Re: Oopsing cryptoapi (or loop device?) on 2.6.*

[Jari Ruusu]

Jari Ruusu wrote:
> Jan Rychter wrote:
> > FWIW, I've just tried loop-AES with 2.4.24, after using cryptoapi for a
> > number of years. My machine froze dead in the midst of copying 2.8GB of
> > data onto my file-backed reiserfs encrypted loopback mount.
> >
> > Since the system didn't ever freeze on me before and since I've had zero
> > problems with cryptoapi, I attribute the freeze to loop-AES.
> >
> > Yes, I know this isn't a good bugreport...
> 
> Is there any particular reason why you insist on using file backed loops?
> 
> File backed loops have hard to fix re-entry problem: GFP_NOFS memory
> allocations that cause dirty pages to written out to file backed loop, will
> have to re-enter the file system anyway to complete the write. This causes
> deadlocks. Same deadlocks are there in mainline loop+cryptoloop combo.
> 
> This is one of the reasons why this is in loop-AES README: "If you can
> choose between device backed and file backed, choose device backed even if
> it means that you have to re-partition your disks."

A quick grep of potential deadlock points:

armas:/usr/src/linux-2.4.24/fs/reiserfs # grep -r GFP_NOFS *
dir.c:              local_buf = reiserfs_kmalloc(d_reclen, GFP_NOFS, inode->i_sb) ;
fix_node.c:         buf = reiserfs_kmalloc(size, GFP_NOFS, tb->tb_sb);
journal.c:  bn = reiserfs_kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS, p_s_sb) ;
journal.c:  bn->data = reiserfs_kmalloc(p_s_sb->s_blocksize, GFP_NOFS, p_s_sb) ;
journal.c:  log_blocks = reiserfs_kmalloc(le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ;
journal.c:  real_blocks = reiserfs_kmalloc(le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ;
journal.c:  /* using GFP_NOFS, GFP_KERNEL could try to flush inodes, which will try
journal.c:  ct = reiserfs_kmalloc(sizeof(struct reiserfs_journal_commit_task), GFP_NOFS, p_s_sb) ;
namei.c:        buffer = reiserfs_kmalloc (buflen, GFP_NOFS, dir->i_sb);
namei.c:    name = reiserfs_kmalloc (item_len, GFP_NOFS, parent_dir->i_sb);
armas:/usr/src/linux-2.4.24/fs/reiserfs #

Many potential deadlock points there in reiserfs.

armas:/usr/src/linux-2.4.24/fs/ext2 # grep -r GFP_NOFS *
armas:/usr/src/linux-2.4.24/fs/ext2 #

File backed loop to a file on ext2 may work.

armas:/usr/src/linux-2.4.24/fs/ext3 # grep -r GFP_NOFS *
inode.c:        page = find_or_create_page(mapping, index, GFP_NOFS);
armas:/usr/src/linux-2.4.24/fs/ext3 #

Potential deadlock point there in ext3.

armas:/usr/src/linux-2.4.24/fs/nfs # grep -r GFP_NOFS *
armas:/usr/src/linux-2.4.24/fs/nfs # 

File backed loop to a file on nfs may work.

armas:/usr/src/linux-2.4.24/fs/smbfs # grep -r GFP_NOFS *
armas:/usr/src/linux-2.4.24/fs/smbfs #

File backed loop to a file on smbfs may work.

-- 
Jari Ruusu  1024R/3A220F51 5B 4B F9 BB D3 3F 52 E9  DB 1D EB E3 24 0E A9 DD
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/