On Thu, Aug 29, 2024 at 07:24:55PM +0800, Zizhi Wo wrote: > friendly ping Sorry, I'm not going to get to this until late September. --D > 在 2024/8/26 11:10, Zizhi Wo 写道: > > Prior to this, I had already sent out a patchset related to xfs fsmap > > bugfix, which mainly introduced "info->end_daddr" to fix omitted extents[1] > > and Darrick had already sent out a patchbomb for merging into stable[2], > > which included my previous patches. > > > > However, I recently discovered two new fsmap problems...What follows is a > > brief description of them: > > > > Patch 1: In this scenario, fsmap lost one block count. The root cause is > > that during the calculation of highkey, the calculation of start_block is > > missing an increment by one, which leads to the last query missing one > > This problem is resolved by adding a sentinel node. > > > > Patch 2: In this scenario, the fsmap query for realtime deivce may display > > extra intervals. This is due to an extra increase in "end_rtb". The issue > > is resolved by adjusting the relevant calculations. And this patch depends > > on the previous patch that introduced "info->end_daddr". > > > > [1] https://lore.kernel.org/all/20240819005320.304211-1-wozizhi@xxxxxxxxxx/ > > [2] https://lore.kernel.org/all/172437083728.56860.10056307551249098606.stgit@frogsfrogsfrogs/ > > > > Zizhi Wo (2): > > xfs: Fix missing block calculations in xfs datadev fsmap > > xfs: Fix incorrect parameter calculation in rt fsmap > > > > fs/xfs/libxfs/xfs_rtbitmap.c | 4 +--- > > fs/xfs/xfs_fsmap.c | 39 +++++++++++++++++++++++++++++++----- > > 2 files changed, 35 insertions(+), 8 deletions(-) > > >
