Re: [External] Re: [PATCH] fs/proc/kcore.c: add mmap interface

[zhoufeng <zhoufeng.zf@bytedance.com>]

在 2021/6/1 上午9:23, Andrew Morton 写道:
> On Wed, 26 May 2021 15:51:42 +0800 Feng zhou <zhoufeng.zf@bytedance.com> wrote:
> 
>> From: ZHOUFENG <zhoufeng.zf@bytedance.com>
>>
>> When we do the kernel monitor, use the DRGN
>> (https://github.com/osandov/drgn) access to kernel data structures,
>> found that the system calls a lot. DRGN is implemented by reading
>> /proc/kcore. After looking at the kcore code, it is found that kcore
>> does not implement mmap, resulting in frequent context switching
>> triggered by read. Therefore, we want to add mmap interface to optimize
>> performance. Since vmalloc and module areas will change with allocation
>> and release, consistency cannot be guaranteed, so mmap interface only
>> maps KCORE_TEXT and KCORE_RAM.
>>
>> ...
>>
>> --- a/fs/proc/kcore.c
>> +++ b/fs/proc/kcore.c
>> @@ -573,11 +573,81 @@ static int release_kcore(struct inode *inode, struct file *file)
>>   	return 0;
>>   }
>>   
>> +static vm_fault_t mmap_kcore_fault(struct vm_fault *vmf)
>> +{
>> +	return VM_FAULT_SIGBUS;
>> +}
>> +
>> +static const struct vm_operations_struct kcore_mmap_ops = {
>> +	.fault = mmap_kcore_fault,
>> +};
>> +
>> +static int mmap_kcore(struct file *file, struct vm_area_struct *vma)
>> +{
>> +	size_t size = vma->vm_end - vma->vm_start;
>> +	u64 start, pfn;
>> +	int nphdr;
>> +	size_t data_offset;
>> +	size_t phdrs_len, notes_len;
>> +	struct kcore_list *m = NULL;
>> +	int ret = 0;
>> +
>> +	down_read(&kclist_lock);
>> +
>> +	get_kcore_size(&nphdr, &phdrs_len, &notes_len, &data_offset);
>> +
>> +	start = kc_offset_to_vaddr(((u64)vma->vm_pgoff << PAGE_SHIFT) -
>> +		((data_offset >> PAGE_SHIFT) << PAGE_SHIFT));
>> +
>> +	list_for_each_entry(m, &kclist_head, list) {
>> +		if (start >= m->addr && size <= m->size)
>> +			break;
>> +	}
>> +
>> +	if (&m->list == &kclist_head) {
>> +		ret = -EINVAL;
>> +		goto out;
>> +	}
>> +
>> +	if (vma->vm_flags & (VM_WRITE | VM_EXEC)) {
>> +		ret = -EPERM;
>> +		goto out;
>> +	}
>> +
>> +	vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
>> +	vma->vm_flags |= VM_MIXEDMAP;
>> +	vma->vm_ops = &kcore_mmap_ops;
>> +
>> +	if (kern_addr_valid(start)) {
>> +		if (m->type == KCORE_RAM || m->type == KCORE_REMAP)
>> +			pfn = __pa(start) >> PAGE_SHIFT;
>> +		else if (m->type == KCORE_TEXT)
>> +			pfn = __pa_symbol(start) >> PAGE_SHIFT;
>> +		else {
>> +			ret = -EFAULT;
>> +			goto out;
>> +		}
>> +
>> +		if (remap_pfn_range(vma, vma->vm_start, pfn, size,
>> +				vma->vm_page_prot)) {
>> +			ret = -EAGAIN;
> 
> EAGAIN seems a strange errno for this case.   The mmap manpage says
> 
>         EAGAIN The file has been locked, or too much  memory  has  been  locked
>                (see setrlimit(2)).
> 
> 
> remap_pfn_range() already returns an errno - why not return whatever
> that code was?
> 

   yes, that's a good idea.

>> +			goto out;
>> +		}
>> +	} else {
>> +		ret = -EFAULT;
>> +	}
>> +
>> +out:
>> +	up_read(&kclist_lock);
>> +	return ret;
>> +}
>> +
>>   static const struct proc_ops kcore_proc_ops = {
>>   	.proc_read	= read_kcore,
>>   	.proc_open	= open_kcore,
>>   	.proc_release	= release_kcore,
>>   	.proc_lseek	= default_llseek,
>> +	.proc_mmap	= mmap_kcore,
>>   };
>>   
>>   /* just remember that we have to update kcore */
> 
> Otherwise looks OK to me.  Please update the changelog to reflect the
> discussion thus far and send a v2?
> 
   OK, I am very happy to do so, I will send a v2 in two days.