Bug#682007: NULL pointer dereference in __fscache_read_or_alloc_pages

[Brian Kroth]

Jonathan Nieder <jrnie...@gmail.com> 2012-10-01 01:25:

Brian Kroth wrote:

Sorry, the labs went into their dormant period and all of my test cases ran
away for the rest of the summer (the find cmd didn't seem to trigger the
__fscache problem), so I hadn't moved any further on this.

Now that they're back, I'm definitely seeing it again (about 20 different
machines in two days last week), so I've started the process of hunting down
a trigger cause again.  I'll let you know if I find something.

Thanks for the update.

The human test cases can work fine for vetting a fix.  I'd also be
interested to hear whether the series I sent was completely borked, so
I'd recommend trying on a test machine for a day or two before putting
such a patched kernel into production, though.

Jonathan

Once again very sorry for the delay :(



Here's what I got when applying the patches.

# git am -3sc $(ls -1 ../patches/fscache_read/[01]*.patch)
Applying: CacheFiles: Fix the marking of cached pages
Applying: CacheFiles: Downgrade the requirements passed to the allocator
Applying: FS-Cache: Check that there are no read ops when cookie relinquished
Applying: CacheFiles: Make some debugging statements conditional
Applying: FS-Cache: Make cookie relinquishment wait for outstanding reads
Applying: FS-Cache: Fix operation state management and accounting
Applying: FS-Cache: Provide proper invalidation
Applying: VFS: Make more complete truncate operation available to CacheFiles
Applying: CacheFiles: Implement invalidation
Applying: NFS: Use FS-Cache invalidation
Using index info to reconstruct a base tree...
Falling back to patching base and 3-way merge...
Auto-merging fs/nfs/fscache.h
Auto-merging fs/nfs/inode.c
Auto-merging fs/nfs/nfs4proc.c
Applying: CacheFiles: Add missing retrieval completions
Applying: FS-Cache: Convert the object event ID #defines into an enum
Applying: FS-Cache: Initialise the object event mask with the calculated mask
Applying: FS-Cache: Don't mask off the object event mask when printing it
Applying: FS-Cache: Limit the number of I/O error reports for a cache
Applying: FS-Cache: Exclusive op submission can BUG if there's been an I/O error
Applying: NFS: nfs_migrate_page() does not wait for FS-Cache to finish with a 
page



Thanks,
Brian

I forgot to disable the DEBUG_INFO and kept filling up my build VMs disk during compile. Then realized I had grabbed the 3.7 rc code, which these patches don't apply against. "git checkout remotes/stable/linux-3.2.y" (results in head c74a5e1fe4d0672936c8fb63d7484dfeaa30669c and 3.2.28), seemed to fix that. Anyways, I just started running that on a machine, so I'll let you know if I noticed anything there first before I think about pushing it to further places.

signature.asc
Description: Digital signature