Let#s make these helpers useful even without _cleanup_ logic, to destory
arbitary fields: make them OK wiht a NULL pointer as input, and always
return one as output.
So far, we would allow certain control characters (NL since
b4346b9a77, TAB since 6294aa76d8), but not others. Having
other control characters in environment variable *value* is expected and widely
used, for various prompts like $LESS, $LESS_TERMCAP_*, and other similar
variables. The typical environment exported by bash already contains a dozen or
so such variables, so programs need to handle them.
We handle then correctly too, for example in 'systemctl show-environment',
since 804ee07c13. But we would still disallow setting such variables
by the user, in unit file Environment= and in set-environment/import-environment
operations. This is unexpected and confusing and doesn't help with anything
because such variables are present in the environment through other means.
When printing such variables, 'show-environment' escapes all special
characters, so variables with control characters are plainly visible.
In other uses, e.g. 'cat -v' can be used in similar fashion. This would already
need to be done to suppress color codes starting with \[.
Note that we still forbid invalid utf-8 with this patch. (Control characters
are valid, since they are valid 7-bit ascii.) I'm not sure if we should do
that, but since people haven't been actually asking for invalid utf-8, and only
for control characters, and invalid utf-8 causes other issues, I think it's OK
to leave this unchanged.
Fixes#4446, https://gitlab.gnome.org/GNOME/gnome-session/-/issues/45.
DNS_HOSTNAME_MAX is used by sd-resolve, but it was defined in
src/shared/dns-domain.h. The library libsystemd should be independent
of source files under src/shared.
Functions defined by us are supposed to return negative errno-style
errors on errors. laccess() is for access() what lstat() is for stat(),
but defined by us as a macro. This led to some confusion regarding error
handling.
Let's return a negative errno code just in case. This means callers can
it use either way: like access(), i.e. checking for a negative return
value + looking at errno, or like our own code, i.e. using the negative
errno code it returns.
Getting the numbers right for all architectures has proven to be a
constant chore. Let's autogenerate the header from the tables that
were imported in one of the previous commits.
Fixes#18074. (Hopefully. I cannot verify this on all architectures.)
To update the lists, or to update the header after template changes:
ninja -C build update-syscall-tables update-syscall-header
Note: the generated file is saved in git. Initially I wanted to only
store the tables in git, and generate the header during each build.
Generation is quick enough, but the header is used in many many
places (wherever missing_syscall.h is included, directly or indirectly),
which means that we would need to declare the dependency in meson, so
the header would be generated early enough. This turned out to be very
noisy. Storing the generated header in version control avoids the hassle.
Sometimes, non-ramfs initrd root are useful. Eg, for kdump, because
initramfs is memory consuming, so mount a compressed image in earlier
initrd, chroot into it then let systemd do the rest of job is a good
solution.
But systemd doesn't recognize the initrd environment if rootfs is not a
temporary fs. This is a reasonable check, because switch-root in initrd
will wipe the whole rootfs, will be a disaster if there are any
misdetect.
So extend SYSTEMD_IN_INITRD environment variable, now it accepts boolean
value and two extra keyword, "auto" and "lenient". "auto" is same as
before, and it's the default value. "lenient" will let systemd bypass
the rootfs check.
The `quote` char variable ectually contains a character, not a pointer
or boolean. hence do an explicit comparison rather than rely on C's
downgrade to bool feature, as per our coding style.
This uses GREEDY_ALLOC_ROUND_UP() to grow the allocation size
exponentially. This should speed allocation loops up a bit, given
that we often call strextend() repeatedly in a loop on the same
buffer.
If users do not enable a service like systemd-time-wait-sync.target
(because they don't want to delay boot for external events, such as an
NTP sync), then timers should still take the the weaker time-set.target
feature into account, so that the clock is at least monotonic.
Hence, order timer units after both of the targets: time-sync.target
*and* time-set.target. That way, the right thing will happen regardless
if people have no NTP server (and thus also no
systemd-time-wait-sync.service or equivalent) or, only have an NTP
server (and no systemd-time-wait-sync.service), or have both.
Ordering after time-set.target is basically "free". The logic it is
backed by should be instant, without communication with the outside
going on. It's useful still so that time servers that implement the
timestamp from /var/ logic can run in later boot.
So the currentl and only fd_is_mount_point() check is actually entirely
bogus: it passes "/" as filename argument, but that's not actually a
a valid filename, but an absolute path.
fd_is_mount_point() is written in a way tha the fd refers to a directory
and the specified path is a file directly below it that shall be
checked. The test call actually violated that rule, but still expected
success.
Let's fix this, and check for this explicitly, and refuse it.
Let's extend the test and move it to test-mountpoint-util.c where the
rest of the tests for related calls are placed.
Replaces: #18004Fixes: #17950
Susant Sahani
Lennart Poettering
Zbigniew Jędrzejewski-Szmek
Yu Watanabe
Kairui Song
Susant Sahani
Matthias Klumpp
Luca Boccassi