| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
| |
For context see this thread:
https://groups.google.com/a/chromium.org/forum/?fromgroups#!topic/chromium-dev/RMcVNGjB4II
TBR=thakis,pkasting,jam
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@146163 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
------------------------------------------------------------------------
r842 | mark@chromium.org | 2011-10-03 15:54:28 -0400 (Mon, 03 Oct 2011) | 11 lines
Use a bootstrap subset port for the inspector, tying the subset to the
lifetime of the task to be monitored, the invoking task. This allows the
bootstrap server (in launchd) to automatically clean up the Mach server
registration when the task being monitored exits, avoiding leaks of
com.Breakpad.Inspector(pid) ports in "launchctl bslist".
BUG=chromium:28547
TEST=Handler should still crash catches, but inspector ports should no longer
show up in "launchctl bslist". They should show up under a subset port in
"launchctl bstree" instead. "launchctl bstree" must be invoked as root.
Review URL: http://breakpad.appspot.com/306001
------------------------------------------------------------------------
This gives each Breakpad instance its own bootstrap subset port as a subset
of the process' bootstrap port, and places the Breakpad inspector port into
the bootstrap subset port.
This completely eliminates leaks of on-demand Mach services advertised via the
bootstrap server.
This also reverts r34318 and r34534, "temporary" (21-month) hacks to mitigate
the leak. The temporary hacks were never completely effective against Breakpad
ports leaking from child processes.
DestructCrashReporter at process shutdown was deemed unnecessary and is being
removed. As this was the last caller to that function, the implementation is
removed as well.
This is addressed in Breakpad rather than Chrome because a mechanism to fix it
without changing the process' bootstrap port was discovered. It was determined
that the rohitfork port is already immune to leaks of the sort that the
Breakpad inspector port experiences. A previous attempt to fix this bug was
r103089, but that caused bug 98550. This strategy is safer.
The forked Breakpad module.cc is updated to match upstream r835, which allows
the unforking of the CFI-disabling code. A new CFI-disabling "-c" option to
dump_syms is now available and used by the dump_product_syms script.
BUG=28547
TEST=1. "launchctl bslist" should no longer show on-demand
com.Breakpad.Inspector ports (in Breakpad-enabled builds with Breakpad
on) ports.
2. "launchctl bstree" (as root) should reveal a bootstrap subset for each
process as a child of the per-user/per-session bootstrap namespace if
crash reporting is on. One com.Breakpad.Inspector port should show up
as a child of each bootstrap subset. As each process exits (even if
mercilessly killed by "kill -9"), the associated subsets and inspector
ports should disappear.
3. Breakpad reports should be generated on crashes. For example,
about:crash and about:inducebrowsercrashforrealz should each cause a
minidump to be written in
~/Library/Application Support/Google/Chrome/Crash Reports
when Breakpad is enabled. This tests that the Breakpad ports are
functioning properly.
4. The browser process should be able to access child process data.
Window:Task Manager should show valid values for the Memory, CPU, and
Network columns for all child processes. This tests that the rohitfork
port is functioning properly. Note that this version of the change
does not change the handling of rohitfork ports at all, but this test
case was useful in a previous version of this patch.
5. Test case from bug 98550: Have a link in a web browser window (in a
different browser) and drag it into Chrome. Expect the drag operation
to operate properly in both the tab strip and in web content. See bug
98550 for details.
6. Unreported test case that failed in r103089: browser relaunch should
work. Visit chrome://flags, change some flags to get the "Relaunch Now"
button at the bottom (you can put the flags back to how you found them
initially), and click "Relaunch Now". The browser should close and then
be relaunched.
Review URL: http://codereview.chromium.org/8120007
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@103778 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Give each process its own bootstrap subset port as a subset of its inherited
bootstrap port, and use the bootstrap subset port as the bootstrap port.
This completely eliminates leaks of on-demand Mach services advertised via the
bootstrap server.
This also reverts r34318 and r34534, "temporary" (21-month) hacks to mitigate
the leak. The temporary hacks were never completely effective against Breakpad
ports leaking from child processes.
DestructCrashReporter at process shutdown was deemed unnecessary and is being
removed. As this was the last caller to that function, the implementation is
removed as well.
This is addressed in Chrome rather than Breakpad to account for the potential
leak of rohitfork ports if the browser process crashes or is mercilessly
killed, because library code messing with the task's bootstrap port doesn't
strike me as kosher, and because the Mac Breakpad code is scheduled to be
replaced with something better that doesn't attempt to leak ports like a sieve
within a couple of months anyway.
BUG=28547, 98550
TEST=1. "launchctl bslist" should no longer show on-demand
com.Breakpad.Inspector ports (in Breakpad-enabled builds with Breakpad
on) or com.google.Chrome.rohitfork, com.google.Chrome.canary.rohitfork,
or org.chromium.Chromium.rohitfork ports.
2. "launchctl bstree" (as root) should reveal a bootstrap subset for the
browser process as a child of the per-user/per-session bootstrap
namespace containing the rohitfork port and browser's Breakpad port if
crash reporting is on. There should also be a bootstrap subset for
each child process as a child of the browser's bootstrap subset. If
crash reporting is on, each child process' bootstrap subset should
contain a Breakpad port.
3. Breakpad reports should be generated on crashes. For example,
about:crash and about:inducebrowsercrashforrealz should each cause a
minidump to be written in
~/Library/Application Support/Google/Chrome/Crash Reports
when Breakpad is enabled. This tests that the Breakpad ports are
functioning properly.
4. The browser process should be able to access child process data.
Window:Task Manager should show valid values for the Memory, CPU, and
Network columns for all child processes. This tests that the rohitfork
port is functioning properly.
Review URL: http://codereview.chromium.org/8059041
TBR=mark@chromium.org
Review URL: http://codereview.chromium.org/8086003
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@103303 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
bootstrap port, and use the bootstrap subset port as the bootstrap port.
This completely eliminates leaks of on-demand Mach services advertised via the
bootstrap server.
This also reverts r34318 and r34534, "temporary" (21-month) hacks to mitigate
the leak. The temporary hacks were never completely effective against Breakpad
ports leaking from child processes.
DestructCrashReporter at process shutdown was deemed unnecessary and is being
removed. As this was the last caller to that function, the implementation is
removed as well.
This is addressed in Chrome rather than Breakpad to account for the potential
leak of rohitfork ports if the browser process crashes or is mercilessly
killed, because library code messing with the task's bootstrap port doesn't
strike me as kosher, and because the Mac Breakpad code is scheduled to be
replaced with something better that doesn't attempt to leak ports like a sieve
within a couple of months anyway.
BUG=28547
TEST=1. "launchctl bslist" should no longer show on-demand
com.Breakpad.Inspector ports (in Breakpad-enabled builds with Breakpad
on) or com.google.Chrome.rohitfork, com.google.Chrome.canary.rohitfork,
or org.chromium.Chromium.rohitfork ports.
2. "launchctl bstree" (as root) should reveal a bootstrap subset for the
browser process as a child of the per-user/per-session bootstrap
namespace containing the rohitfork port and browser's Breakpad port if
crash reporting is on. There should also be a bootstrap subset for
each child process as a child of the browser's bootstrap subset. If
crash reporting is on, each child process' bootstrap subset should
contain a Breakpad port.
3. Breakpad reports should be generated on crashes. For example,
about:crash and about:inducebrowsercrashforrealz should each cause a
minidump to be written in
~/Library/Application Support/Google/Chrome/Crash Reports
when Breakpad is enabled. This tests that the Breakpad ports are
functioning properly.
4. The browser process should be able to access child process data.
Window:Task Manager should show valid values for the Memory, CPU, and
Network columns for all child processes. This tests that the rohitfork
port is functioning properly.
Review URL: http://codereview.chromium.org/8059041
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@103089 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Breakpad support currently resides in chrome/app/breakpad_mac.* this makes it inaccessible to lower level code that would also like to set crash keys e.g. The sandboxing infrastructure.
This CL refactors the crash key reporting code to reside in base/mac.
Logging is also added to the Sandbox code to try to track down the cause of crbug.com/94758.
BUG=95272, 94758
TEST=On official builds crash logs should contain crash keys e.g. OS version.
Review URL: http://codereview.chromium.org/7849011
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@100626 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
|
|
|
|
|
|
|
| |
BUG=78782
TEST=Monitor crashes.
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=81855
Review URL: http://codereview.chromium.org/6861004
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@82065 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
|
|
|
|
|
| |
BUG=50273
TEST=everything still builds, build is 10% faster on windows, same speed on mac/linux
TBR: erg
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@53716 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Enable Apple Crash Reporter for Browser process in Non-Debug builds.
* Introduce mac_util::IsBackgroundProcess().
* Invert logic around Breakpad Enabling/Disabling to make it clearer.
* Honor the kDisableBreakpad switch.
Our new policy:
* Breakpad is enabled/disabled based on the user stats reporting opt-in flag - this hasn't changed.
* Always disable Apple Crash in background processes (plug-in, renderer, etc).
* If Breakpad is enabled browser crashes will be passed to Crash Reporter.
* If Breakpad is NOT enabled browser [or any other foreground app] crashes will be passed to Crash Reporter in Release builds.
Effectively this means that we now enable Crash Reporter for the browser process in Chromium release builds so these will no longer crash silently with no trace.
Review URL: http://codereview.chromium.org/180048
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@24959 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Initialize crash reporting in helper processes such as the renderer process.
Renderer crash reporting stopped working in r23006 when multiple .app bundles
were introduced, because the stats collection and crash reporting preference
is presently accessed via NSUserDefaults, keyed on the bundle ID. The main
browser process and helper processes have distinct bundle IDs. In the new
scheme, only the main browser process consults this preference, and passes it
to helper processes in their command lines.
BUG=19204
TEST=When reporting is enabled, Breakpad should pick up browser and renderer
process crashes;
When reporting is enabled, renderer should not log messages like
[mmdd/hhmmss:WARNING:/path/to/breakpad_mac.mm(47)] Breakpad disabled;
When reporting is disabled, browser and renderer should both log these
messages.
Review URL: http://codereview.chromium.org/165546
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@23509 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
| |
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@20197 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
|
|
|
|
|
|
|
| |
process type.
2)Fixed another logic inversion in SetActiveRendererURL.
3)Change IsCrashReporterEnabled-> IsCrashReporterDisabled to prevent future confusion since that was the question all the callers where askign anyway.
Review URL: http://codereview.chromium.org/113737
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@16707 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
|
|
|
|
|
|
| |
* Link agasint Breakpad from public svn repo.
* Bump DEPS to new svn rev of Breakpad with fixes for 10.5 compilation.
* Fill in code to differentiate between processes types in crash report.
Review URL: http://codereview.chromium.org/88043
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@14245 0039d316-1c4b-4281-b951-d872f2087c98
|
| |
|
|
|
|
|
|
|
|
| |
* We now add all the same Metadata on OS X as we do on Windows.
* Made the code for logging URLs in crash dumps a little more x-platform.
Remove custom Breakpad hacks so we can use an unmodified version of the Framework.
Review URL: http://codereview.chromium.org/55028
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@12811 0039d316-1c4b-4281-b951-d872f2087c98
|
|
|
breakpad_mac.mm still needs some work, there's a bunch of scaffolding in there referring to a custom version of Breakpad I've got going locally.
Review URL: http://codereview.chromium.org/53075
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@12553 0039d316-1c4b-4281-b951-d872f2087c98
|
