| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20240129034304-mutt-send-email-mst@kernel.org>
Date: Mon, 29 Jan 2024 03:46:55 -0500
From: "Michael S. Tsirkin" <mst@...hat.com>
To: Guenter Roeck <linux@...ck-us.net>
Cc: Nathan Chancellor <nathan@...nel.org>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
stable@...r.kernel.org, patches@...ts.linux.dev,
linux-kernel@...r.kernel.org, torvalds@...ux-foundation.org,
llvm@...ts.linux.dev, keescook@...omium.org,
arei.gonglei@...wei.com, jasowang@...hat.com,
virtualization@...ts.linux.dev, linux-crypto@...r.kernel.org,
zhenwei pi <pizhenwei@...edance.com>
Subject: Re: [PATCH 5.10 000/286] 5.10.209-rc1 review
On Fri, Jan 26, 2024 at 03:55:02PM -0800, Guenter Roeck wrote:
> On 1/26/24 14:35, Nathan Chancellor wrote:
> > (slimming up the CC list, I don't think this is too relevant to the
> > wider stable community)
> >
> > On Fri, Jan 26, 2024 at 01:01:15PM -0800, Guenter Roeck wrote:
> > > On 1/26/24 12:34, Nathan Chancellor wrote:
> > > > On Fri, Jan 26, 2024 at 10:17:23AM -0800, Guenter Roeck wrote:
> > > > > On 1/26/24 09:51, Greg Kroah-Hartman wrote:
> > > > > > On Fri, Jan 26, 2024 at 08:46:42AM -0800, Guenter Roeck wrote:
> > > > > > > On 1/22/24 15:55, Greg Kroah-Hartman wrote:
> > > > > > > > This is the start of the stable review cycle for the 5.10.209 release.
> > > > > > > > There are 286 patches in this series, all will be posted as a response
> > > > > > > > to this one. If anyone has any issues with these being applied, please
> > > > > > > > let me know.
> > > > > > > >
> > > > > > > > Responses should be made by Wed, 24 Jan 2024 23:56:49 +0000.
> > > > > > > > Anything received after that time might be too late.
> > > > > > > >
> > > > > > > [ ... ]
> > > > > > >
> > > > > > > > zhenwei pi <pizhenwei@...edance.com>
> > > > > > > > virtio-crypto: implement RSA algorithm
> > > > > > > >
> > > > > > >
> > > > > > > Curious: Why was this (and its subsequent fixes) backported to v5.10.y ?
> > > > > > > It is quite beyond a bug fix. Also, unless I am really missing something,
> > > > > > > the series (or at least this patch) was not applied to v5.15.y, so we now
> > > > > > > have functionality in v5.10.y which is not in v5.15.y.
> > > > > >
> > > > > > See the commit text, it was a dependency of a later fix and documented
> > > > > > as such.
> > > > > >
> > > > > > Having it in 5.10 and not 5.15 is a bit odd, I agree, so patches are
> > > > > > gladly accepted :)
> > > > > >
> > > > >
> > > > > We reverted the entire series from the merge because it results in a build
> > > > > failure for us.
> > > > >
> > > > > In file included from /home/groeck/src/linux-chromeos/drivers/crypto/virtio/virtio_crypto_akcipher_algs.c:10:
> > > > > In file included from /home/groeck/src/linux-chromeos/include/linux/mpi.h:21:
> > > > > In file included from /home/groeck/src/linux-chromeos/include/linux/scatterlist.h:5:
> > > > > In file included from /home/groeck/src/linux-chromeos/include/linux/string.h:293:
> > > > > /home/groeck/src/linux-chromeos/include/linux/fortify-string.h:512:4: error: call to __read_overflow2_field declared with 'warning' attribute: detected read beyond size of field (2nd parameter); maybe use struct_group()? [-Werror,-Wattribute-warning]
> > > > > __read_overflow2_field(q_size_field, size);
> > > >
> > > > For what it's worth, this is likely self inflicted for chromeos-5.10,
> > > > which carries a revert of commit eaafc590053b ("fortify: Explicitly
> > > > disable Clang support") as commit c19861d34c003 ("CHROMIUM: Revert
> > > > "fortify: Explicitly disable Clang support""). I don't see the series
> > > > that added proper support for clang to fortify in 5.18 that ended with
> > > > commit 281d0c962752 ("fortify: Add Clang support") in that ChromeOS
> > > > branch, so this seems somewhat expected.
> > > >
> > >
> > > That explains that ;-). I don't mind if the patches stay in v5.10.y,
> > > we have them reverted anyway.
> > >
> > > The revert was a pure process issue, as you may see when looking into
> > > commit c19861d34c003, so, yes, I agree that it is self-inflicted damage.
> > > Still, that doesn't explain why the problem exists in 5.18+.
> > >
> > > > > I also see that upstream (starting with 6.1) when trying to build it with clang,
> > > > > so I guess it is one of those bug-for-bug compatibility things. I really have
> > > > > no idea what causes it, or why we don't see the problem when building
> > > > > chromeos-6.1 or chromeos-6.6, but (so far) only with chromeos-5.10 after
> > > > > merging 5.10.209 into it. Making things worse, the problem isn't _always_
> > > > > seen. Sometimes I can compile the file in 6.1.y without error, sometimes not.
> > > > > I have no idea what triggers the problem.
> > > >
> > > > Have a .config that reproduces it on upstream? I have not personally
> > > > seen this warning in my build matrix nor has our continuous-integration
> > > > matrix (I don't see it in the warning output at the bottom but that
> > > > could have missed something for some reason) in 6.1:
> > > >
> > >
> > > The following command sequence reproduces the problem for me with all stable
> > > branches starting with 5.18.y (plus mainline).
> > >
> > > rm -rf /tmp/crypto-build
> > > mkdir /tmp/crypto-build
> > > make -j CC=clang-15 mrproper >/dev/null 2>&1
> > > make -j O=/tmp/crypto-build CC=clang-15 allmodconfig >/dev/null 2>&1
> > > make -j O=/tmp/crypto-build W=1 CC=clang-15 drivers/crypto/virtio/virtio_crypto_akcipher_algs.o
> > >
> > > I tried clang versions 14, 15, and 16. This is with my home system running
> > > Ubuntu 22.04, no ChromeOS or Google specifics/internals involved. For clang-15,
> > > the version is
> > >
> > > Ubuntu clang version 15.0.7
> > > Target: x86_64-pc-linux-gnu
> > > Thread model: posix
> > > InstalledDir: /usr/bin
> >
> > Okay interesting, this warning is hidden behind W=1, which our CI does
> > not test with. Looks like it has been that way since the introduction of
> > these checks in f68f2ff91512 ("fortify: Detect struct member overflows
> > in memcpy() at compile-time").
> >
>
> Interestingly the warning is seen in chromeos-5.10, without this patch,
> and without W=1. I guess that must have to do with the revert which is
> finally biting us.
>
> > I think this is a legitimate warning though. It is complaining about the
> > second memcpy() in virtio_crypto_alg_akcipher_init_session():
> >
> > memcpy(&ctrl->u, para, sizeof(ctrl->u));
> >
> > where ctrl is:
> >
> > struct virtio_crypto_op_ctrl_req {
> > struct virtio_crypto_ctrl_header header; /* 0 16 */
> > union {
> > struct virtio_crypto_sym_create_session_req sym_create_session; /* 16 56 */
> > struct virtio_crypto_hash_create_session_req hash_create_session; /* 16 56 */
> > struct virtio_crypto_mac_create_session_req mac_create_session; /* 16 56 */
> > struct virtio_crypto_aead_create_session_req aead_create_session; /* 16 56 */
> > struct virtio_crypto_akcipher_create_session_req akcipher_create_session; /* 16 56 */
> > struct virtio_crypto_destroy_session_req destroy_session; /* 16 56 */
> > __u8 padding[56]; /* 16 56 */
> > } u; /* 16 56 */
> > union {
> > struct virtio_crypto_sym_create_session_req sym_create_session; /* 0 56 */
> > struct virtio_crypto_hash_create_session_req hash_create_session; /* 0 56 */
> > struct virtio_crypto_mac_create_session_req mac_create_session; /* 0 56 */
> > struct virtio_crypto_aead_create_session_req aead_create_session; /* 0 56 */
> > struct virtio_crypto_akcipher_create_session_req akcipher_create_session; /* 0 56 */
> > struct virtio_crypto_destroy_session_req destroy_session; /* 0 56 */
> > __u8 padding[56]; /* 0 56 */
> > };
> >
> >
> > /* size: 72, cachelines: 2, members: 2 */
> > /* last cacheline: 8 bytes */
> > };
> >
> > (so size and p_size_field should be 56) and the type of the para
> > parameter in virtio_crypto_alg_akcipher_init_session() is 'void *' but
> > the para passed by reference to
> > virtio_crypto_alg_akcipher_init_session() in virtio_crypto_rsa_set_key()
> > has a type of 'struct virtio_crypto_akcipher_session_para':
> >
> > struct virtio_crypto_akcipher_session_para {
> > __le32 algo; /* 0 4 */
> > __le32 keytype; /* 4 4 */
> > __le32 keylen; /* 8 4 */
> > union {
> > struct virtio_crypto_rsa_session_para rsa; /* 12 8 */
> > struct virtio_crypto_ecdsa_session_para ecdsa; /* 12 8 */
> > } u; /* 12 8 */
> > union {
> > struct virtio_crypto_rsa_session_para rsa; /* 0 8 */
> > struct virtio_crypto_ecdsa_session_para ecdsa; /* 0 8 */
> > };
> >
> >
> > /* size: 20, cachelines: 1, members: 4 */
> > /* last cacheline: 20 bytes */
> > };
> >
> > (so q_size_field would be 20 if clang were able to do inlining to see
> > through the 'void *'...?), which would result in the
> >
> > __compiletime_lessthan(q_size_field, size)
> >
> > check succeeding and triggering the warning because 20 < 56, so it does
> > seem like there is an overread of the source buffer here? Adding the
>
> Looks like it; I think either the passed 'para' should be of type
> virtio_crypto_akcipher_create_session_req() or it should only copy
> sizeof(struct virtio_crypto_akcipher_session_para) bytes.
>
> Anyway, how did you find that ? Is there a magic trick to find the
> actual code causing the warning ? I am asking because we had seen
> similar warnings before, and it would help to know how to find the
> problematic code.
>
> Thanks,
> Guenter
Cc: zhenwei pi <pizhenwei@...edance.com>
Zhenwei I think you wrote most of the code here.
Can you take a look please?
Stack overflows are plus plus ungood.
> > maintainers of the driver and subsystem in question.
> >
> > Cheers,
> > Nathan
Powered by blists - more mailing lists