summaryrefslogtreecommitdiff
path: root/mm
diff options
context:
space:
mode:
authorBaokun Li <libaokun1@huawei.com>2023-12-13 14:23:24 +0800
committerAndrew Morton <akpm@linux-foundation.org>2023-12-20 13:46:19 -0800
commite2c27b803bb664748e090d99042ac128b3f88d92 (patch)
treea22732678ea3d06854dec5ef041c2a191aa9a5f6 /mm
parentb2325bf860faa2f304a7e188f00cf9f7dc9b5ee8 (diff)
mm/filemap: avoid buffered read/write race to read inconsistent data
The following concurrency may cause the data read to be inconsistent with the data on disk: cpu1 cpu2 ------------------------------|------------------------------ // Buffered write 2048 from 0 ext4_buffered_write_iter generic_perform_write copy_page_from_iter_atomic ext4_da_write_end ext4_da_do_write_end block_write_end __block_commit_write folio_mark_uptodate // Buffered read 4096 from 0 smp_wmb() ext4_file_read_iter set_bit(PG_uptodate, folio_flags) generic_file_read_iter i_size_write // 2048 filemap_read unlock_page(page) filemap_get_pages filemap_get_read_batch folio_test_uptodate(folio) ret = test_bit(PG_uptodate, folio_flags) if (ret) smp_rmb(); // Ensure that the data in page 0-2048 is up-to-date. // New buffered write 2048 from 2048 ext4_buffered_write_iter generic_perform_write copy_page_from_iter_atomic ext4_da_write_end ext4_da_do_write_end block_write_end __block_commit_write folio_mark_uptodate smp_wmb() set_bit(PG_uptodate, folio_flags) i_size_write // 4096 unlock_page(page) isize = i_size_read(inode) // 4096 // Read the latest isize 4096, but without smp_rmb(), there may be // Load-Load disorder resulting in the data in the 2048-4096 range // in the page is not up-to-date. copy_page_to_iter // copyout 4096 In the concurrency above, we read the updated i_size, but there is no read barrier to ensure that the data in the page is the same as the i_size at this point, so we may copy the unsynchronized page out. Hence adding the missing read memory barrier to fix this. This is a Load-Load reordering issue, which only occurs on some weak mem-ordering architectures (e.g. ARM64, ALPHA), but not on strong mem-ordering architectures (e.g. X86). And theoretically the problem doesn't only happen on ext4, filesystems that call filemap_read() but don't hold inode lock (e.g. btrfs, f2fs, ubifs ...) will have this problem, while filesystems with inode lock (e.g. xfs, nfs) won't have this problem. Link: https://lkml.kernel.org/r/20231213062324.739009-1-libaokun1@huawei.com Signed-off-by: Baokun Li <libaokun1@huawei.com> Reviewed-by: Jan Kara <jack@suse.cz> Cc: Andreas Dilger <adilger.kernel@dilger.ca> Cc: Christoph Hellwig <hch@infradead.org> Cc: Dave Chinner <david@fromorbit.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Ritesh Harjani (IBM) <ritesh.list@gmail.com> Cc: Theodore Ts'o <tytso@mit.edu> Cc: yangerkun <yangerkun@huawei.com> Cc: Yu Kuai <yukuai3@huawei.com> Cc: Zhang Yi <yi.zhang@huawei.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Diffstat (limited to 'mm')
-rw-r--r--mm/filemap.c9
1 files changed, 9 insertions, 0 deletions
diff --git a/mm/filemap.c b/mm/filemap.c
index f1c8c278310f..ad5b4aa049a3 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2608,6 +2608,15 @@ ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
/*
+ * Pairs with a barrier in
+ * block_write_end()->mark_buffer_dirty() or other page
+ * dirtying routines like iomap_write_end() to ensure
+ * changes to page contents are visible before we see
+ * increased inode size.
+ */
+ smp_rmb();
+
+ /*
* Once we start copying data, we don't want to be touching any
* cachelines that might be contended:
*/