crash-reporter: Send OS version at time of crash and related improvements

Add meta files to crash directory, enabling OS version at crash time.

Adding meta files also:
1) ensures atomically added crash reports
2) allows us to remove orphaned crash report payload files
   (such as core files)
3) gives us better control over the number of reports in
   crash directory

While we're here, also made these minor changes
1) send board (x86-generic, x86-mario, etc) to crash server
2) send hwclass to crash server
3) Only record crash reports when metrics are enabled.
4) No longer allow crash reporting to staging server.

BUG=6100, 5805, 5624, 6865
TEST=unit tests plus UserCrash,CrashSender,KernelCrash autotests

Change-Id: Ieea9bdc8e0680b379c65b91cc56ca0611dd0f31c

Committed: http://chrome-svn/viewvc/chromeos?view=rev&revision=613fde1

[kmixter1, 2010-09-29 17:19:20]

Add meta files to crash directory, enabling OS version at crash time.

Adding meta files also:
1) ensures atomically added crash reports
2) allows us to remove orphaned crash report payload files
   (such as core files)
3) gives us better control over the number of reports in
   crash directory

While we're here, also made these minor changes
1) send board (x86-generic, x86-mario, etc) to crash server
2) send hwclass to crash server
3) Only record crash reports when metrics are enabled.
4) No longer allow crash reporting to staging server.

BUG=6100,5805,5624
TEST=unit tests plus UserCrash,CrashSender,KernelCrash autotests

[petkov, 2010-09-29 18:46:17]

LGTM

http://codereview.chromium.org/3436029/diff/2001/3003
File crash_collector_test.cc (right):

http://codereview.chromium.org/3436029/diff/2001/3003#newcode173
crash_collector_test.cc:173: " c=d \n");
Add an assignment with an empty value? And an assignment with an empty key?

foo=
=foo

http://codereview.chromium.org/3436029/diff/2001/3003#newcode200
crash_collector_test.cc:200: printf("%s\n", i->second.c_str());
remove?

http://codereview.chromium.org/3436029/diff/2001/3005
File crash_sender (right):

http://codereview.chromium.org/3436029/diff/2001/3005#newcode140
crash_sender:140: local base=$(get_base "$1")
Any chance you'd get two files with the same base and different extensions
(.dmp/.kcrash)?

http://codereview.chromium.org/3436029/diff/2001/3005#newcode272
crash_sender:272: if [ ! -e "$(get_base "${old_file}").meta" ]; then
you call get_base old_file twice here and meta_file is unused.

http://codereview.chromium.org/3436029/diff/2001/3005#newcode309
crash_sender:309: for file in $(${FIND} "${dir}" -mindepth 1 -name \
this loop is a bit confusing but I don't have much better suggestions... maybe
other than assigning the output of find to a string and checking for an empty
string.

[kmixter1, 2010-09-30 01:49:07]

PTAL

http://codereview.chromium.org/3436029/diff/2001/3003
File crash_collector_test.cc (right):

http://codereview.chromium.org/3436029/diff/2001/3003#newcode173
crash_collector_test.cc:173: " c=d \n");
On 2010/09/29 18:46:17, petkov wrote:
> Add an assignment with an empty value? And an assignment with an empty key?
> 
> foo=
> =foo
> 

Done.

http://codereview.chromium.org/3436029/diff/2001/3003#newcode200
crash_collector_test.cc:200: printf("%s\n", i->second.c_str());
On 2010/09/29 18:46:17, petkov wrote:
> remove?
> 

Done.

http://codereview.chromium.org/3436029/diff/2001/3005
File crash_sender (right):

http://codereview.chromium.org/3436029/diff/2001/3005#newcode140
crash_sender:140: local base=$(get_base "$1")
On 2010/09/29 18:46:17, petkov wrote:
> Any chance you'd get two files with the same base and different extensions
> (.dmp/.kcrash)?

Shouldn't happen. kcrash will always start with "kernel" and "dmp" should never.
 Also, kernel crashes always have pid 0.  If that were to happen, we would
consider it a minidump report and effectively ignore the kcrash file.

Added a comment to this effect.

http://codereview.chromium.org/3436029/diff/2001/3005#newcode272
crash_sender:272: if [ ! -e "$(get_base "${old_file}").meta" ]; then
On 2010/09/29 18:46:17, petkov wrote:
> you call get_base old_file twice here and meta_file is unused.
> 

Done.

http://codereview.chromium.org/3436029/diff/2001/3005#newcode309
crash_sender:309: for file in $(${FIND} "${dir}" -mindepth 1 -name \
On 2010/09/29 18:46:17, petkov wrote:
> this loop is a bit confusing but I don't have much better suggestions... maybe
> other than assigning the output of find to a string and checking for an empty
> string.

Thanks, that is nicer.

http://codereview.chromium.org/3436029/diff/12001/13001
File crash_collector.cc (right):

http://codereview.chromium.org/3436029/diff/12001/13001#newcode61
crash_collector.cc:61: std::string CrashCollector::Sanitize(const std::string
&name) {
Function to clean up filenames.

[petkov, 2010-09-30 05:15:59]

LGTM

http://codereview.chromium.org/3436029/diff/12001/13001
File crash_collector.cc (right):

http://codereview.chromium.org/3436029/diff/12001/13001#newcode75
crash_collector.cc:75: std::string sanitized_exec_name = Sanitize(exec_name);
I guess there's an unlikely case where you'd sanitize different exec_names into
the same sanitized name. Maybe that's not a concern anyway as long as the two
don't processes don't crash the same second. :)

http://codereview.chromium.org/3436029/diff/12001/13002
File crash_collector.h (right):

http://codereview.chromium.org/3436029/diff/12001/13002#newcode49
crash_collector.h:49: // Return a filename that has only [a-z0-1_] characters.
[A-Za-z0-9_]
Also, "Replaces characters outside that set with _."