1 # Copyright (C) 2020-2021 all contributors <meta@public-inbox.org>
2 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
4 # In public-inbox <=1.5.0, public-inbox-httpd favored "getline"
5 # response bodies to take a "pull"-based approach to feeding
6 # slow clients (as opposed to a more common "push" model).
8 # In newer versions, public-inbox-httpd supports a backpressure-aware
9 # pull/push model which also accounts for slow git blob storage.
10 # async_next callbacks only run when the DS {wbuf} is drained
11 # async_eml callbacks only run when a blob arrives from git.
13 # We continue to support getline+close for generic PSGI servers.
14 package PublicInbox::GzipFilter;
16 use parent qw(Exporter);
17 use Compress::Raw::Zlib qw(Z_OK);
18 use PublicInbox::CompressNoop;
20 use PublicInbox::GitAsyncCat;
22 our @EXPORT_OK = qw(gzf_maybe);
23 my %OPT = (-WindowBits => 15 + 16, -AppendOutput => 1);
24 my @GZIP_HDRS = qw(Vary Accept-Encoding Content-Encoding gzip);
26 sub new { bless {}, shift } # qspawn filter
28 # for Qspawn if using $env->{'pi-httpd.async'}
30 my ($self, $http_out) = @_;
31 $self->{http_out} = $http_out; # PublicInbox::HTTP::{Chunked,Identity}
36 my ($res_hdr, $env) = @_;
37 if (($env->{HTTP_ACCEPT_ENCODING} // '') =~ /\bgzip\b/) {
38 $env->{'plack.skip-deflater'} = 1;
39 push @$res_hdr, @GZIP_HDRS;
42 PublicInbox::CompressNoop::new();
46 sub gzf_maybe ($$) { bless { gz => gz_or_noop(@_) }, __PACKAGE__ }
49 # $code may be an HTTP response code (e.g. 200) or a CODE ref (mbox_hdr)
50 my ($self, $code, $res_hdr) = @_;
51 if ($self->{env}->{'pi-httpd.async'}) {
52 my $http = $self->{env}->{'psgix.io'}; # PublicInbox::HTTP
53 $http->{forward} = $self;
55 my ($wcb) = @_; # -httpd provided write callback
56 $self->{wcb_args} = [ $code, $res_hdr, $wcb ];
57 $self->can('async_next')->($http); # start stepping
59 } else { # generic PSGI code path
60 ref($code) eq 'CODE' and
61 ($code, $res_hdr) = @{$code->($self)};
62 $self->{gz} //= gz_or_noop($res_hdr, $self->{env});
63 [ $code, $res_hdr, $self ];
68 my ($res_hdr, $env) = @_;
69 return if ($env->{HTTP_ACCEPT_ENCODING} // '') !~ /\bgzip\b/;
70 my $hdr = join("\n", @$res_hdr);
71 return if $hdr !~ m!^Content-Type\n
72 (?:(?:text/(?:html|plain))|
73 application/atom\+xml)\b!ixsm;
74 return if $hdr =~ m!^Content-Encoding\ngzip\n!smi;
75 return if $hdr =~ m!^Content-Length\n[0-9]+\n!smi;
76 return if $hdr =~ m!^Transfer-Encoding\n!smi;
77 # in case Plack::Middleware::Deflater is loaded:
78 return if $env->{'plack.skip-deflater'}++;
79 push @$res_hdr, @GZIP_HDRS;
80 bless {}, __PACKAGE__;
84 my ($gz, $err) = Compress::Raw::Zlib::Deflate->new(%OPT);
85 $err == Z_OK or die "Deflate->new failed: $err";
89 sub gone { # what: search/over/mm
90 my ($ctx, $what) = @_;
91 warn "W: `$ctx->{ibx}->{name}' $what went away unexpectedly\n";
95 # for GetlineBody (via Qspawn) when NOT using $env->{'pi-httpd.async'}
96 # Also used for ->getline callbacks
98 my $self = $_[0]; # $_[1] => input
100 # allocate the zlib context lazily here, instead of in ->new.
101 # Deflate contexts are memory-intensive and this object may
102 # be sitting in the Qspawn limiter queue for a while.
103 my $gz = $self->{gz} //= gzip_or_die();
104 my $zbuf = delete($self->{zbuf});
105 if (defined $_[1]) { # my $buf = $_[1];
106 my $err = $gz->deflate($_[1], $zbuf);
107 die "gzip->deflate: $err" if $err != Z_OK;
108 return $zbuf if length($zbuf) >= 8192;
110 $self->{zbuf} = $zbuf;
112 } else { # undef == EOF
113 my $err = $gz->flush($zbuf);
114 die "gzip->flush: $err" if $err != Z_OK;
121 $self->{http_out} // do {
122 my $args = delete $self->{wcb_args} // return undef;
123 my $wcb = pop @$args; # from PublicInbox:HTTP async
124 # $args->[0] may be \&mbox_hdr or similar
125 $args = $args->[0]->($self) if ref($args->[0]) eq 'CODE';
126 $self->{gz} //= gz_or_noop($args->[1], $self->{env});
127 $self->{http_out} = $wcb->($args); # $wcb->([$code, $hdr_ary])
132 # my $ret = bytes::length($_[1]); # XXX does anybody care?
133 http_out($_[0])->write(translate($_[0], $_[1]));
136 # similar to ->translate; use this when we're sure we know we have
137 # more data to buffer after this
139 my $self = $_[0]; # $_[1] => input
141 my $err = $self->{gz}->deflate($_[1], $self->{zbuf});
142 die "gzip->deflate: $err" if $err != Z_OK;
146 # flushes and returns the final bit of gzipped data
148 my $self = $_[0]; # $_[1] => final input (optional)
149 my $zbuf = delete $self->{zbuf};
150 my $gz = delete $self->{gz};
153 $err = $gz->deflate($_[1], $zbuf);
154 die "gzip->deflate: $err" if $err != Z_OK;
156 $err = $gz->flush($zbuf);
157 die "gzip->flush: $err" if $err != Z_OK;
163 my $http_out = http_out($self) // return;
164 $http_out->write(zflush($self));
165 delete($self->{http_out})->close;
170 if (my $env = $self->{env}) {
172 my $http = $env->{'psgix.io'} or return; # client abort
173 eval { $http->close }; # should hit our close
174 warn "E: error in http->close: $@" if $@;
175 eval { $self->close }; # just in case...
176 warn "E: error in self->close: $@" if $@;
182 # this is public-inbox-httpd-specific
183 sub async_blob_cb { # git->cat_async callback
184 my ($bref, $oid, $type, $size, $self) = @_;
185 my $http = $self->{env}->{'psgix.io'};
186 $http->{forward} or return; # client aborted
187 my $smsg = $self->{smsg} or bail($self, 'BUG: no smsg');
188 if (!defined($oid)) {
189 # it's possible to have TOCTOU if an admin runs
190 # public-inbox-(edit|purge), just move onto the next message
191 warn "E: $smsg->{blob} missing in $self->{ibx}->{inboxdir}\n";
192 return $http->next_step($self->can('async_next'));
194 $smsg->{blob} eq $oid or bail($self, "BUG: $smsg->{blob} != $oid");
195 eval { $self->async_eml(PublicInbox::Eml->new($bref)) };
196 bail($self, "E: async_eml: $@") if $@;
197 if ($self->{-low_prio}) {
198 push(@{$self->{www}->{-low_prio_q}}, $self) == 1 and
199 PublicInbox::DS::requeue($self->{www});
201 $http->next_step($self->can('async_next'));
206 my ($self, $smsg) = @_;
207 ibx_async_cat($self->{ibx}, $smsg->{blob}, \&async_blob_cb, $self);