- Dillo-dev - lists.mailman3.com

BUGFIX set close_on_exec (FD_CLOEXEC)

by Raphael Barabas

hello, in my neverending search for the perfect browser i stumbled across dillo. it looks great and quite usable, altough it's still alpha. unfortunately dillo 0.7.0 won't run on solaris8 (ultra30), so i wrote a fix. turns out every other os is affected by this bug, too. solaris just seems to be less forgiving in this matter :) to compile on solaris you may also need some workarounds, take a look at bugreports 402, 413 and 415. problem: solaris: see bugreport 416, dillo enters endless loop on startup, repeating "IO_write: Changing io->Buf (0)" linux: no immediate effect, although some filedescriptors are in state "open" for too long. i've not really looked into it but this should have negative effects in the long run. reason: in IO.c, function IO_submit, filedescriptor-flag close_on_exec is not set the right way (not at all). on solaris, as a direct result, non-blocking io will accidentally be handled via O_NDELAY instead of O_NONBLOCKING, resulting in an endless loop. patch: *** IO.c.org Sat Mar 1 21:56:13 2003 --- IO.c Sat Mar 1 21:56:54 2003 *************** *** 530,536 **** /* Set FD to background and to close on exec. */ fcntl(r_io->FD, F_SETFL, ! O_NONBLOCK | FD_CLOEXEC | fcntl(r_io->FD, F_GETFL)); if ( r_io->Op == IORead ) { g_io_add_watch(r_io->GioCh, G_IO_IN | G_IO_ERR | G_IO_HUP, --- 530,538 ---- /* Set FD to background and to close on exec. */ fcntl(r_io->FD, F_SETFL, ! O_NONBLOCK | fcntl(r_io->FD, F_GETFL)); ! fcntl(r_io->FD, F_SETFD, ! FD_CLOEXEC | fcntl(r_io->FD, F_GETFD)); if ( r_io->Op == IORead ) { g_io_add_watch(r_io->GioCh, G_IO_IN | G_IO_ERR | G_IO_HUP, disclaimer: this bugfix has been tested briefly on debian gnu/linux 2.4.20 and solaris 8. please review my code for correctness. also note that bugreport 417 has been proven wrong by this. regards, raphael

March 1, 2003

Great work on Dillo! (fwd)

by Jorge Arellano Cid

---------- Forwarded message ---------- Date: Wed, 26 Feb 2003 10:02:07 -0500 From: Alwyn Kerr <alkerr(a)yifan.net> To: jcid(a)softhome.net Subject: Great work on Dillo! Hi there, I'm writing to express how happy I am with Dillo. I'm a HUGE fan of software that is light-weight and Dillo would have to be the smallest, functional browser I've ever used. I find that it is especially useful for browsing documentation that's saved as a stack of HTML pages on disk. I mostly use it for reading the GTK API docs. when developing GTK applications. Anyway, before I'm done I was just wondering if future versions of Dillo might make tabs available for multiple pages? Once again, keep up the good work and please extend my happiness to the rest of the development team. Owen Klan ----------------------------------------------------- http://eo.yifan.net Free POP3/Web Email, File Manager, Calendar and Address Book

March 1, 2003

Re: [Dillo-dev] Loosing bookmarks

by Jorge Arellano Cid

On Wed, 26 Feb 2003, Phillip Warner wrote: > On Wed, 26 Feb 2003 10:17:14 -0300 (CLST) > Jorge Arellano Cid <jcid(a)softhome.net> wrote: > > > The patch: > > > > This patch does two things: > > > > 1.- Keeps a backup of the bookmarks file: bm.txt.bak > > 2.- Avoids not loading the bookmarks file by issuing a > > conditional load before any operation. > > > > Notes: > > * The backup file is made by renaming the original file, so > > if a write error happens the backup remains! > > * The backup is made when the original file has a size greater > > than 256 bytes (just in case!). > > > > > > Please test it as hard as you can. > > > > The test was successful on my end. However, if I kill the server with > -9, the pipe remains and the server will not reload. If I then remove > the pipe, the server loads and responds "Hi browser" but no bookmarks > are written. If I then try to access the bookmarks, I get a blank > page. If I then close dillo, the server also stops. > > Of course, after reloading dillo all is well. > > But, no bookmarks are lost, except the ones that were chosen after the > server was killed. > > Is there a way to overwrite the abandoned pipe (when no server is > running) when restarting the server? I know sylpheed does this. Yes, this situation is problematic. Besides the 'kill -9', a system crash can also leave the socket's filename in /tmp... and the trouble only ends when the user realizes he must remove the socket's filename manually! After some thought, I finally found a way to check these problems using connect(). I'm not sure which error code (errno number) the different OSes will return, so I included: ECONNREFUSED, EBADF, ENOTSOCK and EADDRNOTAVAIL. GNU/Linux returns ECONNREFUSED in this case. ...and I almost forgot that sometime ago (Nov 27), Ben Wolley wrote: > Hello Jorge, > > I just tried out the dpi1 bookmarks plugin, and I really like the concept. > But I started wondering if someone could take advantage of a dpi URL to do > something improper. > > <html> > <body> > <a href="dpi:/bm/modify?operation=add_url2&title=home&url=http%3A%2F%2Fhome&sub > mit=submit.">test</a> > <img src="dpi:/bm/modify?operation=add_url2&title=home2&url=http%3A%2F%2Fhome&submit=submit." > alt="test image"> > </body> > </html> > > When I load that page, not only does the image alter my bookmarks, when I > click on the link the window closes (after altering my bookmarks). This > seems to be an issue similar to what is discussed in the HTTP 1.1 spec > section 15.1.3. > > Can POST data be sent instead? Section 9.5 says that POST should be used > for "extending a database through an append operation", which is what the > plugin actually does. A hacked dpi-request using POST can also be faked, so the decision was to only allow dpi-urls from dpi-generated pages. The attached patch (against dillo-0.7.0) does five things: 1.- Keeps a backup of the bookmarks file: bm.txt.bak 2.- Avoids not loading the bookmarks file by issuing a conditional load before any operation. 3.- Takes care of EINTR while in recv(). 4.- Tests the socket's filename to be responsive and relaunchs the server if necessary. 5.- Allows dpi-requests, but only from dpi-generated pages. Most probably this will be the patch in the bug-fix release. Once again, please test it throughly! Cheers Jorge.-

March 1, 2003