Re: [reiserfs-list] ps strangeness

[Nikita Danilov]

Chris Wedgwood writes:
 > On Mon, Jul 30, 2001 at 07:20:44PM -0700, Florin Andrei wrote:
 > 
 >     If i run "ps ax", sometimes i see this: once i get the correct
 >     list of processes, once i get an empty list.
 > 
 > It sounds like /proc changing underneath the ps application, it
 > happens with other utilities from time to time too (eg. netstat).

That is what going on:

 1. when deleting file reiserfs doesn't care to set i_blocks to 0 and,
    moreover for symlinks it happens to be set to -7 (into unsigned
    field!). It doesn't hurt, because file is being deleted anyway.
 2. in-memory inode is immediately reused for /proc inode.
 3. procfs_read_inode() doesn't bother to clear i_blocks either (neither
    does VFS get_new_inode(), why, Alexander?)
 4. now we have /proc file-system inode with huge i_blocks field.
 5. top(1) calls stat() on this inode. Libc wrapper function decides
    that i_blocks doesn't fit into st_blocks and returns EOVERFLOW.
 6. top shows nothing.

Solutions:

 1. update get_new_inode() to clear i_blocks
 2. update procfs_read_inode() to clear i_blocks
 3. update reiserfs_delete_inode() to clear i_blocks

Patch implementing last option (by Vladimir Saveliev):
--------------------------------------------------
--- linux-2.4.7/fs/reiserfs/inode.c	Sat Jul 21 21:05:18 2001
+++ linux-2.4.7.patched/fs/reiserfs/inode.c	Tue Jul 31 18:55:53 2001
@@ -55,6 +55,7 @@ void reiserfs_delete_inode (struct inode
 	;
     }
     clear_inode (inode); /* note this must go after the journal_end to prevent deadlock */
+    inode->i_blocks = 0;
     unlock_kernel() ;
 }
--------------------------------------------------

 > 
 > There isn't much that can be done about it, other than the
 > applications needs to be a little smarted so they can detect /proc
 > changes part-way-through gathering data.
 > 
 > 
 > 
 >   --cw

Nikita.