Re: Re: [RFC] block_dev:Fix bug when read/write block-device which is larger than 16TB in 32bit-OS.

[majianpeng <majianpeng@gmail.com>]

On 2012-07-24 21:48 Christoph Hellwig <hch@infradead.org> Wrote:
>On Tue, Jul 24, 2012 at 08:44:27PM +0800, majianpeng wrote:
>> On 2012-05-29 16:56 majianpeng <majianpeng@gmail.com> Wrote:
>> >The size of block-device is larger than 16TB, and the os is 32bit.
>> >If the offset of read/write is larger then 16TB. The index of address_space will
>> >overflow and supply data from low offset instead.
>
>We can't support > 16TB block device on 32-bit systems with 4k page
>size, just like we can't support files that large.
>
>For filesystems the s_maxbytes limit of MAX_LFS_FILESIZE takes care of
>that, but it seems like we miss that check for block devices.
>
>The proper fix is to add that check (either via s_maxbytes or by
>checking MAX_LFS_FILESIZE) to generic_write_checks and
>generic_file_aio_read (or a block device specific wrapper)
>
/* Page cache limit. The filesystems should put that into their s_maxbytes 
   limits, otherwise bad things can happen in VM. */ 
#if BITS_PER_LONG==32
#define MAX_LFS_FILESIZE	(((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1) 
#elif BITS_PER_LONG==64
#define MAX_LFS_FILESIZE 	0x7fffffffffffffffUL
#endif

If we used MAX_LFS_FILESIZE to limit the block-device, so in 32bit-os, the size of block is
only 8T -1.
But in function do_generic_file_read():
>>index = *ppos >> PAGE_CACHE_SHIFT;
index is unsigned long type. So the ppos can 16T -1.

But the comment said:
>>/* Page cache limit. The filesystems should put that into their s_maxbytes 
>>   limits, otherwise bad things can happen in VM. */ 
Why ?

Thanks !