On Mon, Sep 18, 2023 at 08:13:30PM +0530, Shreeya Patel wrote: > Hi Everyone, > > syzbot has reported a task hung issue in ext4_fallocate, crash report can be > seen at the bottom of the email. What's the link to the syzbot dashboard URL? This is typically the first line of the reproducer, but it's been snipped out in your reproducer. > > When I tried to reproduce this issue on mainline linux kernel using the > reproducer provided by syzbot, I see an endless loop of following errors :- > > [ 89.689751][ T3922] loop1: detected capacity change from 0 to 512 > [ 89.690514][ T3916] EXT4-fs error (device loop4): ext4_map_blocks:577: > inode #3: block 9: comm poc: lblock 0 mapped to illegal pblock 9 (length 1) > [ 89.694709][ T3890] EXT4-fs error (device loop0): ext4_map_blocks:577: Please note that maliciously corrupted file system is considered low priority by ext4 developers. If this is something which is important to Google, then it needs to fund more headcount so that we have time to take a look at these things. There has been plenty of discussions about how syzbot is effectively a denial of service attack on upstream resources, and the only way we can respond is to down-prioritize such bug reports. This is *especially* true when we receive reports from private syzbot instances where we don't have any of the features provided by syzbot console --- including convenient access to the file system image that was mounted as part of the test, the ability to use #syz test to try out patches --- and more importantly, so we can introduce debugging messages and using the syzbot testing facility to run the tests. If you really are concerned about the threat model of users picking up random USB thumb drives that they found in the parking lot, consider running fsck.ext4 -f on the file system first, and if the file system checker is not able to fix the file system, refuse to mount it. Alternatively, consider using cros-vm, and mounting the file system in the guest-kernel so that the VMM provides an additional sandbox layer. Anyway, please provide a link to a public Syzkaller dashboard report, and we'll take a look at it when we have time... Cheers, - Ted
