Commit f9a78d598e48132a8c6cf9ce31b51163b6bd2f67 refactored the code, but
in the process introduced a socket leak. When connect() is called on a
socket set to non-blocking mode, the returned error code is EINPROGRESS.
In that case, the code never initialises the ev structure and the socket
reference is lost. Given enough time (on my network about a day and a half)
this will eventually lead to a non-responsive server because the process
runs out of fds. Netstat will show an excessive number of sockets stuck in
CLOSE_WAIT state forever and ls /proc/<pid>/fd will confirm a lot of open
sockets.
Initialising the ev struct before a call to connect() ensures that the
socket reference is not lost. Verifying with netstat, one can see the
sockets in TIME_WAIT state for a brief period of time and /proc/<pid>/fd
shows a reasonable number of sockets being open.
The number_of_children child connections number becomes negative (-1) at
server startup due to signal handlers treatments on minidlna.c init().
Opening the webserver status page with no clients connected the opening
connection shows -1 even if client list show 0 for all in the columns.
When connecting the first client, its own connections column goes to 1
and total goes to 0, and so on always one count behind.
Simply allow: number_of_children-- only if not 0
From SF user negan07
support compiling if gettext 0.20 or later is found
AM_GNU_GETTEXT_REQUIRE_VERSION overwrites AM_GNU_GETTEXT_VERSION if
autoreconf supports it, so legacy systems are still supported.
From SF user CvH
loop and event dispatchers. This simplifies code and eliminates a bug,
when kevent dispatcher is called with 0 timeout.
While here, in the main loop call gettimeofday() right after event
dispatcher returns. Otherwise, we are using outdated "timeofday" in
second part of the loop. I don't know any bugs because of that, but
they are possible.
specific code to monitor_inotify.c.
Both monitor_inotify.c and monitor_kqueue.c provide their functions
for adding and removing watches. Prefix these functions with monitor_.
Use newer API for IP_MULTICAST_IF which allows one to specify
interface by index, not by address. Introduced in Linux 3.5, it IMHO should
be available on all systems that declare struct ip_mreqn.
This fixes operation failure when a system has multiple interfaces
with same address, but only on of them is desired. Example:
> grep interface /usr/local/etc/minidlna.conf
network_interface=igb0
> ifconfig igb0
igb0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0 mtu 1500
ether 0c:c4:7a:xx:xx:xx
inet 10.1.10.3 netmask 0xffffff00 broadcast 10.1.10.255
media: Ethernet autoselect (1000baseT <full-duplex>)
status: active
> ifconfig ng0
ng0: flags=88d1<UP,POINTOPOINT,RUNNING,NOARP,SIMPLEX,MULTICAST> metric 0 mtu 1454
inet 10.1.10.3 --> 10.1.10.2 netmask 0xffffffff
In such configuration, ng0 would be chosen before this fix.
Using $USER in the friendly name will display which user minidlna is
running as on the clients, which can be helpful for detecting problems.
Using $LOGNAME on the other hand will display "root" as the username if
minidlna was started using the init script, regardless of which user
minidlna is currently running as.
Originally added by Benoît Knecht <benoit.knecht@fsfe.org>
Define setjmp_buffer as static to avoid the following build failure with
gcc 10 (which defaults to -fno-common):
/home/buildroot/autobuild/instance-1/output-1/host/lib/gcc/arm-buildroot-linux-gnueabihf/10.2.0/../../../../arm-buildroot-linux-gnueabihf/bin/ld: image_utils.o:(.bss+0x0): multiple definition of `setjmp_buffer'; metadata.o:(.bss+0x0): first defined here
collect2: error: ld returned 1 exit status
Fixes:
- http://autobuild.buildroot.org/results/8754bb4f7d749f999d5f8ddfec587470ceec4476
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
The UPnP protocol specification mandates that subscribers can request a
callback to an arbitrary URL. This recently resulted in CVE-2020-12695
(CallStranger) outlining the risk of information disclosure and DoS
attacks.
This commit ensures that the callback URL sent in a SUBSCRIBE request
points to the same IP address that made the request.
If our response buffer fills up to the max buffer size on a UPnP Browse
request, we should return short results rather than erroring out.
Fixes: #314 (Cannot browse MiniDLNA 1.2.1 with VLC 2.2.6 (UPnPError 709: Unsupported or invalid sort criteria))
Add flag for milliseconds conversion for bookmarks and model for Samsung Q
series, which needs this. This allows for bookmark synchronization when 2
or more TVs are using the same DLNA server.
getsyshwaddr assumed that the first ifaddr it came across was the MAC
address, and as such assumes that it has the right length. After
upgrading to OS X 10.13.4, this causes minidlnad to crash on startup due
to tripping stack smash protection -- I'm not sure if the order of
addresses returned previously happened to accidentally hit this
invariant, or if this was always an issue and the stack smash protection
got smarter.
In any event, we just need to look for the AF_LINK address and use that.
As an extra check, we make sure the length is the length we expect to
copy into the target buffer.
There's a report about a nightly "crash" when users download, compile
from source, and replace their distro's mindlnad binary. This is because
the Debian package includes a patch that uses SIGUSR2 to reopen the log
file and sends SIGUSR2 from logrotate instead of just using the
"copytruncate" logrotate option. Then logrotate sends SIGUSR2 at 6:25AM,
which causes us to abort due to the unhandled signal.
I don't want to sacrifice SIGUSR2 just for log rotation, especially when
we already do some reload operations on SIGHUP. So to avoid this
Debian/Ubuntu issue, we'll explicitly ignore SIGUSR2, and add log file
reopening to the SIGHUP handler. Then hopefully a future Debian package
version will remove the SIGUSR2 patch and use SIGHUP instead (or
copytruncate).
Fixes: SF Bug #313 (log rotation kills minidlna service)
The Control4 app apparently sends us badly-formatted SearchCriteria,
where they don't quote the upnp:class value. So we'll add it in for them
if we detect that situation.
Fixes: SF Patches #173 (DLNA search query not as minidlna expects)
There's no point in returning bookmark info without a bookmark. It also
has been reported that certain Samsung models may have issues with
subtitles if we send dcmInfo data. So we'll only return it if we have a
bookmark set.
[minidlna:support-requests] #52 External srt subtitles for Samsung UA48JU7800
Add back SSDP filtering. Relying on the kernel to filter out multicast
traffic from non-member interfaces simply doesn't work the way many
people think it does, so we need to re-introduce manual filtering. But
this time we will use in_pktinfo's ifindex for comparison rather than a
netmask comparison, so SSDP packets from other subnets should still work.
This has theoretical bug if our collection spans different
filesystems and some filesystems support holes, and some not.
If this ever encountered we should use pathconf(2) and cache
its result for directories.
