[tracing] Send smaps file desciptor to child process for tracing

[tracing] Send smaps file desciptor to child process for tracing

The child processes cannot open the /proc/self/smaps due to the
sandboxing. So, for tracing purpose the opened smaps file for child is
sent to the child process via IPC. This file is closed when tracing is
disabled.
Note: The smaps file descriptor is sent only when tracing is enabled
with "disabled-by-default-memory-infra" category and with a heavy dump
trace config.
See design doc: https://goo.gl/v7HgBC.

Security concerns:
Tracing can be enabled only by the user or an extension with "debugger"
permission. When the user has enabled tracing manually, it is assumed
that the renderers can be compromised. When tracing is enabled by a
"debugger" extension, the user sees a pop up on each tab that the
renderer is unsafe. See www.crbug.com/521686.

This CL also sends descriptor of /proc/pid/status file for the child,
so that the process can get its working set size.

BUG=461788

[ssid, 2015-09-09 16:07:43]

ssid@chromium.org changed reviewers:
+ primiano@chromium.org

[ssid, 2015-09-09 16:07:44]

1. Extending the std::istream is has too many methods. I can't find a better way
to read the file other than using while on read().

2. The clear_refs only works for gpu process. The file is owned by root and can
only be written by owner, for all the renderer processes.

3. Opening is done on FILE thread, so begin message has an extra hop. The
ScopedAllowIO is banned by presubmit.

PTAL

[ssid, 2015-10-06 13:05:51]

Cleaned up code and rebased. PTAL.

[ssid, 2015-10-06 14:08:42]

Patchset #4 (id:60001) has been deleted

[ssid, 2015-10-06 21:09:32]

Patchset #3 (id:40001) has been deleted

[ssid, 2015-10-08 09:41:00]

Made changes as discussed, WDYT?

[ssid, 2015-10-08 09:41:27]

Patchset #4 (id:100001) has been deleted

[ssid, 2015-10-21 12:43:42]

friendly reminder to review this patch, thanks

[Primiano Tucci (use gerrit), 2015-10-21 20:21:00]

On 2015/10/21 12:43:42, ssid wrote:
> friendly reminder to review this patch, thanks

Ok will look at this soon.

[Primiano Tucci (use gerrit), 2015-11-05 11:03:58]

I think that given the new approach (dumping from browser)
this can be now closed, as we are not going to send any FD anymore, right?