Scroll to navigation

SYSTEMD-JOURNALD.SERVICE(8) systemd-journald.service SYSTEMD-JOURNALD.SERVICE(8)

NAME

systemd-journald.service, systemd-journald.socket, systemd-journald-dev-log.socket, systemd-journald-audit.socket, systemd-journald - Journal service

SYNOPSIS

systemd-journald.service

systemd-journald.socket

systemd-journald-dev-log.socket

systemd-journald-audit.socket

/usr/lib/systemd/systemd-journald

DESCRIPTION

systemd-journald is a system service that collects and stores logging data. It creates and maintains structured, indexed journals based on logging information that is received from a variety of sources:

•Kernel log messages, via kmsg

•Simple system log messages, via the libc syslog(3) call

•Structured system log messages via the native Journal API, see sd_journal_print(4)

•Standard output and standard error of service units. For further details see below.

•Audit records, originating from the kernel audit subsystem

The daemon will implicitly collect numerous metadata fields for each log messages in a secure and unfakeable way. See systemd.journal-fields(7) for more information about the collected metadata.

Log data collected by the journal is primarily text-based but can also include binary data where necessary. Individual fields making up a log record stored in the journal may be up to 2^64-1 bytes in size.

The journal service stores log data either persistently below /var/log/journal or in a volatile way below /run/log/journal/ (in the latter case it is lost at reboot). By default, log data is stored persistently if /var/log/journal/ exists during boot, with an implicit fallback to volatile storage otherwise. Use Storage= in journald.conf(5) to configure where log data is placed, independently of the existence of /var/log/journal/.

On systems where /var/log/journal/ does not exist yet but where persistent logging is desired (and the default journald.conf is used), it is sufficient to create the directory, and ensure it has the correct access modes and ownership:

mkdir -p /var/log/journal
systemd-tmpfiles --create --prefix /var/log/journal

See journald.conf(5) for information about the configuration of this service.

STREAM LOGGING

The systemd service manager invokes all service processes with standard output and standard error connected to the journal by default. This behaviour may be altered via the StandardOutput=/StandardError= unit file settings, see systemd.exec(5) for details. The journal converts the log byte stream received this way into individual log records, splitting the stream at newline ("\n", ASCII 10) and NUL bytes.

If systemd-journald.service is stopped, the stream connections associated with all services are terminated. Further writes to those streams by the service will result in EPIPE errors. In order to react gracefully in this case it is recommended that programs logging to standard output/error ignore such errors. If the SIGPIPE UNIX signal handler is not blocked or turned off, such write attempts will also result in such process signals being generated, see signal(7). To mitigate this issue, systemd service manager explicitly turns off the SIGPIPE signal for all invoked processes by default (this may be changed for each unit individually via the IgnoreSIGPIPE= option, see systemd.exec(5) for details). After the standard output/standard error streams have been terminated they may not be recovered until the services they are associated with are restarted. Note that during normal operation, systemd-journald.service stores copies of the file descriptors for those streams in the service manager. If systemd-journald.service is restarted using systemctl restart or equivalent operation instead of a pair of separate systemctl stop and systemctl start commands (or equivalent operations), these stream connections are not terminated and survive the restart. It is thus safe to restart systemd-journald.service, but stopping it is not recommended.

Note that the log record metadata for records transferred via such standard output/error streams reflect the metadata of the peer the stream was originally created for. If the stream connection is passed on to other processes (such as further child processes forked off the main service process), the log records will not reflect their metadata, but will continue to describe the original process. This is different from the other logging transports listed above, which are inherently record based and where the metadata is always associated with the individual record.

In addition to the implicit standard output/error logging of services, stream logging is also available via the systemd-cat(1) command line tool.

Currently, the number of parallel log streams systemd-journald will accept is limited to 4096. When this limit is reached further log streams may be established but will receive EPIPE right from the beginning.

SIGNALS

SIGUSR1

Request that journal data from /run/ is flushed to /var/ in order to make it persistent (if this is enabled). This must be used after /var/ is mounted, as otherwise log data from /run is never flushed to /var regardless of the configuration. The journalctl --flush command uses this signal to request flushing of the journal files, and then waits for the operation to complete. See journalctl(1) for details.

SIGUSR2

Request immediate rotation of the journal files. The journalctl --rotate command uses this signal to request journal file rotation.

SIGRTMIN+1

Request that all unwritten log data is written to disk. The journalctl --sync command uses this signal to trigger journal synchronization, and then waits for the operation to complete.

KERNEL COMMAND LINE

A few configuration parameters from journald.conf may be overridden on the kernel command line:

systemd.journald.forward_to_syslog=, systemd.journald.forward_to_kmsg=, systemd.journald.forward_to_console=, systemd.journald.forward_to_wall=

Enables/disables forwarding of collected log messages to syslog, the kernel log buffer, the system console or wall.

See journald.conf(5) for information about these settings.

ACCESS CONTROL

Journal files are, by default, owned and readable by the "systemd-journal" system group but are not writable. Adding a user to this group thus enables her/him to read the journal files.

By default, each logged in user will get her/his own set of journal files in /var/log/journal/. These files will not be owned by the user, however, in order to avoid that the user can write to them directly. Instead, file system ACLs are used to ensure the user gets read access only.

Additional users and groups may be granted access to journal files via file system access control lists (ACL). Distributions and administrators may choose to grant read access to all members of the "wheel" and "adm" system groups with a command such as the following:

# setfacl -Rnm g:wheel:rx,d:g:wheel:rx,g:adm:rx,d:g:adm:rx /var/log/journal/

Note that this command will update the ACLs both for existing journal files and for future journal files created in the /var/log/journal/ directory.

FILES

/etc/systemd/journald.conf

Configure systemd-journald behavior. See journald.conf(5).

/run/log/journal/machine-id/*.journal, /run/log/journal/machine-id/*.journal~, /var/log/journal/machine-id/*.journal, /var/log/journal/machine-id/*.journal~

systemd-journald writes entries to files in /run/log/journal/machine-id/ or /var/log/journal/machine-id/ with the ".journal" suffix. If the daemon is stopped uncleanly, or if the files are found to be corrupted, they are renamed using the ".journal~" suffix, and systemd-journald starts writing to a new file. /run is used when /var/log/journal is not available, or when Storage=volatile is set in the journald.conf(5) configuration file.

When systemd-journald ceases writing to a journal file, it will be renamed to "original-name@suffix.journal" (or "original-name@suffix.journal~"). Such files are "archived" and will not be written to any more.

In general, it is safe to read or copy any journal file (active or archived). journalctl(1) and the functions in the sd-journal(3) library should be able to read all entries that have been fully written.

systemd-journald will automatically remove the oldest archived journal files to limit disk use. See SystemMaxUse= and related settings in journald.conf(5).

/dev/kmsg, /dev/log, /run/systemd/journal/dev-log, /run/systemd/journal/socket, /run/systemd/journal/stdout

Sockets and other paths that systemd-journald will listen on that are visible in the file system. In addition to these, journald can listen for audit events using netlink.

SEE ALSO

systemd(1), journalctl(1), journald.conf(5), systemd.journal-fields(7), sd-journal(3), systemd-coredump(8), setfacl(1), sd_journal_print(4), pydoc systemd.journal

systemd 239