On Thu, Mar 25, 2010 at 4:55 AM, <tytso@xxxxxxx> wrote: > On Wed, Mar 24, 2010 at 10:27:24PM +0300, Evgeniy Ivanov wrote: >> >> Sorry for bothering list with my ext2 questions. >> I got into trouble with my ext2 implementation and filesystem with >> 1024 block size. Sometimes when I write files they're written >> correctly (md5sum is the same as original, i_size is correct either), >> but e2fsck changes i_size to another values (which breaks files). E.g. >> 67445000->67446784 or 67445248->67446784. I see that new sizes are >> numbers of multiples of 1024. >> Strange thing is that I can't reproduce this problem with 2048 and >> 4096 block sizes. I thought the problem was in trash in unused part of >> last block (actually it is zeroed), but then it would be reproduceable >> in fs with another block size. > > E2fsck will adjust i_size if it is smaller than the number of blocks > than you have allocated. So in the case of 67445000->67446784, your > file probably had 65866 1k blocks, and since 67445000 is less than > (655865*1024)+1, e2fsck assumed that your i_size was wrong, and so it > asked for permission to fix it. > > Put another way, if you have 2 blocks in 1k file, and i_size is 1024, > it clearly must be wrong. If it's 1025, maybe we're only using 1 byte > in the last block; but if i_size is less than or equal to 1024, then > why was the 2nd block allocated in the file in the first place? Thank you for your explanation. My problem was in miscalculation of first triple indirect block. I used following thing "triple_ind_s = doub_ind_s + pow(addr_in_block, 2)" and it was a bad idea to use pow() instead of multiplication or shifting. It was ok with gcc (and libc), but caused a problem with ACK (I get value of 1 less, thus each last double indirect block became a hole instead of data). Since that was both in reading and writing md5 sums were correct (and in Linux I checked them only after e2fsck). Funny bug :) -- Evgeniy Ivanov -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
