Hello,
My previous analysis of this issue is wrong. Here's the output of nm:
$ nm guest/guest_pre_init.o
0000000000000310 D _binary_guest_pre_init_end
0000000000000310 A _binary_guest_pre_init_size
0000000000000000 D _binary_guest_pre_init_start
$ nm lkvm | grep guest_pre_init
0000000000233788 D _binary_guest_pre_init_end
0000000000000310 A _binary_guest_pre_init_size
0000000000233478 D _binary_guest_pre_init_start
According to nm manpage:
"A" The symbol's value is absolute, and will not be changed by
further linking.
But debugging lkvm with gdb clearly shows that the value of
_binary_guest_pre_init_size changes and is relocated:
$ rm -rf ~/.lkvm/blah/
$ gdb -q lkvm
gdb$ x/2i kvm_setup_guest_init
0x8c60 <kvm_setup_guest_init>: push rbx
0x8c61 <kvm_setup_guest_init+1>: lea
rcx,[rip+0xffffffffffff76a8] # 0x310
gdb$ r setup blah
[code]
=> 0x55555555cc60 <kvm_setup_guest_init>: push rbx
0x55555555cc61 <kvm_setup_guest_init+1>: lea
rcx,[rip+0xffffffffffff76a8] # 0x555555554310
Breakpoint 1, kvm_setup_guest_init (guestfs_name=0x7fffffffe33a
"blah") at builtin-setup.c:156
So I finally think this is an issue in gcc...
On 01/03/2017 03:57 PM, G. Campana wrote:
> Hi,
>
> I just noticed that a typo is present in the original subject (kvmtool
> is misspelled.) I assume that some of you filter messages according to
> their subject, hence this new mail.
>
> Sorry for the spam.
>
> -------- Forwarded Message --------
> Subject: kmvtool: invalid embedded ELF binaries' size
> Date: Mon, 2 Jan 2017 15:04:50 +0100
> To: kvm@xxxxxxxxxxxxxxx
>
> Hi,
>
> A user of the NoFear project reported a bug (
> https://github.com/cappsule/nofear/issues/6 ) suggesting that kvmtool is
> broken if compiled with a specific gcc version. This issue can be
> reproduced with gcc version 6.2.1 20161124 (Debian 6.2.1-5), but not
> with gcc version 6.1.1 20160802 (Debian 6.1.1-11).
>
> The function extract_file() writes embedded ELF files to the filesystem:
>
> static int extract_file(const char *guestfs_name, const char *filename,
> const void *data, const void *_size)
> ...
> ret = xwrite(fd, data, (size_t)_size);
> ...
> }
>
> extern char _binary_guest_init_start;
> extern char _binary_guest_init_size;
> extern char _binary_guest_pre_init_start;
> extern char _binary_guest_pre_init_size;
>
> I didn't manage to find how _binary_guest_init_size and
> _binary_guest_pre_init_size are created during the link step, but a
> quick debug session shows that they're 4 bytes wide:
>
> (gdb) disass kvm_setup_guest_init
> Dump of assembler code for function kvm_setup_guest_init:
> => 0x00005599b846dc60 <+0>: push rbx
> 0x00005599b846dc61 <+1>: lea rcx,[rip+0xffffffffffff76a8]
> # 0x5599b8465310
> 0x00005599b846dc68 <+8>: lea rdx,[rip+0x22a809] #
> 0x5599b8698478
> 0x00005599b846dc6f <+15>: lea rsi,[rip+0x18a2c] #
> 0x5599b84866a2
> 0x00005599b846dc76 <+22>: mov rbx,rdi
> 0x00005599b846dc79 <+25>: call 0x5599b846d9e0 <extract_file>
> ...
> (gdb) x/xw 0x5599b8465310
> 0x5599b8465310: 0x000004c2
> (gdb) x/xg 0x5599b8465310
> 0x5599b8465310: 0x00000012000004c2
>
> Casting _size to size_t in extract_file() is wrong because _size is 4
> bytes wide. On x64, xwrite thus fails when the 4 bytes following
> _binary_guest_init_size/_binary_guest_pre_init_size are different than 0.
>
> I think that casting _size to unsigned int should fix this issue, but I
> would appreciate if someone can explain how _binary_guest_init_size and
> _binary_guest_pre_init_size are produced by the linker.
>
> Thanks
>
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
