Re: [PATCH] Syscalls: reboot: Add options to the reboot syscall to remount filesystems ro

[Ken Sumrall <ksumrall@android.com>]

Writing a single byte to /proc/sysrq-trigger is an asynchronous
operation, with no obvious way to be informed that it has completed
the remount.  I did actually try this at first, but the reboot
happened before the remount finished.  I could of course add a sleep
of a few seconds, but just how long to wait is not obvious, nor is it
a guarantee that the remount will always finish in the time chosen.
I'm heading down the path of reading /proc/mounts and remounting all
read-wirte filesystems backed by a block device as read-only.

___
Ken


On Thu, Mar 3, 2011 at 3:36 PM, Linus Torvalds
<torvalds@linux-foundation.org> wrote:
> On Thu, Mar 3, 2011 at 10:28 AM, Mark Lord <kernel@teksavvy.com> wrote:
>> On 11-03-03 12:45 PM, Linus Torvalds wrote:
>>>
>>> If you can change whatever user-land process that does the reboot
>>> system call (and clearly you can, since you're adding new commands and
>>> using those), then why the heck don't you just do the remount-ro from
>>> that same user land?
>>
>> Is there a system call for emergency_remount_*() ?
>> I've been writing to /proc/proc/sysrq-trigger to accomplish this.
>
> So I don't know what this has to do with "emergency_remount()" - we're
> talking about a regular controlled shutdown/reset. The fact that the
> patch used the emergency_remount() code seems to be purely an
> implementation issue, nothing more.
>
> And while sysrq-trigger certainly works (when it's enabled, but that's
> true of /proc too, of course), I do think it's a rather odd way of
> solving the problem, when the simple "just remount read-only" is what
> the code actually _wants_ to do.
>
> But you're certainly right that it takes less code to open
> /proc/sysrq-trigger and writing a single byte to it than it does to do
> the straightforward "let's just do the normal mount thing".
>
>                             Linus
>