Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 23 Oct 2019 20:05:49 +0200 (CEST)
From:   Thomas Gleixner <tglx@linutronix.de>
To:     Cyrill Gorcunov <gorcunov@gmail.com>
cc:     LKML <linux-kernel@vger.kernel.org>,
        Ingo Molnar <mingo@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>, linux-mm@kvack.org,
        Catalin Marinas <catalin.marinas@arm.com>, x86@kernel.org,
        Josh Poimboeuf <jpoimboe@redhat.com>
Subject: [PATCH] x86/dumpstack/64: Don't evaluate exception stacks before
 setup
In-Reply-To: <20191023135943.GK12121@uranus.lan>
Message-ID: <alpine.DEB.2.21.1910231950590.1852@nanos.tec.linutronix.de>
References: <20191019114421.GK9698@uranus.lan> <20191022142325.GD12121@uranus.lan> <20191022145619.GE12121@uranus.lan> <alpine.DEB.2.21.1910231457400.2308@nanos.tec.linutronix.de> <alpine.DEB.2.21.1910231533180.2308@nanos.tec.linutronix.de>
 <20191023135943.GK12121@uranus.lan>
User-Agent: Alpine 2.21 (DEB 202 2017-01-01)
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Cyrill reported the following crash:

  BUG: unable to handle page fault for address: 0000000000001ff0
  #PF: supervisor read access in kernel mode
  RIP: 0010:get_stack_info+0xb3/0x148

It turns out that if the stack tracer is invoked before the exception stack
mappings are initialized in_exception_stack() can erroneously classify an
invalid address as an address inside of an exception stack:

    begin = this_cpu_read(cea_exception_stacks);  <- 0
    end = begin + sizeof(exception stacks);

i.e. any address between 0 and end will be considered as exception stack
address and the subsequent code will then try to derefence the resulting
stack frame at a non mapped address.

 end = begin + (unsigned long)ep->size;
     ==> end = 0x2000

 regs = (struct pt_regs *)end - 1;
     ==> regs = 0x2000 - sizeof(struct pt_regs *) = 0x1ff0

 info->next_sp   = (unsigned long *)regs->sp;
     ==> Crashes due to accessing 0x1ff0

Prevent this by checking the validity of the cea_exception_stack base
address and bailing out if it is zero.

Fixes: afcd21dad88b ("x86/dumpstack/64: Use cpu_entry_area instead of orig_ist")
Reported-by: Cyrill Gorcunov <gorcunov@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: stable@vger.kernel.org
---
 arch/x86/kernel/dumpstack_64.c |    7 +++++++
 1 file changed, 7 insertions(+)

--- a/arch/x86/kernel/dumpstack_64.c
+++ b/arch/x86/kernel/dumpstack_64.c
@@ -94,6 +94,13 @@ static bool in_exception_stack(unsigned
 	BUILD_BUG_ON(N_EXCEPTION_STACKS != 6);
 
 	begin = (unsigned long)__this_cpu_read(cea_exception_stacks);
+	/*
+	 * Handle the case where stack trace is collected _before_
+	 * cea_exception_stacks had been initialized.
+	 */
+	if (!begin)
+		return false;
+
 	end = begin + sizeof(struct cea_exception_stacks);
 	/* Bail if @stack is outside the exception stack area. */
 	if (stk < begin || stk >= end)