This applies 3.4 patches to 3.10 if no other match is available
Note: CVE-2017-13245/3.4/0002.patch ends up applied over CVE-2018-10902/3.18/0003.patch
Signed-off-by: Tad <tad@spotco.us>
This fixes CVE-2018-9422 which was primarily added via b56fabac
May still need to be fixed:
16.0/kernel_google_yellowstone
16.0/kernel_xiaomi_msm8937
Signed-off-by: Tad <tad@spotco.us>
It is still actively being used by malware.
This largely handles 3.0, 3.4, and 3.10 kernels.
It works for select 3.18 kernels too.
TODO: need alternate get_file_rcu backport for the following:
15.1/lge_msm8996
15.1/zte_msm8996
16.0/xiaomi_msm8937
17.1/motorola_msm8996
18.1/google_marlin
18.1/lge_msm8996
18.1/oneplus_msm8996
Signed-off-by: Tad <tad@spotco.us>
Drop long non-compiling devices:
- 14.1: n7100, jellypro
- 15.1: himaul, oneplus2
- 16.0: zenfone3, fugu
- 17.1: yellowstone, fugu
- 18.1: bonito, sargo
Drop in favor of 19.1:
- 17.1: bonito, sargo
- 18.1: pro1, aura, sunfish, coral, flame, bramble, redfin
(experimental, but these devices don't currently appear to have any users)
Signed-off-by: Tad <tad@spotco.us>
Change the property too, so it takes effect next update.
Since 16.0 lacks a toggle, this effectively disables the feature for it.
Even devices with 4GB of RAM have usability severely impacted.
Plus some other tweaks/churn
Signed-off-by: Tad <tad@spotco.us>
10+4 devices tested working with bionic hardening patches enabled
but hammerhead and shamu do not boot...
2 of the patches were already found to have issues and disabled
3 other patches were ruled out:
- Stop implicitly marking mappings as mergeable
- Make __stack_chk_guard read-only at runtime
- On 64-bit, zero the leading stack canary byte
Leaves 11+1 patches remaining that need to be tested
But I don't have either of the two known impacted devices.
Signed-off-by: Tad <tad@spotco.us>
This removes many duplicately or wrongly applied patches.
Correctly removed:
- CVE-2011-4132 can apply infinitely
- CVE-2013-2891 can apply infinitely
- CVE-2014-9781 can apply once to fb_cmap_to_user correctly and incorrectly to fb_copy_cmap
- CVE-2015-0571 can apply incorrectly and was disabled in patch repo as a result
- CVE-2016-2475 can apply infinitely
- CVE-2017-0627 can apply infinitely
- CVE-2017-0750 can apply infinitely
- CVE-2017-14875 can apply infinitely
- CVE-2017-14883 can apply infinitely
- CVE-2020-11146 can apply infinitely
- CVE-2020-11608 can apply infinitely
- CVE-2021-42008 can apply infinitely
Questionable (might actually be beneficial to "incorrectly" apply again):
- CVE-2012-6544 can apply once to hci_sock_getsockopt correctly and incorrectly to hci_sock_setsockopt
- CVE-2013-2898 can apply once to sensor_hub_get_feature correctly and incorrectly to sensor_hub_set_feature
- CVE-2015-8575 can apply once to sco_sock_bind correctly and incorrectly to sco_sock_connect
- CVE-2017-8281 can apply once to diagchar_ioctl correctly and incorrectly to diagchar_compat_ioctl
- CVE-2019-10622 can apply once to qdsp_cvp_callback correctly and incorrectly to qdsp_cvs_callback
- CVE-2019-14104 can apply once to cam_context_handle_start/stop_dev and incorrectly to cam_context_handle_crm_process_evt and cam_context_handle_flush_dev
Other notes:
- CVE-2016-6693 can be applied again if it was already applied in combination with CVE-2016-6696
then the dupe check will fail and mark CVE-2016-6696 as already applied, effectively reverting it.
This was seemingly fixed with a hand merged patch in patch repo.
Wrongly removed:
- CVE-2013-2147 is meant for cciss_ioctl32_passthru but is detected in cciss_ioctl32_big_passthru
- CVE-2015-8746 is meant for nfs_v4_2_minor_ops but is detected in nfs_v4_1_minor_ops
- CVE-2021-Misc2/ANY/0043.patch is meant for WLANTL_RxCachedFrames but is detected in WLANTL_RxFrames
Signed-off-by: Tad <tad@spotco.us>
