| 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
| ||
|
Message-ID: <4eb3fffa-8330-ad54-8cbc-2cabf6355c74@loongson.cn>
Date: Tue, 2 Sep 2025 20:15:11 +0800
From: Bibo Mao <maobibo@...ngson.cn>
To: Huacai Chen <chenhuacai@...nel.org>
Cc: Xianglai Li <lixianglai@...ngson.cn>, WANG Xuerui <kernel@...0n.name>,
kvm@...r.kernel.org, loongarch@...ts.linux.dev, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 1/4] LoongArch: KVM: Avoid use copy_from_user with lock
hold in kvm_eiointc_regs_access
On 2025/9/2 下午7:58, Huacai Chen wrote:
> Hi, Bibo,
>
> On Tue, Sep 2, 2025 at 5:49 PM Bibo Mao <maobibo@...ngson.cn> wrote:
>>
>> Function copy_from_user() and copy_to_user() may sleep because of page
>> fault, and they cannot be called in spin_lock hold context. Otherwise there
>> will be possible warning such as:
>>
>> BUG: sleeping function called from invalid context at include/linux/uaccess.h:192
>> in_atomic(): 1, irqs_disabled(): 1, non_block: 0, pid: 6292, name: qemu-system-loo
>> preempt_count: 1, expected: 0
>> RCU nest depth: 0, expected: 0
>> INFO: lockdep is turned off.
>> irq event stamp: 0
>> hardirqs last enabled at (0): [<0000000000000000>] 0x0
>> hardirqs last disabled at (0): [<9000000004c4a554>] copy_process+0x90c/0x1d40
>> softirqs last enabled at (0): [<9000000004c4a554>] copy_process+0x90c/0x1d40
>> softirqs last disabled at (0): [<0000000000000000>] 0x0
>> CPU: 41 UID: 0 PID: 6292 Comm: qemu-system-loo Tainted: G W 6.17.0-rc3+ #31 PREEMPT(full)
>> Tainted: [W]=WARN
>> Stack : 0000000000000076 0000000000000000 9000000004c28264 9000100092ff4000
>> 9000100092ff7b80 9000100092ff7b88 0000000000000000 9000100092ff7cc8
>> 9000100092ff7cc0 9000100092ff7cc0 9000100092ff7a00 0000000000000001
>> 0000000000000001 9000100092ff7b88 947d2f9216a5e8b9 900010008773d880
>> 00000000ffff8b9f fffffffffffffffe 0000000000000ba1 fffffffffffffffe
>> 000000000000003e 900000000825a15b 000010007ad38000 9000100092ff7ec0
>> 0000000000000000 0000000000000000 9000000006f3ac60 9000000007252000
>> 0000000000000000 00007ff746ff2230 0000000000000053 9000200088a021b0
>> 0000555556c9d190 0000000000000000 9000000004c2827c 000055556cfb5f40
>> 00000000000000b0 0000000000000007 0000000000000007 0000000000071c1d
>> Call Trace:
>> [<9000000004c2827c>] show_stack+0x5c/0x180
>> [<9000000004c20fac>] dump_stack_lvl+0x94/0xe4
>> [<9000000004c99c7c>] __might_resched+0x26c/0x290
>> [<9000000004f68968>] __might_fault+0x20/0x88
>> [<ffff800002311de0>] kvm_eiointc_regs_access.isra.0+0x88/0x380 [kvm]
>> [<ffff8000022f8514>] kvm_device_ioctl+0x194/0x290 [kvm]
>> [<900000000506b0d8>] sys_ioctl+0x388/0x1010
>> [<90000000063ed210>] do_syscall+0xb0/0x2d8
>> [<9000000004c25ef8>] handle_syscall+0xb8/0x158
>>
>> Fixes: 1ad7efa552fd5 ("LoongArch: KVM: Add EIOINTC user mode read and write functions")
>> Signed-off-by: Bibo Mao <maobibo@...ngson.cn>
>> ---
>> arch/loongarch/kvm/intc/eiointc.c | 33 ++++++++++++++++++++-----------
>> 1 file changed, 21 insertions(+), 12 deletions(-)
>>
>> diff --git a/arch/loongarch/kvm/intc/eiointc.c b/arch/loongarch/kvm/intc/eiointc.c
>> index 026b139dcff2..2fb5b9c6e8ad 100644
>> --- a/arch/loongarch/kvm/intc/eiointc.c
>> +++ b/arch/loongarch/kvm/intc/eiointc.c
>> @@ -462,19 +462,17 @@ static int kvm_eiointc_ctrl_access(struct kvm_device *dev,
>>
>> static int kvm_eiointc_regs_access(struct kvm_device *dev,
>> struct kvm_device_attr *attr,
>> - bool is_write)
>> + bool is_write, int *data)
>> {
>> int addr, cpu, offset, ret = 0;
>> unsigned long flags;
>> void *p = NULL;
>> - void __user *data;
>> struct loongarch_eiointc *s;
>>
>> s = dev->kvm->arch.eiointc;
>> addr = attr->attr;
>> cpu = addr >> 16;
>> addr &= 0xffff;
>> - data = (void __user *)attr->addr;
>> switch (addr) {
>> case EIOINTC_NODETYPE_START ... EIOINTC_NODETYPE_END:
>> offset = (addr - EIOINTC_NODETYPE_START) / 4;
>> @@ -513,13 +511,10 @@ static int kvm_eiointc_regs_access(struct kvm_device *dev,
>> }
>>
>> spin_lock_irqsave(&s->lock, flags);
>> - if (is_write) {
>> - if (copy_from_user(p, data, 4))
>> - ret = -EFAULT;
>> - } else {
>> - if (copy_to_user(data, p, 4))
>> - ret = -EFAULT;
>> - }
>> + if (is_write)
>> + memcpy(p, data, 4);
>> + else
>> + memcpy(data, p, 4);
> p is a local variable, data is a parameter, they both have nothing to
> do with s, why memcpy need to be protected?
p is pointer to register buffer rather than local variable. When dump
extioi register to user space, maybe one vCPU is writing extioi register
at the same time, so there needs spinlock protection.
>
> After some thinking I found the code was wrong at the first time. The
> real code that needs to be protected is not copy_from_user() or
> memcpy(), but the above switch block.
For switch block in function kvm_eiointc_regs_access() for example, it
is only to get register buffer pointer, not register content. Spinlock
protection is not necessary in switch block.
Regards
Bibo Mao
>
> Other patches have similar problems.
>
> Huacai
>
>> spin_unlock_irqrestore(&s->lock, flags);
>>
>> return ret;
>> @@ -576,9 +571,18 @@ static int kvm_eiointc_sw_status_access(struct kvm_device *dev,
>> static int kvm_eiointc_get_attr(struct kvm_device *dev,
>> struct kvm_device_attr *attr)
>> {
>> + int ret, data;
>> +
>> switch (attr->group) {
>> case KVM_DEV_LOONGARCH_EXTIOI_GRP_REGS:
>> - return kvm_eiointc_regs_access(dev, attr, false);
>> + ret = kvm_eiointc_regs_access(dev, attr, false, &data);
>> + if (ret)
>> + return ret;
>> +
>> + if (copy_to_user((void __user *)attr->addr, &data, 4))
>> + ret = -EFAULT;
>> +
>> + return ret;
>> case KVM_DEV_LOONGARCH_EXTIOI_GRP_SW_STATUS:
>> return kvm_eiointc_sw_status_access(dev, attr, false);
>> default:
>> @@ -589,11 +593,16 @@ static int kvm_eiointc_get_attr(struct kvm_device *dev,
>> static int kvm_eiointc_set_attr(struct kvm_device *dev,
>> struct kvm_device_attr *attr)
>> {
>> + int data;
>> +
>> switch (attr->group) {
>> case KVM_DEV_LOONGARCH_EXTIOI_GRP_CTRL:
>> return kvm_eiointc_ctrl_access(dev, attr);
>> case KVM_DEV_LOONGARCH_EXTIOI_GRP_REGS:
>> - return kvm_eiointc_regs_access(dev, attr, true);
>> + if (copy_from_user(&data, (void __user *)attr->addr, 4))
>> + return -EFAULT;
>> +
>> + return kvm_eiointc_regs_access(dev, attr, true, &data);
>> case KVM_DEV_LOONGARCH_EXTIOI_GRP_SW_STATUS:
>> return kvm_eiointc_sw_status_access(dev, attr, true);
>> default:
>> --
>> 2.39.3
>>
Powered by blists - more mailing lists