patches for DPDK stable branches
 help / color / mirror / Atom feed
From: "Min Hu (Connor)" <>
To: Gaëtan Rivet <>
Cc: <>, <>, <>,
	<>, <>
Subject: Re: [dpdk-stable] [v1, 2/2] bonding: fix PCI address comparison on non-pci ports
Date: Wed, 16 Dec 2020 20:14:53 +0800
Message-ID: <> (raw)
In-Reply-To: <>

     sorry for late reply.
     I know what you mean. But "find_port_id_by_pci_addr"
is one static type funtion. it is only used in the function
"parse_port_id". what you modified in "find_port_id_by_pci_addr"
is totally done before "find_port_id_by_pci_addr" in the function
"parse_port_id". That is, it first find pci_bus, then find
rte_pci_device, at last get port id.
So the bug you described will not happen in current version,
unless others directly use the function "find_port_id_by_pci_addr"
without considering that rte_eth_devices[] may contain non-pci devices.

So, I think the patch is OK to me.

Acked-by: Min Hu (Connor) <>

在 2020/12/7 22:07, Gaëtan Rivet 写道:
> On 26/11/20 10:24 +0800, Min Hu (Connor) wrote:
>> what scenarios may cause bugs in old ways.
>> Could you give an example, thanks.
> Hello,
> For example in the following code:
> -       RTE_ETH_FOREACH_DEV(i) {
> -               pci_dev = RTE_ETH_DEV_TO_PCI(&rte_eth_devices[i]);
> -               eth_pci_addr = &pci_dev->addr;
> All ethdev are iterated, before reading their supposed PCI addresses, and
> comparing those to the one passed in arguments.
> But not all ethdev will be PCI devices, so the cast is incorrect. It will do
> a containerof() on a structure that it supposes contains an rte_device at the
> offset 16 (two pointers accounting for the TAILQ_ENTRY()), but nothing prevents any
> other bus from implementing their device with any other layout.
> So the cast is wrong, but generally it will give out readable memory at least.
> Then the field ((eth_dev)->device)->addr will be read and compared against the input,
> with arbitrary data here.
> A scenario that could cause bug would be when another bus implements a device
> in such a way that it will write plausible binary rep of PCI addresses at the
> same offset, and match a request from a user. The bonding PMD would take the
> device over without properly checking that it is actually a PCI device.
> There have been APIs introduced in the EAL to simplify the iteration of
> devices, especially restricted to buses. Those APIs should be used instead.
> The current implementation is inefficient and wrong. It will work in most cases
> but can still trigger weird issues for users, especially in cases that bonding
> PMD devs won't generally encounter (with setups where multiple buses are used
> with a variety of devices).
> Regards,

  reply	other threads:[~2020-12-16 12:15 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <>
2020-04-17 16:42 ` [dpdk-stable] [PATCH v1 1/2] bonding: fix port id validity check on parsing Gaetan Rivet
2020-11-26  2:45   ` [dpdk-stable] [dpdk-dev][PATCH v1, " Min Hu (Connor)
2020-12-07 13:34     ` [dpdk-stable] [dpdk-dev] [PATCH " Ferruh Yigit
2020-04-17 16:42 ` [dpdk-stable] [PATCH v1 2/2] bonding: fix PCI address comparison on non-pci ports Gaetan Rivet
2020-11-26  2:24   ` [dpdk-stable] [v1, " Min Hu (Connor)
2020-12-07 14:07     ` Gaëtan Rivet
2020-12-16 12:14       ` Min Hu (Connor) [this message]
2020-12-17 10:53         ` Ferruh Yigit

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:

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \ \ \ \ \ \ \ \ \

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

patches for DPDK stable branches

This inbox may be cloned and mirrored by anyone:

	git clone --mirror stable/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 stable stable/ \
	public-inbox-index stable

Example config snippet for mirrors.
Newsgroup available over NNTP:

AGPL code for this site: git clone