Re: Pre-Depends changed for dpkg on GNU/kFreeBSD
On 04/10/14 22:59, Guillem Jover wrote: I've also fixed an issue when kvm_getprocs(3) does not find any pid, which might have also been involved in the errors you where seeing. I've just observed this on kfreebsd-amd64, in 1.17.13 but it is fixed by upgrading to 1.17.16, thanks. Regards, -- Steven Chamberlain ste...@pyro.eu.org -- To UNSUBSCRIBE, email to debian-bsd-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/54372276.1070...@pyro.eu.org
Re: Pre-Depends changed for dpkg on GNU/kFreeBSD
On Sat, 2014-10-04 at 23:59:25 +0200, Guillem Jover wrote: On Fri, 2014-10-03 at 04:45:13 +0200, Guillem Jover wrote: On Thu, 2014-10-02 at 22:03:37 +0100, Steven Chamberlain wrote: No problem. Does that mean you'd happily revert to using linprocfs? If there's no better option, yes. Right now I'm thinking to merge the attached patch for 1.17.14 as a hotfix, and then switch to a pure sysctl(2) implementation for 1.17.15, so that we can get rid of the libkvm dependency. Otherwise revert to linprocfs. Does that sound good? Something else I just realized now is that libkvm pulls in libbsd and libfreebsd-glue into the pseudo-essential set on GNU/kFreeBSD, which makes the situation a bit worse. So as stated above I'll be switching to sysctl(2) later on, but I'm open to a revert if this is deemed problematic too. Thanks for input Steven! s-s-d only needs struct kinfo_proc (either when using libkvm or sysctl(3)), so I just did a bit of digging and it seems the last time ABI was broken for that struct was around 2005. It seems since then upstream has gotten lots better, and added unused space to be claimed whenever the need arises. I'd expect them to switch to a new MIB in case they break ABI for that struct again, as they did with kinfo_file. I'm thinking that I could also make the code fallback to linprocfs at run-time in case the ki_structsize member differs from the size known at build time, which would give an additional safe guard, in case the above does not hold true. On Sun, 2014-10-05 at 08:54:41 +0200, Sven Joachim wrote: On 2014-10-04 23:59 +0200, Guillem Jover wrote: Something else I just realized now is that libkvm pulls in libbsd and libfreebsd-glue into the pseudo-essential set on GNU/kFreeBSD, which makes the situation a bit worse. Both libbsd and libfreebsd-glue are already pulled in by freebsd-utils. Ah, right, thanks Sven for looking it up, still given that libkvm can be easily avoided, I think it's fine to get rid of it from the pseudo-essential set. Thanks, Guillem -- To UNSUBSCRIBE, email to debian-bsd-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/20141008104716.gb9...@gaara.hadrons.org
Re: Pre-Depends changed for dpkg on GNU/kFreeBSD
On 08/10/14 11:47, Guillem Jover wrote: I'm thinking that I could also make the code fallback to linprocfs at run-time in case the ki_structsize member differs from the size known at build time, which would give an additional safe guard, in case the above does not hold true. In that case please also fall back to linprocfs if kvm is simply unavailable at run-time, e.g. no /dev/kmem, perhaps because in a chroot or jail. Regards, -- Steven Chamberlain ste...@pyro.eu.org -- To UNSUBSCRIBE, email to debian-bsd-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/54352c88@pyro.eu.org
Re: Pre-Depends changed for dpkg on GNU/kFreeBSD
On 2014-10-04 23:59 +0200, Guillem Jover wrote: Something else I just realized now is that libkvm pulls in libbsd and libfreebsd-glue into the pseudo-essential set on GNU/kFreeBSD, which makes the situation a bit worse. Both libbsd and libfreebsd-glue are already pulled in by freebsd-utils. Cheers, Sven -- To UNSUBSCRIBE, email to debian-bsd-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/87k34fngzy@turtle.gmx.de
Re: Pre-Depends changed for dpkg on GNU/kFreeBSD
Hi! On Fri, 2014-10-03 at 04:45:13 +0200, Guillem Jover wrote: After checking the s-s-d and libkvm code again now, it seems it does not require any kvm_read(3) at all, so /dev/mem should not be needed either, could you try the attached patch on a jail? (I need to recover my damaged GNU/kFreeBSD system, as now I only have access to porter boxes. :/ ) I just reinstalled my GNU/kFreeBSD system and I tested it outside a jail, and ktrace tells me it does not access /dev/mem anymore. I've also fixed an issue when kvm_getprocs(3) does not find any pid, which might have also been involved in the errors you where seeing. On Thu, 2014-10-02 at 22:03:37 +0100, Steven Chamberlain wrote: No problem. Does that mean you'd happily revert to using linprocfs? If there's no better option, yes. Right now I'm thinking to merge the attached patch for 1.17.14 as a hotfix, and then switch to a pure sysctl(2) implementation for 1.17.15, so that we can get rid of the libkvm dependency. Otherwise revert to linprocfs. Does that sound good? Something else I just realized now is that libkvm pulls in libbsd and libfreebsd-glue into the pseudo-essential set on GNU/kFreeBSD, which makes the situation a bit worse. So as stated above I'll be switching to sysctl(2) later on, but I'm open to a revert if this is deemed problematic too. Thanks for input Steven! Regards, Guillem -- To UNSUBSCRIBE, email to debian-bsd-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/20141004215925.ga18...@gaara.hadrons.org
Pre-Depends changed for dpkg on GNU/kFreeBSD
Hi! In dpkg 1.17.13 I switched start-stop-daemon on GNU/kFreeBSD to use the native kFreeBSD backend using libkvm instead of using the Linux backend through linprocfs. Requiring linprocfs has always seemed somewhat wrong to me, more so when on FreeBSD procfs is actually optional. This means the library is now part of dpkg's Pre-Depends only on GNU/kFreeBSD. But I forgot to bring it up here as per policy §3.5 before the upload. Doing so now, but if there's no consensus, I'll revert the change. Sorry about that. Thanks, Guillem -- To UNSUBSCRIBE, email to debian-bsd-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/20141002173402.ga15...@gaara.hadrons.org
Re: Pre-Depends changed for dpkg on GNU/kFreeBSD
Hi! On 19:34, Guillem Jover wrote: In dpkg 1.17.13 I switched start-stop-daemon on GNU/kFreeBSD to use the native kFreeBSD backend using libkvm instead of using the Linux backend through linprocfs. Ahhh I did wonder about that. start-stop-daemon had problems inside of jails due to this; KVM needs /dev/mem, and that usually should not be available inside a jail. (netstat for example supports KVM but has a fallback method to be able to still work in jails, I think.) Requiring linprocfs has always seemed somewhat wrong to me, more so when on FreeBSD procfs is actually optional. I'm usually uncomfortable seeing userland use libkvm to look at kernel internals, because unlike FreeBSD, we need to support mismatching versions of kernel and userland (e.g. sid chroot on the stable buildds). Most of Debian GNU userland expects linprocfs so, even though it seems kind of lame to a FreeBSD person, it's useful to us as a psuedo-standard interface that is always available (including jails and any properly- constructed chroot). This means the library is now part of dpkg's Pre-Depends only on GNU/kFreeBSD. But I forgot to bring it up here as per policy §3.5 before the upload. Doing so now, but if there's no consensus, I'll revert the change. Sorry about that. No problem. Does that mean you'd happily revert to using linprocfs? Regards, -- Steven Chamberlain ste...@pyro.eu.org -- To UNSUBSCRIBE, email to debian-bsd-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/20141002210336.ga10...@squeeze.pyro.eu.org
Re: Pre-Depends changed for dpkg on GNU/kFreeBSD
On Thu, 2014-10-02 at 22:03:37 +0100, Steven Chamberlain wrote: On 19:34, Guillem Jover wrote: In dpkg 1.17.13 I switched start-stop-daemon on GNU/kFreeBSD to use the native kFreeBSD backend using libkvm instead of using the Linux backend through linprocfs. Ahhh I did wonder about that. start-stop-daemon had problems inside of jails due to this; KVM needs /dev/mem, and that usually should not be available inside a jail. Oh, feel free to file a bug in the future if any such regressions happen again. After checking the s-s-d and libkvm code again now, it seems it does not require any kvm_read(3) at all, so /dev/mem should not be needed either, could you try the attached patch on a jail? (I need to recover my damaged GNU/kFreeBSD system, as now I only have access to porter boxes. :/ ) Requiring linprocfs has always seemed somewhat wrong to me, more so when on FreeBSD procfs is actually optional. I'm usually uncomfortable seeing userland use libkvm to look at kernel internals, because unlike FreeBSD, we need to support mismatching versions of kernel and userland (e.g. sid chroot on the stable buildds). Sure, although because in this case the code will not end up using kvm_read(3), only stuff returned by sysctl(2), it should be safer, or is that not guaranteed to be compatible either between major versions? Most of Debian GNU userland expects linprocfs so, even though it seems kind of lame to a FreeBSD person, it's useful to us as a psuedo-standard interface that is always available (including jails and any properly- constructed chroot). Still, I'd prefer for s-s-d not to be the one requiring this. This means the library is now part of dpkg's Pre-Depends only on GNU/kFreeBSD. But I forgot to bring it up here as per policy §3.5 before the upload. Doing so now, but if there's no consensus, I'll revert the change. Sorry about that. No problem. Does that mean you'd happily revert to using linprocfs? If there's no better option, yes. Right now I'm thinking to merge the attached patch for 1.17.14 as a hotfix, and then switch to a pure sysctl(2) implementation for 1.17.15, so that we can get rid of the libkvm dependency. Otherwise revert to linprocfs. Does that sound good? Thanks, Guillem diff --git a/utils/start-stop-daemon.c b/utils/start-stop-daemon.c index fd08b94..e97e86c 100644 --- a/utils/start-stop-daemon.c +++ b/utils/start-stop-daemon.c @@ -103,6 +103,11 @@ #ifdef HAVE_KVM_H #include kvm.h +#if defined(OSFreeBSD) +#define KVM_MEMFILE /dev/null +#else +#define KVM_MEMFILE NULL +#endif #endif #ifdef _POSIX_PRIORITY_SCHEDULING @@ -1324,7 +1329,7 @@ pid_is_exec(pid_t pid, const struct stat *esb) char **pid_argv_p; char *start_argv_0_p, *end_argv_0_p; - kd = kvm_openfiles(NULL, NULL, NULL, O_RDONLY, errbuf); + kd = kvm_openfiles(NULL, KVM_MEMFILE, NULL, O_RDONLY, errbuf); if (kd == NULL) errx(1, %s, errbuf); kp = kvm_getprocs(kd, KERN_PROC_PID, pid, nentries); @@ -1413,7 +1418,7 @@ pid_is_child(pid_t pid, pid_t ppid) char errbuf[_POSIX2_LINE_MAX]; pid_t proc_ppid; - kd = kvm_openfiles(NULL, NULL, NULL, O_RDONLY, errbuf); + kd = kvm_openfiles(NULL, KVM_MEMFILE, NULL, O_RDONLY, errbuf); if (kd == NULL) errx(1, %s, errbuf); kp = kvm_getprocs(kd, KERN_PROC_PID, pid, nentries); @@ -1475,7 +1480,7 @@ pid_is_user(pid_t pid, uid_t uid) struct kinfo_proc *kp; char errbuf[_POSIX2_LINE_MAX]; - kd = kvm_openfiles(NULL, NULL, NULL, O_RDONLY, errbuf); + kd = kvm_openfiles(NULL, KVM_MEMFILE, NULL, O_RDONLY, errbuf); if (kd == NULL) errx(1, %s, errbuf); kp = kvm_getprocs(kd, KERN_PROC_PID, pid, nentries); @@ -1562,7 +1567,7 @@ pid_is_cmd(pid_t pid, const char *name) struct kinfo_proc *kp; char errbuf[_POSIX2_LINE_MAX], *process_name; - kd = kvm_openfiles(NULL, NULL, NULL, O_RDONLY, errbuf); + kd = kvm_openfiles(NULL, KVM_MEMFILE, NULL, O_RDONLY, errbuf); if (kd == NULL) errx(1, %s, errbuf); kp = kvm_getprocs(kd, KERN_PROC_PID, pid, nentries); @@ -1735,7 +1740,7 @@ do_procinit(void) char errbuf[_POSIX2_LINE_MAX]; enum status_code prog_status = STATUS_DEAD; - kd = kvm_openfiles(NULL, NULL, NULL, O_RDONLY, errbuf); + kd = kvm_openfiles(NULL, KVM_MEMFILE, NULL, O_RDONLY, errbuf); if (kd == NULL) errx(1, %s, errbuf); kp = kvm_getprocs(kd, KERN_PROC_ALL, 0, nentries);
