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, 299 insertions, 0 deletions

diff --git a/fs/btrfs/messages.c b/fs/btrfs/messages.c
new file mode 100644
index 000000000..198d1747c
--- /dev/null
+++ b/fs/btrfs/messages.c
@@ -0,0 +1,299 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "fs.h"
+#include "messages.h"
+#include "discard.h"
+#include "super.h"
+
+#ifdef CONFIG_PRINTK
+
+#define STATE_STRING_PREFACE	" state "
+#define STATE_STRING_BUF_LEN	(sizeof(STATE_STRING_PREFACE) + BTRFS_FS_STATE_COUNT + 1)
+
+/*
+ * Characters to print to indicate error conditions or uncommon filesystem state.
+ * RO is not an error.
+ */
+static const char fs_state_chars[] = {
+	[BTRFS_FS_STATE_REMOUNTING]		= 'M',
+	[BTRFS_FS_STATE_RO]			= 0,
+	[BTRFS_FS_STATE_TRANS_ABORTED]		= 'A',
+	[BTRFS_FS_STATE_LOG_REPLAY_ABORTED]	= 'O',
+	[BTRFS_FS_STATE_DEV_REPLACING]		= 'R',
+	[BTRFS_FS_STATE_DUMMY_FS_INFO]		= 0,
+	[BTRFS_FS_STATE_NO_DATA_CSUMS]		= 'C',
+	[BTRFS_FS_STATE_SKIP_META_CSUMS]	= 'S',
+	[BTRFS_FS_STATE_LOG_CLEANUP_ERROR]	= 'L',
+	[BTRFS_FS_STATE_EMERGENCY_SHUTDOWN]	= 'E',
+};
+
+static void btrfs_state_to_string(const struct btrfs_fs_info *info, char *buf)
+{
+	unsigned int bit;
+	bool states_printed = false;
+	unsigned long fs_state = READ_ONCE(info->fs_state);
+	char *curr = buf;
+
+	memcpy(curr, STATE_STRING_PREFACE, sizeof(STATE_STRING_PREFACE));
+	curr += sizeof(STATE_STRING_PREFACE) - 1;
+
+	if (unlikely(BTRFS_FS_ERROR(info))) {
+		*curr++ = 'E';
+		states_printed = true;
+	}
+
+	for_each_set_bit(bit, &fs_state, sizeof(fs_state)) {
+		WARN_ON_ONCE(bit >= BTRFS_FS_STATE_COUNT);
+		if ((bit < BTRFS_FS_STATE_COUNT) && fs_state_chars[bit]) {
+			*curr++ = fs_state_chars[bit];
+			states_printed = true;
+		}
+	}
+
+	/* If no states were printed, reset the buffer */
+	if (!states_printed)
+		curr = buf;
+
+	*curr++ = 0;
+}
+#endif
+
+/*
+ * Generally the error codes correspond to their respective errors, but there
+ * are a few special cases.
+ *
+ * EUCLEAN: Any sort of corruption that we encounter.  The tree-checker for
+ *          instance will return EUCLEAN if any of the blocks are corrupted in
+ *          a way that is problematic.  We want to reserve EUCLEAN for these
+ *          sort of corruptions.
+ *
+ * EROFS: If we check BTRFS_FS_STATE_ERROR and fail out with a return error, we
+ *        need to use EROFS for this case.  We will have no idea of the
+ *        original failure, that will have been reported at the time we tripped
+ *        over the error.  Each subsequent error that doesn't have any context
+ *        of the original error should use EROFS when handling BTRFS_FS_STATE_ERROR.
+ */
+const char * __attribute_const__ btrfs_decode_error(int error)
+{
+	char *errstr = "unknown";
+
+	switch (error) {
+	case -ENOENT:		/* -2 */
+		errstr = "No such entry";
+		break;
+	case -EIO:		/* -5 */
+		errstr = "IO failure";
+		break;
+	case -ENOMEM:		/* -12*/
+		errstr = "Out of memory";
+		break;
+	case -EEXIST:		/* -17 */
+		errstr = "Object already exists";
+		break;
+	case -ENOSPC:		/* -28 */
+		errstr = "No space left";
+		break;
+	case -EROFS:		/* -30 */
+		errstr = "Readonly filesystem";
+		break;
+	case -EOPNOTSUPP:	/* -95 */
+		errstr = "Operation not supported";
+		break;
+	case -EUCLEAN:		/* -117 */
+		errstr = "Filesystem corrupted";
+		break;
+	case -EDQUOT:		/* -122 */
+		errstr = "Quota exceeded";
+		break;
+	}
+
+	return errstr;
+}
+
+/*
+ * Decodes expected errors from the caller and invokes the appropriate error
+ * response.
+ */
+__cold
+void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
+		       unsigned int line, int error, const char *fmt, ...)
+{
+	struct super_block *sb = fs_info->sb;
+#ifdef CONFIG_PRINTK
+	char statestr[STATE_STRING_BUF_LEN];
+	const char *errstr;
+#endif
+
+#ifdef CONFIG_PRINTK_INDEX
+	printk_index_subsys_emit(
+		"BTRFS: error (device %s%s) in %s:%d: errno=%d %s", KERN_CRIT, fmt);
+#endif
+
+	/*
+	 * Special case: if the error is EROFS, and we're already under
+	 * SB_RDONLY, then it is safe here.
+	 */
+	if (error == -EROFS && sb_rdonly(sb))
+		return;
+
+#ifdef CONFIG_PRINTK
+	errstr = btrfs_decode_error(error);
+	btrfs_state_to_string(fs_info, statestr);
+	if (fmt) {
+		struct va_format vaf;
+		va_list args;
+
+		va_start(args, fmt);
+		vaf.fmt = fmt;
+		vaf.va = &args;
+
+		pr_crit("BTRFS: error (device %s%s) in %s:%d: errno=%d %s (%pV)\n",
+			sb->s_id, statestr, function, line, error, errstr, &vaf);
+		va_end(args);
+	} else {
+		pr_crit("BTRFS: error (device %s%s) in %s:%d: errno=%d %s\n",
+			sb->s_id, statestr, function, line, error, errstr);
+	}
+#endif
+
+	/*
+	 * Today we only save the error info to memory.  Long term we'll also
+	 * send it down to the disk.
+	 */
+	WRITE_ONCE(fs_info->fs_error, error);
+
+	/* Don't go through full error handling during mount. */
+	if (!(sb->s_flags & SB_BORN))
+		return;
+
+	if (sb_rdonly(sb))
+		return;
+
+	btrfs_discard_stop(fs_info);
+
+	/* Handle error by forcing the filesystem readonly. */
+	btrfs_set_sb_rdonly(sb);
+	btrfs_info(fs_info, "forced readonly");
+	/*
+	 * Note that a running device replace operation is not canceled here
+	 * although there is no way to update the progress. It would add the
+	 * risk of a deadlock, therefore the canceling is omitted. The only
+	 * penalty is that some I/O remains active until the procedure
+	 * completes. The next time when the filesystem is mounted writable
+	 * again, the device replace operation continues.
+	 */
+}
+
+#ifdef CONFIG_PRINTK
+static const char * const logtypes[] = {
+	"emergency",
+	"alert",
+	"critical",
+	"error",
+	"warning",
+	"notice",
+	"info",
+	"debug",
+};
+
+/*
+ * Use one ratelimit state per log level so that a flood of less important
+ * messages doesn't cause more important ones to be dropped.
+ */
+static struct ratelimit_state printk_limits[] = {
+	RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
+	RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
+	RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
+	RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
+	RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
+	RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
+	RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
+	RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
+};
+
+__printf(3, 4) __cold
+void _btrfs_printk(const struct btrfs_fs_info *fs_info, unsigned int level, const char *fmt, ...)
+{
+	struct va_format vaf;
+	va_list args;
+	const char *type = logtypes[level];
+	struct ratelimit_state *ratelimit = &printk_limits[level];
+
+#ifdef CONFIG_PRINTK_INDEX
+	printk_index_subsys_emit("%sBTRFS %s (device %s): ", NULL, fmt);
+#endif
+
+	va_start(args, fmt);
+	vaf.fmt = fmt;
+	vaf.va = &args;
+
+	/* Do not ratelimit if CONFIG_BTRFS_DEBUG is enabled. */
+	if (IS_ENABLED(CONFIG_BTRFS_DEBUG) || __ratelimit(ratelimit)) {
+		if (fs_info) {
+			char statestr[STATE_STRING_BUF_LEN];
+
+			btrfs_state_to_string(fs_info, statestr);
+			_printk(KERN_SOH "%dBTRFS %s (device %s%s): %pV\n", level, type,
+				fs_info->sb->s_id, statestr, &vaf);
+		} else {
+			_printk(KERN_SOH "%dBTRFS %s: %pV\n", level, type, &vaf);
+		}
+	}
+
+	va_end(args);
+}
+#endif
+
+#if BITS_PER_LONG == 32
+void __cold btrfs_warn_32bit_limit(struct btrfs_fs_info *fs_info)
+{
+	if (!test_and_set_bit(BTRFS_FS_32BIT_WARN, &fs_info->flags)) {
+		btrfs_warn(fs_info, "reaching 32bit limit for logical addresses");
+		btrfs_warn(fs_info,
+"due to page cache limit on 32bit systems, btrfs can't access metadata at or beyond %lluT",
+			   BTRFS_32BIT_MAX_FILE_SIZE >> 40);
+		btrfs_warn(fs_info,
+			   "please consider upgrading to 64bit kernel/hardware");
+	}
+}
+
+void __cold btrfs_err_32bit_limit(struct btrfs_fs_info *fs_info)
+{
+	if (!test_and_set_bit(BTRFS_FS_32BIT_ERROR, &fs_info->flags)) {
+		btrfs_err(fs_info, "reached 32bit limit for logical addresses");
+		btrfs_err(fs_info,
+"due to page cache limit on 32bit systems, metadata beyond %lluT can't be accessed",
+			  BTRFS_32BIT_MAX_FILE_SIZE >> 40);
+		btrfs_err(fs_info,
+			   "please consider upgrading to 64bit kernel/hardware");
+	}
+}
+#endif
+
+/*
+ * Decode unexpected, fatal errors from the caller, issue an alert, and either
+ * panic or BUGs, depending on mount options.
+ */
+__cold
+void __btrfs_panic(const struct btrfs_fs_info *fs_info, const char *function,
+		   unsigned int line, int error, const char *fmt, ...)
+{
+	char *s_id = "<unknown>";
+	struct va_format vaf = { .fmt = fmt };
+	va_list args;
+
+	if (fs_info)
+		s_id = fs_info->sb->s_id;
+
+	va_start(args, fmt);
+	vaf.va = &args;
+
+	if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR)))
+		panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %pe)\n",
+			s_id, function, line, &vaf, error, ERR_PTR(error));
+
+	btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %pe)",
+		   function, line, &vaf, error, ERR_PTR(error));
+	va_end(args);
+	/* Caller calls BUG() */
+}