Message-ID: <20150115122906.GC18876@bfoster.bfoster>
Date:	Thu, 15 Jan 2015 07:29:06 -0500
From:	Brian Foster <bfoster@...hat.com>
To:	Namjae Jeon <linkinjeon@...il.com>
Cc:	david@...morbit.com, tytso@....edu, linux-fsdevel@...r.kernel.org,
	linux-kernel@...r.kernel.org, linux-ext4@...r.kernel.org,
	xfs@....sgi.com, a.sangwan@...sung.com,
	Namjae Jeon <namjae.jeon@...sung.com>
Subject: Re: [PATCH v8 2/11] xfs: Add support FALLOC_FL_INSERT_RANGE for
 fallocate

On Wed, Jan 14, 2015 at 01:05:17AM +0900, Namjae Jeon wrote:
> From: Namjae Jeon <namjae.jeon@...sung.com>
> 
> This patch implements fallocate's FALLOC_FL_INSERT_RANGE for XFS.
> 
> 1) Make sure that both offset and len are block size aligned.
> 2) Update the i_size of inode by len bytes.
> 3) Compute the file's logical block number against offset. If the computed
>    block number is not the starting block of the extent, split the extent
>    such that the block number is the starting block of the extent.
> 4) Shift all the extents which are lying bewteen [offset, last allocated extent]
>    towards right by len bytes. This step will make a hole of len bytes
>    at offset.
> 
> Signed-off-by: Namjae Jeon <namjae.jeon@...sung.com>
> Signed-off-by: Ashish Sangwan <a.sangwan@...sung.com>
> Cc: Brian Foster<bfoster@...hat.com>
> ---

Fixes look good (I assume nothing else changed between the few nits
called out in v7) and survives overnight fsstress and fsx testing
without any explosions:

Reviewed-by: Brian Foster <bfoster@...hat.com>

> Changelog
> v8:
>  - Fixed review comments from Brian(spelling nit, update comment, etc)
> 
> v7:
>  - Fixed review comments from Brian, important ones are: 
>    a) no need to check for adjacent extent merge
>    b) avoid unnecessary inode logging in case of error.
> 
> v6:
>  - This version is based upon Brian's changes to collapse paths.
>  - Instead of having seperate functions for shifting extents left/right, the
>    current extent shift function is made generic to shift in both directions.
> 
> v5:
>  - remove allocation part.
> 
> v4:
>  - set cur->bc_private.b.allocated to zero before calling xfs_btree_del_cursor.
> 
> v3:
>  - remove XFS_TRANS_RESERVE and assert.
>  - update the comment of blockcount calculation.
>  - use 'if(blockcount)' instead of 'if (got.br_blockcount < blockcount)'.
>  - move insert_file_space() calling under xfs_setattr_size to avoid code duplicate.
> 
> v2:
>  - remove reserved enable.
>  - add xfs_qm_dqattach.
>  - reset blockcount in xfs_bmap_shift_extents_right.
>  - update i_size to avoid data loss before insert_file_space() is called.
>  - use in-memory extent array size that delayed allocation extents
> 
>  fs/xfs/libxfs/xfs_bmap.c |  358 ++++++++++++++++++++++++++++++++++++++++------
>  fs/xfs/libxfs/xfs_bmap.h |   13 +-
>  fs/xfs/xfs_bmap_util.c   |  126 +++++++++++-----
>  fs/xfs/xfs_bmap_util.h   |    2 +
>  fs/xfs/xfs_file.c        |   41 +++++-
>  fs/xfs/xfs_trace.h       |    1 +
>  6 files changed, 457 insertions(+), 84 deletions(-)
> 
> diff --git a/fs/xfs/libxfs/xfs_bmap.c b/fs/xfs/libxfs/xfs_bmap.c
> index b5eb474..3fb2f10 100644
> --- a/fs/xfs/libxfs/xfs_bmap.c
> +++ b/fs/xfs/libxfs/xfs_bmap.c
> @@ -5516,50 +5516,86 @@ xfs_bmse_shift_one(
>  	int				*current_ext,
>  	struct xfs_bmbt_rec_host	*gotp,
>  	struct xfs_btree_cur		*cur,
> -	int				*logflags)
> +	int				*logflags,
> +	enum SHIFT_DIRECTION		SHIFT)
>  {
>  	struct xfs_ifork		*ifp;
>  	xfs_fileoff_t			startoff;
> -	struct xfs_bmbt_rec_host	*leftp;
> +	struct xfs_bmbt_rec_host	*contp;
>  	struct xfs_bmbt_irec		got;
> -	struct xfs_bmbt_irec		left;
> +	struct xfs_bmbt_irec		cont;
>  	int				error;
>  	int				i;
> +	int				total_extents;
>  
>  	ifp = XFS_IFORK_PTR(ip, whichfork);
> +	total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
>  
>  	xfs_bmbt_get_all(gotp, &got);
> -	startoff = got.br_startoff - offset_shift_fsb;
>  
>  	/* delalloc extents should be prevented by caller */
>  	XFS_WANT_CORRUPTED_RETURN(!isnullstartblock(got.br_startblock));
>  
> -	/*
> -	 * Check for merge if we've got an extent to the left, otherwise make
> -	 * sure there's enough room at the start of the file for the shift.
> -	 */
> -	if (*current_ext) {
> -		/* grab the left extent and check for a large enough hole */
> -		leftp = xfs_iext_get_ext(ifp, *current_ext - 1);
> -		xfs_bmbt_get_all(leftp, &left);
> +	if (SHIFT == SHIFT_LEFT) {
> +		startoff = got.br_startoff - offset_shift_fsb;
>  
> -		if (startoff < left.br_startoff + left.br_blockcount)
> +		/*
> +		 * Check for merge if we've got an extent to the left,
> +		 * otherwise make sure there's enough room at the start
> +		 * of the file for the shift.
> +		 */
> +		if (*current_ext) {
> +			/*
> +			 * grab the left extent and check for a large
> +			 * enough hole.
> +			 */
> +			contp = xfs_iext_get_ext(ifp, *current_ext - 1);
> +			xfs_bmbt_get_all(contp, &cont);
> +
> +			if (startoff < cont.br_startoff + cont.br_blockcount)
> +				return -EINVAL;
> +
> +			/* check whether to merge the extent or shift it down */
> +			if (xfs_bmse_can_merge(&cont, &got, offset_shift_fsb)) {
> +				return xfs_bmse_merge(ip, whichfork,
> +						      offset_shift_fsb,
> +						      *current_ext, gotp, contp,
> +						      cur, logflags);
> +			}
> +		} else if (got.br_startoff < offset_shift_fsb)
>  			return -EINVAL;
> +	} else {
> +		startoff = got.br_startoff + offset_shift_fsb;
> +		/*
> +		 * If this is not the last extent in the file, make sure there's
> +		 * enough room between current extent and next extent for
> +		 * accommodating the shift.
> +		 */
> +		if (*current_ext < (total_extents - 1)) {
> +			contp = xfs_iext_get_ext(ifp, *current_ext + 1);
> +			xfs_bmbt_get_all(contp, &cont);
> +			if (startoff + got.br_blockcount > cont.br_startoff)
> +				return -EINVAL;
>  
> -		/* check whether to merge the extent or shift it down */
> -		if (xfs_bmse_can_merge(&left, &got, offset_shift_fsb)) {
> -			return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
> -					      *current_ext, gotp, leftp, cur,
> -					      logflags);
> +			/*
> +			 * Unlike a left shift (which involves a hole punch),
> +			 * a right shift does not modify extent neighbors
> +			 * in any way. We should never find mergeable extents
> +			 * in this scenario. Check anyways and warn if we
> +			 * encounter two extents that could be one.
> +			 */
> +			if (xfs_bmse_can_merge(&got, &cont,  offset_shift_fsb))
> +				WARN_ON_ONCE(1);
>  		}
> -	} else if (got.br_startoff < offset_shift_fsb)
> -		return -EINVAL;
> -
> +	}
>  	/*
>  	 * Increment the extent index for the next iteration, update the start
>  	 * offset of the in-core extent and update the btree if applicable.
>  	 */
> -	(*current_ext)++;
> +	if (SHIFT == SHIFT_LEFT)
> +		(*current_ext)++;
> +	else
> +		(*current_ext)--;
>  	xfs_bmbt_set_startoff(gotp, startoff);
>  	*logflags |= XFS_ILOG_CORE;
>  	if (!cur) {
> @@ -5579,10 +5615,10 @@ xfs_bmse_shift_one(
>  }
>  
>  /*
> - * Shift extent records to the left to cover a hole.
> + * Shift extent records to the left/right to cover/create a hole.
>   *
>   * The maximum number of extents to be shifted in a single operation is
> - * @num_exts. @start_fsb specifies the file offset to start the shift and the
> + * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
>   * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
>   * is the length by which each extent is shifted. If there is no hole to shift
>   * the extents into, this will be considered invalid operation and we abort
> @@ -5592,12 +5628,13 @@ int
>  xfs_bmap_shift_extents(
>  	struct xfs_trans	*tp,
>  	struct xfs_inode	*ip,
> -	xfs_fileoff_t		start_fsb,
> +	xfs_fileoff_t		*next_fsb,
>  	xfs_fileoff_t		offset_shift_fsb,
>  	int			*done,
> -	xfs_fileoff_t		*next_fsb,
> +	xfs_fileoff_t		stop_fsb,
>  	xfs_fsblock_t		*firstblock,
>  	struct xfs_bmap_free	*flist,
> +	enum SHIFT_DIRECTION	SHIFT,
>  	int			num_exts)
>  {
>  	struct xfs_btree_cur		*cur = NULL;
> @@ -5607,10 +5644,11 @@ xfs_bmap_shift_extents(
>  	struct xfs_ifork		*ifp;
>  	xfs_extnum_t			nexts = 0;
>  	xfs_extnum_t			current_ext;
> +	xfs_extnum_t			total_extents;
> +	xfs_extnum_t			stop_extent;
>  	int				error = 0;
>  	int				whichfork = XFS_DATA_FORK;
>  	int				logflags = 0;
> -	int				total_extents;
>  
>  	if (unlikely(XFS_TEST_ERROR(
>  	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
> @@ -5626,6 +5664,7 @@ xfs_bmap_shift_extents(
>  
>  	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
>  	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
> +	ASSERT(SHIFT == SHIFT_LEFT || SHIFT == SHIFT_RIGHT);
>  
>  	ifp = XFS_IFORK_PTR(ip, whichfork);
>  	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
> @@ -5643,43 +5682,85 @@ xfs_bmap_shift_extents(
>  	}
>  
>  	/*
> +	 * There may be delalloc extents in the data fork before the range we
> +	 * are collapsing out, so we cannot use the count of real extents here.
> +	 * Instead we have to calculate it from the incore fork.
> +	 */
> +	total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
> +	if (total_extents == 0) {
> +		*done = 1;
> +		goto del_cursor;
> +	}
> +
> +	/*
> +	 * In case of first right shift, we need to initialize next_fsb
> +	 */
> +	if (*next_fsb == NULLFSBLOCK) {
> +		ASSERT(SHIFT == SHIFT_RIGHT);
> +		gotp = xfs_iext_get_ext(ifp, total_extents - 1);
> +		xfs_bmbt_get_all(gotp, &got);
> +		*next_fsb = got.br_startoff;
> +		if (stop_fsb > *next_fsb) {
> +			*done = 1;
> +			goto del_cursor;
> +		}
> +	}
> +
> +	/* Lookup the extent index at which we have to stop */
> +	if (SHIFT == SHIFT_RIGHT) {
> +		gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent);
> +		/* Make stop_extent exclusive of shift range */
> +		stop_extent--;
> +	} else
> +		stop_extent = total_extents;
> +
> +	/*
>  	 * Look up the extent index for the fsb where we start shifting. We can
>  	 * henceforth iterate with current_ext as extent list changes are locked
>  	 * out via ilock.
>  	 *
>  	 * gotp can be null in 2 cases: 1) if there are no extents or 2)
> -	 * start_fsb lies in a hole beyond which there are no extents. Either
> +	 * *next_fsb lies in a hole beyond which there are no extents. Either
>  	 * way, we are done.
>  	 */
> -	gotp = xfs_iext_bno_to_ext(ifp, start_fsb, &current_ext);
> +	gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, &current_ext);
>  	if (!gotp) {
>  		*done = 1;
>  		goto del_cursor;
>  	}
>  
> -	/*
> -	 * There may be delalloc extents in the data fork before the range we
> -	 * are collapsing out, so we cannot use the count of real extents here.
> -	 * Instead we have to calculate it from the incore fork.
> -	 */
> -	total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
> -	while (nexts++ < num_exts && current_ext < total_extents) {
> +	/* some sanity checking before we finally start shifting extents */
> +	if ((SHIFT == SHIFT_LEFT && current_ext >= stop_extent) ||
> +	     (SHIFT == SHIFT_RIGHT && current_ext <= stop_extent)) {
> +		error = EIO;
> +		goto del_cursor;
> +	}
> +
> +	while (nexts++ < num_exts) {
>  		error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
> -					&current_ext, gotp, cur, &logflags);
> +					   &current_ext, gotp, cur, &logflags,
> +					   SHIFT);
>  		if (error)
>  			goto del_cursor;
> +		/*
> +		 * In case there was an extent merge after shifting extent,
> +		 * extent numbers would change.
> +		 * Update total extent count and grab the next record.
> +		 */
> +		if (SHIFT == SHIFT_LEFT) {
> +			total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
> +			stop_extent = total_extents;
> +		}
>  
> -		/* update total extent count and grab the next record */
> -		total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
> -		if (current_ext >= total_extents)
> +		if (current_ext == stop_extent) {
> +			*done = 1;
> +			*next_fsb = NULLFSBLOCK;
>  			break;
> +		}
>  		gotp = xfs_iext_get_ext(ifp, current_ext);
>  	}
>  
> -	/* Check if we are done */
> -	if (current_ext == total_extents) {
> -		*done = 1;
> -	} else if (next_fsb) {
> +	if (!*done) {
>  		xfs_bmbt_get_all(gotp, &got);
>  		*next_fsb = got.br_startoff;
>  	}
> @@ -5694,3 +5775,192 @@ del_cursor:
>  
>  	return error;
>  }
> +
> +/*
> + * Splits an extent into two extents at split_fsb block that it is
> + * the first block of the current_ext. @current_ext is a target extent
> + * to be split. @split_fsb is a block where the extents is split.
> + * If split_fsb lies in a hole or the first block of extents, just return 0.
> + */
> +STATIC int
> +xfs_bmap_split_extent_at(
> +	struct xfs_trans	*tp,
> +	struct xfs_inode	*ip,
> +	xfs_fileoff_t		split_fsb,
> +	xfs_fsblock_t		*firstfsb,
> +	struct xfs_bmap_free	*free_list)
> +{
> +	int				whichfork = XFS_DATA_FORK;
> +	struct xfs_btree_cur		*cur = NULL;
> +	struct xfs_bmbt_rec_host	*gotp;
> +	struct xfs_bmbt_irec		got;
> +	struct xfs_bmbt_irec		new; /* split extent */
> +	struct xfs_mount		*mp = ip->i_mount;
> +	struct xfs_ifork		*ifp;
> +	xfs_fsblock_t			gotblkcnt; /* new block count for got */
> +	xfs_extnum_t			current_ext;
> +	int				error = 0;
> +	int				logflags = 0;
> +	int				i = 0;
> +
> +	if (unlikely(XFS_TEST_ERROR(
> +	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
> +	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
> +	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
> +		XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
> +				 XFS_ERRLEVEL_LOW, mp);
> +		return -EFSCORRUPTED;
> +	}
> +
> +	if (XFS_FORCED_SHUTDOWN(mp))
> +		return -EIO;
> +
> +	ifp = XFS_IFORK_PTR(ip, whichfork);
> +	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
> +		/* Read in all the extents */
> +		error = xfs_iread_extents(tp, ip, whichfork);
> +		if (error)
> +			return error;
> +	}
> +
> +	gotp = xfs_iext_bno_to_ext(ifp, split_fsb, &current_ext);
> +	/*
> +	 * gotp can be null in 2 cases: 1) if there are no extents
> +	 * or 2) split_fsb lies in a hole beyond which there are
> +	 * no extents. Either way, we are done.
> +	 */
> +	if (!gotp)
> +		return 0;
> +
> +	xfs_bmbt_get_all(gotp, &got);
> +
> +	/*
> +	 * Check split_fsb lies in a hole or the start boundary offset
> +	 * of the extent.
> +	 */
> +	if (got.br_startoff >= split_fsb)
> +		return 0;
> +
> +	gotblkcnt = split_fsb - got.br_startoff;
> +	new.br_startoff = split_fsb;
> +	new.br_startblock = got.br_startblock + gotblkcnt;
> +	new.br_blockcount = got.br_blockcount - gotblkcnt;
> +	new.br_state = got.br_state;
> +
> +	if (ifp->if_flags & XFS_IFBROOT) {
> +		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
> +		cur->bc_private.b.firstblock = *firstfsb;
> +		cur->bc_private.b.flist = free_list;
> +		cur->bc_private.b.flags = 0;
> +	}
> +
> +	if (cur) {
> +		error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
> +				got.br_startblock,
> +				got.br_blockcount,
> +				&i);
> +		if (error)
> +			goto del_cursor;
> +		XFS_WANT_CORRUPTED_GOTO(i == 1, del_cursor);
> +	}
> +
> +	xfs_bmbt_set_blockcount(gotp, gotblkcnt);
> +	got.br_blockcount = gotblkcnt;
> +
> +	logflags = XFS_ILOG_CORE;
> +	if (cur) {
> +		error = xfs_bmbt_update(cur, got.br_startoff,
> +				got.br_startblock,
> +				got.br_blockcount,
> +				got.br_state);
> +		if (error)
> +			goto del_cursor;
> +	} else
> +		logflags |= XFS_ILOG_DEXT;
> +
> +	/* Add new extent */
> +	current_ext++;
> +	xfs_iext_insert(ip, current_ext, 1, &new, 0);
> +	XFS_IFORK_NEXT_SET(ip, whichfork,
> +			   XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
> +
> +	if (cur) {
> +		error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
> +				new.br_startblock, new.br_blockcount,
> +				&i);
> +		if (error)
> +			goto del_cursor;
> +		XFS_WANT_CORRUPTED_GOTO(i == 0, del_cursor);
> +		cur->bc_rec.b.br_state = new.br_state;
> +
> +		error = xfs_btree_insert(cur, &i);
> +		if (error)
> +			goto del_cursor;
> +		XFS_WANT_CORRUPTED_GOTO(i == 1, del_cursor);
> +	}
> +
> +	/*
> +	 * Convert to a btree if necessary.
> +	 */
> +	if (xfs_bmap_needs_btree(ip, whichfork)) {
> +		int tmp_logflags; /* partial log flag return val */
> +
> +		ASSERT(cur == NULL);
> +		error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, free_list,
> +				&cur, 0, &tmp_logflags, whichfork);
> +		logflags |= tmp_logflags;
> +	}
> +
> +del_cursor:
> +	if (cur) {
> +		cur->bc_private.b.allocated = 0;
> +		xfs_btree_del_cursor(cur,
> +				error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
> +	}
> +
> +	if (logflags)
> +		xfs_trans_log_inode(tp, ip, logflags);
> +	return error;
> +}
> +
> +int
> +xfs_bmap_split_extent(
> +	struct xfs_inode        *ip,
> +	xfs_fileoff_t           split_fsb)
> +{
> +	struct xfs_mount        *mp = ip->i_mount;
> +	struct xfs_trans        *tp;
> +	struct xfs_bmap_free    free_list;
> +	xfs_fsblock_t           firstfsb;
> +	int                     committed;
> +	int                     error;
> +
> +	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
> +	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
> +			XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
> +	if (error) {
> +		xfs_trans_cancel(tp, 0);
> +		return error;
> +	}
> +
> +	xfs_ilock(ip, XFS_ILOCK_EXCL);
> +	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
> +
> +	xfs_bmap_init(&free_list, &firstfsb);
> +
> +	error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
> +			&firstfsb, &free_list);
> +	if (error)
> +		goto out;
> +
> +	error = xfs_bmap_finish(&tp, &free_list, &committed);
> +	if (error)
> +		goto out;
> +
> +	return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
> +
> +
> +out:
> +	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
> +	return error;
> +}
> diff --git a/fs/xfs/libxfs/xfs_bmap.h b/fs/xfs/libxfs/xfs_bmap.h
> index 44db6db..d4678dd 100644
> --- a/fs/xfs/libxfs/xfs_bmap.h
> +++ b/fs/xfs/libxfs/xfs_bmap.h
> @@ -135,6 +135,11 @@ static inline void xfs_bmap_init(xfs_bmap_free_t *flp, xfs_fsblock_t *fbp)
>   */
>  #define XFS_BMAP_MAX_SHIFT_EXTENTS	1
>  
> +enum SHIFT_DIRECTION {
> +	SHIFT_LEFT = 0,
> +	SHIFT_RIGHT,
> +};
> +
>  #ifdef DEBUG
>  void	xfs_bmap_trace_exlist(struct xfs_inode *ip, xfs_extnum_t cnt,
>  		int whichfork, unsigned long caller_ip);
> @@ -178,8 +183,10 @@ int	xfs_check_nostate_extents(struct xfs_ifork *ifp, xfs_extnum_t idx,
>  		xfs_extnum_t num);
>  uint	xfs_default_attroffset(struct xfs_inode *ip);
>  int	xfs_bmap_shift_extents(struct xfs_trans *tp, struct xfs_inode *ip,
> -		xfs_fileoff_t start_fsb, xfs_fileoff_t offset_shift_fsb,
> -		int *done, xfs_fileoff_t *next_fsb, xfs_fsblock_t *firstblock,
> -		struct xfs_bmap_free *flist, int num_exts);
> +		xfs_fileoff_t *next_fsb, xfs_fileoff_t offset_shift_fsb,
> +		int *done, xfs_fileoff_t stop_fsb, xfs_fsblock_t *firstblock,
> +		struct xfs_bmap_free *flist, enum SHIFT_DIRECTION SHIFT,
> +		int num_exts);
> +int	xfs_bmap_split_extent(struct xfs_inode *ip, xfs_fileoff_t split_offset);
>  
>  #endif	/* __XFS_BMAP_H__ */
> diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c
> index 22a5dcb..841744c 100644
> --- a/fs/xfs/xfs_bmap_util.c
> +++ b/fs/xfs/xfs_bmap_util.c
> @@ -1376,22 +1376,19 @@ out:
>  }
>  
>  /*
> - * xfs_collapse_file_space()
> - *	This routine frees disk space and shift extent for the given file.
> - *	The first thing we do is to free data blocks in the specified range
> - *	by calling xfs_free_file_space(). It would also sync dirty data
> - *	and invalidate page cache over the region on which collapse range
> - *	is working. And Shift extent records to the left to cover a hole.
> - * RETURNS:
> - *	0 on success
> - *	errno on error
> - *
> + * @next_fsb will keep track of the extent currently undergoing shift.
> + * @stop_fsb will keep track of the extent at which we have to stop.
> + * If we are shifting left, we will start with block (offset + len) and
> + * shift each extent till last extent.
> + * If we are shifting right, we will start with last extent inside file space
> + * and continue until we reach the block corresponding to offset.
>   */
>  int
> -xfs_collapse_file_space(
> -	struct xfs_inode	*ip,
> -	xfs_off_t		offset,
> -	xfs_off_t		len)
> +xfs_shift_file_space(
> +	struct xfs_inode        *ip,
> +	xfs_off_t               offset,
> +	xfs_off_t               len,
> +	enum SHIFT_DIRECTION	SHIFT)
>  {
>  	int			done = 0;
>  	struct xfs_mount	*mp = ip->i_mount;
> @@ -1400,21 +1397,26 @@ xfs_collapse_file_space(
>  	struct xfs_bmap_free	free_list;
>  	xfs_fsblock_t		first_block;
>  	int			committed;
> -	xfs_fileoff_t		start_fsb;
> +	xfs_fileoff_t		stop_fsb;
>  	xfs_fileoff_t		next_fsb;
>  	xfs_fileoff_t		shift_fsb;
>  
> -	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
> +	ASSERT(SHIFT == SHIFT_LEFT || SHIFT == SHIFT_RIGHT);
>  
> -	trace_xfs_collapse_file_space(ip);
> +	if (SHIFT == SHIFT_LEFT) {
> +		next_fsb = XFS_B_TO_FSB(mp, offset + len);
> +		stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size);
> +	} else {
> +		/*
> +		 * If right shift, delegate the work of initialization of
> +		 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
> +		 */
> +		next_fsb = NULLFSBLOCK;
> +		stop_fsb = XFS_B_TO_FSB(mp, offset);
> +	}
>  
> -	next_fsb = XFS_B_TO_FSB(mp, offset + len);
>  	shift_fsb = XFS_B_TO_FSB(mp, len);
>  
> -	error = xfs_free_file_space(ip, offset, len);
> -	if (error)
> -		return error;
> -
>  	/*
>  	 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
>  	 * into the accessible region of the file.
> @@ -1427,20 +1429,23 @@ xfs_collapse_file_space(
>  
>  	/*
>  	 * Writeback and invalidate cache for the remainder of the file as we're
> -	 * about to shift down every extent from the collapse range to EOF. The
> -	 * free of the collapse range above might have already done some of
> -	 * this, but we shouldn't rely on it to do anything outside of the range
> -	 * that was freed.
> +	 * about to shift down every extent from offset to EOF.
>  	 */
>  	error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
> -					     offset + len, -1);
> +					     offset, -1);
>  	if (error)
>  		return error;
>  	error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
> -					(offset + len) >> PAGE_CACHE_SHIFT, -1);
> +					offset >> PAGE_CACHE_SHIFT, -1);
>  	if (error)
>  		return error;
>  
> +	if (SHIFT == SHIFT_RIGHT) {
> +		error = xfs_bmap_split_extent(ip, stop_fsb);
> +		if (error)
> +			return error;
> +	}
> +
>  	while (!error && !done) {
>  		tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
>  		/*
> @@ -1464,7 +1469,7 @@ xfs_collapse_file_space(
>  		if (error)
>  			goto out;
>  
> -		xfs_trans_ijoin(tp, ip, 0);
> +		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
>  
>  		xfs_bmap_init(&free_list, &first_block);
>  
> @@ -1472,10 +1477,9 @@ xfs_collapse_file_space(
>  		 * We are using the write transaction in which max 2 bmbt
>  		 * updates are allowed
>  		 */
> -		start_fsb = next_fsb;
> -		error = xfs_bmap_shift_extents(tp, ip, start_fsb, shift_fsb,
> -				&done, &next_fsb, &first_block, &free_list,
> -				XFS_BMAP_MAX_SHIFT_EXTENTS);
> +		error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb,
> +				&done, stop_fsb, &first_block, &free_list,
> +				SHIFT, XFS_BMAP_MAX_SHIFT_EXTENTS);
>  		if (error)
>  			goto out;
>  
> @@ -1484,18 +1488,70 @@ xfs_collapse_file_space(
>  			goto out;
>  
>  		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
> -		xfs_iunlock(ip, XFS_ILOCK_EXCL);
>  	}
>  
>  	return error;
>  
>  out:
>  	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
> -	xfs_iunlock(ip, XFS_ILOCK_EXCL);
>  	return error;
>  }
>  
>  /*
> + * xfs_collapse_file_space()
> + *	This routine frees disk space and shift extent for the given file.
> + *	The first thing we do is to free data blocks in the specified range
> + *	by calling xfs_free_file_space(). It would also sync dirty data
> + *	and invalidate page cache over the region on which collapse range
> + *	is working. And Shift extent records to the left to cover a hole.
> + * RETURNS:
> + *	0 on success
> + *	errno on error
> + *
> + */
> +int
> +xfs_collapse_file_space(
> +	struct xfs_inode	*ip,
> +	xfs_off_t		offset,
> +	xfs_off_t		len)
> +{
> +	int error;
> +
> +	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
> +	trace_xfs_collapse_file_space(ip);
> +
> +	error = xfs_free_file_space(ip, offset, len);
> +	if (error)
> +		return error;
> +
> +	return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT);
> +}
> +
> +/*
> + * xfs_insert_file_space()
> + *	This routine create hole space by shifting extents for the given file.
> + *	The first thing we do is to sync dirty data and invalidate page cache
> + *	over the region on which insert range is working. And split an extent
> + *	to two extents at given offset by calling xfs_bmap_split_extent.
> + *	And shift all extent records which are laying between [offset,
> + *	last allocated extent] to the right to reserve hole range.
> + * RETURNS:
> + *	0 on success
> + *	errno on error
> + */
> +int
> +xfs_insert_file_space(
> +	struct xfs_inode	*ip,
> +	loff_t			offset,
> +	loff_t			len)
> +{
> +	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
> +	trace_xfs_insert_file_space(ip);
> +
> +	return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT);
> +}
> +
> +/*
>   * We need to check that the format of the data fork in the temporary inode is
>   * valid for the target inode before doing the swap. This is not a problem with
>   * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
> diff --git a/fs/xfs/xfs_bmap_util.h b/fs/xfs/xfs_bmap_util.h
> index 2fdb72d..6cb116c 100644
> --- a/fs/xfs/xfs_bmap_util.h
> +++ b/fs/xfs/xfs_bmap_util.h
> @@ -98,6 +98,8 @@ int	xfs_zero_file_space(struct xfs_inode *ip, xfs_off_t offset,
>  			    xfs_off_t len);
>  int	xfs_collapse_file_space(struct xfs_inode *, xfs_off_t offset,
>  				xfs_off_t len);
> +int	xfs_insert_file_space(struct xfs_inode *, xfs_off_t offset,
> +				xfs_off_t len);
>  
>  /* EOF block manipulation functions */
>  bool	xfs_can_free_eofblocks(struct xfs_inode *ip, bool force);
> diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
> index ac7f1e8..4900cdf 100644
> --- a/fs/xfs/xfs_file.c
> +++ b/fs/xfs/xfs_file.c
> @@ -787,11 +787,13 @@ xfs_file_fallocate(
>  	struct xfs_trans	*tp;
>  	long			error;
>  	loff_t			new_size = 0;
> +	int			do_file_insert = 0;
>  
>  	if (!S_ISREG(inode->i_mode))
>  		return -EINVAL;
>  	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
> -		     FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
> +		     FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
> +		     FALLOC_FL_INSERT_RANGE))
>  		return -EOPNOTSUPP;
>  
>  	xfs_ilock(ip, XFS_IOLOCK_EXCL);
> @@ -821,6 +823,28 @@ xfs_file_fallocate(
>  		error = xfs_collapse_file_space(ip, offset, len);
>  		if (error)
>  			goto out_unlock;
> +	} else if (mode & FALLOC_FL_INSERT_RANGE) {
> +		unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
> +
> +		if (offset & blksize_mask || len & blksize_mask) {
> +			error = -EINVAL;
> +			goto out_unlock;
> +		}
> +
> +		/* Check for wrap through zero */
> +		if (inode->i_size + len > inode->i_sb->s_maxbytes) {
> +			error = -EFBIG;
> +			goto out_unlock;
> +		}
> +
> +		/* Offset should be less than i_size */
> +		if (offset >= i_size_read(inode)) {
> +			error = -EINVAL;
> +			goto out_unlock;
> +		}
> +
> +		new_size = i_size_read(inode) + len;
> +		do_file_insert = 1;
>  	} else {
>  		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
>  		    offset + len > i_size_read(inode)) {
> @@ -852,7 +876,8 @@ xfs_file_fallocate(
>  	if (ip->i_d.di_mode & S_IXGRP)
>  		ip->i_d.di_mode &= ~S_ISGID;
>  
> -	if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE)))
> +	if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE |
> +		      FALLOC_FL_INSERT_RANGE)))
>  		ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
>  
>  	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
> @@ -871,8 +896,20 @@ xfs_file_fallocate(
>  		iattr.ia_valid = ATTR_SIZE;
>  		iattr.ia_size = new_size;
>  		error = xfs_setattr_size(ip, &iattr);
> +		if (error)
> +			goto out_unlock;
>  	}
>  
> +	/*
> +	 * Some operations are performed after the inode size is updated. For
> +	 * example, insert range expands the address space of the file, shifts
> +	 * all subsequent extents to create a hole inside the file. Updating
> +	 * the size first ensures that shifted extents aren't left hanging
> +	 * past EOF in the event of a crash or failure.
> +	 */
> +	if (do_file_insert)
> +		error = xfs_insert_file_space(ip, offset, len);
> +
>  out_unlock:
>  	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
>  	return error;
> diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h
> index 51372e3..7e45fa1 100644
> --- a/fs/xfs/xfs_trace.h
> +++ b/fs/xfs/xfs_trace.h
> @@ -664,6 +664,7 @@ DEFINE_INODE_EVENT(xfs_alloc_file_space);
>  DEFINE_INODE_EVENT(xfs_free_file_space);
>  DEFINE_INODE_EVENT(xfs_zero_file_space);
>  DEFINE_INODE_EVENT(xfs_collapse_file_space);
> +DEFINE_INODE_EVENT(xfs_insert_file_space);
>  DEFINE_INODE_EVENT(xfs_readdir);
>  #ifdef CONFIG_XFS_POSIX_ACL
>  DEFINE_INODE_EVENT(xfs_get_acl);
> -- 
> 1.7.9.5
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists