在 2023/3/22 15:05, Mariusz Tkaczyk 写道: > On Wed, 22 Mar 2023 10:24:41 +0800 > Wu Guanghao <wuguanghao3@xxxxxxxxxx> wrote: > >> 在 2023/3/21 18:18, Mariusz Tkaczyk 写道: >>> On Tue, 21 Mar 2023 16:56:37 +0800 >>> Wu Guanghao <wuguanghao3@xxxxxxxxxx> wrote: >>> >>>> The latest kernel version will not report an error through mdadm >>>> set_disk_faulty. >>>> >>>> $ lsblk >>>> sdb 8:16 0 10G 0 disk >>>> └─md0 9:0 0 19.9G 0 raid0 >>>> sdc 8:32 0 10G 0 disk >>>> └─md0 9:0 0 19.9G 0 raid0 >>>> >>>> old kernel: >>>> ... >>>> $ mdadm /dev/md0 -f /dev/sdb >>>> mdadm: set device faulty failed for /dev/sdb: Device or resource busy >>>> ... >>>> >>>> latest kernel: >>>> ... >>>> $ mdadm /dev/md0 -f /dev/sdb >>>> mdadm: set /dev/sdb faulty in /dev/md0 >>>> ... >>>> >>>> The old kernel judges whether the Faulty flag is set in rdev->flags, >>>> and returns -EBUSY if not. And The latest kernel only return -EBUSY >>>> if the MD_BROKEN flag is set in mddev->flags. raid0 doesn't set >>>> error_handler, so MD_BROKEN will not be set, it will return 0. >>>> >>>> So if error_handler isn't set for a raid type, also return -EBUSY. >>> Hi, >>> Please test with: >>> https://lore.kernel.org/linux-raid/20230306130317.3418-1-mariusz.tkaczyk@xxxxxxxxxxxxxxx/ >>> >>> Thanks, >>> Mariusz >>> >> >> Hi, Mariusz >> >> Are there other patches? There are other problems with this patch. >> https://lore.kernel.org/linux-raid/20230306130317.3418-1-mariusz.tkaczyk@xxxxxxxxxxxxxxx/ >> >> md_submit_bio() >> ... >> // raid0 set disk faulty failed, but MD_BROKEN flag is set， >> // write IO will fail. >> if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) { >> bio_io_error(bio); >> return; >> } >> ... >> >> old kernel: >> ... >> $ mdadm /dev/md0 -f /dev/sdb >> mdadm: set device faulty failed for /dev/sdb: Device or resource busy >> >> $ mkfs.xfs /dev/md0 >> log stripe unit (524288 bytes) is too large (maximum is 256KiB) >> log stripe unit adjusted to 32KiB >> meta-data=/dev/md0 isize=512 agcount=16, agsize=1800064 blks >> = sectsz=512 attr=2, projid32bit=1 >> = crc=1 finobt=1, sparse=1, rmapbt=0 >> = reflink=1 bigtime=0 inobtcount=0 >> data = bsize=4096 blocks=28801024, imaxpct=25 >> = sunit=128 swidth=256 blks >> naming =version 2 bsize=4096 ascii-ci=0, ftype=1 >> log =internal log bsize=4096 blocks=14064, version=2 >> = sectsz=512 sunit=8 blks, lazy-count=1 >> realtime =none extsz=4096 blocks=0, rtextents=0 >> Discarding blocks...Done. >> ... >> >> >> merged patch kernel: >> ... >> # mdadm /dev/md0 -f /dev/sdb >> mdadm: set device faulty failed for /dev/sdb: Device or resource busy >> >> mkfs.xfs /dev/md0 >> log stripe unit (524288 bytes) is too large (maximum is 256KiB) >> log stripe unit adjusted to 32KiB >> meta-data=/dev/md0 isize=512 agcount=8, agsize=65408 blks >> = sectsz=512 attr=2, projid32bit=1 >> = crc=1 finobt=1, sparse=1, rmapbt=0 >> = reflink=1 bigtime=0 inobtcount=0 >> data = bsize=4096 blocks=523264, imaxpct=25 >> = sunit=128 swidth=256 blks >> naming =version 2 bsize=4096 ascii-ci=0, ftype=1 >> log =internal log bsize=4096 blocks=2560, version=2 >> = sectsz=512 sunit=8 blks, lazy-count=1 >> realtime =none extsz=4096 blocks=0, rtextents=0 >> mkfs.xfs: pwrite failed: Input/output error >> ... >> >> > Hi Wu, > Beside the kernel, there are also patches in mdadm. Please check if you have > them all. > https://git.kernel.org/pub/scm/utils/mdadm/mdadm.git/commit/?id=b3e7b7eb1dfedd7cbd9a3800e884941f67d94c96 > https://git.kernel.org/pub/scm/utils/mdadm/mdadm.git/commit/?id=461fae7e7809670d286cc19aac5bfa861c29f93a > https://git.kernel.org/pub/scm/utils/mdadm/mdadm.git/commit/?id=fc6fd4063769f4194c3fb8f77b32b2819e140fb9 > Hi, Mariusz Thanks for your reply, I would test with the above patches. > Some background: > --faulty (-f) is intended to be used by administrators. We cannot rely on > kernel answer because if mdadm will try to set device faulty, it results in > MD_BROKEN and every new IO will be failed (and that is intended change). > > Simply, mdadm must check first if it can remove the drive and that was added by > the mentioned patches. The first patch (the last one) added verification but > brings regression, the next two patches are fixes for omitted scenarios. > > Thanks, > Mariusz > . >
