* [PATCH 0/4] httpd/async fixes for coderepos
@ 2022-12-23 11:05 Eric Wong
2022-12-23 11:05 ` [PATCH 1/4] httpd: avoid crash on cgit -> coderepo 404 fallback Eric Wong
` (3 more replies)
0 siblings, 4 replies; 5+ messages in thread
From: Eric Wong @ 2022-12-23 11:05 UTC (permalink / raw)
To: meta
1/4 is a critical fix. The rest are minor quality-of-life
things I noticed along the way.
Eric Wong (4):
httpd: avoid crash on cgit -> coderepo 404 fallback
httpd/async: remove useless undef
qspawn: shorten life of {hdr_buf} in generic code path
httpd/async + qspawn: rename {fh} fields
lib/PublicInbox/GitHTTPBackend.pm | 6 ++++--
lib/PublicInbox/HTTPD/Async.pm | 23 +++++++++++------------
lib/PublicInbox/Qspawn.pm | 8 ++++----
3 files changed, 19 insertions(+), 18 deletions(-)
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH 1/4] httpd: avoid crash on cgit -> coderepo 404 fallback
2022-12-23 11:05 [PATCH 0/4] httpd/async fixes for coderepos Eric Wong
@ 2022-12-23 11:05 ` Eric Wong
2022-12-23 11:05 ` [PATCH 2/4] httpd/async: remove useless undef Eric Wong
` (2 subsequent siblings)
3 siblings, 0 replies; 5+ messages in thread
From: Eric Wong @ 2022-12-23 11:05 UTC (permalink / raw)
To: meta
A trickled cgit response can cause HTTPD::Async->event_step to
fire an extra time after header parsing. We need to account for
the lack of async_pass call populating ->{fh} and ->{http} in
that case and avoid calling $self->{fh}->write when there's
no {fh}.
---
lib/PublicInbox/GitHTTPBackend.pm | 6 ++++--
lib/PublicInbox/HTTPD/Async.pm | 3 ++-
2 files changed, 6 insertions(+), 3 deletions(-)
diff --git a/lib/PublicInbox/GitHTTPBackend.pm b/lib/PublicInbox/GitHTTPBackend.pm
index 1eb51f27..974a1831 100644
--- a/lib/PublicInbox/GitHTTPBackend.pm
+++ b/lib/PublicInbox/GitHTTPBackend.pm
@@ -128,7 +128,7 @@ sub input_prepare {
{ 0 => $in };
}
-sub parse_cgi_headers {
+sub parse_cgi_headers { # {parse_hdr} for Qspawn
my ($r, $bref, $ctx) = @_;
return r(500) unless defined $r && $r >= 0;
$$bref =~ s/\A(.*?)\r?\n\r?\n//s or return $r == 0 ? r(500) : undef;
@@ -145,7 +145,9 @@ sub parse_cgi_headers {
}
# fallback to WwwCoderepo if cgit 404s. Duplicating $ctx prevents
- # ->finalize from the current Qspawn from using qspawn.wcb
+ # ->finalize from the current Qspawn from using qspawn.wcb.
+ # This makes qspawn skip ->async_pass and causes
+ # PublicInbox::HTTPD::Async::event_step to close shortly after
if ($code == 404 && $ctx->{www} && !$ctx->{_coderepo_tried}++) {
my %ctx = %$ctx;
$ctx{env} = +{ %{$ctx->{env}} };
diff --git a/lib/PublicInbox/HTTPD/Async.pm b/lib/PublicInbox/HTTPD/Async.pm
index cb76cfab..9e592f47 100644
--- a/lib/PublicInbox/HTTPD/Async.pm
+++ b/lib/PublicInbox/HTTPD/Async.pm
@@ -48,7 +48,8 @@ sub event_step {
# this may call async_pass when headers are done
$cb->(my $refcnt_guard = delete $self->{arg});
} elsif (my $sock = $self->{sock}) {
- my $http = $self->{http};
+ # $http may be undef if discarding body output from cgit on 404
+ my $http = $self->{http} or return $self->close;
# $self->{sock} is a read pipe for git-http-backend or cgit
# and 65536 is the default Linux pipe size
my $r = sysread($sock, my $buf, 65536);
^ permalink raw reply related [flat|nested] 5+ messages in thread
* [PATCH 2/4] httpd/async: remove useless undef
2022-12-23 11:05 [PATCH 0/4] httpd/async fixes for coderepos Eric Wong
2022-12-23 11:05 ` [PATCH 1/4] httpd: avoid crash on cgit -> coderepo 404 fallback Eric Wong
@ 2022-12-23 11:05 ` Eric Wong
2022-12-23 11:05 ` [PATCH 3/4] qspawn: shorten life of {hdr_buf} in generic code path Eric Wong
2022-12-23 11:05 ` [PATCH 4/4] httpd/async + qspawn: rename {fh} fields Eric Wong
3 siblings, 0 replies; 5+ messages in thread
From: Eric Wong @ 2022-12-23 11:05 UTC (permalink / raw)
To: meta
Assigning `undef' to a scalar doesn't free it's memory,
we need to call `undef($var)' in the caller. It's also
been pointless since we simplified ->async_pass in commit
b7fbffd1f8c12556 (httpd/async: get rid of ephemeral main_cb, 2019-12-25)
---
lib/PublicInbox/HTTPD/Async.pm | 4 ----
1 file changed, 4 deletions(-)
diff --git a/lib/PublicInbox/HTTPD/Async.pm b/lib/PublicInbox/HTTPD/Async.pm
index 9e592f47..75b3bd50 100644
--- a/lib/PublicInbox/HTTPD/Async.pm
+++ b/lib/PublicInbox/HTTPD/Async.pm
@@ -84,10 +84,6 @@ sub async_pass {
# *_wcb methods respond to ->write (and ->close), not ->print
$fh->write($$bref);
- # we're done with this, free this memory up ASAP since the
- # calls after this may use much memory:
- $$bref = undef;
-
$self->{http} = $http;
$self->{fh} = $fh;
}
^ permalink raw reply related [flat|nested] 5+ messages in thread
* [PATCH 3/4] qspawn: shorten life of {hdr_buf} in generic code path
2022-12-23 11:05 [PATCH 0/4] httpd/async fixes for coderepos Eric Wong
2022-12-23 11:05 ` [PATCH 1/4] httpd: avoid crash on cgit -> coderepo 404 fallback Eric Wong
2022-12-23 11:05 ` [PATCH 2/4] httpd/async: remove useless undef Eric Wong
@ 2022-12-23 11:05 ` Eric Wong
2022-12-23 11:05 ` [PATCH 4/4] httpd/async + qspawn: rename {fh} fields Eric Wong
3 siblings, 0 replies; 5+ messages in thread
From: Eric Wong @ 2022-12-23 11:05 UTC (permalink / raw)
To: meta
No point in keeping the old buffer around if we don't need to.
---
lib/PublicInbox/Qspawn.pm | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/lib/PublicInbox/Qspawn.pm b/lib/PublicInbox/Qspawn.pm
index ef9db43e..da7cd74f 100644
--- a/lib/PublicInbox/Qspawn.pm
+++ b/lib/PublicInbox/Qspawn.pm
@@ -247,9 +247,9 @@ sub psgi_return_init_cb {
delete($self->{hdr_buf}));
} else { # for synchronous PSGI servers
require PublicInbox::GetlineBody;
+ my $buf = delete $self->{hdr_buf};
$r->[2] = PublicInbox::GetlineBody->new($self->{rpipe},
- \&event_step, $self,
- ${$self->{hdr_buf}}, $filter);
+ \&event_step, $self, $$buf, $filter);
$wcb->($r);
}
}
^ permalink raw reply related [flat|nested] 5+ messages in thread
* [PATCH 4/4] httpd/async + qspawn: rename {fh} fields
2022-12-23 11:05 [PATCH 0/4] httpd/async fixes for coderepos Eric Wong
` (2 preceding siblings ...)
2022-12-23 11:05 ` [PATCH 3/4] qspawn: shorten life of {hdr_buf} in generic code path Eric Wong
@ 2022-12-23 11:05 ` Eric Wong
3 siblings, 0 replies; 5+ messages in thread
From: Eric Wong @ 2022-12-23 11:05 UTC (permalink / raw)
To: meta
Use more unique names within the project to minimize confusion
since these packages interact quite a bit and using identical
names leads to needless confusion.
---
lib/PublicInbox/HTTPD/Async.pm | 16 +++++++++-------
lib/PublicInbox/Qspawn.pm | 4 ++--
2 files changed, 11 insertions(+), 9 deletions(-)
diff --git a/lib/PublicInbox/HTTPD/Async.pm b/lib/PublicInbox/HTTPD/Async.pm
index 75b3bd50..e03daafa 100644
--- a/lib/PublicInbox/HTTPD/Async.pm
+++ b/lib/PublicInbox/HTTPD/Async.pm
@@ -37,7 +37,7 @@ sub new {
arg => $arg, # arg for $cb
end_obj => $end_obj, # like END{}, can ->event_step
}, $class;
- my $pp = tied *$io;
+ my $pp = tied *$io; # ProcessPipe
$pp->{fh}->blocking(0) // die "$io->blocking(0): $!";
$self->SUPER::new($io, EPOLLIN);
}
@@ -54,7 +54,7 @@ sub event_step {
# and 65536 is the default Linux pipe size
my $r = sysread($sock, my $buf, 65536);
if ($r) {
- $self->{fh}->write($buf); # may call $http->close
+ $self->{ofh}->write($buf); # may call $http->close
# let other clients get some work done, too
return if $http->{sock}; # !closed
@@ -63,7 +63,7 @@ sub event_step {
return; # EPOLLIN means we'll be notified
}
- # Done! Error handling will happen in $self->{fh}->close
+ # Done! Error handling will happen in $self->{ofh}->close
# called by end_obj->event_step handler
delete $http->{forward};
$self->close; # queues end_obj->event_step to be called
@@ -71,9 +71,11 @@ sub event_step {
}
# once this is called, all data we read is passed to the
-# to the PublicInbox::HTTP instance ($http) via $fh->write
+# to the PublicInbox::HTTP instance ($http) via $ofh->write
+# $ofh is typically PublicInbox::HTTP::{Chunked,Identity}, but
+# may be PublicInbox::GzipFilter or $PublicInbox::Qspawn::qx_fh
sub async_pass {
- my ($self, $http, $fh, $bref) = @_;
+ my ($self, $http, $ofh, $bref) = @_;
# In case the client HTTP connection ($http) dies, it
# will automatically close this ($self) object.
$http->{forward} = $self;
@@ -82,10 +84,10 @@ sub async_pass {
# This is typically PublicInbox:HTTP::{chunked,identity}_wcb,
# but may be PublicInbox::GzipFilter::write. PSGI requires
# *_wcb methods respond to ->write (and ->close), not ->print
- $fh->write($$bref);
+ $ofh->write($$bref);
$self->{http} = $http;
- $self->{fh} = $fh;
+ $self->{ofh} = $ofh;
}
# may be called as $forward->close in PublicInbox::HTTP or EOF (event_step)
diff --git a/lib/PublicInbox/Qspawn.pm b/lib/PublicInbox/Qspawn.pm
index da7cd74f..6e245389 100644
--- a/lib/PublicInbox/Qspawn.pm
+++ b/lib/PublicInbox/Qspawn.pm
@@ -187,7 +187,7 @@ sub event_step {
my ($self, $err) = @_; # $err: $!
warn "psgi_{return,qx} $err" if defined($err);
finish($self);
- my ($fh, $qx_fh) = delete(@$self{qw(fh qx_fh)});
+ my ($fh, $qx_fh) = delete(@$self{qw(qfh qx_fh)});
$fh->close if $fh; # async-only (psgi_return)
}
@@ -242,7 +242,7 @@ sub psgi_return_init_cb {
# done reading headers, handoff to read body
my $fh = $wcb->($r); # scalar @$r == 2
$fh = $filter->attach($fh) if $filter;
- $self->{fh} = $fh;
+ $self->{qfh} = $fh;
$async->async_pass($env->{'psgix.io'}, $fh,
delete($self->{hdr_buf}));
} else { # for synchronous PSGI servers
^ permalink raw reply related [flat|nested] 5+ messages in thread
end of thread, other threads:[~2022-12-23 11:05 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-12-23 11:05 [PATCH 0/4] httpd/async fixes for coderepos Eric Wong
2022-12-23 11:05 ` [PATCH 1/4] httpd: avoid crash on cgit -> coderepo 404 fallback Eric Wong
2022-12-23 11:05 ` [PATCH 2/4] httpd/async: remove useless undef Eric Wong
2022-12-23 11:05 ` [PATCH 3/4] qspawn: shorten life of {hdr_buf} in generic code path Eric Wong
2022-12-23 11:05 ` [PATCH 4/4] httpd/async + qspawn: rename {fh} fields Eric Wong
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).