292374 - Research about complete freeze, loops calling ioctl(n, SNDCTL_DSP_GETOPTR, x)

Status: VERIFIED INCOMPLETE

(Firefox :: General, defect)

Description

[Enrique Perez-Terron]

User-Agent:       Mozilla/5.0 (X11; U; Linux i586; en-US; rv:1.7.7) Gecko/20050416 Fedora/1.0.3-1.3.1 Firefox/1.0.3
Build Identifier: Mozilla/5.0 (X11; U; Linux i586; en-US; rv:1.7.7) Gecko/20050416 Fedora/1.0.3-1.3.1 Firefox/1.0.3

I clicked on a link on http://www.aftenposten.no/, pointing to
http://www.aftenposten.no/nyheter/uriks/article1029268.ece, and browser went
completely unresponsive, not updating window after partial obscuring, and not
showing contents from new page. Cursor was hourglass. CPU consumption low, not zero.



Reproducible: Sometimes

Steps to Reproduce:
Site too dinamic to really reproduce. See Additional info.
Reproduction was consistent until site changed. Then another link reproduced it,
God knows for how long.
1. Point the browser to http://www.aftenposten.no
2. Click the link labeled "Utenriks" in leftmost column, near top.
3.

Actual Results:  
Froze. Hourglass icon. Obscure with another window, and expose - does not paint.
strace showed loop on gettimeofday, ioctl(SNDCTL_DSP_GETOPTR) and
ioctl(*_GETOSPACE), and select(0,0,0,11ms).

Expected Results:  
Should display the page that had been transmitted over the wire.

Process status showed an mplayer plugin had started.

Attaching the browser with "strace -p xxxx" showed a loop thus:

  select(0, NULL, NULL, NULL, {0, 11000}) = 0 (Timeout)
  gettimeofday({1114791124, 416741}, NULL) = 0
  ioctl(35, SNDCTL_DSP_GETOPTR, 0xbff4225c) = 0
  ioctl(35, SNDCTL_DSP_GETOSPACE, 0xbff4225c) = 0

repeating ad infinitum, only the timestamp advancing by approx. 15ms.

Sniffing the network with ethereal, repeating the steps reproduced the freeze,
and ethereal showed a transfer complete with the finishing </html>. However, the
last datagram to cross the wire was my PC sending an ACK. There was no FIN or
RST from the server.

Saving the payload of the exchanges to a .html file, and opening a new browser
to view file:///tmp/the-saved-data.html displayed the page, no problem.  It
appears that most links on the page are absolute links in different hosts, so
the page seemed to load the same ads etc from the same sources. However, I never
know if the ads are different.

Downloaded today's latest nightly trunk build, and ran it with a new profile.
The referring page (the page contining the link) had changed by that time,
sorry. The link was no longer there. (The referring page was
"http://www.aftenposten.no/", the major Norwegian newspaper. Highly dynamic, and
quite crowded.) Since the link was to an article about a US event, I clicked the
generic "Foreign Affairs" link ("Utenriks", left hand margin column, near top)
of the main page, and the browser froze at the spot.

In other words, reproduced with the latest nightly trunk build.

Not knowing what else to do to research it. Please suggest. I don't think
slashing down page complexity is relevant here since the problem was gone as
soon as the page was saved to a file. Seems that the strace offers the only
opportunity. You might have a chance looking at what could make it loop issuing
those ioctls. Perhpas a more extensive strace (attaching before clicking the
link would help you, please ask me to do so if you care looking at it.

Comment 1

[Enrique Perez-Terron]

Attachment: TCP payload of connection, i.e., the HTTP request + response headers + html page.

For what it is worth, this is what I captured in Ethereal while reproducing the
bug. Saving the html part to a file, and pointing the browser to it does diplay

with no hang.

I must admit, I forgot to look for mplayer processes when reproducing. Could
not see signs of any transfer related to such contents in ethereal. In fact,
there were no signs in the ethereal log of the browser trying to access any of
the related urls, ads, gifs, etc.  Wonder if it is all related to the lack of
end-of-file indication from the server.  Does not FF display page
incrementally?	 Does it not access related contents in parallel? In case
anyone want to look at it, I saved the complete sniffer log as well, write me
to request it.

The strace attaching hung browser was repeated with the latest trunk build and
was identical.

Comment 2

[Samuel Sidler (old account; do not CC)]

That site has completely changed in the last two years.

Enrique, if you're able to reproduce this using Firefox 2.0.0.3 or later and a clean profile, please let us know and give clear steps to reproduce including exactly which link caused it.

Comment 3

[Samuel Sidler (old account; do not CC)]

Closing as INCOMPLETE given the site changing. If you can reproduce using using the idea in comment 2, please reopen.