summaryrefslogtreecommitdiffstats
path: root/src/net/udp/slam.c
diff options
context:
space:
mode:
authorMichael Brown2008-06-11 10:41:31 +0200
committerMichael Brown2008-06-11 10:41:31 +0200
commit5d36ec6721070e4efe817a853612837535019718 (patch)
tree2ec2a20ec3220457d202a399dcfee791db77f655 /src/net/udp/slam.c
parent[retry] Add start_timer_fixed() (diff)
downloadipxe-5d36ec6721070e4efe817a853612837535019718.tar.gz
ipxe-5d36ec6721070e4efe817a853612837535019718.tar.xz
ipxe-5d36ec6721070e4efe817a853612837535019718.zip
[slam] Implement SLAM flow control
On any fast network, or with any driver that may drop packets (e.g. Infiniband, which has very small RX rings), the traditional usage of the SLAM protocol will result in enormous numbers of packet drops and a consequent large number of retransmissions. By adapting the client behaviour, we can force the server to act more like a multicast TFTP server, with flow control provided by a single master client. This behaviour should interoperate with any traditional SLAM client (e.g. Etherboot 5.4) on the network. The SLAM protocol isn't actually documented anywhere, so it's hard to define either behaviour as compliant or otherwise.
Diffstat (limited to 'src/net/udp/slam.c')
-rw-r--r--src/net/udp/slam.c192
1 files changed, 85 insertions, 107 deletions
diff --git a/src/net/udp/slam.c b/src/net/udp/slam.c
index f7ce546e..510b9b21 100644
--- a/src/net/udp/slam.c
+++ b/src/net/udp/slam.c
@@ -30,6 +30,7 @@
#include <gpxe/open.h>
#include <gpxe/uri.h>
#include <gpxe/tcpip.h>
+#include <gpxe/timer.h>
#include <gpxe/retry.h>
/** @file
@@ -90,18 +91,18 @@ FEATURE ( FEATURE_PROTOCOL, "SLAM", DHCP_EB_FEATURE_SLAM, 1 );
#define SLAM_DEFAULT_MULTICAST_PORT 10000
/** Maximum SLAM header length */
-#define SLAM_MAX_HEADER_LEN ( 8 /* transaction id */ + 8 /* total_bytes */ + \
- 8 /* block_size */ )
+#define SLAM_MAX_HEADER_LEN ( 7 /* transaction id */ + 7 /* total_bytes */ + \
+ 7 /* block_size */ )
/** Maximum SLAM NACK length
*
- * This is a policy decision. Shorter packets take less time to
- * construct and spew out less debug output, and there's a limit to
- * how useful it is to send a complete missing-block list anyway; if
- * the loss rate is high then we're going to have to retransmit an
- * updated missing-block list anyway.
+ * We only ever send a NACK for a single packet.
*/
-#define SLAM_MAX_NACK_LEN 16
+#define SLAM_MAX_NACK_LEN ( 7 /* block */ + 1 /* #blocks = 1 */ + \
+ 1 /* NUL */ )
+
+/** SLAM slave timeout */
+#define SLAM_SLAVE_TIMEOUT ( 1 * TICKS_PER_SEC )
/** A SLAM request */
struct slam_request {
@@ -115,8 +116,10 @@ struct slam_request {
/** Multicast socket */
struct xfer_interface mc_socket;
- /** NACK timer */
- struct retry_timer timer;
+ /** Master client retry timer */
+ struct retry_timer master_timer;
+ /** Slave client retry timer */
+ struct retry_timer slave_timer;
/** Cached header */
uint8_t header[SLAM_MAX_HEADER_LEN];
@@ -167,8 +170,9 @@ static void slam_finished ( struct slam_request *slam, int rc ) {
sizeof ( slam_disconnect ) );
}
- /* Stop the retry timer */
- stop_timer ( &slam->timer );
+ /* Stop the retry timers */
+ stop_timer ( &slam->master_timer );
+ stop_timer ( &slam->slave_timer );
/* Close all data transfer interfaces */
xfer_nullify ( &slam->socket );
@@ -191,14 +195,14 @@ static void slam_finished ( struct slam_request *slam, int rc ) {
* @v slam SLAM request
* @v iobuf I/O buffer
* @v value Value to add
- * @v reserved Length of reserved space at end of buffer
- * @ret len Length of value, or negative error.
+ * @ret rc Return status code
*
- * Adds a variable-length value to the end of an I/O buffer.
+ * Adds a variable-length value to the end of an I/O buffer. Will
+ * always leave at least one byte of tailroom in the I/O buffer (to
+ * allow space for the terminating NUL).
*/
static int slam_put_value ( struct slam_request *slam,
- struct io_buffer *iobuf, unsigned long value,
- size_t reserved ) {
+ struct io_buffer *iobuf, unsigned long value ) {
uint8_t *data;
size_t len;
unsigned int i;
@@ -207,7 +211,7 @@ static int slam_put_value ( struct slam_request *slam,
* leave at least one byte in the I/O buffer.
*/
len = ( ( flsl ( value ) + 10 ) / 8 );
- if ( ( len + reserved ) > iob_tailroom ( iobuf ) ) {
+ if ( len >= iob_tailroom ( iobuf ) ) {
DBGC2 ( slam, "SLAM %p cannot add %d-byte value\n",
slam, len );
return -ENOBUFS;
@@ -227,72 +231,7 @@ static int slam_put_value ( struct slam_request *slam,
*data |= ( len << 5 );
assert ( value == 0 );
- return len;
-}
-
-/**
- * Build SLAM compressed missing-block list
- *
- * @v slam SLAM request
- * @v iobuf I/O buffer
- * @ret rc Return status code
- */
-static int slam_build_block_list ( struct slam_request *slam,
- struct io_buffer *iobuf ) {
- unsigned long block;
- unsigned long block_count;
- int block_present;
- int last_block_present;
- int len;
- size_t last_len = 0;
- unsigned long last_block_count = 0;
- int rc;
-
- DBGC ( slam, "SLAM %p asking for", slam );
-
- /* Walk bitmap to construct list */
- block_count = 0;
- last_block_present = ( ! 0 );
- for ( block = 0 ; block < slam->num_blocks ; block++ ) {
- block_present = ( !! bitmap_test ( &slam->bitmap, block ) );
- if ( block_present != last_block_present ) {
- if ( ( len = slam_put_value ( slam, iobuf, block_count,
- ( sizeof ( block ) + 1 ) ) ) < 0 )
- goto truncated;
- DBGC ( slam, "%c%ld",
- ( last_block_present ? ' ' : '-' ),
- ( last_block_present ? block : block - 1 ) );
- last_len = len;
- last_block_count = block_count;
- last_block_present = block_present;
- block_count = 0;
- }
- block_count++;
- }
- if ( ( len = slam_put_value ( slam, iobuf, block_count,
- ( sizeof ( block ) + 1 ) ) ) < 0 )
- goto truncated;
- DBGC ( slam, "%c%ld\n", ( last_block_present ? ' ' : '-' ),
- ( last_block_present ? block : block - 1 ) );
-
return 0;
-
- truncated:
- rc = len;
- block -= block_count;
- assert ( last_len != 0 ); /* Cannot truncate on first entry */
- if ( last_block_present ) {
- /* Replace last missing-blocks number */
- DBGC ( slam, "#" );
- iob_unput ( iobuf, last_len );
- block -= last_block_count;
- }
- /* Report all remaining blocks as missing */
- block_count = ( slam->num_blocks - block );
- DBGC ( slam, "-%ld\n", ( slam->num_blocks - 1 ) );
- len = slam_put_value ( slam, iobuf, block_count, 1 );
- assert ( len > 0 );
- return rc;
}
/**
@@ -303,17 +242,14 @@ static int slam_build_block_list ( struct slam_request *slam,
*/
static int slam_tx_nack ( struct slam_request *slam ) {
struct io_buffer *iobuf;
+ unsigned long block;
uint8_t *nul;
-
- DBGC ( slam, "SLAM %p transmitting NACK\n", slam );
+ int rc;
/* Mark NACK as sent, so that we know we have to disconnect later */
slam->nack_sent = 1;
- /* Use the current block size as a good estimate of how much
- * data we can fit in a packet. If we overrun, it seems to be
- * acceptable to drop information anyway.
- */
+ /* Allocate I/O buffer */
iobuf = xfer_alloc_iob ( &slam->socket, SLAM_MAX_NACK_LEN );
if ( ! iobuf ) {
DBGC ( slam, "SLAM %p could not allocate I/O buffer\n",
@@ -321,12 +257,23 @@ static int slam_tx_nack ( struct slam_request *slam ) {
return -ENOMEM;
}
- /* Build block list. (Errors are non-fatal; it just means we
- * couldn't fit the compressed list within the packet.)
+ /* Construct NACK. We always request only a single packet;
+ * this allows us to force multicast-TFTP-style flow control
+ * on the SLAM server, which will otherwise just blast the
+ * data out as fast as it can. On a gigabit network, without
+ * RX checksumming, this would inevitably cause packet drops.
*/
- slam_build_block_list ( slam, iobuf );
-
- /* Add NUL terminator */
+ block = bitmap_first_gap ( &slam->bitmap );
+ if ( block ) {
+ DBGCP ( slam, "SLAM %p transmitting NACK for block %ld\n",
+ slam, block );
+ } else {
+ DBGC ( slam, "SLAM %p transmitting initial NACK\n", slam );
+ }
+ if ( ( rc = slam_put_value ( slam, iobuf, block ) ) != 0 )
+ return rc;
+ if ( ( rc = slam_put_value ( slam, iobuf, 1 ) ) != 0 )
+ return rc;
nul = iob_put ( iobuf, 1 );
*nul = 0;
@@ -335,18 +282,47 @@ static int slam_tx_nack ( struct slam_request *slam ) {
}
/**
- * Handle SLAM retransmission timer expiry
+ * Handle SLAM master client retry timer expiry
*
- * @v timer Retry timer
+ * @v timer Master retry timer
* @v fail Failure indicator
*/
-static void slam_timer_expired ( struct retry_timer *timer, int fail ) {
+static void slam_master_timer_expired ( struct retry_timer *timer,
+ int fail ) {
struct slam_request *slam =
- container_of ( timer, struct slam_request, timer );
+ container_of ( timer, struct slam_request, master_timer );
if ( fail ) {
+ /* Allow timer to stop running. We will terminate the
+ * connection only if the slave timer times out.
+ */
+ DBGC ( slam, "SLAM %p giving up acting as master client\n",
+ slam );
+ } else {
+ /* Retransmit NACK */
+ start_timer ( timer );
+ slam_tx_nack ( slam );
+ }
+}
+
+/**
+ * Handle SLAM slave client retry timer expiry
+ *
+ * @v timer Master retry timer
+ * @v fail Failure indicator
+ */
+static void slam_slave_timer_expired ( struct retry_timer *timer,
+ int fail ) {
+ struct slam_request *slam =
+ container_of ( timer, struct slam_request, slave_timer );
+
+ if ( fail ) {
+ /* Terminate connection */
slam_finished ( slam, -ETIMEDOUT );
} else {
+ /* Try sending a NACK */
+ DBGC ( slam, "SLAM %p trying to become master client\n",
+ slam );
start_timer ( timer );
slam_tx_nack ( slam );
}
@@ -490,9 +466,10 @@ static int slam_mc_socket_deliver ( struct xfer_interface *mc_socket,
size_t len;
int rc;
- /* Hit the timer */
- stop_timer ( &slam->timer );
- start_timer ( &slam->timer );
+ /* Stop the master client timer. Restart the slave client timer. */
+ stop_timer ( &slam->master_timer );
+ stop_timer ( &slam->slave_timer );
+ start_timer_fixed ( &slam->slave_timer, SLAM_SLAVE_TIMEOUT );
/* Read and strip packet header */
if ( ( rc = slam_pull_header ( slam, iobuf ) ) != 0 )
@@ -569,9 +546,9 @@ static int slam_socket_deliver ( struct xfer_interface *socket,
container_of ( socket, struct slam_request, socket );
int rc;
- /* Hit the timer */
- stop_timer ( &slam->timer );
- start_timer ( &slam->timer );
+ /* Restart the master client timer */
+ stop_timer ( &slam->master_timer );
+ start_timer ( &slam->master_timer );
/* Read and strip packet header */
if ( ( rc = slam_pull_header ( slam, iobuf ) ) != 0 )
@@ -755,7 +732,8 @@ static int slam_open ( struct xfer_interface *xfer, struct uri *uri ) {
xfer_init ( &slam->socket, &slam_socket_operations, &slam->refcnt );
xfer_init ( &slam->mc_socket, &slam_mc_socket_operations,
&slam->refcnt );
- slam->timer.expired = slam_timer_expired;
+ slam->master_timer.expired = slam_master_timer_expired;
+ slam->slave_timer.expired = slam_slave_timer_expired;
/* Fake an invalid cached header of { 0x00, ... } */
slam->header_len = 1;
/* Fake parameters for initial NACK */
@@ -792,8 +770,8 @@ static int slam_open ( struct xfer_interface *xfer, struct uri *uri ) {
goto err;
}
- /* Start retry timer */
- start_timer ( &slam->timer );
+ /* Start slave retry timer */
+ start_timer_fixed ( &slam->slave_timer, SLAM_SLAVE_TIMEOUT );
/* Attach to parent interface, mortalise self, and return */
xfer_plug_plug ( &slam->xfer, xfer );