]> git.proxmox.com Git - pve-common.git/blob - src/PVE/RESTEnvironment.pm
fix #1766: compare task starttime numerically
[pve-common.git] / src / PVE / RESTEnvironment.pm
1 package PVE::RESTEnvironment;
2
3 # NOTE: you can/should provide your own specialice class, and
4 # use this a bas class (as example see PVE::RPCEnvironment).
5
6 # we use this singleton class to pass RPC related environment values
7
8 use strict;
9 use warnings;
10 use POSIX qw(:sys_wait_h EINTR);
11 use IO::Handle;
12 use IO::File;
13 use IO::Select;
14 use Fcntl qw(:flock);
15 use PVE::Exception qw(raise raise_perm_exc);
16 use PVE::SafeSyslog;
17 use PVE::Tools;
18 use PVE::INotify;
19 use PVE::ProcFSTools;
20
21
22 my $rest_env;
23
24 # save $SIG{CHLD} handler implementation.
25 # simply set $SIG{CHLD} = $worker_reaper;
26 # and register forked processes with &$register_worker(pid)
27 # Note: using $SIG{CHLD} = 'IGNORE' or $SIG{CHLD} = sub { wait (); } or ...
28 # has serious side effects, because perls built in system() and open()
29 # functions can't get the correct exit status of a child. So we cant use
30 # that (also see perlipc)
31
32 my $WORKER_PIDS;
33 my $WORKER_FLAG = 0;
34
35 my $log_task_result = sub {
36 my ($upid, $user, $status) = @_;
37
38 return if !$rest_env;
39
40 my $msg = 'successful';
41 my $pri = 'info';
42 if ($status != 0) {
43 my $ec = $status >> 8;
44 my $ic = $status & 255;
45 $msg = $ec ? "failed ($ec)" : "interrupted ($ic)";
46 $pri = 'err';
47 }
48
49 my $tlist = $rest_env->active_workers($upid);
50 eval { $rest_env->broadcast_tasklist($tlist); };
51 syslog('err', $@) if $@;
52
53 my $task;
54 foreach my $t (@$tlist) {
55 if ($t->{upid} eq $upid) {
56 $task = $t;
57 last;
58 }
59 }
60 if ($task && $task->{status}) {
61 $msg = $task->{status};
62 }
63
64 $rest_env->log_cluster_msg($pri, $user, "end task $upid $msg");
65 };
66
67 my $worker_reaper = sub {
68 local $!; local $?;
69 foreach my $pid (keys %$WORKER_PIDS) {
70 my $waitpid = waitpid ($pid, WNOHANG);
71 if (defined($waitpid) && ($waitpid == $pid)) {
72 my $info = $WORKER_PIDS->{$pid};
73 if ($info && $info->{upid} && $info->{user}) {
74 &$log_task_result($info->{upid}, $info->{user}, $?);
75 }
76 delete ($WORKER_PIDS->{$pid});
77 }
78 }
79 };
80
81 my $register_worker = sub {
82 my ($pid, $user, $upid) = @_;
83
84 return if !$pid;
85
86 # do not register if already finished
87 my $waitpid = waitpid ($pid, WNOHANG);
88 if (defined($waitpid) && ($waitpid == $pid)) {
89 delete ($WORKER_PIDS->{$pid});
90 return;
91 }
92
93 $WORKER_PIDS->{$pid} = {
94 user => $user,
95 upid => $upid,
96 };
97 };
98
99 # initialize environment - must be called once at program startup
100 sub init {
101 my ($class, $type, %params) = @_;
102
103 $class = ref($class) || $class;
104
105 die "already initialized" if $rest_env;
106
107 die "unknown environment type"
108 if !$type || $type !~ m/^(cli|pub|priv|ha)$/;
109
110 $SIG{CHLD} = $worker_reaper;
111
112 # environment types
113 # cli ... command started fron command line
114 # pub ... access from public server (apache)
115 # priv ... access from private server (pvedaemon)
116 # ha ... access from HA resource manager agent (rgmanager)
117
118 my $self = { type => $type };
119
120 bless $self, $class;
121
122 foreach my $p (keys %params) {
123 if ($p eq 'atfork') {
124 $self->{$p} = $params{$p};
125 } else {
126 die "unknown option '$p'";
127 }
128 }
129
130 $rest_env = $self;
131
132 my ($sysname, $nodename) = POSIX::uname();
133
134 $nodename =~ s/\..*$//; # strip domain part, if any
135
136 $self->{nodename} = $nodename;
137
138 return $self;
139 };
140
141 # convenience function for command line tools
142 sub setup_default_cli_env {
143 my ($class, $username) = @_;
144
145 $class = ref($class) || $class;
146
147 $username //= 'root@pam';
148
149 PVE::INotify::inotify_init();
150
151 my $rpcenv = $class->init('cli');
152 $rpcenv->init_request();
153 $rpcenv->set_language($ENV{LANG});
154 $rpcenv->set_user($username);
155
156 die "please run as root\n"
157 if ($username eq 'root@pam') && ($> != 0);
158 }
159
160 # get the singleton
161 sub get {
162
163 die "REST environment not initialized" if !$rest_env;
164
165 return $rest_env;
166 }
167
168 sub set_client_ip {
169 my ($self, $ip) = @_;
170
171 $self->{client_ip} = $ip;
172 }
173
174 sub get_client_ip {
175 my ($self) = @_;
176
177 return $self->{client_ip};
178 }
179
180 sub set_result_attrib {
181 my ($self, $key, $value) = @_;
182
183 $self->{result_attributes}->{$key} = $value;
184 }
185
186 sub get_result_attrib {
187 my ($self, $key) = @_;
188
189 return $self->{result_attributes}->{$key};
190 }
191
192 sub set_language {
193 my ($self, $lang) = @_;
194
195 # fixme: initialize I18N
196
197 $self->{language} = $lang;
198 }
199
200 sub get_language {
201 my ($self) = @_;
202
203 return $self->{language};
204 }
205
206 sub set_user {
207 my ($self, $user) = @_;
208
209 $self->{user} = $user;
210 }
211
212 sub get_user {
213 my ($self, $noerr) = @_;
214
215 return $self->{user} if defined($self->{user}) || $noerr;
216
217 die "user name not set\n";
218 }
219
220 sub is_worker {
221 my ($class) = @_;
222
223 return $WORKER_FLAG;
224 }
225
226 # read/update list of active workers
227 # we move all finished tasks to the archive index,
228 # but keep aktive and most recent task in the active file.
229 # $nocheck ... consider $new_upid still running (avoid that
230 # we try to read the reult to early.
231 sub active_workers {
232 my ($self, $new_upid, $nocheck) = @_;
233
234 my $lkfn = "/var/log/pve/tasks/.active.lock";
235
236 my $timeout = 10;
237
238 my $code = sub {
239
240 my $tasklist = PVE::INotify::read_file('active');
241
242 my @ta;
243 my $tlist = [];
244 my $thash = {}; # only list task once
245
246 my $check_task = sub {
247 my ($task, $running) = @_;
248
249 if ($running || PVE::ProcFSTools::check_process_running($task->{pid}, $task->{pstart})) {
250 push @$tlist, $task;
251 } else {
252 delete $task->{pid};
253 push @ta, $task;
254 }
255 delete $task->{pstart};
256 };
257
258 foreach my $task (@$tasklist) {
259 my $upid = $task->{upid};
260 next if $thash->{$upid};
261 $thash->{$upid} = $task;
262 &$check_task($task);
263 }
264
265 if ($new_upid && !(my $task = $thash->{$new_upid})) {
266 $task = PVE::Tools::upid_decode($new_upid);
267 $task->{upid} = $new_upid;
268 $thash->{$new_upid} = $task;
269 &$check_task($task, $nocheck);
270 }
271
272
273 @ta = sort { $b->{starttime} <=> $a->{starttime} } @ta;
274
275 my $save = defined($new_upid);
276
277 foreach my $task (@ta) {
278 next if $task->{endtime};
279 $task->{endtime} = time();
280 $task->{status} = PVE::Tools::upid_read_status($task->{upid});
281 $save = 1;
282 }
283
284 my $archive = '';
285 my @arlist = ();
286 foreach my $task (@ta) {
287 if (!$task->{saved}) {
288 $archive .= sprintf("%s %08X %s\n", $task->{upid}, $task->{endtime}, $task->{status});
289 $save = 1;
290 push @arlist, $task;
291 $task->{saved} = 1;
292 }
293 }
294
295 if ($archive) {
296 my $size = 0;
297 my $filename = "/var/log/pve/tasks/index";
298 eval {
299 my $fh = IO::File->new($filename, '>>', 0644) ||
300 die "unable to open file '$filename' - $!\n";
301 PVE::Tools::safe_print($filename, $fh, $archive);
302 $size = -s $fh;
303 close($fh) ||
304 die "unable to close file '$filename' - $!\n";
305 };
306 my $err = $@;
307 if ($err) {
308 syslog('err', $err);
309 foreach my $task (@arlist) { # mark as not saved
310 $task->{saved} = 0;
311 }
312 }
313 my $maxsize = 50000; # about 1000 entries
314 if ($size > $maxsize) {
315 rename($filename, "$filename.1");
316 }
317 }
318
319 # we try to reduce the amount of data
320 # list all running tasks and task and a few others
321 # try to limit to 25 tasks
322 my $max = 25 - scalar(@$tlist);
323 foreach my $task (@ta) {
324 last if $max <= 0;
325 push @$tlist, $task;
326 $max--;
327 }
328
329 PVE::INotify::write_file('active', $tlist) if $save;
330
331 return $tlist;
332 };
333
334 my $res = PVE::Tools::lock_file($lkfn, $timeout, $code);
335 die $@ if $@;
336
337 return $res;
338 }
339
340 my $kill_process_group = sub {
341 my ($pid, $pstart) = @_;
342
343 # send kill to process group (negative pid)
344 my $kpid = -$pid;
345
346 # always send signal to all pgrp members
347 kill(15, $kpid); # send TERM signal
348
349 # give max 5 seconds to shut down
350 for (my $i = 0; $i < 5; $i++) {
351 return if !PVE::ProcFSTools::check_process_running($pid, $pstart);
352 sleep (1);
353 }
354
355 # to be sure
356 kill(9, $kpid);
357 };
358
359 sub check_worker {
360 my ($self, $upid, $killit) = @_;
361
362 my $task = PVE::Tools::upid_decode($upid);
363
364 my $running = PVE::ProcFSTools::check_process_running($task->{pid}, $task->{pstart});
365
366 return 0 if !$running;
367
368 if ($killit) {
369 &$kill_process_group($task->{pid});
370 return 0;
371 }
372
373 return 1;
374 }
375
376 # acts almost as tee: writes an output both to STDOUT and a task log,
377 # we differ as we're worker aware and look also at the childs control pipe,
378 # so we know if the function could be executed successfully or not.
379 my $tee_worker = sub {
380 my ($childfd, $ctrlfd, $taskfh, $cpid) = @_;
381
382 eval {
383 my $int_count = 0;
384 local $SIG{INT} = local $SIG{QUIT} = local $SIG{TERM} = sub {
385 # always send signal to all pgrp members
386 my $kpid = -$cpid;
387 if ($int_count < 3) {
388 kill(15, $kpid); # send TERM signal
389 } else {
390 kill(9, $kpid); # send KILL signal
391 }
392 $int_count++;
393 };
394 local $SIG{PIPE} = sub { die "broken pipe\n"; };
395
396 my $select = new IO::Select;
397 my $fh = IO::Handle->new_from_fd($childfd, 'r');
398 $select->add($fh);
399
400 my $readbuf = '';
401 my $count;
402 while ($select->count) {
403 my @handles = $select->can_read(1);
404 if (scalar(@handles)) {
405 my $count = sysread ($handles[0], $readbuf, 4096);
406 if (!defined ($count)) {
407 my $err = $!;
408 die "sync pipe read error: $err\n";
409 }
410 last if $count == 0; # eof
411
412 print $readbuf;
413 select->flush();
414
415 print $taskfh $readbuf;
416 $taskfh->flush();
417 } else {
418 # some commands daemonize without closing stdout
419 last if !PVE::ProcFSTools::check_process_running($cpid);
420 }
421 }
422
423 # get status (error or OK)
424 POSIX::read($ctrlfd, $readbuf, 4096);
425 if ($readbuf =~ m/^TASK OK\n?$/) {
426 # skip printing to stdout
427 print $taskfh $readbuf;
428 } elsif ($readbuf =~ m/^TASK ERROR: (.*)\n?$/) {
429 print STDERR "$1\n";
430 print $taskfh "\n$readbuf"; # ensure start on new line for webUI
431 } else {
432 die "got unexpected control message: $readbuf\n";
433 }
434 $taskfh->flush();
435 };
436 my $err = $@;
437
438 POSIX::close($childfd);
439 POSIX::close($ctrlfd);
440
441 if ($err) {
442 $err =~ s/\n/ /mg;
443 print STDERR "$err\n";
444 print $taskfh "TASK ERROR: $err\n";
445 }
446 };
447
448 # start long running workers
449 # STDIN is redirected to /dev/null
450 # STDOUT,STDERR are redirected to the filename returned by upid_decode
451 # NOTE: we simulate running in foreground if ($self->{type} eq 'cli')
452 sub fork_worker {
453 my ($self, $dtype, $id, $user, $function, $background) = @_;
454
455 $dtype = 'unknown' if !defined ($dtype);
456 $id = '' if !defined ($id);
457
458 $user = 'root@pve' if !defined ($user);
459
460 my $sync = ($self->{type} eq 'cli' && !$background) ? 1 : 0;
461
462 local $SIG{INT} =
463 local $SIG{QUIT} =
464 local $SIG{PIPE} =
465 local $SIG{TERM} = 'IGNORE';
466
467 my $starttime = time ();
468
469 my @psync = POSIX::pipe();
470 my @csync = POSIX::pipe();
471 my @ctrlfd = POSIX::pipe() if $sync;
472
473 my $node = $self->{nodename};
474
475 my $cpid = fork();
476 die "unable to fork worker - $!" if !defined($cpid);
477
478 my $workerpuid = $cpid ? $cpid : $$;
479
480 my $pstart = PVE::ProcFSTools::read_proc_starttime($workerpuid) ||
481 die "unable to read process start time";
482
483 my $upid = PVE::Tools::upid_encode ({
484 node => $node, pid => $workerpuid, pstart => $pstart,
485 starttime => $starttime, type => $dtype, id => $id, user => $user });
486
487 my $outfh;
488
489 if (!$cpid) { # child
490
491 $0 = "task $upid";
492 $WORKER_FLAG = 1;
493
494 $SIG{INT} = $SIG{QUIT} = $SIG{TERM} = sub { die "received interrupt\n"; };
495
496 $SIG{CHLD} = $SIG{PIPE} = 'DEFAULT';
497
498 # set sess/process group - we want to be able to kill the
499 # whole process group
500 POSIX::setsid();
501
502 POSIX::close ($psync[0]);
503 POSIX::close ($ctrlfd[0]) if $sync;
504 POSIX::close ($csync[1]);
505
506 $outfh = $sync ? $psync[1] : undef;
507 my $resfh = $sync ? $ctrlfd[1] : undef;
508
509 eval {
510 PVE::INotify::inotify_close();
511
512 if (my $atfork = $self->{atfork}) {
513 &$atfork();
514 }
515
516 # same algorythm as used inside SA
517 # STDIN = /dev/null
518 my $fd = fileno (STDIN);
519
520 if (!$sync) {
521 close STDIN;
522 POSIX::close(0) if $fd != 0;
523
524 die "unable to redirect STDIN - $!"
525 if !open(STDIN, "</dev/null");
526
527 $outfh = PVE::Tools::upid_open($upid);
528 $resfh = fileno($outfh);
529 }
530
531
532 # redirect STDOUT
533 $fd = fileno(STDOUT);
534 close STDOUT;
535 POSIX::close (1) if $fd != 1;
536
537 die "unable to redirect STDOUT - $!"
538 if !open(STDOUT, ">&", $outfh);
539
540 STDOUT->autoflush (1);
541
542 # redirect STDERR to STDOUT
543 $fd = fileno (STDERR);
544 close STDERR;
545 POSIX::close(2) if $fd != 2;
546
547 die "unable to redirect STDERR - $!"
548 if !open(STDERR, ">&1");
549
550 STDERR->autoflush(1);
551 };
552 if (my $err = $@) {
553 my $msg = "ERROR: $err";
554 POSIX::write($psync[1], $msg, length ($msg));
555 POSIX::close($psync[1]);
556 POSIX::_exit(1);
557 kill(-9, $$);
558 }
559
560 # sync with parent (signal that we are ready)
561 POSIX::write($psync[1], $upid, length ($upid));
562 POSIX::close($psync[1]) if !$sync; # don't need output pipe if async
563
564 eval {
565 my $readbuf = '';
566 # sync with parent (wait until parent is ready)
567 POSIX::read($csync[0], $readbuf, 4096);
568 die "parent setup error\n" if $readbuf ne 'OK';
569
570 if ($self->{type} eq 'ha') {
571 print "task started by HA resource agent\n";
572 }
573 &$function($upid);
574 };
575 my $err = $@;
576 if ($err) {
577 chomp $err;
578 $err =~ s/\n/ /mg;
579 syslog('err', $err);
580 my $msg = "TASK ERROR: $err\n";
581 POSIX::write($resfh, $msg, length($msg));
582 POSIX::close($resfh) if $sync;
583 POSIX::_exit(-1);
584 } else {
585 my $msg = "TASK OK\n";
586 POSIX::write($resfh, $msg, length($msg));
587 POSIX::close($resfh) if $sync;
588 POSIX::_exit(0);
589 }
590 kill(-9, $$);
591 }
592
593 # parent
594
595 POSIX::close ($psync[1]);
596 POSIX::close ($ctrlfd[1]) if $sync;
597 POSIX::close ($csync[0]);
598
599 my $readbuf = '';
600 # sync with child (wait until child starts)
601 POSIX::read($psync[0], $readbuf, 4096);
602
603 if (!$sync) {
604 POSIX::close($psync[0]);
605 &$register_worker($cpid, $user, $upid);
606 } else {
607 chomp $readbuf;
608 }
609
610 eval {
611 die "got no worker upid - start worker failed\n" if !$readbuf;
612
613 if ($readbuf =~ m/^ERROR:\s*(.+)$/m) {
614 die "starting worker failed: $1\n";
615 }
616
617 if ($readbuf ne $upid) {
618 die "got strange worker upid ('$readbuf' != '$upid') - start worker failed\n";
619 }
620
621 if ($sync) {
622 $outfh = PVE::Tools::upid_open($upid);
623 }
624 };
625 my $err = $@;
626
627 if (!$err) {
628 my $msg = 'OK';
629 POSIX::write($csync[1], $msg, length ($msg));
630 POSIX::close($csync[1]);
631
632 } else {
633 POSIX::close($csync[1]);
634 kill(-9, $cpid); # make sure it gets killed
635 die $err;
636 }
637
638 $self->log_cluster_msg('info', $user, "starting task $upid");
639
640 my $tlist = $self->active_workers($upid, $sync);
641 eval { $self->broadcast_tasklist($tlist); };
642 syslog('err', $@) if $@;
643
644 my $res = 0;
645
646 if ($sync) {
647
648 $tee_worker->($psync[0], $ctrlfd[0], $outfh, $cpid);
649
650 &$kill_process_group($cpid, $pstart); # make sure it gets killed
651
652 close($outfh);
653
654 waitpid($cpid, 0);
655 $res = $?;
656 &$log_task_result($upid, $user, $res);
657 }
658
659 return wantarray ? ($upid, $res) : $upid;
660 }
661
662 # Abstract function
663
664 sub log_cluster_msg {
665 my ($self, $pri, $user, $msg) = @_;
666
667 syslog($pri, "%s", $msg);
668
669 # PVE::Cluster::log_msg($pri, $user, $msg);
670 }
671
672 sub broadcast_tasklist {
673 my ($self, $tlist) = @_;
674
675 # PVE::Cluster::broadcast_tasklist($tlist);
676 }
677
678 sub check_api2_permissions {
679 my ($self, $perm, $username, $param) = @_;
680
681 return 1 if !$username && $perm->{user} eq 'world';
682
683 raise_perm_exc("user != null") if !$username;
684
685 return 1 if $username eq 'root@pam';
686
687 raise_perm_exc('user != root@pam') if !$perm;
688
689 return 1 if $perm->{user} && $perm->{user} eq 'all';
690
691 ##return $self->exec_api2_perm_check($perm->{check}, $username, $param)
692 ##if $perm->{check};
693
694 raise_perm_exc();
695 }
696
697 # init_request - should be called before each REST/CLI request
698 sub init_request {
699 my ($self, %params) = @_;
700
701 $self->{result_attributes} = {}
702
703 # if you nedd more, implement in subclass
704 }
705
706 1;