In the Linux kernel, the following vulnerability has been resolved:
btrfs: always do the basic checks for btrfsqgroupinherit structure
[BUG] Syzbot reports the following regression detected by KASAN:
BUG: KASAN: slab-out-of-bounds in btrfsqgroupinherit+0x42e/0x2e20 fs/btrfs/qgroup.c:3277 Read of size 8 at addr ffff88814628ca50 by task syz-executor318/5171
CPU: 0 PID: 5171 Comm: syz-executor318 Not tainted 6.10.0-rc2-syzkaller-00010-g2ab795141095 #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 04/02/2024 Call Trace: <TASK> _dumpstack lib/dumpstack.c:88 [inline] dumpstacklvl+0x241/0x360 lib/dumpstack.c:114 printaddressdescription mm/kasan/report.c:377 [inline] printreport+0x169/0x550 mm/kasan/report.c:488 kasanreport+0x143/0x180 mm/kasan/report.c:601 btrfsqgroupinherit+0x42e/0x2e20 fs/btrfs/qgroup.c:3277 creatependingsnapshot+0x1359/0x29b0 fs/btrfs/transaction.c:1854 creatependingsnapshots+0x195/0x1d0 fs/btrfs/transaction.c:1922 btrfscommittransaction+0xf20/0x3740 fs/btrfs/transaction.c:2382 createsnapshot+0x6a1/0x9e0 fs/btrfs/ioctl.c:875 btrfsmksubvol+0x58f/0x710 fs/btrfs/ioctl.c:1029 btrfsmksnapshot+0xb5/0xf0 fs/btrfs/ioctl.c:1075 _btrfsioctlsnapcreate+0x387/0x4b0 fs/btrfs/ioctl.c:1340 btrfsioctlsnapcreatev2+0x1f2/0x3a0 fs/btrfs/ioctl.c:1422 btrfsioctl+0x99e/0xc60 vfsioctl fs/ioctl.c:51 [inline] _dosysioctl fs/ioctl.c:907 [inline] _sesysioctl+0xfc/0x170 fs/ioctl.c:893 dosyscallx64 arch/x86/entry/common.c:52 [inline] dosyscall64+0xf3/0x230 arch/x86/entry/common.c:83 entrySYSCALL64afterhwframe+0x77/0x7f RIP: 0033:0x7fcbf1992509 RSP: 002b:00007fcbf1928218 EFLAGS: 00000246 ORIGRAX: 0000000000000010 RAX: ffffffffffffffda RBX: 00007fcbf1a1f618 RCX: 00007fcbf1992509 RDX: 0000000020000280 RSI: 0000000050009417 RDI: 0000000000000003 RBP: 00007fcbf1a1f610 R08: 00007ffea1298e97 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 00007fcbf19eb660 R13: 00000000200002b8 R14: 00007fcbf19e60c0 R15: 0030656c69662f2e </TASK>
And it also pinned it down to commit b5357cb268c4 ("btrfs: qgroup: do not check qgroup inherit if qgroup is disabled").
[CAUSE] That offending commit skips the whole qgroup inherit check if qgroup is not enabled.
But that also skips the very basic checks like numrefcopies/numexclcopies and the structure size checks.
Meaning if a qgroup enable/disable race is happening at the background, and we pass a btrfsqgroupinherit structure when the qgroup is disabled, the check would be completely skipped.
Then at the time of transaction commitment, qgroup is re-enabled and btrfsqgroupinherit() is going to use the incorrect structure and causing the above KASAN error.
[FIX] Make btrfsqgroupcheckinherit() only skip the source qgroup checks. So that even if invalid btrfsqgroup_inherit structure is passed in, we can still reject invalid ones no matter if qgroup is enabled or not.
Furthermore we do already have an extra safety inside btrfsqgroupinherit(), which would just ignore invalid qgroup sources, so even if we only skip the qgroup source check we're still safe.