[linux-block.git] / io_uring / kbuf.c

// SPDX-License-Identifier: GPL-2.0
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/fs.h>
#include <linux/file.h>
#include <linux/mm.h>
#include <linux/slab.h>
#include <linux/namei.h>
#include <linux/poll.h>
#include <linux/io_uring.h>

#include <uapi/linux/io_uring.h>

#include "io_uring.h"
#include "opdef.h"
#include "kbuf.h"

#define IO_BUFFER_LIST_BUF_PER_PAGE (PAGE_SIZE / sizeof(struct io_uring_buf))

#define BGID_ARRAY	64

/* BIDs are addressed by a 16-bit field in a CQE */
#define MAX_BIDS_PER_BGID (1 << 16)

struct kmem_cache *io_buf_cachep;

struct io_provide_buf {
	struct file			*file;
	__u64				addr;
	__u32				len;
	__u32				bgid;
	__u32				nbufs;
	__u16				bid;
};

static inline struct io_buffer_list *io_buffer_get_list(struct io_ring_ctx *ctx,
							unsigned int bgid)
{
	if (ctx->io_bl && bgid < BGID_ARRAY)
		return &ctx->io_bl[bgid];

	return xa_load(&ctx->io_bl_xa, bgid);
}

static int io_buffer_add_list(struct io_ring_ctx *ctx,
			      struct io_buffer_list *bl, unsigned int bgid)
{
	bl->bgid = bgid;
	if (bgid < BGID_ARRAY)
		return 0;

	return xa_err(xa_store(&ctx->io_bl_xa, bgid, bl, GFP_KERNEL));
}

void io_kbuf_recycle_legacy(struct io_kiocb *req, unsigned issue_flags)
{
	struct io_ring_ctx *ctx = req->ctx;
	struct io_buffer_list *bl;
	struct io_buffer *buf;

	/*
	 * For legacy provided buffer mode, don't recycle if we already did
	 * IO to this buffer. For ring-mapped provided buffer mode, we should
	 * increment ring->head to explicitly monopolize the buffer to avoid
	 * multiple use.
	 */
	if (req->flags & REQ_F_PARTIAL_IO)
		return;

	io_ring_submit_lock(ctx, issue_flags);

	buf = req->kbuf;
	bl = io_buffer_get_list(ctx, buf->bgid);
	list_add(&buf->list, &bl->buf_list);
	req->flags &= ~REQ_F_BUFFER_SELECTED;
	req->buf_index = buf->bgid;

	io_ring_submit_unlock(ctx, issue_flags);
	return;
}

unsigned int __io_put_kbuf(struct io_kiocb *req, unsigned issue_flags)
{
	unsigned int cflags;

	/*
	 * We can add this buffer back to two lists:
	 *
	 * 1) The io_buffers_cache list. This one is protected by the
	 *    ctx->uring_lock. If we already hold this lock, add back to this
	 *    list as we can grab it from issue as well.
	 * 2) The io_buffers_comp list. This one is protected by the
	 *    ctx->completion_lock.
	 *
	 * We migrate buffers from the comp_list to the issue cache list
	 * when we need one.
	 */
	if (req->flags & REQ_F_BUFFER_RING) {
		/* no buffers to recycle for this case */
		cflags = __io_put_kbuf_list(req, NULL);
	} else if (issue_flags & IO_URING_F_UNLOCKED) {
		struct io_ring_ctx *ctx = req->ctx;

		spin_lock(&ctx->completion_lock);
		cflags = __io_put_kbuf_list(req, &ctx->io_buffers_comp);
		spin_unlock(&ctx->completion_lock);
	} else {
		lockdep_assert_held(&req->ctx->uring_lock);

		cflags = __io_put_kbuf_list(req, &req->ctx->io_buffers_cache);
	}
	return cflags;
}

static void __user *io_provided_buffer_select(struct io_kiocb *req, size_t *len,
					      struct io_buffer_list *bl)
{
	if (!list_empty(&bl->buf_list)) {
		struct io_buffer *kbuf;

		kbuf = list_first_entry(&bl->buf_list, struct io_buffer, list);
		list_del(&kbuf->list);
		if (*len == 0 || *len > kbuf->len)
			*len = kbuf->len;
		req->flags |= REQ_F_BUFFER_SELECTED;
		req->kbuf = kbuf;
		req->buf_index = kbuf->bid;
		return u64_to_user_ptr(kbuf->addr);
	}
	return NULL;
}

static void __user *io_ring_buffer_select(struct io_kiocb *req, size_t *len,
					  struct io_buffer_list *bl,
					  unsigned int issue_flags)
{
	struct io_uring_buf_ring *br = bl->buf_ring;
	struct io_uring_buf *buf;
	__u16 head = bl->head;

	if (unlikely(smp_load_acquire(&br->tail) == head))
		return NULL;

	head &= bl->mask;
	/* mmaped buffers are always contig */
	if (bl->is_mmap || head < IO_BUFFER_LIST_BUF_PER_PAGE) {
		buf = &br->bufs[head];
	} else {
		int off = head & (IO_BUFFER_LIST_BUF_PER_PAGE - 1);
		int index = head / IO_BUFFER_LIST_BUF_PER_PAGE;
		buf = page_address(bl->buf_pages[index]);
		buf += off;
	}
	if (*len == 0 || *len > buf->len)
		*len = buf->len;
	req->flags |= REQ_F_BUFFER_RING;
	req->buf_list = bl;
	req->buf_index = buf->bid;

	if (issue_flags & IO_URING_F_UNLOCKED || !file_can_poll(req->file)) {
		/*
		 * If we came in unlocked, we have no choice but to consume the
		 * buffer here, otherwise nothing ensures that the buffer won't
		 * get used by others. This does mean it'll be pinned until the
		 * IO completes, coming in unlocked means we're being called from
		 * io-wq context and there may be further retries in async hybrid
		 * mode. For the locked case, the caller must call commit when
		 * the transfer completes (or if we get -EAGAIN and must poll of
		 * retry).
		 */
		req->buf_list = NULL;
		bl->head++;
	}
	return u64_to_user_ptr(buf->addr);
}

void __user *io_buffer_select(struct io_kiocb *req, size_t *len,
			      unsigned int issue_flags)
{
	struct io_ring_ctx *ctx = req->ctx;
	struct io_buffer_list *bl;
	void __user *ret = NULL;

	io_ring_submit_lock(req->ctx, issue_flags);

	bl = io_buffer_get_list(ctx, req->buf_index);
	if (likely(bl)) {
		if (bl->is_mapped)
			ret = io_ring_buffer_select(req, len, bl, issue_flags);
		else
			ret = io_provided_buffer_select(req, len, bl);
	}
	io_ring_submit_unlock(req->ctx, issue_flags);
	return ret;
}

static __cold int io_init_bl_list(struct io_ring_ctx *ctx)
{
	int i;

	ctx->io_bl = kcalloc(BGID_ARRAY, sizeof(struct io_buffer_list),
				GFP_KERNEL);
	if (!ctx->io_bl)
		return -ENOMEM;

	for (i = 0; i < BGID_ARRAY; i++) {
		INIT_LIST_HEAD(&ctx->io_bl[i].buf_list);
		ctx->io_bl[i].bgid = i;
	}

	return 0;
}

static int __io_remove_buffers(struct io_ring_ctx *ctx,
			       struct io_buffer_list *bl, unsigned nbufs)
{
	unsigned i = 0;

	/* shouldn't happen */
	if (!nbufs)
		return 0;

	if (bl->is_mapped) {
		i = bl->buf_ring->tail - bl->head;
		if (bl->is_mmap) {
			folio_put(virt_to_folio(bl->buf_ring));
			bl->buf_ring = NULL;
			bl->is_mmap = 0;
		} else if (bl->buf_nr_pages) {
			int j;

			for (j = 0; j < bl->buf_nr_pages; j++)
				unpin_user_page(bl->buf_pages[j]);
			kvfree(bl->buf_pages);
			bl->buf_pages = NULL;
			bl->buf_nr_pages = 0;
		}
		/* make sure it's seen as empty */
		INIT_LIST_HEAD(&bl->buf_list);
		bl->is_mapped = 0;
		return i;
	}

	/* protects io_buffers_cache */
	lockdep_assert_held(&ctx->uring_lock);

	while (!list_empty(&bl->buf_list)) {
		struct io_buffer *nxt;

		nxt = list_first_entry(&bl->buf_list, struct io_buffer, list);
		list_move(&nxt->list, &ctx->io_buffers_cache);
		if (++i == nbufs)
			return i;
		cond_resched();
	}

	return i;
}

void io_destroy_buffers(struct io_ring_ctx *ctx)
{
	struct io_buffer_list *bl;
	struct list_head *item, *tmp;
	struct io_buffer *buf;
	unsigned long index;
	int i;

	for (i = 0; i < BGID_ARRAY; i++) {
		if (!ctx->io_bl)
			break;
		__io_remove_buffers(ctx, &ctx->io_bl[i], -1U);
	}

	xa_for_each(&ctx->io_bl_xa, index, bl) {
		xa_erase(&ctx->io_bl_xa, bl->bgid);
		__io_remove_buffers(ctx, bl, -1U);
		kfree(bl);
	}

	list_for_each_safe(item, tmp, &ctx->io_buffers_cache) {
		buf = list_entry(item, struct io_buffer, list);
		kmem_cache_free(io_buf_cachep, buf);
	}
}

int io_remove_buffers_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
{
	struct io_provide_buf *p = io_kiocb_to_cmd(req, struct io_provide_buf);
	u64 tmp;

	if (sqe->rw_flags || sqe->addr || sqe->len || sqe->off ||
	    sqe->splice_fd_in)
		return -EINVAL;

	tmp = READ_ONCE(sqe->fd);
	if (!tmp || tmp > MAX_BIDS_PER_BGID)
		return -EINVAL;

	memset(p, 0, sizeof(*p));
	p->nbufs = tmp;
	p->bgid = READ_ONCE(sqe->buf_group);
	return 0;
}

int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags)
{
	struct io_provide_buf *p = io_kiocb_to_cmd(req, struct io_provide_buf);
	struct io_ring_ctx *ctx = req->ctx;
	struct io_buffer_list *bl;
	int ret = 0;

	io_ring_submit_lock(ctx, issue_flags);

	ret = -ENOENT;
	bl = io_buffer_get_list(ctx, p->bgid);
	if (bl) {
		ret = -EINVAL;
		/* can't use provide/remove buffers command on mapped buffers */
		if (!bl->is_mapped)
			ret = __io_remove_buffers(ctx, bl, p->nbufs);
	}
	io_ring_submit_unlock(ctx, issue_flags);
	if (ret < 0)
		req_set_fail(req);
	io_req_set_res(req, ret, 0);
	return IOU_OK;
}

int io_provide_buffers_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
{
	unsigned long size, tmp_check;
	struct io_provide_buf *p = io_kiocb_to_cmd(req, struct io_provide_buf);
	u64 tmp;

	if (sqe->rw_flags || sqe->splice_fd_in)
		return -EINVAL;

	tmp = READ_ONCE(sqe->fd);
	if (!tmp || tmp > MAX_BIDS_PER_BGID)
		return -E2BIG;
	p->nbufs = tmp;
	p->addr = READ_ONCE(sqe->addr);
	p->len = READ_ONCE(sqe->len);

	if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs,
				&size))
		return -EOVERFLOW;
	if (check_add_overflow((unsigned long)p->addr, size, &tmp_check))
		return -EOVERFLOW;

	size = (unsigned long)p->len * p->nbufs;
	if (!access_ok(u64_to_user_ptr(p->addr), size))
		return -EFAULT;

	p->bgid = READ_ONCE(sqe->buf_group);
	tmp = READ_ONCE(sqe->off);
	if (tmp > USHRT_MAX)
		return -E2BIG;
	if (tmp + p->nbufs > MAX_BIDS_PER_BGID)
		return -EINVAL;
	p->bid = tmp;
	return 0;
}

#define IO_BUFFER_ALLOC_BATCH 64

static int io_refill_buffer_cache(struct io_ring_ctx *ctx)
{
	struct io_buffer *bufs[IO_BUFFER_ALLOC_BATCH];
	int allocated;

	/*
	 * Completions that don't happen inline (eg not under uring_lock) will
	 * add to ->io_buffers_comp. If we don't have any free buffers, check
	 * the completion list and splice those entries first.
	 */
	if (!list_empty_careful(&ctx->io_buffers_comp)) {
		spin_lock(&ctx->completion_lock);
		if (!list_empty(&ctx->io_buffers_comp)) {
			list_splice_init(&ctx->io_buffers_comp,
						&ctx->io_buffers_cache);
			spin_unlock(&ctx->completion_lock);
			return 0;
		}
		spin_unlock(&ctx->completion_lock);
	}

	/*
	 * No free buffers and no completion entries either. Allocate a new
	 * batch of buffer entries and add those to our freelist.
	 */

	allocated = kmem_cache_alloc_bulk(io_buf_cachep, GFP_KERNEL_ACCOUNT,
					  ARRAY_SIZE(bufs), (void **) bufs);
	if (unlikely(!allocated)) {
		/*
		 * Bulk alloc is all-or-nothing. If we fail to get a batch,
		 * retry single alloc to be on the safe side.
		 */
		bufs[0] = kmem_cache_alloc(io_buf_cachep, GFP_KERNEL);
		if (!bufs[0])
			return -ENOMEM;
		allocated = 1;
	}

	while (allocated)
		list_add_tail(&bufs[--allocated]->list, &ctx->io_buffers_cache);

	return 0;
}

static int io_add_buffers(struct io_ring_ctx *ctx, struct io_provide_buf *pbuf,
			  struct io_buffer_list *bl)
{
	struct io_buffer *buf;
	u64 addr = pbuf->addr;
	int i, bid = pbuf->bid;

	for (i = 0; i < pbuf->nbufs; i++) {
		if (list_empty(&ctx->io_buffers_cache) &&
		    io_refill_buffer_cache(ctx))
			break;
		buf = list_first_entry(&ctx->io_buffers_cache, struct io_buffer,
					list);
		list_move_tail(&buf->list, &bl->buf_list);
		buf->addr = addr;
		buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT);
		buf->bid = bid;
		buf->bgid = pbuf->bgid;
		addr += pbuf->len;
		bid++;
		cond_resched();
	}

	return i ? 0 : -ENOMEM;
}

int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
{
	struct io_provide_buf *p = io_kiocb_to_cmd(req, struct io_provide_buf);
	struct io_ring_ctx *ctx = req->ctx;
	struct io_buffer_list *bl;
	int ret = 0;

	io_ring_submit_lock(ctx, issue_flags);

	if (unlikely(p->bgid < BGID_ARRAY && !ctx->io_bl)) {
		ret = io_init_bl_list(ctx);
		if (ret)
			goto err;
	}

	bl = io_buffer_get_list(ctx, p->bgid);
	if (unlikely(!bl)) {
		bl = kzalloc(sizeof(*bl), GFP_KERNEL_ACCOUNT);
		if (!bl) {
			ret = -ENOMEM;
			goto err;
		}
		INIT_LIST_HEAD(&bl->buf_list);
		ret = io_buffer_add_list(ctx, bl, p->bgid);
		if (ret) {
			kfree(bl);
			goto err;
		}
	}
	/* can't add buffers via this command for a mapped buffer ring */
	if (bl->is_mapped) {
		ret = -EINVAL;
		goto err;
	}

	ret = io_add_buffers(ctx, p, bl);
err:
	io_ring_submit_unlock(ctx, issue_flags);

	if (ret < 0)
		req_set_fail(req);
	io_req_set_res(req, ret, 0);
	return IOU_OK;
}

static int io_pin_pbuf_ring(struct io_uring_buf_reg *reg,
			    struct io_buffer_list *bl)
{
	struct io_uring_buf_ring *br;
	struct page **pages;
	int nr_pages;

	pages = io_pin_pages(reg->ring_addr,
			     flex_array_size(br, bufs, reg->ring_entries),
			     &nr_pages);
	if (IS_ERR(pages))
		return PTR_ERR(pages);

	br = page_address(pages[0]);
#ifdef SHM_COLOUR
	/*
	 * On platforms that have specific aliasing requirements, SHM_COLOUR
	 * is set and we must guarantee that the kernel and user side align
	 * nicely. We cannot do that if IOU_PBUF_RING_MMAP isn't set and
	 * the application mmap's the provided ring buffer. Fail the request
	 * if we, by chance, don't end up with aligned addresses. The app
	 * should use IOU_PBUF_RING_MMAP instead, and liburing will handle
	 * this transparently.
	 */
	if ((reg->ring_addr | (unsigned long) br) & (SHM_COLOUR - 1)) {
		int i;

		for (i = 0; i < nr_pages; i++)
			unpin_user_page(pages[i]);
		return -EINVAL;
	}
#endif
	bl->buf_pages = pages;
	bl->buf_nr_pages = nr_pages;
	bl->buf_ring = br;
	bl->is_mapped = 1;
	bl->is_mmap = 0;
	return 0;
}

static int io_alloc_pbuf_ring(struct io_uring_buf_reg *reg,
			      struct io_buffer_list *bl)
{
	gfp_t gfp = GFP_KERNEL_ACCOUNT | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP;
	size_t ring_size;
	void *ptr;

	ring_size = reg->ring_entries * sizeof(struct io_uring_buf_ring);
	ptr = (void *) __get_free_pages(gfp, get_order(ring_size));
	if (!ptr)
		return -ENOMEM;

	bl->buf_ring = ptr;
	bl->is_mapped = 1;
	bl->is_mmap = 1;
	return 0;
}

int io_register_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg)
{
	struct io_uring_buf_reg reg;
	struct io_buffer_list *bl, *free_bl = NULL;
	int ret;

	if (copy_from_user(&reg, arg, sizeof(reg)))
		return -EFAULT;

	if (reg.resv[0] || reg.resv[1] || reg.resv[2])
		return -EINVAL;
	if (reg.flags & ~IOU_PBUF_RING_MMAP)
		return -EINVAL;
	if (!(reg.flags & IOU_PBUF_RING_MMAP)) {
		if (!reg.ring_addr)
			return -EFAULT;
		if (reg.ring_addr & ~PAGE_MASK)
			return -EINVAL;
	} else {
		if (reg.ring_addr)
			return -EINVAL;
	}

	if (!is_power_of_2(reg.ring_entries))
		return -EINVAL;

	/* cannot disambiguate full vs empty due to head/tail size */
	if (reg.ring_entries >= 65536)
		return -EINVAL;

	if (unlikely(reg.bgid < BGID_ARRAY && !ctx->io_bl)) {
		int ret = io_init_bl_list(ctx);
		if (ret)
			return ret;
	}

	bl = io_buffer_get_list(ctx, reg.bgid);
	if (bl) {
		/* if mapped buffer ring OR classic exists, don't allow */
		if (bl->is_mapped || !list_empty(&bl->buf_list))
			return -EEXIST;
	} else {
		free_bl = bl = kzalloc(sizeof(*bl), GFP_KERNEL);
		if (!bl)
			return -ENOMEM;
	}

	if (!(reg.flags & IOU_PBUF_RING_MMAP))
		ret = io_pin_pbuf_ring(&reg, bl);
	else
		ret = io_alloc_pbuf_ring(&reg, bl);

	if (!ret) {
		bl->nr_entries = reg.ring_entries;
		bl->mask = reg.ring_entries - 1;

		io_buffer_add_list(ctx, bl, reg.bgid);
		return 0;
	}

	kfree(free_bl);
	return ret;
}

int io_unregister_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg)
{
	struct io_uring_buf_reg reg;
	struct io_buffer_list *bl;

	if (copy_from_user(&reg, arg, sizeof(reg)))
		return -EFAULT;
	if (reg.resv[0] || reg.resv[1] || reg.resv[2])
		return -EINVAL;
	if (reg.flags)
		return -EINVAL;

	bl = io_buffer_get_list(ctx, reg.bgid);
	if (!bl)
		return -ENOENT;
	if (!bl->is_mapped)
		return -EINVAL;

	__io_remove_buffers(ctx, bl, -1U);
	if (bl->bgid >= BGID_ARRAY) {
		xa_erase(&ctx->io_bl_xa, bl->bgid);
		kfree(bl);
	}
	return 0;
}

void *io_pbuf_get_address(struct io_ring_ctx *ctx, unsigned long bgid)
{
	struct io_buffer_list *bl;

	bl = io_buffer_get_list(ctx, bgid);
	if (!bl || !bl->is_mmap)
		return NULL;

	return bl->buf_ring;
}
Commit	Line	Data
3b77495a JA	1	// SPDX-License-Identifier: GPL-2.0
	2	#include <linux/kernel.h>
	3	#include <linux/errno.h>
	4	#include <linux/fs.h>
	5	#include <linux/file.h>
	6	#include <linux/mm.h>
	7	#include <linux/slab.h>
	8	#include <linux/namei.h>
	9	#include <linux/poll.h>
	10	#include <linux/io_uring.h>
	11
	12	#include <uapi/linux/io_uring.h>
	13
3b77495a JA	14	#include "io_uring.h"
	15	#include "opdef.h"
	16	#include "kbuf.h"
	17
	18	#define IO_BUFFER_LIST_BUF_PER_PAGE (PAGE_SIZE / sizeof(struct io_uring_buf))
	19
	20	#define BGID_ARRAY 64
	21
f74c746e GKB	22	/* BIDs are addressed by a 16-bit field in a CQE */
	23	#define MAX_BIDS_PER_BGID (1 << 16)
	24
b3a4dbc8 GKB	25	struct kmem_cache *io_buf_cachep;
b3a4dbc8 GKB	26
3b77495a JA	27	struct io_provide_buf {
	28	struct file *file;
	29	__u64 addr;
	30	__u32 len;
	31	__u32 bgid;
f74c746e	32	__u32 nbufs;
3b77495a JA	33	__u16 bid;
	34	};
	35
	36	static inline struct io_buffer_list io_buffer_get_list(struct io_ring_ctx ctx,
	37	unsigned int bgid)
	38	{
	39	if (ctx->io_bl && bgid < BGID_ARRAY)
	40	return &ctx->io_bl[bgid];
	41
	42	return xa_load(&ctx->io_bl_xa, bgid);
	43	}
	44
024b8fde HX	45	static int io_buffer_add_list(struct io_ring_ctx *ctx,
	46	struct io_buffer_list *bl, unsigned int bgid)
	47	{
	48	bl->bgid = bgid;
	49	if (bgid < BGID_ARRAY)
	50	return 0;
	51
	52	return xa_err(xa_store(&ctx->io_bl_xa, bgid, bl, GFP_KERNEL));
	53	}
	54
	55	void io_kbuf_recycle_legacy(struct io_kiocb *req, unsigned issue_flags)
3b77495a JA	56	{
	57	struct io_ring_ctx *ctx = req->ctx;
	58	struct io_buffer_list *bl;
	59	struct io_buffer *buf;
	60
	61	/*
024b8fde HX	62	* For legacy provided buffer mode, don't recycle if we already did
	63	* IO to this buffer. For ring-mapped provided buffer mode, we should
	64	* increment ring->head to explicitly monopolize the buffer to avoid
	65	* multiple use.
3b77495a	66	*/
024b8fde	67	if (req->flags & REQ_F_PARTIAL_IO)
3b77495a	68	return;
3b77495a JA	69
	70	io_ring_submit_lock(ctx, issue_flags);
	71
	72	buf = req->kbuf;
	73	bl = io_buffer_get_list(ctx, buf->bgid);
	74	list_add(&buf->list, &bl->buf_list);
	75	req->flags &= ~REQ_F_BUFFER_SELECTED;
	76	req->buf_index = buf->bgid;
	77
	78	io_ring_submit_unlock(ctx, issue_flags);
024b8fde	79	return;
3b77495a JA	80	}
3b77495a JA	81
53ccf69b PB	82	unsigned int __io_put_kbuf(struct io_kiocb *req, unsigned issue_flags)
	83	{
	84	unsigned int cflags;
	85
	86	/*
	87	* We can add this buffer back to two lists:
	88	*
	89	* 1) The io_buffers_cache list. This one is protected by the
	90	* ctx->uring_lock. If we already hold this lock, add back to this
	91	* list as we can grab it from issue as well.
	92	* 2) The io_buffers_comp list. This one is protected by the
	93	* ctx->completion_lock.
	94	*
	95	* We migrate buffers from the comp_list to the issue cache list
	96	* when we need one.
	97	*/
	98	if (req->flags & REQ_F_BUFFER_RING) {
	99	/* no buffers to recycle for this case */
	100	cflags = __io_put_kbuf_list(req, NULL);
	101	} else if (issue_flags & IO_URING_F_UNLOCKED) {
	102	struct io_ring_ctx *ctx = req->ctx;
	103
	104	spin_lock(&ctx->completion_lock);
	105	cflags = __io_put_kbuf_list(req, &ctx->io_buffers_comp);
	106	spin_unlock(&ctx->completion_lock);
	107	} else {
	108	lockdep_assert_held(&req->ctx->uring_lock);
	109
	110	cflags = __io_put_kbuf_list(req, &req->ctx->io_buffers_cache);
	111	}
	112	return cflags;
	113	}
	114
3b77495a JA	115	static void __user io_provided_buffer_select(struct io_kiocb req, size_t *len,
	116	struct io_buffer_list *bl)
	117	{
	118	if (!list_empty(&bl->buf_list)) {
	119	struct io_buffer *kbuf;
	120
	121	kbuf = list_first_entry(&bl->buf_list, struct io_buffer, list);
	122	list_del(&kbuf->list);
b8c01559	123	if (len == 0 \|\| len > kbuf->len)
3b77495a JA	124	*len = kbuf->len;
	125	req->flags \|= REQ_F_BUFFER_SELECTED;
	126	req->kbuf = kbuf;
	127	req->buf_index = kbuf->bid;
	128	return u64_to_user_ptr(kbuf->addr);
	129	}
	130	return NULL;
	131	}
	132
	133	static void __user io_ring_buffer_select(struct io_kiocb req, size_t *len,
	134	struct io_buffer_list *bl,
	135	unsigned int issue_flags)
	136	{
	137	struct io_uring_buf_ring *br = bl->buf_ring;
	138	struct io_uring_buf *buf;
	139	__u16 head = bl->head;
	140
	141	if (unlikely(smp_load_acquire(&br->tail) == head))
	142	return NULL;
	143
	144	head &= bl->mask;
c56e022c JA	145	/* mmaped buffers are always contig */
c56e022c JA	146	if (bl->is_mmap \|\| head < IO_BUFFER_LIST_BUF_PER_PAGE) {
3b77495a JA	147	buf = &br->bufs[head];
	148	} else {
	149	int off = head & (IO_BUFFER_LIST_BUF_PER_PAGE - 1);
	150	int index = head / IO_BUFFER_LIST_BUF_PER_PAGE;
	151	buf = page_address(bl->buf_pages[index]);
	152	buf += off;
	153	}
b8c01559	154	if (len == 0 \|\| len > buf->len)
3b77495a JA	155	*len = buf->len;
	156	req->flags \|= REQ_F_BUFFER_RING;
	157	req->buf_list = bl;
	158	req->buf_index = buf->bid;
	159
	160	if (issue_flags & IO_URING_F_UNLOCKED \|\| !file_can_poll(req->file)) {
	161	/*
	162	* If we came in unlocked, we have no choice but to consume the
f09c8643 HX	163	* buffer here, otherwise nothing ensures that the buffer won't
	164	* get used by others. This does mean it'll be pinned until the
	165	* IO completes, coming in unlocked means we're being called from
	166	* io-wq context and there may be further retries in async hybrid
	167	* mode. For the locked case, the caller must call commit when
	168	* the transfer completes (or if we get -EAGAIN and must poll of
	169	* retry).
3b77495a JA	170	*/
	171	req->buf_list = NULL;
	172	bl->head++;
	173	}
	174	return u64_to_user_ptr(buf->addr);
	175	}
	176
	177	void __user io_buffer_select(struct io_kiocb req, size_t *len,
	178	unsigned int issue_flags)
	179	{
	180	struct io_ring_ctx *ctx = req->ctx;
	181	struct io_buffer_list *bl;
	182	void __user *ret = NULL;
	183
	184	io_ring_submit_lock(req->ctx, issue_flags);
	185
	186	bl = io_buffer_get_list(ctx, req->buf_index);
	187	if (likely(bl)) {
25a2c188	188	if (bl->is_mapped)
3b77495a JA	189	ret = io_ring_buffer_select(req, len, bl, issue_flags);
	190	else
	191	ret = io_provided_buffer_select(req, len, bl);
	192	}
	193	io_ring_submit_unlock(req->ctx, issue_flags);
	194	return ret;
	195	}
	196
	197	static __cold int io_init_bl_list(struct io_ring_ctx *ctx)
	198	{
	199	int i;
	200
	201	ctx->io_bl = kcalloc(BGID_ARRAY, sizeof(struct io_buffer_list),
	202	GFP_KERNEL);
	203	if (!ctx->io_bl)
	204	return -ENOMEM;
	205
	206	for (i = 0; i < BGID_ARRAY; i++) {
	207	INIT_LIST_HEAD(&ctx->io_bl[i].buf_list);
	208	ctx->io_bl[i].bgid = i;
	209	}
	210
	211	return 0;
	212	}
	213
	214	static int __io_remove_buffers(struct io_ring_ctx *ctx,
	215	struct io_buffer_list *bl, unsigned nbufs)
	216	{
	217	unsigned i = 0;
	218
	219	/* shouldn't happen */
	220	if (!nbufs)
	221	return 0;
	222
c56e022c	223	if (bl->is_mapped) {
3b77495a	224	i = bl->buf_ring->tail - bl->head;
c56e022c	225	if (bl->is_mmap) {
99a9e0b8	226	folio_put(virt_to_folio(bl->buf_ring));
ceac766a	227	bl->buf_ring = NULL;
c56e022c JA	228	bl->is_mmap = 0;
	229	} else if (bl->buf_nr_pages) {
	230	int j;
	231
	232	for (j = 0; j < bl->buf_nr_pages; j++)
	233	unpin_user_page(bl->buf_pages[j]);
	234	kvfree(bl->buf_pages);
	235	bl->buf_pages = NULL;
	236	bl->buf_nr_pages = 0;
	237	}
3b77495a JA	238	/* make sure it's seen as empty */
3b77495a JA	239	INIT_LIST_HEAD(&bl->buf_list);
25a2c188	240	bl->is_mapped = 0;
3b77495a JA	241	return i;
	242	}
	243
b4a72c05 WL	244	/* protects io_buffers_cache */
	245	lockdep_assert_held(&ctx->uring_lock);
	246
3b77495a JA	247	while (!list_empty(&bl->buf_list)) {
	248	struct io_buffer *nxt;
	249
	250	nxt = list_first_entry(&bl->buf_list, struct io_buffer, list);
b4a72c05	251	list_move(&nxt->list, &ctx->io_buffers_cache);
3b77495a JA	252	if (++i == nbufs)
	253	return i;
	254	cond_resched();
	255	}
3b77495a JA	256
	257	return i;
	258	}
	259
	260	void io_destroy_buffers(struct io_ring_ctx *ctx)
	261	{
	262	struct io_buffer_list *bl;
b3a4dbc8 GKB	263	struct list_head item, tmp;
b3a4dbc8 GKB	264	struct io_buffer *buf;
3b77495a JA	265	unsigned long index;
	266	int i;
	267
	268	for (i = 0; i < BGID_ARRAY; i++) {
	269	if (!ctx->io_bl)
	270	break;
	271	__io_remove_buffers(ctx, &ctx->io_bl[i], -1U);
	272	}
	273
	274	xa_for_each(&ctx->io_bl_xa, index, bl) {
	275	xa_erase(&ctx->io_bl_xa, bl->bgid);
	276	__io_remove_buffers(ctx, bl, -1U);
	277	kfree(bl);
	278	}
	279
b3a4dbc8 GKB	280	list_for_each_safe(item, tmp, &ctx->io_buffers_cache) {
	281	buf = list_entry(item, struct io_buffer, list);
	282	kmem_cache_free(io_buf_cachep, buf);
3b77495a JA	283	}
	284	}
	285
	286	int io_remove_buffers_prep(struct io_kiocb req, const struct io_uring_sqe sqe)
	287	{
f2ccb5ae	288	struct io_provide_buf *p = io_kiocb_to_cmd(req, struct io_provide_buf);
3b77495a JA	289	u64 tmp;
	290
	291	if (sqe->rw_flags \|\| sqe->addr \|\| sqe->len \|\| sqe->off \|\|
	292	sqe->splice_fd_in)
	293	return -EINVAL;
	294
	295	tmp = READ_ONCE(sqe->fd);
f74c746e	296	if (!tmp \|\| tmp > MAX_BIDS_PER_BGID)
3b77495a JA	297	return -EINVAL;
	298
	299	memset(p, 0, sizeof(*p));
	300	p->nbufs = tmp;
	301	p->bgid = READ_ONCE(sqe->buf_group);
	302	return 0;
	303	}
	304
	305	int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags)
	306	{
f2ccb5ae	307	struct io_provide_buf *p = io_kiocb_to_cmd(req, struct io_provide_buf);
3b77495a JA	308	struct io_ring_ctx *ctx = req->ctx;
	309	struct io_buffer_list *bl;
	310	int ret = 0;
	311
	312	io_ring_submit_lock(ctx, issue_flags);
	313
	314	ret = -ENOENT;
	315	bl = io_buffer_get_list(ctx, p->bgid);
	316	if (bl) {
	317	ret = -EINVAL;
	318	/* can't use provide/remove buffers command on mapped buffers */
25a2c188	319	if (!bl->is_mapped)
3b77495a JA	320	ret = __io_remove_buffers(ctx, bl, p->nbufs);
3b77495a JA	321	}
c3b49093	322	io_ring_submit_unlock(ctx, issue_flags);
3b77495a JA	323	if (ret < 0)
3b77495a JA	324	req_set_fail(req);
3b77495a	325	io_req_set_res(req, ret, 0);
c3b49093	326	return IOU_OK;
3b77495a JA	327	}
	328
	329	int io_provide_buffers_prep(struct io_kiocb req, const struct io_uring_sqe sqe)
	330	{
	331	unsigned long size, tmp_check;
f2ccb5ae	332	struct io_provide_buf *p = io_kiocb_to_cmd(req, struct io_provide_buf);
3b77495a JA	333	u64 tmp;
	334
	335	if (sqe->rw_flags \|\| sqe->splice_fd_in)
	336	return -EINVAL;
	337
	338	tmp = READ_ONCE(sqe->fd);
f74c746e	339	if (!tmp \|\| tmp > MAX_BIDS_PER_BGID)
3b77495a JA	340	return -E2BIG;
	341	p->nbufs = tmp;
	342	p->addr = READ_ONCE(sqe->addr);
	343	p->len = READ_ONCE(sqe->len);
	344
	345	if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs,
	346	&size))
	347	return -EOVERFLOW;
	348	if (check_add_overflow((unsigned long)p->addr, size, &tmp_check))
	349	return -EOVERFLOW;
	350
	351	size = (unsigned long)p->len * p->nbufs;
	352	if (!access_ok(u64_to_user_ptr(p->addr), size))
	353	return -EFAULT;
	354
	355	p->bgid = READ_ONCE(sqe->buf_group);
	356	tmp = READ_ONCE(sqe->off);
	357	if (tmp > USHRT_MAX)
	358	return -E2BIG;
f74c746e	359	if (tmp + p->nbufs > MAX_BIDS_PER_BGID)
3851d25c	360	return -EINVAL;
3b77495a JA	361	p->bid = tmp;
	362	return 0;
	363	}
	364
b3a4dbc8 GKB	365	#define IO_BUFFER_ALLOC_BATCH 64
b3a4dbc8 GKB	366
3b77495a JA	367	static int io_refill_buffer_cache(struct io_ring_ctx *ctx)
3b77495a JA	368	{
b3a4dbc8 GKB	369	struct io_buffer *bufs[IO_BUFFER_ALLOC_BATCH];
b3a4dbc8 GKB	370	int allocated;
3b77495a JA	371
	372	/*
	373	* Completions that don't happen inline (eg not under uring_lock) will
	374	* add to ->io_buffers_comp. If we don't have any free buffers, check
	375	* the completion list and splice those entries first.
	376	*/
	377	if (!list_empty_careful(&ctx->io_buffers_comp)) {
	378	spin_lock(&ctx->completion_lock);
	379	if (!list_empty(&ctx->io_buffers_comp)) {
	380	list_splice_init(&ctx->io_buffers_comp,
	381	&ctx->io_buffers_cache);
	382	spin_unlock(&ctx->completion_lock);
	383	return 0;
	384	}
	385	spin_unlock(&ctx->completion_lock);
	386	}
	387
	388	/*
	389	* No free buffers and no completion entries either. Allocate a new
b3a4dbc8	390	* batch of buffer entries and add those to our freelist.
3b77495a	391	*/
3b77495a	392
b3a4dbc8 GKB	393	allocated = kmem_cache_alloc_bulk(io_buf_cachep, GFP_KERNEL_ACCOUNT,
	394	ARRAY_SIZE(bufs), (void **) bufs);
	395	if (unlikely(!allocated)) {
	396	/*
	397	* Bulk alloc is all-or-nothing. If we fail to get a batch,
	398	* retry single alloc to be on the safe side.
	399	*/
	400	bufs[0] = kmem_cache_alloc(io_buf_cachep, GFP_KERNEL);
	401	if (!bufs[0])
	402	return -ENOMEM;
	403	allocated = 1;
3b77495a JA	404	}
3b77495a JA	405
b3a4dbc8 GKB	406	while (allocated)
	407	list_add_tail(&bufs[--allocated]->list, &ctx->io_buffers_cache);
	408
3b77495a JA	409	return 0;
	410	}
	411
	412	static int io_add_buffers(struct io_ring_ctx ctx, struct io_provide_buf pbuf,
	413	struct io_buffer_list *bl)
	414	{
	415	struct io_buffer *buf;
	416	u64 addr = pbuf->addr;
	417	int i, bid = pbuf->bid;
	418
	419	for (i = 0; i < pbuf->nbufs; i++) {
	420	if (list_empty(&ctx->io_buffers_cache) &&
	421	io_refill_buffer_cache(ctx))
	422	break;
	423	buf = list_first_entry(&ctx->io_buffers_cache, struct io_buffer,
	424	list);
	425	list_move_tail(&buf->list, &bl->buf_list);
	426	buf->addr = addr;
	427	buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT);
	428	buf->bid = bid;
	429	buf->bgid = pbuf->bgid;
	430	addr += pbuf->len;
	431	bid++;
	432	cond_resched();
	433	}
	434
	435	return i ? 0 : -ENOMEM;
	436	}
	437
	438	int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
	439	{
f2ccb5ae	440	struct io_provide_buf *p = io_kiocb_to_cmd(req, struct io_provide_buf);
3b77495a JA	441	struct io_ring_ctx *ctx = req->ctx;
	442	struct io_buffer_list *bl;
	443	int ret = 0;
	444
	445	io_ring_submit_lock(ctx, issue_flags);
	446
	447	if (unlikely(p->bgid < BGID_ARRAY && !ctx->io_bl)) {
	448	ret = io_init_bl_list(ctx);
	449	if (ret)
	450	goto err;
	451	}
	452
	453	bl = io_buffer_get_list(ctx, p->bgid);
	454	if (unlikely(!bl)) {
cc18cc5e	455	bl = kzalloc(sizeof(*bl), GFP_KERNEL_ACCOUNT);
3b77495a JA	456	if (!bl) {
	457	ret = -ENOMEM;
	458	goto err;
	459	}
	460	INIT_LIST_HEAD(&bl->buf_list);
	461	ret = io_buffer_add_list(ctx, bl, p->bgid);
	462	if (ret) {
	463	kfree(bl);
	464	goto err;
	465	}
	466	}
	467	/* can't add buffers via this command for a mapped buffer ring */
25a2c188	468	if (bl->is_mapped) {
3b77495a JA	469	ret = -EINVAL;
	470	goto err;
	471	}
	472
	473	ret = io_add_buffers(ctx, p, bl);
	474	err:
c3b49093 PB	475	io_ring_submit_unlock(ctx, issue_flags);
c3b49093 PB	476
3b77495a JA	477	if (ret < 0)
3b77495a JA	478	req_set_fail(req);
3b77495a	479	io_req_set_res(req, ret, 0);
c3b49093	480	return IOU_OK;
3b77495a JA	481	}
3b77495a JA	482
ba56b632 JA	483	static int io_pin_pbuf_ring(struct io_uring_buf_reg *reg,
ba56b632 JA	484	struct io_buffer_list *bl)
3b77495a JA	485	{
3b77495a JA	486	struct io_uring_buf_ring *br;
3b77495a JA	487	struct page **pages;
	488	int nr_pages;
	489
ba56b632 JA	490	pages = io_pin_pages(reg->ring_addr,
	491	flex_array_size(br, bufs, reg->ring_entries),
	492	&nr_pages);
	493	if (IS_ERR(pages))
	494	return PTR_ERR(pages);
	495
	496	br = page_address(pages[0]);
fcb46c0c JA	497	#ifdef SHM_COLOUR
	498	/*
	499	* On platforms that have specific aliasing requirements, SHM_COLOUR
	500	* is set and we must guarantee that the kernel and user side align
	501	* nicely. We cannot do that if IOU_PBUF_RING_MMAP isn't set and
	502	* the application mmap's the provided ring buffer. Fail the request
	503	* if we, by chance, don't end up with aligned addresses. The app
	504	* should use IOU_PBUF_RING_MMAP instead, and liburing will handle
	505	* this transparently.
	506	*/
	507	if ((reg->ring_addr \| (unsigned long) br) & (SHM_COLOUR - 1)) {
	508	int i;
	509
	510	for (i = 0; i < nr_pages; i++)
	511	unpin_user_page(pages[i]);
	512	return -EINVAL;
	513	}
	514	#endif
ba56b632 JA	515	bl->buf_pages = pages;
	516	bl->buf_nr_pages = nr_pages;
	517	bl->buf_ring = br;
25a2c188	518	bl->is_mapped = 1;
c56e022c JA	519	bl->is_mmap = 0;
	520	return 0;
	521	}
	522
	523	static int io_alloc_pbuf_ring(struct io_uring_buf_reg *reg,
	524	struct io_buffer_list *bl)
	525	{
	526	gfp_t gfp = GFP_KERNEL_ACCOUNT \| __GFP_ZERO \| __GFP_NOWARN \| __GFP_COMP;
	527	size_t ring_size;
	528	void *ptr;
	529
	530	ring_size = reg->ring_entries * sizeof(struct io_uring_buf_ring);
	531	ptr = (void *) __get_free_pages(gfp, get_order(ring_size));
	532	if (!ptr)
	533	return -ENOMEM;
	534
	535	bl->buf_ring = ptr;
	536	bl->is_mapped = 1;
	537	bl->is_mmap = 1;
ba56b632 JA	538	return 0;
	539	}
	540
	541	int io_register_pbuf_ring(struct io_ring_ctx ctx, void __user arg)
	542	{
	543	struct io_uring_buf_reg reg;
	544	struct io_buffer_list bl, free_bl = NULL;
	545	int ret;
	546
3b77495a JA	547	if (copy_from_user(&reg, arg, sizeof(reg)))
	548	return -EFAULT;
	549
81cf17cd	550	if (reg.resv[0] \|\| reg.resv[1] \|\| reg.resv[2])
3b77495a	551	return -EINVAL;
c56e022c	552	if (reg.flags & ~IOU_PBUF_RING_MMAP)
3b77495a	553	return -EINVAL;
c56e022c JA	554	if (!(reg.flags & IOU_PBUF_RING_MMAP)) {
	555	if (!reg.ring_addr)
	556	return -EFAULT;
	557	if (reg.ring_addr & ~PAGE_MASK)
	558	return -EINVAL;
	559	} else {
	560	if (reg.ring_addr)
	561	return -EINVAL;
	562	}
	563
3b77495a JA	564	if (!is_power_of_2(reg.ring_entries))
	565	return -EINVAL;
	566
	567	/* cannot disambiguate full vs empty due to head/tail size */
	568	if (reg.ring_entries >= 65536)
	569	return -EINVAL;
	570
	571	if (unlikely(reg.bgid < BGID_ARRAY && !ctx->io_bl)) {
	572	int ret = io_init_bl_list(ctx);
	573	if (ret)
	574	return ret;
	575	}
	576
	577	bl = io_buffer_get_list(ctx, reg.bgid);
	578	if (bl) {
	579	/* if mapped buffer ring OR classic exists, don't allow */
25a2c188	580	if (bl->is_mapped \|\| !list_empty(&bl->buf_list))
3b77495a JA	581	return -EEXIST;
	582	} else {
	583	free_bl = bl = kzalloc(sizeof(*bl), GFP_KERNEL);
	584	if (!bl)
	585	return -ENOMEM;
	586	}
	587
c56e022c JA	588	if (!(reg.flags & IOU_PBUF_RING_MMAP))
	589	ret = io_pin_pbuf_ring(&reg, bl);
	590	else
	591	ret = io_alloc_pbuf_ring(&reg, bl);
3b77495a	592
c56e022c JA	593	if (!ret) {
	594	bl->nr_entries = reg.ring_entries;
	595	bl->mask = reg.ring_entries - 1;
ba56b632	596
c56e022c JA	597	io_buffer_add_list(ctx, bl, reg.bgid);
c56e022c JA	598	return 0;
3b77495a JA	599	}
3b77495a JA	600
c56e022c JA	601	kfree(free_bl);
c56e022c JA	602	return ret;
3b77495a JA	603	}
	604
	605	int io_unregister_pbuf_ring(struct io_ring_ctx ctx, void __user arg)
	606	{
	607	struct io_uring_buf_reg reg;
	608	struct io_buffer_list *bl;
	609
	610	if (copy_from_user(&reg, arg, sizeof(reg)))
	611	return -EFAULT;
81cf17cd JA	612	if (reg.resv[0] \|\| reg.resv[1] \|\| reg.resv[2])
	613	return -EINVAL;
	614	if (reg.flags)
3b77495a JA	615	return -EINVAL;
	616
	617	bl = io_buffer_get_list(ctx, reg.bgid);
	618	if (!bl)
	619	return -ENOENT;
25a2c188	620	if (!bl->is_mapped)
3b77495a JA	621	return -EINVAL;
	622
	623	__io_remove_buffers(ctx, bl, -1U);
	624	if (bl->bgid >= BGID_ARRAY) {
	625	xa_erase(&ctx->io_bl_xa, bl->bgid);
	626	kfree(bl);
	627	}
	628	return 0;
	629	}
c56e022c JA	630
	631	void io_pbuf_get_address(struct io_ring_ctx ctx, unsigned long bgid)
	632	{
	633	struct io_buffer_list *bl;
	634
	635	bl = io_buffer_get_list(ctx, bgid);
	636	if (!bl \|\| !bl->is_mmap)
	637	return NULL;
	638
	639	return bl->buf_ring;
	640	}