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, 434 insertions, 0 deletions

diff --git a/fs/jffs2/build.c b/fs/jffs2/build.c
new file mode 100644
index 000000000..6ae9d6fef
--- /dev/null
+++ b/fs/jffs2/build.c
@@ -0,0 +1,434 @@
+/*
+ * JFFS2 -- Journalling Flash File System, Version 2.
+ *
+ * Copyright © 2001-2007 Red Hat, Inc.
+ * Copyright © 2004-2010 David Woodhouse <dwmw2@infradead.org>
+ *
+ * Created by David Woodhouse <dwmw2@infradead.org>
+ *
+ * For licensing information, see the file 'LICENCE' in this directory.
+ *
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/mtd/mtd.h>
+#include <linux/mm.h> /* kvfree() */
+#include "nodelist.h"
+
+static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *,
+		struct jffs2_inode_cache *, struct jffs2_full_dirent **);
+
+static inline struct jffs2_inode_cache *
+first_inode_chain(int *i, struct jffs2_sb_info *c)
+{
+	for (; *i < c->inocache_hashsize; (*i)++) {
+		if (c->inocache_list[*i])
+			return c->inocache_list[*i];
+	}
+	return NULL;
+}
+
+static inline struct jffs2_inode_cache *
+next_inode(int *i, struct jffs2_inode_cache *ic, struct jffs2_sb_info *c)
+{
+	/* More in this chain? */
+	if (ic->next)
+		return ic->next;
+	(*i)++;
+	return first_inode_chain(i, c);
+}
+
+#define for_each_inode(i, c, ic)			\
+	for (i = 0, ic = first_inode_chain(&i, (c));	\
+	     ic;					\
+	     ic = next_inode(&i, ic, (c)))
+
+
+static void jffs2_build_inode_pass1(struct jffs2_sb_info *c,
+				    struct jffs2_inode_cache *ic,
+				    int *dir_hardlinks)
+{
+	struct jffs2_full_dirent *fd;
+
+	dbg_fsbuild("building directory inode #%u\n", ic->ino);
+
+	/* For each child, increase nlink */
+	for(fd = ic->scan_dents; fd; fd = fd->next) {
+		struct jffs2_inode_cache *child_ic;
+		if (!fd->ino)
+			continue;
+
+		/* we can get high latency here with huge directories */
+
+		child_ic = jffs2_get_ino_cache(c, fd->ino);
+		if (!child_ic) {
+			dbg_fsbuild("child \"%s\" (ino #%u) of dir ino #%u doesn't exist!\n",
+				  fd->name, fd->ino, ic->ino);
+			jffs2_mark_node_obsolete(c, fd->raw);
+			/* Clear the ic/raw union so it doesn't cause problems later. */
+			fd->ic = NULL;
+			continue;
+		}
+
+		/* From this point, fd->raw is no longer used so we can set fd->ic */
+		fd->ic = child_ic;
+		child_ic->pino_nlink++;
+		/* If we appear (at this stage) to have hard-linked directories,
+		 * set a flag to trigger a scan later */
+		if (fd->type == DT_DIR) {
+			child_ic->flags |= INO_FLAGS_IS_DIR;
+			if (child_ic->pino_nlink > 1)
+				*dir_hardlinks = 1;
+		}
+
+		dbg_fsbuild("increased nlink for child \"%s\" (ino #%u)\n", fd->name, fd->ino);
+		/* Can't free scan_dents so far. We might need them in pass 2 */
+	}
+}
+
+/* Scan plan:
+ - Scan physical nodes. Build map of inodes/dirents. Allocate inocaches as we go
+ - Scan directory tree from top down, setting nlink in inocaches
+ - Scan inocaches for inodes with nlink==0
+*/
+static int jffs2_build_filesystem(struct jffs2_sb_info *c)
+{
+	int ret, i, dir_hardlinks = 0;
+	struct jffs2_inode_cache *ic;
+	struct jffs2_full_dirent *fd;
+	struct jffs2_full_dirent *dead_fds = NULL;
+
+	dbg_fsbuild("build FS data structures\n");
+
+	/* First, scan the medium and build all the inode caches with
+	   lists of physical nodes */
+
+	c->flags |= JFFS2_SB_FLAG_SCANNING;
+	ret = jffs2_scan_medium(c);
+	c->flags &= ~JFFS2_SB_FLAG_SCANNING;
+	if (ret)
+		goto exit;
+
+	dbg_fsbuild("scanned flash completely\n");
+	jffs2_dbg_dump_block_lists_nolock(c);
+
+	dbg_fsbuild("pass 1 starting\n");
+	c->flags |= JFFS2_SB_FLAG_BUILDING;
+	/* Now scan the directory tree, increasing nlink according to every dirent found. */
+	for_each_inode(i, c, ic) {
+		if (ic->scan_dents) {
+			jffs2_build_inode_pass1(c, ic, &dir_hardlinks);
+			cond_resched();
+		}
+	}
+
+	dbg_fsbuild("pass 1 complete\n");
+
+	/* Next, scan for inodes with nlink == 0 and remove them. If
+	   they were directories, then decrement the nlink of their
+	   children too, and repeat the scan. As that's going to be
+	   a fairly uncommon occurrence, it's not so evil to do it this
+	   way. Recursion bad. */
+	dbg_fsbuild("pass 2 starting\n");
+
+	for_each_inode(i, c, ic) {
+		if (ic->pino_nlink)
+			continue;
+
+		jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
+		cond_resched();
+	}
+
+	dbg_fsbuild("pass 2a starting\n");
+
+	while (dead_fds) {
+		fd = dead_fds;
+		dead_fds = fd->next;
+
+		ic = jffs2_get_ino_cache(c, fd->ino);
+
+		if (ic)
+			jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
+		jffs2_free_full_dirent(fd);
+	}
+
+	dbg_fsbuild("pass 2a complete\n");
+
+	if (dir_hardlinks) {
+		/* If we detected directory hardlinks earlier, *hopefully*
+		 * they are gone now because some of the links were from
+		 * dead directories which still had some old dirents lying
+		 * around and not yet garbage-collected, but which have
+		 * been discarded above. So clear the pino_nlink field
+		 * in each directory, so that the final scan below can
+		 * print appropriate warnings. */
+		for_each_inode(i, c, ic) {
+			if (ic->flags & INO_FLAGS_IS_DIR)
+				ic->pino_nlink = 0;
+		}
+	}
+	dbg_fsbuild("freeing temporary data structures\n");
+
+	/* Finally, we can scan again and free the dirent structs */
+	for_each_inode(i, c, ic) {
+		while(ic->scan_dents) {
+			fd = ic->scan_dents;
+			ic->scan_dents = fd->next;
+			/* We do use the pino_nlink field to count nlink of
+			 * directories during fs build, so set it to the
+			 * parent ino# now. Now that there's hopefully only
+			 * one. */
+			if (fd->type == DT_DIR) {
+				if (!fd->ic) {
+					/* We'll have complained about it and marked the coresponding
+					   raw node obsolete already. Just skip it. */
+					continue;
+				}
+
+				/* We *have* to have set this in jffs2_build_inode_pass1() */
+				BUG_ON(!(fd->ic->flags & INO_FLAGS_IS_DIR));
+
+				/* We clear ic->pino_nlink ∀ directories' ic *only* if dir_hardlinks
+				 * is set. Otherwise, we know this should never trigger anyway, so
+				 * we don't do the check. And ic->pino_nlink still contains the nlink
+				 * value (which is 1). */
+				if (dir_hardlinks && fd->ic->pino_nlink) {
+					JFFS2_ERROR("child dir \"%s\" (ino #%u) of dir ino #%u is also hard linked from dir ino #%u\n",
+						    fd->name, fd->ino, ic->ino, fd->ic->pino_nlink);
+					/* Should we unlink it from its previous parent? */
+				}
+
+				/* For directories, ic->pino_nlink holds that parent inode # */
+				fd->ic->pino_nlink = ic->ino;
+			}
+			jffs2_free_full_dirent(fd);
+		}
+		ic->scan_dents = NULL;
+		cond_resched();
+	}
+	ret = jffs2_build_xattr_subsystem(c);
+	if (ret)
+		goto exit;
+
+	c->flags &= ~JFFS2_SB_FLAG_BUILDING;
+
+	dbg_fsbuild("FS build complete\n");
+
+	/* Rotate the lists by some number to ensure wear levelling */
+	jffs2_rotate_lists(c);
+
+	ret = 0;
+
+exit:
+	if (ret) {
+		for_each_inode(i, c, ic) {
+			while(ic->scan_dents) {
+				fd = ic->scan_dents;
+				ic->scan_dents = fd->next;
+				jffs2_free_full_dirent(fd);
+			}
+		}
+		jffs2_clear_xattr_subsystem(c);
+	}
+
+	return ret;
+}
+
+static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *c,
+					struct jffs2_inode_cache *ic,
+					struct jffs2_full_dirent **dead_fds)
+{
+	struct jffs2_raw_node_ref *raw;
+	struct jffs2_full_dirent *fd;
+
+	dbg_fsbuild("removing ino #%u with nlink == zero.\n", ic->ino);
+
+	raw = ic->nodes;
+	while (raw != (void *)ic) {
+		struct jffs2_raw_node_ref *next = raw->next_in_ino;
+		dbg_fsbuild("obsoleting node at 0x%08x\n", ref_offset(raw));
+		jffs2_mark_node_obsolete(c, raw);
+		raw = next;
+	}
+
+	if (ic->scan_dents) {
+		int whinged = 0;
+		dbg_fsbuild("inode #%u was a directory which may have children...\n", ic->ino);
+
+		while(ic->scan_dents) {
+			struct jffs2_inode_cache *child_ic;
+
+			fd = ic->scan_dents;
+			ic->scan_dents = fd->next;
+
+			if (!fd->ino) {
+				/* It's a deletion dirent. Ignore it */
+				dbg_fsbuild("child \"%s\" is a deletion dirent, skipping...\n", fd->name);
+				jffs2_free_full_dirent(fd);
+				continue;
+			}
+			if (!whinged)
+				whinged = 1;
+
+			dbg_fsbuild("removing child \"%s\", ino #%u\n", fd->name, fd->ino);
+
+			child_ic = jffs2_get_ino_cache(c, fd->ino);
+			if (!child_ic) {
+				dbg_fsbuild("cannot remove child \"%s\", ino #%u, because it doesn't exist\n",
+						fd->name, fd->ino);
+				jffs2_free_full_dirent(fd);
+				continue;
+			}
+
+			/* Reduce nlink of the child. If it's now zero, stick it on the
+			   dead_fds list to be cleaned up later. Else just free the fd */
+			child_ic->pino_nlink--;
+
+			if (!child_ic->pino_nlink) {
+				dbg_fsbuild("inode #%u (\"%s\") now has no links; adding to dead_fds list.\n",
+					  fd->ino, fd->name);
+				fd->next = *dead_fds;
+				*dead_fds = fd;
+			} else {
+				dbg_fsbuild("inode #%u (\"%s\") has now got nlink %d. Ignoring.\n",
+					  fd->ino, fd->name, child_ic->pino_nlink);
+				jffs2_free_full_dirent(fd);
+			}
+		}
+	}
+
+	/*
+	   We don't delete the inocache from the hash list and free it yet.
+	   The erase code will do that, when all the nodes are completely gone.
+	*/
+}
+
+static void jffs2_calc_trigger_levels(struct jffs2_sb_info *c)
+{
+	uint32_t size;
+
+	/* Deletion should almost _always_ be allowed. We're fairly
+	   buggered once we stop allowing people to delete stuff
+	   because there's not enough free space... */
+	c->resv_blocks_deletion = 2;
+
+	/* Be conservative about how much space we need before we allow writes.
+	   On top of that which is required for deletia, require an extra 2%
+	   of the medium to be available, for overhead caused by nodes being
+	   split across blocks, etc. */
+
+	size = c->flash_size / 50; /* 2% of flash size */
+	size += c->nr_blocks * 100; /* And 100 bytes per eraseblock */
+	size += c->sector_size - 1; /* ... and round up */
+
+	c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
+
+	/* When do we let the GC thread run in the background */
+
+	c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
+
+	/* When do we allow garbage collection to merge nodes to make
+	   long-term progress at the expense of short-term space exhaustion? */
+	c->resv_blocks_gcmerge = c->resv_blocks_deletion + 1;
+
+	/* When do we allow garbage collection to eat from bad blocks rather
+	   than actually making progress? */
+	c->resv_blocks_gcbad = 0;//c->resv_blocks_deletion + 2;
+
+	/* What number of 'very dirty' eraseblocks do we allow before we
+	   trigger the GC thread even if we don't _need_ the space. When we
+	   can't mark nodes obsolete on the medium, the old dirty nodes cause
+	   performance problems because we have to inspect and discard them. */
+	c->vdirty_blocks_gctrigger = c->resv_blocks_gctrigger;
+	if (jffs2_can_mark_obsolete(c))
+		c->vdirty_blocks_gctrigger *= 10;
+
+	/* If there's less than this amount of dirty space, don't bother
+	   trying to GC to make more space. It'll be a fruitless task */
+	c->nospc_dirty_size = c->sector_size + (c->flash_size / 100);
+
+	dbg_fsbuild("trigger levels (size %d KiB, block size %d KiB, %d blocks)\n",
+		    c->flash_size / 1024, c->sector_size / 1024, c->nr_blocks);
+	dbg_fsbuild("Blocks required to allow deletion:    %d (%d KiB)\n",
+		  c->resv_blocks_deletion, c->resv_blocks_deletion*c->sector_size/1024);
+	dbg_fsbuild("Blocks required to allow writes:      %d (%d KiB)\n",
+		  c->resv_blocks_write, c->resv_blocks_write*c->sector_size/1024);
+	dbg_fsbuild("Blocks required to quiesce GC thread: %d (%d KiB)\n",
+		  c->resv_blocks_gctrigger, c->resv_blocks_gctrigger*c->sector_size/1024);
+	dbg_fsbuild("Blocks required to allow GC merges:   %d (%d KiB)\n",
+		  c->resv_blocks_gcmerge, c->resv_blocks_gcmerge*c->sector_size/1024);
+	dbg_fsbuild("Blocks required to GC bad blocks:     %d (%d KiB)\n",
+		  c->resv_blocks_gcbad, c->resv_blocks_gcbad*c->sector_size/1024);
+	dbg_fsbuild("Amount of dirty space required to GC: %d bytes\n",
+		  c->nospc_dirty_size);
+	dbg_fsbuild("Very dirty blocks before GC triggered: %d\n",
+		  c->vdirty_blocks_gctrigger);
+}
+
+int jffs2_do_mount_fs(struct jffs2_sb_info *c)
+{
+	int ret;
+	int i;
+	int size;
+
+	c->free_size = c->flash_size;
+	c->nr_blocks = c->flash_size / c->sector_size;
+	size = sizeof(struct jffs2_eraseblock) * c->nr_blocks;
+#ifndef __ECOS
+	if (jffs2_blocks_use_vmalloc(c))
+		c->blocks = vzalloc(size);
+	else
+#endif
+		c->blocks = kzalloc(size, GFP_KERNEL);
+	if (!c->blocks)
+		return -ENOMEM;
+
+	for (i=0; i<c->nr_blocks; i++) {
+		INIT_LIST_HEAD(&c->blocks[i].list);
+		c->blocks[i].offset = i * c->sector_size;
+		c->blocks[i].free_size = c->sector_size;
+	}
+
+	INIT_LIST_HEAD(&c->clean_list);
+	INIT_LIST_HEAD(&c->very_dirty_list);
+	INIT_LIST_HEAD(&c->dirty_list);
+	INIT_LIST_HEAD(&c->erasable_list);
+	INIT_LIST_HEAD(&c->erasing_list);
+	INIT_LIST_HEAD(&c->erase_checking_list);
+	INIT_LIST_HEAD(&c->erase_pending_list);
+	INIT_LIST_HEAD(&c->erasable_pending_wbuf_list);
+	INIT_LIST_HEAD(&c->erase_complete_list);
+	INIT_LIST_HEAD(&c->free_list);
+	INIT_LIST_HEAD(&c->bad_list);
+	INIT_LIST_HEAD(&c->bad_used_list);
+	c->highest_ino = 1;
+	c->summary = NULL;
+
+	ret = jffs2_sum_init(c);
+	if (ret)
+		goto out_free;
+
+	if (jffs2_build_filesystem(c)) {
+		dbg_fsbuild("build_fs failed\n");
+		jffs2_free_ino_caches(c);
+		jffs2_free_raw_node_refs(c);
+		ret = -EIO;
+		goto out_sum_exit;
+	}
+
+	jffs2_calc_trigger_levels(c);
+
+	return 0;
+
+ out_sum_exit:
+	jffs2_sum_exit(c);
+ out_free:
+	kvfree(c->blocks);
+
+	return ret;
+}