Give linux-*'s ->parse() functions the unmodified device link as well

Signed-off-by: Peter Jones <pjones@redhat.com>

Improve ACPI device path formatting

This factors a bunch of the duplication out to another function, which
also does a better job of it.

Signed-off-by: Peter Jones <pjones@redhat.com>

Fix a makeguids building problem with generics.h.

Guard generics.h with EFIVAR_BUILD_ENVIRONMENT to keep it from
interfering with the makeguids build if libefivar.so isn't around
already.

Signed-off-by: Peter Jones <pjones@redhat.com>

Make path-helpers.c also import fix_coverity.h

Signed-off-by: Peter Jones <pjones@redhat.com>

Make efidp_make_file() have even more, better input constraints.

This is all in the effort to convince coverity that it doesn't
dereference buf when size==0, which it already doesn't.

Signed-off-by: Peter Jones <pjones@redhat.com>

Try to convince covscan that sysfs_read_file() doesn't leak on error.

Basically, covscan gets confused about some of our return paths and
doesn't  think the error condition correlates with not having allocated
(or having freed) the ram we're using to pass the file data back.

Signed-off-by: Peter Jones <pjones@redhat.com>

Promote _make_hd_dn() to make_hd_dn() and get rid of the wrapper.

The wrapper is just hiding what the code's doing, and all the other code
around where we use it does the same thing anyway.  This hopefully
convinces coverity we're not really dereferencing "buf" there unless
it's nonzero.

Signed-off-by: Peter Jones <pjones@redhat.com>

efivar: Fix some types in -L behavior to pacify coverity.

Coverity doesn't realize that efi_well_known_guids is /actually/ an
array, because we didn't tell it so.  So fix the declaration so we've
told it so.

Signed-off-by: Peter Jones <pjones@redhat.com>

Pacify some coverity nits.

Coverity has trouble tracking data flow sometimes, and believes that
sysfs_readlink() and read_sysfs_file() will sometimes return >= 0 when
the buffer has not been filled out.  This changes the check to also test
for a NULL pointer, hopefully pacifying it.

Signed-off-by: Peter Jones <pjones@redhat.com>

Move verbosity headers to be public

Signed-off-by: Peter Jones <pjones@redhat.com>

Move the syntastic file I use out of the repo

Signed-off-by: Peter Jones <pjones@redhat.com>

Bump version to 36

- Add NVDIMM support
- Re-written linux interface parser to handle how devices are
  partitioned better, and for cleaner code, with one file per device
  type.
- lots of verbosity updates
- better CI
- analysis with clang's analyzer as well as coverity
- Better handling of immutable bits in sysfs
- LIBEFIVAR_OPS=help
- lots of code cleanups.

Signed-off-by: Peter Jones <pjones@redhat.com>

Make ccc-analyzer get clang cflags

Signed-off-by: Peter Jones <pjones@redhat.com>

Fix some errors scan-build found.

This fixes a couple of failures to check for NULL in string parsing, and
one failure to free a value in an error path.

Signed-off-by: Peter Jones <pjones@redhat.com>

nvdimm: Use the namespace guid, not just the device guid

This makes long-form device paths for nvdimm be:

NVDIMM(pmem-namespace-guid)/NVDIMM(pmem-label-guid)/HD(...)/File(...)

It also conditionally removes our UUID swizzling, because nobody can
figure out what the correct encoding is, and only byteswaps if
LIBEFIBOOT_SWIZZLE_PMEM_UUID is set (to any value) in the environment.

Signed-off-by: Peter Jones <pjones@redhat.com>

Fix an edge case in finding the sysfs driver link

Signed-off-by: Peter Jones <pjones@redhat.com>

Use the pci node's driver link to detect ATA in some cases

efivarfs: fix some minor immutability bugs

Signed-off-by: Peter Jones <pjones@redhat.com>

Expand tabs on the rest of linux.c

Signed-off-by: Peter Jones <pjones@redhat.com>

Always build the .so.1 local symlink

Signed-off-by: Peter Jones <pjones@redhat.com>

audit efi_error() for wrong newlines.

Signed-off-by: Peter Jones <pjones@redhat.com>

Update our abixml files for new ABI bits and minor changes.

Currently abidiff notices the following:

abidiff \
--suppr abignore \
--headers-dir2 /home/pjones/devel/github.com/efivar/master/src/include/efivar/ \
libefivar.abixml \
libefivar.so
Functions changes summary: 0 Removed, 1 Changed, 3 Added functions
Variables changes summary: 0 Removed, 0 Changed, 0 Added variable

3 Added functions:

  'function FILE* efi_get_logfile()'    {efi_get_logfile@@LIBEFIVAR_1.36}
  'function int efi_get_verbose()'    {efi_get_verbose@@LIBEFIVAR_1.36}
  'function void efi_set_verbose(int, FILE*)'    {efi_set_verbose@@LIBEFIVAR_1.36}

1 function with some indirect sub-type change:

  [C]'function ssize_t efidp_make_acpi_hid_ex(uint8_t*, ssize_t, uint32_t, uint32_t, uint32_t, char*, char*, char*)' at <built-in>:263:1 has some indirect sub-type changes:
    parameter 6 of type 'char*' changed:
      in pointed to type 'char':
        entity changed from 'char' to 'const char'
        type size hasn't changed
    parameter 7 of type 'char*' changed:
      in pointed to type 'char':
        entity changed from 'char' to 'const char'
        type size hasn't changed
    parameter 8 of type 'char*' changed:
      in pointed to type 'char':
        entity changed from 'char' to 'const char'
        type size hasn't changed

The 3 added functions are as they should be.  The
efidp_make_acpi_hid_ex() changes are all from making arguments "const"
that weren't before, and that's fine as well.

Signed-off-by: Peter Jones <pjones@redhat.com>

Add a make target to rebuild our abidw .abixml files

Signed-off-by: Peter Jones <pjones@redhat.com>

Rebuild efivar-35's abixml files with newer abidw.

abidiff is not that good at recognizing its own format changes, so if we
just do the normal abicheck we get a lot of:

  [C]'function ssize_t efidp_make_atapi(uint8_t*, ssize_t, uint16_t, uint16_t, uint16_t)' has some indirect sub-type changes:
    parameter 6 of type 'uint8_t*' was added
    parameter 7 of type 'typedef ssize_t' was added
    parameter 8 of type 'typedef uint16_t' was added
    parameter 9 of type 'typedef uint16_t' was added
    parameter 10 of type 'typedef uint16_t' was added

Which is obviously just plain wrong (efidp_make_atapi()'s arguments have
not changed at all, and it still has only 5 of them.)

So the answer is to re-build efivar-35's .abixml files with newer abidw,
and *then* do abicheck.

Signed-off-by: Peter Jones <pjones@redhat.com>

libefiboot: make everything use "struct device" and kill "struct disk_info"

Signed-off-by: Peter Jones <pjones@redhat.com>

linux.h: Make some decls HIDDEN as they should be

Signed-off-by: Peter Jones <pjones@redhat.com>

Fix decl for get_sector_size()

Signed-off-by: Peter Jones <pjones@redhat.com>

Constify efidp_make_acpi_hid_ex() args and make its prototype more readible

Signed-off-by: Peter Jones <pjones@redhat.com>

Move the PMEM device parser to linux-pmem.c

This won't actually *work* yet, because the infrastructure to use it
(replacing "struct disk_info") won't land for a few more patches.

Signed-off-by: Peter Jones <pjones@redhat.com>

Add a PCI parser that's independent of the other stuff.

This doesn't actually remove the old one, but a later patch will when
it's replacing all that stuff.

This won't actually *work* yet, either, because the infrastructure to
use it (replacing "struct disk_info") won't land for a few more patches.

Signed-off-by: Peter Jones <pjones@redhat.com>

Move SAS parsing to linux-sas.c

This won't actually *work* yet, because the infrastructure to use it
(replacing "struct disk_info") won't land for a few more patches.

Signed-off-by: Peter Jones <pjones@redhat.com>

Move SATA parsing to linux-sata.c

This won't actually *work* yet, because the infrastructure to use it
(replacing "struct disk_info") won't land for a few more patches.

Signed-off-by: Peter Jones <pjones@redhat.com>

Move nvme parsing to linux-nvme.c

This won't actually *work* yet, because the infrastructure to use it
(replacing "struct disk_info") won't land for a few more patches.

Signed-off-by: Peter Jones <pjones@redhat.com>

Move virtblk device parsing to linux-virtblk.c

This won't actually *work* yet, because the infrastructure to use it
(replacing "struct disk_info") won't land for a few more patches.

Signed-off-by: Peter Jones <pjones@redhat.com>

Move i2o device parsing to linux-i2o.c

This won't actually *work* yet, because the infrastructure to use it
(replacing "struct disk_info") won't land for a few more patches.

Signed-off-by: Peter Jones <pjones@redhat.com>

Move traditional SCSI parsing to linux-scsi.c

This won't actually *work* yet, because the infrastructure to use it
(replacing "struct disk_info") won't land for a few more patches.

Signed-off-by: Peter Jones <pjones@redhat.com>

Move ATA parsing to linux-ata.c

This won't actually *work* yet, because the infrastructure to use it
(replacing "struct disk_info") won't land for a few more patches.

Signed-off-by: Peter Jones <pjones@redhat.com>

Add a (not great) macro to print parser debugging arrows.

Signed-off-by: Peter Jones <pjones@redhat.com>

Add "struct device" in preperation for removing "struct disk_info".

Also adds src/linux-*.c to the makefile

Signed-off-by: Peter Jones <pjones@redhat.com>

efi_loadopt_create(): add some debugging

Signed-off-by: Peter Jones <pjones@redhat.com>

ucs2.h: there's no reason ucs2_to_utf8 shouldn't malloc directly.

Signed-off-by: Peter Jones <pjones@redhat.com>

rename the pci_root_info fields to match the struct name

Signed-off-by: Peter Jones <pjones@redhat.com>

Move get_sector_size() to the linux specific code

Signed-off-by: Peter Jones <pjones@redhat.com>

linux.h: add some more sysfs helpers, and improve their error handling.

... this also moves them from util.h

Signed-off-by: Peter Jones <pjones@redhat.com>

More fixups for CONSTRUCTOR/DESTRUCTOR/HIDDEN/PUBLIC/etc.

Signed-off-by: Peter Jones <pjones@redhat.com>

Add some helpers for file path mangling

Signed-off-by: Peter Jones <pjones@redhat.com>

util.h: Add error reporting in read_file()

Signed-off-by: Peter Jones <pjones@redhat.com>

Temporary hack in get_partition_number

This code has been wrong for a while, but it's getting removed, so
really don't bother *fixing* it so much as appeasing newer gcc's
noticing that it's uninitialized.

Signed-off-by: Peter Jones <pjones@redhat.com>

Make LIBEFIVAR_OPS=help work

Signed-off-by: Peter Jones <pjones@redhat.com>

Add some more debug facilities.

Signed-off-by: Peter Jones <pjones@redhat.com>

Add hexdumper

Make all of our Make.* files have vim:ft=make

... except Make.version because it's fully automated.

Signed-off-by: Peter Jones <pjones@redhat.com>

fix some dumb stuff in 'make install' targets

Signed-off-by: Peter Jones <pjones@redhat.com>

gpt.h: minor cleanups (mostly formatting/whitespace)

Signed-off-by: Peter Jones <pjones@redhat.com>

get_partition_info(): fix the return code error path.

Half the code thinks >= 0 is success, half the code things <= 0 is
success...

Signed-off-by: Peter Jones <pjones@redhat.com>

clang: ?e64_to_cpu() and cpu_to_?e64(): cast to uint64_t.

On x86_64, clang makes __builtin_bswap64() return "unsigned long"
instead of "unsigned long long", so using PRIx64 complains all over the
place.  Casting to uint64_t satisfies both compilers.

Signed-off-by: Peter Jones <pjones@redhat.com>

More aggressively use macros for compiler attributes

Signed-off-by: Peter Jones <pjones@redhat.com>

Simplify our include files

Make everything just include src/efivar.h or src/efiboot.h, and let
those sort out the correct things to include.

Signed-off-by: Peter Jones <pjones@redhat.com>

util.h: whitespace cleanup

Signed-off-by: Peter Jones <pjones@redhat.com>

gpt: get rid of ad-hoc warning reporting; use efi_error()

Signed-off-by: Peter Jones <pjones@redhat.com>

gpt: don't use the __ versions of cpu_to_le??()/le??_to_cpu()

Signed-off-by: Peter Jones <pjones@redhat.com>

efivar: fix efi_well_known_guids and efi_well_known_guids_end definitions

From github:

build command: make CROSS_COMPILE=aarch64-linux-gnu- prefix=/home/b58867/Desktop/ls1043ardb/efivar/efivar_arm64 static

compilation error:
aarch64-linux-gnu-gcc -O2 -flto -g3 -I/home/b58867/Desktop/ls1043ardb/efivar/src/include/ -specs=/home/b58867/Desktop/ls1043ardb/efivar/gcc.specs -L. -Wl,-z,muldefs -static -o efivar-static efivar.c dp.static.o dp-acpi.static.o dp-hw.static.o dp-media.static.o dp-message.static.o efivarfs.static.o error.static.o export.static.o guid.static.o guids.static.o guid-symbols.static.o lib.static.o vars.static.o -ldl
guid.c:112:24: error: type of ‘efi_well_known_guids_end’ does not match original declaration [-Werror]
extern struct guidname efi_well_known_guids_end
^
efivar.c:490:28: note: previously declared here
extern struct guidname *efi_well_known_guids_end;
^
efivar.c:489:27: error: type of ‘efi_well_known_guids’ does not match original declaration [-Werror]
extern struct guidname efi_well_known_guids[];
^
guid.c:108:24: note: previously declared here
extern struct guidname efi_well_known_guids
^
lto1: all warnings being treated as errors
lto-wrapper: fatal error: aarch64-linux-gnu-gcc returned 1 exit status

Fixes github issue #102

Signed-off-by: Peter Jones <pjones@redhat.com>

Fix our pkg-config-ldflags invocation.

Signed-off-by: Peter Jones <pjones@redhat.com>

We don't use list.h at all; axe it.

Signed-off-by: Peter Jones <pjones@redhat.com>

Another minor docs typo fix.

Signed-off-by: Peter Jones <pjones@redhat.com>

minor correction to the efi_variable_import and efi_variable_export man-page

rewrite efivarfs_set_variable() [9896c26c7e68-based]

This patch rewrites the efivarfs_set_variable() function, to address the
following issues:

- a size_t value is printed with %zd -- size_t is unsigned, so it should
  be printed with %zu or %zx,

- a VLA is used for storing input of basically unbounded size -- we should
  use a range-checked malloc() call instead,

- the efivarfs file is opened for writing while it may be immutable --
  this is the trickiest issue to resolve,

- passing just O_APPEND|O_CREAT to open() is undefined -- O_WRONLY is
  required, and O_APPEND and (O_CREAT | O_EXCL) should both be independent
  of it (and of each other),

- some error branches call efi_error() without setting errno first,

- the variable is removed on any write failure, even if we didn't create
  the variable -- failed writes are expected to be atomic (from the kernel
  side and from the firmware side) and not to leave behind side effects;
  so only delete the variable on error if we created it.

A small helper function efivarfs_make_fd_mutable() is introduced as well.

(It's best to review the new efivarfs_set_variable() function in its
entirety, with the patch applied, rather than comparing old vs. new, hunk
for hunk.)

Ref: https://bugzilla.redhat.com/show_bug.cgi?id=1516599
Signed-off-by: Laszlo Ersek <lersek@redhat.com>

Silence some trivial clang-analyzer/coverity complaints.

One minor memory leak in /usr/bin/efivar, a couple of places where the
analyzers make assumptions without being able to figure out the data
flow.

Signed-off-by: Peter Jones <pjones@redhat.com>

dp-acpi: improve formatting of AcpiEx() device paths.

This also fixes a minor scan-build complaint about cidlen being assigned
but not used.

Signed-off-by: Peter Jones <pjones@redhat.com>

Add a clang-analyzer scan-build make targets as well.

Signed-off-by: Peter Jones <pjones@redhat.com>

Work around coverity being a pretty lazy and poor tool.

I mean, come on, at least dynamically import the types you've missed
from the version of GCC that's installed on the system and you're
pretending to emulate.

Signed-off-by: Peter Jones <pjones@redhat.com>

Update travis to use some better build scripts

Signed-off-by: Peter Jones <pjones@redhat.com>

Simplify the docker/travis stuff some

Signed-off-by: Peter Jones <pjones@redhat.com>

try to make travis work again...

Bump version to 35

Signed-off-by: Peter Jones <pjones@redhat.com>

Don't use _Generic because gcc 4.x doesn't have it...

Signed-off-by: Peter Jones <pjones@redhat.com>

Make efi_get_variable_exists match the version we put on it.

Signed-off-by: Peter Jones <pjones@redhat.com>

Make non-pull-request builds work.

Signed-off-by: Peter Jones <pjones@redhat.com>

Use https for github urls in travis.

If I don't, I awesomely get:

git remote add remote git@github.com:${remote_repo}
git fetch remote
Host key verification failed.
fatal: Could not read from remote repository.

Signed-off-by: Peter Jones <pjones@redhat.com>

more travis stuf...

Signed-off-by: Peter Jones <pjones@redhat.com>

More travis stuff maybe.

Signed-off-by: Peter Jones <pjones@redhat.com>

Make the log smaller...

Signed-off-by: Peter Jones <pjones@redhat.com>

Some more travis work...

Signed-off-by: Peter Jones <pjones@redhat.com>

Make travis dump some stuff at us.

Signed-off-by: Peter Jones <pjones@redhat.com>

Work around -Werror=stringop-overflow= being daft.

With:

len = strlen(foo + offset) + 1;
buf = calloc(1, len);
if (!buf)
err(1, "out of memory");
strncpy(buf, foo+offset, len);

-Wstringop-overflow complains:

efivar.c:169:2: error: 'strncpy' specified bound depends on the length of the source argument [-Werror=stringop-overflow=]
  strncpy(name_buf, guid_name + name_pos, name_len);
  ^
efivar.c:163:13: note: length computed here
  name_len = strlen(guid_name + name_pos) + 1;
             ^
lto1: all warnings being treated as errors

Which... Duh, so was the allocation it's writing into.  So what?

Signed-off-by: Peter Jones <pjones@redhat.com>

Update travis.yml to be wildly different...

Make 0abba7dc-e516-4167-bbf5-4d9d1c739416 reflect its use in fwupdate

Signed-off-by: Peter Jones <pjones@redhat.com>

makeguids: generalize the efi_guid_empty mechanism to arbitrary aliases.

Signed-off-by: Peter Jones <pjones@redhat.com>

Add efi_get_variable_exists()

Sometimes we just want to know if a variable exists, and don't care about the
attributes or what the variable contains. Use efi_get_variable_attributes()
under the covers to avoid adding yet-another-fvunc to all the backends.

Use __builtin_bswap16() in EFI_GUID and other places.

This makes EFI_GUID() usually not be an expression that has to be in a
function.

Signed-off-by: Peter Jones <pjones@redhat.com>

efivarfs / vars: usleep() before reading from efivarfs if euid != 0

There's a kernel rate limiter on efi variable reads now for
non-root users, and we'd rather just not hit it than have to dig out
from having hit it.  So this adds a 10ms sleep before each read call.

If you do have 50 variables, efibootmgr will do 100 reads, which would
trigger the rate limit.  In that case, this patch adds 1 second (plus
lossage due to calling, etc.), so it should stay just below the
triggering threshold.  That will definitely be /smoother/ than hitting
it, and almost certainly faster as well, because the extra calls will
re-enforce the limit.

Signed-off-by: Peter Jones <pjones@redhat.com>

Bump version to 34

Signed-off-by: Peter Jones <pjones@redhat.com>

Add UX capsule to map for 33 not 32

commit cd732494 fixed UX capsule support, but it mistakingly marked
it for version 32.  It really should be marked for 33.

Add a patch from Helmut Grohne for cross compiling

Originally came from
https://bugs.debian.org/cgi-bin/bugreport.cgi?att=1;bug=843261;filename=efivar_30-1.1.debdiff;msg=5

Clean up superfluous slashes

Use correct paths in pc files

Previously, the paths in the pkgconfig files were hardcoded resulting
in broken include paths on NixOS.

This patch replaces the paths with placeholders that will be changed
during making.

Be more stringent on the criteria for not doing disk probes.

This bit of the nvdimm code stops HD() paths from getting generated
right sometimes.

Signed-off-by: Peter Jones <pjones@redhat.com>

define strndupa if it is not defined

musl does not include strndupa in <string.h>. Define strndupa if it has
not already been defined.

cleanup: found is always true in branch

Remove use of found in conditional, as it is always true.

Bump version to 33

- Add NVDIMM support.
- Doesn't change much, but we need to bump to 33 because of an fwupdate bug.

Signed-off-by: Peter Jones <pjones@redhat.com>