Use crypt_ra to allocate scratch area for password hashing #16981

keszybz · 2020-09-08T14:32:42Z

No description provided.

poettering

I'd really hide the buffer allocation in the inner-most code block. The reusing of the allocated buffers seems a lot complexity in the interface for pretty much no gain, since we typically deal with a single password to hash only, and in the worst case maybe a handful, but this still shouldn't matter at all performance-wise.

Or to say this differently: the actual hashing of the pw is likely a lot more expensive than the malloc() we might do too much, hence it shouldn't matter

src/firstboot/firstboot.c

src/home/user-record-util.c

src/home/homework.c

keszybz · 2020-09-08T16:14:14Z

I made the cached allocation mostly hidden, but I left it in places where we loop over passwords, e.g. in user_record_quality_check_password(). PTAL.

keszybz · 2020-09-08T16:18:16Z

Hmm, the build fails because crypt_ra is not available, strange.

poettering · 2020-09-08T16:18:57Z

CI is not happy...

I still wouldn't bother even for the pw quality check stuff. Yeah, it looks like O(n^2), but given that n is almost always == 1, the malloc() shouldn't matter...

poettering · 2020-09-08T16:21:08Z

oh, uh.

i think i remember now: crypt_ra() is libxcrypt-only. glibc didn't have it. our CIs still use glibcs that did not do the libxcrypt split-out.

I guess this means there needs to be some fallback for a while. crypt + crypt_r are available on glibc, and crypt_rn/crypt_ra are openwall/libxcrypt additions.

keszybz · 2020-09-08T16:41:21Z

I removed the caching of the buffer now. It is only exposed in the low-level hash_password_full() function and used in tests. "Normal" callers use the wrapper which always allocates.

src/shared/libcrypt-util.c

poettering · 2020-09-08T18:06:11Z

meson.build

+crypt_header = conf.get('HAVE_CRYPT_H') == 1 ? \
+               '''#include <crypt.h>''' : '''#include <unistd.h>'''
+foreach ident : [
+        ['crypt_r',           crypt_header],


crypt_r is glibc already, we should not need to test for it.

The impetus for this PR came from centos ci which reports:

20:36:36 archlinux_systemd_ci: Checking for function "crypt_r" with dependency -lcrypt: YES 20:36:36 archlinux_systemd_ci: Checking for function "crypt_rn" with dependency -lcrypt: NO 20:36:36 archlinux_systemd_ci: Checking for function "crypt_ra" with dependency -lcrypt: NO

If we had crypt_rn, there'd be hope to improve things. But since there is neither _ra nor _rn, most likely this will not resolve #16965, unless the glibc version there is updated. I think it's still worthwhile to pursue this PR, since the _ra variant seems to be a much better API. WDYT?

(Am I missing some magic option to "enable" xcrypt? At least in Fedora libxcrypt.rpm only provides libcrypt.so.2, so using -lcrypt is the only way to consume xcrypt.)

Hah, now I see why downgrading glibc on the current Arch Linux fails:

error: failed to commit transaction (conflicting files) libxcrypt: /usr/include/crypt.h exists in filesystem (owned by glibc) libxcrypt: /usr/lib/libcrypt.so exists in filesystem (owned by glibc)

The thing is that in recent glibc updates Arch transitioned from libcrypt in glibc to libcrypt in libxcrypt, see [0][1]. And, of course, the libxcrypt contains the functions you seek:

libxcrypt /usr/share/man/man3/crypt_r.3.gz libxcrypt /usr/share/man/man3/crypt_ra.3.gz libxcrypt /usr/share/man/man3/crypt_rn.3.gz

But right now the upgrade is blocked by the sanitizer issue, but to make it go away we need to upgrade the image, so we're in a chicken-egg situation. I guess I could temporarily disable the test which triggers the issue (TEST-46) and go ahead with the upgrade.

all i wanted to say: there's no point in checking for crypt_r, since our baseline, i.e. old glibc had that since time began.

checking for crypt_ra() definitely makes sense since glibc never had that, and onl libxcrypt has.

crypt_rn is a call we probably don#t care about, we shouldn't test for it hence

keszybz · 2020-09-10T16:41:30Z

As discussed above, I think this is still worth pursuing even if it doesn't fix the immediate issue when libxcrypt is not available. Hopefully the tests will pass now.

poettering · 2020-09-10T16:43:17Z

maybe add a minimal local fallback implementation for crypt_ra() here, that just wraps crypt_r() and does malloc() itself. i.e. call it missing_crypt_ra() and #define it to crypt_ra() if thta's available and otherwise just malloc() locally, and call crypt_r().

(I guess the whole issue here is just fix what this the comment in crypt(3) suggests: "struct crypt_data may be quite large (32kB in this implementation of libcrypt; over 128kB in some other implementations). This is large enough that it may be unwise to allocate it on the stack.")

poettering · 2020-09-10T16:44:59Z

using crypt_ra() is probably more than just an option, given this further comment from the man page:

"Some recently designed hashing methods need even more scratch memory, but the crypt_r interface makes it impossible to change the size of struct crypt_data without breaking binary compatibility. The crypt_rn interface could
accommodate larger allocations for specific hashing methods, but the caller of crypt_rn has no way of knowing how much memory to allocate. crypt_ra does the allocation itself, but can only make a single call to malloc(3)."

So yeah, unless we use crypt_ra() if it is available we would not be compatible with contemporary hashes, and I think we should

poettering · 2020-09-10T16:45:42Z

ah, now i see you just pushed such a fallback...

poettering · 2020-09-10T16:46:59Z

src/shared/libcrypt-util.c

@@ -75,21 +75,23 @@ int make_salt(char **ret) {
 #endif
 }

-int hash_password(const char *password, char **ret) {
+int hash_password_full(const char *password, void **cd_data, int *cd_size, char **ret) {


i still wouldn't expose the cd_data, sounds like needless optimization. i.e it's the kind of optimization you might want to do ina hash breaker or so, but otherwise I really wouldn't bother...

This is exposed here, but is only used in the test. I think it is worthwhile to show that in tests: it should make it much easier to debug issues like #16965.

poettering · 2020-09-10T16:51:11Z

meson.build

+crypt_header = conf.get('HAVE_CRYPT_H') == 1 ? \
+               '''#include <crypt.h>''' : '''#include <unistd.h>'''
+foreach ident : [
+        ['crypt_r',           crypt_header],


all i wanted to say: there's no point in checking for crypt_r, since our baseline, i.e. old glibc had that since time began.

checking for crypt_ra() definitely makes sense since glibc never had that, and onl libxcrypt has.

crypt_rn is a call we probably don#t care about, we shouldn't test for it hence

poettering · 2020-09-10T16:53:20Z

src/shared/libcrypt-util.c

+/* Provide a poor man's fallback that uses a fixed size buffer. */
+
+static char* systemd_crypt_ra(const char *phrase, const char *setting, void **data, int *size) {
+        struct crypt_data cd = {};


i think we should malloc() this, given that the structure might be too large for the stack according to the man page:

"struct crypt_data may be quite large (32kB in this implementation of libcrypt; over 128kB in some other implementations). This is large enough that it may be unwise to allocate it on the stack".

mrc0mmand · 2020-09-10T16:59:55Z

As discussed above, I think this is still worth pursuing even if it doesn't fix the immediate issue when libxcrypt is not available. Hopefully the tests will pass now.

As the issue is reproducible only with libxcrypt (which is not in the current images), the tests should definitely pass. However, once the outstanding reviews in this PR are resolved, I'll run it manually against the upgraded image to see if the issue is indeed gone.

src/home/homework.c

keszybz · 2020-09-15T09:42:46Z

centos7:

10:38:23 --- test-libcrypt-util begin ---
10:38:23 Found container virtualization none.
10:38:23 /* test_hash_password */
10:38:23 ew3bU1.hoKk4o: yes
10:38:23 $1$gc5rWpTB$wK1aul1PyBn9AX1z93stk1: no
10:38:23 $2b$12$BlqcGkB/7BFvNMXKGxDea.5/8D6FTny.cbNcHW/tqcrcyo6ZJd8u2: no
10:38:23 $5$lGhDrcrao9zb5oIK$05KlOVG3ocknx/ThreqXE/gk.XzFFBMTksc4t2CPDUD: no
10:38:23 $6$c7wB/3GiRk0VHf7e$zXJ7hN0aLZapE.iO4mn/oHu6.prsXTUG/5k1AxpgR85ELolyAcaIGRgzfwJs3isTChMDBjnthZyaMCfCNxo9I.: no
10:38:23 $y$j9T$$9cKOWsAm4m97WiYk61lPPibZpy3oaGPIbsL4koRe/XD: no

That machine only supports original descrypt hash, how is that even possible? @mrc0mmand maybe it's missing some library.

They are only used under src/home/, but I want to add tests in test-libcrypt-util.c. And the functions are almost trivial, so I think it is OK to move them to shared.

Now that we wrap crypt_r/ra uses, we can include the header only in libcrypt-util.c.

We always seem to have either libcrypt_r and not the other two, or all three. So the fallback for libcrypt_ra needs to be based on libcrypt_r.

Since the loop to check various xcrypt functions is already in place, adding one more is cheap. And it is nicer to check for the function directly. People like to backport things, so we might get lucky even without having libxcrypt.

Following the style in missing_syscall.h, we use a non-conflicting name for the function and use a macro to map to the real name to the replacement.

…nsupported

I'm tired of trying to figure this out.

…lable On centos7 ci: --- test-libcrypt-util begin --- Found container virtualization none. /* test_hash_password */ ew3bU1.hoKk4o: yes $1$gc5rWpTB$wK1aul1PyBn9AX1z93stk1: no $2b$12$BlqcGkB/7BFvNMXKGxDea.5/8D6FTny.cbNcHW/tqcrcyo6ZJd8u2: no $5$lGhDrcrao9zb5oIK$05KlOVG3ocknx/ThreqXE/gk.XzFFBMTksc4t2CPDUD: no $6$c7wB/3GiRk0VHf7e$zXJ7hN0aLZapE.iO4mn/oHu6.prsXTUG/5k1AxpgR85ELolyAcaIGRgzfwJs3isTChMDBjnthZyaMCfCNxo9I.: no $y$j9T$$9cKOWsAm4m97WiYk61lPPibZpy3oaGPIbsL4koRe/XD: no

mrc0mmand · 2020-09-15T11:00:06Z

centos7:

10:38:23 --- test-libcrypt-util begin ---
10:38:23 Found container virtualization none.
10:38:23 /* test_hash_password */
10:38:23 ew3bU1.hoKk4o: yes
10:38:23 $1$gc5rWpTB$wK1aul1PyBn9AX1z93stk1: no
10:38:23 $2b$12$BlqcGkB/7BFvNMXKGxDea.5/8D6FTny.cbNcHW/tqcrcyo6ZJd8u2: no
10:38:23 $5$lGhDrcrao9zb5oIK$05KlOVG3ocknx/ThreqXE/gk.XzFFBMTksc4t2CPDUD: no
10:38:23 $6$c7wB/3GiRk0VHf7e$zXJ7hN0aLZapE.iO4mn/oHu6.prsXTUG/5k1AxpgR85ELolyAcaIGRgzfwJs3isTChMDBjnthZyaMCfCNxo9I.: no
10:38:23 $y$j9T$$9cKOWsAm4m97WiYk61lPPibZpy3oaGPIbsL4koRe/XD: no

That machine only supports original descrypt hash, how is that even possible? @mrc0mmand maybe it's missing some library.

I suspect this might be caused by non-descrypt methods being available only in the glibc extension or something like that?

From crypt(3) on CentOS 7:

NOTES
   Glibc notes
       The glibc2 version of this function supports additional encryption algorithms.

       If salt is a character string starting with the characters "$id$" followed by a string terminated by "$":

              $id$salt$encrypted

       then instead of using the DES machine, id identifies the encryption method used and this then determines how the rest of the password string is interpreted.  The following values of id are supported:

              ID  | Method
              ─────────────────────────────────────────────────────────
              1   | MD5
              2a  | Blowfish (not in mainline glibc; added in some
                  | Linux distributions)
              5   | SHA-256 (since glibc 2.7)
              6   | SHA-512 (since glibc 2.7)

When I tried $6$ manually on the CI machine, it worked as expected:

#define _GNU_SOURCE
#include <crypt.h>
#include <stdio.h>
#include <string.h>

int main(int argc, char *argv[])
{
	struct crypt_data cc = {};
	const char *hash = "$6$c7wB/3GiRk0VHf7e$zXJ7hN0aLZapE.iO4mn/oHu6.prsXTUG/5k1AxpgR85ELolyAcaIGRgzfwJs3isTChMDBjnthZyaMCfCNxo9I.";
	const char *k;

	k = crypt_r("ppp", hash, &cc);
	if (!k) {
		perror("crypt_r()");
		return 1;
	}

	printf("%s\n", (strcmp(k, hash) == 0) ? "pass" : "fail");

	return 0;
}

# gcc -o main main.c -lcrypt
# ./main 
pass

keszybz · 2020-09-15T11:50:34Z

When I tried $6$ manually on the CI machine, it worked as expected:

I assume that this is on the host itself. The failure happens in nested image in TEST-24-UNITTESTS.

mrc0mmand · 2020-09-15T12:18:43Z

When I tried $6$ manually on the CI machine, it worked as expected:

I assume that this is on the host itself. The failure happens in nested image in TEST-24-UNITTESTS.

Ah, my bad. Will look into that.

mrc0mmand · 2020-09-15T12:58:58Z

@keszybz so, after comparing straces from inside/outside of the VM and playing around with installing various missing libraries, installing /lib64/libfreeblpriv3.so (along with its dependencies) into the VM image does the trick.

E.g.:

diff --git a/test/test-functions b/test/test-functions
index 9893864..c3204a8 100644
--- a/test/test-functions
+++ b/test/test-functions
@@ -446,6 +446,10 @@ setup_basic_environment() {
     if [[ "$IS_BUILT_WITH_ASAN" = "yes" ]]; then
         create_asan_wrapper
     fi
+
+    inst_binary /root/systemd/main /main
+    inst_library /usr/lib64/libfreeblpriv3.so
+    inst /lib64/libfreeblpriv3.chk
 }

(ignore the /root/systemd/main binary)

Before:

-bash-4.2# strace -e open,openat /main
open("/etc/ld.so.cache", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libcrypt.so.1", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libc.so.6", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libfreebl3.so", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libdl.so.2", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libfreeblpriv3.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
open("/usr/lib64/libfreeblpriv3.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
open("/etc/ld.so.cache", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/tls/x86_64/libfreeblpriv3.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
open("/lib64/tls/libfreeblpriv3.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
open("/lib64/x86_64/libfreeblpriv3.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
open("/lib64/libfreeblpriv3.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
open("/usr/lib64/tls/x86_64/libfreeblpriv3.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
open("/usr/lib64/tls/libfreeblpriv3.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
open("/usr/lib64/x86_64/libfreeblpriv3.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
open("/usr/lib64/libfreeblpriv3.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
crypt_r() returned NULL
+++ exited with 1 +++

After:

-bash-4.2# strace -e open,openat /main
open("/etc/ld.so.cache", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libcrypt.so.1", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libc.so.6", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libfreebl3.so", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libdl.so.2", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libfreeblpriv3.so", O_RDONLY|O_CLOEXEC) = 3
open("/etc/ld.so.cache", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libnspr4.so", O_RDONLY|O_CLOEXEC) = 3
open("/dev/urandom", O_RDONLY)          = 3
open("/lib64/libfreeblpriv3.chk", O_RDONLY) = 3
open("/lib64/libfreeblpriv3.so", O_RDONLY) = 3
open("/etc/ld.so.cache", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libnspr4.so", O_RDONLY|O_CLOEXEC) = 3
open("/proc/sys/crypto/fips_enabled", O_RDONLY) = 3
hash comparison: pass
+++ exited with 0 +++

keszybz · 2020-09-15T13:21:22Z

@mrc0mmand wow, thanks for tracking this down. libfreeblpriv3.so is definitely not the first name that would come to mind.

The failing test was caused by missing deps. But I think we do want to test such systems too: this is not something that we verify during the configuration step, and it is possible that users might build on such systems. Also, in principle in the future systemd might be built on a system which only has newer hashes (or just different, e.g. $yg$ ), so it is better not to assume any particular hash is available, so I think we're good here.

poettering · 2020-09-18T17:00:00Z

Hmm, so what's the state of this? I guess we can just merge this? PR looks good to me. Any PPC work to delay this for? If not, let's merge!

keszybz · 2020-09-18T17:07:23Z

I think this is good to go.

bluca · 2020-09-24T14:06:22Z

src/shared/libcrypt-util.c

 #include "missing_stdlib.h"
 #include "random-util.h"
 #include "string-util.h"
 #include "strv.h"

 int make_salt(char **ret) {

-#ifdef XCRYPT_VERSION_MAJOR
+#if HAVE_CRYPT_GENSALT_RA


@keszybz this caused a regression for me - I have crypt_gensalt_ra but not crypt_preferred_method which is used in the same ifdef. I'll send a PR to check for both in meson.build and here.

After systemd#16981 only the presence of crypt_gensalt_ra is checked, but there are cases where that function is available but crypt_preferred_method is not, and they are used in the same ifdef. Add a check for the latter as well.

After #16981 only the presence of crypt_gensalt_ra is checked, but there are cases where that function is available but crypt_preferred_method is not, and they are used in the same ifdef. Add a check for the latter as well.

After systemd#16981 only the presence of crypt_gensalt_ra is checked, but there are cases where that function is available but crypt_preferred_method is not, and they are used in the same ifdef. Add a check for the latter as well.

After systemd/systemd#16981 only the presence of crypt_gensalt_ra is checked, but there are cases where that function is available but crypt_preferred_method is not, and they are used in the same ifdef. Add a check for the latter as well.

keszybz added util-lib firstboot homed homed, homectl, pam_homed labels Sep 8, 2020

poettering requested changes Sep 8, 2020

View reviewed changes

src/firstboot/firstboot.c Outdated Show resolved Hide resolved

src/home/user-record-util.c Outdated Show resolved Hide resolved

src/home/user-record-util.c Outdated Show resolved Hide resolved

src/home/homework.c Outdated Show resolved Hide resolved

poettering added the reviewed/needs-rework 🔨 PR has been reviewed and needs another round of reworks label Sep 8, 2020

poettering linked an issue Sep 8, 2020 that may be closed by this pull request

Possible stack buffer overflow in user_record_make_hashed_password() #16965

Closed

keszybz force-pushed the use-crypt_ra branch 2 times, most recently from 132e357 to 7b884b5 Compare September 8, 2020 16:11

keszybz removed the reviewed/needs-rework 🔨 PR has been reviewed and needs another round of reworks label Sep 8, 2020

keszybz added the ci-fails/needs-rework 🔥 Please rework this, the CI noticed an issue with the PR label Sep 8, 2020

keszybz force-pushed the use-crypt_ra branch from 7b884b5 to 9d74764 Compare September 8, 2020 16:39

poettering reviewed Sep 8, 2020

View reviewed changes

src/shared/libcrypt-util.c Show resolved Hide resolved

keszybz force-pushed the use-crypt_ra branch from 09a2c13 to 365711b Compare September 8, 2020 18:02

poettering reviewed Sep 8, 2020

View reviewed changes

keszybz marked this pull request as draft September 8, 2020 20:44

keszybz force-pushed the use-crypt_ra branch from 365711b to 46f500a Compare September 10, 2020 16:40

keszybz marked this pull request as ready for review September 10, 2020 16:40

keszybz removed the ci-fails/needs-rework 🔥 Please rework this, the CI noticed an issue with the PR label Sep 10, 2020

poettering requested changes Sep 10, 2020

View reviewed changes

poettering added the reviewed/needs-rework 🔨 PR has been reviewed and needs another round of reworks label Sep 10, 2020

yuwata reviewed Sep 15, 2020

View reviewed changes

src/home/homework.c Outdated Show resolved Hide resolved

keszybz added 10 commits September 15, 2020 11:52

Move test_password_{one,many} to libcrypt-util.c

2ae297f

They are only used under src/home/, but I want to add tests in test-libcrypt-util.c. And the functions are almost trivial, so I think it is OK to move them to shared.

Add reciprocal test for password hashing and checking

6016b4b

Make test_password_{one,many} also use crypt_ra()

999b49c

shared/libcrypt-util: include fewer headers

83764f8

Now that we wrap crypt_r/ra uses, we can include the header only in libcrypt-util.c.

meson: test if we have libcrypt_ra

e8644a4

We always seem to have either libcrypt_r and not the other two, or all three. So the fallback for libcrypt_ra needs to be based on libcrypt_r.

shared/libcrypt-util: add fallback for crypt_ra()

9de324c

Following the style in missing_syscall.h, we use a non-conflicting name for the function and use a macro to map to the real name to the replacement.

shared/libcrypt-util: do not refuse passwords if some other hash is u…

35e2282

…nsupported

test-libcrypt-util: skip test on ppc64 with no xcrypt

bd99297

I'm tired of trying to figure this out.

keszybz force-pushed the use-crypt_ra branch from 961afd4 to 5d3fe6f Compare September 15, 2020 09:54

poettering merged commit 3f440b1 into systemd:master Sep 18, 2020

keszybz deleted the use-crypt_ra branch September 20, 2020 08:46

bluca reviewed Sep 24, 2020

View reviewed changes

bluca mentioned this pull request Sep 25, 2020

libcrypt-util: use build-time check for crypt_preferred_method #17163

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use crypt_ra to allocate scratch area for password hashing #16981

Use crypt_ra to allocate scratch area for password hashing #16981

keszybz commented Sep 8, 2020

poettering left a comment

keszybz commented Sep 8, 2020

keszybz commented Sep 8, 2020

poettering commented Sep 8, 2020

poettering commented Sep 8, 2020

keszybz commented Sep 8, 2020

poettering Sep 8, 2020

keszybz Sep 10, 2020

mrc0mmand Sep 10, 2020 •

edited

Loading

poettering Sep 10, 2020

keszybz commented Sep 10, 2020

poettering commented Sep 10, 2020

poettering commented Sep 10, 2020

poettering commented Sep 10, 2020

poettering Sep 10, 2020

keszybz Sep 10, 2020

poettering Sep 10, 2020

poettering Sep 10, 2020

mrc0mmand commented Sep 10, 2020 •

edited

Loading

keszybz commented Sep 15, 2020

mrc0mmand commented Sep 15, 2020

keszybz commented Sep 15, 2020

mrc0mmand commented Sep 15, 2020

mrc0mmand commented Sep 15, 2020

keszybz commented Sep 15, 2020

poettering commented Sep 18, 2020

keszybz commented Sep 18, 2020

bluca Sep 24, 2020

bluca Sep 25, 2020

Use crypt_ra to allocate scratch area for password hashing #16981

Use crypt_ra to allocate scratch area for password hashing #16981

Conversation

keszybz commented Sep 8, 2020

poettering left a comment

Choose a reason for hiding this comment

keszybz commented Sep 8, 2020

keszybz commented Sep 8, 2020

poettering commented Sep 8, 2020

poettering commented Sep 8, 2020

keszybz commented Sep 8, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mrc0mmand Sep 10, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

keszybz commented Sep 10, 2020

poettering commented Sep 10, 2020

poettering commented Sep 10, 2020

poettering commented Sep 10, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mrc0mmand commented Sep 10, 2020 • edited Loading

keszybz commented Sep 15, 2020

mrc0mmand commented Sep 15, 2020

keszybz commented Sep 15, 2020

mrc0mmand commented Sep 15, 2020

mrc0mmand commented Sep 15, 2020

keszybz commented Sep 15, 2020

poettering commented Sep 18, 2020

keszybz commented Sep 18, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mrc0mmand Sep 10, 2020 •

edited

Loading

mrc0mmand commented Sep 10, 2020 •

edited

Loading