introduce support for XDP programs in CPUMAP
Subsystems (9)
v2 cover introduce support for XDP programs in CPUMAP
Similar to what David Ahern proposed in [1] for DEVMAPs, introduce the capability to attach and run a XDP program to CPUMAP entries.
The idea behind this feature is to add the possibility to define on which CPU run the eBPF program if the underlying hw does not support RSS.
I respin patch 1/6 from a previous series sent by David [2].
The functionality has been tested on Marvell Espressobin, i40e and mlx5. Detailed tests results can be found here:
https://github.com/xdp-project/xdp-project/blob/master/areas/cpumap/cpumap04-map-xdp-prog.org
Changes since v1:
- added performance test results
- added kselftest support
- fixed memory accounting with page_pool
- extended xdp_redirect_cpu_user.c to load an external program to perform redirect
- reported ifindex to attached eBPF program
- moved bpf_cpumap_val definition to include/uapi/linux/bpf.h
[1] https://patchwork.ozlabs.org/project/netdev/cover/[email protected]/
[2] https://patchwork.ozlabs.org/project/netdev/patch/[email protected]/
David Ahern (1):
net: Refactor xdp_convert_buff_to_frame
Lorenzo Bianconi (7):
samples/bpf: xdp_redirect_cpu_user: do not update bpf maps in option
loop
cpumap: formalize map value as a named struct
bpf: cpumap: add the possibility to attach an eBPF program to cpumap
bpf: cpumap: implement XDP_REDIRECT for eBPF programs attached to map
entries
libbpf: add SEC name for xdp programs attached to CPUMAP samples/bpf: xdp_redirect_cpu: load a eBPF program on cpumap
selftest: add tests for XDP programs in CPUMAP entries
include/linux/bpf.h | 6 + include/net/xdp.h | 41 ++-- include/trace/events/xdp.h | 16 +- include/uapi/linux/bpf.h | 14 ++ kernel/bpf/cpumap.c | 161 +++++++++++--- net/core/dev.c | 8 + samples/bpf/xdp_redirect_cpu_kern.c | 25 ++- samples/bpf/xdp_redirect_cpu_user.c | 208 ++++++++++++++++-- tools/include/uapi/linux/bpf.h | 14 ++ tools/lib/bpf/libbpf.c | 2 + .../bpf/prog_tests/xdp_cpumap_attach.c | 70 ++++++ .../bpf/progs/test_xdp_with_cpumap_helpers.c | 38 ++++ 12 files changed, 531 insertions(+), 72 deletions(-) create mode 100644 tools/testing/selftests/bpf/prog_tests/xdp_cpumap_attach.c create mode 100644 tools/testing/selftests/bpf/progs/test_xdp_with_cpumap_helpers.c
Signature · 1 line
2.26.2
PATCH v2 1/8 net: Refactor xdp_convert_buff_to_frame
From: David Ahern <[email protected]>
Move the guts of xdp_convert_buff_to_frame to a new helper, xdp_update_frame_from_buff so it can be reused removing code duplication
--- include/net/xdp.h | 35 ++++++++++++++++++++++------------- 1 file changed, 22 insertions(+), 13 deletions(-)
include/net/xdp.h +22 -13
@@ -121,39 +121,48 @@ void xdp_convert_frame_to_buff(struct xdp_frame *frame, struct xdp_buff *xdp) xdp->frame_sz = frame->frame_sz; } -/* Convert xdp_buff to xdp_frame */ static inline -struct xdp_frame *xdp_convert_buff_to_frame (struct xdp_buff *xdp) +int xdp_update_frame_from_buff (struct xdp_buff *xdp, +struct xdp_frame *xdp_frame) { -struct xdp_frame *xdp_frame; -int metasize; -int headroom; - -if (xdp->rxq->mem.type ==MEM_TYPE_XSK_BUFF_POOL ) -return xdp_convert_zc_to_xdp_frame (xdp); +int metasize, headroom;/* Assure headroom is available for storing info */ headroom = xdp->data - xdp->data_hard_start; metasize = xdp->data - xdp->data_meta; metasize = metasize >0 ? metasize :0 ;if (unlikely ((headroom - metasize) <sizeof (*xdp_frame))) -return NULL ; +return -ENOMEM ;/* Catch if driver didn't reserve tailroom for skb_shared_info */ if (unlikely (xdp->data_end >xdp_data_hard_end (xdp))) {XDP_WARN ("Driver BUG: missing reserved tailroom" ); -return NULL ; +return -ENOMEM ; } -/* Store info in top of packet */ - xdp_frame = xdp->data_hard_start; - xdp_frame->data = xdp->data; xdp_frame->len = xdp->data_end - xdp->data; xdp_frame->headroom = headroom -sizeof (*xdp_frame); xdp_frame->metasize = metasize; xdp_frame->frame_sz = xdp->frame_sz; +return 0 ; +} + +/* Convert xdp_buff to xdp_frame */ +static inline +struct xdp_frame *xdp_convert_buff_to_frame (struct xdp_buff *xdp) +{ +struct xdp_frame *xdp_frame; + +if (xdp->rxq->mem.type ==MEM_TYPE_XSK_BUFF_POOL ) +return xdp_convert_zc_to_xdp_frame (xdp); + +/* Store info in top of packet */ + xdp_frame = xdp->data_hard_start; +if (unlikely (xdp_update_frame_from_buff (xdp, xdp_frame) <0 )) +return NULL ; +/* rxq only valid until napi_schedule ends, convert to xdp_mem_info */ xdp_frame->mem = xdp->rxq->mem;
Signature · 1 line
2.26.2
Re: [PATCH v2 bpf-next 1/8] net: Refactor xdp_convert_buff_to_frame
Quoted 32 lines · On Sat, 20 Jun 2020 00:57:17 +0200 Lorenzo Bianconi <[email protected]> wrote:
From: David Ahern <[email protected]>
Move the guts of
xdp_convert_buff_to_frameto a new helper,xdp_update_frame_from_buffso it can be reused removing code duplicationSuggested-by Jesper Dangaard Brouer <[email protected]>Co-developed-by Lorenzo Bianconi <[email protected]>Signed-off-by Lorenzo Bianconi <[email protected]>Signed-off-by David Ahern <[email protected]>--- include/net/xdp.h | 35 ++++++++++++++++++++++------------- 1 file changed, 22 insertions(+), 13 deletions(-)include/net/xdp.h +2
-7@@ -121,39 +121,48 @@ void xdp_convert_frame_to_buff(struct xdp_frame *frame, struct xdp_buff *xdp) xdp->frame_sz = frame->frame_sz; } -/* Convert xdp_buff to xdp_frame */ static inline -struct xdp_frame *xdp_convert_buff_to_frame (struct xdp_buff *xdp) +int xdp_update_frame_from_buff (struct xdp_buff *xdp, +struct xdp_frame *xdp_frame) { -struct xdp_frame *xdp_frame; -int metasize; -int headroom; - -if (xdp->rxq->mem.type ==MEM_TYPE_XSK_BUFF_POOL )
include/net/xdp.h void xdp_convert_frame_to_buff(struct xdp_frame *frame, struct xdp_buff *xdp) · line 135
Quoted 10 lines
include/net/xdp.h +2
-2-return xdp_convert_zc_to_xdp_frame (xdp); +int metasize, headroom;/* Assure headroom is available for storing info */ headroom = xdp->data - xdp->data_hard_start; metasize = xdp->data - xdp->data_meta; metasize = metasize >0 ? metasize :0 ;if (unlikely ((headroom - metasize) <sizeof (*xdp_frame))) -return NULL ; +return -ENOMEM ;
IMHO I think ENOMEM is reserved for memory allocations failures. I think ENOSPC will be more appropriate here (or EOVERFLOW).
include/net/xdp.h void xdp_convert_frame_to_buff(struct xdp_frame *frame, struct xdp_buff *xdp) · line 140
Quoted 6 lines
include/net/xdp.h +1
-1/* Catch if driver didn't reserve tailroom for skb_shared_info */ if (unlikely (xdp->data_end >xdp_data_hard_end (xdp))) {XDP_WARN ("Driver BUG: missing reserved tailroom" ); -return NULL ; +return -ENOMEM ;
Same here.
Quoted 31 lines
include/net/xdp.h +17
-3} -/* Store info in top of packet */ - xdp_frame = xdp->data_hard_start; - xdp_frame->data = xdp->data; xdp_frame->len = xdp->data_end - xdp->data; xdp_frame->headroom = headroom -sizeof (*xdp_frame); xdp_frame->metasize = metasize; xdp_frame->frame_sz = xdp->frame_sz; +return 0 ; +} + +/* Convert xdp_buff to xdp_frame */ +static inline +struct xdp_frame *xdp_convert_buff_to_frame (struct xdp_buff *xdp) +{ +struct xdp_frame *xdp_frame; + +if (xdp->rxq->mem.type ==MEM_TYPE_XSK_BUFF_POOL ) +return xdp_convert_zc_to_xdp_frame (xdp); + +/* Store info in top of packet */ + xdp_frame = xdp->data_hard_start; +if (unlikely (xdp_update_frame_from_buff (xdp, xdp_frame) <0 )) +return NULL ; +/* rxq only valid until napi_schedule ends, convert to xdp_mem_info */ xdp_frame->mem = xdp->rxq->mem;
Signature · 4 lines
Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
Re: [PATCH v2 bpf-next 1/8] net: Refactor xdp_convert_buff_to_frame
Quoted 5 lines
Quoted 2 lines · On Sat, 20 Jun 2020 00:57:17 +0200 Lorenzo Bianconi <[email protected]> wrote:
From: David Ahern <[email protected]>
[...]
Quoted 6 lines
include/net/xdp.h void xdp_convert_frame_to_buff(struct xdp_frame *frame, struct xdp_buff *xdp) · line 135
Quoted 3 lines
include/net/xdp.h +1
-1if (unlikely ((headroom - metasize) <sizeof (*xdp_frame))) -return NULL ; +return -ENOMEM ;IMHO I think ENOMEM is reserved for memory allocations failures. I think ENOSPC will be more appropriate here (or EOVERFLOW).
ack, I will fix it in v3
Regards,
Lorenzo
Quoted 50 lines
include/net/xdp.h void xdp_convert_frame_to_buff(struct xdp_frame *frame, struct xdp_buff *xdp) · line 140
Quoted 7 lines
include/net/xdp.h +1
-1/* Catch if driver didn't reserve tailroom for skb_shared_info */ if (unlikely (xdp->data_end >xdp_data_hard_end (xdp))) {XDP_WARN ("Driver BUG: missing reserved tailroom" ); -return NULL ; +return -ENOMEM ;Same here.
Quoted 31 lines
include/net/xdp.h +17
-3} -/* Store info in top of packet */ - xdp_frame = xdp->data_hard_start; - xdp_frame->data = xdp->data; xdp_frame->len = xdp->data_end - xdp->data; xdp_frame->headroom = headroom -sizeof (*xdp_frame); xdp_frame->metasize = metasize; xdp_frame->frame_sz = xdp->frame_sz; +return 0 ; +} + +/* Convert xdp_buff to xdp_frame */ +static inline +struct xdp_frame *xdp_convert_buff_to_frame (struct xdp_buff *xdp) +{ +struct xdp_frame *xdp_frame; + +if (xdp->rxq->mem.type ==MEM_TYPE_XSK_BUFF_POOL ) +return xdp_convert_zc_to_xdp_frame (xdp); + +/* Store info in top of packet */ + xdp_frame = xdp->data_hard_start; +if (unlikely (xdp_update_frame_from_buff (xdp, xdp_frame) <0 )) +return NULL ; +/* rxq only valid until napi_schedule ends, convert to xdp_mem_info */ xdp_frame->mem = xdp->rxq->mem;Signature · 4 lines
Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
PATCH v2 2/8 samples/bpf: xdp_redirect_cpu_user: do not update bpf maps in option loop
In reply to Lorenzo BianconiDo not update xdp_redirect_cpu maps running while option loop but defer it after all available options have been parsed. This is a preliminary patch to pass the program name we want to attach to the map entries as a user option
--- samples/bpf/xdp_redirect_cpu_user.c | 36 +++++++++++++++++++++-------- 1 file changed, 27 insertions(+), 9 deletions(-)
samples/bpf/xdp_redirect_cpu_user.c +27 -9
@@ -684,6 +684,7 @@ int main(int argc, char **argv)int add_cpu =-1 ;int opt, err;int prog_fd; +int *cpu, i;__u32 qsize; n_cpus =get_nprocs_conf ();
@@ -719,6 +720,13 @@ int main(int argc, char **argv) }mark_cpus_unavailable (); + cpu =malloc (n_cpus *sizeof (int )); +if (!cpu) { +fprintf (stderr,"failed to allocate cpu array\n" ); +return EXIT_FAIL ; + } +memset (cpu,0 , n_cpus *sizeof (int )); +/* Parse commands line args */ while ((opt =getopt_long (argc, argv,"hSd:s:p:q:c:xzF" , long_options, &longindex)) !=-1 ) {
@@ -763,8 +771,7 @@ int main(int argc, char **argv) errno, strerror(errno)); goto error; } -create_cpu_entry (add_cpu, qsize, added_cpus, true); - added_cpus++; + cpu[added_cpus++] = add_cpu;break ;case 'q' : qsize =atoi (optarg);
@@ -775,6 +782,7 @@ int main(int argc, char **argv)case 'h' : error:default : +free (cpu);usage (argv, obj);return EXIT_FAIL_OPTION ; }
@@ -787,16 +795,21 @@ int main(int argc, char **argv)if (ifindex ==-1 ) {fprintf (stderr,"ERR: required option --dev missing\n" );usage (argv, obj); -return EXIT_FAIL_OPTION ; + err =EXIT_FAIL_OPTION ; + goto out; }/* Required option */ if (add_cpu ==-1 ) {fprintf (stderr,"ERR: required option --cpu missing\n" );fprintf (stderr," Specify multiple --cpu option to add more\n" );usage (argv, obj); -return EXIT_FAIL_OPTION ; + err =EXIT_FAIL_OPTION ; + goto out; } +for (i =0 ; i < added_cpus; i++) +create_cpu_entry (cpu[i], qsize, i, true); +/* Remove XDP program when program is interrupted or killed */ signal (SIGINT , int_exit);signal (SIGTERM , int_exit);
@@ -804,27 +817,32 @@ int main(int argc, char **argv) prog =bpf_object__find_program_by_title (obj, prog_name);if (!prog) {fprintf (stderr,"bpf_object__find_program_by_title failed\n" ); -return EXIT_FAIL ; + err =EXIT_FAIL ; + goto out; } prog_fd =bpf_program__fd (prog);if (prog_fd <0 ) {fprintf (stderr,"bpf_program__fd failed\n" ); -return EXIT_FAIL ; + err =EXIT_FAIL ; + goto out; }if (bpf_set_link_xdp_fd (ifindex, prog_fd, xdp_flags) <0 ) {fprintf (stderr,"link set xdp fd failed\n" ); -return EXIT_FAIL_XDP ; + err =EXIT_FAIL_XDP ; + goto out; } err =bpf_obj_get_info_by_fd (prog_fd, &info, &info_len);if (err) {printf ("can't get prog info - %s\n" ,strerror (errno)); -return err; + goto out; } prog_id = info.id;stats_poll (interval, use_separators, prog_name, stress_mode); -return EXIT_OK ; +out: +free (cpu); +return err; }
Signature · 1 line
2.26.2
Re: [PATCH v2 bpf-next 2/8] samples/bpf: xdp_redirect_cpu_user: do not update bpf maps in option loop
Quoted 9 lines · On Sat, 20 Jun 2020 00:57:18 +0200 Lorenzo Bianconi <[email protected]> wrote:
Do not update
xdp_redirect_cpumaps running while option loop but defer it after all available options have been parsed. This is a preliminary patch to pass the program name we want to attach to the map entries as a user optionSigned-off-by Lorenzo Bianconi <[email protected]>--- samples/bpf/xdp_redirect_cpu_user.c | 36 +++++++++++++++++++++-------- 1 file changed, 27 insertions(+), 9 deletions(-)
Signature · 4 lines
Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
PATCH v2 3/8 cpumap: formalize map value as a named struct
In reply to Lorenzo BianconiAs it has been already done for devmap, introduce 'struct bpf_cpumap_val' to formalize the expected values that can be passed in for a CPUMAP. Update cpumap code to use the struct.
--- include/uapi/linux/bpf.h | 9 +++++++++ kernel/bpf/cpumap.c | 25 +++++++++++++------------ tools/include/uapi/linux/bpf.h | 9 +++++++++ 3 files changed, 31 insertions(+), 12 deletions(-)
include/uapi/linux/bpf.h +9 -0
@@ -3774,6 +3774,15 @@ struct bpf_devmap_val { } bpf_prog; }; +/* CPUMAP map-value layout +* +* The struct data-layout of map-value is a configuration interface. +* New members can only be added to the end of this structure. +*/ +struct bpf_cpumap_val { +__u32 qsize;/* queue size */ +}; +enum sk_action {SK_DROP =0 ,SK_PASS ,
kernel/bpf/cpumap.c +13 -12
@@ -52,7 +52,6 @@ struct xdp_bulk_queue {struct bpf_cpu_map_entry {u32 cpu;/* kthread CPU and map index */ int map_id;/* Back reference to map */ -u32 qsize;/* Queue size placeholder for map lookup */ /* XDP can run multiple RX-ring queues, need __percpu enqueue store */ struct xdp_bulk_queue __percpu *bulkq;
@@ -66,6 +65,8 @@ struct bpf_cpu_map_entry {atomic_t refcnt;/* Control when this struct can be free'ed */ struct rcu_head rcu; + +struct bpf_cpumap_val value; };struct bpf_cpu_map {
@@ -307,8 +308,8 @@ static int cpu_map_kthread_run(void *data)return 0 ; } -static struct bpf_cpu_map_entry *__cpu_map_entry_alloc (u32 qsize,u32 cpu, -int map_id) +static struct bpf_cpu_map_entry * +__cpu_map_entry_alloc (struct bpf_cpumap_val *value,u32 cpu,int map_id) {gfp_t gfp =GFP_KERNEL | __GFP_NOWARN;struct bpf_cpu_map_entry *rcpu;
@@ -338,13 +339,13 @@ static struct bpf_cpu_map_entry *__cpu_map_entry_alloc(u32 qsize, u32 cpu,if (!rcpu->queue) goto free_bulkq; - err =ptr_ring_init (rcpu->queue, qsize, gfp); + err =ptr_ring_init (rcpu->queue, value->qsize, gfp);if (err) goto free_queue; rcpu->cpu = cpu; rcpu->map_id = map_id; - rcpu->qsize = qsize; + rcpu->value.qsize = value->qsize;/* Setup kthread */ rcpu->kthread = kthread_create_on_node(cpu_map_kthread_run, rcpu, numa,
@@ -437,12 +438,12 @@ static int cpu_map_update_elem(struct bpf_map *map, void *key, void *value, u64 map_flags) { struct bpf_cpu_map *cmap = container_of(map, struct bpf_cpu_map, map); +struct bpf_cpumap_val cpumap_value = {};struct bpf_cpu_map_entry *rcpu; -/* Array index key correspond to CPU number */ u32 key_cpu = *(u32 *)key; -/* Value is the queue size */ -u32 qsize = *(u32 *)value; + +memcpy (&cpumap_value, value, map->value_size);if (unlikely (map_flags >BPF_EXIST ))return -EINVAL ;
@@ -450,18 +451,18 @@ static int cpu_map_update_elem(struct bpf_map *map, void *key, void *value,return -E2BIG ;if (unlikely (map_flags ==BPF_NOEXIST ))return -EEXIST ; -if (unlikely (qsize >16384 ))/* sanity limit on qsize */ +if (unlikely (cpumap_value.qsize >16384 ))/* sanity limit on qsize */ return -EOVERFLOW ;/* Make sure CPU is a valid possible cpu */ if (key_cpu >= nr_cpumask_bits || !cpu_possible (key_cpu))return -ENODEV ; -if (qsize ==0 ) { +if (cpumap_value.qsize ==0 ) { rcpu =NULL ;/* Same as deleting */ }else {/* Updating qsize cause re-allocation of bpf_cpu_map_entry */ - rcpu =__cpu_map_entry_alloc (qsize, key_cpu, map->id); + rcpu =__cpu_map_entry_alloc (&cpumap_value, key_cpu, map->id);if (!rcpu)return -ENOMEM ; rcpu->cmap = cmap;
@@ -523,7 +524,7 @@ static void *cpu_map_lookup_elem(struct bpf_map *map, void *key)struct bpf_cpu_map_entry *rcpu =__cpu_map_lookup_elem (map, *(u32 *)key); -return rcpu ? &rcpu->qsize :NULL ; +return rcpu ? &rcpu->value :NULL ; }static int cpu_map_get_next_key (struct bpf_map *map,void *key,void *next_key)
tools/include/uapi/linux/bpf.h +9 -0
@@ -3774,6 +3774,15 @@ struct bpf_devmap_val { } bpf_prog; }; +/* CPUMAP map-value layout +* +* The struct data-layout of map-value is a configuration interface. +* New members can only be added to the end of this structure. +*/ +struct bpf_cpumap_val { +__u32 qsize;/* queue size */ +}; +enum sk_action {SK_DROP =0 ,SK_PASS ,
Signature · 1 line
2.26.2
Re: [PATCH v2 bpf-next 3/8] cpumap: formalize map value as a named struct
Quoted 18 lines · On Sat, 20 Jun 2020 00:57:19 +0200 Lorenzo Bianconi <[email protected]> wrote:
As it has been already done for devmap, introduce 'struct
bpf_cpumap_val' to formalize the expected values that can be passed in for a CPUMAP. Update cpumap code to use the struct.Signed-off-by Lorenzo Bianconi <[email protected]>--- include/uapi/linux/bpf.h | 9 +++++++++ kernel/bpf/cpumap.c | 25 +++++++++++++------------ tools/include/uapi/linux/bpf.h | 9 +++++++++ 3 files changed, 31 insertions(+), 12 deletions(-)include/uapi/linux/bpf.h +0
-0@@ -3774,6 +3774,15 @@ struct bpf_devmap_val { } bpf_prog; };
include/uapi/linux/bpf.h struct bpf_devmap_val { · line 3785
Quoted 10 lines
include/uapi/linux/bpf.h +9
-0+/* CPUMAP map-value layout +* +* The struct data-layout of map-value is a configuration interface. +* New members can only be added to the end of this structure. +*/ +struct bpf_cpumap_val { +__u32 qsize;/* queue size */ +}; +
Nitpicking the comment: /* queue size */
It doesn't provide much information to the end-user.
What about changing it to: /* queue size to remote target CPU */ ?
Signature · 4 lines
Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
Re: [PATCH v2 bpf-next 3/8] cpumap: formalize map value as a named struct
Quoted 35 lines
Quoted 18 lines · On Sat, 20 Jun 2020 00:57:19 +0200 Lorenzo Bianconi <[email protected]> wrote:
As it has been already done for devmap, introduce 'struct
bpf_cpumap_val' to formalize the expected values that can be passed in for a CPUMAP. Update cpumap code to use the struct.Signed-off-by Lorenzo Bianconi <[email protected]>--- include/uapi/linux/bpf.h | 9 +++++++++ kernel/bpf/cpumap.c | 25 +++++++++++++------------ tools/include/uapi/linux/bpf.h | 9 +++++++++ 3 files changed, 31 insertions(+), 12 deletions(-)include/uapi/linux/bpf.h +0
-0@@ -3774,6 +3774,15 @@ struct bpf_devmap_val { } bpf_prog; };include/uapi/linux/bpf.h struct bpf_devmap_val { · line 3785
Quoted 10 lines
include/uapi/linux/bpf.h +9
-0+/* CPUMAP map-value layout +* +* The struct data-layout of map-value is a configuration interface. +* New members can only be added to the end of this structure. +*/ +struct bpf_cpumap_val { +__u32 qsize;/* queue size */ +}; +Nitpicking the comment: /* queue size */
It doesn't provide much information to the end-user.
Quoted 1 line
What about changing it to: /* queue size to remote target CPU */
Yes, I agree. I will fix it in v3.
Regards,
Lorenzo
Quoted 8 lines
?
Signature · 4 lines
Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
PATCH v2 4/8 bpf: cpumap: add the possibility to attach an eBPF program to cpumap
In reply to Lorenzo BianconiIntroduce the capability to attach an eBPF program to cpumap entries. The idea behind this feature is to add the possibility to define on which CPU run the eBPF program if the underlying hw does not support RSS. Current supported verdicts are XDP_DROP and XDP_PASS.
This patch has been tested on Marvell ESPRESSObin using xdp_redirect_cpu sample available in the kernel tree to identify possible performance regressions. Results show there are no observable differences in packet-per-second:
$./xdp_redirect_cpu --progname xdp_cpu_map0 --dev eth0 --cpu 1
rx: 354.8 Kpps
rx: 356.0 Kpps
rx: 356.8 Kpps
rx: 356.3 Kpps
rx: 356.6 Kpps
rx: 356.6 Kpps
rx: 356.7 Kpps
rx: 355.8 Kpps
rx: 356.8 Kpps
rx: 356.8 Kpps
--- include/linux/bpf.h | 6 ++ include/net/xdp.h | 5 ++ include/trace/events/xdp.h | 14 ++-- include/uapi/linux/bpf.h | 5 ++ kernel/bpf/cpumap.c | 123 +++++++++++++++++++++++++++++---- net/core/dev.c | 8 +++ tools/include/uapi/linux/bpf.h | 5 ++ 7 files changed, 149 insertions(+), 17 deletions(-)
include/linux/bpf.h +6 -0
@@ -1256,6 +1256,7 @@ struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key);void __cpu_map_flush (void );int cpu_map_enqueue (struct bpf_cpu_map_entry *rcpu,struct xdp_buff *xdp,struct net_device *dev_rx); +bool cpu_map_prog_allowed (struct bpf_map *map);/* Return map's numa specified by userspace */ static inline int bpf_map_attr_numa_node (const union bpf_attr *attr)
@@ -1416,6 +1417,11 @@ static inline int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu,return 0 ; } +static inline bool cpu_map_prog_allowed (struct bpf_map *map) +{ +return false; +} +static inline struct bpf_prog *bpf_prog_get_type_path (const char *name,enum bpf_prog_type type) {
include/net/xdp.h +5 -0
@@ -98,6 +98,11 @@ struct xdp_frame {struct net_device *dev_rx;/* used by cpumap */ }; +struct xdp_cpumap_stats { +unsigned int pass; +unsigned int drop; +}; +/* Clear kernel pointers in xdp_frame */ static inline void xdp_scrub_frame (struct xdp_frame *frame) {
include/trace/events/xdp.h +10 -4
@@ -177,9 +177,9 @@ DEFINE_EVENT(xdp_redirect_template, xdp_redirect_map_err, TRACE_EVENT(xdp_cpumap_kthread, TP_PROTO(int map_id, unsigned int processed, unsigned int drops, - int sched), + int sched, struct xdp_cpumap_stats *xdp_stats), - TP_ARGS(map_id, processed, drops, sched), + TP_ARGS(map_id, processed, drops, sched, xdp_stats), TP_STRUCT__entry( __field(int, map_id)
@@ -188,6 +188,8 @@ TRACE_EVENT(xdp_cpumap_kthread, __field(unsigned int, drops) __field(unsigned int, processed) __field(int, sched) + __field(unsigned int, xdp_pass) + __field(unsigned int, xdp_drop) ), TP_fast_assign(
@@ -197,16 +199,20 @@ TRACE_EVENT(xdp_cpumap_kthread, __entry->drops = drops; __entry->processed = processed; __entry->sched = sched; + __entry->xdp_pass = xdp_stats->pass; + __entry->xdp_drop = xdp_stats->drop; ),TP_printk ("kthread" " cpu=%d map_id=%d action=%s" " processed=%u drops=%u" -" sched=%d" , +" sched=%d" +" xdp_pass=%u xdp_drop=%u" , __entry->cpu, __entry->map_id,__print_symbolic (__entry->act, __XDP_ACT_SYM_TAB), __entry->processed, __entry->drops, - __entry->sched) + __entry->sched, + __entry->xdp_pass, __entry->xdp_drop) ); TRACE_EVENT(xdp_cpumap_enqueue,
include/uapi/linux/bpf.h +5 -0
@@ -226,6 +226,7 @@ enum bpf_attach_type {BPF_CGROUP_INET4_GETSOCKNAME ,BPF_CGROUP_INET6_GETSOCKNAME ,BPF_XDP_DEVMAP , +BPF_XDP_CPUMAP , __MAX_BPF_ATTACH_TYPE };
@@ -3781,6 +3782,10 @@ struct bpf_devmap_val {*/ struct bpf_cpumap_val {__u32 qsize;/* queue size */ +union { +int fd;/* prog fd on map write */ +__u32 id;/* prog id on map read */ + } bpf_prog; };enum sk_action {
kernel/bpf/cpumap.c +110 -13
204 lines not on this page · raw message
net/core/dev.c +8 -0
@@ -5440,6 +5440,8 @@ static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)for (i =0 ; i < new->aux->used_map_cnt; i++) {if (dev_map_can_have_prog (new->aux->used_maps[i]))return -EINVAL ; +if (cpu_map_prog_allowed (new->aux->used_maps[i])) +return -EINVAL ; } }
@@ -8864,6 +8866,12 @@ int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,return -EINVAL ; } +if (prog->expected_attach_type ==BPF_XDP_CPUMAP ) { +NL_SET_ERR_MSG (extack,"BPF_XDP_CPUMAP programs can not be attached to a device" ); +bpf_prog_put (prog); +return -EINVAL ; + } +/* prog->aux->id may be 0 for orphaned device-bound progs */ if (prog->aux->id && prog->aux->id == prog_id) {bpf_prog_put (prog);
tools/include/uapi/linux/bpf.h +5 -0
@@ -226,6 +226,7 @@ enum bpf_attach_type {BPF_CGROUP_INET4_GETSOCKNAME ,BPF_CGROUP_INET6_GETSOCKNAME ,BPF_XDP_DEVMAP , +BPF_XDP_CPUMAP , __MAX_BPF_ATTACH_TYPE };
@@ -3781,6 +3782,10 @@ struct bpf_devmap_val {*/ struct bpf_cpumap_val {__u32 qsize;/* queue size */ +union { +int fd;/* prog fd on map write */ +__u32 id;/* prog id on map read */ + } bpf_prog; };enum sk_action {
Signature · 1 line
2.26.2
Re: [PATCH v2 bpf-next 4/8] bpf: cpumap: add the possibility to attach an eBPF program to cpumap
kernel/bpf/cpumap.c static void put_cpu_map_entry(struct bpf_cpu_map_entry *rcpu) · line 232
Quoted 5 lines · On Sat, 20 Jun 2020 00:57:20 +0200 Lorenzo Bianconi <[email protected]> wrote:
kernel/bpf/cpumap.c +5
-0+static int cpu_map_bpf_prog_run_xdp (struct bpf_cpu_map_entry *rcpu, +void **xdp_frames,int n, +struct xdp_cpumap_stats *stats) +{ +struct xdp_rxq_info rxq;
I do think we can merge this code as-is (and I actually re-factored this code offlist), but I have some optimizations I would like to try out.
E.g. as I've tried to explain over IRC, it will be possible to avoid having to reconstruct xdp_rxq_info here, as we can use the expected_attach_type and remap the BPF instructions to use info from xdp_frame. I want to benchmark it first, to see if it is worth it (we will only save 2 store operations in a likely cache-hot area).
Quoted 51 lines
kernel/bpf/cpumap.c +51
-0+struct bpf_prog *prog; +struct xdp_buff xdp; +int i, nframes =0 ; + +if (!rcpu->prog) +return n; + +xdp_set_return_frame_no_direct (); + xdp.rxq = &rxq; + +rcu_read_lock (); + + prog =READ_ONCE (rcpu->prog); +for (i =0 ; i < n; i++) { +struct xdp_frame *xdpf = xdp_frames[i]; +u32 act; +int err; + + rxq.dev = xdpf->dev_rx; + rxq.mem = xdpf->mem; +/* TODO: report queue_index to xdp_rxq_info */ + +xdp_convert_frame_to_buff (xdpf, &xdp); + + act =bpf_prog_run_xdp (prog, &xdp); +switch (act) { +case XDP_PASS : + err =xdp_update_frame_from_buff (&xdp, xdpf); +if (err <0 ) { +xdp_return_frame (xdpf); + stats->drop++; + }else { + xdp_frames[nframes++] = xdpf; + stats->pass++; + } +break ; +default : +bpf_warn_invalid_xdp_action (act); +/* fallthrough */ +case XDP_DROP : +xdp_return_frame (xdpf); + stats->drop++; +break ; + } + } + +rcu_read_unlock (); +xdp_clear_return_frame_no_direct (); + +return nframes; +}
Signature · 4 lines
Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
Re: [PATCH v2 bpf-next 4/8] bpf: cpumap: add the possibility to attach an eBPF program to cpumap
In reply to Lorenzo Bianconikernel/bpf/cpumap.c static int cpu_map_kthread_run(void *data) · line 341
Quoted 8 lines · On Sat, 20 Jun 2020 00:57:20 +0200 Lorenzo Bianconi <[email protected]> wrote:
kernel/bpf/cpumap.c +3
-1@@ -273,16 +336,20 @@ static int cpu_map_kthread_run(void *data)prefetchw (page); } - m =kmem_cache_alloc_bulk (skbuff_head_cache, gfp, n, skbs); +/* Support running another XDP prog on this CPU */ + nframes =cpu_map_bpf_prog_run_xdp (rcpu, xdp_frames, n, &stats); +
If all frames are dropped my XDP program, then we will call kmem_cache_alloc_bulk() to allocate zero elements. I found this during my testing[1], and I think we should squash my proposed change in[1].
Quoted 7 lines
kernel/bpf/cpumap.c +3
-2+ m =kmem_cache_alloc_bulk (skbuff_head_cache, gfp, + nframes, skbs);if (unlikely (m ==0 )) { -for (i =0 ; i < n; i++) +for (i =0 ; i < nframes; i++) skbs[i] =NULL ;/* effect: xdp_return_frame */ - drops = n;
kernel/bpf/cpumap.c static int cpu_map_kthread_run(void *data) · line 354
Quoted 10 lines
kernel/bpf/cpumap.c +3
-2+ drops += nframes; }local_bh_disable (); -for (i =0 ; i < n; i++) { -struct xdp_frame *xdpf = frames[i]; +for (i =0 ; i < nframes; i++) { +struct xdp_frame *xdpf = xdp_frames[i];struct sk_buff *skb = skbs[i];int ret;
Signature · 4 lines
Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
Re: [PATCH v2 bpf-next 4/8] bpf: cpumap: add the possibility to attach an eBPF program to cpumap
In reply to Lorenzo Bianconikernel/bpf/cpumap.c static int cpu_map_kthread_run(void *data) · line 326
Quoted 7 lines · On Sat, 20 Jun 2020 00:57:20 +0200 Lorenzo Bianconi <[email protected]> wrote:
kernel/bpf/cpumap.c +2
-1@@ -260,10 +322,11 @@ static int cpu_map_kthread_run(void *data)* kthread CPU pinned. Lockless access to ptr_ring * consume side valid as no-resize allowed of queue. */ - n =ptr_ring_consume_batched (rcpu->queue, frames,CPUMAP_BATCH ); + n =ptr_ring_consume_batched (rcpu->queue, xdp_frames, +CPUMAP_BATCH );
This should have used the non-lock variant __ ptr_ring_consume_batched(). It did before, but I can see in git-history that this is this is a bug/issue I introduced myself earlier... ups.
Signature · 4 lines
Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
PATCH v2 5/8 bpf: cpumap: implement XDP_REDIRECT for eBPF programs attached to map entries
In reply to Lorenzo BianconiIntroduce XDP_REDIRECT support for eBPF programs attached to cpumap entries.
This patch has been tested on Marvell ESPRESSObin using a modified version of xdp_redirect_cpu sample in order to attach a XDP program to CPUMAP entries to perform a redirect on the mvneta interface. In particular the following scenario has been tested:
rq (cpu0) --> mvneta - XDP_REDIRECT (cpu0) --> CPUMAP - XDP_REDIRECT (cpu1) --> mvneta
$./xdp_redirect_cpu -p xdp_cpu_map0 -d eth0 -c 1 -e xdp_redirect \
-f xdp_redirect_kern.o -m tx_port -r eth0
tx: 285.2 Kpps rx: 285.2 Kpps
Attacching a simple XDP program on eth0 to perform XDP_TX gives comparable results:
tx: 288.4 Kpps rx: 288.4 Kpps
--- include/net/xdp.h | 1 + include/trace/events/xdp.h | 6 ++++-- kernel/bpf/cpumap.c | 17 +++++++++++++++-- 3 files changed, 20 insertions(+), 4 deletions(-)
include/net/xdp.h +1 -0
@@ -99,6 +99,7 @@ struct xdp_frame { };struct xdp_cpumap_stats { +unsigned int redirect;unsigned int pass;unsigned int drop; };
include/trace/events/xdp.h +4 -2
@@ -190,6 +190,7 @@ TRACE_EVENT(xdp_cpumap_kthread, __field(int, sched) __field(unsigned int, xdp_pass) __field(unsigned int, xdp_drop) + __field(unsigned int, xdp_redirect) ), TP_fast_assign(
@@ -201,18 +202,19 @@ TRACE_EVENT(xdp_cpumap_kthread, __entry->sched = sched; __entry->xdp_pass = xdp_stats->pass; __entry->xdp_drop = xdp_stats->drop; + __entry->xdp_redirect = xdp_stats->redirect; ),TP_printk ("kthread" " cpu=%d map_id=%d action=%s" " processed=%u drops=%u" " sched=%d" -" xdp_pass=%u xdp_drop=%u" , +" xdp_pass=%u xdp_drop=%u xdp_redirect=%u" , __entry->cpu, __entry->map_id,__print_symbolic (__entry->act, __XDP_ACT_SYM_TAB), __entry->processed, __entry->drops, __entry->sched, - __entry->xdp_pass, __entry->xdp_drop) + __entry->xdp_pass, __entry->xdp_drop, __entry->xdp_redirect) ); TRACE_EVENT(xdp_cpumap_enqueue,
kernel/bpf/cpumap.c +15 -2
@@ -240,7 +240,7 @@ static int cpu_map_bpf_prog_run_xdp(struct bpf_cpu_map_entry *rcpu,xdp_set_return_frame_no_direct (); xdp.rxq = &rxq; -rcu_read_lock (); +rcu_read_lock_bh (); prog =READ_ONCE (rcpu->prog);for (i =0 ; i < n; i++) {
@@ -266,6 +266,16 @@ static int cpu_map_bpf_prog_run_xdp(struct bpf_cpu_map_entry *rcpu, stats->pass++; }break ; +case XDP_REDIRECT : + err =xdp_do_redirect (xdpf->dev_rx, &xdp, + prog); +if (unlikely (err)) { +xdp_return_frame (xdpf); + stats->drop++; + }else { + stats->redirect++; + } +break ;default :bpf_warn_invalid_xdp_action (act);/* fallthrough */
@@ -276,7 +286,10 @@ static int cpu_map_bpf_prog_run_xdp(struct bpf_cpu_map_entry *rcpu, } } -rcu_read_unlock (); +if (stats->redirect) +xdp_do_flush_map (); + +rcu_read_unlock_bh ();/* resched point, may call do_softirq() */ xdp_clear_return_frame_no_direct ();return nframes;
Signature · 1 line
2.26.2
PATCH v2 6/8 libbpf: add SEC name for xdp programs attached to CPUMAP
In reply to Lorenzo BianconiAs for DEVMAP, support SEC("xdp_cpumap*") as a short cut for loading the program with type BPF_PROG_TYPE_XDP and expected attach type BPF_XDP_CPUMAP.
--- tools/lib/bpf/libbpf.c | 2 ++ 1 file changed, 2 insertions(+)
tools/lib/bpf/libbpf.c +2 -0
@@ -6655,6 +6655,8 @@ static const struct bpf_sec_def section_defs[] = { .attach_fn = attach_iter),BPF_EAPROG_SEC ("xdp_devmap" ,BPF_PROG_TYPE_XDP ,BPF_XDP_DEVMAP ), +BPF_EAPROG_SEC ("xdp_cpumap" ,BPF_PROG_TYPE_XDP , +BPF_XDP_CPUMAP ),BPF_PROG_SEC ("xdp" ,BPF_PROG_TYPE_XDP ),BPF_PROG_SEC ("perf_event" ,BPF_PROG_TYPE_PERF_EVENT ), BPF_PROG_SEC("lwt_in", BPF_PROG_TYPE_LWT_IN),
Signature · 1 line
2.26.2
Re: [PATCH v2 bpf-next 6/8] libbpf: add SEC name for xdp programs attached to CPUMAP
Quoted 15 lines · Lorenzo Bianconi
As for DEVMAP, support
SEC("xdp_cpumap*")as a short cut for loading the program with typeBPF_PROG_TYPE_XDPand expected attach typeBPF_XDP_CPUMAP.Signed-off-by Lorenzo Bianconi <[email protected]>--- tools/lib/bpf/libbpf.c | 2 ++ 1 file changed, 2 insertions(+)tools/lib/bpf/libbpf.c +0
-0
tools/lib/bpf/libbpf.c static const struct bpf_sec_def section_defs[] = { · line 6659
Quoted 5 lines
tools/lib/bpf/libbpf.c +2
-0@@ -6655,6 +6655,8 @@ static const struct bpf_sec_def section_defs[] = { .attach_fn = attach_iter),BPF_EAPROG_SEC ("xdp_devmap" ,BPF_PROG_TYPE_XDP ,BPF_XDP_DEVMAP ), +BPF_EAPROG_SEC ("xdp_cpumap" ,BPF_PROG_TYPE_XDP , + BPF_XDP_CPUMAP),
I noticed that XDP and a bunch of other progs don't enforce "/" at the end of the prog type prefix. Is there any reason for that? Tracing prog types are pretty consistent about that (except for perf_event, I think). I'd love it if we were trying to converge here. Do you mind switching this to "xdp_cpumap/"? Would be nice to do the same for xdp_devmap, if it's not too late yet.
Quoted 6 lines
tools/lib/bpf/libbpf.c +0
-1BPF_PROG_SEC ("xdp" ,BPF_PROG_TYPE_XDP ),BPF_PROG_SEC ("perf_event" ,BPF_PROG_TYPE_PERF_EVENT ), BPF_PROG_SEC("lwt_in", BPF_PROG_TYPE_LWT_IN), --2.26.2
PATCH v2 7/8 samples/bpf: xdp_redirect_cpu: load a eBPF program on cpumap
In reply to Lorenzo BianconiExtend xdp_redirect_cpu_{usr,kern}.c adding the possibility to load a XDP program on cpumap entries. The following options have been added:
- mprog-name: cpumap entry program name
- mprog-filename: cpumap entry program filename
- redirect-device: output interface if the cpumap program performs a XDP_REDIRECT to an egress interface
- redirect-map: bpf map used to perform XDP_REDIRECT to an egress interface
- mprog-disable: disable loading XDP program on cpumap entries
Add xdp_pass, xdp_drop, xdp_redirect stats accounting
--- samples/bpf/xdp_redirect_cpu_kern.c | 25 ++-- samples/bpf/xdp_redirect_cpu_user.c | 174 +++++++++++++++++++++++++--- 2 files changed, 177 insertions(+), 22 deletions(-)
samples/bpf/xdp_redirect_cpu_kern.c +17 -8
@@ -21,7 +21,7 @@struct { __uint(type, BPF_MAP_TYPE_CPUMAP); __uint(key_size, sizeof(u32)); - __uint(value_size, sizeof(u32)); + __uint(value_size, sizeof(struct bpf_cpumap_val)); __uint(max_entries, MAX_CPUS); } cpu_map SEC(".maps");
@@ -30,6 +30,9 @@ struct datarec {__u64 processed;__u64 dropped;__u64 issue; +__u64 xdp_pass; +__u64 xdp_drop; +__u64 xdp_redirect; };/* Count RX packets, as XDP bpf_prog doesn't get direct TX-success
@@ -692,13 +695,16 @@ int trace_xdp_cpumap_enqueue(struct cpumap_enqueue_ctx *ctx)* Code in: kernel/include/trace/events/xdp.h */ struct cpumap_kthread_ctx { -u64 __pad;// First 8 bytes are not accessible by bpf code -int map_id;// offset:8; size:4; signed:1; -u32 act;// offset:12; size:4; signed:0; -int cpu;// offset:16; size:4; signed:1; -unsigned int drops;// offset:20; size:4; signed:0; -unsigned int processed;// offset:24; size:4; signed:0; -int sched;// offset:28; size:4; signed:1; +u64 __pad;// First 8 bytes are not accessible +int map_id;// offset:8; size:4; signed:1; +u32 act;// offset:12; size:4; signed:0; +int cpu;// offset:16; size:4; signed:1; +unsigned int drops;// offset:20; size:4; signed:0; +unsigned int processed;// offset:24; size:4; signed:0; +int sched;// offset:28; size:4; signed:1; +unsigned int xdp_pass;// offset:32; size:4; signed:0; +unsigned int xdp_drop;// offset:36; size:4; signed:0; +unsigned int xdp_redirect;// offset:40; size:4; signed:0; };SEC ("tracepoint/xdp/xdp_cpumap_kthread" )
@@ -712,6 +718,9 @@ int trace_xdp_cpumap_kthread(struct cpumap_kthread_ctx *ctx)return 0 ; rec->processed += ctx->processed; rec->dropped += ctx->drops; + rec->xdp_pass += ctx->xdp_pass; + rec->xdp_drop += ctx->xdp_drop; + rec->xdp_redirect += ctx->xdp_redirect;/* Count times kthread yielded CPU via schedule call */ if (ctx->sched)
samples/bpf/xdp_redirect_cpu_user.c +160 -14
309 lines not on this page · raw message
Signature · 1 line
2.26.2
PATCH v2 8/8 selftest: add tests for XDP programs in CPUMAP entries
In reply to Lorenzo BianconiSimilar to what have been done for DEVMAP, introduce tests to verify ability to add a XDP program to an entry in a CPUMAP. Verify CPUMAP programs can not be attached to devices as a normal XDP program, and only programs with BPF_XDP_CPUMAP attach type can be loaded in a CPUMAP.
--- .../bpf/prog_tests/xdp_cpumap_attach.c | 70 +++++++++++++++++++ .../bpf/progs/test_xdp_with_cpumap_helpers.c | 38 ++++++++++ 2 files changed, 108 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/xdp_cpumap_attach.c create mode 100644 tools/testing/selftests/bpf/progs/test_xdp_with_cpumap_helpers.c
tools/testing/selftests/bpf/prog_tests/xdp_cpumap_attach.c new file +70 -0
@@ -0,0 +1,70 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <uapi/linux/bpf.h> +#include <linux/if_link.h> +#include <test_progs.h> + +#include "test_xdp_with_cpumap_helpers.skel.h" + +#define IFINDEX_LO 1 + +void test_xdp_with_cpumap_helpers (void ) +{ +struct test_xdp_with_cpumap_helpers *skel; +struct bpf_prog_info info = {}; +struct bpf_cpumap_val val = { + .qsize =192 , + }; +__u32 duration =0 , idx =0 ; +__u32 len =sizeof (info); +int err, prog_fd, map_fd; + + skel =test_xdp_with_cpumap_helpers__open_and_load (); +if (CHECK_FAIL (!skel)) { +perror ("test_xdp_with_cpumap_helpers__open_and_load" ); +return ; + } + +/* can not attach program with cpumaps that allow programs +* as xdp generic +*/ + prog_fd =bpf_program__fd (skel->progs.xdp_redir_prog); + err =bpf_set_link_xdp_fd (IFINDEX_LO , prog_fd,XDP_FLAGS_SKB_MODE ); +CHECK (err ==0 ,"Generic attach of program with 8-byte CPUMAP" , +"should have failed\n" ); + + prog_fd =bpf_program__fd (skel->progs.xdp_dummy_cm); + map_fd =bpf_map__fd (skel->maps.cpu_map); + err =bpf_obj_get_info_by_fd (prog_fd, &info, &len); +if (CHECK_FAIL (err)) + goto out_close; + + val.bpf_prog.fd = prog_fd; + err =bpf_map_update_elem (map_fd, &idx, &val,0 ); +CHECK (err,"Add program to cpumap entry" ,"err %d errno %d\n" , + err, errno); + + err =bpf_map_lookup_elem (map_fd, &idx, &val); +CHECK (err,"Read cpumap entry" ,"err %d errno %d\n" , err, errno); +CHECK (info.id != val.bpf_prog.id,"Expected program id in cpumap entry" , +"expected %u read %u\n" , info.id, val.bpf_prog.id); + +/* can not attach BPF_XDP_CPUMAP program to a device */ + err =bpf_set_link_xdp_fd (IFINDEX_LO , prog_fd,XDP_FLAGS_SKB_MODE ); +CHECK (err ==0 ,"Attach of BPF_XDP_CPUMAP program" , +"should have failed\n" ); + + val.qsize =192 ; + val.bpf_prog.fd =bpf_program__fd (skel->progs.xdp_dummy_prog); + err =bpf_map_update_elem (map_fd, &idx, &val,0 ); +CHECK (err ==0 ,"Add non-BPF_XDP_CPUMAP program to cpumap entry" , +"should have failed\n" ); + +out_close: +test_xdp_with_cpumap_helpers__destroy (skel); +} + +void test_xdp_cpumap_attach (void ) +{ +if (test__start_subtest ("CPUMAP with programs in entries" )) +test_xdp_with_cpumap_helpers (); +}
tools/testing/selftests/bpf/progs/test_xdp_with_cpumap_helpers.c new file +38 -0
@@ -0,0 +1,38 @@ +// SPDX-License-Identifier: GPL-2.0 + +#include <linux/bpf.h> +#include <bpf/bpf_helpers.h> + +struct { + __uint(type, BPF_MAP_TYPE_CPUMAP); + __uint(key_size, sizeof(__u32)); + __uint(value_size, sizeof(struct bpf_cpumap_val)); + __uint(max_entries, 4); +} cpu_map SEC(".maps"); + +SEC("xdp_redir") +int xdp_redir_prog (struct xdp_md *ctx) +{ +return bpf_redirect_map (&cpu_map,1 ,0 ); +} + +SEC("xdp_dummy") +int xdp_dummy_prog (struct xdp_md *ctx) +{ +return XDP_PASS ; +} + +SEC("xdp_cpumap") +int xdp_dummy_cm (struct xdp_md *ctx) +{ +char fmt[] ="devmap redirect: dev %u len %u\n" ; +void *data_end = (void *)(long )ctx->data_end; +void *data = (void *)(long )ctx->data; +unsigned int len = data_end - data; + +bpf_trace_printk (fmt,sizeof (fmt), ctx->ingress_ifindex, len); + +return XDP_PASS ; +} + +char _license[] SEC("license") = "GPL";
Signature · 1 line
2.26.2
Re: [PATCH v2 bpf-next 8/8] selftest: add tests for XDP programs in CPUMAP entries
Quoted 79 lines · Lorenzo Bianconi
Similar to what have been done for DEVMAP, introduce tests to verify ability to add a XDP program to an entry in a CPUMAP. Verify CPUMAP programs can not be attached to devices as a normal XDP program, and only programs with
BPF_XDP_CPUMAPattach type can be loaded in a CPUMAP.Signed-off-by Lorenzo Bianconi <[email protected]>--- .../bpf/prog_tests/xdp_cpumap_attach.c | 70 +++++++++++++++++++ .../bpf/progs/test_xdp_with_cpumap_helpers.c | 38 ++++++++++ 2 files changed, 108 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/xdp_cpumap_attach.c create mode 100644 tools/testing/selftests/bpf/progs/test_xdp_with_cpumap_helpers.ctools/testing/selftests/bpf/prog_tests/xdp_cpumap_attach.c new file +58
-0@@ -0,0 +1,70 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <uapi/linux/bpf.h> +#include <linux/if_link.h> +#include <test_progs.h> + +#include "test_xdp_with_cpumap_helpers.skel.h" + +#define IFINDEX_LO 1 + +void test_xdp_with_cpumap_helpers (void ) +{ +struct test_xdp_with_cpumap_helpers *skel; +struct bpf_prog_info info = {}; +struct bpf_cpumap_val val = { + .qsize =192 , + }; +__u32 duration =0 , idx =0 ; +__u32 len =sizeof (info); +int err, prog_fd, map_fd; + + skel =test_xdp_with_cpumap_helpers__open_and_load (); +if (CHECK_FAIL (!skel)) { +perror ("test_xdp_with_cpumap_helpers__open_and_load" ); +return ; + } + +/* can not attach program with cpumaps that allow programs +* as xdp generic +*/ + prog_fd =bpf_program__fd (skel->progs.xdp_redir_prog); + err =bpf_set_link_xdp_fd (IFINDEX_LO , prog_fd,XDP_FLAGS_SKB_MODE ); +CHECK (err ==0 ,"Generic attach of program with 8-byte CPUMAP" , +"should have failed\n" ); + + prog_fd =bpf_program__fd (skel->progs.xdp_dummy_cm); + map_fd =bpf_map__fd (skel->maps.cpu_map); + err =bpf_obj_get_info_by_fd (prog_fd, &info, &len); +if (CHECK_FAIL (err)) + goto out_close; + + val.bpf_prog.fd = prog_fd; + err =bpf_map_update_elem (map_fd, &idx, &val,0 ); +CHECK (err,"Add program to cpumap entry" ,"err %d errno %d\n" , + err, errno); + + err =bpf_map_lookup_elem (map_fd, &idx, &val); +CHECK (err,"Read cpumap entry" ,"err %d errno %d\n" , err, errno); +CHECK (info.id != val.bpf_prog.id,"Expected program id in cpumap entry" , +"expected %u read %u\n" , info.id, val.bpf_prog.id); + +/* can not attach BPF_XDP_CPUMAP program to a device */ + err =bpf_set_link_xdp_fd (IFINDEX_LO , prog_fd,XDP_FLAGS_SKB_MODE ); +CHECK (err ==0 ,"Attach of BPF_XDP_CPUMAP program" , +"should have failed\n" ); + + val.qsize =192 ; + val.bpf_prog.fd =bpf_program__fd (skel->progs.xdp_dummy_prog); + err =bpf_map_update_elem (map_fd, &idx, &val,0 );
tools/testing/selftests/bpf/prog_tests/xdp_cpumap_attach.c line 68
Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/xdp_cpumap_attach.c new file +10
-0+CHECK (err ==0 ,"Add non-BPF_XDP_CPUMAP program to cpumap entry" , +"should have failed\n" ); + +out_close: +test_xdp_with_cpumap_helpers__destroy (skel); +} + +void test_xdp_cpumap_attach (void ) +{ +if (test__start_subtest ("CPUMAP with programs in entries" ))
These subtest names are supposed to be short and follow test names (i.e., being more or less valid C identifiers). It makes it easier to select or blacklist them (with -t and -b params). So something like cpumap_with_progs or similar would be better in that regard and would make it easier for me to maintain a blacklist of tests/subtests for Travis CI, for instance.
I think there is similarly verbose DEVMAP subtest name, I'd love it to be "simplified" as well... But can't get my hands on everything, unfortunately.
Quoted 31 lines
tools/testing/selftests/bpf/prog_tests/xdp_cpumap_attach.c new file +2
-0+test_xdp_with_cpumap_helpers (); +}tools/testing/selftests/bpf/progs/test_xdp_with_cpumap_helpers.c new file +23
-0@@ -0,0 +1,38 @@ +// SPDX-License-Identifier: GPL-2.0 + +#include <linux/bpf.h> +#include <bpf/bpf_helpers.h> + +struct { + __uint(type, BPF_MAP_TYPE_CPUMAP); + __uint(key_size, sizeof(__u32)); + __uint(value_size, sizeof(struct bpf_cpumap_val)); + __uint(max_entries, 4); +} cpu_map SEC(".maps"); + +SEC("xdp_redir") +int xdp_redir_prog (struct xdp_md *ctx) +{ +return bpf_redirect_map (&cpu_map,1 ,0 ); +} + +SEC("xdp_dummy") +int xdp_dummy_prog (struct xdp_md *ctx) +{ +return XDP_PASS ; +}
tools/testing/selftests/bpf/progs/test_xdp_with_cpumap_helpers.c line 33
Quoted 10 lines
tools/testing/selftests/bpf/progs/test_xdp_with_cpumap_helpers.c new file +10
-0+ +SEC("xdp_cpumap") +int xdp_dummy_cm (struct xdp_md *ctx) +{ +char fmt[] ="devmap redirect: dev %u len %u\n" ; +void *data_end = (void *)(long )ctx->data_end; +void *data = (void *)(long )ctx->data; +unsigned int len = data_end - data; + +bpf_trace_printk (fmt,sizeof (fmt), ctx->ingress_ifindex, len);
Is there any reason to use bpf_trace_printk as opposed to saving ctx->ingress_ifindex into a global variable? bpf_trace_printk isn't really testing anything, just pollutes trace_pipe.
Quoted 8 lines
tools/testing/selftests/bpf/progs/test_xdp_with_cpumap_helpers.c new file +5
-1+ +return XDP_PASS ; +} + +char _license[] SEC("license") = "GPL"; --2.26.2