Re: [PATCH 3/3] PCI: Prevent overflow in proc_bus_pci_write()

[duziming <duziming2@huawei.com>]

在 2025/12/17 18:19, Ilpo Järvinen 写道:
> On Wed, 17 Dec 2025, duziming wrote:
>> 在 2025/12/16 18:57, Ilpo Järvinen 写道:
>>> On Tue, 16 Dec 2025, Ziming Du wrote:
>>>
>>>> When the value of ppos over the INT_MAX, the pos will be
>>> is over
>>>
>>>> set a negtive value which will be pass to get_user() or
>>> set to a negative value which will be passed
>>>
>>>> pci_user_write_config_dword(). And unexpected behavior
>>> Please start the sentence with something else than And.
>>>
>>> Hmm, the lines look rather short too, can you please reflow the changelog
>>> paragraphs to 75 characters.
>> Thanks for the review. I'll reflow the changelog to 75-character lines and
>> avoid
>>
>> starting sentences with 'And' in the next revision.
>>
>>>> such as a softlock happens:
>>>>
>>>>    watchdog: BUG: soft lockup - CPU#0 stuck for 130s! [syz.3.109:3444]
>>>>    Modules linked in:
>>>>    CPU: 0 PID: 3444 Comm: syz.3.109 Not tainted 6.6.0+ #33
>>>>    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
>>>> rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/01/2014
>>>>    RIP: 0010:_raw_spin_unlock_irq+0x17/0x30
>>>>    Code: cc cc cc 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 f3 0f 1e
>>>> fa 0f 1f 44 00 00 e8 52 12 00 00 90 fb 65 ff 0d b1 a1 86 6d <74> 05 e9 42
>>>> 52 00 00 0f 1f 44 00 00 c3 cc cc cc cc 0f 1f 84 00 00
>>>>    RSP: 0018:ffff88816851fb50 EFLAGS: 00000246
>>>>    RAX: 0000000000000001 RBX: 0000000000000000 RCX: ffffffff927daf9b
>>>>    RDX: 0000000000000cfc RSI: 0000000000000046 RDI: ffffffff9a7c7400
>>>>    RBP: 00000000818bb9dc R08: 0000000000000001 R09: ffffed102d0a3f59
>>>>    R10: 0000000000000003 R11: 0000000000000000 R12: 0000000000000000
>>>>    R13: ffff888102220000 R14: ffffffff926d3b10 R15: 00000000210bbb5f
>>>>    FS:  00007ff2d4e56640(0000) GS:ffff8881f5c00000(0000)
>>>> knlGS:0000000000000000
>>>>    CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>>>    CR2: 00000000210bbb5b CR3: 0000000147374002 CR4: 0000000000772ef0
>>>>    PKRU: 00000000
>>>>    Call Trace:
>>>>     <TASK>
>>>>     pci_user_write_config_dword+0x126/0x1f0
>>>>     ? __get_user_nocheck_8+0x20/0x20
>>>>     proc_bus_pci_write+0x273/0x470
>>>>     proc_reg_write+0x1b6/0x280
>>>>     do_iter_write+0x48e/0x790
>>>>     ? import_iovec+0x47/0x90
>>>>     vfs_writev+0x125/0x4a0
>>>>     ? futex_wake+0xed/0x500
>>>>     ? __pfx_vfs_writev+0x10/0x10
>>>>     ? userfaultfd_ioctl+0x131/0x1ae0
>>>>     ? userfaultfd_ioctl+0x131/0x1ae0
>>>>     ? do_futex+0x17e/0x220
>>>>     ? __pfx_do_futex+0x10/0x10
>>>>     ? __fget_files+0x193/0x2b0
>>>>     __x64_sys_pwritev+0x1e2/0x2a0
>>>>     ? __pfx___x64_sys_pwritev+0x10/0x10
>>>>     do_syscall_64+0x59/0x110
>>>>     entry_SYSCALL_64_after_hwframe+0x78/0xe2
>>> Could you please trim the dump so it only contains things relevant to this
>>> issue () (also check trimming in the other patches).
>> Thanks for pointing that out, we'll make sure to only keep the relevant stacks
>> in future patches.
>>>> Fix this by use unsigned int for the pos.
>>>>
>>>> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
>>>> Signed-off-by: Yongqiang Liu <liuyongqiang13@huawei.com>
>>>> Signed-off-by: Ziming Du <duziming2@huawei.com>
>>>> ---
>>>>    drivers/pci/proc.c | 2 +-
>>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/pci/proc.c b/drivers/pci/proc.c
>>>> index 9348a0fb8084..dbec1d4209c9 100644
>>>> --- a/drivers/pci/proc.c
>>>> +++ b/drivers/pci/proc.c
>>>> @@ -113,7 +113,7 @@ static ssize_t proc_bus_pci_write(struct file *file,
>>>> const char __user *buf,
>>>>    {
>>>>    	struct inode *ino = file_inode(file);
>>>>    	struct pci_dev *dev = pde_data(ino);
>>>> -	int pos = *ppos;
>>>> +	unsigned int pos = *ppos;
>>>>    	int size = dev->cfg_size;
>>>>    	int cnt, ret;
>>> So this still throws away some bits compared with the original ppos ?
>> The current approach may lose some precision compared to the original ppos,
>> but a later check ensures  pos
>>
>> remains valid—so any potential information loss is handled safely.
> That's somewhat odd definition of "valid" if a big ppos results in
> a smaller number after the precision loss that is smaller than size.

Oh, I get your concern now. In fact, in previous version, we fixed it 
like this :

diff --git a/drivers/pci/proc.c b/drivers/pci/proc.c
index dbec1d4209c9..200d42feafd8 100644
--- a/drivers/pci/proc.c
+++ b/drivers/pci/proc.c
@@ -113,7 +113,7 @@ static ssize_t proc_bus_pci_write(struct file *file, 
const char __user *buf,
  {
         struct inode *ino = file_inode(file);
         struct pci_dev *dev = pde_data(ino);
-       unsigned int pos = *ppos;
+       int pos = *ppos;
         int size = dev->cfg_size;
         int cnt, ret;

@@ -121,7 +121,7 @@ static ssize_t proc_bus_pci_write(struct file *file, 
const char __user *buf,
         if (ret)
                 return ret;

-       if (pos >= size)
+       if (pos >= size || pos < 0)
                 return 0;
         if (nbytes >= size)
                 nbytes = size;

In addition, we notice that in proc_bus_pci_read(), "unsigned int pos = 
*ppos" might also cause some issues.

>