Re: Missing 'libubus.so' and change of ABI in 19.07.8
On 13/11/2021 18:47, Jo-Philipp Wich wrote: But I really can't see the breaking change, could you please point it out for me? struct ubus_client changed its layout. Ok but isn't that affecting only the binary executable 'ubusd'? I don't see 'libubus.so' including 'ubusd.h' even indirectly... Which makes sense as that struct is for managing a server-to-client, while the shared library only manages client-to-server connections? ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Missing 'libubus.so' and change of ABI in 19.07.8
Greetings. I have two questions about a recent change related to ubus package in 19.07.8. The first question is related to the ABI version change from 20191227 to 20210603. The changeset in the ubus repository the following: git diff 041c9d1..38c7fdd CMakeLists.txt | 4 lua/ubus.c | 5 - ubusd.c| 25 + ubusd.h| 13 ++--- ubusd_main.c | 37 + ubusd_proto.c | 1 + But I really can't see the breaking change, could you please point it out for me? The second question is the change of the library name from 'libubus.so' to 'libubus.so.20210603'. Was this intended? Could we at least add the symlink 'libubus.so'? My take is that both of these changes, together with unsharing the ubus package, were really unnecessary in a stable branch nearing its end of life. Thank you for insights and kind regards. ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
[PATCH ubox] logread: fix erroneous message "Logread connected to" when using 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 streaming to..." is logged.
Also fixed capital letter for "failed to send" message.
Signed-off-by: Giovanni Giacobbi
---
log/logread.c | 10 --
1 file changed, 8 insertions(+), 2 deletions(-)
diff --git a/log/logread.c b/log/logread.c
index a764742..dc00c43 100644
--- a/log/logread.c
+++ b/log/logread.c
@@ -80,41 +80,47 @@ static int check_facility_filter(int f)
}
static const char* getcodetext(int value, CODE *codetable) {
CODE *i;
if (value >= 0)
for (i = codetable; i->c_val != -1; i++)
if (i->c_val == value)
return (i->c_name);
return "";
};
static void log_handle_reconnect(struct uloop_timeout *timeout)
{
sender.fd = usock((log_udp) ? (USOCK_UDP) : (USOCK_TCP), log_ip,
log_port);
if (sender.fd < 0) {
fprintf(stderr, "failed to connect: %m\n");
uloop_timeout_set(&retry, 1000);
} else {
uloop_fd_add(&sender, ULOOP_READ);
- syslog(LOG_INFO, "Logread connected to %s:%s\n", log_ip,
log_port);
+ if (!log_udp) {
+ syslog(LOG_INFO, "Logread connected to %s:%s via
tcp\n", log_ip, log_port);
+ }
+ else if (log_udp == 1) {
+ syslog(LOG_INFO, "Logread streaming to %s:%s via
udp\n", log_ip, log_port);
+ log_udp = 2;
+ }
}
}
static void log_handle_fd(struct uloop_fd *u, unsigned int events)
{
if (u->eof) {
uloop_fd_delete(u);
close(sender.fd);
sender.fd = -1;
uloop_timeout_set(&retry, 1000);
}
}
static int log_notify(struct blob_attr *msg)
{
struct blob_attr *tb[__LOG_MAX];
struct stat s;
char buf[LOG_LINE_SIZE + 128];
char buf_ts[32];
uint32_t p;
@@ -175,41 +181,41 @@ static int log_notify(struct blob_attr *msg)
strncat(buf, hostname, sizeof(buf) - strlen(buf) - 1);
strncat(buf, " ", sizeof(buf) - strlen(buf) - 1);
}
if (log_prefix) {
strncat(buf, log_prefix, sizeof(buf) - strlen(buf) - 1);
strncat(buf, ": ", sizeof(buf) - strlen(buf) - 1);
}
if (blobmsg_get_u32(tb[LOG_SOURCE]) == SOURCE_KLOG)
strncat(buf, "kernel: ", sizeof(buf) - strlen(buf) - 1);
strncat(buf, m, sizeof(buf) - strlen(buf) - 1);
if (log_udp)
err = write(sender.fd, buf, strlen(buf));
else {
size_t buflen = strlen(buf);
if (!log_trailer_null)
buf[buflen] = '\n';
err = send(sender.fd, buf, buflen + 1, 0);
}
if (err < 0) {
- syslog(LOG_INFO, "failed to send log data to %s:%s via
%s\n",
+ syslog(LOG_INFO, "Failed to send log data to %s:%s via
%s\n",
log_ip, log_port, (log_udp) ? ("udp") :
("tcp"));
uloop_fd_delete(&sender);
close(sender.fd);
sender.fd = -1;
uloop_timeout_set(&retry, 1000);
}
} else {
snprintf(buf, sizeof(buf), "%s %s%s.%s%s %s\n",
c, log_timestamp ? buf_ts : "",
getcodetext(LOG_FAC(p) << 3, facilitynames),
getcodetext(LOG_PRI(p), prioritynames),
(blobmsg_get_u32(tb[LOG_SOURCE])) ? ("") : ("
kernel:"), m);
ret = write(sender.fd, buf, strlen(buf));
}
if (log_type == LOG_FILE)
fsync(sender.fd);
return ret;
}
--
2.17.2
___
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [PATCH ubox] logd: fix priviledge dropping order
On 27/07/2021 14:39, Giovanni Giacobbi wrote:
Fixes: 41664054b8b1 ("logd: fix ignored return values in set{gid,uid}")
Signed-off-by: Giovanni Giacobbi
---
log/logd.c | 8
1 file changed, 4 insertions(+), 4 deletions(-)
Let me add some remarks about this change. Comments on each point are
welcome.
1) The way it is now, the daemon is completely broken as it cannot
setgid(2) after setuid(2) to a non-zero uid. This is the minimal fix to
this situation. The bug exists already in 21.02 and master but the
setgid is failing silently (so the egid remains zero).
2) I'm not sure that exit() is the right response in case of failure.
Given the role of this daemon I would consider letting it run as root to
avoid ending up without a syslog.
3) Currently the privilege de-escalation is activated by the sole
presence of the "logd" user. If it doesn't exist, it silently keeps
running as root, while if the user is there but setuid/setgid fails, it
exits. I personally don't like this asymmetric behaviour, i'd rather
have it print a warning (and maybe self-log it to itself as log
critical?) and continue running as root. As an alternative, I would
propose to add a command line switch to indicate the user to switch to,
and then fail if user not found or setgid/setuid fail, so that you know
for sure once you specify the command line switch that either privilege
is dropped, or it doesn't start. Not using the command line switch would
NOT attempt to switch user even if "logd" exists. Ideas? Should I submit
this?
P.S. for the maintainer: If you commit this patch, please fix the typo
in the subject, thx. "logd: fix privilege dropping order"
___
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/mailman/listinfo/openwrt-devel
[PATCH ubox] logd: fix priviledge dropping order
Fixes: 41664054b8b1 ("logd: fix ignored return values in set{gid,uid}")
Signed-off-by: Giovanni Giacobbi
---
log/logd.c | 8
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/log/logd.c b/log/logd.c
index 5d6c458..594b1e0 100644
--- a/log/logd.c
+++ b/log/logd.c
@@ -243,37 +243,37 @@ main(int argc, char **argv)
struct passwd *p = NULL;
signal(SIGPIPE, SIG_IGN);
while ((ch = getopt(argc, argv, "S:")) != -1) {
switch (ch) {
case 'S':
log_size = atoi(optarg);
if (log_size < 1)
log_size = 16;
break;
}
}
log_size *= 1024;
uloop_init();
log_init(log_size);
conn.cb = ubus_connect_handler;
ubus_auto_connect(&conn);
p = getpwnam("logd");
if (p) {
- if (setuid(p->pw_uid) < 0) {
- fprintf(stderr, "setuid() failed: %s\n",
strerror(errno));
+ if (setgid(p->pw_gid) < 0) {
+ fprintf(stderr, "setgid() failed: %s\n",
strerror(errno));
exit(1);
}
- if (setgid(p->pw_gid) < 0) {
- fprintf(stderr, "setgid() failed: %s\n",
strerror(errno));
+ if (setuid(p->pw_uid) < 0) {
+ fprintf(stderr, "setuid() failed: %s\n",
strerror(errno));
exit(1);
}
}
uloop_run();
log_shutdown();
uloop_done();
ubus_auto_shutdown(&conn);
return 0;
}
--
2.17.2
___
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [PATCH 19.07 4/4] treewide: mark selected packages nonshared
On 04/07/2021 14:46, Adrian Schmutzler wrote: PKG_NAME:=json-c PKG_VERSION:=0.12.1 -PKG_RELEASE:=3.1 +PKG_RELEASE:=3.2 I've never seen a non integer release, is there a special reason for this? I've also used this as standard scheme for changes in stable branches. The advantage is that you immediately see up to which version the changes are shared with master, and when it starts to deviate. What would you do then if the upstream version is bumped, but the patchset/Makefile would still differ between the master and branch? In this case, you bump PKG_VERSION to 0.12.2, but if you restart the PKG_RELEASE from 1 for both the branch and the master you end up with two different "0.12.2-1" packages in let's say 21.02 and 19.07. ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
[PATCH urngd] Fix busy loop in case of ioctl(RNDADDENTROPY) failure
On systems where adding entropy to /dev/random using ioctl(RNDADDENTROPY) is
not allowed (notably: docker containers), avoid entering a busy loop that
consumes high cpu, as the poll loop will keep firing.
Cc: Petr Štetiar
---
urngd.c | 7 ++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/urngd.c b/urngd.c
index 35ccdec..410d300 100644
--- a/urngd.c
+++ b/urngd.c
@@ -129,9 +129,14 @@ static size_t gather_entropy(struct urngd *u)
static void low_entropy_cb(struct uloop_fd *ufd, unsigned int events)
{
struct urngd *u = container_of(ufd, struct urngd, rnd_fd);
+ size_t res;
DEBUG(2, DEV_RANDOM " signals low entropy\n");
- gather_entropy(u);
+ res = gather_entropy(u);
+ if (!res) {
+ DEBUG(2, "gather_entropy failed, delaying further attempts\n");
+ sleep(60);
+ }
}
static void urngd_done(struct urngd *u)
--
Giovanni Giacobbi
___
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/mailman/listinfo/openwrt-devel
