memcpy can fail with rewriter tests

[stanfordcox]

The rewriter tests can fail with dyninst 9.3.1 on
ppc linux. For example: with test5_1 at the point of failure I see:

(gdb) bt
#0 .__memcpy_power7 () at ../sysdeps/powerpc/powerpc64/power7/memcpy.S:101
#1 0x00003fffb74f4ecc in memcpy (__len=1080, __src=0x0,
__dest=) at /usr/include/bits/string3.h:51
#2 memoryTracker (b=0x0, s=1080, a=268569680, this=0x12ee5fa0)
at
/usr/src/debug/dyninst-9.3.1/dyninst-9.3.1/dyninstAPI/src/binaryEdit.h:262
#3 BinaryEdit::createMemoryBackingStore (this=this@entry=0x15b73b00,
obj=)
at
/usr/src/debug/dyninst-9.3.1/dyninst-9.3.1/dyninstAPI/src/binaryEdit.C:768
#4 0x00003fffb74f569c in BinaryEdit::openFile
(file="dyninst_cxx_group_test.dyn_g++_64_none_none", mgr=...,
patch=, member=...) at
/usr/src/debug/dyninst-9.3.1/dyninst-9.3.1/dyninstAPI/src/binaryEdit.C:381
#5 0x00003fffb74a5da8 in BPatch_binaryEdit::BPatch_binaryEdit
(this=0x15b70640,
path=0x10197480 "dyninst_cxx_group_test.dyn_g++_64_none_none",
openDependencies=)
at
/usr/src/debug/dyninst-9.3.1/dyninst-9.3.1/dyninstAPI/src/BPatch_binaryEdit.C:90
#6 0x00003fffb7453d58 in BPatch::openBinary (this=,
path=, openDependencies=)
at
/usr/src/debug/dyninst-9.3.1/dyninst-9.3.1/dyninstAPI/src/BPatch.C:1886
#7 0x00003fffb769fb3c in DyninstComponent::group_setup
(this=0x10497650, group=0x1038e710,
params=std::map with 35 elements = {...}) at
/builddir/build/BUILD/dyninst-9.3.1/testsuite-9.3.0/src/dyninst/dyninst_comp.C:216
#8 0x0000000010015298 in executeGroup (group=0x1038e710,
groups=std::vector of length 1717, capacity 2048 = {...},
param=std::map with 35 elements = {...}) at
/builddir/build/BUILD/dyninst-9.3.1/testsuite-9.3.0/src/test_driver.C:486
#9 0x0000000010016b80 in startAllTests (groups=std::vector of length
1717, capacity 2048 = {...},
param=std::map with 35 elements = {...}) at
/builddir/build/BUILD/dyninst-9.3.1/testsuite-9.3.0/src/test_driver.C:666
#10 0x0000000010017d74 in main (argc=5, argv=0x3ffffffff468)
at
/builddir/build/BUILD/dyninst-9.3.1/testsuite-9.3.0/src/test_driver.C:877

(gdb) fra 3
#3 BinaryEdit::createMemoryBackingStore (this=this@entry=0x15b73b00, obj=)
at /usr/src/debug/dyninst-9.3.1/dyninst-9.3.1/dyninstAPI/src/binaryEdit.C:768
768 regs[i]->getPtrToRawData());
(gdb) list
763 continue;
764 }
765 else {
766 newTracker = new memoryTracker(regs[i]->getMemOffset(),
767 regs[i]->getMemSize(),
768 regs[i]->getPtrToRawData());
769
770 }
771 newTracker->alloced = false;
772 if (!memoryTracker_)

[wrwilliams]

In frame 2 (the memoryTracker constructor, in binaryEdit.h), this:

            b_ = malloc(s_);
            memcpy(b_, b, s_);

should become this:

            if(b) {
                b_ = malloc(s_);
                memcpy(b_, b, s_);
            } else {
                b_ = calloc(1, s_);
            }

Have a local patch in testing, but if you can cross-check on ppc64 while I test, that would be handy.

[stanfordcox]

The above binaryEdit.h patch fixes the case I was seeing.

[wrwilliams]

Fixed by #346.