implement a way to abort workers
[pve-access-control.git] / PVE / RPCEnvironment.pm
CommitLineData
2c3a6c0a
DM
1package PVE::RPCEnvironment;
2
3use strict;
4use warnings;
5use POSIX ":sys_wait_h";
6use IO::File;
7use Fcntl qw(:flock);
8use PVE::SafeSyslog;
9use PVE::Tools;
10use PVE::INotify;
11use PVE::Cluster;
12use PVE::ProcFSTools;
13use PVE::AccessControl;
14
15# we use this singleton class to pass RPC related environment values
16
17my $pve_env;
18
19# save $SIG{CHLD} handler implementation.
20# simply set $SIG{CHLD} = $worker_reaper;
21# and register forked processes with &$register_worker(pid)
22# Note: using $SIG{CHLD} = 'IGNORE' or $SIG{CHLD} = sub { wait (); } or ...
23# has serious side effects, because perls built in system() and open()
24# functions can't get the correct exit status of a child. So we cant use
25# that (also see perlipc)
26
27my $WORKER_PIDS;
28
29my $log_task_result = sub {
30 my ($upid, $user, $status) = @_;
31
32 my $msg = 'successful';
33 my $pri = 'info';
34 if ($status != 0) {
35 my $ec = $status >> 8;
36 my $ic = $status & 255;
37 $msg = $ec ? "failed ($ec)" : "interrupted ($ic)";
38 $pri = 'err';
39 }
40 my $tlist = active_workers($upid);
41 PVE::Cluster::broadcast_tasklist($tlist);
42 my $task;
43 foreach my $t (@$tlist) {
44 if ($t->{upid} eq $upid) {
45 $task = $t;
46 last;
47 }
48 }
49 if ($task && $task->{status}) {
50 $msg = $task->{status};
51 }
52 PVE::Cluster::log_msg($pri, $user, "end task $upid $msg");
53};
54
55my $worker_reaper = sub {
56 local $!; local $?;
57 foreach my $pid (keys %$WORKER_PIDS) {
58 my $waitpid = waitpid ($pid, WNOHANG);
59 if (defined($waitpid) && ($waitpid == $pid)) {
60 my $info = $WORKER_PIDS->{$pid};
61 if ($info && $info->{upid} && $info->{user}) {
62 &$log_task_result($info->{upid}, $info->{user}, $?);
63 }
64 delete ($WORKER_PIDS->{$pid});
65 }
66 }
67};
68
69my $register_worker = sub {
70 my ($pid, $user, $upid) = @_;
71
72 return if !$pid;
73
74 # do not register if already finished
75 my $waitpid = waitpid ($pid, WNOHANG);
76 if (defined($waitpid) && ($waitpid == $pid)) {
77 delete ($WORKER_PIDS->{$pid});
78 return;
79 }
80
81 $WORKER_PIDS->{$pid} = {
82 user => $user,
83 upid => $upid,
84 };
85};
86
87# ACL cache
88
89my $compile_acl = sub {
90 my ($self, $user) = @_;
91
92 my $res = {};
93 my $cfg = $self->{user_cfg};
94
95 return undef if !$cfg->{roles};
96
97 if ($user eq 'root@pam') { # root can do anything
98 return {'/' => $cfg->{roles}->{'Administrator'}};
99 }
100
101 foreach my $path (sort keys %{$cfg->{acl}}) {
102 my @ra = PVE::AccessControl::roles($cfg, $user, $path);
103
104 my $privs = {};
105 foreach my $role (@ra) {
106 if (my $privset = $cfg->{roles}->{$role}) {
107 foreach my $p (keys %$privset) {
108 $privs->{$p} = 1;
109 }
110 }
111 }
112
113 $res->{$path} = $privs;
114 }
115
116 return $res;
117};
118
119sub permissions {
120 my ($self, $user, $path) = @_;
121
122 $user = PVE::AccessControl::verify_username($user, 1);
123 return {} if !$user;
124
125 my $cache = $self->{aclcache};
126
127 my $acl = $cache->{$user};
128
129 if (!$acl) {
130 if (!($acl = &$compile_acl($self, $user))) {
131 return {};
132 }
133 $cache->{$user} = $acl;
134 }
135
136 my $perm;
137
138 if (!($perm = $acl->{$path})) {
139 $perm = {};
140 foreach my $p (sort keys %$acl) {
141 my $final = ($path eq $p);
142
143 next if !(($p eq '/') || $final || ($path =~ m|^$p/|));
144
145 $perm = $acl->{$p};
146 }
147 $acl->{$path} = $perm;
148 }
149
150 return $perm;
151}
152
153sub check {
154 my ($self, $user, $path, $privs) = @_;
155
156 my $perm = $self->permissions($user, $path);
157
158 foreach my $priv (@$privs) {
159 return undef if !$perm->{$priv};
160 };
161
162 return 1;
163};
164
165sub user_enabled {
166 my ($self, $user) = @_;
167
168 my $cfg = $self->{user_cfg};
169 return PVE::AccessControl::user_enabled($cfg, $user);
170}
171
172# initialize environment - must be called once at program startup
173sub init {
174 my ($class, $type, %params) = @_;
175
176 $class = ref($class) || $class;
177
178 die "already initialized" if $pve_env;
179
180 die "unknown environment type" if !$type || $type !~ m/^(cli|pub|priv)$/;
181
182 $SIG{CHLD} = $worker_reaper;
183
184 # environment types
185 # cli ... command started fron command line
186 # pub ... access from public server (apache)
187 # priv ... access from private server (pvedaemon)
188
189 my $self = {
190 user_cfg => {},
191 aclcache => {},
192 aclversion => undef,
193 type => $type,
194 };
195
196 bless $self, $class;
197
198 foreach my $p (keys %params) {
199 if ($p eq 'atfork') {
200 $self->{$p} = $params{$p};
201 } else {
202 die "unknown option '$p'";
203 }
204 }
205
206 $pve_env = $self;
207
208 my ($sysname, $nodename) = POSIX::uname();
209
210 $nodename =~ s/\..*$//; # strip domain part, if any
211
212 $self->{nodename} = $nodename;
213
214 return $self;
215};
216
217# get the singleton
218sub get {
219
220 die "not initialized" if !$pve_env;
221
222 return $pve_env;
223}
224
225# init_request - must be called before each RPC request
226sub init_request {
227 my ($self, %params) = @_;
228
229 PVE::Cluster::cfs_update();
230
231 my $userconfig; # we use this for regression tests
232 foreach my $p (keys %params) {
233 if ($p eq 'userconfig') {
234 $userconfig = $params{$p};
235 } else {
236 die "unknown parameter '$p'";
237 }
238 }
239
240 eval {
241 $self->{aclcache} = {};
242 if ($userconfig) {
243 my $ucdata = PVE::Tools::file_get_contents($userconfig);
244 my $cfg = PVE::AccessControl::parse_user_config($userconfig, $ucdata);
245 $self->{user_cfg} = $cfg;
246 } else {
247 my $ucvers = PVE::Cluster::cfs_file_version('user.cfg');
248 if (!$self->{aclcache} || !defined($self->{aclversion}) ||
249 !defined($ucvers) || ($ucvers ne $self->{aclversion})) {
250 $self->{aclversion} = $ucvers;
251 my $cfg = PVE::Cluster::cfs_read_file('user.cfg');
252 $self->{user_cfg} = $cfg;
253 }
254 }
255 };
256 if (my $err = $@) {
257 $self->{user_cfg} = {};
258 die "Unable to load access control list: $err";
259 }
260}
261
262sub set_client_ip {
263 my ($self, $ip) = @_;
264
265 $self->{client_ip} = $ip;
266}
267
268sub get_client_ip {
269 my ($self) = @_;
270
271 return $self->{client_ip};
272}
273
274sub set_result_count {
275 my ($self, $count) = @_;
276
277 $self->{result_count} = $count;
278}
279
280sub get_result_count {
281 my ($self) = @_;
282
283 return $self->{result_count};
284}
285
286sub set_language {
287 my ($self, $lang) = @_;
288
289 # fixme: initialize I18N
290
291 $self->{language} = $lang;
292}
293
294sub get_language {
295 my ($self) = @_;
296
297 return $self->{language};
298}
299
300sub set_user {
301 my ($self, $user) = @_;
302
303 # fixme: get ACLs
304
305 $self->{user} = $user;
306}
307
308sub get_user {
309 my ($self) = @_;
310
311 die "user name not set\n" if !$self->{user};
312
313 return $self->{user};
314}
315
316# read/update list of active workers
317# we move all finished tasks to the archive index,
318# but keep aktive and most recent task in the active file.
319sub active_workers {
320 my ($new_upid) = @_;
321
322 my $lkfn = "/var/log/pve/tasks/.active.lock";
323
324 my $timeout = 10;
325
326 my $code = sub {
327
328 my $tasklist = PVE::INotify::read_file('active');
329
330 my @ta;
331 my $tlist = [];
332 my $thash = {}; # only list task once
333
334 my $check_task = sub {
335 my ($task) = @_;
336
b9e47e57 337 if (PVE::ProcFSTools::check_process_running($task->{pid}, $task->{pstart})) {
2c3a6c0a
DM
338 push @$tlist, $task;
339 } else {
340 delete $task->{pid};
341 push @ta, $task;
342 }
343 delete $task->{pstart};
344 };
345
346 foreach my $task (@$tasklist) {
347 my $upid = $task->{upid};
348 next if $thash->{$upid};
349 $thash->{$upid} = $task;
350 &$check_task($task);
351 }
352
353 if ($new_upid && !(my $task = $thash->{$new_upid})) {
354 $task = PVE::Tools::upid_decode($new_upid);
355 $task->{upid} = $new_upid;
356 $thash->{$new_upid} = $task;
357 &$check_task($task);
358 }
359
360
361 @ta = sort { $b->{starttime} cmp $a->{starttime} } @ta;
362
363 my $save = defined($new_upid);
364
365 foreach my $task (@ta) {
366 next if $task->{endtime};
367 $task->{endtime} = time();
368 $task->{status} = PVE::Tools::upid_read_status($task->{upid});
369 $save = 1;
370 }
371
372 my $archive = '';
373 my @arlist = ();
374 foreach my $task (@ta) {
375 if (!$task->{saved}) {
376 $archive .= sprintf("$task->{upid} %08X $task->{status}\n", $task->{endtime});
377 $save = 1;
378 push @arlist, $task;
379 $task->{saved} = 1;
380 }
381 }
382
383 if ($archive) {
384 my $size = 0;
385 my $filename = "/var/log/pve/tasks/index";
386 eval {
387 my $fh = IO::File->new($filename, '>>', 0644) ||
388 die "unable to open file '$filename' - $!\n";
389 PVE::Tools::safe_print($filename, $fh, $archive);
390 $size = -s $fh;
391 close($fh) ||
392 die "unable to close file '$filename' - $!\n";
393 };
394 my $err = $@;
395 if ($err) {
396 syslog('err', $err);
397 foreach my $task (@arlist) { # mark as not saved
398 $task->{saved} = 0;
399 }
400 }
401 my $maxsize = 50000; # about 1000 entries
402 if ($size > $maxsize) {
403 rename($filename, "$filename.1");
404 }
405 }
406
407 # we try to reduce the amount of data
408 # list all running tasks and task and a few others
409 # try to limit to 25 tasks
410 my $ctime = time();
411 my $max = 25 - scalar(@$tlist);
412 foreach my $task (@ta) {
413 last if $max <= 0;
414 push @$tlist, $task;
415 $max--;
416 }
417
418 PVE::INotify::write_file('active', $tlist) if $save;
419
420 return $tlist;
421 };
422
423 my $res = PVE::Tools::lock_file($lkfn, $timeout, $code);
424 die $@ if $@;
425
426 return $res;
427}
428
b9e47e57
DM
429my $kill_process_group = sub {
430 my ($pid, $pstart) = @_;
431
432 # send kill to process group (negative pid)
433 my $kpid = -$pid;
434
435 # always send signal to all pgrp members
436 kill(15, $kpid); # send TERM signal
437
438 # give max 5 seconds to shut down
439 for (my $i = 0; $i < 5; $i++) {
440 return if !PVE::ProcFSTools::check_process_running($pid, $pstart);
441 sleep (1);
442 }
443
444 # to be sure
445 kill(9, $kpid);
446};
447
448sub check_worker {
449 my ($upid, $killit) = @_;
450
451 my $task = PVE::Tools::upid_decode($upid);
452
453 my $running = PVE::ProcFSTools::check_process_running($task->{pid}, $task->{pstart});
454
455 return 0 if !$running;
456
457 if ($killit) {
458 &$kill_process_group($task->{pid});
459 return 0;
460 }
461
462 return 1;
463}
464
2c3a6c0a
DM
465# start long running workers
466# STDIN is redirected to /dev/null
467# STDOUT,STDERR are redirected to the filename returned by upid_decode
468# NOTE: we simulate running in foreground if ($self->{type} eq 'cli')
469sub fork_worker {
470 my ($self, $dtype, $id, $user, $function) = @_;
471
472 $dtype = 'unknown' if !defined ($dtype);
473 $id = '' if !defined ($id);
474
475 $user = 'root@pve' if !defined ($user);
476
477 my $sync = $self->{type} eq 'cli' ? 1 : 0;
478
479 local $SIG{INT} =
480 local $SIG{QUIT} =
481 local $SIG{PIPE} =
482 local $SIG{TERM} = 'IGNORE';
483
484 my $starttime = time ();
485
486 my @psync = POSIX::pipe();
487 my @csync = POSIX::pipe();
488
489 my $node = $self->{nodename};
490
491 my $cpid = fork();
492 die "unable to fork worker - $!" if !defined($cpid);
493
494 my $workerpuid = $cpid ? $cpid : $$;
495
496 my $pstart = PVE::ProcFSTools::read_proc_starttime($workerpuid) ||
497 die "unable to read process start time";
498
499 my $upid = PVE::Tools::upid_encode ({
500 node => $node, pid => $workerpuid, pstart => $pstart,
501 starttime => $starttime, type => $dtype, id => $id, user => $user });
502
503 my $outfh;
504
505 if (!$cpid) { # child
506
507 $0 = "task $upid";
508
509 $SIG{INT} = $SIG{QUIT} = $SIG{TERM} = sub { die "received interrupt\n"; };
510
511 $SIG{CHLD} = $SIG{PIPE} = 'DEFAULT';
512
513 # set sess/process group - we want to be able to kill the
514 # whole process group
515 POSIX::setsid();
516
517 POSIX::close ($psync[0]);
518 POSIX::close ($csync[1]);
519
520 $outfh = $sync ? $psync[1] : undef;
521
522 eval {
523 PVE::INotify::inotify_close();
524
525 if (my $atfork = $self->{atfork}) {
526 &$atfork();
527 }
528
529 # same algorythm as used inside SA
530 # STDIN = /dev/null
531 my $fd = fileno (STDIN);
532 close STDIN;
533 POSIX::close(0) if $fd != 0;
534
535 die "unable to redirect STDIN - $!"
536 if !open(STDIN, "</dev/null");
537
538 $outfh = PVE::Tools::upid_open($upid) if !$sync;
539
540 # redirect STDOUT
541 $fd = fileno(STDOUT);
542 close STDOUT;
543 POSIX::close (1) if $fd != 1;
544
545 die "unable to redirect STDOUT - $!"
546 if !open(STDOUT, ">&", $outfh);
547
548 STDOUT->autoflush (1);
549
550 # redirect STDERR to STDOUT
551 $fd = fileno (STDERR);
552 close STDERR;
553 POSIX::close(2) if $fd != 2;
554
555 die "unable to redirect STDERR - $!"
556 if !open(STDERR, ">&1");
557
558 STDERR->autoflush(1);
559 };
560 if (my $err = $@) {
561 my $msg = "ERROR: $err";
562 POSIX::write($psync[1], $msg, length ($msg));
563 POSIX::close($psync[1]);
564 POSIX::_exit(1);
b9e47e57 565 kill(-9, $$);
2c3a6c0a
DM
566 }
567
f6f2d51f 568 # sync with parent (signal that we are ready)
2c3a6c0a
DM
569 if ($sync) {
570 print "$upid\n";
571 } else {
572 POSIX::write($psync[1], $upid, length ($upid));
573 POSIX::close($psync[1]);
574 }
575
576 my $readbuf = '';
577 # sync with parent (wait until parent is ready)
578 POSIX::read($csync[0], $readbuf, 4096);
579 die "parent setup error\n" if $readbuf ne 'OK';
580
581 eval { &$function($upid); };
582 my $err = $@;
583 if ($err) {
584 chomp $err;
585 $err =~ s/\n/ /mg;
586 syslog('err', $err);
587 print STDERR "TASK ERROR: $err\n";
588 POSIX::_exit(-1);
589 } else {
590 print STDERR "TASK OK\n";
b9e47e57 591 POSIX::_exit(0);
2c3a6c0a 592 }
b9e47e57 593 kill(-9, $$);
2c3a6c0a
DM
594 }
595
596 # parent
597
598 POSIX::close ($psync[1]);
599 POSIX::close ($csync[0]);
600
601 my $readbuf = '';
602 # sync with child (wait until child starts)
603 POSIX::read($psync[0], $readbuf, 4096);
604
605 if (!$sync) {
606 POSIX::close($psync[0]);
607 &$register_worker($cpid, $user, $upid);
608 } else {
609 chomp $readbuf;
610 }
611
612 eval {
613 die "got no worker upid - start worker failed\n" if !$readbuf;
614
615 if ($readbuf =~ m/^ERROR:\s*(.+)$/m) {
616 die "starting worker failed: $1\n";
617 }
618
619 if ($readbuf ne $upid) {
620 die "got strange worker upid ('$readbuf' != '$upid') - start worker failed\n";
621 }
622
623 if ($sync) {
624 $outfh = PVE::Tools::upid_open($upid);
625 }
626 };
627 my $err = $@;
628
629 if (!$err) {
630 my $msg = 'OK';
631 POSIX::write($csync[1], $msg, length ($msg));
632 POSIX::close($csync[1]);
633
634 } else {
635 POSIX::close($csync[1]);
b9e47e57 636 kill(-9, $cpid); # make sure it gets killed
2c3a6c0a
DM
637 die $err;
638 }
639
640 PVE::Cluster::log_msg('info', $user, "starting task $upid");
641
642 my $tlist = active_workers($upid);
643 PVE::Cluster::broadcast_tasklist($tlist);
644
645 my $res = 0;
646
647 if ($sync) {
648 my $count;
649 my $outbuf = '';
650 eval {
651 local $SIG{INT} =
652 local $SIG{QUIT} =
653 local $SIG{TERM} = sub { die "got interrupt\n"; };
654 local $SIG{PIPE} = sub { die "broken pipe\n"; };
655
656 while (($count = POSIX::read($psync[0], $readbuf, 4096)) && ($count > 0)) {
657 $outbuf .= $readbuf;
658 while ($outbuf =~ s/^(([^\010\r\n]*)(\r|\n|(\010)+|\r\n))//s) {
659 my $line = $1;
660 my $data = $2;
661 if ($data =~ m/^TASK OK$/) {
662 # skip
663 } elsif ($data =~ m/^TASK ERROR: (.+)$/) {
664 print STDERR "$1\n";
665 } else {
666 print $line;
667 }
668 if ($outfh) {
669 print $outfh $line;
f6f2d51f 670 $outfh->flush();
2c3a6c0a
DM
671 }
672 }
673 }
674 };
675 my $err = $@;
676
677 POSIX::close($psync[0]);
678
679 if ($outbuf) { # just to be sure
680 print $outbuf;
681 if ($outfh) {
682 print $outfh $outbuf;
683 }
684 }
685
686 if ($err) {
687 $err =~ s/\n/ /mg;
688 print STDERR "$err\n";
689 if ($outfh) {
690 print $outfh "TASK ERROR: $err\n";
691 }
2c3a6c0a
DM
692 }
693
b9e47e57
DM
694 &$kill_process_group($cpid, $pstart); # make sure it gets killed
695
2c3a6c0a
DM
696 close($outfh);
697
b9e47e57 698 waitpid($cpid, 0);
2c3a6c0a
DM
699 $res = $?;
700 &$log_task_result($upid, $user, $res);
701 }
702
703 return wantarray ? ($upid, $res) : $upid;
704}
705
7061;