package urpm; # $Id$ use strict; use vars qw($VERSION @ISA @EXPORT); use MDK::Common; $VERSION = '4.4'; @ISA = qw(Exporter URPM); @EXPORT = qw(N); use URPM; use URPM::Resolve; use POSIX; #- I18N. eval { require Locale::gettext; setlocale(LC_ALL, ""); Locale::gettext::textdomain("urpmi"); }; sub N { my ($format, @params) = @_; sprintf(eval { Locale::gettext::gettext($format || '') } || $format, @params); } #- tool functions. sub basename { local $_ = shift; s|/*\s*$||; s|.*/||; $_ } sub localtime2changelog { scalar(localtime($_[0])) =~ /(.*) \S+ (\d{4})$/ && "$1 $2" }; #- create a new urpm object. sub new { my ($class) = @_; my $self; $self = bless { config => "/etc/urpmi/urpmi.cfg", skiplist => "/etc/urpmi/skip.list", instlist => "/etc/urpmi/inst.list", statedir => "/var/lib/urpmi", cachedir => "/var/cache/urpmi", media => undef, provides => {}, depslist => [], #- sync: first argument is directory, others are url to fetch. sync => sub { $self->sync_webfetch(@_) }, proxy => get_proxy(), options => {}, fatal => sub { printf STDERR "%s\n", $_[1]; exit($_[0]) }, error => sub { printf STDERR "%s\n", $_[0] }, log => sub { printf STDERR "%s\n", $_[0] }, ui_msg => sub { $self->{log}($_[0]); $self->{ui} and $self->{ui}{msg}->($_[1]) }, }, $class; } sub get_proxy { my $proxy = { http_proxy => undef , ftp_proxy => undef , user => undef, pwd => undef }; local (*F, $_); open F, "/etc/urpmi/proxy.cfg" or return undef; while () { chomp; s/#.*$//; s/^\s*//; s/\s*$//; /^http_proxy\s*=\s*(.*)$/ and $proxy->{http_proxy} = $1, next; /^ftp_proxy\s*=\s*(.*)$/ and $proxy->{ftp_proxy} = $1, next; /^proxy_user\s*=\s*(.*):(.*)$/ and do { $proxy->{user} = $1; $proxy->{pwd} = $2; next; }; next; } close F; $proxy; } sub set_proxy { my $proxy = shift @_; my @res; if (defined $proxy->{proxy}{http_proxy} or defined $proxy->{proxy}{ftp_proxy}) { for ($proxy->{type}) { /wget/ && do { for ($proxy->{proxy}) { if (defined $_->{http_proxy}) { $ENV{http_proxy} = $_->{http_proxy} =~ /^http:/ ? $_->{http_proxy} : "http://$_->{http_proxy}"; } $ENV{ftp_proxy} = $_->{ftp_proxy} if defined $_->{ftp_proxy}; @res = ("--proxy-user=$_->{user}", "--proxy-passwd=$_->{pwd}") if defined $_->{user} && defined $_->{pwd}; } last; }; /curl/ && do { for ($proxy->{proxy}) { push @res, ('-x', $_->{http_proxy}) if defined $_->{http_proxy}; push @res, ('-x', $_->{ftp_proxy}) if defined $_->{ftp_proxy}; push @res, ('-U', "$_->{user}:$_->{pwd}") if defined $_->{user} && defined $_->{pwd}; } last; }; die N("Unknown webfetch `%s' !!!\n", $proxy->{type}); } } return @res; } #- quoting/unquoting a string that may be containing space chars. sub quotespace { local $_ = $_[0] || ''; s/(\s)/\\$1/g; $_ } sub unquotespace { local $_ = $_[0] || ''; s/\\(\s)/$1/g; $_ } #- syncing algorithms, currently is implemented wget and curl methods, #- webfetch is trying to find the best (and one which will work :-) sub sync_webfetch { my $urpm = shift @_; my $options = shift @_; my %files; #- extract files according to protocol supported. #- currently ftp and http protocol are managed by curl or wget, #- ssh and rsync protocol are managed by rsync *AND* ssh. foreach (@_) { /^([^:_]*)[^:]*:/ or die N("unknown protocol defined for %s", $_); push @{$files{$1}}, $_; } if ($files{removable} || $files{file}) { sync_file($options, @{$files{removable} || []}, @{$files{file} || []}); delete @files{qw(removable file)}; } if ($files{ftp} || $files{http} || $files{https}) { my @webfetch = qw(curl wget); my @available_webfetch = grep { -x "/usr/bin/$_" } @webfetch; my $prefered; #- use user default downloader if provided and available if ($urpm->{options}{downloader}) { $prefered = find { $_ eq $urpm->{options}{downloader} } @available_webfetch; } #- else first downloader of @webfetch is the default one $prefered ||= $available_webfetch[0]; if ($prefered eq 'curl') { sync_curl($options, @{$files{ftp} || []}, @{$files{http} || []}, @{$files{https} || []}); } elsif ($prefered eq 'wget') { sync_wget($options, @{$files{ftp} || []}, @{$files{http} || []}, @{$files{https} || []}); } else { die N("no webfetch found, supported webfetch are: %s\n", join(", ", @webfetch)); } delete @files{qw(ftp http https)}; } if ($files{rsync}) { sync_rsync($options, @{$files{rsync} || []}); delete $files{rsync}; } if ($files{ssh}) { my @ssh_files; foreach (@{$files{ssh} || []}) { m|^ssh://([^/]*)(.*)| and push @ssh_files, "$1:$2"; } sync_ssh($options, @ssh_files); delete $files{ssh}; } %files and die N("unable to handle protocol: %s", join ', ', keys %files); } sub propagate_sync_callback { my $options = shift @_; if (ref($options) && $options->{callback}) { my $mode = shift @_; if ($mode =~ /^(start|progress|end)$/) { my $file = shift @_; $file =~ s|([^:]*://[^/:\@]*:)[^/:\@]*(\@.*)|$1xxxx$2|; #- if needed... return $options->{callback}($mode, $file, @_); } else { return $options->{callback}($mode, @_); } } } sub sync_file { my $options = shift @_; foreach (@_) { my ($in) = m!^(?:removable[^:]*|file):/(.*)!; propagate_sync_callback($options, 'start', $_); system("cp", "-p", "-R", $in || $_, ref($options) ? $options->{dir} : $options) and die N("copy failed: %s", $@); propagate_sync_callback($options, 'end', $_); } } sub sync_wget { -x "/usr/bin/wget" or die N("wget is missing\n"); local *WGET; my $options = shift @_; #- force download to be done in cachedir to avoid polluting cwd. my $cwd = `pwd`; chomp $cwd; chdir(ref($options) ? $options->{dir} : $options); my ($buf, $total, $file) = ('', undef, undef); my $wget_pid = open WGET, join(" ", map { "'$_'" } "/usr/bin/wget", (ref($options) && $options->{limit_rate} ? "--limit-rate=$options->{limit_rate}" : ()), (ref($options) && $options->{proxy} ? set_proxy({ type => "wget", proxy => $options->{proxy} }) : ()), (ref($options) && $options->{callback} ? ("--progress=bar:force", "-o", "-") : ref($options) && $options->{quiet} ? "-q" : @{[]}), "--retr-symlinks", "-NP", (ref($options) ? $options->{dir} : $options), @_) . " |"; local $/ = \1; #- read input by only one char, this is slow but very nice (and it works!). while () { $buf .= $_; if ($_ eq "\r" || $_ eq "\n") { if (ref($options) && $options->{callback}) { if ($buf =~ /^--\d\d:\d\d:\d\d--\s+(\S.*)\n/ms) { $file && $file ne $1 and propagate_sync_callback($options, 'end', $file); ! defined $file and propagate_sync_callback($options, 'start', $file = $1); } elsif (defined $file && ! defined $total && $buf =~ /==>\s+RETR/) { $total = ''; } elsif (defined $total && $total eq '' && $buf =~ /^[^:]*:\s+(\d\S*)/) { $total = $1; } elsif (my ($percent, $speed, $eta) = $buf =~ /^\s*(\d+)%.*\s+(\S+)\s+ETA\s+(\S+)\s*[\r\n]$/ms) { if (propagate_sync_callback($options, 'progress', $file, $percent, $total, $eta, $speed) eq 'canceled') { kill 15, $wget_pid; close WGET; return; } if ($_ eq "\n") { propagate_sync_callback($options, 'end', $file); ($total, $file) = (undef, undef); } } } else { ref($options) && $options->{quiet} or print STDERR $buf; } $buf = ''; } } $file and propagate_sync_callback($options, 'end', $file); chdir $cwd; close WGET or die N("wget failed: exited with %d or signal %d\n", $? >> 8, $? & 127); } sub sync_curl { -x "/usr/bin/curl" or die N("curl is missing\n"); local *CURL; my $options = shift @_; #- force download to be done in cachedir to avoid polluting cwd, #- howerver for curl, this is mandatory. my $cwd = `pwd`; chomp $cwd; chdir(ref($options) ? $options->{dir} : $options); my (@ftp_files, @other_files); foreach (@_) { m|^ftp://.*/([^/]*)$| && -s $1 > 8192 and do { push @ftp_files, $_; next }; #- manage time stamp for large file only. push @other_files, $_; } if (@ftp_files) { my ($cur_ftp_file, %ftp_files_info); eval { require Date::Manip }; #- prepare to get back size and time stamp of each file. open CURL, join(" ", map { "'$_'" } "/usr/bin/curl", (ref($options) && $options->{limit_rate} ? ("--limit-rate", $options->{limit_rate}) : ()), (ref($options) && $options->{proxy} ? set_proxy({ type => "curl", proxy => $options->{proxy} }) : ()), "--stderr", "-", "-s", "-I", @ftp_files) . " |"; while () { if (/Content-Length:\s*(\d+)/) { !$cur_ftp_file || exists($ftp_files_info{$cur_ftp_file}{size}) and $cur_ftp_file = shift @ftp_files; $ftp_files_info{$cur_ftp_file}{size} = $1; } if (/Last-Modified:\s*(.*)/) { !$cur_ftp_file || exists($ftp_files_info{$cur_ftp_file}{time}) and $cur_ftp_file = shift @ftp_files; eval { $ftp_files_info{$cur_ftp_file}{time} = Date::Manip::ParseDate($1); $ftp_files_info{$cur_ftp_file}{time} =~ s/(\d{6}).{4}(.*)/$1$2/; #- remove day and hour. }; } } close CURL; #- now analyse size and time stamp according to what already exists here. if (@ftp_files) { #- re-insert back shifted element of ftp_files, because curl output above #- have not been parsed correctly, in doubt download them all. push @ftp_files, keys %ftp_files_info; } else { #- for that, it should be clear ftp_files is empty... else a above work is #- use less. foreach (keys %ftp_files_info) { my ($lfile) = m|/([^/]*)$| or next; #- strange if we can't parse it correctly. my $ltime = eval { Date::Manip::ParseDate(scalar gmtime((stat $1)[9])) }; $ltime =~ s/(\d{6}).{4}(.*)/$1$2/; #- remove day and hour. -s $lfile == $ftp_files_info{$_}{size} && $ftp_files_info{$_}{time} eq $ltime or push @ftp_files, $_; } } } #- http files (and other files) are correctly managed by curl to conditionnal download. #- options for ftp files, -R (-O )* #- options for http files, -R (-z file -O )* if (my @all_files = ((map { ("-O", $_) } @ftp_files), (map { m|/([^/]*)$| ? ("-z", $1, "-O", $_) : @{[]} } @other_files))) { my @l = (@ftp_files, @other_files); my ($buf, $file) = ('', undef); my $curl_pid = open CURL, join(" ", map { "'$_'" } "/usr/bin/curl", (ref($options) && $options->{limit_rate} ? ("--limit-rate", $options->{limit_rate}) : ()), (ref($options) && $options->{proxy} ? set_proxy({ type => "curl", proxy => $options->{proxy} }) : ()), (ref($options) && $options->{quiet} && !$options->{verbose} ? "-s" : @{[]}), "-k", `curl -h` =~ /location-trusted/ ? "--location-trusted" : @{[]}, "-R", "-f", "--stderr", "-", @all_files) . " |"; local $/ = \1; #- read input by only one char, this is slow but very nice (and it works!). while () { $buf .= $_; if ($_ eq "\r" || $_ eq "\n") { if (ref($options) && $options->{callback}) { unless (defined $file) { $file = shift @l; propagate_sync_callback($options, 'start', $file); } if (my ($percent, $total, $eta, $speed) = $buf =~ /^\s*(\d+)\s+(\S+)[^\r\n]*\s+(\S+)\s+(\S+)[\r\n]$/ms) { if (propagate_sync_callback($options, 'progress', $file, $percent, $total, $eta, $speed) eq 'canceled') { kill 15, $curl_pid; close CURL; return; } if ($_ eq "\n") { propagate_sync_callback($options, 'end', $file); $file = undef; } } elsif ($buf =~ /^curl:/) { #- probably an error reported by curl propagate_sync_callback($options, 'error', $file, $buf); } } else { ref($options) && $options->{quiet} or print STDERR $buf; } $buf = ''; } } chdir $cwd; close CURL or die N("curl failed: exited with %d or signal %d\n", $? >> 8, $? & 127); } else { chdir $cwd; } } sub sync_rsync { -x "/usr/bin/rsync" or die N("rsync is missing\n"); my $options = shift @_; #- force download to be done in cachedir to avoid polluting cwd. my $cwd = `pwd`; chomp $cwd; chdir(ref($options) ? $options->{dir} : $options); my $limit_rate = ref($options) && $options->{limit_rate}; for ($limit_rate) { /^(\d+)$/ and $limit_rate = int $1/1024; /^(\d+)[kK]$/ and $limit_rate = $1; /^(\d+)[mM]$/ and $limit_rate = 1024*$1; /^(\d+)[gG]$/ and $limit_rate = 1024*1024*$1; } foreach (@_) { my $count = 10; #- retry count on error (if file exists). my $basename = basename($_); my ($file) = m|^rsync://(.*)| or next; $file =~ /::/ or $file = $_; propagate_sync_callback($options, 'start', $file); do { local (*RSYNC, $_); my $buf = ''; open RSYNC, join(" ", map { "'$_'" } "/usr/bin/rsync", ($limit_rate ? "--bwlimit=$limit_rate" : ()), (ref($options) && $options->{quiet} ? qw(-q) : qw(--progress -v)), if_(ref($options) && $options->{compress}, qw(-z)), qw(--partial --no-whole-file), $file, (ref($options) ? $options->{dir} : $options)) . " |"; local $/ = \1; #- read input by only one char, this is slow but very nice (and it works!). while () { $buf .= $_; if ($_ eq "\r" || $_ eq "\n") { if (ref($options) && $options->{callback}) { if (my ($percent, $speed) = $buf =~ /^\s*\d+\s+(\d+)%\s+(\S+)\s+/) { propagate_sync_callback($options, 'progress', $file, $percent, undef, undef, $speed); } } else { ref($options) && $options->{quiet} or print STDERR $buf; } $buf = ''; } } close RSYNC; } while ($? != 0 && --$count > 0 && -e (ref($options) ? $options->{dir} : $options) . "/$basename"); propagate_sync_callback($options, 'end', $file); } chdir $cwd; $? == 0 or die N("rsync failed: exited with %d or signal %d\n", $? >> 8, $? & 127); } sub sync_ssh { -x "/usr/bin/rsync" or die N("rsync is missing\n"); -x "/usr/bin/ssh" or die N("ssh is missing\n"); my $options = shift @_; #- force download to be done in cachedir to avoid polluting cwd. my $cwd = `pwd`; chomp $cwd; chdir(ref($options) ? $options->{dir} : $options); my $limit_rate = ref($options) && $options->{limit_rate}; for ($limit_rate) { /^(\d+)$/ and $limit_rate = int $1/1024; /^(\d+)[kK]$/ and $limit_rate = $1; /^(\d+)[mM]$/ and $limit_rate = 1024*$1; /^(\d+)[gG]$/ and $limit_rate = 1024*1024*$1; } foreach my $file (@_) { my $count = 10; #- retry count on error (if file exists). my $basename = basename($file); propagate_sync_callback($options, 'start', $file); do { local (*RSYNC, $_); my $buf = ''; open RSYNC, join(" ", map { "'$_'" } "/usr/bin/rsync", ($limit_rate ? "--bwlimit=$limit_rate" : ()), (ref($options) && $options->{quiet} ? qw(-q) : qw(--progress -v)), if_(ref($options) && $options->{compress}, qw(-z)), qw(--partial -e ssh), $file, (ref($options) ? $options->{dir} : $options)) . " |"; local $/ = \1; #- read input by only one char, this is slow but very nice (and it works!). while () { $buf .= $_; if ($_ eq "\r" || $_ eq "\n") { if (ref($options) && $options->{callback}) { if (my ($percent, $speed) = $buf =~ /^\s*\d+\s+(\d+)%\s+(\S+)\s+/) { propagate_sync_callback($options, 'progress', $file, $percent, undef, undef, $speed); } } else { ref($options) && $options->{quiet} or print STDERR $buf; } $buf = ''; } } close RSYNC; } while ($? != 0 && --$count > 0 && -e (ref($options) ? $options->{dir} : $options) . "/$basename"); propagate_sync_callback($options, 'end', $file); } chdir $cwd; $? == 0 or die N("rsync failed: exited with %d or signal %d\n", $? >> 8, $? & 127); } #- default logger suitable for sync operation on STDERR only. sub sync_logger { my ($mode, $file, $percent, $total, $eta, $speed) = @_; if ($mode eq 'start') { print STDERR " $file\n"; } elsif ($mode eq 'progress') { my $text; if (defined $total && defined $eta) { $text = N(" %s%% of %s completed, ETA = %s, speed = %s", $percent, $total, $eta, $speed); } else { $text = N(" %s%% completed, speed = %s", $percent, $speed); } print STDERR $text, " " x (79 - length($text)), "\r"; } elsif ($mode eq 'end') { print STDERR " " x 79, "\r"; } } #- read /etc/urpmi/urpmi.cfg as config file, keep compability with older #- configuration file by examining if one entry is of the form #- { #- ... #- } #- else only this form is used #- #- with sub read_config { my ($urpm, %options) = @_; #- keep in mind if it has been called before. $urpm->{media} and return; $urpm->{media} ||= []; #- check urpmi.cfg content, if the file is old keep track #- of old format used. local (*F, $_); open F, $urpm->{config}; #- no filename can be allowed on some case while () { chomp; s/#.*$//; s/^\s*//; s/\s*$//; $_ eq '{' and do { #- urpmi.cfg global options extension while () { chomp; s/#.*$//; s/^\s*//; s/\s*$//; $_ eq '}' and last; #- check for boolean variables first, and after that valued variables. my ($no, $k, $v); if (($no, $k, $v) = /^(no-)?(verify-rpm|fuzzy|keep|auto|allow-(?:force|nodeps)|(?:pre|post)-clean|excludedocs|compress)(?:\s*:\s*(.*))?$/) { unless (exists($urpm->{options}{$k})) { $urpm->{options}{$k} = $v eq '' || $v =~ /^(yes|on|1)$/i || 0; $no and $urpm->{options}{$k} = ! $urpm->{options}{$k} || 0; } next; } elsif (($k, $v) = /^(limit-rate|excludepath|key[\-_]ids|split-(?:level|length)|priority-upgrade|downloader)\s*:\s*(.*)$/) { unless (exists($urpm->{options}{$k})) { $v =~ /^'([^']*)'$/ and $v = $1; $v =~ /^"([^"]*)"$/ and $v = $1; $urpm->{options}{$k} = $v; } next; } $_ and $urpm->{error}(N("syntax error in config file at line %s", $.)); } exists $urpm->{options}{key_ids} && ! exists $urpm->{options}{'key-ids'} and $urpm->{options}{'key-ids'} = delete $urpm->{options}{key_ids}; next }; /^(.*?[^\\])\s+(?:(.*?[^\\])\s+)?{$/ and do { #- urpmi.cfg format extention my $medium = { name => unquotespace($1), clear_url => unquotespace($2) }; while () { chomp; s/#.*$//; s/^\s*//; s/\s*$//; $_ eq '}' and last; /^(hdlist|list|with_hdlist|removable|md5sum|key[\-_]ids)\s*:\s*(.*)$/ and $medium->{$1} = $2, next; /^(update|ignore|synthesis|virtual)\s*$/ and $medium->{$1} = 1, next; /^modified\s*$/ and next; $_ and $urpm->{error}(N("syntax error in config file at line %s", $.)); } exists $medium->{key_ids} && ! exists $medium->{'key-ids'} and $medium->{'key-ids'} = delete $medium->{key_ids}; $urpm->probe_medium($medium, %options) and push @{$urpm->{media}}, $medium; next }; /^(.*?[^\\])\s+(.*?[^\\])\s+with\s+(.*)$/ and do { #- urpmi.cfg old format for ftp my $medium = { name => unquotespace($1), clear_url => unquotespace($2), with_hdlist => unquotespace($3) }; $urpm->probe_medium($medium, %options) and push @{$urpm->{media}}, $medium; next }; /^(.*?[^\\])\s+(?:(.*?[^\\])\s*)?$/ and do { #- urpmi.cfg old format (assume hdlist..cz2?) my $medium = { name => unquotespace($1), clear_url => unquotespace($2) }; $urpm->probe_medium($medium, %options) and push @{$urpm->{media}}, $medium; next }; $_ and $urpm->{error}(N("syntax error in config file at line %s", $.)); } close F; #- keep in mind when an hdlist/list file is used, really usefull for #- the next probe. my (%hdlists, %lists); foreach (@{$urpm->{media}}) { if ($_->{hdlist}) { exists($hdlists{$_->{hdlist}}) and $_->{ignore} = 1, $urpm->{error}(N("medium \"%s\" trying to use an already used hdlist, medium ignored", $_->{name})); $hdlists{$_->{hdlist}} = undef; } if ($_->{list}) { exists($lists{$_->{list}}) and $_->{ignore} = 1, $urpm->{error}(N("medium \"%s\" trying to use an already used list, medium ignored", $_->{name})); $lists{$_->{list}} = undef; } } #- urpmi.cfg if old is not enough to known the various media, track #- directly into /var/lib/urpmi, foreach (glob("$urpm->{statedir}/hdlist.*")) { if (m|/hdlist\.((.*)\.cz2?)$|) { #- check if it has already been detected above. exists($hdlists{"hdlist.$1"}) and next; #- if not this is a new media to take care if #- there is a list file. if (-s "$urpm->{statedir}/list.$2") { if (exists($lists{"list.$2"})) { $urpm->{error}(N("unable to take care of medium \"%s\" as list file is already used by another medium", $2)); } else { my $medium; foreach (@{$urpm->{media}}) { $_->{name} eq $2 and $medium = $_, last; } $medium and $urpm->{error}(N("unable to use name \"%s\" for unnamed medium because it is already used", $2)), next; $medium = { name => $2, hdlist => "hdlist.$1", list => "list.$2" }; $urpm->probe_medium($medium, %options) and push @{$urpm->{media}}, $medium; } } else { $urpm->{error}(N("unable to take medium \"%s\" into account as no list file [%s] exists", $2, "$urpm->{statedir}/list.$2")); } } else { $urpm->{error}(N("unable to determine medium of this hdlist file [%s]", $_)); } } #- check the presence of hdlist file and list file if necessary. unless ($options{nocheck_access}) { foreach (@{$urpm->{media}}) { $_->{ignore} and next; -r "$urpm->{statedir}/$_->{hdlist}" || -r "$urpm->{statedir}/synthesis.$_->{hdlist}" && $_->{synthesis} or $_->{ignore} = 1, $urpm->{error}(N("unable to access hdlist file of \"%s\", medium ignored", $_->{name})); $_->{list} && -r "$urpm->{statedir}/$_->{list}" || defined $_->{url} or $_->{ignore} = 1, $urpm->{error}(N("unable to access list file of \"%s\", medium ignored", $_->{name})); } } local *MD5SUM; open MD5SUM, "$urpm->{statedir}/MD5SUM"; while () { my ($md5sum, $file) = /(\S*)\s+(.*)/; foreach (@{$urpm->{media}}) { ($_->{synthesis} && "synthesis.").$_->{hdlist} eq $file and $_->{md5sum} = $md5sum, last; } } close MD5SUM; } #- probe medium to be used, take old medium into account too. sub probe_medium { my ($urpm, $medium, %options) = @_; local $_; my $existing_medium; foreach (@{$urpm->{media}}) { $_->{name} eq $medium->{name} and $existing_medium = $_, last; } $existing_medium and $urpm->{error}(N("trying to bypass existing medium \"%s\", avoiding", $medium->{name})), return; $medium->{url} ||= $medium->{clear_url}; if ($medium->{virtual}) { #- a virtual medium need to have an url available without using a list file. if ($medium->{hdlist} || $medium->{list}) { $medium->{ignore} = 1; $urpm->{error}(N("virtual medium \"%s\" should not have defined hdlist or list file, medium ignored", $medium->{name})); } unless ($medium->{url}) { $medium->{ignore} = 1; $urpm->{error}(N("virtual medium \"%s\" should have a clear url, medium ignored", $medium->{name})); } } else { unless ($medium->{ignore} || $medium->{hdlist}) { $medium->{hdlist} = "hdlist.$medium->{name}.cz"; -e "$urpm->{statedir}/$medium->{hdlist}" or $medium->{hdlist} = "hdlist.$medium->{name}.cz2"; -e "$urpm->{statedir}/$medium->{hdlist}" or $medium->{ignore} = 1, $urpm->{error}(N("unable to find hdlist file for \"%s\", medium ignored", $medium->{name})); } unless ($medium->{ignore} || $medium->{list}) { unless (defined $medium->{url}) { $medium->{list} = "list.$medium->{name}"; unless (-e "$urpm->{statedir}/$medium->{list}") { $medium->{ignore} = 1, $urpm->{error}(N("unable to find list file for \"%s\", medium ignored", $medium->{name})); } } } #- there is a little more to do at this point as url is not known, inspect directly list file for it. unless ($medium->{url}) { my %probe; if (-r "$urpm->{statedir}/$medium->{list}") { local *L; open L, "$urpm->{statedir}/$medium->{list}"; while () { #- /./ is end of url marker in list file (typically generated by a #- find . -name "*.rpm" > list #- for exportable list file. m|^(.*)/\./| and $probe{$1} = undef; m|^(.*)/[^/]*$| and $probe{$1} = undef; } close L; } foreach (sort { length($a) <=> length($b) } keys %probe) { if ($medium->{url}) { $medium->{url} eq substr($_, 0, length($medium->{url})) or $medium->{ignore} || $urpm->{error}(N("incoherent list file for \"%s\", medium ignored", $medium->{name})), $medium->{ignore} = 1, last; } else { $medium->{url} = $_; } } unless ($options{nocheck_access}) { $medium->{url} or $medium->{ignore} || $urpm->{error}(N("unable to inspect list file for \"%s\", medium ignored", $medium->{name})), $medium->{ignore} = 1; } } } #- probe removable device. $urpm->probe_removable_device($medium); #- clear URLs for trailing /es. $medium->{url} and $medium->{url} =~ s|(.*?)/*$|$1|; $medium->{clear_url} and $medium->{clear_url} =~ s|(.*?)/*$|$1|; $medium; } #- probe device associated with a removable device. sub probe_removable_device { my ($urpm, $medium) = @_; if ($medium->{url} && $medium->{url} =~ /^removable_?([^_:]*)(?:_[^:]*)?:/) { $medium->{removable} ||= $1 && "/dev/$1"; } else { delete $medium->{removable}; } #- try to find device to open/close for removable medium. if (exists($medium->{removable})) { if (my ($dir) = $medium->{url} =~ m!(?:file|removable)[^:]*:/(.*)!) { my %infos; my @mntpoints = $urpm->find_mntpoints($dir, \%infos); if (@mntpoints > 1) { #- return value is suitable for an hash. $urpm->{log}(N("too many mount points for removable medium \"%s\"", $medium->{name})); $urpm->{log}(N("taking removable device as \"%s\"", join ',', map { $infos{$_}{device} } @mntpoints)); } if (@mntpoints) { if ($medium->{removable} && $medium->{removable} ne $infos{$mntpoints[-1]}{device}) { $urpm->{log}(N("using different removable device [%s] for \"%s\"", $infos{$mntpoints[-1]}{device}, $medium->{name})); } $medium->{removable} = $infos{$mntpoints[-1]}{device}; } else { $urpm->{error}(N("unable to retrieve pathname for removable medium \"%s\"", $medium->{name})); } } else { $urpm->{error}(N("unable to retrieve pathname for removable medium \"%s\"", $medium->{name})); } } } #- write back urpmi.cfg code to allow modification of medium listed. sub write_config { my ($urpm) = @_; #- avoid trashing exiting configuration in this case. $urpm->{media} or return; local (*F, *MD5SUM); open F, ">$urpm->{config}" or $urpm->{fatal}(6, N("unable to write config file [%s]", $urpm->{config})); open MD5SUM, ">$urpm->{statedir}/MD5SUM"; if (%{$urpm->{options} || {}}) { printf F "{\n"; while (my ($k, $v) = each %{$urpm->{options}}) { printf F " %s: %s\n", $k, $v; } printf F "}\n\n"; } foreach my $medium (@{$urpm->{media}}) { printf F "%s %s {\n", quotespace($medium->{name}), quotespace($medium->{clear_url}); foreach (qw(hdlist with_hdlist list removable key-ids priority-upgrade)) { $medium->{$_} and printf F " %s: %s\n", $_, $medium->{$_}; } $medium->{md5sum} and print MD5SUM "$medium->{md5sum} ".($medium->{synthesis} && "synthesis.").$medium->{hdlist}."\n"; foreach (qw(update ignore synthesis modified virtual)) { $medium->{$_} and printf F " %s\n", $_; } printf F "}\n\n"; } close MD5SUM; close F; $urpm->{log}(N("write config file [%s]", $urpm->{config})); #- everything should be synced now. delete $urpm->{modified}; } #- read urpmi.cfg file as well as synthesis file needed. sub configure { my ($urpm, %options) = @_; $urpm->clean; $options{parallel} && $options{usedistrib} and die N("Can't use parallel mode with use-distrib mode"); if ($options{parallel}) { my ($parallel_options, $parallel_handler); #- handle parallel configuration, examine all module available that #- will handle the parallel mode (configuration is /etc/urpmi/parallel.cfg). local ($_, *PARALLEL); open PARALLEL, "/etc/urpmi/parallel.cfg"; while () { chomp; s/#.*$//; s/^\s*//; s/\s*$//; /\s*([^:]*):(.*)/ or $urpm->{error}(N("unable to parse \"%s\" in file [%s]", $_, "/etc/urpmi/parallel.cfg")), next; $1 eq $options{parallel} and $parallel_options = ($parallel_options && "\n") . $2; } close PARALLEL; #- if a configuration options has been found, use it else fatal error. if ($parallel_options) { foreach my $dir (grep { -d $_ } map { "$_/urpm" } @INC) { local *DIR; opendir DIR, $dir; while ($_ = readdir DIR) { -f "$dir/$_" or next; $urpm->{log}->(N("examining parallel handler in file [%s]", "$dir/$_")); eval { require "$dir/$_"; $parallel_handler = $urpm->handle_parallel_options($parallel_options) }; $parallel_handler and last; } closedir DIR; $parallel_handler and last; } } if ($parallel_handler) { if ($parallel_handler->{nodes}) { $urpm->{log}->(N("found parallel handler for nodes: %s", join(', ', keys %{$parallel_handler->{nodes}}))); } if (!$options{media} && $parallel_handler->{media}) { $options{media} = $parallel_handler->{media}; $urpm->{log}->(N("using associated media for parallel mode: %s", $options{media})); } $urpm->{parallel_handler} = $parallel_handler; } else { $urpm->{fatal}(1, N("unable to use parallel option \"%s\"", $options{parallel})); } } else { #- parallel is exclusive against root options. $urpm->{root} = $options{root}; } if ($options{synthesis}) { if ($options{synthesis} ne 'none') { #- synthesis take precedence over media, update options. $options{media} || $options{excludemedia} || $options{sortmedia} || $options{update} || $options{parallel} and $urpm->{fatal}(1, N("--synthesis cannot be used with --media, --excludemedia, --sortmedia, --update or --parallel")); $urpm->parse_synthesis($options{synthesis}); #- synthesis disable the split of transaction (too risky and not usefull). $urpm->{options}{'split-length'} = 0; } } else { if ($options{usedistrib}) { $urpm->{media} = []; $urpm->add_distrib_media("Virtual", $options{usedistrib}, %options, 'virtual' => 1); } else { $urpm->read_config(%options); } if ($options{media}) { delete $_->{modified} foreach @{$urpm->{media} || []}; $urpm->select_media(split ',', $options{media}); foreach (grep { !$_->{modified} } @{$urpm->{media} || []}) { #- this is only a local ignore that will not be saved. $_->{ignore} = 1; } } if ($options{excludemedia}) { delete $_->{modified} foreach @{$urpm->{media} || []}; $urpm->select_media(split ',', $options{excludemedia}); foreach (grep { $_->{modified} } @{$urpm->{media} || []}) { #- this is only a local ignore that will not be saved. $_->{ignore} = 1; } } if ($options{sortmedia}) { delete $_->{modified} foreach @{$urpm->{media} || []}; my @oldmedia = @{$urpm->{media} || []}; my @newmedia; foreach (split ',', $options{sortmedia}) { $urpm->select_media($_); push @newmedia, grep { $_->{modified} } @oldmedia; @oldmedia = grep { !$_->{modified} } @oldmedia; } #- anything not selected should be added as is after the selected one. $urpm->{media} = [ @newmedia, @oldmedia ]; #- clean remaining modified flag. delete $_->{modified} foreach @{$urpm->{media} || []}; } unless ($options{nodepslist}) { my $second_pass; do { foreach (grep { !$_->{ignore} && (!$options{update} || $_->{update}) } @{$urpm->{media} || []}) { delete @$_{qw(start end)}; if ($_->{virtual}) { my $path = $_->{url} =~ m|^file:/*(/[^/].*[^/])/*$| && $1; if ($path) { if ($_->{synthesis}) { $urpm->{log}(N("examining synthesis file [%s]", "$path/$_->{with_hdlist}")); eval { ($_->{start}, $_->{end}) = $urpm->parse_synthesis("$path/$_->{with_hdlist}", callback => $options{callback}) }; } else { $urpm->{log}(N("examining hdlist file [%s]", "$path/$_->{with_hdlist}")); eval { ($_->{start}, $_->{end}) = $urpm->parse_hdlist("$path/$_->{with_hdlist}", packing => 1, callback => $options{callback}) }; #- we need a second pass now. defined $second_pass or $second_pass = 1; } } else { $urpm->{error}(N("virtual medium \"%s\" is not local, medium ignored", $_->{name})); $_->{ignore} = 1; } } else { if ($options{hdlist} && -s "$urpm->{statedir}/$_->{hdlist}" > 32) { $urpm->{log}(N("examining hdlist file [%s]", "$urpm->{statedir}/$_->{hdlist}")); eval { ($_->{start}, $_->{end}) = $urpm->parse_hdlist("$urpm->{statedir}/$_->{hdlist}", packing => 1, callback => $options{callback}) }; } else { $urpm->{log}(N("examining synthesis file [%s]", "$urpm->{statedir}/synthesis.$_->{hdlist}")); eval { ($_->{start}, $_->{end}) = $urpm->parse_synthesis("$urpm->{statedir}/synthesis.$_->{hdlist}", callback => $options{callback}) }; unless (defined $_->{start} && defined $_->{end}) { $urpm->{log}(N("examining hdlist file [%s]", "$urpm->{statedir}/$_->{hdlist}")); eval { ($_->{start}, $_->{end}) = $urpm->parse_hdlist("$urpm->{statedir}/$_->{hdlist}", packing => 1, callback => $options{callback}) }; } } } unless ($_->{ignore}) { unless (defined $_->{start} && defined $_->{end}) { $urpm->{error}(N("problem reading hdlist or synthesis file of medium \"%s\"", $_->{name})); $_->{ignore} = 1; } } } } while ($second_pass && do { require URPM::Build; $urpm->{log}(N("performing second pass to compute dependencies\n")); $urpm->unresolved_provides_clean; $second_pass-- }); } } #- determine package to withdraw (from skip.list file) only if something should be withdrawn. unless ($options{noskipping}) { my %uniq; $urpm->compute_flags($urpm->get_packages_list($urpm->{skiplist}, $options{skip}), skip => 1, callback => sub { my ($urpm, $pkg) = @_; $pkg->is_arch_compat && ! exists $uniq{$pkg->fullname} or return; $uniq{$pkg->fullname} = undef; $urpm->{log}(N("skipping package %s", scalar($pkg->fullname))); }); } unless ($options{noinstalling}) { my %uniq; $urpm->compute_flags($urpm->get_packages_list($urpm->{instlist}, $options{inst}), disable_obsolete => 1, callback => sub { my ($urpm, $pkg) = @_; $pkg->is_arch_compat && ! exists $uniq{$pkg->fullname} or return; $uniq{$pkg->fullname} = undef; $urpm->{log}(N("would install instead of upgrade package %s", scalar($pkg->fullname))); }); } if ($options{bug}) { #- and a dump of rpmdb itself as synthesis file. my $db = URPM::DB::open($options{root}); my $sig_handler = sub { undef $db; exit 3 }; local $SIG{INT} = $sig_handler; local $SIG{QUIT} = $sig_handler; local *RPMDB; $db or $urpm->{fatal}(9, N("unable to open rpmdb")); open RPMDB, "| " . ($ENV{LD_LOADER} || '') . " gzip -9 >'$options{bug}/rpmdb.cz'"; $db->traverse(sub { my ($p) = @_; #- this is not right but may be enough. my $files = join '@', grep { exists($urpm->{provides}{$_}) } $p->files; $p->pack_header; $p->build_info(fileno *RPMDB, $files); }); close RPMDB; } } #- add a new medium, sync the config file accordingly. sub add_medium { my ($urpm, $name, $url, $with_hdlist, %options) = @_; #- make sure configuration has been read. # (Olivier Thauvin) Yes but Why ??? Is this a workaround ? $urpm->{media} or $urpm->read_config; #- if a medium with that name has already been found #- we have to exit now my ($medium); if (defined $options{index_name}) { my $i = $options{index_name}; do { ++$i; undef $medium; foreach (@{$urpm->{media}}) { $_->{name} eq $name.$i and $medium = $_; } } while $medium; $name .= $i; } else { foreach (@{$urpm->{media}}) { $_->{name} eq $name and $medium = $_; } } $medium and $urpm->{fatal}(5, N("medium \"%s\" already exists", $medium->{name})); #- clear URLs for trailing /es. $url =~ s|(.*?)/*$|$1|; #- creating the medium info. if ($options{virtual}) { $url =~ m|^file:/*(/[^/].*)/| or $urpm->{fatal}(1, N("virtual medium need to be local")); $medium = { name => $name, url => $url, update => $options{update}, virtual => 1, modified => 1, }; } else { $medium = { name => $name, url => $url, hdlist => "hdlist.$name.cz", list => "list.$name", update => $options{update}, modified => 1, }; #- check to see if the medium is using file protocol or removable medium. $url =~ m!^(removable[^:]*|file):/(.*)! and $urpm->probe_removable_device($medium); } #- check if a password is visible, if not set clear_url. $url =~ m|([^:]*://[^/:\@]*:)[^/:\@]*(\@.*)| or $medium->{clear_url} = $url; #- all flags once everything has been computed. $with_hdlist and $medium->{with_hdlist} = $with_hdlist; #- create an entry in media list. push @{$urpm->{media}}, $medium; #- keep in mind the database has been modified and base files need to be updated. #- this will be done automatically by transfering modified flag from medium to global. $urpm->{log}(N("added medium %s", $name)); } #- add distribution media, according to url given. sub add_distrib_media { my ($urpm, $name, $url, %options) = @_; my ($hdlists_file); #- make sure configuration has been read. # (Olivier Thauvin): Is this a workaround ? $urpm->{media} or $urpm->read_config; #- try to copy/retrive Mandrake/basehdlists file. if (my ($dir) = $url =~ m!^(?:removable[^:]*|file):/(.*)!) { $hdlists_file = reduce_pathname("$dir/Mandrake/base/hdlists"); $urpm->try_mounting($hdlists_file) or $urpm->{error}(N("unable to access first installation medium")), return; if (-e $hdlists_file) { unlink "$urpm->{cachedir}/partial/hdlists"; $urpm->{log}(N("copying hdlists file...")); system("cp", "-p", "-R", $hdlists_file, "$urpm->{cachedir}/partial/hdlists") ? $urpm->{log}(N("...copying failed")) : $urpm->{log}(N("...copying done")); } else { $urpm->{error}(N("unable to access first installation medium (no Mandrake/base/hdlists file found)")), return; } } else { #- try to get the description if it has been found. unlink "$urpm->{cachedir}/partial/hdlists"; eval { $urpm->{log}(N("retrieving hdlists file...")); $urpm->{sync}({ dir => "$urpm->{cachedir}/partial", quiet => 1, limit_rate => $options{limit_rate}, compress => $options{compress}, proxy => $urpm->{proxy} }, reduce_pathname("$url/Mandrake/base/hdlists")); $urpm->{log}(N("...retrieving done")); }; $@ and $urpm->{log}(N("...retrieving failed: %s", $@)); if (-e "$urpm->{cachedir}/partial/hdlists") { $hdlists_file = "$urpm->{cachedir}/partial/hdlists"; } else { $urpm->{error}(N("unable to access first installation medium (no Mandrake/base/hdlists file found)")), return; } } #- cosmetic update of name if it contains blank char. $name =~ /\s/ and $name .= ' '; #- at this point, we have found an hdlists file, so parse it #- and create all necessary medium according to it. local *HDLISTS; if (open HDLISTS, $hdlists_file) { my $medium = 1; foreach () { chomp; s/\s*#.*$//; /^\s*$/ and next; m/^\s*(?:noauto:)?(hdlist\S*\.cz2?)\s+(\S+)\s*(.*)$/ or $urpm->{error}(N("invalid hdlist description \"%s\" in hdlists file"), $_); my ($hdlist, $rpmsdir, $descr) = ($1, $2, $3); $urpm->add_medium($name ? "$descr ($name$medium)" : $descr, "$url/$rpmsdir", "../base/$hdlist", %options); ++$medium; } close HDLISTS; } else { $urpm->{error}(N("unable to access first installation medium (no Mandrake/base/hdlists file found)")), return; } } sub select_media { my $urpm = shift; my %media; @media{@_} = undef; foreach (@{$urpm->{media}}) { if (exists($media{$_->{name}})) { $media{$_->{name}} = 1; #- keep it mind this one has been selected. #- select medium by setting modified flags, do not check ignore. $_->{modified} = 1; } } #- check if some arguments does not correspond to medium name. #- in such case, try to find the unique medium (or list candidate #- media found). foreach (keys %media) { unless ($media{$_}) { my $q = quotemeta; my (@found, @foundi); foreach my $medium (@{$urpm->{media}}) { $medium->{name} =~ /$q/ and push @found, $medium; $medium->{name} =~ /$q/i and push @foundi, $medium; } if (@found == 1) { $found[0]{modified} = 1; } elsif (@foundi == 1) { $foundi[0]{modified} = 1; } elsif (@found == 0 && @foundi == 0) { $urpm->{error}(N("trying to select nonexistent medium \"%s\"", $_)); } else { #- multiple element in found or foundi list. $urpm->{log}(N("selecting multiple media: %s", join(", ", map { N("\"%s\"", $_->{name}) } (@found ? @found : @foundi)))); #- changed behaviour to select all occurence by default. foreach (@found ? @found : @foundi) { $_->{modified} = 1; } } } } } sub remove_selected_media { my ($urpm) = @_; my @result; foreach (@{$urpm->{media}}) { if ($_->{modified}) { $urpm->{log}(N("removing medium \"%s\"", $_->{name})); #- mark to re-write configuration. $urpm->{modified} = 1; #- remove file associated with this medium. foreach ($_->{hdlist}, $_->{list}, "synthesis.$_->{hdlist}", "descriptions.$_->{name}", "names.$_->{name}", "$_->{name}.cache") { $_ and unlink "$urpm->{statedir}/$_"; } } else { push @result, $_; #- not removed so keep it } } #- restore newer media list. $urpm->{media} = \@result; } #- return list of synthesis or hdlist reference to probe. sub probe_with_try_list { my ($suffix, $probe_with) = @_; my @probe = ("../base/synthesis.hdlist$suffix.cz", "../synthesis.hdlist$suffix.cz", "synthesis.hdlist$suffix.cz"); length($suffix) and unshift @probe, "synthesis.hdlist.cz"; length($suffix) or push @probe, ("../base/synthesis.hdlist1.cz", "../base/synthesis.hdlist2.cz", "../synthesis.hdlist1.cz", "../synthesis.hdlist2.cz", "synthesis.hdlist1.cz", "synthesis.hdlist2.cz"); my @probe_hdlist = ("../base/hdlist$suffix.cz", "../hdlist$suffix.cz", "hdlist$suffix.cz"); length($suffix) and push @probe_hdlist, "hdlist.cz"; length($suffix) or push @probe_hdlist, ("../base/hdlist1.cz", "../base/hdlist2.cz", "../hdlist1.cz", "../hdlist2.cz", "hdlist1.cz", "hdlist2.cz"); if ($probe_with =~ /synthesis/) { push @probe, @probe_hdlist; } else { unshift @probe, @probe_hdlist; } @probe; } #- update urpmi database regarding the current configuration. #- take care of modification and try some trick to bypass #- computational of base files. #- allow options : #- all -> all medium are rebuilded. #- force -> try to force rebuilding base files (1) or hdlist from rpm files (2). #- probe_with -> probe synthesis or hdlist. #- ratio -> use compression ratio (with gzip, default is 4) #- noclean -> keep header directory cleaned. sub update_media { my ($urpm, %options) = @_; #- do not trust existing hdlist and try to recompute them. my ($cleaned_cache, $second_pass); #- take care of some options. $cleaned_cache = !$options{noclean}; #- avoid trashing existing configuration in this case. $urpm->{media} or return; #- now we need additional methods not defined by default in URPM. require URPM::Build; require URPM::Signature; #- get gpg-pubkey signature. $options{nopubkey} or $urpm->exlock_rpm_db; $options{nopubkey} or $urpm->{keys} or $urpm->parse_pubkeys(root => $urpm->{root}); #- lock database if allowed. $options{nolock} or $urpm->exlock_urpmi_db; #- examine each medium to see if one of them need to be updated. #- if this is the case and if not forced, try to use a pre-calculated #- hdlist file else build it from rpm files. $urpm->clean; foreach my $medium (@{$urpm->{media}}) { #- take care of modified medium only or all if all have to be recomputed. $medium->{ignore} and next; $options{forcekey} and delete $medium->{'key-ids'}; #- and create synthesis file associated if it does not already exists... -s "$urpm->{statedir}/synthesis.$medium->{hdlist}" > 32 or $medium->{modified_synthesis} = 1; #- but do not take care of removable media for all. $medium->{modified} ||= $options{all} && $medium->{url} !~ m!^removable://!; unless ($medium->{modified}) { #- the medium is not modified, but for computing dependencies, #- we still need to read it and all synthesis will be written if #- a unresolved provides is found. #- to speed up the process, we only read the synthesis at the begining. delete @$medium{qw(start end)}; if ($medium->{virtual}) { my ($path) = $medium->{url} =~ m|^file:/*(/[^/].*[^/])/*$|; my $with_hdlist_file = "$path/$medium->{with_hdlist}"; if ($path) { if ($medium->{synthesis}) { $urpm->{log}(N("examining synthesis file [%s]", $with_hdlist_file)); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis($with_hdlist_file) }; } else { $urpm->{log}(N("examining hdlist file [%s]", $with_hdlist_file)); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_hdlist($with_hdlist_file, packing => 1) }; } } else { $urpm->{error}(N("virtual medium \"%s\" is not local, medium ignored", $medium->{name})); $_->{ignore} = 1; } } else { $urpm->{log}(N("examining synthesis file [%s]", "$urpm->{statedir}/synthesis.$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis("$urpm->{statedir}/synthesis.$medium->{hdlist}") }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{log}(N("examining hdlist file [%s]", "$urpm->{statedir}/$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_hdlist("$urpm->{statedir}/$medium->{hdlist}", packing => 1) }; } } unless ($medium->{ignore}) { unless (defined $medium->{start} && defined $medium->{end}) { #- this is almost a fatal error, ignore it by default? $urpm->{error}(N("problem reading hdlist or synthesis file of medium \"%s\"", $medium->{name})); $medium->{ignore} = 1; } } next; } #- list of rpm files for this medium, only available for local medium where #- the source hdlist is not used (use force). my ($prefix, $dir, $error, $retrieved_md5sum, @files); #- always delete a remaining list file or pubkey file in cache. foreach (qw(list pubkey)) { unlink "$urpm->{cachedir}/partial/$_"; } #- check to see if the medium is using file protocol or removable medium. if (($prefix, $dir) = $medium->{url} =~ m!^(removable[^:]*|file):/(.*)!) { #- try to figure a possible hdlist_path (or parent directory of searched directory. #- this is used to probe possible hdlist file. my $with_hdlist_dir = reduce_pathname($dir . ($medium->{with_hdlist} ? "/$medium->{with_hdlist}" : "/..")); #- the directory given does not exist and may be accessible #- by mounting some other. try to figure out these directory and #- mount everything necessary. $urpm->try_mounting($options{force} < 2 && ($options{probe_with} || $medium->{with_hdlist}) ? $with_hdlist_dir : $dir) or $urpm->{error}(N("unable to access medium \"%s\", this could happen if you mounted manually the directory when creating the medium.", $medium->{name})), next; #- try to probe for possible with_hdlist parameter, unless #- it is already defined (and valid). if ($options{probe_with} && (!$medium->{with_hdlist} || ! -e "$dir/$medium->{with_hdlist}")) { my ($suffix) = $dir =~ m|RPMS([^/]*)/*$|; foreach (probe_with_try_list($suffix, $options{probe_with})) { if (-s "$dir/$_" > 32) { $medium->{with_hdlist} = $_; last; } } #- redo... $with_hdlist_dir = reduce_pathname($dir . ($medium->{with_hdlist} ? "/$medium->{with_hdlist}" : "/..")); } if ($medium->{virtual}) { #- syncing a virtual medium is very simple, just try to read the file in order to #- determine its type, once a with_hdlist has been found (but is mandatory). if ($medium->{with_hdlist} && -e $with_hdlist_dir) { delete @$medium{qw(start end)}; if ($medium->{synthesis}) { $urpm->{log}(N("examining synthesis file [%s]", $with_hdlist_dir)); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis($with_hdlist_dir); delete $medium->{modified}; $medium->{synthesis} = 1; $urpm->{modified} = 1 }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{log}(N("examining hdlist file [%s]", $with_hdlist_dir)); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_hdlist($with_hdlist_dir, packing => 1); delete @$medium{qw(modified synthesis)}; $urpm->{modified} = 1 }; } } else { $urpm->{log}(N("examining hdlist file [%s]", $with_hdlist_dir)); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_hdlist($with_hdlist_dir, packing => 1); delete @$medium{qw(modified synthesis)}; $urpm->{modified} = 1 }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{log}(N("examining synthesis file [%s]", $with_hdlist_dir)); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis($with_hdlist_dir); delete $medium->{modified}; $medium->{synthesis} = 1; $urpm->{modified} = 1 }; } } unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{error}(N("problem reading hdlist or synthesis file of medium \"%s\"", $medium->{name})); $medium->{ignore} = 1; } } else { $urpm->{error}(N("virtual medium \"%s\" should have valid source hdlist or synthesis, medium ignored", $medium->{name})); $medium->{ignore} = 1; } } #- try to get the description if it has been found. unlink "$urpm->{statedir}/descriptions.$medium->{name}"; if (-e "$dir/../descriptions") { $urpm->{log}(N("copying description file of \"%s\"...", $medium->{name})); system("cp", "-p", "-R", "$dir/../descriptions", "$urpm->{statedir}/descriptions.$medium->{name}") ? $urpm->{log}(N("...copying failed")) : $urpm->{log}(N("...copying done")); } #- examine if a distant MD5SUM file is available. #- this will only be done if $with_hdlist is not empty in order to use #- an existing hdlist or synthesis file, and to check if download was good. #- if no MD5SUM are available, do it as before... #- we can assume at this point a basename is existing, but it needs #- to be checked for being valid, nothing can be deduced if no MD5SUM #- file are present. my $basename = basename($with_hdlist_dir); unless ($medium->{virtual}) { if ($medium->{with_hdlist}) { if (!$options{nomd5sum} && -s reduce_pathname("$with_hdlist_dir/../MD5SUM") > 32) { if ($options{force}) { #- force downloading the file again, else why a force option has been defined ? delete $medium->{md5sum}; } else { unless ($medium->{md5sum}) { $urpm->{log}(N("computing md5sum of existing source hdlist (or synthesis)")); if ($medium->{synthesis}) { -e "$urpm->{statedir}/synthesis.$medium->{hdlist}" and $medium->{md5sum} = (split ' ', `md5sum '$urpm->{statedir}/synthesis.$medium->{hdlist}'`)[0]; } else { -e "$urpm->{statedir}/$medium->{hdlist}" and $medium->{md5sum} = (split ' ', `md5sum '$urpm->{statedir}/$medium->{hdlist}'`)[0]; } } } if ($medium->{md5sum}) { $urpm->{log}(N("examining MD5SUM file")); local (*F, $_); open F, reduce_pathname("$with_hdlist_dir/../MD5SUM"); while () { my ($md5sum, $file) = m|(\S+)\s+(?:\./)?(\S+)| or next; #- keep md5sum got here to check download was ok ! so even if md5sum is not defined, we need #- to compute it, keep it in mind ;) $file eq $basename and $retrieved_md5sum = $md5sum; } close F; #- if an existing hdlist or synthesis file has the same md5sum, we assume the #- file are the same. #- if local md5sum is the same as distant md5sum, this means there is no need to #- download hdlist or synthesis file again. foreach (@{$urpm->{media}}) { if ($_->{md5sum} && $_->{md5sum} eq $retrieved_md5sum) { unlink "$urpm->{cachedir}/partial/$basename"; #- the medium is now considered not modified. $medium->{modified} = 0; #- hdlist or synthesis file must be linked with the other same one. #- a link is better for reducing used size of /var/lib/urpmi. if ($_ ne $medium) { $medium->{md5sum} = $_->{md5sum}; unlink "$urpm->{statedir}/synthesis.$medium->{hdlist}"; unlink "$urpm->{statedir}/$medium->{hdlist}"; symlink "synthesis.$_->{hdlist}", "$urpm->{statedir}/synthesis.$medium->{hdlist}"; symlink $_->{hdlist}, "$urpm->{statedir}/$medium->{hdlist}"; } #- as previously done, just read synthesis file here, this is enough. $urpm->{log}(N("examining synthesis file [%s]", "$urpm->{statedir}/synthesis.$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis("$urpm->{statedir}/synthesis.$medium->{hdlist}") }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{log}(N("examining hdlist file [%s]", "$urpm->{statedir}/$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_hdlist("$urpm->{statedir}/$medium->{hdlist}", packing => 1) }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{error}(N("problem reading hdlist or synthesis file of medium \"%s\"", $medium->{name})); $medium->{ignore} = 1; } } #- no need to continue examining other md5sum. last; } } $medium->{modified} or next; } } #- if the source hdlist is present and we are not forcing using rpms file if ($options{force} < 2 && -e $with_hdlist_dir) { unlink "$urpm->{cachedir}/partial/$medium->{hdlist}"; $urpm->{log}(N("copying source hdlist (or synthesis) of \"%s\"...", $medium->{name})); $options{callback} && $options{callback}('copy', $medium->{name}); if (system("cp", "-p", "-R", $with_hdlist_dir, "$urpm->{cachedir}/partial/$medium->{hdlist}")) { $options{callback} && $options{callback}('failed', $medium->{name}); $urpm->{log}(N("...copying failed")); unlink "$urpm->{cachedir}/partial/$medium->{hdlist}"; #- force error... } else { $options{callback} && $options{callback}('done', $medium->{name}); $urpm->{log}(N("...copying done")); } } -s "$urpm->{cachedir}/partial/$medium->{hdlist}" > 32 or $error = 1, $urpm->{error}(N("copy of [%s] failed", $with_hdlist_dir)); #- keep checking md5sum of file just copied ! (especially on nfs or removable device). if (!$error && $retrieved_md5sum) { $urpm->{log}(N("computing md5sum of copied source hdlist (or synthesis)")); (split ' ', `md5sum '$urpm->{cachedir}/partial/$medium->{hdlist}'`)[0] eq $retrieved_md5sum or $error = 1, $urpm->{error}(N("copy of [%s] failed", $with_hdlist_dir)); } #- check if the file are equals... and no force copy... if (!$error && !$options{force} && -e "$urpm->{statedir}/synthesis.$medium->{hdlist}") { my @sstat = stat "$urpm->{cachedir}/partial/$medium->{hdlist}"; my @lstat = stat "$urpm->{statedir}/$medium->{hdlist}"; if ($sstat[7] == $lstat[7] && $sstat[9] == $lstat[9]) { #- the two files are considered equal here, the medium is so not modified. $medium->{modified} = 0; unlink "$urpm->{cachedir}/partial/$medium->{hdlist}"; #- as previously done, just read synthesis file here, this is enough, but only #- if synthesis exists, else it need to be recomputed. $urpm->{log}(N("examining synthesis file [%s]", "$urpm->{statedir}/synthesis.$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis("$urpm->{statedir}/synthesis.$medium->{hdlist}") }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{log}(N("examining hdlist file [%s]", "$urpm->{statedir}/$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_hdlist("$urpm->{statedir}/$medium->{hdlist}", packing => 1) }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{error}(N("problem reading synthesis file of medium \"%s\"", $medium->{name})); $medium->{ignore} = 1; } } next; } } } else { $options{force} < 2 and $options{force} = 2; } #- if copying hdlist has failed, try to build it directly. if ($error) { $options{force} < 2 and $options{force} = 2; #- clean error state now. $error = undef; } if ($options{force} < 2) { #- examine if a local list file is available (always probed according to with_hdlist #- and check hdlist has not be named very strangely... if ($medium->{hdlist} ne 'list') { my $local_list = $medium->{with_hdlist} =~ /hd(list.*)\.cz2?$/ ? $1 : 'list'; my $path_list = reduce_pathname("$with_hdlist_dir/../$local_list"); -s $path_list or $path_list = "$dir/list"; -s $path_list and system("cp", "-p", "-R", $path_list, "$urpm->{cachedir}/partial/list"); } } else { #- try to find rpm files, use recursive method, added additional #- / after dir to make sure it will be taken into account if this #- is a symlink to a directory. #- make sure rpm filename format is correct and is not a source rpm #- which are not well managed by urpmi. @files = split "\n", `find '$dir/' -name "*.rpm" -print`; #- check files contains something good! if (@files > 0) { #- we need to rebuild from rpm files the hdlist. eval { $urpm->{log}(N("reading rpm files from [%s]", $dir)); my @unresolved_before = grep { ! defined $urpm->{provides}{$_} } keys %{$urpm->{provides} || {}}; $medium->{start} = @{$urpm->{depslist}}; $medium->{headers} = [ $urpm->parse_rpms_build_headers(dir => "$urpm->{cachedir}/headers", rpms => \@files, clean => $cleaned_cache, ) ]; $medium->{end} = $#{$urpm->{depslist}}; if ($medium->{start} > $medium->{end}) { #- an error occured (provided there are files in input. delete $medium->{start}; delete $medium->{end}; die "no rpms read\n"; } else { $cleaned_cache = 0; #- make sure the headers will not be removed for another media. my @unresolved = grep { ! defined $urpm->{provides}{$_} } keys %{$urpm->{provides} || {}}; @unresolved_before == @unresolved or $second_pass = 1; } }; $@ and $error = 1, $urpm->{error}(N("unable to read rpm files from [%s]: %s", $dir, $@)); $error and delete $medium->{headers}; #- do not propagate these. $error or delete $medium->{synthesis}; #- when building hdlist by ourself, drop synthesis property. } else { $error = 1; $urpm->{error}(N("no rpm files found from [%s]", $dir)); } } } #- examine if a local pubkey file is available. if (!$options{nopubkey} && $medium->{hdlist} ne 'pubkey' && !$medium->{'key-ids'}) { my $local_pubkey = $medium->{with_hdlist} =~ /hdlist(.*)\.cz2?$/ ? "pubkey$1" : 'pubkey'; my $path_pubkey = reduce_pathname("$with_hdlist_dir/../$local_pubkey"); -s $path_pubkey or $path_pubkey = "$dir/pubkey"; -s $path_pubkey and system("cp", "-p", "-R", $path_pubkey, "$urpm->{cachedir}/partial/pubkey"); } } else { my $basename; #- try to get the description if it has been found. unlink "$urpm->{cachedir}/partial/descriptions"; if (-e "$urpm->{statedir}/descriptions.$medium->{name}") { rename("$urpm->{statedir}/descriptions.$medium->{name}", "$urpm->{cachedir}/partial/descriptions") or system("mv", "$urpm->{statedir}/descriptions.$medium->{name}", "$urpm->{cachedir}/partial/descriptions"); } eval { $urpm->{sync}({ dir => "$urpm->{cachedir}/partial", quiet => 1, limit_rate => $options{limit_rate}, compress => $options{compress}, proxy => $urpm->{proxy} }, reduce_pathname("$medium->{url}/../descriptions")); }; if (-e "$urpm->{cachedir}/partial/descriptions") { rename("$urpm->{cachedir}/partial/descriptions", "$urpm->{statedir}/descriptions.$medium->{name}") or system("mv", "$urpm->{cachedir}/partial/descriptions", "$urpm->{statedir}/descriptions.$medium->{name}"); } #- examine if a distant MD5SUM file is available. #- this will only be done if $with_hdlist is not empty in order to use #- an existing hdlist or synthesis file, and to check if download was good. #- if no MD5SUM are available, do it as before... if ($medium->{with_hdlist}) { #- we can assume at this point a basename is existing, but it needs #- to be checked for being valid, nothing can be deduced if no MD5SUM #- file are present. $basename = basename($medium->{with_hdlist}); unlink "$urpm->{cachedir}/partial/MD5SUM"; eval { if (!$options{nomd5sum}) { $urpm->{sync}({ dir => "$urpm->{cachedir}/partial", quiet => 1, limit_rate => $options{limit_rate}, compress => $options{compress}, proxy => $urpm->{proxy} }, reduce_pathname("$medium->{url}/$medium->{with_hdlist}/../MD5SUM")); } }; if (!$@ && -s "$urpm->{cachedir}/partial/MD5SUM" > 32) { if ($options{force} >= 2) { #- force downloading the file again, else why a force option has been defined ? delete $medium->{md5sum}; } else { unless ($medium->{md5sum}) { $urpm->{log}(N("computing md5sum of existing source hdlist (or synthesis)")); if ($medium->{synthesis}) { -e "$urpm->{statedir}/synthesis.$medium->{hdlist}" and $medium->{md5sum} = (split ' ', `md5sum '$urpm->{statedir}/synthesis.$medium->{hdlist}'`)[0]; } else { -e "$urpm->{statedir}/$medium->{hdlist}" and $medium->{md5sum} = (split ' ', `md5sum '$urpm->{statedir}/$medium->{hdlist}'`)[0]; } } } if ($medium->{md5sum}) { $urpm->{log}(N("examining MD5SUM file")); local (*F, $_); open F, "$urpm->{cachedir}/partial/MD5SUM"; while () { my ($md5sum, $file) = m|(\S+)\s+(?:\./)?(\S+)| or next; #- keep md5sum got here to check download was ok ! so even if md5sum is not defined, we need #- to compute it, keep it in mind ;) $file eq $basename and $retrieved_md5sum = $md5sum; } close F; #- if an existing hdlist or synthesis file has the same md5sum, we assume the #- file are the same. #- if local md5sum is the same as distant md5sum, this means there is no need to #- download hdlist or synthesis file again. foreach (@{$urpm->{media}}) { if ($_->{md5sum} && $_->{md5sum} eq $retrieved_md5sum) { unlink "$urpm->{cachedir}/partial/$basename"; #- the medium is now considered not modified. $medium->{modified} = 0; #- hdlist or synthesis file must be linked with the other same one. #- a link is better for reducing used size of /var/lib/urpmi. if ($_ ne $medium) { $medium->{md5sum} = $_->{md5sum}; unlink "$urpm->{statedir}/synthesis.$medium->{hdlist}"; unlink "$urpm->{statedir}/$medium->{hdlist}"; symlink "synthesis.$_->{hdlist}", "$urpm->{statedir}/synthesis.$medium->{hdlist}"; symlink $_->{hdlist}, "$urpm->{statedir}/$medium->{hdlist}"; } #- as previously done, just read synthesis file here, this is enough. $urpm->{log}(N("examining synthesis file [%s]", "$urpm->{statedir}/synthesis.$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis("$urpm->{statedir}/synthesis.$medium->{hdlist}") }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{log}(N("examining hdlist file [%s]", "$urpm->{statedir}/$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_hdlist("$urpm->{statedir}/$medium->{hdlist}", packing => 1) }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{error}(N("problem reading synthesis file of medium \"%s\"", $medium->{name})); $medium->{ignore} = 1; } } #- no need to continue examining other md5sum. last; } } $medium->{modified} or next; } } else { #- at this point, we don't if a basename exists and is valid, let probe it later. $basename = undef; } } #- try to probe for possible with_hdlist parameter, unless #- it is already defined (and valid). $urpm->{log}(N("retrieving source hdlist (or synthesis) of \"%s\"...", $medium->{name})); $options{callback} && $options{callback}('retrieve', $medium->{name}); if ($options{probe_with}) { my ($suffix) = $dir =~ m|RPMS([^/]*)/*$|; foreach my $with_hdlist ($medium->{with_hdlist}, probe_with_try_list($suffix, $options{probe_with})) { $basename = basename($with_hdlist) or next; $options{force} and unlink "$urpm->{cachedir}/partial/$basename"; eval { $urpm->{sync}({ dir => "$urpm->{cachedir}/partial", quiet => 0, limit_rate => $options{limit_rate}, compress => $options{compress}, callback => $options{callback}, proxy => $urpm->{proxy} }, reduce_pathname("$medium->{url}/$with_hdlist")); }; if (!$@ && -s "$urpm->{cachedir}/partial/$basename" > 32) { $medium->{with_hdlist} = $with_hdlist; $urpm->{log}(N("found probed hdlist (or synthesis) as %s", $medium->{with_hdlist})); last; #- found a suitable with_hdlist in the list above. } } } else { $basename = basename($medium->{with_hdlist}); #- try to sync (copy if needed) local copy after restored the previous one. $options{force} and unlink "$urpm->{cachedir}/partial/$basename"; unless ($options{force}) { if ($medium->{synthesis}) { -e "$urpm->{statedir}/synthesis.$medium->{hdlist}" and system("cp", "-p", "-R", "$urpm->{statedir}/synthesis.$medium->{hdlist}", "$urpm->{cachedir}/partial/$basename"); } else { -e "$urpm->{statedir}/$medium->{hdlist}" and system("cp", "-p", "-R", "$urpm->{statedir}/$medium->{hdlist}", "$urpm->{cachedir}/partial/$basename"); } } eval { $urpm->{sync}({ dir => "$urpm->{cachedir}/partial", quiet => 0, limit_rate => $options{limit_rate}, compress => $options{compress}, callback => $options{callback}, proxy => $urpm->{proxy} }, reduce_pathname("$medium->{url}/$medium->{with_hdlist}")); }; if ($@) { $urpm->{log}(N("...retrieving failed: %s", $@)); unlink "$urpm->{cachedir}/partial/$basename"; } } #- check downloaded file has right signature. if (-s "$urpm->{cachedir}/partial/$basename" > 32 && $retrieved_md5sum) { $urpm->{log}(N("computing md5sum of retrieved source hdlist (or synthesis)")); unless ((split ' ', `md5sum '$urpm->{cachedir}/partial/$basename'`)[0] eq $retrieved_md5sum) { $urpm->{log}(N("...retrieving failed: %s", N("md5sum mismatch"))); unlink "$urpm->{cachedir}/partial/$basename"; } } if (-s "$urpm->{cachedir}/partial/$basename" > 32) { $options{callback} && $options{callback}('done', $medium->{name}); $urpm->{log}(N("...retrieving done")); unless ($options{force}) { my @sstat = stat "$urpm->{cachedir}/partial/$basename"; my @lstat = stat "$urpm->{statedir}/$medium->{hdlist}"; if ($sstat[7] == $lstat[7] && $sstat[9] == $lstat[9]) { #- the two files are considered equal here, the medium is so not modified. $medium->{modified} = 0; unlink "$urpm->{cachedir}/partial/$basename"; #- as previously done, just read synthesis file here, this is enough. $urpm->{log}(N("examining synthesis file [%s]", "$urpm->{statedir}/synthesis.$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis("$urpm->{statedir}/synthesis.$medium->{hdlist}") }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{log}(N("examining hdlist file [%s]", "$urpm->{statedir}/$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_hdlist("$urpm->{statedir}/$medium->{hdlist}", packing => 1) }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{error}(N("problem reading hdlist or synthesis file of medium \"%s\"", $medium->{name})); $medium->{ignore} = 1; } } next; } } #- the file are different, update local copy. rename("$urpm->{cachedir}/partial/$basename", "$urpm->{cachedir}/partial/$medium->{hdlist}"); #- retrieve of hdlist or synthesis has been successfull, check if a list file is available. #- and check hdlist has not be named very strangely... if ($medium->{hdlist} ne 'list') { my $local_list = $medium->{with_hdlist} =~ /hd(list.*)\.cz2?$/ ? $1 : 'list'; foreach (reduce_pathname("$medium->{url}/$medium->{with_hdlist}/../$local_list"), reduce_pathname("$medium->{url}/list"), ) { eval { $urpm->{sync}({ dir => "$urpm->{cachedir}/partial", quiet => 1, limit_rate => $options{limit_rate}, compress => $options{compress}, proxy => $urpm->{proxy} }, $_); $local_list ne 'list' && -s "$urpm->{cachedir}/partial/$local_list" and rename("$urpm->{cachedir}/partial/$local_list", "$urpm->{cachedir}/partial/list"); }; $@ and unlink "$urpm->{cachedir}/partial/list"; -s "$urpm->{cachedir}/partial/list" and last; } } #- retrieve pubkey file. if (!$options{nopubkey} && $medium->{hdlist} ne 'pubkey' && !$medium->{'key-ids'}) { my $local_pubkey = $medium->{with_hdlist} =~ /hdlist(.*)\.cz2?$/ ? "pubkey$1" : 'pubkey'; foreach (reduce_pathname("$medium->{url}/$medium->{with_hdlist}/../$local_pubkey"), reduce_pathname("$medium->{url}/pubkey"), ) { eval { $urpm->{sync}({ dir => "$urpm->{cachedir}/partial", quiet => 1, limit_rate => $options{limit_rate}, compress => $options{compress}, proxy => $urpm->{proxy} }, $_); $local_pubkey ne 'pubkey' && -s "$urpm->{cachedir}/partial/$local_pubkey" and rename("$urpm->{cachedir}/partial/$local_pubkey", "$urpm->{cachedir}/partial/pubkey"); }; $@ and unlink "$urpm->{cachedir}/partial/pubkey"; -s "$urpm->{cachedir}/partial/pubkey" and last; } } } else { $error = 1; $options{callback} && $options{callback}('failed', $medium->{name}); $urpm->{error}(N("retrieve of source hdlist (or synthesis) failed")); } } #- build list file according to hdlist used. unless ($medium->{headers} || -s "$urpm->{cachedir}/partial/$medium->{hdlist}" > 32) { $error = 1; $urpm->{error}(N("no hdlist file found for medium \"%s\"", $medium->{name})); } unless ($error || $medium->{virtual}) { #- sort list file contents according to id. my %list; if ($medium->{headers}) { #- rpm files have already been read (first pass), there is just a need to #- build list hash. foreach (@files) { m|/([^/]*\.rpm)$| or next; $list{$1} and $urpm->{error}(N("file [%s] already used in the same medium \"%s\"", $1, $medium->{name})), next; $list{$1} = "$prefix:/$_\n"; } } else { #- read first pass hdlist or synthesis, try to open as synthesis, if file #- is larger than 1MB, this is probably an hdlist else a synthesis. #- anyway, if one tries fails, try another mode. $options{callback} && $options{callback}('parse', $medium->{name}); my @unresolved_before = grep { ! defined $urpm->{provides}{$_} } keys %{$urpm->{provides} || {}}; if (!$medium->{synthesis} || -s "$urpm->{cachedir}/partial/$medium->{hdlist}" > 262144) { $urpm->{log}(N("examining hdlist file [%s]", "$urpm->{cachedir}/partial/$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_hdlist("$urpm->{cachedir}/partial/$medium->{hdlist}", 1) }; if (defined $medium->{start} && defined $medium->{end}) { delete $medium->{synthesis}; } else { $urpm->{log}(N("examining synthesis file [%s]", "$urpm->{cachedir}/partial/$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis("$urpm->{cachedir}/partial/$medium->{hdlist}") }; defined $medium->{start} && defined $medium->{end} and $medium->{synthesis} = 1; } } else { $urpm->{log}(N("examining synthesis file [%s]", "$urpm->{cachedir}/partial/$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis("$urpm->{cachedir}/partial/$medium->{hdlist}") }; if (defined $medium->{start} && defined $medium->{end}) { $medium->{synthesis} = 1; } else { $urpm->{log}(N("examining hdlist file [%s]", "$urpm->{cachedir}/partial/$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_hdlist("$urpm->{cachedir}/partial/$medium->{hdlist}", 1) }; defined $medium->{start} && defined $medium->{end} and delete $medium->{synthesis}; } } if (defined $medium->{start} && defined $medium->{end}) { $options{callback} && $options{callback}('done', $medium->{name}); } else { $error = 1; $urpm->{error}(N("unable to parse hdlist file of \"%s\"", $medium->{name})); $options{callback} && $options{callback}('failed', $medium->{name}); #- we will have to read back the current synthesis file unmodified. } unless ($error) { my @unresolved_after = grep { ! defined $urpm->{provides}{$_} } keys %{$urpm->{provides} || {}}; @unresolved_before == @unresolved_after or $second_pass = 1; if ($medium->{hdlist} ne 'list' && -s "$urpm->{cachedir}/partial/list") { local (*F, $_); open F, "$urpm->{cachedir}/partial/list"; while () { m|/([^/]*\.rpm)$| or next; $list{$1} and $urpm->{error}(N("file [%s] already used in the same medium \"%s\"", $1, $medium->{name})), next; $list{$1} = "$medium->{url}/$_"; } close F; } else { #- if url is clear and no relative list file has been downloaded, #- there is no need for a list file. if ($medium->{url} ne $medium->{clear_url}) { foreach ($medium->{start} .. $medium->{end}) { my $filename = $urpm->{depslist}[$_]->filename; $list{$filename} = "$medium->{url}/$filename\n"; } } } } } unless ($error) { if (%list) { #- write list file. local *LIST; #- make sure group and other do not have any access to this file, used to hide passwords. my $mask = umask 077; open LIST, ">$urpm->{cachedir}/partial/$medium->{list}" or $error = 1, $urpm->{error}(N("unable to write list file of \"%s\"", $medium->{name})); umask $mask; print LIST values %list; close LIST; #- check if at least something has been written into list file. if (-s "$urpm->{cachedir}/partial/$medium->{list}") { $urpm->{log}(N("writing list file for medium \"%s\"", $medium->{name})); } else { $error = 1, $urpm->{error}(N("nothing written in list file for \"%s\"", $medium->{name})); } } else { #- the flag is no more necessary. delete $medium->{list}; unlink "$urpm->{statedir}/$medium->{list}"; } } } unless ($error) { #- now... on pubkey if (-s "$urpm->{cachedir}/partial/pubkey") { $urpm->{log}(N("examining pubkey file of \"%s\"...", $medium->{name})); my %key_ids; $urpm->import_needed_pubkeys([ $urpm->parse_armored_file("$urpm->{cachedir}/partial/pubkey") ], root => $urpm->{root}, callback => sub { my (undef, undef, $k, $id, $imported) = @_; if ($id) { $key_ids{$id} = undef; $imported and $urpm->{log}(N("...imported key %s from pubkey file of \"%s\"", $id, $medium->{name})); } else { $urpm->{error}(N("unable to import pubkey file of \"%s\"", $medium->{name})); } }); keys(%key_ids) and $medium->{'key-ids'} = join ',', keys %key_ids; } } unless ($medium->{virtual}) { if ($error) { #- an error has occured for updating the medium, we have to remove tempory files. unlink "$urpm->{cachedir}/partial/$medium->{hdlist}"; $medium->{list} and unlink "$urpm->{cachedir}/partial/$medium->{list}"; #- read default synthesis (we have to make sure nothing get out of depslist). $urpm->{log}(N("examining synthesis file [%s]", "$urpm->{statedir}/synthesis.$medium->{hdlist}")); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis("$urpm->{statedir}/synthesis.$medium->{hdlist}") }; unless (defined $medium->{start} && defined $medium->{end}) { $urpm->{error}(N("problem reading synthesis file of medium \"%s\"", $medium->{name})); $medium->{ignore} = 1; } } else { #- make sure to rebuild base files and clean medium modified state. $medium->{modified} = 0; $urpm->{modified} = 1; #- but use newly created file. unlink "$urpm->{statedir}/$medium->{hdlist}"; $medium->{synthesis} and unlink "$urpm->{statedir}/synthesis.$medium->{hdlist}"; $medium->{list} and unlink "$urpm->{statedir}/$medium->{list}"; unless ($medium->{headers}) { unlink "$urpm->{statedir}/synthesis.$medium->{hdlist}"; unlink "$urpm->{statedir}/$medium->{hdlist}"; rename("$urpm->{cachedir}/partial/$medium->{hdlist}", $medium->{synthesis} ? "$urpm->{statedir}/synthesis.$medium->{hdlist}" : "$urpm->{statedir}/$medium->{hdlist}") or system("mv", "$urpm->{cachedir}/partial/$medium->{hdlist}", $medium->{synthesis} ? "$urpm->{statedir}/synthesis.$medium->{hdlist}" : "$urpm->{statedir}/$medium->{hdlist}"); } if ($medium->{list}) { rename("$urpm->{cachedir}/partial/$medium->{list}", "$urpm->{statedir}/$medium->{list}") or system("mv", "$urpm->{cachedir}/partial/$medium->{list}", "$urpm->{statedir}/$medium->{list}"); } $medium->{md5sum} = $retrieved_md5sum; #- anyway, keep it, the previous one is no more usefull. #- and create synthesis file associated. $medium->{modified_synthesis} = !$medium->{synthesis}; } } } #- some unresolved provides may force to rebuild all synthesis, #- a second pass will be necessary. if ($second_pass) { $urpm->{log}(N("performing second pass to compute dependencies\n")); $urpm->unresolved_provides_clean; } #- second pass consist of reading again synthesis or hdlist. foreach my $medium (@{$urpm->{media}}) { #- take care of modified medium only or all if all have to be recomputed. $medium->{ignore} and next; $options{callback} && $options{callback}('parse', $medium->{name}); #- a modified medium is an invalid medium, we have to read back the previous hdlist #- or synthesis which has not been modified by first pass above. if ($medium->{headers} && !$medium->{modified}) { if ($second_pass) { $urpm->{log}(N("reading headers from medium \"%s\"", $medium->{name})); ($medium->{start}, $medium->{end}) = $urpm->parse_headers(dir => "$urpm->{cachedir}/headers", headers => $medium->{headers}, ); } $urpm->{log}(N("building hdlist [%s]", "$urpm->{statedir}/$medium->{hdlist}")); #- finish building operation of hdlist. $urpm->build_hdlist(start => $medium->{start}, end => $medium->{end}, dir => "$urpm->{cachedir}/headers", hdlist => "$urpm->{statedir}/$medium->{hdlist}", ); #- synthesis need to be created for sure, since the medium has been built from rpm files. $urpm->build_synthesis(start => $medium->{start}, end => $medium->{end}, synthesis => "$urpm->{statedir}/synthesis.$medium->{hdlist}", ); $urpm->{log}(N("built hdlist synthesis file for medium \"%s\"", $medium->{name})); #- keep in mind we have modified database, sure at this point. $urpm->{modified} = 1; } elsif ($medium->{synthesis}) { if ($second_pass) { if ($medium->{virtual}) { my ($path) = $medium->{url} =~ m|^file:/*(/[^/].*[^/])/*$|; my $with_hdlist_file = "$path/$medium->{with_hdlist}"; if ($path) { $urpm->{log}(N("examining synthesis file [%s]", $with_hdlist_file)); eval { ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis($with_hdlist_file) }; } } else { $urpm->{log}(N("examining synthesis file [%s]", "$urpm->{statedir}/synthesis.$medium->{hdlist}")); ($medium->{start}, $medium->{end}) = $urpm->parse_synthesis("$urpm->{statedir}/synthesis.$medium->{hdlist}"); } } } else { if ($second_pass) { $urpm->{log}(N("examining hdlist file [%s]", "$urpm->{statedir}/$medium->{hdlist}")); ($medium->{start}, $medium->{end}) = $urpm->parse_hdlist("$urpm->{statedir}/$medium->{hdlist}", 1); } #- check if synthesis file can be built. if (($second_pass || $medium->{modified_synthesis}) && !$medium->{modified}) { unless ($medium->{virtual}) { $urpm->build_synthesis(start => $medium->{start}, end => $medium->{end}, synthesis => "$urpm->{statedir}/synthesis.$medium->{hdlist}", ); $urpm->{log}(N("built hdlist synthesis file for medium \"%s\"", $medium->{name})); } #- keep in mind we have modified database, sure at this point. $urpm->{modified} = 1; } } $options{callback} && $options{callback}('done', $medium->{name}); } #- clean headers cache directory to remove everything that is no more #- usefull according to depslist used. if ($urpm->{modified}) { if ($options{noclean}) { local (*D, $_); my %headers; opendir D, "$urpm->{cachedir}/headers"; while (defined($_ = readdir D)) { m|^([^/]*-[^-]*-[^-]*\.[^\.]*)(?::\S*)?$| and $headers{$1} = $_; } closedir D; $urpm->{log}(N("found %d headers in cache", scalar(keys %headers))); foreach (@{$urpm->{depslist}}) { delete $headers{$_->fullname}; } $urpm->{log}(N("removing %d obsolete headers in cache", scalar(keys %headers))); foreach (values %headers) { unlink "$urpm->{cachedir}/headers/$_"; } } #- this file is written in any cases. $urpm->write_config; } #- make sure names files are regenerated. foreach (@{$urpm->{media}}) { unlink "$urpm->{statedir}/names.$_->{name}"; if (defined $_->{start} && defined $_->{end}) { local *F; open F, ">$urpm->{statedir}/names.$_->{name}"; foreach ($_->{start} .. $_->{end}) { print F $urpm->{depslist}[$_]->name."\n"; } close F; } } $options{nolock} or $urpm->unlock_urpmi_db; $options{nopubkey} or $urpm->unlock_rpm_db; } #- clean params and depslist computation zone. sub clean { my ($urpm) = @_; $urpm->{depslist} = []; $urpm->{provides} = {}; foreach (@{$urpm->{media} || []}) { delete $_->{start}; delete $_->{end}; } } #- check if supermount is used. sub is_using_supermount { my ($urpm, $device_mntpoint) = @_; local (*F, $_); #- read /etc/fstab and check for existing mount point. open F, "/etc/fstab"; while () { my ($device, $mntpoint, $fstype, $options) = m|^\s*(\S+)\s+(/\S+)\s+(\S+)\s+(\S+)| or next; $mntpoint =~ s,/+,/,g; $mntpoint =~ s,/$,,; if ($fstype eq 'supermount') { $device_mntpoint eq $mntpoint and return 1; $options =~ /^(?:.*[\s,])?dev=([^\s,]+)/ && $device_mntpoint eq $1 and return 1; } } return 0; } #- find used mount point from a pathname, use a optional mode to allow #- filtering according the next operation (mount or umount). sub find_mntpoints { my ($urpm, $dir, $infos) = @_; my ($pdir, $v, %fstab, @mntpoints); local (*F, $_); #- read /etc/fstab and check for existing mount point. open F, "/etc/fstab"; while () { my ($device, $mntpoint, $fstype, $options) = m|^\s*(\S+)\s+(/\S+)\s+(\S+)\s+(\S+)| or next; $mntpoint =~ s,/+,/,g; $mntpoint =~ s,/$,,; $fstab{$mntpoint} = 0; if (ref($infos)) { if ($fstype eq 'supermount') { $options =~ /^(?:.*[\s,])?dev=([^\s,]+)/ and $infos->{$mntpoint} = { mounted => 0, device => $1, fs => $fstype, supermount => 1, }; } else { $infos->{$mntpoint} = { mounted => 0, device => $device, fs => $fstype }; } } } open F, "/etc/mtab"; while () { my ($device, $mntpoint, $fstype, $options) = m|^\s*(\S+)\s+(/\S+)\s+(\S+)\s+(\S+)| or next; $mntpoint =~ s,/+,/,g; $mntpoint =~ s,/$,,; $fstab{$mntpoint} = 1; if (ref($infos)) { if ($fstype eq 'supermount') { $options =~ /^(?:.*[\s,])?dev=([^\s,]+)/ and $infos->{$mntpoint} = { mounted => 1, device => $1, fs => $fstype, supermount => 1, }; } else { $infos->{$mntpoint} = { mounted => 1, device => $device, fs => $fstype }; } } } close F; #- try to follow symlink, too complex symlink graph may not be seen. #- check the possible mount point. my @paths = split '/', $dir; while (defined ($_ = shift @paths)) { length($_) or next; $pdir .= "/$_"; $pdir =~ s,/+,/,g; $pdir =~ s,/$,,; if (exists($fstab{$pdir})) { ref($infos) and push @mntpoints, $pdir; $infos eq 'mount' && ! $fstab{$pdir} and push @mntpoints, $pdir; $infos eq 'umount' && $fstab{$pdir} and unshift @mntpoints, $pdir; #- following symlinks may be dangerous for supermounted device and #- unusefull. #- this means it is assumed no symlink inside a removable device #- will go outside the device itself (or at least will go into #- regular already mounted device like /). #- for simplification we refuse also any other device and #- stop here. last; } elsif (-l $pdir) { while ($v = readlink $pdir) { if ($pdir =~ m|^/|) { $pdir = $v; } else { while ($v =~ m|^\.\./(.*)|) { $v = $1; $pdir =~ s|^(.*)/[^/]+/*|$1|; } $pdir .= "/$v"; } } unshift @paths, split '/', $pdir; $pdir = ''; } } @mntpoints; } #- reduce pathname by removing /.. each time it appears (or . too). sub reduce_pathname { my ($url) = @_; #- clean url to remove any macro (which cannot be solved now). #- take care if this is a true url and not a simple pathname. my ($host, $dir) = $url =~ m|([^:/]*://[^/]*/)?(.*)|; #- remove any multiple /s or trailing /. #- then split all components of pathname. $dir =~ s|/+|/|g; $dir =~ s|/$||; my @paths = split '/', $dir; #- reset $dir, recompose it, and clean trailing / added by algorithm. $dir = ''; foreach (@paths) { if ($_ eq '..') { $dir =~ s|([^/]+)/$|| or $dir .= "../"; } elsif ($_ ne '.') { $dir .= "$_/"; } } $dir =~ s|/$||; $dir ||= '/'; $host . $dir; } #- check for necessity of mounting some directory to get access sub try_mounting { my ($urpm, $dir, $removable) = @_; my %infos; $dir = reduce_pathname($dir); foreach (grep { ! $infos{$_}{mounted} && $infos{$_}{fs} ne 'supermount' } $urpm->find_mntpoints($dir, \%infos)) { $urpm->{log}(N("mounting %s", $_)); `mount '$_' 2>/dev/null`; $removable && $infos{$_}{fs} ne 'supermount' and $urpm->{removable_mounted}{$_} = undef; } -e $dir; } sub try_umounting { my ($urpm, $dir) = @_; my %infos; $dir = reduce_pathname($dir); foreach (reverse grep { $infos{$_}{mounted} && $infos{$_}{fs} ne 'supermount' } $urpm->find_mntpoints($dir, \%infos)) { $urpm->{log}(N("unmounting %s", $_)); `umount '$_' 2>/dev/null`; delete $urpm->{removable_mounted}{$_}; } ! -e $dir; } sub try_umounting_removables { my ($urpm) = @_; foreach (keys %{$urpm->{removable_mounted}}) { $urpm->try_umounting($_); } delete $urpm->{removable_mounted}; } #- relocate depslist array id to use only the most recent packages, #- reorder info hashes to give only access to best packages. sub relocate_depslist_provides { my ($urpm, %options) = @_; my $relocated_entries = $urpm->relocate_depslist; $urpm->{log}($relocated_entries ? N("relocated %s entries in depslist", $relocated_entries) : N("no entries relocated in depslist")); $relocated_entries; } #- register local packages for being installed, keep track of source. sub register_rpms { my ($urpm, @files) = @_; my ($start, $id, $error, %requested); #- examine each rpm and build the depslist for them using current #- depslist and provides environment. $start = @{$urpm->{depslist}}; foreach (@files) { /\.rpm$/ or $error = 1, $urpm->{error}(N("invalid rpm file name [%s]", $_)), next; #- allow url to be given. if (my ($basename) = m|^[^:]*:/.*/([^/]*\.rpm)$|) { unlink "$urpm->{cachedir}/partial/$basename"; eval { $urpm->{log}(N("retrieving rpm file [%s] ...", $_)); $urpm->{sync}({ dir => "$urpm->{cachedir}/partial", quiet => 1, proxy => $urpm->{proxy} }, $_); $urpm->{log}(N("...retrieving done")); $_ = "$urpm->{cachedir}/partial/$basename"; }; $@ and $urpm->{log}(N("...retrieving failed: %s", $@)); } else { -r $_ or $error = 1, $urpm->{error}(N("unable to access rpm file [%s]", $_)), next; } ($id, undef) = $urpm->parse_rpm($_); my $pkg = defined $id && $urpm->{depslist}[$id]; $pkg or $urpm->{error}(N("unable to register rpm file")), next; $urpm->{source}{$id} = $_; } $error and $urpm->{fatal}(2, N("error registering local packages")); defined $id && $start <= $id and @requested{($start .. $id)} = (1) x ($id-$start+1); #- distribute local packages to distant nodes directly in cache of each machine. @files && $urpm->{parallel_handler} and $urpm->{parallel_handler}->parallel_register_rpms(@_); %requested; } #- search packages registered by their name by storing their id into packages hash. sub search_packages { my ($urpm, $packages, $names, %options) = @_; my (%exact, %exact_a, %exact_ra, %found, %foundi); foreach my $v (@$names) { my $qv = quotemeta $v; unless ($options{fuzzy}) { #- try to search through provides. if (my @l = map { $_ && ($options{src} ? $_->arch eq 'src' : $_->is_arch_compat) && ($options{use_provides} || $_->name eq $v) && defined $_->id ? $_ : @{[]} } map { $urpm->{depslist}[$_] } keys %{$urpm->{provides}{$v} || {}}) { #- we assume that if the there is at least one package providing the resource exactly, #- this should be the best ones that is described. #- but we first check if one of the packages has the same name as searched. if (my @l2 = grep { $_->name eq $v } @l) { $exact{$v} = join '|', map { $_->id } @l2; } else { $exact{$v} = join '|', map { $_->id } @l; } next; } } if ($options{use_provides} && $options{fuzzy}) { foreach (keys %{$urpm->{provides}}) { #- search through provides to find if a provide match this one. #- but manages choices correctly (as a provides may be virtual or #- multiply defined. if (/$qv/) { my @list = grep { defined $_ } map { my $pkg = $urpm->{depslist}[$_]; $pkg && ($options{src} ? $pkg->arch eq 'src' : $pkg->arch ne 'src') ? $pkg->id : undef } keys %{$urpm->{provides}{$_} || {}}; @list > 0 and push @{$found{$v}}, join '|', @list; } if (/$qv/i) { my @list = grep { defined $_ } map { my $pkg = $urpm->{depslist}[$_]; $pkg && ($options{src} ? $pkg->arch eq 'src' : $pkg->arch ne 'src') ? $pkg->id : undef } keys %{$urpm->{provides}{$_} || {}}; @list > 0 and push @{$found{$v}}, join '|', @list; } } } foreach my $id (0 .. $#{$urpm->{depslist}}) { my $pkg = $urpm->{depslist}[$id]; ($options{src} ? $pkg->arch eq 'src' : $pkg->is_arch_compat) or next; my $pack_ra = $pkg->name . '-' . $pkg->version; my $pack_a = "$pack_ra-" . $pkg->release; my $pack = "$pack_a." . $pkg->arch; unless ($options{fuzzy}) { if ($pack eq $v) { $exact{$v} = $id; next; } elsif ($pack_a eq $v) { push @{$exact_a{$v}}, $id; next; } elsif ($pack_ra eq $v) { push @{$exact_ra{$v}}, $id; next; } } $pack =~ /$qv/ and push @{$found{$v}}, $id; $pack =~ /$qv/i and push @{$foundi{$v}}, $id; } } my $result = 1; foreach (@$names) { if (defined $exact{$_}) { $packages->{$exact{$_}} = 1; foreach (split '\|', $exact{$_}) { my $pkg = $urpm->{depslist}[$_] or next; $pkg->set_flag_skip(0); #- reset skip flag as manually selected. } } else { #- at this level, we need to search the best package given for a given name, #- always prefer already found package. my %l; foreach (@{$exact_a{$_} || $exact_ra{$_} || $found{$_} || $foundi{$_} || []}) { my $pkg = $urpm->{depslist}[$_]; push @{$l{$pkg->name}}, $pkg; } if (values(%l) == 0) { $urpm->{error}(N("no package named %s", $_)); $result = 0; } elsif (values(%l) > 1 && !$options{all}) { $urpm->{error}(N("The following packages contain %s: %s", $_, "\n".join("\n", sort { $a cmp $b } keys %l))); $result = 0; } else { foreach (values %l) { my $best; foreach (@$_) { if ($best && $best != $_) { $_->compare_pkg($best) > 0 and $best = $_; } else { $best = $_; } } $packages->{$best->id} = 1; $best->set_flag_skip(0); #- reset skip flag as manually selected. } } } } #- return true if no error have been encoutered, else false. $result; } #- do the resolution of dependencies. sub resolve_dependencies { my ($urpm, $state, $requested, %options) = @_; my $need_restart; if ($options{install_src}) { #- only src will be installed, so only update $state->{selected} according #- to src status of files. foreach (%$requested) { my $pkg = $urpm->{depslist}[$_] or next; $pkg->arch eq 'src' or next; $state->{selected}{$_} = undef; } } if ($urpm->{parallel_handler}) { #- build the global synthesis file first. my $file = "$urpm->{cachedir}/partial/parallel.cz"; unlink $file; foreach (@{$urpm->{media}}) { defined $_->{start} && defined $_->{end} or next; system "cat '$urpm->{statedir}/synthesis.$_->{hdlist}' >> $file"; } #- let each node determine what is requested, according to handler given. $urpm->{parallel_handler}->parallel_resolve_dependencies($file, @_); } else { my $db; if ($options{rpmdb}) { $db = new URPM; $db->parse_synthesis($options{rpmdb}); } else { $db = URPM::DB::open($urpm->{root}); $db or $urpm->{fatal}(9, N("unable to open rpmdb")); } my $sig_handler = sub { undef $db; exit 3 }; local $SIG{INT} = $sig_handler; local $SIG{QUIT} = $sig_handler; #- auto select package for upgrading the distribution. $options{auto_select} and $urpm->request_packages_to_upgrade($db, $state, $requested, requested => undef); #- resolve dependencies which will be examined for packages needed to #- updated with a restart needed. $urpm->resolve_requested($db, $state, $requested, %options); if ($options{priority_upgrade} && !$options{rpmdb}) { my (%priority_upgrade, %priority_requested); @priority_upgrade{split ',', $options{priority_upgrade}} = (); #- try to find if a priority upgrade should be tried, this is erwan feature he waited for months :) #- this can be also considered as a special gift... foreach (keys %{$state->{selected}}) { my $pkg = $urpm->{depslist}[$_] or next; exists $priority_upgrade{$pkg->name} or next; $priority_requested{$pkg->id} = undef; } if (%priority_requested) { my %priority_state; $urpm->resolve_requested($db, \%priority_state, \%priority_requested, %options); if (grep { ! exists $priority_state{selected}{$_} } keys %priority_requested) { #- some packages which were selected previously have not been selected, strange! $need_restart = 0; } elsif (grep { ! exists $priority_state{selected}{$_} } keys %{$state->{selected}}) { #- there are other packages to install after this priority transaction. %$state = %priority_state; $need_restart = 1; } } } } #- allow caller to know if it should try to restart. $need_restart; } sub create_transaction { my ($urpm, $state, %options) = @_; if ($urpm->{parallel_handler} || !$options{split_length} || $options{nodeps} || keys %{$state->{selected}} < $options{split_level}) { #- build simplest transaction (no split). $urpm->build_transaction_set(undef, $state, split_length => 0); } else { my $db; if ($options{rpmdb}) { $db = new URPM; $db->parse_synthesis($options{rpmdb}); } else { $db = URPM::DB::open($urpm->{root}); $db or $urpm->{fatal}(9, N("unable to open rpmdb")); } my $sig_handler = sub { undef $db; exit 3 }; local $SIG{INT} = $sig_handler; local $SIG{QUIT} = $sig_handler; #- build transaction set... $urpm->build_transaction_set($db, $state, split_length => $options{split_length}); } } #- get list of package that should not be upgraded. sub get_packages_list { my ($urpm, $file, $extra) = @_; my %val; local ($_, *F); open F, $file; while () { chomp; s/#.*$//; s/^\s*//; s/\s*$//; if (my ($n, $s) = /^([^\s\[]+)(?:\[\*\])?\[?\s*([^\s\]]*\s*[^\s\]]*)/) { $val{$n}{$s} = undef; } } close F; #- additional skipping from given parameter. foreach (split ',', $extra) { if (my ($n, $s) = /^([^\s\[]+)(?:\[\*\])?\[?\s*([^\s\]]*\s*[^\s\]]*)/) { $val{$n}{$s} = undef; } } \%val; } #- for compability... sub get_unwanted_packages { my ($urpm, $skip) = @_; print STDERR "calling obsoleted method urpm::get_unwanted_packages\n"; get_packages_list($urpm->{skiplist}, $skip); } #- select source for package selected. #- according to keys given in the packages hash. #- return a list of list containing the source description for each rpm, #- match exactly the number of medium registered, ignored medium always #- have a null list. sub get_source_packages { my ($urpm, $packages, %options) = @_; my ($id, $error, @list_error, %protected_files, %local_sources, @list, %fullname2id, %file2fullnames, %examined); local (*D, *F, $_); #- build association hash to retrieve id and examine all list files. foreach (keys %$packages) { my $p = $urpm->{depslist}[$_]; if ($urpm->{source}{$_}) { $protected_files{$local_sources{$_} = $urpm->{source}{$_}} = undef; } else { $fullname2id{$p->fullname} = $_.''; } } #- examine each medium to search for packages. #- now get rpm file name in hdlist to match list file. foreach my $pkg (@{$urpm->{depslist} || []}) { $file2fullnames{$pkg->filename}{$pkg->fullname} = undef; } #- examine the local repository, which is trusted (no gpg or pgp signature check but md5 is now done). opendir D, "$urpm->{cachedir}/rpms"; while (defined($_ = readdir D)) { if (my ($filename) = m|^([^/]*\.rpm)$|) { my $filepath = "$urpm->{cachedir}/rpms/$filename"; if (!$options{clean_all} && -s $filepath) { if (keys(%{$file2fullnames{$filename} || {}}) > 1) { $urpm->{error}(N("there are multiple packages with the same rpm filename \"%s\""), $filename); next; } elsif (keys(%{$file2fullnames{$filename} || {}}) == 1) { my ($fullname) = keys(%{$file2fullnames{$filename} || {}}); if (defined($id = delete $fullname2id{$fullname})) { $local_sources{$id} = $filepath; } else { $options{clean_other} && ! exists $protected_files{$filepath} and unlink $filepath; } } else { $options{clean_other} && ! exists $protected_files{$filepath} and unlink $filepath; } } else { #- this file should be removed or is already empty. unlink $filepath; } } #- no error on unknown filename located in cache (because .listing) inherited from old urpmi } closedir D; #- clean download directory, do it here even if this is not the best moment. if ($options{clean_all}) { system("rm", "-rf", "$urpm->{cachedir}/partial"); mkdir "$urpm->{cachedir}/partial"; } foreach my $medium (@{$urpm->{media} || []}) { my (%sources, %list_examined, $list_warning); if (defined $medium->{start} && defined $medium->{end} && !$medium->{ignore}) { #- always prefer a list file is available. my $file = $medium->{list} && "$urpm->{statedir}/$medium->{list}"; if (!$file && $medium->{virtual}) { my ($dir) = $medium->{url} =~ m!^(?:removable[^:]*|file)?:/(.*)!; my $with_hdlist_dir = reduce_pathname($dir . ($medium->{with_hdlist} ? "/$medium->{with_hdlist}" : "/..")); my $local_list = $medium->{with_hdlist} =~ /hd(list.*)\.cz2?$/ ? $1 : 'list'; $file = reduce_pathname("$with_hdlist_dir/../$local_list"); -s $file or $file = "$dir/list"; } if (-r $file) { open F, $file; while () { if (my ($filename) = m|/([^/]*\.rpm)$|) { if (keys(%{$file2fullnames{$filename} || {}}) > 1) { $urpm->{error}(N("there are multiple packages with the same rpm filename \"%s\""), $filename); next; } elsif (keys(%{$file2fullnames{$filename} || {}}) == 1) { my ($fullname) = keys(%{$file2fullnames{$filename} || {}}); defined($id = $fullname2id{$fullname}) and $sources{$id} = $medium->{virtual} ? "$medium->{url}/$_" : $_; $list_examined{$fullname} = $examined{$fullname} = undef; } } else { chomp; $error = 1; $urpm->{error}(N("unable to correctly parse [%s] on value \"%s\"", $file, $_)); last; } } close F; } elsif (-e $file) { # list file exists but isn't readable # report error only if no result found, list files are only readable by root push @list_error, N("unable to access list file of \"%s\", medium ignored", $medium->{name}); next; } if (defined $medium->{url}) { foreach ($medium->{start} .. $medium->{end}) { my $pkg = $urpm->{depslist}[$_]; if (keys(%{$file2fullnames{$pkg->filename} || {}}) > 1) { $urpm->{error}(N("there are multiple packages with the same rpm filename \"%s\""), $pkg->filename); next; } elsif (keys(%{$file2fullnames{$pkg->filename} || {}}) == 1) { my ($fullname) = keys(%{$file2fullnames{$pkg->filename} || {}}); unless (exists($list_examined{$fullname})) { ++$list_warning; defined($id = $fullname2id{$fullname}) and $sources{$id} = "$medium->{url}/".$pkg->filename; $examined{$fullname} = undef; } } } $list_warning && $medium->{list} && -r "$urpm->{statedir}/$medium->{list}" and $urpm->{error}(N("medium \"%s\" uses an invalid list file: mirror is probably not up-to-date, trying to use alternate method", $medium->{name})); } elsif (!%list_examined) { $error = 1; $urpm->{error}(N("medium \"%s\" does not define any location for rpm files", $medium->{name})); } } push @list, \%sources; } #- examine package list to see if a package has not been found. foreach (grep { ! exists($examined{$_}) } keys %fullname2id) { # print list errors only once if any @list_error and map { $urpm->{error}($_) } @list_error; @list_error = (); $error = 1; $urpm->{error}(N("package %s is not found.", $_)); } $error ? @{[]} : (\%local_sources, \@list); } #- download package that may need to be downloaded. #- make sure header are available in the appropriate directory. #- change location to find the right package in the local #- filesystem for only one transaction. #- try to mount/eject removable media here. #- return a list of package ready for rpm. sub download_source_packages { my ($urpm, $local_sources, $list, %options) = @_; my %sources = %$local_sources; my %error_sources; print STDERR "calling obsoleted method urpm::download_source_packages\n"; $urpm->exlock_urpmi_db; $urpm->copy_packages_of_removable_media($list, \%sources, %options) or return; $urpm->download_packages_of_distant_media($list, \%sources, \%error_sources, %options); $urpm->unlock_urpmi_db; %sources, %error_sources; } #- lock policy concerning chroot : # - lock rpm db in chroot # - lock urpmi db in / #- safety rpm db locking mechanism sub exlock_rpm_db { my ($urpm) = @_; #- avoid putting a require on Fcntl ':flock' (which is perl and not perl-base). my ($LOCK_EX, $LOCK_NB) = (2, 4); #- lock urpmi database, but keep lock to wait for an urpmi.update to finish. open RPMLOCK_FILE, ">$urpm->{root}/$urpm->{statedir}/.RPMLOCK"; flock RPMLOCK_FILE, $LOCK_EX|$LOCK_NB or $urpm->{fatal}(7, N("urpmi database locked")); } sub shlock_rpm_db { my ($urpm) = @_; #- avoid putting a require on Fcntl ':flock' (which is perl and not perl-base). my ($LOCK_SH, $LOCK_NB) = (1, 4); #- create the .LOCK file if needed (and if possible) unless (-e "$urpm->{root}/$urpm->{statedir}/.RPMLOCK") { open RPMLOCK_FILE, ">$urpm->{root}/$urpm->{statedir}/.RPMLOCK"; close RPMLOCK_FILE; } #- lock urpmi database, if the LOCK file doesn't exists no share lock. open RPMLOCK_FILE, "$urpm->{root}/$urpm->{statedir}/.RPMLOCK" or return; flock RPMLOCK_FILE, $LOCK_SH|$LOCK_NB or $urpm->{fatal}(7, N("urpmi database locked")); } sub unlock_rpm_db { my ($urpm) = @_; #- avoid putting a require on Fcntl ':flock' (which is perl and not perl-base). my $LOCK_UN = 8; #- now everything is finished. system("sync"); #- release lock on database. flock RPMLOCK_FILE, $LOCK_UN; close RPMLOCK_FILE; } sub exlock_urpmi_db { my ($urpm) = @_; #- avoid putting a require on Fcntl ':flock' (which is perl and not perl-base). my ($LOCK_EX, $LOCK_NB) = (2, 4); #- lock urpmi database, but keep lock to wait for an urpmi.update to finish. open LOCK_FILE, ">$urpm->{statedir}/.LOCK"; flock LOCK_FILE, $LOCK_EX|$LOCK_NB or $urpm->{fatal}(7, N("urpmi database locked")); } sub shlock_urpmi_db { my ($urpm) = @_; #- avoid putting a require on Fcntl ':flock' (which is perl and not perl-base). my ($LOCK_SH, $LOCK_NB) = (1, 4); #- create the .LOCK file if needed (and if possible) unless (-e "$urpm->{statedir}/.LOCK") { open LOCK_FILE, ">$urpm->{statedir}/.LOCK"; close LOCK_FILE; } #- lock urpmi database, if the LOCK file doesn't exists no share lock. open LOCK_FILE, "$urpm->{statedir}/.LOCK" or return; flock LOCK_FILE, $LOCK_SH|$LOCK_NB or $urpm->{fatal}(7, N("urpmi database locked")); } sub unlock_urpmi_db { my ($urpm) = @_; #- avoid putting a require on Fcntl ':flock' (which is perl and not perl-base). my $LOCK_UN = 8; #- now everything is finished. system("sync"); #- release lock on database. flock LOCK_FILE, $LOCK_UN; close LOCK_FILE; } sub copy_packages_of_removable_media { my ($urpm, $list, $sources, %options) = @_; my %removables; #- make sure everything is correct on input... @{$urpm->{media} || []} == @$list or return; #- examine if given medium is already inside a removable device. my $check_notfound = sub { my ($id, $dir, $removable) = @_; $dir and $urpm->try_mounting($dir, $removable); if (!$dir || -e $dir) { foreach (values %{$list->[$id]}) { chomp; m!^(removable_?[^_:]*|file):/(.*/([^/]*))! or next; unless ($dir) { $dir = $2; $urpm->try_mounting($dir, $removable); } -r $2 or return 1; } } else { return 2; } return 0; }; #- removable media have to be examined to keep mounted the one that has #- more package than other (size is better ?). my $examine_removable_medium = sub { my ($id, $device, $copy) = @_; my $medium = $urpm->{media}[$id]; if (my ($prefix, $dir) = $medium->{url} =~ m!^(removable[^:]*|file):/(.*)!) { #- the directory given does not exist or may be accessible #- by mounting some other. try to figure out these directory and #- mount everything necessary. while ($check_notfound->($id, $dir, 'removable')) { $options{ask_for_medium} or $urpm->{fatal}(4, N("medium \"%s\" is not selected", $medium->{name})); $urpm->try_umounting($dir); system("eject", $device); $options{ask_for_medium}($medium->{name}, $medium->{removable}) or $urpm->{fatal}(4, N("medium \"%s\" is not selected", $medium->{name})); } if (-e $dir) { while (my ($i, $url) = each %{$list->[$id]}) { chomp $url; my ($filepath, $filename) = $url =~ m!^(?:removable[^:]*|file):/(.*/([^/]*))! or next; if (-r $filepath) { if ($copy) { #- we should assume a possible buggy removable device... #- first copy in cache, and if the package is still good, transfert it #- to the great rpms cache. unlink "$urpm->{cachedir}/partial/$filename"; if (!system("cp", "-p", "-R", $filepath, "$urpm->{cachedir}/partial") && URPM::verify_rpm("$urpm->{cachedir}/partial/$filename", nosignatures => 1) !~ /NOT OK/) { #- now we can consider the file to be fine. unlink "$urpm->{cachedir}/rpms/$filename"; rename("$urpm->{cachedir}/partial/$filename", "$urpm->{cachedir}/rpms/$filename") or system("mv", "$urpm->{cachedir}/partial/$filename", "$urpm->{cachedir}/rpms/$filename"); -r "$urpm->{cachedir}/rpms/$filename" and $sources->{$i} = "$urpm->{cachedir}/rpms/$filename"; } } else { $sources->{$i} = $filepath; } } unless ($sources->{$i}) { #- fallback to use other method for retrieving the file later. $urpm->{error}(N("unable to read rpm file [%s] from medium \"%s\"", $filepath, $medium->{name})); } } } else { $urpm->{error}(N("medium \"%s\" is not selected", $medium->{name})); } } else { #- we have a removable device that is not removable, well... $urpm->{error}(N("incoherent medium \"%s\" marked removable but not really", $medium->{name})); } }; foreach (0..$#$list) { values %{$list->[$_]} or next; my $medium = $urpm->{media}[$_]; #- examine non removable device but that may be mounted. if ($medium->{removable}) { push @{$removables{$medium->{removable}} ||= []}, $_; } elsif (my ($prefix, $dir) = $medium->{url} =~ m!^(removable[^:]*|file):/(.*)!) { chomp $dir; -e $dir || $urpm->try_mounting($dir) or $urpm->{error}(N("unable to access medium \"%s\"", $medium->{name})), next; } } foreach my $device (keys %removables) { #- here we have only removable device. #- if more than one media use this device, we have to sort #- needed package to copy first the needed rpm files. if (@{$removables{$device}} > 1) { my @sorted_media = sort { values %{$list->[$a]} <=> values %{$list->[$b]} } @{$removables{$device}}; #- check if a removable device is already mounted (and files present). if (my ($already_mounted_medium) = grep { !$check_notfound->($_) } @sorted_media) { @sorted_media = grep { $_ ne $already_mounted_medium } @sorted_media; unshift @sorted_media, $already_mounted_medium; } #- mount all except the biggest one. foreach (@sorted_media[0 .. $#sorted_media-1]) { $examine_removable_medium->($_, $device, 'copy'); } #- now mount the last one... $removables{$device} = [ $sorted_media[-1] ]; } #- mount the removable device, only one or the important one. #- if supermount is used on the device, it is preferable to copy #- the file instead (because it is so slooooow). $examine_removable_medium->($removables{$device}[0], $device, $urpm->is_using_supermount($device) && 'copy'); } 1; } sub download_packages_of_distant_media { my ($urpm, $list, $sources, $error_sources, %options) = @_; #- get back all ftp and http accessible rpms file into the local cache #- if necessary (as used by checksig or any other reasons). foreach (0..$#$list) { my %distant_sources; #- ignore as well medium that contains nothing about the current set of files. values %{$list->[$_]} or next; #- examine all files to know what can be indexed on multiple media. while (my ($i, $url) = each %{$list->[$_]}) { #- it is trusted that the url given is acceptable, so the file can safely be ignored. defined $sources->{$i} and next; if ($url =~ m!^(removable[^:]*|file):/(.*\.rpm)$!) { if (-r $2) { $sources->{$i} = $2; } else { $error_sources->{$i} = $2; } } elsif ($url =~ m|^([^:]*):/(.*/([^/]*\.rpm))$|) { if ($options{force_local} || $1 ne 'ftp' && $1 ne 'http') { #- only ftp and http protocol supported by grpmi. $distant_sources{$i} = "$1:/$2"; } else { $sources->{$i} = "$1:/$2"; } } else { $urpm->{error}(N("malformed input: [%s]", $url)); } } #- download files from the current medium. if (%distant_sources) { eval { $urpm->{log}(N("retrieving rpm files from medium \"%s\"...", $urpm->{media}[$_]{name})); $urpm->{sync}({ dir => "$urpm->{cachedir}/partial", quiet => 0, verbose => $options{verbose}, limit_rate => $options{limit_rate}, compress => $options{compress}, callback => $options{callback}, proxy => $urpm->{proxy} }, values %distant_sources); $urpm->{log}(N("...retrieving done")); }; if ($@) { $urpm->{log}(N("...retrieving failed: %s", $@)); } #- clean files that have not been downloaded, but keep mind there #- has been problem downloading them at least once, this is #- necessary to keep track of failing download in order to #- present the error to the user. foreach my $i (keys %distant_sources) { my ($filename) = $distant_sources{$i} =~ m|/([^/]*\.rpm)$|; if ($filename && -s "$urpm->{cachedir}/partial/$filename" && URPM::verify_rpm("$urpm->{cachedir}/partial/$filename", nosignatures => 1) !~ /NOT OK/) { #- it seems the the file has been downloaded correctly and has been checked to be valid. unlink "$urpm->{cachedir}/rpms/$filename"; rename("$urpm->{cachedir}/partial/$filename", "$urpm->{cachedir}/rpms/$filename") or system("mv", "$urpm->{cachedir}/partial/$filename", "$urpm->{cachedir}/rpms/$filename"); -r "$urpm->{cachedir}/rpms/$filename" and $sources->{$i} = "$urpm->{cachedir}/rpms/$filename"; } unless ($sources->{$i}) { $error_sources->{$i} = $distant_sources{$i}; } } } } #- clean failed download which have succeeded. delete @$error_sources{keys %$sources}; 1; } #- prepare transaction. sub prepare_transaction { my ($urpm, $set, $list, $sources, $transaction_list, $transaction_sources) = @_; foreach my $id (@{$set->{upgrade}}) { my $pkg = $urpm->{depslist}[$id]; foreach (0..$#$list) { exists $list->[$_]{$id} and $transaction_list->[$_]{$id} = $list->[$_]{$id}; } exists $sources->{$id} and $transaction_sources->{$id} = $sources->{$id}; } } #- extract package that should be installed instead of upgraded, #- sources is a hash of id -> source rpm filename. sub extract_packages_to_install { my ($urpm, $sources) = @_; my %inst; foreach (keys %$sources) { my $pkg = $urpm->{depslist}[$_] or next; $pkg->flag_disable_obsolete and $inst{$pkg->id} = delete $sources->{$pkg->id}; } \%inst; } #- install logger (ala rpm) sub install_logger { my ($urpm, $type, $id, $subtype, $amount, $total) = @_; my $pkg = defined $id && $urpm->{depslist}[$id]; my $progress_size = 50; if ($subtype eq 'start') { $urpm->{logger_progress} = 0; if ($type eq 'trans') { $urpm->{logger_id} ||= 0; printf "%-28s", N("Preparing..."); } else { printf "%4d:%-23s", ++$urpm->{logger_id}, ($pkg && $pkg->name); } } elsif ($subtype eq 'stop') { if ($urpm->{logger_progress} < $progress_size) { print '#' x ($progress_size - $urpm->{logger_progress}); print "\n"; } } elsif ($subtype eq 'progress') { my $new_progress = $total > 0 ? int($progress_size * $amount / $total) : $progress_size; if ($new_progress > $urpm->{logger_progress}) { print '#' x ($new_progress - $urpm->{logger_progress}); $urpm->{logger_progress} = $new_progress; $urpm->{logger_progress} == $progress_size and print "\n"; } } } #- install packages according to each hashes (install or upgrade). sub install { my ($urpm, $remove, $install, $upgrade, %options) = @_; #- allow process to be forked now. my $pid; local (*CHILD_RETURNS, *ERROR_OUTPUT, $_); if ($options{fork}) { pipe(CHILD_RETURNS, ERROR_OUTPUT); if ($pid = fork()) { close ERROR_OUTPUT; $urpm->{log}(N("using process %d for executing transaction")); #- now get all errors from the child and return them directly. my @l; while () { chomp; if (/^::logger_id:(\d+)/) { $urpm->{logger_id} = $1; } else { push @l, $_; } } close CHILD_RETURNS; waitpid($pid, 0); #- take care of return code from transaction, an error should be returned directly. $? >> 8 and exit $? >> 8; return @l; } else { close CHILD_RETURNS; } } #- beware this can be a child process or the main process now... my $db = URPM::DB::open($urpm->{root}, !$options{test}); #- open in read/write mode unless testing installation. $db or $urpm->{fatal}(9, N("unable to open rpmdb")); my $trans = $db->create_transaction($urpm->{root}); if ($trans) { $urpm->{log}(N("created transaction for installing on %s (remove=%d, install=%d, upgrade=%d)", $urpm->{root} || '/', scalar(@{$remove || []}), scalar(values %$install), scalar(values %$upgrade))); } else { return N("unable to create transaction"); } my ($update, @l, %file2pkg) = 0; local *F; foreach (@$remove) { if ($trans->remove($_)) { $urpm->{log}(N("removing package %s", $_)); } else { $urpm->{error}(N("unable to remove package %s", $_)); } } foreach my $mode ($install, $upgrade) { foreach (keys %$mode) { my $pkg = $urpm->{depslist}[$_]; $file2pkg{$mode->{$_}} = $pkg; $pkg->update_header($mode->{$_}); if ($trans->add($pkg, update => $update, $options{excludepath} ? (excludepath => [ split ',', $options{excludepath} ]) : ())) { $urpm->{log}(N("adding package %s (id=%d, eid=%d, update=%d, file=%s)", scalar($pkg->fullname), $_, $pkg->id, $update, $mode->{$_})); } else { $urpm->{error}(N("unable to install package %s", $mode->{$_})); } } ++$update; } unless (!$options{nodeps} && (@l = $trans->check(%options)) || !$options{noorder} && (@l = $trans->order)) { #- assume default value for some parameter. $options{delta} ||= 1000; $options{callback_open} ||= sub { my ($data, $type, $id) = @_; open F, $install->{$id} || $upgrade->{$id} or $urpm->{error}(N("unable to access rpm file [%s]", $install->{$id} || $upgrade->{$id})); return fileno F; }; $options{callback_close} ||= sub { close F }; if (keys %$install || keys %$upgrade) { $options{callback_inst} ||= \&install_logger; $options{callback_trans} ||= \&install_logger; } @l = $trans->run($urpm, %options); #- in case of error or testing, do not try to check rpmdb #- for packages being upgraded or not. unless (@l || $options{test}) { #- examine the local repository to delete package which have been installed. if ($options{post_clean_cache}) { foreach (keys %$install, keys %$upgrade) { my $pkg = $urpm->{depslist}[$_]; $db->traverse_tag('name', [ $pkg->name ], sub { my ($p) = @_; $p->fullname eq $pkg->fullname or return; unlink "$urpm->{cachedir}/rpms/".$pkg->filename; }); } } } } #- now exit or return according to current status. if (defined $pid) { print ERROR_OUTPUT "::logger_id:$urpm->{logger_id}\n"; #- allow main urpmi to know transaction numbering... print ERROR_OUTPUT "$_\n" foreach @l; close ERROR_OUTPUT; #- keep safe exit now (with destructor call). exit 0; } else { return @l; } } #- install all files to node as remembered according to resolving done. sub parallel_install { my ($urpm, $remove, $install, $upgrade, %options) = @_; $urpm->{parallel_handler}->parallel_install(@_); } #- find packages to remove. sub find_packages_to_remove { my ($urpm, $state, $l, %options) = @_; if ($urpm->{parallel_handler}) { #- invoke parallel finder. $urpm->{parallel_handler}->parallel_find_remove($urpm, $state, $l, %options, find_packages_to_remove => 1); } else { my $db = URPM::DB::open($options{root}); my (@m, @notfound); $db or $urpm->{fatal}(9, N("unable to open rpmdb")); if (!$options{matches}) { foreach (@$l) { my ($n, $found); #- check if name-version-release may have been given. if (($n) = /^(.*)-[^\-]*-[^\-]*\.[^\.\-]*$/) { $db->traverse_tag('name', [ $n ], sub { my ($p) = @_; $p->fullname eq $_ or return; $urpm->resolve_rejected($db, $state, $p, removed => 1); push @m, scalar $p->fullname; $found = 1; }); $found and next; } #- check if name-version-release may have been given. if (($n) = /^(.*)-[^\-]*-[^\-]*$/) { $db->traverse_tag('name', [ $n ], sub { my ($p) = @_; join('-', ($p->fullname)[0..2]) eq $_ or return; $urpm->resolve_rejected($db, $state, $p, removed => 1); push @m, scalar $p->fullname; $found = 1; }); $found and next; } #- check if name-version may have been given. if (($n) = /^(.*)-[^\-]*$/) { $db->traverse_tag('name', [ $n ], sub { my ($p) = @_; join('-', ($p->fullname)[0..1]) eq $_ or return; $urpm->resolve_rejected($db, $state, $p, removed => 1); push @m, scalar $p->fullname; $found = 1; }); $found and next; } #- check if only name may have been given. $db->traverse_tag('name', [ $_ ], sub { my ($p) = @_; $p->name eq $_ or return; $urpm->resolve_rejected($db, $state, $p, removed => 1); push @m, scalar $p->fullname; $found = 1; }); $found and next; push @notfound, $_; } if (!$options{force} && @notfound && @$l > 1) { $options{callback_notfound} and $options{callback_notfound}->($urpm, @notfound) or return (); } } if ($options{matches} || @notfound) { my $match = join "|", map { quotemeta } @$l; #- reset what has been already found. %$state = (); @m = (); #- search for package that matches, and perform closure again. $db->traverse(sub { my ($p) = @_; $p->fullname =~ /$match/ or return; $urpm->resolve_rejected($db, $state, $p, removed => 1); push @m, scalar $p->fullname; }); if (!$options{force} && @notfound) { if (@m) { $options{callback_fuzzy} and $options{callback_fuzzy}->($urpm, $match, @m) or return (); } else { $options{callback_notfound} and $options{callback_notfound}->($urpm, @notfound) or return (); } } } #- check if something need to be removed. if ($options{callback_base} && %{$state->{rejected} || {}}) { my %basepackages; #- check if a package to be removed is a part of basesystem requires. $db->traverse_tag('whatprovides', [ 'basesystem' ], sub { my ($p) = @_; $basepackages{$p->fullname} = 0; }); foreach (grep { $state->{rejected}{$_}{removed} && !$state->{rejected}{$_}{obsoleted} } keys %{$state->{rejected}}) { exists $basepackages{$_} or next; ++$basepackages{$_}; } grep { $_ } values %basepackages and $options{callback_base}->($urpm, grep { $basepackages{$_} } keys %basepackages) || return (); } } grep { $state->{rejected}{$_}{removed} && !$state->{rejected}{$_}{obsoleted} } keys %{$state->{rejected}}; } #- remove packages from node as remembered according to resolving done. sub parallel_remove { my ($urpm, $remove, %options) = @_; my $state = {}; my $callback = sub { $urpm->{fatal}(1, "internal distributed remove fatal error") }; $urpm->{parallel_handler}->parallel_find_remove($urpm, $state, $remove, %options, callback_notfound => undef, callback_fuzzy => $callback, callback_base => $callback, ); } #- misc functions to help finding ask_unselect and ask_remove elements with their reasons translated. sub unselected_packages { my ($urpm, $state) = @_; grep { $state->{rejected}{$_}{backtrack} } keys %{$state->{rejected} || {}}; } sub translate_why_unselected { my ($urpm, $state, @l) = @_; map { my $rb = $state->{rejected}{$_}{backtrack}; my @froms = keys %{$rb->{closure} || {}}; my @unsatisfied = @{$rb->{unsatisfied} || []}; my $s = join ", ", ((map { N("due to missing %s", $_) } @froms), (map { N("due to unsatisfied %s", $_) } @unsatisfied), $rb->{promote} && !$rb->{keep} ? N("trying to promote %s", join(", ", @{$rb->{promote}})) : @{[]}, $rb->{keep} ? N("in order to keep %s", join(", ", @{$rb->{keep}})) : @{[]}, ); $_ . ($s ? " ($s)" : ''); } @l; } sub removed_packages { my ($urpm, $state) = @_; grep { $state->{rejected}{$_}{removed} && !$state->{rejected}{$_}{obsoleted} } keys %{$state->{rejected} || {}}; } sub translate_why_removed { my ($urpm, $state, @l) = @_; map { my ($from) = keys %{$state->{rejected}{$_}{closure}}; my ($whyk) = keys %{$state->{rejected}{$_}{closure}{$from}}; my ($whyv) = $state->{rejected}{$_}{closure}{$from}{$whyk}; my $frompkg = $urpm->search($from, strict_fullname => 1); my $s; for ($whyk) { /old_requested/ and $s .= N("in order to install %s", $frompkg ? scalar $frompkg->fullname : $from); /unsatisfied/ and do { foreach (@$whyv) { $s and $s .= ', '; if (/([^\[\s]*)(?:\[\*\])?(?:\[|\s+)([^\]]*)\]?$/ && $2 ne '*') { $s .= N("due to unsatisfied %s", "$1 $2"); } else { $s .= N("due to missing %s", $_); } } }; /conflicts/ and $s .= N("due to conflicts with %s", $whyv); /unrequested/ and $s .= N("unrequested"); } #- now insert the reason if available. $_ . ($s ? " ($s)" : ''); } @l; } sub check_sources_signatures { my ($urpm, $sources_install, $sources, %options) = @_; my ($medium, %invalid_sources); foreach my $id (sort { $a <=> $b } keys %$sources_install, keys %$sources) { my $filepath = $sources_install->{$id} || $sources->{$id}; my $verif = URPM::verify_rpm($filepath); if ($verif =~ /NOT OK/) { $invalid_sources{$filepath} = N("Invalid signature (%s)", $verif); } else { #print "+++$medium->{name}:$medium->{start}:$medium->{end}:$id\n"; unless ($medium && defined $medium->{start} && $medium->{start} <= $id && defined $medium->{end} && $id <= $medium->{end}) { $medium = undef; foreach (@{$urpm->{media}}) { defined $_->{start} && $_->{start} <= $id && defined $_->{end} && $id <= $_->{end} and $medium = $_, last; } } my $key_ids = $medium && $medium->{'key-ids'} || $urpm->{options}{'key-ids'}; #- check the key ids of the medium are matching (all) the given key id of the package. if ($key_ids) { my $valid_ids = 0; my $invalid_ids = 0; foreach my $key_id ($verif =~ /#(\S+)/g) { if (grep { hex($_) == hex($key_id) } split /[,\s]+/, $key_ids) { ++$valid_ids; } else { ++$invalid_ids; } } if ($invalid_ids) { $invalid_sources{$filepath} = N("Invalid Key ID (%s)", $verif); } elsif (!$valid_ids) { $invalid_sources{$filepath} = N("Missing signature (%s)", $verif); } } #- invoke check signature callback. $options{callback} and $options{callback}->($urpm, $filepath, %options, id => $id, verif => $verif, why => $invalid_sources{$filepath}); } } map { ($options{basename} ? basename($_) : $_) . ($options{translate} ? ": $invalid_sources{$_}" : "") } sort keys %invalid_sources; } #- get reason of update for packages to be updated #- use all update medias if none given sub get_updates_description { my ($urpm, @update_medias) = @_; my %update_descr; my ($cur, $section); @update_medias or @update_medias = grep { !$_->{ignore} && $_->{update} } @{$urpm->{media}}; foreach (map { cat_("$urpm->{statedir}/descriptions.$_->{name}"), '%package dummy' } @update_medias) { /^%package (.+)/ and do { exists $cur->{importance} && !member($cur->{importance}, qw(security bugfix)) and $cur->{importance} = 'normal'; $update_descr{$_} = $cur foreach @{$cur->{pkgs}}; $cur = {}; $cur->{pkgs} = [ split /\s/, $1 ]; $section = 'pkg'; next; }; /^Updated: (.+)/ && $section eq 'pkg' and $cur->{updated} = $1; /^Importance: (.+)/ && $section eq 'pkg' and $cur->{importance} = $1; /^%pre/ and do { $section = 'pre'; next }; /^%description/ and do { $section = 'description'; next }; $section eq 'pre' and $cur->{pre} .= $_; $section eq 'description' and $cur->{description} .= $_; } \%update_descr; } 1; __END__ =head1 NAME urpm - Mandrake perl tools to handle urpmi database =head1 SYNOPSYS require urpm; my $urpm = new urpm; $urpm->read_config(); $urpm->add_medium('medium_ftp', 'ftp://ftp.mirror/pub/linux/distributions/mandrake-devel/cooker/i586/Mandrake/RPMS', 'synthesis.hdlist.cz', update => 0); $urpm->add_distrib_media('stable', 'removable://mnt/cdrom', update => 1); $urpm->select_media('contrib', 'update'); $urpm->update_media(%options); $urpm->write_config(); my $urpm = new urpm; $urpm->read_config(nocheck_access => $uid > 0); foreach (grep { !$_->{ignore} } @{$urpm->{media} || []}) { $urpm->parse_synthesis($_); } if (@files) { push @names, $urpm->register_rpms(@files); } $urpm->relocate_depslist_provides(); my %packages; @names and $urpm->search_packages(\%packages, [ @names], use_provides => 1); if ($auto_select) { my (%to_remove, %keep_files); $urpm->select_packages_to_upgrade('', \%packages, \%to_remove, \%keep_files, use_parsehdlist => $complete); } $urpm->filter_packages_to_upgrade(\%packages, $ask_choice); $urpm->deselect_unwanted_packages(\%packages); my ($local_sources, $list) = $urpm->get_source_packages(\%packages); my %sources = $urpm->download_source_packages($local_sources, $list, 'force_local', $ask_medium_change); my @rpms_install = grep { $_ !~ /\.src.\.rpm/ } values %{ $urpm->extract_packages_to_install(\%sources) || {}}; my @rpms_upgrade = grep { $_ !~ /\.src.\.rpm/ } values %sources; =head1 DESCRIPTION C is used by urpmi executables to manipulate packages and media on a Linux-Mandrake distribution. =head1 SEE ALSO perl-URPM (obsolete rpmtools) package is used to manipulate at a lower level hdlist and rpm files. =head1 COPYRIGHT Copyright (C) 2000,2001,2002 MandrakeSoft This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. =cut