]> git.proxmox.com Git - efi-boot-shim.git/log
efi-boot-shim.git
9 years agoMokManager: Support SHA224, SHA384, and SHA512
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:46:14 +0000 (11:46 -0400)]
MokManager: Support SHA224, SHA384, and SHA512

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoMokManager: Add more key list safe checks
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:46:14 +0000 (11:46 -0400)]
MokManager: Add more key list safe checks

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoMokManager: fix the return value and type
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
MokManager: fix the return value and type

There are some functions that the return value and the type
didn't match.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoMokManager: Support SHA1 hash in MOK
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
MokManager: Support SHA1 hash in MOK

Add SHA1 hash support and amend the code to make it easier to support
other SHA digests.

9 years agoMokManager: fix the hash list counting in delete
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
MokManager: fix the hash list counting in delete

match_hash() requests the number of keys in a list and it was
mistakenly replaced with the size of the Mok node. This would
made MokManager to remove the whole Mok node instead of one
hash.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoMokManager: calculate the variable size correctly
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
MokManager: calculate the variable size correctly

MokSize of the hash signature list includes the owner GUID,
so we should not add the 16bytes compensation.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoMake shim to check MokXAuth for MOKX reset
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
Make shim to check MokXAuth for MOKX reset

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoVerify the EFI images with MOK blacklist
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
Verify the EFI images with MOK blacklist

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoCopy the MOK blacklist to a RT variable
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
Copy the MOK blacklist to a RT variable

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoMokManager: Write the hash list properly
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
MokManager: Write the hash list properly

also return to the previous entry in the list

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoMokManager: Match all hashes in the list
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
MokManager: Match all hashes in the list

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoMokManager: delete the hash properly
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
MokManager: delete the hash properly

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoMokManager: show the hash list properly
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
MokManager: show the hash list properly

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoSupport MOK blacklist
Gary Ching-Pang Lin [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
Support MOK blacklist

The new blacklist, MokListX, stores the keys and hashes that are
banned.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoFix console_print_box*() parameters.
Peter Jones [Tue, 16 Jun 2015 15:41:32 +0000 (11:41 -0400)]
Fix console_print_box*() parameters.

When we made lib build with the correct CFLAGS, it inherited
-Werror=sign-compare, and I fixed up some parameters on
console_print_box() and console_print_box_at() to avoid sign comparison
errors.

The fixups were *completely wrong*, as some behavior relies on negative
values.  So this fixes them in a completely different way, by casting
appropriately to signed types where we're doing comparisons.

Signed-off-by: Peter Jones <pjones@redhat.com>
9 years agoEnsure that apps launched by shim get correct BS->Exit() behavior
Peter Jones [Thu, 11 Jun 2015 17:25:56 +0000 (13:25 -0400)]
Ensure that apps launched by shim get correct BS->Exit() behavior

Right now applications run by shim get our wrapper for Exit(), but it
doesn't do as much cleanup as it should - shim itself also exits, but
currently is not doing all the cleanup it should be doing.

This changes it so all of shim's cleanup is also performed.

Based on a patch and lots of review from Gary Lin.

Signed-off-by: Peter Jones <pjones@redhat.com>
9 years agoDon't leave in_protocol==1 when shim_verify() isn't enforcing.
Peter Jones [Thu, 11 Jun 2015 17:20:09 +0000 (13:20 -0400)]
Don't leave in_protocol==1 when shim_verify() isn't enforcing.

Right now if shim_verify() sees secure_mode()==0, it exits with
EFI_SUCCESS, but accidentally leaves in_protocol=1.  This means any
other call will have supressed error/warning messages.

That's wrong, so don't do it.

Signed-off-by: Peter Jones <pjones@redhat.com>
9 years agoOnly run MokManager if asked or a security violation occurs.
Peter Jones [Thu, 4 Jun 2015 14:19:30 +0000 (10:19 -0400)]
Only run MokManager if asked or a security violation occurs.

Don't run MokManager on any random error from start_image(second_stage);
only try it if it /is/ the second stage, or if start_image gave us
EFI_SECURITY_VIOLATION.

Signed-off-by: Peter Jones <pjones@redhat.com>
9 years agoMake the build failed with objcopy < 2.24
Gary Ching-Pang Lin [Tue, 12 May 2015 17:52:22 +0000 (13:52 -0400)]
Make the build failed with objcopy < 2.24

The wildcard support was introduced in objcopy since binutils 2.24.
However, objcopy < 2.24 never issues any warning message with the
wildcard and a faulty binary will be generated. This commit makes
the build failed as a notification for the usage of binutils < 2.24.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoUpdate Cryptlib and openssl
Gary Ching-Pang Lin [Tue, 12 May 2015 17:51:02 +0000 (13:51 -0400)]
Update Cryptlib and openssl

Update Cryptlib to r16559 and openssl to 0.9.8zf

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoExplicitly request sysv-style ELF hash sections
Matthew Garrett [Wed, 15 Apr 2015 17:30:52 +0000 (13:30 -0400)]
Explicitly request sysv-style ELF hash sections

We depend on there being a .hash section in the binary, and that's not
the case on distributions that default to building with gnu-style ELF
hashes. Explicitly request sysv-style hashes in order to avoid building
broken binaries.

Signed-off-by: Matthew Garrett <mjg59@coreos.com>
9 years agogcc 5.0 changes some include bits, so copy what arm does on x86.
Peter Jones [Mon, 13 Apr 2015 23:55:25 +0000 (19:55 -0400)]
gcc 5.0 changes some include bits, so copy what arm does on x86.

Basically they messed around with stdarg some and now we need to do it
the other way.

Signed-off-by: Peter Jones <pjones@redhat.com>
9 years agoMake lib/ use the right CFLAGS.
Peter Jones [Mon, 13 Apr 2015 23:55:25 +0000 (19:55 -0400)]
Make lib/ use the right CFLAGS.

Signed-off-by: Peter Jones <pjones@redhat.com>
9 years agoMake lib/ build right with the cflags it should be using...
Peter Jones [Mon, 13 Apr 2015 23:55:25 +0000 (19:55 -0400)]
Make lib/ build right with the cflags it should be using...

... but isn't.

Signed-off-by: Peter Jones <pjones@redhat.com>
9 years agoFix length of allocated buffer for boot option comparison.
Laszlo Ersek [Mon, 13 Apr 2015 23:55:25 +0000 (19:55 -0400)]
Fix length of allocated buffer for boot option comparison.

The following commit:

  commit 4aac8a1179e160397d7ef8f1e3232cfb4f3373d6
  Author: Gary Ching-Pang Lin <glin@suse.com>
  Date:   Thu Mar 6 10:57:02 2014 +0800

    [fallback] Fix the data size for boot option comparison

corrected the data size used for comparison, but also reduced the
allocation so it doesn't include the trailing UTF16LE '\0\0' at the
end of the string, with the result that the trailer of the buffer
containing the string is overwritten, which OVMF detects as memory
corruption.

Increase the size of the storage buffer in a few places to correct
this problem.

Signed-off-by: Richard W.M. Jones <rjones@redhat.com>
Cc: Laszlo Ersek <lersek@redhat.com>
Cc: Gary Ching-Pang Lin <glin@suse.com>
9 years agofallback: Fix comparison between signed and unsigned in debugging code.
Richard W.M. Jones [Mon, 13 Apr 2015 23:55:25 +0000 (19:55 -0400)]
fallback: Fix comparison between signed and unsigned in debugging code.

fallback.c: In function ‘update_boot_order’:
fallback.c:334:17: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]
  for (j = 0 ; j < size / sizeof (CHAR16); j++)
                   ^
fallback.c: In function ‘add_to_boot_list’:
fallback.c:402:16: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]
  for (i = 0; i < s; i++) {
                  ^

Signed-off-by: Richard W.M. Jones <rjones@redhat.com>
9 years agoDon't install our protocols if we're not in secure mode.
Peter Jones [Mon, 13 Apr 2015 23:55:25 +0000 (19:55 -0400)]
Don't install our protocols if we're not in secure mode.

System services haven't been hooked if we're not in secure mode, so
do_exit() will never be called.  In this case shim never gets control
once grub exits, which means if booting fails and the firmware tries
another boot option, it'll attempt to talk to the shim protocol we
installed.

This is wrong, because it is allowed to have been cleared from ram at
this time, since the task it's under has exited.

So just don't install the protocols when we're not enforcing.

This version also has a message and a 2-second stall after calling
start_image(), so that we can tell if we are on the expected return path
of our execution flow.

9 years agoAlign the sections we're loading, and check for validity /after/ discarding.
Peter Jones [Mon, 13 Apr 2015 23:55:25 +0000 (19:55 -0400)]
Align the sections we're loading, and check for validity /after/ discarding.

Turns out a) the codegen on aarch64 generates code that has real
alignment needs, and b) if we check the length of discardable sections
before discarding them, we error for no reason.

So do the error checking in the right order, and always enforce some
alignment because we know we have to.

Signed-off-by: Peter Jones <pjones@redhat.com>
9 years agoAdd nostdinc to the CFLAGS for lib
Gary Ching-Pang Lin [Thu, 11 Dec 2014 14:48:50 +0000 (09:48 -0500)]
Add nostdinc to the CFLAGS for lib

We don't need the headers from the standard include path.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
9 years agoBump version to 0.8
Peter Jones [Mon, 13 Oct 2014 20:41:51 +0000 (16:41 -0400)]
Bump version to 0.8

10 years agoCorrectly reject bad tftp addresses earlier, rather than later.
Peter Jones [Thu, 2 Oct 2014 05:01:54 +0000 (01:01 -0400)]
Correctly reject bad tftp addresses earlier, rather than later.

This check is for end == NULL but was meant to be *end == '\0'.  Without
this change, we'll pass a plausibly bad address (i.e. one with no ']' at
the end) to Mtftp(... READ_FILE ...), which should fail correctly, but
our error messaging will be inconsistent.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoUse -Werror=sign-compare .
Peter Jones [Thu, 2 Oct 2014 05:01:54 +0000 (01:01 -0400)]
Use -Werror=sign-compare .

I'm going to have to fix any errors that have this anyway, so may as
well do it here properly.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoMake another integer compare be signed/unsigned safe as well.
Peter Jones [Thu, 2 Oct 2014 05:01:54 +0000 (01:01 -0400)]
Make another integer compare be signed/unsigned safe as well.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoOOB access when parsing MOK List/Certificates on MOK enrollment
Sebastian Krahmer [Thu, 2 Oct 2014 05:01:54 +0000 (01:01 -0400)]
OOB access when parsing MOK List/Certificates on MOK enrollment

10 years agoshim buffer overflow on ipv6 option parsing
Sebastian Krahmer [Thu, 2 Oct 2014 05:01:54 +0000 (01:01 -0400)]
shim buffer overflow on ipv6 option parsing

10 years agoAnother testplan error.
Peter Jones [Thu, 2 Oct 2014 05:01:46 +0000 (01:01 -0400)]
Another testplan error.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoCryptlib: remove the unused files
Gary Ching-Pang Lin [Thu, 2 Oct 2014 04:10:47 +0000 (00:10 -0400)]
Cryptlib: remove the unused files

I mistakenly added CryptPkcs7VerifyNull.c which may make Pkcs7Verify
always return FALSE. Besides CryptPkcs7VerifyNull.c, there are some
functions we would never use. This commit removes those files to
avoid any potential trouble.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
10 years agoDon't verify images with the empty build key
Gary Ching-Pang Lin [Thu, 2 Oct 2014 04:08:50 +0000 (00:08 -0400)]
Don't verify images with the empty build key

We replaced the build key with an empty file while compiling shim
for our distro. Skip the verification with the empty build key
since this makes no sense.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
10 years agoFix some minor testplan errors.
Peter Jones [Thu, 2 Oct 2014 04:02:43 +0000 (00:02 -0400)]
Fix some minor testplan errors.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoDon't append an empty cert list to MokListRT if vendor_cert_size is 0.
Peter Jones [Thu, 2 Oct 2014 04:02:43 +0000 (00:02 -0400)]
Don't append an empty cert list to MokListRT if vendor_cert_size is 0.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoActually find the relocations correctly and process them that way.
Peter Jones [Wed, 1 Oct 2014 02:51:32 +0000 (22:51 -0400)]
Actually find the relocations correctly and process them that way.

Find the relocations based on the *file* address in the old binary,
because it's only the same as the virtual address some of the time.

Also perform some extra validation before processing it, and don't bail
out in /error/ if both ReloceBase and RelocEnd are null - that condition
is fine.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoRevert header changes
Peter Jones [Wed, 1 Oct 2014 02:49:21 +0000 (22:49 -0400)]
Revert header changes

Revert "Do the same for ia32..."
and "Generate a sane PE header on shim, fallback, and MokManager."
This reverts commit 6744a7ef8eca44948565c3d1244ec931ed3f6fee.
and commit 0e7ba5947eb38b79de2051ecf3b95055e620475c.

These are premature and I can do this without such drastic measures.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoMake list_keys() index variables all be signed.
Peter Jones [Sun, 21 Sep 2014 20:25:28 +0000 (16:25 -0400)]
Make list_keys() index variables all be signed.

We build with -Werror=signed-compare in fedora/rhel rpms, and this
showed up.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoDo the same for ia32...
Peter Jones [Sun, 21 Sep 2014 20:25:27 +0000 (16:25 -0400)]
Do the same for ia32...

Once again, on ia32 this time, we see:

00000120  47 84 00 00 0a 00 00 00  00 00 00 00 00 00 00 00 |G...............|

Which is where the pointer on ia32 for the Base Relocation Table should
be.  It points to 0x8447, which isn't a particularly reasonable address as
numbers go, and happens to have this data there:

00008440  6f 00 6e 00 66 00 69 00  67 00 75 00 72 00 65 00 |o.n.f.i.g.u.r.e.|
00008450  00 00 49 00 50 00 76 00  36 00 28 00 00 00 2c 00 |..I.P.v.6.(...,.|
00008460  25 00 73 00 2c 00 00 00  29 00 00 00 25 00 64 00 |%.s.,...)...%.d.|
00008470  2e 00 25 00 64 00 2e 00  25 00 64 00 2e 00 25 00 |..%.d...%.d...%.|
00008480  64 00 00 00 44 00 48 00  43 00 50 00 00 00 49 00 |d...D.H.C.P...I.|
00008490  50 00 76 00 34 00 28 00  00 00 2c 00 25 00 73 00 |P.v.4.(...,.%.s.|

And so that table is, in theory, this part:

00008447                       00  67 00 75 00 72 00 65 00 |       .g.u.r.e.|
00008450  00                                               |.               |

Which is pretty clearly not a pointer table of any kind.

So give ia32 the same treatment as x86_64, and now all arches work basically
the same.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoGenerate a sane PE header on shim, fallback, and MokManager.
Peter Jones [Sun, 21 Sep 2014 20:25:27 +0000 (16:25 -0400)]
Generate a sane PE header on shim, fallback, and MokManager.

It turns out a7249a65 was masking a second problem - on some binaries,
when we actually don't have any base relocations at all, binutils'
"objcopy --target efi-app-x86_64" is generating a PE header with a base
relocations pointer that happily points into the middle of our text
section.  So with shim processing base relocations correctly, it refuses
to load those binaries.

For example, on one binary I just built:

00000130  00 a0 00 00 0a 00 00 00  00 00 00 00 00 00 00 00 |................|

which says there's a Base Relocation Table at 0xa000 that's 0xa bytes long.
That's here:

0000a000  58 00 29 00 00 00 00 00  48 00 44 00 28 00 50 00 |X.).....H.D.(.P.|
0000a010  61 00 72 00 74 00 25 00  64 00 2c 00 53 00 69 00 |a.r.t.%.d.,.S.i.|
0000a020  67 00 25 00 67 00 29 00  00 00 00 00 00 00 00 00 |g.%.g.).........|
0000a030  48 00 44 00 28 00 50 00  61 00 72 00 74 00 25 00 |H.D.(.P.a.r.t.%.|

So the table is:

0000a000  58 00 29 00 00 00 00 00  48 00                   |X.).....H.      |

That wouldn't be so bad, except those binaries are MokManager.efi,
fallback.efi, and shim.efi, and sometimes they're .reloc, which we're
actually trying to handle correctly now because grub builds with a real
and valid .reloc table.  So though I didn't think there was any hair
left on this yak, more shaving ensues.

With this change, instead of letting objcopy do whatever it likes, we
switch to "-O binary" and merely link in a header that's appropriate for
our binaries.  This is the same method Ard wrote for aarch64, and it
seems to work fine in either place (modulo some minor changes.)

At some point this should be merged into gnu-efi instead of carrying our
own crt0-efi-x86_64.S, but that's a less immediate problem.

I did not need this problem.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoFix our "in_protocol" printing.
Peter Jones [Sun, 21 Sep 2014 20:25:27 +0000 (16:25 -0400)]
Fix our "in_protocol" printing.

When I merged 4bfb13d and fixed the conflicts, I managed to make the
in_protocol test exactly backwards, so that's why we don't currently see
error messages.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoDon't call AuthenticodeVerify if vendor_cert_size is 0.
Peter Jones [Sun, 21 Sep 2014 20:25:27 +0000 (16:25 -0400)]
Don't call AuthenticodeVerify if vendor_cert_size is 0.

Actually check the size of our vendor cert quite early, so that there's
no confusion as to what's going on.

This isn't strictly necessary, in that in all cases if vendor_cert_size
is 0, then AuthenticodeVerify -> Pkcs7Verify() -> d2i_X509() will result
in a NULL "Cert", and it will return FALSE, and we'll reject the
signature, but better to avoid all that code in the first place.  Belt
and suspenders and whatnot.

Based on a patch from https://github.com/TBOpen .

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoValidate computed hash bases/hash sizes more thoroughly.
Peter Jones [Sun, 21 Sep 2014 20:25:20 +0000 (16:25 -0400)]
Validate computed hash bases/hash sizes more thoroughly.

I screwed one of these up when working on 750584c, and it's a real pain
to figure out, so that means we should be validating them.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoMake 64-on-32 maybe work on x86_64.
Peter Jones [Sun, 21 Sep 2014 17:12:03 +0000 (13:12 -0400)]
Make 64-on-32 maybe work on x86_64.

This is mostly based on a patch (https://github.com/mjg59/shim/issues/30)
from https://github.com/TBOpen , which refactors our __LP64__
tests to be tests of the header magic instead.  I've simplified things
by using what we've pre-loaded into "context" and making some helper
functions so the conditionals in most of the code say what they do,
instead of how they work.

Note that we're only allowing that from in_protocol's loader - that is,
we'll let 64-bit grub load a 32-bit kernel or 32-bit grub load a 64-bit
kernel, but 32-bit shim isn't loading a 64-bit grub.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoActually refer to the base relocation table of our loaded image.
Peter Jones [Fri, 19 Sep 2014 13:30:26 +0000 (09:30 -0400)]
Actually refer to the base relocation table of our loaded image.

Currently when we process base relocations, we get the correct Data
Directory pointer from the headers (context->RelocDir), and that header
has been copied into our pristine allocated image when we copied up to
SizeOfHeaders.  But the data it points to has not been mirrored in to
the new image, so it is whatever data AllocPool() gave us.

This patch changes relocate_coff() to refer to the base relocation table
from the image we loaded from disk, but apply the fixups to the new
copy.

I have no idea how x86_64 worked without this, but I can't make aarch64
work without it.  I also don't know how Ard or Leif have seen aarch64
work.  Maybe they haven't?  Leif indicated on irc that they may have
only tested shim with simple "hello world" applications from gnu-efi;
they are certainly much less complex than grub.efi, and are generated
through a different linking process.

My only theory is that we're getting recycled data there pretty reliably
that just makes us /not/ process any relocations, but since our
ImageBase is 0, and I don't think we ever load grub with 0 as its base
virtual address, that doesn't follow.  I'm open to any other ideas
anybody has.

I do know that on x86_64 (and presumably aarch64 as well), we don't
actually start seeing *symptoms* of this bug until the first chunk[0] of
94c9a77f is applied[1].  Once that is applied, relocate_coff() starts
seeing zero[2] for both RelocBase->VirtualAddress and
RelocBase->SizeOfBlock, because RelocBase is a (generated, relative)
pointer that only makes sense in the context of the original binary, not
our partial copy.  Since RelocBase->SizeOfBlock is tested first,
relocate_base() gives us "Reloc block size is invalid"[3] and returns
EFI_UNSUPPORTED.  At that point shim exits with an error.

[0] The second chunk of 94c9a77f patch makes no difference on this
    issue.
[1] I don't see why at all.
[2] Which could really be any value since it's AllocatePool() and not
    AllocateZeroPool() results, but 0 is all I've observed; I think
    AllocatePool() has simply never recycled any memory in my test
    cases.
[3] which is silent because perror() tries to avoid talking because that
    has caused much crashing in the past; work needs to go in to 0.9 for
    this.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoMake sure we don't try to load a binary from a different arch.
Peter Jones [Wed, 27 Aug 2014 20:40:57 +0000 (16:40 -0400)]
Make sure we don't try to load a binary from a different arch.

Since in theory you could, for example, get an x86_64 binary signed that
also behaves as an ARM executable, we should be checking this before
people build on other architectures.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoDon't name something exit().
Peter Jones [Wed, 27 Aug 2014 20:40:57 +0000 (16:40 -0400)]
Don't name something exit().

On aarch64 due to some terrifying include chain we wind up with
Cryptlib's definition of exit here.  I'm not a glutton for punishment,
so I'm just changing the name so it's not coliding.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoHandle empty .reloc section in PE/COFF loader
Ard Biesheuvel [Wed, 27 Aug 2014 15:49:39 +0000 (11:49 -0400)]
Handle empty .reloc section in PE/COFF loader

On archs where no EFI aware objcopy is available, the generated PE/COFF
header contains a .reloc section which is completely empty. Handle this by
- returning early from relocate_coff() with EFI_SUCCESS,
- ignoring discardable sections in the section loader.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
10 years agoFix typo from Ard's old tree 32-bit ARM patch.
Peter Jones [Wed, 27 Aug 2014 15:49:39 +0000 (11:49 -0400)]
Fix typo from Ard's old tree 32-bit ARM patch.

We don't need to .data entries; the second one should be .data*.  He's
since fixed this in his tree, but I'd already pulled it and pushed to
master.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoUpdate openssl to 0.9.8zb
Gary Ching-Pang Lin [Tue, 19 Aug 2014 18:20:23 +0000 (14:20 -0400)]
Update openssl to 0.9.8zb

Also update to Tiano Cryptlib r15802 and remove the execute mode
bits from the C and header files of openssl

10 years agoAdd support for 32-bit ARM
Ard Biesheuvel [Tue, 12 Aug 2014 14:54:05 +0000 (10:54 -0400)]
Add support for 32-bit ARM

This adds support for building the shim for a 32-bit ARM UEFI environment.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
10 years agoAdd support for 64-bit ARM (AArch64)
Ard Biesheuvel [Tue, 12 Aug 2014 14:54:05 +0000 (10:54 -0400)]
Add support for 64-bit ARM (AArch64)

This adds support for building the shim for a 64-bit ARM UEFI environment.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
10 years agoFactor out x86-isms and add cross compile support
Ard Biesheuvel [Tue, 12 Aug 2014 14:54:05 +0000 (10:54 -0400)]
Factor out x86-isms and add cross compile support

This patch cleans up and refactors the Makefiles to better allow new
architectures to be added:
- remove unused Makefile definitions
- import Makefile definitions from top level rather than redefining
- move x86 specific CFLAGS to inside ifeq() blocks
- remove x86 inline asm
- allow $(FORMAT) to be overridden: this is necessary as there exists no
  EFI or PE/COFF aware objcopy for ARM

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
10 years agounhook_system_services: bail on systab == NULL
Ard Biesheuvel [Tue, 12 Aug 2014 14:54:05 +0000 (10:54 -0400)]
unhook_system_services: bail on systab == NULL

Prevent unhook_system_services() from dereferencing a NULL systab, which
may occur if hook_system_services() has never been called.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
10 years agoCryptLib: undefine va_arg and friends before redefining them
Ard Biesheuvel [Tue, 12 Aug 2014 14:54:05 +0000 (10:54 -0400)]
CryptLib: undefine va_arg and friends before redefining them

Upstream GNU-EFI contains changes to efistdarg.h resulting in the va_start,
va_arg and va_end macros to be #defined unconditionally. Make sure we #undef
them before overriding the definitions.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
10 years agoReplace build instructions in README with something not completely wrong.
Peter Jones [Mon, 21 Jul 2014 20:15:07 +0000 (16:15 -0400)]
Replace build instructions in README with something not completely wrong.

These were really, really out of date.

10 years agoUpdate openssl to 0.9.8za
Gary Ching-Pang Lin [Mon, 14 Jul 2014 13:03:36 +0000 (09:03 -0400)]
Update openssl to 0.9.8za

Also update to Tiano Cryptlib r15638

10 years agoSimplify the checking of SB and DB states
Gary Ching-Pang Lin [Wed, 25 Jun 2014 14:58:23 +0000 (10:58 -0400)]
Simplify the checking of SB and DB states

MokSBState and MokDBState are just 1 byte variables, so a UINT8
local variable is sufficient to include the content.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
Conflicts:
shim.c

10 years agoMake sure we default to assuming we're locked down.
Peter Jones [Wed, 25 Jun 2014 14:55:56 +0000 (10:55 -0400)]
Make sure we default to assuming we're locked down.

If "SecureBoot" exists but "SetupMode" does not, assume "SetupMode" says
we're not in Setup Mode.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoCheck the secure variables with the lib functions
Gary Ching-Pang Lin [Wed, 25 Jun 2014 14:55:12 +0000 (10:55 -0400)]
Check the secure variables with the lib functions

There are functions defined in lib to check the secure variables.
Use the functions to shun the duplicate code.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
Conflicts:
shim.c

10 years agoExplain the logic in secure_mode() better.
Peter Jones [Wed, 25 Jun 2014 14:46:52 +0000 (10:46 -0400)]
Explain the logic in secure_mode() better.

I was getting confused reading it, and I wrote it, so clearly it needs
more commentry.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoFree the string from DevicePathToStr
Gary Ching-Pang Lin [Wed, 25 Jun 2014 14:33:25 +0000 (10:33 -0400)]
Free the string from DevicePathToStr

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
Conflicts:
shim.c

10 years agoSilence the functions of shim protocol
Gary Ching-Pang Lin [Wed, 25 Jun 2014 14:30:38 +0000 (10:30 -0400)]
Silence the functions of shim protocol

When grub2 invokes the functions of shim protocol in gfx mode,
OutputString in shim could distort the screen.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
Conflicts:
shim.c

(modified by pjones to include some newer Prints that weren't there when
Gary did the initial work here.)

10 years agoRemove the duplicate calls in lib/console.c
Gary Ching-Pang Lin [Wed, 25 Jun 2014 14:15:31 +0000 (10:15 -0400)]
Remove the duplicate calls in lib/console.c

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
10 years agoNo newline for console_notify
Gary Ching-Pang Lin [Wed, 25 Jun 2014 14:12:43 +0000 (10:12 -0400)]
No newline for console_notify

The newlines are for Print(), not console_notify().

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
Conflicts:
shim.c

10 years agoExclude ca.crt while signing EFI images
Gary Ching-Pang Lin [Wed, 25 Jun 2014 14:03:08 +0000 (10:03 -0400)]
Exclude ca.crt while signing EFI images

If ca.crt was added into the certificate database, ca.crt would be the first
certificate in the signature. Because shim couldn't verify ca.crt with the
embedded shim.cer, it failed to load MokManager.efi.signed and
fallback.efi.signed.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
10 years agoMokManager: handle the error status from ReadKeyStroke
Gary Ching-Pang Lin [Wed, 25 Jun 2014 14:02:18 +0000 (10:02 -0400)]
MokManager: handle the error status from ReadKeyStroke

On some machines, even though the key event was signaled, ReadKeyStroke
still got EFI_NOT_READY. This commit handles the error status to avoid
console_get_keystroke from returning unexpected keys.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
Conflicts:
MokManager.c

10 years agoMokManager: delete the BS+NV variables the right way
Gary Ching-Pang Lin [Wed, 25 Jun 2014 13:57:10 +0000 (09:57 -0400)]
MokManager: delete the BS+NV variables the right way

LibDeleteVariable assumes that the variable is RT+NV and it
won't work on a BS+NV variable.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
10 years agoRemove grubpath in generate_path()
Gary Ching-Pang Lin [Wed, 25 Jun 2014 13:56:27 +0000 (09:56 -0400)]
Remove grubpath in generate_path()

The variable is not used anymore.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
10 years agoCheck the first 4 bytes of the certificate
Gary Ching-Pang Lin [Wed, 25 Jun 2014 13:55:49 +0000 (09:55 -0400)]
Check the first 4 bytes of the certificate

A non-DER encoding x509 certificate may be mistakenly enrolled into
db or MokList. This commit checks the first 4 bytes of the certificate
to ensure that it's DER encoding.

This commit also removes the iteration of the x509 signature list.
Per UEFI SPEC, each x509 signature list contains only one x509 certificate.
Besides, the size of certificate is incorrect. The size of the header must
be substracted from the signature size.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
10 years agoFetch the netboot image from the same device
Gary Ching-Pang Lin [Wed, 25 Jun 2014 13:53:23 +0000 (09:53 -0400)]
Fetch the netboot image from the same device

The previous strategy is to locate the first available PXE_BASE_CODE
protocol and to fetch the second stage image from it, and this may
cause shim to fetch the wrong second stage image, i.e. grub.efi.

Consider the machine with the following boot order:
1. PXE Boot
2. Hard Drive

Assume that the EFI image, e.g. bootx64.efi, in the PXE server is
broken, then "PXE Boot" will fail and fallback to "Hard Drive". While
shim.efi in "Hard Drive" is loaded, it will find the PXE protocol is
available and fetch grub.efi from the PXE server, not grub.efi in the
disk.

This commit checks the DeviceHandle from Loaded Image. If the device
supports PXE, then shim fetches grub.efi with the PXE protocol. Otherwise,
shim loads grub.efi from the disk.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
10 years ago[fallback] Try to boot the first boot option anyway
Gary Ching-Pang Lin [Tue, 13 May 2014 17:30:07 +0000 (13:30 -0400)]
[fallback] Try to boot the first boot option anyway

Some UEFI implementations never care the boot options, so the
restored boot options could be just ignored and this results in
endless reboot. To avoid this situation, this commit makes
fallback.efi to load the first matched boot option even if there
is no boot option to be restored. It may not be perfect, but at
least the bootloader is loaded...

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
10 years ago[fallback] Fix the data size for boot option comparison
Gary Ching-Pang Lin [Tue, 13 May 2014 17:24:12 +0000 (13:24 -0400)]
[fallback] Fix the data size for boot option comparison

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
10 years ago[fallback] Avoid duplicate old BootOrder
Gary Ching-Pang Lin [Tue, 13 May 2014 17:23:41 +0000 (13:23 -0400)]
[fallback] Avoid duplicate old BootOrder

set_boot_order() already copies the old BootOrder to the variable,
bootorder. Besides, we can adjust BootOrder when adding the newly
generated boot option. So, we don't have to copy the old one again
in update_boot_order(). This avoid the duplicate entries in BootOrder.

Signed-off-by: Gary Ching-Pang Lin <glin@suse.com>
10 years agoGet rid of SectionCache in generate_hash(), it is unused.
Peter Jones [Fri, 11 Apr 2014 19:07:45 +0000 (15:07 -0400)]
Get rid of SectionCache in generate_hash(), it is unused.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoKees' patch missed the offset adjustment to PEHdr.
Peter Jones [Fri, 11 Apr 2014 19:05:24 +0000 (15:05 -0400)]
Kees' patch missed the offset adjustment to PEHdr.

In read_header, we adjust context->PEHdr's address by doshdr->e_lfanew.
If we're going to recompute that address, we have to adjust it here
too.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoadditional bounds-checking on section sizes
Kees Cook [Fri, 11 Apr 2014 18:41:22 +0000 (14:41 -0400)]
additional bounds-checking on section sizes

This adds additional bounds-checking on the section sizes. Also adds
-Wsign-compare to the Makefile and replaces some signed variables with
unsigned counteparts for robustness.

Signed-off-by: Kees Cook <kees@ubuntu.com>
10 years agoAllow fallback to use the system's LoadImage/StartImage .
Peter Jones [Fri, 14 Feb 2014 22:48:01 +0000 (17:48 -0500)]
Allow fallback to use the system's LoadImage/StartImage .

Track use of the system's LoadImage(), and when the next StartImage()
call is for an image the system verified, allow that to count as
participating, since it has been verified by the system's db.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoAdd a failure case to the test plan and fix an ordering error.
Peter Jones [Fri, 14 Feb 2014 22:48:01 +0000 (17:48 -0500)]
Add a failure case to the test plan and fix an ordering error.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoAdd a preliminary test plan.
Peter Jones [Fri, 14 Feb 2014 19:08:30 +0000 (14:08 -0500)]
Add a preliminary test plan.

Because you know you wanted a test plan.  You feel it deeply inside.

Note that none of the /negative/ cases are tested yet.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years ago[fallback] Attempt to re-use existing entries when possible.
Peter Jones [Fri, 31 Jan 2014 15:31:10 +0000 (10:31 -0500)]
[fallback] Attempt to re-use existing entries when possible.

Some firmwares seem to ignore our boot entries and put their fallback
entries back on top.  Right now that results in a lot of boot entries
for our stuff, a la https://bugzilla.redhat.com/show_bug.cgi?id=995834 .

Instead of that happening, if we simply find existing entries that match
the entry we would create and move them to the top of the boot order,
the machine will continue to operate in failure mode (which we can't
avoid), but at least we won't create thousands of extra entries.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years ago[fallback] For HD() device paths, use just the media node and later.
Peter Jones [Fri, 31 Jan 2014 15:30:36 +0000 (10:30 -0500)]
[fallback] For HD() device paths, use just the media node and later.

UEFI 2.x section 3.1.2 provides for "short-form device path", where the
first element specified is a "hard drive media device path", so that you
can move a disk around on different buses without invalidating your
device path.  Fallback has not been using this option, though in most
cases efibootmgr has.

Note that we still keep the full device path, because LoadImage()
isn't necessarily the layer where HD() works - one some systems BDS is
responsible for resolving the full path and passes that to LoadImage()
instead.  So we have to do LoadImage() with the full path.

10 years agoError check the right thing in get_variable_attr() when allocating.
Peter Jones [Thu, 21 Nov 2013 16:48:24 +0000 (11:48 -0500)]
Error check the right thing in get_variable_attr() when allocating.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoRewrite directory traversal allocation path so coverity can grok it.
Peter Jones [Thu, 21 Nov 2013 16:48:24 +0000 (11:48 -0500)]
Rewrite directory traversal allocation path so coverity can grok it.

The things we do for our tools.  In this case, make the AllocatePool()
happen outside of a conditional, even though that conditional will
always bee satisfied.  This way coverity won't think we're setting fi
to NULL and passing it to StrCaseCmp.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoInitialize entries before we pass it to another function.
Peter Jones [Thu, 21 Nov 2013 16:48:24 +0000 (11:48 -0500)]
Initialize entries before we pass it to another function.

Coverity scan noticed that entries is uninitialized when we pass its
location to another function.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoFix wrong sizeof().
Peter Jones [Thu, 21 Nov 2013 16:48:24 +0000 (11:48 -0500)]
Fix wrong sizeof().

CHAR16* vs CHAR16**, so the result is the same on all platforms.

Detected by coverity.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoLengths that might be -1 can't be unsigned, Peter.
Peter Jones [Thu, 21 Nov 2013 16:48:24 +0000 (11:48 -0500)]
Lengths that might be -1 can't be unsigned, Peter.

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoFix path generation for Dhcpv4 bootloader.
Peter Jones [Thu, 21 Nov 2013 16:48:24 +0000 (11:48 -0500)]
Fix path generation for Dhcpv4 bootloader.

Right now we always look for e.g. "\grubx64.efi", which is completely
wrong.  This makes it look for the path shim was loaded from and modify
that to end in a sanitized version of our default loader name.

Resolves: rhbz#1032583

Signed-off-by: Peter Jones <pjones@redhat.com>
10 years agoDon't hook system services if shim has no built-in keys
Matthew Garrett [Tue, 19 Nov 2013 15:20:34 +0000 (10:20 -0500)]
Don't hook system services if shim has no built-in keys

Shim should only need to enforce its security policy when its launching
binaries signed with its built-in key. Binaries signed by keys in db or
Mokdb should be able to rely on their own security policy.

Signed-off-by: Matthew Garrett <matthew.garrett@nebula.com>
10 years agoClarify meaning of insecure_mode
Matthew Garrett [Tue, 19 Nov 2013 15:20:34 +0000 (10:20 -0500)]
Clarify meaning of insecure_mode

insecure_mode was intended to indicate that the user had explicity disabled
checks with mokutil, which means it wasn't the opposite of secure_mode().
Change the names to clarify this and don't show the insecure mode message
unless the user has explicitly enabled that mode.

Signed-off-by: Matthew Garrett <matthew.garrett@nebula.com>
10 years agoshim: improve error messages
Andrew Boie [Tue, 12 Nov 2013 15:32:48 +0000 (10:32 -0500)]
shim: improve error messages

%r when used in Print() will show a string representation of
an EFI_STATUS code.

Change-Id: I6db47f5213454603bd66177aca378ad01e9f0bd4
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
10 years agoallow 32-bit compilation with 64-bit compiler
Andrew Boie [Tue, 12 Nov 2013 15:31:59 +0000 (10:31 -0500)]
allow 32-bit compilation with 64-bit compiler

Also removed unused LIB_PATH from some Makefiles.

Change-Id: I7d28d18f7531b51b6121a2ffb88bcaedec57c467
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
10 years agopropagate some path variables
Andrew Boie [Tue, 12 Nov 2013 15:30:53 +0000 (10:30 -0500)]
propagate some path variables

If these are overridden on the command line, pass them along to
the sub-makes.

Change-Id: I531ccb5d2f5e4be8e99d4892cdcfffffc1ad9877
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
10 years agofix fallback.so build dependency
Andrew Boie [Tue, 12 Nov 2013 15:30:30 +0000 (10:30 -0500)]
fix fallback.so build dependency

Exposed during parallel builds

Change-Id: I9867858166dcafd69438f37ee5da14a267ace8f4
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
10 years agofallback.c: fix 32-bit compilation
Andrew Boie [Tue, 12 Nov 2013 15:30:02 +0000 (10:30 -0500)]
fallback.c: fix 32-bit compilation

fh->Read expects pointer to 32-bit int, use UINTN

Change-Id: If1a728efd51a9a24dfcd8123e84bf4c0713491fe
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>