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, 579 insertions, 0 deletions

diff --git a/drivers/mtd/mtdpstore.c b/drivers/mtd/mtdpstore.c
new file mode 100644
index 000000000..9cf3872e3
--- /dev/null
+++ b/drivers/mtd/mtdpstore.c
@@ -0,0 +1,579 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#define dev_fmt(fmt) "mtdoops-pstore: " fmt
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/pstore_blk.h>
+#include <linux/mtd/mtd.h>
+#include <linux/bitops.h>
+#include <linux/slab.h>
+
+static struct mtdpstore_context {
+	int index;
+	struct pstore_blk_config info;
+	struct pstore_device_info dev;
+	struct mtd_info *mtd;
+	unsigned long *rmmap;		/* removed bit map */
+	unsigned long *usedmap;		/* used bit map */
+	/*
+	 * used for panic write
+	 * As there are no block_isbad for panic case, we should keep this
+	 * status before panic to ensure panic_write not failed.
+	 */
+	unsigned long *badmap;		/* bad block bit map */
+} oops_cxt;
+
+static int mtdpstore_block_isbad(struct mtdpstore_context *cxt, loff_t off)
+{
+	int ret;
+	struct mtd_info *mtd = cxt->mtd;
+	u64 blknum;
+
+	off = ALIGN_DOWN(off, mtd->erasesize);
+	blknum = div_u64(off, mtd->erasesize);
+
+	if (test_bit(blknum, cxt->badmap))
+		return true;
+	ret = mtd_block_isbad(mtd, off);
+	if (ret < 0) {
+		dev_err(&mtd->dev, "mtd_block_isbad failed, aborting\n");
+		return ret;
+	} else if (ret > 0) {
+		set_bit(blknum, cxt->badmap);
+		return true;
+	}
+	return false;
+}
+
+static inline int mtdpstore_panic_block_isbad(struct mtdpstore_context *cxt,
+		loff_t off)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	u64 blknum;
+
+	off = ALIGN_DOWN(off, mtd->erasesize);
+	blknum = div_u64(off, mtd->erasesize);
+	return test_bit(blknum, cxt->badmap);
+}
+
+static inline void mtdpstore_mark_used(struct mtdpstore_context *cxt,
+		loff_t off)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	u64 zonenum = div_u64(off, cxt->info.kmsg_size);
+
+	dev_dbg(&mtd->dev, "mark zone %llu used\n", zonenum);
+	set_bit(zonenum, cxt->usedmap);
+}
+
+static inline void mtdpstore_mark_unused(struct mtdpstore_context *cxt,
+		loff_t off)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	u64 zonenum = div_u64(off, cxt->info.kmsg_size);
+
+	dev_dbg(&mtd->dev, "mark zone %llu unused\n", zonenum);
+	clear_bit(zonenum, cxt->usedmap);
+}
+
+static inline void mtdpstore_block_mark_unused(struct mtdpstore_context *cxt,
+		loff_t off)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	u32 zonecnt = mtd->erasesize / cxt->info.kmsg_size;
+	u64 zonenum;
+
+	off = ALIGN_DOWN(off, mtd->erasesize);
+	zonenum = div_u64(off, cxt->info.kmsg_size);
+	while (zonecnt > 0) {
+		dev_dbg(&mtd->dev, "mark zone %llu unused\n", zonenum);
+		clear_bit(zonenum, cxt->usedmap);
+		zonenum++;
+		zonecnt--;
+	}
+}
+
+static inline int mtdpstore_is_used(struct mtdpstore_context *cxt, loff_t off)
+{
+	u64 zonenum = div_u64(off, cxt->info.kmsg_size);
+	u64 blknum = div_u64(off, cxt->mtd->erasesize);
+
+	if (test_bit(blknum, cxt->badmap))
+		return true;
+	return test_bit(zonenum, cxt->usedmap);
+}
+
+static int mtdpstore_block_is_used(struct mtdpstore_context *cxt,
+		loff_t off)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	u32 zonecnt = mtd->erasesize / cxt->info.kmsg_size;
+	u64 zonenum;
+
+	off = ALIGN_DOWN(off, mtd->erasesize);
+	zonenum = div_u64(off, cxt->info.kmsg_size);
+	while (zonecnt > 0) {
+		if (test_bit(zonenum, cxt->usedmap))
+			return true;
+		zonenum++;
+		zonecnt--;
+	}
+	return false;
+}
+
+static int mtdpstore_is_empty(struct mtdpstore_context *cxt, char *buf,
+		size_t size)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	size_t sz;
+	int i;
+
+	sz = min_t(uint32_t, size, mtd->writesize / 4);
+	for (i = 0; i < sz; i++) {
+		if (buf[i] != (char)0xFF)
+			return false;
+	}
+	return true;
+}
+
+static void mtdpstore_mark_removed(struct mtdpstore_context *cxt, loff_t off)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	u64 zonenum = div_u64(off, cxt->info.kmsg_size);
+
+	dev_dbg(&mtd->dev, "mark zone %llu removed\n", zonenum);
+	set_bit(zonenum, cxt->rmmap);
+}
+
+static void mtdpstore_block_clear_removed(struct mtdpstore_context *cxt,
+		loff_t off)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	u32 zonecnt = mtd->erasesize / cxt->info.kmsg_size;
+	u64 zonenum;
+
+	off = ALIGN_DOWN(off, mtd->erasesize);
+	zonenum = div_u64(off, cxt->info.kmsg_size);
+	while (zonecnt > 0) {
+		clear_bit(zonenum, cxt->rmmap);
+		zonenum++;
+		zonecnt--;
+	}
+}
+
+static int mtdpstore_block_is_removed(struct mtdpstore_context *cxt,
+		loff_t off)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	u32 zonecnt = mtd->erasesize / cxt->info.kmsg_size;
+	u64 zonenum;
+
+	off = ALIGN_DOWN(off, mtd->erasesize);
+	zonenum = div_u64(off, cxt->info.kmsg_size);
+	while (zonecnt > 0) {
+		if (test_bit(zonenum, cxt->rmmap))
+			return true;
+		zonenum++;
+		zonecnt--;
+	}
+	return false;
+}
+
+static int mtdpstore_erase_do(struct mtdpstore_context *cxt, loff_t off)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	struct erase_info erase;
+	int ret;
+
+	off = ALIGN_DOWN(off, cxt->mtd->erasesize);
+	dev_dbg(&mtd->dev, "try to erase off 0x%llx\n", off);
+	erase.len = cxt->mtd->erasesize;
+	erase.addr = off;
+	ret = mtd_erase(cxt->mtd, &erase);
+	if (!ret)
+		mtdpstore_block_clear_removed(cxt, off);
+	else
+		dev_err(&mtd->dev, "erase of region [0x%llx, 0x%llx] on \"%s\" failed\n",
+		       (unsigned long long)erase.addr,
+		       (unsigned long long)erase.len, cxt->info.device);
+	return ret;
+}
+
+/*
+ * called while removing file
+ *
+ * Avoiding over erasing, do erase block only when the whole block is unused.
+ * If the block contains valid log, do erase lazily on flush_removed() when
+ * unregister.
+ */
+static ssize_t mtdpstore_erase(size_t size, loff_t off)
+{
+	struct mtdpstore_context *cxt = &oops_cxt;
+
+	if (mtdpstore_block_isbad(cxt, off))
+		return -EIO;
+
+	mtdpstore_mark_unused(cxt, off);
+
+	/* If the block still has valid data, mtdpstore do erase lazily */
+	if (likely(mtdpstore_block_is_used(cxt, off))) {
+		mtdpstore_mark_removed(cxt, off);
+		return 0;
+	}
+
+	/* all zones are unused, erase it */
+	return mtdpstore_erase_do(cxt, off);
+}
+
+/*
+ * What is security for mtdpstore?
+ * As there is no erase for panic case, we should ensure at least one zone
+ * is writable. Otherwise, panic write will fail.
+ * If zone is used, write operation will return -ENOMSG, which means that
+ * pstore/blk will try one by one until gets an empty zone. So, it is not
+ * needed to ensure the next zone is empty, but at least one.
+ */
+static int mtdpstore_security(struct mtdpstore_context *cxt, loff_t off)
+{
+	int ret = 0, i;
+	struct mtd_info *mtd = cxt->mtd;
+	u32 zonenum = (u32)div_u64(off, cxt->info.kmsg_size);
+	u32 zonecnt = (u32)div_u64(cxt->mtd->size, cxt->info.kmsg_size);
+	u32 blkcnt = (u32)div_u64(cxt->mtd->size, cxt->mtd->erasesize);
+	u32 erasesize = cxt->mtd->erasesize;
+
+	for (i = 0; i < zonecnt; i++) {
+		u32 num = (zonenum + i) % zonecnt;
+
+		/* found empty zone */
+		if (!test_bit(num, cxt->usedmap))
+			return 0;
+	}
+
+	/* If there is no any empty zone, we have no way but to do erase */
+	while (blkcnt--) {
+		div64_u64_rem(off + erasesize, cxt->mtd->size, (u64 *)&off);
+
+		if (mtdpstore_block_isbad(cxt, off))
+			continue;
+
+		ret = mtdpstore_erase_do(cxt, off);
+		if (!ret) {
+			mtdpstore_block_mark_unused(cxt, off);
+			break;
+		}
+	}
+
+	if (ret)
+		dev_err(&mtd->dev, "all blocks bad!\n");
+	dev_dbg(&mtd->dev, "end security\n");
+	return ret;
+}
+
+static ssize_t mtdpstore_write(const char *buf, size_t size, loff_t off)
+{
+	struct mtdpstore_context *cxt = &oops_cxt;
+	struct mtd_info *mtd = cxt->mtd;
+	size_t retlen;
+	int ret;
+
+	if (mtdpstore_block_isbad(cxt, off))
+		return -ENOMSG;
+
+	/* zone is used, please try next one */
+	if (mtdpstore_is_used(cxt, off))
+		return -ENOMSG;
+
+	dev_dbg(&mtd->dev, "try to write off 0x%llx size %zu\n", off, size);
+	ret = mtd_write(cxt->mtd, off, size, &retlen, (u_char *)buf);
+	if (ret < 0 || retlen != size) {
+		dev_err(&mtd->dev, "write failure at %lld (%zu of %zu written), err %d\n",
+				off, retlen, size, ret);
+		return -EIO;
+	}
+	mtdpstore_mark_used(cxt, off);
+
+	mtdpstore_security(cxt, off);
+	return retlen;
+}
+
+static inline bool mtdpstore_is_io_error(int ret)
+{
+	return ret < 0 && !mtd_is_bitflip(ret) && !mtd_is_eccerr(ret);
+}
+
+/*
+ * All zones will be read as pstore/blk will read zone one by one when do
+ * recover.
+ */
+static ssize_t mtdpstore_read(char *buf, size_t size, loff_t off)
+{
+	struct mtdpstore_context *cxt = &oops_cxt;
+	struct mtd_info *mtd = cxt->mtd;
+	size_t retlen, done;
+	int ret;
+
+	if (mtdpstore_block_isbad(cxt, off))
+		return -ENOMSG;
+
+	dev_dbg(&mtd->dev, "try to read off 0x%llx size %zu\n", off, size);
+	for (done = 0, retlen = 0; done < size; done += retlen) {
+		retlen = 0;
+
+		ret = mtd_read(cxt->mtd, off + done, size - done, &retlen,
+				(u_char *)buf + done);
+		if (mtdpstore_is_io_error(ret)) {
+			dev_err(&mtd->dev, "read failure at %lld (%zu of %zu read), err %d\n",
+					off + done, retlen, size - done, ret);
+			/* the zone may be broken, try next one */
+			return -ENOMSG;
+		}
+
+		/*
+		 * ECC error. The impact on log data is so small. Maybe we can
+		 * still read it and try to understand. So mtdpstore just hands
+		 * over what it gets and user can judge whether the data is
+		 * valid or not.
+		 */
+		if (mtd_is_eccerr(ret)) {
+			dev_err(&mtd->dev, "ecc error at %lld (%zu of %zu read), err %d\n",
+					off + done, retlen, size - done, ret);
+			/* driver may not set retlen when ecc error */
+			retlen = retlen == 0 ? size - done : retlen;
+		}
+	}
+
+	if (mtdpstore_is_empty(cxt, buf, size))
+		mtdpstore_mark_unused(cxt, off);
+	else
+		mtdpstore_mark_used(cxt, off);
+
+	mtdpstore_security(cxt, off);
+	return retlen;
+}
+
+static ssize_t mtdpstore_panic_write(const char *buf, size_t size, loff_t off)
+{
+	struct mtdpstore_context *cxt = &oops_cxt;
+	struct mtd_info *mtd = cxt->mtd;
+	size_t retlen;
+	int ret;
+
+	if (mtdpstore_panic_block_isbad(cxt, off))
+		return -ENOMSG;
+
+	/* zone is used, please try next one */
+	if (mtdpstore_is_used(cxt, off))
+		return -ENOMSG;
+
+	ret = mtd_panic_write(cxt->mtd, off, size, &retlen, (u_char *)buf);
+	if (ret < 0 || size != retlen) {
+		dev_err(&mtd->dev, "panic write failure at %lld (%zu of %zu read), err %d\n",
+				off, retlen, size, ret);
+		return -EIO;
+	}
+	mtdpstore_mark_used(cxt, off);
+
+	return retlen;
+}
+
+static void mtdpstore_notify_add(struct mtd_info *mtd)
+{
+	int ret;
+	struct mtdpstore_context *cxt = &oops_cxt;
+	struct pstore_blk_config *info = &cxt->info;
+	unsigned long longcnt;
+
+	if (!strcmp(mtd->name, info->device))
+		cxt->index = mtd->index;
+
+	if (mtd->index != cxt->index || cxt->index < 0)
+		return;
+
+	dev_dbg(&mtd->dev, "found matching MTD device %s\n", mtd->name);
+
+	if (mtd->size < info->kmsg_size * 2) {
+		dev_err(&mtd->dev, "MTD partition %d not big enough\n",
+				mtd->index);
+		return;
+	}
+	/*
+	 * kmsg_size must be aligned to 4096 Bytes, which is limited by
+	 * psblk. The default value of kmsg_size is 64KB. If kmsg_size
+	 * is larger than erasesize, some errors will occur since mtdpstore
+	 * is designed on it.
+	 */
+	if (mtd->erasesize < info->kmsg_size) {
+		dev_err(&mtd->dev, "eraseblock size of MTD partition %d too small\n",
+				mtd->index);
+		return;
+	}
+	if (unlikely(info->kmsg_size % mtd->writesize)) {
+		dev_err(&mtd->dev, "record size %lu KB must align to write size %d KB\n",
+				info->kmsg_size / 1024,
+				mtd->writesize / 1024);
+		return;
+	}
+
+	longcnt = BITS_TO_LONGS(div_u64(mtd->size, info->kmsg_size));
+	cxt->rmmap = devm_kcalloc(&mtd->dev, longcnt, sizeof(long), GFP_KERNEL);
+	cxt->usedmap = devm_kcalloc(&mtd->dev, longcnt, sizeof(long), GFP_KERNEL);
+
+	longcnt = BITS_TO_LONGS(div_u64(mtd->size, mtd->erasesize));
+	cxt->badmap = devm_kcalloc(&mtd->dev, longcnt, sizeof(long), GFP_KERNEL);
+
+	if (!cxt->rmmap || !cxt->usedmap || !cxt->badmap)
+		return;
+
+	/* just support dmesg right now */
+	cxt->dev.flags = PSTORE_FLAGS_DMESG;
+	cxt->dev.zone.read = mtdpstore_read;
+	cxt->dev.zone.write = mtdpstore_write;
+	cxt->dev.zone.erase = mtdpstore_erase;
+	cxt->dev.zone.panic_write = mtdpstore_panic_write;
+	cxt->dev.zone.total_size = mtd->size;
+
+	ret = register_pstore_device(&cxt->dev);
+	if (ret) {
+		dev_err(&mtd->dev, "mtd%d register to psblk failed\n",
+				mtd->index);
+		return;
+	}
+	cxt->mtd = mtd;
+	dev_info(&mtd->dev, "Attached to MTD device %d\n", mtd->index);
+}
+
+static int mtdpstore_flush_removed_do(struct mtdpstore_context *cxt,
+		loff_t off, size_t size)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	u_char *buf;
+	int ret;
+	size_t retlen;
+	struct erase_info erase;
+
+	buf = kmalloc(mtd->erasesize, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+
+	/* 1st. read to cache */
+	ret = mtd_read(mtd, off, mtd->erasesize, &retlen, buf);
+	if (mtdpstore_is_io_error(ret))
+		goto free;
+
+	/* 2nd. erase block */
+	erase.len = mtd->erasesize;
+	erase.addr = off;
+	ret = mtd_erase(mtd, &erase);
+	if (ret)
+		goto free;
+
+	/* 3rd. write back */
+	while (size) {
+		unsigned int zonesize = cxt->info.kmsg_size;
+
+		/* there is valid data on block, write back */
+		if (mtdpstore_is_used(cxt, off)) {
+			ret = mtd_write(mtd, off, zonesize, &retlen, buf);
+			if (ret)
+				dev_err(&mtd->dev, "write failure at %lld (%zu of %u written), err %d\n",
+						off, retlen, zonesize, ret);
+		}
+
+		off += zonesize;
+		size -= min_t(unsigned int, zonesize, size);
+	}
+
+free:
+	kfree(buf);
+	return ret;
+}
+
+/*
+ * What does mtdpstore_flush_removed() do?
+ * When user remove any log file on pstore filesystem, mtdpstore should do
+ * something to ensure log file removed. If the whole block is no longer used,
+ * it's nice to erase the block. However if the block still contains valid log,
+ * what mtdpstore can do is to erase and write the valid log back.
+ */
+static int mtdpstore_flush_removed(struct mtdpstore_context *cxt)
+{
+	struct mtd_info *mtd = cxt->mtd;
+	int ret;
+	loff_t off;
+	u32 blkcnt = (u32)div_u64(mtd->size, mtd->erasesize);
+
+	for (off = 0; blkcnt > 0; blkcnt--, off += mtd->erasesize) {
+		ret = mtdpstore_block_isbad(cxt, off);
+		if (ret)
+			continue;
+
+		ret = mtdpstore_block_is_removed(cxt, off);
+		if (!ret)
+			continue;
+
+		ret = mtdpstore_flush_removed_do(cxt, off, mtd->erasesize);
+		if (ret)
+			return ret;
+	}
+	return 0;
+}
+
+static void mtdpstore_notify_remove(struct mtd_info *mtd)
+{
+	struct mtdpstore_context *cxt = &oops_cxt;
+
+	if (mtd->index != cxt->index || cxt->index < 0)
+		return;
+
+	mtdpstore_flush_removed(cxt);
+
+	unregister_pstore_device(&cxt->dev);
+	cxt->mtd = NULL;
+	cxt->index = -1;
+}
+
+static struct mtd_notifier mtdpstore_notifier = {
+	.add	= mtdpstore_notify_add,
+	.remove	= mtdpstore_notify_remove,
+};
+
+static int __init mtdpstore_init(void)
+{
+	int ret;
+	struct mtdpstore_context *cxt = &oops_cxt;
+	struct pstore_blk_config *info = &cxt->info;
+
+	ret = pstore_blk_get_config(info);
+	if (unlikely(ret))
+		return ret;
+
+	if (strlen(info->device) == 0) {
+		pr_err("mtd device must be supplied (device name is empty)\n");
+		return -EINVAL;
+	}
+	if (!info->kmsg_size) {
+		pr_err("no backend enabled (kmsg_size is 0)\n");
+		return -EINVAL;
+	}
+
+	/* Setup the MTD device to use */
+	ret = kstrtoint((char *)info->device, 0, &cxt->index);
+	if (ret)
+		cxt->index = -1;
+
+	register_mtd_user(&mtdpstore_notifier);
+	return 0;
+}
+module_init(mtdpstore_init);
+
+static void __exit mtdpstore_exit(void)
+{
+	unregister_mtd_user(&mtdpstore_notifier);
+}
+module_exit(mtdpstore_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("WeiXiong Liao <liaoweixiong@allwinnertech.com>");
+MODULE_DESCRIPTION("MTD backend for pstore/blk");