Enhance the inode allocator to take a goal inode number as a
paremeter; if it is specified, it takes precedence over Orlov or
parent directory inode allocation algorithms.
The extents migration function uses the goal inode number so that the
extent trees allocated the migration function use the correct flex_bg.
In the future, the goal inode functionality will also be used to
allocate an adjacent inode for the extended attributes.
Also, for testing purposes the goal inode number can be specified via
/sys/fs/{dev}/inode_goal. This can be useful for testing inode
allocation beyond 2^32 blocks on very large filesystems.
Signed-off-by: Andreas Dilger <adilger@sun.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Instead of using a random number to determine the goal parent grop for
the Orlov top directories, use a hash of the directory name. This
allows for repeatable results when trying to benchmark filesystem
layout algorithms.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If the caller isn't planning on modifying the block group descriptors,
there's no need to pass in a pointer to a struct buffer_head. Nuking
this saves a tiny amount of CPU time and stack space usage.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
We have sb_bgl_lock() and ext4_group_info.bb_state
bit spinlock to protech group information. The later is only
used within mballoc code. Consolidate them to use sb_bgl_lock().
This makes the mballoc.c code much simpler and also avoid
confusion with two locks protecting same info.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Ext4's on-line resizing adds a new block group and then, only at the
last step adjusts s_groups_count. However, it's possible on SMP
systems that another CPU could see the updated the s_group_count and
not see the newly initialized data structures for the just-added block
group. For this reason, it's important to insert a SMP read barrier
after reading s_groups_count and before reading any (for example) the
new block group descriptors allowed by the increased value of
s_groups_count.
Unfortunately, we rather blatently violate this locking protocol
documented in fs/ext4/resize.c. Fortunately, (1) on-line resizes
happen relatively rarely, and (2) it seems rare that the filesystem
code will immediately try to use just-added block group before any
memory ordering issues resolve themselves. So apparently problems
here are relatively hard to hit, since ext3 has been vulnerable to the
same issue for years with no one apparently complaining.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Missing braces caused the warning to print more than once.
Signed-Off-By: Chuck Ebbert <cebbert@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Instead of looping over all of the block groups in a flex group
summing their summary statistics, start tracking used_dirs in struct
flex_groups, and use struct flex_groups instead. This should save a
bit of CPU for mkdir-heavy workloads.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reduce pressure on the sb_bgl_lock family of locks by using atomic_t's
to track the number of free blocks and inodes in each flex_group.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The find_group_flex() inode allocator is now only used if the
filesystem is mounted using the "oldalloc" mount option. It is
replaced with the original Orlov allocator that has been updated for
flex_bg filesystems (it should behave the same way if flex_bg is
disabled). The inode allocator now functions by taking into account
each flex_bg group, instead of each block group, when deciding whether
or not it's time to allocate a new directory into a fresh flex_bg.
The block allocator has also been changed so that the first block
group in each flex_bg is preferred for use for storing directory
blocks. This keeps directory blocks close together, which is good for
speeding up e2fsck since large directories are more likely to look
like this:
debugfs: stat /home/tytso/Maildir/cur
Inode: 1844562 Type: directory Mode: 0700 Flags: 0x81000
Generation: 1132745781 Version: 0x00000000:0000ad71
User: 15806 Group: 15806 Size: 1060864
File ACL: 0 Directory ACL: 0
Links: 2 Blockcount: 2072
Fragment: Address: 0 Number: 0 Size: 0
ctime: 0x499c0ff4:164961f4 -- Wed Feb 18 08:41:08 2009
atime: 0x499c0ff4:00000000 -- Wed Feb 18 08:41:08 2009
mtime: 0x49957f51:00000000 -- Fri Feb 13 09:10:25 2009
crtime: 0x499c0f57:00d51440 -- Wed Feb 18 08:38:31 2009
Size of extra inode fields: 28
BLOCKS:
(0):7348651, (1-258):7348654-7348911
TOTAL: 259
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
At the moment there are few restrictions on which flags may be set on
which inodes. Specifically DIRSYNC may only be set on directories and
IMMUTABLE and APPEND may not be set on links. Tighten that to disallow
TOPDIR being set on non-directories and only NODUMP and NOATIME to be set
on non-regular file, non-directories.
Introduces a flags masking function which masks flags based on mode and
use it during inode creation and when flags are set via the ioctl to
facilitate future consistency.
Signed-off-by: Duane Griffin <duaneg@dghda.com>
Acked-by: Andreas Dilger <adilger@sun.com>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
At present INDEX and EXTENTS are the only flags that new ext4 inodes do
NOT inherit from their parent. In addition prevent the flags DIRTY,
ECOMPR, IMAGIC, TOPDIR, HUGE_FILE and EXT_MIGRATE from being inherited.
List inheritable flags explicitly to prevent future flags from
accidentally being inherited.
This fixes the TOPDIR flag inheritance bug reported at
http://bugzilla.kernel.org/show_bug.cgi?id=9866.
Signed-off-by: Duane Griffin <duaneg@dghda.com>
Acked-by: Andreas Dilger <adilger@sun.com>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Use lowercase names of quota functions instead of old uppercase ones.
Signed-off-by: Jan Kara <jack@suse.cz>
Acked-by: Mingming Cao <cmm@us.ibm.com>
CC: linux-ext4@vger.kernel.org
This is a short-term warning, and even printk_ratelimit() can result
in too much noise in system logs. So only print it once as a warning.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
I was seeing fsck errors on inode bitmaps after a 4 thread
dbench run on a 4 cpu machine:
Inode bitmap differences: -50736 -(50752--50753) etc...
I believe that this is because ext4_free_inode() uses atomic
bitops, and although ext4_new_inode() *used* to also use atomic
bitops for synchronization, commit
393418676a changed this to use
the sb_bgl_lock, so that we could also synchronize against
read_inode_bitmap and initialization of uninit inode tables.
However, that change left ext4_free_inode using atomic bitops,
which I think leaves no synchronization between setting &
unsetting bits in the inode table.
The below patch fixes it for me, although I wonder if we're
getting at all heavy-handed with this spinlock...
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This is a workaround for find_group_flex() which badly needs to be
replaced. One of its problems (besides ignoring the Orlov algorithm)
is that it is a bit hyperactive about returning failure under
suspicious circumstances. This can lead to spurious ENOSPC failures
even when there are inodes still available.
Work around this for now by retrying the search using
find_group_other() if find_group_flex() returns -1. If
find_group_other() succeeds when find_group_flex() has failed, log a
warning message.
A better block/inode allocator that will fix this problem for real has
been queued up for the next merge window.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This mount option is largely superfluous, and in fact the way it was
implemented was buggy; if a filesystem which did not have the extents
feature flag was mounted -o extents, the filesystem would attempt to
create and use extents-based file even though the extents feature flag
was not eabled. The simplest thing to do is to nuke the mount option
entirely. It's not all that useful to force the non-creation of new
extent-based files if the filesystem can support it.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
We need to mark the block/inode bitmap beyond the end of the group
with '1'.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@kernel.org
For uninit block group, the on-disk bitmap is not initialized. That
implies we cannot depend on the uptodate flag on the bitmap
buffer_head to find bitmap validity. Use a new buffer_head flag which
would be set after we properly initialize the bitmap. This also
prevents (re-)initializing the uninit group bitmap every time we call
ext4_read_block_bitmap().
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@kernel.org
We need to make sure we update the inode bitmap and clear
EXT4_BG_INODE_UNINIT flag with sb_bgl_lock held, since
ext4_read_inode_bitmap() looks at EXT4_BG_INODE_UNINIT to decide
whether to initialize the inode bitmap each time it is called.
(introduced by commit c806e68f.)
ext4_read_inode_bitmap does:
spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
ext4_init_inode_bitmap(sb, bh, block_group, desc);
and ext4_new_inode does
if (!ext4_set_bit_atomic(sb_bgl_lock(sbi, group),
ino, inode_bitmap_bh->b_data))
......
...
spin_lock(sb_bgl_lock(sbi, group));
gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
i.e., on allocation we update the bitmap then we take the sb_bgl_lock
and clear the EXT4_BG_INODE_UNINIT flag. What can happen is a
parallel ext4_read_inode_bitmap can zero out the bitmap in between
the above ext4_set_bit_atomic and spin_lock(sb_bg_lock..)
The race results in below user visible errors
EXT4-fs error (device sdb1): ext4_free_inode: bit already cleared for inode 168449
EXT4-fs warning (device sdb1): ext4_unlink: Deleting nonexistent file ...
EXT4-fs warning (device sdb1): ext4_rmdir: empty directory has too many links ...
# ls -al /mnt/tmp/f/p369/d3/d6/d39/db2/dee/d10f/d3f/l71
ls: /mnt/tmp/f/p369/d3/d6/d39/db2/dee/d10f/d3f/l71: Stale NFS file handle
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@kernel.org
Rename some variables. We also unlock locks in the reverse order we
acquired as a part of cleanup.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Rename the lower bits with suffix _lo and add helper
to access the values. Also rename bg_itable_unused_hi
to bg_pad as in e2fsprogs.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Nearly all places in the ext3/4 code which uses "unsigned long" is
probably a bug, since on 32-bit systems a ulong a 32-bits, which means
we are wasting stack space on 64-bit systems.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This removes annoying blank syslog entries emitted by ext4_error() or
ext4_warning(), since these functions add their own newline.
Signed-off-by: Nick Warne <nick@ukfsn.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
A few weeks ago I posted a patch for discussion that allowed ext4 to run
without a journal. Since that time I've integrated the excellent
comments from Andreas and fixed several serious bugs. We're currently
running with this patch and generating some performance numbers against
both ext2 (with backported reservations code) and ext4 with and without
a journal. It just so happens that running without a journal is
slightly faster for most everything.
We did
iozone -T -t 4 s 2g -r 256k -T -I -i0 -i1 -i2
which creates 4 threads, each of which create and do reads and writes on
a 2G file, with a buffer size of 256K, using O_DIRECT for all file opens
to bypass the page cache. Results:
ext2 ext4, default ext4, no journal
initial writes 13.0 MB/s 15.4 MB/s 15.7 MB/s
rewrites 13.1 MB/s 15.6 MB/s 15.9 MB/s
reads 15.2 MB/s 16.9 MB/s 17.2 MB/s
re-reads 15.3 MB/s 16.9 MB/s 17.2 MB/s
random readers 5.6 MB/s 5.6 MB/s 5.7 MB/s
random writers 5.1 MB/s 5.3 MB/s 5.4 MB/s
So it seems that, so far, this was a useful exercise.
Signed-off-by: Frank Mayhar <fmayhar@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Conflicts:
security/keys/internal.h
security/keys/process_keys.c
security/keys/request_key.c
Fixed conflicts above by using the non 'tsk' versions.
Signed-off-by: James Morris <jmorris@namei.org>
Wrap access to task credentials so that they can be separated more easily from
the task_struct during the introduction of COW creds.
Change most current->(|e|s|fs)[ug]id to current_(|e|s|fs)[ug]id().
Change some task->e?[ug]id to task_e?[ug]id(). In some places it makes more
sense to use RCU directly rather than a convenient wrapper; these will be
addressed by later patches.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: Stephen Tweedie <sct@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: adilger@sun.com
Cc: linux-ext4@vger.kernel.org
Signed-off-by: James Morris <jmorris@namei.org>
When initializing an uninitialized block group in ext4_new_inode(),
its block group checksum must be re-calculated. This fixes a race
when several threads try to allocate a new inode in an UNINIT'd group.
There is some question whether we need to be initializing the block
bitmap in ext4_new_inode() at all, but for now, if we are going to
init the block group, let's eliminate the race.
Signed-off-by: Frederic Bohe <frederic.bohe@bull.net>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This fixes a bug which caused on-line resizing of filesystems with a
1k blocksize to fail. The root cause of this bug was the fact that if
an uninitalized bitmap block gets read in by userspace (which
e2fsprogs does try to avoid, but can happen when the blocksize is less
than the pagesize and an adjacent blocks is read into memory)
ext4_read_block_bitmap() was erroneously depending on the buffer
uptodate flag to decide whether it needed to initialize the bitmap
block in memory --- i.e., to set the standard set of blocks in use by
a block group (superblock, bitmaps, inode table, etc.). Essentially,
ext4_read_block_bitmap() assumed it was the only routine that might
try to read a block containing a block bitmap, which is simply not
true.
To fix this, ext4_read_block_bitmap() and ext4_read_inode_bitmap()
must always initialize uninitialized bitmap blocks. Once a block or
inode is allocated out of that bitmap, it will be marked as
initialized in the block group descriptor, so in general this won't
result any extra unnecessary work.
Signed-off-by: Frederic Bohe <frederic.bohe@bull.net>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The find_group_flex() function starts with best_flex as the
parent_fbg_group, which happens to have 0 inodes free. Some of the
flex groups searched have free blocks and free inodes, but the
flex_freeb_ratio is < 10, so they're skipped. Then when a group is
compared to the current "best" flex group, it does not have more free
blocks than "best", so it is skipped as well.
This continues until no flex group with free inodes is found which has
a proper ratio or which has more free blocks than the "best" group,
and we're left with a "best" group that has 0 inodes free, and we
return -ENOSPC.
We fix this by changing the logic so that if the current "best" flex
group has no inodes free, and the current one does have room, it is
promoted to the next "best."
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
I noticed when filling a 1T filesystem with 4 threads using the
fs_mark benchmark:
fs_mark -d /mnt/test -D 256 -n 100000 -t 4 -s 20480 -F -S 0
that I occasionally got checksum mismatch errors:
EXT4-fs error (device sdb): ext4_init_inode_bitmap: Checksum bad for group 6935
etc. I'd reliably get 4-5 of them during the run.
It appears that the problem is likely a race to init the bg's
when the uninit_bg feature is enabled.
With the patch below, which adds sb_bgl_locking around initialization,
I was able to complete several runs with no errors or warnings.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
ext4_read_block_bitmap and read_inode_bitmap do essentially
the same thing, and yet they are structured quite differently.
I came across this difference while looking at doing bg locking
during bg initialization.
This patch:
* removes unnecessary casts in the error messages
* renames read_inode_bitmap to ext4_read_inode_bitmap
* and more substantially, restructures the inode bitmap
reading function to be more like the block bitmap counterpart.
The change to the inode bitmap reader simplifies the locking
to be applied in the next patch.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
We've talked for a while about getting rid of any feature-
setting from the kernel; this gets rid of the code which would
set the INCOMPAT_EXTENTS flag on the first file write when mounted
as ext4[dev].
With this patch, if the extents feature is not already set on disk,
then mounting as ext4 will fall back to noextents with a warning,
and if -o extents is explicitly requested, the mount will fail,
also with warning.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch mostly controls the way inode are allocated in order to
make ialloc aware of flex_bg block group grouping. It achieves this
by bypassing the Orlov allocator when block group meta-data are packed
toghether through mke2fs. Since the impact on the block allocator is
minimal, this patch should have little or no effect on other block
allocation algorithms. By controlling the inode allocation, it can
basically control where the initial search for new block begins and
thus indirectly manipulate the block allocator.
This allocator favors data and meta-data locality so the disk will
gradually be filled from block group zero upward. This helps improve
performance by reducing seek time. Since the group of inode tables
within one flex_bg are treated as one giant inode table, uninitialized
block groups would not need to partially initialize as many inode
table as with Orlov which would help fsck time as the filesystem usage
goes up.
Signed-off-by: Jose R. Santos <jrs@us.ibm.com>
Signed-off-by: Valerie Clement <valerie.clement@bull.net>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Since this a non-static function, make it be ext4 specific to avoid
conflicts with potentially other filesystems.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If the orphan node list includes valid, untruncatable nodes with nlink > 0
the ext4_orphan_cleanup loop which attempts to delete them will not do so,
causing it to loop forever. Fix by checking for such nodes in the
ext4_orphan_get function.
This patch fixes the second case (image hdb.20000009.softlockup.gz)
reported in http://bugzilla.kernel.org/show_bug.cgi?id=10882.
Signed-off-by: Duane Griffin <duaneg@dghda.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
We should mark the inode dirty only after initializing the extent
tree. Also if we fail during extent initialization we need
to call DQUOT_FREE_INODE.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Move ext4 headers out of include/linux. This is just the trivial move,
there's some more thing that could be done later.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Mingming Cao <cmm@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
__FUNCTION__ is gcc-specific, use __func__
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch enables extent-formatted normal symlinks. Using extents
format allows a symlink to refer to a block number larger than 2^32
on large filesystems. We still don't enable extent format for fast
symlinks, which are contained in the inode itself.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Mingming Cao <cmm@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
In addition, don't inherit EXT4_EXTENTS_FL from parent directory.
If we have a directory with extent flag set and later mount the file
system with -o noextents, the files created in that directory will also
have extent flag set but we would not have called ext4_ext_tree_init for
them. This will cause error later when we are verifying the extent header
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Mingming Cao <cmm@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Andreas Dilger
Theodore Ts'o
Aneesh Kumar K.V
Chuck Ebbert
Duane Griffin
Jan Kara
Eric Sandeen
Frank Mayhar
Al Viro
James Morris
David Howells
Frederic Bohe
Jose R. Santos
Christoph Hellwig
Harvey Harrison
Marcin Slusarz
Benoit Boissinot