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 --- fs/dlm/ast.c | 275 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 275 insertions(+) create mode 100644 fs/dlm/ast.c (limited to 'fs/dlm/ast.c') diff --git a/fs/dlm/ast.c b/fs/dlm/ast.c new file mode 100644 index 000000000..0fe8d80ce --- /dev/null +++ b/fs/dlm/ast.c @@ -0,0 +1,275 @@ +// SPDX-License-Identifier: GPL-2.0-only +/****************************************************************************** +******************************************************************************* +** +** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. +** Copyright (C) 2004-2010 Red Hat, Inc. All rights reserved. +** +** +******************************************************************************* +******************************************************************************/ + +#include + +#include "dlm_internal.h" +#include "lvb_table.h" +#include "memory.h" +#include "lock.h" +#include "user.h" +#include "ast.h" + +static void dlm_run_callback(uint32_t ls_id, uint32_t lkb_id, int8_t mode, + uint32_t flags, uint8_t sb_flags, int sb_status, + struct dlm_lksb *lksb, + void (*astfn)(void *astparam), + void (*bastfn)(void *astparam, int mode), + void *astparam, const char *res_name, + size_t res_length) +{ + if (flags & DLM_CB_BAST) { + trace_dlm_bast(ls_id, lkb_id, mode, res_name, res_length); + bastfn(astparam, mode); + } else if (flags & DLM_CB_CAST) { + trace_dlm_ast(ls_id, lkb_id, sb_flags, sb_status, res_name, + res_length); + lksb->sb_status = sb_status; + lksb->sb_flags = sb_flags; + astfn(astparam); + } +} + +static void dlm_do_callback(struct dlm_callback *cb) +{ + dlm_run_callback(cb->ls_id, cb->lkb_id, cb->mode, cb->flags, + cb->sb_flags, cb->sb_status, cb->lkb_lksb, + cb->astfn, cb->bastfn, cb->astparam, + cb->res_name, cb->res_length); + dlm_free_cb(cb); +} + +static void dlm_callback_work(struct work_struct *work) +{ + struct dlm_callback *cb = container_of(work, struct dlm_callback, work); + + dlm_do_callback(cb); +} + +bool dlm_may_skip_callback(struct dlm_lkb *lkb, uint32_t flags, int mode, + int status, uint32_t sbflags, int *copy_lvb) +{ + struct dlm_rsb *rsb = lkb->lkb_resource; + struct dlm_ls *ls = rsb->res_ls; + int prev_mode; + + if (copy_lvb) + *copy_lvb = 0; + + if (flags & DLM_CB_BAST) { + /* if cb is a bast, it should be skipped if the blocking mode is + * compatible with the last granted mode + */ + if (lkb->lkb_last_cast_cb_mode != -1) { + if (dlm_modes_compat(mode, lkb->lkb_last_cast_cb_mode)) { + log_debug(ls, "skip %x bast mode %d for cast mode %d", + lkb->lkb_id, mode, + lkb->lkb_last_cast_cb_mode); + return true; + } + } + + /* + * Suppress some redundant basts here, do more on removal. + * Don't even add a bast if the callback just before it + * is a bast for the same mode or a more restrictive mode. + * (the addional > PR check is needed for PR/CW inversion) + */ + if (lkb->lkb_last_cb_mode != -1 && + lkb->lkb_last_cb_flags & DLM_CB_BAST) { + prev_mode = lkb->lkb_last_cb_mode; + + if ((prev_mode == mode) || + (prev_mode > mode && prev_mode > DLM_LOCK_PR)) { + log_debug(ls, "skip %x add bast mode %d for bast mode %d", + lkb->lkb_id, mode, prev_mode); + return true; + } + } + + lkb->lkb_last_bast_time = ktime_get(); + lkb->lkb_last_bast_cb_mode = mode; + } else if (flags & DLM_CB_CAST) { + if (test_bit(DLM_DFL_USER_BIT, &lkb->lkb_dflags)) { + prev_mode = lkb->lkb_last_cast_cb_mode; + + if (!status && lkb->lkb_lksb->sb_lvbptr && + dlm_lvb_operations[prev_mode + 1][mode + 1]) { + if (copy_lvb) + *copy_lvb = 1; + } + } + + lkb->lkb_last_cast_cb_mode = mode; + lkb->lkb_last_cast_time = ktime_get(); + } + + lkb->lkb_last_cb_mode = mode; + lkb->lkb_last_cb_flags = flags; + + return false; +} + +int dlm_get_cb(struct dlm_lkb *lkb, uint32_t flags, int mode, + int status, uint32_t sbflags, + struct dlm_callback **cb) +{ + struct dlm_rsb *rsb = lkb->lkb_resource; + struct dlm_ls *ls = rsb->res_ls; + + *cb = dlm_allocate_cb(); + if (WARN_ON_ONCE(!*cb)) + return -ENOMEM; + + /* for tracing */ + (*cb)->lkb_id = lkb->lkb_id; + (*cb)->ls_id = ls->ls_global_id; + memcpy((*cb)->res_name, rsb->res_name, rsb->res_length); + (*cb)->res_length = rsb->res_length; + + (*cb)->flags = flags; + (*cb)->mode = mode; + (*cb)->sb_status = status; + (*cb)->sb_flags = (sbflags & 0x000000FF); + (*cb)->lkb_lksb = lkb->lkb_lksb; + + return 0; +} + +static int dlm_get_queue_cb(struct dlm_lkb *lkb, uint32_t flags, int mode, + int status, uint32_t sbflags, + struct dlm_callback **cb) +{ + int rv; + + rv = dlm_get_cb(lkb, flags, mode, status, sbflags, cb); + if (rv) + return rv; + + (*cb)->astfn = lkb->lkb_astfn; + (*cb)->bastfn = lkb->lkb_bastfn; + (*cb)->astparam = lkb->lkb_astparam; + INIT_WORK(&(*cb)->work, dlm_callback_work); + + return 0; +} + +void dlm_add_cb(struct dlm_lkb *lkb, uint32_t flags, int mode, int status, + uint32_t sbflags) +{ + struct dlm_rsb *rsb = lkb->lkb_resource; + struct dlm_ls *ls = rsb->res_ls; + struct dlm_callback *cb; + int rv; + + if (test_bit(DLM_DFL_USER_BIT, &lkb->lkb_dflags)) { + dlm_user_add_ast(lkb, flags, mode, status, sbflags); + return; + } + + if (dlm_may_skip_callback(lkb, flags, mode, status, sbflags, NULL)) + return; + + spin_lock_bh(&ls->ls_cb_lock); + if (test_bit(LSFL_CB_DELAY, &ls->ls_flags)) { + rv = dlm_get_queue_cb(lkb, flags, mode, status, sbflags, &cb); + if (!rv) + list_add(&cb->list, &ls->ls_cb_delay); + } else { + if (test_bit(LSFL_SOFTIRQ, &ls->ls_flags)) { + dlm_run_callback(ls->ls_global_id, lkb->lkb_id, mode, flags, + sbflags, status, lkb->lkb_lksb, + lkb->lkb_astfn, lkb->lkb_bastfn, + lkb->lkb_astparam, rsb->res_name, + rsb->res_length); + } else { + rv = dlm_get_queue_cb(lkb, flags, mode, status, sbflags, &cb); + if (!rv) + queue_work(ls->ls_callback_wq, &cb->work); + } + } + spin_unlock_bh(&ls->ls_cb_lock); +} + +int dlm_callback_start(struct dlm_ls *ls) +{ + if (!test_bit(LSFL_FS, &ls->ls_flags) || + test_bit(LSFL_SOFTIRQ, &ls->ls_flags)) + return 0; + + ls->ls_callback_wq = alloc_ordered_workqueue("dlm_callback", + WQ_HIGHPRI | WQ_MEM_RECLAIM); + if (!ls->ls_callback_wq) { + log_print("can't start dlm_callback workqueue"); + return -ENOMEM; + } + return 0; +} + +void dlm_callback_stop(struct dlm_ls *ls) +{ + if (ls->ls_callback_wq) + destroy_workqueue(ls->ls_callback_wq); +} + +void dlm_callback_suspend(struct dlm_ls *ls) +{ + if (!test_bit(LSFL_FS, &ls->ls_flags)) + return; + + spin_lock_bh(&ls->ls_cb_lock); + set_bit(LSFL_CB_DELAY, &ls->ls_flags); + spin_unlock_bh(&ls->ls_cb_lock); + + if (ls->ls_callback_wq) + flush_workqueue(ls->ls_callback_wq); +} + +#define MAX_CB_QUEUE 25 + +void dlm_callback_resume(struct dlm_ls *ls) +{ + struct dlm_callback *cb, *safe; + int count = 0, sum = 0; + bool empty; + + if (!test_bit(LSFL_FS, &ls->ls_flags)) + return; + +more: + spin_lock_bh(&ls->ls_cb_lock); + list_for_each_entry_safe(cb, safe, &ls->ls_cb_delay, list) { + list_del(&cb->list); + if (test_bit(LSFL_SOFTIRQ, &ls->ls_flags)) + dlm_do_callback(cb); + else + queue_work(ls->ls_callback_wq, &cb->work); + + count++; + if (count == MAX_CB_QUEUE) + break; + } + empty = list_empty(&ls->ls_cb_delay); + if (empty) + clear_bit(LSFL_CB_DELAY, &ls->ls_flags); + spin_unlock_bh(&ls->ls_cb_lock); + + sum += count; + if (!empty) { + count = 0; + cond_resched(); + goto more; + } + + if (sum) + log_rinfo(ls, "%s %d", __func__, sum); +} + -- cgit v1.3.1