[PATCH] Add support for the sys_vmsplice syscall

sys_splice() moves data to/from pipes with a file input/output. sys_vmsplice()
moves data to a pipe, with the input being a user address range instead.

This uses an approach suggested by Linus, where we can hold partial ranges
inside the pages[] map. Hopefully this will be useful for network
receive support as well.

Signed-off-by: Jens Axboe <[e-mail]>

arch/ia64/kernel/entry.S

···
 	data8 sys_get_robust_list
 	data8 sys_sync_file_range		// 1300
 	data8 sys_tee
+	data8 sys_vmsplice
 
 	.org sys_call_table + 8*NR_syscalls	// guard against failures to increase NR_syscalls

arch/powerpc/kernel/systbl.S

···
 SYSCALL(unshare)
 SYSCALL(splice)
 SYSCALL(tee)
+SYSCALL(vmsplice)
 
 /*
  * please add new calls to arch/powerpc/platforms/cell/spu_callbacks.c

arch/powerpc/platforms/cell/spu_callbacks.c

···
 	[__NR_unshare]			sys_unshare,
 	[__NR_splice]			sys_splice,
 	[__NR_tee]			sys_tee,
+	[__NR_vmsplice]			sys_vmsplice,
 };
 
 long spu_sys_callback(struct spu_syscall_block *s)

fs/splice.c

···
 #include <linux/buffer_head.h>
 #include <linux/module.h>
 #include <linux/syscalls.h>
+#include <linux/uio.h>
 
 /*
  * Passed to the actors
···
 	unsigned int flags;		/* splice flags */
 	struct file *file;		/* file to read/write */
 	loff_t pos;			/* file position */
+};
+
+struct partial_page {
+	unsigned int offset;
+	unsigned int len;
+};
+
+/*
+ * Passed to move_to_pipe
+ */
+struct splice_pipe_desc {
+	struct page **pages;		/* page map */
+	struct partial_page *partial;	/* pages[] may not be contig */
+	int nr_pages;			/* number of pages in map */
+	unsigned int flags;		/* splice flags */
+	struct pipe_buf_operations *ops;/* ops associated with output pipe */
 };
 
 /*
···
 	kunmap(buf->page);
 }
 
+static void *user_page_pipe_buf_map(struct file *file,
+				    struct pipe_inode_info *pipe,
+				    struct pipe_buffer *buf)
+{
+	return kmap(buf->page);
+}
+
+static void user_page_pipe_buf_unmap(struct pipe_inode_info *pipe,
+				     struct pipe_buffer *buf)
+{
+	kunmap(buf->page);
+}
+
 static void page_cache_pipe_buf_get(struct pipe_inode_info *info,
 				    struct pipe_buffer *buf)
 {
···
 	.get = page_cache_pipe_buf_get,
 };
 
+static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
+				    struct pipe_buffer *buf)
+{
+	return 1;
+}
+
+static struct pipe_buf_operations user_page_pipe_buf_ops = {
+	.can_merge = 0,
+	.map = user_page_pipe_buf_map,
+	.unmap = user_page_pipe_buf_unmap,
+	.release = page_cache_pipe_buf_release,
+	.steal = user_page_pipe_buf_steal,
+	.get = page_cache_pipe_buf_get,
+};
+
 /*
  * Pipe output worker. This sets up our pipe format with the page cache
  * pipe buffer operations. Otherwise very similar to the regular pipe_writev().
  */
-static ssize_t move_to_pipe(struct pipe_inode_info *pipe, struct page **pages,
-			    int nr_pages, unsigned long len,
-			    unsigned int offset, unsigned int flags)
+static ssize_t move_to_pipe(struct pipe_inode_info *pipe,
+			    struct splice_pipe_desc *spd)
 {
-	int ret, do_wakeup, i;
+	int ret, do_wakeup, page_nr;
 
 	ret = 0;
 	do_wakeup = 0;
-	i = 0;
+	page_nr = 0;
 
 	if (pipe->inode)
 		mutex_lock(&pipe->inode->i_mutex);
···
 		if (pipe->nrbufs < PIPE_BUFFERS) {
 			int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1);
 			struct pipe_buffer *buf = pipe->bufs + newbuf;
-			struct page *page = pages[i++];
-			unsigned long this_len;
 
-			this_len = PAGE_CACHE_SIZE - offset;
-			if (this_len > len)
-				this_len = len;
-
-			buf->page = page;
-			buf->offset = offset;
-			buf->len = this_len;
-			buf->ops = &page_cache_pipe_buf_ops;
+			buf->page = spd->pages[page_nr];
+			buf->offset = spd->partial[page_nr].offset;
+			buf->len = spd->partial[page_nr].len;
+			buf->ops = spd->ops;
 			pipe->nrbufs++;
+			page_nr++;
+			ret += buf->len;
+
 			if (pipe->inode)
 				do_wakeup = 1;
 
-			ret += this_len;
-			len -= this_len;
-			offset = 0;
-			if (!--nr_pages)
-				break;
-			if (!len)
+			if (!--spd->nr_pages)
 				break;
 			if (pipe->nrbufs < PIPE_BUFFERS)
 				continue;
···
 			break;
 		}
 
-		if (flags & SPLICE_F_NONBLOCK) {
+		if (spd->flags & SPLICE_F_NONBLOCK) {
 			if (!ret)
 				ret = -EAGAIN;
 			break;
···
 		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 	}
 
-	while (i < nr_pages)
-		page_cache_release(pages[i++]);
+	while (page_nr < spd->nr_pages)
+		page_cache_release(spd->pages[page_nr++]);
 
 	return ret;
 }
···
 			   unsigned int flags)
 {
 	struct address_space *mapping = in->f_mapping;
-	unsigned int loff, offset, nr_pages;
+	unsigned int loff, nr_pages;
 	struct page *pages[PIPE_BUFFERS];
+	struct partial_page partial[PIPE_BUFFERS];
 	struct page *page;
 	pgoff_t index, end_index;
 	loff_t isize;
-	size_t bytes;
-	int i, error;
+	size_t total_len;
+	int error;
+	struct splice_pipe_desc spd = {
+		.pages = pages,
+		.partial = partial,
+		.flags = flags,
+		.ops = &page_cache_pipe_buf_ops,
+	};
 
 	index = *ppos >> PAGE_CACHE_SHIFT;
-	loff = offset = *ppos & ~PAGE_CACHE_MASK;
-	nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
+	loff = *ppos & ~PAGE_CACHE_MASK;
+	nr_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 
 	if (nr_pages > PIPE_BUFFERS)
 		nr_pages = PIPE_BUFFERS;
···
 	 * read-ahead if this is a non-zero offset (we are likely doing small
 	 * chunk splice and the page is already there) for a single page.
 	 */
-	if (!offset || nr_pages > 1)
-		do_page_cache_readahead(mapping, in, index, nr_pages);
+	if (!loff || spd.nr_pages > 1)
+		do_page_cache_readahead(mapping, in, index, spd.nr_pages);
 
 	/*
 	 * Now fill in the holes:
 	 */
 	error = 0;
-	bytes = 0;
-	for (i = 0; i < nr_pages; i++, index++) {
+	total_len = 0;
+	for (spd.nr_pages = 0; spd.nr_pages < nr_pages; spd.nr_pages++, index++) {
 		unsigned int this_len;
 
 		if (!len)
···
 			 */
 			if (end_index == index) {
 				loff = PAGE_CACHE_SIZE - (isize & ~PAGE_CACHE_MASK);
-				if (bytes + loff > isize) {
+				if (total_len + loff > isize) {
 					page_cache_release(page);
 					break;
 				}
 				/*
 				 * force quit after adding this page
 				 */
-				nr_pages = i;
+				nr_pages = spd.nr_pages;
 				this_len = min(this_len, loff);
+				loff = 0;
 			}
 		}
 fill_it:
-		pages[i] = page;
-		bytes += this_len;
+		pages[spd.nr_pages] = page;
+		partial[spd.nr_pages].offset = loff;
+		partial[spd.nr_pages].len = this_len;
 		len -= this_len;
+		total_len += this_len;
 		loff = 0;
 	}
 
-	if (i)
-		return move_to_pipe(pipe, pages, i, bytes, offset, flags);
+	if (spd.nr_pages)
+		return move_to_pipe(pipe, &spd);
 
 	return error;
 }
···
 	}
 
 	return -EINVAL;
+}
+
+/*
+ * Map an iov into an array of pages and offset/length tupples. With the
+ * partial_page structure, we can map several non-contiguous ranges into
+ * our ones pages[] map instead of splitting that operation into pieces.
+ * Could easily be exported as a generic helper for other users, in which
+ * case one would probably want to add a 'max_nr_pages' parameter as well.
+ */
+static int get_iovec_page_array(const struct iovec __user *iov,
+				unsigned int nr_vecs, struct page **pages,
+				struct partial_page *partial)
+{
+	int buffers = 0, error = 0;
+
+	/*
+	 * It's ok to take the mmap_sem for reading, even
+	 * across a "get_user()".
+	 */
+	down_read(&current->mm->mmap_sem);
+
+	while (nr_vecs) {
+		unsigned long off, npages;
+		void __user *base;
+		size_t len;
+		int i;
+
+		/*
+		 * Get user address base and length for this iovec.
+		 */
+		error = get_user(base, &iov->iov_base);
+		if (unlikely(error))
+			break;
+		error = get_user(len, &iov->iov_len);
+		if (unlikely(error))
+			break;
+
+		/*
+		 * Sanity check this iovec. 0 read succeeds.
+		 */
+		if (unlikely(!len))
+			break;
+		error = -EFAULT;
+		if (unlikely(!base))
+			break;
+
+		/*
+		 * Get this base offset and number of pages, then map
+		 * in the user pages.
+		 */
+		off = (unsigned long) base & ~PAGE_MASK;
+		npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
+		if (npages > PIPE_BUFFERS - buffers)
+			npages = PIPE_BUFFERS - buffers;
+
+		error = get_user_pages(current, current->mm,
+				       (unsigned long) base, npages, 0, 0,
+				       &pages[buffers], NULL);
+
+		if (unlikely(error <= 0))
+			break;
+
+		/*
+		 * Fill this contiguous range into the partial page map.
+		 */
+		for (i = 0; i < error; i++) {
+			const int plen = min_t(size_t, len, PAGE_SIZE) - off;
+
+			partial[buffers].offset = off;
+			partial[buffers].len = plen;
+
+			off = 0;
+			len -= plen;
+			buffers++;
+		}
+
+		/*
+		 * We didn't complete this iov, stop here since it probably
+		 * means we have to move some of this into a pipe to
+		 * be able to continue.
+		 */
+		if (len)
+			break;
+
+		/*
+		 * Don't continue if we mapped fewer pages than we asked for,
+		 * or if we mapped the max number of pages that we have
+		 * room for.
+		 */
+		if (error < npages || buffers == PIPE_BUFFERS)
+			break;
+
+		nr_vecs--;
+		iov++;
+	}
+
+	up_read(&current->mm->mmap_sem);
+
+	if (buffers)
+		return buffers;
+
+	return error;
+}
+
+/*
+ * vmsplice splices a user address range into a pipe. It can be thought of
+ * as splice-from-memory, where the regular splice is splice-from-file (or
+ * to file). In both cases the output is a pipe, naturally.
+ *
+ * Note that vmsplice only supports splicing _from_ user memory to a pipe,
+ * not the other way around. Splicing from user memory is a simple operation
+ * that can be supported without any funky alignment restrictions or nasty
+ * vm tricks. We simply map in the user memory and fill them into a pipe.
+ * The reverse isn't quite as easy, though. There are two possible solutions
+ * for that:
+ *
+ *	- memcpy() the data internally, at which point we might as well just
+ *	  do a regular read() on the buffer anyway.
+ *	- Lots of nasty vm tricks, that are neither fast nor flexible (it
+ *	  has restriction limitations on both ends of the pipe).
+ *
+ * Alas, it isn't here.
+ *
+ */
+static long do_vmsplice(struct file *file, const struct iovec __user *iov,
+			unsigned long nr_segs, unsigned int flags)
+{
+	struct pipe_inode_info *pipe = file->f_dentry->d_inode->i_pipe;
+	struct page *pages[PIPE_BUFFERS];
+	struct partial_page partial[PIPE_BUFFERS];
+	struct splice_pipe_desc spd = {
+		.pages = pages,
+		.partial = partial,
+		.flags = flags,
+		.ops = &user_page_pipe_buf_ops,
+	};
+
+	if (unlikely(!pipe))
+		return -EBADF;
+	if (unlikely(nr_segs > UIO_MAXIOV))
+		return -EINVAL;
+	else if (unlikely(!nr_segs))
+		return 0;
+
+	spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial);
+	if (spd.nr_pages <= 0)
+		return spd.nr_pages;
+
+	return move_to_pipe(pipe, &spd);
+}
+
+asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov,
+			     unsigned long nr_segs, unsigned int flags)
+{
+	struct file *file;
+	long error;
+	int fput;
+
+	error = -EBADF;
+	file = fget_light(fd, &fput);
+	if (file) {
+		if (file->f_mode & FMODE_WRITE)
+			error = do_vmsplice(file, iov, nr_segs, flags);
+
+		fput_light(file, fput);
+	}
+
+	return error;
 }
 
 asmlinkage long sys_splice(int fd_in, loff_t __user *off_in,

include/asm-i386/unistd.h

···
 #define __NR_splice		313
 #define __NR_sync_file_range	314
 #define __NR_tee		315
+#define __NR_vmsplice		316
 
-#define NR_syscalls 316
+#define NR_syscalls 317
 
 /*
  * user-visible error numbers are in the range -1 - -128: see

include/asm-ia64/unistd.h

···
 #define __NR_get_robust_list		1299
 #define __NR_sync_file_range		1300
 #define __NR_tee			1301
+#define __NR_vmsplice			1302
 
 #ifdef __KERNEL__
 
 #include <linux/config.h>
 
-#define NR_syscalls			278 /* length of syscall table */
+#define NR_syscalls			279 /* length of syscall table */
 
 #define __ARCH_WANT_SYS_RT_SIGACTION
 

include/asm-powerpc/unistd.h

···
 #define __NR_unshare		282
 #define __NR_splice		283
 #define __NR_tee		284
+#define __NR_vmsplice		285
 
-#define __NR_syscalls		285
+#define __NR_syscalls		286
 
 #ifdef __KERNEL__
 #define __NR__exit __NR_exit

include/asm-x86_64/unistd.h

···
 __SYSCALL(__NR_tee, sys_tee)
 #define __NR_sync_file_range	277
 __SYSCALL(__NR_sync_file_range, sys_sync_file_range)
+#define __NR_vmsplice		278
+__SYSCALL(__NR_vmsplice, sys_vmsplice)
 
-#define __NR_syscall_max __NR_sync_file_range
+#define __NR_syscall_max __NR_vmsplice
 
 #ifndef __NO_STUBS
 

include/linux/syscalls.h

···
 			   int fd_out, loff_t __user *off_out,
 			   size_t len, unsigned int flags);
 
+asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov,
+			     unsigned long nr_segs, unsigned int flags);
+
 asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags);
 
 asmlinkage long sys_sync_file_range(int fd, loff_t offset, loff_t nbytes,