From a5f7a5bb3b7f28ba7e4fa246775b29a0e5537255 Mon Sep 17 00:00:00 2001 From: Jinke Han Date: Tue, 8 Sep 2026 15:37:42 +0800 Subject: x86/kprobes: Fix crash when probing CS CALL instructions 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) Signed-off-by: Jinke Han Signed-off-by: Ingo Molnar Reviewed-by: Masami Hiramatsu (Google) Acked-by: Yafang Shao Acked-by: Borislav Petkov Cc: Peter Zijlstra Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000 --- lib/ts_bm.c | 237 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 237 insertions(+) create mode 100644 lib/ts_bm.c (limited to 'lib/ts_bm.c') diff --git a/lib/ts_bm.c b/lib/ts_bm.c new file mode 100644 index 000000000..676105e84 --- /dev/null +++ b/lib/ts_bm.c @@ -0,0 +1,237 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * lib/ts_bm.c Boyer-Moore text search implementation + * + * Authors: Pablo Neira Ayuso + * + * ========================================================================== + * + * Implements Boyer-Moore string matching algorithm: + * + * [1] A Fast String Searching Algorithm, R.S. Boyer and Moore. + * Communications of the Association for Computing Machinery, + * 20(10), 1977, pp. 762-772. + * https://www.cs.utexas.edu/users/moore/publications/fstrpos.pdf + * + * [2] Handbook of Exact String Matching Algorithms, Thierry Lecroq, 2004 + * http://www-igm.univ-mlv.fr/~lecroq/string/string.pdf + * + * Note: Since Boyer-Moore (BM) performs searches for matchings from right + * to left, it's still possible that a matching could be spread over + * multiple blocks, in that case this algorithm won't find any coincidence. + * + * If you're willing to ensure that such thing won't ever happen, use the + * Knuth-Pratt-Morris (KMP) implementation instead. In conclusion, choose + * the proper string search algorithm depending on your setting. + * + * Say you're using the textsearch infrastructure for filtering, NIDS or + * any similar security focused purpose, then go KMP. Otherwise, if you + * really care about performance, say you're classifying packets to apply + * Quality of Service (QoS) policies, and you don't mind about possible + * matchings spread over multiple fragments, then go BM. + */ + +#include +#include +#include +#include +#include +#include + +/* Alphabet size, use ASCII */ +#define ASIZE 256 + +#if 0 +#define DEBUGP printk +#else +#define DEBUGP(args, format...) +#endif + +struct ts_bm +{ + u8 * pattern; + unsigned int patlen; + unsigned int bad_shift[ASIZE]; + unsigned int good_shift[]; +}; + +static unsigned int matchpat(const u8 *pattern, unsigned int patlen, + const u8 *text, bool icase) +{ + unsigned int i; + + for (i = 0; i < patlen; i++) { + u8 t = *(text-i); + + if (icase) + t = toupper(t); + + if (t != *(pattern-i)) + break; + } + + return i; +} + +static unsigned int bm_find(struct ts_config *conf, struct ts_state *state) +{ + struct ts_bm *bm = ts_config_priv(conf); + unsigned int i, text_len, consumed = state->offset; + const u8 *text; + int bs; + const u8 icase = conf->flags & TS_IGNORECASE; + + for (;;) { + int shift = bm->patlen - 1; + + text_len = conf->get_next_block(consumed, &text, conf, state); + + if (unlikely(text_len == 0)) + break; + + while (shift < text_len) { + DEBUGP("Searching in position %d (%c)\n", + shift, text[shift]); + + i = matchpat(&bm->pattern[bm->patlen-1], bm->patlen, + &text[shift], icase); + if (i == bm->patlen) { + /* London calling... */ + DEBUGP("found!\n"); + return consumed + (shift-(bm->patlen-1)); + } + + bs = bm->bad_shift[text[shift-i]]; + + /* Now jumping to... */ + shift = max_t(int, shift-i+bs, shift+bm->good_shift[i]); + } + consumed += text_len; + } + + return UINT_MAX; +} + +static int subpattern(u8 *pattern, int i, int j, int g) +{ + int x = i+g-1, y = j+g-1, ret = 0; + + while(pattern[x--] == pattern[y--]) { + if (y < 0) { + ret = 1; + break; + } + if (--g == 0) { + ret = pattern[i-1] != pattern[j-1]; + break; + } + } + + return ret; +} + +static void compute_prefix_tbl(struct ts_bm *bm, int flags) +{ + int i, j, g; + + for (i = 0; i < ASIZE; i++) + bm->bad_shift[i] = bm->patlen; + for (i = 0; i < bm->patlen - 1; i++) { + bm->bad_shift[bm->pattern[i]] = bm->patlen - 1 - i; + if (flags & TS_IGNORECASE) + bm->bad_shift[tolower(bm->pattern[i])] + = bm->patlen - 1 - i; + } + + /* Compute the good shift array, used to match reocurrences + * of a subpattern */ + bm->good_shift[0] = 1; + for (i = 1; i < bm->patlen; i++) + bm->good_shift[i] = bm->patlen; + for (i = bm->patlen-1, g = 1; i > 0; g++, i--) { + for (j = i-1; j >= 1-g ; j--) + if (subpattern(bm->pattern, i, j, g)) { + bm->good_shift[g] = bm->patlen-j-g; + break; + } + } +} + +static struct ts_config *bm_init(const void *pattern, unsigned int len, + gfp_t gfp_mask, int flags) +{ + struct ts_config *conf; + struct ts_bm *bm; + int i; + unsigned int prefix_tbl_len; + size_t priv_size; + + /* Zero-length patterns would underflow bm_find()'s initial shift. */ + if (unlikely(!len)) + return ERR_PTR(-EINVAL); + + /* + * bm->pattern is stored immediately after the good_shift[] table. + * Reject lengths that would wrap while sizing either region. + */ + if (unlikely(check_mul_overflow(len, sizeof(*bm->good_shift), + &prefix_tbl_len) || + check_add_overflow(sizeof(*bm), (size_t)len, &priv_size) || + check_add_overflow(priv_size, prefix_tbl_len, &priv_size))) + return ERR_PTR(-EINVAL); + + conf = alloc_ts_config(priv_size, gfp_mask); + if (IS_ERR(conf)) + return conf; + + conf->flags = flags; + bm = ts_config_priv(conf); + bm->patlen = len; + bm->pattern = (u8 *) bm->good_shift + prefix_tbl_len; + if (flags & TS_IGNORECASE) + for (i = 0; i < len; i++) + bm->pattern[i] = toupper(((u8 *)pattern)[i]); + else + memcpy(bm->pattern, pattern, len); + compute_prefix_tbl(bm, flags); + + return conf; +} + +static void *bm_get_pattern(struct ts_config *conf) +{ + struct ts_bm *bm = ts_config_priv(conf); + return bm->pattern; +} + +static unsigned int bm_get_pattern_len(struct ts_config *conf) +{ + struct ts_bm *bm = ts_config_priv(conf); + return bm->patlen; +} + +static struct ts_ops bm_ops = { + .name = "bm", + .find = bm_find, + .init = bm_init, + .get_pattern = bm_get_pattern, + .get_pattern_len = bm_get_pattern_len, + .owner = THIS_MODULE, + .list = LIST_HEAD_INIT(bm_ops.list) +}; + +static int __init init_bm(void) +{ + return textsearch_register(&bm_ops); +} + +static void __exit exit_bm(void) +{ + textsearch_unregister(&bm_ops); +} + +MODULE_DESCRIPTION("Boyer-Moore text search implementation"); +MODULE_LICENSE("GPL"); + +module_init(init_bm); +module_exit(exit_bm); -- cgit v1.3.1