397478 - Lock from ssl_InitSymWrapKeysLock not freed on selfserv shutdown.

Status: RESOLVED FIXED

(NSS :: Libraries, defect, P2)

Description

[Slavomir Katuscak]

Locks created in SSL_ConfigServerSessionIDCache are not freed on selfserv shutdown.

Block in use (biu):
Found block of size 88 bytes at address 0x8085d40 (2.64% of total)
At time of allocation, the call stack was:
    [1] calloc() at 0xb4d308a0
    [2] PR_Calloc() at line 475 in "prmem.c"
    [3] PR_NewLock() at line 174 in "ptsynch.c"
    [4] __nss_InitLock() at line 67 in "nsslocks.c"
    [5] ssl_InitSymWrapKeysLock() at line 3954 in "ssl3con.c"
    [6] SSL_ConfigServerSessionIDCache() at line 1180 in "sslsnce.c"
    [7] main() at line 1899 in "selfserv.c"

Block in use (biu):
Found block of size 88 bytes at address 0x8085cd0 (2.64% of total)
At time of allocation, the call stack was:
    [1] calloc() at 0xb4d308a0
    [2] PR_Calloc() at line 475 in "prmem.c"
    [3] PR_NewLock() at line 174 in "ptsynch.c"
    [4] __nss_InitLock() at line 67 in "nsslocks.c"
    [5] ssl_InitClientSessionCacheLock() at line 74 in "sslnonce.c"
    [6] SSL_ConfigServerSessionIDCache() at line 1178 in "sslsnce.c"
    [7] main() at line 1899 in "selfserv.c"

Comment 1

[Nelson Bolyard (seldom reads bugmail)]

Slavo, in these runs, does selfserv do a clean shutdown without errors?
Does it successfully shut down NSS?
Or is it killed?

Comment 2

[Nelson Bolyard (seldom reads bugmail)]

Not sure if these are library leaks or tools leaks. 
Will assume they're library leaks in libSSL until we know better.

Comment 3

[Slavomir Katuscak]

I always use normal termination (stop message over tstclnt) in memory leak tests. 

Also there is message in log:
selfserv: normal termination

Comment 4

[Nelson Bolyard (seldom reads bugmail)]

Slavo, thanks for recording those important details in this bug.

Comment 5

[Nelson Bolyard (seldom reads bugmail)]

This bug reports two leaks.  The second one is actually the subject of 
another bug, bug 397486.  So this bug is ONLY about the lock initialized
by ssl_InitSymWrapKeysLock(), the first leak reported in comment 0.

Comment 6

[Julien Pierre]

This is a library bug. symWrapKeysLock is never destroyed. It should be done in SSL_ShutdownServerSessionIDCache .

Comment 7

[Julien Pierre]

Removing blocking flag that was in error.

Comment 8

[Nelson Bolyard (seldom reads bugmail)]

Julien, in comment 6, you wrote:
> It should be done in SSL_ShutdownServerSessionIDCache .

Do you see any reason not to do it in SSL3_ShutdownServerCache ?

Comment 9

[Nelson Bolyard (seldom reads bugmail)]

Attachment: patch v1 - Method suggested by Julien

This is the first of two alternative patches.

Comment 10

[Nelson Bolyard (seldom reads bugmail)]

Attachment: patch v2 - simpler alternative method (checked in)

Julien, please choose one of these methods and give it r+.
Thanks.

Comment 11

[Julien Pierre]

Nelson,

That's fine, since SSL3_ShutdownServerCache is called by
SSL_ShutdownServerSessionIDCache.

Comment 12

[Julien Pierre]

The patch for this bug was changed in bug 397486 . The new patch in that bug also fixed this leak, both on the trunk and on the branch. So I'm changing the target for this bug to the branch (3.11.10).