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 --- crypto/simd.c | 264 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 264 insertions(+) create mode 100644 crypto/simd.c (limited to 'crypto/simd.c') diff --git a/crypto/simd.c b/crypto/simd.c new file mode 100644 index 000000000..4e29f7977 --- /dev/null +++ b/crypto/simd.c @@ -0,0 +1,264 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * Shared crypto simd helpers + * + * Copyright (c) 2012 Jussi Kivilinna + * Copyright (c) 2016 Herbert Xu + * Copyright (c) 2019 Google LLC + * + * Based on aesni-intel_glue.c by: + * Copyright (C) 2008, Intel Corp. + * Author: Huang Ying + */ + +/* + * Shared crypto SIMD helpers. These functions dynamically create and register + * an AEAD algorithm that wraps another, internal algorithm. The wrapper + * ensures that the internal algorithm is only executed in a context where SIMD + * instructions are usable, i.e. where may_use_simd() returns true. If SIMD is + * already usable, the wrapper directly calls the internal algorithm. Otherwise + * it defers execution to a workqueue via cryptd. + * + * This is an alternative to the internal algorithm implementing a fallback for + * the !may_use_simd() case itself. + * + * Note that the wrapper algorithm is asynchronous, i.e. it has the + * CRYPTO_ALG_ASYNC flag set. Therefore it won't be found by users who + * explicitly allocate a synchronous algorithm. + */ + +#include +#include +#include +#include +#include +#include +#include + +struct simd_aead_alg { + const char *ialg_name; + struct aead_alg alg; +}; + +struct simd_aead_ctx { + struct cryptd_aead *cryptd_tfm; +}; + +static int simd_aead_setkey(struct crypto_aead *tfm, const u8 *key, + unsigned int key_len) +{ + struct simd_aead_ctx *ctx = crypto_aead_ctx(tfm); + struct crypto_aead *child = &ctx->cryptd_tfm->base; + + crypto_aead_clear_flags(child, CRYPTO_TFM_REQ_MASK); + crypto_aead_set_flags(child, crypto_aead_get_flags(tfm) & + CRYPTO_TFM_REQ_MASK); + return crypto_aead_setkey(child, key, key_len); +} + +static int simd_aead_setauthsize(struct crypto_aead *tfm, unsigned int authsize) +{ + struct simd_aead_ctx *ctx = crypto_aead_ctx(tfm); + struct crypto_aead *child = &ctx->cryptd_tfm->base; + + return crypto_aead_setauthsize(child, authsize); +} + +static int simd_aead_encrypt(struct aead_request *req) +{ + struct crypto_aead *tfm = crypto_aead_reqtfm(req); + struct simd_aead_ctx *ctx = crypto_aead_ctx(tfm); + struct aead_request *subreq; + struct crypto_aead *child; + + subreq = aead_request_ctx(req); + *subreq = *req; + + if (!crypto_simd_usable() || + (in_atomic() && cryptd_aead_queued(ctx->cryptd_tfm))) + child = &ctx->cryptd_tfm->base; + else + child = cryptd_aead_child(ctx->cryptd_tfm); + + aead_request_set_tfm(subreq, child); + + return crypto_aead_encrypt(subreq); +} + +static int simd_aead_decrypt(struct aead_request *req) +{ + struct crypto_aead *tfm = crypto_aead_reqtfm(req); + struct simd_aead_ctx *ctx = crypto_aead_ctx(tfm); + struct aead_request *subreq; + struct crypto_aead *child; + + subreq = aead_request_ctx(req); + *subreq = *req; + + if (!crypto_simd_usable() || + (in_atomic() && cryptd_aead_queued(ctx->cryptd_tfm))) + child = &ctx->cryptd_tfm->base; + else + child = cryptd_aead_child(ctx->cryptd_tfm); + + aead_request_set_tfm(subreq, child); + + return crypto_aead_decrypt(subreq); +} + +static void simd_aead_exit(struct crypto_aead *tfm) +{ + struct simd_aead_ctx *ctx = crypto_aead_ctx(tfm); + + cryptd_free_aead(ctx->cryptd_tfm); +} + +static int simd_aead_init(struct crypto_aead *tfm) +{ + struct simd_aead_ctx *ctx = crypto_aead_ctx(tfm); + struct cryptd_aead *cryptd_tfm; + struct simd_aead_alg *salg; + struct aead_alg *alg; + unsigned reqsize; + + alg = crypto_aead_alg(tfm); + salg = container_of(alg, struct simd_aead_alg, alg); + + cryptd_tfm = cryptd_alloc_aead(salg->ialg_name, CRYPTO_ALG_INTERNAL, + CRYPTO_ALG_INTERNAL); + if (IS_ERR(cryptd_tfm)) + return PTR_ERR(cryptd_tfm); + + ctx->cryptd_tfm = cryptd_tfm; + + reqsize = max(crypto_aead_reqsize(cryptd_aead_child(cryptd_tfm)), + crypto_aead_reqsize(&cryptd_tfm->base)); + reqsize += sizeof(struct aead_request); + + crypto_aead_set_reqsize(tfm, reqsize); + + return 0; +} + +static struct simd_aead_alg *simd_aead_create_compat(struct aead_alg *ialg, + const char *algname, + const char *drvname, + const char *basename) +{ + struct simd_aead_alg *salg; + struct aead_alg *alg; + int err; + + salg = kzalloc_obj(*salg); + if (!salg) { + salg = ERR_PTR(-ENOMEM); + goto out; + } + + salg->ialg_name = basename; + alg = &salg->alg; + + err = -ENAMETOOLONG; + if (snprintf(alg->base.cra_name, CRYPTO_MAX_ALG_NAME, "%s", algname) >= + CRYPTO_MAX_ALG_NAME) + goto out_free_salg; + + if (snprintf(alg->base.cra_driver_name, CRYPTO_MAX_ALG_NAME, "%s", + drvname) >= CRYPTO_MAX_ALG_NAME) + goto out_free_salg; + + alg->base.cra_flags = CRYPTO_ALG_ASYNC | + (ialg->base.cra_flags & CRYPTO_ALG_INHERITED_FLAGS); + alg->base.cra_priority = ialg->base.cra_priority; + alg->base.cra_blocksize = ialg->base.cra_blocksize; + alg->base.cra_alignmask = ialg->base.cra_alignmask; + alg->base.cra_module = ialg->base.cra_module; + alg->base.cra_ctxsize = sizeof(struct simd_aead_ctx); + + alg->ivsize = ialg->ivsize; + alg->maxauthsize = ialg->maxauthsize; + alg->chunksize = ialg->chunksize; + + alg->init = simd_aead_init; + alg->exit = simd_aead_exit; + + alg->setkey = simd_aead_setkey; + alg->setauthsize = simd_aead_setauthsize; + alg->encrypt = simd_aead_encrypt; + alg->decrypt = simd_aead_decrypt; + + err = crypto_register_aead(alg); + if (err) + goto out_free_salg; + +out: + return salg; + +out_free_salg: + kfree(salg); + salg = ERR_PTR(err); + goto out; +} + +static void simd_aead_free(struct simd_aead_alg *salg) +{ + crypto_unregister_aead(&salg->alg); + kfree(salg); +} + +int simd_register_aeads_compat(struct aead_alg *algs, int count, + struct simd_aead_alg **simd_algs) +{ + int err; + int i; + const char *algname; + const char *drvname; + const char *basename; + struct simd_aead_alg *simd; + + for (i = 0; i < count; i++) { + if (WARN_ON(strncmp(algs[i].base.cra_name, "__", 2) || + strncmp(algs[i].base.cra_driver_name, "__", 2))) + return -EINVAL; + } + + err = crypto_register_aeads(algs, count); + if (err) + return err; + + for (i = 0; i < count; i++) { + algname = algs[i].base.cra_name + 2; + drvname = algs[i].base.cra_driver_name + 2; + basename = algs[i].base.cra_driver_name; + simd = simd_aead_create_compat(algs + i, algname, drvname, basename); + err = PTR_ERR(simd); + if (IS_ERR(simd)) + goto err_unregister; + simd_algs[i] = simd; + } + return 0; + +err_unregister: + simd_unregister_aeads(algs, count, simd_algs); + return err; +} +EXPORT_SYMBOL_GPL(simd_register_aeads_compat); + +void simd_unregister_aeads(struct aead_alg *algs, int count, + struct simd_aead_alg **simd_algs) +{ + int i; + + crypto_unregister_aeads(algs, count); + + for (i = 0; i < count; i++) { + if (simd_algs[i]) { + simd_aead_free(simd_algs[i]); + simd_algs[i] = NULL; + } + } +} +EXPORT_SYMBOL_GPL(simd_unregister_aeads); + +MODULE_DESCRIPTION("Shared crypto SIMD helpers"); +MODULE_LICENSE("GPL"); -- cgit v1.3.1