Two variables in ext4_inode_info, i_reserved_meta_blocks and
i_allocated_meta_blocks, are unused. Removing them saves a little
memory per in-memory inode and cleans up clutter in several tracepoints.
Adjust tracepoint output from ext4_alloc_da_blocks() for consistency
and fix a typo and whitespace near these changes.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
Support the GETFSMAP ioctls so that we can use the xfs free space
management tools to probe ext4 as well. Note that this is a partial
implementation -- we only report fixed-location metadata and free space;
everything else is reported as "unknown".
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
DAX page fault path needs to get blocks that are pre-zeroed to avoid
races when two concurrent page faults happen in the same block of a
file. Implement support for this in ext4_map_blocks().
Signed-off-by: Jan Kara <jack@suse.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When dioread_nolock mode is enabled, we grab i_data_sem in
ext4_ext_direct_IO() and therefore we need to instruct _ext4_get_block()
not to grab i_data_sem again using EXT4_GET_BLOCKS_NO_LOCK. However
holding i_data_sem over overwrite direct IO isn't needed these days. We
have exclusion against truncate / hole punching because we increase
i_dio_count under i_mutex in ext4_ext_direct_IO() so once
ext4_file_write_iter() verifies blocks are allocated & written, they are
guaranteed to stay so during the whole direct IO even after we drop
i_mutex.
So we can just remove this locking abuse and the no longer necessary
EXT4_GET_BLOCKS_NO_LOCK flag.
Signed-off-by: Jan Kara <jack@suse.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Remove outdated comments and dead code from ext4_da_reserve_space.
Clean up its trace point, and relocate it to make it more useful.
While we're at it, fix a nearby conditional used to determine if
we have a non-bigalloc file system. It doesn't match usage elsewhere
in the code, and misleadingly suggests that an s_cluster_ratio value
of 0 would be legal.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
This patch implements fallocate's FALLOC_FL_INSERT_RANGE for Ext4.
1) Make sure that both offset and len are block size aligned.
2) Update the i_size of inode by len bytes.
3) Compute the file's logical block number against offset. If the computed
block number is not the starting block of the extent, split the extent
such that the block number is the starting block of the extent.
4) Shift all the extents which are lying between [offset, last allocated extent]
towards right by len bytes. This step will make a hole of len bytes
at offset.
Signed-off-by: Namjae Jeon <namjae.jeon@samsung.com>
Signed-off-by: Ashish Sangwan <a.sangwan@samsung.com>
Pull fourth vfs update from Al Viro:
"d_inode() annotations from David Howells (sat in for-next since before
the beginning of merge window) + four assorted fixes"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
RCU pathwalk breakage when running into a symlink overmounting something
fix I_DIO_WAKEUP definition
direct-io: only inc/dec inode->i_dio_count for file systems
fs/9p: fix readdir()
VFS: assorted d_backing_inode() annotations
VFS: fs/inode.c helpers: d_inode() annotations
VFS: fs/cachefiles: d_backing_inode() annotations
VFS: fs library helpers: d_inode() annotations
VFS: assorted weird filesystems: d_inode() annotations
VFS: normal filesystems (and lustre): d_inode() annotations
VFS: security/: d_inode() annotations
VFS: security/: d_backing_inode() annotations
VFS: net/: d_inode() annotations
VFS: net/unix: d_backing_inode() annotations
VFS: kernel/: d_inode() annotations
VFS: audit: d_backing_inode() annotations
VFS: Fix up some ->d_inode accesses in the chelsio driver
VFS: Cachefiles should perform fs modifications on the top layer only
VFS: AF_UNIX sockets should call mknod on the top layer only
that's the bulk of filesystem drivers dealing with inodes of their own
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Add an optimization for the MS_LAZYTIME mount option so that we will
opportunistically write out any inodes with the I_DIRTY_TIME flag set
in a particular inode table block when we need to update some inode in
that inode table block anyway.
Also add some temporary code so that we can set the lazytime mount
option without needing a modified /sbin/mount program which can set
MS_LAZYTIME. We can eventually make this go away once util-linux has
added support.
Google-Bug-Id: 18297052
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
In this commit we discard the lru algorithm for inodes with extent
status tree because it takes significant effort to maintain a lru list
in extent status tree shrinker and the shrinker can take a long time to
scan this lru list in order to reclaim some objects.
We replace the lru ordering with a simple round-robin. After that we
never need to keep a lru list. That means that the list needn't be
sorted if the shrinker can not reclaim any objects in the first round.
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently extent status tree doesn't cache extent hole when a write
looks up in extent tree to make sure whether a block has been allocated
or not. In this case, we don't put extent hole in extent cache because
later this extent might be removed and a new delayed extent might be
added back. But it will cause a defect when we do a lot of writes. If
we don't put extent hole in extent cache, the following writes also need
to access extent tree to look at whether or not a block has been
allocated. It brings a cache miss. This commit fixes this defect.
Also if the inode doesn't have any extent, this extent hole will be
cached as well.
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
For bigalloc filesystems we have to check whether newly requested inode
block isn't already part of a cluster for which we already have delayed
allocation reservation. This check happens in ext4_ext_map_blocks() and
that function sets EXT4_MAP_FROM_CLUSTER if that's the case. However if
ext4_da_map_blocks() finds in extent cache information about the block,
we don't call into ext4_ext_map_blocks() and thus we always end up
getting new reservation even if the space for cluster is already
reserved. This results in overreservation and premature ENOSPC reports.
Fix the problem by checking for existing cluster reservation already in
ext4_da_map_blocks(). That simplifies the logic and actually allows us
to get rid of the EXT4_MAP_FROM_CLUSTER flag completely.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
This commit adds some statictics in extent status tree shrinker. The
purpose to add these is that we want to collect more details when we
encounter a stall caused by extent status tree shrinker. Here we count
the following statictics:
stats:
the number of all objects on all extent status trees
the number of reclaimable objects on lru list
cache hits/misses
the last sorted interval
the number of inodes on lru list
average:
scan time for shrinking some objects
the number of shrunk objects
maximum:
the inode that has max nr. of objects on lru list
the maximum scan time for shrinking some objects
The output looks like below:
$ cat /proc/fs/ext4/sda1/es_shrinker_info
stats:
28228 objects
6341 reclaimable objects
5281/631 cache hits/misses
586 ms last sorted interval
250 inodes on lru list
average:
153 us scan time
128 shrunk objects
maximum:
255 inode (255 objects, 198 reclaimable)
125723 us max scan time
If the lru list has never been sorted, the following line will not be
printed:
586ms last sorted interval
If there is an empty lru list, the following lines also will not be
printed:
250 inodes on lru list
...
maximum:
255 inode (255 objects, 198 reclaimable)
0 us max scan time
Meanwhile in this commit a new trace point is defined to print some
details in __ext4_es_shrink().
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Jan Kara <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
This commit improves the trace point of extents status tree. We rename
trace_ext4_es_shrink_enter in ext4_es_count() because it is also used
in ext4_es_scan() and we can not identify them from the result.
Further this commit fixes a variable name in trace point in order to
keep consistency with others.
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Jan Kara <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently in ext4 there is quite a mess when it comes to naming
unwritten extents. Sometimes we call it uninitialized and sometimes we
refer to it as unwritten.
The right name for the extent which has been allocated but does not
contain any written data is _unwritten_. Other file systems are
using this name consistently, even the buffer head state refers to it as
unwritten. We need to fix this confusion in ext4.
This commit changes every reference to an uninitialized extent (meaning
allocated but unwritten) to unwritten extent. This includes comments,
function names and variable names. It even covers abbreviation of the
word uninitialized (such as uninit) and some misspellings.
This commit does not change any of the code paths at all. This has been
confirmed by comparing md5sums of the assembly code of each object file
after all the function names were stripped from it.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Currently EXT4_MAP_UNINIT is used in dioread_nolock case to mark the
cases where we're using dioread_nolock and we're writing into either
unallocated, or unwritten extent, because we need to make sure that
any DIO write into that inode will wait for the extent conversion.
However EXT4_MAP_UNINIT is not only entirely misleading name but also
unnecessary because we can check for EXT4_MAP_UNWRITTEN in the
dioread_nolock case instead.
This commit removes EXT4_MAP_UNINIT flag.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
In retrospect, this was a bad way to handle things, since it limited
testing of these patches. We should just get the VFS level changes
merged in first.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Introduce new FALLOC_FL_ZERO_RANGE flag for fallocate. This has the same
functionality as xfs ioctl XFS_IOC_ZERO_RANGE.
It can be used to convert a range of file to zeros preferably without
issuing data IO. Blocks should be preallocated for the regions that span
holes in the file, and the entire range is preferable converted to
unwritten extents
This can be also used to preallocate blocks past EOF in the same way as
with fallocate. Flag FALLOC_FL_KEEP_SIZE which should cause the inode
size to remain the same.
Also add appropriate tracepoints.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch implements fallocate's FALLOC_FL_COLLAPSE_RANGE for Ext4.
The semantics of this flag are following:
1) It collapses the range lying between offset and length by removing any data
blocks which are present in this range and than updates all the logical
offsets of extents beyond "offset + len" to nullify the hole created by
removing blocks. In short, it does not leave a hole.
2) It should be used exclusively. No other fallocate flag in combination.
3) Offset and length supplied to fallocate should be fs block size aligned
in case of xfs and ext4.
4) Collaspe range does not work beyond i_size.
Signed-off-by: Namjae Jeon <namjae.jeon@samsung.com>
Signed-off-by: Ashish Sangwan <a.sangwan@samsung.com>
Tested-by: Dongsu Park <dongsu.park@profitbricks.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
After applied the commit (4a092d73), we have reduced the number of
source files that need to #include ext4_extents.h. But we can do
better.
This commit defines ext4_zeroout_es() in extents.c and move
EXT_MAX_BLOCKS into ext4.h in order not to include ext4_extents.h in
indirect.c and ioctl.c. Meanwhile we just need to include this file in
extent_status.c when ES_AGGRESSIVE_TEST is defined. Otherwise, this
commit removes a duplicated declaration in trace/events/ext4.h.
After applied this patch, we just need to include ext4_extents.h file
in {super,migrate,move_extents,extents}.c, and it is easy for us to
define a new extent disk layout.
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When we read in an extent tree leaf block from disk, arrange to have
all of its entries cached. In nearly all cases the in-memory
representation will be more compact than the on-disk representation in
the buffer cache, and it allows us to get the information without
having to traverse the extent tree for successive extents.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Don't use an unsigned long long for the es_status flags; this requires
that we pass 64-bit values around which is painful on 32-bit systems.
Instead pass the extent status flags around using the low 4 bits of an
unsigned int, and shift them into place when we are reading or writing
es_pblk.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Translate the bitfields used in various flags argument to strings to
make the tracepoint output more human-readable.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Rename ext4_da_writepages() to ext4_writepages() and use it for all
modes. We still need to iterate over all the pages in the case of
data=journalling, but in the case of nodelalloc/data=ordered (which is
what file systems mounted using ext3 backwards compatibility will use)
this will allow us to use a much more efficient I/O submission path.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
There are two issues with current writeback path in ext4. For one we
don't necessarily map complete pages when blocksize < pagesize and
thus needn't do any writeback in one iteration. We always map some
blocks though so we will eventually finish mapping the page. Just if
writeback races with other operations on the file, forward progress is
not really guaranteed. The second problem is that current code
structure makes it hard to associate all the bios to some range of
pages with one io_end structure so that unwritten extents can be
converted after all the bios are finished. This will be especially
difficult later when io_end will be associated with reserved
transaction handle.
We restructure the writeback path to a relatively simple loop which
first prepares extent of pages, then maps one or more extents so that
no page is partially mapped, and once page is fully mapped it is
submitted for IO. We keep all the mapping and IO submission
information in mpage_da_data structure to somewhat reduce stack usage.
Resulting code is somewhat shorter than the old one and hopefully also
easier to read.
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Currently punch hole is disabled in file systems with bigalloc
feature enabled. However the recent changes in punch hole patch should
make it easier to support punching holes on bigalloc enabled file
systems.
This commit changes partial_cluster handling in ext4_remove_blocks(),
ext4_ext_rm_leaf() and ext4_ext_remove_space(). Currently
partial_cluster is unsigned long long type and it makes sure that we
will free the partial cluster if all extents has been released from that
cluster. However it has been specifically designed only for truncate.
With punch hole we can be freeing just some extents in the cluster
leaving the rest untouched. So we have to make sure that we will notice
cluster which still has some extents. To do this I've changed
partial_cluster to be signed long long type. The only scenario where
this could be a problem is when cluster_size == block size, however in
that case there would not be any partial clusters so we're safe. For
bigger clusters the signed type is enough. Now we use the negative value
in partial_cluster to mark such cluster used, hence we know that we must
not free it even if all other extents has been freed from such cluster.
This scenario can be described in simple diagram:
|FFF...FF..FF.UUU|
^----------^
punch hole
. - free space
| - cluster boundary
F - freed extent
U - used extent
Also update respective tracepoints to use signed long long type for
partial_cluster.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
->invalidatepage() aop now accepts range to invalidate so we can make
use of it in all ext4 invalidatepage routines.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
We (Linux Kernel Performance project) found a regression introduced
by commit:
f7fec032aa ext4: track all extent status in extent status tree
The commit causes about 20% performance decrease in fio random write
test. Profiler shows that rb_next() uses a lot of CPU time. The call
stack is:
rb_next
ext4_es_find_delayed_extent
ext4_map_blocks
_ext4_get_block
ext4_get_block_write
__blockdev_direct_IO
ext4_direct_IO
generic_file_direct_write
__generic_file_aio_write
ext4_file_write
aio_rw_vect_retry
aio_run_iocb
do_io_submit
sys_io_submit
system_call_fastpath
io_submit
td_io_getevents
io_u_queued_complete
thread_main
main
__libc_start_main
The cause is that ext4_es_find_delayed_extent() doesn't have an
upper bound, it keeps searching until a delayed extent is found.
When there are a lots of non-delayed entries in the extent state
tree, ext4_es_find_delayed_extent() may uses a lot of CPU time.
Reported-by: LKP project <lkp@linux.intel.com>
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
The only difference between how we handle data=ordered and
data=writeback is a single call to ext4_jbd2_file_inode(). Eliminate
code duplication by factoring out redundant the code paths.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
When the system is under memory pressure, ext4_es_srhink() will get
called very often. So optimize returning the number of items in the
file system's extent status cache by keeping a per-filesystem count,
instead of calculating it each time by scanning all of the inodes in
the extent status cache.
Also rename the slab used for the extent status cache to be
"ext4_extent_status" so it's obviousl the slab in question is created
by ext4.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: Zheng Liu <gnehzuil.liu@gmail.com>
Although extent status is loaded on-demand, we also need to reclaim
extent from the tree when we are under a heavy memory pressure because
in some cases fragmented extent tree causes status tree costs too much
memory.
Here we maintain a lru list in super_block. When the extent status of
an inode is accessed and changed, this inode will be move to the tail
of the list. The inode will be dropped from this list when it is
cleared. In the inode, a counter is added to count the number of
cached objects in extent status tree. Here only written/unwritten/hole
extent is counted because delayed extent doesn't be reclaimed due to
fiemap, bigalloc and seek_data/hole need it. The counter will be
increased as a new extent is allocated, and it will be decreased as a
extent is freed.
In this commit we use normal shrinker framework to reclaim memory from
the status tree. ext4_es_reclaim_extents_count() traverses the lru list
to count the number of reclaimable extents. ext4_es_shrink() tries to
reclaim written/unwritten/hole extents from extent status tree. The
inode that has been shrunk is moved to the tail of lru list.
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: Jan kara <jack@suse.cz>
After tracking all extent status, we already have a extent cache in
memory. Every time we want to lookup a block mapping, we can first
try to lookup it in extent status tree to avoid a potential disk I/O.
A new function called ext4_es_lookup_extent is defined to finish this
work. When we try to lookup a block mapping, we always call
ext4_map_blocks and/or ext4_da_map_blocks. So in these functions we
first try to lookup a block mapping in extent status tree.
A new flag EXT4_GET_BLOCKS_NO_PUT_HOLE is used in ext4_da_map_blocks
in order not to put a hole into extent status tree because this hole
will be converted to delayed extent in the tree immediately.
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: Jan kara <jack@suse.cz>
This commit renames ext4_es_find_extent with ext4_es_find_delayed_extent
and improve this function. First, we split input and output parameter.
Second, this function never return the first block of the next delayed
extent after 'es'.
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: Jan kara <jack@suse.cz>
This commit adds two members in extent_status structure to let it record
physical block and extent status. Here es_pblk is used to record both
of them because physical block only has 48 bits. So extent status could
be stashed into it so that we can save some memory. Now written,
unwritten, delayed and hole are defined as status.
Due to new member is added into extent status tree, all interfaces need
to be adjusted.
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
This commit refines the extent status tree code.
1) A prefix 'es_' is added to to the extent status tree structure
members.
2) Refactored es_remove_extent() so that __es_remove_extent() can be
used by es_insert_extent() to remove the old extent entry(-ies) before
inserting a new one.
3) Rename extent_status_end() to ext4_es_end()
4) ext4_es_can_be_merged() is define to check whether two extents can
be merged or not.
5) Update and clarified comments.
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
This patch adds a tracepoint in ext4_punch_hole.
CC: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
In data=journal mode we don't need delalloc or DIO handling in invalidatepage
and similarly in other modes we don't need the journal handling. So split
invalidatepage implementations.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch adds some tracepoints in extent status tree.
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When we use trace_ext4_ext/ind_map_blocks_exit, print the value of
map->m_flags in order that we can understand the extent's current
status.
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
In trace_ext4_ext_handle_uninitialized_extents we don't care about the
value of map->m_flags because this value is probably 0, and we prefer
to get the value of flags because we can know how to handle this
extent in this function.
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Most hardware architectures require that data (including struct fields)
have to be aligned in memory. To make it happen compiler inserts padding
between struct fields if they are not aligned correctly.
Reorder fields to remove paddings and make structures denser. Making data
smaller saves some memory that is very important for trace events.
Tracing buffer has limited size and making objects smaller we can put more
of them without overflowing the tracing buffer.
To find data struct holes I used 'pahole -H 1 -E -I vmlinux.o' from
'dwarves' package.
Signed-off-by: Anatol Pomozov <anatol.pomozov@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
pa_inode in group_pa is set NULL in ext4_mb_new_group_pa, so
pa_inode should be not referenced.
Reported-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>