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, 534 insertions, 0 deletions

diff --git a/drivers/md/dm-io.c b/drivers/md/dm-io.c
new file mode 100644
index 000000000..777b917be
--- /dev/null
+++ b/drivers/md/dm-io.c
@@ -0,0 +1,534 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2003 Sistina Software
+ * Copyright (C) 2006 Red Hat GmbH
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm-core.h"
+
+#include <linux/device-mapper.h>
+
+#include <linux/bio.h>
+#include <linux/completion.h>
+#include <linux/mempool.h>
+#include <linux/module.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/dm-io.h>
+
+#define DM_MSG_PREFIX "io"
+
+#define DM_IO_MAX_REGIONS	BITS_PER_LONG
+
+struct dm_io_client {
+	mempool_t pool;
+	struct bio_set bios;
+};
+
+/*
+ * Aligning 'struct io' reduces the number of bits required to store
+ * its address.  Refer to store_io_and_region_in_bio() below.
+ */
+struct io {
+	unsigned long error_bits;
+	unsigned long unsup_bits;
+	atomic_t count;
+	struct dm_io_client *client;
+	io_notify_fn callback;
+	void *context;
+	void *vma_invalidate_address;
+	unsigned long vma_invalidate_size;
+} __aligned(DM_IO_MAX_REGIONS);
+
+static struct kmem_cache *_dm_io_cache;
+
+/*
+ * Create a client with mempool and bioset.
+ */
+struct dm_io_client *dm_io_client_create(void)
+{
+	struct dm_io_client *client;
+	unsigned int min_ios = dm_get_reserved_bio_based_ios();
+	int ret;
+
+	client = kzalloc_obj(*client);
+	if (!client)
+		return ERR_PTR(-ENOMEM);
+
+	ret = mempool_init_slab_pool(&client->pool, min_ios, _dm_io_cache);
+	if (ret)
+		goto bad;
+
+	ret = bioset_init(&client->bios, min_ios, 0, BIOSET_NEED_BVECS);
+	if (ret)
+		goto bad;
+
+	return client;
+
+bad:
+	mempool_exit(&client->pool);
+	kfree(client);
+	return ERR_PTR(ret);
+}
+EXPORT_SYMBOL(dm_io_client_create);
+
+void dm_io_client_destroy(struct dm_io_client *client)
+{
+	mempool_exit(&client->pool);
+	bioset_exit(&client->bios);
+	kfree(client);
+}
+EXPORT_SYMBOL(dm_io_client_destroy);
+
+/*
+ *-------------------------------------------------------------------
+ * We need to keep track of which region a bio is doing io for.
+ * To avoid a memory allocation to store just 5 or 6 bits, we
+ * ensure the 'struct io' pointer is aligned so enough low bits are
+ * always zero and then combine it with the region number directly in
+ * bi_private.
+ *-------------------------------------------------------------------
+ */
+static void store_io_and_region_in_bio(struct bio *bio, struct io *io,
+				       unsigned int region)
+{
+	if (unlikely(!IS_ALIGNED((unsigned long)io, DM_IO_MAX_REGIONS))) {
+		DMCRIT("Unaligned struct io pointer %p", io);
+		BUG();
+	}
+
+	bio->bi_private = (void *)((unsigned long)io | region);
+}
+
+static void retrieve_io_and_region_from_bio(struct bio *bio, struct io **io,
+				       unsigned int *region)
+{
+	unsigned long val = (unsigned long)bio->bi_private;
+
+	*io = (void *)(val & -(unsigned long)DM_IO_MAX_REGIONS);
+	*region = val & (DM_IO_MAX_REGIONS - 1);
+}
+
+/*
+ *--------------------------------------------------------------
+ * We need an io object to keep track of the number of bios that
+ * have been dispatched for a particular io.
+ *--------------------------------------------------------------
+ */
+static void complete_io(struct io *io)
+{
+	unsigned long error_bits = io->error_bits;
+	unsigned long unsup_bits = io->unsup_bits;
+	io_notify_fn fn = io->callback;
+	void *context = io->context;
+
+	if (io->vma_invalidate_size)
+		invalidate_kernel_vmap_range(io->vma_invalidate_address,
+					     io->vma_invalidate_size);
+
+	mempool_free(io, &io->client->pool);
+	fn(error_bits, unsup_bits, context);
+}
+
+static void dec_count(struct io *io, unsigned int region, blk_status_t error)
+{
+	if (unlikely(error)) {
+		if (error == BLK_STS_NOTSUPP || error == BLK_STS_INVAL)
+			set_bit(region, &io->unsup_bits);
+		else
+			set_bit(region, &io->error_bits);
+	}
+
+	if (atomic_dec_and_test(&io->count))
+		complete_io(io);
+}
+
+static void endio(struct bio *bio)
+{
+	struct io *io;
+	unsigned int region;
+	blk_status_t error;
+
+	if (bio->bi_status && bio_data_dir(bio) == READ)
+		zero_fill_bio(bio);
+
+	/*
+	 * The bio destructor in bio_put() may use the io object.
+	 */
+	retrieve_io_and_region_from_bio(bio, &io, &region);
+
+	error = bio->bi_status;
+	bio_put(bio);
+
+	dec_count(io, region, error);
+}
+
+/*
+ *--------------------------------------------------------------
+ * These little objects provide an abstraction for getting a new
+ * destination page for io.
+ *--------------------------------------------------------------
+ */
+struct dpages {
+	void (*get_page)(struct dpages *dp,
+			 struct page **p, unsigned long *len, unsigned int *offset);
+	void (*next_page)(struct dpages *dp);
+
+	unsigned int context_u;
+	void *context_ptr;
+
+	struct bio *orig_bio;
+
+	void *vma_invalidate_address;
+	unsigned long vma_invalidate_size;
+};
+
+/*
+ * Functions for getting the pages from a list.
+ */
+static void list_get_page(struct dpages *dp,
+		  struct page **p, unsigned long *len, unsigned int *offset)
+{
+	unsigned int o = dp->context_u;
+	struct page_list *pl = dp->context_ptr;
+
+	*p = pl->page;
+	*len = PAGE_SIZE - o;
+	*offset = o;
+}
+
+static void list_next_page(struct dpages *dp)
+{
+	struct page_list *pl = dp->context_ptr;
+
+	dp->context_ptr = pl->next;
+	dp->context_u = 0;
+}
+
+static void list_dp_init(struct dpages *dp, struct page_list *pl, unsigned int offset)
+{
+	dp->get_page = list_get_page;
+	dp->next_page = list_next_page;
+	dp->context_u = offset;
+	dp->context_ptr = pl;
+}
+
+/*
+ * Functions for getting the pages from a VMA.
+ */
+static void vm_get_page(struct dpages *dp,
+		 struct page **p, unsigned long *len, unsigned int *offset)
+{
+	*p = vmalloc_to_page(dp->context_ptr);
+	*offset = dp->context_u;
+	*len = PAGE_SIZE - dp->context_u;
+}
+
+static void vm_next_page(struct dpages *dp)
+{
+	dp->context_ptr += PAGE_SIZE - dp->context_u;
+	dp->context_u = 0;
+}
+
+static void vm_dp_init(struct dpages *dp, void *data)
+{
+	dp->get_page = vm_get_page;
+	dp->next_page = vm_next_page;
+	dp->context_u = offset_in_page(data);
+	dp->context_ptr = data;
+}
+
+/*
+ * Functions for getting the pages from kernel memory.
+ */
+static void km_get_page(struct dpages *dp, struct page **p, unsigned long *len,
+			unsigned int *offset)
+{
+	*p = virt_to_page(dp->context_ptr);
+	*offset = dp->context_u;
+	*len = PAGE_SIZE - dp->context_u;
+}
+
+static void km_next_page(struct dpages *dp)
+{
+	dp->context_ptr += PAGE_SIZE - dp->context_u;
+	dp->context_u = 0;
+}
+
+static void km_dp_init(struct dpages *dp, void *data)
+{
+	dp->get_page = km_get_page;
+	dp->next_page = km_next_page;
+	dp->context_u = offset_in_page(data);
+	dp->context_ptr = data;
+}
+
+/*
+ *---------------------------------------------------------------
+ * IO routines that accept a list of pages.
+ *---------------------------------------------------------------
+ */
+static void do_region(const blk_opf_t opf, unsigned int region,
+		      struct dm_io_region *where, struct dpages *dp,
+		      struct io *io, unsigned short ioprio)
+{
+	struct bio *bio;
+	struct page *page;
+	unsigned long len;
+	unsigned int offset;
+	unsigned int num_bvecs;
+	sector_t remaining = where->count;
+	struct request_queue *q = bdev_get_queue(where->bdev);
+	sector_t num_sectors;
+	unsigned int special_cmd_max_sectors;
+	const enum req_op op = opf & REQ_OP_MASK;
+
+	/*
+	 * Reject unsupported discard and write same requests.
+	 */
+	if (op == REQ_OP_DISCARD)
+		special_cmd_max_sectors = bdev_max_discard_sectors(where->bdev);
+	else if (op == REQ_OP_WRITE_ZEROES)
+		special_cmd_max_sectors = q->limits.max_write_zeroes_sectors;
+	if ((op == REQ_OP_DISCARD || op == REQ_OP_WRITE_ZEROES) &&
+	    special_cmd_max_sectors == 0) {
+		atomic_inc(&io->count);
+		dec_count(io, region, BLK_STS_NOTSUPP);
+		return;
+	}
+
+	if (dp->orig_bio) {
+		bio = bio_alloc_clone(where->bdev, dp->orig_bio, GFP_NOIO,
+				      &io->client->bios);
+		bio->bi_iter.bi_sector = where->sector;
+		bio->bi_iter.bi_size = where->count << SECTOR_SHIFT;
+		bio->bi_opf = opf;
+		bio->bi_end_io = endio;
+		bio->bi_ioprio = ioprio;
+		store_io_and_region_in_bio(bio, io, region);
+
+		atomic_inc(&io->count);
+		submit_bio(bio);
+		return;
+	}
+
+	/*
+	 * where->count may be zero if op holds a flush and we need to
+	 * send a zero-sized flush.
+	 */
+	do {
+		/*
+		 * Allocate a suitably sized-bio.
+		 */
+		switch (op) {
+		case REQ_OP_DISCARD:
+		case REQ_OP_WRITE_ZEROES:
+			num_bvecs = 0;
+			break;
+		default:
+			num_bvecs = bio_max_segs(dm_sector_div_up(remaining,
+						(PAGE_SIZE >> SECTOR_SHIFT)) + 1);
+		}
+
+		bio = bio_alloc_bioset(where->bdev, num_bvecs, opf, GFP_NOIO,
+				       &io->client->bios);
+		bio->bi_iter.bi_sector = where->sector + (where->count - remaining);
+		bio->bi_end_io = endio;
+		bio->bi_ioprio = ioprio;
+		store_io_and_region_in_bio(bio, io, region);
+
+		if (op == REQ_OP_DISCARD || op == REQ_OP_WRITE_ZEROES) {
+			num_sectors = min_t(sector_t, special_cmd_max_sectors, remaining);
+			bio->bi_iter.bi_size = num_sectors << SECTOR_SHIFT;
+			remaining -= num_sectors;
+		} else {
+			while (remaining) {
+				/*
+				 * Try and add as many pages as possible.
+				 */
+				dp->get_page(dp, &page, &len, &offset);
+				len = min(len, to_bytes(remaining));
+				if (!bio_add_page(bio, page, len, offset))
+					break;
+
+				offset = 0;
+				remaining -= to_sector(len);
+				dp->next_page(dp);
+			}
+		}
+
+		atomic_inc(&io->count);
+		submit_bio(bio);
+		WARN_ON_ONCE(opf & REQ_ATOMIC && remaining);
+	} while (remaining);
+}
+
+static void dispatch_io(blk_opf_t opf, unsigned int num_regions,
+			struct dm_io_region *where, struct dpages *dp,
+			struct io *io, unsigned short ioprio)
+{
+	int i;
+	struct dpages old_pages = *dp;
+
+	BUG_ON(num_regions > DM_IO_MAX_REGIONS);
+
+	/*
+	 * For multiple regions we need to be careful to rewind
+	 * the dp object for each call to do_region.
+	 */
+	for (i = 0; i < num_regions; i++) {
+		*dp = old_pages;
+		if (where[i].count || (opf & REQ_PREFLUSH))
+			do_region(opf, i, where + i, dp, io, ioprio);
+	}
+
+	/*
+	 * Drop the extra reference that we were holding to avoid
+	 * the io being completed too early.
+	 */
+	dec_count(io, 0, 0);
+}
+
+static void async_io(struct dm_io_client *client, unsigned int num_regions,
+		     struct dm_io_region *where, blk_opf_t opf,
+		     struct dpages *dp, io_notify_fn fn, void *context,
+		     unsigned short ioprio)
+{
+	struct io *io;
+
+	io = mempool_alloc(&client->pool, GFP_NOIO);
+	io->error_bits = 0;
+	io->unsup_bits = 0;
+	atomic_set(&io->count, 1); /* see dispatch_io() */
+	io->client = client;
+	io->callback = fn;
+	io->context = context;
+
+	io->vma_invalidate_address = dp->vma_invalidate_address;
+	io->vma_invalidate_size = dp->vma_invalidate_size;
+
+	dispatch_io(opf, num_regions, where, dp, io, ioprio);
+}
+
+struct sync_io {
+	unsigned long error_bits;
+	unsigned long unsup_bits;
+	struct completion wait;
+};
+
+static void sync_io_complete(unsigned long error, unsigned long unsup, void *context)
+{
+	struct sync_io *sio = context;
+
+	sio->error_bits = error;
+	sio->unsup_bits = unsup;
+	complete(&sio->wait);
+}
+
+static int sync_io(struct dm_io_client *client, unsigned int num_regions,
+		   struct dm_io_region *where, blk_opf_t opf, struct dpages *dp,
+		   unsigned long *error_bits, unsigned long *unsup_bits,
+		   unsigned short ioprio)
+{
+	struct sync_io sio;
+
+	init_completion(&sio.wait);
+
+	async_io(client, num_regions, where, opf | REQ_SYNC, dp,
+		 sync_io_complete, &sio, ioprio);
+
+	wait_for_completion_io(&sio.wait);
+
+	if (error_bits)
+		*error_bits = sio.error_bits;
+	if (unsup_bits)
+		*unsup_bits = sio.unsup_bits;
+
+	return sio.error_bits ? -EIO : sio.unsup_bits ? -EOPNOTSUPP : 0;
+}
+
+static int dp_init(struct dm_io_request *io_req, struct dpages *dp,
+		   unsigned long size)
+{
+	/* Set up dpages based on memory type */
+
+	dp->vma_invalidate_address = NULL;
+	dp->vma_invalidate_size = 0;
+	dp->orig_bio = NULL;
+
+	switch (io_req->mem.type) {
+	case DM_IO_PAGE_LIST:
+		list_dp_init(dp, io_req->mem.ptr.pl, io_req->mem.offset);
+		break;
+
+	case DM_IO_BIO:
+		/*
+		 * The destination bios clone this bio's biovec directly, so
+		 * there are no per-page accessors to set up here.
+		 */
+		dp->orig_bio = io_req->mem.ptr.bio;
+		break;
+
+	case DM_IO_VMA:
+		flush_kernel_vmap_range(io_req->mem.ptr.vma, size);
+		if ((io_req->bi_opf & REQ_OP_MASK) == REQ_OP_READ) {
+			dp->vma_invalidate_address = io_req->mem.ptr.vma;
+			dp->vma_invalidate_size = size;
+		}
+		vm_dp_init(dp, io_req->mem.ptr.vma);
+		break;
+
+	case DM_IO_KMEM:
+		km_dp_init(dp, io_req->mem.ptr.addr);
+		break;
+
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+int dm_io(struct dm_io_request *io_req, unsigned int num_regions,
+	  struct dm_io_region *where, unsigned long *sync_error_bits,
+	  unsigned long *sync_unsup_bits, unsigned short ioprio)
+{
+	int r;
+	struct dpages dp;
+
+	if (num_regions > 1 && !op_is_write(io_req->bi_opf)) {
+		WARN_ON(1);
+		return -EIO;
+	}
+
+	r = dp_init(io_req, &dp, (unsigned long)where->count << SECTOR_SHIFT);
+	if (r)
+		return r;
+
+	if (!io_req->notify.fn)
+		return sync_io(io_req->client, num_regions, where,
+			       io_req->bi_opf, &dp, sync_error_bits,
+			       sync_unsup_bits, ioprio);
+
+	async_io(io_req->client, num_regions, where, io_req->bi_opf, &dp,
+		 io_req->notify.fn, io_req->notify.context, ioprio);
+	return 0;
+}
+EXPORT_SYMBOL(dm_io);
+
+int __init dm_io_init(void)
+{
+	_dm_io_cache = KMEM_CACHE(io, 0);
+	if (!_dm_io_cache)
+		return -ENOMEM;
+
+	return 0;
+}
+
+void dm_io_exit(void)
+{
+	kmem_cache_destroy(_dm_io_cache);
+	_dm_io_cache = NULL;
+}