Hi Jan,
On 09/26/2013 10:53 PM, Jan Kara wrote:
Hello,
On Thu 26-09-13 21:32:07, Maxim Patlasov wrote:
While handling punch-hole fallocate, it's useless to truncate page cache
before removing the range from extent tree (or block map in indirect case)
because page cache can be re-populated (by read-ahead or read(2) or mmap-ed
read) immediately after truncating page cache, but before updating extent
tree (or block map). In that case the user will see stale data even after
fallocate is completed.
Yes, this is a known problem. The trouble is there isn't a reliable fix
currently possible. If we don't truncate page cache before removing blocks,
we will have pages in memory being backed by already freed blocks - not
good as that can lead to data corruption. So you should't really remove the
truncation from before we remove the blocks.
I'd like to understand the problem better. Could you please provide any
details about that data corruption? And if it was already discussed
somewhere, please point me to there.
You are right that if punch hole races with page fault or read, we can
create again pages with block mapping which will become stale soon and the
same problem as I wrote above applies. Truncating pagecache after we
removed blocks only narrows the race window but doesn't really fix the
problem.
There seems to be two different problems: 1) pages backed by already
freed blocks; 2) keeping page-cache populated by pages with stale data
after fallocate completes. Your concerns refer to the first problem. My
patch was intended to resolve the second. It seems to me that my patch
really fixes the second problem and it doesn't make things worse w.r.t.
the fisrt problem. Do I miss something?
Thanks,
Maxim
Properly fixing the problem requires significant overhaul in how mmap_sem
is used in page fault. I'm working on patches to do that but it will take
some time.
Honza
Signed-off-by: Maxim Patlasov <mpatlasov@xxxxxxxxxxxxx>
---
fs/ext4/inode.c | 17 +++++++++--------
1 file changed, 9 insertions(+), 8 deletions(-)
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 0d424d7..6b71116 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -3564,14 +3564,6 @@ int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
}
- first_block_offset = round_up(offset, sb->s_blocksize);
- last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
-
- /* Now release the pages and zero block aligned part of pages*/
- if (last_block_offset > first_block_offset)
- truncate_pagecache_range(inode, first_block_offset,
- last_block_offset);
-
/* Wait all existing dio workers, newcomers will block on i_mutex */
ext4_inode_block_unlocked_dio(inode);
inode_dio_wait(inode);
@@ -3621,6 +3613,15 @@ int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
up_write(&EXT4_I(inode)->i_data_sem);
if (IS_SYNC(inode))
ext4_handle_sync(handle);
+
+ first_block_offset = round_up(offset, sb->s_blocksize);
+ last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
+
+ /* Now release the pages and zero block aligned part of pages */
+ if (last_block_offset > first_block_offset)
+ truncate_pagecache_range(inode, first_block_offset,
+ last_block_offset);
+
inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
ext4_mark_inode_dirty(handle, inode);
out_stop:
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
