Commit 8daed767 authored by Daniel Borkmann's avatar Daniel Borkmann
Browse files

Merge branch 'bpf-lookup-devmap' 



Toke Høiland-Jørgensen says:

====================
When using the bpf_redirect_map() helper to redirect packets from XDP, the eBPF
program cannot currently know whether the redirect will succeed, which makes it
impossible to gracefully handle errors. To properly fix this will probably
require deeper changes to the way TX resources are allocated, but one thing that
is fairly straight forward to fix is to allow lookups into devmaps, so programs
can at least know when a redirect is *guaranteed* to fail because there is no
entry in the map. Currently, programs work around this by keeping a shadow map
of another type which indicates whether a map index is valid.

This series contains two changes that are complementary ways to fix this issue:

- Moving the map lookup into the bpf_redirect_map() helper (and caching the
  result), so the helper can return an error if no value is found in the map.
  This includes a refactoring of the devmap and cpumap code to not care about
  the index on enqueue.

- Allowing regular lookups into devmaps from eBPF programs, using the read-only
  flag to make sure they don't change the values.

The performance impact of the series is negligible, in the sense that I cannot
measure it because the variance between test runs is higher than the difference
pre/post series.

Changelog:

v6:
  - Factor out list handling in maps to a helper in list.h (new patch 1)
  - Rename variables in struct bpf_redirect_info (new patch 3 + patch 4)
  - Explain why we are clearing out the map in the info struct on lookup failure
  - Remove unneeded check for forwarding target in tracepoint macro

v5:
  - Rebase on latest bpf-next.
  - Update documentation for bpf_redirect_map() with the new meaning of flags.

v4:
  - Fix a few nits from Andrii
  - Lose the #defines in bpf.h and just compare the flags argument directly to
    XDP_TX in bpf_xdp_redirect_map().

v3:
  - Adopt Jonathan's idea of using the lower two bits of the flag value as the
    return code.
  - Always do the lookup, and cache the result for use in xdp_do_redirect(); to
    achieve this, refactor the devmap and cpumap code to get rid the bitmap for
    selecting which devices to flush.

v2:
  - For patch 1, make it clear that the change works for any map type.
  - For patch 2, just use the new BPF_F_RDONLY_PROG flag to make the return
    value read-only.
====================

Signed-off-by: default avatarDaniel Borkmann <daniel@iogearbox.net>
parents 2d6dbb9a 0cdbb4b0

include/linux/filter.h

+2 −1
Original line number Diff line number Diff line
@@ -578,8 +578,9 @@ struct bpf_skb_data_end { 
};



struct bpf_redirect_info {

	u32 ifindex;

	u32 flags;

	u32 tgt_index;

	void *tgt_value;

	struct bpf_map *map;

	struct bpf_map *map_to_flush;

	u32 kern_flags;

include/linux/list.h

+14 −0
Original line number Diff line number Diff line
@@ -106,6 +106,20 @@ static inline void __list_del(struct list_head * prev, struct list_head * next) 
	WRITE_ONCE(prev->next, next);

}



/*

 * Delete a list entry and clear the 'prev' pointer.

 *

 * This is a special-purpose list clearing method used in the networking code

 * for lists allocated as per-cpu, where we don't want to incur the extra

 * WRITE_ONCE() overhead of a regular list_del_init(). The code that uses this

 * needs to check the node 'prev' pointer instead of calling list_empty().

 */

static inline void __list_del_clearprev(struct list_head *entry)

{

	__list_del(entry->prev, entry->next);

	entry->prev = NULL;

}



/**

 * list_del - deletes entry from list.

 * @entry: the element to delete from the list.

include/trace/events/xdp.h

+2 −3
Original line number Diff line number Diff line
@@ -175,9 +175,8 @@ struct _bpf_dtab_netdev { 
#endif /* __DEVMAP_OBJ_TYPE */



#define devmap_ifindex(fwd, map)				\

	(!fwd ? 0 :						\

	((map->map_type == BPF_MAP_TYPE_DEVMAP) ?		\

	  ((struct _bpf_dtab_netdev *)fwd)->dev->ifindex : 0))

	  ((struct _bpf_dtab_netdev *)fwd)->dev->ifindex : 0)



#define _trace_xdp_redirect_map(dev, xdp, fwd, map, idx)		\

	 trace_xdp_redirect_map(dev, xdp, devmap_ifindex(fwd, map),	\

include/uapi/linux/bpf.h

+5 −2
Original line number Diff line number Diff line
@@ -1571,8 +1571,11 @@ union bpf_attr { 
 * 		but this is only implemented for native XDP (with driver

 * 		support) as of this writing).

 *

 * 		All values for *flags* are reserved for future usage, and must

 * 		be left at zero.

 * 		The lower two bits of *flags* are used as the return code if

 * 		the map lookup fails. This is so that the return value can be

 * 		one of the XDP program return codes up to XDP_TX, as chosen by

 * 		the caller. Any higher bits in the *flags* argument must be

 * 		unset.

 *

 * 		When used to redirect packets to net devices, this helper

 * 		provides a high performance increase over **bpf_redirect**\ ().

kernel/bpf/cpumap.c

+48 −57
Original line number Diff line number Diff line
@@ -32,14 +32,19 @@ 


/* General idea: XDP packets getting XDP redirected to another CPU,

 * will maximum be stored/queued for one driver ->poll() call.  It is

 * guaranteed that setting flush bit and flush operation happen on

 * guaranteed that queueing the frame and the flush operation happen on

 * same CPU.  Thus, cpu_map_flush operation can deduct via this_cpu_ptr()

 * which queue in bpf_cpu_map_entry contains packets.

 */



#define CPU_MAP_BULK_SIZE 8  /* 8 == one cacheline on 64-bit archs */

struct bpf_cpu_map_entry;

struct bpf_cpu_map;



struct xdp_bulk_queue {

	void *q[CPU_MAP_BULK_SIZE];

	struct list_head flush_node;

	struct bpf_cpu_map_entry *obj;

	unsigned int count;

};
@@ -52,6 +57,8 @@ struct bpf_cpu_map_entry { 
	/* XDP can run multiple RX-ring queues, need __percpu enqueue store */

	struct xdp_bulk_queue __percpu *bulkq;



	struct bpf_cpu_map *cmap;



	/* Queue with potential multi-producers, and single-consumer kthread */

	struct ptr_ring *queue;

	struct task_struct *kthread;
@@ -65,23 +72,17 @@ struct bpf_cpu_map { 
	struct bpf_map map;

	/* Below members specific for map type */

	struct bpf_cpu_map_entry **cpu_map;

	unsigned long __percpu *flush_needed;

	struct list_head __percpu *flush_list;

};



static int bq_flush_to_queue(struct bpf_cpu_map_entry *rcpu,

			     struct xdp_bulk_queue *bq, bool in_napi_ctx);



static u64 cpu_map_bitmap_size(const union bpf_attr *attr)

{

	return BITS_TO_LONGS(attr->max_entries) * sizeof(unsigned long);

}

static int bq_flush_to_queue(struct xdp_bulk_queue *bq, bool in_napi_ctx);



static struct bpf_map *cpu_map_alloc(union bpf_attr *attr)

{

	struct bpf_cpu_map *cmap;

	int err = -ENOMEM;

	int ret, cpu;

	u64 cost;

	int ret;



	if (!capable(CAP_SYS_ADMIN))

		return ERR_PTR(-EPERM);
@@ -105,7 +106,7 @@ static struct bpf_map *cpu_map_alloc(union bpf_attr *attr) 


	/* make sure page count doesn't overflow */

	cost = (u64) cmap->map.max_entries * sizeof(struct bpf_cpu_map_entry *);

	cost += cpu_map_bitmap_size(attr) * num_possible_cpus();

	cost += sizeof(struct list_head) * num_possible_cpus();



	/* Notice returns -EPERM on if map size is larger than memlock limit */

	ret = bpf_map_charge_init(&cmap->map.memory, cost);
@@ -114,12 +115,13 @@ static struct bpf_map *cpu_map_alloc(union bpf_attr *attr) 
		goto free_cmap;

	}



	/* A per cpu bitfield with a bit per possible CPU in map  */

	cmap->flush_needed = __alloc_percpu(cpu_map_bitmap_size(attr),

					    __alignof__(unsigned long));

	if (!cmap->flush_needed)

	cmap->flush_list = alloc_percpu(struct list_head);

	if (!cmap->flush_list)

		goto free_charge;



	for_each_possible_cpu(cpu)

		INIT_LIST_HEAD(per_cpu_ptr(cmap->flush_list, cpu));



	/* Alloc array for possible remote "destination" CPUs */

	cmap->cpu_map = bpf_map_area_alloc(cmap->map.max_entries *

					   sizeof(struct bpf_cpu_map_entry *),
@@ -129,7 +131,7 @@ static struct bpf_map *cpu_map_alloc(union bpf_attr *attr) 


	return &cmap->map;

free_percpu:

	free_percpu(cmap->flush_needed);

	free_percpu(cmap->flush_list);

free_charge:

	bpf_map_charge_finish(&cmap->map.memory);

free_cmap:
@@ -334,7 +336,8 @@ static struct bpf_cpu_map_entry *__cpu_map_entry_alloc(u32 qsize, u32 cpu, 
{

	gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;

	struct bpf_cpu_map_entry *rcpu;

	int numa, err;

	struct xdp_bulk_queue *bq;

	int numa, err, i;



	/* Have map->numa_node, but choose node of redirect target CPU */

	numa = cpu_to_node(cpu);
@@ -349,6 +352,11 @@ static struct bpf_cpu_map_entry *__cpu_map_entry_alloc(u32 qsize, u32 cpu, 
	if (!rcpu->bulkq)

		goto free_rcu;



	for_each_possible_cpu(i) {

		bq = per_cpu_ptr(rcpu->bulkq, i);

		bq->obj = rcpu;

	}



	/* Alloc queue */

	rcpu->queue = kzalloc_node(sizeof(*rcpu->queue), gfp, numa);

	if (!rcpu->queue)
@@ -405,7 +413,7 @@ static void __cpu_map_entry_free(struct rcu_head *rcu) 
		struct xdp_bulk_queue *bq = per_cpu_ptr(rcpu->bulkq, cpu);



		/* No concurrent bq_enqueue can run at this point */

		bq_flush_to_queue(rcpu, bq, false);

		bq_flush_to_queue(bq, false);

	}

	free_percpu(rcpu->bulkq);

	/* Cannot kthread_stop() here, last put free rcpu resources */
@@ -488,6 +496,7 @@ static int cpu_map_update_elem(struct bpf_map *map, void *key, void *value, 
		rcpu = __cpu_map_entry_alloc(qsize, key_cpu, map->id);

		if (!rcpu)

			return -ENOMEM;

		rcpu->cmap = cmap;

	}

	rcu_read_lock();

	__cpu_map_entry_replace(cmap, key_cpu, rcpu);
@@ -514,14 +523,14 @@ static void cpu_map_free(struct bpf_map *map) 
	synchronize_rcu();



	/* To ensure all pending flush operations have completed wait for flush

	 * bitmap to indicate all flush_needed bits to be zero on _all_ cpus.

	 * Because the above synchronize_rcu() ensures the map is disconnected

	 * from the program we can assume no new bits will be set.

	 * list be empty on _all_ cpus. Because the above synchronize_rcu()

	 * ensures the map is disconnected from the program we can assume no new

	 * items will be added to the list.

	 */

	for_each_online_cpu(cpu) {

		unsigned long *bitmap = per_cpu_ptr(cmap->flush_needed, cpu);

		struct list_head *flush_list = per_cpu_ptr(cmap->flush_list, cpu);



		while (!bitmap_empty(bitmap, cmap->map.max_entries))

		while (!list_empty(flush_list))

			cond_resched();

	}
@@ -538,7 +547,7 @@ static void cpu_map_free(struct bpf_map *map) 
		/* bq flush and cleanup happens after RCU graze-period */

		__cpu_map_entry_replace(cmap, i, NULL); /* call_rcu */

	}

	free_percpu(cmap->flush_needed);

	free_percpu(cmap->flush_list);

	bpf_map_area_free(cmap->cpu_map);

	kfree(cmap);

}
@@ -590,9 +599,9 @@ const struct bpf_map_ops cpu_map_ops = { 
	.map_check_btf		= map_check_no_btf,

};



static int bq_flush_to_queue(struct bpf_cpu_map_entry *rcpu,

			     struct xdp_bulk_queue *bq, bool in_napi_ctx)

static int bq_flush_to_queue(struct xdp_bulk_queue *bq, bool in_napi_ctx)

{

	struct bpf_cpu_map_entry *rcpu = bq->obj;

	unsigned int processed = 0, drops = 0;

	const int to_cpu = rcpu->cpu;

	struct ptr_ring *q;
@@ -621,6 +630,8 @@ static int bq_flush_to_queue(struct bpf_cpu_map_entry *rcpu, 
	bq->count = 0;

	spin_unlock(&q->producer_lock);



	__list_del_clearprev(&bq->flush_node);



	/* Feedback loop via tracepoints */

	trace_xdp_cpumap_enqueue(rcpu->map_id, processed, drops, to_cpu);

	return 0;
@@ -631,10 +642,11 @@ static int bq_flush_to_queue(struct bpf_cpu_map_entry *rcpu, 
 */

static int bq_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_frame *xdpf)

{

	struct list_head *flush_list = this_cpu_ptr(rcpu->cmap->flush_list);

	struct xdp_bulk_queue *bq = this_cpu_ptr(rcpu->bulkq);



	if (unlikely(bq->count == CPU_MAP_BULK_SIZE))

		bq_flush_to_queue(rcpu, bq, true);

		bq_flush_to_queue(bq, true);



	/* Notice, xdp_buff/page MUST be queued here, long enough for

	 * driver to code invoking us to finished, due to driver
@@ -646,6 +658,10 @@ static int bq_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_frame *xdpf) 
	 * operation, when completing napi->poll call.

	 */

	bq->q[bq->count++] = xdpf;



	if (!bq->flush_node.prev)

		list_add(&bq->flush_node, flush_list);



	return 0;

}
@@ -665,41 +681,16 @@ int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_buff *xdp, 
	return 0;

}



void __cpu_map_insert_ctx(struct bpf_map *map, u32 bit)

{

	struct bpf_cpu_map *cmap = container_of(map, struct bpf_cpu_map, map);

	unsigned long *bitmap = this_cpu_ptr(cmap->flush_needed);



	__set_bit(bit, bitmap);

}



void __cpu_map_flush(struct bpf_map *map)

{

	struct bpf_cpu_map *cmap = container_of(map, struct bpf_cpu_map, map);

	unsigned long *bitmap = this_cpu_ptr(cmap->flush_needed);

	u32 bit;

	struct list_head *flush_list = this_cpu_ptr(cmap->flush_list);

	struct xdp_bulk_queue *bq, *tmp;



	/* The napi->poll softirq makes sure __cpu_map_insert_ctx()

	 * and __cpu_map_flush() happen on same CPU. Thus, the percpu

	 * bitmap indicate which percpu bulkq have packets.

	 */

	for_each_set_bit(bit, bitmap, map->max_entries) {

		struct bpf_cpu_map_entry *rcpu = READ_ONCE(cmap->cpu_map[bit]);

		struct xdp_bulk_queue *bq;



		/* This is possible if entry is removed by user space

		 * between xdp redirect and flush op.

		 */

		if (unlikely(!rcpu))

			continue;



		__clear_bit(bit, bitmap);



		/* Flush all frames in bulkq to real queue */

		bq = this_cpu_ptr(rcpu->bulkq);

		bq_flush_to_queue(rcpu, bq, true);

	list_for_each_entry_safe(bq, tmp, flush_list, flush_node) {

		bq_flush_to_queue(bq, true);



		/* If already running, costs spin_lock_irqsave + smb_mb */

		wake_up_process(rcpu->kthread);

		wake_up_process(bq->obj->kthread);

	}

}

CRA Git | Maintained and supported by SUSTech CRA and CCSE