Re: 6.1 io_uring mmap backport

[Jens Axboe <axboe@kernel.dk>]

[-- Attachment #1: Type: text/plain, Size: 526 bytes --]

On 3/15/25 6:59 AM, Jens Axboe wrote:
> Hi,
> 
> I prepared this series about 6 months ago, but never got around to
> sending it in. In mainline, we got rid of remap_pfn_range() on the
> io_uring side, and this just backports it to 6.1-stable as well.
> This eliminates issues with fragmented memory, and hence it'd
> be nice to have it in 6.1 stable as well.

Just in case 6.1 runs into the same missing include issue as 6.6-stable,
here's a replacement patch 4 for that series as well that adds the
include.

-- 
Jens Axboe

[-- Attachment #2: 0004-io_uring-get-rid-of-remap_pfn_range-for-mapping-ring.patch --]
[-- Type: text/x-patch, Size: 8152 bytes --]

From 8c3571a1ebeca6037fa49f6192fd8fe45c61245c Mon Sep 17 00:00:00 2001
From: Jens Axboe <axboe@kernel.dk>
Date: Wed, 13 Mar 2024 09:56:14 -0600
Subject: [PATCH 4/6] io_uring: get rid of remap_pfn_range() for mapping
 rings/sqes

Commit 3ab1db3c6039e02a9deb9d5091d28d559917a645 upstream.

Rather than use remap_pfn_range() for this and manually free later,
switch to using vm_insert_pages() and have it Just Work.

If possible, allocate a single compound page that covers the range that
is needed. If that works, then we can just use page_address() on that
page. If we fail to get a compound page, allocate single pages and use
vmap() to map them into the kernel virtual address space.

This just covers the rings/sqes, the other remaining user of the mmap
remap_pfn_range() user will be converted separately. Once that is done,
we can kill the old alloc/free code.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/linux/io_uring_types.h |   5 ++
 include/uapi/linux/io_uring.h  |   1 +
 io_uring/io_uring.c            | 133 +++++++++++++++++++++++++++++----
 io_uring/io_uring.h            |   2 +
 4 files changed, 125 insertions(+), 16 deletions(-)

diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 37aeea266ebb..2bd01109e0ec 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -352,6 +352,11 @@ struct io_ring_ctx {
 	unsigned			sq_thread_idle;
 	/* protected by ->completion_lock */
 	unsigned			evfd_last_cq_tail;
+
+	unsigned short			n_ring_pages;
+	unsigned short			n_sqe_pages;
+	struct page			**ring_pages;
+	struct page			**sqe_pages;
 };
 
 enum {
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 9eff86acdfec..8f8e39cd183e 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -379,6 +379,7 @@ enum {
 #define IORING_OFF_SQ_RING		0ULL
 #define IORING_OFF_CQ_RING		0x8000000ULL
 #define IORING_OFF_SQES			0x10000000ULL
+#define IORING_OFF_MMAP_MASK		0xf8000000ULL
 
 /*
  * Filled with the offset for mmap(2)
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index b211feb0d2b1..7ce80510898f 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -71,6 +71,7 @@
 #include <linux/io_uring.h>
 #include <linux/audit.h>
 #include <linux/security.h>
+#include <linux/vmalloc.h>
 #include <asm/shmparam.h>
 
 #define CREATE_TRACE_POINTS
@@ -2513,37 +2514,118 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 	return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
 }
 
-static void io_mem_free(void *ptr)
+static void io_pages_unmap(void *ptr, struct page ***pages,
+			   unsigned short *npages)
 {
-	struct page *page;
+	bool do_vunmap = false;
 
 	if (!ptr)
 		return;
 
-	page = virt_to_head_page(ptr);
-	if (put_page_testzero(page))
-		free_compound_page(page);
+	if (*npages) {
+		struct page **to_free = *pages;
+		int i;
+
+		/*
+		 * Only did vmap for the non-compound multiple page case.
+		 * For the compound page, we just need to put the head.
+		 */
+		if (PageCompound(to_free[0]))
+			*npages = 1;
+		else if (*npages > 1)
+			do_vunmap = true;
+		for (i = 0; i < *npages; i++)
+			put_page(to_free[i]);
+	}
+	if (do_vunmap)
+		vunmap(ptr);
+	kvfree(*pages);
+	*pages = NULL;
+	*npages = 0;
 }
 
 static void io_rings_free(struct io_ring_ctx *ctx)
 {
-	io_mem_free(ctx->rings);
-	io_mem_free(ctx->sq_sqes);
+	io_pages_unmap(ctx->rings, &ctx->ring_pages, &ctx->n_ring_pages);
+	io_pages_unmap(ctx->sq_sqes, &ctx->sqe_pages, &ctx->n_sqe_pages);
 	ctx->rings = NULL;
 	ctx->sq_sqes = NULL;
 }
 
-static void *io_mem_alloc(size_t size)
+static void *io_mem_alloc_compound(struct page **pages, int nr_pages,
+				   size_t size, gfp_t gfp)
+{
+	struct page *page;
+	int i, order;
+
+	order = get_order(size);
+	if (order > 10)
+		return ERR_PTR(-ENOMEM);
+	else if (order)
+		gfp |= __GFP_COMP;
+
+	page = alloc_pages(gfp, order);
+	if (!page)
+		return ERR_PTR(-ENOMEM);
+
+	for (i = 0; i < nr_pages; i++)
+		pages[i] = page + i;
+
+	return page_address(page);
+}
+
+static void *io_mem_alloc_single(struct page **pages, int nr_pages, size_t size,
+				 gfp_t gfp)
 {
-	gfp_t gfp = GFP_KERNEL_ACCOUNT | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP;
 	void *ret;
+	int i;
 
-	ret = (void *) __get_free_pages(gfp, get_order(size));
+	for (i = 0; i < nr_pages; i++) {
+		pages[i] = alloc_page(gfp);
+		if (!pages[i])
+			goto err;
+	}
+
+	ret = vmap(pages, nr_pages, VM_MAP, PAGE_KERNEL);
 	if (ret)
 		return ret;
+err:
+	while (i--)
+		put_page(pages[i]);
 	return ERR_PTR(-ENOMEM);
 }
 
+static void *io_pages_map(struct page ***out_pages, unsigned short *npages,
+			  size_t size)
+{
+	gfp_t gfp = GFP_KERNEL_ACCOUNT | __GFP_ZERO | __GFP_NOWARN;
+	struct page **pages;
+	int nr_pages;
+	void *ret;
+ 
+	nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
+	pages = kvmalloc_array(nr_pages, sizeof(struct page *), gfp);
+	if (!pages)
+		return ERR_PTR(-ENOMEM);
+
+	ret = io_mem_alloc_compound(pages, nr_pages, size, gfp);
+	if (!IS_ERR(ret))
+		goto done;
+
+	ret = io_mem_alloc_single(pages, nr_pages, size, gfp);
+	if (!IS_ERR(ret)) {
+done:
+		*out_pages = pages;
+		*npages = nr_pages;
+		return ret;
+	}
+
+	kvfree(pages);
+	*out_pages = NULL;
+	*npages = 0;
+	return ret;
+}
+
 static unsigned long rings_size(struct io_ring_ctx *ctx, unsigned int sq_entries,
 				unsigned int cq_entries, size_t *sq_offset)
 {
@@ -3125,11 +3207,9 @@ static void *io_uring_validate_mmap_request(struct file *file,
 	switch (offset) {
 	case IORING_OFF_SQ_RING:
 	case IORING_OFF_CQ_RING:
-		ptr = ctx->rings;
-		break;
+		return ctx->rings;
 	case IORING_OFF_SQES:
-		ptr = ctx->sq_sqes;
-		break;
+		return ctx->sq_sqes;
 	default:
 		return ERR_PTR(-EINVAL);
 	}
@@ -3141,11 +3221,22 @@ static void *io_uring_validate_mmap_request(struct file *file,
 	return ptr;
 }
 
+int io_uring_mmap_pages(struct io_ring_ctx *ctx, struct vm_area_struct *vma,
+			struct page **pages, int npages)
+{
+	unsigned long nr_pages = npages;
+
+	vma->vm_flags |= VM_DONTEXPAND;
+	return vm_insert_pages(vma, vma->vm_start, pages, &nr_pages);
+}
+
 #ifdef CONFIG_MMU
 
 static __cold int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
 {
+	struct io_ring_ctx *ctx = file->private_data;
 	size_t sz = vma->vm_end - vma->vm_start;
+	long offset = vma->vm_pgoff << PAGE_SHIFT;
 	unsigned long pfn;
 	void *ptr;
 
@@ -3153,6 +3244,16 @@ static __cold int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
 	if (IS_ERR(ptr))
 		return PTR_ERR(ptr);
 
+	switch (offset & IORING_OFF_MMAP_MASK) {
+	case IORING_OFF_SQ_RING:
+	case IORING_OFF_CQ_RING:
+		return io_uring_mmap_pages(ctx, vma, ctx->ring_pages,
+						ctx->n_ring_pages);
+	case IORING_OFF_SQES:
+		return io_uring_mmap_pages(ctx, vma, ctx->sqe_pages,
+						ctx->n_sqe_pages);
+	}
+
 	pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
 	return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
 }
@@ -3443,7 +3544,7 @@ static __cold int io_allocate_scq_urings(struct io_ring_ctx *ctx,
 	if (size == SIZE_MAX)
 		return -EOVERFLOW;
 
-	rings = io_mem_alloc(size);
+	rings = io_pages_map(&ctx->ring_pages, &ctx->n_ring_pages, size);
 	if (IS_ERR(rings))
 		return PTR_ERR(rings);
 
@@ -3463,7 +3564,7 @@ static __cold int io_allocate_scq_urings(struct io_ring_ctx *ctx,
 		return -EOVERFLOW;
 	}
 
-	ptr = io_mem_alloc(size);
+	ptr = io_pages_map(&ctx->sqe_pages, &ctx->n_sqe_pages, size);
 	if (IS_ERR(ptr)) {
 		io_rings_free(ctx);
 		return PTR_ERR(ptr);
diff --git a/io_uring/io_uring.h b/io_uring/io_uring.h
index a1f679b8199e..886921d2d58d 100644
--- a/io_uring/io_uring.h
+++ b/io_uring/io_uring.h
@@ -41,6 +41,8 @@ bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags
 void __io_commit_cqring_flush(struct io_ring_ctx *ctx);
 
 struct page **io_pin_pages(unsigned long ubuf, unsigned long len, int *npages);
+int io_uring_mmap_pages(struct io_ring_ctx *ctx, struct vm_area_struct *vma,
+			struct page **pages, int npages);
 
 struct file *io_file_get_normal(struct io_kiocb *req, int fd);
 struct file *io_file_get_fixed(struct io_kiocb *req, int fd,
-- 
2.49.0