Hello Christoph, Am 02/25/20 um 20:12 schrieb Christoph Hellwig: > On Tue, Feb 25, 2020 at 06:07:44PM +0100, Daniel Glöckner wrote: >> In cases where dec_in_flight has to requeue the integrity_bio_wait work >> to transfer the rest of the data, the __bi_remaining field of the bio >> might already have been decremented to zero. Reusing the bio without >> reinitializing that counter to 1 can then result in integrity_end_io >> being called too early when the BIO_CHAIN flag is set, f.ex. due to >> blk_queue_split. In our case this triggered the BUG() in >> blk_mq_end_request when the hardware signalled completion of the bio >> after integrity_end_io had modified it. >> >> Signed-off-by: Daniel Glöckner <dg@xxxxxxxxx> > > Drivers have no business poking into these internals. If a bio is > reused the caller needs to use bio_reset instead. bio_reset will reset too many fields. As you can see in the context of the diff, dm-integrity expects f.ex. the values modified by bio_advance to stay intact and the transfer should of course use the same disk and operation. How about doing the atomic_set in bio_remaining_done (in block/bio.c) where the BIO_CHAIN flag is cleared once __bi_remaining hits zero? Or is requeuing a bio without bio_reset really a no-go? In that case a one-liner won't do... Best regards, Daniel -- Besuchen Sie uns auf der Embedded World 2020 in Nürnberg! -> Halle 4, Stand 368 Dipl.-Math. Daniel Glöckner, emlix GmbH, http://www.emlix.com Fon +49 551 30664-0, Fax +49 551 30664-11, Gothaer Platz 3, 37083 Göttingen, Germany Sitz der Gesellschaft: Göttingen, Amtsgericht Göttingen HR B 3160 Geschäftsführung: Heike Jordan, Dr. Uwe Kracke Ust-IdNr.: DE 205 198 055 emlix - your embedded linux partner
