Skip to contentSkip to conversation

gfs2_stuffed_write_end modifying source buffer?

PATCH In review Bob Peterson 2 participants 2 messages Sep 23, 2009 – Oct 1, 2009 1 message not in the archive GFS2 file system

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): 652851863.523171253742277925.JavaMail.root@zmail06.collab.prod.int.phx2.redhat.com

#1 lore raw

PATCH 1/1 gfs2_stuffed_write_end modifying source buffer?

Hi,

Maybe I'm wrong, but this looks like a bug to me: It looks like GFS2's function gfs2_stuffed_write_end is zeroing out portions of the source buffer. So if I create a character array and filled it with "X" then wrote only one byte to a very small file, all the other X's in my buffer would get nuked. Just a theory at this point but perhaps Steve Whitehouse can tell.

Regards,

Bob Peterson
Red Hat File Systems
--

 fs/gfs2/aops.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)
fs/gfs2/aops.c +0 -1
index 7ebae9a..6a23ba2 100644
@@ -801,7 +801,6 @@ static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
 	BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
 	kaddr = kmap_atomic(page, KM_USER0);
 	memcpy(buf + pos, kaddr + pos, copied);
-	memset(kaddr + pos + copied, 0, len - copied);
 	flush_dcache_page(page);
 	kunmap_atomic(kaddr, KM_USER0);
 
#2 lore raw

Hi,

I'm not convinced there is a bug here...

Quoted 22 lines · Bob Peterson

Hi,

Maybe I'm wrong, but this looks like a bug to me: It looks like GFS2's function gfs2_stuffed_write_end is zeroing out portions of the source buffer. So if I create a character array and filled it with "X" then wrote only one byte to a very small file, all the other X's in my buffer would get nuked. Just a theory at this point but perhaps Steve Whitehouse can tell.

Regards,

Bob Peterson
Red Hat File Systems
--

 fs/gfs2/aops.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)
fs/gfs2/aops.c +0 -0
index 7ebae9a..6a23ba2 100644

fs/gfs2/aops.c static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh, · line 804

Quoted 4 lines
fs/gfs2/aops.c +0 -1
@@ -801,7 +801,6 @@ static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
 	BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
 	kaddr = kmap_atomic(page, KM_USER0);
 	memcpy(buf + pos, kaddr + pos, copied);
-	memset(kaddr + pos + copied, 0, len - copied);

pos is the current file offset (which since its stuffed is also a byte offset into this (first) page of the file)
copied is the number of bytes that were copied
len is the number of bytes that we wanted to copy (so for a successful copy, its equal to copied and thus the memset() is a no-op

fs/gfs2/aops.c static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh, · line 806

Quoted 3 lines
fs/gfs2/aops.c +0 -0
 	flush_dcache_page(page);
 	kunmap_atomic(kaddr, KM_USER0);
 

The purpose of that memset() is to deal with the situation where the copy has failed part way though (source page inaccessible) and thus we couldn't complete the operation. If the write was not an extending write, we cannot truncate off the bit which wasn't copied, so it gets zeroed out instead.

Steve.