[Tracker] could crash if many files opened at once

[diver]

If you for example go to /boot/beos/etc/Keymap, select all files and press enter rapidly several times (10-20) Tracker will crash. Before crash you could see several tracker windows with "There was an error resolving a link" message, but folder itself doesn't contain any links at all. Backtrace will follow.

[diver]

back trace

[diver]

This bug is still here in hrev21703.

[diver]

It could also freeze app_server paste from serial port:

vm_page_fault: sending team "/boot/beos/system/Tracker" 0x3d SIGSEGV, ip 0x6dc504 ("libroot.so_seg0ro" +0x6f504) stack trace:

0x002d7b06 (libbe.so_seg0ro + 0xd5b06) 0x002d74ca (libbe.so_seg0ro + 0xd54ca) 0x002d5730 (libbe.so_seg0ro + 0xd3730) 0x00551e9b (libtracker.so_seg0ro + 0xcce9b) 0x00554d9a (libtracker.so_seg0ro + 0xcfd9a) 0x005e5bbb (libtracker.so_seg0ro + 0x160bbb) 0x005e5a00 (libtracker.so_seg0ro + 0x160a00) 0x0068d74c (libroot.so_seg0ro + 0x2074c) 0x7b140fec (LaunchTask_7e4_stack + 0x3ffec)

vm_soft_fault: va 0x0 not covered by area in address space vm_page_fault: vm_soft_fault returned error 'Bad address' on fault at 0x0, ip 0x80089564, write 0, user 0, thread 0x7e4 debug_server: Thread 1518 entered the debugger: After syscall stack trace, current PC 0xffff0102 :

(0x70c6f82c) 0x2caccb _SendMessageC8BMessagelllxbR10BMessenger + 0x233 (0x70c6f8bc) 0x2cae91 _SendMessageC8BMessagelllP8BMessagexx + 0x139 (0x70c6f93c) 0x2d13a5 SendMessageC10BMessengerP8BMessageT1xx + 0x61 (0x70c6f99c) 0x2d6d22 _IsAppRegisteredC7BRosterPC9entry_reflUlPbP8app_info + 0xfa (0x70c6fa7c) 0x2d7609 _LaunchAppC7BRosterPCcPC9entry_refPC5BListiPCPCcPl + 0x1f9 (0x70c6fd8c) 0x2d5730 LaunchC7BRosterPC9entry_refPC8BMessagePl + 0x68 (0x70c6fdec) 0x551e9b GetAttrInfo8BPrivatePC5BNodePCcT2PUlPUl + 0xbab (closest symbol) (0x70c6ff1c) 0x554d9a clQ28BPrivatet24ThreeParamFunctionObject3ZPC9entry_refZPC8BMessageZb + 0x2e (0x70c6ff4c) 0x5e5bbb RunQ28BPrivate6Thread + 0x2b (0x70c6ff7c) 0x5e5a00 RunBinderQ28BPrivate12SimpleThreadPv + 0x28 (0x70c6ffac) 0x68d74c _get_next_team_info + 0x5c (closest symbol)

[diver]

This bug is still with us in hrev22867.

[diver]

Still here in hrev28457, but GDB now chokes and will not give back trace.

[diver]

Still here in hrev32617

[anevilyak]

A minor work-in-progress note: the problems in this ticket seem to stem from Tracker running out of file descriptors while executing all those launch requests in parallel threads, and not gracefully handling those error conditions. Properly fixing this is not going to be trivial though.

[mmlr]

Should be fixed in hrev35351.