[SRU][K/J/I][PATCH 0/1] prevent kernel panic with overlayfs + shiftfs

[Andrea Righi]

BugLink: https://bugs.launchpad.net/bugs/1973620

[Impact]

The patch that we have recently re-introduced to properly support
overlayfs on top of shiftfs can introduce potential kernel panics, for
example:

    BUG: kernel NULL pointer dereference, address: 0000000000000008
    [ 447.039738] #PF: supervisor read access in kernel mode
    [ 447.040369] #PF: error_code(0x0000) - not-present page
    [ 447.041002] PGD 0 P4D 0
    [ 447.041325] Oops: 0000 [#1] SMP NOPTI
    [ 447.041798] CPU: 0 PID: 73766 Comm: sudo Not tainted 5.15.0-28-generic #29~20.04.1-Ubuntu
    [ 447.042800] Hardware name: OpenStack Foundation OpenStack Nova, BIOS Ubuntu-1.8.2-1ubuntu1+esm1 04/01/2014
    [ 447.043979] RIP: 0010:aa_file_perm+0x3a/0x470
    [ 447.044565] Code: 54 53 48 83 ec 68 48 89 7d 80 89 4d 8c 65 48 8b 04 25 28 00 00 00 48 89 45 d0 31 c0 48 63 05 01 0a 19 01 48 03 82 c0 00 00 00 <4c> 8b 68 08 f6 46 40 02 0f 85 d0 00 00 00 41 f6 45 40 02 0f 85 c5
    [ 447.046837] RSP: 0018:ffffaefe80a4bca8 EFLAGS: 00010246
    [ 447.047481] RAX: 0000000000000000 RBX: ffff96e4038abd01 RCX: 0000000000000004
    [ 447.048351] RDX: ffff96e4038abd00 RSI: ffff96e401215eb8 RDI: ffffffff9c22a2ac
    [ 447.049241] RBP: ffffaefe80a4bd38 R08: 0000000000000000 R09: 0000000000000000
    [ 447.050121] R10: 0000000000000000 R11: 0000000000000000 R12: ffff96e401215eb8
    [ 447.051040] R13: ffff96e4038abd00 R14: ffffffff9c22a2ac R15: 0000000000000004
    [ 447.051942] FS: 00007eff3c0f8c80(0000) GS:ffff96e45e400000(0000) knlGS:0000000000000000
    [ 447.052981] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
    [ 447.053696] CR2: 0000000000000008 CR3: 0000000002be2000 CR4: 00000000003506f0
    [ 447.054571] Call Trace:
    [ 447.054883] <TASK>
    [ 447.055154] ? unlock_page_memcg+0x2f/0x40
    [ 447.055668] ? page_remove_rmap+0x4b/0x320
    [ 447.056180] common_file_perm+0x72/0x170
    [ 447.056669] apparmor_file_permission+0x1c/0x20
    [ 447.057237] security_file_permission+0x30/0x1a0
    [ 447.057898] rw_verify_area+0x35/0x60
    [ 447.058392] vfs_read+0x6d/0x1a0
    [ 447.058842] ksys_read+0xb1/0xe0
    [ 447.059276] __x64_sys_read+0x1a/0x20
    [ 447.059732] do_syscall_64+0x5c/0xc0
    [ 447.060183] ? __set_current_blocked+0x3b/0x60
    [ 447.060738] ? exit_to_user_mode_prepare+0x3d/0x1c0
    [ 447.061434] ? syscall_exit_to_user_mode+0x27/0x50
    [ 447.062099] ? do_syscall_64+0x69/0xc0
    [ 447.062603] ? irqentry_exit_to_user_mode+0x9/0x20
    [ 447.063210] ? irqentry_exit+0x19/0x30
    [ 447.063678] ? exc_page_fault+0x89/0x160
    [ 447.064165] ? asm_exc_page_fault+0x8/0x30
    [ 447.064675] entry_SYSCALL_64_after_hwframe+0x44/0xae
    [ 447.065298] RIP: 0033:0x7eff3c2cb002

[Test case]

It is really easy to trigger this specific kernel panic running the lxc
autopackage test.

[Fix]

This bug happens because we don't need to decrement anymore the refcount
for the previous vm_file value in ovl_vm_prfile_set(). So the fix simply
consists of removing the unnecessary fput().

[Regression potential]

This patch affects only overlayfs (only when AUFS is enabled), so we may
see regressions in overlayfs in kernels that have AUFS enabled (focal
hwe and cloud kernels).