bpf: Add support for XDP programs in DEVMAPs
Subsystems (7)
cover bpf: Add support for XDP programs in DEVMAPs
Implementation of Daniel's proposal for allowing DEVMAP entries to be a device index, program id pair. Daniel suggested an fd to specify the program, but that seems odd to me that you insert the value as an fd, but read it back as an id since the fd can be closed.
David Ahern (4):
bpf: Handle 8-byte values in DEVMAP and DEVMAP_HASH
bpf: Add support to attach bpf program to a devmap
xdp: Add xdp_txq_info to xdp_buff
bpftool: Add SEC name for xdp programs attached to device map
include/linux/bpf.h | 5 ++ include/net/xdp.h | 5 ++ include/uapi/linux/bpf.h | 3 + kernel/bpf/devmap.c | 132 +++++++++++++++++++++++++++++---- net/core/dev.c | 18 +++++ net/core/filter.c | 17 +++++ tools/include/uapi/linux/bpf.h | 3 + tools/lib/bpf/libbpf.c | 2 + 8 files changed, 170 insertions(+), 15 deletions(-)
Signature · 1 line
2.21.1 (Apple Git-122.3)
RFC PATCH 1/4 bpf: Handle 8-byte values in DEVMAP and DEVMAP_HASH
Add support to DEVMAP and DEVMAP_HASH to support 8-byte values as a <device index, program id> pair. To do this, a new struct is needed in bpf_dtab_netdev to hold the values to return on lookup.
--- kernel/bpf/devmap.c | 53 ++++++++++++++++++++++++++++++++++----------- 1 file changed, 40 insertions(+), 13 deletions(-)
kernel/bpf/devmap.c +40 -13
@@ -60,12 +60,19 @@ struct xdp_dev_bulk_queue {unsigned int count; }; +/* devmap value can be dev index or dev index + prog id */ +struct dev_map_ext_val { +u32 ifindex;/* must be first for compat with 4-byte values */ +u32 prog_id; +}; +struct bpf_dtab_netdev {struct net_device *dev;/* must be first member, due to tracepoint */ struct hlist_node index_hlist;struct bpf_dtab *dtab;struct rcu_head rcu;unsigned int idx; +struct dev_map_ext_val val; };struct bpf_dtab {
@@ -108,9 +115,13 @@ static int dev_map_init_map(struct bpf_dtab *dtab, union bpf_attr *attr)u64 cost =0 ;int err; -/* check sanity of attributes */ +/* check sanity of attributes. 2 value sizes supported: +* 4 bytes: ifindex +* 8 bytes: ifindex + prog id +*/ if (attr->max_entries ==0 || attr->key_size !=4 || - attr->value_size !=4 || attr->map_flags & ~DEV_CREATE_FLAG_MASK ) + (attr->value_size !=4 && attr->value_size !=8 ) || + attr->map_flags & ~DEV_CREATE_FLAG_MASK )return -EINVAL ;/* Lookup returns a pointer straight to dev->ifindex, so make sure the
@@ -472,18 +483,15 @@ int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, struct sk_buff *skb,static void *dev_map_lookup_elem (struct bpf_map *map,void *key) {struct bpf_dtab_netdev *obj =__dev_map_lookup_elem (map, *(u32 *)key); -struct net_device *dev = obj ? obj->dev :NULL ; -return dev ? &dev->ifindex :NULL ; +return obj ? &obj->val :NULL ; }static void *dev_map_hash_lookup_elem (struct bpf_map *map,void *key) {struct bpf_dtab_netdev *obj =__dev_map_hash_lookup_elem (map, *(u32 *)key); -struct net_device *dev = obj ? obj->dev :NULL ; - -return dev ? &dev->ifindex :NULL ; +return obj ? &obj->val :NULL ; }static void __dev_map_entry_free (struct rcu_head *rcu)
@@ -552,15 +560,18 @@ static struct bpf_dtab_netdev *__dev_map_alloc_node(struct net *net,return ERR_PTR (-ENOMEM ); dev->dev =dev_get_by_index (net, ifindex); -if (!dev->dev) { -kfree (dev); -return ERR_PTR (-EINVAL ); - } +if (!dev->dev) + goto err_out; dev->idx = idx; dev->dtab = dtab; + dev->val.ifindex = ifindex; +return dev; +err_out: +kfree (dev); +return ERR_PTR (-EINVAL ); } static int __dev_map_update_elem(struct net *net, struct bpf_map *map,
@@ -568,8 +579,16 @@ static int __dev_map_update_elem(struct net *net, struct bpf_map *map, { struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);struct bpf_dtab_netdev *dev, *old_dev; -u32 ifindex = *(u32 *)value;u32 i = *(u32 *)key; +u32 ifindex; + +if (map->value_size ==4 ) { + ifindex = *(u32 *)value; + }else { +struct dev_map_ext_val *val = value; + + ifindex = val->ifindex; + }if (unlikely (map_flags >BPF_EXIST ))return -EINVAL ;
@@ -609,10 +628,18 @@ static int __dev_map_hash_update_elem(struct net *net, struct bpf_map *map, { struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);struct bpf_dtab_netdev *dev, *old_dev; -u32 ifindex = *(u32 *)value;u32 idx = *(u32 *)key;unsigned long flags;int err = -EEXIST ; +u32 ifindex; + +if (map->value_size ==4 ) { + ifindex = *(u32 *)value; + }else { +struct dev_map_ext_val *val = value; + + ifindex = val->ifindex; + }if (unlikely (map_flags >BPF_EXIST || !ifindex))return -EINVAL ;
Signature · 1 line
2.21.1 (Apple Git-122.3)
RFC PATCH 2/4 bpf: Add support to attach bpf program to a devmap
In reply to David AhernAdd BPF_XDP_DEVMAP attach type for use with programs associated with a DEVMAP.
DEVMAPs can associate a program with a device entry by setting the value to <index,id> pair. The program associated with the id must have type XDP with expected attach type BPF_XDP_DEVMAP. When a program is associated with a device index, the program is run on an XDP_REDIRECT and before the buffer is added to the per-cpu queue. At this point rxq data is still valid; the next patch adds tx device information allowing the prorgam to see both ingress and egress device indices.
XDP generic is skb based and XDP programs do not work with skb's. Block the use case by walking maps used by a program that is to be attached via xdpgeneric and fail if any of them are DEVMAP / DEVMAP_HASH with 8-bytes values.
--- include/linux/bpf.h | 5 +++ include/uapi/linux/bpf.h | 1 + kernel/bpf/devmap.c | 82 +++++++++++++++++++++++++++++++--- net/core/dev.c | 18 ++++++++ tools/include/uapi/linux/bpf.h | 1 + 5 files changed, 102 insertions(+), 5 deletions(-)
include/linux/bpf.h +5 -0
@@ -1242,6 +1242,7 @@ int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp, struct net_device *dev_rx);int dev_map_generic_redirect (struct bpf_dtab_netdev *dst,struct sk_buff *skb,struct bpf_prog *xdp_prog); +bool dev_map_can_have_prog (struct bpf_map *map);struct bpf_cpu_map_entry *__cpu_map_lookup_elem (struct bpf_map *map,u32 key);void __cpu_map_flush (void );
@@ -1355,6 +1356,10 @@ static inline struct net_device *__dev_map_hash_lookup_elem(struct bpf_map *map {return NULL ; } +static inline bool dev_map_can_have_prog (struct bpf_map *map) +{ +return false; +}static inline void __dev_flush (void ) {
include/uapi/linux/bpf.h +1 -0
@@ -224,6 +224,7 @@ enum bpf_attach_type {BPF_CGROUP_INET6_GETPEERNAME ,BPF_CGROUP_INET4_GETSOCKNAME ,BPF_CGROUP_INET6_GETSOCKNAME , +BPF_XDP_DEVMAP , __MAX_BPF_ATTACH_TYPE };
kernel/bpf/devmap.c +77 -5
182 lines not on this page · raw message
net/core/dev.c +18 -0
@@ -5410,6 +5410,18 @@ static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)struct bpf_prog *new = xdp->prog;int ret =0 ; +if (new) { +u32 i; + +/* generic XDP does not work with DEVMAPs that can +* have a bpf_prog installed +*/ +for (i =0 ; i < new->aux->used_map_cnt; i++) { +if (dev_map_can_have_prog (new->aux->used_maps[i])) +return -EINVAL ; + } + } +switch (xdp->command) {case XDP_SETUP_PROG :rcu_assign_pointer (dev->xdp_prog, new);
@@ -8825,6 +8837,12 @@ int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,return -EINVAL ; } +if (prog->expected_attach_type ==BPF_XDP_DEVMAP ) { +NL_SET_ERR_MSG (extack,"BPF_XDP_DEVMAP 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 +1 -0
@@ -224,6 +224,7 @@ enum bpf_attach_type {BPF_CGROUP_INET6_GETPEERNAME ,BPF_CGROUP_INET4_GETSOCKNAME ,BPF_CGROUP_INET6_GETSOCKNAME , +BPF_XDP_DEVMAP , __MAX_BPF_ATTACH_TYPE };
Signature · 1 line
2.21.1 (Apple Git-122.3)
RFC PATCH 3/4 xdp: Add xdp_txq_info to xdp_buff
In reply to David AhernAdd xdp_txq_info as the Tx counterpart to xdp_rxq_info. At the moment only the device is added. Other fields (queue_index) can be added as use cases arise.
From a UAPI perspective, add egress_ifindex to xdp context.
Update the verifier to only allow accesses to egress_ifindex by XDP programs with BPF_XDP_DEVMAP expected attach type.
--- include/net/xdp.h | 5 +++++ include/uapi/linux/bpf.h | 2 ++ kernel/bpf/devmap.c | 3 +++ net/core/filter.c | 17 +++++++++++++++++ tools/include/uapi/linux/bpf.h | 2 ++ 5 files changed, 29 insertions(+)
include/net/xdp.h +5 -0
@@ -65,6 +65,10 @@ struct xdp_rxq_info {struct xdp_mem_info mem; }____cacheline_aligned ;/* perf critical, avoid false-sharing */ +struct xdp_txq_info { +struct net_device *dev; +}; +struct xdp_buff {void *data;void *data_end;
@@ -72,6 +76,7 @@ struct xdp_buff {void *data_hard_start;unsigned long handle;struct xdp_rxq_info *rxq; +struct xdp_txq_info *txq;u32 frame_sz;/* frame size to deduce data_hard_end/reserved tailroom*/ };
include/uapi/linux/bpf.h +2 -0
@@ -3624,6 +3624,8 @@ struct xdp_md {/* Below access go through struct xdp_rxq_info */ __u32 ingress_ifindex;/* rxq->dev->ifindex */ __u32 rx_queue_index;/* rxq->queue_index */ + +__u32 egress_ifindex;/* txq->dev->ifindex */ };enum sk_action {
kernel/bpf/devmap.c +3 -0
@@ -471,8 +471,11 @@ static struct xdp_buff *dev_map_run_prog(struct net_device *dev,struct xdp_buff *xdp, struct bpf_prog *xdp_prog) { + struct xdp_txq_info txq = { .dev = dev };u32 act; + xdp->txq = &txq; + act =bpf_prog_run_xdp (xdp_prog, xdp);switch (act) {case XDP_DROP :
net/core/filter.c +17 -0
@@ -6994,6 +6994,13 @@ static bool xdp_is_valid_access(int off, int size,const struct bpf_prog *prog, struct bpf_insn_access_aux *info) { + if (prog->expected_attach_type != BPF_XDP_DEVMAP) { + switch (off) { + case offsetof(struct xdp_md, egress_ifindex): + return false; + } + } +if (type ==BPF_WRITE ) {if (bpf_prog_is_dev_bound (prog->aux)) {switch (off) {
@@ -7942,6 +7949,16 @@ static u32 xdp_convert_ctx_access(enum bpf_access_type type, offsetof(struct xdp_rxq_info , queue_index));break ; +case offsetof(struct xdp_md , egress_ifindex): + *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq), + si->dst_reg, si->src_reg, + offsetof(struct xdp_buff , txq)); + *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev), + si->dst_reg, si->dst_reg, + offsetof(struct xdp_txq_info , dev)); + *insn++ =BPF_LDX_MEM (BPF_W , si->dst_reg, si->dst_reg, + offsetof(struct net_device , ifindex)); +break ; }return insn - insn_buf;
tools/include/uapi/linux/bpf.h +2 -0
@@ -3624,6 +3624,8 @@ struct xdp_md {/* Below access go through struct xdp_rxq_info */ __u32 ingress_ifindex;/* rxq->dev->ifindex */ __u32 rx_queue_index;/* rxq->queue_index */ + +__u32 egress_ifindex;/* txq->dev->ifindex */ };enum sk_action {
Signature · 1 line
2.21.1 (Apple Git-122.3)
RFC PATCH 4/4 bpftool: Add SEC name for xdp programs attached to device map
In reply to David AhernSupport SEC("xdp_dm*") as a short cut for loading the program with type BPF_PROG_TYPE_XDP and expected attach type BPF_XDP_DEVMAP.
--- tools/lib/bpf/libbpf.c | 2 ++ 1 file changed, 2 insertions(+)
tools/lib/bpf/libbpf.c +2 -0
@@ -6657,6 +6657,8 @@ static const struct bpf_sec_def section_defs[] = { .expected_attach_type = BPF_TRACE_ITER, .is_attach_btf = true, .attach_fn = attach_iter), +BPF_EAPROG_SEC ("xdp_dm" ,BPF_PROG_TYPE_XDP , +BPF_XDP_DEVMAP ),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
2.21.1 (Apple Git-122.3)
Quoted 4 lines · On Thu, 21 May 2020 19:05:22 -0600 David Ahern <[email protected]> wrote:
Implementation of Daniel's proposal for allowing DEVMAP entries to be a device index, program id pair. Daniel suggested an fd to specify the program, but that seems odd to me that you insert the value as an fd, but read it back as an id since the fd can be closed.
Great that you are working on this :-)
Lorenzo (cc [email protected]) is working on a similar approach for cpumap, please coordinate/Cc him on these patches.
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 RFC bpf-next 1/4] bpf: Handle 8-byte values in DEVMAP and DEVMAP_HASH
In reply to David AhernQuoted 3 lines · On Thu, 21 May 2020 19:05:23 -0600 David Ahern <[email protected]> wrote:
Add support to DEVMAP and
DEVMAP_HASHto support 8-byte values as a <device index, program id> pair. To do this, a new struct is needed inbpf_dtab_netdevto hold the values to return on lookup.
I would like to see us leverage BTF instead of checking on the size attr->value_size. E.g do the sanity check based on BTF. Given I don't know the exact details on how this should be done, I will look into it... I already promised Lorenzo, as we have already discussed this on IRC.
So, you can Lorenzo can go ahead with this approach, and test the use-case. And I'll try to figure out if-and-how we can leverage BTF here. Input from BTF experts will be much appreciated.
Quoted 45 lines
Signed-off-by David Ahern <[email protected]>--- kernel/bpf/devmap.c | 53 ++++++++++++++++++++++++++++++++++----------- 1 file changed, 40 insertions(+), 13 deletions(-)kernel/bpf/devmap.c +13
-2@@ -60,12 +60,19 @@ struct xdp_dev_bulk_queue {unsigned int count; }; +/* devmap value can be dev index or dev index + prog id */ +struct dev_map_ext_val { +u32 ifindex;/* must be first for compat with 4-byte values */ +u32 prog_id; +}; +struct bpf_dtab_netdev {struct net_device *dev;/* must be first member, due to tracepoint */ struct hlist_node index_hlist;struct bpf_dtab *dtab;struct rcu_head rcu;unsigned int idx; +struct dev_map_ext_val val; };struct bpf_dtab {@@ -108,9 +115,13 @@ static int dev_map_init_map(struct bpf_dtab *dtab, union bpf_attr *attr)u64 cost =0 ;int err; -/* check sanity of attributes */ +/* check sanity of attributes. 2 value sizes supported: +* 4 bytes: ifindex +* 8 bytes: ifindex + prog id +*/ if (attr->max_entries ==0 || attr->key_size !=4 || - attr->value_size !=4 || attr->map_flags & ~DEV_CREATE_FLAG_MASK ) + (attr->value_size !=4 && attr->value_size !=8 ) || + attr->map_flags & ~DEV_CREATE_FLAG_MASK )return -EINVAL ;/* Lookup returns a pointer straight to dev->ifindex, so make sure the
[...]
Quoted 39 lines
static int __dev_map_update_elem(struct net *net, struct
bpf_map*map,kernel/bpf/devmap.c +18
-2@@ -568,8 +579,16 @@ static int __dev_map_update_elem(struct net *net, struct bpf_map *map, { struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);struct bpf_dtab_netdev *dev, *old_dev; -u32 ifindex = *(u32 *)value;u32 i = *(u32 *)key; +u32 ifindex; + +if (map->value_size ==4 ) { + ifindex = *(u32 *)value; + }else { +struct dev_map_ext_val *val = value; + + ifindex = val->ifindex; + }if (unlikely (map_flags >BPF_EXIST ))return -EINVAL ;@@ -609,10 +628,18 @@ static int __dev_map_hash_update_elem(struct net *net, struct bpf_map *map, { struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);struct bpf_dtab_netdev *dev, *old_dev; -u32 ifindex = *(u32 *)value;u32 idx = *(u32 *)key;unsigned long flags;int err = -EEXIST ; +u32 ifindex; + +if (map->value_size ==4 ) { + ifindex = *(u32 *)value; + }else { +struct dev_map_ext_val *val = value; + + ifindex = val->ifindex; + }if (unlikely (map_flags >BPF_EXIST || !ifindex))return -EINVAL ;
Signature · 4 lines
Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
Quoted 4 lines · David Ahern
Implementation of Daniel's proposal for allowing DEVMAP entries to be a device index, program id pair. Daniel suggested an fd to specify the program, but that seems odd to me that you insert the value as an fd, but read it back as an id since the fd can be closed.
While I can be sympathetic to the argument that it seems odd, every other API uses FD for insert and returns ID, so why make it different here? Also, the choice has privilege implications, since the CAP_BPF series explicitly makes going from ID->FD a more privileged operation than just querying the ID.
-Toke
Re: [PATCH RFC bpf-next 2/4] bpf: Add support to attach bpf program to a devmap
In reply to David AhernQuoted 181 lines · David Ahern
Add
BPF_XDP_DEVMAPattach type for use with programs associated with a DEVMAP.DEVMAPs can associate a program with a device entry by setting the value to <index,id> pair. The program associated with the id must have type XDP with expected attach type
BPF_XDP_DEVMAP. When a program is associated with a device index, the program is run on anXDP_REDIRECTand before the buffer is added to the per-cpu queue. At this point rxq data is still valid; the next patch adds tx device information allowing the prorgam to see both ingress and egress device indices.XDP generic is skb based and XDP programs do not work with skb's. Block the use case by walking maps used by a program that is to be attached via xdpgeneric and fail if any of them are DEVMAP /
DEVMAP_HASHwith 8-bytes values.Signed-off-by David Ahern <[email protected]>--- include/linux/bpf.h | 5 +++ include/uapi/linux/bpf.h | 1 + kernel/bpf/devmap.c | 82 +++++++++++++++++++++++++++++++--- net/core/dev.c | 18 ++++++++ tools/include/uapi/linux/bpf.h | 1 + 5 files changed, 102 insertions(+), 5 deletions(-)include/linux/bpf.h +5
-0@@ -1242,6 +1242,7 @@ int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp, struct net_device *dev_rx);int dev_map_generic_redirect (struct bpf_dtab_netdev *dst,struct sk_buff *skb,struct bpf_prog *xdp_prog); +bool dev_map_can_have_prog (struct bpf_map *map);struct bpf_cpu_map_entry *__cpu_map_lookup_elem (struct bpf_map *map,u32 key);void __cpu_map_flush (void );@@ -1355,6 +1356,10 @@ static inline struct net_device *__dev_map_hash_lookup_elem(struct bpf_map *map {return NULL ; } +static inline bool dev_map_can_have_prog (struct bpf_map *map) +{ +return false; +}static inline void __dev_flush (void ) {include/uapi/linux/bpf.h +1
-0@@ -224,6 +224,7 @@ enum bpf_attach_type {BPF_CGROUP_INET6_GETPEERNAME ,BPF_CGROUP_INET4_GETSOCKNAME ,BPF_CGROUP_INET6_GETSOCKNAME , +BPF_XDP_DEVMAP , __MAX_BPF_ATTACH_TYPE };kernel/bpf/devmap.c +57
-1@@ -70,6 +70,7 @@ struct bpf_dtab_netdev {struct net_device *dev;/* must be first member, due to tracepoint */ struct hlist_node index_hlist;struct bpf_dtab *dtab; +struct bpf_prog *xdp_prog;struct rcu_head rcu;unsigned int idx;struct dev_map_ext_val val;@@ -228,6 +229,8 @@ static void dev_map_free(struct bpf_map *map)hlist_for_each_entry_safe (dev, next, head, index_hlist) {hlist_del_rcu (&dev->index_hlist); +if (dev->xdp_prog) +bpf_prog_put (dev->xdp_prog);dev_put (dev->dev);kfree (dev); }@@ -242,6 +245,8 @@ static void dev_map_free(struct bpf_map *map)if (!dev)continue ; +if (dev->xdp_prog) +bpf_prog_put (dev->xdp_prog);dev_put (dev->dev);kfree (dev); }@@ -328,6 +333,16 @@ static int dev_map_hash_get_next_key(struct bpf_map *map, void *key,return -ENOENT ; } +bool dev_map_can_have_prog (struct bpf_map *map) +{ +if ((map->map_type ==BPF_MAP_TYPE_DEVMAP || + map->map_type ==BPF_MAP_TYPE_DEVMAP_HASH ) && + map->value_size ==8 ) +return true; + +return false; +} +static int bq_xmit_all (struct xdp_dev_bulk_queue *bq,u32 flags) {struct net_device *dev = bq->dev;@@ -452,6 +467,39 @@ static inline int __xdp_enqueue(struct net_device *dev, struct xdp_buff *xdp,return bq_enqueue (dev, xdpf, dev_rx); } +static struct xdp_buff *dev_map_run_prog (struct net_device *dev, +struct xdp_buff *xdp, +struct bpf_prog *xdp_prog) +{ +u32 act; + + act =bpf_prog_run_xdp (xdp_prog, xdp); +switch (act) { +case XDP_DROP : + fallthrough; +case XDP_PASS : +break ; +case XDP_TX : + fallthrough; +case XDP_REDIRECT : + fallthrough; +default : +bpf_warn_invalid_xdp_action (act); + fallthrough; +case XDP_ABORTED : +trace_xdp_exception (dev, xdp_prog, act); + act =XDP_DROP ; +break ; + } + +if (act ==XDP_DROP ) { +xdp_return_buff (xdp); + xdp =NULL ; + } + +return xdp; +} +int dev_xdp_enqueue (struct net_device *dev,struct xdp_buff *xdp,struct net_device *dev_rx) {@@ -463,6 +511,11 @@ int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp, {struct net_device *dev = dst->dev; +if (dst->xdp_prog) { + xdp =dev_map_run_prog (dev, xdp, dst->xdp_prog); +if (!xdp) +return 0 ; + }return __xdp_enqueue (dev, xdp, dev_rx); }@@ -499,6 +552,8 @@ static void __dev_map_entry_free(struct rcu_head *rcu)struct bpf_dtab_netdev *dev; dev = container_of(rcu, struct bpf_dtab_netdev, rcu); +if (dev->xdp_prog) +bpf_prog_put (dev->xdp_prog);dev_put (dev->dev);kfree (dev); }@@ -549,9 +604,10 @@ static int dev_map_hash_delete_elem(struct bpf_map *map, void *key)static struct bpf_dtab_netdev *__dev_map_alloc_node (struct net *net,struct bpf_dtab *dtab, -u32 ifindex, +u32 ifindex,u32 prog_id,unsigned int idx) { +struct bpf_prog *prog =NULL ;struct bpf_dtab_netdev *dev; dev = kmalloc_node(sizeof(*dev), GFP_ATOMIC | __GFP_NOWARN,
kernel/bpf/devmap.c static struct bpf_dtab_netdev *__dev_map_alloc_node(struct net *net, · line 625
Quoted 7 lines
kernel/bpf/devmap.c +4
-0@@ -563,12 +619,23 @@ static struct bpf_dtab_netdev *__dev_map_alloc_node(struct net *net,if (!dev->dev) goto err_out; +if (prog_id) { + prog =bpf_prog_by_id (prog_id); +if (IS_ERR (prog) || prog->type !=BPF_PROG_TYPE_XDP || + prog->expected_attach_type !=BPF_XDP_DEVMAP )
If the second or third condition fails here you never bpf_prog_put() the prog ref.
Quoted 123 lines
kernel/bpf/devmap.c +16
-4+ goto err_dev; + } + dev->idx = idx; dev->dtab = dtab; + dev->xdp_prog = prog; dev->val.ifindex = ifindex; + dev->val.prog_id = prog_id;return dev; +err_dev: +dev_put (dev->dev); err_out:kfree (dev);return ERR_PTR (-EINVAL );@@ -580,7 +647,7 @@ static int __dev_map_update_elem(struct net *net, struct bpf_map *map, struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);struct bpf_dtab_netdev *dev, *old_dev;u32 i = *(u32 *)key; -u32 ifindex; +u32 ifindex, id =0 ;if (map->value_size ==4 ) { ifindex = *(u32 *)value;@@ -588,6 +655,7 @@ static int __dev_map_update_elem(struct net *net, struct bpf_map *map,struct dev_map_ext_val *val = value; ifindex = val->ifindex; + id = val->prog_id; }if (unlikely (map_flags >BPF_EXIST ))@@ -599,8 +667,11 @@ static int __dev_map_update_elem(struct net *net, struct bpf_map *map,if (!ifindex) { dev =NULL ; +/* can not specify id if ifindex is 0 */ +if (id) +return -EINVAL ; }else { - dev =__dev_map_alloc_node (net, dtab, ifindex, i); + dev =__dev_map_alloc_node (net, dtab, ifindex, id, i);if (IS_ERR (dev))return PTR_ERR (dev); }@@ -630,8 +701,8 @@ static int __dev_map_hash_update_elem(struct net *net, struct bpf_map *map,struct bpf_dtab_netdev *dev, *old_dev;u32 idx = *(u32 *)key;unsigned long flags; +u32 ifindex, id =0 ;int err = -EEXIST ; -u32 ifindex;if (map->value_size ==4 ) { ifindex = *(u32 *)value;@@ -639,6 +710,7 @@ static int __dev_map_hash_update_elem(struct net *net, struct bpf_map *map,struct dev_map_ext_val *val = value; ifindex = val->ifindex; + id = val->prog_id; }if (unlikely (map_flags >BPF_EXIST || !ifindex))@@ -650,7 +722,7 @@ static int __dev_map_hash_update_elem(struct net *net, struct bpf_map *map,if (old_dev && (map_flags &BPF_NOEXIST )) goto out_err; - dev =__dev_map_alloc_node (net, dtab, ifindex, idx); + dev =__dev_map_alloc_node (net, dtab, ifindex, id, idx);if (IS_ERR (dev)) { err =PTR_ERR (dev); goto out_err;net/core/dev.c +18
-0@@ -5410,6 +5410,18 @@ static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)struct bpf_prog *new = xdp->prog;int ret =0 ; +if (new) { +u32 i; + +/* generic XDP does not work with DEVMAPs that can +* have a bpf_prog installed +*/ +for (i =0 ; i < new->aux->used_map_cnt; i++) { +if (dev_map_can_have_prog (new->aux->used_maps[i])) +return -EINVAL ; + } + } +switch (xdp->command) {case XDP_SETUP_PROG :rcu_assign_pointer (dev->xdp_prog, new);@@ -8825,6 +8837,12 @@ int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,return -EINVAL ; } +if (prog->expected_attach_type ==BPF_XDP_DEVMAP ) { +NL_SET_ERR_MSG (extack,"BPF_XDP_DEVMAP 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 +1
-0@@ -224,6 +224,7 @@ enum bpf_attach_type {BPF_CGROUP_INET6_GETPEERNAME ,BPF_CGROUP_INET4_GETSOCKNAME ,BPF_CGROUP_INET6_GETSOCKNAME , +BPF_XDP_DEVMAP , __MAX_BPF_ATTACH_TYPE };Signature · 1 line
2.21.1 (Apple Git-122.3)
Re: [PATCH RFC bpf-next 3/4] xdp: Add xdp_txq_info to xdp_buff
In reply to David AhernQuoted 10 lines · David Ahern
Add
xdp_txq_infoas the Tx counterpart toxdp_rxq_info. At the moment only the device is added. Other fields (queue_index) can be added as use cases arise.From a UAPI perspective, add
egress_ifindexto xdp context.Update the verifier to only allow accesses to
egress_ifindexby XDP programs withBPF_XDP_DEVMAPexpected attach type.Signed-off-by David Ahern <[email protected]>
Nit: seems a bit odd to order the patches like this, I'd have expected this patch to go before #2.
-Toke
Re: [PATCH RFC bpf-next 1/4] bpf: Handle 8-byte values in DEVMAP and DEVMAP_HASH
In reply to Jesper Dangaard BrouerQuoted 13 lines · On Fri, 22 May 2020 14:08:05 +0200 Jesper Dangaard Brouer <[email protected]>…
Quoted 3 lines · On Thu, 21 May 2020 19:05:23 -0600 David Ahern <[email protected]> wrote:
Add support to DEVMAP and
DEVMAP_HASHto support 8-byte values as a <device index, program id> pair. To do this, a new struct is needed inbpf_dtab_netdevto hold the values to return on lookup.I would like to see us leverage BTF instead of checking on the size
attr->value_size. E.g do the sanity check based on BTF. Given I don't know the exact details on how this should be done, I will look into it... I already promised Lorenzo, as we have already discussed this on IRC.
Quoted 3 lines
So, you can Lorenzo can go ahead with this approach, and test the use-case. And I'll try to figure out if-and-how we can leverage BTF here. Input from BTF experts will be much appreciated.
Published my current notes here:
https://github.com/xdp-project/xdp-project/blob/BTF01-notes.public/areas/core/BTF_01_notes.org
And created PR that people can GitHub "subscribe" to, if you are interested:
https://github.com/xdp-project/xdp-project/pull/36
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 RFC bpf-next 2/4] bpf: Add support to attach bpf program to a devmap
In reply to Toke Høiland-JørgensenQuoted 4 lines · Toke Høiland-Jørgensen
Quoted 4 lines
kernel/bpf/devmap.c +0
-0@@ -563,12 +619,23 @@ static struct bpf_dtab_netdev *__dev_map_alloc_node(struct net *net,if (!dev->dev) goto err_out;
Quoted 8 lines
kernel/bpf/devmap.c static struct bpf_dtab_netdev *__dev_map_alloc_node(struct net *net, · line 625
Quoted 4 lines
kernel/bpf/devmap.c +4
-0+if (prog_id) { + prog =bpf_prog_by_id (prog_id); +if (IS_ERR (prog) || prog->type !=BPF_PROG_TYPE_XDP || + prog->expected_attach_type !=BPF_XDP_DEVMAP )If the second or third condition fails here you never
bpf_prog_put()the prog ref.
good catch. Thanks, will fix.
Re: [PATCH RFC bpf-next 3/4] xdp: Add xdp_txq_info to xdp_buff
In reply to Toke Høiland-JørgensenQuoted 8 lines · Toke Høiland-Jørgensen
Quoted 6 lines · David Ahern
Add
xdp_txq_infoas the Tx counterpart toxdp_rxq_info. At the moment only the device is added. Other fields (queue_index) can be added as use cases arise.From a UAPI perspective, add
egress_ifindexto xdp context.
Quoted 8 lines
Quoted 4 lines
Update the verifier to only allow accesses to
egress_ifindexby XDP programs withBPF_XDP_DEVMAPexpected attach type.Signed-off-by David Ahern <[email protected]>Nit: seems a bit odd to order the patches like this, I'd have expected this patch to go before #2.
This patch depends on patch 2, so it needs to follow it.
Quoted 7 lines · Toke Høiland-Jørgensen
Quoted 4 lines · David Ahern
Implementation of Daniel's proposal for allowing DEVMAP entries to be a device index, program id pair. Daniel suggested an fd to specify the program, but that seems odd to me that you insert the value as an fd, but read it back as an id since the fd can be closed.
Quoted 6 lines
While I can be sympathetic to the argument that it seems odd, every other API uses FD for insert and returns ID, so why make it different here? Also, the choice has privilege implications, since the
CAP_BPFseries explicitly makes going fromID->FDa more privileged operation than just querying the ID.
I do not like the model where the kernel changes the value the user pushed down.
Re: [PATCH RFC bpf-next 1/4] bpf: Handle 8-byte values in DEVMAP and DEVMAP_HASH
In reply to Jesper Dangaard BrouerQuoted 20 lines · Jesper Dangaard Brouer
Quoted 13 lines · On Fri, 22 May 2020 14:08:05 +0200 Jesper Dangaard Brouer <[email protected]>…
Quoted 3 lines · On Thu, 21 May 2020 19:05:23 -0600 David Ahern <[email protected]> wrote:
Add support to DEVMAP and
DEVMAP_HASHto support 8-byte values as a <device index, program id> pair. To do this, a new struct is needed inbpf_dtab_netdevto hold the values to return on lookup.I would like to see us leverage BTF instead of checking on the size
attr->value_size. E.g do the sanity check based on BTF. Given I don't know the exact details on how this should be done, I will look into it... I already promised Lorenzo, as we have already discussed this on IRC.Quoted 3 lines
So, you can Lorenzo can go ahead with this approach, and test the use-case. And I'll try to figure out if-and-how we can leverage BTF here. Input from BTF experts will be much appreciated.
Quoted 6 lines
Published my current notes here:
https://github.com/xdp-project/xdp-project/blob/BTF01-notes.public/areas/core/BTF_01_notes.orgAnd created PR that people can GitHub "subscribe" to, if you are interested:
https://github.com/xdp-project/xdp-project/pull/36
thanks for compiling some notes.
Fundamentally, I do not see how this can work for something like the program id where the kernel needs to know not just the type (u32) but that it should take the id and do a lookup - convert the id into a bpf_prog reference.
Quoted 15 lines · David Ahern
Quoted 7 lines · Toke Høiland-Jørgensen
Quoted 4 lines · David Ahern
Implementation of Daniel's proposal for allowing DEVMAP entries to be a device index, program id pair. Daniel suggested an fd to specify the program, but that seems odd to me that you insert the value as an fd, but read it back as an id since the fd can be closed.
Quoted 6 lines
While I can be sympathetic to the argument that it seems odd, every other API uses FD for insert and returns ID, so why make it different here? Also, the choice has privilege implications, since the
CAP_BPFseries explicitly makes going fromID->FDa more privileged operation than just querying the ID.
Quoted 2 lines
I do not like the model where the kernel changes the value the user pushed down.
Yet it's what we do in every other interface where a user needs to supply a program, including in prog array maps. So let's not create a new inconsistent interface here...
-Toke
Quoted 10 lines · On Mon, 25 May 2020 14:15:32 +0200 Toke Høiland-Jørgensen <[email protected]>…
Quoted 8 lines · David Ahern
Quoted 7 lines · Toke Høiland-Jørgensen
Quoted 4 lines · David Ahern
Implementation of Daniel's proposal for allowing DEVMAP entries to be a device index, program id pair. Daniel suggested an fd to specify the program, but that seems odd to me that you insert the value as an fd, but read it back as an id since the fd can be closed.
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
While I can be sympathetic to the argument that it seems odd, every other API uses FD for insert and returns ID, so why make it different here? Also, the choice has privilege implications, since the
CAP_BPFseries explicitly makes going fromID->FDa more privileged operation than just querying the ID.
Sorry, I don't follow.
Can someone explain why is inserting an ID is a privilege problem?
Quoted 7 lines
Quoted 3 lines
I do not like the model where the kernel changes the value the user pushed down.
Yet it's what we do in every other interface where a user needs to supply a program, including in prog array maps. So let's not create a new inconsistent interface here...
I sympathize with Ahern on this. It seems very weird to insert/write one value-type, but read another value-type.
Signature · 4 lines
Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
Quoted 13 lines · Jesper Dangaard Brouer
Quoted 10 lines · On Mon, 25 May 2020 14:15:32 +0200 Toke Høiland-Jørgensen <[email protected]>…
Quoted 8 lines · David Ahern
Quoted 7 lines · Toke Høiland-Jørgensen
Quoted 4 lines · David Ahern
Implementation of Daniel's proposal for allowing DEVMAP entries to be a device index, program id pair. Daniel suggested an fd to specify the program, but that seems odd to me that you insert the value as an fd, but read it back as an id since the fd can be closed.
Quoted 8 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
While I can be sympathetic to the argument that it seems odd, every other API uses FD for insert and returns ID, so why make it different here? Also, the choice has privilege implications, since the
CAP_BPFseries explicitly makes going fromID->FDa more privileged operation than just querying the ID.Sorry, I don't follow.
Can someone explain why is inserting an ID is a privilege problem?
See description here:
https://lore.kernel.org/bpf/[email protected]/
Specifically, this part:
Quoted 15 lines
Consolidating all CAP checks at load time makes security model similar to
open()syscall. Once the user got an FD it can do everything with it. read/write/poll don't check permissions. The same way whenbpf_prog_loadcommand returns an FD the user can do everything (including attaching, detaching, andbpf_test_run).The important design decision is to allow
ID->FDtransition forCAP_SYS_ADMINonly. What it means that user processes can run withCAP_BPFandCAP_NET_ADMINand they will not be able to affect each other unless they pass FDs viascm_rightsor via pinning in bpffs.Quoted 3 lines
Quoted 2 lines
I do not like the model where the kernel changes the value the user pushed down.
Quoted 6 lines
Quoted 3 lines
Yet it's what we do in every other interface where a user needs to supply a program, including in prog array maps. So let's not create a new inconsistent interface here...
I sympathize with Ahern on this. It seems very weird to insert/write one value-type, but read another value-type.
Yeah, I do kinda agree that it's a bit weird. But it's what we do everywhere else, so I think consistency wins out here. There might be an argument that maps should be different (because they're conceptually a read/write data structure not a syscall return value). But again, we already have a map type that takes prog IDs, and that already does the rewriting, so doing it different here would be even weirder...
-Toke
Daniel Borkmann
maintainerQuoted 17 lines · Toke Høiland-Jørgensen
Quoted 16 lines · Jesper Dangaard Brouer
Quoted 14 lines · On Mon, 25 May 2020 14:15:32 +0200 Toke Høiland-Jørgensen <[email protected]>…
Quoted 13 lines · David Ahern
Quoted 12 lines · Toke Høiland-Jørgensen
Quoted 4 lines · David Ahern
Implementation of Daniel's proposal for allowing DEVMAP entries to be a device index, program id pair. Daniel suggested an fd to specify the program, but that seems odd to me that you insert the value as an fd, but read it back as an id since the fd can be closed.
While I can be sympathetic to the argument that it seems odd, every other API uses FD for insert and returns ID, so why make it different here? Also, the choice has privilege implications, since the
CAP_BPFseries explicitly makes going fromID->FDa more privileged operation than just querying the ID.
[...]
Quoted 9 lines
Quoted 2 lines
I sympathize with Ahern on this. It seems very weird to insert/write one value-type, but read another value-type.
Yeah, I do kinda agree that it's a bit weird. But it's what we do everywhere else, so I think consistency wins out here. There might be an argument that maps should be different (because they're conceptually a read/write data structure not a syscall return value). But again, we already have a map type that takes prog IDs, and that already does the rewriting, so doing it different here would be even weirder...
Sorry for the late reply. Agree, it would at least be consistent to what is done in tail call maps, and the XDP netlink API where you have the fd->id in both cases. Either way, quick glance over the patches, the direction of this RFC looks good to me, better fit than the prior XDP egress approaches.
Thanks,
Daniel