[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20241023003958.GU10433@brightrain.aerifal.cx>
Date: Tue, 22 Oct 2024 20:39:58 -0400
From: Rich Felker <dalias@...c.org>
To: Xing Li <lixing@...ngson.cn>
Cc: musl@...ts.openwall.com, wanghongliang@...ngson.cn
Subject: Re: [PATCH] loongarch64: add fstat and newfstatat syscall
support
On Tue, Sep 10, 2024 at 08:09:38AM -0400, Rich Felker wrote:
> On Tue, Sep 10, 2024 at 05:41:12PM +0800, Xing Li wrote:
> > linux kernel add fstat and newfstatat to solve the chromium sandbox, the link as follow:
> > https://lore.kernel.org/loongarch/CAAhV-H7W-Ygn6tXySrip4k3P5xVbVf7GpjOzjXfQvCCbA4r5Wg@mail.gmail.com/T/#t
> >
> > So we add fstat and newfstatat as linux kernel.
> > ---
> > arch/loongarch64/bits/syscall.h.in | 2 ++
> > arch/loongarch64/kstat.h | 21 +++++++++++++++++++++
> > 2 files changed, 23 insertions(+)
> > create mode 100644 arch/loongarch64/kstat.h
> >
> > diff --git a/arch/loongarch64/bits/syscall.h.in b/arch/loongarch64/bits/syscall.h.in
> > index 2afb4ea1..43efb916 100644
> > --- a/arch/loongarch64/bits/syscall.h.in
> > +++ b/arch/loongarch64/bits/syscall.h.in
> > @@ -76,6 +76,8 @@
> > #define __NR_splice 76
> > #define __NR_tee 77
> > #define __NR_readlinkat 78
> > +#define __NR_newfstatat 79
> > +#define __NR_fstat 80
> > #define __NR_sync 81
> > #define __NR_fsync 82
> > #define __NR_fdatasync 83
> > diff --git a/arch/loongarch64/kstat.h b/arch/loongarch64/kstat.h
> > new file mode 100644
> > index 00000000..92625f36
> > --- /dev/null
> > +++ b/arch/loongarch64/kstat.h
> > @@ -0,0 +1,21 @@
> > +struct kstat {
> > + dev_t st_dev;
> > + ino_t st_ino;
> > + mode_t st_mode;
> > + nlink_t st_nlink;
> > + uid_t st_uid;
> > + gid_t st_gid;
> > + dev_t st_rdev;
> > + unsigned long __pad;
> > + off_t st_size;
> > + blksize_t st_blksize;
> > + int __pad2;
> > + blkcnt_t st_blocks;
> > + long st_atime_sec;
> > + long st_atime_nsec;
> > + long st_mtime_sec;
> > + long st_mtime_nsec;
> > + long st_ctime_sec;
> > + long st_ctime_nsec;
> > + unsigned __unused[2];
> > +};
> > --
> > 2.27.0
>
> This can't be accepted as-is because it breaks execution of new
> binaries on old kernels that lack the syscalls. We don't have a
> fallback order that tries statx is legacy stat fails with ENOSYS.
>
> I also don't think this was a good solution to the problem. The same
> problem exists on riscv32 and any future 32-bit archs where legacy
> stat syscalls can't be added (because they're time32-only). statx
> should have just been updated to avoid the need to peek at memory to
> do effective seccomp with it (e.g. by accepting a null pointer for the
> empty path part).
Any thoughts on how this should be fixed? We can add the loongarch
syscall.h entries, but the syscall_arch.h logic would have to disable
them for musl-internal use unless we want to support fallback logic on
archs that might have statx but not fstatat. Since this is not going
to work for most archs, I wonder if we should hold off on a better
solution from kernel side.
Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
