Hi, thanks. I did not review in detail the code below, but since it's a
separate tool,
which obviously you will care about and continue to improve as you discover
bugs, I don't mind committing it as-is and
letting you send followup patches later. So I'll commit it now.
--
Nadav Har'El
On Sat, May 11, 2019 at 3:15 PM Waldek Kozaczuk
wrote:
> Or it could be that at this layer - musl - it is not handled but instead
> at the syscall level in Linux. Either way I think we should fix it
> somewhere.
>
Yes. I prefer that we fix it in the level of the open() system call, please.
In
Or it could be that at this layer - musl - it is not handled but instead at the
syscall level in Linux. Either way I think we should fix it somewhere.
Sent from my iPhone
> On May 11, 2019, at 08:06, Nadav Har'El wrote:
>
>> On Sat, May 11, 2019 at 2:40 PM Waldemar Kozaczuk
>> wrote:
>>
On Sat, May 11, 2019 at 2:40 PM Waldemar Kozaczuk
wrote:
> Signed-off-by: Waldemar Kozaczuk
> ---
> libc/stdio/fopen.c | 5 +
> 1 file changed, 5 insertions(+)
>
> diff --git a/libc/stdio/fopen.c b/libc/stdio/fopen.c
> index 83452407..c6053c21 100644
> --- a/libc/stdio/fopen.c
> +++
Please see my other explanation in the thread email related to the patch
“Enhance getopt functions to work with PIEs”. There I am talking about how
unmodified iperf was failing and it was because of nullptr filename. Based on
strace of iperf3 on Linux fopen on Linux seems to return EFAULT.
Even though the test app, golang and node apps (which use local TLS) work
fine with this patch, the rust programs built as pies do not. Rust uses
thread local heavily so one has to bump up the size of the TLS reservation
in loader.ld.
Here is an example of a crash (seems to be in thread local
Did not mean to hijack this patch thread but it seemed related. In reality
the issue with running unmodified iperf was caused by a bug in our fopen
impl whch I sent a patch for.
On Friday, May 10, 2019 at 4:58:45 PM UTC-4, Waldek Kozaczuk wrote:
>
> When trying to run unmodified iperf3 3.6
Forgot to mention in the patch description that this tool is NOT meant to
install the artifacts to generate the manifest.
On Thursday, May 9, 2019 at 11:53:40 PM UTC-4, Waldek Kozaczuk wrote:
>
> Building OSv images with apps that have to built from source from scratch
> is not fun
> at best.
Signed-off-by: Waldemar Kozaczuk
---
libc/stdio/fopen.c | 5 +
1 file changed, 5 insertions(+)
diff --git a/libc/stdio/fopen.c b/libc/stdio/fopen.c
index 83452407..c6053c21 100644
--- a/libc/stdio/fopen.c
+++ b/libc/stdio/fopen.c
@@ -10,6 +10,11 @@ FILE *fopen(const char *restrict filename,