Avoid runtime linker warning ...

Avoid runtime linker warning 
"Symbol `SSL_ImplementedCiphers' has different size in shared object, consider re-linking"
if SSL_ImplementedCiphers is different size on target system than it was on build system.

BUG=12826
TEST=run on Jaunty, look for "Symbol `SSL_ImplementedCiphers' has different size in shared object" in console output


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=21649

[dank, 2009-06-07 22:03:42]

Paper over a scary but benign warning that happens when
users run Chrome on some system with an NSS other than the one it
was compiled against.

This points out a problem with the NSS ABI; it really should have
a function rather than exposing a global array like that.

Doing a dlopen of a library we link directly to is kind of cheesy, but
seems to work.  One more thing to worry about while packaging, though.

[Michael Moss, 2009-06-08 18:29:28]

http://codereview.chromium.org/118367/diff/1001/1002
File base/nss_init.cc (right):

http://codereview.chromium.org/118367/diff/1001/1002#newcode89
Line 89: FilePath ssl3Path("libssl3.so");
Does it matter if the package only has "libssl3.so.1d", or do we also need the
"libssl3.so" symlink?

http://codereview.chromium.org/118367/diff/1001/1002#newcode99
Line 99: NOTREACHED() << "Can't get list of supported ciphers";
Do we need UnloadNativeLibrary() here?

[Michael Moss, 2009-06-08 18:31:03]

> Does it matter if the package only has "libssl3.so.1d", or do we also need the
"libssl3.so" symlink?

And if so, can you add it to src/chrome/installer/installer.gyp under the
'/usr/<(lib32_dir)/libssl3.so.1d' line, so it's part of this CL?

[wtc, 2009-06-08 23:51:38]

LGTM.

[Evan Martin, 2009-07-23 23:00:11]

http://codereview.chromium.org/118367/diff/1001/1002
File base/nss_init.cc (right):

http://codereview.chromium.org/118367/diff/1001/1002#newcode97
Line 97: "SSL_ImplementedCiphers"));
Took me a sec to realize this isn't a function pointer.  You can just use
dlopen() directly if you want -- we have native_library when we need to share
code with windows.

http://codereview.chromium.org/118367/diff/1001/1002#newcode104
Line 104: for (int i = 0; i < SSL_NumImplementedCiphers; i++) {
Don't we need to look up this value from the library as well?

[wtc, 2009-07-23 23:15:38]

http://codereview.chromium.org/118367/diff/1001/1002
File base/nss_init.cc (right):

http://codereview.chromium.org/118367/diff/1001/1002#newcode104
Line 104: for (int i = 0; i < SSL_NumImplementedCiphers; i++) {
No, because the runtime loader doesn't complain about
SSL_NumImplementedCiphers, which is an integer as opposed to
an array.

[dkegel, 2009-07-23 23:19:31]

On Thu, Jul 23, 2009 at 11:15 PM, <wtc@chromium.org> wrote:
> http://codereview.chromium.org/118367/diff/1001/1002#newcode104
> Line 104: for (int i = 0; i < SSL_NumImplementedCiphers; i++) {
> No, because the runtime loader doesn't complain about
> SSL_NumImplementedCiphers, which is an integer as opposed to
> an array.

I suppose there's some danger dlopen could find a different SSL
library than the runtime linker did...
in which case we'd need to worry about more than SSL_NumImplementedCiphers.

[Evan Martin, 2009-07-23 23:26:02]

http://codereview.chromium.org/118367/diff/1001/1002
File base/nss_init.cc (right):

http://codereview.chromium.org/118367/diff/1001/1002#newcode104
Line 104: for (int i = 0; i < SSL_NumImplementedCiphers; i++) {
On 2009/07/23 23:15:38, wtc wrote:
> No, because the runtime loader doesn't complain about
> SSL_NumImplementedCiphers, which is an integer as opposed to
> an array.

My concern is that this warning is telling us the size of the array is different
than what we expected.  That means that perhaps the number of implemented
ciphers, which we use to index into the array, is also plausibly different.

[Evan Martin, 2009-07-23 23:48:00]

New idea!  (Markus's idea.)

Skip the dlopen(), but use dlsym(RTLD_DEFAULT, "symbol as before").  That means
"give me the symbol from the already-opened library".

[dkegel, 2009-07-23 23:59:35]

oooh, schnazzy.

On Thu, Jul 23, 2009 at 11:48 PM, <evan@chromium.org> wrote:
> New idea! =A0(Markus's idea.)
>
> Skip the dlopen(), but use dlsym(RTLD_DEFAULT, "symbol as before").
> That means "give me the symbol from the already-opened library".
>
> http://codereview.chromium.org/118367
>

[dank, 2009-07-24 00:11:36]

OK, look now.

[Evan Martin, 2009-07-24 00:17:14]

LGTM!  do you know if it fixes the problem?

[wtc, 2009-07-24 00:41:59]

LGTM.  Good solution!

[dkegel, 2009-07-24 13:46:14]

I'm out today, will check in on Monday.

On Thu, Jul 23, 2009 at 5:41 PM, <wtc@chromium.org> wrote:
> LGTM. =A0Good solution!
>
> http://codereview.chromium.org/118367
>