ext4: fix reservation release on invalidatepage for delalloc fs
authorLukas Czerner <lczerner@redhat.com>
Sat, 4 Jul 2015 01:13:55 +0000 (21:13 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Sat, 4 Jul 2015 01:13:55 +0000 (21:13 -0400)
On delalloc enabled file system on invalidatepage operation
in ext4_da_page_release_reservation() we want to clear the delayed
buffer and remove the extent covering the delayed buffer from the extent
status tree.

However currently there is a bug where on the systems with page size >
block size we will always remove extents from the start of the page
regardless where the actual delayed buffers are positioned in the page.
This leads to the errors like this:

EXT4-fs warning (device loop0): ext4_da_release_space:1225:
ext4_da_release_space: ino 13, to_free 1 with only 0 reserved data
blocks

This however can cause data loss on writeback time if the file system is
in ENOSPC condition because we're releasing reservation for someones
else delayed buffer.

Fix this by only removing extents that corresponds to the part of the
page we want to invalidate.

This problem is reproducible by the following fio receipt (however I was
only able to reproduce it with fio-2.1 or older.

[global]
bs=8k
iodepth=1024
iodepth_batch=60
randrepeat=1
size=1m
directory=/mnt/test
numjobs=20
[job1]
ioengine=sync
bs=1k
direct=1
rw=randread
filename=file1:file2
[job2]
ioengine=libaio
rw=randwrite
direct=1
filename=file1:file2
[job3]
bs=1k
ioengine=posixaio
rw=randwrite
direct=1
filename=file1:file2
[job5]
bs=1k
ioengine=sync
rw=randread
filename=file1:file2
[job7]
ioengine=libaio
rw=randwrite
filename=file1:file2
[job8]
ioengine=posixaio
rw=randwrite
filename=file1:file2
[job10]
ioengine=mmap
rw=randwrite
bs=1k
filename=file1:file2
[job11]
ioengine=mmap
rw=randwrite
direct=1
filename=file1:file2

Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: stable@vger.kernel.org
fs/ext4/inode.c

index 4ad73d3..2334e86 100644 (file)
@@ -1330,7 +1330,7 @@ static void ext4_da_page_release_reservation(struct page *page,
                                             unsigned int offset,
                                             unsigned int length)
 {
-       int to_release = 0;
+       int to_release = 0, contiguous_blks = 0;
        struct buffer_head *head, *bh;
        unsigned int curr_off = 0;
        struct inode *inode = page->mapping->host;
@@ -1351,14 +1351,23 @@ static void ext4_da_page_release_reservation(struct page *page,
 
                if ((offset <= curr_off) && (buffer_delay(bh))) {
                        to_release++;
+                       contiguous_blks++;
                        clear_buffer_delay(bh);
+               } else if (contiguous_blks) {
+                       lblk = page->index <<
+                              (PAGE_CACHE_SHIFT - inode->i_blkbits);
+                       lblk += (curr_off >> inode->i_blkbits) -
+                               contiguous_blks;
+                       ext4_es_remove_extent(inode, lblk, contiguous_blks);
+                       contiguous_blks = 0;
                }
                curr_off = next_off;
        } while ((bh = bh->b_this_page) != head);
 
-       if (to_release) {
+       if (contiguous_blks) {
                lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
-               ext4_es_remove_extent(inode, lblk, to_release);
+               lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
+               ext4_es_remove_extent(inode, lblk, contiguous_blks);
        }
 
        /* If we have released all the blocks belonging to a cluster, then we