[CRIU] [PATCH v2 03/28] files: Send/receive fds with pointers on their struct fdinfo_list_entry in receiver

[Pavel Emelyanov]

On 12/05/2016 04:17 PM, Kirill Tkhai wrote:
> On 05.12.2016 16:13, Pavel Emelyanov wrote:
>> On 12/05/2016 04:05 PM, Kirill Tkhai wrote:
>>> On 05.12.2016 15:50, Pavel Emelyanov wrote:
>>>> On 12/05/2016 01:55 PM, Kirill Tkhai wrote:
>>>>> On 05.12.2016 12:50, Pavel Emelyanov wrote:
>>>>>> On 11/30/2016 07:29 PM, Kirill Tkhai wrote:
>>>>>>> For moving to a single transport socket scheme, we need to be able to differ
>>>>>>> fds in receive queue from each other. Add a fle pointer as identifier for that.
>>>>>>>
>>>>>>> v2: Rebase on compel
>>>>>>> Signed-off-by: Kirill Tkhai <ktkhai at virtuozzo.com>
>>>>>>> ---
>>>>>>>  criu/files.c         |   21 +++++++++++++++++++--
>>>>>>>  criu/include/files.h |    1 +
>>>>>>>  criu/pipes.c         |    2 +-
>>>>>>>  criu/sk-unix.c       |    2 +-
>>>>>>>  criu/tty.c           |    2 +-
>>>>>>>  5 files changed, 23 insertions(+), 5 deletions(-)
>>>>>>>
>>>>>>> diff --git a/criu/files.c b/criu/files.c
>>>>>>> index e691558..c3ea2c4 100644
>>>>>>> --- a/criu/files.c
>>>>>>> +++ b/criu/files.c
>>>>>>> @@ -957,6 +957,23 @@ static int open_transport_fd(int pid, struct fdinfo_list_entry *fle)
>>>>>>>  	return -1;
>>>>>>>  }
>>>>>>>  
>>>>>>> +int recv_fd_from_peer(struct fdinfo_list_entry *fle)
>>>>>>> +{
>>>>>>> +	struct fdinfo_list_entry *tmp;
>>>>>>> +	int fd, ret;
>>>>>>> +
>>>>>>> +	ret = recv_fds(fle->fe->fd, &fd, 1, (void *)&tmp, sizeof(struct fdinfo_list_entry *));
>>>>>>> +	if (ret)
>>>>>>> +		return -1;
>>>>>>> +
>>>>>>> +	if (tmp != fle) {
>>>>>>
>>>>>> Can this patch be applied by its own? I mean each pstree-item allocates its
>>>>>> own fle-s, their address cannot match to each other.
>>>>>
>>>>> Yes, it can be applied by its own. Each fle is related to certain process,
>>>>> which is specified in fle->pid.
>>>>
>>>> So the check for tmp != fle will always be true, will it?
>>>
>>> It has to be always false, as we receive the only fd from per-fd transport socket.
>>
>> But how will it be false? You call send_fds() passing a pointer to the sender's
>> fle structure as an argument, then in another task go to this routine and receive
>> this pointer, then compare it the this other task's local pointer. How can
>> a != b be false in this scenario?
> 
> We allocate all fles in shared memory, so the pointers are not local. They are
> linked to each other in collect_fd(). So all tasks work with the same fles,
> we do not copy them to task private address space.

Objects are in shared space, but each task calls collect_fd() which allocates
a set of fle-s for each pstree_item independently. Thus each task has its own
set of pointers on shared fle-s.

-- Pavel

>>>>>>> +		pr_err("Received wrong fle\n");
>>>>>>> +		return -1;
>>>>>>> +	}
>>>>>>> +
>>>>>>> +	return fd;
>>>>>>> +}
>>>>>>> +
>>>>>>>  int send_fd_to_peer(int fd, struct fdinfo_list_entry *fle, int sock)
>>>>>>>  {
>>>>>>>  	struct sockaddr_un saddr;
>>>>>>> @@ -967,7 +984,7 @@ int send_fd_to_peer(int fd, struct fdinfo_list_entry *fle, int sock)
>>>>>>>  	transport_name_gen(&saddr, &len,
>>>>>>>  			futex_get(&fle->real_pid), fle->fe->fd);
>>>>>>>  	pr_info("\t\tSend fd %d to %s\n", fd, saddr.sun_path + 1);
>>>>>>> -	return send_fd(sock, &saddr, len, fd);
>>>>>>> +	return send_fds(sock, &saddr, len, &fd, 1, (void *)&fle, sizeof(struct fdinfo_list_entry *));
>>>>>>>  }
>>>>>>>  
>>>>>>>  static int send_fd_to_self(int fd, struct fdinfo_list_entry *fle, int sock)
>>>>>>> @@ -1077,7 +1094,7 @@ static int receive_fd(int pid, struct fdinfo_list_entry *fle)
>>>>>>>  
>>>>>>>  	pr_info("\tReceive fd for %d\n", fle->fe->fd);
>>>>>>>  
>>>>>>> -	tmp = recv_fd(fle->fe->fd);
>>>>>>> +	tmp = recv_fd_from_peer(fle);
>>>>>>>  	if (tmp < 0) {
>>>>>>>  		pr_err("Can't get fd %d\n", tmp);
>>>>>>>  		return -1;
>>>>>>> diff --git a/criu/include/files.h b/criu/include/files.h
>>>>>>> index 229c0d2..4f40e71 100644
>>>>>>> --- a/criu/include/files.h
>>>>>>> +++ b/criu/include/files.h
>>>>>>> @@ -148,6 +148,7 @@ extern int file_desc_add(struct file_desc *d, u32 id, struct file_desc_ops *ops)
>>>>>>>  extern struct fdinfo_list_entry *file_master(struct file_desc *d);
>>>>>>>  extern struct file_desc *find_file_desc_raw(int type, u32 id);
>>>>>>>  
>>>>>>> +extern int recv_fd_from_peer(struct fdinfo_list_entry *fle);
>>>>>>>  extern int send_fd_to_peer(int fd, struct fdinfo_list_entry *fle, int sock);
>>>>>>>  extern int restore_fown(int fd, FownEntry *fown);
>>>>>>>  extern int rst_file_params(int fd, FownEntry *fown, int flags);
>>>>>>> diff --git a/criu/pipes.c b/criu/pipes.c
>>>>>>> index 9fc87bd..31f436e 100644
>>>>>>> --- a/criu/pipes.c
>>>>>>> +++ b/criu/pipes.c
>>>>>>> @@ -237,7 +237,7 @@ static int recv_pipe_fd(struct pipe_info *pi)
>>>>>>>  
>>>>>>>  	pr_info("\tWaiting fd for %d\n", fd);
>>>>>>>  
>>>>>>> -	tmp = recv_fd(fd);
>>>>>>> +	tmp = recv_fd_from_peer(fle);
>>>>>>>  	if (tmp < 0) {
>>>>>>>  		pr_err("Can't get fd %d\n", tmp);
>>>>>>>  		return -1;
>>>>>>> diff --git a/criu/sk-unix.c b/criu/sk-unix.c
>>>>>>> index 202228d..74bd76d 100644
>>>>>>> --- a/criu/sk-unix.c
>>>>>>> +++ b/criu/sk-unix.c
>>>>>>> @@ -1096,7 +1096,7 @@ static int open_unixsk_pair_slave(struct unix_sk_info *ui)
>>>>>>>  	pr_info("Opening pair slave (id %#x ino %#x peer %#x) on %d\n",
>>>>>>>  			ui->ue->id, ui->ue->ino, ui->ue->peer, fle->fe->fd);
>>>>>>>  
>>>>>>> -	sk = recv_fd(fle->fe->fd);
>>>>>>> +	sk = recv_fd_from_peer(fle);
>>>>>>>  	if (sk < 0) {
>>>>>>>  		pr_err("Can't recv pair slave\n");
>>>>>>>  		return -1;
>>>>>>> diff --git a/criu/tty.c b/criu/tty.c
>>>>>>> index 7822a4f..6d0c8fa 100644
>>>>>>> --- a/criu/tty.c
>>>>>>> +++ b/criu/tty.c
>>>>>>> @@ -923,7 +923,7 @@ static int receive_tty(struct tty_info *info)
>>>>>>>  	fle = file_master(&info->d);
>>>>>>>  	pr_info("\tWaiting tty fd %d (pid %d)\n", fle->fe->fd, fle->pid);
>>>>>>>  
>>>>>>> -	fd = recv_fd(fle->fe->fd);
>>>>>>> +	fd = recv_fd_from_peer(fle);
>>>>>>>  	close(fle->fe->fd);
>>>>>>>  	if (fd < 0) {
>>>>>>>  		pr_err("Can't get fd %d\n", fd);
>>>>>>>
>>>>>>> .
>>>>>>>
>>>>>>
>>>>> .
>>>>>
>>>>
>>> .
>>>
>>
> .
>