Fix double-free corruption
authorPedro Alves <palves@redhat.com>
Tue, 17 Oct 2017 11:41:00 +0000 (12:41 +0100)
committerPedro Alves <palves@redhat.com>
Tue, 17 Oct 2017 11:41:00 +0000 (12:41 +0100)
commit7b7009999ab8daac9db776c850b7df6e1f586334
tree5423e586e7d59f5072bc3d50246ad493ee957df1
parentd3037ba6a398d37585b3d34ff9ed439848ba98a1
Fix double-free corruption

Fixes a double-free regression introduced by commit b7b030adc405
("Return unique_xmalloc_ptr from target_read_stralloc"):

gdb.sum:
  Running src/gdb/testsuite/gdb.base/catch-syscall.exp ...
  ERROR: Process no longer exists

Valgrind shows:

  (gdb) catch syscall
  ==3687== Thread 1:
  ==3687== Invalid free() / delete / delete[] / realloc()
  ==3687==    at 0x4C29CF0: free (vg_replace_malloc.c:530)
  ==3687==    by 0x610862: xfree(void*) (common-utils.c:101)
  ==3687==    by 0x440D5D: gdb::xfree_deleter<char>::operator()(char*) const (gdb_unique_ptr.h:34)
  ==3687==    by 0x446CC6: std::unique_ptr<char, gdb::xfree_deleter<char> >::reset(char*) (unique_ptr.h:344)
  ==3687==    by 0x81BE50: xml_fetch_content_from_file(char const*, void*) (xml-support.c:1042)
  ==3687==    by 0x81DA86: xml_init_syscalls_info(char const*) (xml-syscall.c:366)
  ==3687==    by 0x81DBDD: init_syscalls_info(gdbarch*) (xml-syscall.c:398)
  ==3687==    by 0x81E131: get_syscall_by_number(gdbarch*, int, syscall*) (xml-syscall.c:599)
  ==3687==    by 0x5BE86F: catch_syscall_command_1(char*, int, cmd_list_element*) (break-catch-syscall.c:481)
  ==3687==    by 0x4B46B1: do_sfunc(cmd_list_element*, char*, int) (cli-decode.c:138)
  ==3687==    by 0x4B76B8: cmd_func(cmd_list_element*, char*, int) (cli-decode.c:1952)
  ==3687==    by 0x7E91C7: execute_command(char*, int) (top.c:615)
  ==3687==  Address 0x14332ae0 is 0 bytes inside a block of size 4,096 free'd
  ==3687==    at 0x4C2AB8B: realloc (vg_replace_malloc.c:785)
  ==3687==    by 0x610792: xrealloc (common-utils.c:62)
  ==3687==    by 0x81BE3E: xml_fetch_content_from_file(char const*, void*) (xml-support.c:1042)
  ==3687==    by 0x81DA86: xml_init_syscalls_info(char const*) (xml-syscall.c:366)
  ==3687==    by 0x81DBDD: init_syscalls_info(gdbarch*) (xml-syscall.c:398)
  ==3687==    by 0x81E131: get_syscall_by_number(gdbarch*, int, syscall*) (xml-syscall.c:599)
  ==3687==    by 0x5BE86F: catch_syscall_command_1(char*, int, cmd_list_element*) (break-catch-syscall.c:481)
  ==3687==    by 0x4B46B1: do_sfunc(cmd_list_element*, char*, int) (cli-decode.c:138)
  ==3687==    by 0x4B76B8: cmd_func(cmd_list_element*, char*, int) (cli-decode.c:1952)
  ==3687==    by 0x7E91C7: execute_command(char*, int) (top.c:615)
  ==3687==    by 0x6A422D: command_handler(char*) (event-top.c:583)
  ==3687==    by 0x6A45F2: command_line_handler(char*) (event-top.c:773)
  [...]

The problem is that if xrealloc decides it needs a new memory block,
it frees the previous block/pointer, and then text.reset() frees it
again.

gdb/ChangeLog:
2017-10-17  Pedro Alves  <palves@redhat.com>

* xml-support.c (xml_fetch_content_from_file): Call
unique_ptr::release() instead unique_ptr::get() when passing
through xrealloc.
gdb/ChangeLog
gdb/xml-support.c