Date: Sun, 17 Mar 2013 17:57:43 +0200 From: Konstantin Belousov <kostikbel@gmail.com> To: Pawel Jakub Dawidek <pjd@FreeBSD.org> Cc: freebsd-arch@FreeBSD.org Subject: Re: chflags(2)'s flags argument. Message-ID: <20130317155743.GR3794@kib.kiev.ua> In-Reply-To: <20130317111112.GC1364@garage.freebsd.pl> References: <20130317003559.GA1364@garage.freebsd.pl> <20130317064123.GM3794@kib.kiev.ua> <20130317111112.GC1364@garage.freebsd.pl>
next in thread | previous in thread | raw e-mail | index | archive | help
[-- Attachment #1 --] On Sun, Mar 17, 2013 at 12:11:12PM +0100, Pawel Jakub Dawidek wrote: > On Sun, Mar 17, 2013 at 08:41:23AM +0200, Konstantin Belousov wrote: > > On Sun, Mar 17, 2013 at 01:35:59AM +0100, Pawel Jakub Dawidek wrote: > > > Hi. > > > > > > Currently this is a bit messy: chflags(2) and fchflags(2) take 'flags' > > > argument of type unsigned long and lchflags(2) takes the same argument > > > of type int. At least this is what you can see in manual page and in > > > prototypes of those functions in sys/stat.h. > > > > > > However all of those syscalls are defined in syscalls.master to take the > > > 'flags' argument of type int and this is what they use in kernel. > > > > > > I'd like to proposed the following patch: > > > > > > http://people.freebsd.org/~pjd/patches/chflags_int.patch > > > > > > It changes type of the 'flags' argument from unsigned long to int where > > > possible. > > > > > > I believe this change won't break ABI, as the syscalls (apart from the > > > prototypes) already expect int and I hope in doesn't break API in any > > > really visible and destructive way. If you think otherwise, let me know. > > The patch seems to keep ABI intact for all useful purposes, at least > > on all architectures the FreeBSD supports. A FreeBSD architecture where > > sizeof(int) != sizeof(long), uses register calling conventions. > > Actually I'd rephrase that. If I understand correctly, because we use > register calling conventions on architectures where sizeof(int) != > sizeof(long), this mess is working correctly now. Remember that syscalls > are defined to take int, but prototypes say unsigned long. It needs some untangling. The ABI exported by libc is what I care about when referring to the ABI. And this is the ABI which is not broken due to the reason I stated above. The interface exported by the kernel for consumption of the libc syscall wrappers is dufferent from the normal ABI on the architecture. > > > Please note that API != ABI, and you found a case where the API > > is broken indeed by your change. > > I know it can break API in some rare cases like in chflags(1), but it > results in compilation error (at least with the compilation flags we > use), so can be easly spotted and fixed, hopefully: > > /usr/home/pjd/p4/capkern/bin/chflags/chflags.c: In function 'main': > /usr/home/pjd/p4/capkern/bin/chflags/chflags.c:120: warning: assignment from incompatible pointer type > Project aims to maintain better compatibility then to claim that the changes could be 'spotted'. [-- Attachment #2 --] -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.19 (FreeBSD) iQIcBAEBAgAGBQJRRef3AAoJEJDCuSvBvK1BYtgP/0xDdOCWPvx1o0SOI7iLZCCK L4qeIWieXqmWKmjDmHnPehj4G/AQAXkVr3+jOvUEHEjiP1gFANy6wSIwPDW8l79n WZDmllfOSFTKGIu5khDLKDX25LaZxRtGQm1CJEbiKBVM5ofXCq/jcBvsu+X+GHTx /xqB31JSLF7Obo6cVOljLE+ZkAO17vLptUBhjuH2D6CdeIw8vX3RUvbyD68rlMgV 16H31ViOuK1WJVVl5IS97hBEtrYILtXye0hr9o9QUOywFhJNoZaqRoqMFbl3TDm4 xP/g1nXEM2ncovkwpx9T3FQVNRk9KzELTpmknyYB42P5WPQhOf2d7ieaPpdjV7be iBgT5XZsj1rXALCbUi703+NO5b5xIhFU0PW2iqA4xG9orDkcVDhmSS+ZOcWIwVPe pj/gjaNV2j/eRSdQvFpOv/epXME4zAO5q6SOs8wMMBNFmM4F5Um3mf1O4VN5o2ak SXnorZ8OJSnrmRR39l7/3cbf+wdv1zX0vOJyDsbh4maPAxJaOWI0mECmXRqSHZO6 r7Fg3GsHIWvLrroS7WDnUp83s7IBdVbTidayesv+8d+iy/VMmjQQuV6ZN9StGCsI pQu9WdBd4dpxi2gKPKLZkXQxJCEov3ijITOFuS8Z28Rcm/7SYdsnBEhHItWB/E2X FM6i1b3u9pmQ73AXfqPZ =vsac -----END PGP SIGNATURE-----
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20130317155743.GR3794>