Whenever iterating over a list and removing entries,
llist_for_each_entry_safe must be used instead of llist_for_each_entry.
The comment with "non-consecutive(!)" entries sounds like this is not
needed as long as one is iterating over the list consecutively. I guess
that might have worked with prefetch logic, however the prefetch
function is just a stub in linuxlist.h. (Also prefetch has been removed
from list.h in linux.git e66eed651fd18a961f11cda62f3b5286c8cc4f9f.)
Change-Id: I217e6871afe121edba26e4c6fd1a461e397c9e72
According to TS 101 318, section 5.2.2, a SID frame is identified
by a SID codeword consisting of 79 bits (r34..r112) which are all 1.
Given that there are no gaps in the codeword, we don't really need
to use bitvec_get_bit_pos() and check each field individually.
This brings additional complexity and negatively affects performance.
Instead, let's use bitvec_get_bit_pos() to check all bits in a row.
Change-Id: I678f8ff92317fe87f1aca511a3a062ad898e55cc
Related: SYS#5853
Make sure that osmo_time_cc counts exactly the same, regardless of a
rate_ctr being present or not. Only skip sending counter increments when
there is no rate counter, do not affect anything else.
In osmo-bsc, we are discussing a patch where an lchan redirects
osmo_time_cc counter increments to different rate counters depending on
its current type. In my comments I am also claiming that osmo_time_cc
works the same regardless of a rate_ctr being present or not. Looking at
the code, this is not actually the case.
Before this patch, when there is no rate_ctr, the reported_sum would
freeze, and as soon as a rate_ctr shows up, all counter increments from
the previous reported_sum would be sent to the newly set rate_ctr.
Instead, we want counter increments to simply be lost while there is no
rate_ctr set. IOW, rate_ctr == NULL should mean >/dev/null.
Related: I1b0670c47cb5e0b7776eda89d1e71545ba0e3347 (osmo-bsc)
Change-Id: I380b28d7ab0a6c8aab6c7e2a64bc73cab14863d2
If there's only a single device with matching VID/PID attached,
we don't need to insist that either the path or the address of the
device matches. Those are only needed to disambiguate multiple
devices with identical VID/PID.
Change-Id: I2ef245a56dfcf22758a0216b86d2a5c602ee5588
If an API user has defined a name for this particular
stat, we should consider it unique and not append ip and
port information from the connection.
By appending ip and port information to all tcp stat
names, we end up creating unique stat names every
time a reconnection occurs and the source port changes.
This makes the statistic impossible to track over time
as it is continually using a different name.
A quick example from the field over the course of a
day:
tcp.ipa-rsl-0,r=192.168.55.88.33056<->l=192.168.0.1.3003.tcp.rtt
tcp.ipa-rsl-0,r=192.168.55.88.33311<->l=192.168.0.1.3003.tcp.rtt
tcp.ipa-rsl-0,r=192.168.55.88.35510<->l=192.168.0.1.3003.tcp.rtt
tcp.ipa-rsl-0,r=192.168.55.88.35958<->l=192.168.0.1.3003.tcp.rtt
tcp.ipa-rsl-0,r=192.168.55.88.36110<->l=192.168.0.1.3003.tcp.rtt
tcp.ipa-rsl-0,r=192.168.55.88.39269<->l=192.168.0.1.3003.tcp.rtt
tcp.ipa-rsl-0,r=192.168.55.88.40394<->l=192.168.0.1.3003.tcp.rtt
tcp.ipa-rsl-0,r=192.168.55.88.40397<->l=192.168.0.1.3003.tcp.rtt
tcp.ipa-rsl-0,r=192.168.55.88.42920<->l=192.168.0.1.3003.tcp.rtt
tcp.ipa-rsl-0,r=192.168.55.88.43839<->l=192.168.0.1.3003.tcp.rtt
This change would treat tcp stats like other stats
around the system. A unique name must be set by the
API user. This would let us set a unique name like
the following to avoid the situation above:
bts.0.rsl.0.tcp.rtt
Matching the existing rsl related stats:
bts.0.rsl.delete_ind
bts.0.rsl.ipa_nack
bts.0.rsl.unknown
...they retain a constant name regardless of the underlying
connectivity situation.
Change-Id: Ib04c2f5bfcbd6c19dd87debf1fc053abf0b9bef2
Those are available in 3GPP TS 48.008 version 16.0.0 Release 16, section
3.2.2.17 Cell Identifier. It can be seen that we have a collision
between the osmocom non-standard format and the SAI standard one.
This is because CGI-PS is not really a TS 48.008 Cell Identifier, but only
specified in TS 48.018 and has no ID number assigned. The CGI-PS was
added there because the whole osmo-bsc neighbour configuration works
with CellIds to manage neighbours, so it felt natural to extend the APIs
to also provide means to use CGI-PS format (TS 48.018 even refers 48.008
existance and mentions there's no explicit ID).
At the time this Cell Identifier was added, the firstly available number
(11) was taken, which was of course a really bad idea since newer
versions of the spec can at some point use it, which is the case if one
checks for instance TS 48.008 Release 16 SAI Cell Id.
There no perfect way to fix this bad decision at the time, but the
CGI-PS is only used in osmo-bsc and only for RIM related purposes, so by
changing the ID of CELL_IDENT_WHOLE_GLOBAL_PS, we only break RIM under
some specific CIs being used, and when an osmo-bsc is built against
older libosmocore and then used at runtime against a newer libosmocore
(which should be rare).
Hence, the downside is acceptable, and by moving the new ID number to be
ouside of the spec proto TS 48.008 range (4 bits), we make sure we don't
have the same problem again in the future.
Related: SYS#5838
Fixes: ca33a71ca8
Change-Id: Id25e563febdb7640174540136225f399515a0089
We don't handle this correctly, as we decided to overload the
meaning of 0b1011 in a cell identifier list (defined by 3GPP
as used in SRVCC) with something osmocom proprietary.
Change-Id: I608220e8e5dd5a44f85c6bc5ea04622a2cad24ec
Theoretically, this should improve performance of the code where
we frequently invoke OSMO_ASSERT(), like osmo-msc and osmo-bsc.
Change-Id: I29b32a2477ec92762f8f0ce5e5c5a30810f6abbe
Using do-while is a common practice for complex macros. This allows
invoking OSMO_ASSERT in simple if-else statements without braces.
Change-Id: I42d9c315c9c30bce828564a63c496ee62e5c1431
These macros are built on top of the __builtin_expect() function [1],
which provides the compiler with branch prediction information.
Similar macros exist in the Linux kernel: likely() / unlikely().
[1] https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html
Change-Id: I0b029654ba050f079eed4a0574a3fa8019677067
Shorthand for the INET/INET6 switch() to get/put the addr part, useful
for encoding and decoding message buffers.
Related: OS#5599
Change-Id: Ie9e33bfac525c59c30714663d2bfcc62ec9eeb81
There already are osmo_quote_str_buf() and osmo_quote_str_buf2(), same
for _escape_, but none of them return the snprintf() like string length.
A private function does, publish this in the API.
The returned chars_needed is required to accurately allocate sufficient
size in string functions that call osmo_quote_str/osmo_escape_str. I am
adding such in osmo-upf.git.
Related: SYS#5599
Change-Id: I05d75a40599e3133da099a11e8babaaad0e9493a
The OSMO_SOCKADDR_STR_FMT() and _ARGS() macros properly place square
braces around IPv6 addresses, so that the port nr is clearly
distinguishable.
before: 1:2::3:4:5
after: [1:2::3:4]:5
When using a struct reference, the macro resolves to '(&sastr) ? .. : ..',
which the compiler complains about as "condition is always true". Shim
around that error with a pointer variable.
I considered using osmo_sockaddr_to_str_c() instead, but here in
socket.c we cannot assume that osmo_select_main_ctx() is being used and
hence can't just use OTC_SELECT for log string composition. The
struct osmo_sockaddr_str is a string representation in a local variable,
and hence doesn't need talloc for log strings.
I considered adding log_check_level() around the log string conversion,
but since all of these instances are on LOGL_ERROR, I didn't bother.
Related: SYS#5599
Change-Id: Idbe7582b2b7f14540919e911dad08af6d491f68f
It seems it is insufficient to register file-descriptor call-backs
with libusb_set_pollfd_notifiers(), but we also need to call
libusb_get_pollfds() once to get the initial set of file descriptors
which may have been created already during libusb_init().
As we don't have a libusb context before libusb_init() returns,
we cannot call libusb_set_pollfd_notifiers() early enough to be
active already during libusb_init().
Change-Id: Icf81014d689ffa738719af68120fa2dedbeec689
As was demonstrated in I54bf5e5c036efb1908232fe3d8e8e2989715fbb3,
when the logging is configured to print the filename *after* the
logging message, each logging line contains an artifact - '\0'.
The problem is that the 'len' variable is not updated. Fix this.
Change-Id: I5c920a0d5c1cf45bcdd327b39e33d63346b4f51c
Fixes: I393907b3c9e0cc1145e102328adad0a83ee13a9f
To easily log and print a sockaddr using OTC_SELECT, add
osmo_sockaddr_to_str_c().
Implement osmo_sockaddr_to_str_buf2() using osmo_strbuf, so that we can
return the chars_needed which osmo_sockaddr_to_str_c() uses.
From previous osmo_sockaddr_to_str_buf(), call
osmo_sockaddr_to_str_buf2() and return NULL if the buf_len was
insufficient, to mimick previous behavior. This makes it more
consistently returning NULL for insufficient buf_len, as shown in the
tweak that is needed in socket_test.c. Before osmo_sockaddr_to_str_buf()
would return a truncated port number, now it's all or NULL.
I will use osmo_sockaddr_to_str_c() in the new osmo-upf implementation.
Related: SYS#5599
Change-Id: I12771bf8a021e6785217b1faad03c09ec1cfef0e
This commit demonstrates a bug introduced in [1], which can be
observed when the logging is configured to print the filename
*after* the logging message (LOG_FILENAME_POS_HEADER_END):
logging print file 1 last
^^^^
In this mode, the code in _output_buf() overwrites the '\n' sybmol
contained in the logging message itself by shifting the 'offset'
backwards, and appends the nipped '\n' after the filename info.
The problem is that the 'len' variable is not updated in this case,
so the resulting length includes +1 character - '\0', which gets
printed at the end of every logging line.
Interestingly enough, this problem affects only the wqueue mode.
Change-Id: I54bf5e5c036efb1908232fe3d8e8e2989715fbb3
Related: [1] I393907b3c9e0cc1145e102328adad0a83ee13a9f
In general, it's safe not to use talloc API here because those are
internal allocations, and there are no 'return' statements between
calloc() and free(). However, we don't really need to initialize
the heap memory with 0, so let's use the 'normal' malloc().
Change-Id: I6956cbd83b2999dbcf8e2d210134b0a166c33efb
When the logging framework is not initialized we get an error:
"ERROR: osmo_log_info == NULL! You must call log_init() before
using logging in ..."
There are sometimes situations where some code tries to log before
logging was initialied. This is a problem because the actual log
line with the debug info we need is covered up by the error message
from the logging framework.
Lets introduce a fallback logging function that is called when the
the logging framework is not available. This function can just use
fprintf to output to stderr.
Change-Id: I9b1b0988e02322e3e44fd4ceea3e1bc2d4df3c45
The event names contain '.', and there are spaces ' ' in the state
names. This is a problem since states and events can also be monitored
via the CTRL interface. Unfortunately the CTRL interface does not allow
certain reserved characters. So lets rename the states and event names
to make them compatible with the CTRL interface.
Change-Id: Id19973b56f9d7b1e3d0b0d7c7d0be7beba5428fc
Related OS#4149
Change-Id: I5ebc9ab5b1456fee29aa4e254fae862dc053f0aa
The parameters described in the docstrings for osmo_plmn_from_bcd() do not match the actual parameter list.
Change-Id: Ic0999dbe096a98418db7482bd110e20497d8e4a5
The identifier of stats item STATS_TCP_REORD_SEEN is tcp:sndbuf_limited,
it should be tcp:reord_seen.
Change-Id: If3539ceb570ae784cc9b6567c59da7afd11acf82
Related: OS#5701
This allows init-passive users to get the configured sizes for the RFCIs
and other similar information once engotiated with the peer.
Realted: OS#1937
Change-Id: I63ee780b4aa162ea097410b234e73984000c0965
The configuration defaults for the socket statistics are currently set
to a batch size of 1. This means that only one socket per timer
expiration is scanned. This rate is probably a bit low. To speed things
up a bit we should set the default to 5. Scanning 5 sockets at a time is
still in the affordable range.
Change-Id: I87abc74c00377191f7940c5b8f19d932618fc019
Related: SYS#5701
When picking the end state, looking only at the path metric
is highly suboptimal because in a tail biting code, we _know_ that
whatever treillis path is correct, it must start and end at the same
state. So we only consider path meeting that condition. We know any
path that doesn't isn't the right one. We only fallback to only
path metric if no path met that condition.
Fixes OS#4508
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
Change-Id: I87e51d3880c0fe7bf3d6cd08fd46517a424a230c
The function clearly specified in its documentation that the number of
bytes written to the out buffer were being returned. However, the value
returned was "the number of characters (excluding the terminating null
byte) which would have been written to the final string if enough space
had been available.", aka snprintf-style.
The 2 callers of that function were not expecting it, so if a long
enough buffer was passed, the program asserted.
Closes: OS#5383
Change-Id: I8d71bd1a0dad37606acb8302b05c2ae338112e57