laforge has submitted this change. (
https://gerrit.osmocom.org/c/osmo-e1d/+/27579?usp=email )
Change subject: usb: Deal with truncated ISO IN transfers
......................................................................
usb: Deal with truncated ISO IN transfers
It seems that in some circumstances, an ISO IN transfer can be
truncated by the bus / host. In such situation we'd currently pass
a non-modulo-32 length to the mux_demux (deframer) code, and it ASSERTs
on that. Let's try to handle this more gracefully by substituting
random garbage and letting higher layers deal with massive bit errors.
Related: OS#5490
Change-Id: Ic453325b93b0e12727625a1495a948d96df4b542
---
M src/e1d.h
M src/intf_line.c
M src/usb.c
3 files changed, 45 insertions(+), 0 deletions(-)
Approvals:
manawyrm: Looks good to me, approved
Jenkins Builder: Verified
diff --git a/src/e1d.h b/src/e1d.h
index 1c7c4c4..45ecb82 100644
--- a/src/e1d.h
+++ b/src/e1d.h
@@ -59,6 +59,7 @@
LINE_CTR_RX_REMOTE_A,
LINE_CTR_FRAMES_MUXED_E1T,
LINE_CTR_FRAMES_DEMUXED_E1O,
+ LINE_CTR_USB_ISO_TRUNC,
};
enum e1d_line_stat_item {
diff --git a/src/intf_line.c b/src/intf_line.c
index 3895653..f41214e 100644
--- a/src/intf_line.c
+++ b/src/intf_line.c
@@ -60,6 +60,7 @@
[LINE_CTR_RX_REMOTE_A] ={ "rx:remote_alarm", "Rx Frames
Reporting Remote Alarm"},
[LINE_CTR_FRAMES_MUXED_E1T] = { "tx:frames_muxed", "E1 Tx Frames
multiplexed" },
[LINE_CTR_FRAMES_DEMUXED_E1O] = { "rx:frames_demuxed", "E1 Rx Frames
demultiplexed" },
+ [LINE_CTR_USB_ISO_TRUNC] = { "rx:usb_iso_trunc", "USB ISO
packets truncated" },
};
static const struct rate_ctr_group_desc line_ctrg_desc = {
diff --git a/src/usb.c b/src/usb.c
index 1ff9b43..fcffc12 100644
--- a/src/usb.c
+++ b/src/usb.c
@@ -131,6 +131,33 @@
{
if (len == 0)
return 0;
+
+ if (len < 4) {
+ LOGPLI(flow->line, DE1D, LOGL_ERROR, "IN EP %02x ISO packet
truncated: len = %u\n",
+ flow->ep, len);
+ line_ctr_add(flow->line, LINE_CTR_USB_ISO_TRUNC, 1);
+ return 0;
+ }
+
+ if (len > 4 && (len - 4) % 32) {
+ /* some ISO IN packet was truncated. Apparently this
+ * does happen, see https://osmocom.org/issues/5490 -
+ * there is little we can do here, but instead of the
+ * earlier ASSERT, we just feed some garbage for the
+ * last few timeslots, resulting in bit errors etc. */
+ LOGPLI(flow->line, DE1D, LOGL_ERROR, "IN EP %02x ISO packet
truncated: len-4 = %u\n",
+ flow->ep, len - 4);
+ line_ctr_add(flow->line, LINE_CTR_USB_ISO_TRUNC, 1);
+
+ /* The assumption here is that only the last E1 frame is
+ * truncated, as we have no idea how many E1 frames the
+ * USB device firmware wanted to send us. */
+ len += 32 - (len % 32);
+ /* don't overflow the underlying buffer */
+ if (len > (int) size)
+ len = size;
+ }
+
return e1_line_demux_in(flow->line, buf + 4, len - 4, buf[3] & 0xf);
}
--
To view, visit https://gerrit.osmocom.org/c/osmo-e1d/+/27579?usp=email
To unsubscribe, or for help writing mail filters, visit
https://gerrit.osmocom.org/settings
Gerrit-Project: osmo-e1d
Gerrit-Branch: master
Gerrit-Change-Id: Ic453325b93b0e12727625a1495a948d96df4b542
Gerrit-Change-Number: 27579
Gerrit-PatchSet: 6
Gerrit-Owner: laforge <lafo...@osmocom.org>
Gerrit-Reviewer: Jenkins Builder
Gerrit-Reviewer: laforge <lafo...@osmocom.org>
Gerrit-Reviewer: manawyrm <osmocom.acco...@tbspace.de>
Gerrit-Reviewer: tnt <t...@246tnt.com>
Gerrit-MessageType: merged
