From: Vishnu Dasa <vdasa@vmware.com>
To: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Cc: Dan Carpenter <dan.carpenter@oracle.com>,
Jorgen Hansen <jhansen@vmware.com>,
"arnd@arndb.de" <arnd@arndb.de>,
"gregkh@linuxfoundation.org" <gregkh@linuxfoundation.org>,
"acking@vmware.com" <acking@vmware.com>,
"dtor@vmware.com" <dtor@vmware.com>,
Pv-drivers <Pv-drivers@vmware.com>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
"kernel-janitors@vger.kernel.org"
<kernel-janitors@vger.kernel.org>
Subject: Re: [PATCH 3/3] VMCI: Fix some error handling paths in vmci_guest_probe_device()
Date: Thu, 24 Feb 2022 06:53:10 +0000 [thread overview]
Message-ID: <85DB8130-D600-464E-9717-1A0AFE795EA7@vmware.com> (raw)
In-Reply-To: <1bc15512-d811-b26a-d4f3-e14b30730c00@wanadoo.fr>
> On Feb 11, 2022, at 12:06 PM, Christophe JAILLET <christophe.jaillet@wanadoo.fr> wrote:
>
> Le 11/02/2022 à 15:09, Dan Carpenter a écrit :
>> On Thu, Feb 10, 2022 at 11:27:34PM +0100, Christophe JAILLET wrote:
>>> The 'err_remove_vmci_dev_g' error label is not at the right place.
>>> This could lead to un-released resource.
>>>
>>> There is also a missing label. If pci_alloc_irq_vectors() fails, the
>>> previous vmci_event_subscribe() call must be undone.
>>>
>>> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
>>> ---
>>> Review with GREAT care.
>>>
>>> This patch is a recent rebase of an old patch that has never been
>>> submitted.
>>> This function is huge and modifying its error handling path is error
>>> prone (at least for me).
>>>
>>> The patch is compile-tested only.
>> There is still one bug. Sorry if the line numbers are off.
>
> Thanks for the review Dan.
> Much appreciated.
Thanks, Christophe and Dan!
>
>> drivers/misc/vmw_vmci/vmci_guest.c
>> 705 if (capabilities & VMCI_CAPS_NOTIFICATIONS) {
>> 706 vmci_dev->notification_bitmap = dma_alloc_coherent(
>> ^^^^^
>> Alloc
>> 707 &pdev->dev, PAGE_SIZE, &vmci_dev->notification_base,
>> 708 GFP_KERNEL);
>> 709 if (!vmci_dev->notification_bitmap) {
>> 710 dev_warn(&pdev->dev,
>> 711 "Unable to allocate notification bitmap\n");
>> 712 } else {
>> 713 memset(vmci_dev->notification_bitmap, 0, PAGE_SIZE);
>> 714 caps_in_use |= VMCI_CAPS_NOTIFICATIONS;
>> 715 }
>> 716 }
>> 717
>> 718 if (mmio_base != NULL) {
>> 719 if (capabilities & VMCI_CAPS_DMA_DATAGRAM) {
>> 720 caps_in_use |= VMCI_CAPS_DMA_DATAGRAM;
>> 721 } else {
>> 722 dev_err(&pdev->dev,
>> 723 "Missing capability: VMCI_CAPS_DMA_DATAGRAM\n");
>> 724 error = -ENXIO;
>> 725 goto err_free_data_buffers;
>> This should be goto err_free_notification_bitmap;
>
> Agreed.
> The error handling path still looked odd to me because 2 things were undone without a label between the 2 steps.
> That was it. An err_free_notification_bitmap should be added and used.
> I missed it.
Good catch. This fixes recent code, so a separate patch would be good.
"[PATCH v3 3/8] VMCI: dma dg: detect DMA datagram capability"
>
>> 726 }
>> 727 }
>> On of the rules for error handling is that the unwind code should mirror
>> the allocation code but instead of that this code will have:
>> Alloc:
>> if (capabilities & VMCI_CAPS_NOTIFICATIONS)
>> Free:
>> if (vmci_dev->notification_bitmap)
>> It's the same if statement but you wouldn't really know it from just
>> looking at it so it's confusing.
>
> This one is fine I think. If the allocation of notification_bitmap fails, it is not an error. So it looks fine to test it the way it is done.
> Or we should have both 'if'.
>
Right. And we would need to check 'capabilities & VMCI_CAPS_NOTIFICATIONS',
'caps_in_use & VMCI_CAPS_NOTIFICATIONS' and then
'vmci_dev->notification_bitmap' if we go that route. I think we can leave it as is.
>
>> Whatever... But where this really
>> hurts is with:
>> Alloc:
>> if (vmci_dev->exclusive_vectors) {
>> error = request_irq(pci_irq_vector(pdev, 1), ...
>> Free:
>> free_irq(pci_irq_vector(pdev, 1), vmci_dev);
>> No if statement. It works because it's the last allocation but it's
>> confusing and fragile.
>
> Agreed.
Sorry, I hope I'm not missing something obvious or misunderstanding the point.
But I don't think the problem implied here exists?
If 'request_irq(pci_irq_vector(pdev, 0), ...' fails we goto err_disable_msi and there
is no free_irq in this path. If 'request_irq(pci_irq_vector(pdev, 1), ...' fails then we
goto err_free_irq and we do 'free_irq(pci_irq_vector(pdev, 0), vmci_dev)'. Note that
this is for the previous one without the check for vmci_dev->exclusive_vectors.
>
>> The other question I had was:
>> 882 err_remove_bitmap:
>> 883 if (vmci_dev->notification_bitmap) {
>> 884 vmci_write_reg(vmci_dev, VMCI_CONTROL_RESET, VMCI_CONTROL_ADDR);
>> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>> This doesn't mirror anything in the allocation code so who knows if its
>> done in the correct place/order.
>
> Agreed. It puzzled me as well.
>
> vmci_guest_remove_device() also has this kind of code, but it is not done the same way in this function. It is unconditional and not close to the dma_free_coherent() call.
> Odd.
>
> I won't touch it by myself :)
>
>> 885 dma_free_coherent(&pdev->dev, PAGE_SIZE,
>> 886 vmci_dev->notification_bitmap,
>> 887 vmci_dev->notification_base);
>> 888 }
>> regards,
>> dan carpenter
>
> All your comments are unrelated to my patch and looks like additional fixes.
>
> Until recently, this file was mostly untouched.
> So, let see if a maintainer looks interested in these patches and if he prefers a patch that fixes everything or several patches, maybe easier to review.
>
> Once again, big thanks.
>
> CJ
next prev parent reply other threads:[~2022-02-24 6:53 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-02-10 22:26 [PATCH 0/3] VMCI: Various fixes Christophe JAILLET
2022-02-10 22:27 ` [PATCH 1/3] VMCI: Fix the description of vmci_check_host_caps() Christophe JAILLET
2022-02-17 7:08 ` Vishnu Dasa
2022-02-10 22:27 ` [PATCH 2/3] VMCI: No need to clear memory after a dma_alloc_coherent() call Christophe JAILLET
2022-02-17 7:10 ` Vishnu Dasa
2022-02-10 22:27 ` [PATCH 3/3] VMCI: Fix some error handling paths in vmci_guest_probe_device() Christophe JAILLET
2022-02-11 14:09 ` Dan Carpenter
2022-02-11 20:06 ` Christophe JAILLET
2022-02-24 6:53 ` Vishnu Dasa [this message]
2022-02-24 10:03 ` Dan Carpenter
2022-02-24 16:08 ` Vishnu Dasa
2022-02-24 0:43 ` Vishnu Dasa
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=85DB8130-D600-464E-9717-1A0AFE795EA7@vmware.com \
--to=vdasa@vmware.com \
--cc=Pv-drivers@vmware.com \
--cc=acking@vmware.com \
--cc=arnd@arndb.de \
--cc=christophe.jaillet@wanadoo.fr \
--cc=dan.carpenter@oracle.com \
--cc=dtor@vmware.com \
--cc=gregkh@linuxfoundation.org \
--cc=jhansen@vmware.com \
--cc=kernel-janitors@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).