Re: [PATCH v2 12/16] mm: allow handling of stacked mmap_prepare hooks in more drivers

[Joshua Hahn <joshua.hahnjy@gmail.com>]

On Mon, 16 Mar 2026 21:12:08 +0000 "Lorenzo Stoakes (Oracle)" <ljs@kernel.org> wrote:

> While the conversion of mmap hooks to mmap_prepare is underway, we wil
> encounter situations where mmap hooks need to invoke nested mmap_prepare
> hooks.
> 
> The nesting of mmap hooks is termed 'stacking'.  In order to flexibly
> facilitate the conversion of custom mmap hooks in drivers which stack, we
> must split up the existing compat_vma_mapped() function into two separate
> functions:
> 
> * compat_set_desc_from_vma() - This allows the setting of a vm_area_desc
>   object's fields to the relevant fields of a VMA.

Hello Lorenzo, I hope you are doing well!

Thank you for this patch. I was developing on top of mm-new today and had
an error that I think was caused by this patch. I want to preface this by
saying that I am not at all familiar with this area of the code, so please
do forgive me if I've misinterpreted the crash and mistakenly pointed
at this commit : -)

Here is the crash:

[    1.083795] kernel tried to execute NX-protected page - exploit attempt? (uid: 0)
[    1.083883] BUG: unable to handle page fault for address: ffa00000048efbb8
[    1.083957] #PF: supervisor instruction fetch in kernel mode
[    1.084030] #PF: error_code(0x0011) - permissions violation
[    1.084086] PGD 100000067 P4D 10035f067 PUD 100364067 PMD 441ed9067 PTE 80000004466a3163
[    1.084162] Oops: Oops: 0011 [#1] SMP
[    1.084218] CPU: 0 UID: 0 PID: 305 Comm: mkdir Tainted: G        W   E       7.0.0-rc4-virtme-00442-ge53de5a0302f-dirty #85 PREEMPTLAZY

As you can see, it's on a QEMU instance. I don't think this makes a difference
in the crash, though.

[    1.084321] Tainted: [W]=WARN, [E]=UNSIGNED_MODULE
[    1.084369] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-5.el9 11/05/2023
[    1.084450] RIP: 0010:0xffa00000048efbb8
[    1.084489] Code: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 <40> 12 0e 00 01 00 11 ff d0 fa 8e 04 00 00 a0 ff 80 33 51 02 01 00
[    1.084642] RSP: 0018:ffa00000048ef998 EFLAGS: 00010286
[    1.084692] RAX: ffa00000048efbb8 RBX: ff11000102512cc0 RCX: 000000000000000d
[    1.084766] RDX: ffffffffa06247d0 RSI: ffa00000048efa18 RDI: ff11000102512cc0
[    1.084826] RBP: ffa00000048ef9c8 R08: 0000000000000000 R09: 0000000000000007
[    1.084889] R10: ff110001047d1f08 R11: 00007effdc3d0fff R12: ff110001047d3b00
[    1.084954] R13: ff11000446cae600 R14: ff110001024efe00 R15: ff11000102510a80
[    1.085021] FS:  0000000000000000(0000) GS:ff110004aae72000(0000) knlGS:0000000000000000
[    1.085083] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    1.085136] CR2: ffa00000048efbb8 CR3: 0000000102667001 CR4: 0000000000771ef0
[    1.085201] PKRU: 55555554
[    1.085228] Call Trace:
[    1.085248]  <TASK>
[    1.085274]  ? __compat_vma_mmap+0x8e/0x130
[    1.085318]  ? compat_vma_mmap+0x76/0x80
[    1.085354]  ? mas_alloc_nodes+0xb2/0x110
[    1.085390]  ? backing_file_mmap+0xc3/0xf0
[    1.085426]  ? ovl_mmap+0x41/0x50
[    1.085463]  ? ovl_mmap+0x50/0x50
[    1.085499]  ? __mmap_region+0x7e8/0x1100
[    1.085539]  ? do_mmap+0x49f/0x5e0
[    1.085573]  ? vm_mmap_pgoff+0xef/0x1e0
[    1.085609]  ? ksys_mmap_pgoff+0x15c/0x1f0
[    1.085647]  ? do_syscall_64+0xab/0x980
[    1.085684]  ? entry_SYSCALL_64_after_hwframe+0x4b/0x53
[    1.085730]  </TASK>
[    1.085770] Modules linked in: virtio_mmio(E) 9pnet_virtio(E) 9p(E) 9pnet(E) netfs(E)
[    1.085838] CR2: ffa00000048efbb8
[    1.085874] ---[ end trace 0000000000000000 ]---
[    1.085875] kernel tried to execute NX-protected page - exploit attempt? (uid: 0)
[    1.085918] RIP: 0010:0xffa00000048efbb8
[    1.085921] Code: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 <40> 12 0e 00 01 00 11 ff d0 fa 8e 04 00 00 a0 ff 80 33 51 02 01 00
[    1.085988] BUG: unable to handle page fault for address: ffa00000048f7bb8
[    1.086026] RSP: 0018:ffa00000048ef998 EFLAGS: 00010286
[    1.086166] #PF: supervisor instruction fetch in kernel mode
[    1.086221]
[    1.086267] #PF: error_code(0x0011) - permissions violation
[    1.086321] RAX: ffa00000048efbb8 RBX: ff11000102512cc0 RCX: 000000000000000d
[    1.086348] PGD 100000067
[    1.086394] RDX: ffffffffa06247d0 RSI: ffa00000048efa18 RDI: ff11000102512cc0
[    1.086459] P4D 10035f067
[    1.086486] RBP: ffa00000048ef9c8 R08: 0000000000000000 R09: 0000000000000007
[    1.086550] PUD 100364067
[    1.086577] R10: ff110001047d1f08 R11: 00007effdc3d0fff R12: ff110001047d3b00
[    1.086641] PMD 441ed9067
[    1.086668] R13: ff11000446cae600 R14: ff110001024efe00 R15: ff11000102510a80
[    1.086731] PTE 80000004433d3163
[    1.086764] FS:  0000000000000000(0000) GS:ff110004aae72000(0000) knlGS:0000000000000000
[    1.086829]
[    1.086868] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    1.086931] Oops: Oops: 0011 [#2] SMP
[    1.086958] CR2: ffa00000048efbb8 CR3: 0000000102667001 CR4: 0000000000771ef0
[    1.087015] CPU: 29 UID: 0 PID: 306 Comm: mount Tainted: G      D W   E       7.0.0-rc4-virtme-00442-ge53de5a0302f-dirty #85 PREEMPTLAZY
[    1.087050] PKRU: 55555554
[    1.087115] Tainted: [D]=DIE, [W]=WARN, [E]=UNSIGNED_MODULE
[    1.087207] Kernel panic - not syncing: Fatal exception
[    2.158392] Shutting down cpus with NMI
[    2.158629] Kernel Offset: disabled
[    2.158668] ---[ end Kernel panic - not syncing: Fatal exception ]---

It crashes at compat_vma_mmap, and here is what I think could be the 
potential crash path:

- compat_vma_mmap() creates struct vm_area_desc desc;
  - compat_set_desc_from_vma Doesn't initialize the struct, but instead
    modifies independent fields. I think this is where the behavior
    diverges, since before we would use the C initializer and uninitialized
    variables would be set to 0 (including ommitted ones, like
    action.success_hook or action.error_hook). But action.type = MMAP_NOTHING
  - desc.action.success_hook remains uninitialized in vfs_mmap_prepare
  - mmap_action_complete()
    - Here, We've set action.type to be MMAP_NOTHING, so we have err = 0
    - mmap_action_finish(action, vma, 0)
      - And here, since err == 0, we check action->success_hook (which has
        garbage, therefore it's nonzero) and call action->success_hook(vma)

And I think action->success_hook(vma) where success_hook is uninitialized
stack garbage gets me to where I am.

Again, I'm not too familiar with this area of the kernel, this is just
based on the quick digging that I did. And aplogies again if I'm missing
something ; -) I do think that the uninitialized members could be a problem
though.

Thank you, I hope you have a great day Lorenzo!
Joshua