Hello, On Fri 11-10-24 12:08:58, Baolin Liu wrote: > Greetings， > > This problem is reproduced by our customer using their own testing tool > “run_bug”. When I consulted with a client, the testing tool “run_bug” > used a variety of background programs to benchmark (including memory > pressure, cpu pressure, file cycle manipulation, fsstress Stress testing > tool, postmark program，and so on). > > The recurrence probability is relatively low. OK, thanks for asking! > In response to your query, in ext4_block_write_begin, the new state will > be clear before get block, and the bh that failed get_block will not be > set to new. However, when the page size is greater than the block size, a > page will contain multiple bh. True. I wanted to argue that the buffer_new bit should be either cleared in ext4_block_write_begin() (in case of error) or in ext4_journalled_write_end() (in case of success) but actually ext4_journalled_write_end() misses the clearing. So I think the better solution is like the attached patch. I'll submit it once testing finishes but it would be great if you could test that it fixes your problems as well. Thanks! Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
