jbd2,ext4: add a shrinker to release checkpointed buffers

Current metadata buffer release logic in bdev_try_to_free_page() have
a lot of use-after-free issues when umount filesystem concurrently, and
it is difficult to fix directly because ext4 is the only user of
s_op->bdev_try_to_free_page callback and we may have to add more special
refcount or lock that is only used by ext4 into the common vfs layer,
which is unacceptable.

One better solution is remove the bdev_try_to_free_page callback, but
the real problem is we cannot easily release journal_head on the
checkpointed buffer, so try_to_free_buffers() cannot release buffers and
page under memory pressure, which is more likely to trigger
out-of-memory. So we cannot remove the callback directly before we find
another way to release journal_head.

This patch introduce a shrinker to free journal_head on the checkpointed
transaction. After the journal_head got freed, try_to_free_buffers()
could free buffer properly.

Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Suggested-by: Jan Kara <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20210610112440.3438139-6-yi.zhang@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

3 files changed, 242 insertions, 0 deletions

diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index ad3919dbd49e..7ee2e21537e0 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -1174,6 +1174,7 @@ static void ext4_put_super(struct super_block *sb)
 	ext4_unregister_sysfs(sb);
 
 	if (sbi->s_journal) {
+		jbd2_journal_unregister_shrinker(sbi->s_journal);
 		aborted = is_journal_aborted(sbi->s_journal);
 		err = jbd2_journal_destroy(sbi->s_journal);
 		sbi->s_journal = NULL;
@@ -5186,6 +5187,7 @@ failed_mount_wq:
 	sbi->s_ea_block_cache = NULL;
 
 	if (sbi->s_journal) {
+		jbd2_journal_unregister_shrinker(sbi->s_journal);
 		jbd2_journal_destroy(sbi->s_journal);
 		sbi->s_journal = NULL;
 	}
@@ -5511,6 +5513,12 @@ static int ext4_load_journal(struct super_block *sb,
 		ext4_commit_super(sb);
 	}
 
+	err = jbd2_journal_register_shrinker(journal);
+	if (err) {
+		EXT4_SB(sb)->s_journal = NULL;
+		goto err_out;
+	}
+
 	return 0;
 
 err_out:
diff --git a/fs/jbd2/checkpoint.c b/fs/jbd2/checkpoint.c
index 75a4f622afaf..1abdae44a3d8 100644
--- a/fs/jbd2/checkpoint.c
+++ b/fs/jbd2/checkpoint.c
@@ -80,6 +80,18 @@ static inline void __buffer_relink_io(struct journal_head *jh)
 }
 
 /*
+ * Check a checkpoint buffer could be release or not.
+ *
+ * Requires j_list_lock
+ */
+static inline bool __cp_buffer_busy(struct journal_head *jh)
+{
+	struct buffer_head *bh = jh2bh(jh);
+
+	return (jh->b_transaction || buffer_locked(bh) || buffer_dirty(bh));
+}
+
+/*
  * Try to release a checkpointed buffer from its transaction.
  * Returns 1 if we released it and 2 if we also released the
  * whole transaction.
@@ -459,6 +471,137 @@ static int journal_clean_one_cp_list(struct journal_head *jh, bool destroy)
 }
 
 /*
+ * journal_shrink_one_cp_list
+ *
+ * Find 'nr_to_scan' written-back checkpoint buffers in the given list
+ * and try to release them. If the whole transaction is released, set
+ * the 'released' parameter. Return the number of released checkpointed
+ * buffers.
+ *
+ * Called with j_list_lock held.
+ */
+static unsigned long journal_shrink_one_cp_list(struct journal_head *jh,
+						unsigned long *nr_to_scan,
+						bool *released)
+{
+	struct journal_head *last_jh;
+	struct journal_head *next_jh = jh;
+	unsigned long nr_freed = 0;
+	int ret;
+
+	if (!jh || *nr_to_scan == 0)
+		return 0;
+
+	last_jh = jh->b_cpprev;
+	do {
+		jh = next_jh;
+		next_jh = jh->b_cpnext;
+
+		(*nr_to_scan)--;
+		if (__cp_buffer_busy(jh))
+			continue;
+
+		nr_freed++;
+		ret = __jbd2_journal_remove_checkpoint(jh);
+		if (ret) {
+			*released = true;
+			break;
+		}
+
+		if (need_resched())
+			break;
+	} while (jh != last_jh && *nr_to_scan);
+
+	return nr_freed;
+}
+
+/*
+ * jbd2_journal_shrink_checkpoint_list
+ *
+ * Find 'nr_to_scan' written-back checkpoint buffers in the journal
+ * and try to release them. Return the number of released checkpointed
+ * buffers.
+ *
+ * Called with j_list_lock held.
+ */
+unsigned long jbd2_journal_shrink_checkpoint_list(journal_t *journal,
+						  unsigned long *nr_to_scan)
+{
+	transaction_t *transaction, *last_transaction, *next_transaction;
+	bool released;
+	tid_t first_tid = 0, last_tid = 0, next_tid = 0;
+	tid_t tid = 0;
+	unsigned long nr_freed = 0;
+	unsigned long nr_scanned = *nr_to_scan;
+
+again:
+	spin_lock(&journal->j_list_lock);
+	if (!journal->j_checkpoint_transactions) {
+		spin_unlock(&journal->j_list_lock);
+		goto out;
+	}
+
+	/*
+	 * Get next shrink transaction, resume previous scan or start
+	 * over again. If some others do checkpoint and drop transaction
+	 * from the checkpoint list, we ignore saved j_shrink_transaction
+	 * and start over unconditionally.
+	 */
+	if (journal->j_shrink_transaction)
+		transaction = journal->j_shrink_transaction;
+	else
+		transaction = journal->j_checkpoint_transactions;
+
+	if (!first_tid)
+		first_tid = transaction->t_tid;
+	last_transaction = journal->j_checkpoint_transactions->t_cpprev;
+	next_transaction = transaction;
+	last_tid = last_transaction->t_tid;
+	do {
+		transaction = next_transaction;
+		next_transaction = transaction->t_cpnext;
+		tid = transaction->t_tid;
+		released = false;
+
+		nr_freed += journal_shrink_one_cp_list(transaction->t_checkpoint_list,
+						       nr_to_scan, &released);
+		if (*nr_to_scan == 0)
+			break;
+		if (need_resched() || spin_needbreak(&journal->j_list_lock))
+			break;
+		if (released)
+			continue;
+
+		nr_freed += journal_shrink_one_cp_list(transaction->t_checkpoint_io_list,
+						       nr_to_scan, &released);
+		if (*nr_to_scan == 0)
+			break;
+		if (need_resched() || spin_needbreak(&journal->j_list_lock))
+			break;
+	} while (transaction != last_transaction);
+
+	if (transaction != last_transaction) {
+		journal->j_shrink_transaction = next_transaction;
+		next_tid = next_transaction->t_tid;
+	} else {
+		journal->j_shrink_transaction = NULL;
+		next_tid = 0;
+	}
+
+	spin_unlock(&journal->j_list_lock);
+	cond_resched();
+
+	if (*nr_to_scan && next_tid)
+		goto again;
+out:
+	nr_scanned -= *nr_to_scan;
+	trace_jbd2_shrink_checkpoint_list(journal, first_tid, tid, last_tid,
+					  nr_freed, nr_scanned, next_tid);
+
+	return nr_freed;
+}
+
+/*
  * journal_clean_checkpoint_list
  *
  * Find all the written-back checkpoint buffers in the journal and release them.
@@ -580,6 +723,7 @@ int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
 
 	__buffer_unlink(jh);
 	jh->b_cp_transaction = NULL;
+	percpu_counter_dec(&journal->j_jh_shrink_count);
 	jbd2_journal_put_journal_head(jh);
 
 	/* Is this transaction empty? */
@@ -642,6 +786,7 @@ void __jbd2_journal_insert_checkpoint(struct journal_head *jh,
 		jh->b_cpnext->b_cpprev = jh;
 	}
 	transaction->t_checkpoint_list = jh;
+	percpu_counter_inc(&transaction->t_journal->j_jh_shrink_count);
 }
 
 /*
@@ -657,6 +802,8 @@ void __jbd2_journal_insert_checkpoint(struct journal_head *jh,
 void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction)
 {
 	assert_spin_locked(&journal->j_list_lock);
+
+	journal->j_shrink_transaction = NULL;
 	if (transaction->t_cpnext) {
 		transaction->t_cpnext->t_cpprev = transaction->t_cpprev;
 		transaction->t_cpprev->t_cpnext = transaction->t_cpnext;
diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
index 8b3f5bbd65f9..7c52feb6f753 100644
--- a/fs/jbd2/journal.c
+++ b/fs/jbd2/journal.c
@@ -2051,6 +2051,91 @@ recovery_error:
 }
 
 /**
+ * jbd2_journal_shrink_scan()
+ *
+ * Scan the checkpointed buffer on the checkpoint list and release the
+ * journal_head.
+ */
+static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
+					      struct shrink_control *sc)
+{
+	journal_t *journal = container_of(shrink, journal_t, j_shrinker);
+	unsigned long nr_to_scan = sc->nr_to_scan;
+	unsigned long nr_shrunk;
+	unsigned long count;
+
+	count = percpu_counter_read_positive(&journal->j_jh_shrink_count);
+	trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count);
+
+	nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan);
+
+	count = percpu_counter_read_positive(&journal->j_jh_shrink_count);
+	trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
+
+	return nr_shrunk;
+}
+
+/**
+ * jbd2_journal_shrink_count()
+ *
+ * Count the number of checkpoint buffers on the checkpoint list.
+ */
+static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink,
+					       struct shrink_control *sc)
+{
+	journal_t *journal = container_of(shrink, journal_t, j_shrinker);
+	unsigned long count;
+
+	count = percpu_counter_read_positive(&journal->j_jh_shrink_count);
+	trace_jbd2_shrink_count(journal, sc->nr_to_scan, count);
+
+	return count;
+}
+
+/**
+ * jbd2_journal_register_shrinker()
+ * @journal: Journal to act on.
+ *
+ * Init a percpu counter to record the checkpointed buffers on the checkpoint
+ * list and register a shrinker to release their journal_head.
+ */
+int jbd2_journal_register_shrinker(journal_t *journal)
+{
+	int err;
+
+	journal->j_shrink_transaction = NULL;
+
+	err = percpu_counter_init(&journal->j_jh_shrink_count, 0, GFP_KERNEL);
+	if (err)
+		return err;
+
+	journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan;
+	journal->j_shrinker.count_objects = jbd2_journal_shrink_count;
+	journal->j_shrinker.seeks = DEFAULT_SEEKS;
+	journal->j_shrinker.batch = journal->j_max_transaction_buffers;
+
+	err = register_shrinker(&journal->j_shrinker);
+	if (err) {
+		percpu_counter_destroy(&journal->j_jh_shrink_count);
+		return err;
+	}
+
+	return 0;
+}
+
+/**
+ * jbd2_journal_unregister_shrinker()
+ * @journal: Journal to act on.
+ *
+ * Unregister the checkpointed buffer shrinker and destroy the percpu counter.
+ */
+void jbd2_journal_unregister_shrinker(journal_t *journal)
+{
+	percpu_counter_destroy(&journal->j_jh_shrink_count);
+	unregister_shrinker(&journal->j_shrinker);
+}
+
+/**
  * jbd2_journal_destroy() - Release a journal_t structure.
  * @journal: Journal to act on.
  *
@@ -2122,6 +2207,8 @@ int jbd2_journal_destroy(journal_t *journal)
 		brelse(journal->j_sb_buffer);
 	}
 
+	jbd2_journal_unregister_shrinker(journal);
+
 	if (journal->j_proc_entry)
 		jbd2_stats_proc_exit(journal);
 	iput(journal->j_inode);