[PATCH 17/21] fastboot net: implement fastboot over UDP

[Sascha Hauer]

Hi Daniel,

On Mon, Jun 29, 2020 at 09:50:51PM +0200, Daniel Glöckner wrote:
> Hello Sascha,
> 
> Am 19.06.20 um 09:44 schrieb Sascha Hauer:
> > +struct fastboot_net {
> > +	struct fastboot fastboot;
> > +
> > +	struct net_connection *net_con;
> > +	struct fastboot_header response_header;
> > +	struct poller_struct poller;
> > +	struct work_queue wq;
> > +	u64 host_waits_since;
> > +	u64 last_download_pkt;
> > +	bool sequence_number_seen;
> > +	bool active_download;
> > +	bool reinit;
> > +	bool send_keep_alive;
> > +	enum may_send may_send;
> > +
> > +	IPaddr_t host_addr;
> > +	u16 host_port;
> > +	u8 host_mac[ETH_ALEN];
> > +	u16 sequence_number;
> > +	u16 last_payload_len;
> > +	uchar last_payload[FASTBOOT_MAX_CMD_LEN + sizeof(struct fastboot_header)];
> 
> This is not FASTBOOT_MAX_CMD_LEN. It's the 64 that is strewn around in
> fastboot_tx_print. Adding a new constant FASTBOOT_MAX_MSG_LEN would be
> correct.

There's no check that the packet copied into last_payload has this
maximum size. I switched to storing the packet in an allocated buffer,
with this the define is not needed anymore.

> > +	net_write_ip(&fbn->net_con->ip->daddr, fbn->host_addr);
> > +	fbn->net_con->udp->uh_dport = fbn->host_port;
> > +	net_udp_send(fbn->net_con, fbn->last_payload_len);
> > +
> > +	fbn->may_send = MAY_NOT_SEND;
> 
> You moved that line below net_udp_send. Is there any risk that
> 
> 1. our work queue executes a command which calls fastboot_tx_print
> 2. the net_udp_send caused by that fastboot_tx_print sleeps
> 3. our poller is executed and decides to send a message because
>    may_send is still MAY_SEND_MESSAGE

Ok, changed that.

> > +	fbn->last_download_pkt = get_time_ns();
> > +}
> 
> Although you added that last_download_pkt timeout check to the poller,
> there is still the risk that we will never close download_fd if
> fastboot_net_abort is called (f.ex. by the first fastboot_tx_print
> inside cb_download) before we open download_fd. In that case there
> is no poller to check for the timeout.

I was not able to provoke such a behaviour. It seems that
fastboot_abort() is called often enough that this doesn't happen. In
doubt it happens when the next fastboot session is initiated.

> > +	if (fastboot_data_len >= FASTBOOT_MAX_CMD_LEN) {
> 
> Still off-by-one. Replace >= with >

Ok, fixed.

> > +		ret = poller_register(&fbn->poller, "fastboot");
> > +		if (ret) {
> > +			pr_err("Cannot register poller: %s\n", strerror(-ret));
> > +			return;
> 
> It is not obvious that a second FASTBOOT_INIT will _not_ cause this
> error because fastboot_net_abort unregistered the previous poller.
> I would at least add a comment to the fastboot_net_abort(fbn) line.

Added a comment.

> > +{
> > +	struct fastboot_net *fbn = container_of(poller, struct fastboot_net,
> > +					       poller);
> > +
> > +	if (fbn->active_download && is_timeout(fbn->last_download_pkt, 5 * SECOND)) {
> 
> Should pollers prefer is_timeout_non_interruptible over is_timeout?

Since with the current approach we no longer execute pollers inside of
pollers this shouldn't make a difference.

Sascha

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |