|
On 8/31/17 13:40, Daniel D. Daugherty
wrote:
> src/jdk.jdwp.agent/share/native/libjdwp/debugInit.c
> L574: EXIT_ERROR(map2jvmtiError(serror), "JDWP
Transport failed to initialize");
> This is a new exit code path. Previously the
process
> did not exit here. Why the change in behavior?
I think this new exit code path is going to be a problem.
Here's where startTransport is called:
L722: transport_initialize();
L723: (void)bagEnumerateOver(transports, startTransport,
&arg);
L724:
L725: /*
L726: * Exit with an error only if
L727: * 1) none of the transports was successfully
started, and
L728: * 2) the application has not yet started running
L729: */
L730: if ((arg.error != JDWP_ERROR(NONE)) &&
L731: (arg.startCount == 0) &&
L732: initOnStartup) {
L733: EXIT_ERROR(map2jvmtiError(arg.error), "No
transports initialized");
L734: }
Suppose you have more than one transport in the 'transports'
bag.
If you have an error in the first one's call to
'startTransport',
then you're going to exit at L574 and you won't get to the next
transport in the bag.
The intended error exit point for startTransport is L733.
You are right.
I fixed the issue in a wrong place.
But then I'm puzzled why the was no exit on error at L733.
I wonder if it is because of the initOnStartup.
Now, I've removed this line from the fix.
Will try to reproduce this situation again (I remember, it was
tricky) as a follow up.
Will fill new bug if needed.
Thanks,
Serguei
Dan
On 8/31/17 12:54 PM, Daniel D.
Daugherty wrote:
On
8/29/17 2:44 AM, serguei.spit...@oracle.com wrote:
src/jdk.jdwp.agent/share/native/include/jdwpTransport.h
L202: /* 12: SetTransportConfiguration */
I missed updating this comment also. Perhaps:
/* 12: SetTransportConfiguration added in
JDWPTRANSPORT_VERSION_1_1 */
and add this one before L262:
/* SetTransportConfiguration added in
JDWPTRANSPORT_VERSION_1_1 */
src/jdk.jdwp.agent/share/native/libdt_socket/socketTransport.c
L414: "invalid ip address in
allow option");
Should this "ip" be "IP"?
L435: if (++_peers_cnt >= MAX_PEER_ENTRIES)
{
L436: fprintf(stderr, "Error in allow
option: '%s'\n", allowed_peers);
L437:
RETURN_ERROR(JDWPTRANSPORT_ERROR_ILLEGAL_ARGUMENT,
L438: "exceeded max number of
allowed peers (32)");
L439: }
I think this error block will execute if you happen to
have exactly 32 allowed peers, i.e., "*s == 0" and I
don't think that's what you want.
I think you want the check to be:
if (_peers_cnt >= MAX_PEER_ENTRIES) {
and you want that check to be before L433. Basically,
if the
current count has overflowed, error out. Of course,
you'll
want the "++peer_cnt" increment just before "if (*s ==
0)".
L438: "exceeded max number of
allowed peers (32)");
That literal '32' is a maintenance problem when you
have
MAX_PEER_ENTRIES available.
L444: // advance to next ip block
Should this "ip" be "IP"?
L623: static int option_was_printed = 0;
Variable is not used.
L645: if (err) {
Not your bug, but this if-statement should be:
if (err != JDWPTRANSPORT_ERROR_NONE) {
src/jdk.jdwp.agent/share/native/libjdwp/debugInit.c
L574: EXIT_ERROR(map2jvmtiError(serror), "JDWP
Transport failed to initialize");
This is a new exit code path. Previously the process
did not exit here. Why the change in behavior?
src/jdk.jdwp.agent/share/native/libjdwp/transport.c
L208: jint supported_versions[2] =
{JDWPTRANSPORT_VERSION_1_1, JDWPTRANSPORT_VERSION_1_0};
Please consider adding a comment above this line:
/* If a new version is added here, update 'case
JNI_EVERSION' below. */
L463: info->transportVersion = transportVersion;
Perhaps init name, address and allowed_peers fields to
NULL
here. I don't think jvmtiAllocate() guarantees NULL
init.
L529: cfg.allowed_peers =
info->allowed_peers;
In the 'goto handlerError' case on L534, you are
publishing the
info->allowed_peers in cfg.allowed_peers, but
you're going to
free it. Do you want to NULL out cfg.allowed_peers in
the
'goto handlerError' case?
L602: if (err != JDWPTRANSPORT_ERROR_NONE) {
In this error block, you added the free of 'info'.
Nice catch!
Perhaps add a comment that the name, address and
allowed_peers
fields in 'info' are not allocated in the non-server
case so
they do not need to be freed.
src/jdk.jdwp.agent/share/native/libjdwp/transport.h
No comments.
test/com/sun/jdi/BasicJDWPConnectionTest.java
L173: // Bad mix of allow address value with allow
option '*'
L174: String allowOpt =
",allow=allow=127.0.0.1+*";
Sorry, I'm still puzzled by this test case. With the
description on L173, I would expect L174 to be:
String allowOpt = ",allow=127.0.0.1+allow=*";
Dan
http://cr.openjdk.java.net/~sspitsyn/webrevs/2017/hotspot/8061228-jdi-transport.2.inc/
I think, I've resolved all you comments/suggestions.
The list of allowed peers is still not printed in
socketTransport_accept()
in case of a rejected peer (not sure, if it is very
necessary at this point).
The issue is that the allow option is not available at this
point.
Regenerating it from the _peers array is non-trivial and
error-prone.
I'll try to implement it, if you think it is important.
The nsk.jdi and JTreg jdk_jdi test runs are in progress.
Thanks,
Serguei
On 8/28/17 15:12, serguei.spit...@oracle.com
wrote:
Hi Dan,
Thank you a lot for review!
On 8/28/17 11:00, Daniel D. Daugherty wrote:
Resending with Dmitry's e-mail address
included.
Please delete the previous version.
On 8/22/17 5:22 PM, serguei.spit...@oracle.com
wrote:
Please, review another revision of the fix for the
enhancement:
https://bugs.openjdk.java.net/browse/JDK-8061228
CSR:
https://bugs.openjdk.java.net/browse/CCC-8061228
The SCR is in the DRAFT state.
Joe suggested to consider this CSR approved and gave a
GO for integration.
It will be moved to the right state later when the CSR
tools are ready.
I'm still asking at least one reviewer to look at this
CSR and give a thumbs up.
It is to ensure everything is going in a right
direction.
I'll finalize the CSR after that.
Webrev:
http://cr.openjdk.java.net/~sspitsyn/webrevs/2017/hotspot/8061228-jdi-transport.1/
> http://cr.openjdk.java.net/~sspitsyn/webrevs/2017/hotspot/8061228-jdi-transport.2/
You seems to looked at 8061228-jdi.transport.2 that I
generated
temporarily for myself and which is obsolete now.
The 8061228-jdi.transport.1 was sent for
review and needs to be used.
I will consider and fix all the comments that are still
relevant for v1.
src/jdk.jdwp.agent/share/native/include/jdwpTransport.h
Needs copyright year update.
It was fixed in the The 8061228-jdi.transport.1.
L150: const char* allowed_peers; /* Peers
allowed for connection */
Please consider adding the following comment
above this line:
/* Field added in JDWPTRANSPORT_VERSION_1_1: */
That should provide a hint to future maintainers
about
how to add fields to jdwpTransportConfiguration.
src/jdk.jdwp.agent/share/native/libdt_socket/socketTransport.c
Needs copyright year update.
It was fixed in the The 8061228-jdi.transport.1.
L31: #include <netinet/in.h>
L34: #include <netinet/in.h>
Duplicated includes. Would be easier to spot if
the includes
were sorted, but that doesn't seem to be the
style in the file.
For the includes that you add, can you sort
those? I don't
recommend sorting the existing ones since that
would make this
patch messier.
Nice catch.
Fixed.
L396: while(1) {
Please add space before '('.
Fixed.
L408: // Input is not consumed,
something bad happens
typo: 'happens' -> 'happened'
Fixed.
L410:
RETURN_ERROR(JDWPTRANSPORT_ERROR_ILLEGAL_ARGUMENT,
You don't print the current value of 's' before
this error
return like you did in the previous error
return. Why?
This is printed/fixed in v1.
L421: _peers_cnt += 1;
Why not ++_peers_cnt or _peers_cnt++?
As it is minor, I did not want to fix it to minimize my
incremental webrev.
Fixed now.
I don't see any checks for overflow of
MAX_PEER_ENTRIES in
parseAllowedPeers().
Nice catch.
Fixed.
L590: fprintf(stderr, "ERROR: Peer not
allowed to connect, peers_cnt: %d\n", _peers_cnt);
_peers_cnt is not particular interesting. It
might be
more interesting to print info about the peer
that's
trying to connect and maybe the list of allowed
peers
(one time).
The _peers_cnt value is not printed in the webrev.1.
I agree, it is better to print
I'm not sure in what form to print the details about the
peer that's trying to connect
Should I use something like this:
char buffer[20] = { 0 };
inet_ntop(AF_INET, &(sa.sin_addr), buffer,
len);
socketTransport_accept() is executing a "do {...}
while (socketFD < 0);"
loop with various return points due to errors. Your
new
"if (_peers_cnt > 0)" block short circuits the
logic in the
"if (err) {" block that manages the acceptTimeout
variable
so the time we spent waiting for the connection
won't be
counted against the overall timeout specified by the
caller.
Example:
- Say the caller asks for a 30 second timeout.
- After 25 seconds we get a connection from an
unapproved peer.
- We won't update acceptTimeout (decrement by 25
seconds) so we won't return from the
socketTransport_accept() call for 55 seconds.
I think acceptTimeout management has to be
refactored
to be common to both the not-allowed-peer path and
the error path.
L935: int err;
Can move this variable decl to this line:
L955: err =
parseAllowedPeers(allowed_peers);
Good suggestion - fixed.
src/jdk.jdwp.agent/share/native/libjdwp/debugInit.c
No comments.
src/jdk.jdwp.agent/share/native/libjdwp/transport.c
Needs copyright year update.
It was fixed in the The 8061228-jdi.transport.1.
L150: if (name == NULL) {
You should add a check for parameter 'info'
after this block.
'info' should not be NULL either.
Nice catch - fixed.
L203: size_t i;
This decl can be moved to this line:
L209 for (i = 0; i <
sizeof(supported_versions); ++i) {
It is not a C++ code, so the declaration can not be moved to
the line 203.
At least, some C compilers would not accept it.
L210-214: four space indents should be used.
Fixed.
L224: ERROR_MESSAGE(("transport
doesn't recognize supported versions"));
Perhaps you should also list the supported
versions that were
tried so there's more failure info.
Nice suggestion.
Fixed.
L241: * even if info is already dealocated.
Typo: 'dealocated' -> 'deallocated'
Fixed.
L507-511: four space indents should be used.
L513-523: four space indents should be used.
Fixed.
L527-541: four space indents should be used, but I
don't think
the switch statement is a good idea. That logic
block should
be something like:
err = (*trans)->StartListening(trans,
address, &retAddress);
if (err != JDWPTRANSPORT_ERROR_NONE) {
printLastError(trans, err);
serror = JDWP_ERROR(TRANSPORT_INIT);
goto handleError;
}
if (info->transportVersion >=
JDWPTRANSPORT_VERSION_1_1) {
config.allowed_peers =
info->allowed_peers;
err =
(*trans)->SetTransportConfiguration(trans,
&config);
if (err != JDWPTRANSPORT_ERROR_NONE) {
printLastError(trans, err);
serror = JDWP_ERROR(TRANSPORT_INIT);
goto handleError;
}
}
The error checking block at L544-548 is now
above. Note
that I don't see a reason to error here if the
version
is newer than JDWPTRANSPORT_VERSION_1_1.
Agreed - fixed.
I was thinking about the same refactoring but decided to
keep the original minimize my update.
Also, please, note that the order of calls to SetTransportConfiguration
and StartListening is different in the webrev.1.
src/jdk.jdwp.agent/share/native/libjdwp/transport.h
Needs copyright year update.
It was fixed in the The 8061228-jdi.transport.1.
test/com/sun/jdi/BasicJDWPConnectionTest.java
L32-34 - imports should be sorted.
Fixed.
L165: // Bad mix of option '*' with other
adress values
Typo: 'adress' -> 'address'
Not sure I like that description though.
Perhaps:
// Bad mix of option '*' with bad allow address
value
Fixed.
The address should not be bad, so I've put the 127.0.0.1
there.
It looks like this now:
167 // Bad mix of allow option '*' with allow
address value
168 String allowOpt = ",allow=*+allow=127.0.0.1";
L171: // Bad mix of option '*' with other
adress values
Typo: 'adress' -> 'address'
Not sure I like that description though because
you
don't have a correctly formed '*' option there.
Perhaps:
// Bad mix of bad allow address values with
option '*':
String allowOpt =
",allow=allow=0.0.0.0+allow=*";
Fixed.
The address should not be bad, so I've put the 127.0.0.1
there.
It looks like this now:
173 // Bad mix of allow address value with allow
option '*'
174 String allowOpt = ",allow=allow=127.0.0.1+*";
So you have two bad ones before
the good option '*'. Not
sure if that's what you were really looking for
though...
Right.
A good address must be there, a bad address was used by
mistake.
I think that's it. I still need to review the CSR...
Wow!
Good catches and nice suggestions.
The updated webrev is (one comment has not been resolved
yet):
http://cr.openjdk.java.net/~sspitsyn/webrevs/2017/hotspot/8061228-jdi-transport.2/
The original 8061228-jdi-transport.2 was moved to
8061228-jdi-transport.2.old .
Thanks a lot, Dan!
Serguei
Dan
The lastest webrev from Dmitry:
http://cr.openjdk.java.net/~dsamersoff/JDK-8061228/webrev.18/
Incremental webrev vs the latest webrev from Dmitry:
http://cr.openjdk.java.net/~sspitsyn/webrevs/2017/hotspot/8061228-jdi-transport.1.inc/
Summary:
This enhancement was developed by Dmitry who left the
team.
I don't know what email address to use to CC him at
this point.
I hope, Dmitry will find this discussion and reply
accordingly.
The latest webrev revision from Dmitry was v18
(please, see above).
This revision covers the following:
- Cleanup for versioning negotiation protocol (back
up to the original).
Now the transport library supports both versions
1_0 and 1_1 (newly introduced).
- The transport native interface was changed.
The function SetTransportConfiguration() is
introduced instead of the
StartListeningWithAllow(). It allows to the same
transport library to support
both old and new version of the transport
interface. At this point, the
new structure jdwpTransportConfiguration includes
only one field:
const char* allowed_peers;
But it can be extended in the future if
any other update in configuration
will be required.
- The unit test was updated to provide better
coverage of the corner cases
for 'allow' option introduced by this enhancement.
- Fixes to improve diagnosability.
- A couple of bugs/regressions were fixed so that
all the JDI tests are passed now.
- A cleanup that includes some renaming and
reformatting.
Testing:
Tested new agent flag (allow), with new test:
jdk/test/com/sun/jdi/BasicJDWPConnectionTest.java
Ran the nsk.jdi, nsk.jdwp and jtreg jdk_jdi for both
release and fastdebug builds.
All tests are passed.
Thanks,
Serguei
|
