Currently ext4_ext_walk_space() only takes i_data_sem for read when
searching for the extent at given block with ext4_ext_find_extent().
Then it drops the lock and the extent tree can be changed at will.
However later on we're searching for the 'next' extent, but the extent
tree might already have changed, so the information might not be
accurate.
In fact we can hit BUG_ON(end <= start) if the extent got inserted into
the tree after the one we found and before the block we were searching
for. This has been reproduced by running xfstests 225 in loop on s390x
architecture, but theoretically we could hit this on any other
architecture as well, but probably not as often.
Moreover the extent currently in delayed allocation might be allocated
after we search the extent tree and before we search extent status tree
delayed buffers resulting in those delayed buffers being completely
missed, even though completely written and allocated.
We fix all those problems in several steps:
1. remove unnecessary callback indirection
2. rename functions
ext4_ext_walk_space -> ext4_fill_fiemap_extents
ext4_ext_fiemap_cb -> ext4_find_delayed_extent
3. move fiemap_fill_next_extent() into ext4_fill_fiemap_extents()
4. hold the i_data_sem for:
ext4_ext_find_extent()
ext4_ext_next_allocated_block()
ext4_find_delayed_extent()
5. call fiemap_fill_next_extent after releasing the i_data_sem
6. move path reinitialization into the critical section.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch lets ext4 maintain extent status tree.
Currently it only tracks delay extent status in extent status tree. When a
delay allocation is issued, the related delay extent will be inserted into
extent status tree. When a delay extent is written out or invalidated, it will
be removed from this tree.
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: Allison Henderson <achender@linux.vnet.ibm.com>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When ext4_ext_handle_uninitialized_extents(), we will directly return
from ext4_ext_map_blocks(). The trace point of
trace_ext4_ext_map_blocks_exit isn't called, and the user doesn't see
any result. This patch tries to fix this problem.
Meanwhile in ext4_ext_handle_uninitialized_extents it returns errors
or the number of allocated blocks. So 'ret' variable can be removed
due to previously modifications.
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
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>
We assumed that at the time we call ext4_convert_unwritten_extents_endio()
extent in question is fully inside [map.m_lblk, map->m_len] because
it was already split during submission. But this may not be true due to
a race between writeback vs fallocate.
If extent in question is larger than requested we will split it again.
Special precautions should being done if zeroout required because
[map.m_lblk, map->m_len] already contains valid data.
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
Fallocate should wait for pended ext4_convert_unwritten_extents()
otherwise following race may happen:
ftruncate( ,12288);
fallocate( ,0, 4096)
io_sibmit( ,0, 4096); /* Write to fallocated area, split extent if needed */
fallocate( ,0, 8192); /* Grow extent and broke assumption about extent */
Later kwork completion will do:
->ext4_convert_unwritten_extents (0, 4096)
->ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_IO_CONVERT_EXT);
->ext4_ext_map_blocks() /* Will find new extent: ex = [0,2] !!!!!! */
->ext4_ext_handle_uninitialized_extents()
->ext4_convert_unwritten_extents_endio()
/* convert [0,2] extent to initialized, but only[0,1] was written */
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
BUG #1) All places where we call ext4_flush_completed_IO are broken
because buffered io and DIO/AIO goes through three stages
1) submitted io,
2) completed io (in i_completed_io_list) conversion pended
3) finished io (conversion done)
And by calling ext4_flush_completed_IO we will flush only
requests which were in (2) stage, which is wrong because:
1) punch_hole and truncate _must_ wait for all outstanding unwritten io
regardless to it's state.
2) fsync and nolock_dio_read should also wait because there is
a time window between end_page_writeback() and ext4_add_complete_io()
As result integrity fsync is broken in case of buffered write
to fallocated region:
fsync blkdev_completion
->filemap_write_and_wait_range
->ext4_end_bio
->end_page_writeback
<-- filemap_write_and_wait_range return
->ext4_flush_completed_IO
sees empty i_completed_io_list but pended
conversion still exist
->ext4_add_complete_io
BUG #2) Race window becomes wider due to the 'ext4: completed_io
locking cleanup V4' patch series
This patch make following changes:
1) ext4_flush_completed_io() now first try to flush completed io and when
wait for any outstanding unwritten io via ext4_unwritten_wait()
2) Rename function to more appropriate name.
3) Assert that all callers of ext4_flush_unwritten_io should hold i_mutex to
prevent endless wait
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
Inode is allowed to have empty leaf only if it this is blockless inode.
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
punch_hole is the place where we have to wait for all existing writers
(writeback, aio, dio), but currently we simply flush pended end_io request
which is not sufficient. Other issue is that punch_hole performed w/o i_mutex
held which obviously result in dangerous data corruption due to
write-after-free.
This patch performs following changes:
- Guard punch_hole with i_mutex
- Recheck inode flags under i_mutex
- Block all new dio readers in order to prevent information leak caused by
read-after-free pattern.
- punch_hole now wait for all writers in flight
NOTE: XXX write-after-free race is still possible because new dirty pages
may appear due to mmap(), and currently there is no easy way to stop
writeback while punch_hole is in progress.
[ Fixed error return from ext4_ext_punch_hole() to make sure that we
release i_mutex before returning EPERM or ETXTBUSY -- Ted ]
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Current unwritten extent conversion state-machine is very fuzzy.
- For unknown reason it performs conversion under i_mutex. What for?
My diagnosis:
We already protect extent tree with i_data_sem, truncate and punch_hole
should wait for DIO, so the only data we have to protect is end_io->flags
modification, but only flush_completed_IO and end_io_work modified this
flags and we can serialize them via i_completed_io_lock.
Currently all these games with mutex_trylock result in the following deadlock
truncate: kworker:
ext4_setattr ext4_end_io_work
mutex_lock(i_mutex)
inode_dio_wait(inode) ->BLOCK
DEADLOCK<- mutex_trylock()
inode_dio_done()
#TEST_CASE1_BEGIN
MNT=/mnt_scrach
unlink $MNT/file
fallocate -l $((1024*1024*1024)) $MNT/file
aio-stress -I 100000 -O -s 100m -n -t 1 -c 10 -o 2 -o 3 $MNT/file
sleep 2
truncate -s 0 $MNT/file
#TEST_CASE1_END
Or use 286's xfstests https://github.com/dmonakhov/xfstests/blob/devel/286
This patch makes state machine simple and clean:
(1) xxx_end_io schedule final extent conversion simply by calling
ext4_add_complete_io(), which append it to ei->i_completed_io_list
NOTE1: because of (2A) work should be queued only if
->i_completed_io_list was empty, otherwise the work is scheduled already.
(2) ext4_flush_completed_IO is responsible for handling all pending
end_io from ei->i_completed_io_list
Flushing sequence consists of following stages:
A) LOCKED: Atomically drain completed_io_list to local_list
B) Perform extents conversion
C) LOCKED: move converted io's to to_free list for final deletion
This logic depends on context which we was called from.
D) Final end_io context destruction
NOTE1: i_mutex is no longer required because end_io->flags modification
is protected by ei->ext4_complete_io_lock
Full list of changes:
- Move all completion end_io related routines to page-io.c in order to improve
logic locality
- Move open coded logic from various xx_end_xx routines to ext4_add_complete_io()
- remove EXT4_IO_END_FSYNC
- Improve SMP scalability by removing useless i_mutex which does not
protect io->flags anymore.
- Reduce lock contention on i_completed_io_lock by optimizing list walk.
- Rename ext4_end_io_nolock to end4_end_io and make it static
- Check flush completion status to ext4_ext_punch_hole(). Because it is
not good idea to punch blocks from corrupted inode.
Changes since V3 (in request to Jan's comments):
Fall back to active flush_completed_IO() approach in order to prevent
performance issues with nolocked DIO reads.
Changes since V2:
Fix use-after-free caused by race truncate vs end_io_work
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
ext4_set_io_unwritten_flag() will increment i_unwritten counter, so
once we mark end_io with EXT4_END_IO_UNWRITTEN we have to revert it back
on error path.
- add missed error checks to prevent counter leakage
- ext4_end_io_nolock() will clear EXT4_END_IO_UNWRITTEN flag to signal
that conversion finished.
- add BUG_ON to ext4_free_end_io() to prevent similar leakage in future.
Visible effect of this bug is that unaligned aio_stress may deadlock
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Generic inode has unused i_private pointer which may be used as cur_aio_dio
storage.
TODO: If cur_aio_dio will be passed as an argument to get_block_t this allow
to have concurent AIO_DIO requests.
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Convert cpu_to_leXX(leXX_to_cpu(E1) + E2) to use leXX_add_cpu().
dpatch engine is used to auto generate this patch.
(https://github.com/weiyj/dpatch)
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Remove unused function ext4_ext_check_cache() and merge the code back to
the ext4_ext_in_cache().
Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Do not iterate over data blocks scanning for bh's to forget as they're
never exist. This improves time taken by unlink / truncate syscall.
Tested by continuously truncating file that is being written by dd.
Another test is rm -rf of linux tree while tar unpacks it. With
ordered data mode condition unlikely(!tbh) was always met in
ext4_free_blocks. With journal data mode tbh was found only few times,
so optimisation is also possible.
Unlinking fallocated 60G file after doing sync && echo 3 >
/proc/sys/vm/drop_caches && time rm --help
X86 before (linux 3.6-rc4):
# time rm -f test1
real 0m2.710s
user 0m0.000s
sys 0m1.530s
X86 after:
# time rm -f test1
real 0m0.644s
user 0m0.003s
sys 0m0.060s
MIPS before (linux 2.6.37):
# time rm -f test1
real 0m 4.93s
user 0m 0.00s
sys 0m 4.61s
MIPS after:
# time rm -f test1
real 0m 0.16s
user 0m 0.00s
sys 0m 0.06s
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrey Sidorov <qrxd43@motorola.com>
While performing punch hole for an inode, i_disksize is not changed.
So, there is no need to add the inode to orphan list.
Signed-off-by: Ashish Sangwan <ashish.sangwan2@gmail.com>
Signed-off-by: Namjae Jeon <linkinjeon@gmail.com>
Acked-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Currently in ext4 the length of zero-out chunk is set to 7 file system
blocks. But if an inode has uninitailized extents from using
fallocate to preallocate space, and the workload issues many random
writes, this can cause a fragmented extent tree that will
unnecessarily grow the extent tree.
So create a new sysfs tunable, extent_max_zeroout_kb, which controls
the maximum size where blocks will be zeroed out instead of creating a
new uninitialized extent. The default of this has been sent to 32kb.
CC: Zach Brown <zab@zabbo.net>
CC: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If an inode has more than 4 extents, but then later some of the
extents are merged together, we can optimize the file system by moving
the extents up into the inode, and discarding the extent tree block.
This is important, because if there are a large number of inodes with
an external extent tree blocks where the contents could fit in the
inode, this can significantly increase the fsck time of the file
system.
Google-Bug-Id: 6801242
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Commit 968dee7722: "ext4: fix hole punch failure when depth is greater
than 0" introduced a regression in v3.5.1/v3.6-rc1 which caused kernel
crashes when users ran run "rm -rf" on large directory hierarchy on
ext4 filesystems on RAID devices:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000028
Process rm (pid: 18229, threadinfo ffff8801276bc000, task ffff880123631710)
Call Trace:
[<ffffffff81236483>] ? __ext4_handle_dirty_metadata+0x83/0x110
[<ffffffff812353d3>] ext4_ext_truncate+0x193/0x1d0
[<ffffffff8120a8cf>] ? ext4_mark_inode_dirty+0x7f/0x1f0
[<ffffffff81207e05>] ext4_truncate+0xf5/0x100
[<ffffffff8120cd51>] ext4_evict_inode+0x461/0x490
[<ffffffff811a1312>] evict+0xa2/0x1a0
[<ffffffff811a1513>] iput+0x103/0x1f0
[<ffffffff81196d84>] do_unlinkat+0x154/0x1c0
[<ffffffff8118cc3a>] ? sys_newfstatat+0x2a/0x40
[<ffffffff81197b0b>] sys_unlinkat+0x1b/0x50
[<ffffffff816135e9>] system_call_fastpath+0x16/0x1b
Code: 8b 4d 20 0f b7 41 02 48 8d 04 40 48 8d 04 81 49 89 45 18 0f b7 49 02 48 83 c1 01 49 89 4d 00 e9 ae f8 ff ff 0f 1f 00 49 8b 45 28 <48> 8b 40 28 49 89 45 20 e9 85 f8 ff ff 0f 1f 80 00 00 00
RIP [<ffffffff81233164>] ext4_ext_remove_space+0xa34/0xdf0
This could be reproduced as follows:
The problem in commit 968dee7722 was that caused the variable 'i' to
be left uninitialized if the truncate required more space than was
available in the journal. This resulted in the function
ext4_ext_truncate_extend_restart() returning -EAGAIN, which caused
ext4_ext_remove_space() to restart the truncate operation after
starting a new jbd2 handle.
Reported-by: Maciej Żenczykowski <maze@google.com>
Reported-by: Marti Raudsepp <marti@juffo.org>
Tested-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
Whether to continue removing extents or not is decided by the return
value of function ext4_ext_more_to_rm() which checks 2 conditions:
a) if there are no more indexes to process.
b) if the number of entries are decreased in the header of "depth -1".
In case of hole punch, if the last block to be removed is not part of
the last extent index than this index will not be deleted, hence the
number of valid entries in the extent header of "depth - 1" will
remain as it is and ext4_ext_more_to_rm will return 0 although the
required blocks are not yet removed.
This patch fixes the above mentioned problem as instead of removing
the extents from the end of file, it starts removing the blocks from
the particular extent from which removing blocks is actually required
and continue backward until done.
Signed-off-by: Ashish Sangwan <ashish.sangwan2@gmail.com>
Signed-off-by: Namjae Jeon <linkinjeon@gmail.com>
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
Cc: stable@vger.kernel.org
In this patch, ext4_ext_try_to_merge has been change to merge
an extent both left and right. So we need to update the comment
in here.
Signed-off-by: HaiboLiu <HaiboLiu6@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Ext4 must make sure the transaction to be commited to the disk when
user opens a file with O_(D)SYNC flag and do a fallocate(2) call.
This problem had been reported by Christoph Hellwig in this thread:
http://www.spinics.net/lists/linux-btrfs/msg13621.html
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When truncating a file, we unmap pages from userspace first, as that's
usually more efficient than relying, page by page, on the fallback in
truncate_inode_page() - particularly if the file is mapped many times.
Do the same when punching a hole: 3.4 added truncate_pagecache_range()
to do the unmap and trunc, so use it in ext4_ext_punch_hole(), instead
of calling truncate_inode_pages_range() directly.
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Calculate and verify the checksum for each extent tree block. The
checksum is located in the space immediately after the last possible
ext4_extent in the block. The space is is typically the last 4-8
bytes in the block.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Create a new BH_Verified flag to indicate that we've verified all the
data in a buffer_head for correctness. This allows us to bypass
expensive verification steps when they are not necessary without
missing them when they are.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This fixes a scalability problem reported by Andi Kleen and Tim Chen;
they were quite secretive about the precise nature of their workload,
but they later admitted that it only showed up when they were using a
large sparse file, so the amount of data I/O that was needed was close
to zero. I'm not sure how realistic this is and it's only a
regression if you consider changes made since 2.6.39 to be a
"regression" vis-a-vis the policy regarding post-merge window bug
fixes, but Linus agreed it was worth fixing, so I'm including it in
this pull request.
This also fixes the journalled quota mount options, which I
accidentally broke while I was cleaning up the mount option handling.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
iQIcBAABCAAGBQJPjbs2AAoJENNvdpvBGATw/4UQAOMsxRlzbMnOAmohIBmesJiB
nTPX41dNw0QCXstMFjvSyRbUJI0NZPGg6ZDhbtoQ/c42/izZOVNd7gh7QQYHRCt2
Oqh6WS159P04ixcAe8NgCm5B5AV/C5Er/vSOUZENBaBo430vvrZasifsyUgQnx+b
PxXlYsfPVzaQVeSxCu/68OeiBRLcLwmKZ6MicOaWt9GCNoCsWzgU+/LNskYuscPI
841yQL6/BE7redU2E9qoEn/xjxx57hJOj2iiIAuqGPmNmRQqq3VtvTqNZHldNBLp
Hz5mB3zSZsPg0uwvp+OxrnpP37NQCCn1L64UFIXxqGF47mcCYw7schAoGBtqwGQS
neGUfkzG4beKk7kojyDawvnrrvVn4iCMaIkR1ZjzjPOk+QBPagckrS2nmuObbYzJ
l4lmHq1v8nOh0clxqJPioNp5/Y13sbpEOY4tAa6sLpzKLKXF330RNuwwrKKHB7zo
ZhCvSwVEjmxacgPCPhFJnR3fxtjoXWR8WvJs7H+gZB/QaC8hjEYw6xvvrkw8mAiS
RNe3cYdYAz6kOJdtJrJaMzp/1CYdHydf+0WvNYCQ/d1poPr7uU5wQY61hdm2gFxh
owsbVAOiEFjZWJHqrRRdcg2irTpINJTS3iRe4g/ltcvYzzRSeOcZNWvkKFspq3i8
EUMHRBxLzPMRa+gU6Unm
=jb3f
-----END PGP SIGNATURE-----
Merge tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
Pull ext4 regression fixes from Ted Ts'o:
"This fixes a scalability problem reported by Andi Kleen and Tim Chen;
they were quite secretive about the precise nature of their workload,
but they later admitted that it only showed up when they were using a
large sparse file, so the amount of data I/O that was needed was close
to zero.
I'm not sure how realistic this is and it's only a regression if you
consider changes made since 2.6.39 to be a "regression" vis-a-vis the
policy regarding post-merge window bug fixes, but Linus agreed it was
worth fixing, so I'm including it in this pull request.
This also fixes the journalled quota mount options, which I
accidentally broke while I was cleaning up the mount option handling."
* tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
ext4: fix handling of journalled quota options
ext4: address scalability issue by removing extent cache statistics
Andi Kleen and Tim Chen have reported that under certain circumstances
the extent cache statistics are causing scalability problems due to
cache line bounces.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
->ee_len is __le16, so assigning cpu_to_le32() to it is going to do
Bad Things(tm) on big-endian hosts...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
We are going to remove the EOFBLOCKS_FL flag in the future, so this is
the first part of the removal. We can not remove it entirely just now,
since the e2fsck is still checking for it and it might cause headache to
some people. Instead, remove the restrictive checks now and the rest
later, when the new e2fsck code is out and common enough.
This is also needed because punch hole already breaks the EOFBLOCKS_FL
semantics, so it might cause the some troubles. So simply remove it.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The error message produced by the ext4_ext_rm_leaf() when we are
removing blocks which accidentally ends up inside the existing extent,
is not very helpful, because we would like to also know which extent did
we collide with.
This commit changes the error message to get us also the information
about the extent we are colliding with.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Since the commit 'Rewrite punch hole to use ext4_ext_remove_space()'
reworked the punch hole implementation to use ext4_ext_remove_space()
instead of ext4_ext_map_blocks(), we can remove the code which is no
longer needed from the ext4_ext_map_blocks().
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This commit rewrites ext4 punch hole implementation to use
ext4_ext_remove_space() instead of its home gown way of doing this via
ext4_ext_map_blocks(). There are several reasons for changing this.
Firstly it is quite non obvious that punching hole needs to
ext4_ext_map_blocks() to punch a hole, especially given that this
function should map blocks, not unmap it. It also required a lot of new
code in ext4_ext_map_blocks().
Secondly the design of it is not very effective. The reason is that we
are trying to punch out blocks in ext4_ext_punch_hole() in opposite
direction than in ext4_ext_rm_leaf() which causes the ext4_ext_rm_leaf()
to iterate through the whole tree from the end to the start to find the
requested extent for every extent we are going to punch out.
And finally the current implementation does not use the existing code,
but bring a lot of new code, which is IMO unnecessary since there
already is some infrastructure we can use. Specifically
ext4_ext_remove_space().
This commit changes ext4_ext_remove_space() to accept 'end' parameter so
we can not only truncate to the end of file, but also remove the space
in the middle of the file (punch a hole). Moreover, because the last
block to punch out, might be in the middle of the extent, we have to
split the extent at 'end + 1' so ext4_ext_rm_leaf() can easily either
remove the whole fist part of split extent, or change its size.
ext4_ext_remove_space() is then used to actually remove the space
(extents) from within the hole, instead of ext4_ext_map_blocks().
Note that this also fix the issue with punch hole, where we would forget
to remove empty index blocks from the extent tree, resulting in double
free block error and file system corruption. This is simply because we
now use different code path, where this problem does not exist.
This has been tested with fsx running for several days and xfstests,
plus xfstest #251 with '-o discard' run on the loop image (which
converts discard requestes into punch hole to the backing file). All of
it on 1K and 4K file system block size.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Explicitly test for an extent whose length is zero, and flag that as a
corrupted extent.
This avoids a kernel BUG_ON assertion failure.
Tested: Without this patch, the file system image found in
tests/f_ext_zero_len/image.gz in the latest e2fsprogs sources causes a
kernel panic. With this patch, an ext4 file system error is noted
instead, and the file system is marked as being corrupted.
https://bugzilla.kernel.org/show_bug.cgi?id=42859
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@kernel.org
* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs:
ext2/3/4: delete unneeded includes of module.h
ext{3,4}: Fix potential race when setversion ioctl updates inode
udf: Mark LVID buffer as uptodate before marking it dirty
ext3: Don't warn from writepage when readonly inode is spotted after error
jbd: Remove j_barrier mutex
reiserfs: Force inode evictions before umount to avoid crash
reiserfs: Fix quota mount option parsing
udf: Treat symlink component of type 2 as /
udf: Fix deadlock when converting file from in-ICB one to normal one
udf: Cleanup calling convention of inode_getblk()
ext2: Fix error handling on inode bitmap corruption
ext3: Fix error handling on inode bitmap corruption
ext3: replace ll_rw_block with other functions
ext3: NULL dereference in ext3_evict_inode()
jbd: clear revoked flag on buffers before a new transaction started
ext3: call ext3_mark_recovery_complete() when recovery is really needed
Delete any instances of include module.h that were not strictly
required. In the case of ext2, the declaration of MODULE_LICENSE
etc. were in inode.c but the module_init/exit were in super.c, so
relocate the MODULE_LICENCE/AUTHOR block to super.c which makes it
consistent with ext3 and ext4 at the same time.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Fix ext4_debug format in ext4_ext_handle_uninitialized_extents() and
ext4_end_io_dio().
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
We found performance regression when using bigalloc with "nodelalloc"
(1MB cluster size):
1. mke2fs -C 1048576 -O ^has_journal,bigalloc /dev/sda
2. mount -o nodelalloc /dev/sda /test/
3. time dd if=/dev/zero of=/test/io bs=1048576 count=1024
The "dd" will cost about 2 seconds to finish, but if we mke2fs without
"bigalloc", "dd" will only cost less than 1 second.
The reason is: when using ext4 with "nodelalloc", it will call
ext4_find_delalloc_cluster() nearly everytime it call
ext4_ext_map_blocks(), and ext4_find_delalloc_range() will also scan
all pages in cluster because no buffer is "delayed". A cluster has
256 pages (1MB cluster), so it will scan 256 * 256k pags when creating
a 1G file. That severely hurts the performance.
Therefore, we return immediately from ext4_find_delalloc_range() in
nodelalloc mode, since by definition there can't be any delalloc
pages.
Signed-off-by: Robin Dong <sanbai@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
In get_implied_cluster_alloc(), rr_cluster_end was being
defined and set, but was never used. Removed this.
Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If a file is fallocated on a hole, map->m_lblk + map->m_len may be greater
than ee_block + ee_len.
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@kernel.org
Commit 1939dd84b3 ("ext4: cleanup ext4_ext_grow_indepth code") added a
reference to ext4_extent_header.eh_depth, but forget to pass the value
read through le16_to_cpu. The result is a crash on big-endian
machines, such as this crash on a POWER7 server:
attempt to access beyond end of device
sda8: rw=0, want=776392648163376, limit=168558560
Unable to handle kernel paging request for data at address 0x6b6b6b6b6b6b6bcb
Faulting instruction address: 0xc0000000001f5f38
cpu 0x14: Vector: 300 (Data Access) at [c000001bd1aaecf0]
pc: c0000000001f5f38: .__brelse+0x18/0x60
lr: c0000000002e07a4: .ext4_ext_drop_refs+0x44/0x80
sp: c000001bd1aaef70
msr: 9000000000009032
dar: 6b6b6b6b6b6b6bcb
dsisr: 40000000
current = 0xc000001bd15b8010
paca = 0xc00000000ffe4600
pid = 19911, comm = flush-8:0
enter ? for help
[c000001bd1aaeff0] c0000000002e07a4 .ext4_ext_drop_refs+0x44/0x80
[c000001bd1aaf090] c0000000002e0c58 .ext4_ext_find_extent+0x408/0x4c0
[c000001bd1aaf180] c0000000002e145c .ext4_ext_insert_extent+0x2bc/0x14c0
[c000001bd1aaf2c0] c0000000002e3fb8 .ext4_ext_map_blocks+0x628/0x1710
[c000001bd1aaf420] c0000000002b2974 .ext4_map_blocks+0x224/0x310
[c000001bd1aaf4d0] c0000000002b7f2c .mpage_da_map_and_submit+0xbc/0x490
[c000001bd1aaf5a0] c0000000002b8688 .write_cache_pages_da+0x2c8/0x430
[c000001bd1aaf720] c0000000002b8b28 .ext4_da_writepages+0x338/0x670
[c000001bd1aaf8d0] c000000000157280 .do_writepages+0x40/0x90
[c000001bd1aaf940] c0000000001ea830 .writeback_single_inode+0xe0/0x530
[c000001bd1aafa00] c0000000001eb680 .writeback_sb_inodes+0x210/0x300
[c000001bd1aafb20] c0000000001ebc84 .__writeback_inodes_wb+0xd4/0x140
[c000001bd1aafbe0] c0000000001ebfec .wb_writeback+0x2fc/0x3e0
[c000001bd1aafce0] c0000000001ed770 .wb_do_writeback+0x2f0/0x300
[c000001bd1aafdf0] c0000000001ed848 .bdi_writeback_thread+0xc8/0x340
[c000001bd1aafed0] c0000000000c5494 .kthread+0xb4/0xc0
[c000001bd1aaff90] c000000000021f48 .kernel_thread+0x54/0x70
This is due to getting ext_depth(inode) == 0x101 and therefore running
off the end of the path array in ext4_ext_drop_refs into following
unallocated structures.
This fixes it by adding the necessary le16_to_cpu.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The variable 'block' is removed by commit 750c9c47, so use the
replacement ex_ee_block instead.
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch fixes a syntax error which omits a comma. Besides this,
logical block number is unsigend 32 bits, so printk should use %u
instead %d.
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If an fallocate request fits in EXT_UNINIT_MAX_LEN, then set the
EXT4_GET_BLOCKS_NO_NORMALIZE flag. For larger fallocate requests,
let mballoc.c normalize the request.
This fixes a problem where large requests were being split into
non-contiguous extents due to commit 556b27abf7: ext4: do not
normalize block requests from fallocate.
Testing:
*) Checked that 8.x MB falloc'ed files are still laid down next to
each other (contiguously).
*) Checked that the maximum size extent (127.9MB) is allocated as 1
extent.
*) Checked that a 1GB file is somewhat contiguous (often 5-6
non-contiguous extents now).
*) Checked that a 120MB file can still be falloc'ed even if there are
no single extents large enough to hold it.
Signed-off-by: Greg Harm <gharm@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
EXT4_IO_END_UNWRITTEN flag set and the increase of i_aiodio_unwritten
should be done simultaneously since ext4_end_io_nolock always clear
the flag and decrease the counter in the same time.
We have found some bugs that the flag is set while leaving
i_aiodio_unwritten unchanged(commit 32c80b32c0). So this patch just tries
to create a helper function to wrap them to avoid any future bug.
The idea is inspired by Eric.
Cc: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When ext4_ext_map_blocks() is called by punch_hole, trace should
trace blocks punched out.
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
ext4_ext_insert_extent() (respectively ext4_ext_insert_index())
was using EXT_MAX_EXTENT() (resp. EXT_MAX_INDEX()) to determine
how many entries needed to be moved beyond the insertion point.
In practice this means that (320 - I) * 24 bytes were memmove()'d
when I is the insertion point, rather than (#entries - I) * 24 bytes.
This patch uses EXT_LAST_EXTENT() (resp. EXT_LAST_INDEX()) instead
to only move existing entries. The code flow is also simplified
slightly to highlight similarities and reduce code duplication in
the insertion logic.
This patch reduces system CPU consumption by over 25% on a 4kB
synchronous append DIO write workload when used with the
pre-2.6.39 x86_64 memmove() implementation. With the much faster
2.6.39 memmove() implementation we still see a decrease in
system CPU usage between 2% and 7%.
Note that the ext_debug() output changes with this patch, splitting
some log information between entries. Users of the ext_debug() output
should note that the "move %d" units changed from reporting the number
of bytes moved to reporting the number of entries moved.
Signed-off-by: Eric Gouriou <egouriou@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch introduces a fast path in ext4_ext_convert_to_initialized()
for the case when the conversion can be performed by transferring
the newly initialized blocks from the uninitialized extent into
an adjacent initialized extent. Doing so removes the expensive
invocations of memmove() which occur during extent insertion and
the subsequent merge.
In practice this should be the common case for clients performing
append writes into files pre-allocated via
fallocate(FALLOC_FL_KEEP_SIZE). In such a workload performed via
direct IO and when using a suboptimal implementation of memmove()
(x86_64 prior to the 2.6.39 rewrite), this patch reduces kernel CPU
consumption by 32%.
Two new trace points are added to ext4_ext_convert_to_initialized()
to offer visibility into its operations. No exit trace point has
been added due to the multiplicity of return points. This can be
revisited once the upstream cleanup is backported.
Signed-off-by: Eric Gouriou <egouriou@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When we want to convert the unitialized extent in direct write, we can
either do it in ext4_end_io_nolock(AIO case) or in
ext4_ext_direct_IO(non AIO case) and EXT4_I(inode)->cur_aio_dio is a
guard for ext4_ext_map_blocks to find the right case. In e9e3bcecf,
we mistakenly change it by:
- if (io)
+ if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
io->flag = EXT4_IO_END_UNWRITTEN;
- else
+ atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
+ } else
ext4_set_inode_state(inode,
EXT4_STATE_DIO_UNWRITTEN);
So now if we map 2 blocks, and the first one set the
EXT_IO_END_UNWRITTEN, the 2nd mapping will set inode state because of
the check for the flag. This is wrong.
Cc: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
In ext4_ext_next_allocated_block(), the path[depth] might
have a p_ext that is NULL -- see ext4_ext_binsearch(). In
such a case, dereferencing it will crash the machine.
This patch checks for p_ext == NULL in
ext4_ext_next_allocated_block() before dereferencinging it.
Tested using a hand-crafted an inode with eh_entries == 0 in
an extent block, verified that running FIEMAP on it crashes
without this patch, works fine with it.
Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When allocated is unsigned it breaks the error handling at the end
of the function when we call:
allocated = ext4_split_extent(...);
if (allocated < 0)
err = allocated;
I've made it a signed int instead of unsigned.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
EOFBLOCK_FL should be updated if called w/o FALLOCATE_FL_KEEP_SIZE
Currently it happens only if new extent was allocated.
TESTCASE:
fallocate test_file -n -l4096
fallocate test_file -l4096
Last fallocate cmd has updated size, but keept EOFBLOCK_FL set. And
fsck will complain about that.
Also remove ping pong in ext4_fallocate() in case of new extents,
where ext4_ext_map_blocks() clear EOFBLOCKS bit, and later
ext4_falloc_update_inode() restore it again.
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
- Both callers(truncate and punch_hole) already aligned left end point
so we no longer need split logic here.
- Remove dead duplicated code.
- Call ext4_ext_dirty only after we have updated eh_entries, otherwise
we'll loose entries update. Regression caused by d583fb87a3
266'th testcase in xfstests (http://patchwork.ozlabs.org/patch/120872)
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Currently code make an impression what grow procedure is very complicated
and some mythical paths, blocks are involved. But in fact grow in depth
it relatively simple procedure:
1) Just create new meta block and copy root data to that block.
2) Convert root from extent to index if old depth == 0
3) Update root block pointer
This patch does:
- Reorganize code to make it more self explanatory
- Do not pass path parameter to new_meta_block() in order to
provoke allocation from inode's group because top-level block
should site closer to it's inode, but not to leaf data block.
[ This happens anyway, due to logic in mballoc; we should drop
the path parameter from new_meta_block() entirely. -- tytso ]
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The third parameter to ext4_free_blocks is a struct buffer_head *. This
parameter should be NULL not 0.
This quiets the sparse noise:
warning: Using plain integer as NULL pointer
Signed-off-by: H Hartley Sweeten <hsweeten@visionengravers.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Add a sanity check to make sure ix hasn't gone beyond the valid bounds
of the extent block.
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
ext4_extent_idx.e_block is __le32, so use le32_to_cpu() in
ext4_ext_search_left().
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The comment describing what ext4_ext_search_right() does is incorrect.
We return 0 in *phys when *logical is the 'largest' allocated block,
not smallest.
Fix a few other typos while we're at it.
Cc: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Currently, there exists a race between delayed allocated writes and
the writeback when bigalloc feature is in use. The race was because we
wanted to determine what blocks in a cluster are under delayed
allocation and we were using buffer_delayed(bh) check for it. But, the
writeback codepath clears this bit without any synchronization which
resulted in a race and an ext4 warning similar to:
EXT4-fs (ram1): ext4_da_update_reserve_space: ino 13, used 1 with only 0
reserved data blocks
The race existed in two places.
(1) between ext4_find_delalloc_range() and ext4_map_blocks() when called from
writeback code path.
(2) between ext4_find_delalloc_range() and ext4_da_get_block_prep() (where
buffer_delayed(bh) is set.
To fix (1), this patch introduces a new buffer_head state bit -
BH_Da_Mapped. This bit is set under the protection of
EXT4_I(inode)->i_data_sem when we have actually mapped the delayed
allocated blocks during the writeout time. We can now reliably check
for this bit inside ext4_find_delalloc_range() to determine whether
the reservation for the blocks have already been claimed or not.
To fix (2), it was necessary to set buffer_delay(bh) under the
protection of i_data_sem. So, I extracted the very beginning of
ext4_map_blocks into a new function - ext4_da_map_blocks() - and
performed the required setting of bh_delay bit and the quota
reservation under the protection of i_data_sem. These two fixes makes
the checking of buffer_delay(bh) and buffer_da_mapped(bh) consistent,
thus removing the race.
Tested: I was able to reproduce the problem by running 'dd' and
'fsync' in parallel. Also, xfstests sometimes used to reproduce this
race. After the fix both my test and xfstests were successful and no
race (warning message) was observed.
Google-Bug-Id: 4997027
Signed-off-by: Aditya Kali <adityakali@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch adds some tracepoints in ext4/extents.c and updates a tracepoint in
ext4/inode.c.
Tested: Built and ran the kernel and verified that these tracepoints work.
Also ran xfstests.
Signed-off-by: Aditya Kali <adityakali@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
With bigalloc changes, the i_blocks value was not correctly set (it was still
set to number of blocks being used, but in case of bigalloc, we want i_blocks
to represent the number of clusters being used). Since the quota subsystem sets
the i_blocks value, this patch fixes the quota accounting and makes sure that
the i_blocks value is set correctly.
Signed-off-by: Aditya Kali <adityakali@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When we are truncating (as opposed unlinking) a file, we need to worry
about partial truncates of a file, especially in the light of sparse
files. The changes here make sure that arbitrary truncates of sparse
files works correctly. Yeah, it's messy.
Note that these functions will need to be revisted when the punch
ioctl is integrated --- in fact this commit will probably have merge
conflicts with the punch changes which Allison Henders and the IBM LTC
have been working on. I will need to fix this up when either patch
hits mainline.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If we need to allocate a new block in ext4_ext_map_blocks(), the
function needs to see if the cluster has already been allocated.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
While running extended fsx tests to verify the first
two patches, a similar bug was also found in the
truncate operation.
This bug happens because the truncate routine only zeros
the unblock aligned portion of the last page. This means
that the block aligned portions of the page appearing after
i_size are left unzeroed, and the buffer heads still mapped.
This bug is corrected by using ext4_discard_partial_page_buffers
in the truncate routine to zero the partial page and unmap
the buffer headers.
Signed-off-by: Allison Henderson <achender@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Add debugging information in case jbd2_journal_dirty_metadata() is
called with a buffer_head which didn't have
jbd2_journal_get_write_access() called on it, or if the journal_head
has the wrong transaction in it. In addition, return an error code.
This won't change anything for ocfs2, which will BUG_ON() the non-zero
exit code.
For ext4, the caller of this function is ext4_handle_dirty_metadata(),
and on seeing a non-zero return code, will call __ext4_journal_stop(),
which will print the function and line number of the (buggy) calling
function and abort the journal. This will allow us to recover instead
of bug halting, which is better from a robustness and reliability
point of view.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch fixes a second punch hole bug found by xfstests 127.
This bug happens because punch hole needs to flush the pages
of the hole to avoid race conditions. But if the end of the
hole is in the same page as i_size, the buffer heads beyond
i_size need to be unmapped and the page needs to be zeroed
after it is flushed.
To correct this, the new ext4_discard_partial_page_buffers
routine is used to zero and unmap the partial page
beyond i_size if the end of the hole appears in the same
page as i_size.
The code has also been optimized to set the end of the hole
to the page after i_size if the specified hole exceeds i_size,
and the code that flushes the pages has been simplified.
Signed-off-by: Allison Henderson <achender@linux.vnet.ibm.com>
This patch addresses a bug found by xfstests 75, 112, 127
when blocksize = 1k
This bug happens because the punch hole code only zeros
out non block aligned regions of the page. This means that if the
blocks are smaller than a page, then the block aligned regions of
the page inside the hole are left un-zeroed, and their buffer heads
are still mapped. This bug is corrected by using
ext4_discard_partial_page_buffers to properly zero the partial page
at the head and tail of the hole, and unmap the corresponding buffer
heads
This patch also addresses a bug reported by Lukas while working on a
new patch to add discard support for loop devices using punch hole.
The bug happened because of the first and last block number
needed to be cast to a larger data type before calculating the
byte offset, but since now we only need the byte offsets of the
pages, we no longer even need to be calculating the byte offsets
of the blocks. The code to do the block offset calculations is
removed in this patch.
Signed-off-by: Allison Henderson <achender@linux.vnet.ibm.com>
The logical block number in map.l_blk is a __u32, and so before we
shift it left, by the block size, we neeed cast it to a 64-bit size.
Otherwise i_size can be corrupted on an ENOSPC.
# df -T /mnt/mp1
Filesystem Type 1K-blocks Used Available Use% Mounted on
/dev/sda6 ext4 9843276 153056 9190200 2% /mnt/mp1
# fallocate -o 0 -l 2199023251456 /mnt/mp1/testfile
fallocate: /mnt/mp1/testfile: fallocate failed: No space left on device
# stat /mnt/mp1/testfile
File: `/mnt/mp1/testfile'
Size: 4293656576 Blocks: 19380440 IO Block: 4096 regular file
Device: 806h/2054d Inode: 12 Links: 1
Access: (0644/-rw-r--r--) Uid: ( 0/ root) Gid: ( 0/ root)
Access: 2011-07-25 13:01:31.414490496 +0900
Modify: 2011-07-25 13:01:31.414490496 +0900
Change: 2011-07-25 13:01:31.454490495 +0900
Signed-off-by: Utako Kusaka <u-kusaka@wm.jp.nec.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
--
fs/ext4/extents.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
The old function ext4_ext_rm_idx is used only for truncate case
because it just remove last index in extent-index-block. When punching
hole, it usually needed to remove "middle" index, therefore we must
move indexes which after it forward.
(I create a file with 1 depth extent tree and punch hole in the middle
of it, the last index in index-block strangly gone, so I find out this
bug)
Signed-off-by: Robin Dong <sanbai@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The debug message in ext4_ext_insert_extent before moving extent
is incorrect (the "from xx to xx").
Signed-off-by: Robin Dong <sanbai@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The argument "inode" in function ext4_ext_next_allocated_block looks useless,
so clean it.
Signed-off-by: Robin Dong <sanbai@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If eh_entries is equal to (or greater than) eh_max, the operation of
inserting new extent_idx will make number of entries overflow.
So check eh_entries before inserting the new extent_idx.
Although there is no bug case according the code (function
ext4_ext_insert_index is called by ext4_ext_split and ext4_ext_split
is called only if the index block has free space), the right logic
should be "lookup the capacity before insertion".
Signed-off-by: Robin Dong <sanbai@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch avoids an extraneous lookup of the extent cache
in ext4_ext_map_blocks() when the flag
EXT4_GET_BLOCKS_PUNCH_OUT_EXT is absent.
The existing logic was performing the lookup but not making
use of the result. The patch simply reverses the order of evaluation
in the condition.
Since ext4_ext_in_cache() does not initialize newex on misses, bypassing
its invocation does not introduce any new issue in this regard.
Signed-off-by: Robin Dong <sanbai@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Eric Gouriou <egouriou@google.com>
This patch removes the extra parameter in ext4_ext_remove_space()
which is no longer needed.
Signed-off-by: Allison Henderson <achender@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch optimizes the punch hole operation by skipping the
tree walking code that is used by truncate. Since punch hole
is done through map blocks, the path to the extent is already
known in this function, so we do not need to look it up again.
Signed-off-by: Allison Henderson <achender@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Optimize ext4_ext_insert_extent() by avoiding
ext4_ext_next_leaf_block() when the result is not used/needed.
Signed-off-by: Robin Dong <sanbai@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If eh->eh_entries is smaller than eh->eh_max, the routine will
go to the "repeat" and then go to "has_space" directlly ,
since argument "depth" and "eh" are not even changed.
Therefore, goto "has_space" directly and remove redundant "repeat" tag.
Signed-off-by: Robin Dong <sanbai@taobao.com>
Upon corrupted inode or disk failures, we may fail after we already
allocate some blocks from the inode or take some blocks from the
inode's preallocation list, but before we successfully insert the
corresponding extent to the extent tree. In this case, we should free
any allocated blocks and discard the inode's preallocated blocks
because the entries in the inode's preallocation list may be in an
inconsistent state.
Signed-off-by: Jiaying Zhang <jiayingz@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@kernel.org
The current implementation of ext4_free_blocks() always calls
dquot_free_block This looks quite sensible in the most cases: blocks
to be freed are associated with inode and were accounted in quota and
i_blocks some time ago.
However, there is a case when blocks to free were not accounted by the
time calling ext4_free_blocks() yet:
1. delalloc is on, write_begin pre-allocated some space in quota
2. write-back happens, ext4 allocates some blocks in ext4_ext_map_blocks()
3. then ext4_ext_map_blocks() gets an error (e.g. ENOSPC) from
ext4_ext_insert_extent() and calls ext4_free_blocks().
In this scenario, ext4_free_blocks() calls dquot_free_block() who, in
turn, decrements i_blocks for blocks which were not accounted yet (due
to delalloc) After clean umount, e2fsck reports something like:
> Inode 21, i_blocks is 5080, should be 5128. Fix<y>?
because i_blocks was erroneously decremented as explained above.
The patch fixes the problem by passing the new flag
EXT4_FREE_BLOCKS_NO_QUOT_UPDATE to ext4_free_blocks(), to request
that the dquot_free_block() call be skipped.
Signed-off-by: Maxim Patlasov <maxim.patlasov@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@kernel.org
I found that ext4_ext_find_goal() and ext4_find_near()
share the same code for returning a coloured start block
based on i_block_group.
We can refactor this into a common function so that they
don't diverge in the future.
Thanks to adilger for suggesting the new function name.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
In function ext4_ext_insert_index when eh_entries of curp is
bigger than eh_max, error messages will be printed out, but the content
is about logical and ei_block, that's incorret.
Signed-off-by: Robin Dong <sanbai@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Currently we are not marking the extent as the last one
(FIEMAP_EXTENT_LAST) if there is a hole at the end of the file. This is
because we just do not check for it right now and continue searching for
next extent. But at the point we hit the hole at the end of the file, it
is too late.
This commit adds check for the allocated block in subsequent extent and
if there is no more extents (block = EXT_MAX_BLOCKS) just flag the
current one as the last one.
This behaviour has been spotted unintentionally by 252 xfstest, when the
test hangs out, because of wrong loop condition. However on other
filesystems (like xfs) it will exit anyway, because we notice the last
extent flag and exit.
With this patch xfstest 252 does not hang anymore, ext4 fiemap
implementation still reports bad extent type in some cases, however
this seems to be different issue.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Kazuya Mio reported that he was able to hit BUG_ON(next == lblock)
in ext4_ext_put_gap_in_cache() while creating a sparse file in extent
format and fill the tail of file up to its end. We will hit the BUG_ON
when we write the last block (2^32-1) into the sparse file.
The root cause of the problem lies in the fact that we specifically set
s_maxbytes so that block at s_maxbytes fit into on-disk extent format,
which is 32 bit long. However, we are not storing start and end block
number, but rather start block number and length in blocks. It means
that in order to cover extent from 0 to EXT_MAX_BLOCK we need
EXT_MAX_BLOCK+1 to fit into len (because we counting block 0 as well) -
and it does not.
The only way to fix it without changing the meaning of the struct
ext4_extent members is, as Kazuya Mio suggested, to lower s_maxbytes
by one fs block so we can cover the whole extent we can get by the
on-disk extent format.
Also in many places EXT_MAX_BLOCK is used as length instead of maximum
logical block number as the name suggests, it is all a bit messy. So
this commit renames it to EXT_MAX_BLOCKS and change its usage in some
places to actually be maximum number of blocks in the extent.
The bug which this commit fixes can be reproduced as follows:
dd if=/dev/zero of=/mnt/mp1/file bs=<blocksize> count=1 seek=$((2**32-2))
sync
dd if=/dev/zero of=/mnt/mp1/file bs=<blocksize> count=1 seek=$((2**32-1))
Reported-by: Kazuya Mio <k-mio@sx.jp.nec.com>
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Make ext4_ext_split() get extents to be moved by calculating in a statement
instead of counting in a loop.
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Currently, an fallocate request of size slightly larger than a power of
2 is turned into two block requests, each a power of 2, with the extra
blocks pre-allocated for future use. When an application calls
fallocate, it already has an idea about how large the file may grow so
there is usually little benefit to reserve extra blocks on the
preallocation list. This reduces disk fragmentation.
Tested: fsstress. Also verified manually that fallocat'ed files are
contiguously laid out with this change (whereas without it they begin at
power-of-2 boundaries, leaving blocks in between). CPU usage of
fallocate is not appreciably higher. In a tight fallocate loop, CPU
usage hovers between 5%-8% with this change, and 5%-7% without it.
Using a simulated file system aging program which the file system to
70%, the percentage of free extents larger than 8MB (as measured by
e2freefrag) increased from 38.8% without this change, to 69.4% with
this change.
Signed-off-by: Vivek Haldar <haldar@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch adds new routines: "ext4_punch_hole" "ext4_ext_punch_hole"
and "ext4_ext_check_cache"
fallocate has been modified to call ext4_punch_hole when the punch hole
flag is passed. At the moment, we only support punching holes in
extents, so this routine is pretty much a wrapper for the ext4_ext_punch_hole
routine.
The ext4_ext_punch_hole routine first completes all outstanding writes
with the associated pages, and then releases them. The unblock
aligned data is zeroed, and all blocks in between are punched out.
The ext4_ext_check_cache routine is very similar to ext4_ext_in_cache
except it accepts a ext4_ext_cache parameter instead of a ext4_extent
parameter. This routine is used by ext4_ext_punch_hole to check and
see if a block in a hole that has been cached. The ext4_ext_cache
parameter is necessary because the members ext4_extent structure are
not large enough to hold a 32 bit value. The existing
ext4_ext_in_cache routine has become a wrapper to this new function.
[ext4 punch hole patch series 5/5 v7]
Signed-off-by: Allison Henderson <achender@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Mingming Cao <cmm@us.ibm.com>
This patch adds a new flag to ext4_map_blocks() that specifies the
given range of blocks should be punched out. Extents are first
converted to uninitialized extents before they are punched
out. Because punching a hole may require that the extent be split, it
is possible that the splitting may need more blocks than are
available. To deal with this, use of reserved blocks are enabled to
allow the split to proceed.
The routine then returns the number of blocks successfully
punched out.
[ext4 punch hole patch series 4/5 v7]
Signed-off-by: Allison Henderson <achender@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Mingming Cao <cmm@us.ibm.com>
This patch modifies the truncate routines to support hole punching
Below is a brief summary of the patches changes:
- Added end param to ext_ext4_rm_leaf
This function has been modified to accept an end parameter
which enables it to punch holes in leafs instead of just
truncating them.
- Implemented the "remove head" case in the ext_remove_blocks routine
This routine is used by ext_ext4_rm_leaf to remove the tail
of an extent during a truncate. The new ext_ext4_rm_leaf
routine will now also use it to remove the head of an extent in the
case that the hole covers a region of blocks at the beginning
of an extent.
- Added "end" param to ext4_ext_remove_space routine
This function has been modified to accept a stop parameter, which
is passed through to ext4_ext_rm_leaf.
[ext4 punch hole patch series 3/5 v6]
Signed-off-by: Allison Henderson <achender@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch adds an allocation request flag to the ext4_has_free_blocks
function which enables the use of reserved blocks. This will allow a
punch hole to proceed even if the disk is full. Punching a hole may
require additional blocks to first split the extents.
Because ext4_has_free_blocks is a low level function, the flag needs
to be passed down through several functions listed below:
ext4_ext_insert_extent
ext4_ext_create_new_leaf
ext4_ext_grow_indepth
ext4_ext_split
ext4_ext_new_meta_block
ext4_mb_new_blocks
ext4_claim_free_blocks
ext4_has_free_blocks
[ext4 punch hole patch series 1/5 v7]
Signed-off-by: Allison Henderson <achender@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Mingming Cao <cmm@us.ibm.com>
To get delayed-extent information, ext4_ext_fiemap_cb() looks up
pagecache, it thus collects information starting from a page's
head block.
If blocksize < pagesize, the beginning blocks of a page may lies
before the request range. So ext4_ext_fiemap_cb() should proceed
ignoring them, because they has been handled before. If no mapped
buffer in the range is found in the 1st page, we need to look up
the 2nd page, otherwise delayed-extents after a hole will be ignored.
Without this patch, xfstests 225 will hung on ext4 with 1K block.
Reported-by: Amir Goldstein <amir73il@users.sourceforge.net>
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
ext4_ext_truncate() should not invoke up_write(&EXT4_I(inode)->i_data_sem)
when ext4_orphan_add() returns an error, as it hasn't performed a
down_write() yet. This trivial patch fixes this by moving the up_write()
invocation above the out_stop label.
Signed-off-by: Eric Gouriou <egouriou@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The number of hits and misses for each filesystem is exposed in
/sys/fs/ext4/<dev>/extent_cache_{hits, misses}.
Tested: fsstress, manual checks.
Signed-off-by: Vivek Haldar <haldar@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch addresses bugs found while testing punch hole
with the fsx test. The patch corrects the number of blocks
that are zeroed out while splitting an extent, and also corrects
the return value to return the number of blocks split out, instead
of the number of blocks zeroed out.
This patch has been tested in addition to the following patches:
[Ext4 punch hole v7]
[XFS Tests Punch Hole 1/1 v2] Add Punch Hole Testing to FSX
The test ran successfully for 24 hours.
Signed-off-by: Allison Henderson <achender@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Add two functions: ext4_split_extent_at(), which splits an extent into
two extents at given logical block, and ext4_split_extent() which
splits an extent into three extents.
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Tested-by: Allison Henderson <achender@linux.vnet.ibm.com>
1) Rename ext4_ext_try_to_merge() to ext4_ext_try_to_merge_right().
2) Add a new function ext4_ext_try_to_merge() which tries to merge
an extent both left and right.
3) Use the new function in ext4_ext_convert_unwritten_endio() and
ext4_ext_insert_extent().
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Tested-by: Allison Henderson <achender@linux.vnet.ibm.com>
* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: (43 commits)
ext4: fix a BUG in mb_mark_used during trim.
ext4: unused variables cleanup in fs/ext4/extents.c
ext4: remove redundant set_buffer_mapped() in ext4_da_get_block_prep()
ext4: add more tracepoints and use dev_t in the trace buffer
ext4: don't kfree uninitialized s_group_info members
ext4: add missing space in printk's in __ext4_grp_locked_error()
ext4: add FITRIM to compat_ioctl.
ext4: handle errors in ext4_clear_blocks()
ext4: unify the ext4_handle_release_buffer() api
ext4: handle errors in ext4_rename
jbd2: add COW fields to struct jbd2_journal_handle
jbd2: add the b_cow_tid field to journal_head struct
ext4: Initialize fsync transaction ids in ext4_new_inode()
ext4: Use single thread to perform DIO unwritten convertion
ext4: optimize ext4_bio_write_page() when no extent conversion is needed
ext4: skip orphan cleanup if fs has unknown ROCOMPAT features
ext4: use the nblocks arg to ext4_truncate_restart_trans()
ext4: fix missing iput of root inode for some mount error paths
ext4: make FIEMAP and delayed allocation play well together
ext4: suppress verbose debugging information if malloc-debug is off
...
Fi up conflicts in fs/ext4/super.c due to workqueue changes
- Add more ext4 tracepoints.
- Change ext4 tracepoints to use dev_t field with MAJOR/MINOR macros
so that we can save 4 bytes in the ring buffer on some platforms.
- Add sync_mode to ext4_da_writepages, ext4_da_write_pages, and
ext4_da_writepages_result tracepoints. Also remove for_reclaim
field from ext4_da_writepages since it is usually not very useful.
Signed-off-by: Jiaying Zhang <jiayingz@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Fix the FIEMAP ioctl so that it returns all of the page ranges which
are still subject to delayed allocation. We were missing some cases
if the file was sparse.
Reported by Chris Mason <chris.mason@oracle.com>:
>We've had reports on btrfs that cp is giving us files full of zeros
>instead of actually copying them. It was tracked down to a bug with
>the btrfs fiemap implementation where it was returning holes for
>delalloc ranges.
>
>Newer versions of cp are trusting fiemap to tell it where the holes
>are, which does seem like a pretty neat trick.
>
>I decided to give xfs and ext4 a shot with a few tests cases too, xfs
>passed with all the ones btrfs was getting wrong, and ext4 got the basic
>delalloc case right.
>$ mkfs.ext4 /dev/xxx
>$ mount /dev/xxx /mnt
>$ dd if=/dev/zero of=/mnt/foo bs=1M count=1
>$ fiemap-test foo
>ext: 0 logical: [ 0.. 255] phys: 0.. 255
>flags: 0x007 tot: 256
>
>Horray! But once we throw a hole in, things go bad:
>$ mkfs.ext4 /dev/xxx
>$ mount /dev/xxx /mnt
>$ dd if=/dev/zero of=/mnt/foo bs=1M count=1 seek=1
>$ fiemap-test foo
>< no output >
>
>We've got a delalloc extent after the hole and ext4 fiemap didn't find
>it. If I run sync to kick the delalloc out:
>$sync
>$ fiemap-test foo
>ext: 0 logical: [ 256.. 511] phys: 34048.. 34303
>flags: 0x001 tot: 256
>
>fiemap-test is sitting in my /usr/local/bin, and I have no idea how it
>got there. It's full of pretty comments so I know it isn't mine, but
>you can grab it here:
>
>http://oss.oracle.com/~mason/fiemap-test.c
>
>xfsqa has a fiemap program too.
After Fix, test results are as follows:
ext: 0 logical: [ 256.. 511] phys: 0.. 255
flags: 0x007 tot: 256
ext: 0 logical: [ 256.. 511] phys: 33280.. 33535
flags: 0x001 tot: 256
$ mkfs.ext4 /dev/xxx
$ mount /dev/xxx /mnt
$ dd if=/dev/zero of=/mnt/foo bs=1M count=1 seek=1
$ sync
$ dd if=/dev/zero of=/mnt/foo bs=1M count=1 seek=3
$ dd if=/dev/zero of=/mnt/foo bs=1M count=1 seek=5
$ fiemap-test foo
ext: 0 logical: [ 256.. 511] phys: 33280.. 33535
flags: 0x000 tot: 256
ext: 1 logical: [ 768.. 1023] phys: 0.. 255
flags: 0x006 tot: 256
ext: 2 logical: [ 1280.. 1535] phys: 0.. 255
flags: 0x007 tot: 256
Tested-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch fixes the warning "Using plain integer as NULL pointer",
generated by sparse, by replacing the offending 0s with NULL.
Signed-off-by: Peter Huewe <peterhuewe@gmx.de>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
ext4 has a data corruption case when doing non-block-aligned
asynchronous direct IO into a sparse file, as demonstrated
by xfstest 240.
The root cause is that while ext4 preallocates space in the
hole, mappings of that space still look "new" and
dio_zero_block() will zero out the unwritten portions. When
more than one AIO thread is going, they both find this "new"
block and race to zero out their portion; this is uncoordinated
and causes data corruption.
Dave Chinner fixed this for xfs by simply serializing all
unaligned asynchronous direct IO. I've done the same here.
The difference is that we only wait on conversions, not all IO.
This is a very big hammer, and I'm not very pleased with
stuffing this into ext4_file_write(). But since ext4 is
DIO_LOCKING, we need to serialize it at this high level.
I tried to move this into ext4_ext_direct_IO, but by then
we have the i_mutex already, and we will wait on the
work queue to do conversions - which must also take the
i_mutex. So that won't work.
This was originally exposed by qemu-kvm installing to
a raw disk image with a normal sector-63 alignment. I've
tested a backport of this patch with qemu, and it does
avoid the corruption. It is also quite a lot slower
(14 min for package installs, vs. 8 min for well-aligned)
but I'll take slow correctness over fast corruption any day.
Mingming suggested that we can track outstanding
conversions, and wait on those so that non-sparse
files won't be affected, and I've implemented that here;
unaligned AIO to nonsparse files won't take a perf hit.
[tytso@mit.edu: Keep the mutex as a hashed array instead
of bloating the ext4 inode]
[tytso@mit.edu: Fix up namespace issues so that global
variables are protected with an "ext4_" prefix.]
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Currently all filesystems except XFS implement fallocate asynchronously,
while XFS forced a commit. Both of these are suboptimal - in case of O_SYNC
I/O we really want our allocation on disk, especially for the !KEEP_SIZE
case where we actually grow the file with user-visible zeroes. On the
other hand always commiting the transaction is a bad idea for fast-path
uses of fallocate like for example in recent Samba versions. Given
that block allocation is a data plane operation anyway change it from
an inode operation to a file operation so that we have the file structure
available that lets us check for O_SYNC.
This also includes moving the code around for a few of the filesystems,
and remove the already unnedded S_ISDIR checks given that we only wire
up fallocate for regular files.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Instead of various home grown checks that might need updates for new
flags just check for any bit outside the mask of the features supported
by the filesystem. This makes the check future proof for any newly
added flag.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6: (41 commits)
fs: add documentation on fallocate hole punching
Gfs2: fail if we try to use hole punch
Btrfs: fail if we try to use hole punch
Ext4: fail if we try to use hole punch
Ocfs2: handle hole punching via fallocate properly
XFS: handle hole punching via fallocate properly
fs: add hole punching to fallocate
vfs: pass struct file to do_truncate on O_TRUNC opens (try #2)
fix signedness mess in rw_verify_area() on 64bit architectures
fs: fix kernel-doc for dcache::prepend_path
fs: fix kernel-doc for dcache::d_validate
sanitize ecryptfs ->mount()
switch afs
move internal-only parts of ncpfs headers to fs/ncpfs
switch ncpfs
switch 9p
pass default dentry_operations to mount_pseudo()
switch hostfs
switch affs
switch configfs
...
Ext4 doesn't have the ability to punch holes yet, so make sure we return
EOPNOTSUPP if we try to use hole punching through fallocate. This support can
be added later. Thanks,
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Since check_eofblocks_fl() only uses the m_lblk portion of the map
structure, we may as well pass that directly, rather than passing the
entire map, which IMHO obfuscates what parameters check_eofblocks_fl()
cares about. Not a big deal, but seems tidier and less confusing, to
me.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Ted first found the bug when running 2.6.36 kernel with dioread_nolock
mount option that xfstests #13 complained about wrong file size during fsck.
However, the bug exists in the older kernels as well although it is
somehow harder to trigger.
The problem is that ext4_end_io_work() can happen after we have truncated an
inode to a smaller size. Then when ext4_end_io_work() calls
ext4_convert_unwritten_extents(), we may reallocate some blocks that have
been truncated, so the inode size becomes inconsistent with the allocated
blocks.
The following patch flushes the i_completed_io_list during truncate to reduce
the risk that some pending end_io requests are executed later and convert
already truncated blocks to initialized.
Note that although the fix helps reduce the problem a lot there may still
be a race window between vmtruncate() and ext4_end_io_work(). The fundamental
problem is that if vmtruncate() is called without either i_mutex or i_alloc_sem
held, it can race with an ongoing write request so that the io_end request is
processed later when the corresponding blocks have been truncated.
Ted and I have discussed the problem offline and we saw a few ways to fix
the race completely:
a) We guarantee that i_mutex lock and i_alloc_sem write lock are both hold
whenever vmtruncate() is called. The i_mutex lock prevents any new write
requests from entering writeback and the i_alloc_sem prevents the race
from ext4_page_mkwrite(). Currently we hold both locks if vmtruncate()
is called from do_truncate(), which is probably the most common case.
However, there are places where we may call vmtruncate() without holding
either i_mutex or i_alloc_sem. I would like to ask for other people's
opinions on what locks are expected to be held before calling vmtruncate().
There seems a disagreement among the callers of that function.
b) We change the ext4 write path so that we change the extent tree to contain
the newly allocated blocks and update i_size both at the same time --- when
the write of the data blocks is completed.
c) We add some additional locking to synchronize vmtruncate() and
ext4_end_io_work(). This approach may have performance implications so we
need to be careful.
All of the above proposals may require more substantial changes, so
we may consider to take the following patch as a bandaid.
Signed-off-by: Jiaying Zhang <jiayingz@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
We can encode the ec_type information by using ee_len == 0 to denote
EXT4_EXT_CACHE_NO, ee_start == 0 to denote EXT4_EXT_CACHE_GAP, and if
neither is true, then the cache type must be EXT4_EXT_CACHE_EXTENT.
This allows us to reduce the size of ext4_ext_inode by another 8
bytes. (ec_type is 4 bytes, plus another 4 bytes of padding)
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This fixes a number of places where we used sector_t instead of
ext4_lblk_t for logical blocks, which for ext4 are still 32-bit data
types. No point wasting space in the ext4_inode_info structure, and
requiring 64-bit arithmetic on 32-bit systems, when it isn't
necessary.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
ext4_ext_find_goal() returns an ideal physical block number that the block
allocator tries to allocate first. However, if a required file offset is
smaller than the existing extent's one, ext4_ext_find_goal() returns
a wrong block number because it may overflow at
"block - le32_to_cpu(ex->ee_block)". This patch fixes the problem.
ext4_ext_find_goal() will also return a wrong block number in case
a file offset of the existing extent is too big. In this case,
the ideal physical block number is fixed in ext4_mb_initialize_context(),
so it's no problem.
reproduce:
# dd if=/dev/zero of=/mnt/mp1/tmp bs=127M count=1 oflag=sync
# dd if=/dev/zero of=/mnt/mp1/file bs=512K count=1 seek=1 oflag=sync
# filefrag -v /mnt/mp1/file
Filesystem type is: ef53
File size of /mnt/mp1/file is 1048576 (256 blocks, blocksize 4096)
ext logical physical expected length flags
0 128 67456 128 eof
/mnt/mp1/file: 2 extents found
# rm -rf /mnt/mp1/tmp
# echo $((512*4096)) > /sys/fs/ext4/loop0/mb_stream_req
# dd if=/dev/zero of=/mnt/mp1/file bs=512K count=1 oflag=sync conv=notrunc
result (linux-2.6.37-rc2 + ext4 patch queue):
# filefrag -v /mnt/mp1/file
Filesystem type is: ef53
File size of /mnt/mp1/file is 1048576 (256 blocks, blocksize 4096)
ext logical physical expected length flags
0 0 33280 128
1 128 67456 33407 128 eof
/mnt/mp1/file: 2 extents found
result(apply this patch):
# filefrag -v /mnt/mp1/file
Filesystem type is: ef53
File size of /mnt/mp1/file is 1048576 (256 blocks, blocksize 4096)
ext logical physical expected length flags
0 0 66560 128
1 128 67456 66687 128 eof
/mnt/mp1/file: 2 extents found
Signed-off-by: Kazuya Mio <k-mio@sx.jp.nec.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When I compiled 2.6.36-rc3 kernel with EXT4FS_DEBUG definition, I got
the following compile error.
CC [M] fs/ext4/extents.o
fs/ext4/extents.c: In function 'ext4_fallocate':
fs/ext4/extents.c:3772: error: 'block' undeclared (first use in this function)
fs/ext4/extents.c:3772: error: (Each undeclared identifier is reported only once
fs/ext4/extents.c:3772: error: for each function it appears in.)
make[2]: *** [fs/ext4/extents.o] Error 1
The patch fixes this problem.
Signed-off-by: Kazuya Mio <k-mio@sx.jp.nec.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cleanup namespace leaks from fs/ext4 and the inline trivial functions
ext4_{ext,idx}_pblock() and ext4_{ext,idx}_store_pblock() since the
code size actually shrinks when we make these functions inline,
they're so trivial.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
These functions have no need to be exported beyond file context.
No functions needed to be moved for this commit; just some function
declarations changed to be static and removed from header files.
(A similar patch was submitted by Eric Sandeen, but I wanted to handle
code movement in separate patches to make sure code changes didn't
accidentally get dropped.)
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Call the block I/O layer directly instad of going through the buffer
layer. This should give us much better performance and scalability,
as well as lowering our CPU utilization when doing buffered writeback.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Change ext4_ext_zeroout to use sb_issue_zeroout instead of its
own approach to zero out extents.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
It turns out we have several problems with how EOFBLOCKS_FL is
handled. First of all, there was a fencepost error where we were not
clearing the EOFBLOCKS_FL when fill in the last uninitialized block,
but rather when we allocate the next block _after_ the uninitalized
block. Secondly we were not testing to see if we needed to clear the
EOFBLOCKS_FL when writing to the file O_DIRECT or when were converting
an uninitialized block (which is the most common case).
Google-Bug-Id: 2928259
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
* 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: (40 commits)
ext4: Adding error check after calling ext4_mb_regular_allocator()
ext4: Fix dirtying of journalled buffers in data=journal mode
ext4: re-inline ext4_rec_len_(to|from)_disk functions
jbd2: Remove t_handle_lock from start_this_handle()
jbd2: Change j_state_lock to be a rwlock_t
jbd2: Use atomic variables to avoid taking t_handle_lock in jbd2_journal_stop
ext4: Add mount options in superblock
ext4: force block allocation on quota_off
ext4: fix freeze deadlock under IO
ext4: drop inode from orphan list if ext4_delete_inode() fails
ext4: check to make make sure bd_dev is set before dereferencing it
jbd2: Make barrier messages less scary
ext4: don't print scary messages for allocation failures post-abort
ext4: fix EFBIG edge case when writing to large non-extent file
ext4: fix ext4_get_blocks references
ext4: Always journal quota file modifications
ext4: Fix potential memory leak in ext4_fill_super
ext4: Don't error out the fs if the user tries to make a file too big
ext4: allocate stripe-multiple IOs on stripe boundaries
ext4: move aio completion after unwritten extent conversion
...
Fix up conflicts in fs/ext4/inode.c as per Ted.
Fix up xfs conflicts as per earlier xfs merge.
No real bugs found, just removed some dead code.
Found by gcc 4.6's new warnings.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If i_data_sem was internally dropped due to transaction restart, it is
necessary to restart path look-up because extents tree was possibly
modified by ext4_get_block().
https://bugzilla.kernel.org/show_bug.cgi?id=15827
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Acked-by: Jan Kara <jack@suse.cz>
Dimitry Monakhov discovered an edge case where it was possible for the
EXT4_EOFBLOCKS_FL flag could get cleared unnecessarily. This is true;
I have a test case that can be exercised via downloading and
decompressing the file:
wget ftp://ftp.kernel.org/pub/linux/kernel/people/tytso/ext4-testcases/eofblocks-fl-test-case.img.bz2
bunzip2 eofblocks-fl-test-case.img
dd if=/dev/zero of=eofblocks-fl-test-case.img bs=1k seek=17925 bs=1k count=1 conv=notrunc
However, triggering it in real life is highly unlikely since it
requires an extremely fragmented sparse file with a hole in exactly
the right place in the extent tree. (It actually took quite a bit of
work to generate this test case.) Still, it's nice to get even
extreme corner cases to be correct, so this patch makes sure that we
don't clear the EXT4_EOFBLOCKS_FL incorrectly even in this corner
case.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If the EOFBLOCK_FL flag is set when it should not be and the inode is
zero length, then eh_entries is zero, and ex is NULL, so dereferencing
ex to print ex->ee_block causes a kernel OOPS in
ext4_ext_map_blocks().
On top of that, the error message which is printed isn't very helpful.
So we fix this by printing something more explanatory which doesn't
involve trying to print ex->ee_block.
Addresses-Google-Bug: #2655740
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
At several places we modify EXT4_I(inode)->i_flags without holding
i_mutex (ext4_do_update_inode, ...). These modifications are racy and
we can lose updates to i_flags. So convert handling of i_flags to use
bitops which are atomic.
https://bugzilla.kernel.org/show_bug.cgi?id=15792
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
EXT4_ERROR_INODE() tends to provide better error information and in a
more consistent format. Some errors were not even identifying the inode
or directory which was corrupted, which made them not very useful.
Addresses-Google-Bug: #2507977
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This saves a huge amount of stack space by avoiding unnecesary struct
buffer_head's from being allocated on the stack.
In addition, to make the code easier to understand, collapse and
refactor ext4_get_block(), ext4_get_block_write(),
noalloc_get_block_write(), into a single function.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Jack up ext4_get_blocks() and add a new function, ext4_map_blocks()
which uses a much smaller structure, struct ext4_map_blocks which is
20 bytes, as opposed to a struct buffer_head, which nearly 5 times
bigger on an x86_64 machine. By switching things to use
ext4_map_blocks(), we can save stack space by using ext4_map_blocks()
since we can avoid allocating a struct buffer_head on the stack.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Currently using posix_fallocate one can bypass an RLIMIT_FSIZE limit
and create a file larger than the limit. Add a check for that.
Signed-off-by: Nikanth Karthikesan <knikanth@suse.de>
Signed-off-by: Amit Arora <aarora@in.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The extents code will sometimes zero out blocks and mark them as
initialized instead of splitting an extent into several smaller ones.
This optimization however, causes problems if the extent is beyond
i_size because fsck will complain if there are uninitialized blocks
after i_size as this can not be distinguished from an inode that has
an incorrect i_size field.
https://bugzilla.kernel.org/show_bug.cgi?id=15742
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
