https://bugzilla.kernel.org/show_bug.cgi?id=203947 --- Comment #8 from Darrick J. Wong (djwong+kernel@xxxxxxxxxx) --- Ok, so I reproduced it locally and tracked the crash to this part of xfs_bmapi_read() where we dereference *ifp: if (!(ifp->if_flags & XFS_IFEXTENTS)) { error = xfs_iread_extents(NULL, ip, whichfork); if (error) return error; } Looking at xfs_iformat_fork(), it seems that if there's any kind of error formatting the attr fork it'll free ip->i_afp and set it to NULL, so I think the fix is to add an "if (!afp) return -EIO;" somewhere. Not sure how we actually get to this place, though. fsstress is running bulkstat, which is inactivating an inode with i_nlink == 0 and a corrupt attr fork that won't load. Maybe we hit an inode that had previously gone through unlinked processing after log recovery but was lurking on the mru waiting to be inactivated, but then bulkstat showed up (with its IGET_DONTCACHE) which forced immediate inactivation? -- You are receiving this mail because: You are watching the assignee of the bug.
