[NaCl SDK] Remove syscalls wrappers for chmod and unlink.

[NaCl SDK] Remove syscalls wrappers for chmod and unlink.

This allows applications that are linked with nacl_io
to use the libc-supplied versions of these functions
when nacl_io is not initialized.

Fix some issues with using nacl_io without ppapi present.

Add some tests for using syscalls when nacl_io is not
initialized.

R=binji@chromium.org

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=291036

[binji, 2014-08-20 18:13:35]

lgtm

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
File native_client_sdk/src/libraries/nacl_io/kernel_intercept.cc (right):

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
native_client_sdk/src/libraries/nacl_io/kernel_intercept.cc:147: // succeed.  If
we haven't initialized kernel-intercept yet, they try
s/they/then/

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
File native_client_sdk/src/libraries/nacl_io/kernel_intercept.h (right):

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
native_client_sdk/src/libraries/nacl_io/kernel_intercept.h:47: * pushed state is
restored in the next call to ki_uninit().
or by calling ki_pop_state_for_testing().

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
File native_client_sdk/src/libraries/nacl_io/kernel_wrap_glibc.cc (right):

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
native_client_sdk/src/libraries/nacl_io/kernel_wrap_glibc.cc:526: if
(!REAL(func))              \
I'm not sure I understand. Why wouldn't there be a real pointer for this symbol?

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
File native_client_sdk/src/libraries/nacl_io/kernel_wrap_newlib.cc (right):

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
native_client_sdk/src/libraries/nacl_io/kernel_wrap_newlib.cc:362: int
_real_getcwd(char* pathname, size_t len) {
bionic too?

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/t...
File native_client_sdk/src/tests/nacl_io_test/kernel_wrap_test.cc (right):

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/t...
native_client_sdk/src/tests/nacl_io_test/kernel_wrap_test.cc:658:
ASSERT_NE((char*)NULL, getcwd(dir, 1024));
PATH_MAX?

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/t...
native_client_sdk/src/tests/nacl_io_test/kernel_wrap_test.cc:660: const char*
suffix = "nacl_io_test";
hm, seems like a fragile test. There's no reason we couldn't run this test from
another directory.

[Sam Clegg, 2014-08-21 10:19:34]

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
File native_client_sdk/src/libraries/nacl_io/kernel_intercept.cc (right):

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
native_client_sdk/src/libraries/nacl_io/kernel_intercept.cc:147: // succeed.  If
we haven't initialized kernel-intercept yet, they try
On 2014/08/20 18:13:35, binji wrote:
> s/they/then/

Done.

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
File native_client_sdk/src/libraries/nacl_io/kernel_intercept.h (right):

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
native_client_sdk/src/libraries/nacl_io/kernel_intercept.h:47: * pushed state is
restored in the next call to ki_uninit().
On 2014/08/20 18:13:35, binji wrote:
> or by calling ki_pop_state_for_testing().

Done.

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
File native_client_sdk/src/libraries/nacl_io/kernel_wrap_glibc.cc (right):

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
native_client_sdk/src/libraries/nacl_io/kernel_wrap_glibc.cc:526: if
(!REAL(func))              \
On 2014/08/20 18:13:35, binji wrote:
> I'm not sure I understand. Why wouldn't there be a real pointer for this
symbol?

The real pointers will be NULL if the corresponding IRT interface doesn't exist.

The getcwd implementation in the C library checks for the NULL case and we also
need to do this if we use the real symbols directly.

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
File native_client_sdk/src/libraries/nacl_io/kernel_wrap_newlib.cc (right):

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/l...
native_client_sdk/src/libraries/nacl_io/kernel_wrap_newlib.cc:362: int
_real_getcwd(char* pathname, size_t len) {
On 2014/08/20 18:13:35, binji wrote:
> bionic too?

Done..   looks like we don't have a trybot that even builds the bionic SDK do
we?

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/t...
File native_client_sdk/src/tests/nacl_io_test/kernel_wrap_test.cc (right):

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/t...
native_client_sdk/src/tests/nacl_io_test/kernel_wrap_test.cc:658:
ASSERT_NE((char*)NULL, getcwd(dir, 1024));
On 2014/08/20 18:13:35, binji wrote:
> PATH_MAX?

Done.

https://codereview.chromium.org/474433005/diff/100001/native_client_sdk/src/t...
native_client_sdk/src/tests/nacl_io_test/kernel_wrap_test.cc:660: const char*
suffix = "nacl_io_test";
On 2014/08/20 18:13:35, binji wrote:
> hm, seems like a fragile test. There's no reason we couldn't run this test
from
> another directory.

Got any better ideas?  My goal here is to check that the real getcwd is being
used ..  Maybe I should just skip this test.  Its testing something that will
most likely cause gtest itself to crash on startup if it fails anyway.

[Sam Clegg, 2014-08-21 10:20:01]

Committed patchset #5 manually as 291036 (presubmit successful).