Merge tag 'for-6.6/io_uring-2023-08-28' of git://git.kernel.dk/linux

Pull io_uring updates from Jens Axboe:
 "Fairly quiet round in terms of features, mostly just improvements all
  over the map for existing code. In detail:

   - Initial support for socket operations through io_uring. Latter half
     of this will likely land with the 6.7 kernel, then allowing things
     like get/setsockopt (Breno)

   - Cleanup of the cancel code, and then adding support for canceling
     requests with the opcode as the key (me)

   - Improvements for the io-wq locking (me)

   - Fix affinity setting for SQPOLL based io-wq (me)

   - Remove the io_uring userspace code. These were added initially as
     copies from liburing, but all of them have since bitrotted and are
     way out of date at this point. Rather than attempt to keep them in
     sync, just get rid of them. People will have liburing available
     anyway for these examples. (Pavel)

   - Series improving the CQ/SQ ring caching (Pavel)

   - Misc fixes and cleanups (Pavel, Yue, me)"

* tag 'for-6.6/io_uring-2023-08-28' of git://git.kernel.dk/linux: (47 commits)
  io_uring: move iopoll ctx fields around
  io_uring: move multishot cqe cache in ctx
  io_uring: separate task_work/waiting cache line
  io_uring: banish non-hot data to end of io_ring_ctx
  io_uring: move non aligned field to the end
  io_uring: add option to remove SQ indirection
  io_uring: compact SQ/CQ heads/tails
  io_uring: force inline io_fill_cqe_req
  io_uring: merge iopoll and normal completion paths
  io_uring: reorder cqring_flush and wakeups
  io_uring: optimise extra io_get_cqe null check
  io_uring: refactor __io_get_cqe()
  io_uring: simplify big_cqe handling
  io_uring: cqe init hardening
  io_uring: improve cqe !tracing hot path
  io_uring/rsrc: Annotate struct io_mapped_ubuf with __counted_by
  io_uring/sqpoll: fix io-wq affinity when IORING_SETUP_SQPOLL is used
  io_uring: simplify io_run_task_work_sig return
  io_uring/rsrc: keep one global dummy_ubuf
  io_uring: never overflow io_aux_cqe
  ...

Author: torvalds

MAINTAINERS

@@ -10966,7 +10966,6 @@ F:	include/linux/io_uring_types.h
 F:	include/trace/events/io_uring.h
 F:	include/uapi/linux/io_uring.h
 F:	io_uring/
-F:	tools/io_uring/
 
 IPMI SUBSYSTEM
 M:	Corey Minyard <minyard@acm.org>

include/linux/io_uring.h

@@ -81,6 +81,7 @@ static inline void io_uring_free(struct task_struct *tsk)
 	if (tsk->io_uring)
 		__io_uring_free(tsk);
 }
+int io_uring_cmd_sock(struct io_uring_cmd *cmd, unsigned int issue_flags);
 #else
 static inline int io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw,
 			      struct iov_iter *iter, void *ioucmd)
@@ -116,6 +117,11 @@ static inline const char *io_uring_get_opcode(u8 opcode)
 {
 	return "";
 }
+static inline int io_uring_cmd_sock(struct io_uring_cmd *cmd,
+				    unsigned int issue_flags)
+{
+	return -EOPNOTSUPP;
+}
 #endif
 
 #endif

include/linux/io_uring_types.h

@@ -69,8 +69,8 @@ struct io_uring_task {
 };
 
 struct io_uring {
-	u32 head ____cacheline_aligned_in_smp;
-	u32 tail ____cacheline_aligned_in_smp;
+	u32 head;
+	u32 tail;
 };
 
 /*
@@ -176,7 +176,6 @@ struct io_submit_state {
 	unsigned short		submit_nr;
 	unsigned int		cqes_count;
 	struct blk_plug		plug;
-	struct io_uring_cqe	cqes[16];
 };
 
 struct io_ev_fd {
@@ -205,25 +204,17 @@ struct io_ring_ctx {
 		unsigned int		has_evfd: 1;
 		/* all CQEs should be posted only by the submitter task */
 		unsigned int		task_complete: 1;
+		unsigned int		lockless_cq: 1;
 		unsigned int		syscall_iopoll: 1;
 		unsigned int		poll_activated: 1;
 		unsigned int		drain_disabled: 1;
 		unsigned int		compat: 1;
 
-		enum task_work_notify_mode	notify_method;
+		struct task_struct	*submitter_task;
+		struct io_rings		*rings;
+		struct percpu_ref	refs;
 
-		/*
-		 * If IORING_SETUP_NO_MMAP is used, then the below holds
-		 * the gup'ed pages for the two rings, and the sqes.
-		 */
-		unsigned short		n_ring_pages;
-		unsigned short		n_sqe_pages;
-		struct page		**ring_pages;
-		struct page		**sqe_pages;
-
-		struct io_rings			*rings;
-		struct task_struct		*submitter_task;
-		struct percpu_ref		refs;
+		enum task_work_notify_mode	notify_method;
 	} ____cacheline_aligned_in_smp;
 
 	/* submission data */
@@ -261,31 +252,20 @@ struct io_ring_ctx {
 
 		struct io_buffer_list	*io_bl;
 		struct xarray		io_bl_xa;
-		struct list_head	io_buffers_cache;
 
 		struct io_hash_table	cancel_table_locked;
-		struct list_head	cq_overflow_list;
 		struct io_alloc_cache	apoll_cache;
 		struct io_alloc_cache	netmsg_cache;
-	} ____cacheline_aligned_in_smp;
-
-	/* IRQ completion list, under ->completion_lock */
-	struct io_wq_work_list	locked_free_list;
-	unsigned int		locked_free_nr;
-
-	const struct cred	*sq_creds;	/* cred used for __io_sq_thread() */
-	struct io_sq_data	*sq_data;	/* if using sq thread polling */
-
-	struct wait_queue_head	sqo_sq_wait;
-	struct list_head	sqd_list;
 
-	unsigned long		check_cq;
-
-	unsigned int		file_alloc_start;
-	unsigned int		file_alloc_end;
-
-	struct xarray		personalities;
-	u32			pers_next;
+		/*
+		 * ->iopoll_list is protected by the ctx->uring_lock for
+		 * io_uring instances that don't use IORING_SETUP_SQPOLL.
+		 * For SQPOLL, only the single threaded io_sq_thread() will
+		 * manipulate the list, hence no extra locking is needed there.
+		 */
+		struct io_wq_work_list	iopoll_list;
+		bool			poll_multi_queue;
+	} ____cacheline_aligned_in_smp;
 
 	struct {
 		/*
@@ -298,39 +278,55 @@ struct io_ring_ctx {
 		unsigned		cached_cq_tail;
 		unsigned		cq_entries;
 		struct io_ev_fd	__rcu	*io_ev_fd;
-		struct wait_queue_head	cq_wait;
 		unsigned		cq_extra;
 	} ____cacheline_aligned_in_smp;
 
+	/*
+	 * task_work and async notification delivery cacheline. Expected to
+	 * regularly bounce b/w CPUs.
+	 */
 	struct {
-		spinlock_t		completion_lock;
-
-		bool			poll_multi_queue;
-		atomic_t		cq_wait_nr;
-
-		/*
-		 * ->iopoll_list is protected by the ctx->uring_lock for
-		 * io_uring instances that don't use IORING_SETUP_SQPOLL.
-		 * For SQPOLL, only the single threaded io_sq_thread() will
-		 * manipulate the list, hence no extra locking is needed there.
-		 */
-		struct io_wq_work_list	iopoll_list;
-		struct io_hash_table	cancel_table;
-
 		struct llist_head	work_llist;
-
-		struct list_head	io_buffers_comp;
+		unsigned long		check_cq;
+		atomic_t		cq_wait_nr;
+		atomic_t		cq_timeouts;
+		struct wait_queue_head	cq_wait;
 	} ____cacheline_aligned_in_smp;
 
 	/* timeouts */
 	struct {
 		spinlock_t		timeout_lock;
-		atomic_t		cq_timeouts;
 		struct list_head	timeout_list;
 		struct list_head	ltimeout_list;
 		unsigned		cq_last_tm_flush;
 	} ____cacheline_aligned_in_smp;
 
+	struct io_uring_cqe	completion_cqes[16];
+
+	spinlock_t		completion_lock;
+
+	/* IRQ completion list, under ->completion_lock */
+	struct io_wq_work_list	locked_free_list;
+	unsigned int		locked_free_nr;
+
+	struct list_head	io_buffers_comp;
+	struct list_head	cq_overflow_list;
+	struct io_hash_table	cancel_table;
+
+	const struct cred	*sq_creds;	/* cred used for __io_sq_thread() */
+	struct io_sq_data	*sq_data;	/* if using sq thread polling */
+
+	struct wait_queue_head	sqo_sq_wait;
+	struct list_head	sqd_list;
+
+	unsigned int		file_alloc_start;
+	unsigned int		file_alloc_end;
+
+	struct xarray		personalities;
+	u32			pers_next;
+
+	struct list_head	io_buffers_cache;
+
 	/* Keep this last, we don't need it for the fast path */
 	struct wait_queue_head		poll_wq;
 	struct io_restriction		restrictions;
@@ -374,6 +370,15 @@ struct io_ring_ctx {
 	unsigned			sq_thread_idle;
 	/* protected by ->completion_lock */
 	unsigned			evfd_last_cq_tail;
+
+	/*
+	 * If IORING_SETUP_NO_MMAP is used, then the below holds
+	 * the gup'ed pages for the two rings, and the sqes.
+	 */
+	unsigned short			n_ring_pages;
+	unsigned short			n_sqe_pages;
+	struct page			**ring_pages;
+	struct page			**sqe_pages;
 };
 
 struct io_tw_state {
@@ -409,7 +414,6 @@ enum {
 	REQ_F_SINGLE_POLL_BIT,
 	REQ_F_DOUBLE_POLL_BIT,
 	REQ_F_PARTIAL_IO_BIT,
-	REQ_F_CQE32_INIT_BIT,
 	REQ_F_APOLL_MULTISHOT_BIT,
 	REQ_F_CLEAR_POLLIN_BIT,
 	REQ_F_HASH_LOCKED_BIT,
@@ -479,8 +483,6 @@ enum {
 	REQ_F_PARTIAL_IO	= BIT(REQ_F_PARTIAL_IO_BIT),
 	/* fast poll multishot mode */
 	REQ_F_APOLL_MULTISHOT	= BIT(REQ_F_APOLL_MULTISHOT_BIT),
-	/* ->extra1 and ->extra2 are initialised */
-	REQ_F_CQE32_INIT	= BIT(REQ_F_CQE32_INIT_BIT),
 	/* recvmsg special flag, clear EPOLLIN */
 	REQ_F_CLEAR_POLLIN	= BIT(REQ_F_CLEAR_POLLIN_BIT),
 	/* hashed into ->cancel_hash_locked, protected by ->uring_lock */
@@ -579,13 +581,7 @@ struct io_kiocb {
 	struct io_task_work		io_task_work;
 	unsigned			nr_tw;
 	/* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
-	union {
-		struct hlist_node	hash_node;
-		struct {
-			u64		extra1;
-			u64		extra2;
-		};
-	};
+	struct hlist_node		hash_node;
 	/* internal polling, see IORING_FEAT_FAST_POLL */
 	struct async_poll		*apoll;
 	/* opcode allocated if it needs to store data for async defer */
@@ -595,6 +591,11 @@ struct io_kiocb {
 	/* custom credentials, valid IFF REQ_F_CREDS is set */
 	const struct cred		*creds;
 	struct io_wq_work		work;
+
+	struct {
+		u64			extra1;
+		u64			extra2;
+	} big_cqe;
 };
 
 struct io_overflow_cqe {

include/uapi/linux/io_uring.h

@@ -185,6 +185,11 @@ enum {
  */
 #define IORING_SETUP_REGISTERED_FD_ONLY	(1U << 15)
 
+/*
+ * Removes indirection through the SQ index array.
+ */
+#define IORING_SETUP_NO_SQARRAY		(1U << 16)
+
 enum io_uring_op {
 	IORING_OP_NOP,
 	IORING_OP_READV,
@@ -299,11 +304,15 @@ enum io_uring_op {
  *				request 'user_data'
  * IORING_ASYNC_CANCEL_ANY	Match any request
  * IORING_ASYNC_CANCEL_FD_FIXED	'fd' passed in is a fixed descriptor
+ * IORING_ASYNC_CANCEL_USERDATA	Match on user_data, default for no other key
+ * IORING_ASYNC_CANCEL_OP	Match request based on opcode
  */
 #define IORING_ASYNC_CANCEL_ALL	(1U << 0)
 #define IORING_ASYNC_CANCEL_FD	(1U << 1)
 #define IORING_ASYNC_CANCEL_ANY	(1U << 2)
 #define IORING_ASYNC_CANCEL_FD_FIXED	(1U << 3)
+#define IORING_ASYNC_CANCEL_USERDATA	(1U << 4)
+#define IORING_ASYNC_CANCEL_OP	(1U << 5)
 
 /*
  * send/sendmsg and recv/recvmsg flags (sqe->ioprio)
@@ -697,7 +706,9 @@ struct io_uring_sync_cancel_reg {
 	__s32				fd;
 	__u32				flags;
 	struct __kernel_timespec	timeout;
-	__u64				pad[4];
+	__u8				opcode;
+	__u8				pad[7];
+	__u64				pad2[3];
 };
 
 /*
@@ -717,6 +728,14 @@ struct io_uring_recvmsg_out {
 	__u32 flags;
 };
 
+/*
+ * Argument for IORING_OP_URING_CMD when file is a socket
+ */
+enum {
+	SOCKET_URING_OP_SIOCINQ		= 0,
+	SOCKET_URING_OP_SIOCOUTQ,
+};
+
 #ifdef __cplusplus
 }
 #endif