[PATCH resend] jfs: fix use-after-free in lbmIODone

[PATCH resend] jfs: fix use-after-free in lbmIODone

[Anirudh Rayabharam]

Fix use-after-free by waiting for ongoing IO to complete before freeing
lbufs in lbmLogShutdown. Add a counter in struct jfs_log to keep track
of the number of in-flight IO operations and a wait queue to wait on for
the IO operations to complete.

Reported-by: syzbot+5d2008bd1f1b722ba94e@syzkaller.appspotmail.com
Suggested-by: Hillf Danton <hdanton@sina.com>
Signed-off-by: Anirudh Rayabharam <mail@anirudhrb.com>
---
 fs/jfs/jfs_logmgr.c | 17 ++++++++++++++---
 fs/jfs/jfs_logmgr.h |  2 ++
 2 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/fs/jfs/jfs_logmgr.c b/fs/jfs/jfs_logmgr.c
index 9330eff210e0..82d20c4687aa 100644
--- a/fs/jfs/jfs_logmgr.c
+++ b/fs/jfs/jfs_logmgr.c
@@ -1815,6 +1815,8 @@ static int lbmLogInit(struct jfs_log * log)
 	 */
 	init_waitqueue_head(&log->free_wait);
 
+	init_waitqueue_head(&log->io_waitq);
+
 	log->lbuf_free = NULL;
 
 	for (i = 0; i < LOGPAGES;) {
@@ -1864,6 +1866,7 @@ static void lbmLogShutdown(struct jfs_log * log)
 	struct lbuf *lbuf;
 
 	jfs_info("lbmLogShutdown: log:0x%p", log);
+	wait_event(log->io_waitq, !atomic_read(&log->io_inflight));
 
 	lbuf = log->lbuf_free;
 	while (lbuf) {
@@ -1990,6 +1993,8 @@ static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
 	bio->bi_end_io = lbmIODone;
 	bio->bi_private = bp;
 	bio->bi_opf = REQ_OP_READ;
+
+	atomic_inc(&log->io_inflight);
 	/*check if journaling to disk has been disabled*/
 	if (log->no_integrity) {
 		bio->bi_iter.bi_size = 0;
@@ -2135,6 +2140,7 @@ static void lbmStartIO(struct lbuf * bp)
 	bio->bi_private = bp;
 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
 
+	atomic_inc(&log->io_inflight);
 	/* check if journaling to disk has been disabled */
 	if (log->no_integrity) {
 		bio->bi_iter.bi_size = 0;
@@ -2200,6 +2206,8 @@ static void lbmIODone(struct bio *bio)
 
 	bio_put(bio);
 
+	log = bp->l_log;
+
 	/*
 	 *	pagein completion
 	 */
@@ -2211,7 +2219,7 @@ static void lbmIODone(struct bio *bio)
 		/* wakeup I/O initiator */
 		LCACHE_WAKEUP(&bp->l_ioevent);
 
-		return;
+		goto out;
 	}
 
 	/*
@@ -2230,13 +2238,12 @@ static void lbmIODone(struct bio *bio)
 	INCREMENT(lmStat.pagedone);
 
 	/* update committed lsn */
-	log = bp->l_log;
 	log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor;
 
 	if (bp->l_flag & lbmDIRECT) {
 		LCACHE_WAKEUP(&bp->l_ioevent);
 		LCACHE_UNLOCK(flags);
-		return;
+		goto out;
 	}
 
 	tail = log->wqueue;
@@ -2315,6 +2322,10 @@ static void lbmIODone(struct bio *bio)
 
 		LCACHE_UNLOCK(flags);	/* unlock+enable */
 	}
+
+out:
+	if (atomic_dec_and_test(&log->io_inflight))
+		wake_up(&log->io_waitq);
 }
 
 int jfsIOWait(void *arg)
diff --git a/fs/jfs/jfs_logmgr.h b/fs/jfs/jfs_logmgr.h
index 805877ce5020..3e92fe251f28 100644
--- a/fs/jfs/jfs_logmgr.h
+++ b/fs/jfs/jfs_logmgr.h
@@ -400,6 +400,8 @@ struct jfs_log {
 	uuid_t uuid;		/* 16: 128-bit uuid of log device */
 
 	int no_integrity;	/* 3: flag to disable journaling to disk */
+	atomic_t io_inflight;
+	wait_queue_head_t io_waitq;
 };
 
 /*
-- 
2.26.2

Re: [PATCH resend] jfs: fix use-after-free in lbmIODone

[Greg KH]

On Mon, Mar 22, 2021 at 09:41:47PM +0530, Anirudh Rayabharam wrote:
> Fix use-after-free by waiting for ongoing IO to complete before freeing
> lbufs in lbmLogShutdown. Add a counter in struct jfs_log to keep track
> of the number of in-flight IO operations and a wait queue to wait on for
> the IO operations to complete.
> 
> Reported-by: syzbot+5d2008bd1f1b722ba94e@syzkaller.appspotmail.com
> Suggested-by: Hillf Danton <hdanton@sina.com>
> Signed-off-by: Anirudh Rayabharam <mail@anirudhrb.com>
> ---
>  fs/jfs/jfs_logmgr.c | 17 ++++++++++++++---
>  fs/jfs/jfs_logmgr.h |  2 ++
>  2 files changed, 16 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/jfs/jfs_logmgr.c b/fs/jfs/jfs_logmgr.c
> index 9330eff210e0..82d20c4687aa 100644
> --- a/fs/jfs/jfs_logmgr.c
> +++ b/fs/jfs/jfs_logmgr.c
> @@ -1815,6 +1815,8 @@ static int lbmLogInit(struct jfs_log * log)
>  	 */
>  	init_waitqueue_head(&log->free_wait);
>  
> +	init_waitqueue_head(&log->io_waitq);
> +
>  	log->lbuf_free = NULL;
>  
>  	for (i = 0; i < LOGPAGES;) {
> @@ -1864,6 +1866,7 @@ static void lbmLogShutdown(struct jfs_log * log)
>  	struct lbuf *lbuf;
>  
>  	jfs_info("lbmLogShutdown: log:0x%p", log);
> +	wait_event(log->io_waitq, !atomic_read(&log->io_inflight));
>  
>  	lbuf = log->lbuf_free;
>  	while (lbuf) {
> @@ -1990,6 +1993,8 @@ static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
>  	bio->bi_end_io = lbmIODone;
>  	bio->bi_private = bp;
>  	bio->bi_opf = REQ_OP_READ;
> +
> +	atomic_inc(&log->io_inflight);
>  	/*check if journaling to disk has been disabled*/
>  	if (log->no_integrity) {
>  		bio->bi_iter.bi_size = 0;
> @@ -2135,6 +2140,7 @@ static void lbmStartIO(struct lbuf * bp)
>  	bio->bi_private = bp;
>  	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
>  
> +	atomic_inc(&log->io_inflight);

Why use an atomic for this?  The value can change after you test for it,
as there's no lock involved.

Do you really need to keep track of all of these "inflight"?  That feels
very "heavy" to me.

jfs developers, any ideas?

thanks,

greg k-h