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, 407 insertions, 0 deletions

diff --git a/fs/adfs/map.c b/fs/adfs/map.c
new file mode 100644
index 000000000..9d535a2ca
--- /dev/null
+++ b/fs/adfs/map.c
@@ -0,0 +1,407 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ *  linux/fs/adfs/map.c
+ *
+ *  Copyright (C) 1997-2002 Russell King
+ */
+#include <linux/slab.h>
+#include <linux/statfs.h>
+#include <linux/unaligned.h>
+#include "adfs.h"
+
+/*
+ * The ADFS map is basically a set of sectors.  Each sector is called a
+ * zone which contains a bitstream made up of variable sized fragments.
+ * Each bit refers to a set of bytes in the filesystem, defined by
+ * log2bpmb.  This may be larger or smaller than the sector size, but
+ * the overall size it describes will always be a round number of
+ * sectors.  A fragment id is always idlen bits long.
+ *
+ *  < idlen > <       n        > <1>
+ * +---------+-------//---------+---+
+ * | frag id |  0000....000000  | 1 |
+ * +---------+-------//---------+---+
+ *
+ * The physical disk space used by a fragment is taken from the start of
+ * the fragment id up to and including the '1' bit - ie, idlen + n + 1
+ * bits.
+ *
+ * A fragment id can be repeated multiple times in the whole map for
+ * large or fragmented files.  The first map zone a fragment starts in
+ * is given by fragment id / ids_per_zone - this allows objects to start
+ * from any zone on the disk.
+ *
+ * Free space is described by a linked list of fragments.  Each free
+ * fragment describes free space in the same way as the other fragments,
+ * however, the frag id specifies an offset (in map bits) from the end
+ * of this fragment to the start of the next free fragment.
+ *
+ * Objects stored on the disk are allocated object ids (we use these as
+ * our inode numbers.)  Object ids contain a fragment id and an optional
+ * offset.  This allows a directory fragment to contain small files
+ * associated with that directory.
+ */
+
+/*
+ * For the future...
+ */
+static DEFINE_RWLOCK(adfs_map_lock);
+
+/*
+ * This is fun.  We need to load up to 19 bits from the map at an
+ * arbitrary bit alignment.  (We're limited to 19 bits by F+ version 2).
+ */
+#define GET_FRAG_ID(_map,_start,_idmask)				\
+	({								\
+		unsigned char *_m = _map + (_start >> 3);		\
+		u32 _frag = get_unaligned_le32(_m);			\
+		_frag >>= (_start & 7);					\
+		_frag & _idmask;					\
+	})
+
+/*
+ * return the map bit offset of the fragment frag_id in the zone dm.
+ * Note that the loop is optimised for best asm code - look at the
+ * output of:
+ *  gcc -D__KERNEL__ -O2 -I../../include -o - -S map.c
+ */
+static int lookup_zone(const struct adfs_discmap *dm, const unsigned int idlen,
+		       const u32 frag_id, unsigned int *offset)
+{
+	const unsigned int endbit = dm->dm_endbit;
+	const u32 idmask = (1 << idlen) - 1;
+	unsigned char *map = dm->dm_bh->b_data;
+	unsigned int start = dm->dm_startbit;
+	unsigned int freelink, fragend;
+	u32 frag;
+
+	frag = GET_FRAG_ID(map, 8, idmask & 0x7fff);
+	freelink = frag ? 8 + frag : 0;
+
+	do {
+		frag = GET_FRAG_ID(map, start, idmask);
+
+		fragend = find_next_bit_le(map, endbit, start + idlen);
+		if (fragend >= endbit)
+			goto error;
+
+		if (start == freelink) {
+			freelink += frag & 0x7fff;
+		} else if (frag == frag_id) {
+			unsigned int length = fragend + 1 - start;
+
+			if (*offset < length)
+				return start + *offset;
+			*offset -= length;
+		}
+
+		start = fragend + 1;
+	} while (start < endbit);
+	return -1;
+
+error:
+	printk(KERN_ERR "adfs: oversized fragment 0x%x at 0x%x-0x%x\n",
+		frag, start, fragend);
+	return -1;
+}
+
+/*
+ * Scan the free space map, for this zone, calculating the total
+ * number of map bits in each free space fragment.
+ *
+ * Note: idmask is limited to 15 bits [3.2]
+ */
+static unsigned int
+scan_free_map(struct adfs_sb_info *asb, struct adfs_discmap *dm)
+{
+	const unsigned int endbit = dm->dm_endbit;
+	const unsigned int idlen  = asb->s_idlen;
+	const unsigned int frag_idlen = idlen <= 15 ? idlen : 15;
+	const u32 idmask = (1 << frag_idlen) - 1;
+	unsigned char *map = dm->dm_bh->b_data;
+	unsigned int start = 8, fragend;
+	u32 frag;
+	unsigned long total = 0;
+
+	/*
+	 * get fragment id
+	 */
+	frag = GET_FRAG_ID(map, start, idmask);
+
+	/*
+	 * If the freelink is null, then no free fragments
+	 * exist in this zone.
+	 */
+	if (frag == 0)
+		return 0;
+
+	do {
+		start += frag;
+
+		frag = GET_FRAG_ID(map, start, idmask);
+
+		fragend = find_next_bit_le(map, endbit, start + idlen);
+		if (fragend >= endbit)
+			goto error;
+
+		total += fragend + 1 - start;
+	} while (frag >= idlen + 1);
+
+	if (frag != 0)
+		printk(KERN_ERR "adfs: undersized free fragment\n");
+
+	return total;
+error:
+	printk(KERN_ERR "adfs: oversized free fragment\n");
+	return 0;
+}
+
+static int scan_map(struct adfs_sb_info *asb, unsigned int zone,
+		    const u32 frag_id, unsigned int mapoff)
+{
+	const unsigned int idlen = asb->s_idlen;
+	struct adfs_discmap *dm, *dm_end;
+	int result;
+
+	dm	= asb->s_map + zone;
+	zone	= asb->s_map_size;
+	dm_end	= asb->s_map + zone;
+
+	do {
+		result = lookup_zone(dm, idlen, frag_id, &mapoff);
+
+		if (result != -1)
+			goto found;
+
+		dm ++;
+		if (dm == dm_end)
+			dm = asb->s_map;
+	} while (--zone > 0);
+
+	return -1;
+found:
+	result -= dm->dm_startbit;
+	result += dm->dm_startblk;
+
+	return result;
+}
+
+/*
+ * calculate the amount of free blocks in the map.
+ *
+ *              n=1
+ *  total_free = E(free_in_zone_n)
+ *              nzones
+ */
+void adfs_map_statfs(struct super_block *sb, struct kstatfs *buf)
+{
+	struct adfs_sb_info *asb = ADFS_SB(sb);
+	struct adfs_discrecord *dr = adfs_map_discrecord(asb->s_map);
+	struct adfs_discmap *dm;
+	unsigned int total = 0;
+	unsigned int zone;
+
+	dm   = asb->s_map;
+	zone = asb->s_map_size;
+
+	do {
+		total += scan_free_map(asb, dm++);
+	} while (--zone > 0);
+
+	buf->f_blocks  = adfs_disc_size(dr) >> sb->s_blocksize_bits;
+	buf->f_files   = asb->s_ids_per_zone * asb->s_map_size;
+	buf->f_bavail  =
+	buf->f_bfree   = signed_asl(total, asb->s_map2blk);
+}
+
+int adfs_map_lookup(struct super_block *sb, u32 frag_id, unsigned int offset)
+{
+	struct adfs_sb_info *asb = ADFS_SB(sb);
+	unsigned int zone, mapoff;
+	int result;
+
+	/*
+	 * map & root fragment is special - it starts in the center of the
+	 * disk.  The other fragments start at zone (frag / ids_per_zone)
+	 */
+	if (frag_id == ADFS_ROOT_FRAG)
+		zone = asb->s_map_size >> 1;
+	else
+		zone = frag_id / asb->s_ids_per_zone;
+
+	if (zone >= asb->s_map_size)
+		goto bad_fragment;
+
+	/* Convert sector offset to map offset */
+	mapoff = signed_asl(offset, -asb->s_map2blk);
+
+	read_lock(&adfs_map_lock);
+	result = scan_map(asb, zone, frag_id, mapoff);
+	read_unlock(&adfs_map_lock);
+
+	if (result > 0) {
+		unsigned int secoff;
+
+		/* Calculate sector offset into map block */
+		secoff = offset - signed_asl(mapoff, asb->s_map2blk);
+		return secoff + signed_asl(result, asb->s_map2blk);
+	}
+
+	adfs_error(sb, "fragment 0x%04x at offset %d not found in map",
+		   frag_id, offset);
+	return 0;
+
+bad_fragment:
+	adfs_error(sb, "invalid fragment 0x%04x (zone = %d, max = %d)",
+		   frag_id, zone, asb->s_map_size);
+	return 0;
+}
+
+static unsigned char adfs_calczonecheck(struct super_block *sb, unsigned char *map)
+{
+	unsigned int v0, v1, v2, v3;
+	int i;
+
+	v0 = v1 = v2 = v3 = 0;
+	for (i = sb->s_blocksize - 4; i; i -= 4) {
+		v0 += map[i]     + (v3 >> 8);
+		v3 &= 0xff;
+		v1 += map[i + 1] + (v0 >> 8);
+		v0 &= 0xff;
+		v2 += map[i + 2] + (v1 >> 8);
+		v1 &= 0xff;
+		v3 += map[i + 3] + (v2 >> 8);
+		v2 &= 0xff;
+	}
+	v0 +=           v3 >> 8;
+	v1 += map[1] + (v0 >> 8);
+	v2 += map[2] + (v1 >> 8);
+	v3 += map[3] + (v2 >> 8);
+
+	return v0 ^ v1 ^ v2 ^ v3;
+}
+
+static int adfs_checkmap(struct super_block *sb, struct adfs_discmap *dm)
+{
+	unsigned char crosscheck = 0, zonecheck = 1;
+	int i;
+
+	for (i = 0; i < ADFS_SB(sb)->s_map_size; i++) {
+		unsigned char *map;
+
+		map = dm[i].dm_bh->b_data;
+
+		if (adfs_calczonecheck(sb, map) != map[0]) {
+			adfs_error(sb, "zone %d fails zonecheck", i);
+			zonecheck = 0;
+		}
+		crosscheck ^= map[3];
+	}
+	if (crosscheck != 0xff)
+		adfs_error(sb, "crosscheck != 0xff");
+	return crosscheck == 0xff && zonecheck;
+}
+
+/*
+ * Layout the map - the first zone contains a copy of the disc record,
+ * and the last zone must be limited to the size of the filesystem.
+ */
+static void adfs_map_layout(struct adfs_discmap *dm, unsigned int nzones,
+			    struct adfs_discrecord *dr)
+{
+	unsigned int zone, zone_size;
+	u64 size;
+
+	zone_size = (8 << dr->log2secsize) - le16_to_cpu(dr->zone_spare);
+
+	dm[0].dm_bh       = NULL;
+	dm[0].dm_startblk = 0;
+	dm[0].dm_startbit = 32 + ADFS_DR_SIZE_BITS;
+	dm[0].dm_endbit   = 32 + zone_size;
+
+	for (zone = 1; zone < nzones; zone++) {
+		dm[zone].dm_bh       = NULL;
+		dm[zone].dm_startblk = zone * zone_size - ADFS_DR_SIZE_BITS;
+		dm[zone].dm_startbit = 32;
+		dm[zone].dm_endbit   = 32 + zone_size;
+	}
+
+	size = adfs_disc_size(dr) >> dr->log2bpmb;
+	size -= (nzones - 1) * zone_size - ADFS_DR_SIZE_BITS;
+	dm[nzones - 1].dm_endbit = 32 + size;
+}
+
+static int adfs_map_read(struct adfs_discmap *dm, struct super_block *sb,
+			 unsigned int map_addr, unsigned int nzones)
+{
+	unsigned int zone;
+
+	for (zone = 0; zone < nzones; zone++) {
+		dm[zone].dm_bh = sb_bread(sb, map_addr + zone);
+		if (!dm[zone].dm_bh)
+			return -EIO;
+	}
+
+	return 0;
+}
+
+static void adfs_map_relse(struct adfs_discmap *dm, unsigned int nzones)
+{
+	unsigned int zone;
+
+	for (zone = 0; zone < nzones; zone++)
+		brelse(dm[zone].dm_bh);
+}
+
+struct adfs_discmap *adfs_read_map(struct super_block *sb, struct adfs_discrecord *dr)
+{
+	struct adfs_sb_info *asb = ADFS_SB(sb);
+	struct adfs_discmap *dm;
+	unsigned int map_addr, zone_size, nzones;
+	int ret;
+
+	nzones    = dr->nzones | dr->nzones_high << 8;
+	zone_size = (8 << dr->log2secsize) - le16_to_cpu(dr->zone_spare);
+
+	asb->s_idlen = dr->idlen;
+	asb->s_map_size = nzones;
+	asb->s_map2blk = dr->log2bpmb - dr->log2secsize;
+	asb->s_log2sharesize = dr->log2sharesize;
+	asb->s_ids_per_zone = zone_size / (asb->s_idlen + 1);
+
+	map_addr = (nzones >> 1) * zone_size -
+		     ((nzones > 1) ? ADFS_DR_SIZE_BITS : 0);
+	map_addr = signed_asl(map_addr, asb->s_map2blk);
+
+	dm = kmalloc_objs(*dm, nzones);
+	if (dm == NULL) {
+		adfs_error(sb, "not enough memory");
+		return ERR_PTR(-ENOMEM);
+	}
+
+	adfs_map_layout(dm, nzones, dr);
+
+	ret = adfs_map_read(dm, sb, map_addr, nzones);
+	if (ret) {
+		adfs_error(sb, "unable to read map");
+		goto error_free;
+	}
+
+	if (adfs_checkmap(sb, dm))
+		return dm;
+
+	adfs_error(sb, "map corrupted");
+
+error_free:
+	adfs_map_relse(dm, nzones);
+	kfree(dm);
+	return ERR_PTR(-EIO);
+}
+
+void adfs_free_map(struct super_block *sb)
+{
+	struct adfs_sb_info *asb = ADFS_SB(sb);
+
+	adfs_map_relse(asb->s_map, asb->s_map_size);
+	kfree(asb->s_map);
+}