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>
21 #include <linux/file.h>
22 #include <linux/pagemap.h>
23 #include <linux/splice.h>
24 #include <linux/memcontrol.h>
25 #include <linux/mm_inline.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/export.h>
29 #include <linux/syscalls.h>
30 #include <linux/uio.h>
31 #include <linux/security.h>
32 #include <linux/gfp.h>
33 #include <linux/socket.h>
34 #include <linux/compat.h>
38 * Attempt to steal a page from a pipe buffer. This should perhaps go into
39 * a vm helper function, it's already simplified quite a bit by the
40 * addition of remove_mapping(). If success is returned, the caller may
41 * attempt to reuse this page for another destination.
43 static int page_cache_pipe_buf_steal(struct pipe_inode_info
*pipe
,
44 struct pipe_buffer
*buf
)
46 struct page
*page
= buf
->page
;
47 struct address_space
*mapping
;
51 mapping
= page_mapping(page
);
53 WARN_ON(!PageUptodate(page
));
56 * At least for ext2 with nobh option, we need to wait on
57 * writeback completing on this page, since we'll remove it
58 * from the pagecache. Otherwise truncate wont wait on the
59 * page, allowing the disk blocks to be reused by someone else
60 * before we actually wrote our data to them. fs corruption
63 wait_on_page_writeback(page
);
65 if (page_has_private(page
) &&
66 !try_to_release_page(page
, GFP_KERNEL
))
70 * If we succeeded in removing the mapping, set LRU flag
73 if (remove_mapping(mapping
, page
)) {
74 buf
->flags
|= PIPE_BUF_FLAG_LRU
;
80 * Raced with truncate or failed to remove page from current
81 * address space, unlock and return failure.
88 static void page_cache_pipe_buf_release(struct pipe_inode_info
*pipe
,
89 struct pipe_buffer
*buf
)
92 buf
->flags
&= ~PIPE_BUF_FLAG_LRU
;
96 * Check whether the contents of buf is OK to access. Since the content
97 * is a page cache page, IO may be in flight.
99 static int page_cache_pipe_buf_confirm(struct pipe_inode_info
*pipe
,
100 struct pipe_buffer
*buf
)
102 struct page
*page
= buf
->page
;
105 if (!PageUptodate(page
)) {
109 * Page got truncated/unhashed. This will cause a 0-byte
110 * splice, if this is the first page.
112 if (!page
->mapping
) {
118 * Uh oh, read-error from disk.
120 if (!PageUptodate(page
)) {
126 * Page is ok afterall, we are done.
137 const struct pipe_buf_operations page_cache_pipe_buf_ops
= {
139 .confirm
= page_cache_pipe_buf_confirm
,
140 .release
= page_cache_pipe_buf_release
,
141 .steal
= page_cache_pipe_buf_steal
,
142 .get
= generic_pipe_buf_get
,
145 static int user_page_pipe_buf_steal(struct pipe_inode_info
*pipe
,
146 struct pipe_buffer
*buf
)
148 if (!(buf
->flags
& PIPE_BUF_FLAG_GIFT
))
151 buf
->flags
|= PIPE_BUF_FLAG_LRU
;
152 return generic_pipe_buf_steal(pipe
, buf
);
155 static const struct pipe_buf_operations user_page_pipe_buf_ops
= {
157 .confirm
= generic_pipe_buf_confirm
,
158 .release
= page_cache_pipe_buf_release
,
159 .steal
= user_page_pipe_buf_steal
,
160 .get
= generic_pipe_buf_get
,
163 static void wakeup_pipe_readers(struct pipe_inode_info
*pipe
)
166 if (waitqueue_active(&pipe
->wait
))
167 wake_up_interruptible(&pipe
->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 int ret
= 0, page_nr
= 0;
191 if (unlikely(!pipe
->readers
)) {
192 send_sig(SIGPIPE
, current
, 0);
197 while (pipe
->nrbufs
< pipe
->buffers
) {
198 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
199 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
201 buf
->page
= spd
->pages
[page_nr
];
202 buf
->offset
= spd
->partial
[page_nr
].offset
;
203 buf
->len
= spd
->partial
[page_nr
].len
;
204 buf
->private = spd
->partial
[page_nr
].private;
212 if (!--spd
->nr_pages
)
220 while (page_nr
< spd_pages
)
221 spd
->spd_release(spd
, page_nr
++);
225 EXPORT_SYMBOL_GPL(splice_to_pipe
);
227 ssize_t
add_to_pipe(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
)
231 if (unlikely(!pipe
->readers
)) {
232 send_sig(SIGPIPE
, current
, 0);
234 } else if (pipe
->nrbufs
== pipe
->buffers
) {
237 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
238 pipe
->bufs
[newbuf
] = *buf
;
242 pipe_buf_release(pipe
, buf
);
245 EXPORT_SYMBOL(add_to_pipe
);
247 void spd_release_page(struct splice_pipe_desc
*spd
, unsigned int i
)
249 put_page(spd
->pages
[i
]);
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 buffers
= ACCESS_ONCE(pipe
->buffers
);
260 spd
->nr_pages_max
= buffers
;
261 if (buffers
<= PIPE_DEF_BUFFERS
)
264 spd
->pages
= kmalloc(buffers
* sizeof(struct page
*), GFP_KERNEL
);
265 spd
->partial
= kmalloc(buffers
* sizeof(struct partial_page
), GFP_KERNEL
);
267 if (spd
->pages
&& spd
->partial
)
275 void splice_shrink_spd(struct splice_pipe_desc
*spd
)
277 if (spd
->nr_pages_max
<= PIPE_DEF_BUFFERS
)
285 * generic_file_splice_read - splice data from file to a pipe
286 * @in: file to splice from
287 * @ppos: position in @in
288 * @pipe: pipe to splice to
289 * @len: number of bytes to splice
290 * @flags: splice modifier flags
293 * Will read pages from given file and fill them into a pipe. Can be
294 * used as long as it has more or less sane ->read_iter().
297 ssize_t
generic_file_splice_read(struct file
*in
, loff_t
*ppos
,
298 struct pipe_inode_info
*pipe
, size_t len
,
305 iov_iter_pipe(&to
, ITER_PIPE
| READ
, pipe
, len
);
307 init_sync_kiocb(&kiocb
, in
);
308 kiocb
.ki_pos
= *ppos
;
309 ret
= in
->f_op
->read_iter(&kiocb
, &to
);
311 *ppos
= kiocb
.ki_pos
;
313 } else if (ret
< 0) {
316 iov_iter_advance(&to
, 0); /* to free what was emitted */
318 * callers of ->splice_read() expect -EAGAIN on
319 * "can't put anything in there", rather than -EFAULT.
327 EXPORT_SYMBOL(generic_file_splice_read
);
329 const struct pipe_buf_operations default_pipe_buf_ops
= {
331 .confirm
= generic_pipe_buf_confirm
,
332 .release
= generic_pipe_buf_release
,
333 .steal
= generic_pipe_buf_steal
,
334 .get
= generic_pipe_buf_get
,
337 static int generic_pipe_buf_nosteal(struct pipe_inode_info
*pipe
,
338 struct pipe_buffer
*buf
)
343 /* Pipe buffer operations for a socket and similar. */
344 const struct pipe_buf_operations nosteal_pipe_buf_ops
= {
346 .confirm
= generic_pipe_buf_confirm
,
347 .release
= generic_pipe_buf_release
,
348 .steal
= generic_pipe_buf_nosteal
,
349 .get
= generic_pipe_buf_get
,
351 EXPORT_SYMBOL(nosteal_pipe_buf_ops
);
353 static ssize_t
kernel_readv(struct file
*file
, const struct kvec
*vec
,
354 unsigned long vlen
, loff_t offset
)
362 /* The cast to a user pointer is valid due to the set_fs() */
363 res
= vfs_readv(file
, (const struct iovec __user
*)vec
, vlen
, &pos
, 0);
369 ssize_t
kernel_write(struct file
*file
, const char *buf
, size_t count
,
377 /* The cast to a user pointer is valid due to the set_fs() */
378 res
= vfs_write(file
, (__force
const char __user
*)buf
, count
, &pos
);
383 EXPORT_SYMBOL(kernel_write
);
385 static ssize_t
default_file_splice_read(struct file
*in
, loff_t
*ppos
,
386 struct pipe_inode_info
*pipe
, size_t len
,
389 struct kvec
*vec
, __vec
[PIPE_DEF_BUFFERS
];
392 unsigned int nr_pages
;
393 size_t offset
, dummy
, copied
= 0;
397 if (pipe
->nrbufs
== pipe
->buffers
)
401 * Try to keep page boundaries matching to source pagecache ones -
402 * it probably won't be much help, but...
404 offset
= *ppos
& ~PAGE_MASK
;
406 iov_iter_pipe(&to
, ITER_PIPE
| READ
, pipe
, len
+ offset
);
408 res
= iov_iter_get_pages_alloc(&to
, &pages
, len
+ offset
, &dummy
);
413 nr_pages
= DIV_ROUND_UP(res
, PAGE_SIZE
);
416 if (nr_pages
> PIPE_DEF_BUFFERS
) {
417 vec
= kmalloc(nr_pages
* sizeof(struct kvec
), GFP_KERNEL
);
418 if (unlikely(!vec
)) {
424 pipe
->bufs
[to
.idx
].offset
= offset
;
425 pipe
->bufs
[to
.idx
].len
-= offset
;
427 for (i
= 0; i
< nr_pages
; i
++) {
428 size_t this_len
= min_t(size_t, len
, PAGE_SIZE
- offset
);
429 vec
[i
].iov_base
= page_address(pages
[i
]) + offset
;
430 vec
[i
].iov_len
= this_len
;
435 res
= kernel_readv(in
, vec
, nr_pages
, *ppos
);
444 for (i
= 0; i
< nr_pages
; i
++)
447 iov_iter_advance(&to
, copied
); /* truncates and discards */
452 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
453 * using sendpage(). Return the number of bytes sent.
455 static int pipe_to_sendpage(struct pipe_inode_info
*pipe
,
456 struct pipe_buffer
*buf
, struct splice_desc
*sd
)
458 struct file
*file
= sd
->u
.file
;
459 loff_t pos
= sd
->pos
;
462 if (!likely(file
->f_op
->sendpage
))
465 more
= (sd
->flags
& SPLICE_F_MORE
) ? MSG_MORE
: 0;
467 if (sd
->len
< sd
->total_len
&& pipe
->nrbufs
> 1)
468 more
|= MSG_SENDPAGE_NOTLAST
;
470 return file
->f_op
->sendpage(file
, buf
->page
, buf
->offset
,
471 sd
->len
, &pos
, more
);
474 static void wakeup_pipe_writers(struct pipe_inode_info
*pipe
)
477 if (waitqueue_active(&pipe
->wait
))
478 wake_up_interruptible(&pipe
->wait
);
479 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
483 * splice_from_pipe_feed - feed available data from a pipe to a file
484 * @pipe: pipe to splice from
485 * @sd: information to @actor
486 * @actor: handler that splices the data
489 * This function loops over the pipe and calls @actor to do the
490 * actual moving of a single struct pipe_buffer to the desired
491 * destination. It returns when there's no more buffers left in
492 * the pipe or if the requested number of bytes (@sd->total_len)
493 * have been copied. It returns a positive number (one) if the
494 * pipe needs to be filled with more data, zero if the required
495 * number of bytes have been copied and -errno on error.
497 * This, together with splice_from_pipe_{begin,end,next}, may be
498 * used to implement the functionality of __splice_from_pipe() when
499 * locking is required around copying the pipe buffers to the
502 static int splice_from_pipe_feed(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
507 while (pipe
->nrbufs
) {
508 struct pipe_buffer
*buf
= pipe
->bufs
+ pipe
->curbuf
;
511 if (sd
->len
> sd
->total_len
)
512 sd
->len
= sd
->total_len
;
514 ret
= pipe_buf_confirm(pipe
, buf
);
521 ret
= actor(pipe
, buf
, sd
);
528 sd
->num_spliced
+= ret
;
531 sd
->total_len
-= ret
;
534 pipe_buf_release(pipe
, buf
);
535 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
538 sd
->need_wakeup
= true;
549 * splice_from_pipe_next - wait for some data to splice from
550 * @pipe: pipe to splice from
551 * @sd: information about the splice operation
554 * This function will wait for some data and return a positive
555 * value (one) if pipe buffers are available. It will return zero
556 * or -errno if no more data needs to be spliced.
558 static int splice_from_pipe_next(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
561 * Check for signal early to make process killable when there are
562 * always buffers available
564 if (signal_pending(current
))
567 while (!pipe
->nrbufs
) {
571 if (!pipe
->waiting_writers
&& sd
->num_spliced
)
574 if (sd
->flags
& SPLICE_F_NONBLOCK
)
577 if (signal_pending(current
))
580 if (sd
->need_wakeup
) {
581 wakeup_pipe_writers(pipe
);
582 sd
->need_wakeup
= false;
592 * splice_from_pipe_begin - start splicing from pipe
593 * @sd: information about the splice operation
596 * This function should be called before a loop containing
597 * splice_from_pipe_next() and splice_from_pipe_feed() to
598 * initialize the necessary fields of @sd.
600 static void splice_from_pipe_begin(struct splice_desc
*sd
)
603 sd
->need_wakeup
= false;
607 * splice_from_pipe_end - finish splicing from pipe
608 * @pipe: pipe to splice from
609 * @sd: information about the splice operation
612 * This function will wake up pipe writers if necessary. It should
613 * be called after a loop containing splice_from_pipe_next() and
614 * splice_from_pipe_feed().
616 static void splice_from_pipe_end(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
619 wakeup_pipe_writers(pipe
);
623 * __splice_from_pipe - splice data from a pipe to given actor
624 * @pipe: pipe to splice from
625 * @sd: information to @actor
626 * @actor: handler that splices the data
629 * This function does little more than loop over the pipe and call
630 * @actor to do the actual moving of a single struct pipe_buffer to
631 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
635 ssize_t
__splice_from_pipe(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
640 splice_from_pipe_begin(sd
);
643 ret
= splice_from_pipe_next(pipe
, sd
);
645 ret
= splice_from_pipe_feed(pipe
, sd
, actor
);
647 splice_from_pipe_end(pipe
, sd
);
649 return sd
->num_spliced
? sd
->num_spliced
: ret
;
651 EXPORT_SYMBOL(__splice_from_pipe
);
654 * splice_from_pipe - splice data from a pipe to a file
655 * @pipe: pipe to splice from
656 * @out: file to splice to
657 * @ppos: position in @out
658 * @len: how many bytes to splice
659 * @flags: splice modifier flags
660 * @actor: handler that splices the data
663 * See __splice_from_pipe. This function locks the pipe inode,
664 * otherwise it's identical to __splice_from_pipe().
667 ssize_t
splice_from_pipe(struct pipe_inode_info
*pipe
, struct file
*out
,
668 loff_t
*ppos
, size_t len
, unsigned int flags
,
672 struct splice_desc sd
= {
680 ret
= __splice_from_pipe(pipe
, &sd
, actor
);
687 * iter_file_splice_write - splice data from a pipe to a file
689 * @out: file to write to
690 * @ppos: position in @out
691 * @len: number of bytes to splice
692 * @flags: splice modifier flags
695 * Will either move or copy pages (determined by @flags options) from
696 * the given pipe inode to the given file.
697 * This one is ->write_iter-based.
701 iter_file_splice_write(struct pipe_inode_info
*pipe
, struct file
*out
,
702 loff_t
*ppos
, size_t len
, unsigned int flags
)
704 struct splice_desc sd
= {
710 int nbufs
= pipe
->buffers
;
711 struct bio_vec
*array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
715 if (unlikely(!array
))
720 splice_from_pipe_begin(&sd
);
721 while (sd
.total_len
) {
722 struct iov_iter from
;
726 ret
= splice_from_pipe_next(pipe
, &sd
);
730 if (unlikely(nbufs
< pipe
->buffers
)) {
732 nbufs
= pipe
->buffers
;
733 array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
741 /* build the vector */
743 for (n
= 0, idx
= pipe
->curbuf
; left
&& n
< pipe
->nrbufs
; n
++, idx
++) {
744 struct pipe_buffer
*buf
= pipe
->bufs
+ idx
;
745 size_t this_len
= buf
->len
;
750 if (idx
== pipe
->buffers
- 1)
753 ret
= pipe_buf_confirm(pipe
, buf
);
760 array
[n
].bv_page
= buf
->page
;
761 array
[n
].bv_len
= this_len
;
762 array
[n
].bv_offset
= buf
->offset
;
766 iov_iter_bvec(&from
, ITER_BVEC
| WRITE
, array
, n
,
767 sd
.total_len
- left
);
768 ret
= vfs_iter_write(out
, &from
, &sd
.pos
);
772 sd
.num_spliced
+= ret
;
776 /* dismiss the fully eaten buffers, adjust the partial one */
778 struct pipe_buffer
*buf
= pipe
->bufs
+ pipe
->curbuf
;
779 if (ret
>= buf
->len
) {
782 pipe_buf_release(pipe
, buf
);
783 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
786 sd
.need_wakeup
= true;
796 splice_from_pipe_end(pipe
, &sd
);
801 ret
= sd
.num_spliced
;
806 EXPORT_SYMBOL(iter_file_splice_write
);
808 static int write_pipe_buf(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
809 struct splice_desc
*sd
)
813 loff_t tmp
= sd
->pos
;
815 data
= kmap(buf
->page
);
816 ret
= __kernel_write(sd
->u
.file
, data
+ buf
->offset
, sd
->len
, &tmp
);
822 static ssize_t
default_file_splice_write(struct pipe_inode_info
*pipe
,
823 struct file
*out
, loff_t
*ppos
,
824 size_t len
, unsigned int flags
)
828 ret
= splice_from_pipe(pipe
, out
, ppos
, len
, flags
, write_pipe_buf
);
836 * generic_splice_sendpage - splice data from a pipe to a socket
837 * @pipe: pipe to splice from
838 * @out: socket to write to
839 * @ppos: position in @out
840 * @len: number of bytes to splice
841 * @flags: splice modifier flags
844 * Will send @len bytes from the pipe to a network socket. No data copying
848 ssize_t
generic_splice_sendpage(struct pipe_inode_info
*pipe
, struct file
*out
,
849 loff_t
*ppos
, size_t len
, unsigned int flags
)
851 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_sendpage
);
854 EXPORT_SYMBOL(generic_splice_sendpage
);
857 * Attempt to initiate a splice from pipe to file.
859 static long do_splice_from(struct pipe_inode_info
*pipe
, struct file
*out
,
860 loff_t
*ppos
, size_t len
, unsigned int flags
)
862 ssize_t (*splice_write
)(struct pipe_inode_info
*, struct file
*,
863 loff_t
*, size_t, unsigned int);
865 if (out
->f_op
->splice_write
)
866 splice_write
= out
->f_op
->splice_write
;
868 splice_write
= default_file_splice_write
;
870 return splice_write(pipe
, out
, ppos
, len
, flags
);
874 * Attempt to initiate a splice from a file to a pipe.
876 static long do_splice_to(struct file
*in
, loff_t
*ppos
,
877 struct pipe_inode_info
*pipe
, size_t len
,
880 ssize_t (*splice_read
)(struct file
*, loff_t
*,
881 struct pipe_inode_info
*, size_t, unsigned int);
884 if (unlikely(!(in
->f_mode
& FMODE_READ
)))
887 ret
= rw_verify_area(READ
, in
, ppos
, len
);
888 if (unlikely(ret
< 0))
891 if (unlikely(len
> MAX_RW_COUNT
))
894 if (in
->f_op
->splice_read
)
895 splice_read
= in
->f_op
->splice_read
;
897 splice_read
= default_file_splice_read
;
899 return splice_read(in
, ppos
, pipe
, len
, flags
);
903 * splice_direct_to_actor - splices data directly between two non-pipes
904 * @in: file to splice from
905 * @sd: actor information on where to splice to
906 * @actor: handles the data splicing
909 * This is a special case helper to splice directly between two
910 * points, without requiring an explicit pipe. Internally an allocated
911 * pipe is cached in the process, and reused during the lifetime of
915 ssize_t
splice_direct_to_actor(struct file
*in
, struct splice_desc
*sd
,
916 splice_direct_actor
*actor
)
918 struct pipe_inode_info
*pipe
;
925 * We require the input being a regular file, as we don't want to
926 * randomly drop data for eg socket -> socket splicing. Use the
927 * piped splicing for that!
929 i_mode
= file_inode(in
)->i_mode
;
930 if (unlikely(!S_ISREG(i_mode
) && !S_ISBLK(i_mode
)))
934 * neither in nor out is a pipe, setup an internal pipe attached to
935 * 'out' and transfer the wanted data from 'in' to 'out' through that
937 pipe
= current
->splice_pipe
;
938 if (unlikely(!pipe
)) {
939 pipe
= alloc_pipe_info();
944 * We don't have an immediate reader, but we'll read the stuff
945 * out of the pipe right after the splice_to_pipe(). So set
946 * PIPE_READERS appropriately.
950 current
->splice_pipe
= pipe
;
962 * Don't block on output, we have to drain the direct pipe.
964 sd
->flags
&= ~SPLICE_F_NONBLOCK
;
965 more
= sd
->flags
& SPLICE_F_MORE
;
969 loff_t pos
= sd
->pos
, prev_pos
= pos
;
971 ret
= do_splice_to(in
, &pos
, pipe
, len
, flags
);
972 if (unlikely(ret
<= 0))
976 sd
->total_len
= read_len
;
979 * If more data is pending, set SPLICE_F_MORE
980 * If this is the last data and SPLICE_F_MORE was not set
981 * initially, clears it.
984 sd
->flags
|= SPLICE_F_MORE
;
986 sd
->flags
&= ~SPLICE_F_MORE
;
988 * NOTE: nonblocking mode only applies to the input. We
989 * must not do the output in nonblocking mode as then we
990 * could get stuck data in the internal pipe:
992 ret
= actor(pipe
, sd
);
993 if (unlikely(ret
<= 0)) {
1002 if (ret
< read_len
) {
1003 sd
->pos
= prev_pos
+ ret
;
1009 pipe
->nrbufs
= pipe
->curbuf
= 0;
1015 * If we did an incomplete transfer we must release
1016 * the pipe buffers in question:
1018 for (i
= 0; i
< pipe
->buffers
; i
++) {
1019 struct pipe_buffer
*buf
= pipe
->bufs
+ i
;
1022 pipe_buf_release(pipe
, buf
);
1030 EXPORT_SYMBOL(splice_direct_to_actor
);
1032 static int direct_splice_actor(struct pipe_inode_info
*pipe
,
1033 struct splice_desc
*sd
)
1035 struct file
*file
= sd
->u
.file
;
1037 return do_splice_from(pipe
, file
, sd
->opos
, sd
->total_len
,
1042 * do_splice_direct - splices data directly between two files
1043 * @in: file to splice from
1044 * @ppos: input file offset
1045 * @out: file to splice to
1046 * @opos: output file offset
1047 * @len: number of bytes to splice
1048 * @flags: splice modifier flags
1051 * For use by do_sendfile(). splice can easily emulate sendfile, but
1052 * doing it in the application would incur an extra system call
1053 * (splice in + splice out, as compared to just sendfile()). So this helper
1054 * can splice directly through a process-private pipe.
1057 long do_splice_direct(struct file
*in
, loff_t
*ppos
, struct file
*out
,
1058 loff_t
*opos
, size_t len
, unsigned int flags
)
1060 struct splice_desc sd
= {
1070 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
1073 if (unlikely(out
->f_flags
& O_APPEND
))
1076 ret
= rw_verify_area(WRITE
, out
, opos
, len
);
1077 if (unlikely(ret
< 0))
1080 ret
= splice_direct_to_actor(in
, &sd
, direct_splice_actor
);
1086 EXPORT_SYMBOL(do_splice_direct
);
1088 static int wait_for_space(struct pipe_inode_info
*pipe
, unsigned flags
)
1091 if (unlikely(!pipe
->readers
)) {
1092 send_sig(SIGPIPE
, current
, 0);
1095 if (pipe
->nrbufs
!= pipe
->buffers
)
1097 if (flags
& SPLICE_F_NONBLOCK
)
1099 if (signal_pending(current
))
1100 return -ERESTARTSYS
;
1101 pipe
->waiting_writers
++;
1103 pipe
->waiting_writers
--;
1107 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1108 struct pipe_inode_info
*opipe
,
1109 size_t len
, unsigned int flags
);
1112 * Determine where to splice to/from.
1114 static long do_splice(struct file
*in
, loff_t __user
*off_in
,
1115 struct file
*out
, loff_t __user
*off_out
,
1116 size_t len
, unsigned int flags
)
1118 struct pipe_inode_info
*ipipe
;
1119 struct pipe_inode_info
*opipe
;
1123 ipipe
= get_pipe_info(in
);
1124 opipe
= get_pipe_info(out
);
1126 if (ipipe
&& opipe
) {
1127 if (off_in
|| off_out
)
1130 if (!(in
->f_mode
& FMODE_READ
))
1133 if (!(out
->f_mode
& FMODE_WRITE
))
1136 /* Splicing to self would be fun, but... */
1140 return splice_pipe_to_pipe(ipipe
, opipe
, len
, flags
);
1147 if (!(out
->f_mode
& FMODE_PWRITE
))
1149 if (copy_from_user(&offset
, off_out
, sizeof(loff_t
)))
1152 offset
= out
->f_pos
;
1155 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
1158 if (unlikely(out
->f_flags
& O_APPEND
))
1161 ret
= rw_verify_area(WRITE
, out
, &offset
, len
);
1162 if (unlikely(ret
< 0))
1165 file_start_write(out
);
1166 ret
= do_splice_from(ipipe
, out
, &offset
, len
, flags
);
1167 file_end_write(out
);
1170 out
->f_pos
= offset
;
1171 else if (copy_to_user(off_out
, &offset
, sizeof(loff_t
)))
1181 if (!(in
->f_mode
& FMODE_PREAD
))
1183 if (copy_from_user(&offset
, off_in
, sizeof(loff_t
)))
1190 ret
= wait_for_space(opipe
, flags
);
1192 ret
= do_splice_to(in
, &offset
, opipe
, len
, flags
);
1195 wakeup_pipe_readers(opipe
);
1198 else if (copy_to_user(off_in
, &offset
, sizeof(loff_t
)))
1207 static int iter_to_pipe(struct iov_iter
*from
,
1208 struct pipe_inode_info
*pipe
,
1211 struct pipe_buffer buf
= {
1212 .ops
= &user_page_pipe_buf_ops
,
1217 bool failed
= false;
1219 while (iov_iter_count(from
) && !failed
) {
1220 struct page
*pages
[16];
1225 copied
= iov_iter_get_pages(from
, pages
, ~0UL, 16, &start
);
1231 for (n
= 0; copied
; n
++, start
= 0) {
1232 int size
= min_t(int, copied
, PAGE_SIZE
- start
);
1234 buf
.page
= pages
[n
];
1237 ret
= add_to_pipe(pipe
, &buf
);
1238 if (unlikely(ret
< 0)) {
1241 iov_iter_advance(from
, ret
);
1250 return total
? total
: ret
;
1253 static int pipe_to_user(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
1254 struct splice_desc
*sd
)
1256 int n
= copy_page_to_iter(buf
->page
, buf
->offset
, sd
->len
, sd
->u
.data
);
1257 return n
== sd
->len
? n
: -EFAULT
;
1261 * For lack of a better implementation, implement vmsplice() to userspace
1262 * as a simple copy of the pipes pages to the user iov.
1264 static long vmsplice_to_user(struct file
*file
, const struct iovec __user
*uiov
,
1265 unsigned long nr_segs
, unsigned int flags
)
1267 struct pipe_inode_info
*pipe
;
1268 struct splice_desc sd
;
1270 struct iovec iovstack
[UIO_FASTIOV
];
1271 struct iovec
*iov
= iovstack
;
1272 struct iov_iter iter
;
1274 pipe
= get_pipe_info(file
);
1278 ret
= import_iovec(READ
, uiov
, nr_segs
,
1279 ARRAY_SIZE(iovstack
), &iov
, &iter
);
1283 sd
.total_len
= iov_iter_count(&iter
);
1291 ret
= __splice_from_pipe(pipe
, &sd
, pipe_to_user
);
1300 * vmsplice splices a user address range into a pipe. It can be thought of
1301 * as splice-from-memory, where the regular splice is splice-from-file (or
1302 * to file). In both cases the output is a pipe, naturally.
1304 static long vmsplice_to_pipe(struct file
*file
, const struct iovec __user
*uiov
,
1305 unsigned long nr_segs
, unsigned int flags
)
1307 struct pipe_inode_info
*pipe
;
1308 struct iovec iovstack
[UIO_FASTIOV
];
1309 struct iovec
*iov
= iovstack
;
1310 struct iov_iter from
;
1312 unsigned buf_flag
= 0;
1314 if (flags
& SPLICE_F_GIFT
)
1315 buf_flag
= PIPE_BUF_FLAG_GIFT
;
1317 pipe
= get_pipe_info(file
);
1321 ret
= import_iovec(WRITE
, uiov
, nr_segs
,
1322 ARRAY_SIZE(iovstack
), &iov
, &from
);
1327 ret
= wait_for_space(pipe
, flags
);
1329 ret
= iter_to_pipe(&from
, pipe
, buf_flag
);
1332 wakeup_pipe_readers(pipe
);
1338 * Note that vmsplice only really supports true splicing _from_ user memory
1339 * to a pipe, not the other way around. Splicing from user memory is a simple
1340 * operation that can be supported without any funky alignment restrictions
1341 * or nasty vm tricks. We simply map in the user memory and fill them into
1342 * a pipe. The reverse isn't quite as easy, though. There are two possible
1343 * solutions for that:
1345 * - memcpy() the data internally, at which point we might as well just
1346 * do a regular read() on the buffer anyway.
1347 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1348 * has restriction limitations on both ends of the pipe).
1350 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1353 SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct iovec __user
*, iov
,
1354 unsigned long, nr_segs
, unsigned int, flags
)
1359 if (unlikely(nr_segs
> UIO_MAXIOV
))
1361 else if (unlikely(!nr_segs
))
1367 if (f
.file
->f_mode
& FMODE_WRITE
)
1368 error
= vmsplice_to_pipe(f
.file
, iov
, nr_segs
, flags
);
1369 else if (f
.file
->f_mode
& FMODE_READ
)
1370 error
= vmsplice_to_user(f
.file
, iov
, nr_segs
, flags
);
1378 #ifdef CONFIG_COMPAT
1379 COMPAT_SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct compat_iovec __user
*, iov32
,
1380 unsigned int, nr_segs
, unsigned int, flags
)
1383 struct iovec __user
*iov
;
1384 if (nr_segs
> UIO_MAXIOV
)
1386 iov
= compat_alloc_user_space(nr_segs
* sizeof(struct iovec
));
1387 for (i
= 0; i
< nr_segs
; i
++) {
1388 struct compat_iovec v
;
1389 if (get_user(v
.iov_base
, &iov32
[i
].iov_base
) ||
1390 get_user(v
.iov_len
, &iov32
[i
].iov_len
) ||
1391 put_user(compat_ptr(v
.iov_base
), &iov
[i
].iov_base
) ||
1392 put_user(v
.iov_len
, &iov
[i
].iov_len
))
1395 return sys_vmsplice(fd
, iov
, nr_segs
, flags
);
1399 SYSCALL_DEFINE6(splice
, int, fd_in
, loff_t __user
*, off_in
,
1400 int, fd_out
, loff_t __user
*, off_out
,
1401 size_t, len
, unsigned int, flags
)
1412 if (in
.file
->f_mode
& FMODE_READ
) {
1413 out
= fdget(fd_out
);
1415 if (out
.file
->f_mode
& FMODE_WRITE
)
1416 error
= do_splice(in
.file
, off_in
,
1428 * Make sure there's data to read. Wait for input if we can, otherwise
1429 * return an appropriate error.
1431 static int ipipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1436 * Check ->nrbufs without the inode lock first. This function
1437 * is speculative anyways, so missing one is ok.
1445 while (!pipe
->nrbufs
) {
1446 if (signal_pending(current
)) {
1452 if (!pipe
->waiting_writers
) {
1453 if (flags
& SPLICE_F_NONBLOCK
) {
1466 * Make sure there's writeable room. Wait for room if we can, otherwise
1467 * return an appropriate error.
1469 static int opipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1474 * Check ->nrbufs without the inode lock first. This function
1475 * is speculative anyways, so missing one is ok.
1477 if (pipe
->nrbufs
< pipe
->buffers
)
1483 while (pipe
->nrbufs
>= pipe
->buffers
) {
1484 if (!pipe
->readers
) {
1485 send_sig(SIGPIPE
, current
, 0);
1489 if (flags
& SPLICE_F_NONBLOCK
) {
1493 if (signal_pending(current
)) {
1497 pipe
->waiting_writers
++;
1499 pipe
->waiting_writers
--;
1507 * Splice contents of ipipe to opipe.
1509 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1510 struct pipe_inode_info
*opipe
,
1511 size_t len
, unsigned int flags
)
1513 struct pipe_buffer
*ibuf
, *obuf
;
1515 bool input_wakeup
= false;
1519 ret
= ipipe_prep(ipipe
, flags
);
1523 ret
= opipe_prep(opipe
, flags
);
1528 * Potential ABBA deadlock, work around it by ordering lock
1529 * grabbing by pipe info address. Otherwise two different processes
1530 * could deadlock (one doing tee from A -> B, the other from B -> A).
1532 pipe_double_lock(ipipe
, opipe
);
1535 if (!opipe
->readers
) {
1536 send_sig(SIGPIPE
, current
, 0);
1542 if (!ipipe
->nrbufs
&& !ipipe
->writers
)
1546 * Cannot make any progress, because either the input
1547 * pipe is empty or the output pipe is full.
1549 if (!ipipe
->nrbufs
|| opipe
->nrbufs
>= opipe
->buffers
) {
1550 /* Already processed some buffers, break */
1554 if (flags
& SPLICE_F_NONBLOCK
) {
1560 * We raced with another reader/writer and haven't
1561 * managed to process any buffers. A zero return
1562 * value means EOF, so retry instead.
1569 ibuf
= ipipe
->bufs
+ ipipe
->curbuf
;
1570 nbuf
= (opipe
->curbuf
+ opipe
->nrbufs
) & (opipe
->buffers
- 1);
1571 obuf
= opipe
->bufs
+ nbuf
;
1573 if (len
>= ibuf
->len
) {
1575 * Simply move the whole buffer from ipipe to opipe
1580 ipipe
->curbuf
= (ipipe
->curbuf
+ 1) & (ipipe
->buffers
- 1);
1582 input_wakeup
= true;
1585 * Get a reference to this pipe buffer,
1586 * so we can copy the contents over.
1588 pipe_buf_get(ipipe
, ibuf
);
1592 * Don't inherit the gift flag, we need to
1593 * prevent multiple steals of this page.
1595 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1599 ibuf
->offset
+= obuf
->len
;
1600 ibuf
->len
-= obuf
->len
;
1610 * If we put data in the output pipe, wakeup any potential readers.
1613 wakeup_pipe_readers(opipe
);
1616 wakeup_pipe_writers(ipipe
);
1622 * Link contents of ipipe to opipe.
1624 static int link_pipe(struct pipe_inode_info
*ipipe
,
1625 struct pipe_inode_info
*opipe
,
1626 size_t len
, unsigned int flags
)
1628 struct pipe_buffer
*ibuf
, *obuf
;
1629 int ret
= 0, i
= 0, nbuf
;
1632 * Potential ABBA deadlock, work around it by ordering lock
1633 * grabbing by pipe info address. Otherwise two different processes
1634 * could deadlock (one doing tee from A -> B, the other from B -> A).
1636 pipe_double_lock(ipipe
, opipe
);
1639 if (!opipe
->readers
) {
1640 send_sig(SIGPIPE
, current
, 0);
1647 * If we have iterated all input buffers or ran out of
1648 * output room, break.
1650 if (i
>= ipipe
->nrbufs
|| opipe
->nrbufs
>= opipe
->buffers
)
1653 ibuf
= ipipe
->bufs
+ ((ipipe
->curbuf
+ i
) & (ipipe
->buffers
-1));
1654 nbuf
= (opipe
->curbuf
+ opipe
->nrbufs
) & (opipe
->buffers
- 1);
1657 * Get a reference to this pipe buffer,
1658 * so we can copy the contents over.
1660 pipe_buf_get(ipipe
, ibuf
);
1662 obuf
= opipe
->bufs
+ nbuf
;
1666 * Don't inherit the gift flag, we need to
1667 * prevent multiple steals of this page.
1669 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1671 if (obuf
->len
> len
)
1681 * return EAGAIN if we have the potential of some data in the
1682 * future, otherwise just return 0
1684 if (!ret
&& ipipe
->waiting_writers
&& (flags
& SPLICE_F_NONBLOCK
))
1691 * If we put data in the output pipe, wakeup any potential readers.
1694 wakeup_pipe_readers(opipe
);
1700 * This is a tee(1) implementation that works on pipes. It doesn't copy
1701 * any data, it simply references the 'in' pages on the 'out' pipe.
1702 * The 'flags' used are the SPLICE_F_* variants, currently the only
1703 * applicable one is SPLICE_F_NONBLOCK.
1705 static long do_tee(struct file
*in
, struct file
*out
, size_t len
,
1708 struct pipe_inode_info
*ipipe
= get_pipe_info(in
);
1709 struct pipe_inode_info
*opipe
= get_pipe_info(out
);
1713 * Duplicate the contents of ipipe to opipe without actually
1716 if (ipipe
&& opipe
&& ipipe
!= opipe
) {
1718 * Keep going, unless we encounter an error. The ipipe/opipe
1719 * ordering doesn't really matter.
1721 ret
= ipipe_prep(ipipe
, flags
);
1723 ret
= opipe_prep(opipe
, flags
);
1725 ret
= link_pipe(ipipe
, opipe
, len
, flags
);
1732 SYSCALL_DEFINE4(tee
, int, fdin
, int, fdout
, size_t, len
, unsigned int, flags
)
1743 if (in
.file
->f_mode
& FMODE_READ
) {
1744 struct fd out
= fdget(fdout
);
1746 if (out
.file
->f_mode
& FMODE_WRITE
)
1747 error
= do_tee(in
.file
, out
.file
,