Modified from bug 1248818 comment 11:
Before this patch, if a user had a smart card (PKCS#11 device) with removable
slots, Firefox would launch a thread for each module and loop, calling
SECMOD_WaitForAnyTokenEvent to be alerted to any insertions/removals. At
shutdown, we would call SECMOD_CancelWait, which would cancel any waiting
threads. However, since that involved calling 3rd party code, we really had no
idea if these modules were behaving correctly (and, indeed, they often weren't,
judging by the shutdown crashes we were getting).
The real solution is to stop relying on PKCS#11, but since that's unlikely in
the near future, the next best thing would be to load these modules in a child
process. That way, misbehaving modules don't cause Firefox to hang/crash/etc.
That's a lot of engineering work, though, so what this patch does is avoids the
issue by never calling SECMOD_WaitForAnyTokenEvent (and thus we never have to
call SECMOD_CancelWait, etc.). Instead, every time Firefox performs an operation
that may be affected by a newly added or removed smart card, it first has NSS
refresh its view of any removable slots. This is similar to how we ensure the
loadable roots module has been loaded (see bug 1372656).
MozReview-Commit-ID: JpmLdV7Vvor
--HG--
extra : rebase_source : d3503d19fa9297106d661a017a38c30969fa39b4
Incidentally, this means we can remove certificateUsageVerifyCA and
certificateUsageStatusResponder from CertVerifier, since we no longer use them.
MozReview-Commit-ID: Bbqn8fShfTm
--HG--
extra : rebase_source : 012cb08dcbe33fe889c9f6824959b1a02cd0bdc7
Bug 1364159 introduced an optimization that attempted to avoid reading from the
user's cached certificate database as much as possible when building a verified
certificate chain. Unfortunately this had the side-effect of not preferring root
certificates in path building, which can result in unnecessarily long chains
(which rather defeats the purpose, since it means more signature verifications).
This patch reverts the functionality changes from that bug but keeps the test
that was added (the test didn't directly test the functionality changes - it's
more of a check that path building will query the cached certificate db when
necessary).
MozReview-Commit-ID: I56THTLUytH
--HG--
extra : rebase_source : 7db9597e25b98942450840519d707046cc660781
In a profile, loading the loadable roots PKCS#11 module (i.e. the built-in root
CA module) accounted for about 60% of the time to initialize PSM/NSS. Since we
only need the roots module loaded when we're actually looking for an issuing
certificate or querying a certificate's trust, we can do the load
asynchronously (where it hopefully finishes before we actually need it, because
otherwise we'll have to wait anyway).
MozReview-Commit-ID: JyY6NtpQAUj
--HG--
extra : rebase_source : f63a697b18a409dd042289afa2b727b09f81f19f
GetHostNameRaw() returns a char* string, which is less safe and ergonomic
compared to the Mozilla string classes. GetHostName() can be used instead.
MozReview-Commit-ID: GYvTnISNN35
--HG--
extra : rebase_source : da257f5fba2c26cd92d932c3d1d363458b84a65b
CERT_CreateSubjectCertList is not an inexpensive function call, since it
enumerates the certificate database (i.e. reads from disk a lot). If we're
verifying for a TLS handshake, however, we should already have in memory a
certificate chain sent by the peer (there are some cases where we won't, such as
session resumption (see bug 731478)). If we can, we should use those
certificates before falling back to calling CERT_CreateSubjectCertList.
MozReview-Commit-ID: ASjVGsELb1O
--HG--
extra : rebase_source : 1efc635d4a98079c87f77ef3794e4b2f20eec59f
The default OCSP timeout for soft-fail DV is still 2 seconds. This patch makes
it configurable on the interval (0, 5] seconds.
The default OCSP timeout for EV and hard-fail DV is still 10 seconds. This patch
makes it configurable on the interval (0, 20] seconds.
MozReview-Commit-ID: CPd8pwYrJhj
--HG--
extra : rebase_source : 45bd7d06ea013f0a776ea18be9408dedb18271d8
Unfortunately, this doesn't cover delegated OCSP responder certificates. While
gathering telemetry on the use of SHA-1, we encountered bug 1183822 (basically,
that the method of gathering telemetry was causing OCSP verification failures
due to delegated responders signed with SHA-1). As a temporary solution, we
changed the verifier to always allow SHA-1 for OCSP certificates when verifying
an OCSP response. Consequently, we now have no idea what the compatibility
impact of disabling SHA-1 in OCSP responder certificates will be, so it's
probably not a good idea to do that right now.
Even if someone does manage to forge an OCSP responder certificate using a SHA-1
collision, they will have about as much power as an active network attacker
blocking OCSP requests or injecting bad stapled OCSP responses, so this isn't a
disaster.
MozReview-Commit-ID: 10r23W1APiR
--HG--
extra : rebase_source : dc003c4812677c40882506b1b6b1e1f68d7e6e92
PR_ASSERT() is an unnecessary dependency on NSPR.
We can use MOZ_ASSERT() instead, which accomplishes the same task but doesn't
depend on NSPR.
MozReview-Commit-ID: 9gyWUkv3KxQ
--HG--
extra : rebase_source : 313ce6c8de3db3ce72635e37f09d28316ae02c51
The PR_SetError() + PR_GetError() pattern is error prone and unnecessary.
Also fixes Bug 1254403.
MozReview-Commit-ID: DRI69xY4vxC
--HG--
extra : rebase_source : aa07c0dfb5cc2a203e772b415b7a75b27d9bad3c
Previously PSM would load EV information on-demand (i.e. just before verifying a
certificate). This simplifies this operation, removes a dubious optimization
(loading the EV information on another thread while opening a network
connection), and relocates the loading operation to when we are likely to have
good disk locality (i.e. when we've just loaded the built-in roots module).
This also removes the now-unused MOZ_NO_EV_CERTS build flag.
MozReview-Commit-ID: 8Rnl4ozF95V
--HG--
extra : rebase_source : 5b2e76079c256f7e3c55b1d4ec0d9f654fec44f6
Previously PSM would load EV information on-demand (i.e. just before verifying a
certificate). This simplifies this operation, removes a dubious optimization
(loading the EV information on another thread while opening a network
connection), and relocates the loading operation to when we are likely to have
good disk locality (i.e. when we've just loaded the built-in roots module).
This also removes the now-unused MOZ_NO_EV_CERTS build flag.
MozReview-Commit-ID: 8Rnl4ozF95V
--HG--
extra : rebase_source : 344b68c81af1ed3fb038e4e96c3c50e939d32c3d
The PR_SetError() + PR_GetError() pattern currently used is error prone and
unnecessary. The functions involved can instead return mozilla::pkix::Result,
which is equally expressive and more robust.
MozReview-Commit-ID: Hkd39eqTvds
--HG--
extra : rebase_source : f09e37c6a3a930c30cce003139df86bc84d771ee
The functions aren't necessary now that we have BitwiseCast.
MozReview-Commit-ID: 2nzOuwAop4Y
--HG--
extra : rebase_source : 0cb2c16f484a81b2e77384564973b58ac2d10fb9
The functions aren't necessary now that we have BitwiseCast.
MozReview-Commit-ID: 2nzOuwAop4Y
--HG--
extra : rebase_source : 196449249eec75b8eb10e59662231c3f4e83c268
Previously this implementation would use the expected names of the built-in
module and slot to get a handle on them. This doesn't work on distributions that
use other names. The new implementation searches through the slots from the
default module list for one where PK11_HasRootCerts returns true (which
indicates that NSS considers that slot to contain the default built-in root
list).
MozReview-Commit-ID: LmX27hQfFJU
--HG--
extra : rebase_source : 50383dcc77257fe08ce2c7d908e95cda7c4bbe9d
ScopedCERTCertList is based on Scoped.h, which is deprecated in favour of the
standardised UniquePtr.
Also changes CERTCertList parameters of various functions to make ownership more
explicit.
MozReview-Commit-ID: EXqxTK6inqy
--HG--
extra : transplant_source : %9B%A9a%94%D1%7E%2BTa%9E%9Fu%9F%02%B3%1AT%1B%F1%F6
Before this patch, the default policy for the use of SHA1 in certificate
signatures was "allow all" due to compatibility concerns.
After gathering telemetry, we are confident that we can enforce the policy of
"allow for locally-installed roots" (or certificates valid before 2016) without
too much breakage.
MozReview-Commit-ID: 8GxtgdbaS3P
--HG--
extra : rebase_source : d1bed911f2d5d40229ea06556fee0848668e98b6
When a built-in root certificate has its trust changed from the default value,
the platform has to essentially create a copy of it in the read/write
certificate database with the new trust settings. At that point, the desired
behavior is that the platform still considers that certificate a built-in root.
Before this patch, this would indeed happen for the duration of that run of the
platform, but as soon as it restarted, the certificate in question would only
appear to be from the read/write database, and thus was not considered a
built-in root. This patch changes the test of built-in-ness to explicitly
search the built-in certificate slot for the certificate in question. If found,
it is considered a built-in root.
MozReview-Commit-ID: HCtZpPQVEGZ
--HG--
extra : rebase_source : 759e9c5a7bb14f14a77e62eae2ba40c085f04ccd
When a built-in root certificate has its trust changed from the default value,
the platform has to essentially create a copy of it in the read/write
certificate database with the new trust settings. At that point, the desired
behavior is that the platform still considers that certificate a built-in root.
Before this patch, this would indeed happen for the duration of that run of the
platform, but as soon as it restarted, the certificate in question would only
appear to be from the read/write database, and thus was not considered a
built-in root. This patch changes the test of built-in-ness to explicitly
search the built-in certificate slot for the certificate in question. If found,
it is considered a built-in root.
MozReview-Commit-ID: HCtZpPQVEGZ
--HG--
extra : rebase_source : 898ef37459723f1d8479cfdc58658ccb00e782a9
Before this patch, we were measuring where SHA-1 was being used in TLS
certificates: nowhere, in end-entities, in intermediates, or in both. However,
the possible SHA-1 policies don't differentiate between end-entities and
intermediates and instead depended on whether or not each certificate has a
notBefore value after 2015 (i.e. >= 0:00:00 1 January 2016 UTC). We need to
gather telemetry on the possible policy configurations.
--HG--
extra : rebase_source : 301c821c8de16ffb924cd198dd0a4d3139536019
David Keeler
Cykesiopka
Kai Engert
Tom Tromey
Sergei Chernov
Andrea Marchesini
Jonathan Hao
Carsten "Tomcat" Book
Wes Kocher
Sebastian Hengst
Ryan VanderMeulen
sajitk