Date: Fri, 9 Mar 2007 16:08:46 +0200 From: Nikos Vassiliadis <nvass@teledomenet.gr> To: Modulok <modulok@gmail.com> Cc: freebsd-questions@freebsd.org Subject: Re: Kill a hanged disk i/o process... Message-ID: <200703091608.47529.nvass@teledomenet.gr> In-Reply-To: <64c038660703090528k6974a8c3odf87e4d8a3b368c9@mail.gmail.com> References: <64c038660703080349t3311fa22lf8e6ba736db330ed@mail.gmail.com> <200703091139.51272.nvass@teledomenet.gr> <64c038660703090528k6974a8c3odf87e4d8a3b368c9@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Friday 09 March 2007 15:28, Modulok wrote: > Thank you for your reply, it was quite informative and very much > appreciated, but the underlying question remains un-answered: > > How do you kill a hanged process that (seemingly) cannot be killed because > of the two conditions below? > > -It's hanged, so it's not ever going to self terminate. > -It's a disk i/o process so not even root can kill it. > As I said before disk I/O is irrelevant. > The gentle shutdown solution doesn't work: Even during shutdown the process > cannot be killed: it's hanged, it's disk i/o. > > How do you kill an un-killable process? What makes you believe there is another official way to kill a process? Perhaps you should ask "How do I work-around a situation where my rm, cp, whatever hang forever?", if that's what you are looking for. > -Modulok- > > > On 3/9/07, Nikos Vassiliadis <nvass@teledomenet.gr> wrote: > > > > On Thursday 08 March 2007 13:49, Modulok wrote: > > > To the best of my knowledge, most processes can be killed explicitly > > > by "kill -s KILL;" There are a few which cannot, such as disk i/o > > > processes. The idea here is data integrity. > > > > A process might be in cannot-be-killed condition while > > in kernel e.g. during a system call. That has to do with > > the completion of the system call, not with data integrity. > > The kernel tries to complete what was asked for. > > > > Also, Killing a process with SIGKILL is far from safe. To put > > it in another way "data integrity" can be guaranteed only > > by the program itself. For example it could have a defined > > behavior when it is signaled by e.g. SIGTERM, for example > > clean up data and exit. Or not. It's up to the programmer. > > Sending a SIGKILL will not give that chance. SIGKILL can > > not be handled. It will be terminated as soon as possible. > > > > Also, separate the meanings "data integrity" and "filesystem > > data integrity". The filesystem will be in fine condition when > > a process gets killed by SIGKILL during file I/O, the data in > > the file most probably not. > > > > > > > > On the rare occasion however, (when attempting to recover data from > > > corrupt disks for example), I've had a process invoked by the "cp" > > > command, hang. This poses a significant problem as these processes are > > > disk i/o processes, and as such cannot be terminated (even by root). > > > So, other than physically hitting the reset button on the case, is > > > there a more eloquent method of forcefully halting a hanged disk i/o > > > process? The idea of "you don't want to terminate a disk i/o process, > > > it could corrupt the data" isn't really a good argument, because if > > > the process hangs and I have to punch the reset button anyway what's > > > the difference? > > > > "Pressing the button" will leave your filesystem in a undefined state, > > you are risking filesystem integrity. Keep in mind that while in use > > (open files etc) a filesystem cannot be unmounted. Anyway, try to shut > > the computer down, it's far more gentle than pressing the button. At > > least the rest of the filesystems will be cleanly unmounted. > > > > Is there something in particular you want to achieve? > > > > Nikos > > >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200703091608.47529.nvass>