-
Notifications
You must be signed in to change notification settings - Fork 50
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Fix non-recoverable AF_PACKET TX error
Add a part of upstream AF_PACKET fix. The full patch breaks E2E tests, so only a part of it is used that fixes the stuck TX error and improves diagnostics. Ref: https://gerrit.fd.io/r/c/vpp/+/31078
- Loading branch information
Showing
1 changed file
with
153 additions
and
0 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,153 @@ | ||
| From f5b65cf107e9d4a5ba7e5b2e4d788e178b6b3387 Mon Sep 17 00:00:00 2001 | ||
| From: Nathan Skrzypczak <[email protected]> | ||
| Date: Mon, 1 Feb 2021 17:13:59 +0100 | ||
| Subject: [PATCH] devices: af-packet gso mtu [MODIFIED - [email protected]] | ||
|
|
||
| Type: fix | ||
|
|
||
| Set the GSO flag when buffer length exceeds the | ||
| linux mtu. Don't listen for mtu changes on linux | ||
| side for now. | ||
|
|
||
| This also fixes a TX issue, as we only search for | ||
| valid frames on tx to the extent of n_left, we might | ||
| stay stuck. | ||
|
|
||
| Change-Id: Idf0bdd88990254a614962c2f7bc3e0292ccfd61a | ||
| Signed-off-by: Nathan Skrzypczak <[email protected]> | ||
| --- | ||
| src/vnet/devices/af_packet/device.c | 88 +++++++++++++++++++------- | ||
| 4 files changed, 126 insertions(+), 38 deletions(-) | ||
|
|
||
| diff --git a/src/vnet/devices/af_packet/device.c b/src/vnet/devices/af_packet/device.c | ||
| index 2886fb5b6..b6b99a046 100644 | ||
| --- a/src/vnet/devices/af_packet/device.c | ||
| +++ b/src/vnet/devices/af_packet/device.c | ||
| @@ -67,7 +67,48 @@ format_af_packet_device_name (u8 * s, va_list * args) | ||
| static u8 * | ||
| format_af_packet_device (u8 * s, va_list * args) | ||
| { | ||
| - s = format (s, "Linux PACKET socket interface"); | ||
| + u32 dev_instance = va_arg (*args, u32); | ||
| + u32 indent = format_get_indent (s); | ||
| + int __clib_unused verbose = va_arg (*args, int); | ||
| + | ||
| + af_packet_main_t *apm = &af_packet_main; | ||
| + af_packet_if_t *apif = pool_elt_at_index (apm->interfaces, dev_instance); | ||
| + clib_spinlock_lock_if_init (&apif->lockp); | ||
| + u32 block_size = apif->tx_req->tp_block_size; | ||
| + u32 frame_size = apif->tx_req->tp_frame_size; | ||
| + u32 frame_num = apif->tx_req->tp_frame_nr; | ||
| + int block = 0; | ||
| + u8 *block_start = apif->tx_ring + block * block_size; | ||
| + u32 tx_frame = apif->next_tx_frame; | ||
| + struct tpacket2_hdr *tph; | ||
| + | ||
| + s = format (s, "Linux PACKET socket interface\n"); | ||
| + s = format (s, "%Ublock:%d frame:%d\n", format_white_space, indent, | ||
| + block_size, frame_size); | ||
| + s = format (s, "%Unext frame:%d\n", format_white_space, indent, | ||
| + apif->next_tx_frame); | ||
| + | ||
| + int n_send_req = 0, n_avail = 0, n_sending = 0, n_tot = 0, n_wrong = 0; | ||
| + do | ||
| + { | ||
| + tph = (struct tpacket2_hdr *) (block_start + tx_frame * frame_size); | ||
| + tx_frame = (tx_frame + 1) % frame_num; | ||
| + if (tph->tp_status == 0) | ||
| + n_avail++; | ||
| + else if (tph->tp_status & TP_STATUS_SEND_REQUEST) | ||
| + n_send_req++; | ||
| + else if (tph->tp_status & TP_STATUS_SENDING) | ||
| + n_sending++; | ||
| + else | ||
| + n_wrong++; | ||
| + n_tot++; | ||
| + } | ||
| + while (tx_frame != apif->next_tx_frame); | ||
| + s = format (s, "%Uavailable:%d request:%d sending:%d wrong:%d total:%d\n", | ||
| + format_white_space, indent, n_avail, n_send_req, n_sending, | ||
| + n_wrong, n_tot); | ||
| + | ||
| + clib_spinlock_unlock_if_init (&apif->lockp); | ||
| return s; | ||
| } | ||
|
|
||
| @@ -99,7 +140,7 @@ VNET_DEVICE_CLASS_TX_FN (af_packet_device_class) (vlib_main_t * vm, | ||
| struct tpacket2_hdr *tph; | ||
| u32 frame_not_ready = 0; | ||
|
|
||
| - while (n_left > 0) | ||
| + while (n_left) | ||
| { | ||
| u32 len; | ||
| u32 offset = 0; | ||
| @@ -108,13 +149,17 @@ VNET_DEVICE_CLASS_TX_FN (af_packet_device_class) (vlib_main_t * vm, | ||
| u32 bi = buffers[0]; | ||
| buffers++; | ||
|
|
||
| + nextframe: | ||
| tph = (struct tpacket2_hdr *) (block_start + tx_frame * frame_size); | ||
| - | ||
| - if (PREDICT_FALSE | ||
| - (tph->tp_status & (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING))) | ||
| + if (PREDICT_FALSE (tph->tp_status & | ||
| + (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING))) | ||
| { | ||
| + tx_frame = (tx_frame + 1) % frame_num; | ||
| frame_not_ready++; | ||
| - goto next; | ||
| + /* check if we've exhausted the ring */ | ||
| + if (PREDICT_FALSE (frame_not_ready + n_sent == frame_num)) | ||
| + break; | ||
| + goto nextframe; | ||
| } | ||
|
|
||
| do | ||
| @@ -132,7 +177,7 @@ VNET_DEVICE_CLASS_TX_FN (af_packet_device_class) (vlib_main_t * vm, | ||
| tph->tp_len = tph->tp_snaplen = offset; | ||
| tph->tp_status = TP_STATUS_SEND_REQUEST; | ||
| n_sent++; | ||
| - next: | ||
| + | ||
| tx_frame = (tx_frame + 1) % frame_num; | ||
|
|
||
| /* check if we've exhausted the ring */ | ||
| @@ -142,23 +187,22 @@ VNET_DEVICE_CLASS_TX_FN (af_packet_device_class) (vlib_main_t * vm, | ||
|
|
||
| CLIB_MEMORY_BARRIER (); | ||
|
|
||
| - if (PREDICT_TRUE (n_sent)) | ||
| - { | ||
| - apif->next_tx_frame = tx_frame; | ||
| + apif->next_tx_frame = tx_frame; | ||
|
|
||
| - if (PREDICT_FALSE (sendto (apif->fd, NULL, 0, | ||
| - MSG_DONTWAIT, NULL, 0) == -1)) | ||
| - { | ||
| - /* Uh-oh, drop & move on, but count whether it was fatal or not. | ||
| - * Note that we have no reliable way to properly determine the | ||
| - * disposition of the packets we just enqueued for delivery. | ||
| - */ | ||
| - vlib_error_count (vm, node->node_index, | ||
| - unix_error_is_fatal (errno) ? | ||
| + if (PREDICT_TRUE (n_sent)) | ||
| + if (PREDICT_FALSE (sendto (apif->fd, NULL, 0, MSG_DONTWAIT, NULL, 0) == | ||
| + -1)) | ||
| + { | ||
| + /* Uh-oh, drop & move on, but count whether it was fatal or not. | ||
| + * Note that we have no reliable way to properly determine the | ||
| + * disposition of the packets we just enqueued for delivery. | ||
| + */ | ||
| + vlib_error_count (vm, node->node_index, | ||
| + unix_error_is_fatal (errno) ? | ||
| AF_PACKET_TX_ERROR_TXRING_FATAL : | ||
| - AF_PACKET_TX_ERROR_TXRING_EAGAIN, n_sent); | ||
| - } | ||
| - } | ||
| + AF_PACKET_TX_ERROR_TXRING_EAGAIN, | ||
| + n_sent); | ||
| + } | ||
|
|
||
| clib_spinlock_unlock_if_init (&apif->lockp); | ||
|
|