Received-SPF: pass (google.com: domain of linux-kernel+bounces-57486-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Message-ID: <tencent_1D3AC867D2233D8E19C8CFF3B9A8AA893A05@qq.com>
Date: Thu, 8 Feb 2024 13:34:45 +0800
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.11.0
Subject: Re: [PATCH] eventfd: strictly check the count parameter of
 eventfd_write to avoid inputting illegal strings
To: Eric Biggers <ebiggers@kernel.org>
Cc: Christian Brauner <brauner@kernel.org>, Jens Axboe <axboe@kernel.dk>,
 Alexander Viro <viro@zeniv.linux.org.uk>, Jan Kara <jack@suse.cz>,
 David Woodhouse <dwmw@amazon.co.uk>, Matthew Wilcox <willy@infradead.org>,
 linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
References: <tencent_10AAA44731FFFA493F9F5501521F07DD4D0A@qq.com>
 <20240208043354.GA85799@sol.localdomain>
Content-Language: en-US
From: Wen Yang <wenyang.linux@foxmail.com>
In-Reply-To: <20240208043354.GA85799@sol.localdomain>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit


On 2024/2/8 12:33, Eric Biggers wrote:
> On Wed, Feb 07, 2024 at 12:35:18AM +0800, wenyang.linux@foxmail.com wrote:
>> By checking whether count is equal to sizeof(ucnt), such errors
>> could be detected. It also follows the requirements of the manual.
> Does it?  This is what the eventfd manual page says:
>
>       A write(2) fails with the error EINVAL if the size of the supplied buffer
>       is less than 8 bytes, or if an attempt is made to write the value
>       0xffffffffffffffff.
>
> So, *technically* it doesn't mention the behavior if the size is greater than 8
> bytes.  But one might assume that such writes are accepted, since otherwise it
> would have been mentioned that they're rejected, just like writes < 8 bytes.


Thank you for your commtents.
Although this behavior was not mentioned, it may indeed lead to
undefined performance, such as (we changed char [] to char *):

#include <sys/eventfd.h>

#include <unistd.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>

int main()
{
     //char str[32] = "hello world";
     char *str = "hello world";
     uint64_t value;
     ssize_t size;
     int fd;

     fd = eventfd(0, 0);
     size = write(fd, &str, strlen(str));
     printf("eventfd: test writing a string:%s, size=%ld\n", str, size);
     size = read(fd, &value, sizeof(value));
     printf("eventfd: test reading as uint64, size=%ld, value=0x%lX\n", 
size, value);
     close(fd);

     return 0;
}


$ ./a.out
eventfd: test writing a string:hello world, size=8
eventfd: test reading as uint64, size=8, value=0x560CC0134008

$ ./a.out
eventfd: test writing a string:hello world, size=8
eventfd: test reading as uint64, size=8, value=0x55A3CD373008

$ ./a.out
eventfd: test writing a string:hello world, size=8
eventfd: test reading as uint64, size=8, value=0x55B8D7B99008


--

Best wishes,

Wen


>
> If the validation is indeed going to be made more strict, the manual page will
> need to be fixed alongside it.
>
> - Eric