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, 194 insertions, 0 deletions

diff --git a/fs/fserror.c b/fs/fserror.c
new file mode 100644
index 000000000..1e4d11fd9
--- /dev/null
+++ b/fs/fserror.c
@@ -0,0 +1,194 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (c) 2025 Oracle.  All Rights Reserved.
+ * Author: Darrick J. Wong <djwong@kernel.org>
+ */
+#include <linux/fs.h>
+#include <linux/fsnotify.h>
+#include <linux/mempool.h>
+#include <linux/fserror.h>
+
+#define FSERROR_DEFAULT_EVENT_POOL_SIZE		(32)
+
+static struct mempool fserror_events_pool;
+
+void fserror_mount(struct super_block *sb)
+{
+	/*
+	 * The pending error counter is biased by 1 so that we don't wake_var
+	 * until we're actually trying to unmount.
+	 */
+	refcount_set(&sb->s_pending_errors, 1);
+}
+
+void fserror_unmount(struct super_block *sb)
+{
+	/*
+	 * If we don't drop the pending error count to zero, then wait for it
+	 * to drop below 1, which means that the pending errors cleared and
+	 * hopefully we didn't saturate with 1 billion+ concurrent events.
+	 */
+	if (!refcount_dec_and_test(&sb->s_pending_errors))
+		wait_var_event(&sb->s_pending_errors,
+			       refcount_read(&sb->s_pending_errors) < 1);
+}
+
+static inline void fserror_pending_dec(struct super_block *sb)
+{
+	if (refcount_dec_and_test(&sb->s_pending_errors))
+		wake_up_var(&sb->s_pending_errors);
+}
+
+static inline void fserror_free_event(struct fserror_event *event)
+{
+	fserror_pending_dec(event->sb);
+	mempool_free(event, &fserror_events_pool);
+}
+
+static void fserror_worker(struct work_struct *work)
+{
+	struct fserror_event *event =
+			container_of(work, struct fserror_event, work);
+	struct super_block *sb = event->sb;
+
+	if (sb->s_flags & SB_ACTIVE) {
+		struct fs_error_report report = {
+			/* send positive error number to userspace */
+			.error = -event->error,
+			.inode = event->inode,
+			.sb = event->sb,
+		};
+
+		if (sb->s_op->report_error)
+			sb->s_op->report_error(event);
+
+		fsnotify(FS_ERROR, &report, FSNOTIFY_EVENT_ERROR, NULL, NULL,
+			 NULL, 0);
+	}
+
+	iput(event->inode);
+	fserror_free_event(event);
+}
+
+static inline struct fserror_event *fserror_alloc_event(struct super_block *sb,
+							gfp_t gfp_flags)
+{
+	struct fserror_event *event = NULL;
+
+	/*
+	 * If pending_errors already reached zero or is no longer active,
+	 * the superblock is being deactivated so there's no point in
+	 * continuing.
+	 *
+	 * The order of the check of s_pending_errors and SB_ACTIVE are
+	 * mandated by order of accesses in generic_shutdown_super and
+	 * fserror_unmount.  Barriers are implicitly provided by the refcount
+	 * manipulations in this function and fserror_unmount.
+	 */
+	if (!refcount_inc_not_zero(&sb->s_pending_errors))
+		return NULL;
+	if (!(sb->s_flags & SB_ACTIVE))
+		goto out_pending;
+
+	event = mempool_alloc(&fserror_events_pool, gfp_flags);
+	if (!event)
+		goto out_pending;
+
+	/* mempool_alloc doesn't support GFP_ZERO */
+	memset(event, 0, sizeof(*event));
+	event->sb = sb;
+	INIT_WORK(&event->work, fserror_worker);
+
+	return event;
+
+out_pending:
+	fserror_pending_dec(sb);
+	return NULL;
+}
+
+/**
+ * fserror_report - report a filesystem error of some kind
+ *
+ * @sb:		superblock of the filesystem
+ * @inode:	inode within that filesystem, if applicable
+ * @type:	type of error encountered
+ * @pos:	start of inode range affected, if applicable
+ * @len:	length of inode range affected, if applicable
+ * @error:	error number encountered, must be negative
+ * @gfp:	memory allocation flags for conveying the event to a worker,
+ *		since this function can be called from atomic contexts
+ *
+ * Report details of a filesystem error to the super_operations::report_error
+ * callback if present; and to fsnotify for distribution to userspace.  @sb,
+ * @gfp, @type, and @error must all be specified.  For file I/O errors, the
+ * @inode, @pos, and @len fields must also be specified.  For file metadata
+ * errors, @inode must be specified.  If @inode is not NULL, then @inode->i_sb
+ * must point to @sb.
+ *
+ * Reporting work is deferred to a workqueue to ensure that ->report_error is
+ * called from process context without any locks held.  An active reference to
+ * the inode is maintained until event handling is complete, and unmount will
+ * wait for queued events to drain.
+ */
+void fserror_report(struct super_block *sb, struct inode *inode,
+		    enum fserror_type type, loff_t pos, u64 len, int error,
+		    gfp_t gfp)
+{
+	struct fserror_event *event;
+
+	/* sb and inode must be from the same filesystem */
+	WARN_ON_ONCE(inode && inode->i_sb != sb);
+
+	/* error number must be negative */
+	WARN_ON_ONCE(error >= 0);
+
+	event = fserror_alloc_event(sb, gfp);
+	if (!event)
+		goto lost;
+
+	event->type = type;
+	event->pos = pos;
+	event->len = len;
+	event->error = error;
+
+	/*
+	 * Can't iput from non-sleeping context, so grabbing another reference
+	 * to the inode must be the last thing before submitting the event.
+	 */
+	if (inode) {
+		event->inode = igrab(inode);
+		if (!event->inode)
+			goto lost_event;
+	}
+
+	/*
+	 * Use schedule_work here even if we're already in process context so
+	 * that fsnotify and super_operations::report_error implementations are
+	 * guaranteed to run in process context without any locks held.  Since
+	 * errors are supposed to be rare, the overhead shouldn't kill us any
+	 * more than the failing device will.
+	 */
+	schedule_work(&event->work);
+	return;
+
+lost_event:
+	fserror_free_event(event);
+lost:
+	if (inode)
+		pr_err_ratelimited(
+ "%s: lost file I/O error report for ino %llu type %u pos 0x%llx len 0x%llx error %d",
+		       sb->s_id, inode->i_ino, type, pos, len, error);
+	else
+		pr_err_ratelimited(
+ "%s: lost filesystem error report for type %u error %d",
+		       sb->s_id, type, error);
+}
+EXPORT_SYMBOL_GPL(fserror_report);
+
+static int __init fserror_init(void)
+{
+	return mempool_init_kmalloc_pool(&fserror_events_pool,
+					 FSERROR_DEFAULT_EVENT_POOL_SIZE,
+					 sizeof(struct fserror_event));
+}
+fs_initcall(fserror_init);