2 * "splice": joining two ropes together by interweaving their strands.
4 * This is the "extended pipe" functionality, where a pipe is used as
5 * an arbitrary in-memory buffer. Think of a pipe as a small kernel
6 * buffer that you can use to transfer data from one end to the other.
8 * The traditional unix read/write is extended with a "splice()" operation
9 * that transfers data buffers to or from a pipe buffer.
11 * Named by Larry McVoy, original implementation from Linus, extended by
12 * Jens to support splicing to files, network, direct splicing, etc and
13 * fixing lots of bugs.
15 * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk>
16 * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
17 * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
20 #include <linux/bvec.h>
22 #include <linux/file.h>
23 #include <linux/pagemap.h>
24 #include <linux/splice.h>
25 #include <linux/memcontrol.h>
26 #include <linux/mm_inline.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/export.h>
30 #include <linux/syscalls.h>
31 #include <linux/uio.h>
32 #include <linux/security.h>
33 #include <linux/gfp.h>
34 #include <linux/socket.h>
35 #include <linux/compat.h>
36 #include <linux/sched/signal.h>
41 * Attempt to steal a page from a pipe buffer. This should perhaps go into
42 * a vm helper function, it's already simplified quite a bit by the
43 * addition of remove_mapping(). If success is returned, the caller may
44 * attempt to reuse this page for another destination.
46 static int page_cache_pipe_buf_steal(struct pipe_inode_info
*pipe
,
47 struct pipe_buffer
*buf
)
49 struct page
*page
= buf
->page
;
50 struct address_space
*mapping
;
54 mapping
= page_mapping(page
);
56 WARN_ON(!PageUptodate(page
));
59 * At least for ext2 with nobh option, we need to wait on
60 * writeback completing on this page, since we'll remove it
61 * from the pagecache. Otherwise truncate wont wait on the
62 * page, allowing the disk blocks to be reused by someone else
63 * before we actually wrote our data to them. fs corruption
66 wait_on_page_writeback(page
);
68 if (page_has_private(page
) &&
69 !try_to_release_page(page
, GFP_KERNEL
))
73 * If we succeeded in removing the mapping, set LRU flag
76 if (remove_mapping(mapping
, page
)) {
77 buf
->flags
|= PIPE_BUF_FLAG_LRU
;
83 * Raced with truncate or failed to remove page from current
84 * address space, unlock and return failure.
91 static void page_cache_pipe_buf_release(struct pipe_inode_info
*pipe
,
92 struct pipe_buffer
*buf
)
95 buf
->flags
&= ~PIPE_BUF_FLAG_LRU
;
99 * Check whether the contents of buf is OK to access. Since the content
100 * is a page cache page, IO may be in flight.
102 static int page_cache_pipe_buf_confirm(struct pipe_inode_info
*pipe
,
103 struct pipe_buffer
*buf
)
105 struct page
*page
= buf
->page
;
108 if (!PageUptodate(page
)) {
112 * Page got truncated/unhashed. This will cause a 0-byte
113 * splice, if this is the first page.
115 if (!page
->mapping
) {
121 * Uh oh, read-error from disk.
123 if (!PageUptodate(page
)) {
129 * Page is ok afterall, we are done.
140 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
= {
160 .confirm
= generic_pipe_buf_confirm
,
161 .release
= page_cache_pipe_buf_release
,
162 .steal
= user_page_pipe_buf_steal
,
163 .get
= generic_pipe_buf_get
,
166 static void wakeup_pipe_readers(struct pipe_inode_info
*pipe
)
169 if (waitqueue_active(&pipe
->wait
))
170 wake_up_interruptible(&pipe
->wait
);
171 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
175 * splice_to_pipe - fill passed data into a pipe
176 * @pipe: pipe to fill
180 * @spd contains a map of pages and len/offset tuples, along with
181 * the struct pipe_buf_operations associated with these pages. This
182 * function will link that data to the pipe.
185 ssize_t
splice_to_pipe(struct pipe_inode_info
*pipe
,
186 struct splice_pipe_desc
*spd
)
188 unsigned int spd_pages
= spd
->nr_pages
;
189 int ret
= 0, page_nr
= 0;
194 if (unlikely(!pipe
->readers
)) {
195 send_sig(SIGPIPE
, current
, 0);
200 while (pipe
->nrbufs
< pipe
->buffers
) {
201 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
202 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
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;
215 if (!--spd
->nr_pages
)
223 while (page_nr
< spd_pages
)
224 spd
->spd_release(spd
, page_nr
++);
228 EXPORT_SYMBOL_GPL(splice_to_pipe
);
230 ssize_t
add_to_pipe(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
)
234 if (unlikely(!pipe
->readers
)) {
235 send_sig(SIGPIPE
, current
, 0);
237 } else if (pipe
->nrbufs
== pipe
->buffers
) {
240 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
241 pipe
->bufs
[newbuf
] = *buf
;
245 pipe_buf_release(pipe
, buf
);
248 EXPORT_SYMBOL(add_to_pipe
);
251 * Check if we need to grow the arrays holding pages and partial page
254 int splice_grow_spd(const struct pipe_inode_info
*pipe
, struct splice_pipe_desc
*spd
)
256 unsigned int buffers
= READ_ONCE(pipe
->buffers
);
258 spd
->nr_pages_max
= buffers
;
259 if (buffers
<= PIPE_DEF_BUFFERS
)
262 spd
->pages
= kmalloc(buffers
* sizeof(struct page
*), GFP_KERNEL
);
263 spd
->partial
= kmalloc(buffers
* sizeof(struct partial_page
), GFP_KERNEL
);
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
, ITER_PIPE
| 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
= {
329 .confirm
= generic_pipe_buf_confirm
,
330 .release
= generic_pipe_buf_release
,
331 .steal
= generic_pipe_buf_steal
,
332 .get
= generic_pipe_buf_get
,
335 int generic_pipe_buf_nosteal(struct pipe_inode_info
*pipe
,
336 struct pipe_buffer
*buf
)
341 /* Pipe buffer operations for a socket and similar. */
342 const struct pipe_buf_operations nosteal_pipe_buf_ops
= {
344 .confirm
= generic_pipe_buf_confirm
,
345 .release
= generic_pipe_buf_release
,
346 .steal
= generic_pipe_buf_nosteal
,
347 .get
= generic_pipe_buf_get
,
349 EXPORT_SYMBOL(nosteal_pipe_buf_ops
);
351 static ssize_t
kernel_readv(struct file
*file
, const struct kvec
*vec
,
352 unsigned long vlen
, loff_t offset
)
360 /* The cast to a user pointer is valid due to the set_fs() */
361 res
= vfs_readv(file
, (const struct iovec __user
*)vec
, vlen
, &pos
, 0);
367 static ssize_t
default_file_splice_read(struct file
*in
, loff_t
*ppos
,
368 struct pipe_inode_info
*pipe
, size_t len
,
371 struct kvec
*vec
, __vec
[PIPE_DEF_BUFFERS
];
374 unsigned int nr_pages
;
375 size_t offset
, base
, copied
= 0;
379 if (pipe
->nrbufs
== pipe
->buffers
)
383 * Try to keep page boundaries matching to source pagecache ones -
384 * it probably won't be much help, but...
386 offset
= *ppos
& ~PAGE_MASK
;
388 iov_iter_pipe(&to
, ITER_PIPE
| READ
, pipe
, len
+ offset
);
390 res
= iov_iter_get_pages_alloc(&to
, &pages
, len
+ offset
, &base
);
394 nr_pages
= DIV_ROUND_UP(res
+ base
, PAGE_SIZE
);
397 if (nr_pages
> PIPE_DEF_BUFFERS
) {
398 vec
= kmalloc(nr_pages
* sizeof(struct kvec
), GFP_KERNEL
);
399 if (unlikely(!vec
)) {
405 pipe
->bufs
[to
.idx
].offset
= offset
;
406 pipe
->bufs
[to
.idx
].len
-= offset
;
408 for (i
= 0; i
< nr_pages
; i
++) {
409 size_t this_len
= min_t(size_t, len
, PAGE_SIZE
- offset
);
410 vec
[i
].iov_base
= page_address(pages
[i
]) + offset
;
411 vec
[i
].iov_len
= this_len
;
416 res
= kernel_readv(in
, vec
, nr_pages
, *ppos
);
425 for (i
= 0; i
< nr_pages
; i
++)
428 iov_iter_advance(&to
, copied
); /* truncates and discards */
433 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
434 * using sendpage(). Return the number of bytes sent.
436 static int pipe_to_sendpage(struct pipe_inode_info
*pipe
,
437 struct pipe_buffer
*buf
, struct splice_desc
*sd
)
439 struct file
*file
= sd
->u
.file
;
440 loff_t pos
= sd
->pos
;
443 if (!likely(file
->f_op
->sendpage
))
446 more
= (sd
->flags
& SPLICE_F_MORE
) ? MSG_MORE
: 0;
448 if (sd
->len
< sd
->total_len
&& pipe
->nrbufs
> 1)
449 more
|= MSG_SENDPAGE_NOTLAST
;
451 return file
->f_op
->sendpage(file
, buf
->page
, buf
->offset
,
452 sd
->len
, &pos
, more
);
455 static void wakeup_pipe_writers(struct pipe_inode_info
*pipe
)
458 if (waitqueue_active(&pipe
->wait
))
459 wake_up_interruptible(&pipe
->wait
);
460 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
464 * splice_from_pipe_feed - feed available data from a pipe to a file
465 * @pipe: pipe to splice from
466 * @sd: information to @actor
467 * @actor: handler that splices the data
470 * This function loops over the pipe and calls @actor to do the
471 * actual moving of a single struct pipe_buffer to the desired
472 * destination. It returns when there's no more buffers left in
473 * the pipe or if the requested number of bytes (@sd->total_len)
474 * have been copied. It returns a positive number (one) if the
475 * pipe needs to be filled with more data, zero if the required
476 * number of bytes have been copied and -errno on error.
478 * This, together with splice_from_pipe_{begin,end,next}, may be
479 * used to implement the functionality of __splice_from_pipe() when
480 * locking is required around copying the pipe buffers to the
483 static int splice_from_pipe_feed(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
488 while (pipe
->nrbufs
) {
489 struct pipe_buffer
*buf
= pipe
->bufs
+ pipe
->curbuf
;
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
);
516 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
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
->nrbufs
) {
552 if (!pipe
->waiting_writers
&& sd
->num_spliced
)
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
->buffers
;
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
;
707 ret
= splice_from_pipe_next(pipe
, &sd
);
711 if (unlikely(nbufs
< pipe
->buffers
)) {
713 nbufs
= pipe
->buffers
;
714 array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
722 /* build the vector */
724 for (n
= 0, idx
= pipe
->curbuf
; left
&& n
< pipe
->nrbufs
; n
++, idx
++) {
725 struct pipe_buffer
*buf
= pipe
->bufs
+ idx
;
726 size_t this_len
= buf
->len
;
731 if (idx
== pipe
->buffers
- 1)
734 ret
= pipe_buf_confirm(pipe
, buf
);
741 array
[n
].bv_page
= buf
->page
;
742 array
[n
].bv_len
= this_len
;
743 array
[n
].bv_offset
= buf
->offset
;
747 iov_iter_bvec(&from
, ITER_BVEC
| WRITE
, array
, n
,
748 sd
.total_len
- left
);
749 ret
= vfs_iter_write(out
, &from
, &sd
.pos
, 0);
753 sd
.num_spliced
+= ret
;
757 /* dismiss the fully eaten buffers, adjust the partial one */
759 struct pipe_buffer
*buf
= pipe
->bufs
+ pipe
->curbuf
;
760 if (ret
>= buf
->len
) {
763 pipe_buf_release(pipe
, buf
);
764 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
767 sd
.need_wakeup
= true;
777 splice_from_pipe_end(pipe
, &sd
);
782 ret
= sd
.num_spliced
;
787 EXPORT_SYMBOL(iter_file_splice_write
);
789 static int write_pipe_buf(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
790 struct splice_desc
*sd
)
794 loff_t tmp
= sd
->pos
;
796 data
= kmap(buf
->page
);
797 ret
= __kernel_write(sd
->u
.file
, data
+ buf
->offset
, sd
->len
, &tmp
);
803 static ssize_t
default_file_splice_write(struct pipe_inode_info
*pipe
,
804 struct file
*out
, loff_t
*ppos
,
805 size_t len
, unsigned int flags
)
809 ret
= splice_from_pipe(pipe
, out
, ppos
, len
, flags
, write_pipe_buf
);
817 * generic_splice_sendpage - splice data from a pipe to a socket
818 * @pipe: pipe to splice from
819 * @out: socket to write to
820 * @ppos: position in @out
821 * @len: number of bytes to splice
822 * @flags: splice modifier flags
825 * Will send @len bytes from the pipe to a network socket. No data copying
829 ssize_t
generic_splice_sendpage(struct pipe_inode_info
*pipe
, struct file
*out
,
830 loff_t
*ppos
, size_t len
, unsigned int flags
)
832 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_sendpage
);
835 EXPORT_SYMBOL(generic_splice_sendpage
);
838 * Attempt to initiate a splice from pipe to file.
840 long do_splice_from(struct pipe_inode_info
*pipe
, struct file
*out
,
841 loff_t
*ppos
, size_t len
, unsigned int flags
)
843 ssize_t (*splice_write
)(struct pipe_inode_info
*, struct file
*,
844 loff_t
*, size_t, unsigned int);
846 if (out
->f_op
->splice_write
)
847 splice_write
= out
->f_op
->splice_write
;
849 splice_write
= default_file_splice_write
;
851 return splice_write(pipe
, out
, ppos
, len
, flags
);
853 EXPORT_SYMBOL_GPL(do_splice_from
);
856 * Attempt to initiate a splice from a file to a pipe.
858 long do_splice_to(struct file
*in
, loff_t
*ppos
,
859 struct pipe_inode_info
*pipe
, size_t len
,
862 ssize_t (*splice_read
)(struct file
*, loff_t
*,
863 struct pipe_inode_info
*, size_t, unsigned int);
866 if (unlikely(!(in
->f_mode
& FMODE_READ
)))
869 ret
= rw_verify_area(READ
, in
, ppos
, len
);
870 if (unlikely(ret
< 0))
873 if (unlikely(len
> MAX_RW_COUNT
))
876 if (in
->f_op
->splice_read
)
877 splice_read
= in
->f_op
->splice_read
;
879 splice_read
= default_file_splice_read
;
881 return splice_read(in
, ppos
, pipe
, len
, flags
);
883 EXPORT_SYMBOL_GPL(do_splice_to
);
886 * splice_direct_to_actor - splices data directly between two non-pipes
887 * @in: file to splice from
888 * @sd: actor information on where to splice to
889 * @actor: handles the data splicing
892 * This is a special case helper to splice directly between two
893 * points, without requiring an explicit pipe. Internally an allocated
894 * pipe is cached in the process, and reused during the lifetime of
898 ssize_t
splice_direct_to_actor(struct file
*in
, struct splice_desc
*sd
,
899 splice_direct_actor
*actor
)
901 struct pipe_inode_info
*pipe
;
908 * We require the input being a regular file, as we don't want to
909 * randomly drop data for eg socket -> socket splicing. Use the
910 * piped splicing for that!
912 i_mode
= file_inode(in
)->i_mode
;
913 if (unlikely(!S_ISREG(i_mode
) && !S_ISBLK(i_mode
)))
917 * neither in nor out is a pipe, setup an internal pipe attached to
918 * 'out' and transfer the wanted data from 'in' to 'out' through that
920 pipe
= current
->splice_pipe
;
921 if (unlikely(!pipe
)) {
922 pipe
= alloc_pipe_info();
927 * We don't have an immediate reader, but we'll read the stuff
928 * out of the pipe right after the splice_to_pipe(). So set
929 * PIPE_READERS appropriately.
933 current
->splice_pipe
= pipe
;
945 * Don't block on output, we have to drain the direct pipe.
947 sd
->flags
&= ~SPLICE_F_NONBLOCK
;
948 more
= sd
->flags
& SPLICE_F_MORE
;
950 WARN_ON_ONCE(pipe
->nrbufs
!= 0);
953 unsigned int pipe_pages
;
955 loff_t pos
= sd
->pos
, prev_pos
= pos
;
957 /* Don't try to read more the pipe has space for. */
958 pipe_pages
= pipe
->buffers
- pipe
->nrbufs
;
959 read_len
= min(len
, (size_t)pipe_pages
<< PAGE_SHIFT
);
960 ret
= do_splice_to(in
, &pos
, pipe
, read_len
, flags
);
961 if (unlikely(ret
<= 0))
965 sd
->total_len
= read_len
;
968 * If more data is pending, set SPLICE_F_MORE
969 * If this is the last data and SPLICE_F_MORE was not set
970 * initially, clears it.
973 sd
->flags
|= SPLICE_F_MORE
;
975 sd
->flags
&= ~SPLICE_F_MORE
;
977 * NOTE: nonblocking mode only applies to the input. We
978 * must not do the output in nonblocking mode as then we
979 * could get stuck data in the internal pipe:
981 ret
= actor(pipe
, sd
);
982 if (unlikely(ret
<= 0)) {
991 if (ret
< read_len
) {
992 sd
->pos
= prev_pos
+ ret
;
998 pipe
->nrbufs
= pipe
->curbuf
= 0;
1004 * If we did an incomplete transfer we must release
1005 * the pipe buffers in question:
1007 for (i
= 0; i
< pipe
->buffers
; i
++) {
1008 struct pipe_buffer
*buf
= pipe
->bufs
+ i
;
1011 pipe_buf_release(pipe
, buf
);
1019 EXPORT_SYMBOL(splice_direct_to_actor
);
1021 static int direct_splice_actor(struct pipe_inode_info
*pipe
,
1022 struct splice_desc
*sd
)
1024 struct file
*file
= sd
->u
.file
;
1026 return do_splice_from(pipe
, file
, sd
->opos
, sd
->total_len
,
1031 * do_splice_direct - splices data directly between two files
1032 * @in: file to splice from
1033 * @ppos: input file offset
1034 * @out: file to splice to
1035 * @opos: output file offset
1036 * @len: number of bytes to splice
1037 * @flags: splice modifier flags
1040 * For use by do_sendfile(). splice can easily emulate sendfile, but
1041 * doing it in the application would incur an extra system call
1042 * (splice in + splice out, as compared to just sendfile()). So this helper
1043 * can splice directly through a process-private pipe.
1046 long do_splice_direct(struct file
*in
, loff_t
*ppos
, struct file
*out
,
1047 loff_t
*opos
, size_t len
, unsigned int flags
)
1049 struct splice_desc sd
= {
1059 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
1062 if (unlikely(out
->f_flags
& O_APPEND
))
1065 ret
= rw_verify_area(WRITE
, out
, opos
, len
);
1066 if (unlikely(ret
< 0))
1069 ret
= splice_direct_to_actor(in
, &sd
, direct_splice_actor
);
1075 EXPORT_SYMBOL(do_splice_direct
);
1077 static int wait_for_space(struct pipe_inode_info
*pipe
, unsigned flags
)
1080 if (unlikely(!pipe
->readers
)) {
1081 send_sig(SIGPIPE
, current
, 0);
1084 if (pipe
->nrbufs
!= pipe
->buffers
)
1086 if (flags
& SPLICE_F_NONBLOCK
)
1088 if (signal_pending(current
))
1089 return -ERESTARTSYS
;
1090 pipe
->waiting_writers
++;
1092 pipe
->waiting_writers
--;
1096 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1097 struct pipe_inode_info
*opipe
,
1098 size_t len
, unsigned int flags
);
1101 * Determine where to splice to/from.
1103 static long do_splice(struct file
*in
, loff_t __user
*off_in
,
1104 struct file
*out
, loff_t __user
*off_out
,
1105 size_t len
, unsigned int flags
)
1107 struct pipe_inode_info
*ipipe
;
1108 struct pipe_inode_info
*opipe
;
1112 ipipe
= get_pipe_info(in
);
1113 opipe
= get_pipe_info(out
);
1115 if (ipipe
&& opipe
) {
1116 if (off_in
|| off_out
)
1119 if (!(in
->f_mode
& FMODE_READ
))
1122 if (!(out
->f_mode
& FMODE_WRITE
))
1125 /* Splicing to self would be fun, but... */
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 file_start_write(out
);
1155 ret
= do_splice_from(ipipe
, out
, &offset
, len
, flags
);
1156 file_end_write(out
);
1159 out
->f_pos
= offset
;
1160 else if (copy_to_user(off_out
, &offset
, sizeof(loff_t
)))
1170 if (!(in
->f_mode
& FMODE_PREAD
))
1172 if (copy_from_user(&offset
, off_in
, sizeof(loff_t
)))
1179 ret
= wait_for_space(opipe
, flags
);
1181 unsigned int pipe_pages
;
1183 /* Don't try to read more the pipe has space for. */
1184 pipe_pages
= opipe
->buffers
- opipe
->nrbufs
;
1185 len
= min(len
, (size_t)pipe_pages
<< PAGE_SHIFT
);
1187 ret
= do_splice_to(in
, &offset
, opipe
, len
, flags
);
1191 wakeup_pipe_readers(opipe
);
1194 else if (copy_to_user(off_in
, &offset
, sizeof(loff_t
)))
1203 static int iter_to_pipe(struct iov_iter
*from
,
1204 struct pipe_inode_info
*pipe
,
1207 struct pipe_buffer buf
= {
1208 .ops
= &user_page_pipe_buf_ops
,
1213 bool failed
= false;
1215 while (iov_iter_count(from
) && !failed
) {
1216 struct page
*pages
[16];
1221 copied
= iov_iter_get_pages(from
, pages
, ~0UL, 16, &start
);
1227 for (n
= 0; copied
; n
++, start
= 0) {
1228 int size
= min_t(int, copied
, PAGE_SIZE
- start
);
1230 buf
.page
= pages
[n
];
1233 ret
= add_to_pipe(pipe
, &buf
);
1234 if (unlikely(ret
< 0)) {
1237 iov_iter_advance(from
, ret
);
1246 return total
? total
: ret
;
1249 static int pipe_to_user(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
1250 struct splice_desc
*sd
)
1252 int n
= copy_page_to_iter(buf
->page
, buf
->offset
, sd
->len
, sd
->u
.data
);
1253 return n
== sd
->len
? n
: -EFAULT
;
1257 * For lack of a better implementation, implement vmsplice() to userspace
1258 * as a simple copy of the pipes pages to the user iov.
1260 static long vmsplice_to_user(struct file
*file
, const struct iovec __user
*uiov
,
1261 unsigned long nr_segs
, unsigned int flags
)
1263 struct pipe_inode_info
*pipe
;
1264 struct splice_desc sd
;
1266 struct iovec iovstack
[UIO_FASTIOV
];
1267 struct iovec
*iov
= iovstack
;
1268 struct iov_iter iter
;
1270 pipe
= get_pipe_info(file
);
1274 ret
= import_iovec(READ
, uiov
, nr_segs
,
1275 ARRAY_SIZE(iovstack
), &iov
, &iter
);
1279 sd
.total_len
= iov_iter_count(&iter
);
1287 ret
= __splice_from_pipe(pipe
, &sd
, pipe_to_user
);
1296 * vmsplice splices a user address range into a pipe. It can be thought of
1297 * as splice-from-memory, where the regular splice is splice-from-file (or
1298 * to file). In both cases the output is a pipe, naturally.
1300 static long vmsplice_to_pipe(struct file
*file
, const struct iovec __user
*uiov
,
1301 unsigned long nr_segs
, unsigned int flags
)
1303 struct pipe_inode_info
*pipe
;
1304 struct iovec iovstack
[UIO_FASTIOV
];
1305 struct iovec
*iov
= iovstack
;
1306 struct iov_iter from
;
1308 unsigned buf_flag
= 0;
1310 if (flags
& SPLICE_F_GIFT
)
1311 buf_flag
= PIPE_BUF_FLAG_GIFT
;
1313 pipe
= get_pipe_info(file
);
1317 ret
= import_iovec(WRITE
, uiov
, nr_segs
,
1318 ARRAY_SIZE(iovstack
), &iov
, &from
);
1323 ret
= wait_for_space(pipe
, flags
);
1325 ret
= iter_to_pipe(&from
, pipe
, buf_flag
);
1328 wakeup_pipe_readers(pipe
);
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 SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct iovec __user
*, iov
,
1350 unsigned long, nr_segs
, unsigned int, flags
)
1355 if (unlikely(flags
& ~SPLICE_F_ALL
))
1357 if (unlikely(nr_segs
> UIO_MAXIOV
))
1359 else if (unlikely(!nr_segs
))
1365 if (f
.file
->f_mode
& FMODE_WRITE
)
1366 error
= vmsplice_to_pipe(f
.file
, iov
, nr_segs
, flags
);
1367 else if (f
.file
->f_mode
& FMODE_READ
)
1368 error
= vmsplice_to_user(f
.file
, iov
, nr_segs
, flags
);
1376 #ifdef CONFIG_COMPAT
1377 COMPAT_SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct compat_iovec __user
*, iov32
,
1378 unsigned int, nr_segs
, unsigned int, flags
)
1381 struct iovec __user
*iov
;
1382 if (nr_segs
> UIO_MAXIOV
)
1384 iov
= compat_alloc_user_space(nr_segs
* sizeof(struct iovec
));
1385 for (i
= 0; i
< nr_segs
; i
++) {
1386 struct compat_iovec v
;
1387 if (get_user(v
.iov_base
, &iov32
[i
].iov_base
) ||
1388 get_user(v
.iov_len
, &iov32
[i
].iov_len
) ||
1389 put_user(compat_ptr(v
.iov_base
), &iov
[i
].iov_base
) ||
1390 put_user(v
.iov_len
, &iov
[i
].iov_len
))
1393 return sys_vmsplice(fd
, iov
, nr_segs
, flags
);
1397 SYSCALL_DEFINE6(splice
, int, fd_in
, loff_t __user
*, off_in
,
1398 int, fd_out
, loff_t __user
*, off_out
,
1399 size_t, len
, unsigned int, flags
)
1407 if (unlikely(flags
& ~SPLICE_F_ALL
))
1413 if (in
.file
->f_mode
& FMODE_READ
) {
1414 out
= fdget(fd_out
);
1416 if (out
.file
->f_mode
& FMODE_WRITE
)
1417 error
= do_splice(in
.file
, off_in
,
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 ->nrbufs without the inode lock first. This function
1438 * is speculative anyways, so missing one is ok.
1446 while (!pipe
->nrbufs
) {
1447 if (signal_pending(current
)) {
1453 if (!pipe
->waiting_writers
) {
1454 if (flags
& SPLICE_F_NONBLOCK
) {
1467 * Make sure there's writeable room. Wait for room if we can, otherwise
1468 * return an appropriate error.
1470 static int opipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1475 * Check ->nrbufs without the inode lock first. This function
1476 * is speculative anyways, so missing one is ok.
1478 if (pipe
->nrbufs
< pipe
->buffers
)
1484 while (pipe
->nrbufs
>= pipe
->buffers
) {
1485 if (!pipe
->readers
) {
1486 send_sig(SIGPIPE
, current
, 0);
1490 if (flags
& SPLICE_F_NONBLOCK
) {
1494 if (signal_pending(current
)) {
1498 pipe
->waiting_writers
++;
1500 pipe
->waiting_writers
--;
1508 * Splice contents of ipipe to opipe.
1510 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1511 struct pipe_inode_info
*opipe
,
1512 size_t len
, unsigned int flags
)
1514 struct pipe_buffer
*ibuf
, *obuf
;
1516 bool input_wakeup
= false;
1520 ret
= ipipe_prep(ipipe
, flags
);
1524 ret
= opipe_prep(opipe
, flags
);
1529 * Potential ABBA deadlock, work around it by ordering lock
1530 * grabbing by pipe info address. Otherwise two different processes
1531 * could deadlock (one doing tee from A -> B, the other from B -> A).
1533 pipe_double_lock(ipipe
, opipe
);
1536 if (!opipe
->readers
) {
1537 send_sig(SIGPIPE
, current
, 0);
1543 if (!ipipe
->nrbufs
&& !ipipe
->writers
)
1547 * Cannot make any progress, because either the input
1548 * pipe is empty or the output pipe is full.
1550 if (!ipipe
->nrbufs
|| opipe
->nrbufs
>= opipe
->buffers
) {
1551 /* Already processed some buffers, break */
1555 if (flags
& SPLICE_F_NONBLOCK
) {
1561 * We raced with another reader/writer and haven't
1562 * managed to process any buffers. A zero return
1563 * value means EOF, so retry instead.
1570 ibuf
= ipipe
->bufs
+ ipipe
->curbuf
;
1571 nbuf
= (opipe
->curbuf
+ opipe
->nrbufs
) & (opipe
->buffers
- 1);
1572 obuf
= opipe
->bufs
+ nbuf
;
1574 if (len
>= ibuf
->len
) {
1576 * Simply move the whole buffer from ipipe to opipe
1581 ipipe
->curbuf
= (ipipe
->curbuf
+ 1) & (ipipe
->buffers
- 1);
1583 input_wakeup
= true;
1586 * Get a reference to this pipe buffer,
1587 * so we can copy the contents over.
1589 if (!pipe_buf_get(ipipe
, ibuf
)) {
1597 * Don't inherit the gift flag, we need to
1598 * prevent multiple steals of this page.
1600 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1602 pipe_buf_mark_unmergeable(obuf
);
1606 ibuf
->offset
+= obuf
->len
;
1607 ibuf
->len
-= obuf
->len
;
1617 * If we put data in the output pipe, wakeup any potential readers.
1620 wakeup_pipe_readers(opipe
);
1623 wakeup_pipe_writers(ipipe
);
1629 * Link contents of ipipe to opipe.
1631 static int link_pipe(struct pipe_inode_info
*ipipe
,
1632 struct pipe_inode_info
*opipe
,
1633 size_t len
, unsigned int flags
)
1635 struct pipe_buffer
*ibuf
, *obuf
;
1636 int ret
= 0, i
= 0, nbuf
;
1639 * Potential ABBA deadlock, work around it by ordering lock
1640 * grabbing by pipe info address. Otherwise two different processes
1641 * could deadlock (one doing tee from A -> B, the other from B -> A).
1643 pipe_double_lock(ipipe
, opipe
);
1646 if (!opipe
->readers
) {
1647 send_sig(SIGPIPE
, current
, 0);
1654 * If we have iterated all input buffers or ran out of
1655 * output room, break.
1657 if (i
>= ipipe
->nrbufs
|| opipe
->nrbufs
>= opipe
->buffers
)
1660 ibuf
= ipipe
->bufs
+ ((ipipe
->curbuf
+ i
) & (ipipe
->buffers
-1));
1661 nbuf
= (opipe
->curbuf
+ opipe
->nrbufs
) & (opipe
->buffers
- 1);
1664 * Get a reference to this pipe buffer,
1665 * so we can copy the contents over.
1667 if (!pipe_buf_get(ipipe
, ibuf
)) {
1673 obuf
= opipe
->bufs
+ nbuf
;
1677 * Don't inherit the gift flag, we need to
1678 * prevent multiple steals of this page.
1680 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1682 pipe_buf_mark_unmergeable(obuf
);
1684 if (obuf
->len
> len
)
1694 * return EAGAIN if we have the potential of some data in the
1695 * future, otherwise just return 0
1697 if (!ret
&& ipipe
->waiting_writers
&& (flags
& SPLICE_F_NONBLOCK
))
1704 * If we put data in the output pipe, wakeup any potential readers.
1707 wakeup_pipe_readers(opipe
);
1713 * This is a tee(1) implementation that works on pipes. It doesn't copy
1714 * any data, it simply references the 'in' pages on the 'out' pipe.
1715 * The 'flags' used are the SPLICE_F_* variants, currently the only
1716 * applicable one is SPLICE_F_NONBLOCK.
1718 static long do_tee(struct file
*in
, struct file
*out
, size_t len
,
1721 struct pipe_inode_info
*ipipe
= get_pipe_info(in
);
1722 struct pipe_inode_info
*opipe
= get_pipe_info(out
);
1726 * Duplicate the contents of ipipe to opipe without actually
1729 if (ipipe
&& opipe
&& ipipe
!= opipe
) {
1731 * Keep going, unless we encounter an error. The ipipe/opipe
1732 * ordering doesn't really matter.
1734 ret
= ipipe_prep(ipipe
, flags
);
1736 ret
= opipe_prep(opipe
, flags
);
1738 ret
= link_pipe(ipipe
, opipe
, len
, flags
);
1745 SYSCALL_DEFINE4(tee
, int, fdin
, int, fdout
, size_t, len
, unsigned int, flags
)
1750 if (unlikely(flags
& ~SPLICE_F_ALL
))
1759 if (in
.file
->f_mode
& FMODE_READ
) {
1760 struct fd out
= fdget(fdout
);
1762 if (out
.file
->f_mode
& FMODE_WRITE
)
1763 error
= do_tee(in
.file
, out
.file
,