[PATCHSET 0/3] io_uring: add support for madvise/fadvise

[PATCHSET 0/3] io_uring: add support for madvise/fadvise

[Jens Axboe]

Nothing earth shattering here, just wiring up madvise and fadvise for
io_uring. The madvise part requires a small prep patch.

-- 
Jens Axboe

[PATCH 1/3] io_uring: add IORING_OP_FADVISE

[Jens Axboe]

This adds support for doing fadvise through io_uring. We assume that
WILLNEED doesn't block, but that DONTNEED may block.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c                 | 53 +++++++++++++++++++++++++++++++++++
 include/uapi/linux/io_uring.h |  2 ++
 2 files changed, 55 insertions(+)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 62459a79a61f..0b200a7d4ae0 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -72,6 +72,7 @@
 #include <linux/highmem.h>
 #include <linux/namei.h>
 #include <linux/fsnotify.h>
+#include <linux/fadvise.h>
 
 #define CREATE_TRACE_POINTS
 #include <trace/events/io_uring.h>
@@ -400,6 +401,13 @@ struct io_files_update {
 	u32				offset;
 };
 
+struct io_fadvise {
+	struct file			*file;
+	u64				offset;
+	u32				len;
+	u32				advice;
+};
+
 struct io_async_connect {
 	struct sockaddr_storage		address;
 };
@@ -452,6 +460,7 @@ struct io_kiocb {
 		struct io_open		open;
 		struct io_close		close;
 		struct io_files_update	files_update;
+		struct io_fadvise	fadvise;
 	};
 
 	struct io_async_ctx		*io;
@@ -669,6 +678,10 @@ static const struct io_op_def io_op_defs[] = {
 		.needs_file		= 1,
 		.unbound_nonreg_file	= 1,
 	},
+	{
+		/* IORING_OP_FADVISE */
+		.needs_file		= 1,
+	},
 };
 
 static void io_wq_submit_work(struct io_wq_work **workptr);
@@ -2435,6 +2448,35 @@ static int io_openat(struct io_kiocb *req, struct io_kiocb **nxt,
 	return 0;
 }
 
+static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	if (sqe->ioprio || sqe->buf_index || sqe->addr)
+		return -EINVAL;
+
+	req->fadvise.offset = READ_ONCE(sqe->off);
+	req->fadvise.len = READ_ONCE(sqe->len);
+	req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
+	return 0;
+}
+
+static int io_fadvise(struct io_kiocb *req, struct io_kiocb **nxt,
+		      bool force_nonblock)
+{
+	struct io_fadvise *fa = &req->fadvise;
+	int ret;
+
+	/* DONTNEED may block, others _should_ not */
+	if (fa->advice == POSIX_FADV_DONTNEED && force_nonblock)
+		return -EAGAIN;
+
+	ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
+	if (ret < 0)
+		req_set_fail_links(req);
+	io_cqring_add_event(req, ret);
+	io_put_req_find_next(req, nxt);
+	return 0;
+}
+
 static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 {
 	unsigned lookup_flags;
@@ -3724,6 +3766,9 @@ static int io_req_defer_prep(struct io_kiocb *req,
 	case IORING_OP_STATX:
 		ret = io_statx_prep(req, sqe);
 		break;
+	case IORING_OP_FADVISE:
+		ret = io_fadvise_prep(req, sqe);
+		break;
 	default:
 		printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
 				req->opcode);
@@ -3920,6 +3965,14 @@ static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
 		}
 		ret = io_statx(req, nxt, force_nonblock);
 		break;
+	case IORING_OP_FADVISE:
+		if (sqe) {
+			ret = io_fadvise_prep(req, sqe);
+			if (ret)
+				break;
+		}
+		ret = io_fadvise(req, nxt, force_nonblock);
+		break;
 	default:
 		ret = -EINVAL;
 		break;
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 80f892628e66..f87d8fb42916 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -36,6 +36,7 @@ struct io_uring_sqe {
 		__u32		cancel_flags;
 		__u32		open_flags;
 		__u32		statx_flags;
+		__u32		fadvise_advice;
 	};
 	__u64	user_data;	/* data to be passed back at completion time */
 	union {
@@ -86,6 +87,7 @@ enum {
 	IORING_OP_STATX,
 	IORING_OP_READ,
 	IORING_OP_WRITE,
+	IORING_OP_FADVISE,
 
 	/* this goes last, obviously */
 	IORING_OP_LAST,
-- 
2.24.1

[PATCH 2/3] mm: make do_madvise() available internally

[Jens Axboe]

This is in preparation for enabling this functionality through io_uring.
Add a helper that is just exporting what sys_madvise() does, and have the
system call use it.

No functional changes in this patch.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/linux/mm.h | 1 +
 mm/madvise.c       | 7 ++++++-
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 80a9162b406c..766cad8aaa60 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -2328,6 +2328,7 @@ extern int __do_munmap(struct mm_struct *, unsigned long, size_t,
 		       struct list_head *uf, bool downgrade);
 extern int do_munmap(struct mm_struct *, unsigned long, size_t,
 		     struct list_head *uf);
+extern int do_madvise(unsigned long start, size_t len_in, int behavior);
 
 static inline unsigned long
 do_mmap_pgoff(struct file *file, unsigned long addr,
diff --git a/mm/madvise.c b/mm/madvise.c
index bcdb6a042787..43b47d3fae02 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -1044,7 +1044,7 @@ madvise_behavior_valid(int behavior)
  *  -EBADF  - map exists, but area maps something that isn't a file.
  *  -EAGAIN - a kernel resource was temporarily unavailable.
  */
-SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
+int do_madvise(unsigned long start, size_t len_in, int behavior)
 {
 	unsigned long end, tmp;
 	struct vm_area_struct *vma, *prev;
@@ -1141,3 +1141,8 @@ SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
 
 	return error;
 }
+
+SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
+{
+	return do_madvise(start, len_in, behavior);
+}
-- 
2.24.1

[PATCH 3/3] io_uring: add IORING_OP_MADVISE

[Jens Axboe]

This adds support for doing madvise(2) through io_uring. We assume that
any operation can block, and hence punt everything async. This could be
improved, but hard to make bullet proof. The async punt ensures it's
safe.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c                 | 56 ++++++++++++++++++++++++++++++++++-
 include/uapi/linux/io_uring.h |  1 +
 2 files changed, 56 insertions(+), 1 deletion(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 0b200a7d4ae0..378f97cc2bf2 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -403,7 +403,10 @@ struct io_files_update {
 
 struct io_fadvise {
 	struct file			*file;
-	u64				offset;
+	union {
+		u64			offset;
+		u64			addr;
+	};
 	u32				len;
 	u32				advice;
 };
@@ -682,6 +685,10 @@ static const struct io_op_def io_op_defs[] = {
 		/* IORING_OP_FADVISE */
 		.needs_file		= 1,
 	},
+	{
+		/* IORING_OP_MADVISE */
+		.needs_mm		= 1,
+	},
 };
 
 static void io_wq_submit_work(struct io_wq_work **workptr);
@@ -2448,6 +2455,42 @@ static int io_openat(struct io_kiocb *req, struct io_kiocb **nxt,
 	return 0;
 }
 
+static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
+	if (sqe->ioprio || sqe->buf_index || sqe->off)
+		return -EINVAL;
+
+	req->fadvise.addr = READ_ONCE(sqe->addr);
+	req->fadvise.len = READ_ONCE(sqe->len);
+	req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
+	return 0;
+#else
+	return -EOPNOTSUPP;
+#endif
+}
+
+static int io_madvise(struct io_kiocb *req, struct io_kiocb **nxt,
+		      bool force_nonblock)
+{
+#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
+	struct io_fadvise *fa = &req->fadvise;
+	int ret;
+
+	if (force_nonblock)
+		return -EAGAIN;
+
+	ret = do_madvise(fa->addr, fa->len, fa->advice);
+	if (ret < 0)
+		req_set_fail_links(req);
+	io_cqring_add_event(req, ret);
+	io_put_req_find_next(req, nxt);
+	return 0;
+#else
+	return -EOPNOTSUPP;
+#endif
+}
+
 static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 {
 	if (sqe->ioprio || sqe->buf_index || sqe->addr)
@@ -3769,6 +3812,9 @@ static int io_req_defer_prep(struct io_kiocb *req,
 	case IORING_OP_FADVISE:
 		ret = io_fadvise_prep(req, sqe);
 		break;
+	case IORING_OP_MADVISE:
+		ret = io_madvise_prep(req, sqe);
+		break;
 	default:
 		printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
 				req->opcode);
@@ -3973,6 +4019,14 @@ static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
 		}
 		ret = io_fadvise(req, nxt, force_nonblock);
 		break;
+	case IORING_OP_MADVISE:
+		if (sqe) {
+			ret = io_madvise_prep(req, sqe);
+			if (ret)
+				break;
+		}
+		ret = io_madvise(req, nxt, force_nonblock);
+		break;
 	default:
 		ret = -EINVAL;
 		break;
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index f87d8fb42916..7cb6fe0fccd7 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -88,6 +88,7 @@ enum {
 	IORING_OP_READ,
 	IORING_OP_WRITE,
 	IORING_OP_FADVISE,
+	IORING_OP_MADVISE,
 
 	/* this goes last, obviously */
 	IORING_OP_LAST,
-- 
2.24.1

Re: [PATCH 3/3] io_uring: add IORING_OP_MADVISE

[Kirill A. Shutemov]

On Fri, Jan 10, 2020 at 08:47:39AM -0700, Jens Axboe wrote:
> This adds support for doing madvise(2) through io_uring. We assume that
> any operation can block, and hence punt everything async. This could be
> improved, but hard to make bullet proof. The async punt ensures it's
> safe.
> 
> Signed-off-by: Jens Axboe <axboe@kernel.dk>

How capability checks work with io_uring?

MADV_HWPOISON requires CAP_SYS_ADMIN and I just want to make sure it will
not open a way around.

-- 
 Kirill A. Shutemov

Re: [PATCH 3/3] io_uring: add IORING_OP_MADVISE

[Jens Axboe]

On 1/11/20 4:10 PM, Kirill A. Shutemov wrote:
> On Fri, Jan 10, 2020 at 08:47:39AM -0700, Jens Axboe wrote:
>> This adds support for doing madvise(2) through io_uring. We assume that
>> any operation can block, and hence punt everything async. This could be
>> improved, but hard to make bullet proof. The async punt ensures it's
>> safe.
>>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> 
> How capability checks work with io_uring?
> 
> MADV_HWPOISON requires CAP_SYS_ADMIN and I just want to make sure it will
> not open a way around.

There are two ways the request can get invoked from io_uring:

1) Inline from the system call, personality is the application (of course)
   in that case.

2) Async helper, personality (creds, mm, etc) are inherited from the ring.

So it should be totally safe, and madvise is no different than the other
system calls supported in that regard.

-- 
Jens Axboe

Re: [PATCH 3/3] io_uring: add IORING_OP_MADVISE

[Pavel Begunkov]

[-- Attachment #1.1: Type: text/plain, Size: 3840 bytes --]

On 10/01/2020 18:47, Jens Axboe wrote:
> This adds support for doing madvise(2) through io_uring. We assume that
> any operation can block, and hence punt everything async. This could be
> improved, but hard to make bullet proof. The async punt ensures it's
> safe.
> 
I don't like that it share structs/fields names with fadvise. E.g. madvise's
context is called struct io_fadvise. Could it at least have fadvise_advice filed
in struct io_uring_sqe? io_uring parts of the patchset look good.

Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>

> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>  fs/io_uring.c                 | 56 ++++++++++++++++++++++++++++++++++-
>  include/uapi/linux/io_uring.h |  1 +
>  2 files changed, 56 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index 0b200a7d4ae0..378f97cc2bf2 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -403,7 +403,10 @@ struct io_files_update {
>  
>  struct io_fadvise {
>  	struct file			*file;
> -	u64				offset;
> +	union {
> +		u64			offset;
> +		u64			addr;
> +	};
>  	u32				len;
>  	u32				advice;
>  };
> @@ -682,6 +685,10 @@ static const struct io_op_def io_op_defs[] = {
>  		/* IORING_OP_FADVISE */
>  		.needs_file		= 1,
>  	},
> +	{
> +		/* IORING_OP_MADVISE */
> +		.needs_mm		= 1,
> +	},
>  };
>  
>  static void io_wq_submit_work(struct io_wq_work **workptr);
> @@ -2448,6 +2455,42 @@ static int io_openat(struct io_kiocb *req, struct io_kiocb **nxt,
>  	return 0;
>  }
>  
> +static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
> +{
> +#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
> +	if (sqe->ioprio || sqe->buf_index || sqe->off)
> +		return -EINVAL;
> +
> +	req->fadvise.addr = READ_ONCE(sqe->addr);
> +	req->fadvise.len = READ_ONCE(sqe->len);
> +	req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
> +	return 0;
> +#else
> +	return -EOPNOTSUPP;
> +#endif
> +}
> +
> +static int io_madvise(struct io_kiocb *req, struct io_kiocb **nxt,
> +		      bool force_nonblock)
> +{
> +#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
> +	struct io_fadvise *fa = &req->fadvise;
> +	int ret;
> +
> +	if (force_nonblock)
> +		return -EAGAIN;
> +
> +	ret = do_madvise(fa->addr, fa->len, fa->advice);
> +	if (ret < 0)
> +		req_set_fail_links(req);
> +	io_cqring_add_event(req, ret);
> +	io_put_req_find_next(req, nxt);
> +	return 0;
> +#else
> +	return -EOPNOTSUPP;
> +#endif
> +}
> +
>  static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
>  {
>  	if (sqe->ioprio || sqe->buf_index || sqe->addr)
> @@ -3769,6 +3812,9 @@ static int io_req_defer_prep(struct io_kiocb *req,
>  	case IORING_OP_FADVISE:
>  		ret = io_fadvise_prep(req, sqe);
>  		break;
> +	case IORING_OP_MADVISE:
> +		ret = io_madvise_prep(req, sqe);
> +		break;
>  	default:
>  		printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
>  				req->opcode);
> @@ -3973,6 +4019,14 @@ static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
>  		}
>  		ret = io_fadvise(req, nxt, force_nonblock);
>  		break;
> +	case IORING_OP_MADVISE:
> +		if (sqe) {
> +			ret = io_madvise_prep(req, sqe);
> +			if (ret)
> +				break;
> +		}
> +		ret = io_madvise(req, nxt, force_nonblock);
> +		break;
>  	default:
>  		ret = -EINVAL;
>  		break;
> diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
> index f87d8fb42916..7cb6fe0fccd7 100644
> --- a/include/uapi/linux/io_uring.h
> +++ b/include/uapi/linux/io_uring.h
> @@ -88,6 +88,7 @@ enum {
>  	IORING_OP_READ,
>  	IORING_OP_WRITE,
>  	IORING_OP_FADVISE,
> +	IORING_OP_MADVISE,
>  
>  	/* this goes last, obviously */
>  	IORING_OP_LAST,
> 

-- 
Pavel Begunkov


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH 3/3] io_uring: add IORING_OP_MADVISE

[Jens Axboe]

On 1/12/20 2:39 PM, Pavel Begunkov wrote:
> On 10/01/2020 18:47, Jens Axboe wrote:
>> This adds support for doing madvise(2) through io_uring. We assume that
>> any operation can block, and hence punt everything async. This could be
>> improved, but hard to make bullet proof. The async punt ensures it's
>> safe.
>>
> I don't like that it share structs/fields names with fadvise. E.g. madvise's
> context is called struct io_fadvise. Could it at least have fadvise_advice filed
> in struct io_uring_sqe? io_uring parts of the patchset look good.
> 
> Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>

Thanks, I can add the separate union, not a big deal.

-- 
Jens Axboe