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, 419 insertions, 0 deletions

diff --git a/drivers/mtd/tests/nandbiterrs.c b/drivers/mtd/tests/nandbiterrs.c
new file mode 100644
index 000000000..98d7508f9
--- /dev/null
+++ b/drivers/mtd/tests/nandbiterrs.c
@@ -0,0 +1,419 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright © 2012 NetCommWireless
+ * Iwo Mergler <Iwo.Mergler@netcommwireless.com.au>
+ *
+ * Test for multi-bit error recovery on a NAND page This mostly tests the
+ * ECC controller / driver.
+ *
+ * There are two test modes:
+ *
+ *	0 - artificially inserting bit errors until the ECC fails
+ *	    This is the default method and fairly quick. It should
+ *	    be independent of the quality of the FLASH.
+ *
+ *	1 - re-writing the same pattern repeatedly until the ECC fails.
+ *	    This method relies on the physics of NAND FLASH to eventually
+ *	    generate '0' bits if '1' has been written sufficient times.
+ *	    Depending on the NAND, the first bit errors will appear after
+ *	    1000 or more writes and then will usually snowball, reaching the
+ *	    limits of the ECC quickly.
+ *
+ *	    The test stops after 10000 cycles, should your FLASH be
+ *	    exceptionally good and not generate bit errors before that. Try
+ *	    a different page in that case.
+ *
+ * Please note that neither of these tests will significantly 'use up' any
+ * FLASH endurance. Only a maximum of two erase operations will be performed.
+ */
+
+#define pr_fmt(fmt)	KBUILD_MODNAME ": " fmt
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/mtd/mtd.h>
+#include <linux/err.h>
+#include <linux/mtd/rawnand.h>
+#include <linux/slab.h>
+#include "mtd_test.h"
+
+static int dev;
+module_param(dev, int, S_IRUGO);
+MODULE_PARM_DESC(dev, "MTD device number to use");
+
+static unsigned page_offset;
+module_param(page_offset, uint, S_IRUGO);
+MODULE_PARM_DESC(page_offset, "Page number relative to dev start");
+
+static unsigned seed;
+module_param(seed, uint, S_IRUGO);
+MODULE_PARM_DESC(seed, "Random seed");
+
+static int mode;
+module_param(mode, int, S_IRUGO);
+MODULE_PARM_DESC(mode, "0=incremental errors, 1=overwrite test");
+
+static unsigned max_overwrite = 10000;
+
+static loff_t   offset;     /* Offset of the page we're using. */
+static unsigned eraseblock; /* Eraseblock number for our page. */
+
+/* We assume that the ECC can correct up to a certain number
+ * of biterrors per subpage. */
+static unsigned subsize;  /* Size of subpages */
+static unsigned subcount; /* Number of subpages per page */
+
+static struct mtd_info *mtd;   /* MTD device */
+
+static uint8_t *wbuffer; /* One page write / compare buffer */
+static uint8_t *rbuffer; /* One page read buffer */
+
+/* 'random' bytes from known offsets */
+static uint8_t hash(unsigned offset)
+{
+	unsigned v = offset;
+	unsigned char c;
+	v ^= 0x7f7edfd3;
+	v = v ^ (v >> 3);
+	v = v ^ (v >> 5);
+	v = v ^ (v >> 13);
+	c = v & 0xFF;
+	/* Reverse bits of result. */
+	c = (c & 0x0F) << 4 | (c & 0xF0) >> 4;
+	c = (c & 0x33) << 2 | (c & 0xCC) >> 2;
+	c = (c & 0x55) << 1 | (c & 0xAA) >> 1;
+	return c;
+}
+
+/* Writes wbuffer to page */
+static int write_page(int log)
+{
+	if (log)
+		pr_info("write_page\n");
+
+	return mtdtest_write(mtd, offset, mtd->writesize, wbuffer);
+}
+
+/* Re-writes the data area while leaving the OOB alone. */
+static int rewrite_page(int log)
+{
+	int err = 0;
+	struct mtd_oob_ops ops = { };
+
+	if (log)
+		pr_info("rewrite page\n");
+
+	ops.mode      = MTD_OPS_RAW; /* No ECC */
+	ops.len       = mtd->writesize;
+	ops.retlen    = 0;
+	ops.ooblen    = 0;
+	ops.oobretlen = 0;
+	ops.ooboffs   = 0;
+	ops.datbuf    = wbuffer;
+	ops.oobbuf    = NULL;
+
+	err = mtd_write_oob(mtd, offset, &ops);
+	if (err || ops.retlen != mtd->writesize) {
+		pr_err("error: write_oob failed (%d)\n", err);
+		if (!err)
+			err = -EIO;
+	}
+
+	return err;
+}
+
+/* Reads page into rbuffer. Returns number of corrected bit errors (>=0)
+ * or error (<0) */
+static int read_page(int log)
+{
+	int err = 0;
+	size_t read;
+	struct mtd_ecc_stats oldstats;
+
+	if (log)
+		pr_info("read_page\n");
+
+	/* Saving last mtd stats */
+	memcpy(&oldstats, &mtd->ecc_stats, sizeof(oldstats));
+
+	err = mtd_read(mtd, offset, mtd->writesize, &read, rbuffer);
+	if (!err || err == -EUCLEAN)
+		err = mtd->ecc_stats.corrected - oldstats.corrected;
+
+	if (err < 0 || read != mtd->writesize) {
+		pr_err("error: read failed at %#llx\n", (long long)offset);
+		if (err >= 0)
+			err = -EIO;
+	}
+
+	return err;
+}
+
+/* Verifies rbuffer against random sequence */
+static int verify_page(int log)
+{
+	unsigned i, errs = 0;
+
+	if (log)
+		pr_info("verify_page\n");
+
+	for (i = 0; i < mtd->writesize; i++) {
+		if (rbuffer[i] != hash(i+seed)) {
+			pr_err("Error: page offset %u, expected %02x, got %02x\n",
+				i, hash(i+seed), rbuffer[i]);
+			errs++;
+		}
+	}
+
+	if (errs)
+		return -EIO;
+	else
+		return 0;
+}
+
+#define CBIT(v, n) ((v) & (1 << (n)))
+#define BCLR(v, n) ((v) = (v) & ~(1 << (n)))
+
+/* Finds the first '1' bit in wbuffer starting at offset 'byte'
+ * and sets it to '0'. */
+static int insert_biterror(unsigned byte)
+{
+	int bit;
+
+	while (byte < mtd->writesize) {
+		for (bit = 7; bit >= 0; bit--) {
+			if (CBIT(wbuffer[byte], bit)) {
+				BCLR(wbuffer[byte], bit);
+				pr_info("Inserted biterror @ %u/%u\n", byte, bit);
+				return 0;
+			}
+		}
+		byte++;
+	}
+	pr_err("biterror: Failed to find a '1' bit\n");
+	return -EIO;
+}
+
+/* Writes 'random' data to page and then introduces deliberate bit
+ * errors into the page, while verifying each step. */
+static int incremental_errors_test(void)
+{
+	int err = 0;
+	unsigned i;
+	unsigned errs_per_subpage = 0;
+
+	pr_info("incremental biterrors test\n");
+
+	for (i = 0; i < mtd->writesize; i++)
+		wbuffer[i] = hash(i+seed);
+
+	err = write_page(1);
+	if (err)
+		goto exit;
+
+	while (1) {
+
+		err = rewrite_page(1);
+		if (err)
+			goto exit;
+
+		err = read_page(1);
+		if (err > 0)
+			pr_info("Read reported %d corrected bit errors\n", err);
+		if (err < 0) {
+			pr_err("After %d biterrors per subpage, read reported error %d\n",
+				errs_per_subpage, err);
+			err = 0;
+			goto exit;
+		}
+
+		err = verify_page(1);
+		if (err) {
+			pr_err("ECC failure, read data is incorrect despite read success\n");
+			goto exit;
+		}
+
+		pr_info("Successfully corrected %d bit errors per subpage\n",
+			errs_per_subpage);
+
+		for (i = 0; i < subcount; i++) {
+			err = insert_biterror(i * subsize);
+			if (err < 0)
+				goto exit;
+		}
+		errs_per_subpage++;
+	}
+
+exit:
+	return err;
+}
+
+
+/* Writes 'random' data to page and then re-writes that same data repeatedly.
+   This eventually develops bit errors (bits written as '1' will slowly become
+   '0'), which are corrected as far as the ECC is capable of. */
+static int overwrite_test(void)
+{
+	int err = 0;
+	unsigned i;
+	unsigned max_corrected = 0;
+	unsigned opno = 0;
+	/* We don't expect more than this many correctable bit errors per
+	 * page. */
+	#define MAXBITS 512
+	static unsigned bitstats[MAXBITS]; /* bit error histogram. */
+
+	memset(bitstats, 0, sizeof(bitstats));
+
+	pr_info("overwrite biterrors test\n");
+
+	for (i = 0; i < mtd->writesize; i++)
+		wbuffer[i] = hash(i+seed);
+
+	err = write_page(1);
+	if (err)
+		goto exit;
+
+	while (opno < max_overwrite) {
+
+		err = write_page(0);
+		if (err)
+			break;
+
+		err = read_page(0);
+		if (err >= 0) {
+			if (err >= MAXBITS) {
+				pr_info("Implausible number of bit errors corrected\n");
+				err = -EIO;
+				break;
+			}
+			bitstats[err]++;
+			if (err > max_corrected) {
+				max_corrected = err;
+				pr_info("Read reported %d corrected bit errors\n",
+					err);
+			}
+		} else { /* err < 0 */
+			pr_info("Read reported error %d\n", err);
+			err = 0;
+			break;
+		}
+
+		err = verify_page(0);
+		if (err) {
+			bitstats[max_corrected] = opno;
+			pr_info("ECC failure, read data is incorrect despite read success\n");
+			break;
+		}
+
+		err = mtdtest_relax();
+		if (err)
+			break;
+
+		opno++;
+	}
+
+	/* At this point bitstats[0] contains the number of ops with no bit
+	 * errors, bitstats[1] the number of ops with 1 bit error, etc. */
+	pr_info("Bit error histogram (%d operations total):\n", opno);
+	for (i = 0; i < max_corrected; i++)
+		pr_info("Page reads with %3d corrected bit errors: %d\n",
+			i, bitstats[i]);
+
+exit:
+	return err;
+}
+
+static int __init mtd_nandbiterrs_init(void)
+{
+	int err = 0;
+
+	printk("\n");
+	printk(KERN_INFO "==================================================\n");
+	pr_info("MTD device: %d\n", dev);
+
+	mtd = get_mtd_device(NULL, dev);
+	if (IS_ERR(mtd)) {
+		err = PTR_ERR(mtd);
+		pr_err("error: cannot get MTD device\n");
+		goto exit_mtddev;
+	}
+
+	if (!mtd_type_is_nand(mtd)) {
+		pr_info("this test requires NAND flash\n");
+		err = -ENODEV;
+		goto exit_nand;
+	}
+
+	pr_info("MTD device size %llu, eraseblock=%u, page=%u, oob=%u\n",
+		(unsigned long long)mtd->size, mtd->erasesize,
+		mtd->writesize, mtd->oobsize);
+
+	subsize  = mtd->writesize >> mtd->subpage_sft;
+	subcount = mtd->writesize / subsize;
+
+	pr_info("Device uses %d subpages of %d bytes\n", subcount, subsize);
+
+	offset     = (loff_t)page_offset * mtd->writesize;
+	eraseblock = mtd_div_by_eb(offset, mtd);
+
+	pr_info("Using page=%u, offset=%llu, eraseblock=%u\n",
+		page_offset, offset, eraseblock);
+
+	wbuffer = kmalloc(mtd->writesize, GFP_KERNEL);
+	if (!wbuffer) {
+		err = -ENOMEM;
+		goto exit_wbuffer;
+	}
+
+	rbuffer = kmalloc(mtd->writesize, GFP_KERNEL);
+	if (!rbuffer) {
+		err = -ENOMEM;
+		goto exit_rbuffer;
+	}
+
+	err = mtdtest_erase_eraseblock(mtd, eraseblock);
+	if (err)
+		goto exit_error;
+
+	if (mode == 0)
+		err = incremental_errors_test();
+	else
+		err = overwrite_test();
+
+	if (err)
+		goto exit_error;
+
+	/* We leave the block un-erased in case of test failure. */
+	err = mtdtest_erase_eraseblock(mtd, eraseblock);
+	if (err)
+		goto exit_error;
+
+	err = -EIO;
+	pr_info("finished successfully.\n");
+	printk(KERN_INFO "==================================================\n");
+
+exit_error:
+	kfree(rbuffer);
+exit_rbuffer:
+	kfree(wbuffer);
+exit_wbuffer:
+	/* Nothing */
+exit_nand:
+	put_mtd_device(mtd);
+exit_mtddev:
+	return err;
+}
+
+static void __exit mtd_nandbiterrs_exit(void)
+{
+	return;
+}
+
+module_init(mtd_nandbiterrs_init);
+module_exit(mtd_nandbiterrs_exit);
+
+MODULE_DESCRIPTION("NAND bit error recovery test");
+MODULE_AUTHOR("Iwo Mergler");
+MODULE_LICENSE("GPL");