sd_journal_get_fd, sd_journal_get_events, sd_journal_get_timeout, sd_journal_process, sd_journal_wait, sd_journal_reliable_fd, SD_JOURNAL_NOP, SD_JOURNAL_APPEND, SD_JOURNAL_INVALIDATE — Journal change notification interface
#include <systemd/sd-journal.h>
int sd_journal_get_fd( | sd_journal *j) ; |
int sd_journal_get_events( | sd_journal *j) ; |
int sd_journal_get_timeout( | sd_journal *j, |
uint64_t *timeout_usec) ; |
int sd_journal_process( | sd_journal *j) ; |
int sd_journal_wait( | sd_journal *j, |
uint64_t timeout_usec) ; |
int sd_journal_reliable_fd( | sd_journal *j) ; |
sd_journal_get_fd()
returns a file
descriptor that may be asynchronously polled in an external event
loop and is signaled as soon as the journal changes, because new
entries or files were added, rotation took place, or files have
been deleted, and similar. The file descriptor is suitable for
usage in
poll(2).
Use sd_journal_get_events()
for an events
mask to watch for. The call takes one argument: the journal
context object. Note that not all file systems are capable of
generating the necessary events for wakeups from this file
descriptor for changes to be noticed immediately. In particular
network files systems do not generate suitable file change events
in all cases. Cases like this can be detected with
sd_journal_reliable_fd()
, below.
sd_journal_get_timeout()
will ensure in these
cases that wake-ups happen frequently enough for changes to be
noticed, although with a certain latency.
sd_journal_get_events()
will return the
poll()
mask to wait for. This function will
return a combination of POLLIN
and
POLLOUT
and similar to fill into the
".events
" field of struct
pollfd
.
sd_journal_get_timeout()
will return a
timeout value for usage in poll()
. This
returns a value in microseconds since the epoch of
CLOCK_MONOTONIC
for timing out
poll()
in timeout_usec
.
See
clock_gettime(2)
for details about CLOCK_MONOTONIC
. If there
is no timeout to wait for, this will fill in (uint64_t)
-1
instead. Note that poll()
takes
a relative timeout in milliseconds rather than an absolute timeout
in microseconds. To convert the absolute 'us' timeout into
relative 'ms', use code like the following:
uint64_t t; int msec; sd_journal_get_timeout(m, &t); if (t == (uint64_t) -1) msec = -1; else { struct timespec ts; uint64_t n; clock_gettime(CLOCK_MONOTONIC, &ts); n = (uint64_t) ts.tv_sec * 1000000 + ts.tv_nsec / 1000; msec = t > n ? (int) ((t - n + 999) / 1000) : 0; }
The code above does not do any error checking for brevity's
sake. The calculated msec
integer can be passed
directly as poll()
's timeout
parameter.
After each poll()
wake-up
sd_journal_process()
needs to be called to
process events. This call will also indicate what kind of change
has been detected (see below; note that spurious wake-ups are
possible).
A synchronous alternative for using
sd_journal_get_fd()
,
sd_journal_get_events()
,
sd_journal_get_timeout()
and
sd_journal_process()
is
sd_journal_wait()
. It will synchronously wait
until the journal gets changed. The maximum time this call sleeps
may be controlled with the timeout_usec
parameter. Pass (uint64_t) -1
to wait
indefinitely. Internally this call simply combines
sd_journal_get_fd()
,
sd_journal_get_events()
,
sd_journal_get_timeout()
,
poll()
and
sd_journal_process()
into one.
sd_journal_reliable_fd()
may be used to check whether the wake-up events from
the file descriptor returned by sd_journal_get_fd()
are known to be quickly
triggered. On certain file systems where file change events from the OS are not available (such as NFS)
changes need to be polled for repeatedly, and hence are detected only with a considerable latency. This
call will return a positive value if the journal changes are detected quickly and zero when they need to
be polled for. Note that there is usually no need to invoke this function directly as
sd_journal_get_timeout()
will request appropriate timeouts anyway.
Note that all of the above change notification interfaces do not report changes instantly. Latencies are introduced for multiple reasons: as mentioned certain storage backends require time-based polling, in other cases wake-ups are optimized by coalescing events, and the OS introduces additional IO/CPU scheduling latencies.
sd_journal_get_fd()
returns a valid
file descriptor on success or a negative errno-style error
code.
sd_journal_get_events()
returns a
combination of POLLIN
,
POLLOUT
and suchlike on success or a negative
errno-style error code.
sd_journal_reliable_fd()
returns a
positive integer if the file descriptor returned by
sd_journal_get_fd()
will generate wake-ups
immediately for all journal changes. Returns 0 if there might be a
latency involved.
sd_journal_process()
and sd_journal_wait()
return a negative
errno-style error code, or one of SD_JOURNAL_NOP
, SD_JOURNAL_APPEND
or
SD_JOURNAL_INVALIDATE
on success:
If SD_JOURNAL_NOP
is returned, the journal did not change since the last
invocation.
If SD_JOURNAL_APPEND
is returned, new entries have been appended to the end
of the journal. In this case it is sufficient to simply continue reading at the previous end location of the
journal, to read the newly added entries.
If SD_JOURNAL_INVALIDATE
, journal files were added to or removed from the
set of journal files watched (e.g. due to rotation or vacuuming), and thus entries might have appeared or
disappeared at arbitrary places in the log stream, possibly before or after the previous end of the log
stream. If SD_JOURNAL_INVALIDATE
is returned, live-view UIs that want to reflect on screen
the precise state of the log data on disk should probably refresh their entire display (relative to the cursor of
the log entry on the top of the screen). Programs only interested in a strictly sequential stream of log data may
treat SD_JOURNAL_INVALIDATE
the same way as SD_JOURNAL_APPEND
, thus
ignoring any changes to the log view earlier than the old end of the log stream.
In general, sd_journal_get_fd()
, sd_journal_get_events()
, and
sd_journal_get_timeout()
are not "async signal safe" in the meaning of
signal-safety(7).
Nevertheless, only the first call to any of those three functions performs unsafe operations, so subsequent calls
are safe.
sd_journal_process()
and sd_journal_wait()
are not
safe. sd_journal_reliable_fd()
is safe.
All functions listed here are thread-agnostic and only a single specific thread may operate on a given object during its entire lifetime. It's safe to allocate multiple independent objects and use each from a specific thread in parallel. However, it's not safe to allocate such an object in one thread, and operate or free it from any other, even if locking is used to ensure these threads don't operate on it at the very same time.
These APIs are implemented as a shared
library, which can be compiled and linked to with the
libsystemd
pkg-config(1)
file.
Iterating through the journal, in a live view tracking all changes:
#include <stdio.h> #include <string.h> #include <systemd/sd-journal.h> int main(int argc, char *argv[]) { int r; sd_journal *j; r = sd_journal_open(&j, SD_JOURNAL_LOCAL_ONLY); if (r < 0) { fprintf(stderr, "Failed to open journal: %s\n", strerror(-r)); return 1; } for (;;) { const void *d; size_t l; r = sd_journal_next(j); if (r < 0) { fprintf(stderr, "Failed to iterate to next entry: %s\n", strerror(-r)); break; } if (r == 0) { /* Reached the end, let's wait for changes, and try again */ r = sd_journal_wait(j, (uint64_t) -1); if (r < 0) { fprintf(stderr, "Failed to wait for changes: %s\n", strerror(-r)); break; } continue; } r = sd_journal_get_data(j, "MESSAGE", &d, &l); if (r < 0) { fprintf(stderr, "Failed to read message field: %s\n", strerror(-r)); continue; } printf("%.*s\n", (int) l, (const char*) d); } sd_journal_close(j); return 0; }
Waiting with poll()
(this
example lacks all error checking for the sake of
simplicity):
#include <poll.h> #include <time.h> #include <systemd/sd-journal.h> int wait_for_changes(sd_journal *j) { uint64_t t; int msec; struct pollfd pollfd; sd_journal_get_timeout(j, &t); if (t == (uint64_t) -1) msec = -1; else { struct timespec ts; uint64_t n; clock_gettime(CLOCK_MONOTONIC, &ts); n = (uint64_t) ts.tv_sec * 1000000 + ts.tv_nsec / 1000; msec = t > n ? (int) ((t - n + 999) / 1000) : 0; } pollfd.fd = sd_journal_get_fd(j); pollfd.events = sd_journal_get_events(j); poll(&pollfd, 1, msec); return sd_journal_process(j); }