Skip to content
  1. Feb 10, 2013
  3. Feb 09, 2013
    • Theodore Ts'o's avatar
      ext4: start handle at the last possible moment in ext4_rmdir() · 8dcfaad2
      Theodore Ts'o authored 
      

Don't start the jbd2 transaction handle until after the directory
entry has been found, to minimize the amount of time that a handle is
held active.

Signed-off-by: default avatar"Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: default avatarJan Kara <jack@suse.cz>
      8dcfaad2
    • Theodore Ts'o's avatar
      ext4: start handle at the last possible moment in ext4_unlink() · 931b6864
      Theodore Ts'o authored 
      

Don't start the jbd2 transaction handle until after the directory
entry has been found, to minimize the amount of time that a handle is
held active.

Signed-off-by: default avatar"Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: default avatarJan Kara <jack@suse.cz>
      931b6864
    • Theodore Ts'o's avatar
      ext4: grab page before starting transaction handle in write_begin() · 47564bfb
      Theodore Ts'o authored 
      

The grab_cache_page_write_begin() function can potentially sleep for a
long time, since it may need to do memory allocation which can block
if the system is under significant memory pressure, and because it may
be blocked on page writeback.  If it does take a long time to grab the
page, it's better that we not hold an active jbd2 handle.

So grab a handle on the page first, and _then_ start the transaction
handle.

This commit fixes the following long transaction handle hold time:

postmark-2917  [000] ....   196.435786: jbd2_handle_stats: dev 254,32
   tid 570 type 2 line_no 2541 interval 311 sync 0 requested_blocks 1
   dirtied_blocks 0

Signed-off-by: default avatar"Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: default avatarJan Kara <jack@suse.cz>
      47564bfb
    • Theodore Ts'o's avatar
      ext4: pass context information to jbd2__journal_start() · 9924a92a
      Theodore Ts'o authored 
      

So we can better understand what bits of ext4 are responsible for
long-running jbd2 handles, use jbd2__journal_start() so we can pass
context information for logging purposes.

The recommended way for finding the longer-running handles is:

   T=/sys/kernel/debug/tracing
   EVENT=$T/events/jbd2/jbd2_handle_stats
   echo "interval > 5" > $EVENT/filter
   echo 1 > $EVENT/enable

   ./run-my-fs-benchmark

   cat $T/trace > /tmp/problem-handles

This will list handles that were active for longer than 20ms.  Having
longer-running handles is bad, because a commit started at the wrong
time could stall for those 20+ milliseconds, which could delay an
fsync() or an O_SYNC operation.  Here is an example line from the
trace file describing a handle which lived on for 311 jiffies, or over
1.2 seconds:

postmark-2917  [000] ....   196.435786: jbd2_handle_stats: dev 254,32 
   tid 570 type 2 line_no 2541 interval 311 sync 0 requested_blocks 1
   dirtied_blocks 0

Signed-off-by: default avatar"Theodore Ts'o" <tytso@mit.edu>
      9924a92a
    • Theodore Ts'o's avatar
      ext4: move the jbd2 wrapper functions out of super.c · 722887dd
      Theodore Ts'o authored 
      

Move the jbd2 wrapper functions which start and stop handles out of
super.c, where they don't really logically belong, and into
ext4_jbd2.c.

Signed-off-by: default avatar"Theodore Ts'o" <tytso@mit.edu>
      722887dd
    • Theodore Ts'o's avatar
      jbd2: add tracepoints which provide per-handle statistics · 343d9c28
      Theodore Ts'o authored 
      

Handles which stay open a long time are problematic when it comes time
to close down a transaction so it can be committed.  These tracepoints
will help us determine which ones are the problematic ones, and to
validate whether changes makes things better or worse.

Signed-off-by: default avatar"Theodore Ts'o" <tytso@mit.edu>
      343d9c28
  5. Feb 07, 2013
  7. Feb 05, 2013
    • Theodore Ts'o's avatar
      ext4: optimize mballoc for large allocations · 40ae3487
      Theodore Ts'o authored 
      

The ext4 block allocator only maintains buddy bitmaps for chunks which
are less than or equal to one quarter of a block group.  That is, for
a file aystem with a 1k blocksize, and where the number of blocks in a
block group is 8192 blocks, the largest chunk size tracked by buddy
bitmaps is 2048 blocks.

For a file system with a 4k blocksize, and where the number of blocks
in a block group is 32768 blocks, the largest chunk size tracked by
buddy bitmaps is 8192 blocks.

To work around this code, mballoc.c before this commit would truncate
allocation requests to the number of blocks in a block group minus 10.
Why 10?  Aside from being a completely arbitrary number, it avoids
block allocation to be a power of two larger than 25% of the block
group.  If you try to explicitly fallocate 50% of the block group
size, this will demonstrate the problem; the block allocation code
will scan the all of the blocks in the file system with cr==0 (since
the request is for a natural power of two), but then completely fail
for all blocks groups, since the buddy bitmaps don't track chunk sizes
of 50% of the block group.

To fix this, in these we use ext4_mb_complex_scan_group() instead of
ext4_mb_simple_scan_group().

Signed-off-by: default avatar"Theodore Ts'o" <tytso@mit.edu>
Cc: Andreas Dilger <adilger@dilger.ca>
      40ae3487
  9. Feb 03, 2013
  11. Feb 02, 2013
  13. Jan 30, 2013
  15. Jan 29, 2013
  17. Jan 28, 2013
  19. Jan 25, 2013