xdp-forward-queueing: Update outstanding bytes#497
xdp-forward-queueing: Update outstanding bytes#497luisjira wants to merge 2 commits intoxdp-project:xdp-forward-queueingfrom
Conversation
Signed-off-by: Luis Jira <me@luisjira.com>
Signed-off-by: Luis Jira <me@luisjira.com>
luisjira
force-pushed
the
xdp-forward-queueing-fix
branch
from
4be4a36 to
4847583
Compare
tohojo
left a comment
tohojo
left a comment
There was a problem hiding this comment.
Thank you for the patch; a bunch of nits, see below.
| * pin the link to pin_root_path/xdp_check_return. | ||
| */ | ||
|
|
||
| return_prog = bpf_object__find_program_by_name(skel->obj,"xdp_check_return"); |
There was a problem hiding this comment.
The program is available in the skeleton as skel->progs.xdp_check_return - no need to use bpf_object__find_program_by_name()
| goto end_detach; | ||
| } | ||
|
|
||
| link = bpf_program__attach_raw_tracepoint(return_prog, "xdp_frame_return"); |
There was a problem hiding this comment.
It should be possible to just call bpf_program__attach(), since the target is specified in the section definition in the BPF source file.
| } | ||
|
|
||
| link = bpf_program__attach_raw_tracepoint(return_prog, "xdp_frame_return"); | ||
| if (libbpf_get_error((const void *) link)) { |
There was a problem hiding this comment.
It should be possible to just pass the pointers to libbpf_get_error() without the cast...
| link = bpf_program__attach_raw_tracepoint(return_prog, "xdp_frame_return"); | ||
| if (libbpf_get_error((const void *) link)) { | ||
| pr_warn("Couldn't attach to xdp_frame_return: %s\n", | ||
| strerror(libbpf_get_error((const void *) link))); |
| } | ||
| skel->links.xdp_check_return = link; | ||
|
|
||
| snprintf(pin_link_path, sizeof(pin_link_path), "%s/%s", pin_root_path, |
There was a problem hiding this comment.
This should use the try_snprintf() function (from lib/util), and check the return code. We should probably also disambiguate the pin path right now, with this path bad things will happen if multiple instances of xdp-forward are loaded at the same time.
| for (iface = opt->ifaces; iface; iface = iface->next) | ||
| xdp_program__detach(xdp_prog, iface->ifindex, opt->xdp_mode, 0); | ||
|
|
||
| bpf_link__unpin(link); |
There was a problem hiding this comment.
bpf_link__unpin() will crash if link is NULL
| pr_info("Unloaded from interface %s\n", iface->ifname); | ||
| } | ||
|
|
||
| snprintf(pin_link_path, sizeof(pin_link_path), "%s/%s", pin_root_path, |
There was a problem hiding this comment.
See above wrt using try_snprintf()
| snprintf(pin_link_path, sizeof(pin_link_path), "%s/%s", pin_root_path, | ||
| "xdp_check_return"); | ||
|
|
||
| link = bpf_link__open(pin_link_path); |
There was a problem hiding this comment.
Instead of going through this open/unpin/destroy dance, just unlink the pin file?
| strerror(errno)); | ||
| goto end_detach; | ||
| } | ||
| pr_info("Pinned xdp_check_return to %s\n", pin_link_path); |
There was a problem hiding this comment.
This should probably be pr_debug()
2 participants
At the moment, xdp-forward-queueing includes logic to track the number of outstanding bytes in hardware, but it is unused.
This pull request attaches the frame return logic to the corresponding tracepoint to allow subtracting the completed data. It also adds the counter increment when data is sent.