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/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 bool page_cache_pipe_buf_try_steal(struct pipe_inode_info
*pipe
,
47 struct pipe_buffer
*buf
)
49 struct folio
*folio
= page_folio(buf
->page
);
50 struct address_space
*mapping
;
54 mapping
= folio_mapping(folio
);
56 WARN_ON(!folio_test_uptodate(folio
));
59 * At least for ext2 with nobh option, we need to wait on
60 * writeback completing on this folio, since we'll remove it
61 * from the pagecache. Otherwise truncate wont wait on the
62 * folio, allowing the disk blocks to be reused by someone else
63 * before we actually wrote our data to them. fs corruption
66 folio_wait_writeback(folio
);
68 if (folio_has_private(folio
) &&
69 !filemap_release_folio(folio
, GFP_KERNEL
))
73 * If we succeeded in removing the mapping, set LRU flag
76 if (remove_mapping(mapping
, folio
)) {
77 buf
->flags
|= PIPE_BUF_FLAG_LRU
;
83 * Raced with truncate or failed to remove folio 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
= {
141 .confirm
= page_cache_pipe_buf_confirm
,
142 .release
= page_cache_pipe_buf_release
,
143 .try_steal
= page_cache_pipe_buf_try_steal
,
144 .get
= generic_pipe_buf_get
,
147 static bool user_page_pipe_buf_try_steal(struct pipe_inode_info
*pipe
,
148 struct pipe_buffer
*buf
)
150 if (!(buf
->flags
& PIPE_BUF_FLAG_GIFT
))
153 buf
->flags
|= PIPE_BUF_FLAG_LRU
;
154 return generic_pipe_buf_try_steal(pipe
, buf
);
157 static const struct pipe_buf_operations user_page_pipe_buf_ops
= {
158 .release
= page_cache_pipe_buf_release
,
159 .try_steal
= user_page_pipe_buf_try_steal
,
160 .get
= generic_pipe_buf_get
,
163 static void wakeup_pipe_readers(struct pipe_inode_info
*pipe
)
166 if (waitqueue_active(&pipe
->rd_wait
))
167 wake_up_interruptible(&pipe
->rd_wait
);
168 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
172 * splice_to_pipe - fill passed data into a pipe
173 * @pipe: pipe to fill
177 * @spd contains a map of pages and len/offset tuples, along with
178 * the struct pipe_buf_operations associated with these pages. This
179 * function will link that data to the pipe.
182 ssize_t
splice_to_pipe(struct pipe_inode_info
*pipe
,
183 struct splice_pipe_desc
*spd
)
185 unsigned int spd_pages
= spd
->nr_pages
;
186 unsigned int tail
= pipe
->tail
;
187 unsigned int head
= pipe
->head
;
188 unsigned int mask
= pipe
->ring_size
- 1;
189 int ret
= 0, page_nr
= 0;
194 if (unlikely(!pipe
->readers
)) {
195 send_sig(SIGPIPE
, current
, 0);
200 while (!pipe_full(head
, tail
, pipe
->max_usage
)) {
201 struct pipe_buffer
*buf
= &pipe
->bufs
[head
& mask
];
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;
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
)
232 unsigned int head
= pipe
->head
;
233 unsigned int tail
= pipe
->tail
;
234 unsigned int mask
= pipe
->ring_size
- 1;
237 if (unlikely(!pipe
->readers
)) {
238 send_sig(SIGPIPE
, current
, 0);
240 } else if (pipe_full(head
, tail
, pipe
->max_usage
)) {
243 pipe
->bufs
[head
& mask
] = *buf
;
244 pipe
->head
= head
+ 1;
247 pipe_buf_release(pipe
, buf
);
250 EXPORT_SYMBOL(add_to_pipe
);
253 * Check if we need to grow the arrays holding pages and partial page
256 int splice_grow_spd(const struct pipe_inode_info
*pipe
, struct splice_pipe_desc
*spd
)
258 unsigned int max_usage
= READ_ONCE(pipe
->max_usage
);
260 spd
->nr_pages_max
= max_usage
;
261 if (max_usage
<= PIPE_DEF_BUFFERS
)
264 spd
->pages
= kmalloc_array(max_usage
, sizeof(struct page
*), GFP_KERNEL
);
265 spd
->partial
= kmalloc_array(max_usage
, sizeof(struct partial_page
),
268 if (spd
->pages
&& spd
->partial
)
276 void splice_shrink_spd(struct splice_pipe_desc
*spd
)
278 if (spd
->nr_pages_max
<= PIPE_DEF_BUFFERS
)
286 * Splice data from an O_DIRECT file into pages and then add them to the output
289 ssize_t
direct_splice_read(struct file
*in
, loff_t
*ppos
,
290 struct pipe_inode_info
*pipe
,
291 size_t len
, unsigned int flags
)
298 size_t used
, npages
, chunk
, remain
, reclaim
;
301 /* Work out how much data we can actually add into the pipe */
302 used
= pipe_occupancy(pipe
->head
, pipe
->tail
);
303 npages
= max_t(ssize_t
, pipe
->max_usage
- used
, 0);
304 len
= min_t(size_t, len
, npages
* PAGE_SIZE
);
305 npages
= DIV_ROUND_UP(len
, PAGE_SIZE
);
307 bv
= kzalloc(array_size(npages
, sizeof(bv
[0])) +
308 array_size(npages
, sizeof(struct page
*)), GFP_KERNEL
);
312 pages
= (void *)(bv
+ npages
);
313 npages
= alloc_pages_bulk_array(GFP_USER
, npages
, pages
);
319 remain
= len
= min_t(size_t, len
, npages
* PAGE_SIZE
);
321 for (i
= 0; i
< npages
; i
++) {
322 chunk
= min_t(size_t, PAGE_SIZE
, remain
);
323 bv
[i
].bv_page
= pages
[i
];
325 bv
[i
].bv_len
= chunk
;
330 iov_iter_bvec(&to
, ITER_DEST
, bv
, npages
, len
);
331 init_sync_kiocb(&kiocb
, in
);
332 kiocb
.ki_pos
= *ppos
;
333 ret
= call_read_iter(in
, &kiocb
, &to
);
335 reclaim
= npages
* PAGE_SIZE
;
340 *ppos
= kiocb
.ki_pos
;
342 } else if (ret
< 0) {
344 * callers of ->splice_read() expect -EAGAIN on
345 * "can't put anything in there", rather than -EFAULT.
351 /* Free any pages that didn't get touched at all. */
352 reclaim
/= PAGE_SIZE
;
355 release_pages(pages
+ npages
, reclaim
);
358 /* Push the remaining pages into the pipe. */
359 for (i
= 0; i
< npages
; i
++) {
360 struct pipe_buffer
*buf
= pipe_head_buf(pipe
);
362 chunk
= min_t(size_t, remain
, PAGE_SIZE
);
363 *buf
= (struct pipe_buffer
) {
364 .ops
= &default_pipe_buf_ops
,
365 .page
= bv
[i
].bv_page
,
376 EXPORT_SYMBOL(direct_splice_read
);
379 * generic_file_splice_read - splice data from file to a pipe
380 * @in: file to splice from
381 * @ppos: position in @in
382 * @pipe: pipe to splice to
383 * @len: number of bytes to splice
384 * @flags: splice modifier flags
387 * Will read pages from given file and fill them into a pipe. Can be
388 * used as long as it has more or less sane ->read_iter().
391 ssize_t
generic_file_splice_read(struct file
*in
, loff_t
*ppos
,
392 struct pipe_inode_info
*pipe
, size_t len
,
399 iov_iter_pipe(&to
, ITER_DEST
, pipe
, len
);
400 init_sync_kiocb(&kiocb
, in
);
401 kiocb
.ki_pos
= *ppos
;
402 ret
= call_read_iter(in
, &kiocb
, &to
);
404 *ppos
= kiocb
.ki_pos
;
406 } else if (ret
< 0) {
407 /* free what was emitted */
408 pipe_discard_from(pipe
, to
.start_head
);
410 * callers of ->splice_read() expect -EAGAIN on
411 * "can't put anything in there", rather than -EFAULT.
419 EXPORT_SYMBOL(generic_file_splice_read
);
421 const struct pipe_buf_operations default_pipe_buf_ops
= {
422 .release
= generic_pipe_buf_release
,
423 .try_steal
= generic_pipe_buf_try_steal
,
424 .get
= generic_pipe_buf_get
,
427 /* Pipe buffer operations for a socket and similar. */
428 const struct pipe_buf_operations nosteal_pipe_buf_ops
= {
429 .release
= generic_pipe_buf_release
,
430 .get
= generic_pipe_buf_get
,
432 EXPORT_SYMBOL(nosteal_pipe_buf_ops
);
435 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
436 * using sendpage(). Return the number of bytes sent.
438 static int pipe_to_sendpage(struct pipe_inode_info
*pipe
,
439 struct pipe_buffer
*buf
, struct splice_desc
*sd
)
441 struct file
*file
= sd
->u
.file
;
442 loff_t pos
= sd
->pos
;
445 if (!likely(file
->f_op
->sendpage
))
448 more
= (sd
->flags
& SPLICE_F_MORE
) ? MSG_MORE
: 0;
450 if (sd
->len
< sd
->total_len
&&
451 pipe_occupancy(pipe
->head
, pipe
->tail
) > 1)
452 more
|= MSG_SENDPAGE_NOTLAST
;
454 return file
->f_op
->sendpage(file
, buf
->page
, buf
->offset
,
455 sd
->len
, &pos
, more
);
458 static void wakeup_pipe_writers(struct pipe_inode_info
*pipe
)
461 if (waitqueue_active(&pipe
->wr_wait
))
462 wake_up_interruptible(&pipe
->wr_wait
);
463 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
467 * splice_from_pipe_feed - feed available data from a pipe to a file
468 * @pipe: pipe to splice from
469 * @sd: information to @actor
470 * @actor: handler that splices the data
473 * This function loops over the pipe and calls @actor to do the
474 * actual moving of a single struct pipe_buffer to the desired
475 * destination. It returns when there's no more buffers left in
476 * the pipe or if the requested number of bytes (@sd->total_len)
477 * have been copied. It returns a positive number (one) if the
478 * pipe needs to be filled with more data, zero if the required
479 * number of bytes have been copied and -errno on error.
481 * This, together with splice_from_pipe_{begin,end,next}, may be
482 * used to implement the functionality of __splice_from_pipe() when
483 * locking is required around copying the pipe buffers to the
486 static int splice_from_pipe_feed(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
489 unsigned int head
= pipe
->head
;
490 unsigned int tail
= pipe
->tail
;
491 unsigned int mask
= pipe
->ring_size
- 1;
494 while (!pipe_empty(head
, tail
)) {
495 struct pipe_buffer
*buf
= &pipe
->bufs
[tail
& mask
];
498 if (sd
->len
> sd
->total_len
)
499 sd
->len
= sd
->total_len
;
501 ret
= pipe_buf_confirm(pipe
, buf
);
508 ret
= actor(pipe
, buf
, sd
);
515 sd
->num_spliced
+= ret
;
518 sd
->total_len
-= ret
;
521 pipe_buf_release(pipe
, buf
);
525 sd
->need_wakeup
= true;
535 /* We know we have a pipe buffer, but maybe it's empty? */
536 static inline bool eat_empty_buffer(struct pipe_inode_info
*pipe
)
538 unsigned int tail
= pipe
->tail
;
539 unsigned int mask
= pipe
->ring_size
- 1;
540 struct pipe_buffer
*buf
= &pipe
->bufs
[tail
& mask
];
542 if (unlikely(!buf
->len
)) {
543 pipe_buf_release(pipe
, buf
);
552 * splice_from_pipe_next - wait for some data to splice from
553 * @pipe: pipe to splice from
554 * @sd: information about the splice operation
557 * This function will wait for some data and return a positive
558 * value (one) if pipe buffers are available. It will return zero
559 * or -errno if no more data needs to be spliced.
561 static int splice_from_pipe_next(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
564 * Check for signal early to make process killable when there are
565 * always buffers available
567 if (signal_pending(current
))
571 while (pipe_empty(pipe
->head
, pipe
->tail
)) {
578 if (sd
->flags
& SPLICE_F_NONBLOCK
)
581 if (signal_pending(current
))
584 if (sd
->need_wakeup
) {
585 wakeup_pipe_writers(pipe
);
586 sd
->need_wakeup
= false;
589 pipe_wait_readable(pipe
);
592 if (eat_empty_buffer(pipe
))
599 * splice_from_pipe_begin - start splicing from pipe
600 * @sd: information about the splice operation
603 * This function should be called before a loop containing
604 * splice_from_pipe_next() and splice_from_pipe_feed() to
605 * initialize the necessary fields of @sd.
607 static void splice_from_pipe_begin(struct splice_desc
*sd
)
610 sd
->need_wakeup
= false;
614 * splice_from_pipe_end - finish splicing from pipe
615 * @pipe: pipe to splice from
616 * @sd: information about the splice operation
619 * This function will wake up pipe writers if necessary. It should
620 * be called after a loop containing splice_from_pipe_next() and
621 * splice_from_pipe_feed().
623 static void splice_from_pipe_end(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
626 wakeup_pipe_writers(pipe
);
630 * __splice_from_pipe - splice data from a pipe to given actor
631 * @pipe: pipe to splice from
632 * @sd: information to @actor
633 * @actor: handler that splices the data
636 * This function does little more than loop over the pipe and call
637 * @actor to do the actual moving of a single struct pipe_buffer to
638 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
642 ssize_t
__splice_from_pipe(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
647 splice_from_pipe_begin(sd
);
650 ret
= splice_from_pipe_next(pipe
, sd
);
652 ret
= splice_from_pipe_feed(pipe
, sd
, actor
);
654 splice_from_pipe_end(pipe
, sd
);
656 return sd
->num_spliced
? sd
->num_spliced
: ret
;
658 EXPORT_SYMBOL(__splice_from_pipe
);
661 * splice_from_pipe - splice data from a pipe to a file
662 * @pipe: pipe to splice from
663 * @out: file to splice to
664 * @ppos: position in @out
665 * @len: how many bytes to splice
666 * @flags: splice modifier flags
667 * @actor: handler that splices the data
670 * See __splice_from_pipe. This function locks the pipe inode,
671 * otherwise it's identical to __splice_from_pipe().
674 ssize_t
splice_from_pipe(struct pipe_inode_info
*pipe
, struct file
*out
,
675 loff_t
*ppos
, size_t len
, unsigned int flags
,
679 struct splice_desc sd
= {
687 ret
= __splice_from_pipe(pipe
, &sd
, actor
);
694 * iter_file_splice_write - splice data from a pipe to a file
696 * @out: file to write to
697 * @ppos: position in @out
698 * @len: number of bytes to splice
699 * @flags: splice modifier flags
702 * Will either move or copy pages (determined by @flags options) from
703 * the given pipe inode to the given file.
704 * This one is ->write_iter-based.
708 iter_file_splice_write(struct pipe_inode_info
*pipe
, struct file
*out
,
709 loff_t
*ppos
, size_t len
, unsigned int flags
)
711 struct splice_desc sd
= {
717 int nbufs
= pipe
->max_usage
;
718 struct bio_vec
*array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
722 if (unlikely(!array
))
727 splice_from_pipe_begin(&sd
);
728 while (sd
.total_len
) {
729 struct iov_iter from
;
730 unsigned int head
, tail
, mask
;
734 ret
= splice_from_pipe_next(pipe
, &sd
);
738 if (unlikely(nbufs
< pipe
->max_usage
)) {
740 nbufs
= pipe
->max_usage
;
741 array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
751 mask
= pipe
->ring_size
- 1;
753 /* build the vector */
755 for (n
= 0; !pipe_empty(head
, tail
) && left
&& n
< nbufs
; tail
++) {
756 struct pipe_buffer
*buf
= &pipe
->bufs
[tail
& mask
];
757 size_t this_len
= buf
->len
;
759 /* zero-length bvecs are not supported, skip them */
762 this_len
= min(this_len
, left
);
764 ret
= pipe_buf_confirm(pipe
, buf
);
771 bvec_set_page(&array
[n
], buf
->page
, this_len
,
777 iov_iter_bvec(&from
, ITER_SOURCE
, array
, n
, sd
.total_len
- left
);
778 ret
= vfs_iter_write(out
, &from
, &sd
.pos
, 0);
782 sd
.num_spliced
+= ret
;
786 /* dismiss the fully eaten buffers, adjust the partial one */
789 struct pipe_buffer
*buf
= &pipe
->bufs
[tail
& mask
];
790 if (ret
>= buf
->len
) {
793 pipe_buf_release(pipe
, buf
);
797 sd
.need_wakeup
= true;
807 splice_from_pipe_end(pipe
, &sd
);
812 ret
= sd
.num_spliced
;
817 EXPORT_SYMBOL(iter_file_splice_write
);
820 * generic_splice_sendpage - splice data from a pipe to a socket
821 * @pipe: pipe to splice from
822 * @out: socket to write to
823 * @ppos: position in @out
824 * @len: number of bytes to splice
825 * @flags: splice modifier flags
828 * Will send @len bytes from the pipe to a network socket. No data copying
832 ssize_t
generic_splice_sendpage(struct pipe_inode_info
*pipe
, struct file
*out
,
833 loff_t
*ppos
, size_t len
, unsigned int flags
)
835 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_sendpage
);
838 EXPORT_SYMBOL(generic_splice_sendpage
);
840 static int warn_unsupported(struct file
*file
, const char *op
)
842 pr_debug_ratelimited(
843 "splice %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
844 op
, file
, current
->pid
, current
->comm
);
849 * Attempt to initiate a splice from pipe to file.
851 static long do_splice_from(struct pipe_inode_info
*pipe
, struct file
*out
,
852 loff_t
*ppos
, size_t len
, unsigned int flags
)
854 if (unlikely(!out
->f_op
->splice_write
))
855 return warn_unsupported(out
, "write");
856 return out
->f_op
->splice_write(pipe
, out
, ppos
, len
, flags
);
860 * Attempt to initiate a splice from a file to a pipe.
862 static long do_splice_to(struct file
*in
, loff_t
*ppos
,
863 struct pipe_inode_info
*pipe
, size_t len
,
866 unsigned int p_space
;
869 if (unlikely(!(in
->f_mode
& FMODE_READ
)))
872 /* Don't try to read more the pipe has space for. */
873 p_space
= pipe
->max_usage
- pipe_occupancy(pipe
->head
, pipe
->tail
);
874 len
= min_t(size_t, len
, p_space
<< PAGE_SHIFT
);
876 ret
= rw_verify_area(READ
, in
, ppos
, len
);
877 if (unlikely(ret
< 0))
880 if (unlikely(len
> MAX_RW_COUNT
))
883 if (unlikely(!in
->f_op
->splice_read
))
884 return warn_unsupported(in
, "read");
885 return in
->f_op
->splice_read(in
, ppos
, pipe
, len
, flags
);
889 * splice_direct_to_actor - splices data directly between two non-pipes
890 * @in: file to splice from
891 * @sd: actor information on where to splice to
892 * @actor: handles the data splicing
895 * This is a special case helper to splice directly between two
896 * points, without requiring an explicit pipe. Internally an allocated
897 * pipe is cached in the process, and reused during the lifetime of
901 ssize_t
splice_direct_to_actor(struct file
*in
, struct splice_desc
*sd
,
902 splice_direct_actor
*actor
)
904 struct pipe_inode_info
*pipe
;
910 * We require the input to be seekable, as we don't want to randomly
911 * drop data for eg socket -> socket splicing. Use the piped splicing
914 if (unlikely(!(in
->f_mode
& FMODE_LSEEK
)))
918 * neither in nor out is a pipe, setup an internal pipe attached to
919 * 'out' and transfer the wanted data from 'in' to 'out' through that
921 pipe
= current
->splice_pipe
;
922 if (unlikely(!pipe
)) {
923 pipe
= alloc_pipe_info();
928 * We don't have an immediate reader, but we'll read the stuff
929 * out of the pipe right after the splice_to_pipe(). So set
930 * PIPE_READERS appropriately.
934 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_empty(pipe
->head
, pipe
->tail
));
954 loff_t pos
= sd
->pos
, prev_pos
= pos
;
956 ret
= do_splice_to(in
, &pos
, pipe
, len
, flags
);
957 if (unlikely(ret
<= 0))
961 sd
->total_len
= read_len
;
964 * If more data is pending, set SPLICE_F_MORE
965 * If this is the last data and SPLICE_F_MORE was not set
966 * initially, clears it.
969 sd
->flags
|= SPLICE_F_MORE
;
971 sd
->flags
&= ~SPLICE_F_MORE
;
973 * NOTE: nonblocking mode only applies to the input. We
974 * must not do the output in nonblocking mode as then we
975 * could get stuck data in the internal pipe:
977 ret
= actor(pipe
, sd
);
978 if (unlikely(ret
<= 0)) {
987 if (ret
< read_len
) {
988 sd
->pos
= prev_pos
+ ret
;
994 pipe
->tail
= pipe
->head
= 0;
1000 * If we did an incomplete transfer we must release
1001 * the pipe buffers in question:
1003 for (i
= 0; i
< pipe
->ring_size
; i
++) {
1004 struct pipe_buffer
*buf
= &pipe
->bufs
[i
];
1007 pipe_buf_release(pipe
, buf
);
1015 EXPORT_SYMBOL(splice_direct_to_actor
);
1017 static int direct_splice_actor(struct pipe_inode_info
*pipe
,
1018 struct splice_desc
*sd
)
1020 struct file
*file
= sd
->u
.file
;
1022 return do_splice_from(pipe
, file
, sd
->opos
, sd
->total_len
,
1027 * do_splice_direct - splices data directly between two files
1028 * @in: file to splice from
1029 * @ppos: input file offset
1030 * @out: file to splice to
1031 * @opos: output file offset
1032 * @len: number of bytes to splice
1033 * @flags: splice modifier flags
1036 * For use by do_sendfile(). splice can easily emulate sendfile, but
1037 * doing it in the application would incur an extra system call
1038 * (splice in + splice out, as compared to just sendfile()). So this helper
1039 * can splice directly through a process-private pipe.
1042 long do_splice_direct(struct file
*in
, loff_t
*ppos
, struct file
*out
,
1043 loff_t
*opos
, size_t len
, unsigned int flags
)
1045 struct splice_desc sd
= {
1055 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
1058 if (unlikely(out
->f_flags
& O_APPEND
))
1061 ret
= rw_verify_area(WRITE
, out
, opos
, len
);
1062 if (unlikely(ret
< 0))
1065 ret
= splice_direct_to_actor(in
, &sd
, direct_splice_actor
);
1071 EXPORT_SYMBOL(do_splice_direct
);
1073 static int wait_for_space(struct pipe_inode_info
*pipe
, unsigned flags
)
1076 if (unlikely(!pipe
->readers
)) {
1077 send_sig(SIGPIPE
, current
, 0);
1080 if (!pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
))
1082 if (flags
& SPLICE_F_NONBLOCK
)
1084 if (signal_pending(current
))
1085 return -ERESTARTSYS
;
1086 pipe_wait_writable(pipe
);
1090 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1091 struct pipe_inode_info
*opipe
,
1092 size_t len
, unsigned int flags
);
1094 long splice_file_to_pipe(struct file
*in
,
1095 struct pipe_inode_info
*opipe
,
1097 size_t len
, unsigned int flags
)
1102 ret
= wait_for_space(opipe
, flags
);
1104 ret
= do_splice_to(in
, offset
, opipe
, len
, flags
);
1107 wakeup_pipe_readers(opipe
);
1112 * Determine where to splice to/from.
1114 long do_splice(struct file
*in
, loff_t
*off_in
, struct file
*out
,
1115 loff_t
*off_out
, size_t len
, unsigned int flags
)
1117 struct pipe_inode_info
*ipipe
;
1118 struct pipe_inode_info
*opipe
;
1122 if (unlikely(!(in
->f_mode
& FMODE_READ
) ||
1123 !(out
->f_mode
& FMODE_WRITE
)))
1126 ipipe
= get_pipe_info(in
, true);
1127 opipe
= get_pipe_info(out
, true);
1129 if (ipipe
&& opipe
) {
1130 if (off_in
|| off_out
)
1133 /* Splicing to self would be fun, but... */
1137 if ((in
->f_flags
| out
->f_flags
) & O_NONBLOCK
)
1138 flags
|= SPLICE_F_NONBLOCK
;
1140 return splice_pipe_to_pipe(ipipe
, opipe
, len
, flags
);
1147 if (!(out
->f_mode
& FMODE_PWRITE
))
1151 offset
= out
->f_pos
;
1154 if (unlikely(out
->f_flags
& O_APPEND
))
1157 ret
= rw_verify_area(WRITE
, out
, &offset
, len
);
1158 if (unlikely(ret
< 0))
1161 if (in
->f_flags
& O_NONBLOCK
)
1162 flags
|= SPLICE_F_NONBLOCK
;
1164 file_start_write(out
);
1165 ret
= do_splice_from(ipipe
, out
, &offset
, len
, flags
);
1166 file_end_write(out
);
1169 out
->f_pos
= offset
;
1180 if (!(in
->f_mode
& FMODE_PREAD
))
1187 if (out
->f_flags
& O_NONBLOCK
)
1188 flags
|= SPLICE_F_NONBLOCK
;
1190 ret
= splice_file_to_pipe(in
, opipe
, &offset
, len
, flags
);
1202 static long __do_splice(struct file
*in
, loff_t __user
*off_in
,
1203 struct file
*out
, loff_t __user
*off_out
,
1204 size_t len
, unsigned int flags
)
1206 struct pipe_inode_info
*ipipe
;
1207 struct pipe_inode_info
*opipe
;
1208 loff_t offset
, *__off_in
= NULL
, *__off_out
= NULL
;
1211 ipipe
= get_pipe_info(in
, true);
1212 opipe
= get_pipe_info(out
, true);
1214 if (ipipe
&& off_in
)
1216 if (opipe
&& off_out
)
1220 if (copy_from_user(&offset
, off_out
, sizeof(loff_t
)))
1222 __off_out
= &offset
;
1225 if (copy_from_user(&offset
, off_in
, sizeof(loff_t
)))
1230 ret
= do_splice(in
, __off_in
, out
, __off_out
, len
, flags
);
1234 if (__off_out
&& copy_to_user(off_out
, __off_out
, sizeof(loff_t
)))
1236 if (__off_in
&& copy_to_user(off_in
, __off_in
, sizeof(loff_t
)))
1242 static int iter_to_pipe(struct iov_iter
*from
,
1243 struct pipe_inode_info
*pipe
,
1246 struct pipe_buffer buf
= {
1247 .ops
= &user_page_pipe_buf_ops
,
1253 while (iov_iter_count(from
)) {
1254 struct page
*pages
[16];
1259 left
= iov_iter_get_pages2(from
, pages
, ~0UL, 16, &start
);
1265 n
= DIV_ROUND_UP(left
+ start
, PAGE_SIZE
);
1266 for (i
= 0; i
< n
; i
++) {
1267 int size
= min_t(int, left
, PAGE_SIZE
- start
);
1269 buf
.page
= pages
[i
];
1272 ret
= add_to_pipe(pipe
, &buf
);
1273 if (unlikely(ret
< 0)) {
1274 iov_iter_revert(from
, left
);
1275 // this one got dropped by add_to_pipe()
1286 return total
? total
: ret
;
1289 static int pipe_to_user(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
1290 struct splice_desc
*sd
)
1292 int n
= copy_page_to_iter(buf
->page
, buf
->offset
, sd
->len
, sd
->u
.data
);
1293 return n
== sd
->len
? n
: -EFAULT
;
1297 * For lack of a better implementation, implement vmsplice() to userspace
1298 * as a simple copy of the pipes pages to the user iov.
1300 static long vmsplice_to_user(struct file
*file
, struct iov_iter
*iter
,
1303 struct pipe_inode_info
*pipe
= get_pipe_info(file
, true);
1304 struct splice_desc sd
= {
1305 .total_len
= iov_iter_count(iter
),
1316 ret
= __splice_from_pipe(pipe
, &sd
, pipe_to_user
);
1324 * vmsplice splices a user address range into a pipe. It can be thought of
1325 * as splice-from-memory, where the regular splice is splice-from-file (or
1326 * to file). In both cases the output is a pipe, naturally.
1328 static long vmsplice_to_pipe(struct file
*file
, struct iov_iter
*iter
,
1331 struct pipe_inode_info
*pipe
;
1333 unsigned buf_flag
= 0;
1335 if (flags
& SPLICE_F_GIFT
)
1336 buf_flag
= PIPE_BUF_FLAG_GIFT
;
1338 pipe
= get_pipe_info(file
, true);
1343 ret
= wait_for_space(pipe
, flags
);
1345 ret
= iter_to_pipe(iter
, pipe
, buf_flag
);
1348 wakeup_pipe_readers(pipe
);
1352 static int vmsplice_type(struct fd f
, int *type
)
1356 if (f
.file
->f_mode
& FMODE_WRITE
) {
1357 *type
= ITER_SOURCE
;
1358 } else if (f
.file
->f_mode
& FMODE_READ
) {
1368 * Note that vmsplice only really supports true splicing _from_ user memory
1369 * to a pipe, not the other way around. Splicing from user memory is a simple
1370 * operation that can be supported without any funky alignment restrictions
1371 * or nasty vm tricks. We simply map in the user memory and fill them into
1372 * a pipe. The reverse isn't quite as easy, though. There are two possible
1373 * solutions for that:
1375 * - memcpy() the data internally, at which point we might as well just
1376 * do a regular read() on the buffer anyway.
1377 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1378 * has restriction limitations on both ends of the pipe).
1380 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1383 SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct iovec __user
*, uiov
,
1384 unsigned long, nr_segs
, unsigned int, flags
)
1386 struct iovec iovstack
[UIO_FASTIOV
];
1387 struct iovec
*iov
= iovstack
;
1388 struct iov_iter iter
;
1393 if (unlikely(flags
& ~SPLICE_F_ALL
))
1397 error
= vmsplice_type(f
, &type
);
1401 error
= import_iovec(type
, uiov
, nr_segs
,
1402 ARRAY_SIZE(iovstack
), &iov
, &iter
);
1406 if (!iov_iter_count(&iter
))
1408 else if (type
== ITER_SOURCE
)
1409 error
= vmsplice_to_pipe(f
.file
, &iter
, flags
);
1411 error
= vmsplice_to_user(f
.file
, &iter
, flags
);
1419 SYSCALL_DEFINE6(splice
, int, fd_in
, loff_t __user
*, off_in
,
1420 int, fd_out
, loff_t __user
*, off_out
,
1421 size_t, len
, unsigned int, flags
)
1429 if (unlikely(flags
& ~SPLICE_F_ALL
))
1435 out
= fdget(fd_out
);
1437 error
= __do_splice(in
.file
, off_in
, out
.file
, off_out
,
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 the pipe occupancy without the inode lock first. This function
1456 * is speculative anyways, so missing one is ok.
1458 if (!pipe_empty(pipe
->head
, pipe
->tail
))
1464 while (pipe_empty(pipe
->head
, pipe
->tail
)) {
1465 if (signal_pending(current
)) {
1471 if (flags
& SPLICE_F_NONBLOCK
) {
1475 pipe_wait_readable(pipe
);
1483 * Make sure there's writeable room. Wait for room if we can, otherwise
1484 * return an appropriate error.
1486 static int opipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1491 * Check pipe occupancy without the inode lock first. This function
1492 * is speculative anyways, so missing one is ok.
1494 if (!pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
))
1500 while (pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
)) {
1501 if (!pipe
->readers
) {
1502 send_sig(SIGPIPE
, current
, 0);
1506 if (flags
& SPLICE_F_NONBLOCK
) {
1510 if (signal_pending(current
)) {
1514 pipe_wait_writable(pipe
);
1522 * Splice contents of ipipe to opipe.
1524 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1525 struct pipe_inode_info
*opipe
,
1526 size_t len
, unsigned int flags
)
1528 struct pipe_buffer
*ibuf
, *obuf
;
1529 unsigned int i_head
, o_head
;
1530 unsigned int i_tail
, o_tail
;
1531 unsigned int i_mask
, o_mask
;
1533 bool input_wakeup
= false;
1537 ret
= ipipe_prep(ipipe
, flags
);
1541 ret
= opipe_prep(opipe
, flags
);
1546 * Potential ABBA deadlock, work around it by ordering lock
1547 * grabbing by pipe info address. Otherwise two different processes
1548 * could deadlock (one doing tee from A -> B, the other from B -> A).
1550 pipe_double_lock(ipipe
, opipe
);
1552 i_tail
= ipipe
->tail
;
1553 i_mask
= ipipe
->ring_size
- 1;
1554 o_head
= opipe
->head
;
1555 o_mask
= opipe
->ring_size
- 1;
1560 if (!opipe
->readers
) {
1561 send_sig(SIGPIPE
, current
, 0);
1567 i_head
= ipipe
->head
;
1568 o_tail
= opipe
->tail
;
1570 if (pipe_empty(i_head
, i_tail
) && !ipipe
->writers
)
1574 * Cannot make any progress, because either the input
1575 * pipe is empty or the output pipe is full.
1577 if (pipe_empty(i_head
, i_tail
) ||
1578 pipe_full(o_head
, o_tail
, opipe
->max_usage
)) {
1579 /* Already processed some buffers, break */
1583 if (flags
& SPLICE_F_NONBLOCK
) {
1589 * We raced with another reader/writer and haven't
1590 * managed to process any buffers. A zero return
1591 * value means EOF, so retry instead.
1598 ibuf
= &ipipe
->bufs
[i_tail
& i_mask
];
1599 obuf
= &opipe
->bufs
[o_head
& o_mask
];
1601 if (len
>= ibuf
->len
) {
1603 * Simply move the whole buffer from ipipe to opipe
1608 ipipe
->tail
= i_tail
;
1609 input_wakeup
= true;
1612 opipe
->head
= o_head
;
1615 * Get a reference to this pipe buffer,
1616 * so we can copy the contents over.
1618 if (!pipe_buf_get(ipipe
, ibuf
)) {
1626 * Don't inherit the gift and merge flags, we need to
1627 * prevent multiple steals of this page.
1629 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1630 obuf
->flags
&= ~PIPE_BUF_FLAG_CAN_MERGE
;
1633 ibuf
->offset
+= len
;
1637 opipe
->head
= o_head
;
1647 * If we put data in the output pipe, wakeup any potential readers.
1650 wakeup_pipe_readers(opipe
);
1653 wakeup_pipe_writers(ipipe
);
1659 * Link contents of ipipe to opipe.
1661 static int link_pipe(struct pipe_inode_info
*ipipe
,
1662 struct pipe_inode_info
*opipe
,
1663 size_t len
, unsigned int flags
)
1665 struct pipe_buffer
*ibuf
, *obuf
;
1666 unsigned int i_head
, o_head
;
1667 unsigned int i_tail
, o_tail
;
1668 unsigned int i_mask
, o_mask
;
1672 * Potential ABBA deadlock, work around it by ordering lock
1673 * grabbing by pipe info address. Otherwise two different processes
1674 * could deadlock (one doing tee from A -> B, the other from B -> A).
1676 pipe_double_lock(ipipe
, opipe
);
1678 i_tail
= ipipe
->tail
;
1679 i_mask
= ipipe
->ring_size
- 1;
1680 o_head
= opipe
->head
;
1681 o_mask
= opipe
->ring_size
- 1;
1684 if (!opipe
->readers
) {
1685 send_sig(SIGPIPE
, current
, 0);
1691 i_head
= ipipe
->head
;
1692 o_tail
= opipe
->tail
;
1695 * If we have iterated all input buffers or run out of
1696 * output room, break.
1698 if (pipe_empty(i_head
, i_tail
) ||
1699 pipe_full(o_head
, o_tail
, opipe
->max_usage
))
1702 ibuf
= &ipipe
->bufs
[i_tail
& i_mask
];
1703 obuf
= &opipe
->bufs
[o_head
& o_mask
];
1706 * Get a reference to this pipe buffer,
1707 * so we can copy the contents over.
1709 if (!pipe_buf_get(ipipe
, ibuf
)) {
1718 * Don't inherit the gift and merge flag, we need to prevent
1719 * multiple steals of this page.
1721 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1722 obuf
->flags
&= ~PIPE_BUF_FLAG_CAN_MERGE
;
1724 if (obuf
->len
> len
)
1730 opipe
->head
= o_head
;
1738 * If we put data in the output pipe, wakeup any potential readers.
1741 wakeup_pipe_readers(opipe
);
1747 * This is a tee(1) implementation that works on pipes. It doesn't copy
1748 * any data, it simply references the 'in' pages on the 'out' pipe.
1749 * The 'flags' used are the SPLICE_F_* variants, currently the only
1750 * applicable one is SPLICE_F_NONBLOCK.
1752 long do_tee(struct file
*in
, struct file
*out
, size_t len
, unsigned int flags
)
1754 struct pipe_inode_info
*ipipe
= get_pipe_info(in
, true);
1755 struct pipe_inode_info
*opipe
= get_pipe_info(out
, true);
1758 if (unlikely(!(in
->f_mode
& FMODE_READ
) ||
1759 !(out
->f_mode
& FMODE_WRITE
)))
1763 * Duplicate the contents of ipipe to opipe without actually
1766 if (ipipe
&& opipe
&& ipipe
!= opipe
) {
1767 if ((in
->f_flags
| out
->f_flags
) & O_NONBLOCK
)
1768 flags
|= SPLICE_F_NONBLOCK
;
1771 * Keep going, unless we encounter an error. The ipipe/opipe
1772 * ordering doesn't really matter.
1774 ret
= ipipe_prep(ipipe
, flags
);
1776 ret
= opipe_prep(opipe
, flags
);
1778 ret
= link_pipe(ipipe
, opipe
, len
, flags
);
1785 SYSCALL_DEFINE4(tee
, int, fdin
, int, fdout
, size_t, len
, unsigned int, flags
)
1790 if (unlikely(flags
& ~SPLICE_F_ALL
))
1801 error
= do_tee(in
.file
, out
.file
, len
, flags
);