[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <d096e463-b135-49b1-884b-a3038723ccdd@app.fastmail.com>
Date: Fri, 30 Jun 2023 20:45:15 +0200
From: "Arnd Bergmann" <arnd@...db.de>
To: "Zhangjin Wu" <falcon@...ylab.org>
Cc: linux-kernel@...r.kernel.org, linux-kselftest@...r.kernel.org,
Thomas Weißschuh <linux@...ssschuh.net>,
Thomas Weißschuh <thomas@...ch.de>,
"Willy Tarreau" <w@....eu>
Subject: Re: [PATCH v3 03/14] selftests/nolibc: add _LARGEFILE64_SOURCE for musl
On Fri, Jun 30, 2023, at 20:01, Zhangjin Wu wrote:
> Hi, Arnd
>
>> On Fri, Jun 30, 2023, at 16:44, Zhangjin Wu wrote:
>> > _GNU_SOURCE Implies _LARGEFILE64_SOURCE in glibc, but in musl, the
>> > default configuration doesn't enable _LARGEFILE64_SOURCE.
>> >
>> > From include/dirent.h of musl, getdents64 is provided as getdents when
>> > _LARGEFILE64_SOURCE is defined.
>> >
>> > #if defined(_LARGEFILE64_SOURCE)
>> > ...
>> > #define getdents64 getdents
>> > #endif
>> >
>> > Let's define _LARGEFILE64_SOURCE to fix up this compile error:
>>
>> I think a better solution would be to use the normal getdents() instead
>> of glibc getdents64(), but then define _FILE_OFFSET_BITS=64 to tell
>> glibc to use the modern version of all filesystem syscalls.
>>
>
> Just checked the getdents manpage[1] and the nolibc code, both of glibc and
> nolibc don't provide the getdents() library routine but both of them provide
> the getdents64(), only musl provide getdents() by default.
>
> And as the manpage shows, it is not easy to call getdents() with glibc, we
> need manually call syscall() and define the 'dirent' struct ourselves:
>
> glibc does not provide a wrapper for getdents(); call getdents()
> using syscall(2). In that case you will need to define the
> linux_dirent or linux_dirent64 structure yourself.
>
> And for nolibc, a getdents64() with linux_dirent64 struct (with int64_t offset)
> is provided, there is either no getdents() currently.
>
> This patch aims to let nolibc-test at least compile for musl and therefore we
> can easily check the new tests for musl, glibc and nolibc together.
Ok, I see. Your current approach should be fine then.
> For the 64bit offset related stuff, we'd better delay it in another patchset
> (part of full rv32 support), which will convert the off_t to 64bit by default.
Right, makes sense.
ARnd
Powered by blists - more mailing lists