[linux-block.git] / net / xdp / xdp_umem.c

// SPDX-License-Identifier: GPL-2.0
/* XDP user-space packet buffer
 * Copyright(c) 2018 Intel Corporation.
 */

#include <linux/init.h>
#include <linux/sched/mm.h>
#include <linux/sched/signal.h>
#include <linux/sched/task.h>
#include <linux/uaccess.h>
#include <linux/slab.h>
#include <linux/bpf.h>
#include <linux/mm.h>
#include <linux/netdevice.h>
#include <linux/rtnetlink.h>
#include <linux/idr.h>
#include <linux/vmalloc.h>

#include "xdp_umem.h"
#include "xsk_queue.h"

static DEFINE_IDA(umem_ida);

static void xdp_umem_unpin_pages(struct xdp_umem *umem)
{
	unpin_user_pages_dirty_lock(umem->pgs, umem->npgs, true);

	kvfree(umem->pgs);
	umem->pgs = NULL;
}

static void xdp_umem_unaccount_pages(struct xdp_umem *umem)
{
	if (umem->user) {
		atomic_long_sub(umem->npgs, &umem->user->locked_vm);
		free_uid(umem->user);
	}
}

static void xdp_umem_addr_unmap(struct xdp_umem *umem)
{
	vunmap(umem->addrs);
	umem->addrs = NULL;
}

static int xdp_umem_addr_map(struct xdp_umem *umem, struct page **pages,
			     u32 nr_pages)
{
	umem->addrs = vmap(pages, nr_pages, VM_MAP, PAGE_KERNEL);
	if (!umem->addrs)
		return -ENOMEM;
	return 0;
}

static void xdp_umem_release(struct xdp_umem *umem)
{
	umem->zc = false;
	ida_free(&umem_ida, umem->id);

	xdp_umem_addr_unmap(umem);
	xdp_umem_unpin_pages(umem);

	xdp_umem_unaccount_pages(umem);
	kfree(umem);
}

static void xdp_umem_release_deferred(struct work_struct *work)
{
	struct xdp_umem *umem = container_of(work, struct xdp_umem, work);

	xdp_umem_release(umem);
}

void xdp_get_umem(struct xdp_umem *umem)
{
	refcount_inc(&umem->users);
}

void xdp_put_umem(struct xdp_umem *umem, bool defer_cleanup)
{
	if (!umem)
		return;

	if (refcount_dec_and_test(&umem->users)) {
		if (defer_cleanup) {
			INIT_WORK(&umem->work, xdp_umem_release_deferred);
			schedule_work(&umem->work);
		} else {
			xdp_umem_release(umem);
		}
	}
}

static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
{
	unsigned int gup_flags = FOLL_WRITE;
	long npgs;
	int err;

	umem->pgs = kvcalloc(umem->npgs, sizeof(*umem->pgs), GFP_KERNEL | __GFP_NOWARN);
	if (!umem->pgs)
		return -ENOMEM;

	mmap_read_lock(current->mm);
	npgs = pin_user_pages(address, umem->npgs,
			      gup_flags | FOLL_LONGTERM, &umem->pgs[0], NULL);
	mmap_read_unlock(current->mm);

	if (npgs != umem->npgs) {
		if (npgs >= 0) {
			umem->npgs = npgs;
			err = -ENOMEM;
			goto out_pin;
		}
		err = npgs;
		goto out_pgs;
	}
	return 0;

out_pin:
	xdp_umem_unpin_pages(umem);
out_pgs:
	kvfree(umem->pgs);
	umem->pgs = NULL;
	return err;
}

static int xdp_umem_account_pages(struct xdp_umem *umem)
{
	unsigned long lock_limit, new_npgs, old_npgs;

	if (capable(CAP_IPC_LOCK))
		return 0;

	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
	umem->user = get_uid(current_user());

	do {
		old_npgs = atomic_long_read(&umem->user->locked_vm);
		new_npgs = old_npgs + umem->npgs;
		if (new_npgs > lock_limit) {
			free_uid(umem->user);
			umem->user = NULL;
			return -ENOBUFS;
		}
	} while (atomic_long_cmpxchg(&umem->user->locked_vm, old_npgs,
				     new_npgs) != old_npgs);
	return 0;
}

static int xdp_umem_reg(struct xdp_umem *umem, struct xdp_umem_reg *mr)
{
	u32 npgs_rem, chunk_size = mr->chunk_size, headroom = mr->headroom;
	bool unaligned_chunks = mr->flags & XDP_UMEM_UNALIGNED_CHUNK_FLAG;
	u64 npgs, addr = mr->addr, size = mr->len;
	unsigned int chunks, chunks_rem;
	int err;

	if (chunk_size < XDP_UMEM_MIN_CHUNK_SIZE || chunk_size > PAGE_SIZE) {
		/* Strictly speaking we could support this, if:
		 * - huge pages, or*
		 * - using an IOMMU, or
		 * - making sure the memory area is consecutive
		 * but for now, we simply say "computer says no".
		 */
		return -EINVAL;
	}

	if (mr->flags & ~XDP_UMEM_UNALIGNED_CHUNK_FLAG)
		return -EINVAL;

	if (!unaligned_chunks && !is_power_of_2(chunk_size))
		return -EINVAL;

	if (!PAGE_ALIGNED(addr)) {
		/* Memory area has to be page size aligned. For
		 * simplicity, this might change.
		 */
		return -EINVAL;
	}

	if ((addr + size) < addr)
		return -EINVAL;

	npgs = div_u64_rem(size, PAGE_SIZE, &npgs_rem);
	if (npgs_rem)
		npgs++;
	if (npgs > U32_MAX)
		return -EINVAL;

	chunks = (unsigned int)div_u64_rem(size, chunk_size, &chunks_rem);
	if (chunks == 0)
		return -EINVAL;

	if (!unaligned_chunks && chunks_rem)
		return -EINVAL;

	if (headroom >= chunk_size - XDP_PACKET_HEADROOM)
		return -EINVAL;

	umem->size = size;
	umem->headroom = headroom;
	umem->chunk_size = chunk_size;
	umem->chunks = chunks;
	umem->npgs = (u32)npgs;
	umem->pgs = NULL;
	umem->user = NULL;
	umem->flags = mr->flags;

	INIT_LIST_HEAD(&umem->xsk_dma_list);
	refcount_set(&umem->users, 1);

	err = xdp_umem_account_pages(umem);
	if (err)
		return err;

	err = xdp_umem_pin_pages(umem, (unsigned long)addr);
	if (err)
		goto out_account;

	err = xdp_umem_addr_map(umem, umem->pgs, umem->npgs);
	if (err)
		goto out_unpin;

	return 0;

out_unpin:
	xdp_umem_unpin_pages(umem);
out_account:
	xdp_umem_unaccount_pages(umem);
	return err;
}

struct xdp_umem *xdp_umem_create(struct xdp_umem_reg *mr)
{
	struct xdp_umem *umem;
	int err;

	umem = kzalloc(sizeof(*umem), GFP_KERNEL);
	if (!umem)
		return ERR_PTR(-ENOMEM);

	err = ida_alloc(&umem_ida, GFP_KERNEL);
	if (err < 0) {
		kfree(umem);
		return ERR_PTR(err);
	}
	umem->id = err;

	err = xdp_umem_reg(umem, mr);
	if (err) {
		ida_free(&umem_ida, umem->id);
		kfree(umem);
		return ERR_PTR(err);
	}

	return umem;
}
Commit	Line	Data
c0c77d8f BT	1	// SPDX-License-Identifier: GPL-2.0
	2	/* XDP user-space packet buffer
	3	* Copyright(c) 2018 Intel Corporation.
c0c77d8f BT	4	*/
	5
	6	#include <linux/init.h>
	7	#include <linux/sched/mm.h>
	8	#include <linux/sched/signal.h>
	9	#include <linux/sched/task.h>
	10	#include <linux/uaccess.h>
	11	#include <linux/slab.h>
	12	#include <linux/bpf.h>
	13	#include <linux/mm.h>
84c6b868 JK	14	#include <linux/netdevice.h>
84c6b868 JK	15	#include <linux/rtnetlink.h>
50e74c01	16	#include <linux/idr.h>
624676e7	17	#include <linux/vmalloc.h>
c0c77d8f BT	18
c0c77d8f BT	19	#include "xdp_umem.h"
e61e62b9	20	#include "xsk_queue.h"
c0c77d8f	21
50e74c01 BT	22	static DEFINE_IDA(umem_ida);
50e74c01 BT	23
1c1efc2a	24	static void xdp_umem_unpin_pages(struct xdp_umem *umem)
c9b47cc1	25	{
1c1efc2a	26	unpin_user_pages_dirty_lock(umem->pgs, umem->npgs, true);
84c6b868	27
a720a2a0	28	kvfree(umem->pgs);
1c1efc2a	29	umem->pgs = NULL;
c9b47cc1	30	}
84c6b868	31
1c1efc2a	32	static void xdp_umem_unaccount_pages(struct xdp_umem *umem)
c9b47cc1	33	{
1c1efc2a MK	34	if (umem->user) {
	35	atomic_long_sub(umem->npgs, &umem->user->locked_vm);
	36	free_uid(umem->user);
	37	}
84c6b868 JK	38	}
84c6b868 JK	39
7f7ffa4e MK	40	static void xdp_umem_addr_unmap(struct xdp_umem *umem)
	41	{
	42	vunmap(umem->addrs);
	43	umem->addrs = NULL;
	44	}
	45
	46	static int xdp_umem_addr_map(struct xdp_umem umem, struct page *pages,
	47	u32 nr_pages)
	48	{
	49	umem->addrs = vmap(pages, nr_pages, VM_MAP, PAGE_KERNEL);
	50	if (!umem->addrs)
	51	return -ENOMEM;
	52	return 0;
	53	}
	54
c0c77d8f BT	55	static void xdp_umem_release(struct xdp_umem *umem)
c0c77d8f BT	56	{
c2d3d6a4	57	umem->zc = false;
21f1481a	58	ida_free(&umem_ida, umem->id);
50e74c01	59
7f7ffa4e	60	xdp_umem_addr_unmap(umem);
a49049ea	61	xdp_umem_unpin_pages(umem);
c0c77d8f	62
c0c77d8f	63	xdp_umem_unaccount_pages(umem);
c0c77d8f BT	64	kfree(umem);
	65	}
	66
537cf4e3 MK	67	static void xdp_umem_release_deferred(struct work_struct *work)
	68	{
	69	struct xdp_umem *umem = container_of(work, struct xdp_umem, work);
	70
	71	xdp_umem_release(umem);
	72	}
	73
c0c77d8f BT	74	void xdp_get_umem(struct xdp_umem *umem)
c0c77d8f BT	75	{
d3b42f14	76	refcount_inc(&umem->users);
c0c77d8f BT	77	}
c0c77d8f BT	78
537cf4e3	79	void xdp_put_umem(struct xdp_umem *umem, bool defer_cleanup)
c0c77d8f BT	80	{
	81	if (!umem)
	82	return;
	83
537cf4e3 MK	84	if (refcount_dec_and_test(&umem->users)) {
	85	if (defer_cleanup) {
	86	INIT_WORK(&umem->work, xdp_umem_release_deferred);
	87	schedule_work(&umem->work);
	88	} else {
	89	xdp_umem_release(umem);
	90	}
	91	}
c0c77d8f BT	92	}
c0c77d8f BT	93
07bf2d97	94	static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
c0c77d8f BT	95	{
	96	unsigned int gup_flags = FOLL_WRITE;
	97	long npgs;
	98	int err;
	99
a720a2a0	100	umem->pgs = kvcalloc(umem->npgs, sizeof(*umem->pgs), GFP_KERNEL \| __GFP_NOWARN);
c0c77d8f BT	101	if (!umem->pgs)
	102	return -ENOMEM;
	103
d8ed45c5	104	mmap_read_lock(current->mm);
07bf2d97	105	npgs = pin_user_pages(address, umem->npgs,
932f4a63	106	gup_flags \| FOLL_LONGTERM, &umem->pgs[0], NULL);
d8ed45c5	107	mmap_read_unlock(current->mm);
c0c77d8f BT	108
	109	if (npgs != umem->npgs) {
	110	if (npgs >= 0) {
	111	umem->npgs = npgs;
	112	err = -ENOMEM;
	113	goto out_pin;
	114	}
	115	err = npgs;
	116	goto out_pgs;
	117	}
	118	return 0;
	119
	120	out_pin:
	121	xdp_umem_unpin_pages(umem);
	122	out_pgs:
a720a2a0	123	kvfree(umem->pgs);
c0c77d8f BT	124	umem->pgs = NULL;
	125	return err;
	126	}
	127
	128	static int xdp_umem_account_pages(struct xdp_umem *umem)
	129	{
	130	unsigned long lock_limit, new_npgs, old_npgs;
	131
	132	if (capable(CAP_IPC_LOCK))
	133	return 0;
	134
	135	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
	136	umem->user = get_uid(current_user());
	137
	138	do {
	139	old_npgs = atomic_long_read(&umem->user->locked_vm);
	140	new_npgs = old_npgs + umem->npgs;
	141	if (new_npgs > lock_limit) {
	142	free_uid(umem->user);
	143	umem->user = NULL;
	144	return -ENOBUFS;
	145	}
	146	} while (atomic_long_cmpxchg(&umem->user->locked_vm, old_npgs,
	147	new_npgs) != old_npgs);
	148	return 0;
	149	}
	150
a49049ea	151	static int xdp_umem_reg(struct xdp_umem umem, struct xdp_umem_reg mr)
c0c77d8f	152	{
2b1667e5	153	u32 npgs_rem, chunk_size = mr->chunk_size, headroom = mr->headroom;
c05cd364	154	bool unaligned_chunks = mr->flags & XDP_UMEM_UNALIGNED_CHUNK_FLAG;
b16a87d0	155	u64 npgs, addr = mr->addr, size = mr->len;
2b1667e5	156	unsigned int chunks, chunks_rem;
99e3a236	157	int err;
c0c77d8f	158
bbff2f32	159	if (chunk_size < XDP_UMEM_MIN_CHUNK_SIZE \|\| chunk_size > PAGE_SIZE) {
c0c77d8f BT	160	/* Strictly speaking we could support this, if:
	161	* - huge pages, or*
	162	* - using an IOMMU, or
	163	* - making sure the memory area is consecutive
	164	* but for now, we simply say "computer says no".
	165	*/
	166	return -EINVAL;
	167	}
	168
c2d3d6a4	169	if (mr->flags & ~XDP_UMEM_UNALIGNED_CHUNK_FLAG)
c05cd364 KL	170	return -EINVAL;
	171
	172	if (!unaligned_chunks && !is_power_of_2(chunk_size))
c0c77d8f BT	173	return -EINVAL;
	174
	175	if (!PAGE_ALIGNED(addr)) {
	176	/* Memory area has to be page size aligned. For
	177	* simplicity, this might change.
	178	*/
	179	return -EINVAL;
	180	}
	181
	182	if ((addr + size) < addr)
	183	return -EINVAL;
	184
2b1667e5 BT	185	npgs = div_u64_rem(size, PAGE_SIZE, &npgs_rem);
	186	if (npgs_rem)
	187	npgs++;
b16a87d0 BT	188	if (npgs > U32_MAX)
	189	return -EINVAL;
	190
2b1667e5	191	chunks = (unsigned int)div_u64_rem(size, chunk_size, &chunks_rem);
bbff2f32	192	if (chunks == 0)
c0c77d8f BT	193	return -EINVAL;
c0c77d8f BT	194
2b1667e5 BT	195	if (!unaligned_chunks && chunks_rem)
2b1667e5 BT	196	return -EINVAL;
c0c77d8f	197
99e3a236	198	if (headroom >= chunk_size - XDP_PACKET_HEADROOM)
c0c77d8f BT	199	return -EINVAL;
c0c77d8f BT	200
93ee30f3	201	umem->size = size;
bbff2f32	202	umem->headroom = headroom;
2b43470a	203	umem->chunk_size = chunk_size;
1c1efc2a	204	umem->chunks = chunks;
b16a87d0	205	umem->npgs = (u32)npgs;
c0c77d8f BT	206	umem->pgs = NULL;
c0c77d8f BT	207	umem->user = NULL;
c05cd364	208	umem->flags = mr->flags;
c0c77d8f	209
921b6869	210	INIT_LIST_HEAD(&umem->xsk_dma_list);
d3b42f14	211	refcount_set(&umem->users, 1);
c0c77d8f BT	212
	213	err = xdp_umem_account_pages(umem);
	214	if (err)
044175a0	215	return err;
c0c77d8f	216
07bf2d97	217	err = xdp_umem_pin_pages(umem, (unsigned long)addr);
c0c77d8f BT	218	if (err)
c0c77d8f BT	219	goto out_account;
8aef7340	220
7f7ffa4e MK	221	err = xdp_umem_addr_map(umem, umem->pgs, umem->npgs);
	222	if (err)
	223	goto out_unpin;
	224
2b43470a	225	return 0;
c0c77d8f	226
7f7ffa4e MK	227	out_unpin:
7f7ffa4e MK	228	xdp_umem_unpin_pages(umem);
c0c77d8f BT	229	out_account:
c0c77d8f BT	230	xdp_umem_unaccount_pages(umem);
c0c77d8f BT	231	return err;
c0c77d8f BT	232	}
965a9909	233
a49049ea BT	234	struct xdp_umem xdp_umem_create(struct xdp_umem_reg mr)
	235	{
	236	struct xdp_umem *umem;
	237	int err;
	238
	239	umem = kzalloc(sizeof(*umem), GFP_KERNEL);
	240	if (!umem)
	241	return ERR_PTR(-ENOMEM);
	242
21f1481a	243	err = ida_alloc(&umem_ida, GFP_KERNEL);
50e74c01 BT	244	if (err < 0) {
	245	kfree(umem);
	246	return ERR_PTR(err);
	247	}
	248	umem->id = err;
	249
a49049ea BT	250	err = xdp_umem_reg(umem, mr);
a49049ea BT	251	if (err) {
21f1481a	252	ida_free(&umem_ida, umem->id);
a49049ea BT	253	kfree(umem);
	254	return ERR_PTR(err);
	255	}
	256
	257	return umem;
	258	}