Those two functions are only used by osmo_sock_init2_multiaddr(), which
is only built if HAVE_LIBSCTP is defined. Avoid compiler warning about
unusued function helpers if osmo_sock_init2_multiaddr() is not being
built.
Change-Id: I52769d6b8f70af1a8bda23d60b3230a932e71fab
Since we return error at the start of the function if proto !=
IPPROTO_SCTP, it makes no sense to check for proto != IPPROTO_UDP later
on.
Fixes: CID#205088
Change-Id: Ibba7eacaa9debb77d536d47dc85170c5ee79e479
This API will be used by libosmo-netif's osmo_stream for SCTP sockets,
which in turn will be used by libosmo-sccp to support multi-homed
connections.
Related: OS#3608
Change-Id: Ic8681d9e093216c99c6bca4be81c31ef83688ed1
ubsan will report undefined behavior due to the SUN_LEN macros interaction with a null pointer,
so let's tell ubsan to ignore this function. After carefully reviewing the final publically
availlable drafts of the C99,C11 and C18 standards I can confirm that dereferencing null pointers
is still undefined behavior, as such ubsan will always warn with absolutely every existing compiler
version. Since the sanitizers are periodically synced between llvm and gcc I'm also fairly confident
that rebuilding everything with compiler_rt to use the integrated sanitizers would result in the same message.
I sincerly hope that this explanation provides to be sufficient, If not I'd be willing to show up at
the next llvm dev meeting to provide quotes from actual sanitizer developers to back up these claims.
Change-Id: I0ff445072f1b46390c9f70b21d61c789e39358d5
We have a number of library-internal static global buffers which are
mainly used for various stringification functions. This worked as
all of the related Osmocom programs were strictly single-threaded.
Let's make those buffers at least thread-local. This way every thread
gets their own set of buffers, and it's safe for multiple threads to
execute the same functions once. They're of course still not
re-entrant. If you need re-entrancy, you will need to use the _c()
or _buf() suffix version of those functions and work with your own
(stack or heap) buffers.
Change-Id: I50eb2436a7c1261d79a9d2955584dce92780ca07
osmo_sock_get_name_buf():
In case the getsockname() call is failing for some weird reason,
we shouldn't return an uninitialized, non-zero-terminated string
buffer to the caller, as most callers will be too lazy to test the
return value.
This holds even more true for users of the internal
osmo_sock_get_name2() and osmo_sock_get_name2_c() functions which indeed
very much ignore the return value of osmo_sock_get_name_buf().
Change-Id: I2d56327e96b7a6783cca38b828c5ee74aed776ae
Calling sizeof() on a pointer to dynamically allocated memory would
result in getting size of the pointer (usually 4 or 8 bytes) itself,
but not the size of allocated memory.
Change-Id: I8ffda4dea2b7f9b4b76dfeecad1fab6384c5a62c
Fixes: CID#197629, CID#197628, CID#197627
Fixes: CID#197626, CID#197625, CID#197624
We have a habit of returning static buffers from some functions,
particularly when generating some kind of string values. This is
convenient in terms of memory management, but it comes at the expense
of not being thread-safe, and not allowing for two calls of the
related function within one printf() statement.
Let's introduce _c suffix versions of those functions where the
caller passes in a talloc context from which the output buffer shall
be allocated.
Change-Id: I8481c19b68ff67cfa22abb93c405ebcfcb0ab19b
The naming of these constants dates back to when the code was private
within OpenBSC. Everything else was renamed (bsc_fd -> osmo_fd) at
the time, but somehow the BSC_FD_* defines have been missed at the
time.
Keep compatibility #defines around, but allow us to migrate the
applications to a less confusing naming meanwhile.
Change-Id: Ifae33ed61a7cf0ae54ad487399e7dd2489986436
The function osmo_sock_get_name_buf() can be used to write a string
representation to a user provided memory. Unfortunately the proper
length for the user provided memory is not obvious. To make using
osmo_sock_get_name_buf() more practical, add a define constant that
defines the length of the required memory. Also use this define in
socket.c.
Change-Id: If8be8c2c0d4935da17ab13b2c2127b719ceefbcc
Basically, I am applying code review that I would have given had I not been on
vacation when the last osmo_sock_get_name* stuff was merged.
osmo_sock_get_name2() is so far a static internal function. However, it is
nothing like osmo_sock_get_name(), so instead rename it to
osmo_sock_get_ip_and_port(). Also make it public API, no need to hide it. I'm
adding an "and" in the name to hopefully clarify: "ip_port" vs. "ip_and_port"
-- there already are _get_X_ip_port() functions that only return the port
string, despite "ip" in the name.
Add new public osmo_sock_get_name2(), which is like osmo_sock_get_name(),
except it uses a static string instead of talloc, and omits the braces. This
is most convenient for log statement formats, avoiding dyn allocations.
Add new osmo_sock_get_name_buf(), which is like osmo_sock_get_name2() but
writes to a caller provided char buffer.
Use osmo_sock_get_name_buf() in the implementation of osmo_sock_get_name(),
but use another (non-static) local string buffer, because adding braces is too
complex without talloc_snprintf().
Rationale:
I want to improve the logging of socket errors, e.g. change
DLMGCP ERROR Failed to read: 111/Connection refused (mgcp_client.c:720)
to
DLMGCP ERROR Failed to read: r=10.0.99.2:2427<->l=10.0.99.2:2728: 111='Connection refused' (mgcp_client.c:721)
but it is just not handy to compose logging with the current API:
- osmo_sock_get_name() requires a talloc_free().
- all the others require output buffers.
- the only way to conveniently compose a logging string and,
- notably, the only trivial way to skip the string composition if the logging
level is currently muted, is to have a function that returns a static string:
the new osmo_sock_get_name2().
- (I think the osmo_sock_get_{local,remote}_* convenience wrappers should never
have been added, because they encourage the caller to invoke the same code
twice, for IP addr and port, and throw away one half each time.)
Related: Iae728192f499330d16836d9435648f6b8ed213b6 (osmo-mgw)
Change-Id: I8ad89ac447c9c582742e70d082072bdd40a5a398
It's similar to osmo_sockaddr_to_str_and_uint() but does not require odd
typecasting for AF_INET case. Make osmo_sockaddr_to_str_and_uint() into
wrapper around new function and make sure to check for address family
before typecasting. Also use proper return type.
Change-Id: Ie384483124d407a960ab6732e6a7fd90554389d2
Use INET6_ADDRSTRLEN (46) instead of 64 for IP address buffers, and 6
instead of 16 for port buffers (the highest possible port number is
65535).
Change-Id: Ia25e2f3277ad2f60df31c08d12f42c1e6d2a14a6
Return only the IP or port of either the local or remote connection,
not the whole set of IP and port of both the local and remote
connection like osmo_sock_get_name() does it. This is needed for
OS#2841, where we only want to print the remote IP.
Related: OS#2841
Change-Id: I6803c204771c59a2002bc6a0e6b79c83c35f87e1
The unix(7) man page recommends that sun_path is NUL-terminated
when struct sockaddr_un is passed to a bind() or connect() call.
Non-NUL-terminated paths only need to be dealt with at the
receiving end of a UNIX domain socket.
Commit 896ff6d erroneously assumed otherwise.
This commit almost reverts 896ff6d: It only leaves the added
osmo_strlcpy() overflow check in place.
Change-Id: I6c4ac6b0a0eef4842beae4107f6f09f6cd29172a
Fixes: 896ff6db16
Related: OS#2673
In osmo_sock_unix_init(), add support for non-NUL-terminated unix
socket paths and return an error if the supplied socket path exceeds
the maximum socket path length supported by the operating system.
Change-Id: I19d935e5e3dd7928e6e153c6f5ad7044de726016
Related: OS#2673
When IPPROTO_UDP is used then SO_REUSEADDR omitted since UDP is
connection less we do not have to wait until lingering connections time
out. There were also negative effects such as that two applicatications
could use the same UDP port, normally one of the two applications would
get an error, but with SO_REUSEADDR this is supressed. However, there
are applications (UDP MULTICAST) where two applications must be able to
use the same port. In the osmocom project those are osmo-bts-virtual,
virtphy and gsmtap in general.
Lets introduce a flag that the API user can supply in order to have
SO_REUSEADDR applied.
- Add new flag OSMO_SOCK_F_UDP_REUSEADDR
Change-Id: I94aaf6d5224ab23bde5ea5c4a83569b6145ab32b
Related: OS#3497
When UDP is used as protocol (proto=IPPROTO_DUP), then we should not set
SO_REUSEADDR in the socket option. Because if we do, we allow two
processes to bind on the same UDP port. The errornous situation will be
undetectable to both applications. So lets only set SO_REUSEADDR when we
do not use UDP.
- Add check if we use UDP, if yes do not set SO_REUSEADDR
Change-Id: I4a8ffb8d598aca88801a4a0322944d7cdd8d4047
Related: OS#3441
the return code of the last setsockopt() call in osmo_sock_init() is not
checked. Since all other calls to setsockopt are checked, lets check
this one as well.
- check return code of setsockopt() and close the socket on failure
Change-Id: I96dbccc3bcff35bf39979dbe0c44aadc8ce20c83
Catched by AddressSanitizer in osmo-bts-trx while running tests in
osmo-gsm-tester:
==31738==ERROR: LeakSanitizer: detected memory leaks
Direct leak of 5744 byte(s) in 1 object(s) allocated from:
#0 0x7ff7ec789ed0 in calloc (/usr/lib/x86_64-linux-gnu/libasan.so.3+0xc1ed0)
#1 0x7ff7e952697c (/lib/x86_64-linux-gnu/libc.so.6+0x10297c)
#2 0x7ff7e95274df in getifaddrs (/lib/x86_64-linux-gnu/libc.so.6+0x1034df)
#3 0x7ff7eadcdc8f in osmo_sockaddr_is_local libosmocore/src/socket.c:537
Change-Id: I778d3c1f162abce0595e62670c29c5134bccd28d
After investigating osmo-msc showing this log message and looking at the
code, it's a bit difficult to find out what's going on in the code:
socket.c:224 unable to bind socket: (null):0: Protocol not supported
The root cause was not yet found, but probably SCTP is not enabled in
the kernel of the host running it.
The cod eis most probably failing during socket() and not due to bind
error as the log says, so let's print an error if socket() fails.
Then, if setsockopt fails, we want to still keep trying in case an extra
addr was offered by addrinfo_helper. It is definetly wrong to continue
if setsockopt fails, because then we are skipping the bind(), which is a
fundamental part of what osmo_sock_init2 does.
Then, let's print the bind error when it really happens, and re-write
the extra log at the end if we reach the point at which no suitable addr
is found.
Change-Id: I1854422ad92dadf33ed4d849e15c0380c3bf1626
The function inet_ntoa() stores its result in a static buffer and
returns the pointer. When inet_ntoa() is called subsequently it
overwrite the content of its static buffer with the new result.
Since we osmo_sock_local_ip() is a library function we should use
the more safe variant inet_ntop() in order to prevent unintentionally
overwriting data that the caller might still need. Such an error
would be hard to find.
- Use the more safe inet_ntop() inestead of inet_ntoa()
Change-Id: I9852b57736432032542bd96b6fdd4a2f08fc1f64
The socket that is opend to probe the correct local ip-address is
not closed when the test is done.
- Close socket when it is not needed anymore
Change-Id: I7f3562a344b58f6298d2068314be1626a96e1b1d
Let's fix some erroneous/accidential references to wrong license,
update copyright information where applicable and introduce a
SPDX-License-Identifier to all files.
Change-Id: I39af26c6aaaf5c926966391f6565fc5936be21af
In some cases it is required to know the ip-address of the interface
through that a given remote IP-Address can be reached.
Add function osmo_sock_local_ip() to determine the local ip-address
for a given remote ip-address
Change-Id: I2988cc52b196fc8476703d1287e24cb4a48491c2
If osmo_sock_init2() was used with CONNECT flag but without BIND
flag, an invalid check for "did we create a socket yet" caused
the socket to never be created, and subsequently the entire function
to return an error.
Change-Id: I0206dbb9c5b8f74d7fb088576941b092acd2ca22
Using this option at socket creation, the caller can request disabling
the IP_MULTICAST_ALL socket option.
Change-Id: I5ab5de45c0b64ceb3636ea98245a23defa24ffd4
This introduces a new flag OSMO_SOCK_F_NO_MCAST_LOOP, which can be used
to disable the looping back of multicast packets transmitted throug this
socket to other local sockets on the machine.
As this looping-back is active by default, a single option to deviate
from the default is deemed sufficient.
Change-Id: I24a5b1ebc3f84d2d5d4734e54df50efaea26490b
We had three places at the end of socket initialization functions
calling listen(). Let's unify that and fix some bugs:
* close + return error in case of bad listen() result
* don't call listen() on AF_UNIX SOCK_DGRAM sockets
Change-Id: I7e8dbe3c0486bb3b9810b0add1331e93fc106d82
In a string like
127.0.0.1:2905<->127.0.0.1:60661
it is hard to tell which is the local part. I'd have expected it on the left,
but it is actually on the right.
To avoid doubt and bypass bikesheds on which side should be what, clearly mark
the two sides as remote and local.
(r=127.0.0.1:2905<->l=127.0.0.1:60661)
Change-Id: I43dcc6a1906429bd0955fd7fe2eb5b8495b592d8
Considering the various styles and implications found in the sources, edit
scores of files to follow the same API doc guidelines around the doxygen
grouping and the \file tag.
Many files now show a short description in the generated API doc that was so
far only available as C comment.
The guidelines and reasoning behind it is documented at
https://osmocom.org/projects/cellular-infrastructure/wiki/Guidelines_for_API_documentation
In some instances, remove file comments and add to the corresponding group
instead, to be shared among several files (e.g. bitvec).
Change-Id: Ifa70e77e90462b5eb2b0457c70fd25275910c72b
Especially for short descriptions, it is annoying to have to type \brief for
every single API doc.
Drop all \brief and enable the AUTOBRIEF feature of doxygen, which always takes
the first sentence of an API doc as the brief description.
Change-Id: I11a8a821b065a128108641a2a63fb5a2b1916e87
It's a pity that even with this patch we still are fare away from having
the whole API documented. However, at least we have a more solid
foundation. Updates not only extend the documentation, but also make
sure it is rendered properly in the doxygen HTML.
Change-Id: I1344bd1a6869fb00de7c1899a8db93bba9bafce3
The old osmo_sock_init() function allows only either a bind (for a
server socket), or a connect (for a client socket), but not both
together. So there's no way to have a client socket that is bound to a
specific local IP and/or port, which is needed for some use cases.
Change-Id: Idab124bcca47872f55311a82d6818aed590965e6
socket.c still uses fprintf to output error messages. This commit
replaces the fprintf with proper LOGP messages.
Change-Id: Ia2993415d5f5c33ccd719af239ff59252d11b764
Using this function, one can obtain a human-readable string identifying
the host and port names of the socket.
Change-Id: Ib5de5c7b9effe1b0a363e4473a7be7fa38ca6ef3
For programs like osmo-hnbgw with numerous sockets, the message that some
unspecified connection was refused is not very helpful. Also output the host
and port where an error occured.
Instead of perror, use fprintf(stderr, ..., strerror()) to be able to include a
format string and print host and port as passed to osmo_sock_init().
Change-Id: I8d0343f51310699b78fcb83fd76fd93764acf3dc
Harald Welte
Pau Espin
Eric Wild
Vadim Yanitskiy
Philipp Maier
Oliver Smith
Neels Hofmeyr