From: Yongqiang Yang <xiaoqiangnk@gmail.com>
Subject: Re: [Ext4 punch hole 4/5] Ext4 Punch Hole Support: Enable Punch Hole
Date: Wed, 2 Mar 2011 10:34:10 +0800
Message-ID: <AANLkTinz2d=OqjDez=NZEfZ-eEQe-kMq-ytyzp4kJ3kL@mail.gmail.com>
References: <4D6C634C.8050308@linux.vnet.ibm.com>
	<1299030568.6761.678.camel@mingming-laptop>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: QUOTED-PRINTABLE
Cc: Allison Henderson <achender@linux.vnet.ibm.com>,
	linux-ext4@vger.kernel.org
To: Mingming Cao <cmm@us.ibm.com>
Return-path: <linux-ext4-owner@vger.kernel.org>
Received: from mail-iy0-f174.google.com ([209.85.210.174]:35668 "EHLO
	mail-iy0-f174.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1754185Ab1CBCeL convert rfc822-to-8bit (ORCPT
	<rfc822;linux-ext4@vger.kernel.org>); Tue, 1 Mar 2011 21:34:11 -0500
Received: by iyb26 with SMTP id 26so4506581iyb.19
        for <linux-ext4@vger.kernel.org>; Tue, 01 Mar 2011 18:34:11 -0800 (PST)
In-Reply-To: <1299030568.6761.678.camel@mingming-laptop>
Sender: linux-ext4-owner@vger.kernel.org
List-ID: <linux-ext4.vger.kernel.org>

On Wed, Mar 2, 2011 at 9:49 AM, Mingming Cao <cmm@us.ibm.com> wrote:
> On Mon, 2011-02-28 at 20:09 -0700, Allison Henderson wrote:
>> This patch adds the =A0new "ext4_punch_hole" "ext4_ext_punch_hole" r=
outines.
>>
>> fallocate has been modified to call ext4_punch_hole when the punch h=
ole
>> flag is passed. =A0At the moment, we only support punching holes in
>> extents, so this routine is pretty much a wrapper for the ext4_ext_p=
unch_hole
>> routine.
>>
>> The ext4_ext_punch_hole routine zeros out the pages that are
>> covered by the hole. =A0The blocks to be punched out
>> are then identified as mapped, delayed, or already punched out.
>> The blocks that mapped are the converted to into uninitialized
>> extents. =A0The blocks are then punched out using the
>> "ext4_ext_release_blocks" routine.
>>
>
> All right, I mainly looked at the punch hole over a hole or delayed
> allocation handling part...so my comments below...
>
>> Some minor utility functions have also been added.
>> A new ext4_ext_lookup_hole routine is used by
>> ext4_ext_punch_hole to check if a range of blocks
>> have already been punched out.
>>
>> A new ext4_ext_test_block_flag has also been
>> added to identify the state of a block (ie mapped,
>> delayed, ect)
>>
>> Signed-off-by: Allison Henderson <achender@us.ibm.com>
>> ---
>> :100644 100644 43a5772... aeb86d6... M =A0 =A0 =A0 =A0fs/ext4/ext4.h
>> :100644 100644 efbc3ef... 5713258... M =A0 =A0 =A0 =A0fs/ext4/extent=
s.c
>> :100644 100644 28c9137... 493c908... M =A0 =A0 =A0 =A0fs/ext4/inode.=
c
>> =A0fs/ext4/ext4.h =A0 =A0| =A0 =A02 +
>> =A0fs/ext4/extents.c | =A0321 ++++++++++++++++++++++++++++++++++++++=
++++++++++++++-
>> =A0fs/ext4/inode.c =A0 | =A0 26 +++++
>> =A03 files changed, 345 insertions(+), 4 deletions(-)
>>
>> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
>> index 43a5772..aeb86d6 100644
>> --- a/fs/ext4/ext4.h
>> +++ b/fs/ext4/ext4.h
>> @@ -1729,6 +1729,7 @@ extern int ext4_change_inode_journal_flag(stru=
ct inode *, int);
>> =A0extern int ext4_get_inode_loc(struct inode *, struct ext4_iloc *)=
;
>> =A0extern int ext4_can_truncate(struct inode *inode);
>> =A0extern void ext4_truncate(struct inode *);
>> +extern long =A0ext4_punch_hole(struct inode *inode,loff_t offset, l=
off_t length);
>> =A0extern int ext4_truncate_restart_trans(handle_t *, struct inode *=
, int nblocks);
>> =A0extern void ext4_set_inode_flags(struct inode *);
>> =A0extern void ext4_get_inode_flags(struct ext4_inode_info *);
>> @@ -2066,6 +2067,7 @@ extern int ext4_ext_index_trans_blocks(struct =
inode *inode, int nrblocks,
>> =A0extern int ext4_ext_map_blocks(handle_t *handle, struct inode *in=
ode,
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0struct ex=
t4_map_blocks *map, int flags);
>> =A0extern void ext4_ext_truncate(struct inode *);
>> +extern void ext4_ext_punch_hole(struct inode *inode, loff_t offset,=
 loff_t length);
>> =A0extern void ext4_ext_init(struct super_block *);
>> =A0extern void ext4_ext_release(struct super_block *);
>> =A0extern long ext4_fallocate(struct file *file, int mode, loff_t of=
fset,
>> diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
>> index efbc3ef..5713258 100644
>> --- a/fs/ext4/extents.c
>> +++ b/fs/ext4/extents.c
>> @@ -2776,6 +2776,154 @@ out:
>> =A0}
>>
>> =A0/*
>> + * lookup_hole()
>> + * Returns the numbers of consecutive blocks starting at "start"
>> + * that are not contained within an extent
>> + */
>
> The lookup hole path, IMHO, could be a special flag pass to
> ext4_map_blocks(), reuse existing code, rather adding a new function
> directly inspecting the inode's allocation tree from there.:)
>
>> +static int ext4_ext_lookup_hole(struct inode *inode, ext4_lblk_t st=
art){
>> + =A0 =A0struct super_block *sb =3D inode->i_sb;
>> + =A0 =A0 int depth =3D ext_depth(inode);
>> + =A0 =A0 struct ext4_ext_path *path;
>> + =A0 =A0 struct ext4_extent_header *eh;
>> + =A0 =A0 struct ext4_extent *ex;
>> + =A0 =A0 struct buffer_head *bh;
>> + =A0 =A0 ext4_lblk_t last_block;
>> + =A0 =A0 handle_t *handle;
>> + =A0 =A0 int i, err;
>> +
>> + =A0 =A0 ext_debug("lookup hole since %u\n", start);
>> +
>> + =A0 =A0 /* Make sure start is valid */
>> + =A0 =A0 last_block =3D inode->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
>> + =A0 =A0 if(start >=3D last_block)
>> + =A0 =A0 =A0 =A0 =A0 =A0 return -EIO;
>> +
>> + =A0 =A0 handle =3D ext4_journal_start(inode, depth + 1);
>> + =A0 =A0 if (IS_ERR(handle))
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 return PTR_ERR(handle);
>> +
>> + =A0 =A0 /*
>> + =A0 =A0 =A0* We start scanning from right side, looking for
>> + =A0 =A0 =A0* the left most block contained in the leaf, and
>> + =A0 =A0 =A0* stopping when "start" is crossed.
>> + =A0 =A0 =A0*/
>> + =A0 =A0 depth =3D ext_depth(inode);
>> + =A0 =A0 path =3D kzalloc(sizeof(struct ext4_ext_path) * (depth + 1=
), GFP_NOFS);
>> + =A0 =A0 if (path =3D=3D NULL) {
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 ext4_journal_stop(handle);
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 return -ENOMEM;
>> + =A0 =A0 }
>> + =A0 =A0 path[0].p_depth =3D depth;
>> + =A0 =A0 path[0].p_hdr =3D ext_inode_hdr(inode);
>> + =A0 =A0 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
>> + =A0 =A0 =A0 =A0 =A0 =A0 err =3D -EIO;
>> + =A0 =A0 =A0 =A0 =A0 =A0 goto out;
>> + =A0 =A0 }
>> + =A0 =A0 i =3D err =3D 0;
>> +
>> + =A0 =A0 while (i >=3D 0 && err =3D=3D 0) {
>> + =A0 =A0 =A0 =A0 =A0 =A0 if (i =3D=3D depth) {
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* this is leaf block */
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 eh =3D path[i].p_hdr;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 if (eh !=3D NULL){
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 if (eh->eh=
_entries =3D=3D 0){
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 err =3D -EIO;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 goto out;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 }
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 ex =3D EXT=
_LAST_EXTENT(eh);
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 while (ex =
!=3D NULL && ex >=3D EXT_FIRST_EXTENT(eh)){
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 /*
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0* If the entire extent apears before start
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0* then we have passed the hole.
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0*/
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 if(ex->ee_block + ex->ee_len <=3D start)
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0 =A0 =A0 =A0 goto out;
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 /*
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0* If the start of the extent appears after
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0* or on start, then mark this as the edge
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0* of the hole
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0*/
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 if(ex->ee_block >=3D start)
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0 =A0 =A0 =A0 last_block =3D ex->ee_block;
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 /*
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0* If the extent contains start, then there
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0* is no hole.
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0*/
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 else if(ex->ee_block + ex->ee_len > start){
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0 =A0 =A0 =A0 last_block =3D start;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0 =A0 =A0 =A0 goto out;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 }
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 ex--;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 }
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 }
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* root level has p_bh =3D=
=3D NULL, brelse() eats this */
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 brelse(path[i].p_bh);
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 path[i].p_bh =3D NULL;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 i--;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 continue;
>> + =A0 =A0 =A0 =A0 =A0 =A0 }
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 /* this is index block */
>> + =A0 =A0 =A0 =A0 =A0 =A0 if (!path[i].p_hdr)
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 path[i].p_hdr =3D ext_bloc=
k_hdr(path[i].p_bh);
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 if (!path[i].p_idx) {
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* this level hasn't been =
touched yet */
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 path[i].p_idx =3D EXT_LAST=
_INDEX(path[i].p_hdr);
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 path[i].p_block =3D le16_t=
o_cpu(path[i].p_hdr->eh_entries)+1;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 ext_debug("init index ptr:=
 hdr 0x%p, num %d\n",
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 path[i].p_hdr,
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 le16_to_cpu(path[i].p_hdr-=
>eh_entries));
>> + =A0 =A0 =A0 =A0 =A0 =A0 }
>> + =A0 =A0 =A0 =A0 =A0 =A0 else {
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* we were already here, s=
ee at next index */
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 path[i].p_idx--;
>> + =A0 =A0 =A0 =A0 =A0 =A0 }
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 ext_debug("level %d - index, first 0x%p, c=
ur 0x%p\n",
>> + =A0 =A0 =A0 =A0 =A0 =A0 i, EXT_FIRST_INDEX(path[i].p_hdr),
>> + =A0 =A0 =A0 =A0 =A0 =A0 path[i].p_idx);
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 /* go to the next level */
>> + =A0 =A0 =A0 =A0 =A0 =A0 ext_debug("move to level %d (block %llu)\n=
",
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 i + 1, ext=
4_idx_pblock(path[i].p_idx));
>> + =A0 =A0 =A0 =A0 =A0 =A0 memset(path + i + 1, 0, sizeof(*path));
>> + =A0 =A0 =A0 =A0 =A0 =A0 bh =3D sb_bread(sb, ext4_idx_pblock(path[i=
].p_idx));
>> + =A0 =A0 =A0 =A0 =A0 =A0 if (!bh) {
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 err =3D -EIO;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 break;
>> + =A0 =A0 =A0 =A0 =A0 =A0 }
>> + =A0 =A0 =A0 =A0 =A0 =A0 if (WARN_ON(i + 1 > depth)) {
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 err =3D -EIO;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 break;
>> + =A0 =A0 =A0 =A0 =A0 =A0 }
>> + =A0 =A0 =A0 =A0 =A0 =A0 if (ext4_ext_check(inode, ext_block_hdr(bh=
), depth - i - 1)) {
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 err =3D -EIO;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 break;
>> + =A0 =A0 =A0 =A0 =A0 =A0 }
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 path[i + 1].p_bh =3D bh;
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 i++;
>> +
>> + =A0 =A0 }
>> +out:
>> + =A0 =A0 ext4_ext_drop_refs(path);
>> + =A0 =A0 kfree(path);
>> + =A0 =A0 ext4_journal_stop(handle);
>> +
>> + =A0 =A0 return err ? err : last_block - start;
>> +
>> +}
>> +
>> +/*
>> =A0 * called at mount time
>> =A0 */
>> =A0void ext4_ext_init(struct super_block *sb)
>> @@ -4029,6 +4177,172 @@ next:
>> =A0 =A0 =A0 return ret;
>> =A0}
>>
>> +/*
>> + * ext4_ext_test_block_flag
>> + * Tests the buffer head associated with the given block
>> + * to see if the state contains flag
>> + *
>> + * @inode: =A0The inode of the given file
>> + * @block: =A0The block to test
>> + * @flag: =A0 The flag to check for
>> + *
>> + * Returns 0 on sucess or negative on err
>> + */
>> +static int ext4_ext_test_block_flag(struct inode *inode, ext4_lblk_=
t block, enum bh_state_bits flag){
>> + =A0 =A0 struct buffer_head *bh;
>> + =A0 =A0 struct page *page;
>> + =A0 =A0 struct address_space *mapping =3D inode->i_mapping;
>> + =A0 =A0 loff_t block_offset;
>> + =A0 =A0 int i, ret;
>> + =A0 =A0 unsigned long flag_mask =3D 1 << flag;
>> +
>> + =A0 =A0 block_offset =3D block << EXT4_BLOCK_SIZE_BITS(inode->i_sb=
);
>> + =A0 =A0 page =3D find_or_create_page(mapping, block_offset >> PAGE=
_CACHE_SHIFT,
>> + =A0 =A0 mapping_gfp_mask(mapping) & ~__GFP_FS);
>> +
>> + =A0 =A0 if (!page)
>> + =A0 =A0 =A0 =A0 =A0 =A0 return -EIO;
>> +
>> + =A0 =A0 if (!page_has_buffers(page))
>> + =A0 =A0 =A0 =A0 =A0 =A0 create_empty_buffers(page, EXT4_BLOCK_SIZE=
(inode->i_sb), 0);
>> +
>> + =A0 =A0 /* advance to the buffer that has the block offset =A0*/
>> + =A0 =A0 bh =3D page_buffers(page);
>> + =A0 =A0 for (i =3D 0; i < block_offset; i+=3DEXT4_BLOCK_SIZE(inode=
->i_sb)) {
>> + =A0 =A0 =A0 =A0 =A0 =A0 bh =3D bh->b_this_page;
>> + =A0 =A0 }
>> +
>> + =A0 =A0 if(bh->b_state & flag_mask)
>> + =A0 =A0 =A0 =A0 =A0 =A0 ret =3D 0;
>> + =A0 =A0 else
>> + =A0 =A0 =A0 =A0 =A0 =A0 ret =3D -1;
>> +
>> + =A0 =A0 unlock_page(page);
>> + =A0 =A0 page_cache_release(page);
>> +
>> + =A0 =A0 return ret;
>> +
>> +}
>> +
>> +/*
>> + * ext4_ext_punch_hole
>> + *
>> + * Punches a hole of "length" bytes in a file starting
>> + * at byte "offset"
>> + *
>> + * @inode: =A0The inode of the file to punch a hole in
>> + * @offset: The starting byte offset of the hole
>> + * @length: The length of the hole
>> + *
>> + */
>> +void ext4_ext_punch_hole(struct inode *inode, loff_t offset, loff_t=
 length)
>> +{
>> + =A0 =A0 struct super_block *sb =3D inode->i_sb;
>> + =A0 =A0 ext4_lblk_t first_block, last_block, num_blocks, iblock =3D=
 0;
>> + =A0 =A0 struct address_space *mapping =3D inode->i_mapping;
>> + =A0 =A0 struct ext4_map_blocks map;
>> + =A0 =A0 handle_t *handle;
>> + =A0 =A0 loff_t first_block_offset, last_block_offset, block_len;
>> + =A0 =A0 int get_blocks_flags, err, ret =3D 0;
>> +
>> + =A0 =A0 first_block =3D (offset + sb->s_blocksize - 1)
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 >> EXT4_BLOCK_SIZE_BITS(sb=
);
>> + =A0 =A0 last_block =3D (offset+length) =A0>> EXT4_BLOCK_SIZE_BITS(=
sb);
>> +
>> + =A0 =A0 first_block_offset =3D first_block << EXT4_BLOCK_SIZE_BITS=
(sb);
>> + =A0 =A0 last_block_offset =3D last_block << EXT4_BLOCK_SIZE_BITS(s=
b);
>> +
>> + =A0 =A0 err =3D ext4_writepage_trans_blocks(inode);
>> + =A0 =A0 handle =3D ext4_journal_start(inode, err);
>> + =A0 =A0 if (IS_ERR(handle))
>> + =A0 =A0 =A0 =A0 =A0 =A0 return;
>> +
>> + =A0 =A0 /*
>> + =A0 =A0 =A0* Now we need to zero out the un block aligned data.
>> + =A0 =A0 =A0* If the file is smaller than a block, just
>> + =A0 =A0 =A0* zero out the middle and return
>> + =A0 =A0 =A0*/
>> + =A0 =A0 if(first_block > last_block)
>> + =A0 =A0 =A0 =A0 =A0 =A0 ext4_block_zero_page_range(handle, mapping=
, offset, length);
>> + =A0 =A0 else{
>> + =A0 =A0 =A0 =A0 =A0 =A0 /* zero out the head of the hole before th=
e first block */
>> + =A0 =A0 =A0 =A0 =A0 =A0 block_len =A0=3D first_block_offset - offs=
et;
>> + =A0 =A0 =A0 =A0 =A0 =A0 if(block_len > 0)
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 ext4_block_zero_page_range=
(handle, mapping, offset, block_len);
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 /* zero out the tail of the hole after the=
 last block */
>> + =A0 =A0 =A0 =A0 =A0 =A0 block_len =3D offset + length - last_block=
_offset;
>> + =A0 =A0 =A0 =A0 =A0 =A0 if(block_len > 0)
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 ext4_block_zero_page_range=
(handle, mapping,
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 last_block_offset, block_len);
>> + =A0 =A0 }
>> +
>> + =A0 =A0 /* If there are no blocks to remove, return now */
>> + =A0 =A0 if(first_block >=3D last_block){
>> + =A0 =A0 =A0 =A0 =A0 =A0 ext4_journal_stop(handle);
>> + =A0 =A0 =A0 =A0 =A0 =A0 return;
>> + =A0 =A0 }
>> +
>> + =A0 =A0 /* Clear pages associated with the hole */
>> + =A0 =A0 if (mapping->nrpages)
>> + =A0 =A0 =A0 =A0 =A0 =A0 invalidate_inode_pages2_range(mapping, off=
set >> PAGE_CACHE_SHIFT,
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=
 =A0 =A0 =A0 =A0 =A0 (offset+length) >> PAGE_CACHE_SHIFT );
>> +
>> +
>> + =A0 =A0 /* Loop over all the blocks and identify blocks that need =
to be punched out */
>> + =A0 =A0 iblock =3D first_block;
>> + =A0 =A0 while(iblock < last_block){
>> + =A0 =A0 =A0 =A0 =A0 =A0 map.m_lblk =3D iblock;
>> + =A0 =A0 =A0 =A0 =A0 =A0 map.m_len =3D last_block - iblock;
>> + =A0 =A0 =A0 =A0 =A0 =A0 ret =3D ext4_map_blocks(handle, inode, &ma=
p, 0);
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 /* If the blocks are mapped, release them =
*/
>> + =A0 =A0 =A0 =A0 =A0 =A0 if(ret > 0){
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 num_blocks =3D ret;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 ext4_ext_convert_blocks_un=
init(inode, handle, iblock, num_blocks);
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 ext4_ext_release_blocks(in=
ode, iblock, iblock+num_blocks);
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 goto next;
>> + =A0 =A0 =A0 =A0 =A0 =A0 }
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 /*
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0* If they are not mapped
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0* check to see if they are punched out
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0*/
>> + =A0 =A0 =A0 =A0 =A0 =A0 ret =3D ext4_ext_lookup_hole(inode, iblock=
);
>> + =A0 =A0 =A0 =A0 =A0 =A0 if(ret > 0){
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 num_blocks =3D ret;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 goto next;
>> + =A0 =A0 =A0 =A0 =A0 =A0 }
>> +
>
> I am wondering how ext4 FIEMAP handles hole lookup more efficently?
In ext4 FIEMAP, ext4_ext_walk_space() lookup requested block in extent
tree firstly, and look next  allocated block in extent-tree secondly,
so if the block is not contained in the found extent, then lookup
dirty pages starting from offset of the block in pagecahe.  Next, find
1st mapped block in the found pages, if the 1st mapped block is not
delayed and its block nr is less than or equal to the next allocated
block, then a hole is found.

To lookup a hole, just do as follows.
1. lookup block in the extent tree. if the found extent contains the
request block, then no hole. otherwise, goto 2.
2. lookup the next allocated block.
3. lookup dirty pages in pagecache starting from offset of the block,
then find the 1st mapped block. there are 3 cases.
   a. block number of 1st mapped block is greater than or equal to the
next allocated block, then a hole is found.
   b. block number of 1st mapped block is less than the next allocated =
block,
  check delayed flag, a delayed extent is found.

>
>> + =A0 =A0 =A0 =A0 =A0 =A0 /*
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0* If the block could not be mapped, and
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0* its not already punched out,
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0* check to see if the block is delayed
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0*/
>> + =A0 =A0 =A0 =A0 =A0 =A0 if(ext4_ext_test_block_flag(inode, iblock,=
 BH_Delay) =3D=3D 0){
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 get_blocks_flags =3D EXT4_=
GET_BLOCKS_CREATE | EXT4_GET_BLOCKS_DELALLOC_RESERVE;
>
> Ah... the flags, could you check it again? We might get this wrong.
>
> EXT4_GET_BLOCKS_CREATE | EXT4_GET_BLOCKS_DELALLOC_RESERVE?
>
> these combination means we are plan to do block allocation via delaye=
d
> allocation path. From inode.c. this flag is aim to tell block allocat=
ion
> to takes care of block reservation/release for delayed allocation pat=
ch.
>
>
> we should at least turn off the create flag, and check if
> EXT4_GET_BLOCKS_DELALLOC_RESERVE is also used for delayed extents loo=
k
> up also? Maybe I missed something.
>
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 ret =3D ext4_map_blocks(ha=
ndle, inode, &map, get_blocks_flags);
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* If the blocks are found=
, release them */
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 if(ret > 0){
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 num_blocks=
 =3D ret;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 ext4_ext_r=
elease_blocks(inode, iblock, iblock+num_blocks);
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 goto next;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 }
>
> ext4_ext_release_blocks() is freeing up real storage on disk. For
> delayed allocation case, there are no blocks allocated yet. We should
> call ext4_da_release_space() or similar to free up the blocks reserve=
d
> by delayed allocation.
>
>> + =A0 =A0 =A0 =A0 =A0 =A0 }
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 /* If the block cannot be identified, just=
 skip it */
>> + =A0 =A0 =A0 =A0 =A0 =A0 num_blocks =3D 1;
>> +
>> +next:
>> + =A0 =A0 =A0 =A0 =A0 =A0 iblock+=3Dnum_blocks;
>> + =A0 =A0 }
>> + =A0 =A0 ext4_mark_inode_dirty(handle, inode);
>> +
>> + =A0 =A0 ext4_journal_stop(handle);
>> +
>> +}
>> +
>>
>> =A0static void ext4_falloc_update_inode(struct inode *inode,
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 int mode=
, loff_t new_size, int update_ctime)
>> @@ -4079,10 +4393,6 @@ long ext4_fallocate(struct file *file, int mo=
de, loff_t offset, loff_t len)
>> =A0 =A0 =A0 struct ext4_map_blocks map;
>> =A0 =A0 =A0 unsigned int credits, blkbits =3D inode->i_blkbits;
>>
>> - =A0 =A0 /* We only support the FALLOC_FL_KEEP_SIZE mode */
>> - =A0 =A0 if (mode & ~FALLOC_FL_KEEP_SIZE)
>> - =A0 =A0 =A0 =A0 =A0 =A0 return -EOPNOTSUPP;
>> -
>> =A0 =A0 =A0 /*
>> =A0 =A0 =A0 =A0* currently supporting (pre)allocate mode for extent-=
based
>> =A0 =A0 =A0 =A0* files _only_
>> @@ -4090,6 +4400,9 @@ long ext4_fallocate(struct file *file, int mod=
e, loff_t offset, loff_t len)
>> =A0 =A0 =A0 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 return -EOPNOTSUPP;
>>
>> + =A0 =A0 if (mode & FALLOC_FL_PUNCH_HOLE)
>> + =A0 =A0 =A0 =A0 =A0 =A0 return ext4_punch_hole(inode, offset, len)=
;
>> +
>
> so for other than the three existing mode, we should also return
> EOPNOTSUPP too, isn't?
>
>> =A0 =A0 =A0 map.m_lblk =3D offset >> blkbits;
>> =A0 =A0 =A0 /*
>> =A0 =A0 =A0 =A0* We can't just convert len to max_blocks because
>> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
>> index 28c9137..493c908 100644
>> --- a/fs/ext4/inode.c
>> +++ b/fs/ext4/inode.c
>> @@ -4487,6 +4487,32 @@ int ext4_can_truncate(struct inode *inode)
>> =A0}
>>
>> =A0/*
>> + * ext4_punch_hole: punches a hole in a file by releaseing the bloc=
ks
>> + * associated with the given offset and length
>> + *
>> + * @inode: =A0File inode
>> + * @offset: The offset where the hole will begin
>> + * @len: =A0 =A0The length of the hole
>> + *
>> + * Returns: 0 on sucess or negative on failure
>> + */
>> +
>> +long =A0ext4_punch_hole(struct inode *inode, loff_t offset, loff_t =
length)
>> +{
>> +
>> + =A0 =A0 if (!S_ISREG(inode->i_mode)=3D=3D1)
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0return -ENOTSUPP;
>> +
>> + =A0 =A0 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
>> + =A0 =A0 =A0 =A0 =A0 =A0 //TODO: Add support for non extent hole pu=
nching
>> + =A0 =A0 =A0 =A0 =A0 =A0 return -ENOTSUPP;
>> + =A0 =A0 }
>> +
>> + =A0 =A0 ext4_ext_punch_hole(inode, offset, length);
>> + =A0 =A0 return 0;
>> +}
>> +
>> +/*
>> =A0 * ext4_truncate()
>> =A0 *
>> =A0 * We block out ext4_get_block() block instantiations across the =
entire
>
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4"=
 in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at =A0http://vger.kernel.org/majordomo-info.html
>



--=20
Best Wishes
Yongqiang Yang
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" i=
n
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html