Not everybody has those dirs in the filesystem (and they don't need to).
When creating an installation package using $DESTDIR, it is easy enough to
remove or ignore those directories, but if installing into a real root, it
is ugly to create and remove them. Let's add an option so people can skip
it if they want.
Inspired by #12930.
The functions to retrieve and print process cmdlines were based on the
assumption that they contain printable ASCII, and everything else
should be filtered out. That assumption doesn't hold in today's world,
where people are free to use unicode everywhere.
This replaces the custom cmdline reading code with a more generic approach
using utf8_escape_non_printable_full().
For kernel threads, truncation is done on the parenthesized name, so we'll
get "[worker]", "[worker…]", …, "[w…]", "[…", "…" as we reduce the number of
available columns.
This implementation is most likely slower for very long cmdlines, but I don't
think this is very important. The common case is to have short commandlines,
and should print those properly. Absurdly long cmdlines are the exception,
which needs to be handled correctly and safely, but speed is not too important.
Fixes#12532.
v2:
- use size_t for the number of columns. This change propagates into various
other functions that call get_process_cmdline(), increasing the size of the
patch, but the changes are rather trivial.
When journalctl is compiled with PCRE2 support, let's return a non-zero
exit code when --grep is used and no match for given pattern is found.
This should allow users to use journalctl --grep in scripts instead of
piping journalctl into grep
Fixes#8152
The latter is identical to the former, but becomes a NOP if
/var/log/journal is on the same mount as /, and thus during shutdown
unmounting /var is not necessary and hence we can keep logging until the
very end.
When emitting the calendarspec warning we want to see some color.
Follow-up for 04220fda5c.
Exceptions:
- systemctl, because it has a lot hand-crafted coloring
- tmpfiles, sysusers, stdio-bridge, etc, because they are also used in
services and I'm not sure if this wouldn't mess up something.
We had all kinds of indentation: 2 sp, 3 sp, 4 sp, 8 sp, and mixed.
4 sp was the most common, in particular the majority of scripts under test/
used that. Let's standarize on 4 sp, because many commandlines are long and
there's a lot of nesting, and with 8sp indentation less stuff fits. 4 sp
also seems to be the default indentation, so this will make it less likely
that people will mess up if they don't load the editor config. (I think people
often use vi, and vi has no support to load project-wide configuration
automatically. We distribute a .vimrc file, but it is not loaded by default,
and even the instructions in it seem to discourage its use for security
reasons.)
Also remove the few vim config lines that were left. We should either have them
on all files, or none.
Also remove some strange stuff like '#!/bin/env bash', yikes.
The journal files might not be tiny hence let's write them to /var/tmp/
instead of /tmp. Also, let's turn on NOCOW on the files, as these tests
might apparently be slow on btrfs.
Fixes: #12210
systemd-journal-remote always wrote the boot-id of the device it was running on
to the header of its journal files. When the source had a different boot-id
(because it was generated on a different boot, or a different device), the
boot-ids in the file were inconsistent. The _BOOT_ID field was that of the
source, but the journal file header and each entry object header were that of
the device systemd-journal-remote ran on. This breaks journalctl --list-boots
on any of these files.
Set the boot-id in the header to be that of the source. This also fixes the
entry object headers.
Let's be helpful to static analyzers which care about whether we
knowingly ignore return values. We do in these cases, since they are
usually part of error paths.
Let's unify the two similar code paths to watch /run/systemd/journal.
The code in manager.c is similar, but it uses mkdir_p_label(), and unifying
that would be too much trouble, so let's just adjust the error messages to
be the same.
CID #1400224.
The thread launched in journal_file_set_offline() accesses a memory
mapped file, so it needs to handle SIGBUS. Leave SIGBUS unblocked on the
offlining thread so that it uses the same handler as the main thread.
The result of triggering SIGBUS in a thread where it's blocked is
undefined in Linux. The tested implementations were observed to cause
the default handler to run, taking down the whole journald process.
We can leave SIGBUS unblocked in multiple threads since it's handler is
thread-safe. If SIGBUS is sent to the journald process asynchronously
(i.e. with kill, sigqueue, or raise), either thread handling it will
result in the same behavior: it will install the default handler and
reraise the signal, killing the process.
Fixes: #12042
let's use mkdir_parents() (because its shorter), and 0755 as access
mode, so that things have the access mode tmpfiles.d also suggests.
Prompted by: #11903
The option cursor-file takes a filename as argument. If the file exists and
contains a valid cursor, this is used to start the output after this position.
At the end, the last cursor gets written to the file.
This allows for an easy implementation of a timer that regularly looks in the
journal for some messages.
journalctl --cursor-file err-cursor -b -p err
journalctl --cursor-file audit-cursor -t audit --grep DENIED
Or you might want to walk the journal in steps of 10 messages:
journalctl --cursor-file ./curs -n10 --since=today -t systemd
We could potentially create an unterminated string and then call normal string
operations on it. Let's be more careful: first remove the suffix we ignore anyway,
then find if the string is of acceptable length, and possibly ignore it if it
is too long. The code rejects lengths above 31 bytes. Language names that are
actually used are much shorter, so this doesn't matter much.
In normal use, this allow us to drop dead entries from the cache and reduces
the cache size so that we don't evict entries unnecessarily. The time limit is
there mostly to serve as a guard against malicious logging from many different
PIDs.
This is far from perfect, but should give mostly reasonable values. My
assumption is that if somebody has a few hundred MB of memory, they are
unlikely to have thousands of processes logging. A hundred would already be a
lot. So let's scale the cache size propritionally to the total memory size,
with clamping on both ends.
The formula gives 64 cache entries for each GB of RAM.
Found by inspecting results of running this small program:
int main(int argc, const char **argv) {
for (int i = 1; i < argc; i++) {
FILE *f;
char line[1024], prev[1024], *r;
int lineno;
prev[0] = '\0';
lineno = 1;
f = fopen(argv[i], "r");
if (!f)
exit(1);
do {
r = fgets(line, sizeof(line), f);
if (!r)
break;
if (strcmp(line, prev) == 0)
printf("%s:%d: error: dup %s", argv[i], lineno, line);
lineno++;
strcpy(prev, line);
} while (!feof(f));
fclose(f);
}
}
We immediately read the whole contents into memory, making thigs much more
expensive. Sealed fds should be used instead since they are more efficient
on our side.
We'd first parse all or most of the message, and only then consider if it
is not too large. Also, when encountering a single field over the limit,
we'd still process the preceding part of the message. Let's be stricter,
and check size limits early, and let's refuse the whole message if it fails
any of the size limits.
We allocate a iovec entry for each field, so with many short entries,
our memory usage and processing time can be large, even with a relatively
small message size. Let's refuse overly long entries.
CVE-2018-16865
https://bugzilla.redhat.com/show_bug.cgi?id=1653861
What from I can see, the problem is not from an alloca, despite what the CVE
description says, but from the attack multiplication that comes from creating
many very small iovecs: (void* + size_t) for each three bytes of input message.
This fixes a crash where we would read the commandline, whose length is under
control of the sending program, and then crash when trying to create a stack
allocation for it.
CVE-2018-16864
https://bugzilla.redhat.com/show_bug.cgi?id=1653855
The message actually doesn't get written to disk, because
journal_file_append_entry() returns -E2BIG.
Nitpicky, but we've used a lot of random spacings and names in the past,
but we're trying to be completely consistent on "cgroup vN" now.
Generated by `fd -0 | xargs -0 -n1 sed -ri --follow-symlinks 's/cgroups? ?v?([0-9])/cgroup v\1/gI'`.
I manually ignored places where it's not appropriate to replace (eg.
"cgroup2" fstype and in src/shared/linux).
Continuation of a3ebe5eb620e49f0d24082876cafc7579261e64f:
in other places we sometimes use assert_se(), and sometimes normal error
handling. sigfillset and sigaddset can only fail if mask is NULL (which cannot
happen if we are passing in a reference), or if the signal number is invalid
(which really shouldn't happen when we are using a constant like SIGCHLD. If
SIGCHLD is invalid, we have a bigger problem). So let's simplify things and
always use assert_se() in those cases.
In sigset_add_many() we could conceivably pass an invalid signal, so let's keep
normal error handling here. The caller can do assert_se() around the
sigprocmask_many() call if appropriate.
'>= 0' is used for consistency with the rest of the codebase.
With cgroupsv1 a zombie process is migrated to root cgroup in all
hierarchies. This was changed for unified hierarchy and /proc/PID/cgroup
reports cgroup to which process belonged before it exited.
Be more suspicious about cgroup path reported by the kernel and use
unit_id provided by the log client if the kernel reports that process is
running in the root cgroup.
Users tend to care the most about 'log->unit_id' mapping so systemctl
status can correctly report last log lines. Also we wouldn't be able to
infer anything useful from "/" path anyway.
See: 2e91fa7f6d
This splits out a bunch of functions from fileio.c that have to do with
temporary files. Simply to make the header files a bit shorter, and to
group things more nicely.
No code changes, just some rearranging of source files.
Ideally, coccinelle would strip unnecessary braces too. But I do not see any
option in coccinelle for this, so instead, I edited the patch text using
search&replace to remove the braces. Unfortunately this is not fully automatic,
in particular it didn't deal well with if-else-if-else blocks and ifdefs, so
there is an increased likelikehood be some bugs in such spots.
I also removed part of the patch that coccinelle generated for udev, where we
returns -1 for failure. This should be fixed independently.
This way, we can extend the macro a bit with stuff pulled in from other
headers without this affecting everything which pulls in macro.h, which
is one of our most basic headers.
This is just refactoring, no change in behaviour, in prepartion for
later changes.
It's possible for sscanf to receive strings containing all three fields
and not matching the template at the same time. When this happens the
value of k doesn't change, which basically means that process_audit_string
tries to access memory randomly. Sometimes it works and sometimes it doesn't :-)
See also https://bugzilla.redhat.com/show_bug.cgi?id=1059314.
Now that we don't (mis-)use the env file parser to parse kernel command
lines there's no need anymore to override the used newline character
set. Let's hence drop the argument and just "\n\r" always. This nicely
simplifies our code.
All users of the macro (except for one, in serialize.c), use the macro in
connection with read_line(), so they must include fileio.h. Let's not play
libc games and require multiple header file to be included for the most common
use of a function.
The removal of def.h includes is not exact. I mostly went over the commits that
switch over to use read_line() and add def.h at the same time and reverted the
addition of def.h in those files.
Pretty much everything uses just the first argument, and this doesn't make this
common pattern more complicated, but makes it simpler to pass multiple options.
With lz4 1.8.3, this function can now decompress partial results into a smaller
buffer. The release news don't say anything interesting, but the test case that
was previously failing now works OK.
Fixes#10259.
A test is added. It shows that with *older* lz4, a partial decompression can
occur with the returned size smaller then the requested number of bytes _and_
smaller then the size of the compressed data:
(lz4-libs-1.8.2-1.fc29.x86_64)
Compressed 4194304 → 16464
Decompressed → 4194304
Decompressed partial 12/4194304 → 4194304
Decompressed partial 1/1 → -2 (bad)
Decompressed partial 2/2 → -2 (bad)
Decompressed partial 3/3 → -2 (bad)
Decompressed partial 4/4 → -2 (bad)
Decompressed partial 5/5 → -2 (bad)
Decompressed partial 6/6 → 6 (good)
Decompressed partial 7/7 → 6 (good)
Decompressed partial 8/8 → 6 (good)
Decompressed partial 9/9 → 6 (good)
Decompressed partial 10/10 → 6 (good)
Decompressed partial 11/11 → 6 (good)
Decompressed partial 12/12 → 6 (good)
Decompressed partial 13/13 → 6 (good)
Decompressed partial 14/14 → 6 (good)
Decompressed partial 15/15 → 6 (good)
Decompressed partial 16/16 → 6 (good)
Decompressed partial 17/17 → 6 (good)
Decompressed partial 18/18 → -16459 (bad)
(lz4-libs-1.8.3-1.fc29.x86_64)
Compressed 4194304 → 16464
Decompressed → 4194304
Decompressed partial 12/4194304 → 12
Decompressed partial 1/1 → 1 (good)
Decompressed partial 2/2 → 2 (good)
Decompressed partial 3/3 → 3 (good)
Decompressed partial 4/4 → 4 (good)
...
If we got such a short "successful" decompression in decompress_startswith() as
implemented before this patch, we could be confused and return a false negative
result. But it turns out that this only occurs with small output buffer
sizes. We use greedy_realloc() to manager the buffer, so it is always at least
64 bytes. I couldn't hit a case where decompress_startswith() would actually
return a bogus result. But since the lack of proof is not conclusive, the code
for *older* lz4 is changed too, just to be safe. We cannot rule out that on a
different architecture or with some unlucky compressed string we could hit this
corner case.
The fallback code is guarded by a version check. The check uses a function not
the compile-time define, because there was no soversion bump in lz4 or new
symbols, and we could be compiled against a newer lz4 and linked at runtime
with an older one. (This happens routinely e.g. when somebody upgrades a subset
of distro packages.)
I thought this might fail with lz4 < 1.8.3, but it seems that because of
greedy_realloc, we always use a buffer that is large enough, and it always
passes.
lz4-r130 was released on May 29th, 2015. Let's drop the work-around for older
versions. In particular, we won't test any new code against those ancient
releases, so we shouldn't pretend they are supported.
Before this when asked for rotation we'd only rotate files we have open
anyway. However there might be a number of other files on disk that are
active (i.e. not archived yet) but not open. Let's take care of those
too, so that rotation is always comprehensive, and the user gets the
guarantee that afterthe rotation all stored data is in archived files.
Fixes: #1017
journalctl --vacuum-*= only vacuums archived files. To archive all
active files the rotate operation is used. Let's add a new switch that
combines both, so that the user a single command to first move all
running journal files into archival and then vacuum them.
See: #1017
Let's split out the part that actually renames the file in case we can't
open it into a new function journal_file_dispose().
This way we can reuse the function in other cases where we want to open
a file but can't.
Let's split the function in three: the part where we archive the old
file into journal_file_archive(), and the part where we initiate the
deferred closing into journal_file_initiate_close().
journal_file_rotate() then simply becomes a wrapper around these two
calls, and the opening of the new journal file.
This useful so that we can archive journal files without having to open
new ones, i.e. to do only the archival part of the rotation, without the
rotation part.
Let's store array sizes and indexes in size_t. And let's count numbers
of files in uint64_t (simply because that is the type of the input
parameter for this of the function)
Add LogRateLimitIntervalSec= and LogRateLimitBurst= options for
services. If provided, these values get passed to the journald
client context, and those values are used in the rate limiting
function in the journal over the the journald.conf values.
Part of #10230
This makes use of rlimit_nofile_bump() in all tools that access the
journal. In some cases this replaces older code to achieve this, and
others we add it in where it was missing.
All over the place we define local variables for the various sockopts
that take a bool-like "int" value. Sometimes they are const, sometimes
static, sometimes both, sometimes neither.
Let's clean this up, introduce a common const variable "const_int_one"
(as well as one matching "const_int_zero") and use it everywhere, all
acorss the codebase.
Also, while we are at it, beef it up, by adding json-seq support (i.e.
https://tools.ietf.org/html/rfc7464). This is particularly useful in
conjunction with jq's --seq switch.
Let's remove redundancy and not advertise "journalctl --new-id128"
anymore, now that we have "systemd-id128 new" in a proper tool.
This allows us to reduce the overly large journalctl command set a bit.
Note that this just removes the --help and man text, the call remains
available for compat reasons.
The raison d'etre for this program is printing machine-app-specific IDs. We
provide a library function for that, but not a convenient API. We can hardly
ask people to quickly hack their own C programs or call libsystemd through CFFI
in python or another scripting language if they just want to print an ID.
Verb 'new' was already available as 'journalctl --new-id128', but this makes
it more discoverable.
v2:
- rename binary to systemd-id128
- make --app-specific= into a switch that applies to boot-id and machine-id
This makes it so that tests no longer need to know the absolute paths to the
source and build dirs, instead using the systemd-runtest.env file to get these
paths when running from the build tree.
Confirmed that test-catalog works on `ninja test`, when called standalone and
also when the environment file is not present, in which case it will use the
installed location under /usr/lib/systemd/catalog.
The location can now also be overridden for this test by setting the
$SYSTEMD_CATALOG_DIR environment variable.
This is a bit like the info link in most of GNU's --help texts, but we
don't do info but man pages, and we make them properly clickable on
terminal supporting that, because awesome.
I think it's generally advisable to link up our (brief) --help texts and
our (more comprehensive) man pages a bit, so this should be an easy and
straight-forward way to do it.
The function replaces a couple commas, a semicolon and the final newline with
zero bytes in the string passed to it. The 'const' seems to have been added
by accident during a bulk edit (more specifically 3b3154df7e).
Even if built without gcrypt, show the relevant options in help message.
Otherwise, the help message diverges from the man page or suggestions
by the shell completion.
Looked for definitions of functions using the *_compare_func() suffix.
Tested:
- Unit tests passed (ninja -C build/ test)
- Installed this build and booted with it.
This is useful if someone wants to recreate the original syslog datagram. We
already include timestamp information as _SOURCE_REALTIME_TIMESTAMP=, and in
normal use that timestamp, converted back to the form used by syslog
(Mth dd HH:MM:SS) would usually give the value. But there are various
circumstances where this might not be true. Most obviously, if the datagram is
sent a bit later after being prepared, the time is rounded to the nearest
second, and it might be off. This is especially bad around New Year when the
syslog timestamp wraps around. Then the same timezone and locale need to be
used to recreate the original timestamp. In the end doing this reliably is
complicated, and it seems much easier to just unconditionally include the
original timestamp.
If the original timestamp cannot be located, we store the full log line.
This way, it should be always possible to recreate the original input.
Example:
MESSAGE=x
SYSLOG_TIMESTAMP=Sep 15 15:07:58
SYSLOG_RAW
^]^@^@^@^@^@^@^@<13>Sep 15 15:07:58 HOST: x^@y
_PID=3318
_SOURCE_REALTIME_TIMESTAMP=1530743976393553
Fixes#2398.
These lines are generally out-of-date, incomplete and unnecessary. With
SPDX and git repository much more accurate and fine grained information
about licensing and authorship is available, hence let's drop the
per-file copyright notice. Of course, removing copyright lines of others
is problematic, hence this commit only removes my own lines and leaves
all others untouched. It might be nicer if sooner or later those could
go away too, making git the only and accurate source of authorship
information.
This part of the copyright blurb stems from the GPL use recommendations:
https://www.gnu.org/licenses/gpl-howto.en.html
The concept appears to originate in times where version control was per
file, instead of per tree, and was a way to glue the files together.
Ultimately, we nowadays don't live in that world anymore, and this
information is entirely useless anyway, as people are very welcome to
copy these files into any projects they like, and they shouldn't have to
change bits that are part of our copyright header for that.
hence, let's just get rid of this old cruft, and shorten our codebase a
bit.
They are not needed, because anything that is non-zero is converted
to true.
C11:
> 6.3.1.2: When any scalar value is converted to _Bool, the result is 0 if the
> value compares equal to 0; otherwise, the result is 1.
https://stackoverflow.com/questions/31551888/casting-int-to-bool-in-c-c
The refactor in #9200 inadvertently dropped the variable assignment to
traverse the device and its hierarchy in add_matches_for_device().
This was uncovered by Coverity (CID #1393310).
Fix that by restoring the assignment.
Tested: `journalctl /dev/sda` now filters journalctl output again.
This is mostly fall-out from d1a1f0aaf0,
however some cases are older bugs.
There might be more issues lurking, this was a simple grep for "%m"
across the tree, with all lines removed that mention "errno" at all.
The journal verification functions would reject such an entry. It would probably
still display fine (because we prefer _SOURCE_REALTIME_TIMESTAMP= if present), but
it seems wrong to create an entry that would not pass verification.
The code to open journal files seems like the wrong place to enforce this. We
already check during boot and refuse to boot if machine-id is missing, no need
to enforce this here. In particular, it seems better to write logs from
journald even if they are not completely functional rather than refuse to
operate at all, and systemd-journal-remote also writes journal files and may
even be run on a system without systemd at all.
The docker image that oss-fuzz uses has an empty /etc/machine-id. Obviously
this is an error in the docker, but docker is fact of life, and it seems better
for systemd-journal-remote to work in such an incomplete environment.
In journal_file_set_online() the offline thread doesn't need to be
joined if it's been canceled before actually reaching the phase of
writing the offline state.
When dealing with a large number of template instances, for example
when launching daemons per VRF, it is hard for operators to correlate
log lines to arguments.
Add a new with-unit mode which, if available, prefixes unit and user
unit names when displaying its log messages instead of the syslog
identifier. It will also use the full timestamp with timezones, like
the short-full mode.
Most our other parsing functions do this, let's do this here too,
internally we accept that anyway. Also, the closely related
load_env_file() and load_env_file_pairs() also do this, so let's be
systematic.
This makes most header files easier to look at. Also Emacs gets really
slow when browsing through large sections of overly long prototypes,
which is much improved by this macro.
We should probably not do something similar with too many other cases,
as macros like this might help readability for some, but make it worse
for others. But I think given the complexity of this specific prototype
and how often we use it, it's worth doing.
This simplifies the use of tempfiles in tests and fixes "leaked"
temporary files in test-fileio, test-catalog, test-conf-parser.
Not the whole tree is converted.
Configuration through environment variable is inconvenient with meson, because
they cannot be convieniently changed and/or are not preserved during
reconfiguration (https://github.com/mesonbuild/meson/issues/1503).
This adds -Dvalgrind=true/false, which has the advantage that it can be set
at any time with meson configure -Dvalgrind=... and ninja will rebuild targets
as necessary. Additional minor advantages are better consistency with the
options for hashmap debugging, and typo avoidance with '#if' instead of '#ifdef'.
Previously we were a bit sloppy with the index and size types of arrays,
we'd regularly use unsigned. While I don't think this ever resulted in
real issues I think we should be more careful there and follow a
stricter regime: unless there's a strong reason not to use size_t for
array sizes and indexes, size_t it should be. Any allocations we do
ultimately will use size_t anyway, and converting forth and back between
unsigned and size_t will always be a source of problems.
Note that on 32bit machines "unsigned" and "size_t" are equivalent, and
on 64bit machines our arrays shouldn't grow that large anyway, and if
they do we have a problem, however that kind of overly large allocation
we have protections for usually, but for overflows we do not have that
so much, hence let's add it.
So yeah, it's a story of the current code being already "good enough",
but I think some extra type hygiene is better.
This patch tries to be comprehensive, but it probably isn't and I missed
a few cases. But I guess we can cover that later as we notice it. Among
smaller fixes, this changes:
1. strv_length()' return type becomes size_t
2. the unit file changes array size becomes size_t
3. DNS answer and query array sizes become size_t
Fixes: https://bugs.freedesktop.org/show_bug.cgi?id=76745
This drops a good number of type-specific _cleanup_ macros, and patches
all users to just use the generic ones.
In most recent code we abstained from defining type-specific macros, and
this basically removes all those added already, with the exception of
the really low-level ones.
Having explicit macros for this is not too useful, as the expression
without the extra macro is generally just 2ch wider. We should generally
emphesize generic code, unless there are really good reasons for
specific code, hence let's follow this in this case too.
Note that _cleanup_free_ and similar really low-level, libc'ish, Linux
API'ish macros continue to be defined, only the really high-level OO
ones are dropped. From now on this should really be the rule: for really
low-level stuff, such as memory allocation, fd handling and so one, go
ahead and define explicit per-type macros, but for high-level, specific
program code, just use the generic _cleanup_() macro directly, in order
to keep things simple and as readable as possible for the uninitiated.
Note that before this patch some of the APIs (notable libudev ones) were
already used with the high-level macros at some places and with the
generic _cleanup_ macro at others. With this patch we hence unify on the
latter.
Double newlines (i.e. one empty lines) are great to structure code. But
let's avoid triple newlines (i.e. two empty lines), quadruple newlines,
quintuple newlines, …, that's just spurious whitespace.
It's an easy way to drop 121 lines of code, and keeps the coding style
of our sources a bit tigther.
Files which are installed as-is (any .service and other unit files, .conf
files, .policy files, etc), are left as is. My assumption is that SPDX
identifiers are not yet that well known, so it's better to retain the
extended header to avoid any doubt.
I also kept any copyright lines. We can probably remove them, but it'd nice to
obtain explicit acks from all involved authors before doing that.
It might happen that we try to bisect through a chain of offset arrays in the
journal whose last element was just allocated but no item yet written
to. In that case that array will be all NUL, but it might still end up
in our array chain cache. If it does, we cannot use it for bisection,
since for bisection we need to know the value of the first entry in that
array, but if it's uninitialized it does not have a first value. Hence,
as a simple fix, in this unlikely case, simply ignore the chain cache.
This is supposed to fix the issue pointed out in #8432, but in a more
permissive way, as this case isn't strictly a badly formatted journal
but actually a valid state (though one within a very short time window),
and we should make the best of it, and handle it gracefully.
Background: in each journal file entries are linked up in large arrays
of offsets. In each array the entries are strictly ordered by the
offsets of the entries, which permits search by bisection. These arrays
are allocated with a fixed size and then filled up as entries are added
to the journal file. If an array is fully filled up, a new array
(double in size as the old one) is appended to the journal file, and
linked up. This means, the journal file will contain a series of chained
up arrays, each time doubling in size, and strictly ordered. When
looking for an entry we maintain a "chain cache", which allows us to
bypass traversing the chain in full if we look for entries close to each
other in a short time. With the fix above we make sure we don't
erroneously use a chain cache item that doesn't carry enough information
for this bisection to work.
Original issue identified (with patch) by @Kxuan.
Replaces: #8432
