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, 243 insertions, 0 deletions

diff --git a/fs/hfs/bitmap.c b/fs/hfs/bitmap.c
new file mode 100644
index 000000000..5e84833a4
--- /dev/null
+++ b/fs/hfs/bitmap.c
@@ -0,0 +1,243 @@
+/*
+ *  linux/fs/hfs/bitmap.c
+ *
+ * Copyright (C) 1996-1997  Paul H. Hargrove
+ * (C) 2003 Ardis Technologies <roman@ardistech.com>
+ * This file may be distributed under the terms of the GNU General Public License.
+ *
+ * Based on GPLed code Copyright (C) 1995  Michael Dreher
+ *
+ * This file contains the code to modify the volume bitmap:
+ * search/set/clear bits.
+ */
+
+#include "hfs_fs.h"
+
+/*
+ * hfs_find_zero_bit()
+ *
+ * Description:
+ *  Given a block of memory, its length in bits, and a starting bit number,
+ *  determine the number of the first zero bits (in left-to-right ordering)
+ *  in that range.
+ *
+ *  Returns >= 'size' if no zero bits are found in the range.
+ *
+ *  Accesses memory in 32-bit aligned chunks of 32-bits and thus
+ *  may read beyond the 'size'th bit.
+ */
+static u32 hfs_find_set_zero_bits(__be32 *bitmap, u32 size, u32 offset, u32 *max)
+{
+	__be32 *curr, *end;
+	u32 mask, start, len, n;
+	__be32 val;
+	int i;
+
+	len = *max;
+	if (!len)
+		return size;
+
+	curr = bitmap + (offset / 32);
+	end = bitmap + ((size + 31) / 32);
+
+	/* scan the first partial u32 for zero bits */
+	val = *curr;
+	if (~val) {
+		n = be32_to_cpu(val);
+		i = offset % 32;
+		mask = (1U << 31) >> i;
+		for (; i < 32; mask >>= 1, i++) {
+			if (!(n & mask))
+				goto found;
+		}
+	}
+
+	/* scan complete u32s for the first zero bit */
+	while (++curr < end) {
+		val = *curr;
+		if (~val) {
+			n = be32_to_cpu(val);
+			mask = 1 << 31;
+			for (i = 0; i < 32; mask >>= 1, i++) {
+				if (!(n & mask))
+					goto found;
+			}
+		}
+	}
+	return size;
+
+found:
+	start = (curr - bitmap) * 32 + i;
+	if (start >= size)
+		return start;
+	/* do any partial u32 at the start */
+	len = min(size - start, len);
+	while (1) {
+		n |= mask;
+		if (++i >= 32)
+			break;
+		mask >>= 1;
+		if (!--len || n & mask)
+			goto done;
+	}
+	if (!--len)
+		goto done;
+	*curr++ = cpu_to_be32(n);
+	/* do full u32s */
+	while (1) {
+		n = be32_to_cpu(*curr);
+		if (len < 32)
+			break;
+		if (n) {
+			len = 32;
+			break;
+		}
+		*curr++ = cpu_to_be32(0xffffffff);
+		len -= 32;
+	}
+	/* do any partial u32 at end */
+	mask = 1U << 31;
+	for (i = 0; i < len; i++) {
+		if (n & mask)
+			break;
+		n |= mask;
+		mask >>= 1;
+	}
+done:
+	*curr = cpu_to_be32(n);
+	*max = (curr - bitmap) * 32 + i - start;
+	return start;
+}
+
+/*
+ * hfs_vbm_search_free()
+ *
+ * Description:
+ *   Search for 'num_bits' consecutive cleared bits in the bitmap blocks of
+ *   the hfs MDB. 'mdb' had better be locked or the returned range
+ *   may be no longer free, when this functions returns!
+ *   XXX Currently the search starts from bit 0, but it should start with
+ *   the bit number stored in 's_alloc_ptr' of the MDB.
+ * Input Variable(s):
+ *   struct hfs_mdb *mdb: Pointer to the hfs MDB
+ *   u16 *num_bits: Pointer to the number of cleared bits
+ *     to search for
+ * Output Variable(s):
+ *   u16 *num_bits: The number of consecutive clear bits of the
+ *     returned range. If the bitmap is fragmented, this will be less than
+ *     requested and it will be zero, when the disk is full.
+ * Returns:
+ *   The number of the first bit of the range of cleared bits which has been
+ *   found. When 'num_bits' is zero, this is invalid!
+ * Preconditions:
+ *   'mdb' points to a "valid" (struct hfs_mdb).
+ *   'num_bits' points to a variable of type (u16), which contains
+ *	the number of cleared bits to find.
+ * Postconditions:
+ *   'num_bits' is set to the length of the found sequence.
+ */
+u32 hfs_vbm_search_free(struct super_block *sb, u32 goal, u32 *num_bits)
+{
+	void *bitmap;
+	u32 pos;
+
+	/* make sure we have actual work to perform */
+	if (!*num_bits)
+		return 0;
+
+	mutex_lock(&HFS_SB(sb)->bitmap_lock);
+	bitmap = HFS_SB(sb)->bitmap;
+
+	pos = hfs_find_set_zero_bits(bitmap, HFS_SB(sb)->fs_ablocks, goal, num_bits);
+	if (pos >= HFS_SB(sb)->fs_ablocks) {
+		if (goal)
+			pos = hfs_find_set_zero_bits(bitmap, goal, 0, num_bits);
+		if (pos >= HFS_SB(sb)->fs_ablocks) {
+			*num_bits = pos = 0;
+			goto out;
+		}
+	}
+
+	hfs_dbg("pos %u, num_bits %u\n", pos, *num_bits);
+	HFS_SB(sb)->free_ablocks -= *num_bits;
+	hfs_bitmap_dirty(sb);
+out:
+	mutex_unlock(&HFS_SB(sb)->bitmap_lock);
+	return pos;
+}
+
+
+/*
+ * hfs_clear_vbm_bits()
+ *
+ * Description:
+ *   Clear the requested bits in the volume bitmap of the hfs filesystem
+ * Input Variable(s):
+ *   struct hfs_mdb *mdb: Pointer to the hfs MDB
+ *   u16 start: The offset of the first bit
+ *   u16 count: The number of bits
+ * Output Variable(s):
+ *   None
+ * Returns:
+ *    0: no error
+ *   -1: One of the bits was already clear.  This is a strange
+ *	 error and when it happens, the filesystem must be repaired!
+ *   -2: One or more of the bits are out of range of the bitmap.
+ * Preconditions:
+ *   'mdb' points to a "valid" (struct hfs_mdb).
+ * Postconditions:
+ *   Starting with bit number 'start', 'count' bits in the volume bitmap
+ *   are cleared. The affected bitmap blocks are marked "dirty", the free
+ *   block count of the MDB is updated and the MDB is marked dirty.
+ */
+int hfs_clear_vbm_bits(struct super_block *sb, u16 start, u16 count)
+{
+	__be32 *curr;
+	u32 mask;
+	int i, len;
+
+	/* is there any actual work to be done? */
+	if (!count)
+		return 0;
+
+	hfs_dbg("start %u, count %u\n", start, count);
+	/* are all of the bits in range? */
+	if ((start + count) > HFS_SB(sb)->fs_ablocks)
+		return -2;
+
+	mutex_lock(&HFS_SB(sb)->bitmap_lock);
+	/* bitmap is always on a 32-bit boundary */
+	curr = HFS_SB(sb)->bitmap + (start / 32);
+	len = count;
+
+	/* do any partial u32 at the start */
+	i = start % 32;
+	if (i) {
+		int j = 32 - i;
+		mask = 0xffffffffU << j;
+		if (j > count) {
+			mask |= 0xffffffffU >> (i + count);
+			*curr &= cpu_to_be32(mask);
+			goto out;
+		}
+		*curr++ &= cpu_to_be32(mask);
+		count -= j;
+	}
+
+	/* do full u32s */
+	while (count >= 32) {
+		*curr++ = 0;
+		count -= 32;
+	}
+	/* do any partial u32 at end */
+	if (count) {
+		mask = 0xffffffffU >> count;
+		*curr &= cpu_to_be32(mask);
+	}
+out:
+	HFS_SB(sb)->free_ablocks += len;
+	mutex_unlock(&HFS_SB(sb)->bitmap_lock);
+	hfs_bitmap_dirty(sb);
+
+	return 0;
+}