Skip to contentSkip to conversation

iptables: flush stdout after every verbose log.

PATCH Applied Maciej Żenczykowski 2 participants 7 messages Apr 21, 2020 – May 11, 2020

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 iptables: flush stdout after every verbose log.

From: Maciej Żenczykowski <[email protected]>

Ensures that each logged line is flushed to stdout after it's written, and not held in any buffer.

Places to modify found via:
git grep -C5 'fputs[(]buffer, stdout[)];'

On Android iptables-restore -v is run as netd daemon's child process and fed actions via pipe. '#PING' is used to verify the child is still responsive, and thus needs to be unbuffered.

Luckily if you're running iptables-restore in verbose mode you probably either don't care about performance or - like Android
- actually need this.

Test: builds, required on Android for ip6?tables-restore netd
subprocess health monitoring.

---
 iptables/iptables-restore.c | 4 +++-
 iptables/xtables-restore.c  | 4 +++-
 2 files changed, 6 insertions(+), 2 deletions(-)
iptables/iptables-restore.c +3 -1
index b0a51d49..fea04842 100644
@@ -178,8 +178,10 @@ ip46tables_restore_main(const struct iptables_restore_cb *cb,
 		if (buffer[0] == '\n')
 			continue;
 		else if (buffer[0] == '#') {
-			if (verbose)
+			if (verbose) {
 				fputs(buffer, stdout);
+				fflush(stdout);
+			}
 			continue;
 		} else if ((strcmp(buffer, "COMMIT\n") == 0) && (in_table)) {
 			if (!testing) {
iptables/xtables-restore.c +3 -1
index c472ac9b..8c25e5b2 100644
@@ -85,8 +85,10 @@ static void xtables_restore_parse_line(struct nft_handle *h,
 	if (buffer[0] == '\n')
 		return;
 	else if (buffer[0] == '#') {
-		if (verbose)
+		if (verbose) {
 			fputs(buffer, stdout);
+			fflush(stdout);
+		}
 		return;
 	} else if (state->in_table &&
 		   (strncmp(buffer, "COMMIT", 6) == 0) &&
Signature · 1 line
2.26.1.301.g55bc3eb7cb9-goog
#1 lore raw
Quoted 8 lines · Maciej Żenczykowski

From: Maciej Żenczykowski <[email protected]>

Ensures that each logged line is flushed to stdout after it's written, and not held in any buffer.

Places to modify found via:
git grep -C5 'fputs[(]buffer, stdout[)];'

Quoted 7 lines

On Android iptables-restore -v is run as netd daemon's child process and fed actions via pipe. '#PING' is used to verify the child is still responsive, and thus needs to be unbuffered.

Luckily if you're running iptables-restore in verbose mode you probably either don't care about performance or - like Android
- actually need this.

Could you check if this slows down iptables-restore?

Thank you.

#2 lore raw
Quoted 1 line

Could you check if this slows down iptables-restore?

per the iptables-restore man page
-v, --verbose
Print additional debug info during ruleset processing.

Well, if you run it with verbose mode enabled you probably don't care about performance all that much...
And it only does anything if you're feeding in comments, which again - is unlikely.

iptables-save produces two comment lines per table, so you're likely to have a grand total of 8 of these lines
(if you have raw/nat/mangle/filter tables all listed), and if you feed that in to iptables-restore -v then
you end up calling flush 8 times, which triggers 8 extra write() syscalls - which aren't exactly expensive to begin with...

So I think this is a non issue.

(I could change it so that only "#PING" flushes, which is the actual comment netd sends for healthchecking,
but I just don't think it matters)

#3 lore raw
Quoted 8 lines · Maciej Żenczykowski
Quoted 1 line

Could you check if this slows down iptables-restore?

per the iptables-restore man page
-v, --verbose
Print additional debug info during ruleset processing.

Well, if you run it with verbose mode enabled you probably don't care about performance all that much...

Thanks for explaining.

How long has this been broken? I mean, netd has been there for quite a while interacting with iptables. However, the existing behaviour was not a problem? Or a recent bug?

#4 lore raw

I don't think it's ever been broken per say since Android has it's own copy of everything,
and there are local changes to iptables (that I'm trying to cleanup/drop/upstream),
so we've carried this patch for years.

Quoted 16 lines · Pablo Neira Ayuso
Quoted 8 lines · Maciej Żenczykowski
Quoted 1 line

Could you check if this slows down iptables-restore?

per the iptables-restore man page
-v, --verbose
Print additional debug info during ruleset processing.

Well, if you run it with verbose mode enabled you probably don't care about performance all that much...

Thanks for explaining.

How long has this been broken? I mean, netd has been there for quite a while interacting with iptables. However, the existing behaviour was not a problem? Or a recent bug?

maintainer Applied
#5 lore raw
In reply to Maciej Żenczykowski
Quoted 8 lines · Maciej Żenczykowski

From: Maciej Żenczykowski <[email protected]>

Ensures that each logged line is flushed to stdout after it's written, and not held in any buffer.

Places to modify found via:
git grep -C5 'fputs[(]buffer, stdout[)];'

Quoted 7 lines

On Android iptables-restore -v is run as netd daemon's child process and fed actions via pipe. '#PING' is used to verify the child is still responsive, and thus needs to be unbuffered.

Luckily if you're running iptables-restore in verbose mode you probably either don't care about performance or - like Android
- actually need this.

Applied, thanks for explaning.