[PATCH] fuzz: avoid building twice, when running on gitlab

[PATCH] fuzz: avoid building twice, when running on gitlab

[Alexander Bulekov]

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>
---

From a couple test runs it looks like this can shave off 15-20 minutes.

 scripts/oss-fuzz/build.sh | 24 +++++++++++++-----------
 1 file changed, 13 insertions(+), 11 deletions(-)

diff --git a/scripts/oss-fuzz/build.sh b/scripts/oss-fuzz/build.sh
index c1af43fded..98b56e0521 100755
--- a/scripts/oss-fuzz/build.sh
+++ b/scripts/oss-fuzz/build.sh
@@ -73,17 +73,19 @@ if ! make "-j$(nproc)" qemu-fuzz-i386; then
           "\nFor example: CC=clang CXX=clang++ $0"
 fi
 
-for i in $(ldd ./qemu-fuzz-i386 | cut -f3 -d' '); do
-    cp "$i" "$DEST_DIR/lib/"
-done
-rm qemu-fuzz-i386
-
-# Build a second time to build the final binary with correct rpath
-../configure --disable-werror --cc="$CC" --cxx="$CXX" --enable-fuzzing \
-    --prefix="$DEST_DIR" --bindir="$DEST_DIR" --datadir="$DEST_DIR/data/" \
-    --extra-cflags="$EXTRA_CFLAGS" --extra-ldflags="-Wl,-rpath,\$ORIGIN/lib" \
-    --target-list="i386-softmmu"
-make "-j$(nproc)" qemu-fuzz-i386 V=1
+if [ "$GITLAB_CI" != "true" ]; then
+    for i in $(ldd ./qemu-fuzz-i386 | cut -f3 -d' '); do
+        cp "$i" "$DEST_DIR/lib/"
+    done
+    rm qemu-fuzz-i386
+
+    # Build a second time to build the final binary with correct rpath
+    ../configure --disable-werror --cc="$CC" --cxx="$CXX" --enable-fuzzing \
+        --prefix="$DEST_DIR" --bindir="$DEST_DIR" --datadir="$DEST_DIR/data/" \
+        --extra-cflags="$EXTRA_CFLAGS" --extra-ldflags="-Wl,-rpath,\$ORIGIN/lib" \
+        --target-list="i386-softmmu"
+    make "-j$(nproc)" qemu-fuzz-i386 V=1
+fi
 
 # Copy over the datadir
 cp  -r ../pc-bios/ "$DEST_DIR/pc-bios"
-- 
2.30.2

Re: [PATCH] fuzz: avoid building twice, when running on gitlab

[Darren Kenny]

On Monday, 2021-08-09 at 07:16:21 -04, Alexander Bulekov wrote:
> 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>

> ---
>
> From a couple test runs it looks like this can shave off 15-20 minutes.
>
>  scripts/oss-fuzz/build.sh | 24 +++++++++++++-----------
>  1 file changed, 13 insertions(+), 11 deletions(-)
>
> diff --git a/scripts/oss-fuzz/build.sh b/scripts/oss-fuzz/build.sh
> index c1af43fded..98b56e0521 100755
> --- a/scripts/oss-fuzz/build.sh
> +++ b/scripts/oss-fuzz/build.sh
> @@ -73,17 +73,19 @@ if ! make "-j$(nproc)" qemu-fuzz-i386; then
>            "\nFor example: CC=clang CXX=clang++ $0"
>  fi
>  
> -for i in $(ldd ./qemu-fuzz-i386 | cut -f3 -d' '); do
> -    cp "$i" "$DEST_DIR/lib/"
> -done
> -rm qemu-fuzz-i386
> -
> -# Build a second time to build the final binary with correct rpath
> -../configure --disable-werror --cc="$CC" --cxx="$CXX" --enable-fuzzing \
> -    --prefix="$DEST_DIR" --bindir="$DEST_DIR" --datadir="$DEST_DIR/data/" \
> -    --extra-cflags="$EXTRA_CFLAGS" --extra-ldflags="-Wl,-rpath,\$ORIGIN/lib" \
> -    --target-list="i386-softmmu"
> -make "-j$(nproc)" qemu-fuzz-i386 V=1
> +if [ "$GITLAB_CI" != "true" ]; then
> +    for i in $(ldd ./qemu-fuzz-i386 | cut -f3 -d' '); do
> +        cp "$i" "$DEST_DIR/lib/"
> +    done
> +    rm qemu-fuzz-i386
> +
> +    # Build a second time to build the final binary with correct rpath
> +    ../configure --disable-werror --cc="$CC" --cxx="$CXX" --enable-fuzzing \
> +        --prefix="$DEST_DIR" --bindir="$DEST_DIR" --datadir="$DEST_DIR/data/" \
> +        --extra-cflags="$EXTRA_CFLAGS" --extra-ldflags="-Wl,-rpath,\$ORIGIN/lib" \
> +        --target-list="i386-softmmu"
> +    make "-j$(nproc)" qemu-fuzz-i386 V=1
> +fi
>  
>  # Copy over the datadir
>  cp  -r ../pc-bios/ "$DEST_DIR/pc-bios"
> -- 
> 2.30.2

Re: [PATCH] fuzz: avoid building twice, when running on gitlab

[Peter Maydell]

On Mon, 9 Aug 2021 at 12:18, Alexander Bulekov <alxndr@bu.edu> wrote:
>
> 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>
> ---
>
> From a couple test runs it looks like this can shave off 15-20 minutes.
>
>  scripts/oss-fuzz/build.sh | 24 +++++++++++++-----------
>  1 file changed, 13 insertions(+), 11 deletions(-)

I tried a test run with this, but it still hit the 1 hour timeout:

https://gitlab.com/qemu-project/qemu/-/pipelines/350387482

-- PMM

Re: [PATCH] fuzz: avoid building twice, when running on gitlab

[Alexander Bulekov]

On 210809 1925, Peter Maydell wrote:
> On Mon, 9 Aug 2021 at 12:18, Alexander Bulekov <alxndr@bu.edu> wrote:
> >
> > 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>
> > ---
> >
> > From a couple test runs it looks like this can shave off 15-20 minutes.
> >
> >  scripts/oss-fuzz/build.sh | 24 +++++++++++++-----------
> >  1 file changed, 13 insertions(+), 11 deletions(-)
> 
> I tried a test run with this, but it still hit the 1 hour timeout:
> 
> https://gitlab.com/qemu-project/qemu/-/pipelines/350387482

It also timed out for me with a 120 minute timeout:
https://gitlab.com/a1xndr/qemu/-/jobs/1488160601

The log has almost exactly the same number of lines as yours, so I'm
guessing one of the qtests is timing out with --enable-sanitizers .

-Alex

> 
> -- PMM

Re: [PATCH] fuzz: avoid building twice, when running on gitlab

[Alexander Bulekov]

On 210809 1506, Alexander Bulekov wrote:
> On 210809 1925, Peter Maydell wrote:
> > On Mon, 9 Aug 2021 at 12:18, Alexander Bulekov <alxndr@bu.edu> wrote:
> > >
> > > 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>
> > > ---
> > >
> > > From a couple test runs it looks like this can shave off 15-20 minutes.
> > >
> > >  scripts/oss-fuzz/build.sh | 24 +++++++++++++-----------
> > >  1 file changed, 13 insertions(+), 11 deletions(-)
> > 
> > I tried a test run with this, but it still hit the 1 hour timeout:
> > 
> > https://gitlab.com/qemu-project/qemu/-/pipelines/350387482
> 
> It also timed out for me with a 120 minute timeout:
> https://gitlab.com/a1xndr/qemu/-/jobs/1488160601
> 
> The log has almost exactly the same number of lines as yours, so I'm
> guessing one of the qtests is timing out with --enable-sanitizers .
> 
> -Alex
> 

Building locally:
$ CC=clang-11 CXX=clang++-11 ../configure --enable-fuzzing \
    --enable-debug --enable-sanitizers
$ make check-qtest-i386 check-unit

Same as on gitlab, this times out shortly after outputting
"sh: 1: exec: ./storage-daemon/qemu-storage-daemon: not found"

Manually running qos-test, the same way check-qtest-i386 invokes it:

$ QTEST_QEMU_BINARY=./qemu-system-i386 QTEST_QEMU_STORAGE_DAEMON_BINARY=./storage-daemon/qemu-storage-daemon tests/qtest/qos-test --tap -k -m quick < /dev/null

# starting vhost-user backend: exec ./storage-daemon/qemu-storage-daemon --blockdev driver=file,node-name=disk0,filename=qtest.XRAzzu --export type=vhost-user-blk,id=disk0,addr.type=unix,addr.path=/tmp/qtest-94561-sock.NdKWpt,node-name=disk0,writable=on,num-queues=1
sh: 1: exec: ./storage-daemon/qemu-storage-daemon: not found
# starting QEMU: exec ./qemu-system-i386 -qtest unix:/tmp/qtest-94561.sock -qtest-log /dev/null -chardev socket,path=/tmp/qtest-94561.qmp,id=char0 -mon chardev=char0,mode=control -display none -M pc  -device vhost-user-blk-pci,id=drv0,chardev=char1,addr=4.0 -object memory-backend-memfd,id=mem,size=256M,share=on  -M memory-backend=mem -m 256M -chardev socket id=char1,path=/tmp/qtest-94561-sock.NdKWpt  -accel qtest

*timeout*

Ok, lets try to manually build ./storage-daemon/qemu-storage-daemon
$ make ./storage-daemon/qemu-storage-daemon

And rerun the tests...
$ QTEST_QEMU_BINARY=./qemu-system-i386 QTEST_QEMU_STORAGE_DAEMON_BINARY=./storage-daemon/qemu-storage-daemon tests/qtest/qos-test --tap -k -m quick < /dev/null

No timeout... Still not sure why ./storage-daemon/qemu-storage-daemon isn't
being built by make check, and how to fix that.
-Alex

> > 
> > -- PMM

Re: [PATCH] fuzz: avoid building twice, when running on gitlab

[Peter Maydell]

On Mon, 9 Aug 2021 at 20:30, Alexander Bulekov <alxndr@bu.edu> wrote:
>
> On 210809 1506, Alexander Bulekov wrote:
> > On 210809 1925, Peter Maydell wrote:
> > > On Mon, 9 Aug 2021 at 12:18, Alexander Bulekov <alxndr@bu.edu> wrote:
> > > >
> > > > 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>
> > > > ---
> > > >
> > > > From a couple test runs it looks like this can shave off 15-20 minutes.
> > > >
> > > >  scripts/oss-fuzz/build.sh | 24 +++++++++++++-----------
> > > >  1 file changed, 13 insertions(+), 11 deletions(-)
> > >
> > > I tried a test run with this, but it still hit the 1 hour timeout:
> > >
> > > https://gitlab.com/qemu-project/qemu/-/pipelines/350387482
> >
> > It also timed out for me with a 120 minute timeout:
> > https://gitlab.com/a1xndr/qemu/-/jobs/1488160601
> >
> > The log has almost exactly the same number of lines as yours, so I'm
> > guessing one of the qtests is timing out with --enable-sanitizers .

>
> Building locally:
> $ CC=clang-11 CXX=clang++-11 ../configure --enable-fuzzing \
>     --enable-debug --enable-sanitizers
> $ make check-qtest-i386 check-unit
>
> Same as on gitlab, this times out shortly after outputting
> "sh: 1: exec: ./storage-daemon/qemu-storage-daemon: not found"
>
> Manually running qos-test, the same way check-qtest-i386 invokes it:
>
> $ QTEST_QEMU_BINARY=./qemu-system-i386 QTEST_QEMU_STORAGE_DAEMON_BINARY=./storage-daemon/qemu-storage-daemon tests/qtest/qos-test --tap -k -m quick < /dev/null
>
> # starting vhost-user backend: exec ./storage-daemon/qemu-storage-daemon --blockdev driver=file,node-name=disk0,filename=qtest.XRAzzu --export type=vhost-user-blk,id=disk0,addr.type=unix,addr.path=/tmp/qtest-94561-sock.NdKWpt,node-name=disk0,writable=on,num-queues=1
> sh: 1: exec: ./storage-daemon/qemu-storage-daemon: not found
> # starting QEMU: exec ./qemu-system-i386 -qtest unix:/tmp/qtest-94561.sock -qtest-log /dev/null -chardev socket,path=/tmp/qtest-94561.qmp,id=char0 -mon chardev=char0,mode=control -display none -M pc  -device vhost-user-blk-pci,id=drv0,chardev=char1,addr=4.0 -object memory-backend-memfd,id=mem,size=256M,share=on  -M memory-backend=mem -m 256M -chardev socket id=char1,path=/tmp/qtest-94561-sock.NdKWpt  -accel qtest
>
> *timeout*

vhost-user timing out in realize I suspect. I see that as
an intermittent hang in non-sanitizer configs.

vhost-user folk: Can we either look at fixing this or else disable
the test ? (Stack backtraces available in the other email thread.)

thanks
-- PMM

Re: [PATCH] fuzz: avoid building twice, when running on gitlab

[Philippe Mathieu-Daudé]

+Coiby Xu & qemu-block@

On 8/9/21 9:36 PM, Peter Maydell wrote:
> On Mon, 9 Aug 2021 at 20:30, Alexander Bulekov <alxndr@bu.edu> wrote:
>>
>> On 210809 1506, Alexander Bulekov wrote:
>>> On 210809 1925, Peter Maydell wrote:
>>>> On Mon, 9 Aug 2021 at 12:18, Alexander Bulekov <alxndr@bu.edu> wrote:
>>>>>
>>>>> 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>
>>>>> ---
>>>>>
>>>>> From a couple test runs it looks like this can shave off 15-20 minutes.
>>>>>
>>>>>  scripts/oss-fuzz/build.sh | 24 +++++++++++++-----------
>>>>>  1 file changed, 13 insertions(+), 11 deletions(-)
>>>>
>>>> I tried a test run with this, but it still hit the 1 hour timeout:
>>>>
>>>> https://gitlab.com/qemu-project/qemu/-/pipelines/350387482
>>>
>>> It also timed out for me with a 120 minute timeout:
>>> https://gitlab.com/a1xndr/qemu/-/jobs/1488160601
>>>
>>> The log has almost exactly the same number of lines as yours, so I'm
>>> guessing one of the qtests is timing out with --enable-sanitizers .
> 
>>
>> Building locally:
>> $ CC=clang-11 CXX=clang++-11 ../configure --enable-fuzzing \
>>     --enable-debug --enable-sanitizers
>> $ make check-qtest-i386 check-unit
>>
>> Same as on gitlab, this times out shortly after outputting
>> "sh: 1: exec: ./storage-daemon/qemu-storage-daemon: not found"
>>
>> Manually running qos-test, the same way check-qtest-i386 invokes it:
>>
>> $ QTEST_QEMU_BINARY=./qemu-system-i386 QTEST_QEMU_STORAGE_DAEMON_BINARY=./storage-daemon/qemu-storage-daemon tests/qtest/qos-test --tap -k -m quick < /dev/null
>>
>> # starting vhost-user backend: exec ./storage-daemon/qemu-storage-daemon --blockdev driver=file,node-name=disk0,filename=qtest.XRAzzu --export type=vhost-user-blk,id=disk0,addr.type=unix,addr.path=/tmp/qtest-94561-sock.NdKWpt,node-name=disk0,writable=on,num-queues=1
>> sh: 1: exec: ./storage-daemon/qemu-storage-daemon: not found
>> # starting QEMU: exec ./qemu-system-i386 -qtest unix:/tmp/qtest-94561.sock -qtest-log /dev/null -chardev socket,path=/tmp/qtest-94561.qmp,id=char0 -mon chardev=char0,mode=control -display none -M pc  -device vhost-user-blk-pci,id=drv0,chardev=char1,addr=4.0 -object memory-backend-memfd,id=mem,size=256M,share=on  -M memory-backend=mem -m 256M -chardev socket id=char1,path=/tmp/qtest-94561-sock.NdKWpt  -accel qtest
>>
>> *timeout*
> 
> vhost-user timing out in realize I suspect. I see that as
> an intermittent hang in non-sanitizer configs.
> 
> vhost-user folk: Can we either look at fixing this or else disable
> the test ? (Stack backtraces available in the other email thread.)
> 
> thanks
> -- PMM
> 

Re: [PATCH] fuzz: avoid building twice, when running on gitlab

[Coiby Xu]

On Tue, Aug 10, 2021 at 07:01:45AM +0200, Philippe Mathieu-Daudé wrote:
>+Coiby Xu & qemu-block@

Thanks for adding me to the Cc list.

>
>On 8/9/21 9:36 PM, Peter Maydell wrote:
>> On Mon, 9 Aug 2021 at 20:30, Alexander Bulekov <alxndr@bu.edu> wrote:
>>>
>>> On 210809 1506, Alexander Bulekov wrote:
>>>> On 210809 1925, Peter Maydell wrote:
>>>>> On Mon, 9 Aug 2021 at 12:18, Alexander Bulekov <alxndr@bu.edu> wrote:
>>>>>>
>>>>>> 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>
>>>>>> ---
>>>>>>
>>>>>> From a couple test runs it looks like this can shave off 15-20 minutes.
>>>>>>
>>>>>>  scripts/oss-fuzz/build.sh | 24 +++++++++++++-----------
>>>>>>  1 file changed, 13 insertions(+), 11 deletions(-)
>>>>>
>>>>> I tried a test run with this, but it still hit the 1 hour timeout:
>>>>>
>>>>> https://gitlab.com/qemu-project/qemu/-/pipelines/350387482
>>>>
>>>> It also timed out for me with a 120 minute timeout:
>>>> https://gitlab.com/a1xndr/qemu/-/jobs/1488160601
>>>>
>>>> The log has almost exactly the same number of lines as yours, so I'm
>>>> guessing one of the qtests is timing out with --enable-sanitizers .
>>
>>>
>>> Building locally:
>>> $ CC=clang-11 CXX=clang++-11 ../configure --enable-fuzzing \
>>>     --enable-debug --enable-sanitizers
>>> $ make check-qtest-i386 check-unit
>>>
>>> Same as on gitlab, this times out shortly after outputting
>>> "sh: 1: exec: ./storage-daemon/qemu-storage-daemon: not found"
>>>
>>> Manually running qos-test, the same way check-qtest-i386 invokes it:
>>>
>>> $ QTEST_QEMU_BINARY=./qemu-system-i386 QTEST_QEMU_STORAGE_DAEMON_BINARY=./storage-daemon/qemu-storage-daemon tests/qtest/qos-test --tap -k -m quick < /dev/null
>>>
>>> # starting vhost-user backend: exec ./storage-daemon/qemu-storage-daemon --blockdev driver=file,node-name=disk0,filename=qtest.XRAzzu --export type=vhost-user-blk,id=disk0,addr.type=unix,addr.path=/tmp/qtest-94561-sock.NdKWpt,node-name=disk0,writable=on,num-queues=1
>>> sh: 1: exec: ./storage-daemon/qemu-storage-daemon: not found

The error is different from the previous issue of intermittent hang.
This time the hang is caused by missing qemu-storage-daemon and I
guess the hang could happen reliably each time. The reason of missing 
qemu-storage-daemon is the test doesn't add qemu-storage-daemon as
dependency. If we run `make`, qemu-storage-daemon would be built. But if 
`make check-qtest-i386` is run directly, qemu-storage-daemon wouldn't be 
built. I'll send a patch to make vhost-user-blk test depends on
emu-storage-daemon. 

>>> # starting QEMU: exec ./qemu-system-i386 -qtest unix:/tmp/qtest-94561.sock -qtest-log /dev/null -chardev socket,path=/tmp/qtest-94561.qmp,id=char0 -mon chardev=char0,mode=control -display none -M pc  -device vhost-user-blk-pci,id=drv0,chardev=char1,addr=4.0 -object memory-backend-memfd,id=mem,size=256M,share=on  -M memory-backend=mem -m 256M -chardev socket id=char1,path=/tmp/qtest-94561-sock.NdKWpt  -accel qtest
>>>
>>> *timeout*
>>
>> vhost-user timing out in realize I suspect. I see that as
>> an intermittent hang in non-sanitizer configs.
>>
>> vhost-user folk: Can we either look at fixing this or else disable
>> the test ? (Stack backtraces available in the other email thread.)
>>
>> thanks
>> -- PMM
>>
>

-- 
Best regards,
Coiby