hfsplus corrupts filesystems >2TB
A Debian bug report <http://bugs.debian.org/550010> alerted me to the fact that hfsplus is not using enough bits for some sector calculations.
hfsplus_get_block() does:
u32 ablock, dblock, mask; ... map_bh(bh_result, sb, (dblock << HFSPLUS_SB(sb).fs_shift) + HFSPLUS_SB(sb).blockoffset + (iblock & mask));
which results in overflow when the sector number is >2^32. Now it might be sufficient to change the last line to:
map_bh(bh_result, sb, ((sector_t)dblock << HFSPLUS_SB(sb).fs_shift) + HFSPLUS_SB(sb).blockoffset + (iblock & mask));
but there may be many other places where u32 must be changed to sector_t.
For Debian's stable release, I'm intending to prevent mounting volumes larger than 2^32 sectors (2TB). Is anyone interested in fixing this properly or should I submit the same change for mainline?
Ben.
Signature · 2 lines
Ben Hutchings To err is human; to really foul things up requires a computer.
PATCH 1/1 hfsplus: Refuse to mount volumes larger than 2TB
As found in <http://bugs.debian.org/550010>, hfsplus is using type u32 rather than sector_t for some sector number calculations.
In particular, hfsplus_get_block() does:
u32 ablock, dblock, mask;
...
map_bh(bh_result, sb, (dblock << HFSPLUS_SB(sb).fs_shift) + HFSPLUS_SB(sb).blockoffset + (iblock & mask));I am not confident that I can find and fix all cases where a sector number may be truncated. For now, avoid data loss by refusing to mount HFS+ volumes with more than 2^32 sectors (2TB).
---
fs/hfsplus/wrapper.c +4 -0
@@ -99,6 +99,10 @@if (hfsplus_get_last_session (sb, &part_start, &part_size))return -EINVAL ; +if (part_start + part_size >0x100000000 ) { +pr_err ("hfs: volumes larger than 2TB are not supported yet\n" ); +return -EINVAL ; + }while (1 ) { bh =sb_bread512 (sb, part_start +HFSPLUS_VOLHEAD_SECTOR , vhdr);if (!bh)
Signature · 2 lines
Ben Hutchings Quantity is no substitute for quality, but it's the only one we've got.
Re: [PATCH] hfsplus: Refuse to mount volumes larger than 2TB
Quoted 19 lines · On Sun, 11 Oct 2009 03:11:27 +0100 Ben Hutchings <[email protected]> wrote:
As found in <http://bugs.debian.org/550010>, hfsplus is using type u32 rather than
sector_tfor some sector number calculations.In particular,
hfsplus_get_block()does:u32 ablock, dblock, mask; ... map_bh(bh_result, sb, (dblock << HFSPLUS_SB(sb).fs_shift) + HFSPLUS_SB(sb).blockoffset + (iblock & mask));I am not confident that I can find and fix all cases where a sector number may be truncated. For now, avoid data loss by refusing to mount HFS+ volumes with more than 2^32 sectors (2TB).
Signed-off-by Ben Hutchings <[email protected]>---
fs/hfsplus/wrapper.c line 105
Quoted 7 lines
fs/hfsplus/wrapper.c +4
-0@@ -99,6 +99,10 @@if (hfsplus_get_last_session (sb, &part_start, &part_size))return -EINVAL ; +if (part_start + part_size >0x100000000 ) { +pr_err ("hfs: volumes larger than 2TB are not supported yet\n" ); +return -EINVAL ; + }
part_start and part_size are sector_t. This code will do weird overflow things when sector_t is 32-bit. Also 32-bit compilers will get upset at the excessively large hex constant.
This should fix both issues:
fs/hfsplus/wrapper.c renamed from fs/hfsplus/wrapper.c~hfsplus-refuse-to-mount-volumes-larger-than-2tb-fix +1 -1
@@ -99,7 +99,7 @@ int hfsplus_read_wrapper(struct super_blif (hfsplus_get_last_session (sb, &part_start, &part_size))return -EINVAL ; -if (part_start + part_size >0x100000000 ) { +if ((u64 )part_start + part_size >0x100000000ULL ) {pr_err ("hfs: volumes larger than 2TB are not supported yet\n" );return -EINVAL ; }
_
Re: [PATCH] hfsplus: Refuse to mount volumes larger than 2TB
On Sun, 2009-10-11 at 00:51 -0700, Andrew Morton wrote:
[...]
Quoted 4 lines
Quoted 4 lines
fs/hfsplus/wrapper.c +0
-0@@ -99,6 +99,10 @@
Quoted 9 lines
fs/hfsplus/wrapper.c line 105
Quoted 6 lines
fs/hfsplus/wrapper.c +4
-0if (hfsplus_get_last_session (sb, &part_start, &part_size))return -EINVAL ; +if (part_start + part_size >0x100000000 ) { +pr_err ("hfs: volumes larger than 2TB are not supported yet\n" ); +return -EINVAL ; + }
part_startandpart_sizearesector_t. This code will do weird overflow things whensector_tis 32-bit.
Sorry, I forgot CONFIG_LBD is still optional.
Quoted 1 line
Also 32-bit compilers will get upset at the excessively large hex constant.
[...]
Good point.
Ben.
Signature · 3 lines
Ben Hutchings DNRC Motto: I can please only one person per day. Today is not your day. Tomorrow isn't looking good either.