We renamed the options recently, with the intention of keeping the old
names as supported but deprecated for some time, before removal. This
was done with the configure script options, but in the unitd binary, we
accidentally removed the old names, causing some unintended breakage.
Keep support for the old names, albeit with a deprecation message to
stderr, for some time, until we decide to remove them.
Fixes: 5a37171f73 ("Added default values for pathnames.")
Closes: <https://github.com/nginx/unit/issues/876>
Reported-by: El RIDO <elrido@gmx.net>
Acked-by: Liam Crilly <liam@nginx.com>
Acked-by: Artem Konev <a.konev@f5.com>
Acked-by: Timo Stark <t.stark@nginx.com>
Reviewed-by: Andrew Clayton <a.clayton@nginx.com>
Cc: Andrei Zeliankou <zelenkov@nginx.com>
Signed-off-by: Alejandro Colomar <alx@nginx.com>
In BSD systems, it's usually </var/db> or some other dir under </var>
that is not </var/lib>, so $statedir is a more generic name. See
hier(7).
Reported-by: Andrei Zeliankou <zelenkov@nginx.com>
Reported-by: Zhidao Hong <z.hong@f5.com>
Reviewed-by: Konstantin Pavlov <thresh@nginx.com>
Reviewed-by: Andrew Clayton <a.clayton@nginx.com>
Cc: Liam Crilly <liam@nginx.com>
Signed-off-by: Alejandro Colomar <alx@nginx.com>
If we don't remove the Unix domain listen socket file then when Unit
restarts it get an error like
2023/02/25 23:10:11 [alert] 36388#36388 bind(\"unix:/tmp/unit.sock\") failed (98: Address already in use)
This patch makes use of the listen_sockets array, that is already
allocated in the main process but never populated, to place the Unix
domain listen sockets into.
At shutdown we can then loop through this array and unlink(2) any Unix
domain sockets found therein.
Closes: <https://github.com/nginx/unit/issues/792>
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
This allows one to simply run `./configure` and expect it to
produce sane defaults for an install.
Previously, without specifying `--prefix=...`, `make install`
would simply fail, recommending to set `--prefix` or `DESTDIR`,
but that recommendation was incomplete at best, since it didn't
set many of the subdirs needed for a good organization.
Setting `DESTDIR` was even worse, since that shouldn't even affect
an installation (it is required to be transparent to the
installation).
/usr/local is the historic Unix standard path to use for
installations from source made manually by the admin of the
system. Some package managers (Homebrew, I'm looking specifically
at you) have abused that path to install their things, but 1) it's
not our fault that someone else incorrectly abuses that path (and
they seem to be fixing it for newer archs; e.g., they started
using /opt/homebrew for Apple Silicon), 2) there's no better path
than /usr/local, 3) we still allow changing it for systems where
this might not be the desired path (MacOS Intel with hombrew), and
4) it's _the standard_.
See a related conversation with Ingo (OpenBSD maintainer):
On 7/27/22 16:16, Ingo Schwarze wrote:
> Hi Alejandro,
[...]
>
> Alejandro Colomar wrote on Sun, Jul 24, 2022 at 07:07:18PM +0200:
>> On 7/24/22 16:57, Ingo Schwarze wrote:
>>> Alejandro Colomar wrote on Sun, Jul 24, 2022 at 01:20:46PM +0200:
>
>>>> /usr/local is for sysadmins to build from source;
>
>>> Doing that is *very* strongly discouraged on OpenBSD.
>
>> I guess that's why the directory was reused in the BSDs to install ports
>> (probably ports were installed by the sysadmin there, and by extension,
>> ports are now always installed there, but that's just a guess).
>
> Maybe. In any case, the practice of using /usr/local for packages
> created from ports is significantly older than the recommendation
> to refrain from using upstream "make install" outside the ports
> framework.
>
> * The FreeBSD ports framework was started by Jordan Hubbard in 1993.
> * The ports framework was ported from FreeBSD to OpenBSD
> by Niklas Hallqvist in 1996.
> * NetBSD pkgsrc was forked from FreeBSD ports by Alistair G. Crooks
> and Hubert Feyrer in 1997.
>
> I failed to quickly find Jordan's original version, but rev. 1.1
> of /usr/ports/infrastructure/mk/bsd.port.mk in OpenBSD (dated Jun 3
> 22:47:10 1996 UTC) already said
>
> LOCALBASE ?= /usr/local
> PREFIX ?= ${LOCALBASE}
>
[...]
>> I had a discussion in NGINX Unit about it, and
>> the decission for now has been: "support prefix=/usr/local for default
>> manual installation through the Makefile, and let BSD users adjust to
>> their preferred path".
>
> That's an *excellent* solution for the task, thanks for doing it
> the right way. By setting PREFIX=/usr/local by default in the
> upstream Makefile, you are minimizing the work for *BSD porters.
>
> The BSD ports frameworks will typically run the upstreak "make install"
> with the variable DESTDIR set to a custom value, for example
>
> DESTDIR=/usr/ports/pobj/groff-1.23.0/fake-amd64
>
> so if the upstream Makefile sets PREFIX=/usr/local ,
> that's perfect, everything gets installed to the right place
> without an intervention by the person doing the porting.
>
> Of course, if the upstream Makefile would use some other PREFIX,
> that would not be a huge obstacle. All we have to do in that case
> is pass the option --prefix=/usr/local to the ./configure script,
> or something equivalent if the software isn't using GNU configure.
>
>> We were concerned that we might get collisions
>> with the BSD port also installing in /usr/local, but that's the least
>> evil (and considering BSD users don't typically run `make install`, it's
>> not so bad).
>
> It's not bad at all. It's perfect.
>
> Of course, if a user wants to install *without* the ports framework,
> they have to provide their own --prefix. But that's not an issue
> because it is easy to do, and installing without a port is discouraged
> anyway.
===
Directory variables should never contain a trailing slash (I've
learned that the hard way, where some things would break
unexpectedly). Especially, make(1) is likely to have problems
when things have double slashes or a trailing slash, since it
treats filenames as text strings. I've removed the trailing slash
from the prefix, and added it to the derivate variables just after
the prefix. pkg-config(1) also expects directory variables to have
no trailing slash.
===
I also removed the code that would set variables as depending on
the prefix if they didn't start with a slash, because that is a
rather non-obvious behavior, and things should not always depend
on prefix, but other dirs such as $(runstatedir), so if we keep
a similar behavior it would be very unreliable. Better keep
variables intact if set, or use the default if unset.
===
Print the real defaults for ./configure --help, rather than the actual
values.
===
I used a subdirectory under the standard /var/lib for NXT_STATE,
instead of a homemade "state" dir that does the same thing.
===
Modified the Makefile to create some dirs that weren't being
created, and also remove those that weren't being removed in
uninstall, probably because someone forgot to add them.
===
Add new options for setting the new variables, and rename some to be
consistent with the standard names. Keep the old ones at configuration
time for compatibility, but mark them as deprecated. Don't keep the old
ones at exec time.
===
A summary of the default config is:
Unit configuration summary:
bin directory: ............. "/usr/local/bin"
sbin directory: ............ "/usr/local/sbin"
lib directory: ............. "/usr/local/lib"
include directory: ......... "/usr/local/include"
man pages directory: ....... "/usr/local/share/man"
modules directory: ......... "/usr/local/lib/unit/modules"
state directory: ........... "/usr/local/var/lib/unit"
tmp directory: ............. "/tmp"
pid file: .................. "/usr/local/var/run/unit/unit.pid"
log file: .................. "/usr/local/var/log/unit/unit.log"
control API socket: ........ "unix:/usr/local/var/run/unit/control.unit.sock"
Link: <https://www.gnu.org/prep/standards/html_node/Directory-Variables.html>
Link: <https://refspecs.linuxfoundation.org/FHS_3.0/fhs/index.html>
Reviewed-by: Artem Konev <a.konev@f5.com>
Reviewed-by: Andrew Clayton <a.clayton@nginx.com>
Tested-by: Andrew Clayton <a.clayton@nginx.com>
Reviewed-by: Konstantin Pavlov <thresh@nginx.com>
Signed-off-by: Alejandro Colomar <alx@nginx.com>
@alejandro-colomar reported an issue on GitHub whereby Unit would fail
to start due to not being able to create the control socket (a Unix
Domain Socket)
2022/08/05 20:12:22 [alert] 21613#21613 bind(6,
unix:/opt/local/unit/var/run/unit/control.unit.sock.tmp)
failed (2: No such file or directory)
This could happen if the control socket was set to a directory that
doesn't exist. A common place to put the control socket would be under
/run/unit, and while /run will exist, /run/unit may well not (/run
is/should be cleared on each boot).
The pid file would also generally go under /run/unit, though this is
created after the control socket, however it could go someplace else so
we should also ensure its directory exists.
This commit will try to create the pid file and control sockets parent
directory. In some cases the user will need to ensure that the rest of
the path already exists.
This adds a new nxt_fs_mkdir_parent() function that given a full path
to a file (or directory), strips the last component off before passing
the remaining directory path to nxt_fs_mkdir().
Cc: Konstantin Pavlov <thresh@nginx.com>
Closes: <https://github.com/nginx/unit/issues/742>
Reported-by: Alejandro Colomar <alx@nginx.com>
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Tested-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
Registering an isolated PID in the global PID hash is wrong
because it can be duplicated. Isolated processes are stored only
in the children list until the response for the WHOAMI message is
processed and the global PID is discovered.
To remove isolated siblings, a pointer to the children list is
introduced in the nxt_process_init_t struct.
This closes#633 issue on GitHub.
Generic process-to-process shared memory exchange is no more required. Here,
it is transformed into a router-to-application pattern. The outgoing shared
memory segments collection is now the property of the application structure.
The applications connect to the router only, and the process only needs to group
the ports.
The application process needs to request the port from the router instead of the
latter pushing the port before sending a request to the application. This is
required to simplify the communication between the router and the application
and to prepare the router to use the application shared port and then the queue.
Currently, the router exits without waiting for the worker threads to stop.
There is a short gap between the runtime memory pool's free and the exit, during
which a worker thread may try to access a runtime structure. In turn, this may
cause a crash. For now, it is better to keep this memory allocated.
The process abstraction has changed to:
setup(task, process)
start(task, process_data)
prefork(task, process, mp)
The prefork() occurs in the main process right before fork.
The file src/nxt_main_process.c is completely free of process
specific logic.
The creation of a process now supports a PROCESS_CREATED state. The
The setup() function of each process can set its state to either
created or ready. If created, a MSG_PROCESS_CREATED is sent to main
process, where external setup can be done (required for rootfs under
container).
The core processes (discovery, controller and router) doesn't need
external setup, then they all proceeds to their start() function
straight away.
In the case of applications, the load of the module happens at the
process setup() time and The module's init() function has changed
to be the start() of the process.
The module API has changed to:
setup(task, process, conf)
start(task, data)
As a direct benefit of the PROCESS_CREATED message, the clone(2) of
processes using pid namespaces now doesn't need to create a pipe
to make the child block until parent setup uid/gid mappings nor it
needs to receive the child pid.
One of the ways to detect Unit's startup and subsequent readiness to accept
commands relies on waiting for the control socket file to be created.
Earlier, it was unreliable due to a race condition between the client's
connect() and the daemon's listen() calls after the socket's bind() call.
Now, unix domain listening sockets are created with a nxt_listen_socket_create()
call as follows:
s = socket();
unlink("path/to/socket.tmp")
bind(s, "path/to/socket.tmp");
listen(s);
rename("path/to/socket.tmp", "path/to/socket");
This eliminates a time-lapse when the socket file is already created but nobody
is listening on it yet, which therefore prevents the condition described above.
Also, it allows reliably detecting whether the socket is being used or simply
wasn't cleaned after the daemon stopped abruptly. A successful connection to
the socket file means the daemon has been started; otherwise, the file can be
overwritten.
This is required to avoid include cycles, as some nxt_clone_* functions
depend on the credential structures, but nxt_process depends on clone
structures.
- Introduced nxt_runtime_process_port_create().
- Moved nxt_process_use() into nxt_process.c from nxt_runtime.c.
- Renamed nxt_runtime_process_remove_pid() as nxt_runtime_process_remove().
- Some public functions transformed to static.
This closes#327 issue on GitHub.
There was a typo: nxt_queue_head() used instead of nxt_queue_first() in
connection iteration loop. This prevents idle connection close on quit.
This closes#334 issue on GitHub.
Thanks to 洪志道 (Hong Zhi Dao).
Unit master process restarts the router if the router accidentally dies.
New router process receives the configuration from controller and starts
configured applications. The information of running applications cannot
be transferred to router because currently there is no persistent application
identifier. To avoid orphan application processes started by died router,
master process stops all currently running applications once it receives
SIGCHLD for router process.
The bug had appeared in 5cc5002a788e when process type has been
converted to bitmask. This commit reverts the type back to a number.
This commit is related to #131 issue on GitHub.
This also fixes#101 issue on GitHub. The function previously used to
parse IPv6 address of control socket was broken. Now the working function
is used instead.
Two different router threads may send different requests to single
application worker. In this case shared memory fds from worker
to router will be send over 2 different router ports. These fds
will be received and processed by different threads in any order.
This patch made possible to add incoming shared memory segments in
arbitrary order. Additionally, array and memory pool are no longer
used to store segments because of pool's single threaded nature.
Custom array-like structure nxt_port_mmaps_t introduced.
