mft.c

linux 内核源代码

 * it is dirty in the inode meta data rather than the data page cache of the
 * inode, and thus there are no data pages that need writing out.  Therefore, a
 * full mark_inode_dirty() is overkill.  A mark_inode_dirty_sync(), on the
 * other hand, is not sufficient, because I_DIRTY_DATASYNC needs to be set to
 * ensure ->write_inode is called from generic_osync_inode() and this needs to
 * happen or the file data would not necessarily hit the device synchronously,
 * even though the vfs inode has the O_SYNC flag set.  Also, I_DIRTY_DATASYNC
 * simply "feels" better than just I_DIRTY_SYNC, since the file data has not
 * actually hit the block device yet, which is not what I_DIRTY_SYNC on its own
 * would suggest.
 */
void __mark_mft_record_dirty(ntfs_inode *ni)
{
	ntfs_inode *base_ni;

	ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
	BUG_ON(NInoAttr(ni));
	mark_ntfs_record_dirty(ni->page, ni->page_ofs);
	/* Determine the base vfs inode and mark it dirty, too. */
	mutex_lock(&ni->extent_lock);
	if (likely(ni->nr_extents >= 0))
		base_ni = ni;
	else
		base_ni = ni->ext.base_ntfs_ino;
	mutex_unlock(&ni->extent_lock);
	__mark_inode_dirty(VFS_I(base_ni), I_DIRTY_SYNC | I_DIRTY_DATASYNC);
}

static const char *ntfs_please_email = "Please email "
		"[e-mail] and say that you saw "
		"this message.  Thank you.";

/**
 * ntfs_sync_mft_mirror_umount - synchronise an mft record to the mft mirror
 * @vol:	ntfs volume on which the mft record to synchronize resides
 * @mft_no:	mft record number of mft record to synchronize
 * @m:		mapped, mst protected (extent) mft record to synchronize
 *
 * Write the mapped, mst protected (extent) mft record @m with mft record
 * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol,
 * bypassing the page cache and the $MFTMirr inode itself.
 *
 * This function is only for use at umount time when the mft mirror inode has
 * already been disposed off.  We BUG() if we are called while the mft mirror
 * inode is still attached to the volume.
 *
 * On success return 0.  On error return -errno.
 *
 * NOTE:  This function is not implemented yet as I am not convinced it can
 * actually be triggered considering the sequence of commits we do in super.c::
 * ntfs_put_super().  But just in case we provide this place holder as the
 * alternative would be either to BUG() or to get a NULL pointer dereference
 * and Oops.
 */
static int ntfs_sync_mft_mirror_umount(ntfs_volume *vol,
		const unsigned long mft_no, MFT_RECORD *m)
{
	BUG_ON(vol->mftmirr_ino);
	ntfs_error(vol->sb, "Umount time mft mirror syncing is not "
			"implemented yet.  %s", ntfs_please_email);
	return -EOPNOTSUPP;
}

/**
 * ntfs_sync_mft_mirror - synchronize an mft record to the mft mirror
 * @vol:	ntfs volume on which the mft record to synchronize resides
 * @mft_no:	mft record number of mft record to synchronize
 * @m:		mapped, mst protected (extent) mft record to synchronize
 * @sync:	if true, wait for i/o completion
 *
 * Write the mapped, mst protected (extent) mft record @m with mft record
 * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol.
 *
 * On success return 0.  On error return -errno and set the volume errors flag
 * in the ntfs volume @vol.
 *
 * NOTE:  We always perform synchronous i/o and ignore the @sync parameter.
 *
 * TODO:  If @sync is false, want to do truly asynchronous i/o, i.e. just
 * schedule i/o via ->writepage or do it via kntfsd or whatever.
 */
int ntfs_sync_mft_mirror(ntfs_volume *vol, const unsigned long mft_no,
		MFT_RECORD *m, int sync)
{
	struct page *page;
	unsigned int blocksize = vol->sb->s_blocksize;
	int max_bhs = vol->mft_record_size / blocksize;
	struct buffer_head *bhs[max_bhs];
	struct buffer_head *bh, *head;
	u8 *kmirr;
	runlist_element *rl;
	unsigned int block_start, block_end, m_start, m_end, page_ofs;
	int i_bhs, nr_bhs, err = 0;
	unsigned char blocksize_bits = vol->sb->s_blocksize_bits;

	ntfs_debug("Entering for inode 0x%lx.", mft_no);
	BUG_ON(!max_bhs);
	if (unlikely(!vol->mftmirr_ino)) {
		/* This could happen during umount... */
		err = ntfs_sync_mft_mirror_umount(vol, mft_no, m);
		if (likely(!err))
			return err;
		goto err_out;
	}
	/* Get the page containing the mirror copy of the mft record @m. */
	page = ntfs_map_page(vol->mftmirr_ino->i_mapping, mft_no >>
			(PAGE_CACHE_SHIFT - vol->mft_record_size_bits));
	if (IS_ERR(page)) {
		ntfs_error(vol->sb, "Failed to map mft mirror page.");
		err = PTR_ERR(page);
		goto err_out;
	}
	lock_page(page);
	BUG_ON(!PageUptodate(page));
	ClearPageUptodate(page);
	/* Offset of the mft mirror record inside the page. */
	page_ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK;
	/* The address in the page of the mirror copy of the mft record @m. */
	kmirr = page_address(page) + page_ofs;
	/* Copy the mst protected mft record to the mirror. */
	memcpy(kmirr, m, vol->mft_record_size);
	/* Create uptodate buffers if not present. */
	if (unlikely(!page_has_buffers(page))) {
		struct buffer_head *tail;

		bh = head = alloc_page_buffers(page, blocksize, 1);
		do {
			set_buffer_uptodate(bh);
			tail = bh;
			bh = bh->b_this_page;
		} while (bh);
		tail->b_this_page = head;
		attach_page_buffers(page, head);
	}
	bh = head = page_buffers(page);
	BUG_ON(!bh);
	rl = NULL;
	nr_bhs = 0;
	block_start = 0;
	m_start = kmirr - (u8*)page_address(page);
	m_end = m_start + vol->mft_record_size;
	do {
		block_end = block_start + blocksize;
		/* If the buffer is outside the mft record, skip it. */
		if (block_end <= m_start)
			continue;
		if (unlikely(block_start >= m_end))
			break;
		/* Need to map the buffer if it is not mapped already. */
		if (unlikely(!buffer_mapped(bh))) {
			VCN vcn;
			LCN lcn;
			unsigned int vcn_ofs;

			bh->b_bdev = vol->sb->s_bdev;
			/* Obtain the vcn and offset of the current block. */
			vcn = ((VCN)mft_no << vol->mft_record_size_bits) +
					(block_start - m_start);
			vcn_ofs = vcn & vol->cluster_size_mask;
			vcn >>= vol->cluster_size_bits;
			if (!rl) {
				down_read(&NTFS_I(vol->mftmirr_ino)->
						runlist.lock);
				rl = NTFS_I(vol->mftmirr_ino)->runlist.rl;
				/*
				 * $MFTMirr always has the whole of its runlist
				 * in memory.
				 */
				BUG_ON(!rl);
			}
			/* Seek to element containing target vcn. */
			while (rl->length && rl[1].vcn <= vcn)
				rl++;
			lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
			/* For $MFTMirr, only lcn >= 0 is a successful remap. */
			if (likely(lcn >= 0)) {
				/* Setup buffer head to correct block. */
				bh->b_blocknr = ((lcn <<
						vol->cluster_size_bits) +
						vcn_ofs) >> blocksize_bits;
				set_buffer_mapped(bh);
			} else {
				bh->b_blocknr = -1;
				ntfs_error(vol->sb, "Cannot write mft mirror "
						"record 0x%lx because its "
						"location on disk could not "
						"be determined (error code "
						"%lli).", mft_no,
						(long long)lcn);
				err = -EIO;
			}
		}
		BUG_ON(!buffer_uptodate(bh));
		BUG_ON(!nr_bhs && (m_start != block_start));
		BUG_ON(nr_bhs >= max_bhs);
		bhs[nr_bhs++] = bh;
		BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end));
	} while (block_start = block_end, (bh = bh->b_this_page) != head);
	if (unlikely(rl))
		up_read(&NTFS_I(vol->mftmirr_ino)->runlist.lock);
	if (likely(!err)) {
		/* Lock buffers and start synchronous write i/o on them. */
		for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
			struct buffer_head *tbh = bhs[i_bhs];

			if (unlikely(test_set_buffer_locked(tbh)))
				BUG();
			BUG_ON(!buffer_uptodate(tbh));
			clear_buffer_dirty(tbh);
			get_bh(tbh);
			tbh->b_end_io = end_buffer_write_sync;
			submit_bh(WRITE, tbh);
		}
		/* Wait on i/o completion of buffers. */
		for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
			struct buffer_head *tbh = bhs[i_bhs];

			wait_on_buffer(tbh);
			if (unlikely(!buffer_uptodate(tbh))) {
				err = -EIO;
				/*
				 * Set the buffer uptodate so the page and
				 * buffer states do not become out of sync.
				 */
				set_buffer_uptodate(tbh);
			}
		}
	} else /* if (unlikely(err)) */ {
		/* Clean the buffers. */
		for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++)
			clear_buffer_dirty(bhs[i_bhs]);
	}
	/* Current state: all buffers are clean, unlocked, and uptodate. */
	/* Remove the mst protection fixups again. */
	post_write_mst_fixup((NTFS_RECORD*)kmirr);
	flush_dcache_page(page);
	SetPageUptodate(page);
	unlock_page(page);
	ntfs_unmap_page(page);
	if (likely(!err)) {
		ntfs_debug("Done.");
	} else {
		ntfs_error(vol->sb, "I/O error while writing mft mirror "
				"record 0x%lx!", mft_no);
err_out:
		ntfs_error(vol->sb, "Failed to synchronize $MFTMirr (error "
				"code %i).  Volume will be left marked dirty "
				"on umount.  Run ntfsfix on the partition "
				"after umounting to correct this.", -err);
		NVolSetErrors(vol);
	}
	return err;
}

/**
 * write_mft_record_nolock - write out a mapped (extent) mft record
 * @ni:		ntfs inode describing the mapped (extent) mft record
 * @m:		mapped (extent) mft record to write
 * @sync:	if true, wait for i/o completion
 *
 * Write the mapped (extent) mft record @m described by the (regular or extent)
 * ntfs inode @ni to backing store.  If the mft record @m has a counterpart in
 * the mft mirror, that is also updated.
 *
 * We only write the mft record if the ntfs inode @ni is dirty and the first
 * buffer belonging to its mft record is dirty, too.  We ignore the dirty state
 * of subsequent buffers because we could have raced with
 * fs/ntfs/aops.c::mark_ntfs_record_dirty().
 *
 * On success, clean the mft record and return 0.  On error, leave the mft
 * record dirty and return -errno.
 *
 * NOTE:  We always perform synchronous i/o and ignore the @sync parameter.
 * However, if the mft record has a counterpart in the mft mirror and @sync is
 * true, we write the mft record, wait for i/o completion, and only then write
 * the mft mirror copy.  This ensures that if the system crashes either the mft
 * or the mft mirror will contain a self-consistent mft record @m.  If @sync is
 * false on the other hand, we start i/o on both and then wait for completion
 * on them.  This provides a speedup but no longer guarantees that you will end
 * up with a self-consistent mft record in the case of a crash but if you asked
 * for asynchronous writing you probably do not care about that anyway.
 *
 * TODO:  If @sync is false, want to do truly asynchronous i/o, i.e. just
 * schedule i/o via ->writepage or do it via kntfsd or whatever.
 */
int write_mft_record_nolock(ntfs_inode *ni, MFT_RECORD *m, int sync)
{
	ntfs_volume *vol = ni->vol;
	struct page *page = ni->page;
	unsigned int blocksize = vol->sb->s_blocksize;
	unsigned char blocksize_bits = vol->sb->s_blocksize_bits;
	int max_bhs = vol->mft_record_size / blocksize;
	struct buffer_head *bhs[max_bhs];
	struct buffer_head *bh, *head;
	runlist_element *rl;
	unsigned int block_start, block_end, m_start, m_end;
	int i_bhs, nr_bhs, err = 0;

	ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
	BUG_ON(NInoAttr(ni));
	BUG_ON(!max_bhs);
	BUG_ON(!PageLocked(page));
	/*
	 * If the ntfs_inode is clean no need to do anything.  If it is dirty,
	 * mark it as clean now so that it can be redirtied later on if needed.
	 * There is no danger of races since the caller is holding the locks
	 * for the mft record @m and the page it is in.
	 */
	if (!NInoTestClearDirty(ni))
		goto done;
	bh = head = page_buffers(page);
	BUG_ON(!bh);
	rl = NULL;
	nr_bhs = 0;
	block_start = 0;
	m_start = ni->page_ofs;
	m_end = m_start + vol->mft_record_size;
	do {
		block_end = block_start + blocksize;
		/* If the buffer is outside the mft record, skip it. */
		if (block_end <= m_start)
			continue;
		if (unlikely(block_start >= m_end))
			break;
		/*
		 * If this block is not the first one in the record, we ignore
		 * the buffer's dirty state because we could have raced with a
		 * parallel mark_ntfs_record_dirty().
		 */
		if (block_start == m_start) {
			/* This block is the first one in the record. */
			if (!buffer_dirty(bh)) {
				BUG_ON(nr_bhs);
				/* Clean records are not written out. */
				break;
			}
		}
		/* Need to map the buffer if it is not mapped already. */
		if (unlikely(!buffer_mapped(bh))) {
			VCN vcn;
			LCN lcn;
			unsigned int vcn_ofs;

			bh->b_bdev = vol->sb->s_bdev;
			/* Obtain the vcn and offset of the current block. */
			vcn = ((VCN)ni->mft_no << vol->mft_record_size_bits) +
					(block_start - m_start);
			vcn_ofs = vcn & vol->cluster_size_mask;
			vcn >>= vol->cluster_size_bits;
			if (!rl) {
				down_read(&NTFS_I(vol->mft_ino)->runlist.lock);
				rl = NTFS_I(vol->mft_ino)->runlist.rl;
				BUG_ON(!rl);
			}
			/* Seek to element containing target vcn. */
			while (rl->length && rl[1].vcn <= vcn)
				rl++;
			lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
			/* For $MFT, only lcn >= 0 is a successful remap. */
			if (likely(lcn >= 0)) {
				/* Setup buffer head to correct block. */
				bh->b_blocknr = ((lcn <<
						vol->cluster_size_bits) +
						vcn_ofs) >> blocksize_bits;
				set_buffer_mapped(bh);
			} else {
				bh->b_blocknr = -1;
				ntfs_error(vol->sb, "Cannot write mft record "
						"0x%lx because its location "
						"on disk could not be "
						"determined (error code %lli).",
						ni->mft_no, (long long)lcn);
				err = -EIO;
			}
		}
		BUG_ON(!buffer_uptodate(bh));
		BUG_ON(!nr_bhs && (m_start != block_start));
		BUG_ON(nr_bhs >= max_bhs);
		bhs[nr_bhs++] = bh;
		BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end));
	} while (block_start = block_end, (bh = bh->b_this_page) != head);
	if (unlikely(rl))
		up_read(&NTFS_I(vol->mft_ino)->runlist.lock);