The IScsiHexToBin() function documents the EFI_BUFFER_TOO_SMALL return
condition, but never actually checks whether the decoded buffer fits into
the caller-provided room (i.e., the input value of "BinLength"), and
EFI_BUFFER_TOO_SMALL is never returned. The decoding of "HexStr" can
overflow "BinBuffer".
This is remotely exploitable, as shown in a subsequent patch, which adds
error checking to the IScsiHexToBin() call sites. This issue allows the
target to compromise the initiator.
Introduce EFI_BAD_BUFFER_SIZE, in addition to the existent
EFI_BUFFER_TOO_SMALL, for reporting a special case of the buffer overflow,
plus actually catch the buffer overflow.
Cc: Jiaxin Wu <jiaxin.wu@intel.com>
Cc: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Cc: Philippe Mathieu-Daudé <philmd@redhat.com>
Cc: Siyuan Fu <siyuan.fu@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=3356
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20210608121259.32451-10-lersek@redhat.com>
The IScsiHexToBin() function has the following parser issues:
(1) If the *subject sequence* in "HexStr" is empty, the function returns
EFI_SUCCESS (with "BinLength" set to 0 on output). Such inputs should
be rejected.
(2) The function mis-handles a "HexStr" that ends with a stray nibble. For
example, if "HexStr" is "0xABC", the function decodes it to the bytes
{0xAB, 0x0C}, sets "BinLength" to 2 on output, and returns
EFI_SUCCESS. Such inputs should be rejected.
(3) If an invalid hex char is found in "HexStr", the function treats it as
end-of-hex-string, and returns EFI_SUCCESS. Such inputs should be
rejected.
All of the above cases are remotely triggerable, as shown in a subsequent
patch, which adds error checking to the IScsiHexToBin() call sites. While
the initiator is not immediately compromised, incorrectly parsing CHAP_R
from the target, in case of mutual authentication, is not great.
Extend the interface contract of IScsiHexToBin() with
EFI_INVALID_PARAMETER, for reporting issues (1) through (3), and implement
the new checks.
Cc: Jiaxin Wu <jiaxin.wu@intel.com>
Cc: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Cc: Philippe Mathieu-Daudé <philmd@redhat.com>
Cc: Siyuan Fu <siyuan.fu@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=3356
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20210608121259.32451-9-lersek@redhat.com>
Considering IScsiBinToHex():
> if (((*HexLength) - 3) < BinLength * 2) {
> *HexLength = BinLength * 2 + 3;
> }
the following subexpressions are problematic:
(*HexLength) - 3
BinLength * 2
BinLength * 2 + 3
The first one may wrap under zero, the latter two may wrap over
MAX_UINT32.
Rewrite the calculation using SafeIntLib.
While at it, change the type of the "Index" variable from UINTN to UINT32.
The largest "Index"-based value that we calculate is
Index * 2 + 2 (with (Index == BinLength))
Because the patch makes
BinLength * 2 + 3
safe to calculate in UINT32, using UINT32 for
Index * 2 + 2 (with (Index == BinLength))
is safe too. Consistently using UINT32 improves readability.
This patch is best reviewed with "git show -W".
The integer overflows that this patch fixes are theoretical; a subsequent
patch in the series will audit the IScsiBinToHex() call sites, and show
that none of them can fail.
Cc: Jiaxin Wu <jiaxin.wu@intel.com>
Cc: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Cc: Philippe Mathieu-Daudé <philmd@redhat.com>
Cc: Siyuan Fu <siyuan.fu@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=3356
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20210608121259.32451-6-lersek@redhat.com>
The "ISCSI_CHAP_AUTH_DATA.OutChallenge" field is declared as a UINT8 array
with ISCSI_CHAP_AUTH_MAX_LEN (1024) elements. However, when the challenge
is generated and formatted, only ISCSI_CHAP_RSP_LEN (16) octets are used
in the array.
Change the array size to ISCSI_CHAP_RSP_LEN, and remove the (now unused)
ISCSI_CHAP_AUTH_MAX_LEN macro.
Remove the "ISCSI_CHAP_AUTH_DATA.OutChallengeLength" field, which is
superfluous too.
Most importantly, explain in a new comment *why* tying the challenge size
to the digest size (ISCSI_CHAP_RSP_LEN) has always made sense. (See also
Linux kernel commit 19f5f88ed779, "scsi: target: iscsi: tie the challenge
length to the hash digest size", 2019-11-06.) For sure, the motivation
that the new comment now explains has always been there, and has always
been the same, for IScsiDxe; it's just that now we spell it out too.
No change in peer-visible behavior.
Cc: Jiaxin Wu <jiaxin.wu@intel.com>
Cc: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Cc: Philippe Mathieu-Daudé <philmd@redhat.com>
Cc: Siyuan Fu <siyuan.fu@intel.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=3356
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Message-Id: <20210608121259.32451-4-lersek@redhat.com>
BZ: https://bugzilla.tianocore.org/show_bug.cgi?id=2917
Add NETWORK_HTTP_ENABLE macro and separate HttpDxe
and HttpUtilitiesDxe drivers from
HTTP_NETWORK_HTTP_BOOT_ENABLE macro.
Current NETWORK_HTTP_BOOT_ENABLE macro is defined to enable HTTP
boot feature in POST, this macro is not only enabling HTTP Boot
related modules but also enabling other generic HTTP modules
such as HttpDxe, HttpUtilitiesDxe and DnsDxe.
These HTTP base drivers would not be only used by HTTP boot
when we introduce the use case of Redfish implementation over
HTTP to edk2.
We should have a dedicate macro to enable generic HTTP functions
on Network stack and additionally provide NETWORK_HTTP_BOOT_ENABLE
for HTTP boot functionality for the use case that platform doesn't
require HTTP boot.
Signed-off-by: Abner Chang <abner.chang@hpe.com>
Cc: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Cc: Jiaxin Wu <jiaxin.wu@intel.com>
Cc: Siyuan Fu <siyuan.fu@intel.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Cc: Nickle Wang <nickle.wang@hpe.com>
Cc: Peter O'Hanley <peter.ohanley@hpe.com>
Reviewed-by: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
REF: https://bugzilla.tianocore.org/show_bug.cgi?id=3003
There is a plan to make MD5 disable as default.
The new MACRO ENABLE_MD5_DEPRECATED_INTERFACES
would be introduced to enable MD5. Make the
definition ahead of the change to avoid build
error after the MACRO changed.
1. Add the NetworkBuildOptions.dsc.inc to define
the MACRO for build (support: GCC, INTEL, MSFT and
RVCT)
2. Add the BuildOption file to Network.dsc.inc
Cc: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Cc: Jiaxin Wu <jiaxin.wu@intel.com>
Cc: Siyuan Fu <siyuan.fu@intel.com>
Signed-off-by: Zhichao Gao <zhichao.gao@intel.com>
Message-Id: <20201112055558.2348-5-zhichao.gao@intel.com>
[lersek@redhat.com: clean up comments in "NetworkBuildOptions.dsc.inc"]
[lersek@redhat.com: hoist "BuildOptions" above "Components" in
"Network.dsc.inc" for bug compat with edk2-platforms]
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Siyuan Fu <siyuan.fu@intel.com>
REF:https://bugzilla.tianocore.org/show_bug.cgi?id=1562
The current SnpDxe implementation registers its ExitBootServices event
notification function (SnpNotifyExitBootServices ()) at TPL_NOTIFY. This
function calls PxeShutdown() which issues an UNDI shutdown operation.
Ultimately, this may invoke Shutdown() in EFI_SIMPLE_NETWORK_PROTOCOL.
The UEFI specification 2.8A Table 27 "TPL Restrictions" restricts the
TPL
for Simple Network Protocol to <= TPL_CALLBACK. In addition, it has been
observed in some 3rd party UNDI drivers to cause an issue further down
the call stack if the TPL is higher than TPL_CALLBACK on invocation.
Therefore, this commit changes the TPL of SnpNotifyExitBootServices() to
TPL_CALLBACK.
Cc: Siyuan Fu <siyuan.fu@intel.com>
Cc: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Cc: Jiaxin Wu <jiaxin.wu@intel.com>
Signed-off-by: Michael Kubacki <michael.kubacki@microsoft.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Maciej Rabeda <maciej.rabeda@linux.intel.com>
When DHCP is misconfigured on a network segment, such that two DHCP
servers attempt to reply to requests (and therefore race with each other),
the edk2 PXE client can confuse itself.
In PxeBcDhcp4BootInfo() / PxeBcDhcp6BootInfo(), the client may refer to a
DHCP reply packet as an "earlier" packet from the "same" DHCP server, when
in reality both packets are unrelated, and arrive from different DHCP
servers.
While the edk2 PXE client can do nothing to fix this, it should at least
not ASSERT() -- ASSERT() is for catching programming errors (violations of
invariants that are under the control of the programmer). ASSERT()s should
in particular not refer to external data (such as network packets). What's
more, in RELEASE builds, we get NULL pointer references.
Check the problem conditions with actual "if"s, and return
EFI_PROTOCOL_ERROR. This will trickle out to PxeBcLoadBootFile(), and be
reported as "PXE-E99: Unexpected network error".
Cc: Jiaxin Wu <jiaxin.wu@intel.com>
Cc: Maciej Rabeda <maciej.rabeda@linux.intel.com>
Cc: Philippe Mathieu-Daudé <philmd@redhat.com>
Cc: Siyuan Fu <siyuan.fu@intel.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Message-Id: <20200331004749.16128-1-lersek@redhat.com>
Reviewed-by: Siyuan Fu <siyuan.fu@intel.com>
Reviewed-by: Maciej Rabeda <maciej.rabeda@linux.intel.com>
