linux-kernel - [PATCH 4.12 14/65] usb-storage: fix deadlock involving host lock and scsi

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <20170815011942.957515846@linuxfoundation.org>
Date:   Mon, 14 Aug 2017 18:19:05 -0700
From:   Greg Kroah-Hartman <gregkh@...uxfoundation.org>
To:     linux-kernel@...r.kernel.org
Cc:     Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        stable@...r.kernel.org, Alan Stern <stern@...land.harvard.edu>,
        Christoph Hellwig <hch@....de>
Subject: [PATCH 4.12 14/65] usb-storage: fix deadlock involving host lock and scsi_done

4.12-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Alan Stern <stern@...land.harvard.edu>

commit 8b52291a0743fc4db4a7495c846a6f31ee84d282 upstream.

Christoph Hellwig says that since version 4.12, the kernel switched to
using blk-mq by default.  The old code used a softirq for handling
request completions, but blk-mq can handle completions in the caller's
context.  This may cause a problem for usb-storage, because it invokes
the ->scsi_done callback while holding the host lock, and the
completion routine sometimes tries to acquire the same lock (when
running the error handler, for example).

The consequence is that the existing code will sometimes deadlock upon
error completion of a SCSI command (with a lockdep warning).

This is easy enough to fix, since usb-storage doesn't really need to
hold the host lock while the callback runs.  It was simpler to write
it that way, but moving the call outside the locked region is pretty
easy and there's no downside.  That's what this patch does.

Signed-off-by: Alan Stern <stern@...land.harvard.edu>
Reported-and-tested-by: Arthur Marsh <arthur.marsh@...ernode.on.net>
CC: Christoph Hellwig <hch@....de>
Signed-off-by: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@...uxfoundation.org>

---
 drivers/usb/storage/usb.c |   18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)

--- a/drivers/usb/storage/usb.c
+++ b/drivers/usb/storage/usb.c
@@ -315,6 +315,7 @@ static int usb_stor_control_thread(void
 {
 	struct us_data *us = (struct us_data *)__us;
 	struct Scsi_Host *host = us_to_host(us);
+	struct scsi_cmnd *srb;
 
 	for (;;) {
 		usb_stor_dbg(us, "*** thread sleeping\n");
@@ -330,6 +331,7 @@ static int usb_stor_control_thread(void
 		scsi_lock(host);
 
 		/* When we are called with no command pending, we're done */
+		srb = us->srb;
 		if (us->srb == NULL) {
 			scsi_unlock(host);
 			mutex_unlock(&us->dev_mutex);
@@ -398,14 +400,11 @@ static int usb_stor_control_thread(void
 		/* lock access to the state */
 		scsi_lock(host);
 
-		/* indicate that the command is done */
-		if (us->srb->result != DID_ABORT << 16) {
-			usb_stor_dbg(us, "scsi cmd done, result=0x%x\n",
-				     us->srb->result);
-			us->srb->scsi_done(us->srb);
-		} else {
+		/* was the command aborted? */
+		if (us->srb->result == DID_ABORT << 16) {
 SkipForAbort:
 			usb_stor_dbg(us, "scsi command aborted\n");
+			srb = NULL;	/* Don't call srb->scsi_done() */
 		}
 
 		/*
@@ -429,6 +428,13 @@ SkipForAbort:
 
 		/* unlock the device pointers */
 		mutex_unlock(&us->dev_mutex);
+
+		/* now that the locks are released, notify the SCSI core */
+		if (srb) {
+			usb_stor_dbg(us, "scsi cmd done, result=0x%x\n",
+					srb->result);
+			srb->scsi_done(srb);
+		}
 	} /* for (;;) */
 
 	/* Wait until we are told to stop */