]> git.ipfire.org Git - thirdparty/kernel/stable-queue.git/commitdiff
4.9-stable patches
authorGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 15 Jan 2019 10:38:22 +0000 (11:38 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 15 Jan 2019 10:38:22 +0000 (11:38 +0100)
added patches:
sunrpc-use-after-free-in-svc_process_common.patch

queue-4.9/series
queue-4.9/sunrpc-use-after-free-in-svc_process_common.patch [new file with mode: 0644]

index 4eccfed49cf1909cbc422e29e83e9087e122aae2..8313b491c5321520f562ba8d54cf258c9ecfbc22 100644 (file)
@@ -13,3 +13,4 @@ rbd-don-t-return-0-on-unmap-if-rbd_dev_flag_removing-is-set.patch
 ext4-make-sure-enough-credits-are-reserved-for-dioread_nolock-writes.patch
 ext4-fix-a-potential-fiemap-page-fault-deadlock-w-inline_data.patch
 ext4-avoid-kernel-warning-when-writing-the-superblock-to-a-dead-device.patch
+sunrpc-use-after-free-in-svc_process_common.patch
diff --git a/queue-4.9/sunrpc-use-after-free-in-svc_process_common.patch b/queue-4.9/sunrpc-use-after-free-in-svc_process_common.patch
new file mode 100644 (file)
index 0000000..da2d230
--- /dev/null
@@ -0,0 +1,146 @@
+From d4b09acf924b84bae77cad090a9d108e70b43643 Mon Sep 17 00:00:00 2001
+From: Vasily Averin <vvs@virtuozzo.com>
+Date: Mon, 24 Dec 2018 14:44:52 +0300
+Subject: sunrpc: use-after-free in svc_process_common()
+
+From: Vasily Averin <vvs@virtuozzo.com>
+
+commit d4b09acf924b84bae77cad090a9d108e70b43643 upstream.
+
+if node have NFSv41+ mounts inside several net namespaces
+it can lead to use-after-free in svc_process_common()
+
+svc_process_common()
+        /* Setup reply header */
+        rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp); <<< HERE
+
+svc_process_common() can use incorrect rqstp->rq_xprt,
+its caller function bc_svc_process() takes it from serv->sv_bc_xprt.
+The problem is that serv is global structure but sv_bc_xprt
+is assigned per-netnamespace.
+
+According to Trond, the whole "let's set up rqstp->rq_xprt
+for the back channel" is nothing but a giant hack in order
+to work around the fact that svc_process_common() uses it
+to find the xpt_ops, and perform a couple of (meaningless
+for the back channel) tests of xpt_flags.
+
+All we really need in svc_process_common() is to be able to run
+rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr()
+
+Bruce J Fields points that this xpo_prep_reply_hdr() call
+is an awfully roundabout way just to do "svc_putnl(resv, 0);"
+in the tcp case.
+
+This patch does not initialiuze rqstp->rq_xprt in bc_svc_process(),
+now it calls svc_process_common() with rqstp->rq_xprt = NULL.
+
+To adjust reply header svc_process_common() just check
+rqstp->rq_prot and calls svc_tcp_prep_reply_hdr() for tcp case.
+
+To handle rqstp->rq_xprt = NULL case in functions called from
+svc_process_common() patch intruduces net namespace pointer
+svc_rqst->rq_bc_net and adjust SVC_NET() definition.
+Some other function was also adopted to properly handle described case.
+
+Signed-off-by: Vasily Averin <vvs@virtuozzo.com>
+Cc: stable@vger.kernel.org
+Fixes: 23c20ecd4475 ("NFS: callback up - users counting cleanup")
+Signed-off-by: J. Bruce Fields <bfields@redhat.com>
+v2: - added lost extern svc_tcp_prep_reply_hdr()
+    - dropped trace_svc_process() changes
+    - context fixes in svc_process_common()
+Signed-off-by: Vasily Averin <vvs@virtuozzo.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+---
+ include/linux/sunrpc/svc.h |    5 ++++-
+ net/sunrpc/svc.c           |    9 ++++++---
+ net/sunrpc/svc_xprt.c      |    5 +++--
+ net/sunrpc/svcsock.c       |    2 +-
+ 4 files changed, 14 insertions(+), 7 deletions(-)
+
+--- a/include/linux/sunrpc/svc.h
++++ b/include/linux/sunrpc/svc.h
+@@ -291,9 +291,12 @@ struct svc_rqst {
+       struct svc_cacherep *   rq_cacherep;    /* cache info */
+       struct task_struct      *rq_task;       /* service thread */
+       spinlock_t              rq_lock;        /* per-request lock */
++      struct net              *rq_bc_net;     /* pointer to backchannel's
++                                               * net namespace
++                                               */
+ };
+-#define SVC_NET(svc_rqst)     (svc_rqst->rq_xprt->xpt_net)
++#define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net)
+ /*
+  * Rigorous type checking on sockaddr type conversions
+--- a/net/sunrpc/svc.c
++++ b/net/sunrpc/svc.c
+@@ -1137,6 +1137,8 @@ void svc_printk(struct svc_rqst *rqstp,
+ static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
+ #endif
++extern void svc_tcp_prep_reply_hdr(struct svc_rqst *);
++
+ /*
+  * Common routine for processing the RPC request.
+  */
+@@ -1166,7 +1168,8 @@ svc_process_common(struct svc_rqst *rqst
+       clear_bit(RQ_DROPME, &rqstp->rq_flags);
+       /* Setup reply header */
+-      rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
++      if (rqstp->rq_prot == IPPROTO_TCP)
++              svc_tcp_prep_reply_hdr(rqstp);
+       svc_putu32(resv, rqstp->rq_xid);
+@@ -1312,7 +1315,7 @@ svc_process_common(struct svc_rqst *rqst
+       return 0;
+  close:
+-      if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
++      if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
+               svc_close_xprt(rqstp->rq_xprt);
+       dprintk("svc: svc_process close\n");
+       return 0;
+@@ -1439,10 +1442,10 @@ bc_svc_process(struct svc_serv *serv, st
+       dprintk("svc: %s(%p)\n", __func__, req);
+       /* Build the svc_rqst used by the common processing routine */
+-      rqstp->rq_xprt = serv->sv_bc_xprt;
+       rqstp->rq_xid = req->rq_xid;
+       rqstp->rq_prot = req->rq_xprt->prot;
+       rqstp->rq_server = serv;
++      rqstp->rq_bc_net = req->rq_xprt->xprt_net;
+       rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
+       memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
+--- a/net/sunrpc/svc_xprt.c
++++ b/net/sunrpc/svc_xprt.c
+@@ -510,10 +510,11 @@ out:
+  */
+ void svc_reserve(struct svc_rqst *rqstp, int space)
+ {
++      struct svc_xprt *xprt = rqstp->rq_xprt;
++
+       space += rqstp->rq_res.head[0].iov_len;
+-      if (space < rqstp->rq_reserved) {
+-              struct svc_xprt *xprt = rqstp->rq_xprt;
++      if (xprt && space < rqstp->rq_reserved) {
+               atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
+               rqstp->rq_reserved = space;
+--- a/net/sunrpc/svcsock.c
++++ b/net/sunrpc/svcsock.c
+@@ -1195,7 +1195,7 @@ static int svc_tcp_sendto(struct svc_rqs
+ /*
+  * Setup response header. TCP has a 4B record length field.
+  */
+-static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
++void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
+ {
+       struct kvec *resv = &rqstp->rq_res.head[0];