- Update from version 2.8.3 to 2.8.4
- Update of rootfile
- sobump requires shipping of collectd
- Changelog
2.8.4
- Bug fixes for fallout possible due to "fightwarn" effort in 2.8.0+:
* In `usbhid-ups` sources, introduced optional `HU_FLAG_PARAM_REQUIRED` for
`setvar()` or `instcmd()` handling (and a `HU_TYPE_CMD_PARAM_REQUIRED`
shortcut) for setting in the mapping table flags, to specify variables
or instant commands that require an argument (either from caller or a
non-`NULL` default in the run-time table after device data discovery);
if the flag is not set, a zero value is assumed. Incomplete code was a
regression of NUT v2.8.3 causing some instant commands to fail. [#2860,
#2955]
- Fix fallout of development in NUT v2.8.0 and/or v2.8.1 and/or v2.8.2 and/or
v2.8.3:
* Fixed a regression in recipes of NUT v2.8.3 release (as compared to
v2.8.2), where `configure --with-docs=all` no longer failed a run
of the `configure` script when some of the required rendering tools
were not in fact available. [#2842, fixed by #2921]
* Some recipe improvements in earlier releases led to `make check` always
running a spelling check (if tools are available), even if the explicit
`configure --disable-spellcheck` option was used. Now it would not run
if disabled (e.g. to speed up CI builds in scenarios that focus on other
aspects of the code base), although developers can still use the explicit
`make spellcheck*` goals, when tools are in fact available. [#2973]
* A change in `Makefile.am` recipes to evaluate some driver names in the
`DRIVERLIST` variables inspected by `configure` script, rather than
having all their names hard-coded like before, led to inability to
`configure --with-drivers=dummy-ups`. [#2825, #2927, fixed by PR #2929]
* A problem noted with `upsdrvquery` (since NUT v2.8.1) message logging
at high debug verbosity levels (5+) with very large blocks of content
has exposed a deficiency in variable-argument handling, and specifically
adaptive resizing of the output buffer or truncation of logged inputs
(which is something NUT code tried to do since the beginning of time),
and could lead to "segmentation fault" crashes on some platforms.
[issue #2948, PR #2963]
* Documentation build recipes overly zealously pre-processed source files,
which was not applicable for each and every document type we have (e.g.
binary images for illustrations); this caused grief with some toolkits.
[issue #2989]
- common code:
* Revised common `writepid()` to use `altpidpath()` as location for the
PID file creation, if the default `rootpidpath()` is not accessible
(e.g. daemon was not initially started as `root`). Likewise updated
short PID file based signal sending to consult both locations. [#1717]
* Linux may report a `/proc/X/exe` symlink with an embedded "(deleted)"
suffix, if the binary was removed (or replaced) since the running process
started. This confused our code which verifies that when it is sending a
signal to a PID, that PID does reflect the expected NUT program. [#3021]
* Refactored NUT "common" sources to reference `nut_version.h` macros from
a smaller C source file, to minimize the compilation unit size impacted
by development iterations. [issue #2097]
* Common code hardening: added sanity-checking for dynamically constructed
or selected formatting strings with variable-argument list methods
(typically used with log printing, `dstate` setting, etc.) [#2450, #3016]
- Warn if `%n` formatting string is used -- it is deprecated in some
newer distros due to security concerns.
* Refactored repetitive implementations of `inet_ntopSS()` (nee
`inet_ntopW()` in `upsd.c`) and `inet_ntopAI()` methods into `common.c`,
so now they can be re-used or expanded more easily. [#2916]
- `upsd` updates:
* Fixed two bugs about printing the "further (ignored) addresses resolved
for this name": the way to extract IP address string was not portable
and misfired on some platforms, and the way to print had a theoretical
potential for buffer overflow. [#2915]
* Print arguments of a processed command into the debug log, to help track
down what unsupported queries are about, etc. (but only endeavor to spend
time, RAM and CPU on this if debug verbosity is high enough). Hide the
sensitive commands' parameters unless verbosity is unusually high. [#3023]
- `upsdrvquery` API updates [#2969]:
* Added `upsdrvquery_oneshot_conn()` for issuing one-shot queries using an
existing `udq_pipe_conn_t *` connection. The caller manages the
connection's lifecycle, and the function includes a best-effort call to
restore broadcast mode after the query to return the connection as it was.
* Added `upsdrvquery_oneshot_sockfn()` for initiating one-shot queries using
a socket filename. Shares internal logic with the existing
`upsdrvquery_oneshot()`, which uses a UPS and driver name, respectively.
* Introduced `upsdrvquery_restore_broadcast()` to explicitly restore
broadcast mode (`BROADCAST 1`) on a connection, helping return it to a
consistent and talkative state.
* Revised connection ownership handling: internal functions like
`upsdrvquery_prepare()` and `upsdrvquery_request()` no longer close
connections they do not own. Responsibility for cleanup is now delegated to
the caller to avoid unintended side effects and better align with expected
usage patterns.
- common driver code:
* Update reports of failed socket file creation, to help troubleshooting
some error cases in the field. [#2959]
* Removed workarounds trying to migrate legacy driver raised `ALARM`
status tokens into modern `alarm_*` function logic. Rather, we keep
supporting them as separate from the modern logic, seeing as `upsmon`
does not care where the token itself was raised for its notifications.
Driver-code related test-cases were updated to reflect these changes.
[issue #2928, PRs #2931 and #2934]
* Introduced some macros in `drivers/upshandler.h` for common syslog level
definitions and message wording for beginning and failing `instcmd()` or
`setvar()` operations consistently in different drivers. As a related
change, operations that intend to turn off or restart the load, or can
do that by side effect (e.g. calibration if batteries are old or dead),
would explicitly `upslogx(LOG_CRIT,...)` by default before commencing.
[#2957]
* Fixed a couple of ancient memory leaks: one "shared" during driver
program initialization, and one specific to `dummy-ups` wind-down. [#2972]
* Added a `suggest_NDE_conflict()` method so drivers which lack access
to the expected device can consistently suggest that this may be because
of running both an NDE-wrapped service unit and a manually launched
driver program at the same time. Currently added to `libusb{0,1}.c`
code, but may later be expanded to e.g. serial drivers and other media,
when their behavior in such situations gets identified. [follow-up to
issue #477, PR #3041]
- `apc_modbus` driver updates:
* The time stamp and inter-frame delay accounting was fixed, alleviating
one of the problems reported in issue #2609. [PR #2982]
* Fix missing variables due to mismatching format string. [PR #3013]
- `bcmxcp` driver updates:
* The latching on to a previous replace battery status was fixed, with its
alarm state variable now correctly being reset; previously a factually
replaced battery did not clear the alarm and the whole driver needed to
be restarted. [issue #2999, PR #3002]
- `clone`, `clone-outlet`, `nhs_ser` driver and `nutdrv_qx_ablerex`
subdriver updates:
* Refactored to follow modern handling of status and alarm conditions,
aligning with current driver design practices. This includes fixing
copy-paste related issues in alarm reporting and removing some alarm
messages that should instead be reflected as status flags. [#2936]
- `dummy-ups` driver updates:
* A new instruction `ALARM` was added for the `Dummy Mode` operation
of the driver, enabling simulation of UPS alarm states more closely
in line with modern, real-world UPS driver implementations. This
follows the updated principle of keeping alarm states decoupled from
the `ups.status` variable, with alarms now raised via common alarm
functions rather than direct manipulation. [issue #2928, PR #2936]
- `nutdrv_qx` driver updates:
* Added support for "preprocess"/"process" methods called from mapping tables
to report back to the driver that an argument value was not supported,
so `setvar()` or `instcmd()` can not proceed safely and should return
`STAT_SET_CONVERSION_FAILED` or `STAT_INSTCMD_CONVERSION_FAILED`. [#3017]
* Introduced `innovart33` protocol support for Ippon Innova RT 3/3 topology
UPSes. [#2938]
* Updated `megatec` protocol for more detailed responses to `I` query
which may return `ups.serial` (after a shorter `device.mfr`) and the
`battery.runtime` (after a shorter `device.model`). Note that the
expected response is shorter than in other dialects (38 vs. 39 bytes),
so if this change breaks anything for your UPS that reported the values
above correctly (e.g. the `ups.firmware` version becomes shorter or
none of these are reported), please let NUT developers know. [#2980]
* Revised `voltronic` protocol to suppress alarm "UPS is in ECO Mode",
using "buzzword mode" settings more correctly than in the previous
iteration, shipped in NUT v2.8.3 release (as PR #2750 for issue #2708).
[issue #2494]
* Introduced a `voltronic-axpert` subdriver for Voltronic Axpert inverters
which speak the P30 protocol, currently in a highly experimental state:
with initial support for query commands, but most values are "hidden"
from default NUT builds by being defined in `experimental.*` namespace,
and should also be enabled by `configure --with-unmapped-data-points`.
Development was based on work done in the Voltronic Sunny subdriver in
https://github.com/nickma82/nut/tree/nutdrv_qx_voltronic-sunny_rebased%2Bcommand
[#1407]
- `phoenixcontact_modbus` driver updates:
* Added more settings that can be tuned -- support for shutdown variables,
UPS mode selector, PC reset delay after main power recovers, and
automatic switch to battery mode (and back) if main power is below
or above a defined threshold (see the new "Configurable Values" section
in the man page). They can be configured via `default.*` values in
`ups.conf`. [#2986]
- `pijuice` driver updates:
* Converted to NUT standard use of `status_set()` with single-token values.
[issue #2708]
- `snmp-ups` driver updates:
* Added support for "fun"/"nuf" methods called from mapping tables to
report back to the driver that an argument value was not supported,
so `setvar()` or `instcmd()` can not proceed safely and should return
`STAT_SET_CONVERSION_FAILED` or `STAT_INSTCMD_CONVERSION_FAILED`. [#3017]
* Fixed `ups.test.date` to be semi-static in `apc-mib` mapping, so it
would be queried more than once per driver up-time. [issue #3011]
* Fixed debug-logging around `SU_FLAG_STATIC` entries to clarify when
they get skipped. [issue #3011]
- `usbhid-ups` driver updates:
* Added support for "fun"/"nuf" methods called from mapping tables to
report back to the driver that an argument value was not supported,
so `setvar()` or `instcmd()` can not proceed safely and should return
`STAT_SET_CONVERSION_FAILED` or `STAT_INSTCMD_CONVERSION_FAILED`. [#3017]
* `hid_ups_walk(HU_WALKMODE_INIT)`: report if exactly one of "fun" or "nuf"
dynamic value mapping methods is defined in a one-line table, and this
may preclude reads/writes of that variable. [#2956]
* The `cps-hid` subdriver's existing mechanism for fixing broken report
descriptors was extended to cover a newly reported case of nominal UPS
power being incorrectly reported due to an unrealistically low maximum
threshold, as seen with a EC850LCD device. [issue #2917, PR #2919]
* Further revision of "ECO mode" related code in `mge-hid` subdriver,
following up from work started for NUT v2.8.3 release. [PR #2956]
* Added APC BVKxxxM2 and BKxxxM2-CH to list of devices where
`lbrb_log_delay_sec=N` may be necessary to address spurious LOWBATT
and REPLACEBATT events. [PR #2942, PR #3007, issue #2347, issue #3006]
- New NUT drivers:
* Introduced a `ve-direct` driver for Victron Energy UPS/solar panels
monitoring. Most specific reported values are in an `experimental.*`
namespace, as a community we need to come up with standard naming for
those via `docs/nut-names.txt`. [#440]
* Introduced a `nutdrv_hashx` driver for numerous devices from Ablerex,
Atlantis Land, Epyc, Infosec, ION, PowerWalker, Right Power Technology,
Salicru, UPS Solutions and other vendors (originally shipped with a
"PowerMaster+", "PowerMaster" or "PowerGuide" software companion suite).
This seems to be a protocol developed by Cyber Energy for serial-port
devices, subsequently used by different vendors in their own products
or re-branded Cyber Energy creations. [#2940]
* Introduced a `failover` driver for monitoring multiple UPS driver sockets
and seamless switching out of UPS data in a failover situation, includes
support for end-to-end tracked instant commands and also variable updating.
[#2962]
* Introduced USB (`powervar_cx_usb`) and Serial (`powervar_cx_ser`) drivers
for Powervar CUSPP protocol, tested with GTS (USB) and UPM (USB, Serial)
models. [#2988]
- The `nut-driver-enumerator.sh` script (NDE) updates:
* Now NDE internally tracks dependency of one driver on another one that
should be locally running to serve the "original" data points (`clone`,
`clone-outlet`, `dummy-ups`, `failover`). It should create "soft"
dependencies between respective service instances to order their
start-up sequence. [#2962]
* Fixed NDE to not consider "masked" systemd units as non-existent or
as syntactically failed instantiated unit names. [#3033]
- NUT Monitor GUI:
* Ported Python 3 version to Qt6, now shipped alongside Qt5 for systems
with either or both, maximizing compatibility with old and new setups.
[#2946]
- `upsmon` client:
* Clearer debug logging of `SHUTDOWNCMD` and `NOTIFYCMD` that would be used
(or warnings that none was set); flush output buffers after these messages
and after each main loop cycle, so any emitted text is seen in a timely
manner. [issue #3003, PR #3008]
- The `nutshutdown` script (end-game integration for UPS power-off in case
of FSD initiated by `upsmon`) was updated to consider `MODE=none` set in
`nut.conf` and bail out quietly. [issue #2935, PR #3008]
- Manual page recipes and contents:
* Introduced handling (possibly rewriting) for man page section "Overviews,
conventions, and miscellaneous" (commonly number 7), to deliver support
for `man nut` queries (NUT overview manual page also created). [#2945]
* A new `configure --with-docs-man-dir-as-base` option was introduced so
that directories for man page sections can now be automatically named
as either "base" number of the section (e.g. `man1`) or by full section
name (`man1m`), as different OS distributions have different preferences
in this regard. [#2950]
* Option to `configure --enable-docs-man-for-progs-built-only` was added,
to differentiate NUT builds that deliver man pages for only built programs
(legacy default) or for all of them (as needed for docs sites). [#2976]
* Option to `configure --enable-docs-changelog` was added, specifically
to allow developer iterations to not waste CPU time rebuilding the huge
`ChangeLog*` files whenever their Git index changes. [#3019]
* Options to `configure --with-docs-changelog-start` and/or
`configure --with-docs-changelog-end` were added to allow developers
to customize the size of `ChangeLog*` files when they are generated.
Default starting value is `auto` which applies the legacy default
`v2.6.0` to release/pre-release builds, or when local Git version info
could not be retrieved, and the most-recent release tag (or `master`
as fallback) for usual build iterations. Default ending value is `HEAD`
for the current git commit at the moment the ChangeLog is (re-)generated.
Balancing against the option to not build `ChangeLog*` files at all,
this couple allows quicker builds that exercise all relevant recipe
code paths. [#3019]
- Extended the `gitlog2changelog.py` helper script to report start/end commits
actually used, and to allow callers to tweak them better (not only `HEAD`
for the end of range); this may be of interest to other projects which use
this script. Allow `configure` to disable generation of either certain
`ChangeLog*` rendering formats or completely, to speed up developer
iterations (much time is wasted when dev-testing new code, due to git
index changes if NUT was configured to build with documentation). [#3019]
- The `BUILD_TYPE=default-all-errors ci_build.sh` script handling was
revised to simplify code, and to default in CI builds to a quicker
mode which randomly mixes the selected SSL, USB and UNMAPPED variants
(and relies on the dozens of NUT CI farm runs per iteration to likely
cover all possible combinations), which should roughly halve the CI
build times. Default activity for developer builds should remain as
it was -- to try each such "axis" sequentially. [#2973]
- Revised generation of links to external manual pages in HTML rendering
of NUT manual pages (previous recipe iterations left DocBook XML `ulink`
tag "as is", which was not understood by web browsers).
[follow-up to PR #2797]
- Made the distro-dependent URL template for man pages configurable.
[follow-up to PR #2797]
- Revised `make install-as-root` to fall back to legacy ways of enabling
services, if `systemctl preset-all` fails (assumed due to a systemd 252
bug). [#3022]
- Added a `make check-parallel-builds` recipe to help troubleshoot recipes
in sub-directories, and improved build-ability of existing NUT sources
starting from scratch there. This is a workflow useful for NUT development
(e.g. to focus only on drivers, or tests, or nut-scanner) but not so much
for end-user packaging where everything builds from the root directory.
[PR #3030, follows up from PR #2825, highlights why issue #2584 better
be solved]
- Revised `appveyor.yml` to run CI builds faster (forfeit MSYS2 ecosystem
updates and some other steps) and more likely fit in one-hour allocation.
Also have it install `mingw-w64-x86_64-python-pyqt6` so the `NUT-Monitor`
application can get packaged (would need a capable Python run-time though).
[#3046]
Signed-off-by: Adolf Belka <adolf.belka@ipfire.org> Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
projects / ipfire-2.x.git / commit
