Git log: commit 59b4acd004cc51b3d8664a8ceb0082efd4f74d09
Author: Simon McVittie <smcv@debian.org>
Date: Thu Jul 31 14:34:13 2025 +0100
Update changelog
commit 0ce8863e59654f288cd132c31b5eaca49adbf665
Merge: 035bbe0 0d2cc42
Author: Simon McVittie <smcv@debian.org>
Date: Thu Jul 31 14:32:07 2025 +0100
Merge branch 'debian/trixie' into debian/latest
commit 0d2cc4230f6d90c3e860fbc48f98eb1d5360d8db
Merge: b95c755 074dea8
Author: Simon McVittie <smcv@debian.org>
Date: Thu Jul 31 14:31:42 2025 +0100
Merge remote-tracking branch 'origin/debian/trixie' into debian/trixie
commit 035bbe0e9166b674fb39307c0bc74850d4bb3992
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Sun Feb 23 14:03:53 2025 -0500
Build-Depends on gir1.2-tracker-3.0 instead of libtracker-sparql-3.0-dev
This actually needed the fix from tinysparql 3.8.2-5 first
commit 074dea82f785407d6d662c0039b5c7079e0c1928
Author: Laurent Bigonville <bigon@debian.org>
Date: Sun Jan 28 12:04:38 2024 +0100
Import Debian changes 3.4.6-3
tracker-miners (3.4.6-3) unstable; urgency=medium
.
* Allow getsockopt in seccomp filter (Closes: #1057617)
commit f10a7d1d9fd39f5cf8d03d977f0f1ff0e5b7cc84
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Fri Dec 1 08:13:39 2023 -0500
releasing package tracker-miners version 3.4.6-2
commit e3ef2a5093e00766f5befc7c3a8bc19eb512e4da
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Fri Dec 1 08:13:00 2023 -0500
Stop using debian/control.in and dh_gnome_clean
commit 461ec039fd7498e1e9364b7cd90d0885fae90095
Author: Matthew Ruffell <matthew.ruffell@canonical.com>
Date: Fri Dec 1 08:12:04 2023 -0500
Purge stale systemd unit dependency symbolic links
commit 2eee0a86241484dbe1895f33d62d6cb577ab428e
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Tue Oct 31 13:10:19 2023 -0400
releasing package tracker-miners version 3.4.6-1
commit be0de128aa405099eb7752d2d5a9a0608ed3c4e0
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Tue Oct 31 13:08:09 2023 -0400
Drop patches applied in new release
commit 5d265e034055cc0bdcfef8bee977dd220570c786
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Tue Oct 31 13:06:31 2023 -0400
New upstream release
commit dc3256ac99e5031f422e3872a3d946a17ca2f746
Merge: a22a8d8 5c3d5fa
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Tue Oct 31 13:06:31 2023 -0400
Update upstream source from tag 'upstream/3.4.6'
Update to upstream version '3.4.6'
with Debian dir a05899f74fb7f28c83367395eaf12f96d3a3d0c3
commit 5c3d5fa341d6d3f7d8f0bdb42db86b66fdd471a1
Merge: 2176094 9bba9cf
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Tue Oct 31 13:06:21 2023 -0400
New upstream version 3.4.6
commit 9bba9cf0d669ad4f08c6fb0caab4ff0a8a4d680e
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Oct 31 13:26:31 2023 +0100
Release 3.4.6
commit 59c229f36b46cb3908c3946d2a8c84f013c5ec31
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Thu Apr 13 10:37:36 2023 +0200
tracker-extract: Add missing break to remove a single element
The extractor process listens to notifier events in order to a)
know there's missing elements, in order to build its small cache
(nowadays: 200) of elements to extract, or b) notice files being
deleted, which might be part of the currently cached items.
The loop doing the latter was somewhat broken, trying to remove
elements from a GQueue while iterating it. Since cached elements
must be just once in the list at most, this is just missing a
break.
Closes: https://gitlab.gnome.org/GNOME/tracker-miners/-/issues/265
commit 1f9b49553899e389063db4474b57a94ff45bec8b
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Wed Oct 25 11:21:17 2023 +0200
tracker-extract: Disable GST registry updates in extractor process
Force disable the attempt to update the GStreamer registry from the extractor
process, creation of files is softly forbidden by the seccomp sandbox anyway.
This will have the nice side effect of avoiding GStreamer from attempting to
load changed/new modules at gst_init() time (and the crazy shit some of them
do during plugin initialization), while the less nice side effect (the extractor
not finding old/stale plugins, and being unable to use new ones) should have been
largely avoided by the previous commit.
commit 22a30091f59f66ded7ed0341a7f555c3cc7787f0
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Mon Oct 23 19:17:41 2023 +0200
tracker-miner-fs: Preempt GStreamer registry file creation
Initialize gstreamer in tracker-miner-fs, mostly as a means to
ensure the registry file is guaranteed to exist and be up-to-date
when the tracker-extract-3 process gets to start up and require the
use of GStreamer plugins.
commit 796d185b65cfb08b760da0bce07d4536361d66f0
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Oct 17 16:10:11 2023 +0200
libtracker-miners-common: Forbid some more open()/openat() flags
There are some nasty combinations with O_RDONLY that may already
trigger unintended results, like O_CREAT and O_TRUNC. Avoid any
combination that sounds off with O_RDONLY.
commit df97325e6df9daa5146e908ffdc3f7fe787ba20d
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Oct 17 14:28:29 2023 +0200
libtracker-miners-common: Disallow close/dup2/dup3 on standard I/O FDs
As an additional measure, forbid these FDs to be replaced by anything.
commit 23b4dc3d57b8c929f781dfbe1daa79db67f13c73
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Oct 17 14:12:33 2023 +0200
libtracker-miners-common: Drop ioctl rules for stdout/stderr
These were added in the first instance of the sandbox, I do not
remember what they were for, and they seem largely unnecessary
nowadays.
commit 13354f7df8fccf802f3bc03f7658873822aedbef
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Oct 17 13:11:07 2023 +0200
libtracker-miners-common: Allow restart_syscall syscall
This syscall may happen after SIGCONT if a previous SIGSTOP caught
the process mid-syscall. This is a plausible situation with gdb, and
through coredumpd/abrtd/etc.
Closes: https://gitlab.gnome.org/GNOME/tracker-miners/-/issues/288
commit 7131203b6d9b6394797e11c5ac03301efea5fbef
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Oct 17 12:48:42 2023 +0200
tracker-extract: Initialize modules also before commandline extraction
This allows the gstreamer module to block plugins through the registry
in those paths too.
commit 28248dfabcdb5f452c1c07b797d7aa55562c8a28
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Oct 17 09:31:33 2023 +0200
libtracker-miners-common: Allow tgkill on self's process
This is the syscall underneath abort(), assert(), etc. Let this
syscall through for the purpose of killing self's process, and
don't hide the actual errors.
Closes: https://gitlab.gnome.org/GNOME/tracker-miners/-/issues/287
commit 02cf66a56a3db1b896e1b4f2f1824dc7d13d8e0f
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Oct 17 12:52:32 2023 +0200
libtracker-miners-common: Allow NETLINK_KOBJECT_UEVENT access
Sadly, it remains extremely finicky to provide detailed rules with
differing actions for socket() through libseccomp(), due to the way this
syscall is (was?) wrapped through the multiplexed socketcall() syscall on
some architectures.
Since we cannot provide different SCMP_ACT_* values, and we cannot let
AF_NETLINK/NETLINK_KOBJECT_UEVENT requests fail with SIGSYS (e.g.
video4linux2 gstreamer plugin wants udev access and will trigger this
right in gst_init()), go with SCMP_ACT_ALLOW for this specific combination,
along with AF_LOCAL.
This kind of socket is effectively readonly to unprivileged users and
local by definition, so it does not seem a big risk to allow, the only
lingering question being "why should we allow this".
Closes: https://gitlab.gnome.org/GNOME/tracker-miners/-/issues/283
commit 6070c144353ca30a68f12e3691921fe9ab525294
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Mon Oct 16 11:15:21 2023 +0200
libtracker-miners-common: Allow some more syscalls for i686
These are seen on Debian i686 with the default dependencies pulled
from apt-get build-dep tracker-miners. There's mainly 32-bit syscall
variants (getgid32, fstatfs64), some fancy stuff so far unseen in 64-bit
(timerfd_create), and for some unfathomable reason, gstreamer openNI2
module using chdir() to change the CWD.
Everything is harmless though, so go with that.
Closes: https://gitlab.gnome.org/GNOME/tracker-miners/-/issues/284
commit 4a7e9142b93b9344f670d504d39bb7cd8714b530
Author: psykose <alice@ayaya.dev>
Date: Tue Oct 17 13:29:56 2023 +0000
libtracker-miners-common: use macro stringify instead of G_STRINGIFY
G_STRINGIFY performs macro expansion; this means that on musl, the LFS64
interface define of
#define getdents64 getdents
gets expanded to 'getdents', so the getdents64 syscall becomes not
allowed. using the preprocessor #stringify does not expand the name, so
this works as expected.
closes #285
commit 338a15e822008e21d2fbb08fb72a53d9766a4b0b
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Sun Oct 8 00:02:30 2023 +0200
libtracker-miners-common: Add more seccomp rules
Seen on debian/ubuntu on some arches. Make mkdirat error out
the same way than mkdir, and allow name_to_handle_at() as
that should be innocuous.
Closes: https://gitlab.gnome.org/GNOME/tracker-miners/-/issues/281
commit 9aa5d2693ced4b66a7abdb875203ec3e3d34a96d
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Oct 3 12:47:57 2023 +0200
libtracker-miners-common: Improve "bail out" error loading seccomp rules
We are out in the dark if we happen to add syscalls that do not exist on
obscure architectures. Keep track of the rules being added, so we can
provide a more useful error if we fall in this situation.
commit f3f7b54446a6238acbf38d8f02e5aeba4814afd0
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Oct 3 12:44:42 2023 +0200
libtracker-miners-common: Add custom rules through a define
Bring some more consistence between our ALLOW/ERROR_RULE defines,
and the "custom" rules where we check syscall arguments.
commit a22a8d8fddfb8c0bccb1c41fb9acb822ae37cf61
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Wed Oct 18 11:26:10 2023 -0400
releasing package tracker-miners version 3.4.5-3
commit ad95ff72958b81cdea3c4bacd9c61500b117e0de
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Wed Oct 18 10:20:27 2023 -0400
Cherry-pick proposed patches to fix build on various architectures
Closes: #1053238
commit 30f45b0b327062f115ccfe9731561c922701e075
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Tue Oct 10 07:51:10 2023 -0400
releasing package tracker-miners version 3.4.5-2
commit 5543d9eeb3aa122e0a57e8edd057bfc2be8c7168
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Thu Oct 5 14:12:56 2023 -0400
Cherry-pick build fixes
commit 456a90d5a693df5be95af07a5aff2ea464b2e320
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Thu Sep 28 17:26:26 2023 -0400
New upstream release
commit b40b6ff5931e1251b5fa4d4a07cc2bb5da5faba1
Merge: 9321e19 2176094
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Thu Sep 28 17:26:26 2023 -0400
Update upstream source from tag 'upstream/3.4.5'
Update to upstream version '3.4.5'
with Debian dir e961e7cecdb7a9132cdbb8349f96a08ca8183b03
commit 2176094069a53cfdea8eb95acffc2574ccec3301
Merge: 1759f84 3a46114
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Thu Sep 28 17:26:19 2023 -0400
New upstream version 3.4.5
commit 9321e19e5daeb06118e73f58403282b93e46a774
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Thu Sep 28 17:25:54 2023 -0400
debian/gbp.conf, debian/watch: Watch for 3.4 versions only
commit 3a46114a034485b3911c014d945766c6d5863f51
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Thu Sep 28 22:32:07 2023 +0200
Release 3.4.5
commit 3efd984fcd7ced2a0253cb608486e05c8ad611f4
Merge: 2dddde2 c7f9938
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Sep 26 19:48:48 2023 +0000
Merge branch 'wip/extend-seccomp-3.4' into 'tracker-miners-3.4'
Backport improved seccomp jail (3.4)
See merge request https://gitlab.gnome.org/GNOME/tracker-miners/-/merge_requests/482
commit c7f9938399cd6af005fb6a997aeae3061d5cbdce
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Fri Sep 22 23:14:38 2023 +0200
tracker-extract: Extend seccomp jail to full process
Currently, our main thread is exempted from the seccomp jail.
This was so we could do some menial tasks (e.g. persistence handling
to recover from runtime errors, or error reports on failed extraction)
without caring much about plugging seccomp holes.
It may be preferable to extend the seccomp jail to the full process
instead, so do that. Now the only thing happening prior to setting
up the seccomp jail is the setting up of nice/scheduler/ioprio
priorities. Everything else, and every thread spawned afterwards is
covered by seccomp.
commit 8650aad79d87f4d7df023e2ecbb6668b1d8fba5a
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Fri Sep 22 23:26:38 2023 +0200
libtracker-miners-common: Extend seccomp rules
The plan is to extend the seccomp jail so it affects the full
tracker-extract-3 process. With the changes in the previous
commits we've removed the need for filesystem write access.
We have some remaining outliers, that we're largely sorting
out with rules to error out softly (instead of through SIGSYS).
The only new allowed syscalls are fstatfs and prlimit64 with a
NULL new_limit struct.
commit a8395d834017c676e2c3eb03e1e310e57f2d4f4e
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Sun Sep 24 23:27:56 2023 +0200
tracker-extract: Disable GstRegistry forking
This is going nowhere with the sandbox. Also disable some more
needless GST plugins.
commit 3f5fabbd712bc11009289d6d1b4f89b24f4bee46
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Sun Sep 24 19:18:52 2023 +0200
tracker-extract: Avoid file access for persistence
Use a memfd_create() FD, maintained and kept alive by tracker-miner-fs-3.
This FD is obtained through D-Bus, and used for temporary storage. Since
processing of files in the extractor is largely linear nowadays, this
also simplifies the persistent storage to store a single file.
commit 873c6581243324eadeb294db6c4093442837c19a
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Sun Sep 24 14:20:01 2023 +0200
tracker-extract: Handle configuration through D-Bus
Add an interface on tracker-miner-fs-3 so that tracker-extract-3
can get the relevant settings without using DConf/GSettings directly.
This replaces all settings usage from tracker-extract-3.
commit a5d8dceda36b7616549682c64f968f493e0fd94b
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Sun Sep 24 16:54:39 2023 +0200
tracker-extract: Move text allow list handling to tracker-miner-fs-3
This is something that can be streamlined without delayed handling from
tracker-extract-3. But also avoids having to poke for this setting in
that process.
commit 992ed96c77514294cd2b505dbc6bb3d33b7b98d0
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Sun Sep 24 15:53:22 2023 +0200
tracker-extract: Drop handling of wait-for-miner-fs
Even though this setting is off by default, that is the stock
behavior of tracker-miner-fs, by activating the tracker-extract-3
D-Bus name after going idle.
Furthermore, enabling this setting will have clunky interaction
with the current behavior since 3.1.0 that tracker-miner-fs-3
forwards the tracker-extract-3 status (commit bd3ce694d7), since
tracker-extract-3 activity will make the tracker-miner-fs-3 status
"non-idle", which will pause the extractor, which will make the miner
idle, which will unpause the extractor, ...
It's arguable that we should keep supporting this as a setting
altogether, so just drop the tracker-extract-3 side code handling
this setting.
commit b78a19cf88f92a12b29a32d32e497fa64c6045a5
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Sun Sep 24 11:50:46 2023 +0200
tracker-extract: Move error report handling to tracker-miner-fs
These error reports contain useful information in case metadata
extraction failed for some files. However, it would be nice to
have the tracker-extract-3 process uninvolved with file creation.
To make it so for error reports, use a D-Bus signal on a new
interface at the tracker-extract-3 process, that is picked up by
tracker-miner-fs-3. The deletion of stale error reports is already
performed by tracker-miner-fs-3, and the `tracker3 status` CLI as
a last resort.
commit 28b311e0144de6db998441dbff73556190f4f2c8
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Sat Sep 23 10:42:37 2023 +0200
tracker-extract: Drop SIGINT/SIGTERM handlers
These bring some questions if using seccomp for the full
process (e.g. requiring additional syscalls, or glib spawning
a thread for it), and are not really mandatory since there's
no requirements for a clean exit.
The only thing that is somewhat lost is ease at valgrinding
with noise from things "definitely lost" in the abrupt termination,
but that does already require manually disabling the seccomp
jail, it's not a big stretch to pile up more local hacks, or
ignore the noise.
commit 2dddde2bd3e1725428561a2fd9986e98d29913f0
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Tue Aug 29 16:30:17 2023 +0200
Release 3.4.4
commit 5835b1ad774b6f0c097730cb0e6ba6633014b57c
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Mon Aug 21 08:33:04 2023 -0400
releasing package tracker-miners version 3.4.3-2
commit 4cf9439e755b6ae919f3905bbf3d64aabaf844f2
Author: Daniel van Vugt <daniel.van.vugt@canonical.com>
Date: Thu Apr 13 17:34:18 2023 +0800
Add tracker-extract-Add-missing-break-to-remove-a-single.patch
to fix SIGSEGV in decorator_item_cache_remove
LP: #2012638
Gbp-Dch: Full
commit 66d7f9d1659fbd4b137eb04b2e124899f601f53e
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Mon Aug 21 08:25:19 2023 -0400
Cherry-pick 2 patches to improve tracker systemd services
- Fix tracker-extract.service WantedBy target so that gvfsd has
access to KRB5CCNAME. Thanks Denison Barbosa (LP: #1779890)
- Don't start the daemons for non-root system users like gdm
and lightdm (LP: #2007755)
Gbp-Dch: Full
commit 2baa1386ac1ae34b0e2fff54764e2828d7905064
Author: Jeremy Bícha <jeremy.bicha@canonical.com>
Date: Mon Aug 21 08:16:26 2023 -0400
debian/gbp.conf: Branch for trixie
tracker-miners 3.5+ can't be uploaded to Unstable until
we fix the tracker build test on 32-bit architectures
commit a298c838a5650eb3a02e07ddde55ff7c7af78654
Merge: 83054c8 5fa7041
Author: Sam Thursfield <sam@afuera.me.uk>
Date: Fri Aug 18 10:00:33 2023 +0000
Merge branch 'wip/carlosg/btrfs-subvolumes-3.4' into 'tracker-miners-3.4'
libtracker-miners-common: Get BTRFS subvolume ID through ioctl()
See merge request https://gitlab.gnome.org/GNOME/tracker-miners/-/merge_requests/470
commit 5fa70410da07b2ed227ef0f0a8a869dfd1ff5869
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Mon Apr 17 15:13:03 2023 +0200
libtracker-miners-common: Allow BTRFS_IOC_INO_LOOKUP ioctl in seccomp
We do get resource identifiers from some extractor modules (e.g. for
flac songs). We should make it able to get a subvolume ID without triggering
seccomp.
(cherry-picked from commit e7b19f9a6ec4c6e479304c4f2583a1b31dad9752)
commit 2f02240646f9b9c2ca070ccbf71111aa9ba9e56c
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Mon Apr 17 14:21:40 2023 +0200
libtracker-miners-common: Get BTRFS subvolume ID through ioctl()
Use the BTRFS_IOC_INO_LOOKUP ioctl on a file FD, in order to fetch
its BTRFS subvolume ID. According to BTRFS documentation this is
necessary to get a filesystem-wide unique identifier as we expect:
https://btrfs.readthedocs.io/en/latest/Subvolumes.html#inode-numbers
Commit ff52970f997e only handled subvolumes that are known through
mountpoint info, this in addition works for subvolumes that are
known only to btrfs (e.g. `btrfs subvolume create ...` with no
mount bound to it), with the small drawback that we have to poke
every file for their subvolume through ioctl(), these kernel/fs
queries should be fast though.
Since subvolumes as mounts are also indirectly handled through
the ioctl(), it is not necessary anymore to track the "subvol"
option in mtab.
(cherry-picked from commit 4083d3eb53254a23797cc246a4cb48e750f01737)
commit 8d376c52345a32da6b56393ac54ca10d34fd44cd
Author: Carlos Garnacho <carlosg@gnome.org>
Date: Mon Apr 17 14:08:51 2023 +0200
build: Detect BTRFS_IOC_INO_LOOKUP ioctl at build time
Check whether we are building against a system that has btrfs
and all its relevant headers in place. We will need this ioctl
to get the subvolume ID of each file.
(cherry-picked from commit c4eb04fb540511f51c2f49c0a62ee992442db7d7)
