x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

When using eBPF to probe CS CALL instructions within a function,
a crash can be triggered.

The eBPF tool probes offset 257 of the __hrtimer_run_queues()
function:

<__hrtimer_run_queues+249>:  nopl   0x0(%rax,%rax,1)
<__hrtimer_run_queues+254>:  mov    %r14,%rdi
<__hrtimer_run_queues+257>:  cs call <__x86_indirect_thunk_r12>
<__hrtimer_run_queues+263>:  mov    %eax,%r12d
<__hrtimer_run_queues+266>:  xchg   %ax,%ax
<__hrtimer_run_queues+268>:  mov    %r13,%rdi

Which triggers this crash:

  BUG: unable to handle page fault for address: 00000000000f41c9
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] SMP NOPTI
  CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P
  RIP: 0010:__hrtimer_run_queues+0x106/0x230

Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is
at the 6th byte of the above CS CALL instruction. Since the CS CALL
instruction occupies 6 bytes, the exception occurred in the middle of that
call instruction.

The root cause is that when using eBPF tools to probe in the middle of a
function, a kprobe with INT3 is used as the underlying implementation.

During single-step emulation of the original CALL instruction,
int3_emulate_call() assumes that the probed CALL instruction is 5 bytes
long. However, the actual CS-prefixed CALL instruction occupies 6 bytes,
so it constructs an incorrect exception return address. When the CPU
returns from the kprobe handler, the next instruction to be executed is at
the address of the last byte of that CS CALL instruction. Coincidentally,
starting from that address, the CPU fetches and decodes a completely
different instruction, which ultimately triggers a kernel crash.

Fix the issue by using the actual instruction length obtained from
the instruction decoder when constructing the exception return
address, rather than relying on the hardcoded CALL_INSN_SIZE macro.

[ mingo: Refined the changelog ]

Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

1 files changed, 241 insertions, 0 deletions

diff --git a/net/ipv4/tcp_yeah.c b/net/ipv4/tcp_yeah.c
new file mode 100644
index 000000000..9e581154f
--- /dev/null
+++ b/net/ipv4/tcp_yeah.c
@@ -0,0 +1,241 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ *
+ *   YeAH TCP
+ *
+ * For further details look at:
+ *   https://web.archive.org/web/20080316215752/http://wil.cs.caltech.edu/pfldnet2007/paper/YeAH_TCP.pdf
+ *
+ */
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/skbuff.h>
+#include <linux/inet_diag.h>
+
+#include <net/tcp.h>
+
+#include "tcp_vegas.h"
+
+#define TCP_YEAH_ALPHA       80 /* number of packets queued at the bottleneck */
+#define TCP_YEAH_GAMMA        1 /* fraction of queue to be removed per rtt */
+#define TCP_YEAH_DELTA        3 /* log minimum fraction of cwnd to be removed on loss */
+#define TCP_YEAH_EPSILON      1 /* log maximum fraction to be removed on early decongestion */
+#define TCP_YEAH_PHY          8 /* maximum delta from base */
+#define TCP_YEAH_RHO         16 /* minimum number of consecutive rtt to consider competition on loss */
+#define TCP_YEAH_ZETA        50 /* minimum number of state switches to reset reno_count */
+
+#define TCP_SCALABLE_AI_CNT	 100U
+
+/* YeAH variables */
+struct yeah {
+	struct vegas vegas;	/* must be first */
+
+	/* YeAH */
+	u32 lastQ;
+	u32 doing_reno_now;
+
+	u32 reno_count;
+	u32 fast_count;
+};
+
+static void tcp_yeah_init(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct yeah *yeah = inet_csk_ca(sk);
+
+	tcp_vegas_init(sk);
+
+	yeah->doing_reno_now = 0;
+	yeah->lastQ = 0;
+
+	yeah->reno_count = 2;
+
+	/* Ensure the MD arithmetic works.  This is somewhat pedantic,
+	 * since I don't think we will see a cwnd this large. :) */
+	tp->snd_cwnd_clamp = min_t(u32, tp->snd_cwnd_clamp, 0xffffffff/128);
+}
+
+static void tcp_yeah_cong_avoid(struct sock *sk, u32 ack, u32 acked)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct yeah *yeah = inet_csk_ca(sk);
+
+	if (!tcp_is_cwnd_limited(sk))
+		return;
+
+	if (tcp_in_slow_start(tp)) {
+		acked = tcp_slow_start(tp, acked);
+		if (!acked)
+			goto do_vegas;
+	}
+
+	if (!yeah->doing_reno_now) {
+		/* Scalable */
+		tcp_cong_avoid_ai(tp, min(tcp_snd_cwnd(tp), TCP_SCALABLE_AI_CNT),
+				  acked);
+	} else {
+		/* Reno */
+		tcp_cong_avoid_ai(tp, tcp_snd_cwnd(tp), acked);
+	}
+
+	/* The key players are v_vegas.beg_snd_una and v_beg_snd_nxt.
+	 *
+	 * These are so named because they represent the approximate values
+	 * of snd_una and snd_nxt at the beginning of the current RTT. More
+	 * precisely, they represent the amount of data sent during the RTT.
+	 * At the end of the RTT, when we receive an ACK for v_beg_snd_nxt,
+	 * we will calculate that (v_beg_snd_nxt - v_vegas.beg_snd_una) outstanding
+	 * bytes of data have been ACKed during the course of the RTT, giving
+	 * an "actual" rate of:
+	 *
+	 *     (v_beg_snd_nxt - v_vegas.beg_snd_una) / (rtt duration)
+	 *
+	 * Unfortunately, v_vegas.beg_snd_una is not exactly equal to snd_una,
+	 * because delayed ACKs can cover more than one segment, so they
+	 * don't line up yeahly with the boundaries of RTTs.
+	 *
+	 * Another unfortunate fact of life is that delayed ACKs delay the
+	 * advance of the left edge of our send window, so that the number
+	 * of bytes we send in an RTT is often less than our cwnd will allow.
+	 * So we keep track of our cwnd separately, in v_beg_snd_cwnd.
+	 */
+do_vegas:
+	if (after(ack, yeah->vegas.beg_snd_nxt)) {
+		/* We do the Vegas calculations only if we got enough RTT
+		 * samples that we can be reasonably sure that we got
+		 * at least one RTT sample that wasn't from a delayed ACK.
+		 * If we only had 2 samples total,
+		 * then that means we're getting only 1 ACK per RTT, which
+		 * means they're almost certainly delayed ACKs.
+		 * If  we have 3 samples, we should be OK.
+		 */
+
+		if (yeah->vegas.cntRTT > 2) {
+			u32 rtt, queue;
+			u64 bw;
+
+			/* We have enough RTT samples, so, using the Vegas
+			 * algorithm, we determine if we should increase or
+			 * decrease cwnd, and by how much.
+			 */
+
+			/* Pluck out the RTT we are using for the Vegas
+			 * calculations. This is the min RTT seen during the
+			 * last RTT. Taking the min filters out the effects
+			 * of delayed ACKs, at the cost of noticing congestion
+			 * a bit later.
+			 */
+			rtt = yeah->vegas.minRTT;
+
+			/* Compute excess number of packets above bandwidth
+			 * Avoid doing full 64 bit divide.
+			 */
+			bw = tcp_snd_cwnd(tp);
+			bw *= rtt - yeah->vegas.baseRTT;
+			do_div(bw, rtt);
+			queue = bw;
+
+			if (queue > TCP_YEAH_ALPHA ||
+			    rtt - yeah->vegas.baseRTT > (yeah->vegas.baseRTT / TCP_YEAH_PHY)) {
+				if (queue > TCP_YEAH_ALPHA &&
+				    tcp_snd_cwnd(tp) > yeah->reno_count) {
+					u32 reduction = min(queue / TCP_YEAH_GAMMA ,
+							    tcp_snd_cwnd(tp) >> TCP_YEAH_EPSILON);
+
+					tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) - reduction);
+
+					tcp_snd_cwnd_set(tp, max(tcp_snd_cwnd(tp),
+								 yeah->reno_count));
+
+					WRITE_ONCE(tp->snd_ssthresh,
+						   tcp_snd_cwnd(tp));
+				}
+
+				if (yeah->reno_count <= 2)
+					yeah->reno_count = max(tcp_snd_cwnd(tp)>>1, 2U);
+				else
+					yeah->reno_count++;
+
+				yeah->doing_reno_now = min(yeah->doing_reno_now + 1,
+							   0xffffffU);
+			} else {
+				yeah->fast_count++;
+
+				if (yeah->fast_count > TCP_YEAH_ZETA) {
+					yeah->reno_count = 2;
+					yeah->fast_count = 0;
+				}
+
+				yeah->doing_reno_now = 0;
+			}
+
+			yeah->lastQ = queue;
+		}
+
+		/* Save the extent of the current window so we can use this
+		 * at the end of the next RTT.
+		 */
+		yeah->vegas.beg_snd_una  = yeah->vegas.beg_snd_nxt;
+		yeah->vegas.beg_snd_nxt  = tp->snd_nxt;
+		yeah->vegas.beg_snd_cwnd = tcp_snd_cwnd(tp);
+
+		/* Wipe the slate clean for the next RTT. */
+		yeah->vegas.cntRTT = 0;
+		yeah->vegas.minRTT = 0x7fffffff;
+	}
+}
+
+static u32 tcp_yeah_ssthresh(struct sock *sk)
+{
+	const struct tcp_sock *tp = tcp_sk(sk);
+	struct yeah *yeah = inet_csk_ca(sk);
+	u32 reduction;
+
+	if (yeah->doing_reno_now < TCP_YEAH_RHO) {
+		reduction = yeah->lastQ;
+
+		reduction = min(reduction, max(tcp_snd_cwnd(tp)>>1, 2U));
+
+		reduction = max(reduction, tcp_snd_cwnd(tp) >> TCP_YEAH_DELTA);
+	} else
+		reduction = max(tcp_snd_cwnd(tp)>>1, 2U);
+
+	yeah->fast_count = 0;
+	yeah->reno_count = max(yeah->reno_count>>1, 2U);
+
+	return max_t(int, tcp_snd_cwnd(tp) - reduction, 2);
+}
+
+static struct tcp_congestion_ops tcp_yeah __read_mostly = {
+	.init		= tcp_yeah_init,
+	.ssthresh	= tcp_yeah_ssthresh,
+	.undo_cwnd      = tcp_reno_undo_cwnd,
+	.cong_avoid	= tcp_yeah_cong_avoid,
+	.set_state	= tcp_vegas_state,
+	.cwnd_event	= tcp_vegas_cwnd_event,
+	.cwnd_event_tx_start = tcp_vegas_cwnd_event_tx_start,
+	.get_info	= tcp_vegas_get_info,
+	.pkts_acked	= tcp_vegas_pkts_acked,
+
+	.owner		= THIS_MODULE,
+	.name		= "yeah",
+};
+
+static int __init tcp_yeah_register(void)
+{
+	BUILD_BUG_ON(sizeof(struct yeah) > ICSK_CA_PRIV_SIZE);
+	tcp_register_congestion_control(&tcp_yeah);
+	return 0;
+}
+
+static void __exit tcp_yeah_unregister(void)
+{
+	tcp_unregister_congestion_control(&tcp_yeah);
+}
+
+module_init(tcp_yeah_register);
+module_exit(tcp_yeah_unregister);
+
+MODULE_AUTHOR("Angelo P. Castellani");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("YeAH TCP");