xref: /openbmc/linux/fs/splice.c (revision a0ce2f0a)
15274f052SJens Axboe /*
25274f052SJens Axboe  * "splice": joining two ropes together by interweaving their strands.
35274f052SJens Axboe  *
45274f052SJens Axboe  * This is the "extended pipe" functionality, where a pipe is used as
55274f052SJens Axboe  * an arbitrary in-memory buffer. Think of a pipe as a small kernel
65274f052SJens Axboe  * buffer that you can use to transfer data from one end to the other.
75274f052SJens Axboe  *
85274f052SJens Axboe  * The traditional unix read/write is extended with a "splice()" operation
95274f052SJens Axboe  * that transfers data buffers to or from a pipe buffer.
105274f052SJens Axboe  *
115274f052SJens Axboe  * Named by Larry McVoy, original implementation from Linus, extended by
12c2058e06SJens Axboe  * Jens to support splicing to files, network, direct splicing, etc and
13c2058e06SJens Axboe  * fixing lots of bugs.
145274f052SJens Axboe  *
150fe23479SJens Axboe  * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk>
16c2058e06SJens Axboe  * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
17c2058e06SJens Axboe  * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
185274f052SJens Axboe  *
195274f052SJens Axboe  */
20be297968SChristoph Hellwig #include <linux/bvec.h>
215274f052SJens Axboe #include <linux/fs.h>
225274f052SJens Axboe #include <linux/file.h>
235274f052SJens Axboe #include <linux/pagemap.h>
24d6b29d7cSJens Axboe #include <linux/splice.h>
2508e552c6SKAMEZAWA Hiroyuki #include <linux/memcontrol.h>
265274f052SJens Axboe #include <linux/mm_inline.h>
275abc97aaSJens Axboe #include <linux/swap.h>
284f6f0bd2SJens Axboe #include <linux/writeback.h>
29630d9c47SPaul Gortmaker #include <linux/export.h>
304f6f0bd2SJens Axboe #include <linux/syscalls.h>
31912d35f8SJens Axboe #include <linux/uio.h>
3229ce2058SJames Morris #include <linux/security.h>
335a0e3ad6STejun Heo #include <linux/gfp.h>
3435f9c09fSEric Dumazet #include <linux/socket.h>
3576b021d0SAl Viro #include <linux/compat.h>
36174cd4b1SIngo Molnar #include <linux/sched/signal.h>
37174cd4b1SIngo Molnar 
3806ae43f3SAl Viro #include "internal.h"
395274f052SJens Axboe 
4083f9135bSJens Axboe /*
4183f9135bSJens Axboe  * Attempt to steal a page from a pipe buffer. This should perhaps go into
4283f9135bSJens Axboe  * a vm helper function, it's already simplified quite a bit by the
4383f9135bSJens Axboe  * addition of remove_mapping(). If success is returned, the caller may
4483f9135bSJens Axboe  * attempt to reuse this page for another destination.
4583f9135bSJens Axboe  */
4676ad4d11SJens Axboe static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
475abc97aaSJens Axboe 				     struct pipe_buffer *buf)
485abc97aaSJens Axboe {
495abc97aaSJens Axboe 	struct page *page = buf->page;
509e94cd4fSJens Axboe 	struct address_space *mapping;
515abc97aaSJens Axboe 
529e0267c2SJens Axboe 	lock_page(page);
539e0267c2SJens Axboe 
549e94cd4fSJens Axboe 	mapping = page_mapping(page);
559e94cd4fSJens Axboe 	if (mapping) {
565abc97aaSJens Axboe 		WARN_ON(!PageUptodate(page));
575abc97aaSJens Axboe 
58ad8d6f0aSJens Axboe 		/*
599e94cd4fSJens Axboe 		 * At least for ext2 with nobh option, we need to wait on
609e94cd4fSJens Axboe 		 * writeback completing on this page, since we'll remove it
619e94cd4fSJens Axboe 		 * from the pagecache.  Otherwise truncate wont wait on the
629e94cd4fSJens Axboe 		 * page, allowing the disk blocks to be reused by someone else
639e94cd4fSJens Axboe 		 * before we actually wrote our data to them. fs corruption
649e94cd4fSJens Axboe 		 * ensues.
65ad8d6f0aSJens Axboe 		 */
66ad8d6f0aSJens Axboe 		wait_on_page_writeback(page);
67ad8d6f0aSJens Axboe 
68266cf658SDavid Howells 		if (page_has_private(page) &&
69266cf658SDavid Howells 		    !try_to_release_page(page, GFP_KERNEL))
70ca39d651SJens Axboe 			goto out_unlock;
714f6f0bd2SJens Axboe 
729e94cd4fSJens Axboe 		/*
739e94cd4fSJens Axboe 		 * If we succeeded in removing the mapping, set LRU flag
749e94cd4fSJens Axboe 		 * and return good.
759e94cd4fSJens Axboe 		 */
769e94cd4fSJens Axboe 		if (remove_mapping(mapping, page)) {
771432873aSJens Axboe 			buf->flags |= PIPE_BUF_FLAG_LRU;
785abc97aaSJens Axboe 			return 0;
795abc97aaSJens Axboe 		}
809e94cd4fSJens Axboe 	}
819e94cd4fSJens Axboe 
829e94cd4fSJens Axboe 	/*
839e94cd4fSJens Axboe 	 * Raced with truncate or failed to remove page from current
849e94cd4fSJens Axboe 	 * address space, unlock and return failure.
859e94cd4fSJens Axboe 	 */
86ca39d651SJens Axboe out_unlock:
879e94cd4fSJens Axboe 	unlock_page(page);
889e94cd4fSJens Axboe 	return 1;
899e94cd4fSJens Axboe }
905abc97aaSJens Axboe 
9176ad4d11SJens Axboe static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
925274f052SJens Axboe 					struct pipe_buffer *buf)
935274f052SJens Axboe {
9409cbfeafSKirill A. Shutemov 	put_page(buf->page);
951432873aSJens Axboe 	buf->flags &= ~PIPE_BUF_FLAG_LRU;
965274f052SJens Axboe }
975274f052SJens Axboe 
980845718dSJens Axboe /*
990845718dSJens Axboe  * Check whether the contents of buf is OK to access. Since the content
1000845718dSJens Axboe  * is a page cache page, IO may be in flight.
1010845718dSJens Axboe  */
102cac36bb0SJens Axboe static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
1035274f052SJens Axboe 				       struct pipe_buffer *buf)
1045274f052SJens Axboe {
1055274f052SJens Axboe 	struct page *page = buf->page;
10649d0b21bSJens Axboe 	int err;
1075274f052SJens Axboe 
1085274f052SJens Axboe 	if (!PageUptodate(page)) {
10949d0b21bSJens Axboe 		lock_page(page);
1105274f052SJens Axboe 
11149d0b21bSJens Axboe 		/*
11249d0b21bSJens Axboe 		 * Page got truncated/unhashed. This will cause a 0-byte
11373d62d83SIngo Molnar 		 * splice, if this is the first page.
11449d0b21bSJens Axboe 		 */
1155274f052SJens Axboe 		if (!page->mapping) {
11649d0b21bSJens Axboe 			err = -ENODATA;
11749d0b21bSJens Axboe 			goto error;
1185274f052SJens Axboe 		}
1195274f052SJens Axboe 
12049d0b21bSJens Axboe 		/*
12173d62d83SIngo Molnar 		 * Uh oh, read-error from disk.
12249d0b21bSJens Axboe 		 */
12349d0b21bSJens Axboe 		if (!PageUptodate(page)) {
12449d0b21bSJens Axboe 			err = -EIO;
12549d0b21bSJens Axboe 			goto error;
12649d0b21bSJens Axboe 		}
12749d0b21bSJens Axboe 
12849d0b21bSJens Axboe 		/*
129f84d7519SJens Axboe 		 * Page is ok afterall, we are done.
13049d0b21bSJens Axboe 		 */
13149d0b21bSJens Axboe 		unlock_page(page);
13249d0b21bSJens Axboe 	}
13349d0b21bSJens Axboe 
134f84d7519SJens Axboe 	return 0;
13549d0b21bSJens Axboe error:
13649d0b21bSJens Axboe 	unlock_page(page);
137f84d7519SJens Axboe 	return err;
13870524490SJens Axboe }
13970524490SJens Axboe 
140708e3508SHugh Dickins const struct pipe_buf_operations page_cache_pipe_buf_ops = {
1415274f052SJens Axboe 	.can_merge = 0,
142cac36bb0SJens Axboe 	.confirm = page_cache_pipe_buf_confirm,
1435274f052SJens Axboe 	.release = page_cache_pipe_buf_release,
1445abc97aaSJens Axboe 	.steal = page_cache_pipe_buf_steal,
145f84d7519SJens Axboe 	.get = generic_pipe_buf_get,
1465274f052SJens Axboe };
1475274f052SJens Axboe 
148912d35f8SJens Axboe static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
149912d35f8SJens Axboe 				    struct pipe_buffer *buf)
150912d35f8SJens Axboe {
1517afa6fd0SJens Axboe 	if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
152912d35f8SJens Axboe 		return 1;
1537afa6fd0SJens Axboe 
1541432873aSJens Axboe 	buf->flags |= PIPE_BUF_FLAG_LRU;
155330ab716SJens Axboe 	return generic_pipe_buf_steal(pipe, buf);
156912d35f8SJens Axboe }
157912d35f8SJens Axboe 
158d4c3cca9SEric Dumazet static const struct pipe_buf_operations user_page_pipe_buf_ops = {
159912d35f8SJens Axboe 	.can_merge = 0,
160cac36bb0SJens Axboe 	.confirm = generic_pipe_buf_confirm,
161912d35f8SJens Axboe 	.release = page_cache_pipe_buf_release,
162912d35f8SJens Axboe 	.steal = user_page_pipe_buf_steal,
163f84d7519SJens Axboe 	.get = generic_pipe_buf_get,
164912d35f8SJens Axboe };
165912d35f8SJens Axboe 
166825cdcb1SNamhyung Kim static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
167825cdcb1SNamhyung Kim {
168825cdcb1SNamhyung Kim 	smp_mb();
169825cdcb1SNamhyung Kim 	if (waitqueue_active(&pipe->wait))
170825cdcb1SNamhyung Kim 		wake_up_interruptible(&pipe->wait);
171825cdcb1SNamhyung Kim 	kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
172825cdcb1SNamhyung Kim }
173825cdcb1SNamhyung Kim 
174932cc6d4SJens Axboe /**
175932cc6d4SJens Axboe  * splice_to_pipe - fill passed data into a pipe
176932cc6d4SJens Axboe  * @pipe:	pipe to fill
177932cc6d4SJens Axboe  * @spd:	data to fill
178932cc6d4SJens Axboe  *
179932cc6d4SJens Axboe  * Description:
18079685b8dSRandy Dunlap  *    @spd contains a map of pages and len/offset tuples, along with
181932cc6d4SJens Axboe  *    the struct pipe_buf_operations associated with these pages. This
182932cc6d4SJens Axboe  *    function will link that data to the pipe.
183932cc6d4SJens Axboe  *
18483f9135bSJens Axboe  */
185d6b29d7cSJens Axboe ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
186912d35f8SJens Axboe 		       struct splice_pipe_desc *spd)
1875274f052SJens Axboe {
18800de00bdSJens Axboe 	unsigned int spd_pages = spd->nr_pages;
1898924feffSAl Viro 	int ret = 0, page_nr = 0;
1905274f052SJens Axboe 
191d6785d91SRabin Vincent 	if (!spd_pages)
192d6785d91SRabin Vincent 		return 0;
193d6785d91SRabin Vincent 
1948924feffSAl Viro 	if (unlikely(!pipe->readers)) {
1955274f052SJens Axboe 		send_sig(SIGPIPE, current, 0);
1965274f052SJens Axboe 		ret = -EPIPE;
1978924feffSAl Viro 		goto out;
1985274f052SJens Axboe 	}
1995274f052SJens Axboe 
2008924feffSAl Viro 	while (pipe->nrbufs < pipe->buffers) {
20135f3d14dSJens Axboe 		int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
2023a326a2cSIngo Molnar 		struct pipe_buffer *buf = pipe->bufs + newbuf;
2035274f052SJens Axboe 
204912d35f8SJens Axboe 		buf->page = spd->pages[page_nr];
205912d35f8SJens Axboe 		buf->offset = spd->partial[page_nr].offset;
206912d35f8SJens Axboe 		buf->len = spd->partial[page_nr].len;
207497f9625SJens Axboe 		buf->private = spd->partial[page_nr].private;
208912d35f8SJens Axboe 		buf->ops = spd->ops;
2095a81e6a1SMiklos Szeredi 		buf->flags = 0;
2107afa6fd0SJens Axboe 
2116f767b04SJens Axboe 		pipe->nrbufs++;
212912d35f8SJens Axboe 		page_nr++;
213912d35f8SJens Axboe 		ret += buf->len;
214912d35f8SJens Axboe 
215912d35f8SJens Axboe 		if (!--spd->nr_pages)
2165274f052SJens Axboe 			break;
2175274f052SJens Axboe 	}
2185274f052SJens Axboe 
21929e35094SLinus Torvalds 	if (!ret)
22029e35094SLinus Torvalds 		ret = -EAGAIN;
22129e35094SLinus Torvalds 
2228924feffSAl Viro out:
22300de00bdSJens Axboe 	while (page_nr < spd_pages)
224bbdfc2f7SJens Axboe 		spd->spd_release(spd, page_nr++);
2255274f052SJens Axboe 
2265274f052SJens Axboe 	return ret;
2275274f052SJens Axboe }
2282b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(splice_to_pipe);
2295274f052SJens Axboe 
23079fddc4eSAl Viro ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
23179fddc4eSAl Viro {
23279fddc4eSAl Viro 	int ret;
23379fddc4eSAl Viro 
23479fddc4eSAl Viro 	if (unlikely(!pipe->readers)) {
23579fddc4eSAl Viro 		send_sig(SIGPIPE, current, 0);
23679fddc4eSAl Viro 		ret = -EPIPE;
23779fddc4eSAl Viro 	} else if (pipe->nrbufs == pipe->buffers) {
23879fddc4eSAl Viro 		ret = -EAGAIN;
23979fddc4eSAl Viro 	} else {
24079fddc4eSAl Viro 		int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
24179fddc4eSAl Viro 		pipe->bufs[newbuf] = *buf;
24279fddc4eSAl Viro 		pipe->nrbufs++;
24379fddc4eSAl Viro 		return buf->len;
24479fddc4eSAl Viro 	}
245a779638cSMiklos Szeredi 	pipe_buf_release(pipe, buf);
24679fddc4eSAl Viro 	return ret;
24779fddc4eSAl Viro }
24879fddc4eSAl Viro EXPORT_SYMBOL(add_to_pipe);
24979fddc4eSAl Viro 
25035f3d14dSJens Axboe /*
25135f3d14dSJens Axboe  * Check if we need to grow the arrays holding pages and partial page
25235f3d14dSJens Axboe  * descriptions.
25335f3d14dSJens Axboe  */
254047fe360SEric Dumazet int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
25535f3d14dSJens Axboe {
2566aa7de05SMark Rutland 	unsigned int buffers = READ_ONCE(pipe->buffers);
257047fe360SEric Dumazet 
258047fe360SEric Dumazet 	spd->nr_pages_max = buffers;
259047fe360SEric Dumazet 	if (buffers <= PIPE_DEF_BUFFERS)
26035f3d14dSJens Axboe 		return 0;
26135f3d14dSJens Axboe 
2626da2ec56SKees Cook 	spd->pages = kmalloc_array(buffers, sizeof(struct page *), GFP_KERNEL);
2636da2ec56SKees Cook 	spd->partial = kmalloc_array(buffers, sizeof(struct partial_page),
2646da2ec56SKees Cook 				     GFP_KERNEL);
26535f3d14dSJens Axboe 
26635f3d14dSJens Axboe 	if (spd->pages && spd->partial)
26735f3d14dSJens Axboe 		return 0;
26835f3d14dSJens Axboe 
26935f3d14dSJens Axboe 	kfree(spd->pages);
27035f3d14dSJens Axboe 	kfree(spd->partial);
27135f3d14dSJens Axboe 	return -ENOMEM;
27235f3d14dSJens Axboe }
27335f3d14dSJens Axboe 
274047fe360SEric Dumazet void splice_shrink_spd(struct splice_pipe_desc *spd)
27535f3d14dSJens Axboe {
276047fe360SEric Dumazet 	if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
27735f3d14dSJens Axboe 		return;
27835f3d14dSJens Axboe 
27935f3d14dSJens Axboe 	kfree(spd->pages);
28035f3d14dSJens Axboe 	kfree(spd->partial);
28135f3d14dSJens Axboe }
28235f3d14dSJens Axboe 
28383f9135bSJens Axboe /**
28483f9135bSJens Axboe  * generic_file_splice_read - splice data from file to a pipe
28583f9135bSJens Axboe  * @in:		file to splice from
286932cc6d4SJens Axboe  * @ppos:	position in @in
28783f9135bSJens Axboe  * @pipe:	pipe to splice to
28883f9135bSJens Axboe  * @len:	number of bytes to splice
28983f9135bSJens Axboe  * @flags:	splice modifier flags
29083f9135bSJens Axboe  *
291932cc6d4SJens Axboe  * Description:
292932cc6d4SJens Axboe  *    Will read pages from given file and fill them into a pipe. Can be
29382c156f8SAl Viro  *    used as long as it has more or less sane ->read_iter().
294932cc6d4SJens Axboe  *
29583f9135bSJens Axboe  */
296cbb7e577SJens Axboe ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
297cbb7e577SJens Axboe 				 struct pipe_inode_info *pipe, size_t len,
298cbb7e577SJens Axboe 				 unsigned int flags)
2995274f052SJens Axboe {
30082c156f8SAl Viro 	struct iov_iter to;
30182c156f8SAl Viro 	struct kiocb kiocb;
30282c156f8SAl Viro 	int idx, ret;
303be64f884SBoaz Harrosh 
304aa563d7bSDavid Howells 	iov_iter_pipe(&to, READ, pipe, len);
30582c156f8SAl Viro 	idx = to.idx;
30682c156f8SAl Viro 	init_sync_kiocb(&kiocb, in);
30782c156f8SAl Viro 	kiocb.ki_pos = *ppos;
308bb7462b6SMiklos Szeredi 	ret = call_read_iter(in, &kiocb, &to);
309723590edSMiklos Szeredi 	if (ret > 0) {
31082c156f8SAl Viro 		*ppos = kiocb.ki_pos;
311723590edSMiklos Szeredi 		file_accessed(in);
31282c156f8SAl Viro 	} else if (ret < 0) {
31382c156f8SAl Viro 		to.idx = idx;
31482c156f8SAl Viro 		to.iov_offset = 0;
31582c156f8SAl Viro 		iov_iter_advance(&to, 0); /* to free what was emitted */
31682c156f8SAl Viro 		/*
31782c156f8SAl Viro 		 * callers of ->splice_read() expect -EAGAIN on
31882c156f8SAl Viro 		 * "can't put anything in there", rather than -EFAULT.
31982c156f8SAl Viro 		 */
32082c156f8SAl Viro 		if (ret == -EFAULT)
32182c156f8SAl Viro 			ret = -EAGAIN;
322723590edSMiklos Szeredi 	}
3235274f052SJens Axboe 
3245274f052SJens Axboe 	return ret;
3255274f052SJens Axboe }
326059a8f37SJens Axboe EXPORT_SYMBOL(generic_file_splice_read);
327059a8f37SJens Axboe 
328241699cdSAl Viro const struct pipe_buf_operations default_pipe_buf_ops = {
3296818173bSMiklos Szeredi 	.can_merge = 0,
3306818173bSMiklos Szeredi 	.confirm = generic_pipe_buf_confirm,
3316818173bSMiklos Szeredi 	.release = generic_pipe_buf_release,
3326818173bSMiklos Szeredi 	.steal = generic_pipe_buf_steal,
3336818173bSMiklos Szeredi 	.get = generic_pipe_buf_get,
3346818173bSMiklos Szeredi };
3356818173bSMiklos Szeredi 
33628a625cbSMiklos Szeredi static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe,
33728a625cbSMiklos Szeredi 				    struct pipe_buffer *buf)
33828a625cbSMiklos Szeredi {
33928a625cbSMiklos Szeredi 	return 1;
34028a625cbSMiklos Szeredi }
34128a625cbSMiklos Szeredi 
34228a625cbSMiklos Szeredi /* Pipe buffer operations for a socket and similar. */
34328a625cbSMiklos Szeredi const struct pipe_buf_operations nosteal_pipe_buf_ops = {
34428a625cbSMiklos Szeredi 	.can_merge = 0,
34528a625cbSMiklos Szeredi 	.confirm = generic_pipe_buf_confirm,
34628a625cbSMiklos Szeredi 	.release = generic_pipe_buf_release,
34728a625cbSMiklos Szeredi 	.steal = generic_pipe_buf_nosteal,
34828a625cbSMiklos Szeredi 	.get = generic_pipe_buf_get,
34928a625cbSMiklos Szeredi };
35028a625cbSMiklos Szeredi EXPORT_SYMBOL(nosteal_pipe_buf_ops);
35128a625cbSMiklos Szeredi 
352523ac9afSAl Viro static ssize_t kernel_readv(struct file *file, const struct kvec *vec,
3536818173bSMiklos Szeredi 			    unsigned long vlen, loff_t offset)
3546818173bSMiklos Szeredi {
3556818173bSMiklos Szeredi 	mm_segment_t old_fs;
3566818173bSMiklos Szeredi 	loff_t pos = offset;
3576818173bSMiklos Szeredi 	ssize_t res;
3586818173bSMiklos Szeredi 
3596818173bSMiklos Szeredi 	old_fs = get_fs();
3606818173bSMiklos Szeredi 	set_fs(get_ds());
3616818173bSMiklos Szeredi 	/* The cast to a user pointer is valid due to the set_fs() */
362793b80efSChristoph Hellwig 	res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos, 0);
3636818173bSMiklos Szeredi 	set_fs(old_fs);
3646818173bSMiklos Szeredi 
3656818173bSMiklos Szeredi 	return res;
3666818173bSMiklos Szeredi }
3676818173bSMiklos Szeredi 
36882c156f8SAl Viro static ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
3696818173bSMiklos Szeredi 				 struct pipe_inode_info *pipe, size_t len,
3706818173bSMiklos Szeredi 				 unsigned int flags)
3716818173bSMiklos Szeredi {
372523ac9afSAl Viro 	struct kvec *vec, __vec[PIPE_DEF_BUFFERS];
373523ac9afSAl Viro 	struct iov_iter to;
374523ac9afSAl Viro 	struct page **pages;
3756818173bSMiklos Szeredi 	unsigned int nr_pages;
37613c0f52bSAl Viro 	size_t offset, base, copied = 0;
3776818173bSMiklos Szeredi 	ssize_t res;
3786818173bSMiklos Szeredi 	int i;
3796818173bSMiklos Szeredi 
380523ac9afSAl Viro 	if (pipe->nrbufs == pipe->buffers)
381523ac9afSAl Viro 		return -EAGAIN;
382523ac9afSAl Viro 
383523ac9afSAl Viro 	/*
384523ac9afSAl Viro 	 * Try to keep page boundaries matching to source pagecache ones -
385523ac9afSAl Viro 	 * it probably won't be much help, but...
386523ac9afSAl Viro 	 */
387523ac9afSAl Viro 	offset = *ppos & ~PAGE_MASK;
388523ac9afSAl Viro 
389aa563d7bSDavid Howells 	iov_iter_pipe(&to, READ, pipe, len + offset);
390523ac9afSAl Viro 
39113c0f52bSAl Viro 	res = iov_iter_get_pages_alloc(&to, &pages, len + offset, &base);
392523ac9afSAl Viro 	if (res <= 0)
39335f3d14dSJens Axboe 		return -ENOMEM;
39435f3d14dSJens Axboe 
39513c0f52bSAl Viro 	nr_pages = DIV_ROUND_UP(res + base, PAGE_SIZE);
396523ac9afSAl Viro 
39735f3d14dSJens Axboe 	vec = __vec;
398523ac9afSAl Viro 	if (nr_pages > PIPE_DEF_BUFFERS) {
3996da2ec56SKees Cook 		vec = kmalloc_array(nr_pages, sizeof(struct kvec), GFP_KERNEL);
400523ac9afSAl Viro 		if (unlikely(!vec)) {
401523ac9afSAl Viro 			res = -ENOMEM;
402523ac9afSAl Viro 			goto out;
403523ac9afSAl Viro 		}
40435f3d14dSJens Axboe 	}
40535f3d14dSJens Axboe 
406523ac9afSAl Viro 	pipe->bufs[to.idx].offset = offset;
407523ac9afSAl Viro 	pipe->bufs[to.idx].len -= offset;
4086818173bSMiklos Szeredi 
409523ac9afSAl Viro 	for (i = 0; i < nr_pages; i++) {
410523ac9afSAl Viro 		size_t this_len = min_t(size_t, len, PAGE_SIZE - offset);
411523ac9afSAl Viro 		vec[i].iov_base = page_address(pages[i]) + offset;
4126818173bSMiklos Szeredi 		vec[i].iov_len = this_len;
4136818173bSMiklos Szeredi 		len -= this_len;
4146818173bSMiklos Szeredi 		offset = 0;
4156818173bSMiklos Szeredi 	}
4166818173bSMiklos Szeredi 
417523ac9afSAl Viro 	res = kernel_readv(in, vec, nr_pages, *ppos);
418523ac9afSAl Viro 	if (res > 0) {
419523ac9afSAl Viro 		copied = res;
4206818173bSMiklos Szeredi 		*ppos += res;
421523ac9afSAl Viro 	}
4226818173bSMiklos Szeredi 
42335f3d14dSJens Axboe 	if (vec != __vec)
42435f3d14dSJens Axboe 		kfree(vec);
425523ac9afSAl Viro out:
426523ac9afSAl Viro 	for (i = 0; i < nr_pages; i++)
427523ac9afSAl Viro 		put_page(pages[i]);
428523ac9afSAl Viro 	kvfree(pages);
429523ac9afSAl Viro 	iov_iter_advance(&to, copied);	/* truncates and discards */
4306818173bSMiklos Szeredi 	return res;
4316818173bSMiklos Szeredi }
4326818173bSMiklos Szeredi 
4335274f052SJens Axboe /*
4344f6f0bd2SJens Axboe  * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
435016b661eSJens Axboe  * using sendpage(). Return the number of bytes sent.
4365274f052SJens Axboe  */
43776ad4d11SJens Axboe static int pipe_to_sendpage(struct pipe_inode_info *pipe,
4385274f052SJens Axboe 			    struct pipe_buffer *buf, struct splice_desc *sd)
4395274f052SJens Axboe {
4406a14b90bSJens Axboe 	struct file *file = sd->u.file;
4415274f052SJens Axboe 	loff_t pos = sd->pos;
442a8adbe37SMichał Mirosław 	int more;
4435274f052SJens Axboe 
44472c2d531SAl Viro 	if (!likely(file->f_op->sendpage))
445a8adbe37SMichał Mirosław 		return -EINVAL;
446a8adbe37SMichał Mirosław 
44735f9c09fSEric Dumazet 	more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
448ae62ca7bSEric Dumazet 
449ae62ca7bSEric Dumazet 	if (sd->len < sd->total_len && pipe->nrbufs > 1)
45035f9c09fSEric Dumazet 		more |= MSG_SENDPAGE_NOTLAST;
451ae62ca7bSEric Dumazet 
452a8adbe37SMichał Mirosław 	return file->f_op->sendpage(file, buf->page, buf->offset,
453f84d7519SJens Axboe 				    sd->len, &pos, more);
4545274f052SJens Axboe }
4555274f052SJens Axboe 
456b3c2d2ddSMiklos Szeredi static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
457b3c2d2ddSMiklos Szeredi {
458b3c2d2ddSMiklos Szeredi 	smp_mb();
459b3c2d2ddSMiklos Szeredi 	if (waitqueue_active(&pipe->wait))
460b3c2d2ddSMiklos Szeredi 		wake_up_interruptible(&pipe->wait);
461b3c2d2ddSMiklos Szeredi 	kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
462b3c2d2ddSMiklos Szeredi }
463b3c2d2ddSMiklos Szeredi 
464b3c2d2ddSMiklos Szeredi /**
465b3c2d2ddSMiklos Szeredi  * splice_from_pipe_feed - feed available data from a pipe to a file
466b3c2d2ddSMiklos Szeredi  * @pipe:	pipe to splice from
467b3c2d2ddSMiklos Szeredi  * @sd:		information to @actor
468b3c2d2ddSMiklos Szeredi  * @actor:	handler that splices the data
469b3c2d2ddSMiklos Szeredi  *
470b3c2d2ddSMiklos Szeredi  * Description:
471b3c2d2ddSMiklos Szeredi  *    This function loops over the pipe and calls @actor to do the
472b3c2d2ddSMiklos Szeredi  *    actual moving of a single struct pipe_buffer to the desired
473b3c2d2ddSMiklos Szeredi  *    destination.  It returns when there's no more buffers left in
474b3c2d2ddSMiklos Szeredi  *    the pipe or if the requested number of bytes (@sd->total_len)
475b3c2d2ddSMiklos Szeredi  *    have been copied.  It returns a positive number (one) if the
476b3c2d2ddSMiklos Szeredi  *    pipe needs to be filled with more data, zero if the required
477b3c2d2ddSMiklos Szeredi  *    number of bytes have been copied and -errno on error.
478b3c2d2ddSMiklos Szeredi  *
479b3c2d2ddSMiklos Szeredi  *    This, together with splice_from_pipe_{begin,end,next}, may be
480b3c2d2ddSMiklos Szeredi  *    used to implement the functionality of __splice_from_pipe() when
481b3c2d2ddSMiklos Szeredi  *    locking is required around copying the pipe buffers to the
482b3c2d2ddSMiklos Szeredi  *    destination.
483b3c2d2ddSMiklos Szeredi  */
48496f9bc8fSAl Viro static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
485b3c2d2ddSMiklos Szeredi 			  splice_actor *actor)
486b3c2d2ddSMiklos Szeredi {
487b3c2d2ddSMiklos Szeredi 	int ret;
488b3c2d2ddSMiklos Szeredi 
489b3c2d2ddSMiklos Szeredi 	while (pipe->nrbufs) {
490b3c2d2ddSMiklos Szeredi 		struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
491b3c2d2ddSMiklos Szeredi 
492b3c2d2ddSMiklos Szeredi 		sd->len = buf->len;
493b3c2d2ddSMiklos Szeredi 		if (sd->len > sd->total_len)
494b3c2d2ddSMiklos Szeredi 			sd->len = sd->total_len;
495b3c2d2ddSMiklos Szeredi 
496fba597dbSMiklos Szeredi 		ret = pipe_buf_confirm(pipe, buf);
497a8adbe37SMichał Mirosław 		if (unlikely(ret)) {
498b3c2d2ddSMiklos Szeredi 			if (ret == -ENODATA)
499b3c2d2ddSMiklos Szeredi 				ret = 0;
500b3c2d2ddSMiklos Szeredi 			return ret;
501b3c2d2ddSMiklos Szeredi 		}
502a8adbe37SMichał Mirosław 
503a8adbe37SMichał Mirosław 		ret = actor(pipe, buf, sd);
504a8adbe37SMichał Mirosław 		if (ret <= 0)
505a8adbe37SMichał Mirosław 			return ret;
506a8adbe37SMichał Mirosław 
507b3c2d2ddSMiklos Szeredi 		buf->offset += ret;
508b3c2d2ddSMiklos Szeredi 		buf->len -= ret;
509b3c2d2ddSMiklos Szeredi 
510b3c2d2ddSMiklos Szeredi 		sd->num_spliced += ret;
511b3c2d2ddSMiklos Szeredi 		sd->len -= ret;
512b3c2d2ddSMiklos Szeredi 		sd->pos += ret;
513b3c2d2ddSMiklos Szeredi 		sd->total_len -= ret;
514b3c2d2ddSMiklos Szeredi 
515b3c2d2ddSMiklos Szeredi 		if (!buf->len) {
516a779638cSMiklos Szeredi 			pipe_buf_release(pipe, buf);
51735f3d14dSJens Axboe 			pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
518b3c2d2ddSMiklos Szeredi 			pipe->nrbufs--;
5196447a3cfSAl Viro 			if (pipe->files)
520b3c2d2ddSMiklos Szeredi 				sd->need_wakeup = true;
521b3c2d2ddSMiklos Szeredi 		}
522b3c2d2ddSMiklos Szeredi 
523b3c2d2ddSMiklos Szeredi 		if (!sd->total_len)
524b3c2d2ddSMiklos Szeredi 			return 0;
525b3c2d2ddSMiklos Szeredi 	}
526b3c2d2ddSMiklos Szeredi 
527b3c2d2ddSMiklos Szeredi 	return 1;
528b3c2d2ddSMiklos Szeredi }
529b3c2d2ddSMiklos Szeredi 
530b3c2d2ddSMiklos Szeredi /**
531b3c2d2ddSMiklos Szeredi  * splice_from_pipe_next - wait for some data to splice from
532b3c2d2ddSMiklos Szeredi  * @pipe:	pipe to splice from
533b3c2d2ddSMiklos Szeredi  * @sd:		information about the splice operation
534b3c2d2ddSMiklos Szeredi  *
535b3c2d2ddSMiklos Szeredi  * Description:
536b3c2d2ddSMiklos Szeredi  *    This function will wait for some data and return a positive
537b3c2d2ddSMiklos Szeredi  *    value (one) if pipe buffers are available.  It will return zero
538b3c2d2ddSMiklos Szeredi  *    or -errno if no more data needs to be spliced.
539b3c2d2ddSMiklos Szeredi  */
54096f9bc8fSAl Viro static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
541b3c2d2ddSMiklos Szeredi {
542c725bfceSJan Kara 	/*
543c725bfceSJan Kara 	 * Check for signal early to make process killable when there are
544c725bfceSJan Kara 	 * always buffers available
545c725bfceSJan Kara 	 */
546c725bfceSJan Kara 	if (signal_pending(current))
547c725bfceSJan Kara 		return -ERESTARTSYS;
548c725bfceSJan Kara 
549b3c2d2ddSMiklos Szeredi 	while (!pipe->nrbufs) {
550b3c2d2ddSMiklos Szeredi 		if (!pipe->writers)
551b3c2d2ddSMiklos Szeredi 			return 0;
552b3c2d2ddSMiklos Szeredi 
553b3c2d2ddSMiklos Szeredi 		if (!pipe->waiting_writers && sd->num_spliced)
554b3c2d2ddSMiklos Szeredi 			return 0;
555b3c2d2ddSMiklos Szeredi 
556b3c2d2ddSMiklos Szeredi 		if (sd->flags & SPLICE_F_NONBLOCK)
557b3c2d2ddSMiklos Szeredi 			return -EAGAIN;
558b3c2d2ddSMiklos Szeredi 
559b3c2d2ddSMiklos Szeredi 		if (signal_pending(current))
560b3c2d2ddSMiklos Szeredi 			return -ERESTARTSYS;
561b3c2d2ddSMiklos Szeredi 
562b3c2d2ddSMiklos Szeredi 		if (sd->need_wakeup) {
563b3c2d2ddSMiklos Szeredi 			wakeup_pipe_writers(pipe);
564b3c2d2ddSMiklos Szeredi 			sd->need_wakeup = false;
565b3c2d2ddSMiklos Szeredi 		}
566b3c2d2ddSMiklos Szeredi 
567b3c2d2ddSMiklos Szeredi 		pipe_wait(pipe);
568b3c2d2ddSMiklos Szeredi 	}
569b3c2d2ddSMiklos Szeredi 
570b3c2d2ddSMiklos Szeredi 	return 1;
571b3c2d2ddSMiklos Szeredi }
572b3c2d2ddSMiklos Szeredi 
573b3c2d2ddSMiklos Szeredi /**
574b3c2d2ddSMiklos Szeredi  * splice_from_pipe_begin - start splicing from pipe
575b80901bbSRandy Dunlap  * @sd:		information about the splice operation
576b3c2d2ddSMiklos Szeredi  *
577b3c2d2ddSMiklos Szeredi  * Description:
578b3c2d2ddSMiklos Szeredi  *    This function should be called before a loop containing
579b3c2d2ddSMiklos Szeredi  *    splice_from_pipe_next() and splice_from_pipe_feed() to
580b3c2d2ddSMiklos Szeredi  *    initialize the necessary fields of @sd.
581b3c2d2ddSMiklos Szeredi  */
58296f9bc8fSAl Viro static void splice_from_pipe_begin(struct splice_desc *sd)
583b3c2d2ddSMiklos Szeredi {
584b3c2d2ddSMiklos Szeredi 	sd->num_spliced = 0;
585b3c2d2ddSMiklos Szeredi 	sd->need_wakeup = false;
586b3c2d2ddSMiklos Szeredi }
587b3c2d2ddSMiklos Szeredi 
588b3c2d2ddSMiklos Szeredi /**
589b3c2d2ddSMiklos Szeredi  * splice_from_pipe_end - finish splicing from pipe
590b3c2d2ddSMiklos Szeredi  * @pipe:	pipe to splice from
591b3c2d2ddSMiklos Szeredi  * @sd:		information about the splice operation
592b3c2d2ddSMiklos Szeredi  *
593b3c2d2ddSMiklos Szeredi  * Description:
594b3c2d2ddSMiklos Szeredi  *    This function will wake up pipe writers if necessary.  It should
595b3c2d2ddSMiklos Szeredi  *    be called after a loop containing splice_from_pipe_next() and
596b3c2d2ddSMiklos Szeredi  *    splice_from_pipe_feed().
597b3c2d2ddSMiklos Szeredi  */
59896f9bc8fSAl Viro static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
599b3c2d2ddSMiklos Szeredi {
600b3c2d2ddSMiklos Szeredi 	if (sd->need_wakeup)
601b3c2d2ddSMiklos Szeredi 		wakeup_pipe_writers(pipe);
602b3c2d2ddSMiklos Szeredi }
603b3c2d2ddSMiklos Szeredi 
604932cc6d4SJens Axboe /**
605932cc6d4SJens Axboe  * __splice_from_pipe - splice data from a pipe to given actor
606932cc6d4SJens Axboe  * @pipe:	pipe to splice from
607932cc6d4SJens Axboe  * @sd:		information to @actor
608932cc6d4SJens Axboe  * @actor:	handler that splices the data
609932cc6d4SJens Axboe  *
610932cc6d4SJens Axboe  * Description:
611932cc6d4SJens Axboe  *    This function does little more than loop over the pipe and call
612932cc6d4SJens Axboe  *    @actor to do the actual moving of a single struct pipe_buffer to
613932cc6d4SJens Axboe  *    the desired destination. See pipe_to_file, pipe_to_sendpage, or
614932cc6d4SJens Axboe  *    pipe_to_user.
615932cc6d4SJens Axboe  *
61683f9135bSJens Axboe  */
617c66ab6faSJens Axboe ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
618c66ab6faSJens Axboe 			   splice_actor *actor)
6195274f052SJens Axboe {
620b3c2d2ddSMiklos Szeredi 	int ret;
6215274f052SJens Axboe 
622b3c2d2ddSMiklos Szeredi 	splice_from_pipe_begin(sd);
623b3c2d2ddSMiklos Szeredi 	do {
624c2489e07SJan Kara 		cond_resched();
625b3c2d2ddSMiklos Szeredi 		ret = splice_from_pipe_next(pipe, sd);
626b3c2d2ddSMiklos Szeredi 		if (ret > 0)
627b3c2d2ddSMiklos Szeredi 			ret = splice_from_pipe_feed(pipe, sd, actor);
628b3c2d2ddSMiklos Szeredi 	} while (ret > 0);
629b3c2d2ddSMiklos Szeredi 	splice_from_pipe_end(pipe, sd);
6305274f052SJens Axboe 
631b3c2d2ddSMiklos Szeredi 	return sd->num_spliced ? sd->num_spliced : ret;
6325274f052SJens Axboe }
63340bee44eSMark Fasheh EXPORT_SYMBOL(__splice_from_pipe);
6345274f052SJens Axboe 
635932cc6d4SJens Axboe /**
636932cc6d4SJens Axboe  * splice_from_pipe - splice data from a pipe to a file
637932cc6d4SJens Axboe  * @pipe:	pipe to splice from
638932cc6d4SJens Axboe  * @out:	file to splice to
639932cc6d4SJens Axboe  * @ppos:	position in @out
640932cc6d4SJens Axboe  * @len:	how many bytes to splice
641932cc6d4SJens Axboe  * @flags:	splice modifier flags
642932cc6d4SJens Axboe  * @actor:	handler that splices the data
643932cc6d4SJens Axboe  *
644932cc6d4SJens Axboe  * Description:
6452933970bSMiklos Szeredi  *    See __splice_from_pipe. This function locks the pipe inode,
646932cc6d4SJens Axboe  *    otherwise it's identical to __splice_from_pipe().
647932cc6d4SJens Axboe  *
648932cc6d4SJens Axboe  */
6496da61809SMark Fasheh ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
6506da61809SMark Fasheh 			 loff_t *ppos, size_t len, unsigned int flags,
6516da61809SMark Fasheh 			 splice_actor *actor)
6526da61809SMark Fasheh {
6536da61809SMark Fasheh 	ssize_t ret;
654c66ab6faSJens Axboe 	struct splice_desc sd = {
655c66ab6faSJens Axboe 		.total_len = len,
656c66ab6faSJens Axboe 		.flags = flags,
657c66ab6faSJens Axboe 		.pos = *ppos,
6586a14b90bSJens Axboe 		.u.file = out,
659c66ab6faSJens Axboe 	};
6606da61809SMark Fasheh 
66161e0d47cSMiklos Szeredi 	pipe_lock(pipe);
662c66ab6faSJens Axboe 	ret = __splice_from_pipe(pipe, &sd, actor);
66361e0d47cSMiklos Szeredi 	pipe_unlock(pipe);
6646da61809SMark Fasheh 
6656da61809SMark Fasheh 	return ret;
6666da61809SMark Fasheh }
6676da61809SMark Fasheh 
6686da61809SMark Fasheh /**
6698d020765SAl Viro  * iter_file_splice_write - splice data from a pipe to a file
6708d020765SAl Viro  * @pipe:	pipe info
6718d020765SAl Viro  * @out:	file to write to
6728d020765SAl Viro  * @ppos:	position in @out
6738d020765SAl Viro  * @len:	number of bytes to splice
6748d020765SAl Viro  * @flags:	splice modifier flags
6758d020765SAl Viro  *
6768d020765SAl Viro  * Description:
6778d020765SAl Viro  *    Will either move or copy pages (determined by @flags options) from
6788d020765SAl Viro  *    the given pipe inode to the given file.
6798d020765SAl Viro  *    This one is ->write_iter-based.
6808d020765SAl Viro  *
6818d020765SAl Viro  */
6828d020765SAl Viro ssize_t
6838d020765SAl Viro iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
6848d020765SAl Viro 			  loff_t *ppos, size_t len, unsigned int flags)
6858d020765SAl Viro {
6868d020765SAl Viro 	struct splice_desc sd = {
6878d020765SAl Viro 		.total_len = len,
6888d020765SAl Viro 		.flags = flags,
6898d020765SAl Viro 		.pos = *ppos,
6908d020765SAl Viro 		.u.file = out,
6918d020765SAl Viro 	};
6928d020765SAl Viro 	int nbufs = pipe->buffers;
6938d020765SAl Viro 	struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
6948d020765SAl Viro 					GFP_KERNEL);
6958d020765SAl Viro 	ssize_t ret;
6968d020765SAl Viro 
6978d020765SAl Viro 	if (unlikely(!array))
6988d020765SAl Viro 		return -ENOMEM;
6998d020765SAl Viro 
7008d020765SAl Viro 	pipe_lock(pipe);
7018d020765SAl Viro 
7028d020765SAl Viro 	splice_from_pipe_begin(&sd);
7038d020765SAl Viro 	while (sd.total_len) {
7048d020765SAl Viro 		struct iov_iter from;
7058d020765SAl Viro 		size_t left;
7068d020765SAl Viro 		int n, idx;
7078d020765SAl Viro 
7088d020765SAl Viro 		ret = splice_from_pipe_next(pipe, &sd);
7098d020765SAl Viro 		if (ret <= 0)
7108d020765SAl Viro 			break;
7118d020765SAl Viro 
7128d020765SAl Viro 		if (unlikely(nbufs < pipe->buffers)) {
7138d020765SAl Viro 			kfree(array);
7148d020765SAl Viro 			nbufs = pipe->buffers;
7158d020765SAl Viro 			array = kcalloc(nbufs, sizeof(struct bio_vec),
7168d020765SAl Viro 					GFP_KERNEL);
7178d020765SAl Viro 			if (!array) {
7188d020765SAl Viro 				ret = -ENOMEM;
7198d020765SAl Viro 				break;
7208d020765SAl Viro 			}
7218d020765SAl Viro 		}
7228d020765SAl Viro 
7238d020765SAl Viro 		/* build the vector */
7248d020765SAl Viro 		left = sd.total_len;
7258d020765SAl Viro 		for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) {
7268d020765SAl Viro 			struct pipe_buffer *buf = pipe->bufs + idx;
7278d020765SAl Viro 			size_t this_len = buf->len;
7288d020765SAl Viro 
7298d020765SAl Viro 			if (this_len > left)
7308d020765SAl Viro 				this_len = left;
7318d020765SAl Viro 
7328d020765SAl Viro 			if (idx == pipe->buffers - 1)
7338d020765SAl Viro 				idx = -1;
7348d020765SAl Viro 
735fba597dbSMiklos Szeredi 			ret = pipe_buf_confirm(pipe, buf);
7368d020765SAl Viro 			if (unlikely(ret)) {
7378d020765SAl Viro 				if (ret == -ENODATA)
7388d020765SAl Viro 					ret = 0;
7398d020765SAl Viro 				goto done;
7408d020765SAl Viro 			}
7418d020765SAl Viro 
7428d020765SAl Viro 			array[n].bv_page = buf->page;
7438d020765SAl Viro 			array[n].bv_len = this_len;
7448d020765SAl Viro 			array[n].bv_offset = buf->offset;
7458d020765SAl Viro 			left -= this_len;
7468d020765SAl Viro 		}
7478d020765SAl Viro 
748aa563d7bSDavid Howells 		iov_iter_bvec(&from, WRITE, array, n, sd.total_len - left);
749abbb6589SChristoph Hellwig 		ret = vfs_iter_write(out, &from, &sd.pos, 0);
7508d020765SAl Viro 		if (ret <= 0)
7518d020765SAl Viro 			break;
7528d020765SAl Viro 
7538d020765SAl Viro 		sd.num_spliced += ret;
7548d020765SAl Viro 		sd.total_len -= ret;
755dbe4e192SChristoph Hellwig 		*ppos = sd.pos;
7568d020765SAl Viro 
7578d020765SAl Viro 		/* dismiss the fully eaten buffers, adjust the partial one */
7588d020765SAl Viro 		while (ret) {
7598d020765SAl Viro 			struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
7608d020765SAl Viro 			if (ret >= buf->len) {
7618d020765SAl Viro 				ret -= buf->len;
7628d020765SAl Viro 				buf->len = 0;
763a779638cSMiklos Szeredi 				pipe_buf_release(pipe, buf);
7648d020765SAl Viro 				pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
7658d020765SAl Viro 				pipe->nrbufs--;
7668d020765SAl Viro 				if (pipe->files)
7678d020765SAl Viro 					sd.need_wakeup = true;
7688d020765SAl Viro 			} else {
7698d020765SAl Viro 				buf->offset += ret;
7708d020765SAl Viro 				buf->len -= ret;
7718d020765SAl Viro 				ret = 0;
7728d020765SAl Viro 			}
7738d020765SAl Viro 		}
7748d020765SAl Viro 	}
7758d020765SAl Viro done:
7768d020765SAl Viro 	kfree(array);
7778d020765SAl Viro 	splice_from_pipe_end(pipe, &sd);
7788d020765SAl Viro 
7798d020765SAl Viro 	pipe_unlock(pipe);
7808d020765SAl Viro 
7818d020765SAl Viro 	if (sd.num_spliced)
7828d020765SAl Viro 		ret = sd.num_spliced;
7838d020765SAl Viro 
7848d020765SAl Viro 	return ret;
7858d020765SAl Viro }
7868d020765SAl Viro 
7878d020765SAl Viro EXPORT_SYMBOL(iter_file_splice_write);
7888d020765SAl Viro 
789b2858d7dSMiklos Szeredi static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
790b2858d7dSMiklos Szeredi 			  struct splice_desc *sd)
7910b0a47f5SMiklos Szeredi {
792b2858d7dSMiklos Szeredi 	int ret;
793b2858d7dSMiklos Szeredi 	void *data;
79406ae43f3SAl Viro 	loff_t tmp = sd->pos;
795b2858d7dSMiklos Szeredi 
796fbb32750SAl Viro 	data = kmap(buf->page);
79706ae43f3SAl Viro 	ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
798fbb32750SAl Viro 	kunmap(buf->page);
799b2858d7dSMiklos Szeredi 
800b2858d7dSMiklos Szeredi 	return ret;
8010b0a47f5SMiklos Szeredi }
8020b0a47f5SMiklos Szeredi 
8030b0a47f5SMiklos Szeredi static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
8040b0a47f5SMiklos Szeredi 					 struct file *out, loff_t *ppos,
8050b0a47f5SMiklos Szeredi 					 size_t len, unsigned int flags)
8060b0a47f5SMiklos Szeredi {
807b2858d7dSMiklos Szeredi 	ssize_t ret;
8080b0a47f5SMiklos Szeredi 
809b2858d7dSMiklos Szeredi 	ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
810b2858d7dSMiklos Szeredi 	if (ret > 0)
811b2858d7dSMiklos Szeredi 		*ppos += ret;
8120b0a47f5SMiklos Szeredi 
813b2858d7dSMiklos Szeredi 	return ret;
8140b0a47f5SMiklos Szeredi }
8150b0a47f5SMiklos Szeredi 
81683f9135bSJens Axboe /**
81783f9135bSJens Axboe  * generic_splice_sendpage - splice data from a pipe to a socket
818932cc6d4SJens Axboe  * @pipe:	pipe to splice from
81983f9135bSJens Axboe  * @out:	socket to write to
820932cc6d4SJens Axboe  * @ppos:	position in @out
82183f9135bSJens Axboe  * @len:	number of bytes to splice
82283f9135bSJens Axboe  * @flags:	splice modifier flags
82383f9135bSJens Axboe  *
824932cc6d4SJens Axboe  * Description:
82583f9135bSJens Axboe  *    Will send @len bytes from the pipe to a network socket. No data copying
82683f9135bSJens Axboe  *    is involved.
82783f9135bSJens Axboe  *
82883f9135bSJens Axboe  */
8293a326a2cSIngo Molnar ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
830cbb7e577SJens Axboe 				loff_t *ppos, size_t len, unsigned int flags)
8315274f052SJens Axboe {
83200522fb4SJens Axboe 	return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
8335274f052SJens Axboe }
8345274f052SJens Axboe 
835059a8f37SJens Axboe EXPORT_SYMBOL(generic_splice_sendpage);
836a0f06780SJeff Garzik 
83783f9135bSJens Axboe /*
83883f9135bSJens Axboe  * Attempt to initiate a splice from pipe to file.
83983f9135bSJens Axboe  */
8403a326a2cSIngo Molnar static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
841cbb7e577SJens Axboe 			   loff_t *ppos, size_t len, unsigned int flags)
8425274f052SJens Axboe {
8430b0a47f5SMiklos Szeredi 	ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
8440b0a47f5SMiklos Szeredi 				loff_t *, size_t, unsigned int);
8455274f052SJens Axboe 
84672c2d531SAl Viro 	if (out->f_op->splice_write)
8470b0a47f5SMiklos Szeredi 		splice_write = out->f_op->splice_write;
848cc56f7deSChangli Gao 	else
8490b0a47f5SMiklos Szeredi 		splice_write = default_file_splice_write;
8500b0a47f5SMiklos Szeredi 
851500368f7SAl Viro 	return splice_write(pipe, out, ppos, len, flags);
8525274f052SJens Axboe }
8535274f052SJens Axboe 
85483f9135bSJens Axboe /*
85583f9135bSJens Axboe  * Attempt to initiate a splice from a file to a pipe.
85683f9135bSJens Axboe  */
857cbb7e577SJens Axboe static long do_splice_to(struct file *in, loff_t *ppos,
858cbb7e577SJens Axboe 			 struct pipe_inode_info *pipe, size_t len,
859cbb7e577SJens Axboe 			 unsigned int flags)
8605274f052SJens Axboe {
8616818173bSMiklos Szeredi 	ssize_t (*splice_read)(struct file *, loff_t *,
8626818173bSMiklos Szeredi 			       struct pipe_inode_info *, size_t, unsigned int);
8635274f052SJens Axboe 	int ret;
8645274f052SJens Axboe 
86549570e9bSJens Axboe 	if (unlikely(!(in->f_mode & FMODE_READ)))
8665274f052SJens Axboe 		return -EBADF;
8675274f052SJens Axboe 
868cbb7e577SJens Axboe 	ret = rw_verify_area(READ, in, ppos, len);
8695274f052SJens Axboe 	if (unlikely(ret < 0))
8705274f052SJens Axboe 		return ret;
8715274f052SJens Axboe 
87203cc0789SAl Viro 	if (unlikely(len > MAX_RW_COUNT))
87303cc0789SAl Viro 		len = MAX_RW_COUNT;
87403cc0789SAl Viro 
87572c2d531SAl Viro 	if (in->f_op->splice_read)
8766818173bSMiklos Szeredi 		splice_read = in->f_op->splice_read;
877cc56f7deSChangli Gao 	else
8786818173bSMiklos Szeredi 		splice_read = default_file_splice_read;
8796818173bSMiklos Szeredi 
8806818173bSMiklos Szeredi 	return splice_read(in, ppos, pipe, len, flags);
8815274f052SJens Axboe }
8825274f052SJens Axboe 
883932cc6d4SJens Axboe /**
884932cc6d4SJens Axboe  * splice_direct_to_actor - splices data directly between two non-pipes
885932cc6d4SJens Axboe  * @in:		file to splice from
886932cc6d4SJens Axboe  * @sd:		actor information on where to splice to
887932cc6d4SJens Axboe  * @actor:	handles the data splicing
888932cc6d4SJens Axboe  *
889932cc6d4SJens Axboe  * Description:
890932cc6d4SJens Axboe  *    This is a special case helper to splice directly between two
891932cc6d4SJens Axboe  *    points, without requiring an explicit pipe. Internally an allocated
892932cc6d4SJens Axboe  *    pipe is cached in the process, and reused during the lifetime of
893932cc6d4SJens Axboe  *    that process.
894932cc6d4SJens Axboe  *
895c66ab6faSJens Axboe  */
896c66ab6faSJens Axboe ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
897c66ab6faSJens Axboe 			       splice_direct_actor *actor)
898b92ce558SJens Axboe {
899b92ce558SJens Axboe 	struct pipe_inode_info *pipe;
900b92ce558SJens Axboe 	long ret, bytes;
901b92ce558SJens Axboe 	umode_t i_mode;
902c66ab6faSJens Axboe 	size_t len;
9030ff28d9fSChristophe Leroy 	int i, flags, more;
904b92ce558SJens Axboe 
905b92ce558SJens Axboe 	/*
906b92ce558SJens Axboe 	 * We require the input being a regular file, as we don't want to
907b92ce558SJens Axboe 	 * randomly drop data for eg socket -> socket splicing. Use the
908b92ce558SJens Axboe 	 * piped splicing for that!
909b92ce558SJens Axboe 	 */
910496ad9aaSAl Viro 	i_mode = file_inode(in)->i_mode;
911b92ce558SJens Axboe 	if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
912b92ce558SJens Axboe 		return -EINVAL;
913b92ce558SJens Axboe 
914b92ce558SJens Axboe 	/*
915b92ce558SJens Axboe 	 * neither in nor out is a pipe, setup an internal pipe attached to
916b92ce558SJens Axboe 	 * 'out' and transfer the wanted data from 'in' to 'out' through that
917b92ce558SJens Axboe 	 */
918b92ce558SJens Axboe 	pipe = current->splice_pipe;
91949570e9bSJens Axboe 	if (unlikely(!pipe)) {
9207bee130eSAl Viro 		pipe = alloc_pipe_info();
921b92ce558SJens Axboe 		if (!pipe)
922b92ce558SJens Axboe 			return -ENOMEM;
923b92ce558SJens Axboe 
924b92ce558SJens Axboe 		/*
925b92ce558SJens Axboe 		 * We don't have an immediate reader, but we'll read the stuff
92600522fb4SJens Axboe 		 * out of the pipe right after the splice_to_pipe(). So set
927b92ce558SJens Axboe 		 * PIPE_READERS appropriately.
928b92ce558SJens Axboe 		 */
929b92ce558SJens Axboe 		pipe->readers = 1;
930b92ce558SJens Axboe 
931b92ce558SJens Axboe 		current->splice_pipe = pipe;
932b92ce558SJens Axboe 	}
933b92ce558SJens Axboe 
934b92ce558SJens Axboe 	/*
93573d62d83SIngo Molnar 	 * Do the splice.
936b92ce558SJens Axboe 	 */
937b92ce558SJens Axboe 	ret = 0;
938b92ce558SJens Axboe 	bytes = 0;
939c66ab6faSJens Axboe 	len = sd->total_len;
940c66ab6faSJens Axboe 	flags = sd->flags;
941c66ab6faSJens Axboe 
942c66ab6faSJens Axboe 	/*
943c66ab6faSJens Axboe 	 * Don't block on output, we have to drain the direct pipe.
944c66ab6faSJens Axboe 	 */
945c66ab6faSJens Axboe 	sd->flags &= ~SPLICE_F_NONBLOCK;
9460ff28d9fSChristophe Leroy 	more = sd->flags & SPLICE_F_MORE;
947b92ce558SJens Axboe 
94817614445SDarrick J. Wong 	WARN_ON_ONCE(pipe->nrbufs != 0);
94917614445SDarrick J. Wong 
950b92ce558SJens Axboe 	while (len) {
95151a92c0fSJens Axboe 		size_t read_len;
952a82c53a0STom Zanussi 		loff_t pos = sd->pos, prev_pos = pos;
953b92ce558SJens Axboe 
95417614445SDarrick J. Wong 		/* Don't try to read more the pipe has space for. */
95517614445SDarrick J. Wong 		read_len = min_t(size_t, len,
95617614445SDarrick J. Wong 				 (pipe->buffers - pipe->nrbufs) << PAGE_SHIFT);
95717614445SDarrick J. Wong 		ret = do_splice_to(in, &pos, pipe, read_len, flags);
95851a92c0fSJens Axboe 		if (unlikely(ret <= 0))
959b92ce558SJens Axboe 			goto out_release;
960b92ce558SJens Axboe 
961b92ce558SJens Axboe 		read_len = ret;
962c66ab6faSJens Axboe 		sd->total_len = read_len;
963b92ce558SJens Axboe 
964b92ce558SJens Axboe 		/*
9650ff28d9fSChristophe Leroy 		 * If more data is pending, set SPLICE_F_MORE
9660ff28d9fSChristophe Leroy 		 * If this is the last data and SPLICE_F_MORE was not set
9670ff28d9fSChristophe Leroy 		 * initially, clears it.
9680ff28d9fSChristophe Leroy 		 */
9690ff28d9fSChristophe Leroy 		if (read_len < len)
9700ff28d9fSChristophe Leroy 			sd->flags |= SPLICE_F_MORE;
9710ff28d9fSChristophe Leroy 		else if (!more)
9720ff28d9fSChristophe Leroy 			sd->flags &= ~SPLICE_F_MORE;
9730ff28d9fSChristophe Leroy 		/*
974b92ce558SJens Axboe 		 * NOTE: nonblocking mode only applies to the input. We
975b92ce558SJens Axboe 		 * must not do the output in nonblocking mode as then we
976b92ce558SJens Axboe 		 * could get stuck data in the internal pipe:
977b92ce558SJens Axboe 		 */
978c66ab6faSJens Axboe 		ret = actor(pipe, sd);
979a82c53a0STom Zanussi 		if (unlikely(ret <= 0)) {
980a82c53a0STom Zanussi 			sd->pos = prev_pos;
981b92ce558SJens Axboe 			goto out_release;
982a82c53a0STom Zanussi 		}
983b92ce558SJens Axboe 
984b92ce558SJens Axboe 		bytes += ret;
985b92ce558SJens Axboe 		len -= ret;
986bcd4f3acSJens Axboe 		sd->pos = pos;
987b92ce558SJens Axboe 
988a82c53a0STom Zanussi 		if (ret < read_len) {
989a82c53a0STom Zanussi 			sd->pos = prev_pos + ret;
99051a92c0fSJens Axboe 			goto out_release;
991b92ce558SJens Axboe 		}
992a82c53a0STom Zanussi 	}
993b92ce558SJens Axboe 
9949e97198dSJens Axboe done:
995b92ce558SJens Axboe 	pipe->nrbufs = pipe->curbuf = 0;
9969e97198dSJens Axboe 	file_accessed(in);
997b92ce558SJens Axboe 	return bytes;
998b92ce558SJens Axboe 
999b92ce558SJens Axboe out_release:
1000b92ce558SJens Axboe 	/*
1001b92ce558SJens Axboe 	 * If we did an incomplete transfer we must release
1002b92ce558SJens Axboe 	 * the pipe buffers in question:
1003b92ce558SJens Axboe 	 */
100435f3d14dSJens Axboe 	for (i = 0; i < pipe->buffers; i++) {
1005b92ce558SJens Axboe 		struct pipe_buffer *buf = pipe->bufs + i;
1006b92ce558SJens Axboe 
1007a779638cSMiklos Szeredi 		if (buf->ops)
1008a779638cSMiklos Szeredi 			pipe_buf_release(pipe, buf);
1009b92ce558SJens Axboe 	}
1010b92ce558SJens Axboe 
10119e97198dSJens Axboe 	if (!bytes)
10129e97198dSJens Axboe 		bytes = ret;
1013b92ce558SJens Axboe 
10149e97198dSJens Axboe 	goto done;
1015c66ab6faSJens Axboe }
1016c66ab6faSJens Axboe EXPORT_SYMBOL(splice_direct_to_actor);
1017c66ab6faSJens Axboe 
1018c66ab6faSJens Axboe static int direct_splice_actor(struct pipe_inode_info *pipe,
1019c66ab6faSJens Axboe 			       struct splice_desc *sd)
1020c66ab6faSJens Axboe {
10216a14b90bSJens Axboe 	struct file *file = sd->u.file;
1022c66ab6faSJens Axboe 
10237995bd28SAl Viro 	return do_splice_from(pipe, file, sd->opos, sd->total_len,
10242cb4b05eSChangli Gao 			      sd->flags);
1025c66ab6faSJens Axboe }
1026c66ab6faSJens Axboe 
1027932cc6d4SJens Axboe /**
1028932cc6d4SJens Axboe  * do_splice_direct - splices data directly between two files
1029932cc6d4SJens Axboe  * @in:		file to splice from
1030932cc6d4SJens Axboe  * @ppos:	input file offset
1031932cc6d4SJens Axboe  * @out:	file to splice to
1032acdb37c3SRandy Dunlap  * @opos:	output file offset
1033932cc6d4SJens Axboe  * @len:	number of bytes to splice
1034932cc6d4SJens Axboe  * @flags:	splice modifier flags
1035932cc6d4SJens Axboe  *
1036932cc6d4SJens Axboe  * Description:
1037932cc6d4SJens Axboe  *    For use by do_sendfile(). splice can easily emulate sendfile, but
1038932cc6d4SJens Axboe  *    doing it in the application would incur an extra system call
1039932cc6d4SJens Axboe  *    (splice in + splice out, as compared to just sendfile()). So this helper
1040932cc6d4SJens Axboe  *    can splice directly through a process-private pipe.
1041932cc6d4SJens Axboe  *
1042932cc6d4SJens Axboe  */
1043c66ab6faSJens Axboe long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
10447995bd28SAl Viro 		      loff_t *opos, size_t len, unsigned int flags)
1045c66ab6faSJens Axboe {
1046c66ab6faSJens Axboe 	struct splice_desc sd = {
1047c66ab6faSJens Axboe 		.len		= len,
1048c66ab6faSJens Axboe 		.total_len	= len,
1049c66ab6faSJens Axboe 		.flags		= flags,
1050c66ab6faSJens Axboe 		.pos		= *ppos,
10516a14b90bSJens Axboe 		.u.file		= out,
10527995bd28SAl Viro 		.opos		= opos,
1053c66ab6faSJens Axboe 	};
105451a92c0fSJens Axboe 	long ret;
1055c66ab6faSJens Axboe 
105618c67cb9SAl Viro 	if (unlikely(!(out->f_mode & FMODE_WRITE)))
105718c67cb9SAl Viro 		return -EBADF;
105818c67cb9SAl Viro 
105918c67cb9SAl Viro 	if (unlikely(out->f_flags & O_APPEND))
106018c67cb9SAl Viro 		return -EINVAL;
106118c67cb9SAl Viro 
106218c67cb9SAl Viro 	ret = rw_verify_area(WRITE, out, opos, len);
106318c67cb9SAl Viro 	if (unlikely(ret < 0))
106418c67cb9SAl Viro 		return ret;
106518c67cb9SAl Viro 
1066c66ab6faSJens Axboe 	ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
106751a92c0fSJens Axboe 	if (ret > 0)
1068a82c53a0STom Zanussi 		*ppos = sd.pos;
106951a92c0fSJens Axboe 
1070c66ab6faSJens Axboe 	return ret;
1071b92ce558SJens Axboe }
10721c118596SMiklos Szeredi EXPORT_SYMBOL(do_splice_direct);
1073b92ce558SJens Axboe 
10748924feffSAl Viro static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags)
10758924feffSAl Viro {
107652bce911SLinus Torvalds 	for (;;) {
107752bce911SLinus Torvalds 		if (unlikely(!pipe->readers)) {
107852bce911SLinus Torvalds 			send_sig(SIGPIPE, current, 0);
107952bce911SLinus Torvalds 			return -EPIPE;
108052bce911SLinus Torvalds 		}
108152bce911SLinus Torvalds 		if (pipe->nrbufs != pipe->buffers)
108252bce911SLinus Torvalds 			return 0;
10838924feffSAl Viro 		if (flags & SPLICE_F_NONBLOCK)
10848924feffSAl Viro 			return -EAGAIN;
10858924feffSAl Viro 		if (signal_pending(current))
10868924feffSAl Viro 			return -ERESTARTSYS;
10878924feffSAl Viro 		pipe->waiting_writers++;
10888924feffSAl Viro 		pipe_wait(pipe);
10898924feffSAl Viro 		pipe->waiting_writers--;
10908924feffSAl Viro 	}
10918924feffSAl Viro }
10928924feffSAl Viro 
10937c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
10947c77f0b3SMiklos Szeredi 			       struct pipe_inode_info *opipe,
10957c77f0b3SMiklos Szeredi 			       size_t len, unsigned int flags);
1096ddac0d39SJens Axboe 
1097ddac0d39SJens Axboe /*
109883f9135bSJens Axboe  * Determine where to splice to/from.
109983f9135bSJens Axboe  */
1100529565dcSIngo Molnar static long do_splice(struct file *in, loff_t __user *off_in,
1101529565dcSIngo Molnar 		      struct file *out, loff_t __user *off_out,
1102529565dcSIngo Molnar 		      size_t len, unsigned int flags)
11035274f052SJens Axboe {
11047c77f0b3SMiklos Szeredi 	struct pipe_inode_info *ipipe;
11057c77f0b3SMiklos Szeredi 	struct pipe_inode_info *opipe;
11067995bd28SAl Viro 	loff_t offset;
1107a4514ebdSJens Axboe 	long ret;
11085274f052SJens Axboe 
110971993e62SLinus Torvalds 	ipipe = get_pipe_info(in);
111071993e62SLinus Torvalds 	opipe = get_pipe_info(out);
11117c77f0b3SMiklos Szeredi 
11127c77f0b3SMiklos Szeredi 	if (ipipe && opipe) {
11137c77f0b3SMiklos Szeredi 		if (off_in || off_out)
11147c77f0b3SMiklos Szeredi 			return -ESPIPE;
11157c77f0b3SMiklos Szeredi 
11167c77f0b3SMiklos Szeredi 		if (!(in->f_mode & FMODE_READ))
11177c77f0b3SMiklos Szeredi 			return -EBADF;
11187c77f0b3SMiklos Szeredi 
11197c77f0b3SMiklos Szeredi 		if (!(out->f_mode & FMODE_WRITE))
11207c77f0b3SMiklos Szeredi 			return -EBADF;
11217c77f0b3SMiklos Szeredi 
11227c77f0b3SMiklos Szeredi 		/* Splicing to self would be fun, but... */
11237c77f0b3SMiklos Szeredi 		if (ipipe == opipe)
11247c77f0b3SMiklos Szeredi 			return -EINVAL;
11257c77f0b3SMiklos Szeredi 
11267c77f0b3SMiklos Szeredi 		return splice_pipe_to_pipe(ipipe, opipe, len, flags);
11277c77f0b3SMiklos Szeredi 	}
11287c77f0b3SMiklos Szeredi 
11297c77f0b3SMiklos Szeredi 	if (ipipe) {
1130529565dcSIngo Molnar 		if (off_in)
1131529565dcSIngo Molnar 			return -ESPIPE;
1132b92ce558SJens Axboe 		if (off_out) {
113319c9a49bSChangli Gao 			if (!(out->f_mode & FMODE_PWRITE))
1134b92ce558SJens Axboe 				return -EINVAL;
1135cbb7e577SJens Axboe 			if (copy_from_user(&offset, off_out, sizeof(loff_t)))
1136b92ce558SJens Axboe 				return -EFAULT;
11377995bd28SAl Viro 		} else {
11387995bd28SAl Viro 			offset = out->f_pos;
11397995bd28SAl Viro 		}
1140529565dcSIngo Molnar 
114118c67cb9SAl Viro 		if (unlikely(!(out->f_mode & FMODE_WRITE)))
114218c67cb9SAl Viro 			return -EBADF;
114318c67cb9SAl Viro 
114418c67cb9SAl Viro 		if (unlikely(out->f_flags & O_APPEND))
114518c67cb9SAl Viro 			return -EINVAL;
114618c67cb9SAl Viro 
114718c67cb9SAl Viro 		ret = rw_verify_area(WRITE, out, &offset, len);
114818c67cb9SAl Viro 		if (unlikely(ret < 0))
114918c67cb9SAl Viro 			return ret;
115018c67cb9SAl Viro 
1151500368f7SAl Viro 		file_start_write(out);
11527995bd28SAl Viro 		ret = do_splice_from(ipipe, out, &offset, len, flags);
1153500368f7SAl Viro 		file_end_write(out);
1154a4514ebdSJens Axboe 
11557995bd28SAl Viro 		if (!off_out)
11567995bd28SAl Viro 			out->f_pos = offset;
11577995bd28SAl Viro 		else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
1158a4514ebdSJens Axboe 			ret = -EFAULT;
1159a4514ebdSJens Axboe 
1160a4514ebdSJens Axboe 		return ret;
1161529565dcSIngo Molnar 	}
11625274f052SJens Axboe 
11637c77f0b3SMiklos Szeredi 	if (opipe) {
1164529565dcSIngo Molnar 		if (off_out)
1165529565dcSIngo Molnar 			return -ESPIPE;
1166b92ce558SJens Axboe 		if (off_in) {
116719c9a49bSChangli Gao 			if (!(in->f_mode & FMODE_PREAD))
1168b92ce558SJens Axboe 				return -EINVAL;
1169cbb7e577SJens Axboe 			if (copy_from_user(&offset, off_in, sizeof(loff_t)))
1170b92ce558SJens Axboe 				return -EFAULT;
11717995bd28SAl Viro 		} else {
11727995bd28SAl Viro 			offset = in->f_pos;
11737995bd28SAl Viro 		}
1174529565dcSIngo Molnar 
11758924feffSAl Viro 		pipe_lock(opipe);
11768924feffSAl Viro 		ret = wait_for_space(opipe, flags);
11778924feffSAl Viro 		if (!ret)
11787995bd28SAl Viro 			ret = do_splice_to(in, &offset, opipe, len, flags);
11798924feffSAl Viro 		pipe_unlock(opipe);
11808924feffSAl Viro 		if (ret > 0)
11818924feffSAl Viro 			wakeup_pipe_readers(opipe);
11827995bd28SAl Viro 		if (!off_in)
11837995bd28SAl Viro 			in->f_pos = offset;
11847995bd28SAl Viro 		else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
1185a4514ebdSJens Axboe 			ret = -EFAULT;
1186a4514ebdSJens Axboe 
1187a4514ebdSJens Axboe 		return ret;
1188529565dcSIngo Molnar 	}
11895274f052SJens Axboe 
11905274f052SJens Axboe 	return -EINVAL;
11915274f052SJens Axboe }
11925274f052SJens Axboe 
119379fddc4eSAl Viro static int iter_to_pipe(struct iov_iter *from,
119479fddc4eSAl Viro 			struct pipe_inode_info *pipe,
119579fddc4eSAl Viro 			unsigned flags)
1196912d35f8SJens Axboe {
119779fddc4eSAl Viro 	struct pipe_buffer buf = {
119879fddc4eSAl Viro 		.ops = &user_page_pipe_buf_ops,
119979fddc4eSAl Viro 		.flags = flags
120079fddc4eSAl Viro 	};
120179fddc4eSAl Viro 	size_t total = 0;
120279fddc4eSAl Viro 	int ret = 0;
120379fddc4eSAl Viro 	bool failed = false;
120479fddc4eSAl Viro 
120579fddc4eSAl Viro 	while (iov_iter_count(from) && !failed) {
120679fddc4eSAl Viro 		struct page *pages[16];
1207db85a9ebSAl Viro 		ssize_t copied;
1208db85a9ebSAl Viro 		size_t start;
120979fddc4eSAl Viro 		int n;
1210912d35f8SJens Axboe 
121179fddc4eSAl Viro 		copied = iov_iter_get_pages(from, pages, ~0UL, 16, &start);
121279fddc4eSAl Viro 		if (copied <= 0) {
121379fddc4eSAl Viro 			ret = copied;
121479fddc4eSAl Viro 			break;
121579fddc4eSAl Viro 		}
1216912d35f8SJens Axboe 
121779fddc4eSAl Viro 		for (n = 0; copied; n++, start = 0) {
1218db85a9ebSAl Viro 			int size = min_t(int, copied, PAGE_SIZE - start);
121979fddc4eSAl Viro 			if (!failed) {
122079fddc4eSAl Viro 				buf.page = pages[n];
122179fddc4eSAl Viro 				buf.offset = start;
122279fddc4eSAl Viro 				buf.len = size;
122379fddc4eSAl Viro 				ret = add_to_pipe(pipe, &buf);
122479fddc4eSAl Viro 				if (unlikely(ret < 0)) {
122579fddc4eSAl Viro 					failed = true;
122679fddc4eSAl Viro 				} else {
122779fddc4eSAl Viro 					iov_iter_advance(from, ret);
122879fddc4eSAl Viro 					total += ret;
122979fddc4eSAl Viro 				}
123079fddc4eSAl Viro 			} else {
123179fddc4eSAl Viro 				put_page(pages[n]);
123279fddc4eSAl Viro 			}
1233db85a9ebSAl Viro 			copied -= size;
1234912d35f8SJens Axboe 		}
1235912d35f8SJens Axboe 	}
123679fddc4eSAl Viro 	return total ? total : ret;
1237912d35f8SJens Axboe }
1238912d35f8SJens Axboe 
12396a14b90bSJens Axboe static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
12406a14b90bSJens Axboe 			struct splice_desc *sd)
12416a14b90bSJens Axboe {
12426130f531SAl Viro 	int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
12436130f531SAl Viro 	return n == sd->len ? n : -EFAULT;
12446a14b90bSJens Axboe }
12456a14b90bSJens Axboe 
12466a14b90bSJens Axboe /*
12476a14b90bSJens Axboe  * For lack of a better implementation, implement vmsplice() to userspace
12486a14b90bSJens Axboe  * as a simple copy of the pipes pages to the user iov.
12496a14b90bSJens Axboe  */
125087a3002aSAl Viro static long vmsplice_to_user(struct file *file, struct iov_iter *iter,
125187a3002aSAl Viro 			     unsigned int flags)
12526a14b90bSJens Axboe {
125387a3002aSAl Viro 	struct pipe_inode_info *pipe = get_pipe_info(file);
125487a3002aSAl Viro 	struct splice_desc sd = {
125587a3002aSAl Viro 		.total_len = iov_iter_count(iter),
125687a3002aSAl Viro 		.flags = flags,
125787a3002aSAl Viro 		.u.data = iter
125887a3002aSAl Viro 	};
125987a3002aSAl Viro 	long ret = 0;
12606a14b90bSJens Axboe 
12616a14b90bSJens Axboe 	if (!pipe)
12626a14b90bSJens Axboe 		return -EBADF;
12636a14b90bSJens Axboe 
1264345995faSAl Viro 	if (sd.total_len) {
12656130f531SAl Viro 		pipe_lock(pipe);
12666130f531SAl Viro 		ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
126761e0d47cSMiklos Szeredi 		pipe_unlock(pipe);
1268345995faSAl Viro 	}
12696a14b90bSJens Axboe 
12706a14b90bSJens Axboe 	return ret;
12716a14b90bSJens Axboe }
12726a14b90bSJens Axboe 
1273912d35f8SJens Axboe /*
1274912d35f8SJens Axboe  * vmsplice splices a user address range into a pipe. It can be thought of
1275912d35f8SJens Axboe  * as splice-from-memory, where the regular splice is splice-from-file (or
1276912d35f8SJens Axboe  * to file). In both cases the output is a pipe, naturally.
1277912d35f8SJens Axboe  */
127887a3002aSAl Viro static long vmsplice_to_pipe(struct file *file, struct iov_iter *iter,
127987a3002aSAl Viro 			     unsigned int flags)
1280912d35f8SJens Axboe {
1281ddac0d39SJens Axboe 	struct pipe_inode_info *pipe;
128287a3002aSAl Viro 	long ret = 0;
128379fddc4eSAl Viro 	unsigned buf_flag = 0;
128479fddc4eSAl Viro 
128579fddc4eSAl Viro 	if (flags & SPLICE_F_GIFT)
128679fddc4eSAl Viro 		buf_flag = PIPE_BUF_FLAG_GIFT;
1287912d35f8SJens Axboe 
128871993e62SLinus Torvalds 	pipe = get_pipe_info(file);
1289ddac0d39SJens Axboe 	if (!pipe)
1290912d35f8SJens Axboe 		return -EBADF;
1291912d35f8SJens Axboe 
12928924feffSAl Viro 	pipe_lock(pipe);
12938924feffSAl Viro 	ret = wait_for_space(pipe, flags);
129479fddc4eSAl Viro 	if (!ret)
129587a3002aSAl Viro 		ret = iter_to_pipe(iter, pipe, buf_flag);
12968924feffSAl Viro 	pipe_unlock(pipe);
12978924feffSAl Viro 	if (ret > 0)
12988924feffSAl Viro 		wakeup_pipe_readers(pipe);
129935f3d14dSJens Axboe 	return ret;
1300912d35f8SJens Axboe }
1301912d35f8SJens Axboe 
130287a3002aSAl Viro static int vmsplice_type(struct fd f, int *type)
130387a3002aSAl Viro {
130487a3002aSAl Viro 	if (!f.file)
130587a3002aSAl Viro 		return -EBADF;
130687a3002aSAl Viro 	if (f.file->f_mode & FMODE_WRITE) {
130787a3002aSAl Viro 		*type = WRITE;
130887a3002aSAl Viro 	} else if (f.file->f_mode & FMODE_READ) {
130987a3002aSAl Viro 		*type = READ;
131087a3002aSAl Viro 	} else {
131187a3002aSAl Viro 		fdput(f);
131287a3002aSAl Viro 		return -EBADF;
131387a3002aSAl Viro 	}
131487a3002aSAl Viro 	return 0;
131587a3002aSAl Viro }
131687a3002aSAl Viro 
13176a14b90bSJens Axboe /*
13186a14b90bSJens Axboe  * Note that vmsplice only really supports true splicing _from_ user memory
13196a14b90bSJens Axboe  * to a pipe, not the other way around. Splicing from user memory is a simple
13206a14b90bSJens Axboe  * operation that can be supported without any funky alignment restrictions
13216a14b90bSJens Axboe  * or nasty vm tricks. We simply map in the user memory and fill them into
13226a14b90bSJens Axboe  * a pipe. The reverse isn't quite as easy, though. There are two possible
13236a14b90bSJens Axboe  * solutions for that:
13246a14b90bSJens Axboe  *
13256a14b90bSJens Axboe  *	- memcpy() the data internally, at which point we might as well just
13266a14b90bSJens Axboe  *	  do a regular read() on the buffer anyway.
13276a14b90bSJens Axboe  *	- Lots of nasty vm tricks, that are neither fast nor flexible (it
13286a14b90bSJens Axboe  *	  has restriction limitations on both ends of the pipe).
13296a14b90bSJens Axboe  *
13306a14b90bSJens Axboe  * Currently we punt and implement it as a normal copy, see pipe_to_user().
13316a14b90bSJens Axboe  *
13326a14b90bSJens Axboe  */
133387a3002aSAl Viro static long do_vmsplice(struct file *f, struct iov_iter *iter, unsigned int flags)
1334912d35f8SJens Axboe {
13353d6ea290SAl Viro 	if (unlikely(flags & ~SPLICE_F_ALL))
13363d6ea290SAl Viro 		return -EINVAL;
133787a3002aSAl Viro 
133887a3002aSAl Viro 	if (!iov_iter_count(iter))
13396a14b90bSJens Axboe 		return 0;
13406a14b90bSJens Axboe 
134187a3002aSAl Viro 	if (iov_iter_rw(iter) == WRITE)
134287a3002aSAl Viro 		return vmsplice_to_pipe(f, iter, flags);
134387a3002aSAl Viro 	else
134487a3002aSAl Viro 		return vmsplice_to_user(f, iter, flags);
1345912d35f8SJens Axboe }
1346912d35f8SJens Axboe 
134787a3002aSAl Viro SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, uiov,
134830cfe4efSDominik Brodowski 		unsigned long, nr_segs, unsigned int, flags)
134930cfe4efSDominik Brodowski {
135087a3002aSAl Viro 	struct iovec iovstack[UIO_FASTIOV];
135187a3002aSAl Viro 	struct iovec *iov = iovstack;
135287a3002aSAl Viro 	struct iov_iter iter;
135387a3002aSAl Viro 	long error;
135487a3002aSAl Viro 	struct fd f;
135587a3002aSAl Viro 	int type;
135687a3002aSAl Viro 
135787a3002aSAl Viro 	f = fdget(fd);
135887a3002aSAl Viro 	error = vmsplice_type(f, &type);
135987a3002aSAl Viro 	if (error)
136087a3002aSAl Viro 		return error;
136187a3002aSAl Viro 
136287a3002aSAl Viro 	error = import_iovec(type, uiov, nr_segs,
136387a3002aSAl Viro 			     ARRAY_SIZE(iovstack), &iov, &iter);
136487a3002aSAl Viro 	if (!error) {
136587a3002aSAl Viro 		error = do_vmsplice(f.file, &iter, flags);
136687a3002aSAl Viro 		kfree(iov);
136787a3002aSAl Viro 	}
136887a3002aSAl Viro 	fdput(f);
136987a3002aSAl Viro 	return error;
137030cfe4efSDominik Brodowski }
137130cfe4efSDominik Brodowski 
137276b021d0SAl Viro #ifdef CONFIG_COMPAT
137376b021d0SAl Viro COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
137476b021d0SAl Viro 		    unsigned int, nr_segs, unsigned int, flags)
137576b021d0SAl Viro {
137687a3002aSAl Viro 	struct iovec iovstack[UIO_FASTIOV];
137787a3002aSAl Viro 	struct iovec *iov = iovstack;
137887a3002aSAl Viro 	struct iov_iter iter;
137987a3002aSAl Viro 	long error;
138087a3002aSAl Viro 	struct fd f;
138187a3002aSAl Viro 	int type;
138287a3002aSAl Viro 
138387a3002aSAl Viro 	f = fdget(fd);
138487a3002aSAl Viro 	error = vmsplice_type(f, &type);
138587a3002aSAl Viro 	if (error)
138687a3002aSAl Viro 		return error;
138787a3002aSAl Viro 
138887a3002aSAl Viro 	error = compat_import_iovec(type, iov32, nr_segs,
138987a3002aSAl Viro 			     ARRAY_SIZE(iovstack), &iov, &iter);
139087a3002aSAl Viro 	if (!error) {
139187a3002aSAl Viro 		error = do_vmsplice(f.file, &iter, flags);
139287a3002aSAl Viro 		kfree(iov);
139376b021d0SAl Viro 	}
139487a3002aSAl Viro 	fdput(f);
139587a3002aSAl Viro 	return error;
139676b021d0SAl Viro }
139776b021d0SAl Viro #endif
139876b021d0SAl Viro 
1399836f92adSHeiko Carstens SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1400836f92adSHeiko Carstens 		int, fd_out, loff_t __user *, off_out,
1401836f92adSHeiko Carstens 		size_t, len, unsigned int, flags)
14025274f052SJens Axboe {
14032903ff01SAl Viro 	struct fd in, out;
14045274f052SJens Axboe 	long error;
14055274f052SJens Axboe 
14065274f052SJens Axboe 	if (unlikely(!len))
14075274f052SJens Axboe 		return 0;
14085274f052SJens Axboe 
14093d6ea290SAl Viro 	if (unlikely(flags & ~SPLICE_F_ALL))
14103d6ea290SAl Viro 		return -EINVAL;
14113d6ea290SAl Viro 
14125274f052SJens Axboe 	error = -EBADF;
14132903ff01SAl Viro 	in = fdget(fd_in);
14142903ff01SAl Viro 	if (in.file) {
14152903ff01SAl Viro 		if (in.file->f_mode & FMODE_READ) {
14162903ff01SAl Viro 			out = fdget(fd_out);
14172903ff01SAl Viro 			if (out.file) {
14182903ff01SAl Viro 				if (out.file->f_mode & FMODE_WRITE)
14192903ff01SAl Viro 					error = do_splice(in.file, off_in,
14202903ff01SAl Viro 							  out.file, off_out,
1421529565dcSIngo Molnar 							  len, flags);
14222903ff01SAl Viro 				fdput(out);
14235274f052SJens Axboe 			}
14245274f052SJens Axboe 		}
14252903ff01SAl Viro 		fdput(in);
14265274f052SJens Axboe 	}
14275274f052SJens Axboe 	return error;
14285274f052SJens Axboe }
142970524490SJens Axboe 
143070524490SJens Axboe /*
1431aadd06e5SJens Axboe  * Make sure there's data to read. Wait for input if we can, otherwise
1432aadd06e5SJens Axboe  * return an appropriate error.
1433aadd06e5SJens Axboe  */
14347c77f0b3SMiklos Szeredi static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1435aadd06e5SJens Axboe {
1436aadd06e5SJens Axboe 	int ret;
1437aadd06e5SJens Axboe 
1438aadd06e5SJens Axboe 	/*
1439aadd06e5SJens Axboe 	 * Check ->nrbufs without the inode lock first. This function
1440aadd06e5SJens Axboe 	 * is speculative anyways, so missing one is ok.
1441aadd06e5SJens Axboe 	 */
1442aadd06e5SJens Axboe 	if (pipe->nrbufs)
1443aadd06e5SJens Axboe 		return 0;
1444aadd06e5SJens Axboe 
1445aadd06e5SJens Axboe 	ret = 0;
144661e0d47cSMiklos Szeredi 	pipe_lock(pipe);
1447aadd06e5SJens Axboe 
1448aadd06e5SJens Axboe 	while (!pipe->nrbufs) {
1449aadd06e5SJens Axboe 		if (signal_pending(current)) {
1450aadd06e5SJens Axboe 			ret = -ERESTARTSYS;
1451aadd06e5SJens Axboe 			break;
1452aadd06e5SJens Axboe 		}
1453aadd06e5SJens Axboe 		if (!pipe->writers)
1454aadd06e5SJens Axboe 			break;
1455aadd06e5SJens Axboe 		if (!pipe->waiting_writers) {
1456aadd06e5SJens Axboe 			if (flags & SPLICE_F_NONBLOCK) {
1457aadd06e5SJens Axboe 				ret = -EAGAIN;
1458aadd06e5SJens Axboe 				break;
1459aadd06e5SJens Axboe 			}
1460aadd06e5SJens Axboe 		}
1461aadd06e5SJens Axboe 		pipe_wait(pipe);
1462aadd06e5SJens Axboe 	}
1463aadd06e5SJens Axboe 
146461e0d47cSMiklos Szeredi 	pipe_unlock(pipe);
1465aadd06e5SJens Axboe 	return ret;
1466aadd06e5SJens Axboe }
1467aadd06e5SJens Axboe 
1468aadd06e5SJens Axboe /*
1469aadd06e5SJens Axboe  * Make sure there's writeable room. Wait for room if we can, otherwise
1470aadd06e5SJens Axboe  * return an appropriate error.
1471aadd06e5SJens Axboe  */
14727c77f0b3SMiklos Szeredi static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1473aadd06e5SJens Axboe {
1474aadd06e5SJens Axboe 	int ret;
1475aadd06e5SJens Axboe 
1476aadd06e5SJens Axboe 	/*
1477aadd06e5SJens Axboe 	 * Check ->nrbufs without the inode lock first. This function
1478aadd06e5SJens Axboe 	 * is speculative anyways, so missing one is ok.
1479aadd06e5SJens Axboe 	 */
148035f3d14dSJens Axboe 	if (pipe->nrbufs < pipe->buffers)
1481aadd06e5SJens Axboe 		return 0;
1482aadd06e5SJens Axboe 
1483aadd06e5SJens Axboe 	ret = 0;
148461e0d47cSMiklos Szeredi 	pipe_lock(pipe);
1485aadd06e5SJens Axboe 
148635f3d14dSJens Axboe 	while (pipe->nrbufs >= pipe->buffers) {
1487aadd06e5SJens Axboe 		if (!pipe->readers) {
1488aadd06e5SJens Axboe 			send_sig(SIGPIPE, current, 0);
1489aadd06e5SJens Axboe 			ret = -EPIPE;
1490aadd06e5SJens Axboe 			break;
1491aadd06e5SJens Axboe 		}
1492aadd06e5SJens Axboe 		if (flags & SPLICE_F_NONBLOCK) {
1493aadd06e5SJens Axboe 			ret = -EAGAIN;
1494aadd06e5SJens Axboe 			break;
1495aadd06e5SJens Axboe 		}
1496aadd06e5SJens Axboe 		if (signal_pending(current)) {
1497aadd06e5SJens Axboe 			ret = -ERESTARTSYS;
1498aadd06e5SJens Axboe 			break;
1499aadd06e5SJens Axboe 		}
1500aadd06e5SJens Axboe 		pipe->waiting_writers++;
1501aadd06e5SJens Axboe 		pipe_wait(pipe);
1502aadd06e5SJens Axboe 		pipe->waiting_writers--;
1503aadd06e5SJens Axboe 	}
1504aadd06e5SJens Axboe 
150561e0d47cSMiklos Szeredi 	pipe_unlock(pipe);
1506aadd06e5SJens Axboe 	return ret;
1507aadd06e5SJens Axboe }
1508aadd06e5SJens Axboe 
1509aadd06e5SJens Axboe /*
15107c77f0b3SMiklos Szeredi  * Splice contents of ipipe to opipe.
15117c77f0b3SMiklos Szeredi  */
15127c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
15137c77f0b3SMiklos Szeredi 			       struct pipe_inode_info *opipe,
15147c77f0b3SMiklos Szeredi 			       size_t len, unsigned int flags)
15157c77f0b3SMiklos Szeredi {
15167c77f0b3SMiklos Szeredi 	struct pipe_buffer *ibuf, *obuf;
15177c77f0b3SMiklos Szeredi 	int ret = 0, nbuf;
15187c77f0b3SMiklos Szeredi 	bool input_wakeup = false;
15197c77f0b3SMiklos Szeredi 
15207c77f0b3SMiklos Szeredi 
15217c77f0b3SMiklos Szeredi retry:
15227c77f0b3SMiklos Szeredi 	ret = ipipe_prep(ipipe, flags);
15237c77f0b3SMiklos Szeredi 	if (ret)
15247c77f0b3SMiklos Szeredi 		return ret;
15257c77f0b3SMiklos Szeredi 
15267c77f0b3SMiklos Szeredi 	ret = opipe_prep(opipe, flags);
15277c77f0b3SMiklos Szeredi 	if (ret)
15287c77f0b3SMiklos Szeredi 		return ret;
15297c77f0b3SMiklos Szeredi 
15307c77f0b3SMiklos Szeredi 	/*
15317c77f0b3SMiklos Szeredi 	 * Potential ABBA deadlock, work around it by ordering lock
15327c77f0b3SMiklos Szeredi 	 * grabbing by pipe info address. Otherwise two different processes
15337c77f0b3SMiklos Szeredi 	 * could deadlock (one doing tee from A -> B, the other from B -> A).
15347c77f0b3SMiklos Szeredi 	 */
15357c77f0b3SMiklos Szeredi 	pipe_double_lock(ipipe, opipe);
15367c77f0b3SMiklos Szeredi 
15377c77f0b3SMiklos Szeredi 	do {
15387c77f0b3SMiklos Szeredi 		if (!opipe->readers) {
15397c77f0b3SMiklos Szeredi 			send_sig(SIGPIPE, current, 0);
15407c77f0b3SMiklos Szeredi 			if (!ret)
15417c77f0b3SMiklos Szeredi 				ret = -EPIPE;
15427c77f0b3SMiklos Szeredi 			break;
15437c77f0b3SMiklos Szeredi 		}
15447c77f0b3SMiklos Szeredi 
15457c77f0b3SMiklos Szeredi 		if (!ipipe->nrbufs && !ipipe->writers)
15467c77f0b3SMiklos Szeredi 			break;
15477c77f0b3SMiklos Szeredi 
15487c77f0b3SMiklos Szeredi 		/*
15497c77f0b3SMiklos Szeredi 		 * Cannot make any progress, because either the input
15507c77f0b3SMiklos Szeredi 		 * pipe is empty or the output pipe is full.
15517c77f0b3SMiklos Szeredi 		 */
155235f3d14dSJens Axboe 		if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
15537c77f0b3SMiklos Szeredi 			/* Already processed some buffers, break */
15547c77f0b3SMiklos Szeredi 			if (ret)
15557c77f0b3SMiklos Szeredi 				break;
15567c77f0b3SMiklos Szeredi 
15577c77f0b3SMiklos Szeredi 			if (flags & SPLICE_F_NONBLOCK) {
15587c77f0b3SMiklos Szeredi 				ret = -EAGAIN;
15597c77f0b3SMiklos Szeredi 				break;
15607c77f0b3SMiklos Szeredi 			}
15617c77f0b3SMiklos Szeredi 
15627c77f0b3SMiklos Szeredi 			/*
15637c77f0b3SMiklos Szeredi 			 * We raced with another reader/writer and haven't
15647c77f0b3SMiklos Szeredi 			 * managed to process any buffers.  A zero return
15657c77f0b3SMiklos Szeredi 			 * value means EOF, so retry instead.
15667c77f0b3SMiklos Szeredi 			 */
15677c77f0b3SMiklos Szeredi 			pipe_unlock(ipipe);
15687c77f0b3SMiklos Szeredi 			pipe_unlock(opipe);
15697c77f0b3SMiklos Szeredi 			goto retry;
15707c77f0b3SMiklos Szeredi 		}
15717c77f0b3SMiklos Szeredi 
15727c77f0b3SMiklos Szeredi 		ibuf = ipipe->bufs + ipipe->curbuf;
157335f3d14dSJens Axboe 		nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
15747c77f0b3SMiklos Szeredi 		obuf = opipe->bufs + nbuf;
15757c77f0b3SMiklos Szeredi 
15767c77f0b3SMiklos Szeredi 		if (len >= ibuf->len) {
15777c77f0b3SMiklos Szeredi 			/*
15787c77f0b3SMiklos Szeredi 			 * Simply move the whole buffer from ipipe to opipe
15797c77f0b3SMiklos Szeredi 			 */
15807c77f0b3SMiklos Szeredi 			*obuf = *ibuf;
15817c77f0b3SMiklos Szeredi 			ibuf->ops = NULL;
15827c77f0b3SMiklos Szeredi 			opipe->nrbufs++;
158335f3d14dSJens Axboe 			ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
15847c77f0b3SMiklos Szeredi 			ipipe->nrbufs--;
15857c77f0b3SMiklos Szeredi 			input_wakeup = true;
15867c77f0b3SMiklos Szeredi 		} else {
15877c77f0b3SMiklos Szeredi 			/*
15887c77f0b3SMiklos Szeredi 			 * Get a reference to this pipe buffer,
15897c77f0b3SMiklos Szeredi 			 * so we can copy the contents over.
15907c77f0b3SMiklos Szeredi 			 */
15917bf2d1dfSMiklos Szeredi 			pipe_buf_get(ipipe, ibuf);
15927c77f0b3SMiklos Szeredi 			*obuf = *ibuf;
15937c77f0b3SMiklos Szeredi 
15947c77f0b3SMiklos Szeredi 			/*
15957c77f0b3SMiklos Szeredi 			 * Don't inherit the gift flag, we need to
15967c77f0b3SMiklos Szeredi 			 * prevent multiple steals of this page.
15977c77f0b3SMiklos Szeredi 			 */
15987c77f0b3SMiklos Szeredi 			obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
15997c77f0b3SMiklos Szeredi 
1600a0ce2f0aSJann Horn 			pipe_buf_mark_unmergeable(obuf);
1601a0ce2f0aSJann Horn 
16027c77f0b3SMiklos Szeredi 			obuf->len = len;
16037c77f0b3SMiklos Szeredi 			opipe->nrbufs++;
16047c77f0b3SMiklos Szeredi 			ibuf->offset += obuf->len;
16057c77f0b3SMiklos Szeredi 			ibuf->len -= obuf->len;
16067c77f0b3SMiklos Szeredi 		}
16077c77f0b3SMiklos Szeredi 		ret += obuf->len;
16087c77f0b3SMiklos Szeredi 		len -= obuf->len;
16097c77f0b3SMiklos Szeredi 	} while (len);
16107c77f0b3SMiklos Szeredi 
16117c77f0b3SMiklos Szeredi 	pipe_unlock(ipipe);
16127c77f0b3SMiklos Szeredi 	pipe_unlock(opipe);
16137c77f0b3SMiklos Szeredi 
16147c77f0b3SMiklos Szeredi 	/*
16157c77f0b3SMiklos Szeredi 	 * If we put data in the output pipe, wakeup any potential readers.
16167c77f0b3SMiklos Szeredi 	 */
1617825cdcb1SNamhyung Kim 	if (ret > 0)
1618825cdcb1SNamhyung Kim 		wakeup_pipe_readers(opipe);
1619825cdcb1SNamhyung Kim 
16207c77f0b3SMiklos Szeredi 	if (input_wakeup)
16217c77f0b3SMiklos Szeredi 		wakeup_pipe_writers(ipipe);
16227c77f0b3SMiklos Szeredi 
16237c77f0b3SMiklos Szeredi 	return ret;
16247c77f0b3SMiklos Szeredi }
16257c77f0b3SMiklos Szeredi 
16267c77f0b3SMiklos Szeredi /*
162770524490SJens Axboe  * Link contents of ipipe to opipe.
162870524490SJens Axboe  */
162970524490SJens Axboe static int link_pipe(struct pipe_inode_info *ipipe,
163070524490SJens Axboe 		     struct pipe_inode_info *opipe,
163170524490SJens Axboe 		     size_t len, unsigned int flags)
163270524490SJens Axboe {
163370524490SJens Axboe 	struct pipe_buffer *ibuf, *obuf;
1634aadd06e5SJens Axboe 	int ret = 0, i = 0, nbuf;
163570524490SJens Axboe 
163670524490SJens Axboe 	/*
163770524490SJens Axboe 	 * Potential ABBA deadlock, work around it by ordering lock
163861e0d47cSMiklos Szeredi 	 * grabbing by pipe info address. Otherwise two different processes
163970524490SJens Axboe 	 * could deadlock (one doing tee from A -> B, the other from B -> A).
164070524490SJens Axboe 	 */
164161e0d47cSMiklos Szeredi 	pipe_double_lock(ipipe, opipe);
164270524490SJens Axboe 
1643aadd06e5SJens Axboe 	do {
164470524490SJens Axboe 		if (!opipe->readers) {
164570524490SJens Axboe 			send_sig(SIGPIPE, current, 0);
164670524490SJens Axboe 			if (!ret)
164770524490SJens Axboe 				ret = -EPIPE;
164870524490SJens Axboe 			break;
164970524490SJens Axboe 		}
165070524490SJens Axboe 
165170524490SJens Axboe 		/*
1652aadd06e5SJens Axboe 		 * If we have iterated all input buffers or ran out of
1653aadd06e5SJens Axboe 		 * output room, break.
165470524490SJens Axboe 		 */
165535f3d14dSJens Axboe 		if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
1656aadd06e5SJens Axboe 			break;
1657aadd06e5SJens Axboe 
165835f3d14dSJens Axboe 		ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
165935f3d14dSJens Axboe 		nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
166070524490SJens Axboe 
166170524490SJens Axboe 		/*
166270524490SJens Axboe 		 * Get a reference to this pipe buffer,
166370524490SJens Axboe 		 * so we can copy the contents over.
166470524490SJens Axboe 		 */
16657bf2d1dfSMiklos Szeredi 		pipe_buf_get(ipipe, ibuf);
166670524490SJens Axboe 
166770524490SJens Axboe 		obuf = opipe->bufs + nbuf;
166870524490SJens Axboe 		*obuf = *ibuf;
166970524490SJens Axboe 
16707afa6fd0SJens Axboe 		/*
16717afa6fd0SJens Axboe 		 * Don't inherit the gift flag, we need to
16727afa6fd0SJens Axboe 		 * prevent multiple steals of this page.
16737afa6fd0SJens Axboe 		 */
16747afa6fd0SJens Axboe 		obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
16757afa6fd0SJens Axboe 
1676a0ce2f0aSJann Horn 		pipe_buf_mark_unmergeable(obuf);
1677a0ce2f0aSJann Horn 
167870524490SJens Axboe 		if (obuf->len > len)
167970524490SJens Axboe 			obuf->len = len;
168070524490SJens Axboe 
168170524490SJens Axboe 		opipe->nrbufs++;
168270524490SJens Axboe 		ret += obuf->len;
168370524490SJens Axboe 		len -= obuf->len;
1684aadd06e5SJens Axboe 		i++;
1685aadd06e5SJens Axboe 	} while (len);
168670524490SJens Axboe 
168702cf01aeSJens Axboe 	/*
168802cf01aeSJens Axboe 	 * return EAGAIN if we have the potential of some data in the
168902cf01aeSJens Axboe 	 * future, otherwise just return 0
169002cf01aeSJens Axboe 	 */
169102cf01aeSJens Axboe 	if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
169202cf01aeSJens Axboe 		ret = -EAGAIN;
169302cf01aeSJens Axboe 
169461e0d47cSMiklos Szeredi 	pipe_unlock(ipipe);
169561e0d47cSMiklos Szeredi 	pipe_unlock(opipe);
169670524490SJens Axboe 
1697aadd06e5SJens Axboe 	/*
1698aadd06e5SJens Axboe 	 * If we put data in the output pipe, wakeup any potential readers.
1699aadd06e5SJens Axboe 	 */
1700825cdcb1SNamhyung Kim 	if (ret > 0)
1701825cdcb1SNamhyung Kim 		wakeup_pipe_readers(opipe);
170270524490SJens Axboe 
170370524490SJens Axboe 	return ret;
170470524490SJens Axboe }
170570524490SJens Axboe 
170670524490SJens Axboe /*
170770524490SJens Axboe  * This is a tee(1) implementation that works on pipes. It doesn't copy
170870524490SJens Axboe  * any data, it simply references the 'in' pages on the 'out' pipe.
170970524490SJens Axboe  * The 'flags' used are the SPLICE_F_* variants, currently the only
171070524490SJens Axboe  * applicable one is SPLICE_F_NONBLOCK.
171170524490SJens Axboe  */
171270524490SJens Axboe static long do_tee(struct file *in, struct file *out, size_t len,
171370524490SJens Axboe 		   unsigned int flags)
171470524490SJens Axboe {
171571993e62SLinus Torvalds 	struct pipe_inode_info *ipipe = get_pipe_info(in);
171671993e62SLinus Torvalds 	struct pipe_inode_info *opipe = get_pipe_info(out);
1717aadd06e5SJens Axboe 	int ret = -EINVAL;
171870524490SJens Axboe 
171970524490SJens Axboe 	/*
1720aadd06e5SJens Axboe 	 * Duplicate the contents of ipipe to opipe without actually
1721aadd06e5SJens Axboe 	 * copying the data.
172270524490SJens Axboe 	 */
1723aadd06e5SJens Axboe 	if (ipipe && opipe && ipipe != opipe) {
1724aadd06e5SJens Axboe 		/*
1725aadd06e5SJens Axboe 		 * Keep going, unless we encounter an error. The ipipe/opipe
1726aadd06e5SJens Axboe 		 * ordering doesn't really matter.
1727aadd06e5SJens Axboe 		 */
17287c77f0b3SMiklos Szeredi 		ret = ipipe_prep(ipipe, flags);
1729aadd06e5SJens Axboe 		if (!ret) {
17307c77f0b3SMiklos Szeredi 			ret = opipe_prep(opipe, flags);
173102cf01aeSJens Axboe 			if (!ret)
1732aadd06e5SJens Axboe 				ret = link_pipe(ipipe, opipe, len, flags);
1733aadd06e5SJens Axboe 		}
1734aadd06e5SJens Axboe 	}
173570524490SJens Axboe 
1736aadd06e5SJens Axboe 	return ret;
173770524490SJens Axboe }
173870524490SJens Axboe 
1739836f92adSHeiko Carstens SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
174070524490SJens Axboe {
17412903ff01SAl Viro 	struct fd in;
17422903ff01SAl Viro 	int error;
174370524490SJens Axboe 
17443d6ea290SAl Viro 	if (unlikely(flags & ~SPLICE_F_ALL))
17453d6ea290SAl Viro 		return -EINVAL;
17463d6ea290SAl Viro 
174770524490SJens Axboe 	if (unlikely(!len))
174870524490SJens Axboe 		return 0;
174970524490SJens Axboe 
175070524490SJens Axboe 	error = -EBADF;
17512903ff01SAl Viro 	in = fdget(fdin);
17522903ff01SAl Viro 	if (in.file) {
17532903ff01SAl Viro 		if (in.file->f_mode & FMODE_READ) {
17542903ff01SAl Viro 			struct fd out = fdget(fdout);
17552903ff01SAl Viro 			if (out.file) {
17562903ff01SAl Viro 				if (out.file->f_mode & FMODE_WRITE)
17572903ff01SAl Viro 					error = do_tee(in.file, out.file,
17582903ff01SAl Viro 							len, flags);
17592903ff01SAl Viro 				fdput(out);
176070524490SJens Axboe 			}
176170524490SJens Axboe 		}
17622903ff01SAl Viro  		fdput(in);
176370524490SJens Axboe  	}
176470524490SJens Axboe 
176570524490SJens Axboe 	return error;
176670524490SJens Axboe }
1767