Re: [RFC PATCH] btrfs: scrub: don't call calc_sector_number on failed bios

[Josef Bacik <josef@toxicpanda.com>]

On Mon, Jun 10, 2024 at 04:42:29PM +0200, Johannes Thumshirn wrote:
> When running fstests' test-case btrfs/060 with MKFS_OPTIONS="-O rst"
> to force the creation of a RAID stripe tree even on regular devices I
> hit the follwoing ASSERT() in calc_sector_number():
> 
>     ASSERT(i < stripe->nr_sectors);
> 
> This ASSERT() is triggered, because we cannot find the page in the
> passed in bio_vec in the scrub_stripe's pages[] array.
> 
> When having a closer look at the stripe using drgn on the vmcore dump
> and comparing the stripe's pages to the bio_vec's pages it is evident
> that we cannot find it as first_bvec's bv_page is NULL:
> 
>     >>> stripe = t.stack_trace()[13]['stripe']
>     >>> print(stripe.pages)
>     (struct page *[16]){ 0xffffea000418b280, 0xffffea00043051c0,
>     0xffffea000430ed00, 0xffffea00040fcc00, 0xffffea000441fc80,
>     0xffffea00040fc980, 0xffffea00040fc9c0, 0xffffea00040fc940,
>     0xffffea0004223040, 0xffffea00043a1940, 0xffffea00040fea80,
>     0xffffea00040a5740, 0xffffea0004490f40, 0xffffea00040f7dc0,
>     0xffffea00044985c0, 0xffffea00040f7d80 }
>     >>> bio = t.stack_trace()[12]['bbio'].bio
>     >>> print(bio.bi_io_vec)
>     *(struct bio_vec *)0xffff88810632bc00 = {
>             .bv_page = (struct page *)0x0,
>             .bv_len = (unsigned int)0,
>             .bv_offset = (unsigned int)0,
>     }
> 
> Upon closer inspection of the bio itself we see that bio->bi_status is
> 10, which corresponds to BLK_STS_IOERR:
> 
>     >>> print(bio)
>     (struct bio){
>             .bi_next = (struct bio *)0x0,
>             .bi_bdev = (struct block_device *)0x0,
>             .bi_opf = (blk_opf_t)0,
>             .bi_flags = (unsigned short)0,
>             .bi_ioprio = (unsigned short)0,
>             .bi_write_hint = (enum rw_hint)WRITE_LIFE_NOT_SET,
>             .bi_status = (blk_status_t)10,
>             .__bi_remaining = (atomic_t){
>                     .counter = (int)1,
>             },
>             .bi_iter = (struct bvec_iter){
>             	 .bi_sector = (sector_t)2173056,
>                     .bi_size = (unsigned int)0,
>                     .bi_idx = (unsigned int)0,
>                     .bi_bvec_done = (unsigned int)0,
>             },
>             .bi_cookie = (blk_qc_t)4294967295,
>             .__bi_nr_segments = (unsigned int)4294967295,
>             .bi_end_io = (bio_end_io_t *)0x0,
>             .bi_private = (void *)0x0,
>             .bi_integrity = (struct bio_integrity_payload *)0x0,
>             .bi_vcnt = (unsigned short)0,
>             .bi_max_vecs = (unsigned short)16,
>             .__bi_cnt = (atomic_t){
>                     .counter = (int)1,
>             },
>             .bi_io_vec = (struct bio_vec *)0xffff88810632bc00,
>             .bi_pool = (struct bio_set *)btrfs_bioset+0x0 = 0xffffffff82c85620,
>             .bi_inline_vecs = (struct bio_vec []){},
>     }
> 
> So only call calc_sector_number when we know the bio completed without error.
> 
> Cc: Qu Wenru <wqu@suse.com>
> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>

Reviewed-by: Josef Bacik <josef@toxicpanda.com>

Thanks,

Josef