|
| 1 | +- commits: |
| 2 | + - subject: Update layers/meta-balena to 8352e3480e96f4503361752cfe72aca9622250a3 |
| 3 | + hash: 4755880de6f929fdeb9f5b28ff4e75ef60aa80d3 |
| 4 | + body: Update layers/meta-balena |
| 5 | + footer: |
| 6 | + Changelog-entry: Update layers/meta-balena to 8352e3480e96f4503361752cfe72aca9622250a3 |
| 7 | + changelog-entry: Update layers/meta-balena to 8352e3480e96f4503361752cfe72aca9622250a3 |
| 8 | + author: Self-hosted Renovate Bot |
| 9 | + nested: |
| 10 | + - commits: |
| 11 | + - subject: "os-helpers-tpm2: fix empty efivar reads" |
| 12 | + hash: 3f9d048f964f922c9bfaeee4939197cb31c8eef6 |
| 13 | + body: > |
| 14 | + When reading from efi variables in hostapp-update hooks during |
| 15 | + rollback, |
| 16 | + |
| 17 | + tcgtool will improperly read zero bytes from efivar files. This |
| 18 | + results |
| 19 | + |
| 20 | + in an improper calculation of the PCR 7 digest, and an |
| 21 | + unbootable |
| 22 | + |
| 23 | + system. |
| 24 | + |
| 25 | + |
| 26 | + Read the file contents, skipping the first four bytes that are |
| 27 | + |
| 28 | + attributes, and pipe the data directly to tcgtool to work around |
| 29 | + this. |
| 30 | + footer: |
| 31 | + Change-type: patch |
| 32 | + change-type: patch |
| 33 | + Signed-off-by: Joseph Kogut < [email protected]> |
| 34 | + signed-off-by: Joseph Kogut < [email protected]> |
| 35 | + author: Joseph Kogut |
| 36 | + nested: [] |
| 37 | + - subject: "rollback-health: bind mount EFI partition in old_rootfs" |
| 38 | + hash: 8761c1d6536d86249757f37925ae40d905716cc3 |
| 39 | + body: > |
| 40 | + Some hooks, such as 0-signed-update, will attempt to read files |
| 41 | + from the |
| 42 | + |
| 43 | + EFI system partition, such as combined policy binaries. |
| 44 | + |
| 45 | + |
| 46 | + Bind mount the EFI partition into old_rootfs before running |
| 47 | + hooks to |
| 48 | + |
| 49 | + ensure this is available. |
| 50 | + footer: |
| 51 | + Change-type: patch |
| 52 | + change-type: patch |
| 53 | + Signed-off-by: Joseph Kogut < [email protected]> |
| 54 | + signed-off-by: Joseph Kogut < [email protected]> |
| 55 | + author: Joseph Kogut |
| 56 | + nested: [] |
| 57 | + - subject: "rollback-health: mount securityfs in old_rootfs" |
| 58 | + hash: e87838103af3c5ee05c60b7105353f987af47e61 |
| 59 | + body: > |
| 60 | + When rollback-health runs, a failing healthcheck causes the |
| 61 | + |
| 62 | + hostapp-update hooks to be run from the inactive partition, to |
| 63 | + make the |
| 64 | + |
| 65 | + inactive system bootable again. |
| 66 | + |
| 67 | + |
| 68 | + The 0-signed-update hook, which updates the sealing policy for |
| 69 | + secure |
| 70 | + |
| 71 | + boot enabled systems, reads from the securityfs mounted at |
| 72 | + |
| 73 | + /sys/kernel/security in order to parse the TPM event log. |
| 74 | + |
| 75 | + |
| 76 | + If this filesystem isn't mounted, the hook will improperly |
| 77 | + detect that |
| 78 | + |
| 79 | + the TPM event log isn't available, and unneccessarily create a |
| 80 | + combined |
| 81 | + |
| 82 | + policy when a single PCR policy would suffice. |
| 83 | + |
| 84 | + |
| 85 | + Mount this filesystem in old_rootfs before chrooting to fix |
| 86 | + this. |
| 87 | + footer: |
| 88 | + Change-type: patch |
| 89 | + change-type: patch |
| 90 | + Signed-off-by: Joseph Kogut < [email protected]> |
| 91 | + signed-off-by: Joseph Kogut < [email protected]> |
| 92 | + author: Joseph Kogut |
| 93 | + nested: [] |
| 94 | + - subject: "os-helpers-tpm2: compute_pcr7 w/ events post separator" |
| 95 | + hash: e4f8f6f9c79f994f873bc930bfa028f2161a29d9 |
| 96 | + body: > |
| 97 | + In commit 1c19ebb, we append digests from the TPM event log |
| 98 | + |
| 99 | + corresponding to events that are logged before EV_SEPARATOR. For |
| 100 | + |
| 101 | + instance, parsing the event log on a typical system for event |
| 102 | + types, the |
| 103 | + |
| 104 | + output looks like this: |
| 105 | + |
| 106 | + |
| 107 | + EV_EFI_VARIABLE_DRIVER_CONFIG (SecureBoot) |
| 108 | + |
| 109 | + EV_EFI_VARIABLE_DRIVER_CONFIG (PK) |
| 110 | + |
| 111 | + EV_EFI_VARIABLE_DRIVER_CONFIG (KEK) |
| 112 | + |
| 113 | + EV_EFI_VARIABLE_DRIVER_CONFIG (db) |
| 114 | + |
| 115 | + EV_EFI_VARIABLE_DRIVER_CONFIG (dbx) |
| 116 | + |
| 117 | + EV_SEPARATOR |
| 118 | + |
| 119 | + |
| 120 | + This system requires no merging of event log digests. |
| 121 | + |
| 122 | + |
| 123 | + On systems that measure EFI binaries (mostly only QEMU w/ edk2), |
| 124 | + we also |
| 125 | + |
| 126 | + get this: |
| 127 | + |
| 128 | + |
| 129 | + EV_EFI_VARIABLE_DRIVER_CONFIG (SecureBoot) |
| 130 | + |
| 131 | + EV_EFI_VARIABLE_DRIVER_CONFIG (PK) |
| 132 | + |
| 133 | + EV_EFI_VARIABLE_DRIVER_CONFIG (KEK) |
| 134 | + |
| 135 | + EV_EFI_VARIABLE_DRIVER_CONFIG (db) |
| 136 | + |
| 137 | + EV_EFI_VARIABLE_DRIVER_CONFIG (dbx) |
| 138 | + |
| 139 | + EV_SEPARATOR |
| 140 | + |
| 141 | + EV_EFI_VARIABLE_AUTHORITY (bootx64.efi) |
| 142 | + |
| 143 | + EV_EFI_VARIABLE_AUTHORITY (bzImage) |
| 144 | + |
| 145 | + |
| 146 | + Again, this requires no merging. We compute the signature of the |
| 147 | + |
| 148 | + relevant EFI binaries ourselves, as they're subject to change |
| 149 | + during |
| 150 | + |
| 151 | + hostapp-update. |
| 152 | + |
| 153 | + |
| 154 | + However, we've also seen event logs like this: |
| 155 | + |
| 156 | + |
| 157 | + EV_EFI_VARIABLE_DRIVER_CONFIG (SecureBoot) |
| 158 | + |
| 159 | + EV_EFI_VARIABLE_DRIVER_CONFIG (PK) |
| 160 | + |
| 161 | + EV_EFI_VARIABLE_DRIVER_CONFIG (KEK) |
| 162 | + |
| 163 | + EV_EFI_VARIABLE_DRIVER_CONFIG (db) |
| 164 | + |
| 165 | + EV_EFI_VARIABLE_DRIVER_CONFIG (dbx) |
| 166 | + |
| 167 | + EV_EFI_ACTION (DMA Protection Disabled) |
| 168 | + |
| 169 | + EV_SEPARATOR |
| 170 | + |
| 171 | + |
| 172 | + This case is the one we've handled previously, by reading from |
| 173 | + the event |
| 174 | + |
| 175 | + log and appending event digests before EV_SEPARATOR. We stopped |
| 176 | + at |
| 177 | + |
| 178 | + EV_SEPARATOR because we weren't parsing event types previously, |
| 179 | + and this |
| 180 | + |
| 181 | + digest is a constant that's easily recognized. However, we've |
| 182 | + since |
| 183 | + |
| 184 | + encountered systems that have unexpected events *after* |
| 185 | + EV_SEPARATOR, as |
| 186 | + |
| 187 | + shown below. |
| 188 | + |
| 189 | + |
| 190 | + EV_EFI_VARIABLE_DRIVER_CONFIG (SecureBoot) |
| 191 | + |
| 192 | + EV_EFI_VARIABLE_DRIVER_CONFIG (PK) |
| 193 | + |
| 194 | + EV_EFI_VARIABLE_DRIVER_CONFIG (KEK) |
| 195 | + |
| 196 | + EV_EFI_VARIABLE_DRIVER_CONFIG (db) |
| 197 | + |
| 198 | + EV_EFI_VARIABLE_DRIVER_CONFIG (dbx) |
| 199 | + |
| 200 | + EV_SEPARATOR |
| 201 | + |
| 202 | + Unknown event type (?!) |
| 203 | + |
| 204 | + |
| 205 | + In order to handle this, parse digests and event types into |
| 206 | + temporary |
| 207 | + |
| 208 | + files and iterate through them together. We only stop appending |
| 209 | + digests |
| 210 | + |
| 211 | + from the event log when the next event type is |
| 212 | + EV_EFI_VARIABLE_AUTHORITY |
| 213 | + |
| 214 | + (EFI binary signature) or we hit the end of the list. This |
| 215 | + should |
| 216 | + |
| 217 | + account for all possible variations. |
| 218 | + footer: |
| 219 | + Change-type: patch |
| 220 | + change-type: patch |
| 221 | + Signed-off-by: Joseph Kogut < [email protected]> |
| 222 | + signed-off-by: Joseph Kogut < [email protected]> |
| 223 | + author: Joseph Kogut |
| 224 | + nested: [] |
| 225 | + version: meta-balena-5.3.15 |
| 226 | + title: "" |
| 227 | + date: 2024-06-05T00:15:59.506Z |
| 228 | + version: 5.3.15 |
| 229 | + title: "" |
| 230 | + date: 2024-06-05T04:36:04.969Z |
1 | 231 | - commits:
|
2 | 232 | - subject: Update layers/meta-balena to 9e95be489587395fec796dc6969fc7d78cf81f77
|
3 | 233 | hash: b4458fb7118bc81a98e2a29f24ce0cc6bb058fe3
|
|
0 commit comments