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, 368 insertions, 0 deletions

diff --git a/fs/hpfs/ea.c b/fs/hpfs/ea.c
new file mode 100644
index 000000000..4664f9ab0
--- /dev/null
+++ b/fs/hpfs/ea.c
@@ -0,0 +1,368 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *  linux/fs/hpfs/ea.c
+ *
+ *  Mikulas Patocka (mikulas@artax.karlin.mff.cuni.cz), 1998-1999
+ *
+ *  handling extended attributes
+ */
+
+#include "hpfs_fn.h"
+
+/* Remove external extended attributes. ano specifies whether a is a 
+   direct sector where eas starts or an anode */
+
+void hpfs_ea_ext_remove(struct super_block *s, secno a, int ano, unsigned len)
+{
+	unsigned pos = 0;
+	while (pos < len) {
+		char ex[4 + 255 + 1 + 8];
+		struct extended_attribute *ea = (struct extended_attribute *)ex;
+		if (pos + 4 > len) {
+			hpfs_error(s, "EAs don't end correctly, %s %08x, len %08x",
+				ano ? "anode" : "sectors", a, len);
+			return;
+		}
+		if (hpfs_ea_read(s, a, ano, pos, 4, ex)) return;
+		if (ea_indirect(ea)) {
+			if (ea_valuelen(ea) != 8) {
+				hpfs_error(s, "ea_indirect(ea) set while ea->valuelen!=8, %s %08x, pos %08x",
+					ano ? "anode" : "sectors", a, pos);
+				return;
+			}
+			if (hpfs_ea_read(s, a, ano, pos + 4, ea->namelen + 9, ex+4))
+				return;
+			hpfs_ea_remove(s, ea_sec(ea), ea_in_anode(ea), ea_len(ea));
+		}
+		pos += ea->namelen + ea_valuelen(ea) + 5;
+	}
+	if (!ano) hpfs_free_sectors(s, a, (len+511) >> 9);
+	else {
+		struct buffer_head *bh;
+		struct anode *anode;
+		if ((anode = hpfs_map_anode(s, a, &bh))) {
+			hpfs_remove_btree(s, GET_BTREE_PTR(&anode->btree));
+			brelse(bh);
+			hpfs_free_sectors(s, a, 1);
+		}
+	}
+}
+
+static char *get_indirect_ea(struct super_block *s, int ano, secno a, int size)
+{
+	char *ret;
+	if (!(ret = kmalloc(size + 1, GFP_NOFS))) {
+		pr_err("out of memory for EA\n");
+		return NULL;
+	}
+	if (hpfs_ea_read(s, a, ano, 0, size, ret)) {
+		kfree(ret);
+		return NULL;
+	}
+	ret[size] = 0;
+	return ret;
+}
+
+static void set_indirect_ea(struct super_block *s, int ano, secno a,
+			    const char *data, int size)
+{
+	hpfs_ea_write(s, a, ano, 0, size, data);
+}
+
+/* Read an extended attribute named 'key' into the provided buffer */
+
+int hpfs_read_ea(struct super_block *s, struct fnode *fnode, char *key,
+		char *buf, int size)
+{
+	unsigned pos;
+	int ano, len;
+	secno a;
+	char ex[4 + 255 + 1 + 8];
+	struct extended_attribute *ea;
+	struct extended_attribute *ea_end = fnode_end_ea(fnode);
+	for (ea = fnode_ea(fnode); ea < ea_end; ea = next_ea(ea))
+		if (!strcmp(ea->name, key)) {
+			if (ea_indirect(ea))
+				goto indirect;
+			if (ea_valuelen(ea) >= size)
+				return -EINVAL;
+			memcpy(buf, ea_data(ea), ea_valuelen(ea));
+			buf[ea_valuelen(ea)] = 0;
+			return 0;
+		}
+	a = le32_to_cpu(fnode->ea_secno);
+	len = le32_to_cpu(fnode->ea_size_l);
+	ano = fnode_in_anode(fnode);
+	pos = 0;
+	while (pos < len) {
+		ea = (struct extended_attribute *)ex;
+		if (pos + 4 > len) {
+			hpfs_error(s, "EAs don't end correctly, %s %08x, len %08x",
+				ano ? "anode" : "sectors", a, len);
+			return -EIO;
+		}
+		if (hpfs_ea_read(s, a, ano, pos, 4, ex)) return -EIO;
+		if (hpfs_ea_read(s, a, ano, pos + 4, ea->namelen + 1 + (ea_indirect(ea) ? 8 : 0), ex + 4))
+			return -EIO;
+		if (!strcmp(ea->name, key)) {
+			if (ea_indirect(ea))
+				goto indirect;
+			if (ea_valuelen(ea) >= size)
+				return -EINVAL;
+			if (hpfs_ea_read(s, a, ano, pos + 4 + ea->namelen + 1, ea_valuelen(ea), buf))
+				return -EIO;
+			buf[ea_valuelen(ea)] = 0;
+			return 0;
+		}
+		pos += ea->namelen + ea_valuelen(ea) + 5;
+	}
+	return -ENOENT;
+indirect:
+	if (ea_len(ea) >= size)
+		return -EINVAL;
+	if (hpfs_ea_read(s, ea_sec(ea), ea_in_anode(ea), 0, ea_len(ea), buf))
+		return -EIO;
+	buf[ea_len(ea)] = 0;
+	return 0;
+}
+
+/* Read an extended attribute named 'key' */
+char *hpfs_get_ea(struct super_block *s, struct fnode *fnode, char *key, int *size)
+{
+	char *ret;
+	unsigned pos;
+	int ano, len;
+	secno a;
+	struct extended_attribute *ea;
+	struct extended_attribute *ea_end = fnode_end_ea(fnode);
+	for (ea = fnode_ea(fnode); ea < ea_end; ea = next_ea(ea))
+		if (!strcmp(ea->name, key)) {
+			if (ea_indirect(ea))
+				return get_indirect_ea(s, ea_in_anode(ea), ea_sec(ea), *size = ea_len(ea));
+			if (!(ret = kmalloc((*size = ea_valuelen(ea)) + 1, GFP_NOFS))) {
+				pr_err("out of memory for EA\n");
+				return NULL;
+			}
+			memcpy(ret, ea_data(ea), ea_valuelen(ea));
+			ret[ea_valuelen(ea)] = 0;
+			return ret;
+		}
+	a = le32_to_cpu(fnode->ea_secno);
+	len = le32_to_cpu(fnode->ea_size_l);
+	ano = fnode_in_anode(fnode);
+	pos = 0;
+	while (pos < len) {
+		char ex[4 + 255 + 1 + 8];
+		ea = (struct extended_attribute *)ex;
+		if (pos + 4 > len) {
+			hpfs_error(s, "EAs don't end correctly, %s %08x, len %08x",
+				ano ? "anode" : "sectors", a, len);
+			return NULL;
+		}
+		if (hpfs_ea_read(s, a, ano, pos, 4, ex)) return NULL;
+		if (hpfs_ea_read(s, a, ano, pos + 4, ea->namelen + 1 + (ea_indirect(ea) ? 8 : 0), ex + 4))
+			return NULL;
+		if (!strcmp(ea->name, key)) {
+			if (ea_indirect(ea))
+				return get_indirect_ea(s, ea_in_anode(ea), ea_sec(ea), *size = ea_len(ea));
+			if (!(ret = kmalloc((*size = ea_valuelen(ea)) + 1, GFP_NOFS))) {
+				pr_err("out of memory for EA\n");
+				return NULL;
+			}
+			if (hpfs_ea_read(s, a, ano, pos + 4 + ea->namelen + 1, ea_valuelen(ea), ret)) {
+				kfree(ret);
+				return NULL;
+			}
+			ret[ea_valuelen(ea)] = 0;
+			return ret;
+		}
+		pos += ea->namelen + ea_valuelen(ea) + 5;
+	}
+	return NULL;
+}
+
+/* 
+ * Update or create extended attribute 'key' with value 'data'. Note that
+ * when this ea exists, it MUST have the same size as size of data.
+ * This driver can't change sizes of eas ('cause I just don't need it).
+ */
+
+void hpfs_set_ea(struct inode *inode, struct fnode *fnode, const char *key,
+		 const char *data, int size)
+{
+	fnode_secno fno = inode->i_ino;
+	struct super_block *s = inode->i_sb;
+	unsigned pos;
+	int ano, len;
+	secno a;
+	unsigned char h[4];
+	struct extended_attribute *ea;
+	struct extended_attribute *ea_end = fnode_end_ea(fnode);
+	for (ea = fnode_ea(fnode); ea < ea_end; ea = next_ea(ea))
+		if (!strcmp(ea->name, key)) {
+			if (ea_indirect(ea)) {
+				if (ea_len(ea) == size)
+					set_indirect_ea(s, ea_in_anode(ea), ea_sec(ea), data, size);
+			} else if (ea_valuelen(ea) == size) {
+				memcpy(ea_data(ea), data, size);
+			}
+			return;
+		}
+	a = le32_to_cpu(fnode->ea_secno);
+	len = le32_to_cpu(fnode->ea_size_l);
+	ano = fnode_in_anode(fnode);
+	pos = 0;
+	while (pos < len) {
+		char ex[4 + 255 + 1 + 8];
+		ea = (struct extended_attribute *)ex;
+		if (pos + 4 > len) {
+			hpfs_error(s, "EAs don't end correctly, %s %08x, len %08x",
+				ano ? "anode" : "sectors", a, len);
+			return;
+		}
+		if (hpfs_ea_read(s, a, ano, pos, 4, ex)) return;
+		if (hpfs_ea_read(s, a, ano, pos + 4, ea->namelen + 1 + (ea_indirect(ea) ? 8 : 0), ex + 4))
+			return;
+		if (!strcmp(ea->name, key)) {
+			if (ea_indirect(ea)) {
+				if (ea_len(ea) == size)
+					set_indirect_ea(s, ea_in_anode(ea), ea_sec(ea), data, size);
+			}
+			else {
+				if (ea_valuelen(ea) == size)
+					hpfs_ea_write(s, a, ano, pos + 4 + ea->namelen + 1, size, data);
+			}
+			return;
+		}
+		pos += ea->namelen + ea_valuelen(ea) + 5;
+	}
+	if (!le16_to_cpu(fnode->ea_offs)) {
+		/*if (le16_to_cpu(fnode->ea_size_s)) {
+			hpfs_error(s, "fnode %08x: ea_size_s == %03x, ea_offs == 0",
+				inode->i_ino, le16_to_cpu(fnode->ea_size_s));
+			return;
+		}*/
+		fnode->ea_offs = cpu_to_le16(0xc4);
+	}
+	if (le16_to_cpu(fnode->ea_offs) < 0xc4 || le16_to_cpu(fnode->ea_offs) + le16_to_cpu(fnode->acl_size_s) + le16_to_cpu(fnode->ea_size_s) > 0x200) {
+		hpfs_error(s, "fnode %08llx: ea_offs == %03x, ea_size_s == %03x",
+			inode->i_ino,
+			le16_to_cpu(fnode->ea_offs), le16_to_cpu(fnode->ea_size_s));
+		return;
+	}
+	if ((le16_to_cpu(fnode->ea_size_s) || !le32_to_cpu(fnode->ea_size_l)) &&
+	     le16_to_cpu(fnode->ea_offs) + le16_to_cpu(fnode->acl_size_s) + le16_to_cpu(fnode->ea_size_s) + strlen(key) + size + 5 <= 0x200) {
+		ea = fnode_end_ea(fnode);
+		*(char *)ea = 0;
+		ea->namelen = strlen(key);
+		ea->valuelen_lo = size;
+		ea->valuelen_hi = size >> 8;
+		strcpy(ea->name, key);
+		memcpy(ea_data(ea), data, size);
+		fnode->ea_size_s = cpu_to_le16(le16_to_cpu(fnode->ea_size_s) + strlen(key) + size + 5);
+		goto ret;
+	}
+	/* Most the code here is 99.9993422% unused. I hope there are no bugs.
+	   But what .. HPFS.IFS has also bugs in ea management. */
+	if (le16_to_cpu(fnode->ea_size_s) && !le32_to_cpu(fnode->ea_size_l)) {
+		secno n;
+		struct buffer_head *bh;
+		char *data;
+		if (!(n = hpfs_alloc_sector(s, fno, 1, 0))) return;
+		if (!(data = hpfs_get_sector(s, n, &bh))) {
+			hpfs_free_sectors(s, n, 1);
+			return;
+		}
+		memcpy(data, fnode_ea(fnode), le16_to_cpu(fnode->ea_size_s));
+		fnode->ea_size_l = cpu_to_le32(le16_to_cpu(fnode->ea_size_s));
+		fnode->ea_size_s = cpu_to_le16(0);
+		fnode->ea_secno = cpu_to_le32(n);
+		fnode->flags &= ~FNODE_anode;
+		mark_buffer_dirty(bh);
+		brelse(bh);
+	}
+	pos = le32_to_cpu(fnode->ea_size_l) + 5 + strlen(key) + size;
+	len = (le32_to_cpu(fnode->ea_size_l) + 511) >> 9;
+	if (pos >= 30000) goto bail;
+	while (((pos + 511) >> 9) > len) {
+		if (!len) {
+			secno q = hpfs_alloc_sector(s, fno, 1, 0);
+			if (!q) goto bail;
+			fnode->ea_secno = cpu_to_le32(q);
+			fnode->flags &= ~FNODE_anode;
+			len++;
+		} else if (!fnode_in_anode(fnode)) {
+			if (hpfs_alloc_if_possible(s, le32_to_cpu(fnode->ea_secno) + len)) {
+				len++;
+			} else {
+				/* Aargh... don't know how to create ea anodes :-( */
+				/*struct buffer_head *bh;
+				struct anode *anode;
+				anode_secno a_s;
+				if (!(anode = hpfs_alloc_anode(s, fno, &a_s, &bh)))
+					goto bail;
+				anode->up = cpu_to_le32(fno);
+				anode->btree.fnode_parent = 1;
+				anode->btree.n_free_nodes--;
+				anode->btree.n_used_nodes++;
+				anode->btree.first_free = cpu_to_le16(le16_to_cpu(anode->btree.first_free) + 12);
+				anode->u.external[0].disk_secno = cpu_to_le32(le32_to_cpu(fnode->ea_secno));
+				anode->u.external[0].file_secno = cpu_to_le32(0);
+				anode->u.external[0].length = cpu_to_le32(len);
+				mark_buffer_dirty(bh);
+				brelse(bh);
+				fnode->flags |= FNODE_anode;
+				fnode->ea_secno = cpu_to_le32(a_s);*/
+				secno new_sec;
+				int i;
+				if (!(new_sec = hpfs_alloc_sector(s, fno, 1, 1 - ((pos + 511) >> 9))))
+					goto bail;
+				for (i = 0; i < len; i++) {
+					struct buffer_head *bh1, *bh2;
+					void *b1, *b2;
+					if (!(b1 = hpfs_map_sector(s, le32_to_cpu(fnode->ea_secno) + i, &bh1, len - i - 1))) {
+						hpfs_free_sectors(s, new_sec, (pos + 511) >> 9);
+						goto bail;
+					}
+					if (!(b2 = hpfs_get_sector(s, new_sec + i, &bh2))) {
+						brelse(bh1);
+						hpfs_free_sectors(s, new_sec, (pos + 511) >> 9);
+						goto bail;
+					}
+					memcpy(b2, b1, 512);
+					brelse(bh1);
+					mark_buffer_dirty(bh2);
+					brelse(bh2);
+				}
+				hpfs_free_sectors(s, le32_to_cpu(fnode->ea_secno), len);
+				fnode->ea_secno = cpu_to_le32(new_sec);
+				len = (pos + 511) >> 9;
+			}
+		}
+		if (fnode_in_anode(fnode)) {
+			if (hpfs_add_sector_to_btree(s, le32_to_cpu(fnode->ea_secno),
+						     0, len) != -1) {
+				len++;
+			} else {
+				goto bail;
+			}
+		}
+	}
+	h[0] = 0;
+	h[1] = strlen(key);
+	h[2] = size & 0xff;
+	h[3] = size >> 8;
+	if (hpfs_ea_write(s, le32_to_cpu(fnode->ea_secno), fnode_in_anode(fnode), le32_to_cpu(fnode->ea_size_l), 4, h)) goto bail;
+	if (hpfs_ea_write(s, le32_to_cpu(fnode->ea_secno), fnode_in_anode(fnode), le32_to_cpu(fnode->ea_size_l) + 4, h[1] + 1, key)) goto bail;
+	if (hpfs_ea_write(s, le32_to_cpu(fnode->ea_secno), fnode_in_anode(fnode), le32_to_cpu(fnode->ea_size_l) + 5 + h[1], size, data)) goto bail;
+	fnode->ea_size_l = cpu_to_le32(pos);
+	ret:
+	hpfs_i(inode)->i_ea_size += 5 + strlen(key) + size;
+	return;
+	bail:
+	if (le32_to_cpu(fnode->ea_secno))
+		if (fnode_in_anode(fnode)) hpfs_truncate_btree(s, le32_to_cpu(fnode->ea_secno), 1, (le32_to_cpu(fnode->ea_size_l) + 511) >> 9);
+		else hpfs_free_sectors(s, le32_to_cpu(fnode->ea_secno) + ((le32_to_cpu(fnode->ea_size_l) + 511) >> 9), len - ((le32_to_cpu(fnode->ea_size_l) + 511) >> 9));
+	else fnode->ea_secno = fnode->ea_size_l = cpu_to_le32(0);
+}
+