Add on-demand SECOID table initialization as a workaround for Android PGO build bug
Categories
(NSS :: Libraries, defect, P1)
Tracking
(Not tracked)
People
(Reporter: jschanck, Assigned: jschanck)
References
Details
(Whiteboard: [nss-fx])
Attachments
(1 external file)
The patch for Bug 1754746 changes how we build and link the PSM tlsserver test binaries, and now we're seeing a segfault in GenerateOCSPResponse that only manifests in Android builds with Profile Guided Optimization (PGO) enabled.
Treeherder: https://treeherder.mozilla.org/jobs?repo=try&revision=22fcbc2a1499121635ca4c9800a701629263e3d7&searchStr=X2
*** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***
Build fingerprint: 'Android/sdk_phone_x86_64/generic_x86_64:7.0/NYC/4174735:userdebug/test-keys'
Revision: '0'
ABI: 'x86_64'
pid: 8158, tid: 8158, name: GenerateOCSPRes >>> /data/local/tmp/test_root/xpcb/GenerateOCSPResponse <<<
signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0x10
rax 0000000000000008 rbx 00007fffb1e1e158 rcx 0000000000010fc3 rdx 0000000000000002
rsi 00000000000003a8 rdi 00000000000003a8
r8 0000000000000100 r9 00002d3b23da3f3f r10 0000000000000002 r11 00007fffb1e1e1e8
r12 0000000000000000 r13 0000000000000010 r14 00000000000003a8 r15 00000000ffffffff
cs 0000000000000033 ss 000000000000002b
rip 000073feebeb0204 rbp 00007fffb1e1df20 rsp 00007fffb1e1def0 eflags 0000000000000206
backtrace:
#00 pc 0000000000123204 /data/local/tmp/test_root/xpcb/libnss3.so (offset 0x15000)
#01 pc 000000000011d385 /data/local/tmp/test_root/xpcb/libnss3.so (offset 0x15000)
#02 pc 000000000001624e /data/local/tmp/test_root/xpcb/libsoftokn3.so (offset 0x3000)
#03 pc 00000000000420e3 /data/local/tmp/test_root/xpcb/GenerateOCSPResponse (pk11_TokenKeyGenWithFlagsAndKeyType+844)
#04 pc 0000000000040730 /data/local/tmp/test_root/xpcb/GenerateOCSPResponse (pk11_RawPBEKeyGenWithKeyType+176)
#05 pc 0000000000040867 /data/local/tmp/test_root/xpcb/GenerateOCSPResponse (PK11_PBEKeyGen+303)
#06 pc 000000000002ef21 /data/local/tmp/test_root/xpcb/GenerateOCSPResponse (PK11_ExportEncryptedPrivKeyInfoV2+237)
#07 pc 000000000002f1b3 /data/local/tmp/test_root/xpcb/GenerateOCSPResponse (PK11_ExportEncryptedPrivKeyInfo+27)
#08 pc 0000000000020815 /data/local/tmp/test_root/xpcb/GenerateOCSPResponse (_ZN7mozilla4pkix4test17CreateTestKeyPairENS1_22TestPublicKeyAlgorithmERKNSt6__ndk110unique_ptrI18SECKEYPublicKeyStr17ScopedMaybeDeleteIS5_EEERKNS4_I19SECKEYPrivateKeyStrS6_ISB_EEE+125)
#09 pc 0000000000018fd1 /data/local/tmp/test_root/xpcb/GenerateOCSPResponse (_Z22GetOCSPResponseForType16OCSPResponseTypeRKNSt6__ndk110unique_ptrI18CERTCertificateStrN7mozilla33UniqueCERTCertificateDeletePolicyEEERKNS1_I11PLArenaPoolNS3_29UniquePLArenaPoolDeletePolicyEEEPKcl+1225)
#10 pc 0000000000018923 /data/local/tmp/test_root/xpcb/GenerateOCSPResponse (main+267)
#11 pc 000000000001c8d4 /system/lib64/libc.so (__libc_init+84)
#12 pc 00000000000187e7 /data/local/tmp/test_root/xpcb/GenerateOCSPResponse (_start_main+55)
#13 pc 00000000000187ab /data/local/tmp/test_root/xpcb/GenerateOCSPResponse (_start+11)
I recognized the 0x10 fault address from crashes in SECOID_FindOID, and I was able to confirm that the segfault in GenerateOCSPResponse is in SECOID_FindOIDByMechanism using a debugger.
We should identify and fix the root cause as part of the ongoing review of nssutil, but in the meantime I'd like to add a workaround where we perform a null check in the SECOID_FindOID functions and initialize the hash tables on-demand if necessary. I've confirmed that this fixes the problem in try.
| Assignee | ||
Comment 1•4 years ago
|
||
| Assignee | ||
Comment 2•4 years ago
|
||
Description
•