On Thu, Feb 21, 2019 at 10:54:53AM +0800, Su Hua wrote:
> Hi Carlos
> I am using the kernel version 4.4.163. There is a problem that umount
> is in the D state after the disk is unplugged. This problem was solved
> after version 4.7. Now that we have not upgraded the kernel version, I
> want to try backport on this 4.4.163 first. I added these two patches
> separately. But there are still problems, can you still recall where i
> need to change?
Hello Su. My apologies for the late response.
> patch：
> [1]e6b3bb78962e6 ("xfs: add "fail at unmount" error handling
> configuration")
This alone can't do much without the fixes in buffer writeback code.
> [2]9356413 Release buffer locks in case of IO error
This patch specifically was a 'hack' to prove a point where the problem was, but
it never was the proper fix.
I believe the patches you are looking for are (from Linus's tree):
373b0589dc8d
842f6e9f7862
d3a304b62921
0b80ae6ed131
Cheers
> This is the information I used to communicate with Brian.
> [3]https://www.spinics.net/lists/linux-xfs/msg24740.html
> problem：
> ...
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> Feb 20 18:07:56 node-1 kernel: XFS (sdc): xfs_imap_to_bp:
> xfs_trans_read_buf() returned error -5.
> ...
> This is where I modified according to the second patch.
> 440 /*
> 441 * Someone else is already flushing the inode. Nothing we can
> do
> 442 * here but wait for the flush to finish and remove the item
> from
> 443 * the AIL.
> 444 */
> 445 if (!xfs_iflock_nowait(ip)) {
> 446
> 447 if (lip->li_mountp->m_flags & XFS_MOUNT_UNMOUNTING) {
> 448 int error;
> 449 struct xfs_dinode *dip;
> 450
> 451 error = xfs_imap_to_bp(ip->i_mount, NULL,
> &ip->i_imap, &dip,
> 452 &bp, XBF_TRYLOCK, 0);
> 453 if (error) {
> 454 rval = XFS_ITEM_FLUSHING;
> 455 goto out_unlock;
> 456 }
> 457
> 458 if (!(bp->b_flags & XBF_WRITE_FAIL)) {
> 459 rval = XFS_ITEM_FLUSHING;
> 460 xfs_buf_relse(bp);
> 461 goto out_unlock;
> 462 }
> 463
> 464 if (!xfs_buf_delwri_queue(bp, buffer_list))
> 465 rval = XFS_ITEM_FLUSHING;
> 466
> 467 xfs_buf_relse(bp);
> 468 } else
> 469 rval = XFS_ITEM_FLUSHING;
> 470 goto out_unlock;
> 471 }
> regards
> su
>
> References
>
> 1. http://kernel.suse.com/cgit/kernel/commit/?h=stable&id=e6b3bb78962e65c4ad125598755cfbf2a8779e86
> 2. https://patchwork.kernel.org/patch/9356413/
> 3. https://www.spinics.net/lists/linux-xfs/msg24740.html
--
Carlos
