When pkcs7_verify_sig_chain() is building the certificate chain for a
SignerInfo using the certificates in the PKCS#7 message, it is passing
the wrong arguments to public_key_verify_signature(). Consequently,
when the next certificate is supposed to be used to verify the previous
certificate, the next certificate is actually used to verify itself.
An attacker can use this bug to create a bogus certificate chain that
has no cryptographic relationship between the beginning and end.
Fortunately I couldn't quite find a way to use this to bypass the
overall signature verification, though it comes very close. Here's the
reasoning: due to the bug, every certificate in the chain beyond the
first actually has to be self-signed (where "self-signed" here refers to
the actual key and signature; an attacker might still manipulate the
certificate fields such that the self_signed flag doesn't actually get
set, and thus the chain doesn't end immediately). But to pass trust
validation (pkcs7_validate_trust()), either the SignerInfo or one of the
certificates has to actually be signed by a trusted key. Since only
self-signed certificates can be added to the chain, the only way for an
attacker to introduce a trusted signature is to include a self-signed
trusted certificate.
But, when pkcs7_validate_trust_one() reaches that certificate, instead
of trying to verify the signature on that certificate, it will actually
look up the corresponding trusted key, which will succeed, and then try
to verify the *previous* certificate, which will fail. Thus, disaster
is narrowly averted (as far as I could tell).
Fixes: 6c2dc5ae4a ("X.509: Extract signature digest and make self-signed cert checks earlier")
Cc: <stable@vger.kernel.org> # v4.7+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Add test vectors for Speck64-XTS, generated in userspace using C code.
The inputs were borrowed from the AES-XTS test vectors, with key lengths
adjusted.
xts-speck64-neon passes these tests. However, they aren't currently
applicable for the generic XTS template, as that only supports a 128-bit
block size.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Add test vectors for Speck128-XTS, generated in userspace using C code.
The inputs were borrowed from the AES-XTS test vectors.
Both xts(speck128-generic) and xts-speck128-neon pass these tests.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Export the Speck constants and transform context and the ->setkey(),
->encrypt(), and ->decrypt() functions so that they can be reused by the
ARM NEON implementation of Speck-XTS. The generic key expansion code
will be reused because it is not performance-critical and is not
vectorizable, while the generic encryption and decryption functions are
needed as fallbacks and for the XTS tweak encryption.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Add a generic implementation of Speck, including the Speck128 and
Speck64 variants. Speck is a lightweight block cipher that can be much
faster than AES on processors that don't have AES instructions.
We are planning to offer Speck-XTS (probably Speck128/256-XTS) as an
option for dm-crypt and fscrypt on Android, for low-end mobile devices
with older CPUs such as ARMv7 which don't have the Cryptography
Extensions. Currently, such devices are unencrypted because AES is not
fast enough, even when the NEON bit-sliced implementation of AES is
used. Other AES alternatives such as Twofish, Threefish, Camellia,
CAST6, and Serpent aren't fast enough either; it seems that only a
modern ARX cipher can provide sufficient performance on these devices.
This is a replacement for our original proposal
(https://patchwork.kernel.org/patch/10101451/) which was to offer
ChaCha20 for these devices. However, the use of a stream cipher for
disk/file encryption with no space to store nonces would have been much
more insecure than we thought initially, given that it would be used on
top of flash storage as well as potentially on top of F2FS, neither of
which is guaranteed to overwrite data in-place.
Speck has been somewhat controversial due to its origin. Nevertheless,
it has a straightforward design (it's an ARX cipher), and it appears to
be the leading software-optimized lightweight block cipher currently,
with the most cryptanalysis. It's also easy to implement without side
channels, unlike AES. Moreover, we only intend Speck to be used when
the status quo is no encryption, due to AES not being fast enough.
We've also considered a novel length-preserving encryption mode based on
ChaCha20 and Poly1305. While theoretically attractive, such a mode
would be a brand new crypto construction and would be more complicated
and difficult to implement efficiently in comparison to Speck-XTS.
There is confusion about the byte and word orders of Speck, since the
original paper doesn't specify them. But we have implemented it using
the orders the authors recommended in a correspondence with them. The
test vectors are taken from the original paper but were mapped to byte
arrays using the recommended byte and word orders.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The RSA private key for the first form should have
version, prime1, prime2, exponent1, exponent2, coefficient
values 0.
With non-zero values for prime1,2, exponent 1,2 and coefficient
the Intel QAT driver will assume that values are provided for the
private key second form. This will result in signature verification
failures for modules where QAT device is present and the modules
are signed with rsa,sha256.
Cc: <stable@vger.kernel.org>
Signed-off-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Signed-off-by: Conor McLoughlin <conor.mcloughlin@intel.com>
Reviewed-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This module registers crc32 and crc32c algorithms that use the
optional CRC32[bhwd] and CRC32C[bhwd] instructions in MIPSr6 cores.
Signed-off-by: Marcin Nowakowski <marcin.nowakowski@mips.com>
Signed-off-by: James Hogan <jhogan@kernel.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: linux-mips@linux-mips.org
Cc: linux-crypto@vger.kernel.org
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Patchwork: https://patchwork.linux-mips.org/patch/18601/
[jhogan@kernel.org: Add CRYPTO_ALG_OPTIONAL_KEY flag on Eric Biggers'
suggestion, due to commit a208fa8f33 ("crypto: hash - annotate
algorithms taking optional key") in v4.16-rc1]
The crypto engine could actually only enqueue hash and ablkcipher request.
This patch permit it to enqueue any type of crypto_async_request.
Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Tested-by: Fabien Dessenne <fabien.dessenne@st.com>
Tested-by: Fabien Dessenne <fabien.dessenne@st.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
After checking all possible call chains to crypto_report here,
my tool finds that crypto_report is never called in atomic context.
And crypto_report calls crypto_alg_match which calls down_read,
thus it proves again that crypto_report can call functions which may sleep.
Thus GFP_ATOMIC is not necessary, and it can be replaced with GFP_KERNEL.
This is found by a static analysis tool named DCNS written by myself.
Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
After checking all possible call chains to kzalloc here,
my tool finds that this kzalloc is never called in atomic context.
Thus GFP_ATOMIC is not necessary, and it can be replaced with GFP_KERNEL.
This is found by a static analysis tool named DCNS written by myself.
Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
There is no need for ahash_mcryptd_{update,final,finup,digest}(); we
should just call crypto_ahash_*() directly.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Export and import are mandatory in async hash. As drivers were
rewritten, drop empty wrappers and correct init of ahash transformation.
Signed-off-by: Kamil Konieczny <k.konieczny@partner.samsung.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Pull crypto fixes from Herbert Xu:
"This fixes the following issues:
- oversize stack frames on mn10300 in sha3-generic
- warning on old compilers in sha3-generic
- API error in sun4i_ss_prng
- potential dead-lock in sun4i_ss_prng
- null-pointer dereference in sha512-mb
- endless loop when DECO acquire fails in caam
- kernel oops when hashing empty message in talitos"
* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
crypto: sun4i_ss_prng - convert lock to _bh in sun4i_ss_prng_generate
crypto: sun4i_ss_prng - fix return value of sun4i_ss_prng_generate
crypto: caam - fix endless loop when DECO acquire fails
crypto: sha3-generic - Use __optimize to support old compilers
compiler-gcc.h: __nostackprotector needs gcc-4.4 and up
compiler-gcc.h: Introduce __optimize function attribute
crypto: sha3-generic - deal with oversize stack frames
crypto: talitos - fix Kernel Oops on hashing an empty file
crypto: sha512-mb - initialize pending lengths correctly
This is the mindless scripted replacement of kernel use of POLL*
variables as described by Al, done by this script:
for V in IN OUT PRI ERR RDNORM RDBAND WRNORM WRBAND HUP RDHUP NVAL MSG; do
L=`git grep -l -w POLL$V | grep -v '^t' | grep -v /um/ | grep -v '^sa' | grep -v '/poll.h$'|grep -v '^D'`
for f in $L; do sed -i "-es/^\([^\"]*\)\(\<POLL$V\>\)/\\1E\\2/" $f; done
done
with de-mangling cleanups yet to come.
NOTE! On almost all architectures, the EPOLL* constants have the same
values as the POLL* constants do. But they keyword here is "almost".
For various bad reasons they aren't the same, and epoll() doesn't
actually work quite correctly in some cases due to this on Sparc et al.
The next patch from Al will sort out the final differences, and we
should be all done.
Scripted-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
As reported by kbuild test robot, the optimized SHA3 C implementation
compiles to mn10300 code that uses a disproportionate amount of stack
space, i.e.,
crypto/sha3_generic.c: In function 'keccakf':
crypto/sha3_generic.c:147:1: warning: the frame size of 1232 bytes is larger than 1024 bytes [-Wframe-larger-than=]
As kindly diagnosed by Arnd, this does not only occur when building for
the mn10300 architecture (which is what the report was about) but also
for h8300, and builds for other 32-bit architectures show an increase in
stack space utilization as well.
Given that SHA3 operates on 64-bit quantities, and keeps a state matrix
of 25 64-bit words, it is not surprising that 32-bit architectures with
few general purpose registers are impacted the most by this, and it is
therefore reasonable to implement a workaround that distinguishes between
32-bit and 64-bit architectures.
Arnd figured out that taking the round calculation out of the loop, and
inlining it explicitly but only on 64-bit architectures preserves most
of the performance gain achieved by the rewrite, and also gets rid of
the excessive use of stack space.
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Suggested-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Pull crypto updates from Herbert Xu:
"API:
- Enforce the setting of keys for keyed aead/hash/skcipher
algorithms.
- Add multibuf speed tests in tcrypt.
Algorithms:
- Improve performance of sha3-generic.
- Add native sha512 support on arm64.
- Add v8.2 Crypto Extentions version of sha3/sm3 on arm64.
- Avoid hmac nesting by requiring underlying algorithm to be unkeyed.
- Add cryptd_max_cpu_qlen module parameter to cryptd.
Drivers:
- Add support for EIP97 engine in inside-secure.
- Add inline IPsec support to chelsio.
- Add RevB core support to crypto4xx.
- Fix AEAD ICV check in crypto4xx.
- Add stm32 crypto driver.
- Add support for BCM63xx platforms in bcm2835 and remove bcm63xx.
- Add Derived Key Protocol (DKP) support in caam.
- Add Samsung Exynos True RNG driver.
- Add support for Exynos5250+ SoCs in exynos PRNG driver"
* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6: (166 commits)
crypto: picoxcell - Fix error handling in spacc_probe()
crypto: arm64/sha512 - fix/improve new v8.2 Crypto Extensions code
crypto: arm64/sm3 - new v8.2 Crypto Extensions implementation
crypto: arm64/sha3 - new v8.2 Crypto Extensions implementation
crypto: testmgr - add new testcases for sha3
crypto: sha3-generic - export init/update/final routines
crypto: sha3-generic - simplify code
crypto: sha3-generic - rewrite KECCAK transform to help the compiler optimize
crypto: sha3-generic - fixes for alignment and big endian operation
crypto: aesni - handle zero length dst buffer
crypto: artpec6 - remove select on non-existing CRYPTO_SHA384
hwrng: bcm2835 - Remove redundant dev_err call in bcm2835_rng_probe()
crypto: stm32 - remove redundant dev_err call in stm32_cryp_probe()
crypto: axis - remove unnecessary platform_get_resource() error check
crypto: testmgr - test misuse of result in ahash
crypto: inside-secure - make function safexcel_try_push_requests static
crypto: aes-generic - fix aes-generic regression on powerpc
crypto: chelsio - Fix indentation warning
crypto: arm64/sha1-ce - get rid of literal pool
crypto: arm64/sha2-ce - move the round constant table to .rodata section
...
Pull poll annotations from Al Viro:
"This introduces a __bitwise type for POLL### bitmap, and propagates
the annotations through the tree. Most of that stuff is as simple as
'make ->poll() instances return __poll_t and do the same to local
variables used to hold the future return value'.
Some of the obvious brainos found in process are fixed (e.g. POLLIN
misspelled as POLL_IN). At that point the amount of sparse warnings is
low and most of them are for genuine bugs - e.g. ->poll() instance
deciding to return -EINVAL instead of a bitmap. I hadn't touched those
in this series - it's large enough as it is.
Another problem it has caught was eventpoll() ABI mess; select.c and
eventpoll.c assumed that corresponding POLL### and EPOLL### were
equal. That's true for some, but not all of them - EPOLL### are
arch-independent, but POLL### are not.
The last commit in this series separates userland POLL### values from
the (now arch-independent) kernel-side ones, converting between them
in the few places where they are copied to/from userland. AFAICS, this
is the least disruptive fix preserving poll(2) ABI and making epoll()
work on all architectures.
As it is, it's simply broken on sparc - try to give it EPOLLWRNORM and
it will trigger only on what would've triggered EPOLLWRBAND on other
architectures. EPOLLWRBAND and EPOLLRDHUP, OTOH, are never triggered
at all on sparc. With this patch they should work consistently on all
architectures"
* 'misc.poll' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (37 commits)
make kernel-side POLL... arch-independent
eventpoll: no need to mask the result of epi_item_poll() again
eventpoll: constify struct epoll_event pointers
debugging printk in sg_poll() uses %x to print POLL... bitmap
annotate poll(2) guts
9p: untangle ->poll() mess
->si_band gets POLL... bitmap stored into a user-visible long field
ring_buffer_poll_wait() return value used as return value of ->poll()
the rest of drivers/*: annotate ->poll() instances
media: annotate ->poll() instances
fs: annotate ->poll() instances
ipc, kernel, mm: annotate ->poll() instances
net: annotate ->poll() instances
apparmor: annotate ->poll() instances
tomoyo: annotate ->poll() instances
sound: annotate ->poll() instances
acpi: annotate ->poll() instances
crypto: annotate ->poll() instances
block: annotate ->poll() instances
x86: annotate ->poll() instances
...
Pull block updates from Jens Axboe:
"This is the main pull request for block IO related changes for the
4.16 kernel. Nothing major in this pull request, but a good amount of
improvements and fixes all over the map. This contains:
- BFQ improvements, fixes, and cleanups from Angelo, Chiara, and
Paolo.
- Support for SMR zones for deadline and mq-deadline from Damien and
Christoph.
- Set of fixes for bcache by way of Michael Lyle, including fixes
from himself, Kent, Rui, Tang, and Coly.
- Series from Matias for lightnvm with fixes from Hans Holmberg,
Javier, and Matias. Mostly centered around pblk, and the removing
rrpc 1.2 in preparation for supporting 2.0.
- A couple of NVMe pull requests from Christoph. Nothing major in
here, just fixes and cleanups, and support for command tracing from
Johannes.
- Support for blk-throttle for tracking reads and writes separately.
From Joseph Qi. A few cleanups/fixes also for blk-throttle from
Weiping.
- Series from Mike Snitzer that enables dm to register its queue more
logically, something that's alwways been problematic on dm since
it's a stacked device.
- Series from Ming cleaning up some of the bio accessor use, in
preparation for supporting multipage bvecs.
- Various fixes from Ming closing up holes around queue mapping and
quiescing.
- BSD partition fix from Richard Narron, fixing a problem where we
can't mount newer (10/11) FreeBSD partitions.
- Series from Tejun reworking blk-mq timeout handling. The previous
scheme relied on atomic bits, but it had races where we would think
a request had timed out if it to reused at the wrong time.
- null_blk now supports faking timeouts, to enable us to better
exercise and test that functionality separately. From me.
- Kill the separate atomic poll bit in the request struct. After
this, we don't use the atomic bits on blk-mq anymore at all. From
me.
- sgl_alloc/free helpers from Bart.
- Heavily contended tag case scalability improvement from me.
- Various little fixes and cleanups from Arnd, Bart, Corentin,
Douglas, Eryu, Goldwyn, and myself"
* 'for-4.16/block' of git://git.kernel.dk/linux-block: (186 commits)
block: remove smart1,2.h
nvme: add tracepoint for nvme_complete_rq
nvme: add tracepoint for nvme_setup_cmd
nvme-pci: introduce RECONNECTING state to mark initializing procedure
nvme-rdma: remove redundant boolean for inline_data
nvme: don't free uuid pointer before printing it
nvme-pci: Suspend queues after deleting them
bsg: use pr_debug instead of hand crafted macros
blk-mq-debugfs: don't allow write on attributes with seq_operations set
nvme-pci: Fix queue double allocations
block: Set BIO_TRACE_COMPLETION on new bio during split
blk-throttle: use queue_is_rq_based
block: Remove kblockd_schedule_delayed_work{,_on}()
blk-mq: Avoid that blk_mq_delay_run_hw_queue() introduces unintended delays
blk-mq: Rename blk_mq_request_direct_issue() into blk_mq_request_issue_directly()
lib/scatterlist: Fix chaining support in sgl_alloc_order()
blk-throttle: track read and write request individually
block: add bdev_read_only() checks to common helpers
block: fail op_is_write() requests to read-only partitions
blk-throttle: export io_serviced_recursive, io_service_bytes_recursive
...
All current SHA3 test cases are smaller than the SHA3 block size, which
means not all code paths are being exercised. So add a new test case to
each variant, and make one of the existing test cases chunked.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
To allow accelerated implementations to fall back to the generic
routines, e.g., in contexts where a SIMD based implementation is
not allowed to run, expose the generic SHA3 init/update/final
routines to other modules.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In preparation of exposing the generic SHA3 implementation to other
versions as a fallback, simplify the code, and remove an inconsistency
in the output handling (endian swabbing rsizw words of state before
writing the output does not make sense)
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The way the KECCAK transform is currently coded involves many references
into the state array using indexes that are calculated at runtime using
simple but non-trivial arithmetic. This forces the compiler to treat the
state matrix as an array in memory rather than keep it in registers,
which results in poor performance.
So instead, let's rephrase the algorithm using fixed array indexes only.
This helps the compiler keep the state matrix in registers, resulting
in the following speedup (SHA3-256 performance in cycles per byte):
before after speedup
Intel Core i7 @ 2.0 GHz (2.9 turbo) 100.6 35.7 2.8x
Cortex-A57 @ 2.0 GHz (64-bit mode) 101.6 12.7 8.0x
Cortex-A53 @ 1.0 GHz 224.4 15.8 14.2x
Cortex-A57 @ 2.0 GHz (32-bit mode) 201.8 63.0 3.2x
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Ensure that the input is byte swabbed before injecting it into the
SHA3 transform. Use the get_unaligned() accessor for this so that
we don't perform unaligned access inadvertently on architectures
that do not support that.
Cc: <stable@vger.kernel.org>
Fixes: 53964b9ee6 ("crypto: sha3 - Add SHA-3 hash algorithm")
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Async hash operations can use result pointer in final/finup/digest,
but not in init/update/export/import, so test it for misuse.
Signed-off-by: Kamil Konieczny <k.konieczny@partner.samsung.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
My last bugfix added -Os on the command line, which unfortunately caused
a build regression on powerpc in some configurations.
I've done some more analysis of the original problem and found slightly
different workaround that avoids this regression and also results in
better performance on gcc-7.0: -fcode-hoisting is an optimization step
that got added in gcc-7 and that for all gcc-7 versions causes worse
performance.
This disables -fcode-hoisting on all compilers that understand the option.
For gcc-7.1 and 7.2 I found the same performance as my previous patch
(using -Os), in gcc-7.0 it was even better. On gcc-8 I could see no
change in performance from this patch. In theory, code hoisting should
not be able make things better for the AES cipher, so leaving it
disabled for gcc-8 only serves to simplify the Makefile change.
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Link: https://www.mail-archive.com/linux-crypto@vger.kernel.org/msg30418.html
Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83356
Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83651
Fixes: 148b974dee ("crypto: aes-generic - build with -Os on gcc-7+")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Pull crypto fix from Herbert Xu:
"This fixes a NULL pointer dereference in crypto_remove_spawns that can
be triggered through af_alg"
* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
crypto: algapi - fix NULL dereference in crypto_remove_spawns()
Convert salsa20-asm from the deprecated "blkcipher" API to the
"skcipher" API, in the process fixing it up to use the generic helpers.
This allows removing the salsa20_keysetup() and salsa20_ivsetup()
assembly functions, which aren't performance critical; the C versions do
just fine.
This also fixes the same bug that salsa20-generic had, where the state
array was being maintained directly in the transform context rather than
on the stack or in the request context. Thus, if multiple threads used
the same Salsa20 transform concurrently they produced the wrong results.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Export the Salsa20 constants, transform context, and initialization
functions so that they can be reused by the x86 implementation.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Convert salsa20-generic from the deprecated "blkcipher" API to the
"skcipher" API, in the process fixing it up to be thread-safe (as the
crypto API expects) by maintaining each request's state separately from
the transform context.
Also remove the unnecessary cra_alignmask and tighten validation of the
key size by accepting only 16 or 32 bytes, not anything in between.
These changes bring the code close to the way chacha20-generic does
things, so hopefully it will be easier to maintain in the future.
However, the way Salsa20 interprets the IV is still slightly different;
that was not changed.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
While testing other changes, I discovered that gcc-7.2.1 produces badly
optimized code for aes_encrypt/aes_decrypt. This is especially true when
CONFIG_UBSAN_SANITIZE_ALL is enabled, where it leads to extremely
large stack usage that in turn might cause kernel stack overflows:
crypto/aes_generic.c: In function 'aes_encrypt':
crypto/aes_generic.c:1371:1: warning: the frame size of 4880 bytes is larger than 2048 bytes [-Wframe-larger-than=]
crypto/aes_generic.c: In function 'aes_decrypt':
crypto/aes_generic.c:1441:1: warning: the frame size of 4864 bytes is larger than 2048 bytes [-Wframe-larger-than=]
I verified that this problem exists on all architectures that are
supported by gcc-7.2, though arm64 in particular is less affected than
the others. I also found that gcc-7.1 and gcc-8 do not show the extreme
stack usage but still produce worse code than earlier versions for this
file, apparently because of optimization passes that generally provide
a substantial improvement in object code quality but understandably fail
to find any shortcuts in the AES algorithm.
Possible workarounds include
a) disabling -ftree-pre and -ftree-sra optimizations, this was an earlier
patch I tried, which reliably fixed the stack usage, but caused a
serious performance regression in some versions, as later testing
found.
b) disabling UBSAN on this file or all ciphers, as suggested by Ard
Biesheuvel. This would lead to massively better crypto performance in
UBSAN-enabled kernels and avoid the stack usage, but there is a concern
over whether we should exclude arbitrary files from UBSAN at all.
c) Forcing the optimization level in a different way. Similar to a),
but rather than deselecting specific optimization stages,
this now uses "gcc -Os" for this file, regardless of the
CONFIG_CC_OPTIMIZE_FOR_PERFORMANCE/SIZE option. This is a reliable
workaround for the stack consumption on all architecture, and I've
retested the performance results now on x86, cycles/byte (lower is
better) for cbc(aes-generic) with 256 bit keys:
-O2 -Os
gcc-6.3.1 14.9 15.1
gcc-7.0.1 14.7 15.3
gcc-7.1.1 15.3 14.7
gcc-7.2.1 16.8 15.9
gcc-8.0.0 15.5 15.6
This implements the option c) by enabling forcing -Os on all compiler
versions starting with gcc-7.1. As a workaround for PR83356, it would
only be needed for gcc-7.2+ with UBSAN enabled, but since it also shows
better performance on gcc-7.1 without UBSAN, it seems appropriate to
use the faster version here as well.
Side note: during testing, I also played with the AES code in libressl,
which had a similar performance regression from gcc-6 to gcc-7.2,
but was three times slower overall. It might be interesting to
investigate that further and possibly port the Linux implementation
into that.
Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83356
Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83651
Cc: Richard Biener <rguenther@suse.de>
Cc: Jakub Jelinek <jakub@gcc.gnu.org>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Similar to what was done for the hash API, update the AEAD API to track
whether each transform has been keyed, and reject encryption/decryption
if a key is needed but one hasn't been set.
This isn't quite as important as the equivalent fix for the hash API
because AEADs always require a key, so are unlikely to be used without
one. Still, tracking the key will prevent accidental unkeyed use.
algif_aead also had to track the key anyway, so the new flag replaces
that and slightly simplifies the algif_aead implementation.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Similar to what was done for the hash API, update the skcipher API to
track whether each transform has been keyed, and reject
encryption/decryption if a key is needed but one hasn't been set.
This isn't as important as the equivalent fix for the hash API because
symmetric ciphers almost always require a key (the "null cipher" is the
only exception), so are unlikely to be used without one. Still,
tracking the key will prevent accidental unkeyed use. algif_skcipher
also had to track the key anyway, so the new flag replaces that and
simplifies the algif_skcipher implementation.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Now that the crypto API prevents a keyed hash from being used without
setting the key, there's no need for GHASH to do this check itself.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently, almost none of the keyed hash algorithms check whether a key
has been set before proceeding. Some algorithms are okay with this and
will effectively just use a key of all 0's or some other bogus default.
However, others will severely break, as demonstrated using
"hmac(sha3-512-generic)", the unkeyed use of which causes a kernel crash
via a (potentially exploitable) stack buffer overflow.
A while ago, this problem was solved for AF_ALG by pairing each hash
transform with a 'has_key' bool. However, there are still other places
in the kernel where userspace can specify an arbitrary hash algorithm by
name, and the kernel uses it as unkeyed hash without checking whether it
is really unkeyed. Examples of this include:
- KEYCTL_DH_COMPUTE, via the KDF extension
- dm-verity
- dm-crypt, via the ESSIV support
- dm-integrity, via the "internal hash" mode with no key given
- drbd (Distributed Replicated Block Device)
This bug is especially bad for KEYCTL_DH_COMPUTE as that requires no
privileges to call.
Fix the bug for all users by adding a flag CRYPTO_TFM_NEED_KEY to the
->crt_flags of each hash transform that indicates whether the transform
still needs to be keyed or not. Then, make the hash init, import, and
digest functions return -ENOKEY if the key is still needed.
The new flag also replaces the 'has_key' bool which algif_hash was
previously using, thereby simplifying the algif_hash implementation.
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
We need to consistently enforce that keyed hashes cannot be used without
setting the key. To do this we need a reliable way to determine whether
a given hash algorithm is keyed or not. AF_ALG currently does this by
checking for the presence of a ->setkey() method. However, this is
actually slightly broken because the CRC-32 algorithms implement
->setkey() but can also be used without a key. (The CRC-32 "key" is not
actually a cryptographic key but rather represents the initial state.
If not overridden, then a default initial state is used.)
Prepare to fix this by introducing a flag CRYPTO_ALG_OPTIONAL_KEY which
indicates that the algorithm has a ->setkey() method, but it is not
required to be called. Then set it on all the CRC-32 algorithms.
The same also applies to the Adler-32 implementation in Lustre.
Also, the cryptd and mcryptd templates have to pass through the flag
from their underlying algorithm.
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Since Poly1305 requires a nonce per invocation, the Linux kernel
implementations of Poly1305 don't use the crypto API's keying mechanism
and instead expect the key and nonce as the first 32 bytes of the data.
But ->setkey() is still defined as a stub returning an error code. This
prevents Poly1305 from being used through AF_ALG and will also break it
completely once we start enforcing that all crypto API users (not just
AF_ALG) call ->setkey() if present.
Fix it by removing crypto_poly1305_setkey(), leaving ->setkey as NULL.
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
When the mcryptd template is used to wrap an unkeyed hash algorithm,
don't install a ->setkey() method to the mcryptd instance. This change
is necessary for mcryptd to keep working with unkeyed hash algorithms
once we start enforcing that ->setkey() is called when present.
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
When the cryptd template is used to wrap an unkeyed hash algorithm,
don't install a ->setkey() method to the cryptd instance. This change
is necessary for cryptd to keep working with unkeyed hash algorithms
once we start enforcing that ->setkey() is called when present.
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Templates that use an shash spawn can use crypto_shash_alg_has_setkey()
to determine whether the underlying algorithm requires a key or not.
But there was no corresponding function for ahash spawns. Add it.
Note that the new function actually has to support both shash and ahash
algorithms, since the ahash API can be used with either.
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
There seems to be a cut-n-paste bug with the name of the buffer being
free'd, xoutbuf should be used instead of axbuf.
Detected by CoverityScan, CID#1463420 ("Copy-paste error")
Fixes: 427988d981 ("crypto: tcrypt - add multibuf aead speed test")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Trivial fix to spelling mistakes in pr_err error message text.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The user space interface allows specifying the type and mask field used
to allocate the cipher. Only a subset of the possible flags are intended
for user space. Therefore, white-list the allowed flags.
In case the user space caller uses at least one non-allowed flag, EINVAL
is returned.
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
When char is signed, storing the values 0xba (186) and 0xad (173) in the
`guard` array produces signed overflow. Change the type of `guard` to
static unsigned char to correct undefined behavior and reduce function
stack usage.
Signed-off-by: Joey Pabalinas <joeypabalinas@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use the sgl_alloc() and sgl_free() functions instead of open coding
these functions.
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Pull crypto fixes from Herbert Xu:
"This fixes the following issues:
- racy use of ctx->rcvused in af_alg
- algif_aead crash in chacha20poly1305
- freeing bogus pointer in pcrypt
- build error on MIPS in mpi
- memory leak in inside-secure
- memory overwrite in inside-secure
- NULL pointer dereference in inside-secure
- state corruption in inside-secure
- build error without CRYPTO_GF128MUL in chelsio
- use after free in n2"
* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
crypto: inside-secure - do not use areq->result for partial results
crypto: inside-secure - fix request allocations in invalidation path
crypto: inside-secure - free requests even if their handling failed
crypto: inside-secure - per request invalidation
lib/mpi: Fix umul_ppmm() for MIPS64r6
crypto: pcrypt - fix freeing pcrypt instances
crypto: n2 - cure use after free
crypto: af_alg - Fix race around ctx->rcvused by making it atomic_t
crypto: chacha20poly1305 - validate the digest size
crypto: chelsio - select CRYPTO_GF128MUL
Now that nothing in poly1305-generic assumes any special alignment,
remove the cra_alignmask so that the crypto API does not have to
unnecessarily align the buffers.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently the only part of poly1305-generic which is assuming special
alignment is the part where the final digest is written. Switch this
over to the unaligned access macros so that we'll be able to remove the
cra_alignmask.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
There is a message posted to the crypto notifier chain when an algorithm
is unregistered, and when a template is registered or unregistered. But
nothing is listening for those messages; currently there are only
listeners for the algorithm request and registration messages.
Get rid of these unused notifications for now.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Reference counters should use refcount_t rather than atomic_t, since the
refcount_t implementation can prevent overflows, reducing the
exploitability of reference leak bugs. crypto_alg.cra_refcount is a
reference counter with the usual semantics, so switch it over to
refcount_t.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
syzkaller triggered a NULL pointer dereference in crypto_remove_spawns()
via a program that repeatedly and concurrently requests AEADs
"authenc(cmac(des3_ede-asm),pcbc-aes-aesni)" and hashes "cmac(des3_ede)"
through AF_ALG, where the hashes are requested as "untested"
(CRYPTO_ALG_TESTED is set in ->salg_mask but clear in ->salg_feat; this
causes the template to be instantiated for every request).
Although AF_ALG users really shouldn't be able to request an "untested"
algorithm, the NULL pointer dereference is actually caused by a
longstanding race condition where crypto_remove_spawns() can encounter
an instance which has had spawn(s) "grabbed" but hasn't yet been
registered, resulting in ->cra_users still being NULL.
We probably should properly initialize ->cra_users earlier, but that
would require updating many templates individually. For now just fix
the bug in a simple way that can easily be backported: make
crypto_remove_spawns() treat a NULL ->cra_users list as empty.
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The performance of some aead tfm providers is affected by
the amount of parallelism possible with the processing.
Introduce an async aead concurrent multiple buffer
processing speed test to be able to test performance of such
tfm providers.
Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The performance of some skcipher tfm providers is affected by
the amount of parallelism possible with the processing.
Introduce an async skcipher concurrent multiple buffer
processing speed test to be able to test performance of such
tfm providers.
Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The multi buffer concurrent requests ahash speed test only
supported the cycles mode. Add support for the so called
jiffies mode that test performance of bytes/sec.
We only add support for digest mode at the moment.
Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
For multiple buffers speed tests, the number of buffers, or
requests, used actually sets the level of parallelism a tfm
provider may utilize to hide latency. The existing number
(of 8) is good for some software based providers but not
enough for many HW providers with deep FIFOs.
Add a module parameter that allows setting the number of
multiple buffers/requests used, leaving the default at 8.
Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The AEAD speed test pretended to support decryption, however that support
was broken as decryption requires a valid auth field which the test did
not provide.
Fix this by running the encryption path once with inout/output sgls
switched to calculate the auth field prior to performing decryption
speed tests.
Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The multi buffer ahash speed test was allocating multiple
buffers for use with the multiple outstanding requests
it was starting but never actually using them (except
to free them), instead using a different single statically
allocated buffer for all requests.
Fix this by actually using the allocated buffers for the test.
It is noted that it may seem tempting to instead remove the
allocation and free of the multiple buffers and leave things as
they are since this is a hash test where the input is read
only. However, after consideration I believe that multiple
buffers better reflect real life scenario with regard
to data cache and TLB behaviours etc.
Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Pull crypto fixes from Herbert Xu:
"This fixes the following issues:
- fix chacha20 crash on zero-length input due to unset IV
- fix potential race conditions in mcryptd with spinlock
- only wait once at top of algif recvmsg to avoid inconsistencies
- fix potential use-after-free in algif_aead/algif_skcipher"
* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
crypto: af_alg - fix race accessing cipher request
crypto: mcryptd - protect the per-CPU queue with a lock
crypto: af_alg - wait for data at beginning of recvmsg
crypto: skcipher - set walk.iv for zero-length inputs
This patch remove two unused variable and some dead "code" using it.
Fixes: 92932d03c2 ("crypto: seqiv - Remove AEAD compatibility code")
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch remove two unused variable and some dead "code" using it.
Fixes: 66008d4230 ("crypto: echainiv - Remove AEAD compatibility code")
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The comment in gf128mul_x8_ble() was copy-and-pasted from gf128mul.h and
makes no sense in the new context. Remove it.
Cc: Harsh Jain <harsh@chelsio.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Since commit 499a66e6b6 ("crypto: null - Remove default null
blkcipher"), crypto_get_default_null_skcipher2() and
crypto_put_default_null_skcipher2() are the same as their non-2
equivalents. So switch callers of the "2" versions over to the original
versions and remove the "2" versions.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
crypto_larval_lookup() is not used outside of crypto/api.c, so unexport
it and mark it 'static'.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
pcrypt is using the old way of freeing instances, where the ->free()
method specified in the 'struct crypto_template' is passed a pointer to
the 'struct crypto_instance'. But the crypto_instance is being
kfree()'d directly, which is incorrect because the memory was actually
allocated as an aead_instance, which contains the crypto_instance at a
nonzero offset. Thus, the wrong pointer was being kfree()'d.
Fix it by switching to the new way to free aead_instance's where the
->free() method is specified in the aead_instance itself.
Reported-by: syzbot <syzkaller@googlegroups.com>
Fixes: 0496f56065 ("crypto: pcrypt - Add support for new AEAD interface")
Cc: <stable@vger.kernel.org> # v4.2+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This variable was increased and decreased without any protection.
Result was an occasional misscount and negative wrap around resulting
in false resource allocation failures.
Fixes: 7d2c3f54e6 ("crypto: af_alg - remove locking in async callback")
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Reviewed-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
If the rfc7539 template was instantiated with a hash algorithm with
digest size larger than 16 bytes (POLY1305_DIGEST_SIZE), then the digest
overran the 'tag' buffer in 'struct chachapoly_req_ctx', corrupting the
subsequent memory, including 'cryptlen'. This caused a crash during
crypto_skcipher_decrypt().
Fix it by, when instantiating the template, requiring that the
underlying hash algorithm has the digest size expected for Poly1305.
Reproducer:
#include <linux/if_alg.h>
#include <sys/socket.h>
#include <unistd.h>
int main()
{
int algfd, reqfd;
struct sockaddr_alg addr = {
.salg_type = "aead",
.salg_name = "rfc7539(chacha20,sha256)",
};
unsigned char buf[32] = { 0 };
algfd = socket(AF_ALG, SOCK_SEQPACKET, 0);
bind(algfd, (void *)&addr, sizeof(addr));
setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, sizeof(buf));
reqfd = accept(algfd, 0, 0);
write(reqfd, buf, 16);
read(reqfd, buf, 16);
}
Reported-by: syzbot <syzkaller@googlegroups.com>
Fixes: 71ebc4d1b2 ("crypto: chacha20poly1305 - Add a ChaCha20-Poly1305 AEAD construction, RFC7539")
Cc: <stable@vger.kernel.org> # v4.2+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Pull crypto fixes from Herbert Xu:
"This push fixes the following issues:
- buffer overread in RSA
- potential use after free in algif_aead.
- error path null pointer dereference in af_alg
- forbid combinations such as hmac(hmac(sha3)) which may crash
- crash in salsa20 due to incorrect API usage"
* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
crypto: salsa20 - fix blkcipher_walk API usage
crypto: hmac - require that the underlying hash algorithm is unkeyed
crypto: af_alg - fix NULL pointer dereference in
crypto: algif_aead - fix reference counting of null skcipher
crypto: rsa - fix buffer overread when stripping leading zeroes
The cryptd_max_cpu_qlen module parameter is local to the source and does
not need to be in global scope, so make it static.
Cleans up sparse warning:
crypto/cryptd.c:35:14: warning: symbol 'cryptd_max_cpu_qlen' was not
declared. Should it be static?
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
When invoking an asynchronous cipher operation, the invocation of the
callback may be performed before the subsequent operations in the
initial code path are invoked. The callback deletes the cipher request
data structure which implies that after the invocation of the
asynchronous cipher operation, this data structure must not be accessed
any more.
The setting of the return code size with the request data structure must
therefore be moved before the invocation of the asynchronous cipher
operation.
Fixes: e870456d8e ("crypto: algif_skcipher - overhaul memory management")
Fixes: d887c52d6a ("crypto: algif_aead - overhaul memory management")
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: <stable@vger.kernel.org> # v4.14+
Signed-off-by: Stephan Mueller <smueller@chronox.de>
Acked-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
mcryptd_enqueue_request() grabs the per-CPU queue struct and protects
access to it with disabled preemption. Then it schedules a worker on the
same CPU. The worker in mcryptd_queue_worker() guards access to the same
per-CPU variable with disabled preemption.
If we take CPU-hotplug into account then it is possible that between
queue_work_on() and the actual invocation of the worker the CPU goes
down and the worker will be scheduled on _another_ CPU. And here the
preempt_disable() protection does not work anymore. The easiest thing is
to add a spin_lock() to guard access to the list.
Another detail: mcryptd_queue_worker() is not processing more than
MCRYPTD_BATCH invocation in a row. If there are still items left, then
it will invoke queue_work() to proceed with more later. *I* would
suggest to simply drop that check because it does not use a system
workqueue and the workqueue is already marked as "CPU_INTENSIVE". And if
preemption is required then the scheduler should do it.
However if queue_work() is used then the work item is marked as CPU
unbound. That means it will try to run on the local CPU but it may run
on another CPU as well. Especially with CONFIG_DEBUG_WQ_FORCE_RR_CPU=y.
Again, the preempt_disable() won't work here but lock which was
introduced will help.
In order to keep work-item on the local CPU (and avoid RR) I changed it
to queue_work_on().
Cc: stable@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The wait for data is a non-atomic operation that can sleep and therefore
potentially release the socket lock. The release of the socket lock
allows another thread to modify the context data structure. The waiting
operation for new data therefore must be called at the beginning of
recvmsg. This prevents a race condition where checks of the members of
the context data structure are performed by recvmsg while there is a
potential for modification of these values.
Fixes: e870456d8e ("crypto: algif_skcipher - overhaul memory management")
Fixes: d887c52d6a ("crypto: algif_aead - overhaul memory management")
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: <stable@vger.kernel.org> # v4.14+
Signed-off-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
All the ChaCha20 algorithms as well as the ARM bit-sliced AES-XTS
algorithms call skcipher_walk_virt(), then access the IV (walk.iv)
before checking whether any bytes need to be processed (walk.nbytes).
But if the input is empty, then skcipher_walk_virt() doesn't set the IV,
and the algorithms crash trying to use the uninitialized IV pointer.
Fix it by setting the IV earlier in skcipher_walk_virt(). Also fix it
for the AEAD walk functions.
This isn't a perfect solution because we can't actually align the IV to
->cra_alignmask unless there are bytes to process, for one because the
temporary buffer for the aligned IV is freed by skcipher_walk_done(),
which is only called when there are bytes to process. Thus, algorithms
that require aligned IVs will still need to avoid accessing the IV when
walk.nbytes == 0. Still, many algorithms/architectures are fine with
IVs having any alignment, and even for those that aren't, a misaligned
pointer bug is much less severe than an uninitialized pointer bug.
This change also matches the behavior of the older blkcipher_walk API.
Fixes: 0cabf2af6f ("crypto: skcipher - Fix crash on zero-length input")
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: <stable@vger.kernel.org> # v4.14+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
->pkey_algo used to be an enum, but was changed to a string by commit
4e8ae72a75 ("X.509: Make algo identifiers text instead of enum"). But
two comparisons were not updated. Fix them to use strcmp().
This bug broke signature verification in certain configurations,
depending on whether the string constants were deduplicated or not.
Fixes: 4e8ae72a75 ("X.509: Make algo identifiers text instead of enum")
Cc: <stable@vger.kernel.org> # v4.6+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Use crypto_shash_digest() instead of crypto_shash_init() followed by
crypto_shash_finup(). (For simplicity only; they are equivalent.)
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
In public_key_verify_signature(), if akcipher_request_alloc() fails, we
return -ENOMEM. But that error code was set 25 lines above, and by
accident someone could easily insert new code in between that assigns to
'ret', which would introduce a signature verification bypass. Make the
code clearer by moving the -ENOMEM down to where it is used.
Additionally, the callers of public_key_verify_signature() only consider
a negative return value to be an error. This means that if any positive
return value is accidentally introduced deeper in the call stack (e.g.
'return EBADMSG' instead of 'return -EBADMSG' somewhere in RSA),
signature verification will be bypassed. Make things more robust by
having public_key_verify_signature() warn about positive errors and
translate them into -EINVAL.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Use crypto_shash_digest() instead of crypto_shash_init() followed by
crypto_shash_finup(). (For simplicity only; they are equivalent.)
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
pkcs7_validate_trust_one() used 'x509->next == x509' to identify a
self-signed certificate. That's wrong; ->next is simply the link in the
linked list of certificates in the PKCS#7 message. It should be
checking ->signer instead. Fix it.
Fortunately this didn't actually matter because when we re-visited
'x509' on the next iteration via 'x509->signer', it was already seen and
not verified, so we returned -ENOKEY anyway.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <james.l.morris@oracle.com>
If pkcs7_check_authattrs() returns an error code, we should pass that
error code on, rather than using ENOMEM.
Fixes: 99db443506 ("PKCS#7: Appropriately restrict authenticated attributes and content type")
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <james.l.morris@oracle.com>
Adding a specially crafted X.509 certificate whose subjectPublicKey
ASN.1 value is zero-length caused x509_extract_key_data() to set the
public key size to SIZE_MAX, as it subtracted the nonexistent BIT STRING
metadata byte. Then, x509_cert_parse() called kmemdup() with that bogus
size, triggering the WARN_ON_ONCE() in kmalloc_slab().
This appears to be harmless, but it still must be fixed since WARNs are
never supposed to be user-triggerable.
Fix it by updating x509_cert_parse() to validate that the value has a
BIT STRING metadata byte, and that the byte is 0 which indicates that
the number of bits in the bitstring is a multiple of 8.
It would be nice to handle the metadata byte in asn1_ber_decoder()
instead. But that would be tricky because in the general case a BIT
STRING could be implicitly tagged, and/or could legitimately have a
length that is not a whole number of bytes.
Here was the WARN (cleaned up slightly):
WARNING: CPU: 1 PID: 202 at mm/slab_common.c:971 kmalloc_slab+0x5d/0x70 mm/slab_common.c:971
Modules linked in:
CPU: 1 PID: 202 Comm: keyctl Tainted: G B 4.14.0-09238-g1d3b78bbc6e9 #26
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.11.0-20171110_100015-anatol 04/01/2014
task: ffff880033014180 task.stack: ffff8800305c8000
Call Trace:
__do_kmalloc mm/slab.c:3706 [inline]
__kmalloc_track_caller+0x22/0x2e0 mm/slab.c:3726
kmemdup+0x17/0x40 mm/util.c:118
kmemdup include/linux/string.h:414 [inline]
x509_cert_parse+0x2cb/0x620 crypto/asymmetric_keys/x509_cert_parser.c:106
x509_key_preparse+0x61/0x750 crypto/asymmetric_keys/x509_public_key.c:174
asymmetric_key_preparse+0xa4/0x150 crypto/asymmetric_keys/asymmetric_type.c:388
key_create_or_update+0x4d4/0x10a0 security/keys/key.c:850
SYSC_add_key security/keys/keyctl.c:122 [inline]
SyS_add_key+0xe8/0x290 security/keys/keyctl.c:62
entry_SYSCALL_64_fastpath+0x1f/0x96
Fixes: 42d5ec27f8 ("X.509: Add an ASN.1 decoder")
Cc: <stable@vger.kernel.org> # v3.7+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <james.l.morris@oracle.com>
When chacha20_block() outputs the keystream block, it uses 'u32' stores
directly. However, the callers (crypto/chacha20_generic.c and
drivers/char/random.c) declare the keystream buffer as a 'u8' array,
which is not guaranteed to have the needed alignment.
Fix it by having both callers declare the keystream as a 'u32' array.
For now this is preferable to switching over to the unaligned access
macros because chacha20_block() is only being used in cases where we can
easily control the alignment (stack buffers).
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Now that crypto_chacha20_setkey() and crypto_chacha20_init() use the
unaligned access macros and crypto_xor() also accepts unaligned buffers,
there is no need to have a cra_alignmask set for chacha20-generic.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The generic ChaCha20 implementation has a cra_alignmask of 3, which
ensures that the key passed into crypto_chacha20_setkey() and the IV
passed into crypto_chacha20_init() are 4-byte aligned. However, these
functions are also called from the ARM and ARM64 implementations of
ChaCha20, which intentionally do not have a cra_alignmask set. This is
broken because 32-bit words are being loaded from potentially-unaligned
buffers without the unaligned access macros.
Fix it by using the unaligned access macros when loading the key and IV.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The four 32-bit constants for the initial state of ChaCha20 were loaded
from a char array which is not guaranteed to have the needed alignment.
Fix it by just assigning the constants directly instead.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Make the cryptd queue length configurable. We recently had customer where this
needed to be tuned to accommodate the aesni_intel module and prevent packet
drop.
Signed-off-by: Jon Maxwell <jmaxwell37@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Preempt counter APIs have been split out, currently, hardirq.h just
includes irq_enter/exit APIs which are not used by crypto at all.
So, remove the unused hardirq.h.
Signed-off-by: Yang Shi <yang.s@alibaba-inc.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: linux-crypto@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
On 32-bit (e.g. with m68k-linux-gnu-gcc-4.1):
crypto/keywrap.c: In function ‘crypto_kw_decrypt’:
crypto/keywrap.c:191: warning: integer constant is too large for ‘long’ type
crypto/keywrap.c: In function ‘crypto_kw_encrypt’:
crypto/keywrap.c:224: warning: integer constant is too large for ‘long’ type
Fixes: 9e49451d7a ("crypto: keywrap - simplify code")
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Reviewed-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
A few years ago the FSF moved and "59 Temple Place" is wrong. Having this
still in our source files feels old and unmaintained.
Let's take the license statement serious and not confuse users.
As https://www.gnu.org/licenses/gpl-howto.html suggests, we replace the
postal address with "<http://www.gnu.org/licenses/>".
Signed-off-by: Martin Kepplinger <martink@posteo.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
If crypto_get_default_rng returns an error, the
function ecc_gen_privkey should return an error.
Instead, it currently tries to use the default_rng
nevertheless, thus creating a kernel panic with a
NULL pointer dereference.
Returning the error directly, as was supposedly
intended when looking at the code, fixes this.
Signed-off-by: Pierre Ducroquet <pinaraf@pinaraf.info>
Reviewed-by: PrasannaKumar Muralidharan <prasannatsmkumar@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In case buffer length is a multiple of PAGE_SIZE,
the S/G table is incorrectly generated.
Fix this by handling buflen = k * PAGE_SIZE separately.
Signed-off-by: Robert Baronescu <robert.baronescu@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
When asked to encrypt or decrypt 0 bytes, both the generic and x86
implementations of Salsa20 crash in blkcipher_walk_done(), either when
doing 'kfree(walk->buffer)' or 'free_page((unsigned long)walk->page)',
because walk->buffer and walk->page have not been initialized.
The bug is that Salsa20 is calling blkcipher_walk_done() even when
nothing is in 'walk.nbytes'. But blkcipher_walk_done() is only meant to
be called when a nonzero number of bytes have been provided.
The broken code is part of an optimization that tries to make only one
call to salsa20_encrypt_bytes() to process inputs that are not evenly
divisible by 64 bytes. To fix the bug, just remove this "optimization"
and use the blkcipher_walk API the same way all the other users do.
Reproducer:
#include <linux/if_alg.h>
#include <sys/socket.h>
#include <unistd.h>
int main()
{
int algfd, reqfd;
struct sockaddr_alg addr = {
.salg_type = "skcipher",
.salg_name = "salsa20",
};
char key[16] = { 0 };
algfd = socket(AF_ALG, SOCK_SEQPACKET, 0);
bind(algfd, (void *)&addr, sizeof(addr));
reqfd = accept(algfd, 0, 0);
setsockopt(algfd, SOL_ALG, ALG_SET_KEY, key, sizeof(key));
read(reqfd, key, sizeof(key));
}
Reported-by: syzbot <syzkaller@googlegroups.com>
Fixes: eb6f13eb9f ("[CRYPTO] salsa20_generic: Fix multi-page processing")
Cc: <stable@vger.kernel.org> # v2.6.25+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Because the HMAC template didn't check that its underlying hash
algorithm is unkeyed, trying to use "hmac(hmac(sha3-512-generic))"
through AF_ALG or through KEYCTL_DH_COMPUTE resulted in the inner HMAC
being used without having been keyed, resulting in sha3_update() being
called without sha3_init(), causing a stack buffer overflow.
This is a very old bug, but it seems to have only started causing real
problems when SHA-3 support was added (requires CONFIG_CRYPTO_SHA3)
because the innermost hash's state is ->import()ed from a zeroed buffer,
and it just so happens that other hash algorithms are fine with that,
but SHA-3 is not. However, there could be arch or hardware-dependent
hash algorithms also affected; I couldn't test everything.
Fix the bug by introducing a function crypto_shash_alg_has_setkey()
which tests whether a shash algorithm is keyed. Then update the HMAC
template to require that its underlying hash algorithm is unkeyed.
Here is a reproducer:
#include <linux/if_alg.h>
#include <sys/socket.h>
int main()
{
int algfd;
struct sockaddr_alg addr = {
.salg_type = "hash",
.salg_name = "hmac(hmac(sha3-512-generic))",
};
char key[4096] = { 0 };
algfd = socket(AF_ALG, SOCK_SEQPACKET, 0);
bind(algfd, (const struct sockaddr *)&addr, sizeof(addr));
setsockopt(algfd, SOL_ALG, ALG_SET_KEY, key, sizeof(key));
}
Here was the KASAN report from syzbot:
BUG: KASAN: stack-out-of-bounds in memcpy include/linux/string.h:341 [inline]
BUG: KASAN: stack-out-of-bounds in sha3_update+0xdf/0x2e0 crypto/sha3_generic.c:161
Write of size 4096 at addr ffff8801cca07c40 by task syzkaller076574/3044
CPU: 1 PID: 3044 Comm: syzkaller076574 Not tainted 4.14.0-mm1+ #25
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
__dump_stack lib/dump_stack.c:17 [inline]
dump_stack+0x194/0x257 lib/dump_stack.c:53
print_address_description+0x73/0x250 mm/kasan/report.c:252
kasan_report_error mm/kasan/report.c:351 [inline]
kasan_report+0x25b/0x340 mm/kasan/report.c:409
check_memory_region_inline mm/kasan/kasan.c:260 [inline]
check_memory_region+0x137/0x190 mm/kasan/kasan.c:267
memcpy+0x37/0x50 mm/kasan/kasan.c:303
memcpy include/linux/string.h:341 [inline]
sha3_update+0xdf/0x2e0 crypto/sha3_generic.c:161
crypto_shash_update+0xcb/0x220 crypto/shash.c:109
shash_finup_unaligned+0x2a/0x60 crypto/shash.c:151
crypto_shash_finup+0xc4/0x120 crypto/shash.c:165
hmac_finup+0x182/0x330 crypto/hmac.c:152
crypto_shash_finup+0xc4/0x120 crypto/shash.c:165
shash_digest_unaligned+0x9e/0xd0 crypto/shash.c:172
crypto_shash_digest+0xc4/0x120 crypto/shash.c:186
hmac_setkey+0x36a/0x690 crypto/hmac.c:66
crypto_shash_setkey+0xad/0x190 crypto/shash.c:64
shash_async_setkey+0x47/0x60 crypto/shash.c:207
crypto_ahash_setkey+0xaf/0x180 crypto/ahash.c:200
hash_setkey+0x40/0x90 crypto/algif_hash.c:446
alg_setkey crypto/af_alg.c:221 [inline]
alg_setsockopt+0x2a1/0x350 crypto/af_alg.c:254
SYSC_setsockopt net/socket.c:1851 [inline]
SyS_setsockopt+0x189/0x360 net/socket.c:1830
entry_SYSCALL_64_fastpath+0x1f/0x96
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
af_alg_free_areq_sgls()
If allocating the ->tsgl member of 'struct af_alg_async_req' failed,
during cleanup we dereferenced the NULL ->tsgl pointer in
af_alg_free_areq_sgls(), because ->tsgl_entries was nonzero.
Fix it by only freeing the ->tsgl list if it is non-NULL.
This affected both algif_skcipher and algif_aead.
Fixes: e870456d8e ("crypto: algif_skcipher - overhaul memory management")
Fixes: d887c52d6a ("crypto: algif_aead - overhaul memory management")
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: <stable@vger.kernel.org> # v4.14+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In the AEAD interface for AF_ALG, the reference to the "null skcipher"
held by each tfm was being dropped in the wrong place -- when each
af_alg_ctx was freed instead of when the aead_tfm was freed. As
discovered by syzkaller, a specially crafted program could use this to
cause the null skcipher to be freed while it is still in use.
Fix it by dropping the reference in the right place.
Fixes: 72548b093e ("crypto: algif_aead - copy AAD from src to dst")
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: <stable@vger.kernel.org> # v4.14+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In rsa_get_n(), if the buffer contained all 0's and "FIPS mode" is
enabled, we would read one byte past the end of the buffer while
scanning the leading zeroes. Fix it by checking 'n_sz' before '!*ptr'.
This bug was reachable by adding a specially crafted key of type
"asymmetric" (requires CONFIG_RSA and CONFIG_X509_CERTIFICATE_PARSER).
KASAN report:
BUG: KASAN: slab-out-of-bounds in rsa_get_n+0x19e/0x1d0 crypto/rsa_helper.c:33
Read of size 1 at addr ffff88003501a708 by task keyctl/196
CPU: 1 PID: 196 Comm: keyctl Not tainted 4.14.0-09238-g1d3b78bbc6e9 #26
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.11.0-20171110_100015-anatol 04/01/2014
Call Trace:
rsa_get_n+0x19e/0x1d0 crypto/rsa_helper.c:33
asn1_ber_decoder+0x82a/0x1fd0 lib/asn1_decoder.c:328
rsa_set_pub_key+0xd3/0x320 crypto/rsa.c:278
crypto_akcipher_set_pub_key ./include/crypto/akcipher.h:364 [inline]
pkcs1pad_set_pub_key+0xae/0x200 crypto/rsa-pkcs1pad.c:117
crypto_akcipher_set_pub_key ./include/crypto/akcipher.h:364 [inline]
public_key_verify_signature+0x270/0x9d0 crypto/asymmetric_keys/public_key.c:106
x509_check_for_self_signed+0x2ea/0x480 crypto/asymmetric_keys/x509_public_key.c:141
x509_cert_parse+0x46a/0x620 crypto/asymmetric_keys/x509_cert_parser.c:129
x509_key_preparse+0x61/0x750 crypto/asymmetric_keys/x509_public_key.c:174
asymmetric_key_preparse+0xa4/0x150 crypto/asymmetric_keys/asymmetric_type.c:388
key_create_or_update+0x4d4/0x10a0 security/keys/key.c:850
SYSC_add_key security/keys/keyctl.c:122 [inline]
SyS_add_key+0xe8/0x290 security/keys/keyctl.c:62
entry_SYSCALL_64_fastpath+0x1f/0x96
Allocated by task 196:
__do_kmalloc mm/slab.c:3711 [inline]
__kmalloc_track_caller+0x118/0x2e0 mm/slab.c:3726
kmemdup+0x17/0x40 mm/util.c:118
kmemdup ./include/linux/string.h:414 [inline]
x509_cert_parse+0x2cb/0x620 crypto/asymmetric_keys/x509_cert_parser.c:106
x509_key_preparse+0x61/0x750 crypto/asymmetric_keys/x509_public_key.c:174
asymmetric_key_preparse+0xa4/0x150 crypto/asymmetric_keys/asymmetric_type.c:388
key_create_or_update+0x4d4/0x10a0 security/keys/key.c:850
SYSC_add_key security/keys/keyctl.c:122 [inline]
SyS_add_key+0xe8/0x290 security/keys/keyctl.c:62
entry_SYSCALL_64_fastpath+0x1f/0x96
Fixes: 5a7de97309 ("crypto: rsa - return raw integers for the ASN.1 parser")
Cc: <stable@vger.kernel.org> # v4.8+
Cc: Tudor Ambarus <tudor-dan.ambarus@nxp.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: James Morris <james.l.morris@oracle.com>
Reviewed-by: David Howells <dhowells@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The skcipher_walk_aead_common function calls scatterwalk_copychunks on
the input and output walks to skip the associated data. If the AD end
at an SG list entry boundary, then after these calls the walks will
still be pointing to the end of the skipped region.
These offsets are later checked for alignment in skcipher_walk_next,
so the skcipher_walk may detect the alignment incorrectly.
This patch fixes it by calling scatterwalk_done after the copychunks
calls to ensure that the offsets refer to the right SG list entry.
Fixes: b286d8b1a6 ("crypto: skcipher - Add skcipher walk interface")
Cc: <stable@vger.kernel.org>
Signed-off-by: Ondrej Mosnacek <omosnacek@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
