inode.c

ocfs1.4.1 oracle分布式文件系统

	 * get to ->delete_inode(), each node tries to convert it's
	 * lock to an exclusive. Trylocks are serialized by the inode
	 * meta data lock. If the upconvert suceeds, we know the inode
	 * is no longer live and can be deleted.
	 *
	 * Though we call this with the meta data lock held, the
	 * trylock keeps us from ABBA deadlock.
	 */
	status = ocfs2_try_open_lock(inode, 1);
	if (status == -EAGAIN) {
		status = 0;
		mlog(0, "Skipping delete of %llu because it is in use on "
		     "other nodes\n", (unsigned long long)oi->ip_blkno);
		goto bail;
	}
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	*wipe = 1;
	mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
	     (unsigned long long)oi->ip_blkno,
	     le16_to_cpu(di->i_orphaned_slot));

bail:
	return status;
}

/* Support function for ocfs2_delete_inode. Will help us keep the
 * inode data in a consistent state for clear_inode. Always truncates
 * pages, optionally sync's them first. */
static void ocfs2_cleanup_delete_inode(struct inode *inode,
				       int sync_data)
{
	mlog(0, "Cleanup inode %llu, sync = %d\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
	if (sync_data)
		write_inode_now(inode, 1);
	truncate_inode_pages(&inode->i_data, 0);
}

void ocfs2_delete_inode(struct inode *inode)
{
	int wipe, status;
	sigset_t blocked, oldset;
	struct buffer_head *di_bh = NULL;

	mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);

	if (is_bad_inode(inode)) {
		mlog(0, "Skipping delete of bad inode\n");
		goto bail;
	}

	if (!ocfs2_inode_is_valid_to_delete(inode)) {
		/* It's probably not necessary to truncate_inode_pages
		 * here but we do it for safety anyway (it will most
		 * likely be a no-op anyway) */
		ocfs2_cleanup_delete_inode(inode, 0);
		goto bail;
	}

	/* We want to block signals in delete_inode as the lock and
	 * messaging paths may return us -ERESTARTSYS. Which would
	 * cause us to exit early, resulting in inodes being orphaned
	 * forever. */
	sigfillset(&blocked);
	status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
	if (status < 0) {
		mlog_errno(status);
		ocfs2_cleanup_delete_inode(inode, 1);
		goto bail;
	}

	/* Lock down the inode. This gives us an up to date view of
	 * it's metadata (for verification), and allows us to
	 * serialize delete_inode on multiple nodes.
	 *
	 * Even though we might be doing a truncate, we don't take the
	 * allocation lock here as it won't be needed - nobody will
	 * have the file open.
	 */
	status = ocfs2_inode_lock(inode, &di_bh, 1);
	if (status < 0) {
		if (status != -ENOENT)
			mlog_errno(status);
		ocfs2_cleanup_delete_inode(inode, 0);
		goto bail_unblock;
	}

	/* Query the cluster. This will be the final decision made
	 * before we go ahead and wipe the inode. */
	status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
	if (!wipe || status < 0) {
		/* Error and remote inode busy both mean we won't be
		 * removing the inode, so they take almost the same
		 * path. */
		if (status < 0)
			mlog_errno(status);

		/* Someone in the cluster has disallowed a wipe of
		 * this inode, or it was never completely
		 * orphaned. Write out the pages and exit now. */
		ocfs2_cleanup_delete_inode(inode, 1);
		goto bail_unlock_inode;
	}

	ocfs2_cleanup_delete_inode(inode, 0);

	status = ocfs2_wipe_inode(inode, di_bh);
	if (status < 0) {
		if (status != -EDEADLK)
			mlog_errno(status);
		goto bail_unlock_inode;
	}

	/*
	 * Mark the inode as successfully deleted.
	 *
	 * This is important for ocfs2_clear_inode() as it will check
	 * this flag and skip any checkpointing work
	 *
	 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
	 * the LVB for other nodes.
	 */
	OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;

bail_unlock_inode:
	ocfs2_inode_unlock(inode, 1);
	brelse(di_bh);
bail_unblock:
	status = sigprocmask(SIG_SETMASK, &oldset, NULL);
	if (status < 0)
		mlog_errno(status);
bail:
	clear_inode(inode);
	mlog_exit_void();
}

void ocfs2_clear_inode(struct inode *inode)
{
	int status;
	struct ocfs2_inode_info *oi = OCFS2_I(inode);

	mlog_entry_void();

	if (!inode)
		goto bail;

	mlog(0, "Clearing inode: %llu, nlink = %u\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);

	mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
			"Inode=%lu\n", inode->i_ino);

	/* To preven remote deletes we hold open lock before, now it
	 * is time to unlock PR and EX open locks. */
	ocfs2_open_unlock(inode);

	/* Do these before all the other work so that we don't bounce
	 * the downconvert thread while waiting to destroy the locks. */
	ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
	ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
	ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);

	/* We very well may get a clear_inode before all an inodes
	 * metadata has hit disk. Of course, we can't drop any cluster
	 * locks until the journal has finished with it. The only
	 * exception here are successfully wiped inodes - their
	 * metadata can now be considered to be part of the system
	 * inodes from which it came. */
	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
		ocfs2_checkpoint_inode(inode);

	mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
			"Clear inode of %llu, inode has io markers\n",
			(unsigned long long)oi->ip_blkno);

	ocfs2_extent_map_trunc(inode, 0);

	status = ocfs2_drop_inode_locks(inode);
	if (status < 0)
		mlog_errno(status);

	ocfs2_lock_res_free(&oi->ip_rw_lockres);
	ocfs2_lock_res_free(&oi->ip_inode_lockres);
	ocfs2_lock_res_free(&oi->ip_open_lockres);

	ocfs2_metadata_cache_purge(inode);

	mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
			"Clear inode of %llu, inode has %u cache items\n",
			(unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);

	mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
			"Clear inode of %llu, inode has a bad flag\n",
			(unsigned long long)oi->ip_blkno);

	mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
			"Clear inode of %llu, inode is locked\n",
			(unsigned long long)oi->ip_blkno);

	mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
			"Clear inode of %llu, io_mutex is locked\n",
			(unsigned long long)oi->ip_blkno);
	mutex_unlock(&oi->ip_io_mutex);

	/*
	 * down_trylock() returns 0, down_write_trylock() returns 1
	 * kernel 1, world 0
	 */
	mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
			"Clear inode of %llu, alloc_sem is locked\n",
			(unsigned long long)oi->ip_blkno);
	up_write(&oi->ip_alloc_sem);

	mlog_bug_on_msg(oi->ip_open_count,
			"Clear inode of %llu has open count %d\n",
			(unsigned long long)oi->ip_blkno, oi->ip_open_count);

	/* Clear all other flags. */
	oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
	oi->ip_created_trans = 0;
	oi->ip_last_trans = 0;
	oi->ip_dir_start_lookup = 0;
	oi->ip_blkno = 0ULL;

bail:
	mlog_exit_void();
}

/* Called under inode_lock, with no more references on the
 * struct inode, so it's safe here to check the flags field
 * and to manipulate i_nlink without any other locks. */
void ocfs2_drop_inode(struct inode *inode)
{
	struct ocfs2_inode_info *oi = OCFS2_I(inode);

	mlog_entry_void();

	mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
	     (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);

	if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
		generic_delete_inode(inode);
	else
		generic_drop_inode(inode);

	mlog_exit_void();
}

/*
 * TODO: this should probably be merged into ocfs2_get_block
 *
 * However, you now need to pay attention to the cont_prepare_write()
 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
 * expects never to extend).
 */
struct buffer_head *ocfs2_bread(struct inode *inode,
				int block, int *err, int reada)
{
	struct buffer_head *bh = NULL;
	int tmperr;
	u64 p_blkno;
	int readflags = OCFS2_BH_CACHED;

	if (reada)
		readflags |= OCFS2_BH_READAHEAD;

	if (((u64)block << inode->i_sb->s_blocksize_bits) >=
	    i_size_read(inode)) {
		BUG_ON(!reada);
		return NULL;
	}

	down_read(&OCFS2_I(inode)->ip_alloc_sem);
	tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
					     NULL);
	up_read(&OCFS2_I(inode)->ip_alloc_sem);
	if (tmperr < 0) {
		mlog_errno(tmperr);
		goto fail;
	}

	tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
				  readflags, inode);
	if (tmperr < 0)
		goto fail;

	tmperr = 0;

	*err = 0;
	return bh;

fail:
	if (bh) {
		brelse(bh);
		bh = NULL;
	}
	*err = -EIO;
	return NULL;
}

/*
 * This is called from our getattr.
 */
int ocfs2_inode_revalidate(struct dentry *dentry)
{
	struct inode *inode = dentry->d_inode;
	int status = 0;

	mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
		   inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);

	if (!inode) {
		mlog(0, "eep, no inode!\n");
		status = -ENOENT;
		goto bail;
	}

	spin_lock(&OCFS2_I(inode)->ip_lock);
	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
		spin_unlock(&OCFS2_I(inode)->ip_lock);
		mlog(0, "inode deleted!\n");
		status = -ENOENT;
		goto bail;
	}
	spin_unlock(&OCFS2_I(inode)->ip_lock);

	/* Let ocfs2_inode_lock do the work of updating our struct
	 * inode for us. */
	status = ocfs2_inode_lock(inode, NULL, 0);
	if (status < 0) {
		if (status != -ENOENT)
			mlog_errno(status);
		goto bail;
	}
	ocfs2_inode_unlock(inode, 0);
bail:
	mlog_exit(status);

	return status;
}

/*
 * Updates a disk inode from a
 * struct inode.
 * Only takes ip_lock.
 */
int ocfs2_mark_inode_dirty(handle_t *handle,
			   struct inode *inode,
			   struct buffer_head *bh)
{
	int status;
	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;

	mlog_entry("(inode %llu)\n",
		   (unsigned long long)OCFS2_I(inode)->ip_blkno);

	status = ocfs2_journal_access(handle, inode, bh,
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto leave;
	}

	spin_lock(&OCFS2_I(inode)->ip_lock);
	fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
	ocfs2_get_inode_flags(OCFS2_I(inode));
	fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
	fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
	spin_unlock(&OCFS2_I(inode)->ip_lock);

	fe->i_size = cpu_to_le64(i_size_read(inode));
	fe->i_links_count = cpu_to_le16(inode->i_nlink);
	fe->i_uid = cpu_to_le32(inode->i_uid);
	fe->i_gid = cpu_to_le32(inode->i_gid);
	fe->i_mode = cpu_to_le16(inode->i_mode);
	fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
	fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
	fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
	fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);

	status = ocfs2_journal_dirty(handle, bh);
	if (status < 0)
		mlog_errno(status);

	status = 0;
leave:

	mlog_exit(status);
	return status;
}

/*
 *
 * Updates a struct inode from a disk inode.
 * does no i/o, only takes ip_lock.
 */
void ocfs2_refresh_inode(struct inode *inode,
			 struct ocfs2_dinode *fe)
{
	spin_lock(&OCFS2_I(inode)->ip_lock);

	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
	ocfs2_set_inode_flags(inode);
	i_size_write(inode, le64_to_cpu(fe->i_size));
	inode->i_nlink = le16_to_cpu(fe->i_links_count);
	inode->i_uid = le32_to_cpu(fe->i_uid);
	inode->i_gid = le32_to_cpu(fe->i_gid);
	inode->i_mode = le16_to_cpu(fe->i_mode);
	if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
		inode->i_blocks = 0;
	else
		inode->i_blocks = ocfs2_inode_sector_count(inode);
	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);

	spin_unlock(&OCFS2_I(inode)->ip_lock);
}