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 int page_cache_pipe_buf_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 .steal
= page_cache_pipe_buf_steal
,
145 .get
= generic_pipe_buf_get
,
148 static int user_page_pipe_buf_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_steal(pipe
, buf
);
158 static const struct pipe_buf_operations user_page_pipe_buf_ops
= {
159 .confirm
= generic_pipe_buf_confirm
,
160 .release
= page_cache_pipe_buf_release
,
161 .steal
= user_page_pipe_buf_steal
,
162 .get
= generic_pipe_buf_get
,
165 static void wakeup_pipe_readers(struct pipe_inode_info
*pipe
)
168 if (waitqueue_active(&pipe
->wait
))
169 wake_up_interruptible(&pipe
->wait
);
170 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
174 * splice_to_pipe - fill passed data into a pipe
175 * @pipe: pipe to fill
179 * @spd contains a map of pages and len/offset tuples, along with
180 * the struct pipe_buf_operations associated with these pages. This
181 * function will link that data to the pipe.
184 ssize_t
splice_to_pipe(struct pipe_inode_info
*pipe
,
185 struct splice_pipe_desc
*spd
)
187 unsigned int spd_pages
= spd
->nr_pages
;
188 int ret
= 0, page_nr
= 0;
193 if (unlikely(!pipe
->readers
)) {
194 send_sig(SIGPIPE
, current
, 0);
199 while (pipe
->nrbufs
< pipe
->buffers
) {
200 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
201 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
203 buf
->page
= spd
->pages
[page_nr
];
204 buf
->offset
= spd
->partial
[page_nr
].offset
;
205 buf
->len
= spd
->partial
[page_nr
].len
;
206 buf
->private = spd
->partial
[page_nr
].private;
214 if (!--spd
->nr_pages
)
222 while (page_nr
< spd_pages
)
223 spd
->spd_release(spd
, page_nr
++);
227 EXPORT_SYMBOL_GPL(splice_to_pipe
);
229 ssize_t
add_to_pipe(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
)
233 if (unlikely(!pipe
->readers
)) {
234 send_sig(SIGPIPE
, current
, 0);
236 } else if (pipe
->nrbufs
== pipe
->buffers
) {
239 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
240 pipe
->bufs
[newbuf
] = *buf
;
244 pipe_buf_release(pipe
, buf
);
247 EXPORT_SYMBOL(add_to_pipe
);
250 * Check if we need to grow the arrays holding pages and partial page
253 int splice_grow_spd(const struct pipe_inode_info
*pipe
, struct splice_pipe_desc
*spd
)
255 unsigned int buffers
= READ_ONCE(pipe
->buffers
);
257 spd
->nr_pages_max
= buffers
;
258 if (buffers
<= PIPE_DEF_BUFFERS
)
261 spd
->pages
= kmalloc_array(buffers
, sizeof(struct page
*), GFP_KERNEL
);
262 spd
->partial
= kmalloc_array(buffers
, sizeof(struct partial_page
),
265 if (spd
->pages
&& spd
->partial
)
273 void splice_shrink_spd(struct splice_pipe_desc
*spd
)
275 if (spd
->nr_pages_max
<= PIPE_DEF_BUFFERS
)
283 * generic_file_splice_read - splice data from file to a pipe
284 * @in: file to splice from
285 * @ppos: position in @in
286 * @pipe: pipe to splice to
287 * @len: number of bytes to splice
288 * @flags: splice modifier flags
291 * Will read pages from given file and fill them into a pipe. Can be
292 * used as long as it has more or less sane ->read_iter().
295 ssize_t
generic_file_splice_read(struct file
*in
, loff_t
*ppos
,
296 struct pipe_inode_info
*pipe
, size_t len
,
303 iov_iter_pipe(&to
, READ
, pipe
, len
);
305 init_sync_kiocb(&kiocb
, in
);
306 kiocb
.ki_pos
= *ppos
;
307 ret
= call_read_iter(in
, &kiocb
, &to
);
309 *ppos
= kiocb
.ki_pos
;
311 } else if (ret
< 0) {
314 iov_iter_advance(&to
, 0); /* to free what was emitted */
316 * callers of ->splice_read() expect -EAGAIN on
317 * "can't put anything in there", rather than -EFAULT.
325 EXPORT_SYMBOL(generic_file_splice_read
);
327 const struct pipe_buf_operations default_pipe_buf_ops
= {
328 .confirm
= generic_pipe_buf_confirm
,
329 .release
= generic_pipe_buf_release
,
330 .steal
= generic_pipe_buf_steal
,
331 .get
= generic_pipe_buf_get
,
334 int generic_pipe_buf_nosteal(struct pipe_inode_info
*pipe
,
335 struct pipe_buffer
*buf
)
340 /* Pipe buffer operations for a socket and similar. */
341 const struct pipe_buf_operations nosteal_pipe_buf_ops
= {
342 .confirm
= generic_pipe_buf_confirm
,
343 .release
= generic_pipe_buf_release
,
344 .steal
= generic_pipe_buf_nosteal
,
345 .get
= generic_pipe_buf_get
,
347 EXPORT_SYMBOL(nosteal_pipe_buf_ops
);
349 static ssize_t
kernel_readv(struct file
*file
, const struct kvec
*vec
,
350 unsigned long vlen
, loff_t offset
)
358 /* The cast to a user pointer is valid due to the set_fs() */
359 res
= vfs_readv(file
, (const struct iovec __user
*)vec
, vlen
, &pos
, 0);
365 static ssize_t
default_file_splice_read(struct file
*in
, loff_t
*ppos
,
366 struct pipe_inode_info
*pipe
, size_t len
,
369 struct kvec
*vec
, __vec
[PIPE_DEF_BUFFERS
];
372 unsigned int nr_pages
;
373 size_t offset
, base
, copied
= 0;
377 if (pipe
->nrbufs
== pipe
->buffers
)
381 * Try to keep page boundaries matching to source pagecache ones -
382 * it probably won't be much help, but...
384 offset
= *ppos
& ~PAGE_MASK
;
386 iov_iter_pipe(&to
, READ
, pipe
, len
+ offset
);
388 res
= iov_iter_get_pages_alloc(&to
, &pages
, len
+ offset
, &base
);
392 nr_pages
= DIV_ROUND_UP(res
+ base
, PAGE_SIZE
);
395 if (nr_pages
> PIPE_DEF_BUFFERS
) {
396 vec
= kmalloc_array(nr_pages
, sizeof(struct kvec
), GFP_KERNEL
);
397 if (unlikely(!vec
)) {
403 pipe
->bufs
[to
.idx
].offset
= offset
;
404 pipe
->bufs
[to
.idx
].len
-= offset
;
406 for (i
= 0; i
< nr_pages
; i
++) {
407 size_t this_len
= min_t(size_t, len
, PAGE_SIZE
- offset
);
408 vec
[i
].iov_base
= page_address(pages
[i
]) + offset
;
409 vec
[i
].iov_len
= this_len
;
414 res
= kernel_readv(in
, vec
, nr_pages
, *ppos
);
423 for (i
= 0; i
< nr_pages
; i
++)
426 iov_iter_advance(&to
, copied
); /* truncates and discards */
431 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
432 * using sendpage(). Return the number of bytes sent.
434 static int pipe_to_sendpage(struct pipe_inode_info
*pipe
,
435 struct pipe_buffer
*buf
, struct splice_desc
*sd
)
437 struct file
*file
= sd
->u
.file
;
438 loff_t pos
= sd
->pos
;
441 if (!likely(file
->f_op
->sendpage
))
444 more
= (sd
->flags
& SPLICE_F_MORE
) ? MSG_MORE
: 0;
446 if (sd
->len
< sd
->total_len
&& pipe
->nrbufs
> 1)
447 more
|= MSG_SENDPAGE_NOTLAST
;
449 return file
->f_op
->sendpage(file
, buf
->page
, buf
->offset
,
450 sd
->len
, &pos
, more
);
453 static void wakeup_pipe_writers(struct pipe_inode_info
*pipe
)
456 if (waitqueue_active(&pipe
->wait
))
457 wake_up_interruptible(&pipe
->wait
);
458 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
462 * splice_from_pipe_feed - feed available data from a pipe to a file
463 * @pipe: pipe to splice from
464 * @sd: information to @actor
465 * @actor: handler that splices the data
468 * This function loops over the pipe and calls @actor to do the
469 * actual moving of a single struct pipe_buffer to the desired
470 * destination. It returns when there's no more buffers left in
471 * the pipe or if the requested number of bytes (@sd->total_len)
472 * have been copied. It returns a positive number (one) if the
473 * pipe needs to be filled with more data, zero if the required
474 * number of bytes have been copied and -errno on error.
476 * This, together with splice_from_pipe_{begin,end,next}, may be
477 * used to implement the functionality of __splice_from_pipe() when
478 * locking is required around copying the pipe buffers to the
481 static int splice_from_pipe_feed(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
486 while (pipe
->nrbufs
) {
487 struct pipe_buffer
*buf
= pipe
->bufs
+ pipe
->curbuf
;
490 if (sd
->len
> sd
->total_len
)
491 sd
->len
= sd
->total_len
;
493 ret
= pipe_buf_confirm(pipe
, buf
);
500 ret
= actor(pipe
, buf
, sd
);
507 sd
->num_spliced
+= ret
;
510 sd
->total_len
-= ret
;
513 pipe_buf_release(pipe
, buf
);
514 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
517 sd
->need_wakeup
= true;
528 * splice_from_pipe_next - wait for some data to splice from
529 * @pipe: pipe to splice from
530 * @sd: information about the splice operation
533 * This function will wait for some data and return a positive
534 * value (one) if pipe buffers are available. It will return zero
535 * or -errno if no more data needs to be spliced.
537 static int splice_from_pipe_next(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
540 * Check for signal early to make process killable when there are
541 * always buffers available
543 if (signal_pending(current
))
546 while (!pipe
->nrbufs
) {
550 if (!pipe
->waiting_writers
&& sd
->num_spliced
)
553 if (sd
->flags
& SPLICE_F_NONBLOCK
)
556 if (signal_pending(current
))
559 if (sd
->need_wakeup
) {
560 wakeup_pipe_writers(pipe
);
561 sd
->need_wakeup
= false;
571 * splice_from_pipe_begin - start splicing from pipe
572 * @sd: information about the splice operation
575 * This function should be called before a loop containing
576 * splice_from_pipe_next() and splice_from_pipe_feed() to
577 * initialize the necessary fields of @sd.
579 static void splice_from_pipe_begin(struct splice_desc
*sd
)
582 sd
->need_wakeup
= false;
586 * splice_from_pipe_end - finish splicing from pipe
587 * @pipe: pipe to splice from
588 * @sd: information about the splice operation
591 * This function will wake up pipe writers if necessary. It should
592 * be called after a loop containing splice_from_pipe_next() and
593 * splice_from_pipe_feed().
595 static void splice_from_pipe_end(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
598 wakeup_pipe_writers(pipe
);
602 * __splice_from_pipe - splice data from a pipe to given actor
603 * @pipe: pipe to splice from
604 * @sd: information to @actor
605 * @actor: handler that splices the data
608 * This function does little more than loop over the pipe and call
609 * @actor to do the actual moving of a single struct pipe_buffer to
610 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
614 ssize_t
__splice_from_pipe(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
619 splice_from_pipe_begin(sd
);
622 ret
= splice_from_pipe_next(pipe
, sd
);
624 ret
= splice_from_pipe_feed(pipe
, sd
, actor
);
626 splice_from_pipe_end(pipe
, sd
);
628 return sd
->num_spliced
? sd
->num_spliced
: ret
;
630 EXPORT_SYMBOL(__splice_from_pipe
);
633 * splice_from_pipe - splice data from a pipe to a file
634 * @pipe: pipe to splice from
635 * @out: file to splice to
636 * @ppos: position in @out
637 * @len: how many bytes to splice
638 * @flags: splice modifier flags
639 * @actor: handler that splices the data
642 * See __splice_from_pipe. This function locks the pipe inode,
643 * otherwise it's identical to __splice_from_pipe().
646 ssize_t
splice_from_pipe(struct pipe_inode_info
*pipe
, struct file
*out
,
647 loff_t
*ppos
, size_t len
, unsigned int flags
,
651 struct splice_desc sd
= {
659 ret
= __splice_from_pipe(pipe
, &sd
, actor
);
666 * iter_file_splice_write - splice data from a pipe to a file
668 * @out: file to write to
669 * @ppos: position in @out
670 * @len: number of bytes to splice
671 * @flags: splice modifier flags
674 * Will either move or copy pages (determined by @flags options) from
675 * the given pipe inode to the given file.
676 * This one is ->write_iter-based.
680 iter_file_splice_write(struct pipe_inode_info
*pipe
, struct file
*out
,
681 loff_t
*ppos
, size_t len
, unsigned int flags
)
683 struct splice_desc sd
= {
689 int nbufs
= pipe
->buffers
;
690 struct bio_vec
*array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
694 if (unlikely(!array
))
699 splice_from_pipe_begin(&sd
);
700 while (sd
.total_len
) {
701 struct iov_iter from
;
705 ret
= splice_from_pipe_next(pipe
, &sd
);
709 if (unlikely(nbufs
< pipe
->buffers
)) {
711 nbufs
= pipe
->buffers
;
712 array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
720 /* build the vector */
722 for (n
= 0, idx
= pipe
->curbuf
; left
&& n
< pipe
->nrbufs
; n
++, idx
++) {
723 struct pipe_buffer
*buf
= pipe
->bufs
+ idx
;
724 size_t this_len
= buf
->len
;
729 if (idx
== pipe
->buffers
- 1)
732 ret
= pipe_buf_confirm(pipe
, buf
);
739 array
[n
].bv_page
= buf
->page
;
740 array
[n
].bv_len
= this_len
;
741 array
[n
].bv_offset
= buf
->offset
;
745 iov_iter_bvec(&from
, WRITE
, array
, n
, sd
.total_len
- left
);
746 ret
= vfs_iter_write(out
, &from
, &sd
.pos
, 0);
750 sd
.num_spliced
+= ret
;
754 /* dismiss the fully eaten buffers, adjust the partial one */
756 struct pipe_buffer
*buf
= pipe
->bufs
+ pipe
->curbuf
;
757 if (ret
>= buf
->len
) {
760 pipe_buf_release(pipe
, buf
);
761 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
764 sd
.need_wakeup
= true;
774 splice_from_pipe_end(pipe
, &sd
);
779 ret
= sd
.num_spliced
;
784 EXPORT_SYMBOL(iter_file_splice_write
);
786 static int write_pipe_buf(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
787 struct splice_desc
*sd
)
791 loff_t tmp
= sd
->pos
;
793 data
= kmap(buf
->page
);
794 ret
= __kernel_write(sd
->u
.file
, data
+ buf
->offset
, sd
->len
, &tmp
);
800 static ssize_t
default_file_splice_write(struct pipe_inode_info
*pipe
,
801 struct file
*out
, loff_t
*ppos
,
802 size_t len
, unsigned int flags
)
806 ret
= splice_from_pipe(pipe
, out
, ppos
, len
, flags
, write_pipe_buf
);
814 * generic_splice_sendpage - splice data from a pipe to a socket
815 * @pipe: pipe to splice from
816 * @out: socket to write to
817 * @ppos: position in @out
818 * @len: number of bytes to splice
819 * @flags: splice modifier flags
822 * Will send @len bytes from the pipe to a network socket. No data copying
826 ssize_t
generic_splice_sendpage(struct pipe_inode_info
*pipe
, struct file
*out
,
827 loff_t
*ppos
, size_t len
, unsigned int flags
)
829 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_sendpage
);
832 EXPORT_SYMBOL(generic_splice_sendpage
);
835 * Attempt to initiate a splice from pipe to file.
837 long do_splice_from(struct pipe_inode_info
*pipe
, struct file
*out
,
838 loff_t
*ppos
, size_t len
, unsigned int flags
)
840 ssize_t (*splice_write
)(struct pipe_inode_info
*, struct file
*,
841 loff_t
*, size_t, unsigned int);
843 if (out
->f_op
->splice_write
)
844 splice_write
= out
->f_op
->splice_write
;
846 splice_write
= default_file_splice_write
;
848 return splice_write(pipe
, out
, ppos
, len
, flags
);
850 EXPORT_SYMBOL_GPL(do_splice_from
);
853 * Attempt to initiate a splice from a file to a pipe.
855 long do_splice_to(struct file
*in
, loff_t
*ppos
,
856 struct pipe_inode_info
*pipe
, size_t len
,
859 ssize_t (*splice_read
)(struct file
*, loff_t
*,
860 struct pipe_inode_info
*, size_t, unsigned int);
863 if (unlikely(!(in
->f_mode
& FMODE_READ
)))
866 ret
= rw_verify_area(READ
, in
, ppos
, len
);
867 if (unlikely(ret
< 0))
870 if (unlikely(len
> MAX_RW_COUNT
))
873 if (in
->f_op
->splice_read
)
874 splice_read
= in
->f_op
->splice_read
;
876 splice_read
= default_file_splice_read
;
878 return splice_read(in
, ppos
, pipe
, len
, flags
);
880 EXPORT_SYMBOL_GPL(do_splice_to
);
883 * splice_direct_to_actor - splices data directly between two non-pipes
884 * @in: file to splice from
885 * @sd: actor information on where to splice to
886 * @actor: handles the data splicing
889 * This is a special case helper to splice directly between two
890 * points, without requiring an explicit pipe. Internally an allocated
891 * pipe is cached in the process, and reused during the lifetime of
895 ssize_t
splice_direct_to_actor(struct file
*in
, struct splice_desc
*sd
,
896 splice_direct_actor
*actor
)
898 struct pipe_inode_info
*pipe
;
905 * We require the input being a regular file, as we don't want to
906 * randomly drop data for eg socket -> socket splicing. Use the
907 * piped splicing for that!
909 i_mode
= file_inode(in
)->i_mode
;
910 if (unlikely(!S_ISREG(i_mode
) && !S_ISBLK(i_mode
)))
914 * neither in nor out is a pipe, setup an internal pipe attached to
915 * 'out' and transfer the wanted data from 'in' to 'out' through that
917 pipe
= current
->splice_pipe
;
918 if (unlikely(!pipe
)) {
919 pipe
= alloc_pipe_info();
924 * We don't have an immediate reader, but we'll read the stuff
925 * out of the pipe right after the splice_to_pipe(). So set
926 * PIPE_READERS appropriately.
930 current
->splice_pipe
= pipe
;
942 * Don't block on output, we have to drain the direct pipe.
944 sd
->flags
&= ~SPLICE_F_NONBLOCK
;
945 more
= sd
->flags
& SPLICE_F_MORE
;
947 WARN_ON_ONCE(pipe
->nrbufs
!= 0);
950 unsigned int pipe_pages
;
952 loff_t pos
= sd
->pos
, prev_pos
= pos
;
954 /* Don't try to read more the pipe has space for. */
955 pipe_pages
= pipe
->buffers
- pipe
->nrbufs
;
956 read_len
= min(len
, (size_t)pipe_pages
<< PAGE_SHIFT
);
957 ret
= do_splice_to(in
, &pos
, pipe
, read_len
, flags
);
958 if (unlikely(ret
<= 0))
962 sd
->total_len
= read_len
;
965 * If more data is pending, set SPLICE_F_MORE
966 * If this is the last data and SPLICE_F_MORE was not set
967 * initially, clears it.
970 sd
->flags
|= SPLICE_F_MORE
;
972 sd
->flags
&= ~SPLICE_F_MORE
;
974 * NOTE: nonblocking mode only applies to the input. We
975 * must not do the output in nonblocking mode as then we
976 * could get stuck data in the internal pipe:
978 ret
= actor(pipe
, sd
);
979 if (unlikely(ret
<= 0)) {
988 if (ret
< read_len
) {
989 sd
->pos
= prev_pos
+ ret
;
995 pipe
->nrbufs
= pipe
->curbuf
= 0;
1001 * If we did an incomplete transfer we must release
1002 * the pipe buffers in question:
1004 for (i
= 0; i
< pipe
->buffers
; i
++) {
1005 struct pipe_buffer
*buf
= pipe
->bufs
+ i
;
1008 pipe_buf_release(pipe
, buf
);
1016 EXPORT_SYMBOL(splice_direct_to_actor
);
1018 static int direct_splice_actor(struct pipe_inode_info
*pipe
,
1019 struct splice_desc
*sd
)
1021 struct file
*file
= sd
->u
.file
;
1023 return do_splice_from(pipe
, file
, sd
->opos
, sd
->total_len
,
1028 * do_splice_direct - splices data directly between two files
1029 * @in: file to splice from
1030 * @ppos: input file offset
1031 * @out: file to splice to
1032 * @opos: output file offset
1033 * @len: number of bytes to splice
1034 * @flags: splice modifier flags
1037 * For use by do_sendfile(). splice can easily emulate sendfile, but
1038 * doing it in the application would incur an extra system call
1039 * (splice in + splice out, as compared to just sendfile()). So this helper
1040 * can splice directly through a process-private pipe.
1043 long do_splice_direct(struct file
*in
, loff_t
*ppos
, struct file
*out
,
1044 loff_t
*opos
, size_t len
, unsigned int flags
)
1046 struct splice_desc sd
= {
1056 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
1059 if (unlikely(out
->f_flags
& O_APPEND
))
1062 ret
= rw_verify_area(WRITE
, out
, opos
, len
);
1063 if (unlikely(ret
< 0))
1066 ret
= splice_direct_to_actor(in
, &sd
, direct_splice_actor
);
1072 EXPORT_SYMBOL(do_splice_direct
);
1074 static int wait_for_space(struct pipe_inode_info
*pipe
, unsigned flags
)
1077 if (unlikely(!pipe
->readers
)) {
1078 send_sig(SIGPIPE
, current
, 0);
1081 if (pipe
->nrbufs
!= pipe
->buffers
)
1083 if (flags
& SPLICE_F_NONBLOCK
)
1085 if (signal_pending(current
))
1086 return -ERESTARTSYS
;
1087 pipe
->waiting_writers
++;
1089 pipe
->waiting_writers
--;
1093 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1094 struct pipe_inode_info
*opipe
,
1095 size_t len
, unsigned int flags
);
1098 * Determine where to splice to/from.
1100 static long do_splice(struct file
*in
, loff_t __user
*off_in
,
1101 struct file
*out
, loff_t __user
*off_out
,
1102 size_t len
, unsigned int flags
)
1104 struct pipe_inode_info
*ipipe
;
1105 struct pipe_inode_info
*opipe
;
1109 ipipe
= get_pipe_info(in
);
1110 opipe
= get_pipe_info(out
);
1112 if (ipipe
&& opipe
) {
1113 if (off_in
|| off_out
)
1116 if (!(in
->f_mode
& FMODE_READ
))
1119 if (!(out
->f_mode
& FMODE_WRITE
))
1122 /* Splicing to self would be fun, but... */
1126 if ((in
->f_flags
| out
->f_flags
) & O_NONBLOCK
)
1127 flags
|= SPLICE_F_NONBLOCK
;
1129 return splice_pipe_to_pipe(ipipe
, opipe
, len
, flags
);
1136 if (!(out
->f_mode
& FMODE_PWRITE
))
1138 if (copy_from_user(&offset
, off_out
, sizeof(loff_t
)))
1141 offset
= out
->f_pos
;
1144 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
1147 if (unlikely(out
->f_flags
& O_APPEND
))
1150 ret
= rw_verify_area(WRITE
, out
, &offset
, len
);
1151 if (unlikely(ret
< 0))
1154 if (in
->f_flags
& O_NONBLOCK
)
1155 flags
|= SPLICE_F_NONBLOCK
;
1157 file_start_write(out
);
1158 ret
= do_splice_from(ipipe
, out
, &offset
, len
, flags
);
1159 file_end_write(out
);
1162 out
->f_pos
= offset
;
1163 else if (copy_to_user(off_out
, &offset
, sizeof(loff_t
)))
1173 if (!(in
->f_mode
& FMODE_PREAD
))
1175 if (copy_from_user(&offset
, off_in
, sizeof(loff_t
)))
1181 if (out
->f_flags
& O_NONBLOCK
)
1182 flags
|= SPLICE_F_NONBLOCK
;
1185 ret
= wait_for_space(opipe
, flags
);
1187 unsigned int pipe_pages
;
1189 /* Don't try to read more the pipe has space for. */
1190 pipe_pages
= opipe
->buffers
- opipe
->nrbufs
;
1191 len
= min(len
, (size_t)pipe_pages
<< PAGE_SHIFT
);
1193 ret
= do_splice_to(in
, &offset
, opipe
, len
, flags
);
1197 wakeup_pipe_readers(opipe
);
1200 else if (copy_to_user(off_in
, &offset
, sizeof(loff_t
)))
1209 static int iter_to_pipe(struct iov_iter
*from
,
1210 struct pipe_inode_info
*pipe
,
1213 struct pipe_buffer buf
= {
1214 .ops
= &user_page_pipe_buf_ops
,
1219 bool failed
= false;
1221 while (iov_iter_count(from
) && !failed
) {
1222 struct page
*pages
[16];
1227 copied
= iov_iter_get_pages(from
, pages
, ~0UL, 16, &start
);
1233 for (n
= 0; copied
; n
++, start
= 0) {
1234 int size
= min_t(int, copied
, PAGE_SIZE
- start
);
1236 buf
.page
= pages
[n
];
1239 ret
= add_to_pipe(pipe
, &buf
);
1240 if (unlikely(ret
< 0)) {
1243 iov_iter_advance(from
, ret
);
1252 return total
? total
: ret
;
1255 static int pipe_to_user(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
1256 struct splice_desc
*sd
)
1258 int n
= copy_page_to_iter(buf
->page
, buf
->offset
, sd
->len
, sd
->u
.data
);
1259 return n
== sd
->len
? n
: -EFAULT
;
1263 * For lack of a better implementation, implement vmsplice() to userspace
1264 * as a simple copy of the pipes pages to the user iov.
1266 static long vmsplice_to_user(struct file
*file
, struct iov_iter
*iter
,
1269 struct pipe_inode_info
*pipe
= get_pipe_info(file
);
1270 struct splice_desc sd
= {
1271 .total_len
= iov_iter_count(iter
),
1282 ret
= __splice_from_pipe(pipe
, &sd
, pipe_to_user
);
1290 * vmsplice splices a user address range into a pipe. It can be thought of
1291 * as splice-from-memory, where the regular splice is splice-from-file (or
1292 * to file). In both cases the output is a pipe, naturally.
1294 static long vmsplice_to_pipe(struct file
*file
, struct iov_iter
*iter
,
1297 struct pipe_inode_info
*pipe
;
1299 unsigned buf_flag
= 0;
1301 if (flags
& SPLICE_F_GIFT
)
1302 buf_flag
= PIPE_BUF_FLAG_GIFT
;
1304 pipe
= get_pipe_info(file
);
1309 ret
= wait_for_space(pipe
, flags
);
1311 ret
= iter_to_pipe(iter
, pipe
, buf_flag
);
1314 wakeup_pipe_readers(pipe
);
1318 static int vmsplice_type(struct fd f
, int *type
)
1322 if (f
.file
->f_mode
& FMODE_WRITE
) {
1324 } else if (f
.file
->f_mode
& FMODE_READ
) {
1334 * Note that vmsplice only really supports true splicing _from_ user memory
1335 * to a pipe, not the other way around. Splicing from user memory is a simple
1336 * operation that can be supported without any funky alignment restrictions
1337 * or nasty vm tricks. We simply map in the user memory and fill them into
1338 * a pipe. The reverse isn't quite as easy, though. There are two possible
1339 * solutions for that:
1341 * - memcpy() the data internally, at which point we might as well just
1342 * do a regular read() on the buffer anyway.
1343 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1344 * has restriction limitations on both ends of the pipe).
1346 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1349 static long do_vmsplice(struct file
*f
, struct iov_iter
*iter
, unsigned int flags
)
1351 if (unlikely(flags
& ~SPLICE_F_ALL
))
1354 if (!iov_iter_count(iter
))
1357 if (iov_iter_rw(iter
) == WRITE
)
1358 return vmsplice_to_pipe(f
, iter
, flags
);
1360 return vmsplice_to_user(f
, iter
, flags
);
1363 SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct iovec __user
*, uiov
,
1364 unsigned long, nr_segs
, unsigned int, flags
)
1366 struct iovec iovstack
[UIO_FASTIOV
];
1367 struct iovec
*iov
= iovstack
;
1368 struct iov_iter iter
;
1374 error
= vmsplice_type(f
, &type
);
1378 error
= import_iovec(type
, uiov
, nr_segs
,
1379 ARRAY_SIZE(iovstack
), &iov
, &iter
);
1381 error
= do_vmsplice(f
.file
, &iter
, flags
);
1388 #ifdef CONFIG_COMPAT
1389 COMPAT_SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct compat_iovec __user
*, iov32
,
1390 unsigned int, nr_segs
, unsigned int, flags
)
1392 struct iovec iovstack
[UIO_FASTIOV
];
1393 struct iovec
*iov
= iovstack
;
1394 struct iov_iter iter
;
1400 error
= vmsplice_type(f
, &type
);
1404 error
= compat_import_iovec(type
, iov32
, nr_segs
,
1405 ARRAY_SIZE(iovstack
), &iov
, &iter
);
1407 error
= do_vmsplice(f
.file
, &iter
, flags
);
1415 SYSCALL_DEFINE6(splice
, int, fd_in
, loff_t __user
*, off_in
,
1416 int, fd_out
, loff_t __user
*, off_out
,
1417 size_t, len
, unsigned int, flags
)
1425 if (unlikely(flags
& ~SPLICE_F_ALL
))
1431 if (in
.file
->f_mode
& FMODE_READ
) {
1432 out
= fdget(fd_out
);
1434 if (out
.file
->f_mode
& FMODE_WRITE
)
1435 error
= do_splice(in
.file
, off_in
,
1447 * Make sure there's data to read. Wait for input if we can, otherwise
1448 * return an appropriate error.
1450 static int ipipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1455 * Check ->nrbufs without the inode lock first. This function
1456 * is speculative anyways, so missing one is ok.
1464 while (!pipe
->nrbufs
) {
1465 if (signal_pending(current
)) {
1471 if (!pipe
->waiting_writers
) {
1472 if (flags
& SPLICE_F_NONBLOCK
) {
1485 * Make sure there's writeable room. Wait for room if we can, otherwise
1486 * return an appropriate error.
1488 static int opipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1493 * Check ->nrbufs without the inode lock first. This function
1494 * is speculative anyways, so missing one is ok.
1496 if (pipe
->nrbufs
< pipe
->buffers
)
1502 while (pipe
->nrbufs
>= pipe
->buffers
) {
1503 if (!pipe
->readers
) {
1504 send_sig(SIGPIPE
, current
, 0);
1508 if (flags
& SPLICE_F_NONBLOCK
) {
1512 if (signal_pending(current
)) {
1516 pipe
->waiting_writers
++;
1518 pipe
->waiting_writers
--;
1526 * Splice contents of ipipe to opipe.
1528 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1529 struct pipe_inode_info
*opipe
,
1530 size_t len
, unsigned int flags
)
1532 struct pipe_buffer
*ibuf
, *obuf
;
1534 bool input_wakeup
= false;
1538 ret
= ipipe_prep(ipipe
, flags
);
1542 ret
= opipe_prep(opipe
, flags
);
1547 * Potential ABBA deadlock, work around it by ordering lock
1548 * grabbing by pipe info address. Otherwise two different processes
1549 * could deadlock (one doing tee from A -> B, the other from B -> A).
1551 pipe_double_lock(ipipe
, opipe
);
1554 if (!opipe
->readers
) {
1555 send_sig(SIGPIPE
, current
, 0);
1561 if (!ipipe
->nrbufs
&& !ipipe
->writers
)
1565 * Cannot make any progress, because either the input
1566 * pipe is empty or the output pipe is full.
1568 if (!ipipe
->nrbufs
|| opipe
->nrbufs
>= opipe
->buffers
) {
1569 /* Already processed some buffers, break */
1573 if (flags
& SPLICE_F_NONBLOCK
) {
1579 * We raced with another reader/writer and haven't
1580 * managed to process any buffers. A zero return
1581 * value means EOF, so retry instead.
1588 ibuf
= ipipe
->bufs
+ ipipe
->curbuf
;
1589 nbuf
= (opipe
->curbuf
+ opipe
->nrbufs
) & (opipe
->buffers
- 1);
1590 obuf
= opipe
->bufs
+ nbuf
;
1592 if (len
>= ibuf
->len
) {
1594 * Simply move the whole buffer from ipipe to opipe
1599 ipipe
->curbuf
= (ipipe
->curbuf
+ 1) & (ipipe
->buffers
- 1);
1601 input_wakeup
= true;
1604 * Get a reference to this pipe buffer,
1605 * so we can copy the contents over.
1607 if (!pipe_buf_get(ipipe
, ibuf
)) {
1615 * Don't inherit the gift flag, we need to
1616 * prevent multiple steals of this page.
1618 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1620 pipe_buf_mark_unmergeable(obuf
);
1624 ibuf
->offset
+= obuf
->len
;
1625 ibuf
->len
-= obuf
->len
;
1635 * If we put data in the output pipe, wakeup any potential readers.
1638 wakeup_pipe_readers(opipe
);
1641 wakeup_pipe_writers(ipipe
);
1647 * Link contents of ipipe to opipe.
1649 static int link_pipe(struct pipe_inode_info
*ipipe
,
1650 struct pipe_inode_info
*opipe
,
1651 size_t len
, unsigned int flags
)
1653 struct pipe_buffer
*ibuf
, *obuf
;
1654 int ret
= 0, i
= 0, nbuf
;
1657 * Potential ABBA deadlock, work around it by ordering lock
1658 * grabbing by pipe info address. Otherwise two different processes
1659 * could deadlock (one doing tee from A -> B, the other from B -> A).
1661 pipe_double_lock(ipipe
, opipe
);
1664 if (!opipe
->readers
) {
1665 send_sig(SIGPIPE
, current
, 0);
1672 * If we have iterated all input buffers or ran out of
1673 * output room, break.
1675 if (i
>= ipipe
->nrbufs
|| opipe
->nrbufs
>= opipe
->buffers
)
1678 ibuf
= ipipe
->bufs
+ ((ipipe
->curbuf
+ i
) & (ipipe
->buffers
-1));
1679 nbuf
= (opipe
->curbuf
+ opipe
->nrbufs
) & (opipe
->buffers
- 1);
1682 * Get a reference to this pipe buffer,
1683 * so we can copy the contents over.
1685 if (!pipe_buf_get(ipipe
, ibuf
)) {
1691 obuf
= opipe
->bufs
+ nbuf
;
1695 * Don't inherit the gift flag, we need to
1696 * prevent multiple steals of this page.
1698 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1700 pipe_buf_mark_unmergeable(obuf
);
1702 if (obuf
->len
> len
)
1712 * return EAGAIN if we have the potential of some data in the
1713 * future, otherwise just return 0
1715 if (!ret
&& ipipe
->waiting_writers
&& (flags
& SPLICE_F_NONBLOCK
))
1722 * If we put data in the output pipe, wakeup any potential readers.
1725 wakeup_pipe_readers(opipe
);
1731 * This is a tee(1) implementation that works on pipes. It doesn't copy
1732 * any data, it simply references the 'in' pages on the 'out' pipe.
1733 * The 'flags' used are the SPLICE_F_* variants, currently the only
1734 * applicable one is SPLICE_F_NONBLOCK.
1736 static long do_tee(struct file
*in
, struct file
*out
, size_t len
,
1739 struct pipe_inode_info
*ipipe
= get_pipe_info(in
);
1740 struct pipe_inode_info
*opipe
= get_pipe_info(out
);
1744 * Duplicate the contents of ipipe to opipe without actually
1747 if (ipipe
&& opipe
&& ipipe
!= opipe
) {
1748 if ((in
->f_flags
| out
->f_flags
) & O_NONBLOCK
)
1749 flags
|= SPLICE_F_NONBLOCK
;
1752 * Keep going, unless we encounter an error. The ipipe/opipe
1753 * ordering doesn't really matter.
1755 ret
= ipipe_prep(ipipe
, flags
);
1757 ret
= opipe_prep(opipe
, flags
);
1759 ret
= link_pipe(ipipe
, opipe
, len
, flags
);
1766 SYSCALL_DEFINE4(tee
, int, fdin
, int, fdout
, size_t, len
, unsigned int, flags
)
1771 if (unlikely(flags
& ~SPLICE_F_ALL
))
1780 if (in
.file
->f_mode
& FMODE_READ
) {
1781 struct fd out
= fdget(fdout
);
1783 if (out
.file
->f_mode
& FMODE_WRITE
)
1784 error
= do_tee(in
.file
, out
.file
,