xfs_aops.c

linux-2.6.15.6

/*
 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
 * All Rights Reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it would be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write the Free Software Foundation,
 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 */
#include "xfs.h"
#include "xfs_bit.h"
#include "xfs_log.h"
#include "xfs_inum.h"
#include "xfs_sb.h"
#include "xfs_ag.h"
#include "xfs_dir.h"
#include "xfs_dir2.h"
#include "xfs_trans.h"
#include "xfs_dmapi.h"
#include "xfs_mount.h"
#include "xfs_bmap_btree.h"
#include "xfs_alloc_btree.h"
#include "xfs_ialloc_btree.h"
#include "xfs_dir_sf.h"
#include "xfs_dir2_sf.h"
#include "xfs_attr_sf.h"
#include "xfs_dinode.h"
#include "xfs_inode.h"
#include "xfs_alloc.h"
#include "xfs_btree.h"
#include "xfs_error.h"
#include "xfs_rw.h"
#include "xfs_iomap.h"
#include <linux/mpage.h>
#include <linux/writeback.h>

STATIC void xfs_count_page_state(struct page *, int *, int *, int *);
STATIC void xfs_convert_page(struct inode *, struct page *, xfs_iomap_t *,
		struct writeback_control *wbc, void *, int, int);

#if defined(XFS_RW_TRACE)
void
xfs_page_trace(
	int		tag,
	struct inode	*inode,
	struct page	*page,
	int		mask)
{
	xfs_inode_t	*ip;
	bhv_desc_t	*bdp;
	vnode_t		*vp = LINVFS_GET_VP(inode);
	loff_t		isize = i_size_read(inode);
	loff_t		offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
	int		delalloc = -1, unmapped = -1, unwritten = -1;

	if (page_has_buffers(page))
		xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);

	bdp = vn_bhv_lookup(VN_BHV_HEAD(vp), &xfs_vnodeops);
	ip = XFS_BHVTOI(bdp);
	if (!ip->i_rwtrace)
		return;

	ktrace_enter(ip->i_rwtrace,
		(void *)((unsigned long)tag),
		(void *)ip,
		(void *)inode,
		(void *)page,
		(void *)((unsigned long)mask),
		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
		(void *)((unsigned long)((isize >> 32) & 0xffffffff)),
		(void *)((unsigned long)(isize & 0xffffffff)),
		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
		(void *)((unsigned long)(offset & 0xffffffff)),
		(void *)((unsigned long)delalloc),
		(void *)((unsigned long)unmapped),
		(void *)((unsigned long)unwritten),
		(void *)NULL,
		(void *)NULL);
}
#else
#define xfs_page_trace(tag, inode, page, mask)
#endif

/*
 * Schedule IO completion handling on a xfsdatad if this was
 * the final hold on this ioend.
 */
STATIC void
xfs_finish_ioend(
	xfs_ioend_t		*ioend)
{
	if (atomic_dec_and_test(&ioend->io_remaining))
		queue_work(xfsdatad_workqueue, &ioend->io_work);
}

STATIC void
xfs_destroy_ioend(
	xfs_ioend_t		*ioend)
{
	vn_iowake(ioend->io_vnode);
	mempool_free(ioend, xfs_ioend_pool);
}

/*
 * Issue transactions to convert a buffer range from unwritten
 * to written extents.
 */
STATIC void
xfs_end_bio_unwritten(
	void			*data)
{
	xfs_ioend_t		*ioend = data;
	vnode_t			*vp = ioend->io_vnode;
	xfs_off_t		offset = ioend->io_offset;
	size_t			size = ioend->io_size;
	struct buffer_head	*bh, *next;
	int			error;

	if (ioend->io_uptodate)
		VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error);

	/* ioend->io_buffer_head is only non-NULL for buffered I/O */
	for (bh = ioend->io_buffer_head; bh; bh = next) {
		next = bh->b_private;

		bh->b_end_io = NULL;
		clear_buffer_unwritten(bh);
		end_buffer_async_write(bh, ioend->io_uptodate);
	}

	xfs_destroy_ioend(ioend);
}

/*
 * Allocate and initialise an IO completion structure.
 * We need to track unwritten extent write completion here initially.
 * We'll need to extend this for updating the ondisk inode size later
 * (vs. incore size).
 */
STATIC xfs_ioend_t *
xfs_alloc_ioend(
	struct inode		*inode)
{
	xfs_ioend_t		*ioend;

	ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);

	/*
	 * Set the count to 1 initially, which will prevent an I/O
	 * completion callback from happening before we have started
	 * all the I/O from calling the completion routine too early.
	 */
	atomic_set(&ioend->io_remaining, 1);
	ioend->io_uptodate = 1; /* cleared if any I/O fails */
	ioend->io_vnode = LINVFS_GET_VP(inode);
	ioend->io_buffer_head = NULL;
	atomic_inc(&ioend->io_vnode->v_iocount);
	ioend->io_offset = 0;
	ioend->io_size = 0;

	INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten, ioend);

	return ioend;
}

void
linvfs_unwritten_done(
	struct buffer_head	*bh,
	int			uptodate)
{
	xfs_ioend_t		*ioend = bh->b_private;
	static spinlock_t	unwritten_done_lock = SPIN_LOCK_UNLOCKED;
	unsigned long		flags;

	ASSERT(buffer_unwritten(bh));
	bh->b_end_io = NULL;

	if (!uptodate)
		ioend->io_uptodate = 0;

	/*
	 * Deep magic here.  We reuse b_private in the buffer_heads to build
	 * a chain for completing the I/O from user context after we've issued
	 * a transaction to convert the unwritten extent.
	 */
	spin_lock_irqsave(&unwritten_done_lock, flags);
	bh->b_private = ioend->io_buffer_head;
	ioend->io_buffer_head = bh;
	spin_unlock_irqrestore(&unwritten_done_lock, flags);

	xfs_finish_ioend(ioend);
}

STATIC int
xfs_map_blocks(
	struct inode		*inode,
	loff_t			offset,
	ssize_t			count,
	xfs_iomap_t		*mapp,
	int			flags)
{
	vnode_t			*vp = LINVFS_GET_VP(inode);
	int			error, nmaps = 1;

	VOP_BMAP(vp, offset, count, flags, mapp, &nmaps, error);
	if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
		VMODIFY(vp);
	return -error;
}

/*
 * Finds the corresponding mapping in block @map array of the
 * given @offset within a @page.
 */
STATIC xfs_iomap_t *
xfs_offset_to_map(
	struct page		*page,
	xfs_iomap_t		*iomapp,
	unsigned long		offset)
{
	loff_t			full_offset;	/* offset from start of file */

	ASSERT(offset < PAGE_CACHE_SIZE);

	full_offset = page->index;		/* NB: using 64bit number */
	full_offset <<= PAGE_CACHE_SHIFT;	/* offset from file start */
	full_offset += offset;			/* offset from page start */

	if (full_offset < iomapp->iomap_offset)
		return NULL;
	if (iomapp->iomap_offset + (iomapp->iomap_bsize -1) >= full_offset)
		return iomapp;
	return NULL;
}

STATIC void
xfs_map_at_offset(
	struct page		*page,
	struct buffer_head	*bh,
	unsigned long		offset,
	int			block_bits,
	xfs_iomap_t		*iomapp)
{
	xfs_daddr_t		bn;
	loff_t			delta;
	int			sector_shift;

	ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
	ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
	ASSERT(iomapp->iomap_bn != IOMAP_DADDR_NULL);

	delta = page->index;
	delta <<= PAGE_CACHE_SHIFT;
	delta += offset;
	delta -= iomapp->iomap_offset;
	delta >>= block_bits;

	sector_shift = block_bits - BBSHIFT;
	bn = iomapp->iomap_bn >> sector_shift;
	bn += delta;
	BUG_ON(!bn && !(iomapp->iomap_flags & IOMAP_REALTIME));
	ASSERT((bn << sector_shift) >= iomapp->iomap_bn);

	lock_buffer(bh);
	bh->b_blocknr = bn;
	bh->b_bdev = iomapp->iomap_target->pbr_bdev;
	set_buffer_mapped(bh);
	clear_buffer_delay(bh);
}

/*
 * Look for a page at index which is unlocked and contains our
 * unwritten extent flagged buffers at its head.  Returns page
 * locked and with an extra reference count, and length of the
 * unwritten extent component on this page that we can write,
 * in units of filesystem blocks.
 */
STATIC struct page *
xfs_probe_unwritten_page(
	struct address_space	*mapping,
	pgoff_t			index,
	xfs_iomap_t		*iomapp,
	xfs_ioend_t		*ioend,
	unsigned long		max_offset,
	unsigned long		*fsbs,
	unsigned int            bbits)
{
	struct page		*page;

	page = find_trylock_page(mapping, index);
	if (!page)
		return NULL;
	if (PageWriteback(page))
		goto out;

	if (page->mapping && page_has_buffers(page)) {
		struct buffer_head	*bh, *head;
		unsigned long		p_offset = 0;

		*fsbs = 0;
		bh = head = page_buffers(page);
		do {
			if (!buffer_unwritten(bh) || !buffer_uptodate(bh))
				break;
			if (!xfs_offset_to_map(page, iomapp, p_offset))
				break;
			if (p_offset >= max_offset)
				break;
			xfs_map_at_offset(page, bh, p_offset, bbits, iomapp);
			set_buffer_unwritten_io(bh);
			bh->b_private = ioend;
			p_offset += bh->b_size;
			(*fsbs)++;
		} while ((bh = bh->b_this_page) != head);

		if (p_offset)
			return page;
	}

out:
	unlock_page(page);
	return NULL;
}

/*
 * Look for a page at index which is unlocked and not mapped
 * yet - clustering for mmap write case.
 */
STATIC unsigned int
xfs_probe_unmapped_page(
	struct address_space	*mapping,
	pgoff_t			index,
	unsigned int		pg_offset)
{
	struct page		*page;
	int			ret = 0;

	page = find_trylock_page(mapping, index);
	if (!page)
		return 0;
	if (PageWriteback(page))
		goto out;

	if (page->mapping && PageDirty(page)) {
		if (page_has_buffers(page)) {
			struct buffer_head	*bh, *head;

			bh = head = page_buffers(page);
			do {
				if (buffer_mapped(bh) || !buffer_uptodate(bh))
					break;
				ret += bh->b_size;
				if (ret >= pg_offset)
					break;
			} while ((bh = bh->b_this_page) != head);
		} else
			ret = PAGE_CACHE_SIZE;
	}

out:
	unlock_page(page);
	return ret;
}

STATIC unsigned int
xfs_probe_unmapped_cluster(
	struct inode		*inode,
	struct page		*startpage,
	struct buffer_head	*bh,
	struct buffer_head	*head)
{
	pgoff_t			tindex, tlast, tloff;
	unsigned int		pg_offset, len, total = 0;
	struct address_space	*mapping = inode->i_mapping;

	/* First sum forwards in this page */
	do {
		if (buffer_mapped(bh) || !buffer_uptodate(bh))
			break;
		total += bh->b_size;
	} while ((bh = bh->b_this_page) != head);

	/* If we reached the end of the page, sum forwards in
	 * following pages.
	 */
	if (bh == head) {
		tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
		/* Prune this back to avoid pathological behavior */
		tloff = min(tlast, startpage->index + 64);
		for (tindex = startpage->index + 1; tindex < tloff; tindex++) {
			len = xfs_probe_unmapped_page(mapping, tindex,
							PAGE_CACHE_SIZE);
			if (!len)
				return total;
			total += len;
		}
		if (tindex == tlast &&
		    (pg_offset = i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
			total += xfs_probe_unmapped_page(mapping,
							tindex, pg_offset);
		}
	}
	return total;
}

/*
 * Probe for a given page (index) in the inode and test if it is delayed
 * and without unwritten buffers.  Returns page locked and with an extra
 * reference count.
 */
STATIC struct page *
xfs_probe_delalloc_page(
	struct inode		*inode,
	pgoff_t			index)
{
	struct page		*page;

	page = find_trylock_page(inode->i_mapping, index);
	if (!page)
		return NULL;
	if (PageWriteback(page))
		goto out;

	if (page->mapping && page_has_buffers(page)) {
		struct buffer_head	*bh, *head;
		int			acceptable = 0;

		bh = head = page_buffers(page);
		do {
			if (buffer_unwritten(bh)) {
				acceptable = 0;
				break;
			} else if (buffer_delay(bh)) {
				acceptable = 1;
			}
		} while ((bh = bh->b_this_page) != head);

		if (acceptable)
			return page;
	}