From ab3f0063c48c26c927851b6767824e35a716d878 Mon Sep 17 00:00:00 2001 From: Jakub Kicinski Date: Fri, 3 Nov 2017 13:56:17 -0700 Subject: bpf: offload: add infrastructure for loading programs for a specific netdev The fact that we don't know which device the program is going to be used on is quite limiting in current eBPF infrastructure. We have to reverse or limit the changes which kernel makes to the loaded bytecode if we want it to be offloaded to a networking device. We also have to invent new APIs for debugging and troubleshooting support. Make it possible to load programs for a specific netdev. This helps us to bring the debug information closer to the core eBPF infrastructure (e.g. we will be able to reuse the verifer log in device JIT). It allows device JITs to perform translation on the original bytecode. __bpf_prog_get() when called to get a reference for an attachment point will now refuse to give it if program has a device assigned. Following patches will add a version of that function which passes the expected netdev in. @type argument in __bpf_prog_get() is renamed to attach_type to make it clearer that it's only set on attachment. All calls to ndo_bpf are protected by rtnl, only verifier callbacks are not. We need a wait queue to make sure netdev doesn't get destroyed while verifier is still running and calling its driver. Signed-off-by: Jakub Kicinski Reviewed-by: Simon Horman Reviewed-by: Quentin Monnet Signed-off-by: David S. Miller --- kernel/bpf/offload.c | 182 +++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 182 insertions(+) create mode 100644 kernel/bpf/offload.c (limited to 'kernel/bpf/offload.c') diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c new file mode 100644 index 000000000000..5553e0e2f8b1 --- /dev/null +++ b/kernel/bpf/offload.c @@ -0,0 +1,182 @@ +#include +#include +#include +#include +#include +#include +#include + +/* protected by RTNL */ +static LIST_HEAD(bpf_prog_offload_devs); + +int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr) +{ + struct net *net = current->nsproxy->net_ns; + struct bpf_dev_offload *offload; + + if (!capable(CAP_SYS_ADMIN)) + return -EPERM; + + if (attr->prog_flags) + return -EINVAL; + + offload = kzalloc(sizeof(*offload), GFP_USER); + if (!offload) + return -ENOMEM; + + offload->prog = prog; + init_waitqueue_head(&offload->verifier_done); + + rtnl_lock(); + offload->netdev = __dev_get_by_index(net, attr->prog_target_ifindex); + if (!offload->netdev) { + rtnl_unlock(); + kfree(offload); + return -EINVAL; + } + + prog->aux->offload = offload; + list_add_tail(&offload->offloads, &bpf_prog_offload_devs); + rtnl_unlock(); + + return 0; +} + +static int __bpf_offload_ndo(struct bpf_prog *prog, enum bpf_netdev_command cmd, + struct netdev_bpf *data) +{ + struct net_device *netdev = prog->aux->offload->netdev; + + ASSERT_RTNL(); + + if (!netdev) + return -ENODEV; + if (!netdev->netdev_ops->ndo_bpf) + return -EOPNOTSUPP; + + data->command = cmd; + + return netdev->netdev_ops->ndo_bpf(netdev, data); +} + +int bpf_prog_offload_verifier_prep(struct bpf_verifier_env *env) +{ + struct netdev_bpf data = {}; + int err; + + data.verifier.prog = env->prog; + + rtnl_lock(); + err = __bpf_offload_ndo(env->prog, BPF_OFFLOAD_VERIFIER_PREP, &data); + if (err) + goto exit_unlock; + + env->dev_ops = data.verifier.ops; + + env->prog->aux->offload->dev_state = true; + env->prog->aux->offload->verifier_running = true; +exit_unlock: + rtnl_unlock(); + return err; +} + +static void __bpf_prog_offload_destroy(struct bpf_prog *prog) +{ + struct bpf_dev_offload *offload = prog->aux->offload; + struct netdev_bpf data = {}; + + data.offload.prog = prog; + + if (offload->verifier_running) + wait_event(offload->verifier_done, !offload->verifier_running); + + if (offload->dev_state) + WARN_ON(__bpf_offload_ndo(prog, BPF_OFFLOAD_DESTROY, &data)); + + offload->dev_state = false; + list_del_init(&offload->offloads); + offload->netdev = NULL; +} + +void bpf_prog_offload_destroy(struct bpf_prog *prog) +{ + struct bpf_dev_offload *offload = prog->aux->offload; + + offload->verifier_running = false; + wake_up(&offload->verifier_done); + + rtnl_lock(); + __bpf_prog_offload_destroy(prog); + rtnl_unlock(); + + kfree(offload); +} + +static int bpf_prog_offload_translate(struct bpf_prog *prog) +{ + struct bpf_dev_offload *offload = prog->aux->offload; + struct netdev_bpf data = {}; + int ret; + + data.offload.prog = prog; + + offload->verifier_running = false; + wake_up(&offload->verifier_done); + + rtnl_lock(); + ret = __bpf_offload_ndo(prog, BPF_OFFLOAD_TRANSLATE, &data); + rtnl_unlock(); + + return ret; +} + +static unsigned int bpf_prog_warn_on_exec(const void *ctx, + const struct bpf_insn *insn) +{ + WARN(1, "attempt to execute device eBPF program on the host!"); + return 0; +} + +int bpf_prog_offload_compile(struct bpf_prog *prog) +{ + prog->bpf_func = bpf_prog_warn_on_exec; + + return bpf_prog_offload_translate(prog); +} + +const struct bpf_prog_ops bpf_offload_prog_ops = { +}; + +static int bpf_offload_notification(struct notifier_block *notifier, + ulong event, void *ptr) +{ + struct net_device *netdev = netdev_notifier_info_to_dev(ptr); + struct bpf_dev_offload *offload, *tmp; + + ASSERT_RTNL(); + + switch (event) { + case NETDEV_UNREGISTER: + list_for_each_entry_safe(offload, tmp, &bpf_prog_offload_devs, + offloads) { + if (offload->netdev == netdev) + __bpf_prog_offload_destroy(offload->prog); + } + break; + default: + break; + } + return NOTIFY_OK; +} + +static struct notifier_block bpf_offload_notifier = { + .notifier_call = bpf_offload_notification, +}; + +static int __init bpf_offload_init(void) +{ + register_netdevice_notifier(&bpf_offload_notifier); + return 0; +} + +subsys_initcall(bpf_offload_init); -- cgit v1.2.3 From bd601b6ada11fdfb9e277f24ad2eb54bc599156b Mon Sep 17 00:00:00 2001 From: Jakub Kicinski Date: Fri, 3 Nov 2017 13:56:18 -0700 Subject: bpf: report offload info to user space Extend struct bpf_prog_info to contain information about program being bound to a device. Since the netdev may get destroyed while program still exists we need a flag to indicate the program is loaded for a device, even if the device is gone. Signed-off-by: Jakub Kicinski Reviewed-by: Simon Horman Reviewed-by: Quentin Monnet Signed-off-by: David S. Miller --- kernel/bpf/offload.c | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'kernel/bpf/offload.c') diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c index 5553e0e2f8b1..2816feb38be1 100644 --- a/kernel/bpf/offload.c +++ b/kernel/bpf/offload.c @@ -144,6 +144,18 @@ int bpf_prog_offload_compile(struct bpf_prog *prog) return bpf_prog_offload_translate(prog); } +u32 bpf_prog_offload_ifindex(struct bpf_prog *prog) +{ + struct bpf_dev_offload *offload = prog->aux->offload; + u32 ifindex; + + rtnl_lock(); + ifindex = offload->netdev ? offload->netdev->ifindex : 0; + rtnl_unlock(); + + return ifindex; +} + const struct bpf_prog_ops bpf_offload_prog_ops = { }; -- cgit v1.2.3 From 13a9c48a85ccf1417b527975c0a12b47fbfaf625 Mon Sep 17 00:00:00 2001 From: Jakub Kicinski Date: Mon, 20 Nov 2017 15:21:51 -0800 Subject: bpf: offload: add comment warning developers about double destroy Offload state may get destroyed either because the device for which it was constructed is going away, or because the refcount of bpf program itself has reached 0. In both of those cases we will call __bpf_prog_offload_destroy() to unlink the offload from the device. We may in fact call it twice, which works just fine, but we should make clear this is intended and caution others trying to extend the function. Signed-off-by: Jakub Kicinski Reviewed-by: Quentin Monnet Acked-by: Alexei Starovoitov Acked-by: Daniel Borkmann Signed-off-by: Daniel Borkmann --- kernel/bpf/offload.c | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'kernel/bpf/offload.c') diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c index 2816feb38be1..fd696d3dd429 100644 --- a/kernel/bpf/offload.c +++ b/kernel/bpf/offload.c @@ -85,6 +85,10 @@ static void __bpf_prog_offload_destroy(struct bpf_prog *prog) struct bpf_dev_offload *offload = prog->aux->offload; struct netdev_bpf data = {}; + /* Caution - if netdev is destroyed before the program, this function + * will be called twice. + */ + data.offload.prog = prog; if (offload->verifier_running) -- cgit v1.2.3 From 649f11dcd19a5f0d00fdbc760fbdccdd98e56a43 Mon Sep 17 00:00:00 2001 From: Jakub Kicinski Date: Mon, 20 Nov 2017 15:21:52 -0800 Subject: bpf: offload: limit offload to cls_bpf and xdp programs only We are currently only allowing attachment of device-bound cls_bpf and XDP programs. Make this restriction explicit in the BPF offload code. This way we can potentially reuse the ifindex field in the future. Since XDP and cls_bpf programs can only be loaded by admin, we can drop the explicit capability check from offload code. Signed-off-by: Jakub Kicinski Reviewed-by: Quentin Monnet Acked-by: Alexei Starovoitov Acked-by: Daniel Borkmann Signed-off-by: Daniel Borkmann --- kernel/bpf/offload.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'kernel/bpf/offload.c') diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c index fd696d3dd429..ac187f9ee182 100644 --- a/kernel/bpf/offload.c +++ b/kernel/bpf/offload.c @@ -14,8 +14,9 @@ int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr) struct net *net = current->nsproxy->net_ns; struct bpf_dev_offload *offload; - if (!capable(CAP_SYS_ADMIN)) - return -EPERM; + if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS && + attr->prog_type != BPF_PROG_TYPE_XDP) + return -EINVAL; if (attr->prog_flags) return -EINVAL; -- cgit v1.2.3 From 1f6f4cb7ba219b00a3fa9afe8049fa16444d8b52 Mon Sep 17 00:00:00 2001 From: Jakub Kicinski Date: Mon, 20 Nov 2017 15:21:53 -0800 Subject: bpf: offload: rename the ifindex field bpf_target_prog seems long and clunky, rename it to prog_ifindex. We don't want to call this field just ifindex, because maps may need a similar field in the future and bpf_attr members for programs and maps are unnamed. Signed-off-by: Jakub Kicinski Reviewed-by: Quentin Monnet Acked-by: Alexei Starovoitov Acked-by: Daniel Borkmann Signed-off-by: Daniel Borkmann --- kernel/bpf/offload.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel/bpf/offload.c') diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c index ac187f9ee182..a778e5df7e26 100644 --- a/kernel/bpf/offload.c +++ b/kernel/bpf/offload.c @@ -29,7 +29,7 @@ int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr) init_waitqueue_head(&offload->verifier_done); rtnl_lock(); - offload->netdev = __dev_get_by_index(net, attr->prog_target_ifindex); + offload->netdev = __dev_get_by_index(net, attr->prog_ifindex); if (!offload->netdev) { rtnl_unlock(); kfree(offload); -- cgit v1.2.3 From 62c71b45e8537b8cb746cc929ea05ba0258e0b5a Mon Sep 17 00:00:00 2001 From: Jakub Kicinski Date: Mon, 20 Nov 2017 15:21:57 -0800 Subject: bpf: offload: ignore namespace moves We are currently destroying the device offload state when device moves to another net namespace. This doesn't break with current NFP code, because offload state is not used on program removal, but it's not correct behaviour. Ignore the device unregister notifications on namespace move. Signed-off-by: Jakub Kicinski Signed-off-by: Daniel Borkmann --- kernel/bpf/offload.c | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'kernel/bpf/offload.c') diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c index a778e5df7e26..d4267c674fec 100644 --- a/kernel/bpf/offload.c +++ b/kernel/bpf/offload.c @@ -174,6 +174,10 @@ static int bpf_offload_notification(struct notifier_block *notifier, switch (event) { case NETDEV_UNREGISTER: + /* ignore namespace changes */ + if (netdev->reg_state != NETREG_UNREGISTERING) + break; + list_for_each_entry_safe(offload, tmp, &bpf_prog_offload_devs, offloads) { if (offload->netdev == netdev) -- cgit v1.2.3 From 1ee640095f049e7ac4ec36b985abada497b98cc2 Mon Sep 17 00:00:00 2001 From: Jakub Kicinski Date: Mon, 20 Nov 2017 15:21:59 -0800 Subject: bpf: revert report offload info to user space This reverts commit bd601b6ada11 ("bpf: report offload info to user space"). The ifindex by itself is not sufficient, we should provide information on which network namespace this ifindex belongs to. After considering some options we concluded that it's best to just remove this API for now, and rework it in -next. Signed-off-by: Jakub Kicinski Signed-off-by: Daniel Borkmann --- kernel/bpf/offload.c | 12 ------------ 1 file changed, 12 deletions(-) (limited to 'kernel/bpf/offload.c') diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c index d4267c674fec..68ec884440b7 100644 --- a/kernel/bpf/offload.c +++ b/kernel/bpf/offload.c @@ -149,18 +149,6 @@ int bpf_prog_offload_compile(struct bpf_prog *prog) return bpf_prog_offload_translate(prog); } -u32 bpf_prog_offload_ifindex(struct bpf_prog *prog) -{ - struct bpf_dev_offload *offload = prog->aux->offload; - u32 ifindex; - - rtnl_lock(); - ifindex = offload->netdev ? offload->netdev->ifindex : 0; - rtnl_unlock(); - - return ifindex; -} - const struct bpf_prog_ops bpf_offload_prog_ops = { }; -- cgit v1.2.3