| Message ID | 20240815092141.1223238-1-chizhiling@163.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | ocfs2: fix unexpected zeroing of virtual disk | expand |
On 8/15/24 17:21, Chi Zhiling wrote: > From: Chi Zhiling <chizhiling@kylinos.cn> > > In a guest virtual machine, we found that there is unexpected data > zeroing problem detected occassionly: > > XFS (vdb): Mounting V5 Filesystem > XFS (vdb): Ending clean mount > XFS (vdb): Metadata CRC error detected at xfs_refcountbt_read_verify+0x2c/0xf0, xfs_refcountbt block 0x200028 > XFS (vdb): Unmount and run xfs_repair > XFS (vdb): First 128 bytes of corrupted metadata buffer: > 00000000e0cd2f5e: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ > 00000000cafd57f5: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ > 00000000d0298d7d: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ > 00000000f0698484: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ > 00000000adb789a7: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ > 000000005292b878: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ > 00000000885b4700: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ > 00000000fd4b4df7: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ > XFS (vdb): metadata I/O error in "xfs_trans_read_buf_map" at daddr 0x200028 len 8 error 74 > XFS (vdb): Error -117 recovering leftover CoW allocations. > XFS (vdb): xfs_do_force_shutdown(0x8) called from line 994 of file fs/xfs/xfs_mount.c. Return address = 000000003a53523a > XFS (vdb): Corruption of in-memory data detected. Shutting down filesystem > XFS (vdb): Please umount the filesystem and rectify the problem(s) > > It turns out that the root cause is from the physical host machine. > More specifically, it is caused by the ocfs2. > > when the page_size is 64k, the block should advance by 16 each time > instead of 1. > This will lead to a wrong mapping from the page to the disk, which > will zero some adjacent part of the disk. > > Suggested-by: Shida Zhang <zhangshida@kylinos.cn> > Signed-off-by: Chi Zhiling <chizhiling@kylinos.cn> > --- > fs/ocfs2/aops.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c > index d6c985cc6353..1fea43c33b6b 100644 > --- a/fs/ocfs2/aops.c > +++ b/fs/ocfs2/aops.c > @@ -1187,7 +1187,7 @@ static int ocfs2_write_cluster(struct address_space *mapping, > > /* This is the direct io target page. */ > if (wc->w_pages[i] == NULL) { > - p_blkno++; > + p_blkno += (1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits)); > continue; > } > Looks good to me. Signed-off-by: Heming Zhao <heming.zhao@suse.com>
Looks good. Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com> BTW, ocfs2 hasn't been tested thoroughly under 64k page, so I'm afraid there are other bugs when running ocfs2 under 64k page. On 8/18/24 6:31 PM, Heming Zhao wrote: > On 8/15/24 17:21, Chi Zhiling wrote: >> From: Chi Zhiling <chizhiling@kylinos.cn> >> >> In a guest virtual machine, we found that there is unexpected data >> zeroing problem detected occassionly: >> >> XFS (vdb): Mounting V5 Filesystem >> XFS (vdb): Ending clean mount >> XFS (vdb): Metadata CRC error detected at xfs_refcountbt_read_verify+0x2c/0xf0, xfs_refcountbt block 0x200028 >> XFS (vdb): Unmount and run xfs_repair >> XFS (vdb): First 128 bytes of corrupted metadata buffer: >> 00000000e0cd2f5e: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >> 00000000cafd57f5: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >> 00000000d0298d7d: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >> 00000000f0698484: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >> 00000000adb789a7: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >> 000000005292b878: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >> 00000000885b4700: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >> 00000000fd4b4df7: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >> XFS (vdb): metadata I/O error in "xfs_trans_read_buf_map" at daddr 0x200028 len 8 error 74 >> XFS (vdb): Error -117 recovering leftover CoW allocations. >> XFS (vdb): xfs_do_force_shutdown(0x8) called from line 994 of file fs/xfs/xfs_mount.c. Return address = 000000003a53523a >> XFS (vdb): Corruption of in-memory data detected. Shutting down filesystem >> XFS (vdb): Please umount the filesystem and rectify the problem(s) >> >> It turns out that the root cause is from the physical host machine. >> More specifically, it is caused by the ocfs2. >> >> when the page_size is 64k, the block should advance by 16 each time >> instead of 1. >> This will lead to a wrong mapping from the page to the disk, which >> will zero some adjacent part of the disk. >> >> Suggested-by: Shida Zhang <zhangshida@kylinos.cn> >> Signed-off-by: Chi Zhiling <chizhiling@kylinos.cn> >> --- >> fs/ocfs2/aops.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c >> index d6c985cc6353..1fea43c33b6b 100644 >> --- a/fs/ocfs2/aops.c >> +++ b/fs/ocfs2/aops.c >> @@ -1187,7 +1187,7 @@ static int ocfs2_write_cluster(struct address_space *mapping, >> /* This is the direct io target page. */ >> if (wc->w_pages[i] == NULL) { >> - p_blkno++; >> + p_blkno += (1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits)); >> continue; >> } >> > > Looks good to me. > Signed-off-by: Heming Zhao <heming.zhao@suse.com>
Sorry, I just realized I posted the wrong tag "Signed-off-by: Heming Zhao ...". The correct should be Reviewed-by: Heming Zhao <heming.zhao@suse.com>. On 8/19/24 10:32, Joseph Qi wrote: > Looks good. > Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com> > > BTW, ocfs2 hasn't been tested thoroughly under 64k page, so I'm afraid > there are other bugs when running ocfs2 under 64k page. > > On 8/18/24 6:31 PM, Heming Zhao wrote: >> On 8/15/24 17:21, Chi Zhiling wrote: >>> From: Chi Zhiling <chizhiling@kylinos.cn> >>> >>> In a guest virtual machine, we found that there is unexpected data >>> zeroing problem detected occassionly: >>> >>> XFS (vdb): Mounting V5 Filesystem >>> XFS (vdb): Ending clean mount >>> XFS (vdb): Metadata CRC error detected at xfs_refcountbt_read_verify+0x2c/0xf0, xfs_refcountbt block 0x200028 >>> XFS (vdb): Unmount and run xfs_repair >>> XFS (vdb): First 128 bytes of corrupted metadata buffer: >>> 00000000e0cd2f5e: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >>> 00000000cafd57f5: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >>> 00000000d0298d7d: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >>> 00000000f0698484: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >>> 00000000adb789a7: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >>> 000000005292b878: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >>> 00000000885b4700: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >>> 00000000fd4b4df7: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >>> XFS (vdb): metadata I/O error in "xfs_trans_read_buf_map" at daddr 0x200028 len 8 error 74 >>> XFS (vdb): Error -117 recovering leftover CoW allocations. >>> XFS (vdb): xfs_do_force_shutdown(0x8) called from line 994 of file fs/xfs/xfs_mount.c. Return address = 000000003a53523a >>> XFS (vdb): Corruption of in-memory data detected. Shutting down filesystem >>> XFS (vdb): Please umount the filesystem and rectify the problem(s) >>> >>> It turns out that the root cause is from the physical host machine. >>> More specifically, it is caused by the ocfs2. >>> >>> when the page_size is 64k, the block should advance by 16 each time >>> instead of 1. >>> This will lead to a wrong mapping from the page to the disk, which >>> will zero some adjacent part of the disk. >>> >>> Suggested-by: Shida Zhang <zhangshida@kylinos.cn> >>> Signed-off-by: Chi Zhiling <chizhiling@kylinos.cn> >>> --- >>> fs/ocfs2/aops.c | 2 +- >>> 1 file changed, 1 insertion(+), 1 deletion(-) >>> >>> diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c >>> index d6c985cc6353..1fea43c33b6b 100644 >>> --- a/fs/ocfs2/aops.c >>> +++ b/fs/ocfs2/aops.c >>> @@ -1187,7 +1187,7 @@ static int ocfs2_write_cluster(struct address_space *mapping, >>> /* This is the direct io target page. */ >>> if (wc->w_pages[i] == NULL) { >>> - p_blkno++; >>> + p_blkno += (1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits)); >>> continue; >>> } >>> >> >> Looks good to me. >> Signed-off-by: Heming Zhao <heming.zhao@suse.com>
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c index d6c985cc6353..1fea43c33b6b 100644 --- a/fs/ocfs2/aops.c +++ b/fs/ocfs2/aops.c @@ -1187,7 +1187,7 @@ static int ocfs2_write_cluster(struct address_space *mapping, /* This is the direct io target page. */ if (wc->w_pages[i] == NULL) { - p_blkno++; + p_blkno += (1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits)); continue; }