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, 381 insertions, 0 deletions

diff --git a/fs/fat/cache.c b/fs/fat/cache.c
new file mode 100644
index 000000000..1b87354e2
--- /dev/null
+++ b/fs/fat/cache.c
@@ -0,0 +1,381 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *  linux/fs/fat/cache.c
+ *
+ *  Written 1992,1993 by Werner Almesberger
+ *
+ *  Mar 1999. AV. Changed cache, so that it uses the starting cluster instead
+ *	of inode number.
+ *  May 1999. AV. Fixed the bogosity with FAT32 (read "FAT28"). Fscking lusers.
+ */
+
+#include <linux/slab.h>
+#include "fat.h"
+
+/* this must be > 0. */
+#define FAT_MAX_CACHE	8
+
+struct fat_cache {
+	struct list_head cache_list;
+	int nr_contig;	/* number of contiguous clusters */
+	int fcluster;	/* cluster number in the file. */
+	int dcluster;	/* cluster number on disk. */
+};
+
+struct fat_cache_id {
+	unsigned int id;
+	int nr_contig;
+	int fcluster;
+	int dcluster;
+};
+
+static struct kmem_cache *fat_cache_cachep;
+
+static void init_once(void *foo)
+{
+	struct fat_cache *cache = (struct fat_cache *)foo;
+
+	INIT_LIST_HEAD(&cache->cache_list);
+}
+
+int __init fat_cache_init(void)
+{
+	fat_cache_cachep = kmem_cache_create("fat_cache",
+				sizeof(struct fat_cache),
+				0, SLAB_RECLAIM_ACCOUNT,
+				init_once);
+	if (fat_cache_cachep == NULL)
+		return -ENOMEM;
+	return 0;
+}
+
+void fat_cache_destroy(void)
+{
+	kmem_cache_destroy(fat_cache_cachep);
+}
+
+static inline struct fat_cache *fat_cache_alloc(void)
+{
+	return kmem_cache_alloc(fat_cache_cachep, GFP_NOFS);
+}
+
+static inline void fat_cache_free(struct fat_cache *cache)
+{
+	BUG_ON(!list_empty(&cache->cache_list));
+	kmem_cache_free(fat_cache_cachep, cache);
+}
+
+static inline void fat_cache_update_lru(struct inode *inode,
+					struct fat_cache *cache)
+{
+	if (MSDOS_I(inode)->cache_lru.next != &cache->cache_list)
+		list_move(&cache->cache_list, &MSDOS_I(inode)->cache_lru);
+}
+
+static int fat_cache_lookup(struct inode *inode, int fclus,
+			    struct fat_cache_id *cid,
+			    int *cached_fclus, int *cached_dclus)
+{
+	static struct fat_cache nohit = { .fcluster = 0, };
+
+	struct fat_cache *hit = &nohit, *p;
+	int offset = -1;
+
+	spin_lock(&MSDOS_I(inode)->cache_lru_lock);
+	list_for_each_entry(p, &MSDOS_I(inode)->cache_lru, cache_list) {
+		/* Find the cache of "fclus" or nearest cache. */
+		if (p->fcluster <= fclus && hit->fcluster < p->fcluster) {
+			hit = p;
+			if ((hit->fcluster + hit->nr_contig) < fclus) {
+				offset = hit->nr_contig;
+			} else {
+				offset = fclus - hit->fcluster;
+				break;
+			}
+		}
+	}
+	if (hit != &nohit) {
+		fat_cache_update_lru(inode, hit);
+
+		cid->id = MSDOS_I(inode)->cache_valid_id;
+		cid->nr_contig = hit->nr_contig;
+		cid->fcluster = hit->fcluster;
+		cid->dcluster = hit->dcluster;
+		*cached_fclus = cid->fcluster + offset;
+		*cached_dclus = cid->dcluster + offset;
+	}
+	spin_unlock(&MSDOS_I(inode)->cache_lru_lock);
+
+	return offset;
+}
+
+static struct fat_cache *fat_cache_merge(struct inode *inode,
+					 struct fat_cache_id *new)
+{
+	struct fat_cache *p;
+
+	list_for_each_entry(p, &MSDOS_I(inode)->cache_lru, cache_list) {
+		/* Find the same part as "new" in cluster-chain. */
+		if (p->fcluster == new->fcluster) {
+			BUG_ON(p->dcluster != new->dcluster);
+			if (new->nr_contig > p->nr_contig)
+				p->nr_contig = new->nr_contig;
+			return p;
+		}
+	}
+	return NULL;
+}
+
+static void fat_cache_add(struct inode *inode, struct fat_cache_id *new)
+{
+	struct fat_cache *cache, *tmp;
+
+	if (new->fcluster == -1) /* dummy cache */
+		return;
+
+	spin_lock(&MSDOS_I(inode)->cache_lru_lock);
+	if (new->id != FAT_CACHE_VALID &&
+	    new->id != MSDOS_I(inode)->cache_valid_id)
+		goto out;	/* this cache was invalidated */
+
+	cache = fat_cache_merge(inode, new);
+	if (cache == NULL) {
+		if (MSDOS_I(inode)->nr_caches < FAT_MAX_CACHE) {
+			MSDOS_I(inode)->nr_caches++;
+			spin_unlock(&MSDOS_I(inode)->cache_lru_lock);
+
+			tmp = fat_cache_alloc();
+			if (!tmp) {
+				spin_lock(&MSDOS_I(inode)->cache_lru_lock);
+				MSDOS_I(inode)->nr_caches--;
+				spin_unlock(&MSDOS_I(inode)->cache_lru_lock);
+				return;
+			}
+
+			spin_lock(&MSDOS_I(inode)->cache_lru_lock);
+			cache = fat_cache_merge(inode, new);
+			if (cache != NULL) {
+				MSDOS_I(inode)->nr_caches--;
+				fat_cache_free(tmp);
+				goto out_update_lru;
+			}
+			cache = tmp;
+		} else {
+			struct list_head *p = MSDOS_I(inode)->cache_lru.prev;
+			cache = list_entry(p, struct fat_cache, cache_list);
+		}
+		cache->fcluster = new->fcluster;
+		cache->dcluster = new->dcluster;
+		cache->nr_contig = new->nr_contig;
+	}
+out_update_lru:
+	fat_cache_update_lru(inode, cache);
+out:
+	spin_unlock(&MSDOS_I(inode)->cache_lru_lock);
+}
+
+/*
+ * Cache invalidation occurs rarely, thus the LRU chain is not updated. It
+ * fixes itself after a while.
+ */
+static void __fat_cache_inval_inode(struct inode *inode)
+{
+	struct msdos_inode_info *i = MSDOS_I(inode);
+	struct fat_cache *cache;
+
+	while (!list_empty(&i->cache_lru)) {
+		cache = list_entry(i->cache_lru.next,
+				   struct fat_cache, cache_list);
+		list_del_init(&cache->cache_list);
+		i->nr_caches--;
+		fat_cache_free(cache);
+	}
+	/* Update. The copy of caches before this id is discarded. */
+	i->cache_valid_id++;
+	if (i->cache_valid_id == FAT_CACHE_VALID)
+		i->cache_valid_id++;
+}
+
+void fat_cache_inval_inode(struct inode *inode)
+{
+	spin_lock(&MSDOS_I(inode)->cache_lru_lock);
+	__fat_cache_inval_inode(inode);
+	spin_unlock(&MSDOS_I(inode)->cache_lru_lock);
+}
+
+static inline int cache_contiguous(struct fat_cache_id *cid, int dclus)
+{
+	cid->nr_contig++;
+	return ((cid->dcluster + cid->nr_contig) == dclus);
+}
+
+static inline void cache_init(struct fat_cache_id *cid, int fclus, int dclus)
+{
+	cid->id = FAT_CACHE_VALID;
+	cid->fcluster = fclus;
+	cid->dcluster = dclus;
+	cid->nr_contig = 0;
+}
+
+int fat_get_cluster(struct inode *inode, int cluster, int *fclus, int *dclus)
+{
+	struct super_block *sb = inode->i_sb;
+	struct msdos_sb_info *sbi = MSDOS_SB(sb);
+	const int limit = sb->s_maxbytes >> sbi->cluster_bits;
+	struct fat_entry fatent;
+	struct fat_cache_id cid;
+	int nr;
+
+	BUG_ON(MSDOS_I(inode)->i_start == 0);
+
+	*fclus = 0;
+	*dclus = MSDOS_I(inode)->i_start;
+	if (!fat_valid_entry(sbi, *dclus)) {
+		fat_fs_error_ratelimit(sb,
+			"%s: invalid start cluster (i_pos %lld, start %08x)",
+			__func__, MSDOS_I(inode)->i_pos, *dclus);
+		return -EIO;
+	}
+	if (cluster == 0)
+		return 0;
+
+	if (fat_cache_lookup(inode, cluster, &cid, fclus, dclus) < 0) {
+		/*
+		 * dummy, always not contiguous
+		 * This is reinitialized by cache_init(), later.
+		 */
+		cache_init(&cid, -1, -1);
+	}
+
+	fatent_init(&fatent);
+	while (*fclus < cluster) {
+		/* prevent the infinite loop of cluster chain */
+		if (*fclus > limit) {
+			fat_fs_error_ratelimit(sb,
+				"%s: detected the cluster chain loop (i_pos %lld)",
+				__func__, MSDOS_I(inode)->i_pos);
+			nr = -EIO;
+			goto out;
+		}
+
+		nr = fat_ent_read(inode, &fatent, *dclus);
+		if (nr < 0)
+			goto out;
+		else if (nr == FAT_ENT_FREE) {
+			fat_fs_error_ratelimit(sb,
+				"%s: invalid cluster chain (i_pos %lld)",
+				__func__, MSDOS_I(inode)->i_pos);
+			nr = -EIO;
+			goto out;
+		} else if (nr == FAT_ENT_EOF) {
+			fat_cache_add(inode, &cid);
+			goto out;
+		}
+		(*fclus)++;
+		*dclus = nr;
+		if (!cache_contiguous(&cid, *dclus))
+			cache_init(&cid, *fclus, *dclus);
+	}
+	nr = 0;
+	fat_cache_add(inode, &cid);
+out:
+	fatent_brelse(&fatent);
+	return nr;
+}
+
+static int fat_bmap_cluster(struct inode *inode, int cluster)
+{
+	struct super_block *sb = inode->i_sb;
+	int ret, fclus, dclus;
+
+	if (MSDOS_I(inode)->i_start == 0)
+		return 0;
+
+	ret = fat_get_cluster(inode, cluster, &fclus, &dclus);
+	if (ret < 0)
+		return ret;
+	else if (ret == FAT_ENT_EOF) {
+		fat_fs_error(sb, "%s: request beyond EOF (i_pos %lld)",
+			     __func__, MSDOS_I(inode)->i_pos);
+		return -EIO;
+	}
+	return dclus;
+}
+
+int fat_get_mapped_cluster(struct inode *inode, sector_t sector,
+			   sector_t last_block,
+			   unsigned long *mapped_blocks, sector_t *bmap)
+{
+	struct super_block *sb = inode->i_sb;
+	struct msdos_sb_info *sbi = MSDOS_SB(sb);
+	int cluster, offset;
+
+	cluster = sector >> (sbi->cluster_bits - sb->s_blocksize_bits);
+	offset  = sector & (sbi->sec_per_clus - 1);
+	cluster = fat_bmap_cluster(inode, cluster);
+	if (cluster < 0)
+		return cluster;
+	else if (cluster) {
+		*bmap = fat_clus_to_blknr(sbi, cluster) + offset;
+		*mapped_blocks = sbi->sec_per_clus - offset;
+		if (*mapped_blocks > last_block - sector)
+			*mapped_blocks = last_block - sector;
+	}
+
+	return 0;
+}
+
+static int is_exceed_eof(struct inode *inode, sector_t sector,
+			 sector_t *last_block, int create)
+{
+	struct super_block *sb = inode->i_sb;
+	const unsigned long blocksize = sb->s_blocksize;
+	const unsigned char blocksize_bits = sb->s_blocksize_bits;
+
+	*last_block = (i_size_read(inode) + (blocksize - 1)) >> blocksize_bits;
+	if (sector >= *last_block) {
+		if (!create)
+			return 1;
+
+		/*
+		 * ->mmu_private can access on only allocation path.
+		 * (caller must hold ->i_mutex)
+		 */
+		*last_block = (MSDOS_I(inode)->mmu_private + (blocksize - 1))
+			>> blocksize_bits;
+		if (sector >= *last_block)
+			return 1;
+	}
+
+	return 0;
+}
+
+int fat_bmap(struct inode *inode, sector_t sector, sector_t *phys,
+	     unsigned long *mapped_blocks, int create, bool from_bmap)
+{
+	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
+	sector_t last_block;
+
+	*phys = 0;
+	*mapped_blocks = 0;
+	if (!is_fat32(sbi) && (inode->i_ino == MSDOS_ROOT_INO)) {
+		if (sector < (sbi->dir_entries >> sbi->dir_per_block_bits)) {
+			*phys = sector + sbi->dir_start;
+			*mapped_blocks = 1;
+		}
+		return 0;
+	}
+
+	if (!from_bmap) {
+		if (is_exceed_eof(inode, sector, &last_block, create))
+			return 0;
+	} else {
+		last_block = inode->i_blocks >>
+				(inode->i_sb->s_blocksize_bits - 9);
+		if (sector >= last_block)
+			return 0;
+	}
+
+	return fat_get_mapped_cluster(inode, sector, last_block, mapped_blocks,
+				      phys);
+}