1#!./perl 2 3# If a read or write is interrupted by a signal, Perl will call the 4# signal handler and then attempt to restart the call. If the handler does 5# something nasty like close the handle or pop layers, make sure that the 6# read/write handles this gracefully (for some definition of 'graceful': 7# principally, don't segfault). 8 9BEGIN { 10 chdir 't' if -d 't'; 11 require './test.pl'; 12 set_up_inc('../lib'); 13 skip_all_without_dynamic_extension('Fcntl'); 14} 15 16use warnings; 17use strict; 18use Config; 19 20my $piped; 21eval { 22 pipe my $in, my $out; 23 $piped = 1; 24}; 25if (!$piped) { 26 skip_all('pipe not implemented'); 27 exit 0; 28} 29unless (exists $Config{'d_alarm'}) { 30 skip_all('alarm not implemented'); 31 exit 0; 32} 33 34# XXX for some reason the stdio layer doesn't seem to interrupt 35# write system call when the alarm triggers. This makes the tests 36# hang. 37 38if (exists $ENV{PERLIO} && $ENV{PERLIO} =~ /stdio/ ) { 39 skip_all('stdio not supported for this script'); 40 exit 0; 41} 42 43# on Win32, alarm() won't interrupt the read/write call. 44# Similar issues with VMS. 45# On FreeBSD, writes to pipes of 8192 bytes or more use a mechanism 46# that is not interruptible (see perl #85842 and #84688). 47# "close during print" also hangs on Solaris 8 (but not 10 or 11). 48# 49# Also skip on release builds, to avoid other possibly problematic 50# platforms 51 52my ($osmajmin) = $Config{osvers} =~ /^(\d+\.\d+)/; 53if ($^O eq 'VMS' || $^O eq 'MSWin32' || $^O eq 'cygwin' || $^O =~ /freebsd/ || $^O eq 'midnightbsd' || 54 ($^O eq 'solaris' && $Config{osvers} eq '2.8') || $^O eq 'nto' || 55 ($^O eq 'darwin' && $osmajmin < 9) || 56 ((int($]*1000) & 1) == 0) 57) { 58 skip_all('various portability issues'); 59 exit 0; 60} 61 62 63 64my ($in, $out, $st, $sigst, $buf); 65 66plan(tests => 10); 67 68 69# make two handles that will always block 70 71sub fresh_io { 72 close $in if $in; close $out if $out; 73 undef $in; undef $out; # use fresh handles each time 74 pipe $in, $out; 75 $sigst = ""; 76} 77 78$SIG{PIPE} = 'IGNORE'; 79 80# close during read 81 82fresh_io; 83$SIG{ALRM} = sub { $sigst = close($in) ? "ok" : "nok" }; 84alarm(1); 85$st = read($in, $buf, 1); 86alarm(0); 87is($sigst, 'ok', 'read/close: sig handler close status'); 88ok(!$st, 'read/close: read status'); 89ok(!close($in), 'read/close: close status'); 90 91# die during read 92 93fresh_io; 94$SIG{ALRM} = sub { die }; 95alarm(1); 96$st = eval { read($in, $buf, 1) }; 97alarm(0); 98ok(!$st, 'read/die: read status'); 99ok(close($in), 'read/die: close status'); 100 101SKIP: { 102 skip "Tests hang on older versions of Darwin", 5 103 if $^O eq 'darwin' && $osmajmin < 16; 104 105 # This used to be 1_000_000, but on Linux/ppc64 (POWER7) this kept 106 # consistently failing. At exactly 0x100000 it started passing 107 # again. Now we're asking the kernel what the pipe buffer is, and if 108 # that fails, hoping this number is bigger than any pipe buffer. 109 my $surely_this_arbitrary_number_is_fine = (eval { 110 use Fcntl qw(F_GETPIPE_SZ); 111 fcntl($out, F_GETPIPE_SZ, 0); 112 } || 0xfffff) + 1; 113 114 # close during print 115 116 fresh_io; 117 $SIG{ALRM} = sub { $sigst = close($out) ? "ok" : "nok" }; 118 $buf = "a" x $surely_this_arbitrary_number_is_fine . "\n"; 119 select $out; $| = 1; select STDOUT; 120 alarm(1); 121 $st = print $out $buf; 122 alarm(0); 123 is($sigst, 'nok', 'print/close: sig handler close status'); 124 ok(!$st, 'print/close: print status'); 125 ok(!close($out), 'print/close: close status'); 126 127 # die during print 128 129 fresh_io; 130 $SIG{ALRM} = sub { die }; 131 $buf = "a" x $surely_this_arbitrary_number_is_fine . "\n"; 132 select $out; $| = 1; select STDOUT; 133 alarm(1); 134 $st = eval { print $out $buf }; 135 alarm(0); 136 ok(!$st, 'print/die: print status'); 137 # the close will hang since there's data to flush, so use alarm 138 alarm(1); 139 ok(!eval {close($out)}, 'print/die: close status'); 140 alarm(0); 141 142 # close during close 143 144 # Apparently there's nothing in standard Linux that can cause an 145 # EINTR in close(2); but run the code below just in case it does on some 146 # platform, just to see if it segfaults. 147 fresh_io; 148 $SIG{ALRM} = sub { $sigst = close($in) ? "ok" : "nok" }; 149 alarm(1); 150 close $in; 151 alarm(0); 152 153 # die during close 154 155 fresh_io; 156 $SIG{ALRM} = sub { die }; 157 alarm(1); 158 eval { close $in }; 159 alarm(0); 160} 161 162# vim: ts=4 sts=4 sw=4: 163