summaryrefslogtreecommitdiff
path: root/util/cbfstool/flashmap/valstr.h
diff options
context:
space:
mode:
authorSol Boucher <solb@chromium.org>2015-04-03 09:13:04 -0700
committerPatrick Georgi <pgeorgi@google.com>2015-05-08 19:45:21 +0200
commit636cc858bb6979cbd51abaa1d75842146b0c29f7 (patch)
tree6d92118cfe4d23006056aa7a50bd6e8074de72e9 /util/cbfstool/flashmap/valstr.h
parentb11be321a895aa70adeaa8cb92fcfcd5dcbd748c (diff)
downloadcoreboot-636cc858bb6979cbd51abaa1d75842146b0c29f7.tar.xz
cbfstool: Make the add action choose an aligned entries capacity
This fixes an inconsistency between `cbfstool create` and `cbfstool add` that was resulting in confusing claims about the amount of free space at the end of a CBFS. Calls to `cbfstool add` check whether a file fits under a given empty file entry by testing whether it would collide with the beginning of the *subsequent* file header; thus, if a file's end is unaligned, its reported size will not match the actual available capacity. Although deleted entries always end on an alignment boundary because `cbfstool remove` expands them to fill the available space, `cbfstool create` doesn't necessarily size a new entries region to result in an empty entry with an aligned end. This problem never resulted in clobbering important data because cbfstool would blindly reserve 64B (or the selected alignment) of free space immediately after the all-inclusive empty file entry. This change alters the way this reservation is reported: only the overhang past the alignment is used as hidden padding, and the empty entry's capacity is always reported such that it ends at an aligned address. Much of the time that went into this patch was spent building trust in the trickery cbfstool employs to avoid explicitly tracking the image's total capacity for entries, so below are two proofs of correctness to save others time and discourage inadvertent breakage: OBSERVATION (A): A check in cbfs_image_create() guarantees that an aligned CBFS empty file header is small enough that it won't cross another aligned address. OBSERVATION (B): In cbfs_image_create(), the initial empty entry is sized such that its contents end on an aligned address. THM. 1: Placing a new file within an empty entry located below an existing file entry will never leave an aligned flash address containing neither the beginning of a file header nor part of a file. We can prove this by contradiction: assume a newly-added file neither fills to the end of the preexisting empty entry nor leaves room for another aligned empty header after it. Then the first aligned address after the end of the newly-inserted file... - CASE 1: ...already contains a preexisting file entry header. + Then that address contains a file header. - CASE 2: ...does not already house a file entry header. + Then because CBFS content doesn't fall outside headers, the area between there and the *next* aligned address after that is unused. + By (A), we can fit a file header without clobbering anything. + Then that address now contains a file header. THM. 2: Placing a new file in an empty entry at the very end of the image such that it fits, but leaves no room for a final header, is guaranteed not to change the total amount of space for entries, even if that new file is later removed from the CBFS. Again, we use contradiction: assume that creating such a file causes a permanent... - CASE 1: ...increase in the amount of available space. + Then the combination of the inserted file, its header, and any padding must have exceeded the empty entry in size enough for it to cross at least one additional aligned address, since aligned addresses are how the limit on an entry's capacity is determined. + But adding the file couldn't have caused us to write past any further aligned addresses because they are the boundary's used when verifying that sufficient capacity exists; furthermore, by (B), no entry can ever terminate beyond where the initial empty entry did when the CBFS was first created. + Then the creation of the file did not result in a space increase. - CASE 2: ...decrease in the amount of available space. + Then the end of the new file entry crosses at least one fewer aligned address than did the empty file entry. + Then by (A), there is room to place a new file entry that describes the remaining available space at the first available aligned address. + Then there is now a new record showing the same amount of available space. + Then the creation of the file did not result in a space decrease. BUG=chromium:473726 TEST=Had the following conversation with cbfstool: $ ./cbfstool test.image create -s 0x100000 -m arm Created CBFS image (capacity = 1048408 bytes) $ ./cbfstool test.image print test.image: 1024 kB, bootblocksize 0, romsize 1048576, offset 0x40 alignment: 64 bytes, architecture: arm Name Offset Type Size (empty) 0x40 null 1048408 $ dd if=/dev/zero of=toobigmed.bin bs=1048409 count=1 1+0 records in 1+0 records out 1048409 bytes (1.0 MB) copied, 0.0057865 s, 181 MB/s $ ./cbfstool test.image add -t 0x50 -f toobigmed.bin -n toobig E: Could not add [toobigmed.bin, 1048409 bytes (1023 KB)@0x0]; too big? E: Failed to add 'toobigmed.bin' into ROM image. $ truncate -s -1 toobigmed.bin $ ./cbfstool test.image add -t 0x50 -f toobigmed.bin -n toobig $ ./cbfstool test.image print test.image: 1024 kB, bootblocksize 0, romsize 1048576, offset 0x40 alignment: 64 bytes, architecture: arm Name Offset Type Size toobig 0x40 raw 1048408 $ ./cbfstool test.image remove -n toobig $ ./cbfstool test.image print test.image: 1024 kB, bootblocksize 0, romsize 1048576, offset 0x40 alignment: 64 bytes, architecture: arm Name Offset Type Size (empty) 0x40 deleted 1048408 $ ./cbfstool test.image print test.image: 1024 kB, bootblocksize 0, romsize 1048576, offset 0x40 alignment: 64 bytes, architecture: arm Name Offset Type Size (empty) 0x40 deleted 1048408 BRANCH=None Change-Id: I118743e37469ef0226970decc900db5d9b92c5df Signed-off-by: Sol Boucher <solb@chromium.org> Original-Commit-Id: e317ddca14bc36bc36e6406b758378c88e9ae04e Original-Change-Id: I294ee489b4918646c359b06aa1581918f2d8badc Original-Signed-off-by: Sol Boucher <solb@chromium.org> Original-Reviewed-on: https://chromium-review.googlesource.com/263962 Original-Reviewed-by: Hung-Te Lin <hungte@chromium.org> Original-Reviewed-by: Stefan Reinauer <reinauer@google.com> Reviewed-on: http://review.coreboot.org/9939 Tested-by: build bot (Jenkins)
Diffstat (limited to 'util/cbfstool/flashmap/valstr.h')
0 files changed, 0 insertions, 0 deletions