https://bugs.kde.org/show_bug.cgi?id=393387
Milian Wolff <m...@milianw.de> changed:
What |Removed |Added
----------------------------------------------------------------------------
Latest Commit| |https://commits.kde.org/hea
| |ptrack/95a17c6a407a50b4da66
| |d97521adf94c2a1ae219
Resolution|--- |FIXED
Status|CONFIRMED |RESOLVED
--- Comment #5 from Milian Wolff <m...@milianw.de> ---
Git commit 95a17c6a407a50b4da66d97521adf94c2a1ae219 by Milian Wolff.
Committed on 27/04/2018 at 11:18.
Pushed by mwolff into branch 'master'.
Do not use stdio for buffered I/O
Instead, implement a custom buffer scheme that uses snprintf
and write directly. This way, we have the write syscall under
total control and can properly handle EINTR. See also:
http://unix.derkeiler.com/Newsgroups/comp.unix.programmer/2006-05/msg00314.html
This seems to fix the issue where the data file is corrupted. It
seems to be due to a SIGCHLD signal being received while we were
inside an fprintf syscall:
perf trace --no-syscalls record -m 16M \
-e signal:signal_deliver \
-e signal:signal_generate \
-e sched:sched_process_exit \
--call-graph dwarf \
./AppRun -s test
...
# child process of kdevelop exits:
2443.428 sched:sched_process_exit:comm=docker pid=15924 prio=120
# SIGCHILD is handled by a thread that's inside fprintf:
2444.617 signal:signal_deliver:sig=17 errno=0 code=1 sa_handler=7fa121163570
sa_flags=4000005
get_signal ([kernel.kallsyms])
__GI___libc_write
(/usr/lib/libc-2.26.so)
_IO_new_file_write (inlined)
new_do_write (/usr/lib/libc-2.26.so)
_IO_new_do_write (inlined)
_IO_new_file_xsputn (inlined)
_IO_vfprintf_internal (inlined)
___fprintf_chk (inlined)
fprintf (inlined)
writeLine<long unsigned int, unsigned
int, long unsigned int> (inlined)
handleMalloc (inlined)
...
# eventually heaptrack_interpret receives partially written data:
failed to parse line: t 7fa11060b9- 30a09e0
heaptrack stats:
allocations: 534879
leaked allocations: 150033
temporary allocations: 78765
# and then exists:
3580.135 sched:sched_process_exit:comm=heaptrack_inter pid=15913 prio=120
# leading to a SIGPIPE:
3580.628 signal:signal_generate:sig=13 errno=0 code=0 comm=kdevelop pid=15915
grp=0 res=1
__send_signal ([kernel.kallsyms])
__send_signal ([kernel.kallsyms])
__GI___libc_write
(/usr/lib/libc-2.26.so)
_IO_new_file_write (inlined)
new_do_write (/usr/lib/libc-2.26.so)
_IO_new_do_write (inlined)
_IO_new_file_xsputn (inlined)
_IO_vfprintf_internal (inlined)
___fprintf_chk (inlined)
fprintf (inlined)
writeLine<char const*, long unsigned
int> (inlined)
It seems to me as if this is the explanation for this issue.
Ultimately it boils down to fprintf not being save for use when
you may receive sigchild leading to EINTR.
M +9 -1 src/track/heaptrack_inject.cpp
M +113 -86 src/track/libheaptrack.cpp
M +1 -1 src/track/libheaptrack.h
M +23 -20 tests/auto/tst_libheaptrack.cpp
https://commits.kde.org/heaptrack/95a17c6a407a50b4da66d97521adf94c2a1ae219
--
You are receiving this mail because:
You are watching all bug changes.
