(&ret->b_uptodate_lock); 3001c7b92516SChristoph Lameter preempt_disable(); 3002c7b92516SChristoph Lameter __this_cpu_inc(bh_accounting.nr); 30031da177e4SLinus Torvalds recalc_bh_state(); 3004c7b92516SChristoph Lameter preempt_enable(); 30051da177e4SLinus Torvalds } 30061da177e4SLinus Torvalds return ret; 30071da177e4SLinus Torvalds } 30081da177e4SLinus Torvalds EXPORT_SYMBOL(alloc_buffer_head); 30091da177e4SLinus Torvalds 30101da177e4SLinus Torvalds void free_buffer_head(struct buffer_head *bh) 30111da177e4SLinus Torvalds { 30121da177e4SLinus Torvalds BUG_ON(!list_empty(&bh->b_assoc_buffers)); 30131da177e4SLinus Torvalds kmem_cache_free(bh_cachep, bh); 3014c7b92516SChristoph Lameter preempt_disable(); 3015c7b92516SChristoph Lameter __this_cpu_dec(bh_accounting.nr); 30161da177e4SLinus Torvalds recalc_bh_state(); 3017c7b92516SChristoph Lameter preempt_enable(); 30181da177e4SLinus Torvalds } 30191da177e4SLinus Torvalds EXPORT_SYMBOL(free_buffer_head); 30201da177e4SLinus Torvalds 3021fc4d24c9SSebastian Andrzej Siewior static int buffer_exit_cpu_dead(unsigned int cpu) 30221da177e4SLinus Torvalds { 30231da177e4SLinus Torvalds int i; 30241da177e4SLinus Torvalds struct bh_lru *b = &per_cpu(bh_lrus, cpu); 30251da177e4SLinus Torvalds 30261da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 30271da177e4SLinus Torvalds brelse(b->bhs[i]); 30281da177e4SLinus Torvalds b->bhs[i] = NULL; 30291da177e4SLinus Torvalds } 3030c7b92516SChristoph Lameter this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); 30318a143426SEric Dumazet per_cpu(bh_accounting, cpu).nr = 0; 3032fc4d24c9SSebastian Andrzej Siewior return 0; 30331da177e4SLinus Torvalds } 30341da177e4SLinus Torvalds 3035389d1b08SAneesh Kumar K.V /** 3036a6b91919SRandy Dunlap * bh_uptodate_or_lock - Test whether the buffer is uptodate 3037389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3038389d1b08SAneesh Kumar K.V * 3039389d1b08SAneesh Kumar K.V * Return true if the buffer is up-to-date and false, 3040389d1b08SAneesh Kumar K.V * with the buffer locked, if not. 3041389d1b08SAneesh Kumar K.V */ 3042389d1b08SAneesh Kumar K.V int bh_uptodate_or_lock(struct buffer_head *bh) 3043389d1b08SAneesh Kumar K.V { 3044389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) { 3045389d1b08SAneesh Kumar K.V lock_buffer(bh); 3046389d1b08SAneesh Kumar K.V if (!buffer_uptodate(bh)) 3047389d1b08SAneesh Kumar K.V return 0; 3048389d1b08SAneesh Kumar K.V unlock_buffer(bh); 3049389d1b08SAneesh Kumar K.V } 3050389d1b08SAneesh Kumar K.V return 1; 3051389d1b08SAneesh Kumar K.V } 3052389d1b08SAneesh Kumar K.V EXPORT_SYMBOL(bh_uptodate_or_lock); 3053389d1b08SAneesh Kumar K.V 3054389d1b08SAneesh Kumar K.V /** 3055fdee117eSZhang Yi * __bh_read - Submit read for a locked buffer 3056389d1b08SAneesh Kumar K.V * @bh: struct buffer_head 3057fdee117eSZhang Yi * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ 3058fdee117eSZhang Yi * @wait: wait until reading finish 3059389d1b08SAneesh Kumar K.V * 3060fdee117eSZhang Yi * Returns zero on success or don't wait, and -EIO on error. 3061389d1b08SAneesh Kumar K.V */ 3062fdee117eSZhang Yi int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait) 3063389d1b08SAneesh Kumar K.V { 3064fdee117eSZhang Yi int ret = 0; 3065389d1b08SAneesh Kumar K.V 3066fdee117eSZhang Yi BUG_ON(!buffer_locked(bh)); 3067389d1b08SAneesh Kumar K.V 3068389d1b08SAneesh Kumar K.V get_bh(bh); 3069389d1b08SAneesh Kumar K.V bh->b_end_io = end_buffer_read_sync; 3070fdee117eSZhang Yi submit_bh(REQ_OP_READ | op_flags, bh); 3071fdee117eSZhang Yi if (wait) { 3072389d1b08SAneesh Kumar K.V wait_on_buffer(bh); 3073fdee117eSZhang Yi if (!buffer_uptodate(bh)) 3074fdee117eSZhang Yi ret = -EIO; 3075389d1b08SAneesh Kumar K.V } 3076fdee117eSZhang Yi return ret; 3077fdee117eSZhang Yi } 3078fdee117eSZhang Yi EXPORT_SYMBOL(__bh_read); 3079fdee117eSZhang Yi 3080fdee117eSZhang Yi /** 3081fdee117eSZhang Yi * __bh_read_batch - Submit read for a batch of unlocked buffers 3082fdee117eSZhang Yi * @nr: entry number of the buffer batch 3083fdee117eSZhang Yi * @bhs: a batch of struct buffer_head 3084fdee117eSZhang Yi * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ 3085fdee117eSZhang Yi * @force_lock: force to get a lock on the buffer if set, otherwise drops any 3086fdee117eSZhang Yi * buffer that cannot lock. 3087fdee117eSZhang Yi * 3088fdee117eSZhang Yi * Returns zero on success or don't wait, and -EIO on error. 3089fdee117eSZhang Yi */ 3090fdee117eSZhang Yi void __bh_read_batch(int nr, struct buffer_head *bhs[], 3091fdee117eSZhang Yi blk_opf_t op_flags, bool force_lock) 3092fdee117eSZhang Yi { 3093fdee117eSZhang Yi int i; 3094fdee117eSZhang Yi 3095fdee117eSZhang Yi for (i = 0; i < nr; i++) { 3096fdee117eSZhang Yi struct buffer_head *bh = bhs[i]; 3097fdee117eSZhang Yi 3098fdee117eSZhang Yi if (buffer_uptodate(bh)) 3099fdee117eSZhang Yi continue; 3100fdee117eSZhang Yi 3101fdee117eSZhang Yi if (force_lock) 3102fdee117eSZhang Yi lock_buffer(bh); 3103fdee117eSZhang Yi else 3104fdee117eSZhang Yi if (!trylock_buffer(bh)) 3105fdee117eSZhang Yi continue; 3106fdee117eSZhang Yi 3107fdee117eSZhang Yi if (buffer_uptodate(bh)) { 3108fdee117eSZhang Yi unlock_buffer(bh); 3109fdee117eSZhang Yi continue; 3110fdee117eSZhang Yi } 3111fdee117eSZhang Yi 3112fdee117eSZhang Yi bh->b_end_io = end_buffer_read_sync; 3113fdee117eSZhang Yi get_bh(bh); 3114fdee117eSZhang Yi submit_bh(REQ_OP_READ | op_flags, bh); 3115fdee117eSZhang Yi } 3116fdee117eSZhang Yi } 3117fdee117eSZhang Yi EXPORT_SYMBOL(__bh_read_batch); 3118389d1b08SAneesh Kumar K.V 31191da177e4SLinus Torvalds void __init buffer_init(void) 31201da177e4SLinus Torvalds { 312143be594aSZhang Yanfei unsigned long nrpages; 3122fc4d24c9SSebastian Andrzej Siewior int ret; 31231da177e4SLinus Torvalds 3124b98938c3SChristoph Lameter bh_cachep = kmem_cache_create("buffer_head", 3125b98938c3SChristoph Lameter sizeof(struct buffer_head), 0, 3126b98938c3SChristoph Lameter (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| 3127b98938c3SChristoph Lameter SLAB_MEM_SPREAD), 3128019b4d12SRichard Kennedy NULL); 31291da177e4SLinus Torvalds 31301da177e4SLinus Torvalds /* 31311da177e4SLinus Torvalds * Limit the bh occupancy to 10% of ZONE_NORMAL 31321da177e4SLinus Torvalds */ 31331da177e4SLinus Torvalds nrpages = (nr_free_buffer_pages() * 10) / 100; 31341da177e4SLinus Torvalds max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); 3135fc4d24c9SSebastian Andrzej Siewior ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", 3136fc4d24c9SSebastian Andrzej Siewior NULL, buffer_exit_cpu_dead); 3137fc4d24c9SSebastian Andrzej Siewior WARN_ON(ret < 0); 31381da177e4SLinus Torvalds } 3139