1 files changed, 264 insertions, 314 deletions

diff --git a/io_uring/kbuf.c b/io_uring/kbuf.c
index 18df5a9d2f5e..8e72de7712ac 100644
--- a/io_uring/kbuf.c
+++ b/io_uring/kbuf.c
@@ -7,6 +7,7 @@
 #include <linux/slab.h>
 #include <linux/namei.h>
 #include <linux/poll.h>
+#include <linux/vmalloc.h>
 #include <linux/io_uring.h>
 
 #include <uapi/linux/io_uring.h>
@@ -14,10 +15,7 @@
 #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
+#include "memmap.h"
 
 /* BIDs are addressed by a 16-bit field in a CQE */
 #define MAX_BIDS_PER_BGID (1 << 16)
@@ -33,29 +31,12 @@ struct io_provide_buf {
 	__u16				bid;
 };
 
-struct io_buf_free {
-	struct hlist_node		list;
-	void				*mem;
-	size_t				size;
-	int				inuse;
-};
-
-static struct io_buffer_list *__io_buffer_get_list(struct io_ring_ctx *ctx,
-						   struct io_buffer_list *bl,
-						   unsigned int bgid)
-{
-	if (bl && bgid < BGID_ARRAY)
-		return &bl[bgid];
-
-	return xa_load(&ctx->io_bl_xa, bgid);
-}
-
 static inline struct io_buffer_list *io_buffer_get_list(struct io_ring_ctx *ctx,
 							unsigned int bgid)
 {
 	lockdep_assert_held(&ctx->uring_lock);
 
-	return __io_buffer_get_list(ctx, ctx->io_bl, bgid);
+	return xa_load(&ctx->io_bl_xa, bgid);
 }
 
 static int io_buffer_add_list(struct io_ring_ctx *ctx,
@@ -64,14 +45,10 @@ static int io_buffer_add_list(struct io_ring_ctx *ctx,
 	/*
 	 * Store buffer group ID and finally mark the list as visible.
 	 * The normal lookup doesn't care about the visibility as we're
-	 * always under the ->uring_lock, but the RCU lookup from mmap does.
+	 * always under the ->uring_lock, but lookups from mmap do.
 	 */
 	bl->bgid = bgid;
-	smp_store_release(&bl->is_ready, 1);
-
-	if (bgid < BGID_ARRAY)
-		return 0;
-
+	guard(mutex)(&ctx->mmap_lock);
 	return xa_err(xa_store(&ctx->io_bl_xa, bgid, bl, GFP_KERNEL));
 }
 
@@ -81,15 +58,6 @@ bool io_kbuf_recycle_legacy(struct io_kiocb *req, unsigned issue_flags)
 	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 false;
-
 	io_ring_submit_lock(ctx, issue_flags);
 
 	buf = req->kbuf;
@@ -102,10 +70,8 @@ bool io_kbuf_recycle_legacy(struct io_kiocb *req, unsigned issue_flags)
 	return true;
 }
 
-unsigned int __io_put_kbuf(struct io_kiocb *req, unsigned issue_flags)
+void __io_put_kbuf(struct io_kiocb *req, int len, unsigned issue_flags)
 {
-	unsigned int cflags;
-
 	/*
 	 * We can add this buffer back to two lists:
 	 *
@@ -118,21 +84,17 @@ unsigned int __io_put_kbuf(struct io_kiocb *req, unsigned issue_flags)
 	 * 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) {
+	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);
+		__io_put_kbuf_list(req, len, &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);
+		__io_put_kbuf_list(req, len, &req->ctx->io_buffers_cache);
 	}
-	return cflags;
 }
 
 static void __user *io_provided_buffer_select(struct io_kiocb *req, size_t *len,
@@ -145,6 +107,8 @@ static void __user *io_provided_buffer_select(struct io_kiocb *req, size_t *len,
 		list_del(&kbuf->list);
 		if (*len == 0 || *len > kbuf->len)
 			*len = kbuf->len;
+		if (list_empty(&bl->buf_list))
+			req->flags |= REQ_F_BL_EMPTY;
 		req->flags |= REQ_F_BUFFER_SELECTED;
 		req->kbuf = kbuf;
 		req->buf_index = kbuf->bid;
@@ -153,34 +117,46 @@ static void __user *io_provided_buffer_select(struct io_kiocb *req, size_t *len,
 	return NULL;
 }
 
+static int io_provided_buffers_select(struct io_kiocb *req, size_t *len,
+				      struct io_buffer_list *bl,
+				      struct iovec *iov)
+{
+	void __user *buf;
+
+	buf = io_provided_buffer_select(req, len, bl);
+	if (unlikely(!buf))
+		return -ENOBUFS;
+
+	iov[0].iov_base = buf;
+	iov[0].iov_len = *len;
+	return 1;
+}
+
 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;
+	__u16 tail, head = bl->head;
 	struct io_uring_buf *buf;
-	__u16 head = bl->head;
+	void __user *ret;
 
-	if (unlikely(smp_load_acquire(&br->tail) == head))
+	tail = smp_load_acquire(&br->tail);
+	if (unlikely(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 (head + 1 == tail)
+		req->flags |= REQ_F_BL_EMPTY;
+
+	buf = io_ring_head_to_buf(br, head, bl->mask);
 	if (*len == 0 || *len > buf->len)
 		*len = buf->len;
-	req->flags |= REQ_F_BUFFER_RING;
+	req->flags |= REQ_F_BUFFER_RING | REQ_F_BUFFERS_COMMIT;
 	req->buf_list = bl;
 	req->buf_index = buf->bid;
+	ret = u64_to_user_ptr(buf->addr);
 
-	if (issue_flags & IO_URING_F_UNLOCKED || !file_can_poll(req->file)) {
+	if (issue_flags & IO_URING_F_UNLOCKED || !io_file_can_poll(req)) {
 		/*
 		 * If we came in unlocked, we have no choice but to consume the
 		 * buffer here, otherwise nothing ensures that the buffer won't
@@ -191,10 +167,10 @@ static void __user *io_ring_buffer_select(struct io_kiocb *req, size_t *len,
 		 * the transfer completes (or if we get -EAGAIN and must poll of
 		 * retry).
 		 */
+		io_kbuf_commit(req, bl, *len, 1);
 		req->buf_list = NULL;
-		bl->head++;
 	}
-	return u64_to_user_ptr(buf->addr);
+	return ret;
 }
 
 void __user *io_buffer_select(struct io_kiocb *req, size_t *len,
@@ -208,7 +184,7 @@ void __user *io_buffer_select(struct io_kiocb *req, size_t *len,
 
 	bl = io_buffer_get_list(ctx, req->buf_index);
 	if (likely(bl)) {
-		if (bl->is_mapped)
+		if (bl->flags & IOBL_BUF_RING)
 			ret = io_ring_buffer_select(req, len, bl, issue_flags);
 		else
 			ret = io_provided_buffer_select(req, len, bl);
@@ -217,40 +193,153 @@ void __user *io_buffer_select(struct io_kiocb *req, size_t *len,
 	return ret;
 }
 
-static __cold int io_init_bl_list(struct io_ring_ctx *ctx)
+/* cap it at a reasonable 256, will be one page even for 4K */
+#define PEEK_MAX_IMPORT		256
+
+static int io_ring_buffers_peek(struct io_kiocb *req, struct buf_sel_arg *arg,
+				struct io_buffer_list *bl)
 {
-	struct io_buffer_list *bl;
-	int i;
+	struct io_uring_buf_ring *br = bl->buf_ring;
+	struct iovec *iov = arg->iovs;
+	int nr_iovs = arg->nr_iovs;
+	__u16 nr_avail, tail, head;
+	struct io_uring_buf *buf;
 
-	bl = kcalloc(BGID_ARRAY, sizeof(struct io_buffer_list), GFP_KERNEL);
-	if (!bl)
-		return -ENOMEM;
+	tail = smp_load_acquire(&br->tail);
+	head = bl->head;
+	nr_avail = min_t(__u16, tail - head, UIO_MAXIOV);
+	if (unlikely(!nr_avail))
+		return -ENOBUFS;
+
+	buf = io_ring_head_to_buf(br, head, bl->mask);
+	if (arg->max_len) {
+		u32 len = READ_ONCE(buf->len);
 
-	for (i = 0; i < BGID_ARRAY; i++) {
-		INIT_LIST_HEAD(&bl[i].buf_list);
-		bl[i].bgid = i;
+		if (unlikely(!len))
+			return -ENOBUFS;
+		/*
+		 * Limit incremental buffers to 1 segment. No point trying
+		 * to peek ahead and map more than we need, when the buffers
+		 * themselves should be large when setup with
+		 * IOU_PBUF_RING_INC.
+		 */
+		if (bl->flags & IOBL_INC) {
+			nr_avail = 1;
+		} else {
+			size_t needed;
+
+			needed = (arg->max_len + len - 1) / len;
+			needed = min_not_zero(needed, (size_t) PEEK_MAX_IMPORT);
+			if (nr_avail > needed)
+				nr_avail = needed;
+		}
 	}
 
-	smp_store_release(&ctx->io_bl, bl);
-	return 0;
+	/*
+	 * only alloc a bigger array if we know we have data to map, eg not
+	 * a speculative peek operation.
+	 */
+	if (arg->mode & KBUF_MODE_EXPAND && nr_avail > nr_iovs && arg->max_len) {
+		iov = kmalloc_array(nr_avail, sizeof(struct iovec), GFP_KERNEL);
+		if (unlikely(!iov))
+			return -ENOMEM;
+		if (arg->mode & KBUF_MODE_FREE)
+			kfree(arg->iovs);
+		arg->iovs = iov;
+		nr_iovs = nr_avail;
+	} else if (nr_avail < nr_iovs) {
+		nr_iovs = nr_avail;
+	}
+
+	/* set it to max, if not set, so we can use it unconditionally */
+	if (!arg->max_len)
+		arg->max_len = INT_MAX;
+
+	req->buf_index = buf->bid;
+	do {
+		u32 len = buf->len;
+
+		/* truncate end piece, if needed, for non partial buffers */
+		if (len > arg->max_len) {
+			len = arg->max_len;
+			if (!(bl->flags & IOBL_INC))
+				buf->len = len;
+		}
+
+		iov->iov_base = u64_to_user_ptr(buf->addr);
+		iov->iov_len = len;
+		iov++;
+
+		arg->out_len += len;
+		arg->max_len -= len;
+		if (!arg->max_len)
+			break;
+
+		buf = io_ring_head_to_buf(br, ++head, bl->mask);
+	} while (--nr_iovs);
+
+	if (head == tail)
+		req->flags |= REQ_F_BL_EMPTY;
+
+	req->flags |= REQ_F_BUFFER_RING;
+	req->buf_list = bl;
+	return iov - arg->iovs;
 }
 
-/*
- * Mark the given mapped range as free for reuse
- */
-static void io_kbuf_mark_free(struct io_ring_ctx *ctx, struct io_buffer_list *bl)
+int io_buffers_select(struct io_kiocb *req, struct buf_sel_arg *arg,
+		      unsigned int issue_flags)
 {
-	struct io_buf_free *ibf;
+	struct io_ring_ctx *ctx = req->ctx;
+	struct io_buffer_list *bl;
+	int ret = -ENOENT;
+
+	io_ring_submit_lock(ctx, issue_flags);
+	bl = io_buffer_get_list(ctx, req->buf_index);
+	if (unlikely(!bl))
+		goto out_unlock;
 
-	hlist_for_each_entry(ibf, &ctx->io_buf_list, list) {
-		if (bl->buf_ring == ibf->mem) {
-			ibf->inuse = 0;
-			return;
+	if (bl->flags & IOBL_BUF_RING) {
+		ret = io_ring_buffers_peek(req, arg, bl);
+		/*
+		 * Don't recycle these buffers if we need to go through poll.
+		 * Nobody else can use them anyway, and holding on to provided
+		 * buffers for a send/write operation would happen on the app
+		 * side anyway with normal buffers. Besides, we already
+		 * committed them, they cannot be put back in the queue.
+		 */
+		if (ret > 0) {
+			req->flags |= REQ_F_BUFFERS_COMMIT | REQ_F_BL_NO_RECYCLE;
+			io_kbuf_commit(req, bl, arg->out_len, ret);
 		}
+	} else {
+		ret = io_provided_buffers_select(req, &arg->out_len, bl, arg->iovs);
+	}
+out_unlock:
+	io_ring_submit_unlock(ctx, issue_flags);
+	return ret;
+}
+
+int io_buffers_peek(struct io_kiocb *req, struct buf_sel_arg *arg)
+{
+	struct io_ring_ctx *ctx = req->ctx;
+	struct io_buffer_list *bl;
+	int ret;
+
+	lockdep_assert_held(&ctx->uring_lock);
+
+	bl = io_buffer_get_list(ctx, req->buf_index);
+	if (unlikely(!bl))
+		return -ENOENT;
+
+	if (bl->flags & IOBL_BUF_RING) {
+		ret = io_ring_buffers_peek(req, arg, bl);
+		if (ret > 0)
+			req->flags |= REQ_F_BUFFERS_COMMIT;
+		return ret;
 	}
 
-	/* can't happen... */
-	WARN_ON_ONCE(1);
+	/* don't support multiple buffer selections for legacy */
+	return io_provided_buffers_select(req, &arg->max_len, bl, arg->iovs);
 }
 
 static int __io_remove_buffers(struct io_ring_ctx *ctx,
@@ -262,28 +351,12 @@ static int __io_remove_buffers(struct io_ring_ctx *ctx,
 	if (!nbufs)
 		return 0;
 
-	if (bl->is_mapped) {
+	if (bl->flags & IOBL_BUF_RING) {
 		i = bl->buf_ring->tail - bl->head;
-		if (bl->is_mmap) {
-			/*
-			 * io_kbuf_list_free() will free the page(s) at
-			 * ->release() time.
-			 */
-			io_kbuf_mark_free(ctx, bl);
-			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;
-		}
+		io_free_region(ctx, &bl->region);
 		/* make sure it's seen as empty */
 		INIT_LIST_HEAD(&bl->buf_list);
-		bl->is_mapped = 0;
+		bl->flags &= ~IOBL_BUF_RING;
 		return i;
 	}
 
@@ -303,24 +376,29 @@ static int __io_remove_buffers(struct io_ring_ctx *ctx,
 	return i;
 }
 
+static void io_put_bl(struct io_ring_ctx *ctx, struct io_buffer_list *bl)
+{
+	__io_remove_buffers(ctx, bl, -1U);
+	kfree(bl);
+}
+
 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);
-	}
+	while (1) {
+		unsigned long index = 0;
 
-	xa_for_each(&ctx->io_bl_xa, index, bl) {
-		xa_erase(&ctx->io_bl_xa, bl->bgid);
-		__io_remove_buffers(ctx, bl, -1U);
-		kfree_rcu(bl, rcu);
+		scoped_guard(mutex, &ctx->mmap_lock) {
+			bl = xa_find(&ctx->io_bl_xa, &index, ULONG_MAX, XA_PRESENT);
+			if (bl)
+				xa_erase(&ctx->io_bl_xa, bl->bgid);
+		}
+		if (!bl)
+			break;
+		io_put_bl(ctx, bl);
 	}
 
 	/*
@@ -337,6 +415,13 @@ void io_destroy_buffers(struct io_ring_ctx *ctx)
 	}
 }
 
+static void io_destroy_bl(struct io_ring_ctx *ctx, struct io_buffer_list *bl)
+{
+	scoped_guard(mutex, &ctx->mmap_lock)
+		WARN_ON_ONCE(xa_erase(&ctx->io_bl_xa, bl->bgid) != bl);
+	io_put_bl(ctx, bl);
+}
+
 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);
@@ -370,7 +455,7 @@ int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags)
 	if (bl) {
 		ret = -EINVAL;
 		/* can't use provide/remove buffers command on mapped buffers */
-		if (!bl->is_mapped)
+		if (!(bl->flags & IOBL_BUF_RING))
 			ret = __io_remove_buffers(ctx, bl, p->nbufs);
 	}
 	io_ring_submit_unlock(ctx, issue_flags);
@@ -498,12 +583,6 @@ int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
 
 	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);
@@ -514,21 +593,12 @@ int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
 		INIT_LIST_HEAD(&bl->buf_list);
 		ret = io_buffer_add_list(ctx, bl, p->bgid);
 		if (ret) {
-			/*
-			 * Doesn't need rcu free as it was never visible, but
-			 * let's keep it consistent throughout. Also can't
-			 * be a lower indexed array group, as adding one
-			 * where lookup failed cannot happen.
-			 */
-			if (p->bgid >= BGID_ARRAY)
-				kfree_rcu(bl, rcu);
-			else
-				WARN_ON_ONCE(1);
+			kfree(bl);
 			goto err;
 		}
 	}
 	/* can't add buffers via this command for a mapped buffer ring */
-	if (bl->is_mapped) {
+	if (bl->flags & IOBL_BUF_RING) {
 		ret = -EINVAL;
 		goto err;
 	}
@@ -543,123 +613,14 @@ err:
 	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 i, 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);
-
-	/*
-	 * Apparently some 32-bit boxes (ARM) will return highmem pages,
-	 * which then need to be mapped. We could support that, but it'd
-	 * complicate the code and slowdown the common cases quite a bit.
-	 * So just error out, returning -EINVAL just like we did on kernels
-	 * that didn't support mapped buffer rings.
-	 */
-	for (i = 0; i < nr_pages; i++)
-		if (PageHighMem(pages[i]))
-			goto error_unpin;
-
-	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))
-		goto error_unpin;
-#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;
-error_unpin:
-	for (i = 0; i < nr_pages; i++)
-		unpin_user_page(pages[i]);
-	kvfree(pages);
-	return -EINVAL;
-}
-
-/*
- * See if we have a suitable region that we can reuse, rather than allocate
- * both a new io_buf_free and mem region again. We leave it on the list as
- * even a reused entry will need freeing at ring release.
- */
-static struct io_buf_free *io_lookup_buf_free_entry(struct io_ring_ctx *ctx,
-						    size_t ring_size)
-{
-	struct io_buf_free *ibf, *best = NULL;
-	size_t best_dist;
-
-	hlist_for_each_entry(ibf, &ctx->io_buf_list, list) {
-		size_t dist;
-
-		if (ibf->inuse || ibf->size < ring_size)
-			continue;
-		dist = ibf->size - ring_size;
-		if (!best || dist < best_dist) {
-			best = ibf;
-			if (!dist)
-				break;
-			best_dist = dist;
-		}
-	}
-
-	return best;
-}
-
-static int io_alloc_pbuf_ring(struct io_ring_ctx *ctx,
-			      struct io_uring_buf_reg *reg,
-			      struct io_buffer_list *bl)
-{
-	struct io_buf_free *ibf;
-	size_t ring_size;
-	void *ptr;
-
-	ring_size = reg->ring_entries * sizeof(struct io_uring_buf_ring);
-
-	/* Reuse existing entry, if we can */
-	ibf = io_lookup_buf_free_entry(ctx, ring_size);
-	if (!ibf) {
-		ptr = io_mem_alloc(ring_size);
-		if (IS_ERR(ptr))
-			return PTR_ERR(ptr);
-
-		/* Allocate and store deferred free entry */
-		ibf = kmalloc(sizeof(*ibf), GFP_KERNEL_ACCOUNT);
-		if (!ibf) {
-			io_mem_free(ptr);
-			return -ENOMEM;
-		}
-		ibf->mem = ptr;
-		ibf->size = ring_size;
-		hlist_add_head(&ibf->list, &ctx->io_buf_list);
-	}
-	ibf->inuse = 1;
-	bl->buf_ring = ibf->mem;
-	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;
+	struct io_uring_region_desc rd;
+	struct io_uring_buf_ring *br;
+	unsigned long mmap_offset;
+	unsigned long ring_size;
 	int ret;
 
 	lockdep_assert_held(&ctx->uring_lock);
@@ -669,56 +630,68 @@ int io_register_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg)
 
 	if (reg.resv[0] || reg.resv[1] || reg.resv[2])
 		return -EINVAL;
-	if (reg.flags & ~IOU_PBUF_RING_MMAP)
+	if (reg.flags & ~(IOU_PBUF_RING_MMAP | IOU_PBUF_RING_INC))
 		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))
+		if (bl->flags & IOBL_BUF_RING || !list_empty(&bl->buf_list))
 			return -EEXIST;
-	} else {
-		free_bl = bl = kzalloc(sizeof(*bl), GFP_KERNEL);
-		if (!bl)
-			return -ENOMEM;
+		io_destroy_bl(ctx, bl);
 	}
 
-	if (!(reg.flags & IOU_PBUF_RING_MMAP))
-		ret = io_pin_pbuf_ring(&reg, bl);
-	else
-		ret = io_alloc_pbuf_ring(ctx, &reg, bl);
+	free_bl = bl = kzalloc(sizeof(*bl), GFP_KERNEL);
+	if (!bl)
+		return -ENOMEM;
 
-	if (!ret) {
-		bl->nr_entries = reg.ring_entries;
-		bl->mask = reg.ring_entries - 1;
+	mmap_offset = (unsigned long)reg.bgid << IORING_OFF_PBUF_SHIFT;
+	ring_size = flex_array_size(br, bufs, reg.ring_entries);
 
-		io_buffer_add_list(ctx, bl, reg.bgid);
-		return 0;
+	memset(&rd, 0, sizeof(rd));
+	rd.size = PAGE_ALIGN(ring_size);
+	if (!(reg.flags & IOU_PBUF_RING_MMAP)) {
+		rd.user_addr = reg.ring_addr;
+		rd.flags |= IORING_MEM_REGION_TYPE_USER;
+	}
+	ret = io_create_region_mmap_safe(ctx, &bl->region, &rd, mmap_offset);
+	if (ret)
+		goto fail;
+	br = io_region_get_ptr(&bl->region);
+
+#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.flags & IOU_PBUF_RING_MMAP) &&
+	    ((reg.ring_addr | (unsigned long)br) & (SHM_COLOUR - 1))) {
+		ret = -EINVAL;
+		goto fail;
 	}
+#endif
 
-	kfree_rcu(free_bl, rcu);
+	bl->nr_entries = reg.ring_entries;
+	bl->mask = reg.ring_entries - 1;
+	bl->flags |= IOBL_BUF_RING;
+	bl->buf_ring = br;
+	if (reg.flags & IOU_PBUF_RING_INC)
+		bl->flags |= IOBL_INC;
+	io_buffer_add_list(ctx, bl, reg.bgid);
+	return 0;
+fail:
+	io_free_region(ctx, &bl->region);
+	kfree(free_bl);
 	return ret;
 }
 
@@ -739,14 +712,13 @@ int io_unregister_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg)
 	bl = io_buffer_get_list(ctx, reg.bgid);
 	if (!bl)
 		return -ENOENT;
-	if (!bl->is_mapped)
+	if (!(bl->flags & IOBL_BUF_RING))
 		return -EINVAL;
 
-	__io_remove_buffers(ctx, bl, -1U);
-	if (bl->bgid >= BGID_ARRAY) {
+	scoped_guard(mutex, &ctx->mmap_lock)
 		xa_erase(&ctx->io_bl_xa, bl->bgid);
-		kfree_rcu(bl, rcu);
-	}
+
+	io_put_bl(ctx, bl);
 	return 0;
 }
 
@@ -766,7 +738,7 @@ int io_register_pbuf_status(struct io_ring_ctx *ctx, void __user *arg)
 	bl = io_buffer_get_list(ctx, buf_status.buf_group);
 	if (!bl)
 		return -ENOENT;
-	if (!bl->is_mapped)
+	if (!(bl->flags & IOBL_BUF_RING))
 		return -EINVAL;
 
 	buf_status.head = bl->head;
@@ -776,37 +748,15 @@ int io_register_pbuf_status(struct io_ring_ctx *ctx, void __user *arg)
 	return 0;
 }
 
-void *io_pbuf_get_address(struct io_ring_ctx *ctx, unsigned long bgid)
+struct io_mapped_region *io_pbuf_get_region(struct io_ring_ctx *ctx,
+					    unsigned int bgid)
 {
 	struct io_buffer_list *bl;
 
-	bl = __io_buffer_get_list(ctx, smp_load_acquire(&ctx->io_bl), bgid);
+	lockdep_assert_held(&ctx->mmap_lock);
 
-	if (!bl || !bl->is_mmap)
-		return NULL;
-	/*
-	 * Ensure the list is fully setup. Only strictly needed for RCU lookup
-	 * via mmap, and in that case only for the array indexed groups. For
-	 * the xarray lookups, it's either visible and ready, or not at all.
-	 */
-	if (!smp_load_acquire(&bl->is_ready))
+	bl = xa_load(&ctx->io_bl_xa, bgid);
+	if (!bl || !(bl->flags & IOBL_BUF_RING))
 		return NULL;
-
-	return bl->buf_ring;
-}
-
-/*
- * Called at or after ->release(), free the mmap'ed buffers that we used
- * for memory mapped provided buffer rings.
- */
-void io_kbuf_mmap_list_free(struct io_ring_ctx *ctx)
-{
-	struct io_buf_free *ibf;
-	struct hlist_node *tmp;
-
-	hlist_for_each_entry_safe(ibf, tmp, &ctx->io_buf_list, list) {
-		hlist_del(&ibf->list);
-		io_mem_free(ibf->mem);
-		kfree(ibf);
-	}
+	return &bl->region;
 }