By default, -fsanitize=fuzzer instruments all code with coverage
information. However, this means that libfuzzer will track coverage over
hundreds of source files that are unrelated to virtual-devices. This
means that libfuzzer will optimize inputs for coverage observed in timer
code, memory APIs etc. This slows down the fuzzer and stores many inputs
that are not relevant to the actual virtual-devices.
With this change, clang versions that support the
"-fsanitize-coverage-allowlist" will only instrument a subset of the
compiled code, that is directly related to virtual-devices.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
On oss-fuzz, we build twice, to put together a build that is portable to
the runner containers. On gitlab ci, this is wasteful and contributes to
timeouts on the build-oss-fuzz job. Avoid building twice on gitlab, at
the remote cost of potentially missing some cases that break oss-fuzz
builds.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20210809111621.54454-1-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
While we only use stdin, the chardev is named 'stdio'.
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <20210602170759.2500248-4-f4bug@amsat.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Currently, bash and C crash reproducers are be built manually. This is a
problem, as we want to integrate reproducers into the tree, for
regression testing. This patch adds a script that converts a sequence of
QTest commands into a pasteable Bash reproducer, or a libqtest-based C
program. This will try to wrap pasteable reproducers to 72 chars, but
the generated C code will not have nice formatting. Therefore, the C
output of this script should be piped through an auto-formatter, such as
clang-format
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Qiuhao Li <Qiuhao.Li@outlook.com>
Message-Id: <SYCPR01MB3502E9F6EB06DEDCD484F738FCBA9@SYCPR01MB3502.ausprd01.prod.outlook.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
If previous write commands write the same length of data with the same step,
we view it as a hint.
Signed-off-by: Qiuhao Li <Qiuhao.Li@outlook.com>
Reviewed-by: Alexander Bulekov <alxndr@bu.edu>
Tested-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <SYCPR01MB3502480AD07811A6A49B8FEAFCAB0@SYCPR01MB3502.ausprd01.prod.outlook.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
-M1: remove IO commands iteratively
-M2: try setting bits in operand of write/out to zero
Signed-off-by: Qiuhao Li <Qiuhao.Li@outlook.com>
Reviewed-by: Alexander Bulekov <alxndr@bu.edu>
Tested-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <SYCPR01MB350204C52E7A39E6B0EEC870FCAB0@SYCPR01MB3502.ausprd01.prod.outlook.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Simplifying the crash cases by opportunistically setting bits in operands of
out/write to zero may help to debug, since usually bit one means turn on or
trigger a function while zero is the default turn-off setting.
Tested bug https://bugs.launchpad.net/qemu/+bug/1908062
Signed-off-by: Qiuhao Li <Qiuhao.Li@outlook.com>
Reviewed-by: Alexander Bulekov <alxndr@bu.edu>
Tested-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <SYCPR01MB3502C84B6346A3E3DE708C7BFCAB0@SYCPR01MB3502.ausprd01.prod.outlook.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Now we use a one-time scan and remove strategy in the minimizer,
which is not suitable for timing dependent instructions.
For example, instruction A will indicate an address where the config
chunk locates, and instruction B will make the configuration active.
If we have the following instruction sequence:
...
A1
B1
A2
B2
...
A2 and B2 are the actual instructions that trigger the bug.
If we scan from top to bottom, after we remove A1, the behavior of B1
might be unknowable, including not to crash the program. But we will
successfully remove B1 later cause A2 and B2 will crash the process
anyway:
...
A1
A2
B2
...
Now one more trimming will remove A1.
In the perfect case, we would need to be able to remove A and B (or C!) at
the same time. But for now, let's just add a loop around the minimizer.
Since we only remove instructions, this iterative algorithm is converging.
Tested with Bug 1908062.
Signed-off-by: Qiuhao Li <Qiuhao.Li@outlook.com>
Reviewed-by: Alexander Bulekov <alxndr@bu.edu>
Tested-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <SYCPR01MB350263004448040ACCB9A9F1FCAB0@SYCPR01MB3502.ausprd01.prod.outlook.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Currently, we split the write commands' data from the middle. If it does not
work, try to move the pivot left by one byte and retry until there is no
space.
But, this method has two flaws:
1. It may fail to trim all unnecessary bytes on the right side.
For example, there is an IO write command:
write addr uuxxxxuu
u is the unnecessary byte for the crash. Unlike ram write commands, in most
case, a split IO write won't trigger the same crash, So if we split from the
middle, we will get:
write addr uu (will be removed in next round)
write addr xxxxuu
For xxxxuu, since split it from the middle and retry to the leftmost byte
won't get the same crash, we will be stopped from removing the last two
bytes.
2. The algorithm complexity is O(n) since we move the pivot byte by byte.
To solve the first issue, we can try a symmetrical position on the right if
we fail on the left. As for the second issue, instead moving by one byte, we
can approach the boundary exponentially, achieving O(log(n)).
Give an example:
xxxxuu len=6
+
|
+
xxx,xuu 6/2=3 fail
+
+--------------+-------------+
| |
+ +
xx,xxuu 6/2^2=1 fail xxxxu,u 6-1=5 success
+ +
+------------------+----+ |
| | +-------------+ u removed
+ +
xx,xxu 5/2=2 fail xxxx,u 6-2=4 success
+
|
+-----------+ u removed
In some rare cases, this algorithm will fail to trim all unnecessary bytes:
xxxxxxxxxuxxxxxx
xxxxxxxx-xuxxxxxx Fail
xxxx-xxxxxuxxxxxx Fail
xxxxxxxxxuxx-xxxx Fail
...
I think the trade-off is worth it.
Signed-off-by: Qiuhao Li <Qiuhao.Li@outlook.com>
Reviewed-by: Alexander Bulekov <alxndr@bu.edu>
Tested-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <SYCPR01MB3502D26F1BEB680CBBC169E5FCAB0@SYCPR01MB3502.ausprd01.prod.outlook.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Instead of removing IO instructions one by one, we can try deleting multiple
instructions at once. According to the locality of reference, we double the
number of instructions to remove for the next round and recover it to one
once we fail.
This patch is usually significant for large input.
Test with quadrupled trace input at:
https://bugs.launchpad.net/qemu/+bug/1890333/comments/1
Patched 1/6 version:
real 0m45.904s
user 0m16.874s
sys 0m10.042s
Refined version:
real 0m11.412s
user 0m6.888s
sys 0m3.325s
Signed-off-by: Qiuhao Li <Qiuhao.Li@outlook.com>
Reviewed-by: Alexander Bulekov <alxndr@bu.edu>
Tested-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <SYCPR01MB350280A67BB55C3FADF173E3FCAB0@SYCPR01MB3502.ausprd01.prod.outlook.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
We spend much time waiting for the timeout program during the minimization
process until it passes a time limit. This patch hacks the CLOSED (indicates
the redirection file closed) notification in QTest's output if it doesn't
crash.
Test with quadrupled trace input at:
https://bugs.launchpad.net/qemu/+bug/1890333/comments/1
Original version:
real 1m37.246s
user 0m13.069s
sys 0m8.399s
Refined version:
real 0m45.904s
user 0m16.874s
sys 0m10.042s
Note:
Sometimes the mutated or the same trace may trigger a different crash
summary (second-to-last line) but indicates the same bug. For example, Bug
1910826 [1], which will trigger a stack overflow, may output summaries
like:
SUMMARY: AddressSanitizer: stack-overflow
/home/qiuhao/hack/qemu/build/../softmmu/physmem.c:488 in
flatview_do_translate
or
SUMMARY: AddressSanitizer: stack-overflow
(/home/qiuhao/hack/qemu/build/qemu-system-i386+0x27ca049) in __asan_memcpy
Etc.
If we use the whole summary line as the token, we may be prevented from
further minimization. So in this patch, we only use the first three words
which indicate the type of crash:
SUMMARY: AddressSanitizer: stack-overflow
[1] https://bugs.launchpad.net/qemu/+bug/1910826
Signed-off-by: Qiuhao Li <Qiuhao.Li@outlook.com>
Reviewed-by: Alexander Bulekov <alxndr@bu.edu>
Tested-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <SYCPR01MB350251DC04003450348FAF68FCAB0@SYCPR01MB3502.ausprd01.prod.outlook.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
We switched to hardlinks in
a942f64cc4 ("scripts/oss-fuzz: use hardlinks instead of copying")
The motivation was to conserve space (50 fuzzers built with ASAN, can
weigh close to 9 GB).
Unfortunately, OSS-Fuzz (partially) treated the underlying copy of the
fuzzer as a standalone fuzzer. To attempt to fix, we tried:
f8b8f37463 ("scripts/oss-fuzz: rename bin/qemu-fuzz-i386")
This was also not a complete fix, because though OSS-Fuzz
ignores the renamed fuzzer, the underlying ClusterFuzz, doesn't:
https://storage.googleapis.com/clusterfuzz-builds/qemu/targets.list.addresshttps://oss-fuzz-build-logs.storage.googleapis.com/log-9bfb55f9-1c20-4aa6-a49c-ede12864eeb2.txt
(clusterfuzz still lists qemu-fuzz-i386.base as a fuzzer)
This change keeps the hard-links, but makes them all point to a file
with a qemu-fuzz-i386-target-.. name. If we have targets, A, B, C, the
result will be:
qemu-fuzz-i386-target-A (base file)
qemu-fuzz-i386-target-B -> qemu-fuzz-i386-target-A
qemu-fuzz-i386-target-C -> qemu-fuzz-i386-target-A
The result should be that every file that looks like a fuzzer to
OSS-Fuzz/ClusterFuzz, can run as a fuzzer (we don't have a separate base
copy). Unfortunately, there is not simple way to test this locally.
In the future, it might be worth it to link the majority of QEMU in as a
shared-object (see https://github.com/google/oss-fuzz/issues/4575 )
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <20201108171136.160607-1-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
OSS-Fuzz changed the way it scans for fuzzers in $DEST_DIR. The new code
also scans subdirectories for fuzzers. This means that OSS-Fuzz is
considering bin/qemu-fuzz-i386 as an independent fuzzer (it is not - it
requires a --fuzz-target argument). This has led to coverage-build
failures and false crash reports. To work around this, we take advantage
of OSS-Fuzz' filename extension check - OSS-Fuzz will not run anything
that has an extension that is not ".exe":
https://github.com/google/oss-fuzz/blob/master/infra/utils.py#L115
Reported-by: OSS-Fuzz (Issue 26725)
Reported-by: OSS-Fuzz (Issue 26679)
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <20201101212245.185819-1-alxndr@bu.edu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
generic-fuzz is not a standalone fuzzer - it requires some env variables
to be set. On oss-fuzz, we set these with some predefined
generic-fuzz-{...} targets, that are thin wrappers around generic-fuzz.
Do not make a link for the generic-fuzz from the oss-fuzz build, so
oss-fuzz does not treat it as a standalone fuzzer.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <20201023150746.107063-18-alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
[thuth: Reformatted one comment to stay within the 80 columns limit]
Signed-off-by: Thomas Huth <thuth@redhat.com>
Prior to this, fuzzers in the output oss-fuzz directory were exactly
the same executable, with a different name to do argv[0]-based
fuzz-target selection. This is a waste of space, especially since these
binaries can weigh many MB.
Instead of copying, use hard links, to cut down on wasted space. We need
to place the primary copy of the executable into DEST_DIR, since this is
a separate file-system on oss-fuzz. We should not place it directly into
$DEST_DIR, since oss-fuzz will treat it as an independent fuzzer and try
to run it for fuzzing. Instead, we create a DEST_DIR/bin directory to
store the primary copy.
Suggested-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <20201023150746.107063-17-alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Once we find a crash, we can convert it into a QTest trace. Usually this
trace will contain many operations that are unneeded to reproduce the
crash. This script tries to minimize the crashing trace, by removing
operations and trimming QTest bufwrite(write addr len data...) commands.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20201023150746.107063-12-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
The generic-fuzzer uses hooks to fulfill DMA requests just-in-time.
This means that if we try to use QTEST_LOG=1 to build a reproducer, the
DMA writes will be logged _after_ the in/out/read/write that triggered
the DMA read. To work work around this, the generic-fuzzer annotates
these just-in time DMA fulfilments with a tag that we can use to
discern them. This script simply iterates over a raw qtest
trace (including log messages, errors, timestamps etc), filters it and
re-orders it so that DMA fulfillments are placed directly _before_ the
qtest command that will cause the DMA access.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20201023150746.107063-11-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
I found that there are many spelling errors in the comments of qemu,
so I used the spellcheck tool to check the spelling errors
and finally found some spelling errors in the scripts folder.
Signed-off-by: zhaolichang <zhaolichang@huawei.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <20200917075029.313-5-zhaolichang@huawei.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Prior to this change,
readelf -d build/out/qemu/qemu-fuzz-i386-target-virtio-net-slirp
...
0x000000000000000f (RPATH) Library rpath: ['$$ORIGIN/lib':$ORIGIN/migration:$ORIGIN/]
As of 1a4db552d8 ("ninjatool: quote dollars in variables"), we don't
need to manually double the dollars. Also, remove the single-quotes as
they are copied into the rpath.
After this change:
0x000000000000000f (RPATH) Library rpath: [$ORIGIN/lib:$ORIGIN/migration:$ORIGIN/]
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <20200902142657.112879-3-alxndr@bu.edu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The binaries move to the root directory, e.g. qemu-system-i386 or
qemu-arm. This requires changes to qtests, CI, etc.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Meson build scripts will only include qemu-fuzz-TARGET rules if configured
with --enable-fuzzing, and that takes care of adding -fsanitize=fuzzer.
Therefore we can just specify the configure option and stop modifying
the CFLAGS and CONFIG_FUZZ options in the "make" invocation.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The build.sh script only copies qemu-fuzz-i386 to the destination folder,
so we can speed up the compilation step quite a bit by not compiling the
other targets here.
Signed-off-by: Thomas Huth <thuth@redhat.com>
It is neater to keep this in the QEMU repo, since any change that
requires an update to the oss-fuzz build configuration, can make the
necessary changes in the same series.
Suggested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20200612055145.12101-1-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
