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, 324 insertions, 0 deletions

diff --git a/fs/adfs/dir_f.c b/fs/adfs/dir_f.c
new file mode 100644
index 000000000..05e963402
--- /dev/null
+++ b/fs/adfs/dir_f.c
@@ -0,0 +1,324 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ *  linux/fs/adfs/dir_f.c
+ *
+ * Copyright (C) 1997-1999 Russell King
+ *
+ *  E and F format directory handling
+ */
+#include "adfs.h"
+#include "dir_f.h"
+
+/*
+ * Read an (unaligned) value of length 1..4 bytes
+ */
+static inline unsigned int adfs_readval(unsigned char *p, int len)
+{
+	unsigned int val = 0;
+
+	switch (len) {
+	case 4:		val |= p[3] << 24;
+		fallthrough;
+	case 3:		val |= p[2] << 16;
+		fallthrough;
+	case 2:		val |= p[1] << 8;
+		fallthrough;
+	default:	val |= p[0];
+	}
+	return val;
+}
+
+static inline void adfs_writeval(unsigned char *p, int len, unsigned int val)
+{
+	switch (len) {
+	case 4:		p[3] = val >> 24;
+		fallthrough;
+	case 3:		p[2] = val >> 16;
+		fallthrough;
+	case 2:		p[1] = val >> 8;
+		fallthrough;
+	default:	p[0] = val;
+	}
+}
+
+#define ror13(v) ((v >> 13) | (v << 19))
+
+#define dir_u8(idx)				\
+	({ int _buf = idx >> blocksize_bits;	\
+	   int _off = idx - (_buf << blocksize_bits);\
+	  *(u8 *)(bh[_buf]->b_data + _off);	\
+	})
+
+#define dir_u32(idx)				\
+	({ int _buf = idx >> blocksize_bits;	\
+	   int _off = idx - (_buf << blocksize_bits);\
+	  *(__le32 *)(bh[_buf]->b_data + _off);	\
+	})
+
+#define bufoff(_bh,_idx)			\
+	({ int _buf = _idx >> blocksize_bits;	\
+	   int _off = _idx - (_buf << blocksize_bits);\
+	  (void *)(_bh[_buf]->b_data + _off);	\
+	})
+
+/*
+ * There are some algorithms that are nice in
+ * assembler, but a bitch in C...  This is one
+ * of them.
+ */
+static u8
+adfs_dir_checkbyte(const struct adfs_dir *dir)
+{
+	struct buffer_head * const *bh = dir->bh;
+	const int blocksize_bits = dir->sb->s_blocksize_bits;
+	union { __le32 *ptr32; u8 *ptr8; } ptr, end;
+	u32 dircheck = 0;
+	int last = 5 - 26;
+	int i = 0;
+
+	/*
+	 * Accumulate each word up to the last whole
+	 * word of the last directory entry.  This
+	 * can spread across several buffer heads.
+	 */
+	do {
+		last += 26;
+		do {
+			dircheck = le32_to_cpu(dir_u32(i)) ^ ror13(dircheck);
+
+			i += sizeof(u32);
+		} while (i < (last & ~3));
+	} while (dir_u8(last) != 0);
+
+	/*
+	 * Accumulate the last few bytes.  These
+	 * bytes will be within the same bh.
+	 */
+	if (i != last) {
+		ptr.ptr8 = bufoff(bh, i);
+		end.ptr8 = ptr.ptr8 + last - i;
+
+		do {
+			dircheck = *ptr.ptr8++ ^ ror13(dircheck);
+		} while (ptr.ptr8 < end.ptr8);
+	}
+
+	/*
+	 * The directory tail is in the final bh
+	 * Note that contary to the RISC OS PRMs,
+	 * the first few bytes are NOT included
+	 * in the check.  All bytes are in the
+	 * same bh.
+	 */
+	ptr.ptr8 = bufoff(bh, 2008);
+	end.ptr8 = ptr.ptr8 + 36;
+
+	do {
+		__le32 v = *ptr.ptr32++;
+		dircheck = le32_to_cpu(v) ^ ror13(dircheck);
+	} while (ptr.ptr32 < end.ptr32);
+
+	return (dircheck ^ (dircheck >> 8) ^ (dircheck >> 16) ^ (dircheck >> 24)) & 0xff;
+}
+
+static int adfs_f_validate(struct adfs_dir *dir)
+{
+	struct adfs_dirheader *head = dir->dirhead;
+	struct adfs_newdirtail *tail = dir->newtail;
+
+	if (head->startmasseq != tail->endmasseq ||
+	    tail->dirlastmask || tail->reserved[0] || tail->reserved[1] ||
+	    (memcmp(&head->startname, "Nick", 4) &&
+	     memcmp(&head->startname, "Hugo", 4)) ||
+	    memcmp(&head->startname, &tail->endname, 4) ||
+	    adfs_dir_checkbyte(dir) != tail->dircheckbyte)
+		return -EIO;
+
+	return 0;
+}
+
+/* Read and check that a directory is valid */
+static int adfs_f_read(struct super_block *sb, u32 indaddr, unsigned int size,
+		       struct adfs_dir *dir)
+{
+	const unsigned int blocksize_bits = sb->s_blocksize_bits;
+	int ret;
+
+	if (size && size != ADFS_NEWDIR_SIZE)
+		return -EIO;
+
+	ret = adfs_dir_read_buffers(sb, indaddr, ADFS_NEWDIR_SIZE, dir);
+	if (ret)
+		return ret;
+
+	dir->dirhead = bufoff(dir->bh, 0);
+	dir->newtail = bufoff(dir->bh, 2007);
+
+	if (adfs_f_validate(dir))
+		goto bad_dir;
+
+	dir->parent_id = adfs_readval(dir->newtail->dirparent, 3);
+
+	return 0;
+
+bad_dir:
+	adfs_error(sb, "dir %06x is corrupted", indaddr);
+	adfs_dir_relse(dir);
+
+	return -EIO;
+}
+
+/*
+ * convert a disk-based directory entry to a Linux ADFS directory entry
+ */
+static inline void
+adfs_dir2obj(struct adfs_dir *dir, struct object_info *obj,
+	struct adfs_direntry *de)
+{
+	unsigned int name_len;
+
+	for (name_len = 0; name_len < ADFS_F_NAME_LEN; name_len++) {
+		if (de->dirobname[name_len] < ' ')
+			break;
+
+		obj->name[name_len] = de->dirobname[name_len];
+	}
+
+	obj->name_len =	name_len;
+	obj->indaddr  = adfs_readval(de->dirinddiscadd, 3);
+	obj->loadaddr = adfs_readval(de->dirload, 4);
+	obj->execaddr = adfs_readval(de->direxec, 4);
+	obj->size     = adfs_readval(de->dirlen,  4);
+	obj->attr     = de->newdiratts;
+
+	adfs_object_fixup(dir, obj);
+}
+
+/*
+ * convert a Linux ADFS directory entry to a disk-based directory entry
+ */
+static inline void
+adfs_obj2dir(struct adfs_direntry *de, struct object_info *obj)
+{
+	adfs_writeval(de->dirinddiscadd, 3, obj->indaddr);
+	adfs_writeval(de->dirload, 4, obj->loadaddr);
+	adfs_writeval(de->direxec, 4, obj->execaddr);
+	adfs_writeval(de->dirlen,  4, obj->size);
+	de->newdiratts = obj->attr;
+}
+
+/*
+ * get a directory entry.  Note that the caller is responsible
+ * for holding the relevant locks.
+ */
+static int
+__adfs_dir_get(struct adfs_dir *dir, int pos, struct object_info *obj)
+{
+	struct adfs_direntry de;
+	int ret;
+
+	ret = adfs_dir_copyfrom(&de, dir, pos, 26);
+	if (ret)
+		return ret;
+
+	if (!de.dirobname[0])
+		return -ENOENT;
+
+	adfs_dir2obj(dir, obj, &de);
+
+	return 0;
+}
+
+static int
+adfs_f_setpos(struct adfs_dir *dir, unsigned int fpos)
+{
+	if (fpos >= ADFS_NUM_DIR_ENTRIES)
+		return -ENOENT;
+
+	dir->pos = 5 + fpos * 26;
+	return 0;
+}
+
+static int
+adfs_f_getnext(struct adfs_dir *dir, struct object_info *obj)
+{
+	unsigned int ret;
+
+	ret = __adfs_dir_get(dir, dir->pos, obj);
+	if (ret == 0)
+		dir->pos += 26;
+
+	return ret;
+}
+
+static int adfs_f_iterate(struct adfs_dir *dir, struct dir_context *ctx)
+{
+	struct object_info obj;
+	int pos = 5 + (ctx->pos - 2) * 26;
+
+	while (ctx->pos < 2 + ADFS_NUM_DIR_ENTRIES) {
+		if (__adfs_dir_get(dir, pos, &obj))
+			break;
+		if (!dir_emit(ctx, obj.name, obj.name_len,
+			      obj.indaddr, DT_UNKNOWN))
+			break;
+		pos += 26;
+		ctx->pos++;
+	}
+	return 0;
+}
+
+static int adfs_f_update(struct adfs_dir *dir, struct object_info *obj)
+{
+	struct adfs_direntry de;
+	int offset, ret;
+
+	offset = 5 - (int)sizeof(de);
+
+	do {
+		offset += sizeof(de);
+		ret = adfs_dir_copyfrom(&de, dir, offset, sizeof(de));
+		if (ret) {
+			adfs_error(dir->sb, "error reading directory entry");
+			return -ENOENT;
+		}
+		if (!de.dirobname[0]) {
+			adfs_error(dir->sb, "unable to locate entry to update");
+			return -ENOENT;
+		}
+	} while (adfs_readval(de.dirinddiscadd, 3) != obj->indaddr);
+
+	/* Update the directory entry with the new object state */
+	adfs_obj2dir(&de, obj);
+
+	/* Write the directory entry back to the directory */
+	return adfs_dir_copyto(dir, offset, &de, 26);
+}
+
+static int adfs_f_commit(struct adfs_dir *dir)
+{
+	int ret;
+
+	/* Increment directory sequence number */
+	dir->dirhead->startmasseq += 1;
+	dir->newtail->endmasseq += 1;
+
+	/* Update directory check byte */
+	dir->newtail->dircheckbyte = adfs_dir_checkbyte(dir);
+
+	/* Make sure the directory still validates correctly */
+	ret = adfs_f_validate(dir);
+	if (ret)
+		adfs_msg(dir->sb, KERN_ERR, "error: update broke directory");
+
+	return ret;
+}
+
+const struct adfs_dir_ops adfs_f_dir_ops = {
+	.read		= adfs_f_read,
+	.iterate	= adfs_f_iterate,
+	.setpos		= adfs_f_setpos,
+	.getnext	= adfs_f_getnext,
+	.update		= adfs_f_update,
+	.commit		= adfs_f_commit,
+};