1 // SPDX-License-Identifier: GPL-2.0-only
3 * "splice": joining two ropes together by interweaving their strands.
5 * This is the "extended pipe" functionality, where a pipe is used as
6 * an arbitrary in-memory buffer. Think of a pipe as a small kernel
7 * buffer that you can use to transfer data from one end to the other.
9 * The traditional unix read/write is extended with a "splice()" operation
10 * that transfers data buffers to or from a pipe buffer.
12 * Named by Larry McVoy, original implementation from Linus, extended by
13 * Jens to support splicing to files, network, direct splicing, etc and
14 * fixing lots of bugs.
16 * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk>
17 * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
18 * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
21 #include <linux/bvec.h>
23 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/splice.h>
26 #include <linux/memcontrol.h>
27 #include <linux/mm_inline.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/export.h>
31 #include <linux/syscalls.h>
32 #include <linux/uio.h>
33 #include <linux/security.h>
34 #include <linux/gfp.h>
35 #include <linux/socket.h>
36 #include <linux/compat.h>
37 #include <linux/sched/signal.h>
42 * Attempt to steal a page from a pipe buffer. This should perhaps go into
43 * a vm helper function, it's already simplified quite a bit by the
44 * addition of remove_mapping(). If success is returned, the caller may
45 * attempt to reuse this page for another destination.
47 static bool page_cache_pipe_buf_try_steal(struct pipe_inode_info
*pipe
,
48 struct pipe_buffer
*buf
)
50 struct page
*page
= buf
->page
;
51 struct address_space
*mapping
;
55 mapping
= page_mapping(page
);
57 WARN_ON(!PageUptodate(page
));
60 * At least for ext2 with nobh option, we need to wait on
61 * writeback completing on this page, since we'll remove it
62 * from the pagecache. Otherwise truncate wont wait on the
63 * page, allowing the disk blocks to be reused by someone else
64 * before we actually wrote our data to them. fs corruption
67 wait_on_page_writeback(page
);
69 if (page_has_private(page
) &&
70 !try_to_release_page(page
, GFP_KERNEL
))
74 * If we succeeded in removing the mapping, set LRU flag
77 if (remove_mapping(mapping
, page
)) {
78 buf
->flags
|= PIPE_BUF_FLAG_LRU
;
84 * Raced with truncate or failed to remove page from current
85 * address space, unlock and return failure.
92 static void page_cache_pipe_buf_release(struct pipe_inode_info
*pipe
,
93 struct pipe_buffer
*buf
)
96 buf
->flags
&= ~PIPE_BUF_FLAG_LRU
;
100 * Check whether the contents of buf is OK to access. Since the content
101 * is a page cache page, IO may be in flight.
103 static int page_cache_pipe_buf_confirm(struct pipe_inode_info
*pipe
,
104 struct pipe_buffer
*buf
)
106 struct page
*page
= buf
->page
;
109 if (!PageUptodate(page
)) {
113 * Page got truncated/unhashed. This will cause a 0-byte
114 * splice, if this is the first page.
116 if (!page
->mapping
) {
122 * Uh oh, read-error from disk.
124 if (!PageUptodate(page
)) {
130 * Page is ok afterall, we are done.
141 const struct pipe_buf_operations page_cache_pipe_buf_ops
= {
142 .confirm
= page_cache_pipe_buf_confirm
,
143 .release
= page_cache_pipe_buf_release
,
144 .try_steal
= page_cache_pipe_buf_try_steal
,
145 .get
= generic_pipe_buf_get
,
148 static bool user_page_pipe_buf_try_steal(struct pipe_inode_info
*pipe
,
149 struct pipe_buffer
*buf
)
151 if (!(buf
->flags
& PIPE_BUF_FLAG_GIFT
))
154 buf
->flags
|= PIPE_BUF_FLAG_LRU
;
155 return generic_pipe_buf_try_steal(pipe
, buf
);
158 static const struct pipe_buf_operations user_page_pipe_buf_ops
= {
159 .release
= page_cache_pipe_buf_release
,
160 .try_steal
= user_page_pipe_buf_try_steal
,
161 .get
= generic_pipe_buf_get
,
164 static void wakeup_pipe_readers(struct pipe_inode_info
*pipe
)
167 if (waitqueue_active(&pipe
->rd_wait
))
168 wake_up_interruptible(&pipe
->rd_wait
);
169 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
173 * splice_to_pipe - fill passed data into a pipe
174 * @pipe: pipe to fill
178 * @spd contains a map of pages and len/offset tuples, along with
179 * the struct pipe_buf_operations associated with these pages. This
180 * function will link that data to the pipe.
183 ssize_t
splice_to_pipe(struct pipe_inode_info
*pipe
,
184 struct splice_pipe_desc
*spd
)
186 unsigned int spd_pages
= spd
->nr_pages
;
187 unsigned int tail
= pipe
->tail
;
188 unsigned int head
= pipe
->head
;
189 unsigned int mask
= pipe
->ring_size
- 1;
190 int ret
= 0, page_nr
= 0;
195 if (unlikely(!pipe
->readers
)) {
196 send_sig(SIGPIPE
, current
, 0);
201 while (!pipe_full(head
, tail
, pipe
->max_usage
)) {
202 struct pipe_buffer
*buf
= &pipe
->bufs
[head
& mask
];
204 buf
->page
= spd
->pages
[page_nr
];
205 buf
->offset
= spd
->partial
[page_nr
].offset
;
206 buf
->len
= spd
->partial
[page_nr
].len
;
207 buf
->private = spd
->partial
[page_nr
].private;
216 if (!--spd
->nr_pages
)
224 while (page_nr
< spd_pages
)
225 spd
->spd_release(spd
, page_nr
++);
229 EXPORT_SYMBOL_GPL(splice_to_pipe
);
231 ssize_t
add_to_pipe(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
)
233 unsigned int head
= pipe
->head
;
234 unsigned int tail
= pipe
->tail
;
235 unsigned int mask
= pipe
->ring_size
- 1;
238 if (unlikely(!pipe
->readers
)) {
239 send_sig(SIGPIPE
, current
, 0);
241 } else if (pipe_full(head
, tail
, pipe
->max_usage
)) {
244 pipe
->bufs
[head
& mask
] = *buf
;
245 pipe
->head
= head
+ 1;
248 pipe_buf_release(pipe
, buf
);
251 EXPORT_SYMBOL(add_to_pipe
);
254 * Check if we need to grow the arrays holding pages and partial page
257 int splice_grow_spd(const struct pipe_inode_info
*pipe
, struct splice_pipe_desc
*spd
)
259 unsigned int max_usage
= READ_ONCE(pipe
->max_usage
);
261 spd
->nr_pages_max
= max_usage
;
262 if (max_usage
<= PIPE_DEF_BUFFERS
)
265 spd
->pages
= kmalloc_array(max_usage
, sizeof(struct page
*), GFP_KERNEL
);
266 spd
->partial
= kmalloc_array(max_usage
, sizeof(struct partial_page
),
269 if (spd
->pages
&& spd
->partial
)
277 void splice_shrink_spd(struct splice_pipe_desc
*spd
)
279 if (spd
->nr_pages_max
<= PIPE_DEF_BUFFERS
)
287 * generic_file_splice_read - splice data from file to a pipe
288 * @in: file to splice from
289 * @ppos: position in @in
290 * @pipe: pipe to splice to
291 * @len: number of bytes to splice
292 * @flags: splice modifier flags
295 * Will read pages from given file and fill them into a pipe. Can be
296 * used as long as it has more or less sane ->read_iter().
299 ssize_t
generic_file_splice_read(struct file
*in
, loff_t
*ppos
,
300 struct pipe_inode_info
*pipe
, size_t len
,
308 iov_iter_pipe(&to
, READ
, pipe
, len
);
310 init_sync_kiocb(&kiocb
, in
);
311 kiocb
.ki_pos
= *ppos
;
312 ret
= call_read_iter(in
, &kiocb
, &to
);
314 *ppos
= kiocb
.ki_pos
;
316 } else if (ret
< 0) {
319 iov_iter_advance(&to
, 0); /* to free what was emitted */
321 * callers of ->splice_read() expect -EAGAIN on
322 * "can't put anything in there", rather than -EFAULT.
330 EXPORT_SYMBOL(generic_file_splice_read
);
332 const struct pipe_buf_operations default_pipe_buf_ops
= {
333 .release
= generic_pipe_buf_release
,
334 .try_steal
= generic_pipe_buf_try_steal
,
335 .get
= generic_pipe_buf_get
,
338 /* Pipe buffer operations for a socket and similar. */
339 const struct pipe_buf_operations nosteal_pipe_buf_ops
= {
340 .release
= generic_pipe_buf_release
,
341 .get
= generic_pipe_buf_get
,
343 EXPORT_SYMBOL(nosteal_pipe_buf_ops
);
345 static ssize_t
kernel_readv(struct file
*file
, const struct kvec
*vec
,
346 unsigned long vlen
, loff_t offset
)
354 /* The cast to a user pointer is valid due to the set_fs() */
355 res
= vfs_readv(file
, (const struct iovec __user
*)vec
, vlen
, &pos
, 0);
361 static ssize_t
default_file_splice_read(struct file
*in
, loff_t
*ppos
,
362 struct pipe_inode_info
*pipe
, size_t len
,
365 struct kvec
*vec
, __vec
[PIPE_DEF_BUFFERS
];
368 unsigned int nr_pages
;
370 size_t offset
, base
, copied
= 0;
374 if (pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
))
378 * Try to keep page boundaries matching to source pagecache ones -
379 * it probably won't be much help, but...
381 offset
= *ppos
& ~PAGE_MASK
;
383 iov_iter_pipe(&to
, READ
, pipe
, len
+ offset
);
385 res
= iov_iter_get_pages_alloc(&to
, &pages
, len
+ offset
, &base
);
389 nr_pages
= DIV_ROUND_UP(res
+ base
, PAGE_SIZE
);
392 if (nr_pages
> PIPE_DEF_BUFFERS
) {
393 vec
= kmalloc_array(nr_pages
, sizeof(struct kvec
), GFP_KERNEL
);
394 if (unlikely(!vec
)) {
400 mask
= pipe
->ring_size
- 1;
401 pipe
->bufs
[to
.head
& mask
].offset
= offset
;
402 pipe
->bufs
[to
.head
& mask
].len
-= offset
;
404 for (i
= 0; i
< nr_pages
; i
++) {
405 size_t this_len
= min_t(size_t, len
, PAGE_SIZE
- offset
);
406 vec
[i
].iov_base
= page_address(pages
[i
]) + offset
;
407 vec
[i
].iov_len
= this_len
;
412 res
= kernel_readv(in
, vec
, nr_pages
, *ppos
);
421 for (i
= 0; i
< nr_pages
; i
++)
424 iov_iter_advance(&to
, copied
); /* truncates and discards */
429 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
430 * using sendpage(). Return the number of bytes sent.
432 static int pipe_to_sendpage(struct pipe_inode_info
*pipe
,
433 struct pipe_buffer
*buf
, struct splice_desc
*sd
)
435 struct file
*file
= sd
->u
.file
;
436 loff_t pos
= sd
->pos
;
439 if (!likely(file
->f_op
->sendpage
))
442 more
= (sd
->flags
& SPLICE_F_MORE
) ? MSG_MORE
: 0;
444 if (sd
->len
< sd
->total_len
&&
445 pipe_occupancy(pipe
->head
, pipe
->tail
) > 1)
446 more
|= MSG_SENDPAGE_NOTLAST
;
448 return file
->f_op
->sendpage(file
, buf
->page
, buf
->offset
,
449 sd
->len
, &pos
, more
);
452 static void wakeup_pipe_writers(struct pipe_inode_info
*pipe
)
455 if (waitqueue_active(&pipe
->wr_wait
))
456 wake_up_interruptible(&pipe
->wr_wait
);
457 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
461 * splice_from_pipe_feed - feed available data from a pipe to a file
462 * @pipe: pipe to splice from
463 * @sd: information to @actor
464 * @actor: handler that splices the data
467 * This function loops over the pipe and calls @actor to do the
468 * actual moving of a single struct pipe_buffer to the desired
469 * destination. It returns when there's no more buffers left in
470 * the pipe or if the requested number of bytes (@sd->total_len)
471 * have been copied. It returns a positive number (one) if the
472 * pipe needs to be filled with more data, zero if the required
473 * number of bytes have been copied and -errno on error.
475 * This, together with splice_from_pipe_{begin,end,next}, may be
476 * used to implement the functionality of __splice_from_pipe() when
477 * locking is required around copying the pipe buffers to the
480 static int splice_from_pipe_feed(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
483 unsigned int head
= pipe
->head
;
484 unsigned int tail
= pipe
->tail
;
485 unsigned int mask
= pipe
->ring_size
- 1;
488 while (!pipe_empty(head
, tail
)) {
489 struct pipe_buffer
*buf
= &pipe
->bufs
[tail
& mask
];
492 if (sd
->len
> sd
->total_len
)
493 sd
->len
= sd
->total_len
;
495 ret
= pipe_buf_confirm(pipe
, buf
);
502 ret
= actor(pipe
, buf
, sd
);
509 sd
->num_spliced
+= ret
;
512 sd
->total_len
-= ret
;
515 pipe_buf_release(pipe
, buf
);
519 sd
->need_wakeup
= true;
530 * splice_from_pipe_next - wait for some data to splice from
531 * @pipe: pipe to splice from
532 * @sd: information about the splice operation
535 * This function will wait for some data and return a positive
536 * value (one) if pipe buffers are available. It will return zero
537 * or -errno if no more data needs to be spliced.
539 static int splice_from_pipe_next(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
542 * Check for signal early to make process killable when there are
543 * always buffers available
545 if (signal_pending(current
))
548 while (pipe_empty(pipe
->head
, pipe
->tail
)) {
555 if (sd
->flags
& SPLICE_F_NONBLOCK
)
558 if (signal_pending(current
))
561 if (sd
->need_wakeup
) {
562 wakeup_pipe_writers(pipe
);
563 sd
->need_wakeup
= false;
573 * splice_from_pipe_begin - start splicing from pipe
574 * @sd: information about the splice operation
577 * This function should be called before a loop containing
578 * splice_from_pipe_next() and splice_from_pipe_feed() to
579 * initialize the necessary fields of @sd.
581 static void splice_from_pipe_begin(struct splice_desc
*sd
)
584 sd
->need_wakeup
= false;
588 * splice_from_pipe_end - finish splicing from pipe
589 * @pipe: pipe to splice from
590 * @sd: information about the splice operation
593 * This function will wake up pipe writers if necessary. It should
594 * be called after a loop containing splice_from_pipe_next() and
595 * splice_from_pipe_feed().
597 static void splice_from_pipe_end(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
600 wakeup_pipe_writers(pipe
);
604 * __splice_from_pipe - splice data from a pipe to given actor
605 * @pipe: pipe to splice from
606 * @sd: information to @actor
607 * @actor: handler that splices the data
610 * This function does little more than loop over the pipe and call
611 * @actor to do the actual moving of a single struct pipe_buffer to
612 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
616 ssize_t
__splice_from_pipe(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
621 splice_from_pipe_begin(sd
);
624 ret
= splice_from_pipe_next(pipe
, sd
);
626 ret
= splice_from_pipe_feed(pipe
, sd
, actor
);
628 splice_from_pipe_end(pipe
, sd
);
630 return sd
->num_spliced
? sd
->num_spliced
: ret
;
632 EXPORT_SYMBOL(__splice_from_pipe
);
635 * splice_from_pipe - splice data from a pipe to a file
636 * @pipe: pipe to splice from
637 * @out: file to splice to
638 * @ppos: position in @out
639 * @len: how many bytes to splice
640 * @flags: splice modifier flags
641 * @actor: handler that splices the data
644 * See __splice_from_pipe. This function locks the pipe inode,
645 * otherwise it's identical to __splice_from_pipe().
648 ssize_t
splice_from_pipe(struct pipe_inode_info
*pipe
, struct file
*out
,
649 loff_t
*ppos
, size_t len
, unsigned int flags
,
653 struct splice_desc sd
= {
661 ret
= __splice_from_pipe(pipe
, &sd
, actor
);
668 * iter_file_splice_write - splice data from a pipe to a file
670 * @out: file to write to
671 * @ppos: position in @out
672 * @len: number of bytes to splice
673 * @flags: splice modifier flags
676 * Will either move or copy pages (determined by @flags options) from
677 * the given pipe inode to the given file.
678 * This one is ->write_iter-based.
682 iter_file_splice_write(struct pipe_inode_info
*pipe
, struct file
*out
,
683 loff_t
*ppos
, size_t len
, unsigned int flags
)
685 struct splice_desc sd
= {
691 int nbufs
= pipe
->max_usage
;
692 struct bio_vec
*array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
696 if (unlikely(!array
))
701 splice_from_pipe_begin(&sd
);
702 while (sd
.total_len
) {
703 struct iov_iter from
;
704 unsigned int head
, tail
, mask
;
708 ret
= splice_from_pipe_next(pipe
, &sd
);
712 if (unlikely(nbufs
< pipe
->max_usage
)) {
714 nbufs
= pipe
->max_usage
;
715 array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
725 mask
= pipe
->ring_size
- 1;
727 /* build the vector */
729 for (n
= 0; !pipe_empty(head
, tail
) && left
&& n
< nbufs
; tail
++, n
++) {
730 struct pipe_buffer
*buf
= &pipe
->bufs
[tail
& mask
];
731 size_t this_len
= buf
->len
;
736 ret
= pipe_buf_confirm(pipe
, buf
);
743 array
[n
].bv_page
= buf
->page
;
744 array
[n
].bv_len
= this_len
;
745 array
[n
].bv_offset
= buf
->offset
;
749 iov_iter_bvec(&from
, WRITE
, array
, n
, sd
.total_len
- left
);
750 ret
= vfs_iter_write(out
, &from
, &sd
.pos
, 0);
754 sd
.num_spliced
+= ret
;
758 /* dismiss the fully eaten buffers, adjust the partial one */
761 struct pipe_buffer
*buf
= &pipe
->bufs
[tail
& mask
];
762 if (ret
>= buf
->len
) {
765 pipe_buf_release(pipe
, buf
);
769 sd
.need_wakeup
= true;
779 splice_from_pipe_end(pipe
, &sd
);
784 ret
= sd
.num_spliced
;
789 EXPORT_SYMBOL(iter_file_splice_write
);
791 static int write_pipe_buf(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
792 struct splice_desc
*sd
)
796 loff_t tmp
= sd
->pos
;
798 data
= kmap(buf
->page
);
799 ret
= __kernel_write(sd
->u
.file
, data
+ buf
->offset
, sd
->len
, &tmp
);
805 static ssize_t
default_file_splice_write(struct pipe_inode_info
*pipe
,
806 struct file
*out
, loff_t
*ppos
,
807 size_t len
, unsigned int flags
)
811 ret
= splice_from_pipe(pipe
, out
, ppos
, len
, flags
, write_pipe_buf
);
819 * generic_splice_sendpage - splice data from a pipe to a socket
820 * @pipe: pipe to splice from
821 * @out: socket to write to
822 * @ppos: position in @out
823 * @len: number of bytes to splice
824 * @flags: splice modifier flags
827 * Will send @len bytes from the pipe to a network socket. No data copying
831 ssize_t
generic_splice_sendpage(struct pipe_inode_info
*pipe
, struct file
*out
,
832 loff_t
*ppos
, size_t len
, unsigned int flags
)
834 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_sendpage
);
837 EXPORT_SYMBOL(generic_splice_sendpage
);
840 * Attempt to initiate a splice from pipe to file.
842 static long do_splice_from(struct pipe_inode_info
*pipe
, struct file
*out
,
843 loff_t
*ppos
, size_t len
, unsigned int flags
)
845 if (out
->f_op
->splice_write
)
846 return out
->f_op
->splice_write(pipe
, out
, ppos
, len
, flags
);
847 return default_file_splice_write(pipe
, out
, ppos
, len
, flags
);
851 * Attempt to initiate a splice from a file to a pipe.
853 static long do_splice_to(struct file
*in
, loff_t
*ppos
,
854 struct pipe_inode_info
*pipe
, size_t len
,
859 if (unlikely(!(in
->f_mode
& FMODE_READ
)))
862 ret
= rw_verify_area(READ
, in
, ppos
, len
);
863 if (unlikely(ret
< 0))
866 if (unlikely(len
> MAX_RW_COUNT
))
869 if (in
->f_op
->splice_read
)
870 return in
->f_op
->splice_read(in
, ppos
, pipe
, len
, flags
);
871 return default_file_splice_read(in
, ppos
, pipe
, len
, flags
);
875 * splice_direct_to_actor - splices data directly between two non-pipes
876 * @in: file to splice from
877 * @sd: actor information on where to splice to
878 * @actor: handles the data splicing
881 * This is a special case helper to splice directly between two
882 * points, without requiring an explicit pipe. Internally an allocated
883 * pipe is cached in the process, and reused during the lifetime of
887 ssize_t
splice_direct_to_actor(struct file
*in
, struct splice_desc
*sd
,
888 splice_direct_actor
*actor
)
890 struct pipe_inode_info
*pipe
;
897 * We require the input being a regular file, as we don't want to
898 * randomly drop data for eg socket -> socket splicing. Use the
899 * piped splicing for that!
901 i_mode
= file_inode(in
)->i_mode
;
902 if (unlikely(!S_ISREG(i_mode
) && !S_ISBLK(i_mode
)))
906 * neither in nor out is a pipe, setup an internal pipe attached to
907 * 'out' and transfer the wanted data from 'in' to 'out' through that
909 pipe
= current
->splice_pipe
;
910 if (unlikely(!pipe
)) {
911 pipe
= alloc_pipe_info();
916 * We don't have an immediate reader, but we'll read the stuff
917 * out of the pipe right after the splice_to_pipe(). So set
918 * PIPE_READERS appropriately.
922 current
->splice_pipe
= pipe
;
934 * Don't block on output, we have to drain the direct pipe.
936 sd
->flags
&= ~SPLICE_F_NONBLOCK
;
937 more
= sd
->flags
& SPLICE_F_MORE
;
939 WARN_ON_ONCE(!pipe_empty(pipe
->head
, pipe
->tail
));
942 unsigned int p_space
;
944 loff_t pos
= sd
->pos
, prev_pos
= pos
;
946 /* Don't try to read more the pipe has space for. */
947 p_space
= pipe
->max_usage
-
948 pipe_occupancy(pipe
->head
, pipe
->tail
);
949 read_len
= min_t(size_t, len
, p_space
<< PAGE_SHIFT
);
950 ret
= do_splice_to(in
, &pos
, pipe
, read_len
, flags
);
951 if (unlikely(ret
<= 0))
955 sd
->total_len
= read_len
;
958 * If more data is pending, set SPLICE_F_MORE
959 * If this is the last data and SPLICE_F_MORE was not set
960 * initially, clears it.
963 sd
->flags
|= SPLICE_F_MORE
;
965 sd
->flags
&= ~SPLICE_F_MORE
;
967 * NOTE: nonblocking mode only applies to the input. We
968 * must not do the output in nonblocking mode as then we
969 * could get stuck data in the internal pipe:
971 ret
= actor(pipe
, sd
);
972 if (unlikely(ret
<= 0)) {
981 if (ret
< read_len
) {
982 sd
->pos
= prev_pos
+ ret
;
988 pipe
->tail
= pipe
->head
= 0;
994 * If we did an incomplete transfer we must release
995 * the pipe buffers in question:
997 for (i
= 0; i
< pipe
->ring_size
; i
++) {
998 struct pipe_buffer
*buf
= &pipe
->bufs
[i
];
1001 pipe_buf_release(pipe
, buf
);
1009 EXPORT_SYMBOL(splice_direct_to_actor
);
1011 static int direct_splice_actor(struct pipe_inode_info
*pipe
,
1012 struct splice_desc
*sd
)
1014 struct file
*file
= sd
->u
.file
;
1016 return do_splice_from(pipe
, file
, sd
->opos
, sd
->total_len
,
1021 * do_splice_direct - splices data directly between two files
1022 * @in: file to splice from
1023 * @ppos: input file offset
1024 * @out: file to splice to
1025 * @opos: output file offset
1026 * @len: number of bytes to splice
1027 * @flags: splice modifier flags
1030 * For use by do_sendfile(). splice can easily emulate sendfile, but
1031 * doing it in the application would incur an extra system call
1032 * (splice in + splice out, as compared to just sendfile()). So this helper
1033 * can splice directly through a process-private pipe.
1036 long do_splice_direct(struct file
*in
, loff_t
*ppos
, struct file
*out
,
1037 loff_t
*opos
, size_t len
, unsigned int flags
)
1039 struct splice_desc sd
= {
1049 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
1052 if (unlikely(out
->f_flags
& O_APPEND
))
1055 ret
= rw_verify_area(WRITE
, out
, opos
, len
);
1056 if (unlikely(ret
< 0))
1059 ret
= splice_direct_to_actor(in
, &sd
, direct_splice_actor
);
1065 EXPORT_SYMBOL(do_splice_direct
);
1067 static int wait_for_space(struct pipe_inode_info
*pipe
, unsigned flags
)
1070 if (unlikely(!pipe
->readers
)) {
1071 send_sig(SIGPIPE
, current
, 0);
1074 if (!pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
))
1076 if (flags
& SPLICE_F_NONBLOCK
)
1078 if (signal_pending(current
))
1079 return -ERESTARTSYS
;
1084 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1085 struct pipe_inode_info
*opipe
,
1086 size_t len
, unsigned int flags
);
1089 * Determine where to splice to/from.
1091 long do_splice(struct file
*in
, loff_t __user
*off_in
,
1092 struct file
*out
, loff_t __user
*off_out
,
1093 size_t len
, unsigned int flags
)
1095 struct pipe_inode_info
*ipipe
;
1096 struct pipe_inode_info
*opipe
;
1100 if (unlikely(!(in
->f_mode
& FMODE_READ
) ||
1101 !(out
->f_mode
& FMODE_WRITE
)))
1104 ipipe
= get_pipe_info(in
, true);
1105 opipe
= get_pipe_info(out
, true);
1107 if (ipipe
&& opipe
) {
1108 if (off_in
|| off_out
)
1111 /* Splicing to self would be fun, but... */
1115 if ((in
->f_flags
| out
->f_flags
) & O_NONBLOCK
)
1116 flags
|= SPLICE_F_NONBLOCK
;
1118 return splice_pipe_to_pipe(ipipe
, opipe
, len
, flags
);
1125 if (!(out
->f_mode
& FMODE_PWRITE
))
1127 if (copy_from_user(&offset
, off_out
, sizeof(loff_t
)))
1130 offset
= out
->f_pos
;
1133 if (unlikely(out
->f_flags
& O_APPEND
))
1136 ret
= rw_verify_area(WRITE
, out
, &offset
, len
);
1137 if (unlikely(ret
< 0))
1140 if (in
->f_flags
& O_NONBLOCK
)
1141 flags
|= SPLICE_F_NONBLOCK
;
1143 file_start_write(out
);
1144 ret
= do_splice_from(ipipe
, out
, &offset
, len
, flags
);
1145 file_end_write(out
);
1148 out
->f_pos
= offset
;
1149 else if (copy_to_user(off_out
, &offset
, sizeof(loff_t
)))
1159 if (!(in
->f_mode
& FMODE_PREAD
))
1161 if (copy_from_user(&offset
, off_in
, sizeof(loff_t
)))
1167 if (out
->f_flags
& O_NONBLOCK
)
1168 flags
|= SPLICE_F_NONBLOCK
;
1171 ret
= wait_for_space(opipe
, flags
);
1173 unsigned int p_space
;
1175 /* Don't try to read more the pipe has space for. */
1176 p_space
= opipe
->max_usage
- pipe_occupancy(opipe
->head
, opipe
->tail
);
1177 len
= min_t(size_t, len
, p_space
<< PAGE_SHIFT
);
1179 ret
= do_splice_to(in
, &offset
, opipe
, len
, flags
);
1183 wakeup_pipe_readers(opipe
);
1186 else if (copy_to_user(off_in
, &offset
, sizeof(loff_t
)))
1195 static int iter_to_pipe(struct iov_iter
*from
,
1196 struct pipe_inode_info
*pipe
,
1199 struct pipe_buffer buf
= {
1200 .ops
= &user_page_pipe_buf_ops
,
1205 bool failed
= false;
1207 while (iov_iter_count(from
) && !failed
) {
1208 struct page
*pages
[16];
1213 copied
= iov_iter_get_pages(from
, pages
, ~0UL, 16, &start
);
1219 for (n
= 0; copied
; n
++, start
= 0) {
1220 int size
= min_t(int, copied
, PAGE_SIZE
- start
);
1222 buf
.page
= pages
[n
];
1225 ret
= add_to_pipe(pipe
, &buf
);
1226 if (unlikely(ret
< 0)) {
1229 iov_iter_advance(from
, ret
);
1238 return total
? total
: ret
;
1241 static int pipe_to_user(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
1242 struct splice_desc
*sd
)
1244 int n
= copy_page_to_iter(buf
->page
, buf
->offset
, sd
->len
, sd
->u
.data
);
1245 return n
== sd
->len
? n
: -EFAULT
;
1249 * For lack of a better implementation, implement vmsplice() to userspace
1250 * as a simple copy of the pipes pages to the user iov.
1252 static long vmsplice_to_user(struct file
*file
, struct iov_iter
*iter
,
1255 struct pipe_inode_info
*pipe
= get_pipe_info(file
, true);
1256 struct splice_desc sd
= {
1257 .total_len
= iov_iter_count(iter
),
1268 ret
= __splice_from_pipe(pipe
, &sd
, pipe_to_user
);
1276 * vmsplice splices a user address range into a pipe. It can be thought of
1277 * as splice-from-memory, where the regular splice is splice-from-file (or
1278 * to file). In both cases the output is a pipe, naturally.
1280 static long vmsplice_to_pipe(struct file
*file
, struct iov_iter
*iter
,
1283 struct pipe_inode_info
*pipe
;
1285 unsigned buf_flag
= 0;
1287 if (flags
& SPLICE_F_GIFT
)
1288 buf_flag
= PIPE_BUF_FLAG_GIFT
;
1290 pipe
= get_pipe_info(file
, true);
1295 ret
= wait_for_space(pipe
, flags
);
1297 ret
= iter_to_pipe(iter
, pipe
, buf_flag
);
1300 wakeup_pipe_readers(pipe
);
1304 static int vmsplice_type(struct fd f
, int *type
)
1308 if (f
.file
->f_mode
& FMODE_WRITE
) {
1310 } else if (f
.file
->f_mode
& FMODE_READ
) {
1320 * Note that vmsplice only really supports true splicing _from_ user memory
1321 * to a pipe, not the other way around. Splicing from user memory is a simple
1322 * operation that can be supported without any funky alignment restrictions
1323 * or nasty vm tricks. We simply map in the user memory and fill them into
1324 * a pipe. The reverse isn't quite as easy, though. There are two possible
1325 * solutions for that:
1327 * - memcpy() the data internally, at which point we might as well just
1328 * do a regular read() on the buffer anyway.
1329 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1330 * has restriction limitations on both ends of the pipe).
1332 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1335 static long do_vmsplice(struct file
*f
, struct iov_iter
*iter
, unsigned int flags
)
1337 if (unlikely(flags
& ~SPLICE_F_ALL
))
1340 if (!iov_iter_count(iter
))
1343 if (iov_iter_rw(iter
) == WRITE
)
1344 return vmsplice_to_pipe(f
, iter
, flags
);
1346 return vmsplice_to_user(f
, iter
, flags
);
1349 SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct iovec __user
*, uiov
,
1350 unsigned long, nr_segs
, unsigned int, flags
)
1352 struct iovec iovstack
[UIO_FASTIOV
];
1353 struct iovec
*iov
= iovstack
;
1354 struct iov_iter iter
;
1360 error
= vmsplice_type(f
, &type
);
1364 error
= import_iovec(type
, uiov
, nr_segs
,
1365 ARRAY_SIZE(iovstack
), &iov
, &iter
);
1367 error
= do_vmsplice(f
.file
, &iter
, flags
);
1374 #ifdef CONFIG_COMPAT
1375 COMPAT_SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct compat_iovec __user
*, iov32
,
1376 unsigned int, nr_segs
, unsigned int, flags
)
1378 struct iovec iovstack
[UIO_FASTIOV
];
1379 struct iovec
*iov
= iovstack
;
1380 struct iov_iter iter
;
1386 error
= vmsplice_type(f
, &type
);
1390 error
= compat_import_iovec(type
, iov32
, nr_segs
,
1391 ARRAY_SIZE(iovstack
), &iov
, &iter
);
1393 error
= do_vmsplice(f
.file
, &iter
, flags
);
1401 SYSCALL_DEFINE6(splice
, int, fd_in
, loff_t __user
*, off_in
,
1402 int, fd_out
, loff_t __user
*, off_out
,
1403 size_t, len
, unsigned int, flags
)
1411 if (unlikely(flags
& ~SPLICE_F_ALL
))
1417 out
= fdget(fd_out
);
1419 error
= do_splice(in
.file
, off_in
, out
.file
, off_out
,
1429 * Make sure there's data to read. Wait for input if we can, otherwise
1430 * return an appropriate error.
1432 static int ipipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1437 * Check the pipe occupancy without the inode lock first. This function
1438 * is speculative anyways, so missing one is ok.
1440 if (!pipe_empty(pipe
->head
, pipe
->tail
))
1446 while (pipe_empty(pipe
->head
, pipe
->tail
)) {
1447 if (signal_pending(current
)) {
1453 if (flags
& SPLICE_F_NONBLOCK
) {
1465 * Make sure there's writeable room. Wait for room if we can, otherwise
1466 * return an appropriate error.
1468 static int opipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1473 * Check pipe occupancy without the inode lock first. This function
1474 * is speculative anyways, so missing one is ok.
1476 if (!pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
))
1482 while (pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
)) {
1483 if (!pipe
->readers
) {
1484 send_sig(SIGPIPE
, current
, 0);
1488 if (flags
& SPLICE_F_NONBLOCK
) {
1492 if (signal_pending(current
)) {
1504 * Splice contents of ipipe to opipe.
1506 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1507 struct pipe_inode_info
*opipe
,
1508 size_t len
, unsigned int flags
)
1510 struct pipe_buffer
*ibuf
, *obuf
;
1511 unsigned int i_head
, o_head
;
1512 unsigned int i_tail
, o_tail
;
1513 unsigned int i_mask
, o_mask
;
1515 bool input_wakeup
= false;
1519 ret
= ipipe_prep(ipipe
, flags
);
1523 ret
= opipe_prep(opipe
, flags
);
1528 * Potential ABBA deadlock, work around it by ordering lock
1529 * grabbing by pipe info address. Otherwise two different processes
1530 * could deadlock (one doing tee from A -> B, the other from B -> A).
1532 pipe_double_lock(ipipe
, opipe
);
1534 i_tail
= ipipe
->tail
;
1535 i_mask
= ipipe
->ring_size
- 1;
1536 o_head
= opipe
->head
;
1537 o_mask
= opipe
->ring_size
- 1;
1542 if (!opipe
->readers
) {
1543 send_sig(SIGPIPE
, current
, 0);
1549 i_head
= ipipe
->head
;
1550 o_tail
= opipe
->tail
;
1552 if (pipe_empty(i_head
, i_tail
) && !ipipe
->writers
)
1556 * Cannot make any progress, because either the input
1557 * pipe is empty or the output pipe is full.
1559 if (pipe_empty(i_head
, i_tail
) ||
1560 pipe_full(o_head
, o_tail
, opipe
->max_usage
)) {
1561 /* Already processed some buffers, break */
1565 if (flags
& SPLICE_F_NONBLOCK
) {
1571 * We raced with another reader/writer and haven't
1572 * managed to process any buffers. A zero return
1573 * value means EOF, so retry instead.
1580 ibuf
= &ipipe
->bufs
[i_tail
& i_mask
];
1581 obuf
= &opipe
->bufs
[o_head
& o_mask
];
1583 if (len
>= ibuf
->len
) {
1585 * Simply move the whole buffer from ipipe to opipe
1590 ipipe
->tail
= i_tail
;
1591 input_wakeup
= true;
1594 opipe
->head
= o_head
;
1597 * Get a reference to this pipe buffer,
1598 * so we can copy the contents over.
1600 if (!pipe_buf_get(ipipe
, ibuf
)) {
1608 * Don't inherit the gift and merge flags, we need to
1609 * prevent multiple steals of this page.
1611 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1612 obuf
->flags
&= ~PIPE_BUF_FLAG_CAN_MERGE
;
1615 ibuf
->offset
+= len
;
1619 opipe
->head
= o_head
;
1629 * If we put data in the output pipe, wakeup any potential readers.
1632 wakeup_pipe_readers(opipe
);
1635 wakeup_pipe_writers(ipipe
);
1641 * Link contents of ipipe to opipe.
1643 static int link_pipe(struct pipe_inode_info
*ipipe
,
1644 struct pipe_inode_info
*opipe
,
1645 size_t len
, unsigned int flags
)
1647 struct pipe_buffer
*ibuf
, *obuf
;
1648 unsigned int i_head
, o_head
;
1649 unsigned int i_tail
, o_tail
;
1650 unsigned int i_mask
, o_mask
;
1654 * Potential ABBA deadlock, work around it by ordering lock
1655 * grabbing by pipe info address. Otherwise two different processes
1656 * could deadlock (one doing tee from A -> B, the other from B -> A).
1658 pipe_double_lock(ipipe
, opipe
);
1660 i_tail
= ipipe
->tail
;
1661 i_mask
= ipipe
->ring_size
- 1;
1662 o_head
= opipe
->head
;
1663 o_mask
= opipe
->ring_size
- 1;
1666 if (!opipe
->readers
) {
1667 send_sig(SIGPIPE
, current
, 0);
1673 i_head
= ipipe
->head
;
1674 o_tail
= opipe
->tail
;
1677 * If we have iterated all input buffers or run out of
1678 * output room, break.
1680 if (pipe_empty(i_head
, i_tail
) ||
1681 pipe_full(o_head
, o_tail
, opipe
->max_usage
))
1684 ibuf
= &ipipe
->bufs
[i_tail
& i_mask
];
1685 obuf
= &opipe
->bufs
[o_head
& o_mask
];
1688 * Get a reference to this pipe buffer,
1689 * so we can copy the contents over.
1691 if (!pipe_buf_get(ipipe
, ibuf
)) {
1700 * Don't inherit the gift and merge flag, we need to prevent
1701 * multiple steals of this page.
1703 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1704 obuf
->flags
&= ~PIPE_BUF_FLAG_CAN_MERGE
;
1706 if (obuf
->len
> len
)
1712 opipe
->head
= o_head
;
1720 * If we put data in the output pipe, wakeup any potential readers.
1723 wakeup_pipe_readers(opipe
);
1729 * This is a tee(1) implementation that works on pipes. It doesn't copy
1730 * any data, it simply references the 'in' pages on the 'out' pipe.
1731 * The 'flags' used are the SPLICE_F_* variants, currently the only
1732 * applicable one is SPLICE_F_NONBLOCK.
1734 long do_tee(struct file
*in
, struct file
*out
, size_t len
, unsigned int flags
)
1736 struct pipe_inode_info
*ipipe
= get_pipe_info(in
, true);
1737 struct pipe_inode_info
*opipe
= get_pipe_info(out
, true);
1740 if (unlikely(!(in
->f_mode
& FMODE_READ
) ||
1741 !(out
->f_mode
& FMODE_WRITE
)))
1745 * Duplicate the contents of ipipe to opipe without actually
1748 if (ipipe
&& opipe
&& ipipe
!= opipe
) {
1749 if ((in
->f_flags
| out
->f_flags
) & O_NONBLOCK
)
1750 flags
|= SPLICE_F_NONBLOCK
;
1753 * Keep going, unless we encounter an error. The ipipe/opipe
1754 * ordering doesn't really matter.
1756 ret
= ipipe_prep(ipipe
, flags
);
1758 ret
= opipe_prep(opipe
, flags
);
1760 ret
= link_pipe(ipipe
, opipe
, len
, flags
);
1767 SYSCALL_DEFINE4(tee
, int, fdin
, int, fdout
, size_t, len
, unsigned int, flags
)
1772 if (unlikely(flags
& ~SPLICE_F_ALL
))
1783 error
= do_tee(in
.file
, out
.file
, len
, flags
);