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, 543 insertions, 0 deletions

diff --git a/fs/udf/directory.c b/fs/udf/directory.c
new file mode 100644
index 000000000..6b4ac7464
--- /dev/null
+++ b/fs/udf/directory.c
@@ -0,0 +1,543 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * directory.c
+ *
+ * PURPOSE
+ *	Directory related functions
+ *
+ */
+
+#include "udfdecl.h"
+#include "udf_i.h"
+
+#include <linux/fs.h>
+#include <linux/string.h>
+#include <linux/bio.h>
+#include <linux/crc-itu-t.h>
+#include <linux/iversion.h>
+
+static int udf_verify_fi(struct udf_fileident_iter *iter)
+{
+	unsigned int len;
+
+	if (iter->fi.descTag.tagIdent != cpu_to_le16(TAG_IDENT_FID)) {
+		udf_err(iter->dir->i_sb,
+			"directory (ino %llu) has entry at pos %llu with incorrect tag %x\n",
+			iter->dir->i_ino, (unsigned long long)iter->pos,
+			le16_to_cpu(iter->fi.descTag.tagIdent));
+		return -EFSCORRUPTED;
+	}
+	len = udf_dir_entry_len(&iter->fi);
+	if (le16_to_cpu(iter->fi.lengthOfImpUse) & 3) {
+		udf_err(iter->dir->i_sb,
+			"directory (ino %llu) has entry at pos %llu with unaligned length of impUse field\n",
+			iter->dir->i_ino, (unsigned long long)iter->pos);
+		return -EFSCORRUPTED;
+	}
+	/*
+	 * This is in fact allowed by the spec due to long impUse field but
+	 * we don't support it. If there is real media with this large impUse
+	 * field, support can be added.
+	 */
+	if (len > 1 << iter->dir->i_blkbits) {
+		udf_err(iter->dir->i_sb,
+			"directory (ino %llu) has too big (%u) entry at pos %llu\n",
+			iter->dir->i_ino, len, (unsigned long long)iter->pos);
+		return -EFSCORRUPTED;
+	}
+	if (iter->pos + len > iter->dir->i_size) {
+		udf_err(iter->dir->i_sb,
+			"directory (ino %llu) has entry past directory size at pos %llu\n",
+			iter->dir->i_ino, (unsigned long long)iter->pos);
+		return -EFSCORRUPTED;
+	}
+	if (udf_dir_entry_len(&iter->fi) !=
+	    sizeof(struct tag) + le16_to_cpu(iter->fi.descTag.descCRCLength)) {
+		udf_err(iter->dir->i_sb,
+			"directory (ino %llu) has entry where CRC length (%u) does not match entry length (%u)\n",
+			iter->dir->i_ino,
+			(unsigned)le16_to_cpu(iter->fi.descTag.descCRCLength),
+			(unsigned)(udf_dir_entry_len(&iter->fi) -
+							sizeof(struct tag)));
+		return -EFSCORRUPTED;
+	}
+	return 0;
+}
+
+static int udf_copy_fi(struct udf_fileident_iter *iter)
+{
+	struct udf_inode_info *iinfo = UDF_I(iter->dir);
+	u32 blksize = 1 << iter->dir->i_blkbits;
+	u32 off, len, nameoff;
+	int err;
+
+	/* Skip copying when we are at EOF */
+	if (iter->pos >= iter->dir->i_size) {
+		iter->name = NULL;
+		return 0;
+	}
+	if (iter->dir->i_size < iter->pos + sizeof(struct fileIdentDesc)) {
+		udf_err(iter->dir->i_sb,
+			"directory (ino %llu) has entry straddling EOF\n",
+			iter->dir->i_ino);
+		return -EFSCORRUPTED;
+	}
+	if (iinfo->i_alloc_type == ICBTAG_FLAG_AD_IN_ICB) {
+		memcpy(&iter->fi, iinfo->i_data + iinfo->i_lenEAttr + iter->pos,
+		       sizeof(struct fileIdentDesc));
+		err = udf_verify_fi(iter);
+		if (err < 0)
+			return err;
+		iter->name = iinfo->i_data + iinfo->i_lenEAttr + iter->pos +
+			sizeof(struct fileIdentDesc) +
+			le16_to_cpu(iter->fi.lengthOfImpUse);
+		return 0;
+	}
+
+	off = iter->pos & (blksize - 1);
+	len = min_t(u32, sizeof(struct fileIdentDesc), blksize - off);
+	memcpy(&iter->fi, iter->bh[0]->b_data + off, len);
+	if (len < sizeof(struct fileIdentDesc))
+		memcpy((char *)(&iter->fi) + len, iter->bh[1]->b_data,
+		       sizeof(struct fileIdentDesc) - len);
+	err = udf_verify_fi(iter);
+	if (err < 0)
+		return err;
+
+	/* Handle directory entry name */
+	nameoff = off + sizeof(struct fileIdentDesc) +
+				le16_to_cpu(iter->fi.lengthOfImpUse);
+	if (off + udf_dir_entry_len(&iter->fi) <= blksize) {
+		iter->name = iter->bh[0]->b_data + nameoff;
+	} else if (nameoff >= blksize) {
+		iter->name = iter->bh[1]->b_data + (nameoff - blksize);
+	} else {
+		iter->name = iter->namebuf;
+		len = blksize - nameoff;
+		memcpy(iter->name, iter->bh[0]->b_data + nameoff, len);
+		memcpy(iter->name + len, iter->bh[1]->b_data,
+		       iter->fi.lengthFileIdent - len);
+	}
+	return 0;
+}
+
+/* Readahead 8k once we are at 8k boundary */
+static void udf_readahead_dir(struct udf_fileident_iter *iter)
+{
+	unsigned int ralen = 16 >> (iter->dir->i_blkbits - 9);
+	struct buffer_head *tmp, *bha[16];
+	int i, num;
+	udf_pblk_t blk;
+
+	if (iter->loffset & (ralen - 1))
+		return;
+
+	if (iter->loffset + ralen > (iter->elen >> iter->dir->i_blkbits))
+		ralen = (iter->elen >> iter->dir->i_blkbits) - iter->loffset;
+	num = 0;
+	for (i = 0; i < ralen; i++) {
+		blk = udf_get_lb_pblock(iter->dir->i_sb, &iter->eloc,
+					iter->loffset + i);
+		tmp = sb_getblk(iter->dir->i_sb, blk);
+		if (tmp && !buffer_uptodate(tmp) && !buffer_locked(tmp))
+			bha[num++] = tmp;
+		else
+			brelse(tmp);
+	}
+	if (num) {
+		bh_readahead_batch(num, bha, REQ_RAHEAD);
+		for (i = 0; i < num; i++)
+			brelse(bha[i]);
+	}
+}
+
+static struct buffer_head *udf_fiiter_bread_blk(struct udf_fileident_iter *iter)
+{
+	udf_pblk_t blk;
+
+	udf_readahead_dir(iter);
+	blk = udf_get_lb_pblock(iter->dir->i_sb, &iter->eloc, iter->loffset);
+	return sb_bread(iter->dir->i_sb, blk);
+}
+
+/*
+ * Updates loffset to point to next directory block; eloc, elen & epos are
+ * updated if we need to traverse to the next extent as well.
+ */
+static int udf_fiiter_advance_blk(struct udf_fileident_iter *iter)
+{
+	int8_t etype = -1;
+	int err = 0;
+
+	iter->loffset++;
+	if (iter->loffset < DIV_ROUND_UP(iter->elen, 1<<iter->dir->i_blkbits))
+		return 0;
+
+	iter->loffset = 0;
+	err = udf_next_aext(iter->dir, &iter->epos, &iter->eloc,
+			    &iter->elen, &etype, 1);
+	if (err < 0)
+		return err;
+	else if (err == 0 || etype != (EXT_RECORDED_ALLOCATED >> 30)) {
+		if (iter->pos == iter->dir->i_size) {
+			iter->elen = 0;
+			return 0;
+		}
+		udf_err(iter->dir->i_sb,
+			"extent after position %llu not allocated in directory (ino %llu)\n",
+			(unsigned long long)iter->pos, iter->dir->i_ino);
+		return -EFSCORRUPTED;
+	}
+	return 0;
+}
+
+static int udf_fiiter_load_bhs(struct udf_fileident_iter *iter)
+{
+	int blksize = 1 << iter->dir->i_blkbits;
+	int off = iter->pos & (blksize - 1);
+	int err;
+	struct fileIdentDesc *fi;
+
+	/* Is there any further extent we can map from? */
+	if (!iter->bh[0] && iter->elen) {
+		iter->bh[0] = udf_fiiter_bread_blk(iter);
+		if (!iter->bh[0]) {
+			err = -ENOMEM;
+			goto out_brelse;
+		}
+		if (!buffer_uptodate(iter->bh[0])) {
+			err = -EIO;
+			goto out_brelse;
+		}
+	}
+	/* There's no next block so we are done */
+	if (iter->pos >= iter->dir->i_size)
+		return 0;
+	/* Need to fetch next block as well? */
+	if (off + sizeof(struct fileIdentDesc) > blksize)
+		goto fetch_next;
+	fi = (struct fileIdentDesc *)(iter->bh[0]->b_data + off);
+	/* Need to fetch next block to get name? */
+	if (off + udf_dir_entry_len(fi) > blksize) {
+fetch_next:
+		err = udf_fiiter_advance_blk(iter);
+		if (err)
+			goto out_brelse;
+		iter->bh[1] = udf_fiiter_bread_blk(iter);
+		if (!iter->bh[1]) {
+			err = -ENOMEM;
+			goto out_brelse;
+		}
+		if (!buffer_uptodate(iter->bh[1])) {
+			err = -EIO;
+			goto out_brelse;
+		}
+	}
+	return 0;
+out_brelse:
+	brelse(iter->bh[0]);
+	brelse(iter->bh[1]);
+	iter->bh[0] = iter->bh[1] = NULL;
+	return err;
+}
+
+int udf_fiiter_init(struct udf_fileident_iter *iter, struct inode *dir,
+		    loff_t pos)
+{
+	struct udf_inode_info *iinfo = UDF_I(dir);
+	int err = 0;
+	int8_t etype;
+
+	iter->dir = dir;
+	iter->bh[0] = iter->bh[1] = NULL;
+	iter->pos = pos;
+	iter->elen = 0;
+	iter->epos.bh = NULL;
+	iter->name = NULL;
+	/*
+	 * When directory is verified, we don't expect directory iteration to
+	 * fail and it can be difficult to undo without corrupting filesystem.
+	 * So just do not allow memory allocation failures here.
+	 */
+	iter->namebuf = kmalloc(UDF_NAME_LEN_CS0, GFP_KERNEL | __GFP_NOFAIL);
+
+	if (iinfo->i_alloc_type == ICBTAG_FLAG_AD_IN_ICB) {
+		err = udf_copy_fi(iter);
+		goto out;
+	}
+
+	err = inode_bmap(dir, iter->pos >> dir->i_blkbits, &iter->epos,
+			 &iter->eloc, &iter->elen, &iter->loffset, &etype);
+	if (err <= 0 || etype != (EXT_RECORDED_ALLOCATED >> 30)) {
+		if (pos == dir->i_size)
+			return 0;
+		udf_err(dir->i_sb,
+			"position %llu not allocated in directory (ino %llu)\n",
+			(unsigned long long)pos, dir->i_ino);
+		err = -EFSCORRUPTED;
+		goto out;
+	}
+	err = udf_fiiter_load_bhs(iter);
+	if (err < 0)
+		goto out;
+	err = udf_copy_fi(iter);
+out:
+	if (err < 0)
+		udf_fiiter_release(iter);
+	return err;
+}
+
+int udf_fiiter_advance(struct udf_fileident_iter *iter)
+{
+	unsigned int oldoff, len;
+	int blksize = 1 << iter->dir->i_blkbits;
+	int err;
+
+	oldoff = iter->pos & (blksize - 1);
+	len = udf_dir_entry_len(&iter->fi);
+	iter->pos += len;
+	if (UDF_I(iter->dir)->i_alloc_type != ICBTAG_FLAG_AD_IN_ICB) {
+		if (oldoff + len >= blksize) {
+			brelse(iter->bh[0]);
+			iter->bh[0] = NULL;
+			/* Next block already loaded? */
+			if (iter->bh[1]) {
+				iter->bh[0] = iter->bh[1];
+				iter->bh[1] = NULL;
+			} else {
+				err = udf_fiiter_advance_blk(iter);
+				if (err < 0)
+					return err;
+			}
+		}
+		err = udf_fiiter_load_bhs(iter);
+		if (err < 0)
+			return err;
+	}
+	return udf_copy_fi(iter);
+}
+
+void udf_fiiter_release(struct udf_fileident_iter *iter)
+{
+	iter->dir = NULL;
+	brelse(iter->bh[0]);
+	brelse(iter->bh[1]);
+	iter->bh[0] = iter->bh[1] = NULL;
+	kfree(iter->namebuf);
+	iter->namebuf = NULL;
+}
+
+static void udf_copy_to_bufs(void *buf1, int len1, void *buf2, int len2,
+			     int off, void *src, int len)
+{
+	int copy;
+
+	if (off >= len1) {
+		off -= len1;
+	} else {
+		copy = min(off + len, len1) - off;
+		memcpy(buf1 + off, src, copy);
+		src += copy;
+		len -= copy;
+		off = 0;
+	}
+	if (len > 0) {
+		if (WARN_ON_ONCE(off + len > len2 || !buf2))
+			return;
+		memcpy(buf2 + off, src, len);
+	}
+}
+
+static uint16_t udf_crc_fi_bufs(void *buf1, int len1, void *buf2, int len2,
+				int off, int len)
+{
+	int copy;
+	uint16_t crc = 0;
+
+	if (off >= len1) {
+		off -= len1;
+	} else {
+		copy = min(off + len, len1) - off;
+		crc = crc_itu_t(crc, buf1 + off, copy);
+		len -= copy;
+		off = 0;
+	}
+	if (len > 0) {
+		if (WARN_ON_ONCE(off + len > len2 || !buf2))
+			return 0;
+		crc = crc_itu_t(crc, buf2 + off, len);
+	}
+	return crc;
+}
+
+static void udf_copy_fi_to_bufs(char *buf1, int len1, char *buf2, int len2,
+				int off, struct fileIdentDesc *fi,
+				uint8_t *impuse, uint8_t *name)
+{
+	uint16_t crc;
+	int fioff = off;
+	int crcoff = off + sizeof(struct tag);
+	unsigned int crclen = udf_dir_entry_len(fi) - sizeof(struct tag);
+	char zeros[UDF_NAME_PAD] = {};
+	int endoff = off + udf_dir_entry_len(fi);
+
+	udf_copy_to_bufs(buf1, len1, buf2, len2, off, fi,
+			 sizeof(struct fileIdentDesc));
+	off += sizeof(struct fileIdentDesc);
+	if (impuse)
+		udf_copy_to_bufs(buf1, len1, buf2, len2, off, impuse,
+				 le16_to_cpu(fi->lengthOfImpUse));
+	off += le16_to_cpu(fi->lengthOfImpUse);
+	if (name) {
+		udf_copy_to_bufs(buf1, len1, buf2, len2, off, name,
+				 fi->lengthFileIdent);
+		off += fi->lengthFileIdent;
+		udf_copy_to_bufs(buf1, len1, buf2, len2, off, zeros,
+				 endoff - off);
+	}
+
+	crc = udf_crc_fi_bufs(buf1, len1, buf2, len2, crcoff, crclen);
+	fi->descTag.descCRC = cpu_to_le16(crc);
+	fi->descTag.descCRCLength = cpu_to_le16(crclen);
+	fi->descTag.tagChecksum = udf_tag_checksum(&fi->descTag);
+
+	udf_copy_to_bufs(buf1, len1, buf2, len2, fioff, fi, sizeof(struct tag));
+}
+
+void udf_fiiter_write_fi(struct udf_fileident_iter *iter, uint8_t *impuse)
+{
+	struct udf_inode_info *iinfo = UDF_I(iter->dir);
+	void *buf1, *buf2 = NULL;
+	int len1, len2 = 0, off;
+	int blksize = 1 << iter->dir->i_blkbits;
+
+	off = iter->pos & (blksize - 1);
+	if (iinfo->i_alloc_type == ICBTAG_FLAG_AD_IN_ICB) {
+		buf1 = iinfo->i_data + iinfo->i_lenEAttr;
+		len1 = iter->dir->i_size;
+	} else {
+		buf1 = iter->bh[0]->b_data;
+		len1 = blksize;
+		if (iter->bh[1]) {
+			buf2 = iter->bh[1]->b_data;
+			len2 = blksize;
+		}
+	}
+
+	udf_copy_fi_to_bufs(buf1, len1, buf2, len2, off, &iter->fi, impuse,
+			    iter->name == iter->namebuf ? iter->name : NULL);
+
+	if (iinfo->i_alloc_type == ICBTAG_FLAG_AD_IN_ICB) {
+		mark_inode_dirty(iter->dir);
+	} else {
+		mmb_mark_buffer_dirty(iter->bh[0], &iinfo->i_metadata_bhs);
+		if (iter->bh[1])
+			mmb_mark_buffer_dirty(iter->bh[1],
+					      &iinfo->i_metadata_bhs);
+	}
+	inode_inc_iversion(iter->dir);
+}
+
+void udf_fiiter_update_elen(struct udf_fileident_iter *iter, uint32_t new_elen)
+{
+	struct udf_inode_info *iinfo = UDF_I(iter->dir);
+	int diff = new_elen - iter->elen;
+
+	/* Skip update when we already went past the last extent */
+	if (!iter->elen)
+		return;
+	iter->elen = new_elen;
+	if (iinfo->i_alloc_type == ICBTAG_FLAG_AD_SHORT)
+		iter->epos.offset -= sizeof(struct short_ad);
+	else if (iinfo->i_alloc_type == ICBTAG_FLAG_AD_LONG)
+		iter->epos.offset -= sizeof(struct long_ad);
+	udf_write_aext(iter->dir, &iter->epos, &iter->eloc, iter->elen, 1);
+	iinfo->i_lenExtents += diff;
+	mark_inode_dirty(iter->dir);
+}
+
+/* Append new block to directory. @iter is expected to point at EOF */
+int udf_fiiter_append_blk(struct udf_fileident_iter *iter)
+{
+	struct udf_inode_info *iinfo = UDF_I(iter->dir);
+	int blksize = 1 << iter->dir->i_blkbits;
+	struct buffer_head *bh;
+	sector_t block;
+	uint32_t old_elen = iter->elen;
+	int err;
+	int8_t etype;
+
+	if (WARN_ON_ONCE(iinfo->i_alloc_type == ICBTAG_FLAG_AD_IN_ICB))
+		return -EINVAL;
+
+	/* Round up last extent in the file */
+	udf_fiiter_update_elen(iter, ALIGN(iter->elen, blksize));
+
+	/* Allocate new block and refresh mapping information */
+	block = iinfo->i_lenExtents >> iter->dir->i_blkbits;
+	bh = udf_bread(iter->dir, block, 1, &err);
+	if (!bh) {
+		udf_fiiter_update_elen(iter, old_elen);
+		return err;
+	}
+	err = inode_bmap(iter->dir, block, &iter->epos, &iter->eloc, &iter->elen,
+		   &iter->loffset, &etype);
+	if (err <= 0 || etype != (EXT_RECORDED_ALLOCATED >> 30)) {
+		udf_err(iter->dir->i_sb,
+			"block %llu not allocated in directory (ino %llu)\n",
+			(unsigned long long)block, iter->dir->i_ino);
+		return -EFSCORRUPTED;
+	}
+	if (!(iter->pos & (blksize - 1))) {
+		brelse(iter->bh[0]);
+		iter->bh[0] = bh;
+	} else {
+		iter->bh[1] = bh;
+	}
+	return 0;
+}
+
+struct short_ad *udf_get_fileshortad(uint8_t *ptr, int maxoffset, uint32_t *offset,
+			      int inc)
+{
+	struct short_ad *sa;
+
+	if ((!ptr) || (!offset)) {
+		pr_err("%s: invalidparms\n", __func__);
+		return NULL;
+	}
+
+	if ((*offset + sizeof(struct short_ad)) > maxoffset)
+		return NULL;
+	else {
+		sa = (struct short_ad *)ptr;
+		if (sa->extLength == 0)
+			return NULL;
+	}
+
+	if (inc)
+		*offset += sizeof(struct short_ad);
+	return sa;
+}
+
+struct long_ad *udf_get_filelongad(uint8_t *ptr, int maxoffset, uint32_t *offset, int inc)
+{
+	struct long_ad *la;
+
+	if ((!ptr) || (!offset)) {
+		pr_err("%s: invalidparms\n", __func__);
+		return NULL;
+	}
+
+	if ((*offset + sizeof(struct long_ad)) > maxoffset)
+		return NULL;
+	else {
+		la = (struct long_ad *)ptr;
+		if (la->extLength == 0)
+			return NULL;
+	}
+
+	if (inc)
+		*offset += sizeof(struct long_ad);
+	return la;
+}