Setting `DEBUG_MEM_LEVEL` in the build phase prevents
`setmemdebug()` to call `getargnum`, which would use `dracut-getarg`
in the initrd runtime phase.
No more:
```
modules.d/99base/dracut-lib.sh: line 217: dracut-getarg: command not found
```
Patch 2fabaaa62d changed the behaviour for `dash`
under the assumption, that dash does not take parameters for `.` aka
`source`. Although this is true, the original positional parameters of
the `source_all` function are still in place, so everything is
fine with the old way of sourcing.
The kernel has an odd way to handle `"` surrounded parameters.
To handle the parameters as the kernel would do, no simple shell script
suffices, so a new utility `dracut-util` is introduced. Written in "C"
it handles `dracut-getarg` and `dracut-getargs` as the old shell script
functions `_dogetarg` and `_dogetargs` would.
Library file modules.d/99base/dracut-lib.sh contains initialization code
that makes sure certain directories and links are created when a system
runs from the produced initramfs. This file is normally sourced only
from initramfs scripts. However, modules.d/99base/module-setup.sh
includes it also when an initramfs image is being created to gain access
to some of the functions in this library. This can result in creating
unexpected directories and links on the current root file system when
the dracut command is executed.
For instance, during a system installation when dracut is invoked to
create an initial initramfs, the target sysroot might not contain
directory /run/initramfs but have /var/log. This situation results in
the code in modules.d/99base/dracut-lib.sh creating an unexpected link
/var/log/log that points to /run/initramfs/log. This link is then also
present on the installed system.
The patch fixes the problem by skipping the described logic in
modules.d/99base/dracut-lib.sh when the PREFIX variable is non-empty
which indicates that a target initramfs is being prepared. The variable
is set by modules.d/99base/module-setup.sh prior to including
dracut-lib.sh.
In commit 49c4172 all shell based memory tracing functions are removed,
there are some left over. Remove them as well.
Signed-off-by: Kairui Song <kasong@redhat.com>
This feature could be off loaded to memstrack, which have better
accurecy, better performance, and have more detailed tracing features.
Also simplify make_trace_mem a bit.
And currently rd.memdebug=4 is unstable, fails from time to time.
According to POSIX.1-2017, 2.6.2 Parameter Expansion:
${parameter%[word]} [...] The word shall be expanded to produce a
pattern.
This means if word contains variables that itself contain special
characters like asterisks or backslashes, these are treated as pattern
characters unless the variable is quoted. Try e.g. the following example
in bash, dash or (busybox) ash:
i='a\c'; j='\'; echo "${i%$j*}"
This prints "a\c" because "$j*" is expanded to "\*", escaping the
asterisk. In contrast,
i='a\c'; j='\'; echo "${i%"$j"*}"
produces the expected result "a" because the backslash is not specially
treated any more after quoting.
The quotes that this commit adds have been previously removed in commit
f9c96cf56f, citing issues with busybox
hush without further specifying the actual error. I tested a recent
busybox build (upstream commit 9aa751b08ab03d6396f86c3df77937a19687981b)
and couldn't find any problems. Note that the above example always
produces "a\c" in hush regardless of quoting $j, making hush unsuitable
for use with dracut, but using quotes in parameter expansions generally
works.
The unquoted variables break the "rd.luks.uuid/name" kernel command line
options in dracut 050 because
str_replace "$luksname" '\' '\\'
in modules.d/90crypt/parse-crypt.sh is not able to escape the
backslashes any more, see GH-723, GH-727: backslashes in the
systemd-cryptsetup@.service unit name stay unescaped for use in udev
(cf. commit 0f6d93eb9d), leading to
failures in starting the unit.
This partially reverts commit f9c96cf56f.
For the shell scripts, new environment variables were introduced.
dracutsysrootdir is the root directory, file existence checks use it.
DRACUT_LDCONFIG can override ldconfig with a different one that works
on the sysroot with foreign binaries.
DRACUT_LDD can override ldd with a different one that works
with foreign binaries.
DRACUT_TESTBIN can override /bin/sh. A cross-compiled sysroot
may use symlinks that are valid only when running on the target
so a real file must be provided that exist in the sysroot.
DRACUT_INSTALL now supports debugging dracut-install in itself
when run by dracut but without debugging the dracut scripts.
E.g. DRACUT_INSTALL="valgrind dracut-install or
DRACUT_INSTALL="dracut-install --debug".
DRACUT_COMPRESS_BZIP2, DRACUT_COMPRESS_LBZIP2, DRACUT_COMPRESS_LZMA,
DRACUT_COMPRESS_XZ, DRACUT_COMPRESS_GZIP, DRACUT_COMPRESS_PIGZ,
DRACUT_COMPRESS_LZOP, DRACUT_COMPRESS_ZSTD, DRACUT_COMPRESS_LZ4,
DRACUT_COMPRESS_CAT: All of the compression utilities may be
overridden, to support the native binaries in non-standard places.
DRACUT_ARCH overrides "uname -m".
SYSTEMD_VERSION overrides "systemd --version".
The dracut-install utility was overhauled to support sysroot via
a new option -r and fixes for clang-analyze. It supports
cross-compiler-ldd from
https://gist.github.com/jerome-pouiller/c403786c1394f53f44a3b61214489e6f
DRACUT_INSTALL_PATH was introduced so dracut-install can work with
a different PATH. In a cross-compiled environment (e.g. Yocto), PATH
points to natively built binaries that are not in the host's /bin,
/usr/bin, etc. dracut-install still needs plain /bin and /usr/bin
that are relative to the cross-compiled sysroot.
The hashmap pool allocate_tile/deallocate_tile code was removed
because clang-analyze showed errors in it. hashmap_copy was removed
because it wasn't used and clang-analyze showed errors in it.
DRACUT_INSTALL_LOG_TARGET and DRACUT_INSTALL_LOG_LEVEL were
introduced so dracut-install can use different settings from
DRACUT_LOG_TARGET and DRACUT_LOG_LEVEL.
Signed-off-by: Böszörményi Zoltán <zboszor@pr.hu>
I'd like to rework CoreOS Ignition (which runs in the initramfs)
to include some values from the *real* `/etc/os-release` in
HTTP headers.
Looking at this, it turns out dracut eats almost all of the useful
information from it. I don't think `dracut` should be the `ID`
here...dracut's not an OS itself, it's a way to *build* little
operating systems. It'd be kind of like if Fedora's Koji
injected itself into `/etc/os-release`.
This code dates back a long time; not sure of all the rationale
behind it.
I changed it so that we keep extending the VERSION/PRETTY_NAME
with the dracut version, but otherwise "pass through" the
rest of the real `/etc/os-release` we were built from unchanged.
When DRACUT_SYSTEMD is set and DRACUT_QUIET=yes, vinfo returns 1. This
is a problem for hooks which end with vinfo, as then the hook returns 1.
Especially problematic if this is a shutdown hook, as it will be
restarted again and again.
This commit fixes that.
Signed-off-by: Arnaud Rebillout <arnaud.rebillout@collabora.com>
The commit 9f3c31cd8d ("99base: enable initqueue if extra devices are added")
only covers 'dracut --add-device' case, but it did not cover 'dracut --mount'
case, which causes the kdump failure in the Amazon virtual machine.
Lets make sure that the initqueue is enabled in both cases in order to wake up
the device in time.
Reported-by: Xiao Liang <xiliang@redhat.com>
Signed-off-by: Lianbo Jiang <lijiang@redhat.com>
- allow emergency login on every console
specified in the kernel cmdline
- require password for hostonly images
- emergency mode: Manually multiplex emergency infos
This will bring all vital information to all ttys specified
as console devices, regardless of wether they hold the C flag.
Reference: FATE#325386
Reference: #449
On my system the following initrd-release is generated:
...
VERSION="4 dracut-048 dracut-048"
...
VERSION is not defined in /etc/os-release, so the variable is
concatenated with its previous value:
* "4" comes from the kernel build system since dracut is called from the
kernel install hook ("4" is a major kernel version);
* first "dracut-048" comes from the "systemd-initrd" module;
* second "dracut-048" comes from the "base" module.
When extra devices are added, initqueue should be enabled to make sure
those devices are present, so following services and routines could
use those devices.
See PR #442 for more detail.
If a process (maybe plymouth) was still pinning /oldroot, then shutdown
would
- kill -9 $pid
- umount_a
- umount_a
in a very short timeframe. A small sleep hopefully lets the scheduler free
up /oldroot in the mean time.
JobRunningTimeoutSec now affects how long can start jobs for device
units stay in the "running" state. Disabling default job timeout via
JobTimeoutSec=0 doesn't disable running state timeout. We need to set
running state timeout as well.
Note that doing this the other way around has effect on generic timeout,
i.e. disabling running state timeout disables generic timeout. But doing
it this way we would create implicit dependency on fairly new
systemd-234. However, by setting both options we don't create dependency
on specific systemd version.
Extend "rd.memdebug" to "4", and "make_trace_mem" to "4+:komem".
Add new "cleanup_trace_mem" to cleanup the trace if active.
Signed-off-by: Xunlei Pang <xlpang@redhat.com>
The current method for memory debug is to use "rd.memdebug=[0-3]",
it is not enough for debugging kernel modules. For example, when we
want to find out which kernel module consumes a large amount of memory,
"rd.memdebug=[0-3]" won't help too much.
A better way is needed to achieve this requirement, this is useful for
kdump OOM debugging.
The principle of this patch is to use kernel trace to track slab and
buddy allocation calls during kernel module loading(module_init), thus
we can analyze all the trace data and get the total memory consumption.
As for large slab allocation, it will probably fall into buddy allocation,
thus tracing "mm_page_alloc" alone should be enough for the purpose(this
saves quite some trace buffer memory, also large free is quite unlikey
during module loading, we neglect those memory free events).
The trace events include memory calls under "tracing/events/":
kmem/mm_page_alloc
We also inpect the following events to detect the module loading:
module/module_load
module/module_put
Since we use filters to trace events, the final trace data size won't
be too big. Users can adjust the trace buffer size via "trace_buf_size"
kernel boot command line as needed.
We can get the module name and task pid from "module_load" event which
also mark the beginning of the loading, and module_put called by the
same task pid implies the end of the loading. So the memory events
recorded in between by the same task pid are consumed by this module
during loading(i.e. modprobe or module_init()).
With these information, we can record the rough total memory(the larger,
the more precise the result will be) consumption involved by each kernel
module loading.
Thus we introduce this shell script to find out which kernel module
consumes a large amount of memory during loading. Use "rd.memdebug=4"
as the tigger.
After applying this patch and specifying "rd.memdebug=4", during booting
it will print out something extra like below:
0 pages consumed by "pata_acpi"
0 pages consumed by "ata_generic"
1 pages consumed by "drm"
0 pages consumed by "ttm"
0 pages consumed by "drm_kms_helper"
835 pages consumed by "qxl"
0 pages consumed by "mii"
6 pages consumed by "8139cp"
0 pages consumed by "virtio"
0 pages consumed by "virtio_ring"
9 pages consumed by "virtio_pci"
1 pages consumed by "8139too"
0 pages consumed by "serio_raw"
0 pages consumed by "crc32c_intel"
199 pages consumed by "virtio_console"
0 pages consumed by "libcrc32c"
9 pages consumed by "xfs"
From the print, we see clearly that "qxl" consumed the most memory.
This file will be installed as a separate executable named "tracekomem"
in the following patch.
Signed-off-by: Xunlei Pang <xlpang@redhat.com>
If the initramfs was built with prefix=/run/... /run can't be mounted
with noexec, otherwise no binary can be run.
Guard against it by looking where /bin/sh is really located.
How to reproduce:
host# ./dracut.sh -o 'dracut-systemd systemd systemd-initrd' --local -f ./initramfs.img
host# qemu-system-x86_64 -initrd ./initramfs.img \
-append 'root=/dev/sda1 rd.cmdline=ask rd.hostonly=0' \
...
Enter additional kernel command line parameter (end with ctrl-d or .)
> rd.break
> .
...
There is no "Break before switch_root"
...
Signed-off-by: Evgeny Vereshchagin <evvers@ya.ru>
crypt/parse-crypt.sh generate initqueue job which always call
dev_unit_name() with an argument beginning with "-". This results
in the following error:
dracut-initqueue[307]: + systemd-escape -p -cfb4aa43-2f02-4c6b-a313-60ea99288087
dracut-initqueue[307]: systemd-escape: invalid option -- 'c'
Harald Hoyer