| Message ID | 20170624003719.GA5670@debian.home |
|---|---|
| State | Not Applicable, archived |
| Headers | show |
On Fri 23-06-17 21:37:21, Ernesto A. Fernández wrote: > Buffer heads referencing indirect blocks may not be released if the file > is truncated at the right time. This happens because ext2_get_branch() > returns NULL when it finds the whole chain of indirect blocks already > set, and when truncate alters the chain this value of NULL is > treated as the address of the last head to be released. Handle this in the > same way as it's done after the got_it label. > > Signed-off-by: Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com> > --- > Despite trying I haven't actually managed to trigger this race. I tested > the patch by inducing regular failures in verify_chain(). Thanks! I've merged the patch to my tree. Honza > > fs/ext2/inode.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/fs/ext2/inode.c b/fs/ext2/inode.c > index 2dcbd56..30163d0 100644 > --- a/fs/ext2/inode.c > +++ b/fs/ext2/inode.c > @@ -659,6 +659,7 @@ static int ext2_get_blocks(struct inode *inode, > */ > err = -EAGAIN; > count = 0; > + partial = chain + depth - 1; > break; > } > blk = le32_to_cpu(*(chain[depth-1].p + count)); > -- > 2.1.4 > >
diff --git a/fs/ext2/inode.c b/fs/ext2/inode.c index 2dcbd56..30163d0 100644 --- a/fs/ext2/inode.c +++ b/fs/ext2/inode.c @@ -659,6 +659,7 @@ static int ext2_get_blocks(struct inode *inode, */ err = -EAGAIN; count = 0; + partial = chain + depth - 1; break; } blk = le32_to_cpu(*(chain[depth-1].p + count));
Buffer heads referencing indirect blocks may not be released if the file is truncated at the right time. This happens because ext2_get_branch() returns NULL when it finds the whole chain of indirect blocks already set, and when truncate alters the chain this value of NULL is treated as the address of the last head to be released. Handle this in the same way as it's done after the got_it label. Signed-off-by: Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com> --- Despite trying I haven't actually managed to trigger this race. I tested the patch by inducing regular failures in verify_chain(). fs/ext2/inode.c | 1 + 1 file changed, 1 insertion(+)