percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Replace mark_inode_dirty() as nilfs_mark_inode_dirty()
to reduce deep function calls.
Signed-off-by: Jiro SEKIBA <jir@unicus.jp>
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
It is redundant to call mark_inode_dirty() in nilfs_new_inode() because
all caller of nilfs_new_inode() will call mark_inode_dirty()
after calling nilfs_new_inode() directly or indirectly in transaction.
Signed-off-by: Jiro SEKIBA <jir@unicus.jp>
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Previously, nilfs_bmap_add_blocks() and nilfs_bmap_sub_blocks() called
mark_inode_dirty() after they changed the number of data blocks.
This moves these calls outside bmap outermost functions like
nilfs_bmap_insert() or nilfs_bmap_truncate().
This will mitigate overhead for truncate or delete operation since
they repeatedly remove set of blocks. Nearly 10 percent improvement
was observed for removal of a large file:
# dd if=/dev/zero of=/test/aaa bs=1M count=512
# time rm /test/aaa
real 2.968s -> 2.705s
Further optimization may be possible by eliminating these
mark_inode_dirty() uses though I avoid mixing separate changes here.
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
This lock is eliminable because inodes on the buffer can be updated
independently. Although a log writer also fills in bmap data on the
on-disk inodes, this update is exclusively done by a log writer lock.
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
The comment says, "Caller of this function MUST lock s_inode_lock",
however just above the comment, it locks s_inode_lock in the function.
Signed-off-by: Jiro SEKIBA <jir@unicus.jp>
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
The i_dir_start_lookup field in nilfs_inode_info objects should be
cleared when the objects are allocated, but the the initialization was
missing in case of reading from disk. This adds the initialization.
Since the variable just gives a start page on directory lookups, the
bug was nonfatal until now.
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
The __nilfs_read_inode function is ignoring the error code returned
from nilfs_read_inode_common(), and wrongly delivers a success code
(zero) when it escapes from the function in erroneous cases.
This adds the missing error handling.
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Although get_block() callback function can return extent of contiguous
blocks with bh->b_size, nilfs_get_block() function did not support
this feature.
This adds contiguous lookup feature to the block mapping codes of
nilfs, and allows the nilfs_get_blocks() function to return the extent
information by applying the feature.
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
This adds a missing sync_page method which unplugs bio requests when
waiting for page locks. This will improve read performance of nilfs.
Here is a measurement result using dd command.
Without this patch:
# mount -t nilfs2 /dev/sde1 /test
# dd if=/test/aaa of=/dev/null bs=512k
1024+0 records in
1024+0 records out
536870912 bytes (537 MB) copied, 6.00688 seconds, 89.4 MB/s
With this patch:
# mount -t nilfs2 /dev/sde1 /test
# dd if=/test/aaa of=/dev/null bs=512k
1024+0 records in
1024+0 records out
536870912 bytes (537 MB) copied, 3.54998 seconds, 151 MB/s
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Hi,
I introduced "is_partially_uptodate" aops for NILFS2.
A page can have multiple buffers and even if a page is not uptodate, some buffers
can be uptodate on pagesize != blocksize environment.
This aops checks that all buffers which correspond to a part of a file
that we want to read are uptodate. If so, we do not have to issue actual
read IO to HDD even if a page is not uptodate because the portion we
want to read are uptodate.
"block_is_partially_uptodate" function is already used by ext2/3/4.
With the following patch random read/write mixed workloads or random read after
random write workloads can be optimized and we can get performance improvement.
I did a performance test using the sysbench.
1 --file-block-size=8K --file-total-size=2G --file-test-mode=rndrw --file-fsync-freq=0 --fil
e-rw-ratio=1 run
-2.6.30-rc5
Test execution summary:
total time: 151.2907s
total number of events: 200000
total time taken by event execution: 2409.8387
per-request statistics:
min: 0.0000s
avg: 0.0120s
max: 0.9306s
approx. 95 percentile: 0.0439s
Threads fairness:
events (avg/stddev): 12500.0000/238.52
execution time (avg/stddev): 150.6149/0.01
-2.6.30-rc5-patched
Test execution summary:
total time: 140.8828s
total number of events: 200000
total time taken by event execution: 2240.8577
per-request statistics:
min: 0.0000s
avg: 0.0112s
max: 0.8750s
approx. 95 percentile: 0.0418s
Threads fairness:
events (avg/stddev): 12500.0000/218.43
execution time (avg/stddev): 140.0536/0.01
arch: ia64
pagesize: 16k
Thanks.
Signed-off-by: Hisashi Hifumi <hifumi.hisashi@oss.ntt.co.jp>
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
After a review of user's feedback for finding out other compatibility
issues, I found nilfs improperly initializes timestamps in inode;
CURRENT_TIME was used there instead of CURRENT_TIME_SEC even though nilfs
didn't have nanosecond timestamps on disk. A few users gave us the report
that the tar program sometimes failed to expand symbolic links on nilfs,
and it turned out to be the cause.
Instead of applying the above displacement, I've decided to support
nanosecond timestamps on this occation. Fortunetaly, a needless 64-bit
field was in the nilfs_inode struct, and I found it's available for this
purpose without impact for the users.
So, this will do the enhancement and resolve the tar problem.
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The sketch file is a file to mark checkpoints with user data. It was
experimentally introduced in the original implementation, and now
obsolete. The file was handled differently with regular files; the file
size got truncated when a checkpoint was created.
This stops the special treatment and will treat it as a regular file.
Most users are not affected because mkfs.nilfs2 no longer makes this file.
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pekka Enberg advised me:
> It would be nice if BUG(), BUG_ON(), and panic() calls would be
> converted to proper error handling using WARN_ON() calls. The BUG()
> call in nilfs_cpfile_delete_checkpoints(), for example, looks to be
> triggerable from user-space via the ioctl() system call.
This will follow the comment and keep them to a minimum.
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pekka Enberg pointed out that double error handlings found after
nilfs_transaction_end() can be avoided by separating abort operation:
OK, I don't understand this. The only way nilfs_transaction_end() can
fail is if we have NILFS_TI_SYNC set and we fail to construct the
segment. But why do we want to construct a segment if we don't commit?
I guess what I'm asking is why don't we have a separate
nilfs_transaction_abort() function that can't fail for the erroneous
case to avoid this double error value tracking thing?
This does the separation and renames nilfs_transaction_end() to
nilfs_transaction_commit() for clarification.
Since, some calls of these functions were used just for exclusion control
against the segment constructor, they are replaced with semaphore
operations.
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Chris Mason pointed out that there is a missed sync issue in
nilfs_writepages():
On Wed, 17 Dec 2008 21:52:55 -0500, Chris Mason wrote:
> It looks like nilfs_writepage ignores WB_SYNC_NONE, which is used by
> do_sync_mapping_range().
where WB_SYNC_NONE in do_sync_mapping_range() was replaced with
WB_SYNC_ALL by Nick's patch (commit:
ee53a891f4).
This fixes the problem by letting nilfs_writepages() write out the log of
file data within the range if sync_mode is WB_SYNC_ALL.
This involves removal of nilfs_file_aio_write() which was previously
needed to ensure O_SYNC sync writes.
Cc: Chris Mason <chris.mason@oracle.com>
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
