| Message ID | 20221122020404.3476063-2-Jason@zx2c4.com (mailing list archive) |
|---|---|
| State | Not Applicable |
| Delegated to: | Herbert Xu |
| Headers | show |
| Series | Use EFI variables for random seed | expand |
On Tue, Nov 22, 2022 at 03:04:00AM +0100, Jason A. Donenfeld wrote: > In anticipation of putting random seeds in EFI variables, it's important > that the random GUID namespace of variables remains hidden from > userspace. We accomplish this by not populating efivarfs with entries > from that GUID, as well as denying the creation of new ones in that > GUID. What's the concern here? Booting an older kernel would allow a malicious actor to either read the seed variable or set it to a value under their control, so we can't guarantee that the information is secret.
Hi, On Sun, Nov 27, 2022 at 09:00:40PM +0000, Matthew Garrett wrote: > On Tue, Nov 22, 2022 at 03:04:00AM +0100, Jason A. Donenfeld wrote: > > In anticipation of putting random seeds in EFI variables, it's important > > that the random GUID namespace of variables remains hidden from > > userspace. We accomplish this by not populating efivarfs with entries > > from that GUID, as well as denying the creation of new ones in that > > GUID. > > What's the concern here? Booting an older kernel would allow a malicious > actor to either read the seed variable or set it to a value under their > control, so we can't guarantee that the information is secret. The security model is the same as that of random seed files, on, say, BSD. If you remove the hard drive or change the operating system or what have you, then sure, you can fiddle with the seed and read it. But the running operating system shouldn't show it to you if it can help it. Consider, for example, systemd's use of EFI variables for the SystemToken. There, they have PID 1 take care of chmod'ing it before other processes start. But of course a different OS or even EFI shell could just read it. So, think of this as just basic runtime safety -- like what people do when they set the umask before writing a random seed file -- rather than some type of ultimate secrecy. (And either way, the larger picture is that it's much more important to get as much random data from as many sources as possible as soon as possible, rather than being overly paranoid about every one single source that we start excluding sources. A plethora of sources is better off here.) Jason
diff --git a/fs/efivarfs/inode.c b/fs/efivarfs/inode.c index 939e5e242b98..617f3ad2485e 100644 --- a/fs/efivarfs/inode.c +++ b/fs/efivarfs/inode.c @@ -91,6 +91,10 @@ static int efivarfs_create(struct user_namespace *mnt_userns, struct inode *dir, err = guid_parse(dentry->d_name.name + namelen + 1, &var->var.VendorGuid); if (err) goto out; + if (guid_equal(&var->var.VendorGuid, &LINUX_EFI_RANDOM_SEED_TABLE_GUID)) { + err = -EPERM; + goto out; + } if (efivar_variable_is_removable(var->var.VendorGuid, dentry->d_name.name, namelen)) diff --git a/fs/efivarfs/super.c b/fs/efivarfs/super.c index 6780fc81cc11..07e82e246666 100644 --- a/fs/efivarfs/super.c +++ b/fs/efivarfs/super.c @@ -116,6 +116,9 @@ static int efivarfs_callback(efi_char16_t *name16, efi_guid_t vendor, int err = -ENOMEM; bool is_removable = false; + if (guid_equal(&vendor, &LINUX_EFI_RANDOM_SEED_TABLE_GUID)) + return 0; + entry = kzalloc(sizeof(*entry), GFP_KERNEL); if (!entry) return err;
In anticipation of putting random seeds in EFI variables, it's important that the random GUID namespace of variables remains hidden from userspace. We accomplish this by not populating efivarfs with entries from that GUID, as well as denying the creation of new ones in that GUID. Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com> --- fs/efivarfs/inode.c | 4 ++++ fs/efivarfs/super.c | 3 +++ 2 files changed, 7 insertions(+)