Make sure to never access beyond bounds of either EFI partition name
or DOS partition name. This situation is happening:
part.h: disk_partition_t->name is 32-byte long
part_efi.h: gpt_entry->partition_name is 36-bytes long
The loop in part_efi.c copies over 36 bytes and thus accesses beyond
the disk_partition_t->name .
Fix this by picking the shortest of source and destination arrays and
make sure the destination array is cleared so the trailing bytes are
zeroed-out and don't cause issues with string manipulation.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Tom Rini <trini@ti.com>
Cc: Simon Glass <sjg@chromium.org>
For ISO we check the block size of the device if this is != the CD sector
size we assume that the device has no ISO partition.
Signed-off-by: Egbert Eich <eich@suse.com>
Disks beyond 2T in size use blocksizes of 4096 bytes. However a lot of
code in u-boot still assumes a 512 byte blocksize.
This patch fixes the handling of GPTs.
Signed-off-by: Egbert Eich <eich@suse.com>
Devices that used to have a whole disk FAT filesystem but got then
partitioned will most likely still have a FAT or FAT32 signature
in the first sector as this sector does not get overwritten by
a partitioning tool (otherwise the tool would risk to kill the mbr).
The current partition search algorithm will erronously detects such
a device as a raw FAT device.
Instead of looking for the FAT or FAT32 signatures immediately we
use the same algorithm as used by the Linux kernel and first check
for a valid boot indicator flag on each of the 4 partitions.
If the value of this flag is invalid for the first entry we then
do the raw partition check.
If the flag for any higher partition is wrong we assume the device
is neiter a MBR nor PBR device.
Signed-off-by: Egbert Eich <eich@suse.com>
start_sect is not aligned to a 4 byte boundary thus causing exceptions
on ARM platforms. Access this field via the get_unaligned_le32 macro.
Signed-off-by: Marc Dietrich <marvin24@gmx.de>
This set of ifdefs is used in a number of places. Move its definition
somewhere common so it doesn't have to be repeated.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Acked-by: Simon Glass <sjg@chromium.org>
Acked-by: Tom Rini <trini@ti.com>
Signed-off-by: Tom Warren <twarren@nvidia.com>
The restoration of GPT table (both primary and secondary) is now possible.
Function 'gpt_restore' presents example of partition restoration process.
Signed-off-by: Lukasz Majewski <l.majewski@samsung.com>
Signed-off-by: Piotr Wilczek <p.wilczek@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Custom definitions of le_XX_to_int functions have been replaced with
standard ones, defined at <compiler.h>
Replacement of several GPT related structures members with ones
indicating its endianness and proper size.
Signed-off-by: Chang Hyun Park <heartinpiece@outlook.com>
Signed-off-by: Lukasz Majewski <l.majewski@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
This move is necessary to export gpt header and GPT partition entries to be
used with other commands or subsystems.
Additionally the part_efi.h file has been cleaned-up to supress checkpatch's
warnings.
Signed-off-by: Lukasz Majewski <l.majewski@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
In order to calculate the capability, we use the below expression to check:
((dev_desc->lba * dev_desc->blksz)>0L)
If the capability is greater than 4GB (e.g. 8GB = 8 * 1024 * 104 * 1024),
the result will overflow, the low 32bit may be zero.
Therefore, change to check each variable to fix this potential issue.
Signed-off-by: Jerry Huang <Chang-Ming.Huang@freescale.com>
This change addresses a few printf-formatting errors, and a typecast
error.
Signed-off-by: Taylor Hutt <thutt@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Currently, if the disk partition code is compiled with all of the parition
types compiled out, it hits an #error which stops the build. This change
adjusts that file so that those functions will fall through to their defaults
in those cases instead of breaking the build. These functions are needed
because other code calls them, and that code is needed because other config
options are overly broad and bring in support we don't need along with
support we do.
Also reduce repetition of the 6-term #ifdef throughout the file.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
When get_device_and_partition() finds a disk without a partition table,
under some conditions, it "returns" a disk_partition_t that describes
the entire raw disk. Make sure to initialize all fields in the partition
descriptor in that case.
The value chosen for name is just some arbitrary descriptive string.
The value chosen for info matches the check at the end of
get_device_and_partition(). However, it's probably not that important;
it's not obvious that the value is really used.
Reported-by: Benoît Thébaudeau <benoit.thebaudeau@advansee.com>
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Reviewed-by: Benoît Thébaudeau <benoit.thebaudeau@advansee.com>
Signed-off-by: Tom Rini <trini@ti.com>
This information may be useful to compare against command "part uuid",
or if you want to manually paste the information into the kernel
command-line.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
[trini: print_one_part / print_part_dos output strings didn't quite
match before the changes]
Signed-off-by: Tom Rini <trini@ti.com>
A partition is considered bootable if it either has the "legacy BIOS
bootable" flag set, or if the partition type UUID matches the standard
"system" type.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
When printing the EFI partition table, print the raw attributes. Convert
struct gpt_entry_attributes to a union to allow raw access.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Add no_block_io_protocol and legacy_bios_bootable attribute definitions.
These are sourced from UEFI Spec 2.3, page 105, table 19. Credits to the
libparted source for the specification pointer.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
When printing the partition table, print the partition type UUID and the
individual partition UUID. Do this unconditionally, since partition UUIDs
are useful.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
The partition name is a long variable-length string. Move it last on
the line to ensure consistent layout and that the entries align with
the "header" line. Also, surround it in quotes, so if it's empty, it's
obvious that something is still being printed.
Also, change the case of the LBA numbers; lower-case looks nicer in my
opinion, and will be more consistent with the UUID printing that is
added later in this series.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Simplify the partition printing loop in print_part_efi() to bail out
early when the first invalid partition is found, rather than indenting
the whole body of the loop. This simplifies later patches.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
When no valid partitions are found, guarantee that we return -1. This
most likely already happens, since the most recent get_partition_info()
will have returned an error. However, it's best to be explicit.
Remove an unnecessary assignment of ret=0 in the success case; this value
is over-written with the processed partition ID later.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Logically, a disk that contains a raw FAT filesystem does not in fact
have a partition table. However, test_part_dos() was claiming that such
disks did in fact have a DOS-style partition table. This caused
get_device_and_partition() not to return a whole-disk disk_partition_t,
since part_type != PART_TYPE_UNKNOWN.
part_dos.c's print_partition_extended() detected the raw FAT filesystem
condition and printed a fake partition table that encompassed the whole
disk.
However, part_dos.c's get_partition_info_extended() did not return any
valid partitions in this case. This combination caused
get_device_and_partition() not to find any valid partitions, and hence
to return an error.
Fix test_part_dos() not to claim that raw FAT filesystems are DOS
partition tables. In turn, this causes get_device_and_partition() to
return a whole-disk disk_partition_t, and hence the following commands
work:
fatls mmc 0 /
fatls mmc 0:auto /
An alternative would be to modify print_partition_extended() to detect
raw FAT filesystems, just like print_partition_extended() does, and to
return a fake partition in this case. However, this seems logically
incorrect, and also duplicates code, since get_device_and_partition()
falls back to returning a whole-disk partition when there is no partition
table on the device.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Commit 10a37fd "disk: get_device_and_partition() "auto" partition"
prevented the use of "-" on the command-line to request fallback to the
$bootdevice environment variable instead. This patch allows that, or an
empty string "" to be used.
Tested:
setenv bootfile /boot/zImage
setenv bootdevice 0:1
ext2load mmc 0:1
ext2load mmc -
ext2load mmc ""
Signed-off-by: Stephen Warren <swarren@nvidia.com>
The MSDOS/MBR partition table includes a 32-bit unique ID, often referred
to as the NT disk signature. When combined with a partition number within
the table, this can form a unique ID similar in concept to EFI/GPT's
partition UUID.
This patch generates UUIDs in the format 0002dd75-01, which matches the
format expected by the Linux kernel.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Each EFI partition table entry contains a UUID. Extend U-Boot's struct
disk_partition to be able to store this information, and modify
get_partition_info_efi() to fill it in.
The implementation of uuid_string() was derived from the Linux kernel,
tag v3.6-rc4 file lib/vsprintf.c function uuid_string().
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Enhance get_partition_info_efi() to range-check the partition number.
This prevents invalid partitions being accessed, and prevents access
beyond the end of the gpt_pte[] array.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Rework get_device_and_partition() to:
a) Implement a new partition ID of "auto", which requests that U-Boot
search for the first "bootable" partition, and fall back to the first
valid partition if none is found. This way, users don't need to
specify an explicit partition in their commands.
b) Make use of get_device().
c) Add parameter to indicate whether returning a whole device is
acceptable, or whether a partition is mandatory.
d) Make error-checking of the user's device-/partition-specification
more complete. In particular, if strtoul() doesn't convert all
characters, it's an error rather than just ignored.
The resultant device/partition returned by the function will be as
follows, based on whether the disk has a partition table (ptable) or not,
and whether the calling command allows the whole device to be returned
or not.
(D and P are integers, P >= 1)
D
D:
No ptable:
!allow_whole_dev: error
allow_whole_dev: device D
ptable:
device D partition 1
D:0
!allow_whole_dev: error
allow_whole_dev: device D
D:P
No ptable: error
ptable: device D partition P
D:auto
No ptable:
!allow_whole_dev: error
allow_whole_dev: device D
ptable:
first partition in device D with bootable flag set.
If none, first valid paratition in device D.
Note: In order to review this patch, it's probably easiest to simply
look at the file contents post-application, rather than reading the
patch itself.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
[swarren: Rob implemented scanning for bootable partitions. I fixed a
couple of issues there, switched the syntax to ":auto", added the
error-checking rework, and ":0" syntax for the whole device]
Signed-off-by: Stephen Warren <swarren@nvidia.com>
This patch introduces function get_device(). This looks up a
block_dev_desc_t from an interface name (e.g. mmc) and device number
(e.g. 0). This function is essentially the non-partition-specific
prefix of get_device_and_partition().
Signed-off-by: Stephen Warren <swarren@nvidia.com>
All block device related commands (scsiboot, fatload, ext2ls, etc.) have
simliar duplicated device and partition parsing and selection code. This
adds a common function to replace various implementations.
The new function has an enhancement over current versions. If no device
or partition is specified on the command line, the bootdevice env variable
will be used (scsiboot does this).
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Determine which partitions are bootable/active. In the partition listing,
print "Boot" for partitions with the bootable/active flag set.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Now that get_device_and_partition() always calls get_partition_info()
when disk.c is compiled, we must always compile the function, rather
than ifdef it away.
The implementation must be conditional based on CONFIG_CMD_* etc., since
that's what e.g. part_dos.c uses to ifdef out get_partition_info_dos();
CONFIG_DOS_PARTITION can be enabled even without those commands being
enabled.
Technically, this change is required before Rob's "disk/part: introduce
get_device_and_partition" patch. However, at least when the compiler
optimizer is turned on, it isn't required before then in practice,
since get_device_and_partition() calls get_dev(), which is stubbed out
in disk.c under exactly the same conditions that get_partition_info()
is not compiled, and hence the compiler never generates code for the
call to the missing function. However, in my later patch "disk:
get_device_and_partition() "auto" partition and cleanup", the optimizer
doesn't succeed at this, and may attempt to reference the undefined
function.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
This patch forces the correct alignment for DMA operations of buffers used by
part_mac.c.
Signed-off-by: Benoît Thébaudeau <benoit.thebaudeau@advansee.com>
Cc: Wolfgang Denk <wd@denx.de>
The patch below fixes device enumeration through the U-Boot API.
Device enumeration crashes when the system in question doesn't
have any RAM mapped to address zero (I discovered this on a
BeagleBone board), since the enumeration calls get_dev with a
NULL ifname sometimes which then gets passed down to strncmp().
This fix simply ensures that get_dev returns NULL when invoked
with a NULL ifname.
Signed-off-by: Tim Kientzle <kientzle@freebsd.org>
Signed-off-by: Anatolij Gustschin <agust@denx.de>
Fix errors noticed after enabling CONFIG_EFI_PARTITION
for the OMAP3 EVM board:
part_efi.c: In function 'print_part_efi':
part_efi.c:133:5: warning: passing argument 3 of 'is_gpt_valid'
from incompatible pointer type
part_efi.c:95:12: note: expected 'struct gpt_header *' but arg
ument is of type 'struct gpt_header **'
part_efi.c: In function 'get_partition_info_efi':
part_efi.c:173:4: warning: passing argument 3 of 'is_gpt_valid
' from incompatible pointer type
part_efi.c:95:12: note: expected 'struct gpt_header *' but arg
ument is of type 'struct gpt_header **'
part_efi.c: In function 'alloc_read_gpt_entries':
part_efi.c:384:18: error: 'CONFIG_SYS_CACHELINE_SIZE' undeclare
d (first use in this function)
Signed-off-by: Sanjeev Premi <premi@ti.com>
Cc: Tom Rini <tom.rini@gmail.com>
Cc: Anton staaf <robotboy@chromium.org>
Signed-off-by: Sandeep Paulraj <s-paulraj@ti.com>
Commit deb5ca8027 "disk: part_efi: fix
**pgpt_pte == NULL" modified the code to pass "&gpt_head" to
is_gpt_valid() rather than the previous "gpt_head". However, gpt_head
is a pointer to the buffer, not the actual buffer, since it was allocated
using ALLOC_CACHE_ALIGN_BUFFER. This caused is_gpt_valid() to read the
disk block onto the stack rather than into the buffer, causing the
code to fail.
This change reverts that portion of the commit mentioned above.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Acked-by: Doug Anderson <dianders@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
Acked-by: Mike Frysinger <vapier@gentoo.org>
Tested-by: Anton Staaf <robotboy@chromium.org>
Code was setting **pgpt_pte == NULL, which meant that the pointer
to the gpt_pte would be stored at RAM address 00000000. This 'worked'
on T20 (SDRAM starts @ 0x00000000), but hung gpt/EFI access on T30
(SDRAM starts @ 0x80000000).
Signed-off-by: Tom Warren <twarren@nvidia.com>
Signed-off-by: Doug Anderson <dianders@chromium.org>
Currently part_efi.c allocates buffers for the gpt_header, the
legacy_mbr, and the pte (partition table entry) that may be
incorrectly aligned for DMA operations.
This patch uses ALLOC_CACHE_ALIGN_BUFFER for the stack allocated
buffers and memalign to replace the malloc of the pte.
Signed-off-by: Anton Staaf <robotboy@chromium.org>
Cc: Lukasz Majewski <l.majewski@samsung.com>
Cc: Mike Frysinger <vapier@gentoo.org>
Cc: Albert ARIBAUD <albert.u.boot@aribaud.net>
Acked-by: Mike Frysinger <vapier@gentoo.org>
Apple iPod nanos have sector sizes of 2 or 4 KiB, which crashes U-Boot when it
tries to read the MBR into 512-byte buffer situated on stack. Instead use the
variable length arrays to be safe with any large sector size.
Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
If we don't want to build support for any partition types we can now
add #undef CONFIG_PARTITIONS in a board config file to keep this from
being compiled in. Otherwise boards assume this is compiled in by
default
Signed-off-by: Matthew McClintock <msm@freescale.com>
If the param pass to get_dev is not the one defined in the block_drvr,
it could make uboot becomes unstable, for it would continue run after
search complete the block_drvr table.
Signed-off-by: Lei Wen <leiwen@marvell.com>
Before this commit, weak symbols were not overridden by non-weak symbols
found in archive libraries when linking with recent versions of
binutils. As stated in the System V ABI, "the link editor does not
extract archive members to resolve undefined weak symbols".
This commit changes all Makefiles to use partial linking (ld -r) instead
of creating library archives, which forces all symbols to participate in
linking, allowing non-weak symbols to override weak symbols as intended.
This approach is also used by Linux, from which the gmake function
cmd_link_o_target (defined in config.mk and used in all Makefiles) is
inspired.
The name of each former library archive is preserved except for
extensions which change from ".a" to ".o". This commit updates
references accordingly where needed, in particular in some linker
scripts.
This commit reveals board configurations that exclude some features but
include source files that depend these disabled features in the build,
resulting in undefined symbols. Known such cases include:
- disabling CMD_NET but not CMD_NFS;
- enabling CONFIG_OF_LIBFDT but not CONFIG_QE.
Signed-off-by: Sebastien Carlier <sebastien.carlier@gmail.com>
By now, the majority of architectures have working relocation
support, so the few remaining architectures have become exceptions.
To make this more obvious, we make working relocation now the default
case, and flag the remaining cases with CONFIG_NEEDS_MANUAL_RELOC.
Signed-off-by: Wolfgang Denk <wd@denx.de>
Tested-by: Heiko Schocher <hs@denx.de>
Tested-by: Reinhard Meyer <u-boot@emk-elektronik.de>
