The timeout mechanism won't work if SIGALRM is blocked. This changes
unmasks SIGALRM when the timer is installed. This doesn't completely
solve the problem, as the fuzzer could trigger some device activity that
re-masks SIGALRM. However, there are currently no inputs on OSS-Fuzz
that re-mask SIGALRM and timeout. If that turns out to be a real issue,
we could try to hook sigmask-type calls, or use a separate timer thread.
Based-on: <20210713150037.9297-1-alxndr@bu.edu>
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Using ITIMER_VIRTUAL is a bad idea, if the fuzzer hits a blocking
syscall - e.g. ppoll with a NULL timespec. This causes timeout issues
while fuzzing some block-device code. Fix that by using wall-clock time.
This might cause inputs to timeout sometimes due to scheduling
effects/ambient load, but it is better than bringing the entire fuzzing
process to a halt.
Based-on: <20210713150037.9297-1-alxndr@bu.edu>
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
We have some configs for devices such as the AC97 and ES1370 that were
not matching memory-regions correctly, because the configs provided
lowercase names. To resolve these problems and prevent them from
occurring again in the future, convert both the pattern and names to
lower-case, prior to checking for a match.
Suggested-by: Darren Kenny <darren.kenny@oracle.com>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Using a custom timeout is useful to continue fuzzing complex devices,
even after we run into some slow code-path. However, simply adding a
fixed timeout to each input effectively caps the maximum input
length/number of operations at some artificial value. There are two
major problems with this:
1. Some code might only be reachable through long IO sequences.
2. Longer inputs can actually be _better_ for performance. While the
raw number of fuzzer executions decreases with larger inputs, the
number of MMIO/PIO/DMA operation/second actually increases, since
were are speding proportionately less time fork()ing.
With this change, we keep the custom-timeout, but we renew it, prior to
each MMIO/PIO/DMA operation. Thus, we time-out only when a specific
operation takes a long time.
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
The code mistakenly relied on address_space_translate to store the
length remaining until the next memory-region. We care about this
because when there is RAM or sparse-memory neighboring on an MMIO
region, we should only write up to the border, to prevent inadvertently
invoking MMIO handlers within the DMA callback.
However address_space_translate_internal only stores the length until
the end of the MemoryRegion if memory_region_is_ram(mr). Otherwise
the *len is left unmodified. This caused some false-positive issues,
where the fuzzer found a way to perform a nested MMIO write through a
DMA callback on an [address, length] that started within sparse memory
and spanned some device MMIO regions.
To fix this, write to sparse memory in small chunks of
memory_access_size (similar to the underlying address_space_write code),
which will prevent accidentally hitting MMIO handlers through large
writes.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Use memory_region_size() to get the MemoryRegion size,
and display it with the '0x' prefix.
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20210612195842.1595999-1-f4bug@amsat.org>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Stop including exec/address-spaces.h in files that don't need it.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20210416171314.2074665-5-thuth@redhat.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
The function flatview_for_each_range() calls a callback for each
range in a FlatView. Currently the callback gets the start and
length of the range and the MemoryRegion involved, but not the offset
within the MemoryRegion. Add this to the callback's arguments; we're
going to want it for a new use in the next commit.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20210318174823.18066-4-peter.maydell@linaro.org
The return value of the flatview_cb callback passed to the
flatview_for_each_range() function is zero if the iteration through
the ranges should continue, or non-zero to break out of it. Use a
bool for this rather than int.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20210318174823.18066-2-peter.maydell@linaro.org
The generic-fuzzer often provides randomized DMA addresses to
virtual-devices. For a 64-bit address-space, the chance of these
randomized addresses coinciding with RAM regions, is fairly small. Even
though the fuzzer's instrumentation eventually finds valid addresses,
this can take some-time, and slows-down fuzzing progress (especially,
when multiple DMA buffers are involved). To work around this, create
"fake" sparse-memory that spans all of the 64-bit address-space. Adjust
the DMA call-back to populate this sparse memory, correspondingly
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
I noticed that with a sufficiently small timeout, the fuzzer fork-server
sometimes locks up. On closer inspection, the issue appeared to be
caused by entering our SIGALRM handler, while libfuzzer is in it's crash
handlers. Because libfuzzer relies on pipe communication with an
external child process to print out stack-traces, we shouldn't exit
early, and leave an orphan child. Check for children in the SIGALRM
handler to avoid this issue.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Acked-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
For some device configurations, it is useful to configure some
resources, and adjust QEMU arguments at runtime, prior to fuzzing. This
patch adds an "argfunc" to generic the generic_fuzz_config. When
specified, it is responsible for configuring the resources and returning
a string containing the corresponding QEMU arguments. This can be useful
for targets that rely on e.g.:
* a temporary qcow2 image
* a temporary directory
* an unused TCP port used to bind the VNC server
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20210117230924.449676-2-alxndr@bu.edu>
We passed an is_write flag to the fuzz_dma_read_cb function to
differentiate between the mapped DMA regions that need to be populated
with fuzzed data, and those that don't. We simply passed through the
address_space_map is_write parameter. The goal was to cut down on
unnecessarily populating mapped DMA regions, when they are not read
from.
Unfortunately, nothing precludes code from reading from regions mapped
with is_write=true. For example, see:
https://lists.gnu.org/archive/html/qemu-devel/2021-01/msg04729.html
This patch removes the is_write parameter to fuzz_dma_read_cb. As a
result, we will fill all mapped DMA regions with fuzzed data, ignoring
the specified transfer direction.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20210120060255.558535-1-alxndr@bu.edu>
Prior to this patch, the fuzzer found inputs to map PCI device BARs and
enable the device. While it is nice that the fuzzer can do this, it
added significant overhead, since the fuzzer needs to map all the
BARs (regenerating the memory topology), at the start of each input.
With this patch, we do this once, before fuzzing, mitigating some of
this overhead.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20201221181203.1853-1-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
The generic fuzzer can find double-fetch bugs. However:
* We currently have no good way of producing qemu-system reproducers for
double-fetch bugs. Even if we can get developers to run the binary-blob
reproducers with the qemu-fuzz builds, we currently don't have a minimizer for
these reproducers, so they are usually not easy to follow.
* Often times the fuzzer will provide a reproducer containing a
double-fetch for a bug that can be reproduced without double-fetching.
Until we find a way to build nice double-fetch reproducers that
developers are willing to look at, lets tell OSS-Fuzz to avoid
double-fetches.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20201202164214.93867-1-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
The code did not add offsets to FlatRange bases, so we did not fuzz
offsets within device MemoryRegions.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20201029172901.534442-4-alxndr@bu.edu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We should be checking that the device is trying to read from RAM, before
filling the region with data. Otherwise, we will try to populate
nonsensical addresses in RAM for callbacks on PIO/MMIO reads. We did
this originally, however the final version I sent had the line commented
out..
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20201029172901.534442-3-alxndr@bu.edu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This code had all sorts of issues. We used a loop similar to
address_space_write_rom, but I did not remove a "break" that only made
sense in the context of the switch statement in the original code. Then,
after the loop, we did a separate qtest_memwrite over the entire DMA
access range, defeating the purpose of the loop. Additionally, we
increment the buf pointer, and then try to g_free() it. Fix these
problems.
Reported-by: OSS-Fuzz (Issue 26725)
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reported-by: OSS-Fuzz (Issue 26691)
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20201029172901.534442-2-alxndr@bu.edu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We call get_generic_fuzz_configs, which fills an array with
predefined {name, args, objects} triples. For each of these, we add a
new FuzzTarget, that uses a small wrapper to set
QEMU_FUZZ_{ARGS,OBJECTS} to the corresponding predefined values.
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <20201023150746.107063-16-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <20201023150746.107063-10-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
This new operation is used in the next commit, which concatenates two
fuzzer-generated inputs. With this operation, we can prevent the second
input from clobbering the PCI configuration performed by the first.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20201023150746.107063-9-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
When a virtual-device tries to access some buffer in memory over DMA, we
add call-backs into the fuzzer(next commit). The fuzzer checks verifies
that the DMA request maps to a physical RAM address and fills the memory
with fuzzer-provided data. The patterns that we use to fill this memory
are specified using add_dma_pattern and clear_dma_patterns operations.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20201023150746.107063-5-alxndr@bu.edu>
[thuth: Reformatted one comment according to the QEMU coding style]
Signed-off-by: Thomas Huth <thuth@redhat.com>
This patch compares TYPE_PCI_DEVICE objects against the user-provided
matching pattern. If there is a match, we use some hacks and leverage
QOS to map each possible BAR for that device. Now fuzzed inputs might be
converted to pci_read/write commands which target specific. This means
that we can fuzz a particular device's PCI configuration space,
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20201023150746.107063-4-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
This is a generic fuzzer designed to fuzz a virtual device's
MemoryRegions, as long as they exist within the Memory or Port IO (if it
exists) AddressSpaces. The fuzzer's input is interpreted into a sequence
of qtest commands (outb, readw, etc). The interpreted commands are
separated by a magic seaparator, which should be easy for the fuzzer to
guess. Without ASan, the separator can be specified as a "dictionary
value" using the -dict argument (see libFuzzer documentation).
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Message-Id: <20201023150746.107063-3-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
