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, 454 insertions, 0 deletions

diff --git a/fs/smb/server/unicode.c b/fs/smb/server/unicode.c
new file mode 100644
index 000000000..85e679174
--- /dev/null
+++ b/fs/smb/server/unicode.c
@@ -0,0 +1,454 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ *   Some of the source code in this file came from fs/cifs/cifs_unicode.c
+ *
+ *   Copyright (c) International Business Machines  Corp., 2000,2009
+ *   Modified by Steve French (sfrench@us.ibm.com)
+ *   Modified by Namjae Jeon (linkinjeon@kernel.org)
+ */
+#include <linux/fs.h>
+#include <linux/slab.h>
+#include <linux/unaligned.h>
+#include "glob.h"
+#include "unicode.h"
+#include "smb_common.h"
+
+/*
+ * cifs_mapchar() - convert a host-endian char to proper char in codepage
+ * @target:	where converted character should be copied
+ * @from:	host-endian source string
+ * @cp:		codepage to which character should be converted
+ * @mapchar:	should character be mapped according to mapchars mount option?
+ *
+ * This function handles the conversion of a single character. It is the
+ * responsibility of the caller to ensure that the target buffer is large
+ * enough to hold the result of the conversion (at least NLS_MAX_CHARSET_SIZE).
+ *
+ * Return:	string length after conversion
+ */
+static int
+cifs_mapchar(char *target, const __u16 *from, const struct nls_table *cp,
+	     bool mapchar)
+{
+	int len = 1;
+	__u16 src_char;
+
+	src_char = *from;
+
+	if (!mapchar)
+		goto cp_convert;
+
+	/*
+	 * BB: Cannot handle remapping UNI_SLASH until all the calls to
+	 *     build_path_from_dentry are modified, as they use slash as
+	 *     separator.
+	 */
+	switch (src_char) {
+	case UNI_COLON:
+		*target = ':';
+		break;
+	case UNI_ASTERISK:
+		*target = '*';
+		break;
+	case UNI_QUESTION:
+		*target = '?';
+		break;
+	case UNI_PIPE:
+		*target = '|';
+		break;
+	case UNI_GRTRTHAN:
+		*target = '>';
+		break;
+	case UNI_LESSTHAN:
+		*target = '<';
+		break;
+	default:
+		goto cp_convert;
+	}
+
+out:
+	return len;
+
+cp_convert:
+	len = cp->uni2char(src_char, target, NLS_MAX_CHARSET_SIZE);
+	if (len <= 0)
+		goto surrogate_pair;
+
+	goto out;
+
+surrogate_pair:
+	/* convert SURROGATE_PAIR and IVS */
+	if (strcmp(cp->charset, "utf8"))
+		goto unknown;
+	len = utf16s_to_utf8s(from, 3, UTF16_LITTLE_ENDIAN, target, 6);
+	if (len <= 0)
+		goto unknown;
+	return len;
+
+unknown:
+	*target = '?';
+	len = 1;
+	goto out;
+}
+
+/*
+ * smb_utf16_bytes() - compute converted string length
+ * @from:	pointer to input string
+ * @maxbytes:	input string length
+ * @codepage:	destination codepage
+ *
+ * Walk a utf16le string and return the number of bytes that the string will
+ * be after being converted to the given charset, not including any null
+ * termination required. Don't walk past maxbytes in the source buffer.
+ *
+ * Return:	string length after conversion
+ */
+static int smb_utf16_bytes(const __le16 *from, int maxbytes,
+			   const struct nls_table *codepage)
+{
+	int i, j;
+	int charlen, outlen = 0;
+	int maxwords = maxbytes / 2;
+	char tmp[NLS_MAX_CHARSET_SIZE];
+	__u16 ftmp[3];
+
+	for (i = 0; i < maxwords; i++) {
+		ftmp[0] = get_unaligned_le16(&from[i]);
+		if (ftmp[0] == 0)
+			break;
+		for (j = 1; j <= 2; j++) {
+			if (i + j < maxwords)
+				ftmp[j] = get_unaligned_le16(&from[i + j]);
+			else
+				ftmp[j] = 0;
+		}
+
+		charlen = cifs_mapchar(tmp, ftmp, codepage, 0);
+		if (charlen > 0)
+			outlen += charlen;
+		else
+			outlen++;
+	}
+
+	return outlen;
+}
+
+/*
+ * smb_from_utf16() - convert utf16le string to local charset
+ * @to:		destination buffer
+ * @from:	source buffer
+ * @tolen:	destination buffer size (in bytes)
+ * @fromlen:	source buffer size (in bytes)
+ * @codepage:	codepage to which characters should be converted
+ * @mapchar:	should characters be remapped according to the mapchars option?
+ *
+ * Convert a little-endian utf16le string (as sent by the server) to a string
+ * in the provided codepage. The tolen and fromlen parameters are to ensure
+ * that the code doesn't walk off of the end of the buffer (which is always
+ * a danger if the alignment of the source buffer is off). The destination
+ * string is always properly null terminated and fits in the destination
+ * buffer. Returns the length of the destination string in bytes (including
+ * null terminator).
+ *
+ * Note that some windows versions actually send multiword UTF-16 characters
+ * instead of straight UTF16-2. The linux nls routines however aren't able to
+ * deal with those characters properly. In the event that we get some of
+ * those characters, they won't be translated properly.
+ *
+ * Return:	string length after conversion
+ */
+static int smb_from_utf16(char *to, const __le16 *from, int tolen, int fromlen,
+			  const struct nls_table *codepage, bool mapchar)
+{
+	int i, j, charlen, safelen;
+	int outlen = 0;
+	int nullsize = nls_nullsize(codepage);
+	int fromwords = fromlen / 2;
+	char tmp[NLS_MAX_CHARSET_SIZE];
+	__u16 ftmp[3];	/* ftmp[3] = 3array x 2bytes = 6bytes UTF-16 */
+
+	/*
+	 * because the chars can be of varying widths, we need to take care
+	 * not to overflow the destination buffer when we get close to the
+	 * end of it. Until we get to this offset, we don't need to check
+	 * for overflow however.
+	 */
+	safelen = tolen - (NLS_MAX_CHARSET_SIZE + nullsize);
+
+	for (i = 0; i < fromwords; i++) {
+		ftmp[0] = get_unaligned_le16(&from[i]);
+		if (ftmp[0] == 0)
+			break;
+		for (j = 1; j <= 2; j++) {
+			if (i + j < fromwords)
+				ftmp[j] = get_unaligned_le16(&from[i + j]);
+			else
+				ftmp[j] = 0;
+		}
+
+		/*
+		 * check to see if converting this character might make the
+		 * conversion bleed into the null terminator
+		 */
+		if (outlen >= safelen) {
+			charlen = cifs_mapchar(tmp, ftmp, codepage, mapchar);
+			if ((outlen + charlen) > (tolen - nullsize))
+				break;
+		}
+
+		/* put converted char into 'to' buffer */
+		charlen = cifs_mapchar(&to[outlen], ftmp, codepage, mapchar);
+		outlen += charlen;
+
+		/*
+		 * charlen (=bytes of UTF-8 for 1 character)
+		 * 4bytes UTF-8(surrogate pair) is charlen=4
+		 * (4bytes UTF-16 code)
+		 * 7-8bytes UTF-8(IVS) is charlen=3+4 or 4+4
+		 * (2 UTF-8 pairs divided to 2 UTF-16 pairs)
+		 */
+		if (charlen == 4)
+			i++;
+		else if (charlen >= 5)
+			/* 5-6bytes UTF-8 */
+			i += 2;
+	}
+
+	/* properly null-terminate string */
+	for (i = 0; i < nullsize; i++)
+		to[outlen++] = 0;
+
+	return outlen;
+}
+
+/*
+ * smb_strtoUTF16() - Convert character string to unicode string
+ * @to:		destination buffer
+ * @from:	source buffer
+ * @len:	destination buffer size (in bytes)
+ * @codepage:	codepage to which characters should be converted
+ *
+ * Return:	string length after conversion
+ */
+int smb_strtoUTF16(__le16 *to, const char *from, int len,
+		   const struct nls_table *codepage)
+{
+	int charlen;
+	int i;
+	wchar_t wchar_to; /* needed to quiet sparse */
+
+	/* special case for utf8 to handle no plane0 chars */
+	if (!strcmp(codepage->charset, "utf8")) {
+		/*
+		 * convert utf8 -> utf16, we assume we have enough space
+		 * as caller should have assumed conversion does not overflow
+		 * in destination len is length in wchar_t units (16bits)
+		 */
+		i  = utf8s_to_utf16s(from, len, UTF16_LITTLE_ENDIAN,
+				     (wchar_t *)to, len);
+
+		/* if success terminate and exit */
+		if (i >= 0)
+			goto success;
+		/*
+		 * if fails fall back to UCS encoding as this
+		 * function should not return negative values
+		 * currently can fail only if source contains
+		 * invalid encoded characters
+		 */
+	}
+
+	for (i = 0; len > 0 && *from; i++, from += charlen, len -= charlen) {
+		charlen = codepage->char2uni(from, len, &wchar_to);
+		if (charlen < 1) {
+			/* A question mark */
+			wchar_to = 0x003f;
+			charlen = 1;
+		}
+		put_unaligned_le16(wchar_to, &to[i]);
+	}
+
+success:
+	put_unaligned_le16(0, &to[i]);
+	return i;
+}
+
+/*
+ * smb_strndup_from_utf16() - copy a string from wire format to the local
+ *		codepage
+ * @src:	source string
+ * @maxlen:	don't walk past this many bytes in the source string
+ * @is_unicode:	is this a unicode string?
+ * @codepage:	destination codepage
+ *
+ * Take a string given by the server, convert it to the local codepage and
+ * put it in a new buffer. Returns a pointer to the new string or NULL on
+ * error.
+ *
+ * Return:	destination string buffer or error ptr
+ */
+char *smb_strndup_from_utf16(const char *src, const int maxlen,
+			     const bool is_unicode,
+			     const struct nls_table *codepage)
+{
+	int len, ret;
+	char *dst;
+
+	if (is_unicode) {
+		len = smb_utf16_bytes((__le16 *)src, maxlen, codepage);
+		len += nls_nullsize(codepage);
+		dst = kmalloc(len, KSMBD_DEFAULT_GFP);
+		if (!dst)
+			return ERR_PTR(-ENOMEM);
+		ret = smb_from_utf16(dst, (__le16 *)src, len, maxlen, codepage,
+				     false);
+		if (ret < 0) {
+			kfree(dst);
+			return ERR_PTR(-EINVAL);
+		}
+	} else {
+		len = strnlen(src, maxlen);
+		len++;
+		dst = kmalloc(len, KSMBD_DEFAULT_GFP);
+		if (!dst)
+			return ERR_PTR(-ENOMEM);
+		strscpy(dst, src, len);
+	}
+
+	return dst;
+}
+
+/*
+ * Convert 16 bit Unicode pathname to wire format from string in current code
+ * page. Conversion may involve remapping up the six characters that are
+ * only legal in POSIX-like OS (if they are present in the string). Path
+ * names are little endian 16 bit Unicode on the wire
+ */
+/*
+ * smbConvertToUTF16() - convert string from local charset to utf16
+ * @target:	destination buffer
+ * @source:	source buffer
+ * @srclen:	source buffer size (in bytes)
+ * @cp:		codepage to which characters should be converted
+ * @mapchar:	should characters be remapped according to the mapchars option?
+ *
+ * Convert 16 bit Unicode pathname to wire format from string in current code
+ * page. Conversion may involve remapping up the six characters that are
+ * only legal in POSIX-like OS (if they are present in the string). Path
+ * names are little endian 16 bit Unicode on the wire
+ *
+ * Return:	char length after conversion
+ */
+int smbConvertToUTF16(__le16 *target, const char *source, int srclen,
+		      const struct nls_table *cp, int mapchars)
+{
+	int i, j, charlen;
+	char src_char;
+	__le16 dst_char;
+	wchar_t tmp;
+	wchar_t wchar_to[6];	/* UTF-16 */
+	int ret;
+	unicode_t u;
+
+	if (!mapchars)
+		return smb_strtoUTF16(target, source, srclen, cp);
+
+	for (i = 0, j = 0; i < srclen; j++) {
+		src_char = source[i];
+		charlen = 1;
+		switch (src_char) {
+		case 0:
+			put_unaligned(0, &target[j]);
+			return j;
+		case ':':
+			dst_char = cpu_to_le16(UNI_COLON);
+			break;
+		case '*':
+			dst_char = cpu_to_le16(UNI_ASTERISK);
+			break;
+		case '?':
+			dst_char = cpu_to_le16(UNI_QUESTION);
+			break;
+		case '<':
+			dst_char = cpu_to_le16(UNI_LESSTHAN);
+			break;
+		case '>':
+			dst_char = cpu_to_le16(UNI_GRTRTHAN);
+			break;
+		case '|':
+			dst_char = cpu_to_le16(UNI_PIPE);
+			break;
+		/*
+		 * FIXME: We can not handle remapping backslash (UNI_SLASH)
+		 * until all the calls to build_path_from_dentry are modified,
+		 * as they use backslash as separator.
+		 */
+		default:
+			charlen = cp->char2uni(source + i, srclen - i, &tmp);
+			dst_char = cpu_to_le16(tmp);
+
+			/*
+			 * if no match, use question mark, which at least in
+			 * some cases serves as wild card
+			 */
+			if (charlen > 0)
+				goto ctoUTF16;
+
+			/* convert SURROGATE_PAIR */
+			if (strcmp(cp->charset, "utf8"))
+				goto unknown;
+			if (*(source + i) & 0x80) {
+				charlen = utf8_to_utf32(source + i, 6, &u);
+				if (charlen < 0)
+					goto unknown;
+			} else
+				goto unknown;
+			ret  = utf8s_to_utf16s(source + i, charlen,
+					UTF16_LITTLE_ENDIAN,
+					wchar_to, 6);
+			if (ret < 0)
+				goto unknown;
+
+			i += charlen;
+			dst_char = cpu_to_le16(*wchar_to);
+			if (charlen <= 3)
+				/* 1-3bytes UTF-8 to 2bytes UTF-16 */
+				put_unaligned(dst_char, &target[j]);
+			else if (charlen == 4) {
+				/*
+				 * 4bytes UTF-8(surrogate pair) to 4bytes UTF-16
+				 * 7-8bytes UTF-8(IVS) divided to 2 UTF-16
+				 * (charlen=3+4 or 4+4)
+				 */
+				put_unaligned(dst_char, &target[j]);
+				dst_char = cpu_to_le16(*(wchar_to + 1));
+				j++;
+				put_unaligned(dst_char, &target[j]);
+			} else if (charlen >= 5) {
+				/* 5-6bytes UTF-8 to 6bytes UTF-16 */
+				put_unaligned(dst_char, &target[j]);
+				dst_char = cpu_to_le16(*(wchar_to + 1));
+				j++;
+				put_unaligned(dst_char, &target[j]);
+				dst_char = cpu_to_le16(*(wchar_to + 2));
+				j++;
+				put_unaligned(dst_char, &target[j]);
+			}
+			continue;
+
+unknown:
+			dst_char = cpu_to_le16(0x003f);
+			charlen = 1;
+		}
+
+ctoUTF16:
+		/*
+		 * character may take more than one byte in the source string,
+		 * but will take exactly two bytes in the target string
+		 */
+		i += charlen;
+		put_unaligned(dst_char, &target[j]);
+	}
+
+	return j;
+}