journal.c

linux 内核源代码

	return (struct reiserfs_journal_cnode *)0;
}

/*
** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
** being overwritten by a replay after crashing.
**
** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
** sure you never write the block without logging it.
**
** next_zero_bit is a suggestion about the next block to try for find_forward.
** when bl is rejected because it is set in a journal list bitmap, we search
** for the next zero bit in the bitmap that rejected bl.  Then, we return that
** through next_zero_bit for find_forward to try.
**
** Just because we return something in next_zero_bit does not mean we won't
** reject it on the next call to reiserfs_in_journal
**
*/
int reiserfs_in_journal(struct super_block *p_s_sb,
			unsigned int bmap_nr, int bit_nr, int search_all,
			b_blocknr_t * next_zero_bit)
{
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
	struct reiserfs_journal_cnode *cn;
	struct reiserfs_list_bitmap *jb;
	int i;
	unsigned long bl;

	*next_zero_bit = 0;	/* always start this at zero. */

	PROC_INFO_INC(p_s_sb, journal.in_journal);
	/* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
	 ** if we crash before the transaction that freed it commits,  this transaction won't
	 ** have committed either, and the block will never be written
	 */
	if (search_all) {
		for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
			PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
			jb = journal->j_list_bitmap + i;
			if (jb->journal_list && jb->bitmaps[bmap_nr] &&
			    test_bit(bit_nr,
				     (unsigned long *)jb->bitmaps[bmap_nr]->
				     data)) {
				*next_zero_bit =
				    find_next_zero_bit((unsigned long *)
						       (jb->bitmaps[bmap_nr]->
							data),
						       p_s_sb->s_blocksize << 3,
						       bit_nr + 1);
				return 1;
			}
		}
	}

	bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
	/* is it in any old transactions? */
	if (search_all
	    && (cn =
		get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
		return 1;
	}

	/* is it in the current transaction.  This should never happen */
	if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
		BUG();
		return 1;
	}

	PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
	/* safe for reuse */
	return 0;
}

/* insert cn into table
*/
static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
				       struct reiserfs_journal_cnode *cn)
{
	struct reiserfs_journal_cnode *cn_orig;

	cn_orig = journal_hash(table, cn->sb, cn->blocknr);
	cn->hnext = cn_orig;
	cn->hprev = NULL;
	if (cn_orig) {
		cn_orig->hprev = cn;
	}
	journal_hash(table, cn->sb, cn->blocknr) = cn;
}

/* lock the current transaction */
static inline void lock_journal(struct super_block *p_s_sb)
{
	PROC_INFO_INC(p_s_sb, journal.lock_journal);
	down(&SB_JOURNAL(p_s_sb)->j_lock);
}

/* unlock the current transaction */
static inline void unlock_journal(struct super_block *p_s_sb)
{
	up(&SB_JOURNAL(p_s_sb)->j_lock);
}

static inline void get_journal_list(struct reiserfs_journal_list *jl)
{
	jl->j_refcount++;
}

static inline void put_journal_list(struct super_block *s,
				    struct reiserfs_journal_list *jl)
{
	if (jl->j_refcount < 1) {
		reiserfs_panic(s, "trans id %lu, refcount at %d",
			       jl->j_trans_id, jl->j_refcount);
	}
	if (--jl->j_refcount == 0)
		kfree(jl);
}

/*
** this used to be much more involved, and I'm keeping it just in case things get ugly again.
** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
** transaction.
*/
static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
					   struct reiserfs_journal_list *jl)
{

	struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
	if (jb) {
		cleanup_bitmap_list(p_s_sb, jb);
	}
	jl->j_list_bitmap->journal_list = NULL;
	jl->j_list_bitmap = NULL;
}

static int journal_list_still_alive(struct super_block *s,
				    unsigned long trans_id)
{
	struct reiserfs_journal *journal = SB_JOURNAL(s);
	struct list_head *entry = &journal->j_journal_list;
	struct reiserfs_journal_list *jl;

	if (!list_empty(entry)) {
		jl = JOURNAL_LIST_ENTRY(entry->next);
		if (jl->j_trans_id <= trans_id) {
			return 1;
		}
	}
	return 0;
}

/*
 * If page->mapping was null, we failed to truncate this page for
 * some reason.  Most likely because it was truncated after being
 * logged via data=journal.
 *
 * This does a check to see if the buffer belongs to one of these
 * lost pages before doing the final put_bh.  If page->mapping was
 * null, it tries to free buffers on the page, which should make the
 * final page_cache_release drop the page from the lru.
 */
static void release_buffer_page(struct buffer_head *bh)
{
	struct page *page = bh->b_page;
	if (!page->mapping && !TestSetPageLocked(page)) {
		page_cache_get(page);
		put_bh(bh);
		if (!page->mapping)
			try_to_free_buffers(page);
		unlock_page(page);
		page_cache_release(page);
	} else {
		put_bh(bh);
	}
}

static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
{
	char b[BDEVNAME_SIZE];

	if (buffer_journaled(bh)) {
		reiserfs_warning(NULL,
				 "clm-2084: pinned buffer %lu:%s sent to disk",
				 bh->b_blocknr, bdevname(bh->b_bdev, b));
	}
	if (uptodate)
		set_buffer_uptodate(bh);
	else
		clear_buffer_uptodate(bh);

	unlock_buffer(bh);
	release_buffer_page(bh);
}

static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
{
	if (uptodate)
		set_buffer_uptodate(bh);
	else
		clear_buffer_uptodate(bh);
	unlock_buffer(bh);
	put_bh(bh);
}

static void submit_logged_buffer(struct buffer_head *bh)
{
	get_bh(bh);
	bh->b_end_io = reiserfs_end_buffer_io_sync;
	clear_buffer_journal_new(bh);
	clear_buffer_dirty(bh);
	if (!test_clear_buffer_journal_test(bh))
		BUG();
	if (!buffer_uptodate(bh))
		BUG();
	submit_bh(WRITE, bh);
}

static void submit_ordered_buffer(struct buffer_head *bh)
{
	get_bh(bh);
	bh->b_end_io = reiserfs_end_ordered_io;
	clear_buffer_dirty(bh);
	if (!buffer_uptodate(bh))
		BUG();
	submit_bh(WRITE, bh);
}

static int submit_barrier_buffer(struct buffer_head *bh)
{
	get_bh(bh);
	bh->b_end_io = reiserfs_end_ordered_io;
	clear_buffer_dirty(bh);
	if (!buffer_uptodate(bh))
		BUG();
	return submit_bh(WRITE_BARRIER, bh);
}

static void check_barrier_completion(struct super_block *s,
				     struct buffer_head *bh)
{
	if (buffer_eopnotsupp(bh)) {
		clear_buffer_eopnotsupp(bh);
		disable_barrier(s);
		set_buffer_uptodate(bh);
		set_buffer_dirty(bh);
		sync_dirty_buffer(bh);
	}
}

#define CHUNK_SIZE 32
struct buffer_chunk {
	struct buffer_head *bh[CHUNK_SIZE];
	int nr;
};

static void write_chunk(struct buffer_chunk *chunk)
{
	int i;
	get_fs_excl();
	for (i = 0; i < chunk->nr; i++) {
		submit_logged_buffer(chunk->bh[i]);
	}
	chunk->nr = 0;
	put_fs_excl();
}

static void write_ordered_chunk(struct buffer_chunk *chunk)
{
	int i;
	get_fs_excl();
	for (i = 0; i < chunk->nr; i++) {
		submit_ordered_buffer(chunk->bh[i]);
	}
	chunk->nr = 0;
	put_fs_excl();
}

static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
			spinlock_t * lock, void (fn) (struct buffer_chunk *))
{
	int ret = 0;
	BUG_ON(chunk->nr >= CHUNK_SIZE);
	chunk->bh[chunk->nr++] = bh;
	if (chunk->nr >= CHUNK_SIZE) {
		ret = 1;
		if (lock)
			spin_unlock(lock);
		fn(chunk);
		if (lock)
			spin_lock(lock);
	}
	return ret;
}

static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
static struct reiserfs_jh *alloc_jh(void)
{
	struct reiserfs_jh *jh;
	while (1) {
		jh = kmalloc(sizeof(*jh), GFP_NOFS);
		if (jh) {
			atomic_inc(&nr_reiserfs_jh);
			return jh;
		}
		yield();
	}
}

/*
 * we want to free the jh when the buffer has been written
 * and waited on
 */
void reiserfs_free_jh(struct buffer_head *bh)
{
	struct reiserfs_jh *jh;

	jh = bh->b_private;
	if (jh) {
		bh->b_private = NULL;
		jh->bh = NULL;
		list_del_init(&jh->list);
		kfree(jh);
		if (atomic_read(&nr_reiserfs_jh) <= 0)
			BUG();
		atomic_dec(&nr_reiserfs_jh);
		put_bh(bh);
	}
}

static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
			   int tail)
{
	struct reiserfs_jh *jh;

	if (bh->b_private) {
		spin_lock(&j->j_dirty_buffers_lock);
		if (!bh->b_private) {
			spin_unlock(&j->j_dirty_buffers_lock);
			goto no_jh;
		}
		jh = bh->b_private;
		list_del_init(&jh->list);
	} else {
	      no_jh:
		get_bh(bh);
		jh = alloc_jh();
		spin_lock(&j->j_dirty_buffers_lock);
		/* buffer must be locked for __add_jh, should be able to have
		 * two adds at the same time
		 */
		BUG_ON(bh->b_private);
		jh->bh = bh;
		bh->b_private = jh;
	}
	jh->jl = j->j_current_jl;
	if (tail)
		list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
	else {
		list_add_tail(&jh->list, &jh->jl->j_bh_list);
	}
	spin_unlock(&j->j_dirty_buffers_lock);
	return 0;
}

int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
{
	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
}
int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
{
	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
}

#define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
static int write_ordered_buffers(spinlock_t * lock,
				 struct reiserfs_journal *j,
				 struct reiserfs_journal_list *jl,
				 struct list_head *list)
{
	struct buffer_head *bh;
	struct reiserfs_jh *jh;
	int ret = j->j_errno;
	struct buffer_chunk chunk;
	struct list_head tmp;
	INIT_LIST_HEAD(&tmp);

	chunk.nr = 0;
	spin_lock(lock);
	while (!list_empty(list)) {
		jh = JH_ENTRY(list->next);
		bh = jh->bh;
		get_bh(bh);
		if (test_set_buffer_locked(bh)) {
			if (!buffer_dirty(bh)) {
				list_move(&jh->list, &tmp);
				goto loop_next;
			}
			spin_unlock(lock);
			if (chunk.nr)
				write_ordered_chunk(&chunk);
			wait_on_buffer(bh);
			cond_resched();
			spin_lock(lock);
			goto loop_next;
		}
		/* in theory, dirty non-uptodate buffers should never get here,
		 * but the upper layer io error paths still have a few quirks.
		 * Handle them here as gracefully as we can
		 */
		if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
			clear_buffer_dirty(bh);
			ret = -EIO;
		}
		if (buffer_dirty(bh)) {
			list_move(&jh->list, &tmp);
			add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
		} else {
			reiserfs_free_jh(bh);
			unlock_buffer(bh);
		}
	      loop_next:
		put_bh(bh);
		cond_resched_lock(lock);
	}
	if (chunk.nr) {
		spin_unlock(lock);
		write_ordered_chunk(&chunk);
		spin_lock(lock);
	}
	while (!list_empty(&tmp)) {
		jh = JH_ENTRY(tmp.prev);
		bh = jh->bh;
		get_bh(bh);
		reiserfs_free_jh(bh);

		if (buffer_locked(bh)) {
			spin_unlock(lock);
			wait_on_buffer(bh);
			spin_lock(lock);
		}
		if (!buffer_uptodate(bh)) {
			ret = -EIO;
		}
		/* ugly interaction with invalidatepage here.
		 * reiserfs_invalidate_page will pin any buffer that has a valid
		 * journal head from an older transaction.  If someone else sets
		 * our buffer dirty after we write it in the first loop, and
		 * then someone truncates the page away, nobody will ever write
		 * the buffer. We're safe if we write the page one last time
		 * after freeing the journal header.
		 */
		if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
			spin_unlock(lock);
			ll_rw_block(WRITE, 1, &bh);
			spin_lock(lock);
		}
		put_bh(bh);
		cond_resched_lock(lock);
	}
	spin_unlock(lock);
	return ret;