On 11/11/12 7:54 PM, Andreas Dilger wrote:

On 2012-11-11, at 3:27 PM, George Spelvin wrote:

quoted

I'm using v1.43-WIP-2012-09-22-10-g41bf599, last commit Oct. 14.

I'm trying to create a file system with 64bit support and specify a
maximum resize limit of 64 TiB = 2^34 blocks = 17179869184.

(gdb) run -n -t ext4 -O 64bit -E resize=4294967296 /dev/md1
Starting program: /root/e2fsprogs/misc/mke2fs -n -t ext4 -O 64bit -E resize=4294967295 /dev/md1
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
mke2fs 1.43-WIP (22-Sep-2012)

Program received signal SIGFPE, Arithmetic exception.
0x0000000000405f5a in parse_extended_opts (opts=<optimized out>, 
  param=0x64e200) at mke2fs.c:800
800                             gdpb = EXT2_DESC_PER_BLOCK(param);

This is definitely a bug in the code to do a divide-by-zero.

However, it should be pointed out that the "resize" option does not
make sense for filesystems larger than 16TB.  The mechanism used for
resizing beyond 16TB is different and does not need to reserve blocks.

In fairness to the reporter, nothing in the existing ext4 documentation,
AFAICT, mentions this.  (but then -O 64bit isn't really documented at all)

And given that the poor reporter is re-making his whole filesystem just
because he found out that he can't grow past 16T:
	"(wow, was *that* a nasty surprise)"
it's understandable that he's trying to give it a rather large resize=
value this time around.

This is one of those dark corners of weird behavior that could really use
some formal docs, at least.  :(

-Eric

Cheers, Andreas.

quoted

The issue is that 

#define EXT2_DESC_PER_BLOCK(s)          (EXT2_BLOCK_SIZE(s) / EXT2_DESC_SIZE(s))
#define EXT2_DESC_SIZE(s)                                                \
     ((EXT2_SB(s)->s_feature_incompat & EXT4_FEATURE_INCOMPAT_64BIT) ? \
      (s)->s_desc_size : EXT2_MIN_DESC_SIZE)

and s_desc_size is 0 because parse_extended_opts is called from PRS which
is called very early in main() at line 2320, while s_desc_size is set up
in ext2fs_initialize, which is not called from main() until mke2fs.c:2353.

As a temporary workaround, I notice that ext2fs_initialize sets s_desc_size to
the fixed value EXT2_MIN_DESC_SIZE_64BIT, so I changed the #define as follows:

#define EXT2_DESC_SIZE(s)                                                \
     ((EXT2_SB(s)->s_feature_incompat & EXT4_FEATURE_INCOMPAT_64BIT) ? \
      (s)->s_desc_size ?: EXT2_MIN_DESC_SIZE_64BIT : EXT2_MIN_DESC_SIZE)

... which seems to work.


(One point that occurred to me while wrestling with this is that the
default resize limit of initial size * 1000 should perhaps be clamped
to 2^32 if 64bit is not enabled.)
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Cheers, Andreas





--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html