[PATCH 1/7] fs/aio: Add explicit RCU grace period when freeing kioctx

[PATCH 1/7] fs/aio: Add explicit RCU grace period when freeing kioctx

[Tejun Heo]

While fixing refcounting, e34ecee2ae79 ("aio: Fix a trinity splat")
incorrectly removed explicit RCU grace period before freeing kioctx.
The intention seems to be depending on the internal RCU grace periods
of percpu_ref; however, percpu_ref uses a different flavor of RCU,
sched-RCU.  This can lead to kioctx being freed while RCU read
protected dereferences are still in progress.

Fix it by updating free_ioctx() to go through call_rcu() explicitly.

v2: Comment added to explain double bouncing.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Jann Horn <jannh@google.com>
Fixes: e34ecee2ae79 ("aio: Fix a trinity splat")
Cc: Kent Overstreet <kent.overstreet@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: stable@vger.kernel.org
---
 fs/aio.c | 23 +++++++++++++++++++----
 1 file changed, 19 insertions(+), 4 deletions(-)

diff --git a/fs/aio.c b/fs/aio.c
index a062d75..eb2e0cf 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -115,7 +115,8 @@ struct kioctx {
 	struct page		**ring_pages;
 	long			nr_pages;
 
-	struct work_struct	free_work;
+	struct rcu_head		free_rcu;
+	struct work_struct	free_work;	/* see free_ioctx() */
 
 	/*
 	 * signals when all in-flight requests are done
@@ -588,6 +589,12 @@ static int kiocb_cancel(struct aio_kiocb *kiocb)
 	return cancel(&kiocb->common);
 }
 
+/*
+ * free_ioctx() should be RCU delayed to synchronize against the RCU
+ * protected lookup_ioctx() and also needs process context to call
+ * aio_free_ring(), so the double bouncing through kioctx->free_rcu and
+ * ->free_work.
+ */
 static void free_ioctx(struct work_struct *work)
 {
 	struct kioctx *ctx = container_of(work, struct kioctx, free_work);
@@ -601,6 +608,14 @@ static void free_ioctx(struct work_struct *work)
 	kmem_cache_free(kioctx_cachep, ctx);
 }
 
+static void free_ioctx_rcufn(struct rcu_head *head)
+{
+	struct kioctx *ctx = container_of(head, struct kioctx, free_rcu);
+
+	INIT_WORK(&ctx->free_work, free_ioctx);
+	schedule_work(&ctx->free_work);
+}
+
 static void free_ioctx_reqs(struct percpu_ref *ref)
 {
 	struct kioctx *ctx = container_of(ref, struct kioctx, reqs);
@@ -609,8 +624,8 @@ static void free_ioctx_reqs(struct percpu_ref *ref)
 	if (ctx->rq_wait && atomic_dec_and_test(&ctx->rq_wait->count))
 		complete(&ctx->rq_wait->comp);
 
-	INIT_WORK(&ctx->free_work, free_ioctx);
-	schedule_work(&ctx->free_work);
+	/* Synchronize against RCU protected table->table[] dereferences */
+	call_rcu(&ctx->free_rcu, free_ioctx_rcufn);
 }
 
 /*
@@ -838,7 +853,7 @@ static int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx,
 	table->table[ctx->id] = NULL;
 	spin_unlock(&mm->ioctx_lock);
 
-	/* percpu_ref_kill() will do the necessary call_rcu() */
+	/* free_ioctx_reqs() will do the necessary RCU synchronization */
 	wake_up_all(&ctx->wait);
 
 	/*
-- 
2.9.5

[PATCH 2/7] fs/aio: Use RCU accessors for kioctx_table->table[]

[Tejun Heo]

While converting ioctx index from a list to a table, db446a08c23d
("aio: convert the ioctx list to table lookup v3") missed tagging
kioctx_table->table[] as an array of RCU pointers and using the
appropriate RCU accessors.  This introduces a small window in the
lookup path where init and access may race.

Mark kioctx_table->table[] with __rcu and use the approriate RCU
accessors when using the field.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Jann Horn <jannh@google.com>
Fixes: db446a08c23d ("aio: convert the ioctx list to table lookup v3")
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: stable@vger.kernel.org
---
 fs/aio.c | 21 +++++++++++----------
 1 file changed, 11 insertions(+), 10 deletions(-)

diff --git a/fs/aio.c b/fs/aio.c
index eb2e0cf..6bcd3fb 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -68,9 +68,9 @@ struct aio_ring {
 #define AIO_RING_PAGES	8
 
 struct kioctx_table {
-	struct rcu_head	rcu;
-	unsigned	nr;
-	struct kioctx	*table[];
+	struct rcu_head		rcu;
+	unsigned		nr;
+	struct kioctx __rcu	*table[];
 };
 
 struct kioctx_cpu {
@@ -330,7 +330,7 @@ static int aio_ring_mremap(struct vm_area_struct *vma)
 	for (i = 0; i < table->nr; i++) {
 		struct kioctx *ctx;
 
-		ctx = table->table[i];
+		ctx = rcu_dereference(table->table[i]);
 		if (ctx && ctx->aio_ring_file == file) {
 			if (!atomic_read(&ctx->dead)) {
 				ctx->user_id = ctx->mmap_base = vma->vm_start;
@@ -666,9 +666,9 @@ static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm)
 	while (1) {
 		if (table)
 			for (i = 0; i < table->nr; i++)
-				if (!table->table[i]) {
+				if (!rcu_access_pointer(table->table[i])) {
 					ctx->id = i;
-					table->table[i] = ctx;
+					rcu_assign_pointer(table->table[i], ctx);
 					spin_unlock(&mm->ioctx_lock);
 
 					/* While kioctx setup is in progress,
@@ -849,8 +849,8 @@ static int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx,
 	}
 
 	table = rcu_dereference_raw(mm->ioctx_table);
-	WARN_ON(ctx != table->table[ctx->id]);
-	table->table[ctx->id] = NULL;
+	WARN_ON(ctx != rcu_access_pointer(table->table[ctx->id]));
+	RCU_INIT_POINTER(table->table[ctx->id], NULL);
 	spin_unlock(&mm->ioctx_lock);
 
 	/* free_ioctx_reqs() will do the necessary RCU synchronization */
@@ -895,7 +895,8 @@ void exit_aio(struct mm_struct *mm)
 
 	skipped = 0;
 	for (i = 0; i < table->nr; ++i) {
-		struct kioctx *ctx = table->table[i];
+		struct kioctx *ctx =
+			rcu_dereference_protected(table->table[i], true);
 
 		if (!ctx) {
 			skipped++;
@@ -1084,7 +1085,7 @@ static struct kioctx *lookup_ioctx(unsigned long ctx_id)
 	if (!table || id >= table->nr)
 		goto out;
 
-	ctx = table->table[id];
+	ctx = rcu_dereference(table->table[id]);
 	if (ctx && ctx->user_id == ctx_id) {
 		percpu_ref_get(&ctx->users);
 		ret = ctx;
-- 
2.9.5