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, 361 insertions, 0 deletions

diff --git a/drivers/mtd/mtdblock.c b/drivers/mtd/mtdblock.c
new file mode 100644
index 000000000..c9143711f
--- /dev/null
+++ b/drivers/mtd/mtdblock.c
@@ -0,0 +1,361 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Direct MTD block device access
+ *
+ * Copyright © 1999-2010 David Woodhouse <dwmw2@infradead.org>
+ * Copyright © 2000-2003 Nicolas Pitre <nico@fluxnic.net>
+ */
+
+#include <linux/fs.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+#include <linux/vmalloc.h>
+
+#include <linux/mtd/mtd.h>
+#include <linux/mtd/blktrans.h>
+#include <linux/mutex.h>
+#include <linux/major.h>
+
+
+struct mtdblk_dev {
+	struct mtd_blktrans_dev mbd;
+	int count;
+	struct mutex cache_mutex;
+	unsigned char *cache_data;
+	unsigned long cache_offset;
+	unsigned int cache_size;
+	enum { STATE_EMPTY, STATE_CLEAN, STATE_DIRTY } cache_state;
+};
+
+/*
+ * Cache stuff...
+ *
+ * Since typical flash erasable sectors are much larger than what Linux's
+ * buffer cache can handle, we must implement read-modify-write on flash
+ * sectors for each block write requests.  To avoid over-erasing flash sectors
+ * and to speed things up, we locally cache a whole flash sector while it is
+ * being written to until a different sector is required.
+ */
+
+static int erase_write (struct mtd_info *mtd, unsigned long pos,
+			unsigned int len, const char *buf)
+{
+	struct erase_info erase;
+	size_t retlen;
+	int ret;
+
+	/*
+	 * First, let's erase the flash block.
+	 */
+	erase.addr = pos;
+	erase.len = len;
+
+	ret = mtd_erase(mtd, &erase);
+	if (ret) {
+		printk (KERN_WARNING "mtdblock: erase of region [0x%lx, 0x%x] "
+				     "on \"%s\" failed\n",
+			pos, len, mtd->name);
+		return ret;
+	}
+
+	/*
+	 * Next, write the data to flash.
+	 */
+
+	ret = mtd_write(mtd, pos, len, &retlen, buf);
+	if (ret)
+		return ret;
+	if (retlen != len)
+		return -EIO;
+	return 0;
+}
+
+
+static int write_cached_data (struct mtdblk_dev *mtdblk)
+{
+	struct mtd_info *mtd = mtdblk->mbd.mtd;
+	int ret;
+
+	if (mtdblk->cache_state != STATE_DIRTY)
+		return 0;
+
+	pr_debug("mtdblock: writing cached data for \"%s\" "
+			"at 0x%lx, size 0x%x\n", mtd->name,
+			mtdblk->cache_offset, mtdblk->cache_size);
+
+	ret = erase_write (mtd, mtdblk->cache_offset,
+			   mtdblk->cache_size, mtdblk->cache_data);
+
+	/*
+	 * Here we could arguably set the cache state to STATE_CLEAN.
+	 * However this could lead to inconsistency since we will not
+	 * be notified if this content is altered on the flash by other
+	 * means.  Let's declare it empty and leave buffering tasks to
+	 * the buffer cache instead.
+	 *
+	 * If this cache_offset points to a bad block, data cannot be
+	 * written to the device. Clear cache_state to avoid writing to
+	 * bad blocks repeatedly.
+	 */
+	if (ret == 0 || ret == -EIO)
+		mtdblk->cache_state = STATE_EMPTY;
+	return ret;
+}
+
+
+static int do_cached_write (struct mtdblk_dev *mtdblk, unsigned long pos,
+			    int len, const char *buf)
+{
+	struct mtd_info *mtd = mtdblk->mbd.mtd;
+	unsigned int sect_size = mtdblk->cache_size;
+	size_t retlen;
+	int ret;
+
+	pr_debug("mtdblock: write on \"%s\" at 0x%lx, size 0x%x\n",
+		mtd->name, pos, len);
+
+	if (!sect_size)
+		return mtd_write(mtd, pos, len, &retlen, buf);
+
+	while (len > 0) {
+		unsigned long sect_start = (pos/sect_size)*sect_size;
+		unsigned int offset = pos - sect_start;
+		unsigned int size = sect_size - offset;
+		if( size > len )
+			size = len;
+
+		if (size == sect_size) {
+			/*
+			 * We are covering a whole sector.  Thus there is no
+			 * need to bother with the cache while it may still be
+			 * useful for other partial writes.
+			 */
+			ret = erase_write (mtd, pos, size, buf);
+			if (ret)
+				return ret;
+		} else {
+			/* Partial sector: need to use the cache */
+
+			if (mtdblk->cache_state == STATE_DIRTY &&
+			    mtdblk->cache_offset != sect_start) {
+				ret = write_cached_data(mtdblk);
+				if (ret)
+					return ret;
+			}
+
+			if (mtdblk->cache_state == STATE_EMPTY ||
+			    mtdblk->cache_offset != sect_start) {
+				/* fill the cache with the current sector */
+				mtdblk->cache_state = STATE_EMPTY;
+				ret = mtd_read(mtd, sect_start, sect_size,
+					       &retlen, mtdblk->cache_data);
+				if (ret && !mtd_is_bitflip(ret))
+					return ret;
+				if (retlen != sect_size)
+					return -EIO;
+
+				mtdblk->cache_offset = sect_start;
+				mtdblk->cache_size = sect_size;
+				mtdblk->cache_state = STATE_CLEAN;
+			}
+
+			/* write data to our local cache */
+			memcpy (mtdblk->cache_data + offset, buf, size);
+			mtdblk->cache_state = STATE_DIRTY;
+		}
+
+		buf += size;
+		pos += size;
+		len -= size;
+	}
+
+	return 0;
+}
+
+
+static int do_cached_read (struct mtdblk_dev *mtdblk, unsigned long pos,
+			   int len, char *buf)
+{
+	struct mtd_info *mtd = mtdblk->mbd.mtd;
+	unsigned int sect_size = mtdblk->cache_size;
+	size_t retlen;
+	int ret;
+
+	pr_debug("mtdblock: read on \"%s\" at 0x%lx, size 0x%x\n",
+			mtd->name, pos, len);
+
+	if (!sect_size) {
+		ret = mtd_read(mtd, pos, len, &retlen, buf);
+		if (ret && !mtd_is_bitflip(ret))
+			return ret;
+		return 0;
+	}
+
+	while (len > 0) {
+		unsigned long sect_start = (pos/sect_size)*sect_size;
+		unsigned int offset = pos - sect_start;
+		unsigned int size = sect_size - offset;
+		if (size > len)
+			size = len;
+
+		/*
+		 * Check if the requested data is already cached
+		 * Read the requested amount of data from our internal cache if it
+		 * contains what we want, otherwise we read the data directly
+		 * from flash.
+		 */
+		if (mtdblk->cache_state != STATE_EMPTY &&
+		    mtdblk->cache_offset == sect_start) {
+			memcpy (buf, mtdblk->cache_data + offset, size);
+		} else {
+			ret = mtd_read(mtd, pos, size, &retlen, buf);
+			if (ret && !mtd_is_bitflip(ret))
+				return ret;
+			if (retlen != size)
+				return -EIO;
+		}
+
+		buf += size;
+		pos += size;
+		len -= size;
+	}
+
+	return 0;
+}
+
+static int mtdblock_readsect(struct mtd_blktrans_dev *dev,
+			      unsigned long block, char *buf)
+{
+	struct mtdblk_dev *mtdblk = container_of(dev, struct mtdblk_dev, mbd);
+	return do_cached_read(mtdblk, block<<9, 512, buf);
+}
+
+static int mtdblock_writesect(struct mtd_blktrans_dev *dev,
+			      unsigned long block, char *buf)
+{
+	struct mtdblk_dev *mtdblk = container_of(dev, struct mtdblk_dev, mbd);
+	if (unlikely(!mtdblk->cache_data && mtdblk->cache_size)) {
+		mtdblk->cache_data = vmalloc(mtdblk->mbd.mtd->erasesize);
+		if (!mtdblk->cache_data)
+			return -EINTR;
+		/* -EINTR is not really correct, but it is the best match
+		 * documented in man 2 write for all cases.  We could also
+		 * return -EAGAIN sometimes, but why bother?
+		 */
+	}
+	return do_cached_write(mtdblk, block<<9, 512, buf);
+}
+
+static int mtdblock_open(struct mtd_blktrans_dev *mbd)
+{
+	struct mtdblk_dev *mtdblk = container_of(mbd, struct mtdblk_dev, mbd);
+
+	pr_debug("mtdblock_open\n");
+
+	if (mtdblk->count) {
+		mtdblk->count++;
+		return 0;
+	}
+
+	if (mtd_type_is_nand(mbd->mtd))
+		pr_warn_ratelimited("%s: MTD device '%s' is NAND, please consider using UBI block devices instead.\n",
+			mbd->tr->name, mbd->mtd->name);
+
+	/* OK, it's not open. Create cache info for it */
+	mtdblk->count = 1;
+	mutex_init(&mtdblk->cache_mutex);
+	mtdblk->cache_state = STATE_EMPTY;
+	if (!(mbd->mtd->flags & MTD_NO_ERASE) && mbd->mtd->erasesize) {
+		mtdblk->cache_size = mbd->mtd->erasesize;
+		mtdblk->cache_data = NULL;
+	}
+
+	pr_debug("ok\n");
+
+	return 0;
+}
+
+static void mtdblock_release(struct mtd_blktrans_dev *mbd)
+{
+	struct mtdblk_dev *mtdblk = container_of(mbd, struct mtdblk_dev, mbd);
+
+	pr_debug("mtdblock_release\n");
+
+	mutex_lock(&mtdblk->cache_mutex);
+	write_cached_data(mtdblk);
+	mutex_unlock(&mtdblk->cache_mutex);
+
+	if (!--mtdblk->count) {
+		/*
+		 * It was the last usage. Free the cache, but only sync if
+		 * opened for writing.
+		 */
+		if (mbd->writable)
+			mtd_sync(mbd->mtd);
+		vfree(mtdblk->cache_data);
+	}
+
+	pr_debug("ok\n");
+}
+
+static int mtdblock_flush(struct mtd_blktrans_dev *dev)
+{
+	struct mtdblk_dev *mtdblk = container_of(dev, struct mtdblk_dev, mbd);
+	int ret;
+
+	mutex_lock(&mtdblk->cache_mutex);
+	ret = write_cached_data(mtdblk);
+	mutex_unlock(&mtdblk->cache_mutex);
+	mtd_sync(dev->mtd);
+	return ret;
+}
+
+static void mtdblock_add_mtd(struct mtd_blktrans_ops *tr, struct mtd_info *mtd)
+{
+	struct mtdblk_dev *dev = kzalloc_obj(*dev);
+
+	if (!dev)
+		return;
+
+	dev->mbd.mtd = mtd;
+	dev->mbd.devnum = mtd->index;
+
+	dev->mbd.size = mtd->size >> 9;
+	dev->mbd.tr = tr;
+
+	if (!(mtd->flags & MTD_WRITEABLE))
+		dev->mbd.readonly = 1;
+
+	if (add_mtd_blktrans_dev(&dev->mbd))
+		kfree(dev);
+}
+
+static void mtdblock_remove_dev(struct mtd_blktrans_dev *dev)
+{
+	del_mtd_blktrans_dev(dev);
+}
+
+static struct mtd_blktrans_ops mtdblock_tr = {
+	.name		= "mtdblock",
+	.major		= MTD_BLOCK_MAJOR,
+	.part_bits	= 0,
+	.blksize 	= 512,
+	.open		= mtdblock_open,
+	.flush		= mtdblock_flush,
+	.release	= mtdblock_release,
+	.readsect	= mtdblock_readsect,
+	.writesect	= mtdblock_writesect,
+	.add_mtd	= mtdblock_add_mtd,
+	.remove_dev	= mtdblock_remove_dev,
+	.owner		= THIS_MODULE,
+};
+
+module_mtd_blktrans(mtdblock_tr);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Nicolas Pitre <nico@fluxnic.net> et al.");
+MODULE_DESCRIPTION("Caching read/erase/writeback block device emulation access to MTD devices");