FAILED: patch "[PATCH] SUNRPC: Copy the deferred RPC Call from the head buffer" failed to apply to 5.10-stable tree

The patch below does not apply to the 5.10-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

To reproduce the conflict and resubmit, you may use the following commands:

git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-5.10.y
git checkout FETCH_HEAD
git cherry-pick -x 2192a56b0686599358909c3e12e9a366f5b5d452
# <resolve conflicts, build, test, etc.>
git commit -s
git send-email --to '<[email protected]>' --in-reply-to '2026110458-walnut-distant-f403@gregkh' --subject-prefix 'PATCH 5.10.y' 'HEAD^..'

Possible dependencies:

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From 2192a56b0686599358909c3e12e9a366f5b5d452 Mon Sep 17 00:00:00 2001
From: Chuck Lever <[email protected]>
Date: Wed, 2 Sep 2026 09:31:01 -0400
Subject: [PATCH] SUNRPC: Copy the deferred RPC Call from the head buffer

svc_defer() snapshots the RPC Call for replay once a pending cache upcall resolves. It copies the Call starting (rq_arg.len - head[0].iov_len) bytes before head[0].iov_base, a back-up that once covered a transport header. No transport has had one since commit 983084b2672c ("SUNRPC: Remove svc_rqst::rq_xprt_hlen"), so the two lengths are meant to be equal.

An RPCSEC_GSS Call breaks that assumption. Unwrapping the integrity service truncates rq_arg.len by the length of the trailing checksum, and unwrapping the privacy service leaves rq_arg.len short by an amount the client sets with trailing bytes after the wrap token. Neither adjusts head[0].iov_len. Both lengths are size_t, so the subtraction underflows and the copy starts past head[0].iov_base. Every deferred integrity Call replays a shifted snapshot.

A deferred replay skips GSS verification, so a shifted snapshot runs as an authenticated Call. With the privacy service the offset is client-chosen. A forged Call placed in that window, naming another user's GSS context handle, runs under that identity.

Copy the Call from head[0].iov_base. Refuse to defer a Call whose length runs past the head buffer, so the copy cannot read out of bounds.

index 68310521ddb9..d5634dd6d6cc 100644
@@ -1299,13 +1299,8 @@ static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
 }
 
 /*
- * Save the request off for later processing. The request buffer looks
- * like this:
- *
- * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
- *
- * This code can only handle requests that consist of an xprt-header
- * and rpc-header.
+ * Save the request off for later processing. Only a Call that fits
+ * entirely in rq_arg.head[0] can be deferred.
  */
 static struct cache_deferred_req *svc_defer(struct cache_req *req)
 {

@@ -1318,8 +1313,11 @@ static struct cache_deferred_req *svc_defer(struct cache_req *req)
 		dr = rqstp->rq_deferred;
 		rqstp->rq_deferred = NULL;
 	} else {
-		size_t skip;
 		size_t size;
+
+		if (rqstp->rq_arg.len > rqstp->rq_arg.head[0].iov_len)
+			return NULL;
+
 		/* FIXME maybe discard if size too large */
 		size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
 		dr = kmalloc(size, GFP_KERNEL);

@@ -1333,9 +1331,7 @@ static struct cache_deferred_req *svc_defer(struct cache_req *req)
 		dr->daddr = rqstp->rq_daddr;
 		dr->argslen = rqstp->rq_arg.len >> 2;
 
-		/* back up head to the start of the buffer and copy */
-		skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
-		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
+		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base,
 		       dr->argslen << 2);
 	}
 	dr->xprt_ctxt = rqstp->rq_xprt_ctxt;

@@ -1358,17 +1354,13 @@ static noinline int svc_deferred_recv(struct svc_rqst *rqstp)
 
 	trace_svc_defer_recv(dr);
 
-	/* setup iov_base past transport header */
 	rqstp->rq_arg.head[0].iov_base = dr->args;
-	/* The iov_len does not include the transport header bytes */
 	rqstp->rq_arg.head[0].iov_len = dr->argslen << 2;
 	rqstp->rq_arg.page_len = 0;
-	/* The rq_arg.len includes the transport header bytes */
 	rqstp->rq_arg.len     = dr->argslen << 2;
 	rqstp->rq_prot        = dr->prot;
 	memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
 	rqstp->rq_addrlen     = dr->addrlen;
-	/* Save off transport header len in case we get deferred again */
 	rqstp->rq_daddr       = dr->daddr;
 	rqstp->rq_xprt_ctxt   = dr->xprt_ctxt;
 

From: Chuck Lever <[email protected]>

svc_defer() snapshots the RPC Call for replay once a pending cache upcall resolves. It copies the Call starting (rq_arg.len - head[0].iov_len) bytes before head[0].iov_base, a back-up that once covered a transport header. No transport has had one since commit 983084b2672c ("SUNRPC: Remove svc_rqst::rq_xprt_hlen"), so the two lengths are meant to be equal.

An RPCSEC_GSS Call breaks that assumption. Unwrapping the integrity service truncates rq_arg.len by the length of the trailing checksum, and unwrapping the privacy service leaves rq_arg.len short by an amount the client sets with trailing bytes after the wrap token. Neither adjusts head[0].iov_len. Both lengths are size_t, so the subtraction underflows and the copy starts past head[0].iov_base. Every deferred integrity Call replays a shifted snapshot.

A deferred replay skips GSS verification, so a shifted snapshot runs as an authenticated Call. With the privacy service the offset is client-chosen. A forged Call placed in that window, naming another user's GSS context handle, runs under that identity.

Copy the Call from head[0].iov_base. Refuse to defer a Call whose length runs past the head buffer, so the copy cannot read out of bounds.

---
 net/sunrpc/svc_xprt.c | 18 +++++++-----------
 1 file changed, 7 insertions(+), 11 deletions(-)

index 60782504ad3e7..a8bd4f7f9299d 100644
@@ -1181,13 +1181,8 @@ static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
 }
 
 /*
- * Save the request off for later processing. The request buffer looks
- * like this:
- *
- * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
- *
- * This code can only handle requests that consist of an xprt-header
- * and rpc-header.
+ * Save the request off for later processing. Only a Call that fits
+ * entirely in rq_arg.head[0] can be deferred.
  */
 static struct cache_deferred_req *svc_defer(struct cache_req *req)
 {

@@ -1200,8 +1195,11 @@ static struct cache_deferred_req *svc_defer(struct cache_req *req)
 		dr = rqstp->rq_deferred;
 		rqstp->rq_deferred = NULL;
 	} else {
-		size_t skip;
 		size_t size;
+
+		if (rqstp->rq_arg.len > rqstp->rq_arg.head[0].iov_len)
+			return NULL;
+
 		/* FIXME maybe discard if size too large */
 		size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
 		dr = kmalloc(size, GFP_KERNEL);

@@ -1216,9 +1214,7 @@ static struct cache_deferred_req *svc_defer(struct cache_req *req)
 		dr->argslen = rqstp->rq_arg.len >> 2;
 		dr->xprt_hlen = rqstp->rq_xprt_hlen;
 
-		/* back up head to the start of the buffer and copy */
-		skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
-		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
+		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base,
 		       dr->argslen << 2);
 	}
 	trace_svc_defer(rqstp);

2.50.1

From: Chuck Lever <[email protected]>

commit 2192a56b0686599358909c3e12e9a366f5b5d452 upstream.

svc_defer() snapshots the RPC Call for replay once a pending cache upcall resolves. It copies the Call starting (rq_arg.len - head[0].iov_len) bytes before head[0].iov_base, a back-up that once covered a transport header. No transport has had one since commit 983084b2672c ("SUNRPC: Remove svc_rqst::rq_xprt_hlen"), so the two lengths are meant to be equal.

An RPCSEC_GSS Call breaks that assumption. Unwrapping the integrity service truncates rq_arg.len by the length of the trailing checksum, and unwrapping the privacy service leaves rq_arg.len short by an amount the client sets with trailing bytes after the wrap token. Neither adjusts head[0].iov_len. Both lengths are size_t, so the subtraction underflows and the copy starts past head[0].iov_base. Every deferred integrity Call replays a shifted snapshot.

A deferred replay skips GSS verification, so a shifted snapshot runs as an authenticated Call. With the privacy service the offset is client-chosen. A forged Call placed in that window, naming another user's GSS context handle, runs under that identity.

Copy the Call from head[0].iov_base. Refuse to defer a Call whose length runs past the head buffer, so the copy cannot read out of bounds.

[doebel: 5.10 lacks 983084b2672c; keep xprt_hlen and its comments in
svc_deferred_recv(). The removed skip is zero for all 5.10 transports.]

---
v2:
- Add the "commit <sha1> upstream." line.
- Document why the backport differs from upstream.

 net/sunrpc/svc_xprt.c | 18 +++++++-----------
 1 file changed, 7 insertions(+), 11 deletions(-)

index 60782504ad3e7..a8bd4f7f9299d 100644
@@ -1181,13 +1181,8 @@ static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
 }
 
 /*
- * Save the request off for later processing. The request buffer looks
- * like this:
- *
- * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
- *
- * This code can only handle requests that consist of an xprt-header
- * and rpc-header.
+ * Save the request off for later processing. Only a Call that fits
+ * entirely in rq_arg.head[0] can be deferred.
  */
 static struct cache_deferred_req *svc_defer(struct cache_req *req)
 {

@@ -1200,8 +1195,11 @@ static struct cache_deferred_req *svc_defer(struct cache_req *req)
 		dr = rqstp->rq_deferred;
 		rqstp->rq_deferred = NULL;
 	} else {
-		size_t skip;
 		size_t size;
+
+		if (rqstp->rq_arg.len > rqstp->rq_arg.head[0].iov_len)
+			return NULL;
+
 		/* FIXME maybe discard if size too large */
 		size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
 		dr = kmalloc(size, GFP_KERNEL);

@@ -1216,9 +1214,7 @@ static struct cache_deferred_req *svc_defer(struct cache_req *req)
 		dr->argslen = rqstp->rq_arg.len >> 2;
 		dr->xprt_hlen = rqstp->rq_xprt_hlen;
 
-		/* back up head to the start of the buffer and copy */
-		skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
-		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
+		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base,
 		       dr->argslen << 2);
 	}
 	trace_svc_defer(rqstp);

2.50.1

On 5.10 the RPC header is already consumed from head[0] before svc_defer() runs, so the premise behind the new check in 2192a56b0686 ("SUNRPC: Copy the deferred RPC Call from the head buffer"), rq_arg.len == head[0].iov_len, doesn't hold. With it, nearly every request would be refused deferral.

Could you rework the 5.10 version so it doesn't change that behavior and send a v3?

Thanks,
Sasha

From: Chuck Lever <[email protected]>

commit 2192a56b0686599358909c3e12e9a366f5b5d452 upstream.

svc_defer() snapshots the RPC Call for replay once a pending cache upcall resolves. It copies the Call starting (rq_arg.len - head[0].iov_len) bytes before head[0].iov_base, a back-up that once covered a transport header. No transport has had one since commit 983084b2672c ("SUNRPC: Remove svc_rqst::rq_xprt_hlen"), so the two lengths are meant to be equal.

An RPCSEC_GSS Call breaks that assumption. Unwrapping the integrity service truncates rq_arg.len by the length of the trailing checksum, and unwrapping the privacy service leaves rq_arg.len short by an amount the client sets with trailing bytes after the wrap token. Neither adjusts head[0].iov_len. Both lengths are size_t, so the subtraction underflows and the copy starts past head[0].iov_base. Every deferred integrity Call replays a shifted snapshot.

A deferred replay skips GSS verification, so a shifted snapshot runs as an authenticated Call. With the privacy service the offset is client-chosen. A forged Call placed in that window, naming another user's GSS context handle, runs under that identity.

Copy the Call from head[0].iov_base. Refuse to defer a Call whose length runs past the head buffer, so the copy cannot read out of bounds.

[doebel: 5.10 backport. Here rq_arg.len > head[0].iov_len for every
Call at svc_defer() time, so the upstream guard and copy-from-iov_base cannot apply; keep the skip-based snapshot that svc_deferred_recv() expects (5.10 still has rq_xprt_hlen). Only gss_unwrap_kerberos_v2() breaks the len - iov_len invariant, by resetting buf->len; refuse to defer after a GSS integrity/privacy unwrap (rq_auth_slack != 0) or when len < head[0].iov_len.]

---
v3:
- Rework for 5.10's accounting per Sasha's feedback: keep the skip-based snapshot (5.10 still has rq_xprt_hlen and svc_deferred_recv() depends on it); refuse deferral only after a GSS integrity/privacy unwrap (rq_auth_slack != 0) or when len < head[0].iov_len, since gss_unwrap_kerberos_v2() resets buf->len and breaks the len - iov_len invariant.
v2:
- Add the "commit <sha1> upstream." line.
- Document why the backport differs from upstream.

 net/sunrpc/svc_xprt.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

index 60782504ad3e7..8970abdfaab46 100644
@@ -1202,6 +1202,20 @@ static struct cache_deferred_req *svc_defer(struct cache_req *req)
 	} else {
 		size_t skip;
 		size_t size;
+
+		/*
+		 * A GSS integrity/privacy unwrap can reset rq_arg.len so
+		 * it no longer measures from the start of the receive
+		 * buffer (see gss_unwrap_kerberos_v2()). In that state
+		 * len - iov_len does not locate the buffer start, and a
+		 * snapshot taken here would be replayed without GSS
+		 * verification. Refuse to defer; the caller fails the
+		 * request and the client retries.
+		 */
+		if (rqstp->rq_auth_slack ||
+		    rqstp->rq_arg.len < rqstp->rq_arg.head[0].iov_len)
+			return NULL;
+
 		/* FIXME maybe discard if size too large */
 		size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
 		dr = kmalloc(size, GFP_KERNEL);

2.50.1