diff mbox series

[PULL,24/31] fuzz: support for fork-based fuzzing.

Message ID 20200222085030.1760640-25-stefanha@redhat.com
State New
Headers show
Series None | expand

Commit Message

Stefan Hajnoczi Feb. 22, 2020, 8:50 a.m. UTC
From: Alexander Bulekov <alxndr@bu.edu>

fork() is a simple way to ensure that state does not leak in between
fuzzing runs. Unfortunately, the fuzzer mutation engine relies on
bitmaps which contain coverage information for each fuzzing run, and
these bitmaps should be copied from the child to the parent(where the
mutation occurs). These bitmaps are created through compile-time
instrumentation and they are not shared with fork()-ed processes, by
default. To address this, we create a shared memory region, adjust its
size and map it _over_ the counter region. Furthermore, libfuzzer
doesn't generally expose the globals that specify the location of the
counters/coverage bitmap. As a workaround, we rely on a custom linker
script which forces all of the bitmaps we care about to be placed in a
contiguous region, which is easy to locate and mmap over.

Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-id: 20200220041118.23264-16-alxndr@bu.edu
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 tests/qtest/fuzz/Makefile.include |  5 +++
 tests/qtest/fuzz/fork_fuzz.c      | 55 +++++++++++++++++++++++++++++++
 tests/qtest/fuzz/fork_fuzz.h      | 23 +++++++++++++
 tests/qtest/fuzz/fork_fuzz.ld     | 37 +++++++++++++++++++++
 4 files changed, 120 insertions(+)
 create mode 100644 tests/qtest/fuzz/fork_fuzz.c
 create mode 100644 tests/qtest/fuzz/fork_fuzz.h
 create mode 100644 tests/qtest/fuzz/fork_fuzz.ld

Comments

Eric Blake Feb. 22, 2020, 11:34 a.m. UTC | #1
On 2/22/20 2:50 AM, Stefan Hajnoczi wrote:
> From: Alexander Bulekov <alxndr@bu.edu>
> 
> fork() is a simple way to ensure that state does not leak in between
> fuzzing runs. Unfortunately, the fuzzer mutation engine relies on
> bitmaps which contain coverage information for each fuzzing run, and
> these bitmaps should be copied from the child to the parent(where the
> mutation occurs). These bitmaps are created through compile-time
> instrumentation and they are not shared with fork()-ed processes, by
> default. To address this, we create a shared memory region, adjust its
> size and map it _over_ the counter region. Furthermore, libfuzzer
> doesn't generally expose the globals that specify the location of the
> counters/coverage bitmap. As a workaround, we rely on a custom linker
> script which forces all of the bitmaps we care about to be placed in a
> contiguous region, which is easy to locate and mmap over.
> 
> Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
> Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
> Message-id: 20200220041118.23264-16-alxndr@bu.edu
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> ---

Random drive-by observation:

> +++ b/tests/qtest/fuzz/fork_fuzz.ld
> @@ -0,0 +1,37 @@
> +/* We adjust linker script modification to place all of the stuff that needs to
> + * persist across fuzzing runs into a contiguous seciton of memory. Then, it is

section
Stefan Hajnoczi Feb. 24, 2020, 11:35 a.m. UTC | #2
On Sat, Feb 22, 2020 at 05:34:29AM -0600, Eric Blake wrote:
> On 2/22/20 2:50 AM, Stefan Hajnoczi wrote:
> > From: Alexander Bulekov <alxndr@bu.edu>
> > 
> > fork() is a simple way to ensure that state does not leak in between
> > fuzzing runs. Unfortunately, the fuzzer mutation engine relies on
> > bitmaps which contain coverage information for each fuzzing run, and
> > these bitmaps should be copied from the child to the parent(where the
> > mutation occurs). These bitmaps are created through compile-time
> > instrumentation and they are not shared with fork()-ed processes, by
> > default. To address this, we create a shared memory region, adjust its
> > size and map it _over_ the counter region. Furthermore, libfuzzer
> > doesn't generally expose the globals that specify the location of the
> > counters/coverage bitmap. As a workaround, we rely on a custom linker
> > script which forces all of the bitmaps we care about to be placed in a
> > contiguous region, which is easy to locate and mmap over.
> > 
> > Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
> > Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
> > Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
> > Message-id: 20200220041118.23264-16-alxndr@bu.edu
> > Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> > ---
> 
> Random drive-by observation:
> 
> > +++ b/tests/qtest/fuzz/fork_fuzz.ld
> > @@ -0,0 +1,37 @@
> > +/* We adjust linker script modification to place all of the stuff that needs to
> > + * persist across fuzzing runs into a contiguous seciton of memory. Then, it is
> 
> section

Thanks, Eric!

Alex, please send follow-up patches to fix this typo and the 80
character line limit issues identified by patchew (see patch email reply
to this email thread).

Stefan
Alexander Bulekov Feb. 27, 2020, 2:50 a.m. UTC | #3
On 200224 1135, Stefan Hajnoczi wrote:
> On Sat, Feb 22, 2020 at 05:34:29AM -0600, Eric Blake wrote:
> > On 2/22/20 2:50 AM, Stefan Hajnoczi wrote:
> > > From: Alexander Bulekov <alxndr@bu.edu>
> > > 
> > > fork() is a simple way to ensure that state does not leak in between
> > > fuzzing runs. Unfortunately, the fuzzer mutation engine relies on
> > > bitmaps which contain coverage information for each fuzzing run, and
> > > these bitmaps should be copied from the child to the parent(where the
> > > mutation occurs). These bitmaps are created through compile-time
> > > instrumentation and they are not shared with fork()-ed processes, by
> > > default. To address this, we create a shared memory region, adjust its
> > > size and map it _over_ the counter region. Furthermore, libfuzzer
> > > doesn't generally expose the globals that specify the location of the
> > > counters/coverage bitmap. As a workaround, we rely on a custom linker
> > > script which forces all of the bitmaps we care about to be placed in a
> > > contiguous region, which is easy to locate and mmap over.
> > > 
> > > Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
> > > Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
> > > Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
> > > Message-id: 20200220041118.23264-16-alxndr@bu.edu
> > > Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> > > ---
> > 
> > Random drive-by observation:
> > 
> > > +++ b/tests/qtest/fuzz/fork_fuzz.ld
> > > @@ -0,0 +1,37 @@
> > > +/* We adjust linker script modification to place all of the stuff that needs to
> > > + * persist across fuzzing runs into a contiguous seciton of memory. Then, it is
> > 
> > section
> 
> Thanks, Eric!
> 
> Alex, please send follow-up patches to fix this typo and the 80
> character line limit issues identified by patchew (see patch email reply
> to this email thread).

Thank you Eric, Stefan!
Just sent out some fixes.

> Stefan
diff mbox series

Patch

diff --git a/tests/qtest/fuzz/Makefile.include b/tests/qtest/fuzz/Makefile.include
index 8632bb89f4..a90915d56d 100644
--- a/tests/qtest/fuzz/Makefile.include
+++ b/tests/qtest/fuzz/Makefile.include
@@ -2,5 +2,10 @@  QEMU_PROG_FUZZ=qemu-fuzz-$(TARGET_NAME)$(EXESUF)
 
 fuzz-obj-y += tests/qtest/libqtest.o
 fuzz-obj-y += tests/qtest/fuzz/fuzz.o # Fuzzer skeleton
+fuzz-obj-y += tests/qtest/fuzz/fork_fuzz.o
 
 FUZZ_CFLAGS += -I$(SRC_PATH)/tests -I$(SRC_PATH)/tests/qtest
+
+# Linker Script to force coverage-counters into known regions which we can mark
+# shared
+FUZZ_LDFLAGS += -Xlinker -T$(SRC_PATH)/tests/qtest/fuzz/fork_fuzz.ld
diff --git a/tests/qtest/fuzz/fork_fuzz.c b/tests/qtest/fuzz/fork_fuzz.c
new file mode 100644
index 0000000000..2bd0851903
--- /dev/null
+++ b/tests/qtest/fuzz/fork_fuzz.c
@@ -0,0 +1,55 @@ 
+/*
+ * Fork-based fuzzing helpers
+ *
+ * Copyright Red Hat Inc., 2019
+ *
+ * Authors:
+ *  Alexander Bulekov   <alxndr@bu.edu>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+#include "qemu/osdep.h"
+#include "fork_fuzz.h"
+
+
+void counter_shm_init(void)
+{
+    char *shm_path = g_strdup_printf("/qemu-fuzz-cntrs.%d", getpid());
+    int fd = shm_open(shm_path, O_CREAT | O_RDWR, S_IRUSR | S_IWUSR);
+    g_free(shm_path);
+
+    if (fd == -1) {
+        perror("Error: ");
+        exit(1);
+    }
+    if (ftruncate(fd, &__FUZZ_COUNTERS_END - &__FUZZ_COUNTERS_START) == -1) {
+        perror("Error: ");
+        exit(1);
+    }
+    /* Copy what's in the counter region to the shm.. */
+    void *rptr = mmap(NULL ,
+            &__FUZZ_COUNTERS_END - &__FUZZ_COUNTERS_START,
+            PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
+    memcpy(rptr,
+           &__FUZZ_COUNTERS_START,
+           &__FUZZ_COUNTERS_END - &__FUZZ_COUNTERS_START);
+
+    munmap(rptr, &__FUZZ_COUNTERS_END - &__FUZZ_COUNTERS_START);
+
+    /* And map the shm over the counter region */
+    rptr = mmap(&__FUZZ_COUNTERS_START,
+            &__FUZZ_COUNTERS_END - &__FUZZ_COUNTERS_START,
+            PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED, fd, 0);
+
+    close(fd);
+
+    if (!rptr) {
+        perror("Error: ");
+        exit(1);
+    }
+}
+
+
diff --git a/tests/qtest/fuzz/fork_fuzz.h b/tests/qtest/fuzz/fork_fuzz.h
new file mode 100644
index 0000000000..9ecb8b58ef
--- /dev/null
+++ b/tests/qtest/fuzz/fork_fuzz.h
@@ -0,0 +1,23 @@ 
+/*
+ * Fork-based fuzzing helpers
+ *
+ * Copyright Red Hat Inc., 2019
+ *
+ * Authors:
+ *  Alexander Bulekov   <alxndr@bu.edu>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+#ifndef FORK_FUZZ_H
+#define FORK_FUZZ_H
+
+extern uint8_t __FUZZ_COUNTERS_START;
+extern uint8_t __FUZZ_COUNTERS_END;
+
+void counter_shm_init(void);
+
+#endif
+
diff --git a/tests/qtest/fuzz/fork_fuzz.ld b/tests/qtest/fuzz/fork_fuzz.ld
new file mode 100644
index 0000000000..b23a59f194
--- /dev/null
+++ b/tests/qtest/fuzz/fork_fuzz.ld
@@ -0,0 +1,37 @@ 
+/* We adjust linker script modification to place all of the stuff that needs to
+ * persist across fuzzing runs into a contiguous seciton of memory. Then, it is
+ * easy to re-map the counter-related memory as shared.
+*/
+
+SECTIONS
+{
+  .data.fuzz_start : ALIGN(4K)
+  {
+      __FUZZ_COUNTERS_START = .;
+      __start___sancov_cntrs = .;
+      *(_*sancov_cntrs);
+      __stop___sancov_cntrs = .;
+
+      /* Lowest stack counter */
+      *(__sancov_lowest_stack);
+  }
+  .data.fuzz_ordered :
+  {
+      /* Coverage counters. They're not necessary for fuzzing, but are useful
+       * for analyzing the fuzzing performance
+       */
+      __start___llvm_prf_cnts = .;
+      *(*llvm_prf_cnts);
+      __stop___llvm_prf_cnts = .;
+
+      /* Internal Libfuzzer TracePC object which contains the ValueProfileMap */
+      FuzzerTracePC*(.bss*);
+  }
+  .data.fuzz_end : ALIGN(4K)
+  {
+      __FUZZ_COUNTERS_END = .;
+  }
+}
+/* Dont overwrite the SECTIONS in the default linker script. Instead insert the
+ * above into the default script */
+INSERT AFTER .data;