[PATCH] scsi: sg: Fix mismatch in sg_get_rq_mark

From: firogm
Date: Tue May 16 2017 - 10:52:53 EST


From: Firo Yang <firogm@xxxxxxxxx>

This bug was reported by Andrey Konovalov with syzkaller:

Call Trace:
sg_finish_rem_req+0x2a6/0x320 drivers/scsi/sg.c:1839
sg_new_read+0x3c/0x430 drivers/scsi/sg.c:567
sg_read+0x866/0x1830 drivers/scsi/sg.c:455
__vfs_read+0x5c3/0x750 fs/read_write.c:450
vfs_read+0x11e/0x350 fs/read_write.c:473
SYSC_read fs/read_write.c:589
SyS_read+0xfb/0x230 fs/read_write.c:582
entry_SYSCALL_64_fastpath+0x1f/0xbe arch/x86/entry/entry_64.S:204

If there is nothing matched in the loop of list_for_each_entry,
at last the iteration variable 'resp' will point to
list_entry(&sfp->rq_list, typeof(resp), entry); it's not a valid
Sg_request. Finnally, 'resp' will be wrongly returned to the caller
function.

To fix it, add another return variable to record the correct
match and return it to the caller function.

Signed-off-by: Firo Yang <firogm@xxxxxxxxx>
---
drivers/scsi/sg.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c
index 0a38ba0..74a9188 100644
--- a/drivers/scsi/sg.c
+++ b/drivers/scsi/sg.c
@@ -2065,7 +2065,7 @@ sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
static Sg_request *
sg_get_rq_mark(Sg_fd * sfp, int pack_id)
{
- Sg_request *resp;
+ Sg_request *resp, *ret_resp = NULL;
unsigned long iflags;

write_lock_irqsave(&sfp->rq_list_lock, iflags);
@@ -2074,11 +2074,12 @@ sg_get_rq_mark(Sg_fd * sfp, int pack_id)
if ((1 == resp->done) && (!resp->sg_io_owned) &&
((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
resp->done = 2; /* guard against other readers */
+ ret_resp = resp;
break;
}
}
write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
- return resp;
+ return ret_resp;
}

/* always adds to end of list */
--
2.9.4