mft.c

linux 内核源代码

	if (!nr_bhs)
		goto done;
	if (unlikely(err))
		goto cleanup_out;
	/* Apply the mst protection fixups. */
	err = pre_write_mst_fixup((NTFS_RECORD*)m, vol->mft_record_size);
	if (err) {
		ntfs_error(vol->sb, "Failed to apply mst fixups!");
		goto cleanup_out;
	}
	flush_dcache_mft_record_page(ni);
	/* 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);
	}
	/* Synchronize the mft mirror now if not @sync. */
	if (!sync && ni->mft_no < vol->mftmirr_size)
		ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync);
	/* 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.
			 */
			if (PageUptodate(page))
				set_buffer_uptodate(tbh);
		}
	}
	/* If @sync, now synchronize the mft mirror. */
	if (sync && ni->mft_no < vol->mftmirr_size)
		ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync);
	/* Remove the mst protection fixups again. */
	post_write_mst_fixup((NTFS_RECORD*)m);
	flush_dcache_mft_record_page(ni);
	if (unlikely(err)) {
		/* I/O error during writing.  This is really bad! */
		ntfs_error(vol->sb, "I/O error while writing mft record "
				"0x%lx!  Marking base inode as bad.  You "
				"should unmount the volume and run chkdsk.",
				ni->mft_no);
		goto err_out;
	}
done:
	ntfs_debug("Done.");
	return 0;
cleanup_out:
	/* Clean the buffers. */
	for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++)
		clear_buffer_dirty(bhs[i_bhs]);
err_out:
	/*
	 * Current state: all buffers are clean, unlocked, and uptodate.
	 * The caller should mark the base inode as bad so that no more i/o
	 * happens.  ->clear_inode() will still be invoked so all extent inodes
	 * and other allocated memory will be freed.
	 */
	if (err == -ENOMEM) {
		ntfs_error(vol->sb, "Not enough memory to write mft record.  "
				"Redirtying so the write is retried later.");
		mark_mft_record_dirty(ni);
		err = 0;
	} else
		NVolSetErrors(vol);
	return err;
}

/**
 * ntfs_may_write_mft_record - check if an mft record may be written out
 * @vol:	[IN]  ntfs volume on which the mft record to check resides
 * @mft_no:	[IN]  mft record number of the mft record to check
 * @m:		[IN]  mapped mft record to check
 * @locked_ni:	[OUT] caller has to unlock this ntfs inode if one is returned
 *
 * Check if the mapped (base or extent) mft record @m with mft record number
 * @mft_no belonging to the ntfs volume @vol may be written out.  If necessary
 * and possible the ntfs inode of the mft record is locked and the base vfs
 * inode is pinned.  The locked ntfs inode is then returned in @locked_ni.  The
 * caller is responsible for unlocking the ntfs inode and unpinning the base
 * vfs inode.
 *
 * Return 'true' if the mft record may be written out and 'false' if not.
 *
 * The caller has locked the page and cleared the uptodate flag on it which
 * means that we can safely write out any dirty mft records that do not have
 * their inodes in icache as determined by ilookup5() as anyone
 * opening/creating such an inode would block when attempting to map the mft
 * record in read_cache_page() until we are finished with the write out.
 *
 * Here is a description of the tests we perform:
 *
 * If the inode is found in icache we know the mft record must be a base mft
 * record.  If it is dirty, we do not write it and return 'false' as the vfs
 * inode write paths will result in the access times being updated which would
 * cause the base mft record to be redirtied and written out again.  (We know
 * the access time update will modify the base mft record because Windows
 * chkdsk complains if the standard information attribute is not in the base
 * mft record.)
 *
 * If the inode is in icache and not dirty, we attempt to lock the mft record
 * and if we find the lock was already taken, it is not safe to write the mft
 * record and we return 'false'.
 *
 * If we manage to obtain the lock we have exclusive access to the mft record,
 * which also allows us safe writeout of the mft record.  We then set
 * @locked_ni to the locked ntfs inode and return 'true'.
 *
 * Note we cannot just lock the mft record and sleep while waiting for the lock
 * because this would deadlock due to lock reversal (normally the mft record is
 * locked before the page is locked but we already have the page locked here
 * when we try to lock the mft record).
 *
 * If the inode is not in icache we need to perform further checks.
 *
 * If the mft record is not a FILE record or it is a base mft record, we can
 * safely write it and return 'true'.
 *
 * We now know the mft record is an extent mft record.  We check if the inode
 * corresponding to its base mft record is in icache and obtain a reference to
 * it if it is.  If it is not, we can safely write it and return 'true'.
 *
 * We now have the base inode for the extent mft record.  We check if it has an
 * ntfs inode for the extent mft record attached and if not it is safe to write
 * the extent mft record and we return 'true'.
 *
 * The ntfs inode for the extent mft record is attached to the base inode so we
 * attempt to lock the extent mft record and if we find the lock was already
 * taken, it is not safe to write the extent mft record and we return 'false'.
 *
 * If we manage to obtain the lock we have exclusive access to the extent mft
 * record, which also allows us safe writeout of the extent mft record.  We
 * set the ntfs inode of the extent mft record clean and then set @locked_ni to
 * the now locked ntfs inode and return 'true'.
 *
 * Note, the reason for actually writing dirty mft records here and not just
 * relying on the vfs inode dirty code paths is that we can have mft records
 * modified without them ever having actual inodes in memory.  Also we can have
 * dirty mft records with clean ntfs inodes in memory.  None of the described
 * cases would result in the dirty mft records being written out if we only
 * relied on the vfs inode dirty code paths.  And these cases can really occur
 * during allocation of new mft records and in particular when the
 * initialized_size of the $MFT/$DATA attribute is extended and the new space
 * is initialized using ntfs_mft_record_format().  The clean inode can then
 * appear if the mft record is reused for a new inode before it got written
 * out.
 */
bool ntfs_may_write_mft_record(ntfs_volume *vol, const unsigned long mft_no,
		const MFT_RECORD *m, ntfs_inode **locked_ni)
{
	struct super_block *sb = vol->sb;
	struct inode *mft_vi = vol->mft_ino;
	struct inode *vi;
	ntfs_inode *ni, *eni, **extent_nis;
	int i;
	ntfs_attr na;

	ntfs_debug("Entering for inode 0x%lx.", mft_no);
	/*
	 * Normally we do not return a locked inode so set @locked_ni to NULL.
	 */
	BUG_ON(!locked_ni);
	*locked_ni = NULL;
	/*
	 * Check if the inode corresponding to this mft record is in the VFS
	 * inode cache and obtain a reference to it if it is.
	 */
	ntfs_debug("Looking for inode 0x%lx in icache.", mft_no);
	na.mft_no = mft_no;
	na.name = NULL;
	na.name_len = 0;
	na.type = AT_UNUSED;
	/*
	 * Optimize inode 0, i.e. $MFT itself, since we have it in memory and
	 * we get here for it rather often.
	 */
	if (!mft_no) {
		/* Balance the below iput(). */
		vi = igrab(mft_vi);
		BUG_ON(vi != mft_vi);
	} else {
		/*
		 * Have to use ilookup5_nowait() since ilookup5() waits for the
		 * inode lock which causes ntfs to deadlock when a concurrent
		 * inode write via the inode dirty code paths and the page
		 * dirty code path of the inode dirty code path when writing
		 * $MFT occurs.
		 */
		vi = ilookup5_nowait(sb, mft_no, (test_t)ntfs_test_inode, &na);
	}
	if (vi) {
		ntfs_debug("Base inode 0x%lx is in icache.", mft_no);
		/* The inode is in icache. */
		ni = NTFS_I(vi);
		/* Take a reference to the ntfs inode. */
		atomic_inc(&ni->count);
		/* If the inode is dirty, do not write this record. */
		if (NInoDirty(ni)) {
			ntfs_debug("Inode 0x%lx is dirty, do not write it.",
					mft_no);
			atomic_dec(&ni->count);
			iput(vi);
			return false;
		}
		ntfs_debug("Inode 0x%lx is not dirty.", mft_no);
		/* The inode is not dirty, try to take the mft record lock. */
		if (unlikely(!mutex_trylock(&ni->mrec_lock))) {
			ntfs_debug("Mft record 0x%lx is already locked, do "
					"not write it.", mft_no);
			atomic_dec(&ni->count);
			iput(vi);
			return false;
		}
		ntfs_debug("Managed to lock mft record 0x%lx, write it.",
				mft_no);
		/*
		 * The write has to occur while we hold the mft record lock so
		 * return the locked ntfs inode.
		 */
		*locked_ni = ni;
		return true;
	}
	ntfs_debug("Inode 0x%lx is not in icache.", mft_no);
	/* The inode is not in icache. */
	/* Write the record if it is not a mft record (type "FILE"). */
	if (!ntfs_is_mft_record(m->magic)) {
		ntfs_debug("Mft record 0x%lx is not a FILE record, write it.",
				mft_no);
		return true;
	}
	/* Write the mft record if it is a base inode. */
	if (!m->base_mft_record) {
		ntfs_debug("Mft record 0x%lx is a base record, write it.",
				mft_no);
		return true;
	}
	/*
	 * This is an extent mft record.  Check if the inode corresponding to
	 * its base mft record is in icache and obtain a reference to it if it
	 * is.
	 */
	na.mft_no = MREF_LE(m->base_mft_record);
	ntfs_debug("Mft record 0x%lx is an extent record.  Looking for base "
			"inode 0x%lx in icache.", mft_no, na.mft_no);
	if (!na.mft_no) {
		/* Balance the below iput(). */
		vi = igrab(mft_vi);
		BUG_ON(vi != mft_vi);
	} else
		vi = ilookup5_nowait(sb, na.mft_no, (test_t)ntfs_test_inode,
				&na);
	if (!vi) {
		/*
		 * The base inode is not in icache, write this extent mft
		 * record.
		 */
		ntfs_debug("Base inode 0x%lx is not in icache, write the "
				"extent record.", na.mft_no);
		return true;
	}
	ntfs_debug("Base inode 0x%lx is in icache.", na.mft_no);
	/*
	 * The base inode is in icache.  Check if it has the extent inode
	 * corresponding to this extent mft record attached.
	 */
	ni = NTFS_I(vi);
	mutex_lock(&ni->extent_lock);
	if (ni->nr_extents <= 0) {
		/*
		 * The base inode has no attached extent inodes, write this
		 * extent mft record.
		 */
		mutex_unlock(&ni->extent_lock);
		iput(vi);
		ntfs_debug("Base inode 0x%lx has no attached extent inodes, "
				"write the extent record.", na.mft_no);
		return true;
	}
	/* Iterate over the attached extent inodes. */
	extent_nis = ni->ext.extent_ntfs_inos;
	for (eni = NULL, i = 0; i < ni->nr_extents; ++i) {
		if (mft_no == extent_nis[i]->mft_no) {
			/*
			 * Found the extent inode corresponding to this extent
			 * mft record.
			 */
			eni = extent_nis[i];
			break;
		}
	}
	/*
	 * If the extent inode was not attached to the base inode, write this
	 * extent mft record.
	 */
	if (!eni) {
		mutex_unlock(&ni->extent_lock);
		iput(vi);
		ntfs_debug("Extent inode 0x%lx is not attached to its base "
				"inode 0x%lx, write the extent record.",
				mft_no, na.mft_no);
		return true;
	}
	ntfs_debug("Extent inode 0x%lx is attached to its base inode 0x%lx.",
			mft_no, na.mft_no);
	/* Take a reference to the extent ntfs inode. */
	atomic_inc(&eni->count);
	mutex_unlock(&ni->extent_lock);
	/*
	 * Found the extent inode coresponding to this extent mft record.
	 * Try to take the mft record lock.
	 */
	if (unlikely(!mutex_trylock(&eni->mrec_lock))) {
		atomic_dec(&eni->count);
		iput(vi);
		ntfs_debug("Extent mft record 0x%lx is already locked, do "
				"not write it.", mft_no);
		return false;
	}
	ntfs_debug("Managed to lock extent mft record 0x%lx, write it.",
			mft_no);
	if (NInoTestClearDirty(eni))
		ntfs_debug("Extent inode 0x%lx is dirty, marking it clean.",
				mft_no);
	/*
	 * The write has to occur while we hold the mft record lock so return
	 * the locked extent ntfs inode.
	 */
	*locked_ni = eni;
	return true;
}

static const char *es = "  Leaving inconsistent metadata.  Unmount and run "
		"chkdsk.";

/**
 * ntfs_mft_bitmap_find_and_alloc_free_rec_nolock - see name
 * @vol:	volume on which to search for a free mft record
 * @base_ni:	open base inode if allocating an extent mft record or NULL
 *
 * Search for a free mft record in the mft bitmap attribute on the ntfs volume
 * @vol.
 *
 * If @base_ni is NULL start the search at the default allocator position.
 *
 * If @base_ni is not NULL start the search at the mft record after the base
 * mft record @base_ni.
 *
 * Return the free mft record on success and -errno on error.  An error code of
 * -ENOSPC means that there are no free mft records in the currently
 * initialized mft bitmap.
 *
 * Locking: Caller must hold vol->mftbmp_lock for writing.
 */
static int ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(ntfs_volume *vol,
		ntfs_inode *base_ni)
{
	s64 pass_end, ll, data_pos, pass_start, ofs, bit;
	unsigned long flags;
	struct address_space *mftbmp_mapping;
	u8 *buf, *byte;
	struct page *page;
	unsigned int page_ofs, size;
	u8 pass, b;

	ntfs_debug("Searching for free mft record in the currently "
			"initialized mft bitmap.");
	mftbmp_mapping = vol->mftbmp_ino->i_mapping;
	/*
	 * Set the end of the pass making sure we do not overflow the mft
	 * bitmap.
	 */
	read_lock_irqsave(&NTFS_I(vol->mft_ino)->size_lock, flags);