project/ubox.git
7 months agokmodloader: fix insmod path logic master
Tony Ambardar [Fri, 26 Apr 2024 23:21:25 +0000 (16:21 -0700)]
kmodloader: fix insmod path logic

Correct a small typo that breaks logic for insmod path selection.

Fixes: 4c5c45c6beac ("kmodloader: fix invalid write during insmod, CodeQL warnings")
Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
8 months agokmodloader: support duplicate module names and aliases
Tony Ambardar [Sat, 27 Jan 2024 11:41:56 +0000 (03:41 -0800)]
kmodloader: support duplicate module names and aliases

Allow duplicate keys in the 'modules' avl tree, fixing an indirect memory
leak observed during testing. Given avl keys treat '_' and '-' equally,
some modules may have aliases duplicating their own names under this
mapping, so skip adding these redundant avl tree nodes in alloc_module().

With the above support, update 'modinfo' to print details for all modules
matching the cmdline name, rather than only the first. This is the expected
behaviour and consistent with upstream. The behaviour of other kmodloader
applets, such as boot-time loading or modprobe, remains unchanged.

Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
8 months agokmodloader: fix memleak adding to avl tree
Tony Ambardar [Wed, 24 Jan 2024 08:17:06 +0000 (00:17 -0800)]
kmodloader: fix memleak adding to avl tree

Check avl_insert() return value and free node data on error, otherwise this
is observed to directly leak memory.

Fixes: 6e3c6dcf922e ("kmodloader: add module alias awareness")
Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
8 months agokmodloader: fix invalid write during insmod, CodeQL warnings
Tony Ambardar [Thu, 25 Jan 2024 02:22:56 +0000 (18:22 -0800)]
kmodloader: fix invalid write during insmod, CodeQL warnings

Running 'insmod <module>' (without module options) tries to write an empty
option string (1 byte) to a mis-allocated zero-size memory block:

  ==381835== Command: ./insmod unix
  ==381835==
  ==381835== Invalid write of size 1
  ==381835==    at 0x10A874: main_insmod (kmodloader.c:944)
  ==381835==    by 0x10A874: main (kmodloader.c:1383)
  ==381835==  Address 0x4ab6f60 is 0 bytes after a block of size 0 alloc'd
  ==381835==    at 0x4848899: malloc (in vgpreload_memcheck-amd64-linux.so)
  ==381835==    by 0x10A856: main_insmod (kmodloader.c:937)
  ==381835==    by 0x10A856: main (kmodloader.c:1383)

Change main_insmod() to allocate at least 1 byte in this case for the
null-termination. Also rename local 'options' -> 'opts' to avoid confusion
with similar global, as suggested by CodeQL, and clarify pathname logic.

Fixes: c105f22064d6 ("kmodloader: eliminate some hardcoded buffer sizes")
Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
8 months agokmodloader: improve memory accounting and deallocation
Tony Ambardar [Thu, 25 Jan 2024 02:20:37 +0000 (18:20 -0800)]
kmodloader: improve memory accounting and deallocation

Add missing calls to free module_folders and modules data during cleanup,
and ensure that resources are consistently freed on exit paths.

Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
8 months agoCI: add CodeQL workflow test
Christian Marangi [Sat, 2 Mar 2024 16:23:47 +0000 (17:23 +0100)]
CI: add CodeQL workflow test

Add CodeQL workflow test to catch security vulnerabilities and enable
suggestion for better code.

All the dependency comes from the build-ubox workflow but all is built
on ubuntu instead of alpine image.

CodeQL queries enabled: security-and-quality

Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
8 months agoCI: add build test for ubox
Christian Marangi [Sat, 2 Mar 2024 16:11:42 +0000 (17:11 +0100)]
CI: add build test for ubox

Add build test for ubox. Use alpine container and install libubox,
libubus, libudebug, ucode and libuci as dependency.

Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
8 months ago.gitignore: don't ignore .github directory
Christian Marangi [Sat, 2 Mar 2024 16:10:55 +0000 (17:10 +0100)]
.gitignore: don't ignore .github directory

Don't ignore .github special directory as it does contain specific files
for github CI workflows.

Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
10 months agokmodloader: fix invalid read outside mapped region
Tony Ambardar [Mon, 22 Jan 2024 08:48:37 +0000 (00:48 -0800)]
kmodloader: fix invalid read outside mapped region

Code parsing .modinfo data skips over null sequences without checking
bounds and may read past mapped memory, potentially triggering SIGSEGV.

Fixes: https://github.com/openwrt/openwrt/issues/14463
Fixes: d6e6825c4697 ("add support for module handling")
Refer: 9371411715c8 ("kmodloader: fix out-of-bound access when parsing .modinfo")
Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
10 months agokmodloader: fix TOCTOU problem with scan_builtin_modules
Christian Marangi [Mon, 22 Jan 2024 00:13:11 +0000 (01:13 +0100)]
kmodloader: fix TOCTOU problem with scan_builtin_modules

Fix TOCTOU problem with scan_builtin_modules by opening the file pointer
only once in module_folders scan.

Fix Coverity Report CID 1586645:  Security best practices violations
(TOCTOU).

Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
10 months agokmodloader: fix memory leak in print_modinfo
Christian Marangi [Sun, 21 Jan 2024 23:49:44 +0000 (00:49 +0100)]
kmodloader: fix memory leak in print_modinfo

Fix memory leak in print_modinfo reported by Coverity Report.
It seems there is a logic error and duplicated string in print_modinfo
is never freed. On top of this if the while loop terminates early the
just allocated duplicated string is also never freed.

Rework the function to correctly free the duplicated string.

Fix CID 1586644: Resource leaks (RESOURCE_LEAK).

Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
10 months agokmodloader: Fix NULL pointer dereferences error
Christian Marangi [Sun, 21 Jan 2024 23:39:50 +0000 (00:39 +0100)]
kmodloader: Fix NULL pointer dereferences error

Fix NULL pointer dereferences reported by Coverity Scan.
Return early instead of jumping to out and trying to fclose a NULL
pointer.

Fix
- CID 1586642: Null pointer dereferences (FORWARD_NULL)
- CID 1586641: Null pointer dereferences (FORWARD_NULL)

Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
10 months agokmodloader: cleanup duplicated mmap() code
Tony Ambardar [Tue, 19 Dec 2023 07:15:12 +0000 (23:15 -0800)]
kmodloader: cleanup duplicated mmap() code

Refactor code from get_module_info() and print_modinfo() to locate modinfo
data, call mmap() and handle errors, into new mmap_modinfo(). This drops
~30 source lines and shrinks the compressed binary ~100 bytes.

Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
10 months agokmodloader: add modinfo support for builtin modules
Tony Ambardar [Thu, 14 Dec 2023 07:18:53 +0000 (23:18 -0800)]
kmodloader: add modinfo support for builtin modules

Enable parsing of 'modules.builtin.modinfo' for better handling of builtin
modules, including support for aliases and showing basic builtin module
details, e.g.

  root@OpenWrt:/# modinfo unix
  name:           unix
  filename:       (builtin)
  alias:          net-pf-1
  license:        GPL

  root@OpenWrt:/# modprobe net-pf-1 && echo SUCCESS || echo FAIL
  SUCCESS

Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
10 months agokmodloader: add basic support for builtin modules
Tony Ambardar [Thu, 14 Dec 2023 07:18:30 +0000 (23:18 -0800)]
kmodloader: add basic support for builtin modules

Enable parsing of 'modules.builtin' if present, for improved handling of
loadable and builtin modules similar to most other distros. In particular,
this allows 'modprobe' to return success whether a module is loadable or
built-in, a useful feature for consistent presence-testing. For example:

  root@OpenWrt:/# rmmod sch_fq_codel
  module is builtin

  root@OpenWrt:/# modinfo sch_cake
  filename:       /lib/modules/6.1.65/sch_cake.ko
  license:        Dual BSD/GPL
  depends:
  intree:         Y
  name:           sch_cake
  vermagic:       6.1.65 SMP mod_unload MIPS32_R2 32BIT

  root@OpenWrt:/# modprobe sch_fq_codel && echo SUCCESS || echo FAIL
  SUCCESS
  root@OpenWrt:/# modprobe sch_cake && echo SUCCESS || echo FAIL
  SUCCESS

Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
10 months agokmodloader: support loadable module parameters in modinfo
Tony Ambardar [Fri, 1 Dec 2023 09:40:17 +0000 (01:40 -0800)]
kmodloader: support loadable module parameters in modinfo

Current OpenWrt loadable modules embed details of parameters accepted on
loading, but these aren't shown to users. Enable modinfo to print this
information like most other distros. For example:

  root@OpenWrt:/# modinfo mac80211
  filename:       /lib/modules/6.1.65/mac80211.ko
  license:        GPL
  depends:        cfg80211,compat
  name:           mac80211
  vermagic:       6.1.65 SMP mod_unload MIPS32_R2 32BIT
  parm:           minstrel_vht_only (bool)
  parm:           max_nullfunc_tries (int)
  parm:           max_probe_tries (int)
  parm:           beacon_loss_count (int)
  parm:           probe_wait_ms (int)
  parm:           ieee80211_default_rc_algo (charp)

Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
10 months agokmodloader: fix memory leak in scan_loaded_modules()
Tony Ambardar [Mon, 15 Jan 2024 20:29:48 +0000 (12:29 -0800)]
kmodloader: fix memory leak in scan_loaded_modules()

Commit 4c7b720b9c63 fixed some error-handling paths but introduced another
memory leak; update and fix.

Fixes: 4c7b720b9c63 ("kmodloader: fix GCC fanalyzer warnings")
Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
11 months agokmodloader: remove unneeded uname() call
Tony Ambardar [Thu, 14 Dec 2023 07:17:24 +0000 (23:17 -0800)]
kmodloader: remove unneeded uname() call

This is not used in scan_modules_folder() so drop.

Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
12 months agologd: add udebug support
Felix Fietkau [Thu, 30 Nov 2023 12:42:09 +0000 (13:42 +0100)]
logd: add udebug support

Signed-off-by: Felix Fietkau <nbd@nbd.name>
22 months agologd: add support for subscribing to the log object
Felix Fietkau [Thu, 19 Jan 2023 12:06:19 +0000 (13:06 +0100)]
logd: add support for subscribing to the log object

Provides easier access to log messages

Signed-off-by: Felix Fietkau <nbd@nbd.name>
2 years agokmodloader: fix GCC fanalyzer warnings
Rosen Penev [Mon, 18 Jul 2022 22:12:38 +0000 (15:12 -0700)]
kmodloader: fix GCC fanalyzer warnings

memory leaks and missing NULL checks.

Signed-off-by: Rosen Penev <rosenp@gmail.com>
2 years agokmodloader: fix bad realloc usage
Rosen Penev [Mon, 18 Jul 2022 22:12:37 +0000 (15:12 -0700)]
kmodloader: fix bad realloc usage

Both cppcheck and gcc's -fanalyzer complain here that realloc is being
used improperly.

Signed-off-by: Rosen Penev <rosenp@gmail.com>
2 years agokmodloader: fix compilation warning with not checking return of asprintf
Christian Marangi [Sun, 17 Jul 2022 00:59:32 +0000 (02:59 +0200)]
kmodloader: fix compilation warning with not checking return of asprintf

Fix the following compilation warning:

kmodloader.c: In function 'main_loader':
kmodloader.c:1027:41: error: ignoring return value of 'asprintf' declared with attribute 'warn_unused_result' [-Werror=unused-result]
make[1]: *** [package/Makefile:116: package/system/ubox/compile] Error 1
 1027 |                                         asprintf(&m->opts, "%s %s", prev, opts);
      |                                         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

While at it rework the function to not duplicate too much code with the
error handling.

Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
2 years agokmodloader: support loading module options and blacklist from /etc/modules.conf
Felix Fietkau [Thu, 6 Jan 2022 14:58:11 +0000 (15:58 +0100)]
kmodloader: support loading module options and blacklist from /etc/modules.conf

This has the same format as /etc/modules.d on a regular linux system

Signed-off-by: Felix Fietkau <nbd@nbd.name>
3 years agologread: fix erroneous message "Logread connected to" with udp
Giovanni Giacobbi [Tue, 27 Jul 2021 13:15:26 +0000 (15:15 +0200)]
logread: fix erroneous message "Logread connected to" with udp

When streaming the syslog messages via udp, the socket connection
always succeeds by definition, but it can still fail to send. In such
case, the syslog keep repeating the following two messages:

    failed to send log data to ip:port via udp
    Logread connected to ip:port

With this change, only one initial message "Logread connected to..." is
logged.

Also fixed capital letter for "failed to send" message.

Signed-off-by: Giovanni Giacobbi <giovanni@giacobbi.net>
Signed-off-by: Daniel Golle <daniel@makrotopia.org> (string de-duplication)
3 years agologd: fix privilege dropping order
Giovanni Giacobbi [Tue, 27 Jul 2021 12:39:28 +0000 (14:39 +0200)]
logd: fix privilege dropping order

Fixes: 41664054b8b1 ("logd: fix ignored return values in set{gid,uid}")
Signed-off-by: Giovanni Giacobbi <giovanni@giacobbi.net>
3 years agologd: fix ignored return values in set{gid,uid}
Petr Štetiar [Mon, 22 Mar 2021 17:22:14 +0000 (18:22 +0100)]
logd: fix ignored return values in set{gid,uid}

Ubuntu 18.05 with gcc-7.5 yields following error:

 ubox/log/logd.c:263:3: error: ignoring return value of ‘setuid’, declared with attribute warn_unused_result [-Werror=unused-result]
   setuid(p->pw_uid);
   ^~~~~~~~~~~~~~~~~
 ubox/log/logd.c:264:3: error: ignoring return value of ‘setgid’, declared with attribute warn_unused_result [-Werror=unused-result]
   setgid(p->pw_gid);
   ^~~~~~~~~~~~~~~~~

Fixes: 9ef886819dd4 ("logd: self-degrade to 'logd' user after opening pipes")
Signed-off-by: Petr Štetiar <ynezz@true.cz>
4 years agologd: self-degrade to 'logd' user after opening pipes
Daniel Golle [Mon, 19 Oct 2020 15:18:22 +0000 (16:18 +0100)]
logd: self-degrade to 'logd' user after opening pipes

Signed-off-by: Daniel Golle <daniel@makrotopia.org>
4 years agokmodloader: added -a arg to modprobe
Gerard Ryan [Tue, 31 Dec 2019 02:01:35 +0000 (12:01 +1000)]
kmodloader: added -a arg to modprobe

-a treats all non-op trailing arguments as module names
and attempts to load all of them. This behaviour mirrors the behaviour
of the respective -a in /tools/modprobe.c from https://git.kernel.org.

This is primarily to satiate the multiple modules passed by
docker/libnetwork.

Signed-off-by: Gerard Ryan <G.M0N3Y.2503@gmail.com>
4 years agokmodloader: print an error when no kernel module dir can be found
Paul Fertser [Wed, 18 Dec 2019 16:39:42 +0000 (19:39 +0300)]
kmodloader: print an error when no kernel module dir can be found

Do not just silently fail when kernel modules are nowhere to be
found. This is essential for diagnosing mismatched kernel version
failures.

Signed-off-by: Paul Fertser <fercerpav@gmail.com>
5 years agologread: add option to filter for facilities
Zefir Kurtisi [Thu, 25 Jul 2019 15:43:15 +0000 (17:43 +0200)]
logread: add option to filter for facilities

This adds filtering options for facilities as follows:
    -z <facility> handle only messages with given facility (0-23), repeatable
    -Z <facility> ignore messages with given facility (0-23), repeatable

With that
* 'logread -z 2 -z 16' will display all MAIL and LOCAL0 messages, while
* 'logread -Z 2 -Z 16' will display all but MAIL and LOCAL0 messages

Signed-off-by: Zefir Kurtisi <zefir.kurtisi@neratec.com>
5 years agokmodloader: added -v arg to modeprobe
Gerard Ryan [Sun, 20 Oct 2019 06:01:11 +0000 (16:01 +1000)]
kmodloader: added -v arg to modeprobe

This is primarily to satiate the usage by docker/libnetwork.

Signed-off-by: Gerard Ryan <G.M0N3Y.2503@gmail.com>
5 years agokmodloader: Increase path array size to make it always fit
Hauke Mehrtens [Sat, 15 Jun 2019 21:49:28 +0000 (23:49 +0200)]
kmodloader: Increase path array size to make it always fit

When DEF_MOD_PATH (10 bytes), s (max 255 bytes) and ver.release (max 64
bytes) are all written into path, 256 bytes will not fit, increase the
available size. All given sizes are given without terminating NULL byte
and we add one byte to store the NULL byte.

GCC 9.1 warns about this condition and we treat warnings as errors in
ubox.

Reported-by: Joseph Benden <joe@benden.us>
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
5 years agokmodloader: fix and optimize loading of failed modules
Pavel Merzlyakov [Mon, 18 Mar 2019 10:53:01 +0000 (12:53 +0200)]
kmodloader: fix and optimize loading of failed modules

1) Restore functionality which was lost in commit 876c7f5b.
   Again at boot time kmodloader can load all modules (/etc/modules.d/*)
   even if dependency information is completely missing.
   This functionality is important in case of hidden dependency (not symbol dependency).
   For example, in kernel 4.4.60 is hidden dependency between nf_nat_ipv6 and nf_conntrack_ipv6.
   We can't load nf_nat_ipv6 before nf_conntrack_ipv6 and modinfo do not show this dependency.
   Two sequential load attempts of nf_nat_ipv6 may not be enough (in my case it's definitely not enough).
   nf_nat_ipv4 has a similar problem.

2) Reduce count of attempts to load failed modules.
   Now kmodloader try to load failed modules after all others are loaded.

main_loader: Count of failed and successful attempts to load nf_nat_ipv6.ko (kernel 4.4.60) depend on ubox version:
   COMMIT        FAILED      SUCCESSFUL     TOTAL
   128bc35f      53          1              54
   876c7f5b      2           0              2
   this          1           1              2

Signed-off-by: Pavel Merzlyakov <pavel.merzlyakov@gmail.com>
Acked-by: Jo-Philipp Wich <jo@mein.io>
5 years agokmodloader: increase module name length
Sergiy Kibrik [Fri, 22 Feb 2019 06:57:52 +0000 (06:57 +0000)]
kmodloader: increase module name length

Otherwise modules with long names, e.g. lttng (>32 chars) can't be loaded.

Signed-off-by: Sergiy Kibrik <sakib@darkstar.site>
5 years agokmodloader: load_modprobe: abort after 2 attempts
Stijn Tintel [Tue, 18 Dec 2018 19:24:28 +0000 (21:24 +0200)]
kmodloader: load_modprobe: abort after 2 attempts

The loop in load_modprobe causes dmesg to be flooded with errors when a
module fails to load; e.g.:
- gpio-nct5104d: Unsupported device 0xffff (246 times on x86 in Qemu)
- jitterentropy: Initialization failed with host not compliant with
  requirements: 2 (229 times on brcm2708/bcm2708)

On brcm2708/bcm2708 this also causes vmalloc allocation failures with
kernel 4.14 when running headless. If kmod-drm-vc4 and
kmod-video-bcm2835 are included in the image, they will be loaded before
brcmfmac, and brcmfmac will fail to load:

kmodloader: vmalloc: allocation failure: 249856 bytes,
mode:0x14000c0(GFP_KERNEL), nodemask=(null)

Break the loop after 2 failed attempts to fix these issues.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Signed-off-by: Stijn Tintel <stijn@linux-ipv6.be>
6 years agologread: fix reconnect logd logic
John Crispin [Wed, 14 Feb 2018 08:08:54 +0000 (09:08 +0100)]
logread: fix reconnect logd logic

Signed-off-by: John Crispin <john@phrozen.org>
6 years agologrea: move the code setting up the request blob out of the main loop
John Crispin [Wed, 14 Feb 2018 07:47:03 +0000 (08:47 +0100)]
logrea: move the code setting up the request blob out of the main loop

Signed-off-by: John Crispin <john@phrozen.org>
6 years agologread: move output connection setup code out of main loop
John Crispin [Wed, 14 Feb 2018 07:44:27 +0000 (08:44 +0100)]
logread: move output connection setup code out of main loop

Signed-off-by: John Crispin <john@phrozen.org>
6 years agologread: cleanup pid file handling
John Crispin [Wed, 14 Feb 2018 07:40:13 +0000 (08:40 +0100)]
logread: cleanup pid file handling

Signed-off-by: John Crispin <john@phrozen.org>
6 years agoubox: Replace strerror(errno) with %m format.
Rosen Penev [Sat, 13 Jan 2018 20:46:04 +0000 (12:46 -0800)]
ubox: Replace strerror(errno) with %m format.

Saves 872 bytes on Ubuntu 16.04. No change in functionality.

Signed-off-by: Rosen Penev <rosenp@gmail.com>
7 years agoubox: Remove unnecessary memset calls.
Rosen Penev [Tue, 7 Nov 2017 20:22:09 +0000 (12:22 -0800)]
ubox: Remove unnecessary memset calls.

Change to calloc instead. Less verbose.

Signed-off-by: Rosen Penev <rosenp@gmail.com>
7 years agologd: move stripping of newlines to log_add()
Felix Fietkau [Mon, 6 Nov 2017 10:48:52 +0000 (11:48 +0100)]
logd: move stripping of newlines to log_add()

Signed-off-by: Felix Fietkau <nbd@nbd.name>
7 years agologread: fix line buffer size
Felix Fietkau [Mon, 6 Nov 2017 10:47:57 +0000 (11:47 +0100)]
logread: fix line buffer size

Signed-off-by: Felix Fietkau <nbd@nbd.name>
7 years agologread: remove leftover debug code
Felix Fietkau [Mon, 6 Nov 2017 10:46:24 +0000 (11:46 +0100)]
logread: remove leftover debug code

Signed-off-by: Felix Fietkau <nbd@nbd.name>
7 years agologd: enforce line length limit for ubus based log messages as well
Felix Fietkau [Mon, 6 Nov 2017 10:34:31 +0000 (11:34 +0100)]
logd: enforce line length limit for ubus based log messages as well

Signed-off-by: Felix Fietkau <nbd@nbd.name>
7 years agologread: use oneshot mode without -f, wait for logd to close
Felix Fietkau [Mon, 6 Nov 2017 10:28:33 +0000 (11:28 +0100)]
logread: use oneshot mode without -f, wait for logd to close

Fixes a race condition in dealing with larger amounts of log data

Signed-off-by: Felix Fietkau <nbd@nbd.name>
7 years agologd: implement oneshot mode for stream log read
Felix Fietkau [Mon, 6 Nov 2017 10:27:46 +0000 (11:27 +0100)]
logd: implement oneshot mode for stream log read

This will terminate the connection after all log messages have been
sent. Useful for logread without follow

Signed-off-by: Felix Fietkau <nbd@nbd.name>
7 years agologread: terminate after EOF
Felix Fietkau [Mon, 6 Nov 2017 10:15:39 +0000 (11:15 +0100)]
logread: terminate after EOF

Signed-off-by: Felix Fietkau <nbd@nbd.name>
7 years agologd: only create pipe in stream mode
Felix Fietkau [Mon, 6 Nov 2017 10:11:49 +0000 (11:11 +0100)]
logd: only create pipe in stream mode

Signed-off-by: Felix Fietkau <nbd@nbd.name>
7 years agologd: use uloop instead of ustream_fd for syslog
Felix Fietkau [Mon, 6 Nov 2017 09:46:38 +0000 (10:46 +0100)]
logd: use uloop instead of ustream_fd for syslog

Using a stream buffer for a datagram socket makes no sense. This change
fixes dealing with line buffer truncation on large incoming messages

Signed-off-by: Felix Fietkau <nbd@nbd.name>
7 years agolog/syslog: Add missing static to two structs.
Rosen Penev [Mon, 30 Oct 2017 19:05:30 +0000 (12:05 -0700)]
log/syslog: Add missing static to two structs.

On x86, binary size goes down by 16 bytes.

Signed-off-by: Rosen Penev <rosenp@gmail.com>
7 years agoubox: Replace { 0 } with {}.
Rosen Penev [Mon, 30 Oct 2017 19:05:31 +0000 (12:05 -0700)]
ubox: Replace { 0 } with {}.

The latter is more compatible with older GCC versions like 4.9. Additionally, initializing with { 0 } produces a warning that only the first field was initialized.

Signed-off-by: Rosen Penev <rosenp@gmail.com>
7 years agokmodloader: log error message in case of out of memory
Hans Dedecker [Thu, 31 Aug 2017 12:08:11 +0000 (14:08 +0200)]
kmodloader: log error message in case of out of memory

Log "out of memory" error message in case of OOM

Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
7 years agokmodloader: lift restriction on module alias info
Hans Dedecker [Wed, 30 Aug 2017 11:32:43 +0000 (13:32 +0200)]
kmodloader: lift restriction on module alias info

kmodloader has a restriction of storing only 32 aliases for a given
module; as modules can have easily more than 32 aliases let's remove
the restriction by using a dynamic allocation mechanism when retrieving
the aliases.
While at it also check the get_module_info return value setting an error
code in case NULL is returned.

Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
7 years agokmodloader: fix possible segfaults
Hans Dedecker [Wed, 30 Aug 2017 13:09:49 +0000 (15:09 +0200)]
kmodloader: fix possible segfaults

Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
7 years agokmodloader: fix endianess check
Hans Dedecker [Wed, 30 Aug 2017 13:23:58 +0000 (15:23 +0200)]
kmodloader: fix endianess check

Use __BYTE_ORDER macro to check system endianess as the macro's
__LITTLE_ENDIAN, __BIG_ENDIAN are defined irrespective of the
sytem endianess leading to an endianess check failure on big
endian systems.

Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Acked-by: John Crispin <john@phrozen.org>
7 years agokmodloader: Check module endian before loading
Florian Fainelli [Fri, 30 Jun 2017 23:34:22 +0000 (16:34 -0700)]
kmodloader: Check module endian before loading

If we manage to accidentally get a module installed that does not match
the system endianess (there is currently nothing preventing that in
OpenWrt/LEDE) we will hit a segmentation fault when doing this:

sh = (Elf32_Shdr *) (map + e->e_shoff);

because e->e_shoff will be wrongly encoded and this most likely will
result in an invalid offset to de-reference from.

Update elf_find_section() to check that the build time endianess of
kmodloader matches that of the module, otherwise nothing would work
really.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
7 years agokmodloader/get_module_info: initialized aliases to make it more clean
Alexander Couzens [Sun, 11 Jun 2017 10:33:34 +0000 (12:33 +0200)]
kmodloader/get_module_info: initialized aliases to make it more clean

Even when the callee looks onto the length is a good idea to zero
the points.

Found-by: Coverity Scan #1412541
Signed-off-by: Alexander Couzens <lynxis@fe80.eu>
7 years agokmodloader: insmod: fix a memoryleak in error case
Alexander Couzens [Sun, 11 Jun 2017 10:20:59 +0000 (12:20 +0200)]
kmodloader: insmod: fix a memoryleak in error case

Found-by: Coverity Scan #1412469
Signed-off-by: Alexander Couzens <lynxis@fe80.eu>
7 years agokmodloader/get_module_name: null-terminate the string
Alexander Couzens [Sun, 11 Jun 2017 10:13:05 +0000 (12:13 +0200)]
kmodloader/get_module_name: null-terminate the string

Found-by: Coverity Scan #1412291
Signed-off-by: Alexander Couzens <lynxis@fe80.eu>
7 years agosyslog: remove unnecessary sizeof struct between messages
Dan Bugnar [Fri, 10 Mar 2017 10:06:29 +0000 (12:06 +0200)]
syslog: remove unnecessary sizeof struct between messages

The next message needs to be written after the data of current message.
This was adding "sizeof(struct log_head)" bytes between messages.

Signed-off-by: Dan Bugnar <danutbug@gmail.com>
7 years agokmodloader: modprobe: return 0 for loaded modules
Yousong Zhou [Fri, 3 Mar 2017 02:32:48 +0000 (10:32 +0800)]
kmodloader: modprobe: return 0 for loaded modules

This is the default behaviour with modprobe from kmod package [1] unless
it is explicitly told that the module is to be loaded for --first-time

 [1] http://www.kernel.org/pub/linux/utils/kernel/kmod/

Fixes FS#433

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
7 years agokmodloader: don't store aliases info in struct module
Yousong Zhou [Thu, 23 Feb 2017 03:04:50 +0000 (11:04 +0800)]
kmodloader: don't store aliases info in struct module

This also fixes FS#544 as the possible address alignment issue should
now disappear

function                                             old     new   delta
alloc_module                                         398     245    -153
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-153)           Total: -153 bytes

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
7 years agocmake: fix typo
Felix Fietkau [Tue, 21 Feb 2017 13:11:03 +0000 (14:11 +0100)]
cmake: fix typo

Fixes: fce9382b4b99 ("cmake: Check for getrandom system call")
Signed-off-by: Felix Fietkau <nbd@nbd.name>
7 years agokmodloader: fix not being able to find some modules
Nathan Hintz [Tue, 21 Feb 2017 05:14:24 +0000 (05:14 +0000)]
kmodloader: fix not being able to find some modules

kmodloader is using slightly different criteria for ordering the AVL tree
versus what it uses to traverse it.  This sometimes results in not being
able to find some modules.

Reference: https://bugs.lede-project.org/index.php?do=details&task_id=443

Signed-off-by: Nathan Hintz <nlhintz@hotmail.com>
7 years agocmake: Check for getrandom system call
Florian Fainelli [Sun, 5 Feb 2017 02:41:34 +0000 (18:41 -0800)]
cmake: Check for getrandom system call

In case we are building against a kernel that is too old and does not
support SYS_getrandom, error out with a message indicating so.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
7 years agokmodloader: support '-q' quiet option
Kevin Darbyshire-Bryant [Sun, 29 Jan 2017 09:23:15 +0000 (09:23 +0000)]
kmodloader: support '-q' quiet option

The kernel opportunistically attempts to load modules in advanced with
'predicted' module names.  Often these modules don't exist and hence
kmodloader produces lots of logfile noise.  The kernel commandline to
modprobe from kworker proceses is '-q -- modulename' where '-q' means
quiet. Support suppressing that noise.

Signed-off-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
7 years agoubox: Add an option for more accurate timestamps in log
Wojciech Dubowik [Mon, 23 Jan 2017 10:30:14 +0000 (11:30 +0100)]
ubox: Add an option for more accurate timestamps in log

Systemwide timestamps for syslog messages with ms accuracy are usefull
for debugging timing problems.

Signed-off-by: Wojciech Dubowik <Wojciech.Dubowik@neratec.com>
7 years agokmodloader: make insert_module() idempotent
Yousong Zhou [Fri, 13 Jan 2017 17:00:35 +0000 (01:00 +0800)]
kmodloader: make insert_module() idempotent

To fix spurious error messages in the following situation

 1. scan loaded modules
 2. load wireguard.ko and the module itself will request xt_hashlimit to
    be loaded
 3. xt_hashlimit is still in PROBE state here so we also try to load it,
    but init_module() returns EEXIST

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
7 years agokmodloader: add module alias awareness
Yousong Zhou [Fri, 13 Jan 2017 17:00:34 +0000 (01:00 +0800)]
kmodloader: add module alias awareness

To achieve this, the following changes are made

 - scan module folders before scaning loaded modules
 - struct module was splited into struct module_node for the avl tree
   and struct module for storing actual info about modules

The other minor fix is that module.opts is now freed in free_modules()
when appliable

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
7 years agokmodloader: fix out-of-bound access when parsing .modinfo
Yousong Zhou [Fri, 13 Jan 2017 17:00:33 +0000 (01:00 +0800)]
kmodloader: fix out-of-bound access when parsing .modinfo

Fixes output of "modinfo nf_conntrack_ipv4"

    module:         /lib/modules/4.4.40/nf_conntrack_ipv4.ko
    license:        GPL
    alias:          ip_conntrack
    alias:          nf_conntrack-2
    depends:        nf_conntrack,nf_defrag_ipv4
    src:            %pI4 dst=%pI4

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
7 years agokmodloader: modprobe: skip possible command line arguments
Yousong Zhou [Fri, 13 Jan 2017 17:00:32 +0000 (01:00 +0800)]
kmodloader: modprobe: skip possible command line arguments

The kernel may invocate user mode modprobe with the following scheme

    modprobe -q -- <module_name>

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
7 years agokmodloader: log to kmsg when loading directories of modules
Yousong Zhou [Fri, 13 Jan 2017 17:00:31 +0000 (01:00 +0800)]
kmodloader: log to kmsg when loading directories of modules

syslog may not be ready yet before PREINIT

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
7 years agokmodloader: remove redundant glob wildcard char
Yousong Zhou [Fri, 13 Jan 2017 17:00:30 +0000 (01:00 +0800)]
kmodloader: remove redundant glob wildcard char

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
7 years agoubox: Initialize conditionally uninitialized variabled
Rosen Penev [Sun, 18 Dec 2016 00:48:47 +0000 (16:48 -0800)]
ubox: Initialize conditionally uninitialized variabled

fixes false positive warning from "-Wconditional-uninitialized in clang"

Signed-off by: Rosen Penev <rosenp@gmail.com>

7 years agoubox: Fix some memory leaks
Rosen Penev [Sun, 18 Dec 2016 01:12:28 +0000 (17:12 -0800)]
ubox: Fix some memory leaks

Avoids leaking memory when exiting early.

Signed-off by: Rosen <rosenp@gmail.com>

7 years agokmodloader: Fix typo in error message
Florian Fainelli [Sun, 4 Dec 2016 02:09:39 +0000 (18:09 -0800)]
kmodloader: Fix typo in error message

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
8 years agologd: put non-streamed log entries into one result message
Felix Fietkau [Mon, 26 Sep 2016 14:28:52 +0000 (16:28 +0200)]
logd: put non-streamed log entries into one result message

Signed-off-by: Felix Fietkau <nbd@nbd.name>
8 years agoubox/logd: free ubus context on exit
Alexandru Ardelean [Sat, 24 Sep 2016 08:03:13 +0000 (11:03 +0300)]
ubox/logd: free ubus context on exit

Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>
8 years agoubox/logd: free regex's on log_shutdown() call
Alexandru Ardelean [Sat, 24 Sep 2016 08:03:12 +0000 (11:03 +0300)]
ubox/logd: free regex's on log_shutdown() call

Signed-off-by: Alexandru Ardelean <ardeleanalex@gmail.com>
8 years agologd: fix default initialization of line count
Felix Fietkau [Thu, 22 Sep 2016 18:20:29 +0000 (20:20 +0200)]
logd: fix default initialization of line count

Signed-off-by: Felix Fietkau <nbd@nbd.name>
8 years agologd: add support for sending log messages as replies
Felix Fietkau [Thu, 22 Sep 2016 14:55:28 +0000 (16:55 +0200)]
logd: add support for sending log messages as replies

Useful for JSON-RPC access via uhttpd

Signed-off-by: Felix Fietkau <nbd@nbd.name>
8 years agoadd getrandom to .gitignore
Felix Fietkau [Thu, 22 Sep 2016 14:39:49 +0000 (16:39 +0200)]
add getrandom to .gitignore

Signed-off-by: Felix Fietkau <nbd@nbd.name>
8 years agogetrandom: fix indentation, add missing return statement
Felix Fietkau [Fri, 5 Aug 2016 11:08:53 +0000 (13:08 +0200)]
getrandom: fix indentation, add missing return statement

Signed-off-by: Felix Fietkau <nbd@nbd.name>
8 years agofix segfault in kmodloader when no modules path was present
John Crispin [Wed, 17 Aug 2016 02:32:43 +0000 (04:32 +0200)]
fix segfault in kmodloader when no modules path was present

Signed-off-by: John Crispin <john@phrozen.org>
8 years agovalidate: fix datatype handling of default values
Jo-Philipp Wich [Sat, 16 Jul 2016 12:52:36 +0000 (14:52 +0200)]
validate: fix datatype handling of default values

Properly handle the datatype of default values when the corresponding uci
value is unset. Also consolidate the three default value code branches into
a single one.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>
8 years agokmodloader: fix lsmod depends output
Hauke Mehrtens [Wed, 13 Jul 2016 14:55:20 +0000 (16:55 +0200)]
kmodloader: fix lsmod depends output

Without this patch only the first dependency is shown, with this patch
all module dependencies are show.

Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
8 years agocmake: Find uci header file
Florian Fainelli [Fri, 1 Jul 2016 23:05:23 +0000 (16:05 -0700)]
cmake: Find uci header file

Add a CMake FIND_PATH and INCLUDE_DIRECTORIES searching for uci.h. Some
external toolchains which do not include standard locations would fail
to find the header otherwise.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
8 years agogetrandom: add helper for getrandom() syscall
Etienne CHAMPETIER [Thu, 9 Jun 2016 13:03:00 +0000 (13:03 +0000)]
getrandom: add helper for getrandom() syscall

getrandom() was introduced in version 3.17 of the Linux kernel.
By default getrandom() block until /dev/urandom pool has been initialized
and then read from it.
Read buffer is 256 bytes so getrandom() calls always succeed.

First usage will be to save a seed for /dev/urandom.

Signed-off-by: Etienne CHAMPETIER <champetier.etienne@gmail.com>
8 years agokmodloader: fix NULL pointer deref in case m->depends is missing
Felix Fietkau [Mon, 7 Mar 2016 16:34:31 +0000 (17:34 +0100)]
kmodloader: fix NULL pointer deref in case m->depends is missing

Observed on an invalid module

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
8 years agokmodloader: clean up memory mapping and fix fd leaks in error handling
Jo-Philipp Wich [Tue, 9 Feb 2016 12:49:55 +0000 (13:49 +0100)]
kmodloader: clean up memory mapping and fix fd leaks in error handling

Signed-off-by: Jo-Philipp Wich <jow@openwrt.org>
8 years agokmodloader: fix open() error handling
Jo-Philipp Wich [Tue, 9 Feb 2016 07:53:28 +0000 (08:53 +0100)]
kmodloader: fix open() error handling

Fixes kmodloader on systems without stdio and prevents fd leaks in case
descriptor zero is opened.

Signed-off-by: Jo-Philipp Wich <jow@openwrt.org>
8 years agologread: ensure that the len < sizeof(struct blob_attr) check runs before trying...
Felix Fietkau [Sun, 24 Jan 2016 08:19:53 +0000 (09:19 +0100)]
logread: ensure that the len < sizeof(struct blob_attr) check runs before trying to access the data

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
9 years agofix gcc format security error
Hauke Mehrtens [Sun, 22 Nov 2015 14:45:27 +0000 (15:45 +0100)]
fix gcc format security error

Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
9 years agolog: allow filtering messages with a regexp pattern
Rafał Miłecki [Tue, 14 Jul 2015 06:43:55 +0000 (08:43 +0200)]
log: allow filtering messages with a regexp pattern

This allows printing/streaming/writing messages matching a specified
basic regular expression only.

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
9 years agologd: optimize notify function
Luka Perkov [Tue, 21 Apr 2015 17:04:46 +0000 (19:04 +0200)]
logd: optimize notify function

No need to construct every time the same blob_buf for each client in the list.

Signed-off-by: Luka Perkov <luka@openwrt.org>
9 years agolog: make valgrind less complain by freeing allocated memory
Luka Perkov [Tue, 21 Apr 2015 17:04:45 +0000 (19:04 +0200)]
log: make valgrind less complain by freeing allocated memory

Signed-off-by: Luka Perkov <luka@openwrt.org>
9 years agolog: fix whitespaces
Luka Perkov [Tue, 21 Apr 2015 17:04:44 +0000 (19:04 +0200)]
log: fix whitespaces

Signed-off-by: Luka Perkov <luka@openwrt.org>
9 years agolog: remove unused callbacks
Luka Perkov [Tue, 21 Apr 2015 17:04:43 +0000 (19:04 +0200)]
log: remove unused callbacks

Signed-off-by: Luka Perkov <luka@openwrt.org>