[Webkit-unassigned] [Bug 146793] New: [GTK] Web processes should not have unlimited access to memory

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=146793

            Bug ID: 146793
           Summary: [GTK] Web processes should not have unlimited access
                    to memory
    Classification: Unclassified
           Product: WebKit
           Version: 528+ (Nightly build)
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: Normal
          Priority: P2
         Component: WebKit Gtk
          Assignee: webkit-unassigned at lists.webkit.org
          Reporter: mcatanzaro at igalia.com

Web processes should not have unlimited access to memory.

A quick example of why not: open a bunch of bugs on Red Hat Bugzilla in new tabs. [1] If you open ~20 bugs, at least one (usually about three for me) should run out of control, allocating memory until your computer begins to swap excessively. Mine hangs for an hour, forcing me to power off. This happens to me multiple times per day. (Curious if any folks from Apple can reproduce this in Safari.) Even though it's a bug that should be fixed, other similar bugs exist, like [2] and [3], or might exist in the future, so we should be robust to this by setting a memory limit for the web process.

On Linux the options for limiting memory are RLIMIT_AS and RLIMIT_DATA. RLIMIT_RSS exists but doesn't do anything; even if it worked, it wouldn't be useful for the problem I want to solve, which is swapping. RLIMIT_DATA is not useful in practice since it doesn't affect mmap(), which is what we use to allocate memory, and also what malloc() uses. So RLIMIT_AS is the only option. I've picked a 5 GB address space limit, which should work for a long time in the future. In practice, this limits web processes to ~1.4 GB of memory on my machine. A well-behaved web process uses 50-100 MiB (although I've seen them go as high as ~250 MiB after a sufficiently long time loading many pages and leaking much memory), so this is so wildly higher than necessary that it shouldn't hurt normal operation. (Of course, we might need to increase it in the distant future.)

[1] https://bugzilla.redhat.com/buglist.cgi?component=epiphany&product=Fedora
[2] https://bugs.webkit.org/show_bug.cgi?id=139847
[3] https://bugs.webkit.org/show_bug.cgi?id=126122

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.webkit.org/pipermail/webkit-unassigned/attachments/20150709/b53667cf/attachment.html>