MIME-Version: 1.0
In-Reply-To: <CAE_wzQ9ZVDiDGP4k_2i2KL4JxtRn_S7gjFXUTYZncMLx2m77gQ@mail.gmail.com>
References: <1490256755-889-1-git-send-email-jeffy.chen@rock-chips.com>
 <1490256755-889-3-git-send-email-jeffy.chen@rock-chips.com>
 <CAL_Jsq+xxG6w6J+Zz-kcJ+rA9jJomyHFiazHzcPHuQizn-4-wg@mail.gmail.com> <CAE_wzQ9ZVDiDGP4k_2i2KL4JxtRn_S7gjFXUTYZncMLx2m77gQ@mail.gmail.com>
From: Bjorn Helgaas <bhelgaas@google.com>
Date: Tue, 4 Apr 2017 14:18:04 -0500
Message-ID: <CAErSpo6CZ1hoHyzEnWBi7VDr51bLurF4-4t=-v4jH27kOpzYrg@mail.gmail.com>
Subject: Re: [PATCH v2 2/2] of/pci: Fix memory leak in of_pci_get_host_bridge_resources
To: Dmitry Torokhov <dtor@chromium.org>
Cc: Rob Herring <robh@kernel.org>, Jeffy Chen <jeffy.chen@rock-chips.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        toshi.kani@hpe.com, Shawn Lin <shawn.lin@rock-chips.com>,
        Brian Norris <briannorris@chromium.org>,
        Doug Anderson <dianders@chromium.org>,
        Frank Rowand <frowand.list@gmail.com>,
        "devicetree@vger.kernel.org" <devicetree@vger.kernel.org>
Content-Type: text/plain; charset=UTF-8
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 3727
Lines: 87

On Thu, Mar 23, 2017 at 5:58 PM, Dmitry Torokhov <dtor@chromium.org> wrote:
> On Thu, Mar 23, 2017 at 3:07 PM, Rob Herring <robh@kernel.org> wrote:
>> On Thu, Mar 23, 2017 at 3:12 AM, Jeffy Chen <jeffy.chen@rock-chips.com> wrote:
>>> Currently we only free the allocated resource struct when error.
>>> This would cause memory leak after pci_free_resource_list.
>>>
>>> Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com>
>>> ---
>>>
>>> Changes in v2:
>>> Don't change the resource_list_create_entry's behavior.
>>>
>>>  drivers/of/of_pci.c | 57 +++++++++++++++++++++++------------------------------
>>>  1 file changed, 25 insertions(+), 32 deletions(-)
>>>
>>> diff --git a/drivers/of/of_pci.c b/drivers/of/of_pci.c
>>> index 0ee42c3..a0ec246 100644
>>> --- a/drivers/of/of_pci.c
>>> +++ b/drivers/of/of_pci.c
>>> @@ -190,8 +190,7 @@ int of_pci_get_host_bridge_resources(struct device_node *dev,
>>>                         struct list_head *resources, resource_size_t *io_base)
>>>  {
>>>         struct resource_entry *window;
>>> -       struct resource *res;
>>> -       struct resource *bus_range;
>>> +       struct resource res;
>>>         struct of_pci_range range;
>>>         struct of_pci_range_parser parser;
>>>         char range_type[4];
>>> @@ -200,24 +199,24 @@ int of_pci_get_host_bridge_resources(struct device_node *dev,
>>>         if (io_base)
>>>                 *io_base = (resource_size_t)OF_BAD_ADDR;
>>>
>>> -       bus_range = kzalloc(sizeof(*bus_range), GFP_KERNEL);
>>> -       if (!bus_range)
>>> -               return -ENOMEM;
>>> -
>>>         pr_info("host bridge %s ranges:\n", dev->full_name);
>>>
>>> -       err = of_pci_parse_bus_range(dev, bus_range);
>>> +       err = of_pci_parse_bus_range(dev, &res);
>>>         if (err) {
>>> -               bus_range->start = busno;
>>> -               bus_range->end = bus_max;
>>> -               bus_range->flags = IORESOURCE_BUS;
>>> -               pr_info("  No bus range found for %s, using %pR\n",
>>> -                       dev->full_name, bus_range);
>>> +               res.start = busno;
>>> +               res.end = bus_max;
>>> +               res.flags = IORESOURCE_BUS;
>>> +               pr_info("  No bus range found for %s\n", dev->full_name);
>>>         } else {
>>> -               if (bus_range->end > bus_range->start + bus_max)
>>> -                       bus_range->end = bus_range->start + bus_max;
>>> +               if (res.end > res.start + bus_max)
>>> +                       res.end = res.start + bus_max;
>>> +       }
>>> +       window = pci_add_resource(resources, NULL);
>>> +       if (!window) {
>>> +               err = -ENOMEM;
>>> +               goto parse_failed;
>>>         }
>>> -       pci_add_resource(resources, bus_range);
>>> +       *window->res = res;
>>
>> Well, now this seems racy. You add a blank resource to the list first
>> and then fill it in.
>>
>
> Huh? There is absolutely no guarantees for concurrent access here.
> pcI_add_resource_offset() first adds a resource and then modifies
> offset. Here we add an empty resource and then fill it in.

I don't really like this pattern either.  Even if there's no actual
racy behavior, it takes more analysis than necessary to figure that
out.

pci_add_resource_offset() allocates a resource list entry, sets the
offset, then adds it to the list.  It doesn't update a resource entry
that might be visible to anybody else.  Here we do update a resource
that is already visible to others because it's already on the list.

Bjorn

BTW, please CC linux-pci on the entire series so it's easier to
review.  I don't know where you envision having this applied, but I
only apply things to the PCI tree after they appear on linux-pci.