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, 486 insertions, 0 deletions

diff --git a/fs/erofs/decompressor.c b/fs/erofs/decompressor.c
new file mode 100644
index 000000000..d387b27c4
--- /dev/null
+++ b/fs/erofs/decompressor.c
@@ -0,0 +1,486 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2019 HUAWEI, Inc.
+ *             https://www.huawei.com/
+ * Copyright (C) 2024 Alibaba Cloud
+ */
+#include "compress.h"
+#include <linux/lz4.h>
+
+static int z_erofs_load_lz4_config(struct super_block *sb,
+			    struct erofs_super_block *dsb, void *data, int size)
+{
+	struct erofs_sb_info *sbi = EROFS_SB(sb);
+	struct z_erofs_lz4_cfgs *lz4 = data;
+	u16 distance;
+
+	if (lz4) {
+		if (size < sizeof(struct z_erofs_lz4_cfgs)) {
+			erofs_err(sb, "invalid lz4 cfgs, size=%u", size);
+			return -EINVAL;
+		}
+		sbi->lz4.max_pclusterblks = le16_to_cpu(lz4->max_pclusterblks);
+		if (!sbi->lz4.max_pclusterblks) {
+			sbi->lz4.max_pclusterblks = 1;	/* reserved case */
+		} else if (sbi->lz4.max_pclusterblks >
+			   erofs_blknr(sb, Z_EROFS_PCLUSTER_MAX_SIZE)) {
+			erofs_err(sb, "too large lz4 pclusterblks %u",
+				  sbi->lz4.max_pclusterblks);
+			return -EINVAL;
+		}
+	} else {
+		distance = le16_to_cpu(dsb->u1.lz4_max_distance);
+		if (!distance && !erofs_sb_has_lz4_0padding(sbi))
+			return 0;
+		sbi->lz4.max_pclusterblks = 1;
+		sbi->available_compr_algs = 1 << Z_EROFS_COMPRESSION_LZ4;
+	}
+	return z_erofs_gbuf_growsize(sbi->lz4.max_pclusterblks);
+}
+
+/*
+ * Fill all gaps with bounce pages if it's a sparse page list (for example some
+ * folios are already uptodate and thus can be mapped into userspace). Also
+ * check if pages are physically consecutive, which can be seen for moderate CR.
+ */
+static int z_erofs_oneshot_prepare_dstpages(struct z_erofs_decompress_req *rq,
+					    struct page **pagepool)
+{
+	void *kaddr = NULL;
+	unsigned int i;
+
+	for (i = 0; i < rq->outpages; ++i) {
+		struct page *page, *victim;
+
+		page = rq->out[i];
+		if (page) {
+			if (!PageHighMem(page)) {
+				if (!i) {
+					kaddr = page_address(page);
+					continue;
+				}
+				if (kaddr &&
+				    kaddr + PAGE_SIZE == page_address(page)) {
+					kaddr += PAGE_SIZE;
+					continue;
+				}
+			}
+		} else {
+			victim = __erofs_allocpage(pagepool, rq->gfp, true);
+			if (!victim)
+				return -ENOMEM;
+			set_page_private(victim, Z_EROFS_SHORTLIVED_PAGE);
+			rq->out[i] = victim;
+		}
+		kaddr = NULL;
+	}
+	return kaddr ? 1 : 0;
+}
+
+static void *z_erofs_lz4_handle_overlap(const struct z_erofs_decompress_req *rq,
+			void *inpage, void *out, unsigned int *inputmargin,
+			int *maptype, bool may_inplace)
+{
+	unsigned int oend, omargin, cnt, i;
+	struct page **in;
+	void *src;
+
+	/*
+	 * If in-place I/O isn't used, for example, the bounce compressed cache
+	 * can hold data for incomplete read requests. Just map the compressed
+	 * buffer as well and decompress directly.
+	 */
+	if (!rq->inplace_io) {
+		if (rq->inpages <= 1) {
+			*maptype = 0;
+			return inpage;
+		}
+		kunmap_local(inpage);
+		src = erofs_vm_map_ram(rq->in, rq->inpages);
+		if (!src)
+			return ERR_PTR(-ENOMEM);
+		*maptype = 1;
+		return src;
+	}
+	/*
+	 * Then, deal with in-place I/Os. The reasons why in-place I/O is useful
+	 * are: (1) It minimizes memory footprint during the I/O submission,
+	 * which is useful for slow storage (including network devices and
+	 * low-end HDDs/eMMCs) but with a lot inflight I/Os; (2) If in-place
+	 * decompression can also be applied, it will reuse the unique buffer so
+	 * that no extra CPU D-cache is polluted with temporary compressed data
+	 * for extreme performance.
+	 */
+	oend = rq->pageofs_out + rq->outputsize;
+	omargin = PAGE_ALIGN(oend) - oend;
+	if (!rq->partial_decoding && may_inplace &&
+	    omargin >= LZ4_DECOMPRESS_INPLACE_MARGIN(rq->inputsize)) {
+		for (i = 0; i < rq->inpages; ++i)
+			if (rq->out[rq->outpages - rq->inpages + i] !=
+			    rq->in[i])
+				break;
+		if (i >= rq->inpages) {
+			kunmap_local(inpage);
+			*maptype = 3;
+			return out + ((rq->outpages - rq->inpages) << PAGE_SHIFT);
+		}
+	}
+	/*
+	 * If in-place decompression can't be applied, copy compressed data that
+	 * may potentially overlap during decompression to a per-CPU buffer.
+	 */
+	src = z_erofs_get_gbuf(rq->inpages);
+	if (!src) {
+		DBG_BUGON(1);
+		kunmap_local(inpage);
+		return ERR_PTR(-EFAULT);
+	}
+
+	for (i = 0, in = rq->in; i < rq->inputsize; i += cnt, ++in) {
+		cnt = min_t(u32, rq->inputsize - i, PAGE_SIZE - *inputmargin);
+		if (!inpage)
+			inpage = kmap_local_page(*in);
+		memcpy(src + i, inpage + *inputmargin, cnt);
+		kunmap_local(inpage);
+		inpage = NULL;
+		*inputmargin = 0;
+	}
+	*maptype = 2;
+	return src;
+}
+
+/*
+ * Get the exact on-disk size of the compressed data:
+ *  - For LZ4, it should apply if the zero_padding feature is on (5.3+);
+ *  - For others, zero_padding is enabled all the time.
+ */
+const char *z_erofs_fixup_insize(struct z_erofs_decompress_req *rq,
+				 const char *padbuf, unsigned int padbufsize)
+{
+	const char *padend;
+
+	padend = memchr_inv(padbuf, 0, padbufsize);
+	if (!padend)
+		return "compressed data start not found";
+	rq->inputsize -= padend - padbuf;
+	rq->pageofs_in += padend - padbuf;
+	return NULL;
+}
+
+static const char *__z_erofs_lz4_decompress(struct z_erofs_decompress_req *rq,
+					    u8 *dst)
+{
+	bool may_inplace = false;
+	unsigned int inputmargin;
+	u8 *out, *headpage, *src;
+	const char *reason;
+	int ret, maptype;
+
+	headpage = kmap_local_page(*rq->in);
+	reason = z_erofs_fixup_insize(rq, headpage + rq->pageofs_in,
+			min_t(unsigned int, rq->inputsize,
+			      rq->sb->s_blocksize - rq->pageofs_in));
+	if (reason) {
+		kunmap_local(headpage);
+		return reason;
+	}
+	may_inplace = !((rq->pageofs_in + rq->inputsize) &
+			(rq->sb->s_blocksize - 1));
+
+	inputmargin = rq->pageofs_in;
+	src = z_erofs_lz4_handle_overlap(rq, headpage, dst, &inputmargin,
+					 &maptype, may_inplace);
+	if (IS_ERR(src))
+		return ERR_CAST(src);
+
+	out = dst + rq->pageofs_out;
+	if (rq->partial_decoding)
+		ret = LZ4_decompress_safe_partial(src + inputmargin, out,
+				rq->inputsize, rq->outputsize, rq->outputsize);
+	else
+		ret = LZ4_decompress_safe(src + inputmargin, out,
+					  rq->inputsize, rq->outputsize);
+	if (ret == rq->outputsize)
+		reason = NULL;
+	else if (ret < 0)
+		reason = "corrupted compressed data";
+	else
+		reason = "unexpected end of stream";
+
+	if (!maptype) {
+		kunmap_local(headpage);
+	} else if (maptype == 1) {
+		vm_unmap_ram(src, rq->inpages);
+	} else if (maptype == 2) {
+		z_erofs_put_gbuf(src);
+	} else if (maptype != 3) {
+		DBG_BUGON(1);
+		return ERR_PTR(-EFAULT);
+	}
+	return reason;
+}
+
+static const char *z_erofs_lz4_decompress(struct z_erofs_decompress_req *rq,
+					  struct page **pagepool)
+{
+	unsigned int dst_maptype;
+	const char *reason;
+	void *dst;
+	int ret;
+
+	/* one optimized fast path only for non bigpcluster cases yet */
+	if (rq->inpages == 1 && rq->outpages == 1 && !rq->inplace_io) {
+		DBG_BUGON(!*rq->out);
+		dst = kmap_local_page(*rq->out);
+		dst_maptype = 0;
+	} else {
+		/* general decoding path which can be used for all cases */
+		ret = z_erofs_oneshot_prepare_dstpages(rq, pagepool);
+		if (ret < 0)
+			return ERR_PTR(ret);
+		if (ret > 0) {
+			dst = page_address(*rq->out);
+			dst_maptype = 1;
+		} else {
+			dst = erofs_vm_map_ram(rq->out, rq->outpages);
+			if (!dst)
+				return ERR_PTR(-ENOMEM);
+			dst_maptype = 2;
+		}
+	}
+	reason = __z_erofs_lz4_decompress(rq, dst);
+	if (!dst_maptype)
+		kunmap_local(dst);
+	else if (dst_maptype == 2)
+		vm_unmap_ram(dst, rq->outpages);
+	return reason;
+}
+
+static const char *z_erofs_transform_plain(struct z_erofs_decompress_req *rq,
+					   struct page **pagepool)
+{
+	const unsigned int nrpages_in = rq->inpages, nrpages_out = rq->outpages;
+	const unsigned int bs = rq->sb->s_blocksize;
+	unsigned int cur = 0, ni = 0, no, pi, po, insz, cnt;
+	u8 *kin;
+
+	if (rq->outputsize > rq->inputsize)
+		return ERR_PTR(-EOPNOTSUPP);
+	if (rq->alg == Z_EROFS_COMPRESSION_INTERLACED) {
+		cur = bs - (rq->pageofs_out & (bs - 1));
+		DBG_BUGON(rq->pageofs_in & (bs - 1));
+		pi = (rq->pageofs_in + rq->inputsize - cur) & ~PAGE_MASK;
+		cur = min(cur, rq->outputsize);
+		if (cur && rq->out[0]) {
+			kin = kmap_local_page(rq->in[nrpages_in - 1]);
+			if (rq->out[0] == rq->in[nrpages_in - 1])
+				memmove(kin + rq->pageofs_out, kin + pi, cur);
+			else
+				memcpy_to_page(rq->out[0], rq->pageofs_out,
+					       kin + pi, cur);
+			kunmap_local(kin);
+		}
+		rq->outputsize -= cur;
+	}
+
+	for (; rq->outputsize; rq->pageofs_in = 0, cur += insz, ni++) {
+		insz = min(PAGE_SIZE - rq->pageofs_in, rq->outputsize);
+		rq->outputsize -= insz;
+		if (!rq->in[ni])
+			continue;
+		kin = kmap_local_page(rq->in[ni]);
+		pi = 0;
+		do {
+			no = (rq->pageofs_out + cur + pi) >> PAGE_SHIFT;
+			po = (rq->pageofs_out + cur + pi) & ~PAGE_MASK;
+			DBG_BUGON(no >= nrpages_out);
+			cnt = min(insz - pi, PAGE_SIZE - po);
+			if (rq->out[no] == rq->in[ni])
+				memmove(kin + po,
+					kin + rq->pageofs_in + pi, cnt);
+			else if (rq->out[no])
+				memcpy_to_page(rq->out[no], po,
+					       kin + rq->pageofs_in + pi, cnt);
+			pi += cnt;
+		} while (pi < insz);
+		kunmap_local(kin);
+	}
+	DBG_BUGON(ni > nrpages_in);
+	return NULL;
+}
+
+const char *z_erofs_stream_switch_bufs(struct z_erofs_stream_dctx *dctx,
+				void **dst, void **src, struct page **pgpl)
+{
+	struct z_erofs_decompress_req *rq = dctx->rq;
+	struct page **pgo, *tmppage;
+	unsigned int j;
+
+	if (!dctx->avail_out) {
+		if (++dctx->no >= rq->outpages || !rq->outputsize)
+			return "insufficient space for decompressed data";
+
+		if (dctx->kout)
+			kunmap_local(dctx->kout);
+		dctx->avail_out = min(rq->outputsize, PAGE_SIZE - rq->pageofs_out);
+		rq->outputsize -= dctx->avail_out;
+		pgo = &rq->out[dctx->no];
+		if (!*pgo && rq->fillgaps) {		/* deduped */
+			*pgo = erofs_allocpage(pgpl, rq->gfp);
+			if (!*pgo) {
+				dctx->kout = NULL;
+				return ERR_PTR(-ENOMEM);
+			}
+			set_page_private(*pgo, Z_EROFS_SHORTLIVED_PAGE);
+		}
+		if (*pgo) {
+			dctx->kout = kmap_local_page(*pgo);
+			*dst = dctx->kout + rq->pageofs_out;
+		} else {
+			*dst = dctx->kout = NULL;
+		}
+		rq->pageofs_out = 0;
+	}
+
+	if (dctx->inbuf_pos == dctx->inbuf_sz && rq->inputsize) {
+		if (++dctx->ni >= rq->inpages)
+			return "invalid compressed data";
+		if (dctx->kout) /* unlike kmap(), take care of the orders */
+			kunmap_local(dctx->kout);
+		kunmap_local(dctx->kin);
+
+		dctx->inbuf_sz = min_t(u32, rq->inputsize, PAGE_SIZE);
+		rq->inputsize -= dctx->inbuf_sz;
+		dctx->kin = kmap_local_page(rq->in[dctx->ni]);
+		*src = dctx->kin;
+		dctx->bounced = false;
+		if (dctx->kout) {
+			j = (u8 *)*dst - dctx->kout;
+			dctx->kout = kmap_local_page(rq->out[dctx->no]);
+			*dst = dctx->kout + j;
+		}
+		dctx->inbuf_pos = 0;
+	}
+
+	/*
+	 * Handle overlapping: Use the given bounce buffer if the input data is
+	 * under processing; Or utilize short-lived pages from the on-stack page
+	 * pool, where pages are shared among the same request.  Note that only
+	 * a few inplace I/O pages need to be doubled.
+	 */
+	if (!dctx->bounced && rq->out[dctx->no] == rq->in[dctx->ni]) {
+		memcpy(dctx->bounce, *src, dctx->inbuf_sz);
+		*src = dctx->bounce;
+		dctx->bounced = true;
+	}
+
+	for (j = dctx->ni + 1; j < rq->inpages; ++j) {
+		if (rq->out[dctx->no] != rq->in[j])
+			continue;
+		tmppage = erofs_allocpage(pgpl, rq->gfp);
+		if (!tmppage)
+			return ERR_PTR(-ENOMEM);
+		set_page_private(tmppage, Z_EROFS_SHORTLIVED_PAGE);
+		copy_highpage(tmppage, rq->in[j]);
+		rq->in[j] = tmppage;
+	}
+	return NULL;
+}
+
+const struct z_erofs_decompressor *z_erofs_decomp[] = {
+	[Z_EROFS_COMPRESSION_SHIFTED] = &(const struct z_erofs_decompressor) {
+		.decompress = z_erofs_transform_plain,
+		.name = "shifted"
+	},
+	[Z_EROFS_COMPRESSION_INTERLACED] = &(const struct z_erofs_decompressor) {
+		.decompress = z_erofs_transform_plain,
+		.name = "interlaced"
+	},
+	[Z_EROFS_COMPRESSION_LZ4] = &(const struct z_erofs_decompressor) {
+		.config = z_erofs_load_lz4_config,
+		.decompress = z_erofs_lz4_decompress,
+		.init = z_erofs_gbuf_init,
+		.exit = z_erofs_gbuf_exit,
+		.name = "lz4"
+	},
+#ifdef CONFIG_EROFS_FS_ZIP_LZMA
+	[Z_EROFS_COMPRESSION_LZMA] = &z_erofs_lzma_decomp,
+#endif
+#ifdef CONFIG_EROFS_FS_ZIP_DEFLATE
+	[Z_EROFS_COMPRESSION_DEFLATE] = &z_erofs_deflate_decomp,
+#endif
+#ifdef CONFIG_EROFS_FS_ZIP_ZSTD
+	[Z_EROFS_COMPRESSION_ZSTD] = &z_erofs_zstd_decomp,
+#endif
+};
+
+int z_erofs_parse_cfgs(struct super_block *sb, struct erofs_super_block *dsb)
+{
+	struct erofs_sb_info *sbi = EROFS_SB(sb);
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
+	unsigned long algs, alg;
+	erofs_off_t offset;
+	int size, ret = 0;
+
+	if (!erofs_sb_has_compr_cfgs(sbi))
+		return z_erofs_load_lz4_config(sb, dsb, NULL, 0);
+
+	algs = le16_to_cpu(dsb->u1.available_compr_algs);
+	sbi->available_compr_algs = algs;
+	if (algs & ~Z_EROFS_ALL_COMPR_ALGS) {
+		erofs_err(sb, "unidentified algorithms %lx, please upgrade kernel",
+			  algs & ~Z_EROFS_ALL_COMPR_ALGS);
+		return -EOPNOTSUPP;
+	}
+
+	(void)erofs_init_metabuf(&buf, sb, false);
+	offset = EROFS_SUPER_OFFSET + sbi->sb_size;
+	for_each_set_bit(alg, &algs, Z_EROFS_COMPRESSION_MAX) {
+		const struct z_erofs_decompressor *dec = z_erofs_decomp[alg];
+		void *data;
+
+		data = erofs_read_metadata(sb, &buf, &offset, &size);
+		if (IS_ERR(data)) {
+			ret = PTR_ERR(data);
+			break;
+		}
+
+		if (dec && dec->config) {
+			ret = dec->config(sb, dsb, data, size);
+		} else {
+			erofs_err(sb, "algorithm %ld isn't enabled on this kernel",
+				  alg);
+			ret = -EOPNOTSUPP;
+		}
+		kfree(data);
+		if (ret)
+			break;
+	}
+	erofs_put_metabuf(&buf);
+	return ret;
+}
+
+int __init z_erofs_init_decompressor(void)
+{
+	int i, err;
+
+	for (i = 0; i < Z_EROFS_COMPRESSION_MAX; ++i) {
+		err = z_erofs_decomp[i] ? z_erofs_decomp[i]->init() : 0;
+		if (err) {
+			while (i--)
+				if (z_erofs_decomp[i])
+					z_erofs_decomp[i]->exit();
+			return err;
+		}
+	}
+	return 0;
+}
+
+void z_erofs_exit_decompressor(void)
+{
+	int i;
+
+	for (i = 0; i < Z_EROFS_COMPRESSION_MAX; ++i)
+		if (z_erofs_decomp[i])
+			z_erofs_decomp[i]->exit();
+}