On Mon, May 24, 2021 at 07:57:48AM -0400, Brian Foster wrote:

On Sat, May 22, 2021 at 03:45:11PM +0800, Ming Lei wrote:

quoted

On Mon, May 17, 2021 at 01:17:20PM -0400, Brian Foster wrote:

quoted

The iomap ioend mechanism has the ability to construct very large,
contiguous bios and/or bio chains. This has been reported to lead to

BTW, it is actually wrong to complete a large bio chains in
iomap_finish_ioend(), which may risk in bio allocation deadlock, cause
bio_alloc_bioset() relies on bio submission to make forward progress. But
it becomes not true when all chained bios are freed just after the whole
ioend is done since all chained bios(except for the one embedded in ioend)
are allocated from same bioset(fs_bio_set).

Interesting. Do you have a reproducer (or error report) for this? Is it

No, but the theory has been applied for long time.

addressed by the next patch, or are further changes required?

Your patchset can't address the issue.


Thanks,
Ming