RFR: 8343344: Windows attach logic failed to handle a failed open on a pipe [v2]

[Alex Menkov]

On Tue, 5 Nov 2024 10:29:38 GMT, Kevin Walls <kevinw at openjdk.org> wrote:

> List of tests doesn't mention tier4 where the failures were first noticed?

Good point. Run it now, but it takes time to complete.

> When PipeChannel::open logs error, and returns false, then Win32AttachOperation::open_pipe returns that value, then Win32AttachListener::dequeue() also logs that there was an error? Is that message just a less detailed duplicate? (it states whether it's v1 or v2, but maybe that's not that interesting?)
> In the v2 case, Win32AttachListener::dequeue() will log if open succeeded, if !op->read_request(). But AttachOperation::read_request mostly logs its own errors as well. I think it's only this one: 639 if (buffer_size < 0) { 640 return false; ...where it does not log when returning false. Maybe it should log then, and Win32AttachListener::dequeue() does not need to log?

Ok, done.
Also added error logging in PipeChannel:: read and write (AttachOperation::read_request_data does not logs IO errors)

> While are are here, is it possible to address these at the same time? open/src/hotspot/share/services/attachListener.cpp
> typo "ot" -> "or" 631 case ATTACH_API_V2: // 000000 632 if (AttachListener::get_supported_version() < 2) { 633 log_error(attach)("Failed to read request: v2 is unsupported ot disabled");

Fixed.

> exact -> "exactly" 649 // Must contain exact 'buffer_size' bytes.

Fixed.


> Also, is the line 631 comment, with 3 args, misleading? (moving away from exactly 3 arguments...)

Fixed.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/21888#issuecomment-2458147001