linux-kernel - Re: [PATCH v3 1/2] mm/vmalloc: Add a safer version of find_vm

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <20230908004717.GC4088026@google.com>
Date:   Fri, 8 Sep 2023 00:47:17 +0000
From:   Joel Fernandes <joel@...lfernandes.org>
To:     Vlastimil Babka <vbabka@...e.cz>,
        Andrew Morton <akpm@...ux-foundation.org>
Cc:     linux-kernel@...r.kernel.org, Uladzislau Rezki <urezki@...il.com>,
        Christoph Hellwig <hch@...radead.org>,
        Lorenzo Stoakes <lstoakes@...il.com>,
        "Paul E. McKenney" <paulmck@...nel.org>,
        Zhen Lei <thunder.leizhen@...weicloud.com>,
        rcu@...r.kernel.org, Zqiang <qiang.zhang1211@...il.com>,
        stable@...r.kernel.org, linux-mm@...ck.org
Subject: Re: [PATCH v3 1/2] mm/vmalloc: Add a safer version of find_vm_area()
 for debug

On Thu, Sep 07, 2023 at 08:53:14AM +0200, Vlastimil Babka wrote:
> Hi,
> 
> On 9/4/23 20:08, Joel Fernandes (Google) wrote:
> > It is unsafe to dump vmalloc area information when trying to do so from
> > some contexts. Add a safer trylock version of the same function to do a
> > best-effort VMA finding and use it from vmalloc_dump_obj().
> 
> I was a bit confused by the subject which suggests a new function is added,
> but it seems open-coded in its only caller. I assume it's due to evolution
> of the series. Something like:
> 
> mm/vmalloc: use trylock for vmap_area_lock in vmalloc_dump_obj()
> 
> ?
> 
> I also notice it's trying hard to copy everything from "vm" to temporary
> variables before unlocking, presumably to prevent use-after-free, so should
> that be also mentioned in the changelog?

Apologies for the less-than-ideal changelog. Andrew would you mind replacing
the merged patch with the below one instead? It just contains non-functional
changes to change log and an additional code comment/print. Thanks!

---8<-----------------------

From: "Joel Fernandes (Google)" <joel@...lfernandes.org>
Subject: [PATCH v3.1] mm/vmalloc: Add a safer inlined version of find_vm_area() for
 debug

It is unsafe to dump vmalloc area information when trying to do so from
some contexts such as PREEMPT_RT or from an IRQ handler that interrupted
a vmap_area_lock-held region. Add a safer and inlined trylock version of
find_vm_area() to do a best-effort VMA finding and use it from
vmalloc_dump_obj().

While the vmap_area_lock is held, copy interesting attributes from the
vm_struct before unlocking.

[applied test robot feedback on unused function fix.]
[applied Uladzislau feedback on locking.]
[applied Vlastimil and Lorenzo feedback on changelog, comment and print
improvements]

Reported-by: Zhen Lei <thunder.leizhen@...weicloud.com>
Cc: Paul E. McKenney <paulmck@...nel.org>
Cc: rcu@...r.kernel.org
Cc: Zqiang <qiang.zhang1211@...il.com>
Reviewed-by: Uladzislau Rezki (Sony) <urezki@...il.com>
Fixes: 98f180837a89 ("mm: Make mem_dump_obj() handle vmalloc() memory")
Cc: stable@...r.kernel.org
Signed-off-by: Joel Fernandes (Google) <joel@...lfernandes.org>
---
 mm/vmalloc.c | 34 ++++++++++++++++++++++++++++++----
 1 file changed, 30 insertions(+), 4 deletions(-)

diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 93cf99aba335..990a0d5efba8 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -4274,14 +4274,40 @@ void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
 #ifdef CONFIG_PRINTK
 bool vmalloc_dump_obj(void *object)
 {
-	struct vm_struct *vm;
 	void *objp = (void *)PAGE_ALIGN((unsigned long)object);
+	const void *caller;
+	struct vm_struct *vm;
+	struct vmap_area *va;
+	unsigned long addr;
+	unsigned int nr_pages;
+
+	/*
+	 * Use trylock as we don't want to contend since this is debug code and
+	 * we might run this code in contexts like PREEMPT_RT where spinlock
+	 * contention may result in sleeping, or from an IRQ handler which
+	 * might interrupt a vmap_area_lock-held critical section.
+	 */
+	if (!spin_trylock(&vmap_area_lock)) {
+		pr_cont(" [couldn't acquire vmap_area_lock]\n");
+		return false;
+	}
+	va = __find_vmap_area((unsigned long)objp, &vmap_area_root);
+	if (!va) {
+		spin_unlock(&vmap_area_lock);
+		return false;
+	}
 
-	vm = find_vm_area(objp);
-	if (!vm)
+	vm = va->vm;
+	if (!vm) {
+		spin_unlock(&vmap_area_lock);
 		return false;
+	}
+	addr = (unsigned long)vm->addr;
+	caller = vm->caller;
+	nr_pages = vm->nr_pages;
+	spin_unlock(&vmap_area_lock);
 	pr_cont(" %u-page vmalloc region starting at %#lx allocated at %pS\n",
-		vm->nr_pages, (unsigned long)vm->addr, vm->caller);
+		nr_pages, addr, caller);
 	return true;
 }
 #endif
-- 
2.42.0.283.g2d96d420d3-goog