Samuel Thibault, le ven. 13 mars 2026 08:41:11 +0100, a ecrit: > Samuel Thibault, le dim. 08 mars 2026 11:06:22 +0100, a ecrit: > > Michael Kelly, le dim. 08 mars 2026 06:04:00 +0000, a ecrit: > > > Hi Samuel, > > > > > > On 07/03/2026 21:39, Samuel Thibault wrote: > > > > Samuel Thibault, le sam. 07 mars 2026 21:06:38 +0100, a ecrit: > > > > > Samuel Thibault, le lun. 02 mars 2026 01:24:02 +0100, a ecrit: > > > > > > I have gotten yet another odd file corruption in a buildd chroot, > > > > > > after > > > > > > building haskell-xml-html-qq_0.1.0.1-6+b1, /usr/bin/mkdir became: > > > > > > > > > > > > 00000000 00 00 02 ea 01 00 00 00 01 00 00 00 93 2a e0 d7 > > > > > > |.............*..| > > > > > > 00000010 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > > > > > |................| > > > > > Just realizing... These are little-endian values, and 0xea020000 > > > > > is... EXT2_XATTR_BLOCK_MAGIC ! > > > > > > > > > > That's very informative! We must have block allocation issues in our > > > > > xattr code. > > > > I guess the problem is when we happen to drop an xattr before the > > > > content got flushed to disk, I have a patch that I'll try to check. > > > > > > Would this also explain the corruption to the shared library > > > in libHShedgehog-classes where only 16 byte sections of pages were > > > replaced > > > to all zero at offsets 0x50, 0x70 and 0xff0 without the preceding > > > EXT2_XATTR_BLOCK_MAGIC? > > > > No, the case I'm after writes a whole page. > > I'm however seeing the same kind of issue in the haskell packages > builds, making the symbol table corrupted or the assembly source code > corrupted, as mentioned earlier in the thread. > > Since I only notice this in haskell, that's possibly related to signal > delivery.
(and I only notice it on amd64) Samuel
