There are no mmap_cache_get() users that actually deviate prot
from the JournalFile's f->prot.
So there's no point in making this a separate parameter to
mmap_cache_get(), nor is there any need to store it in
JournalFile's f->prot.
Instead just pass it to mmap_cache_add_fd() at MMapFileDescriptor
creation, storing it in there for the mmap() callers, which
already receive MMapFileDescriptor *.
For functions receiving both an MMapFileDescriptor * and prot,
the prot argument has been simply removed and call sites updated.
Formalizing this fd:prot binding at the public API also enables
discarding the prot check in window_matches(), which is a hot
function on long window lists, so a minor CPU efficiency gain
should be had there as seen with the past removal of the fd
check. Unnoticable for uncached journals, but maybe a little
runtime improvement when cached in specific circumstances.
window_matches_fd() has also been simplified to treat the
MMapFileDescrptor * as equivalent to its fd and prot.
This adds a new (incompatible) feature to journal files: if enabled the
hash function used for the hash tables is no longer jenkins hash with a
zero key, but siphash keyed by the file uuid that is included in the
file header anyway. This should make our hash tables more robust against
collision attacks, as long as the attacker has no read access to the
journal files. We switch from jenkins to siphash simply because it's
more well-known and we standardize for the rest of our codebase onto it.
This is hardening in order to make collision attacks harder for clients
that can forge log messages but have no read access to the logs. It has
no effect on clients that have read access.
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.
These lines are generally out-of-date, incomplete and unnecessary.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.
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.
Previously the compression threshold was hardcoded to 512, which meant that
smaller values wouldn't be compressed. This left some storage savings on the
table, so instead, we make that number tunable.
This adds proper handling of IN_Q_OVERFLOW: when the inotify queue runs
over we'll reiterate all directories we are looking at. At the same time
we'll mark all files and directories we encounter that way with a
generation counter we first increased. All files and directories not
marked like this are then unloaded.
With this logic we do the best when the inotify queue overflows: we
synchronize our in-memory state again with what's on disk.
This contains some refactoring of the directory logic, to share more
code between uuid directories and "root" directories and generally make
things a bit more readable by splitting things up into smaller bits.
See: #7998#8032
The advantage is that is the name is mispellt, cpp will warn us.
$ git grep -Ee "conf.set\('(HAVE|ENABLE)_" -l|xargs sed -r -i "s/conf.set\('(HAVE|ENABLE)_/conf.set10('\1_/"
$ git grep -Ee '#ifn?def (HAVE|ENABLE)' -l|xargs sed -r -i 's/#ifdef (HAVE|ENABLE)/#if \1/; s/#ifndef (HAVE|ENABLE)/#if ! \1/;'
$ git grep -Ee 'if.*defined\(HAVE' -l|xargs sed -i -r 's/defined\((HAVE_[A-Z0-9_]*)\)/\1/g'
$ git grep -Ee 'if.*defined\(ENABLE' -l|xargs sed -i -r 's/defined\((ENABLE_[A-Z0-9_]*)\)/\1/g'
+ manual changes to meson.build
squash! build-sys: use #if Y instead of #ifdef Y everywhere
v2:
- fix incorrect setting of HAVE_LIBIDN2
This way we have a MMapFileDescriptor reference external to the cache,
and can supply the handle directly to mmap_cache_get(), eliminating
hashmap lookups entirely from the hot path.
The only code path which makes a journal durable is via
journal_file_set_offline().
When we perform a rotate the journal's header->state is being set to
STATE_ARCHIVED prior to journal_file_set_offline() being called.
In journal_file_set_offline(), we short-circuit the entire offline when
f->header->state != STATE_ONLINE.
This all results in none of the journal_file_set_offline() fsync() calls
being reached when rotate archives a journal, so archived journals are
never explicitly made durable.
What we do now is instead of setting the f->header->state to
STATE_ARCHIVED directly in journal_file_rotate() prior to
journal_file_close(), we set an archive flag in f->archive for the
journal_file_set_offline() machinery to honor by committing
STATE_ARCHIVED instead of STATE_OFFLINE when set.
Prior to this, rotated journals were never getting fsync() explicitly
performed on them, since journal_file_set_offline() short-circuited.
Obviously this is undesirable, and depends entirely on the underlying
filesystem as to how much durability was achieved when simply closing
the file.
Note that this problem existed prior to the recent asynchronous fsync
changes, but those changes do facilitate our performing this durable
offline on rotate without blocking, regardless of the underlying
filesystem sync-on-close semantics.
Also, expose this via the "journalctl --file=-" syntax for STDIN. This feature
remains undocumented though, as it is probably not too useful in real-life as
this still requires fds that support mmaping and seeking, i.e. does not work
for pipes, for which reading from STDIN is most commonly used.
When we rotate journals, we must set offline and close the current one,
but don't generally need to wait for this to complete.
Instead, we'll initiate an asynchronous offline via
journal_file_set_offline(oldfile, false), and add the file to a
per-server set of deferred closes to be closed later when they
won't block.
There's one complication however; journal_file_open() via
journal_file_verify_header() assumes that any writable journal in the
online state is the product of an unclean shutdown or other form of
corruption.
Thus there's a need for journal_file_open() to be aware of deferred
closes and synchronize with their completion when opening preexisting
journals for writing. To facilitate this the deferred closes set is
supplied to the journal_file_open() function where the deferred closes
may be closed synchronously before verifying the header in such
circumstances.
This adds a wait flag to journal_file_set_offline(), when false the offline is
performed asynchronously in a separate thread.
When wait is true, if an asynchronous offline is already in-progress it is
restarted and waited for. Otherwise the offline is performed synchronously
without the use of a thread.
journal_file_set_online() cancels or waits for the asynchronous offline to
complete if in-flight, depending on where in the offline process the thread
happens to be. If the thread is in the fsync() phase, it is cancelled and
waiting is unnecessary. Otherwise, the thread is joined before proceeding.
A new offline_state member is added to JournalFile which is used via
atomic operations for communicating between the offline thread and the
journal_file_set_{offline,online}() functions.
Let's make sure we handle compression errors properly, and don't
misunderstand an error for success.
Also, let's actually compress things if lz4 is enabled.
Fixes#1662.
Implement a maximum limit on number of journal files to keep around.
Enforcing a limit is useful on this since our performance when viewing
pays a heavy penalty for each journal file to interleve. This setting is
turned on now by default, and set to 100.
Also, actully implement what 348ced9097
promised: use whatever we find on disk at startup as lower bound on how
much disk space we can use. That commit introduced some provisions to
implement this, but actually never did.
This also adds "journalctl --vacuum-files=" to vacuum files on disk by
their number explicitly.
When a new journal file is created we write the header first, then sync
and only then create the data and field hash tables in them. That means
to other processes it might appear that the files have a valid header
but not data and field hash tables. Our reader code should be able to
deal with this.
With this change we'll not map the two hash tables right-away after
opening a file for reading anymore (because that will of course fail if
the objects are missing), but delay this until the first time we access
them. On top of that, when we want to look something up in the hash
tables and we notice they aren't initialized yet, we consider them
empty.
This improves handling of some journal files reported in #487.
Commit 668c965af "journal: skipping of exhausted journal files is bad if
direction changed" fixed a correctness issue, but it also significantly
limited the cases where the optimization that skips exhausted journal
files could apply.
As a result, some journalctl queries are much slower in v219 than in v218.
(e.g. queries where a "--since" cutoff should have quickly eliminated
older journal files from consideration, but didn't.)
If already in the initial iteration find_location_with_matches() finds
no entry, the journal file's location is not updated. This is fine,
except that:
- We must update at least f->last_direction. The optimization relies on
it. Let's separate that from journal_file_save_location() and update
it immediately after the direction checks.
- The optimization was conditional on "f->current_offset > 0", but it
would always be 0 in this scenario. This check is unnecessary for the
optimization.
Our write pattern is quite awful for CoW file systems (btrfs...), as we
keep updating file parts in the beginning of the file. This results in
fragmented journal files. Hence: when rotating files, defragment them,
since at that point we know that no further write accesses will be made.
Even though we use fallocate() it appears that file systems like btrfs
will trigger SIGBUS on certain low-disk-space situation. We should
handle that, hence catch the signal, add it to a list of invalidated
pages, and replace the page with an empty memory area. After each write
check if SIGBUS was triggered, and consider the write invalid if it was.
This should make journald a lot more robust with file systems where
fallocate() is not reliable, for example all CoW file systems
(btrfs...), where changing written data can fail with disk full errors.
https://bugzilla.redhat.com/show_bug.cgi?id=1045810
If from a previous iteration we know we are at the end of a journal
file, don't bother looking into the file again. This is complicated by
the fact that the EOF does not have to be permanent (think of
"journalctl -f"). So we also check if the number of entries in the
journal file changed.
This optimization has a similar effect as "journal: optimize iteration:
skip whole files behind current location" had.
In next_beyond_location() when we find a candidate entry in a journal
file, save its location information in struct JournalFile.
The purpose of remembering the locations of candidate entries is to be
able to save work in the next iteration. This patch does only the
remembering part.
LOCATION_SEEK means the location identifies a candidate entry.
When a winner is picked from among candidates, it becomes
LOCATION_DISCRETE.
LOCATION_TAIL here signifies we've iterated the file to the end (or the
beginning in the case of reversed direction).