linux-kernel - Re: [PATCH] PCI/MSI: fix page fault when msi_populate

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Tue, 12 Oct 2021 10:47:20 +0800
From:   "wanghai (M)" <wanghai38@...wei.com>
To:     Barry Song <21cnbao@...il.com>
CC:     Bjorn Helgaas <helgaas@...nel.org>,
        Bjorn Helgaas <bhelgaas@...gle.com>,
        Marc Zyngier <maz@...nel.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        "Barry Song" <song.bao.hua@...ilicon.com>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        <linux-pci@...r.kernel.org>, LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] PCI/MSI: fix page fault when msi_populate_sysfs() failed


在 2021/10/12 10:39, Barry Song 写道:
> On Tue, Oct 12, 2021 at 3:25 PM wanghai (M) <wanghai38@...wei.com> wrote:
>>
>> 在 2021/10/12 10:09, Bjorn Helgaas 写道:
>>> On Tue, Oct 12, 2021 at 09:59:40AM +0800, wanghai (M) wrote:
>>>> 在 2021/10/12 1:11, Bjorn Helgaas 写道:
>>>>> For v2, please note "git log --oneline drivers/pci/msi.c" and make
>>>>> your patch follow the style, including capitalization.
>>>>>
>>>>> On Mon, Oct 11, 2021 at 05:15:28PM +0800, wanghai (M) wrote:
>>>>>> 在 2021/10/11 16:52, Barry Song 写道:
>>>>>>> On Mon, Oct 11, 2021 at 9:24 PM Wang Hai <wanghai38@...wei.com> wrote:
>>>>>>>> I got a page fault report when doing fault injection test:
>>>>> When you send v2, can you include information about how you injected
>>>>> the fault?  If it's easy, others can reproduce the failure that way.
>>>> Sorry, the reproduction needs to be based on the fault injection framework
>>>> provided by Hulk Robot. I don't know how the framework is implemented.
>>>>
>>>> The way to reproduce this is to do a fault injection to make
>>>> 'msi_attrs = kcalloc() in msi_populate_sysfs()' fail when insmod
>>>> 9pnet_virtio.ko.
>>>>
>>>> I sent v2 yesterday, can you help review it?
>>>> https://lore.kernel.org/linux-pci/20211011130837.766323-1-wanghai38@huawei.com/
>>>>>>>> BUG: unable to handle page fault for address: fffffffffffffff4
>>>>>>>> ...
>>>>>>>> RIP: 0010:sysfs_remove_groups+0x25/0x60
>>>>>>>> ...
>>>>>>>> Call Trace:
>>>>>>>>      msi_destroy_sysfs+0x30/0xa0
>>>>>>>>      free_msi_irqs+0x11d/0x1b0
>>>>>>>>      __pci_enable_msix_range+0x67f/0x760
>>>>>>>>      pci_alloc_irq_vectors_affinity+0xe7/0x170
>>>>>>>>      vp_find_vqs_msix+0x129/0x560
>>>>>>>>      vp_find_vqs+0x52/0x230
>>>>>>>>      vp_modern_find_vqs+0x47/0xb0
>>>>>>>>      p9_virtio_probe+0xa1/0x460 [9pnet_virtio]
>>>>>>>>      virtio_dev_probe+0x1ed/0x2e0
>>>>>>>>      really_probe+0x1c7/0x400
>>>>>>>>      __driver_probe_device+0xa4/0x120
>>>>>>>>      driver_probe_device+0x32/0xe0
>>>>>>>>      __driver_attach+0xbf/0x130
>>>>>>>>      bus_for_each_dev+0xbb/0x110
>>>>>>>>      driver_attach+0x27/0x30
>>>>>>>>      bus_add_driver+0x1d9/0x270
>>>>>>>>      driver_register+0xa9/0x180
>>>>>>>>      register_virtio_driver+0x31/0x50
>>>>>>>>      p9_virtio_init+0x3c/0x1000 [9pnet_virtio]
>>>>>>>>      do_one_initcall+0x7b/0x380
>>>>>>>>      do_init_module+0x5f/0x21e
>>>>>>>>      load_module+0x265c/0x2c60
>>>>>>>>      __do_sys_finit_module+0xb0/0xf0
>>>>>>>>      __x64_sys_finit_module+0x1a/0x20
>>>>>>>>      do_syscall_64+0x34/0xb0
>>>>>>>>      entry_SYSCALL_64_after_hwframe+0x44/0xae
>>>>>>>>
>>>>>>>> When populating msi_irqs sysfs failed in msi_capability_init() or
>>>>>>>> msix_capability_init(), dev->msi_irq_groups will point to ERR_PTR(...).
>>>>>>>> This will cause a page fault when destroying the wrong
>>>>>>>> dev->msi_irq_groups in free_msi_irqs().
>>>>>>>>
>>>>>>>> Fix this by setting dev->msi_irq_groups to NULL when msi_populate_sysfs()
>>>>>>>> failed.
>>>>>>>>
>>>>>>>> Fixes: 2f170814bdd2 ("genirq/msi: Move MSI sysfs handling from PCI to MSI core")
>>>>>>>> Reported-by: Hulk Robot <hulkci@...wei.com>
>>>>> What exactly was reported by the Hulk Robot?  Did it really do the
>>>>> fault injection and report the page fault?
>>>> Yes, it reported the error and provided a way to reproduce it
>>> Great, can you include a link to that report then?
>>> .
>> Currently hulk robot is still in the process of continuous improvement and
>> is not open to the public for the time being, so you can not access our
>> links at the moment. We will open it in the future when it is perfected.
> hi hai, would you like to put some information in the commit log like
> if  'msi_attrs = kcalloc() in msi_populate_sysfs()' fails, blah, blah, blah...
>
> It seems this can make things a bit clearer to me. Anyway, it doesn't matter
> too much. The fix is correct.
Okay, I'll refine the commit log and resend a patch
>> --
>> Wang Hai
>>
> Thanks
> barry
> .
>
-- 
Wang Hai