Re: [Qemu-devel] [PATCH] fcntl64 fix

[Stuart Anderson]

OK, I think I finally have it all sorted out. Sorry if I sounded dense
along the way.. there were multiple variable, which increases the number
of possible combinations quickly.

The patch from Kirill is needed, and makes things better. One thing I
notice with it is that we now handle TARGET_F_GETLK64 in two places,
first in the case for TARGET_NR_fcntl64 (around line 4300), and then
again in do_fcntl(), which is called in the default case of the first
location. Once difference between the two locations is wether or not
the case for EABI is handled.

In addition to Kirill's patch, my original patch for target_eabi_flock64
is still needed as well as an expanded version of the revised patch I
sent later that does target->host strcture mapping for the F_GETLK*
cases.

I have used the fcntl test sets out of the Linux Test Projects to
measure with an without the different parts of these patches. With
the entire set (Which is attached), 16 of the 18 test sets pass
completely, and a significant portion of test14 (one of the two that
don't pass completely) passes as well. The tests in test14 that fail
may be do to a problem with a syscall other than fcntl(), but I haven't
completely resulved it yet. Without my portion of the patch, the results
are much worse (maybe half-ish are passing).

There is something interesting about test18 (the other one that doesn't
pass). It intentionally passes in a bad value (-1) as the 3rd argument
to fcntl(). It is testing wether it will get EFAULT. With these fixes,
qemu will SEGV as it tries to convert the struct flock (or struct
flock64) from target->host, and encounters the bad address that was
passed in. The initial SEGV is caught, but the handler for it then
SEGVs again. Ideally, we could detect that we are inside an emulated
system call, and be able to just return the EFAULT.

I ran the LTP tests for both old ABI and EABI, and got the same results.


Attached is the combined patch for fcntl().




                                Stuart

Stuart R. Anderson                               address@hidden
Network & Software Engineering                   http://www.netsweng.com/
1024D/37A79149:                                  0791 D3B8 9A4C 2CDC A31F
                                                 BD03 0A62 E534 37A7 9149

patch
Description: fcntl() combined patch