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>
35 #include <linux/aio.h>
39 * Attempt to steal a page from a pipe buffer. This should perhaps go into
40 * a vm helper function, it's already simplified quite a bit by the
41 * addition of remove_mapping(). If success is returned, the caller may
42 * attempt to reuse this page for another destination.
44 static int page_cache_pipe_buf_steal(struct pipe_inode_info
*pipe
,
45 struct pipe_buffer
*buf
)
47 struct page
*page
= buf
->page
;
48 struct address_space
*mapping
;
52 mapping
= page_mapping(page
);
54 WARN_ON(!PageUptodate(page
));
57 * At least for ext2 with nobh option, we need to wait on
58 * writeback completing on this page, since we'll remove it
59 * from the pagecache. Otherwise truncate wont wait on the
60 * page, allowing the disk blocks to be reused by someone else
61 * before we actually wrote our data to them. fs corruption
64 wait_on_page_writeback(page
);
66 if (page_has_private(page
) &&
67 !try_to_release_page(page
, GFP_KERNEL
))
71 * If we succeeded in removing the mapping, set LRU flag
74 if (remove_mapping(mapping
, page
)) {
75 buf
->flags
|= PIPE_BUF_FLAG_LRU
;
81 * Raced with truncate or failed to remove page from current
82 * address space, unlock and return failure.
89 static void page_cache_pipe_buf_release(struct pipe_inode_info
*pipe
,
90 struct pipe_buffer
*buf
)
92 page_cache_release(buf
->page
);
93 buf
->flags
&= ~PIPE_BUF_FLAG_LRU
;
97 * Check whether the contents of buf is OK to access. Since the content
98 * is a page cache page, IO may be in flight.
100 static int page_cache_pipe_buf_confirm(struct pipe_inode_info
*pipe
,
101 struct pipe_buffer
*buf
)
103 struct page
*page
= buf
->page
;
106 if (!PageUptodate(page
)) {
110 * Page got truncated/unhashed. This will cause a 0-byte
111 * splice, if this is the first page.
113 if (!page
->mapping
) {
119 * Uh oh, read-error from disk.
121 if (!PageUptodate(page
)) {
127 * Page is ok afterall, we are done.
138 const struct pipe_buf_operations page_cache_pipe_buf_ops
= {
140 .confirm
= page_cache_pipe_buf_confirm
,
141 .release
= page_cache_pipe_buf_release
,
142 .steal
= page_cache_pipe_buf_steal
,
143 .get
= generic_pipe_buf_get
,
146 static int user_page_pipe_buf_steal(struct pipe_inode_info
*pipe
,
147 struct pipe_buffer
*buf
)
149 if (!(buf
->flags
& PIPE_BUF_FLAG_GIFT
))
152 buf
->flags
|= PIPE_BUF_FLAG_LRU
;
153 return generic_pipe_buf_steal(pipe
, buf
);
156 static const struct pipe_buf_operations user_page_pipe_buf_ops
= {
158 .confirm
= generic_pipe_buf_confirm
,
159 .release
= page_cache_pipe_buf_release
,
160 .steal
= user_page_pipe_buf_steal
,
161 .get
= generic_pipe_buf_get
,
164 static void wakeup_pipe_readers(struct pipe_inode_info
*pipe
)
167 if (waitqueue_active(&pipe
->wait
))
168 wake_up_interruptible(&pipe
->wait
);
169 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
173 * splice_to_pipe - fill passed data into a pipe
174 * @pipe: pipe to fill
178 * @spd contains a map of pages and len/offset tuples, along with
179 * the struct pipe_buf_operations associated with these pages. This
180 * function will link that data to the pipe.
183 ssize_t
splice_to_pipe(struct pipe_inode_info
*pipe
,
184 struct splice_pipe_desc
*spd
)
186 unsigned int spd_pages
= spd
->nr_pages
;
187 int ret
, do_wakeup
, page_nr
;
196 if (!pipe
->readers
) {
197 send_sig(SIGPIPE
, current
, 0);
203 if (pipe
->nrbufs
< pipe
->buffers
) {
204 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
205 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
207 buf
->page
= spd
->pages
[page_nr
];
208 buf
->offset
= spd
->partial
[page_nr
].offset
;
209 buf
->len
= spd
->partial
[page_nr
].len
;
210 buf
->private = spd
->partial
[page_nr
].private;
212 if (spd
->flags
& SPLICE_F_GIFT
)
213 buf
->flags
|= PIPE_BUF_FLAG_GIFT
;
222 if (!--spd
->nr_pages
)
224 if (pipe
->nrbufs
< pipe
->buffers
)
230 if (spd
->flags
& SPLICE_F_NONBLOCK
) {
236 if (signal_pending(current
)) {
244 if (waitqueue_active(&pipe
->wait
))
245 wake_up_interruptible_sync(&pipe
->wait
);
246 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
250 pipe
->waiting_writers
++;
252 pipe
->waiting_writers
--;
258 wakeup_pipe_readers(pipe
);
260 while (page_nr
< spd_pages
)
261 spd
->spd_release(spd
, page_nr
++);
266 void spd_release_page(struct splice_pipe_desc
*spd
, unsigned int i
)
268 page_cache_release(spd
->pages
[i
]);
272 * Check if we need to grow the arrays holding pages and partial page
275 int splice_grow_spd(const struct pipe_inode_info
*pipe
, struct splice_pipe_desc
*spd
)
277 unsigned int buffers
= ACCESS_ONCE(pipe
->buffers
);
279 spd
->nr_pages_max
= buffers
;
280 if (buffers
<= PIPE_DEF_BUFFERS
)
283 spd
->pages
= kmalloc(buffers
* sizeof(struct page
*), GFP_KERNEL
);
284 spd
->partial
= kmalloc(buffers
* sizeof(struct partial_page
), GFP_KERNEL
);
286 if (spd
->pages
&& spd
->partial
)
294 void splice_shrink_spd(struct splice_pipe_desc
*spd
)
296 if (spd
->nr_pages_max
<= PIPE_DEF_BUFFERS
)
304 __generic_file_splice_read(struct file
*in
, loff_t
*ppos
,
305 struct pipe_inode_info
*pipe
, size_t len
,
308 struct address_space
*mapping
= in
->f_mapping
;
309 unsigned int loff
, nr_pages
, req_pages
;
310 struct page
*pages
[PIPE_DEF_BUFFERS
];
311 struct partial_page partial
[PIPE_DEF_BUFFERS
];
313 pgoff_t index
, end_index
;
316 struct splice_pipe_desc spd
= {
319 .nr_pages_max
= PIPE_DEF_BUFFERS
,
321 .ops
= &page_cache_pipe_buf_ops
,
322 .spd_release
= spd_release_page
,
325 if (splice_grow_spd(pipe
, &spd
))
328 index
= *ppos
>> PAGE_CACHE_SHIFT
;
329 loff
= *ppos
& ~PAGE_CACHE_MASK
;
330 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
331 nr_pages
= min(req_pages
, spd
.nr_pages_max
);
334 * Lookup the (hopefully) full range of pages we need.
336 spd
.nr_pages
= find_get_pages_contig(mapping
, index
, nr_pages
, spd
.pages
);
337 index
+= spd
.nr_pages
;
340 * If find_get_pages_contig() returned fewer pages than we needed,
341 * readahead/allocate the rest and fill in the holes.
343 if (spd
.nr_pages
< nr_pages
)
344 page_cache_sync_readahead(mapping
, &in
->f_ra
, in
,
345 index
, req_pages
- spd
.nr_pages
);
348 while (spd
.nr_pages
< nr_pages
) {
350 * Page could be there, find_get_pages_contig() breaks on
353 page
= find_get_page(mapping
, index
);
356 * page didn't exist, allocate one.
358 page
= page_cache_alloc_cold(mapping
);
362 error
= add_to_page_cache_lru(page
, mapping
, index
,
364 if (unlikely(error
)) {
365 page_cache_release(page
);
366 if (error
== -EEXIST
)
371 * add_to_page_cache() locks the page, unlock it
372 * to avoid convoluting the logic below even more.
377 spd
.pages
[spd
.nr_pages
++] = page
;
382 * Now loop over the map and see if we need to start IO on any
383 * pages, fill in the partial map, etc.
385 index
= *ppos
>> PAGE_CACHE_SHIFT
;
386 nr_pages
= spd
.nr_pages
;
388 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
389 unsigned int this_len
;
395 * this_len is the max we'll use from this page
397 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
398 page
= spd
.pages
[page_nr
];
400 if (PageReadahead(page
))
401 page_cache_async_readahead(mapping
, &in
->f_ra
, in
,
402 page
, index
, req_pages
- page_nr
);
405 * If the page isn't uptodate, we may need to start io on it
407 if (!PageUptodate(page
)) {
411 * Page was truncated, or invalidated by the
412 * filesystem. Redo the find/create, but this time the
413 * page is kept locked, so there's no chance of another
414 * race with truncate/invalidate.
416 if (!page
->mapping
) {
418 page
= find_or_create_page(mapping
, index
,
419 mapping_gfp_mask(mapping
));
425 page_cache_release(spd
.pages
[page_nr
]);
426 spd
.pages
[page_nr
] = page
;
429 * page was already under io and is now done, great
431 if (PageUptodate(page
)) {
437 * need to read in the page
439 error
= mapping
->a_ops
->readpage(in
, page
);
440 if (unlikely(error
)) {
442 * We really should re-lookup the page here,
443 * but it complicates things a lot. Instead
444 * lets just do what we already stored, and
445 * we'll get it the next time we are called.
447 if (error
== AOP_TRUNCATED_PAGE
)
455 * i_size must be checked after PageUptodate.
457 isize
= i_size_read(mapping
->host
);
458 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
459 if (unlikely(!isize
|| index
> end_index
))
463 * if this is the last page, see if we need to shrink
464 * the length and stop
466 if (end_index
== index
) {
470 * max good bytes in this page
472 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
477 * force quit after adding this page
479 this_len
= min(this_len
, plen
- loff
);
483 spd
.partial
[page_nr
].offset
= loff
;
484 spd
.partial
[page_nr
].len
= this_len
;
492 * Release any pages at the end, if we quit early. 'page_nr' is how far
493 * we got, 'nr_pages' is how many pages are in the map.
495 while (page_nr
< nr_pages
)
496 page_cache_release(spd
.pages
[page_nr
++]);
497 in
->f_ra
.prev_pos
= (loff_t
)index
<< PAGE_CACHE_SHIFT
;
500 error
= splice_to_pipe(pipe
, &spd
);
502 splice_shrink_spd(&spd
);
507 * generic_file_splice_read - splice data from file to a pipe
508 * @in: file to splice from
509 * @ppos: position in @in
510 * @pipe: pipe to splice to
511 * @len: number of bytes to splice
512 * @flags: splice modifier flags
515 * Will read pages from given file and fill them into a pipe. Can be
516 * used as long as the address_space operations for the source implements
520 ssize_t
generic_file_splice_read(struct file
*in
, loff_t
*ppos
,
521 struct pipe_inode_info
*pipe
, size_t len
,
527 isize
= i_size_read(in
->f_mapping
->host
);
528 if (unlikely(*ppos
>= isize
))
531 left
= isize
- *ppos
;
532 if (unlikely(left
< len
))
535 ret
= __generic_file_splice_read(in
, ppos
, pipe
, len
, flags
);
543 EXPORT_SYMBOL(generic_file_splice_read
);
545 static const struct pipe_buf_operations default_pipe_buf_ops
= {
547 .confirm
= generic_pipe_buf_confirm
,
548 .release
= generic_pipe_buf_release
,
549 .steal
= generic_pipe_buf_steal
,
550 .get
= generic_pipe_buf_get
,
553 static int generic_pipe_buf_nosteal(struct pipe_inode_info
*pipe
,
554 struct pipe_buffer
*buf
)
559 /* Pipe buffer operations for a socket and similar. */
560 const struct pipe_buf_operations nosteal_pipe_buf_ops
= {
562 .confirm
= generic_pipe_buf_confirm
,
563 .release
= generic_pipe_buf_release
,
564 .steal
= generic_pipe_buf_nosteal
,
565 .get
= generic_pipe_buf_get
,
567 EXPORT_SYMBOL(nosteal_pipe_buf_ops
);
569 static ssize_t
kernel_readv(struct file
*file
, const struct iovec
*vec
,
570 unsigned long vlen
, loff_t offset
)
578 /* The cast to a user pointer is valid due to the set_fs() */
579 res
= vfs_readv(file
, (const struct iovec __user
*)vec
, vlen
, &pos
);
585 ssize_t
kernel_write(struct file
*file
, const char *buf
, size_t count
,
593 /* The cast to a user pointer is valid due to the set_fs() */
594 res
= vfs_write(file
, (__force
const char __user
*)buf
, count
, &pos
);
599 EXPORT_SYMBOL(kernel_write
);
601 ssize_t
default_file_splice_read(struct file
*in
, loff_t
*ppos
,
602 struct pipe_inode_info
*pipe
, size_t len
,
605 unsigned int nr_pages
;
606 unsigned int nr_freed
;
608 struct page
*pages
[PIPE_DEF_BUFFERS
];
609 struct partial_page partial
[PIPE_DEF_BUFFERS
];
610 struct iovec
*vec
, __vec
[PIPE_DEF_BUFFERS
];
615 struct splice_pipe_desc spd
= {
618 .nr_pages_max
= PIPE_DEF_BUFFERS
,
620 .ops
= &default_pipe_buf_ops
,
621 .spd_release
= spd_release_page
,
624 if (splice_grow_spd(pipe
, &spd
))
629 if (spd
.nr_pages_max
> PIPE_DEF_BUFFERS
) {
630 vec
= kmalloc(spd
.nr_pages_max
* sizeof(struct iovec
), GFP_KERNEL
);
635 offset
= *ppos
& ~PAGE_CACHE_MASK
;
636 nr_pages
= (len
+ offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
638 for (i
= 0; i
< nr_pages
&& i
< spd
.nr_pages_max
&& len
; i
++) {
641 page
= alloc_page(GFP_USER
);
646 this_len
= min_t(size_t, len
, PAGE_CACHE_SIZE
- offset
);
647 vec
[i
].iov_base
= (void __user
*) page_address(page
);
648 vec
[i
].iov_len
= this_len
;
655 res
= kernel_readv(in
, vec
, spd
.nr_pages
, *ppos
);
666 for (i
= 0; i
< spd
.nr_pages
; i
++) {
667 this_len
= min_t(size_t, vec
[i
].iov_len
, res
);
668 spd
.partial
[i
].offset
= 0;
669 spd
.partial
[i
].len
= this_len
;
671 __free_page(spd
.pages
[i
]);
677 spd
.nr_pages
-= nr_freed
;
679 res
= splice_to_pipe(pipe
, &spd
);
686 splice_shrink_spd(&spd
);
690 for (i
= 0; i
< spd
.nr_pages
; i
++)
691 __free_page(spd
.pages
[i
]);
696 EXPORT_SYMBOL(default_file_splice_read
);
699 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
700 * using sendpage(). Return the number of bytes sent.
702 static int pipe_to_sendpage(struct pipe_inode_info
*pipe
,
703 struct pipe_buffer
*buf
, struct splice_desc
*sd
)
705 struct file
*file
= sd
->u
.file
;
706 loff_t pos
= sd
->pos
;
709 if (!likely(file
->f_op
->sendpage
))
712 more
= (sd
->flags
& SPLICE_F_MORE
) ? MSG_MORE
: 0;
714 if (sd
->len
< sd
->total_len
&& pipe
->nrbufs
> 1)
715 more
|= MSG_SENDPAGE_NOTLAST
;
717 return file
->f_op
->sendpage(file
, buf
->page
, buf
->offset
,
718 sd
->len
, &pos
, more
);
721 static void wakeup_pipe_writers(struct pipe_inode_info
*pipe
)
724 if (waitqueue_active(&pipe
->wait
))
725 wake_up_interruptible(&pipe
->wait
);
726 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
730 * splice_from_pipe_feed - feed available data from a pipe to a file
731 * @pipe: pipe to splice from
732 * @sd: information to @actor
733 * @actor: handler that splices the data
736 * This function loops over the pipe and calls @actor to do the
737 * actual moving of a single struct pipe_buffer to the desired
738 * destination. It returns when there's no more buffers left in
739 * the pipe or if the requested number of bytes (@sd->total_len)
740 * have been copied. It returns a positive number (one) if the
741 * pipe needs to be filled with more data, zero if the required
742 * number of bytes have been copied and -errno on error.
744 * This, together with splice_from_pipe_{begin,end,next}, may be
745 * used to implement the functionality of __splice_from_pipe() when
746 * locking is required around copying the pipe buffers to the
749 static int splice_from_pipe_feed(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
754 while (pipe
->nrbufs
) {
755 struct pipe_buffer
*buf
= pipe
->bufs
+ pipe
->curbuf
;
756 const struct pipe_buf_operations
*ops
= buf
->ops
;
759 if (sd
->len
> sd
->total_len
)
760 sd
->len
= sd
->total_len
;
762 ret
= buf
->ops
->confirm(pipe
, buf
);
769 ret
= actor(pipe
, buf
, sd
);
776 sd
->num_spliced
+= ret
;
779 sd
->total_len
-= ret
;
783 ops
->release(pipe
, buf
);
784 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
787 sd
->need_wakeup
= true;
798 * splice_from_pipe_next - wait for some data to splice from
799 * @pipe: pipe to splice from
800 * @sd: information about the splice operation
803 * This function will wait for some data and return a positive
804 * value (one) if pipe buffers are available. It will return zero
805 * or -errno if no more data needs to be spliced.
807 static int splice_from_pipe_next(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
809 while (!pipe
->nrbufs
) {
813 if (!pipe
->waiting_writers
&& sd
->num_spliced
)
816 if (sd
->flags
& SPLICE_F_NONBLOCK
)
819 if (signal_pending(current
))
822 if (sd
->need_wakeup
) {
823 wakeup_pipe_writers(pipe
);
824 sd
->need_wakeup
= false;
834 * splice_from_pipe_begin - start splicing from pipe
835 * @sd: information about the splice operation
838 * This function should be called before a loop containing
839 * splice_from_pipe_next() and splice_from_pipe_feed() to
840 * initialize the necessary fields of @sd.
842 static void splice_from_pipe_begin(struct splice_desc
*sd
)
845 sd
->need_wakeup
= false;
849 * splice_from_pipe_end - finish splicing from pipe
850 * @pipe: pipe to splice from
851 * @sd: information about the splice operation
854 * This function will wake up pipe writers if necessary. It should
855 * be called after a loop containing splice_from_pipe_next() and
856 * splice_from_pipe_feed().
858 static void splice_from_pipe_end(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
861 wakeup_pipe_writers(pipe
);
865 * __splice_from_pipe - splice data from a pipe to given actor
866 * @pipe: pipe to splice from
867 * @sd: information to @actor
868 * @actor: handler that splices the data
871 * This function does little more than loop over the pipe and call
872 * @actor to do the actual moving of a single struct pipe_buffer to
873 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
877 ssize_t
__splice_from_pipe(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
882 splice_from_pipe_begin(sd
);
884 ret
= splice_from_pipe_next(pipe
, sd
);
886 ret
= splice_from_pipe_feed(pipe
, sd
, actor
);
888 splice_from_pipe_end(pipe
, sd
);
890 return sd
->num_spliced
? sd
->num_spliced
: ret
;
892 EXPORT_SYMBOL(__splice_from_pipe
);
895 * splice_from_pipe - splice data from a pipe to a file
896 * @pipe: pipe to splice from
897 * @out: file to splice to
898 * @ppos: position in @out
899 * @len: how many bytes to splice
900 * @flags: splice modifier flags
901 * @actor: handler that splices the data
904 * See __splice_from_pipe. This function locks the pipe inode,
905 * otherwise it's identical to __splice_from_pipe().
908 ssize_t
splice_from_pipe(struct pipe_inode_info
*pipe
, struct file
*out
,
909 loff_t
*ppos
, size_t len
, unsigned int flags
,
913 struct splice_desc sd
= {
921 ret
= __splice_from_pipe(pipe
, &sd
, actor
);
928 * iter_file_splice_write - splice data from a pipe to a file
930 * @out: file to write to
931 * @ppos: position in @out
932 * @len: number of bytes to splice
933 * @flags: splice modifier flags
936 * Will either move or copy pages (determined by @flags options) from
937 * the given pipe inode to the given file.
938 * This one is ->write_iter-based.
942 iter_file_splice_write(struct pipe_inode_info
*pipe
, struct file
*out
,
943 loff_t
*ppos
, size_t len
, unsigned int flags
)
945 struct splice_desc sd
= {
951 int nbufs
= pipe
->buffers
;
952 struct bio_vec
*array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
956 if (unlikely(!array
))
961 splice_from_pipe_begin(&sd
);
962 while (sd
.total_len
) {
963 struct iov_iter from
;
968 ret
= splice_from_pipe_next(pipe
, &sd
);
972 if (unlikely(nbufs
< pipe
->buffers
)) {
974 nbufs
= pipe
->buffers
;
975 array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
983 /* build the vector */
985 for (n
= 0, idx
= pipe
->curbuf
; left
&& n
< pipe
->nrbufs
; n
++, idx
++) {
986 struct pipe_buffer
*buf
= pipe
->bufs
+ idx
;
987 size_t this_len
= buf
->len
;
992 if (idx
== pipe
->buffers
- 1)
995 ret
= buf
->ops
->confirm(pipe
, buf
);
1002 array
[n
].bv_page
= buf
->page
;
1003 array
[n
].bv_len
= this_len
;
1004 array
[n
].bv_offset
= buf
->offset
;
1009 iov_iter_bvec(&from
, ITER_BVEC
| WRITE
, array
, n
,
1010 sd
.total_len
- left
);
1013 init_sync_kiocb(&kiocb
, out
);
1014 kiocb
.ki_pos
= sd
.pos
;
1015 kiocb
.ki_nbytes
= sd
.total_len
- left
;
1018 ret
= out
->f_op
->write_iter(&kiocb
, &from
);
1019 if (-EIOCBQUEUED
== ret
)
1020 ret
= wait_on_sync_kiocb(&kiocb
);
1025 sd
.num_spliced
+= ret
;
1026 sd
.total_len
-= ret
;
1027 *ppos
= sd
.pos
= kiocb
.ki_pos
;
1029 /* dismiss the fully eaten buffers, adjust the partial one */
1031 struct pipe_buffer
*buf
= pipe
->bufs
+ pipe
->curbuf
;
1032 if (ret
>= buf
->len
) {
1033 const struct pipe_buf_operations
*ops
= buf
->ops
;
1037 ops
->release(pipe
, buf
);
1038 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
1041 sd
.need_wakeup
= true;
1051 splice_from_pipe_end(pipe
, &sd
);
1056 ret
= sd
.num_spliced
;
1061 EXPORT_SYMBOL(iter_file_splice_write
);
1063 static int write_pipe_buf(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
1064 struct splice_desc
*sd
)
1068 loff_t tmp
= sd
->pos
;
1070 data
= kmap(buf
->page
);
1071 ret
= __kernel_write(sd
->u
.file
, data
+ buf
->offset
, sd
->len
, &tmp
);
1077 static ssize_t
default_file_splice_write(struct pipe_inode_info
*pipe
,
1078 struct file
*out
, loff_t
*ppos
,
1079 size_t len
, unsigned int flags
)
1083 ret
= splice_from_pipe(pipe
, out
, ppos
, len
, flags
, write_pipe_buf
);
1091 * generic_splice_sendpage - splice data from a pipe to a socket
1092 * @pipe: pipe to splice from
1093 * @out: socket to write to
1094 * @ppos: position in @out
1095 * @len: number of bytes to splice
1096 * @flags: splice modifier flags
1099 * Will send @len bytes from the pipe to a network socket. No data copying
1103 ssize_t
generic_splice_sendpage(struct pipe_inode_info
*pipe
, struct file
*out
,
1104 loff_t
*ppos
, size_t len
, unsigned int flags
)
1106 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_sendpage
);
1109 EXPORT_SYMBOL(generic_splice_sendpage
);
1112 * Attempt to initiate a splice from pipe to file.
1114 static long do_splice_from(struct pipe_inode_info
*pipe
, struct file
*out
,
1115 loff_t
*ppos
, size_t len
, unsigned int flags
)
1117 ssize_t (*splice_write
)(struct pipe_inode_info
*, struct file
*,
1118 loff_t
*, size_t, unsigned int);
1120 if (out
->f_op
->splice_write
)
1121 splice_write
= out
->f_op
->splice_write
;
1123 splice_write
= default_file_splice_write
;
1125 return splice_write(pipe
, out
, ppos
, len
, flags
);
1129 * Attempt to initiate a splice from a file to a pipe.
1131 static long do_splice_to(struct file
*in
, loff_t
*ppos
,
1132 struct pipe_inode_info
*pipe
, size_t len
,
1135 ssize_t (*splice_read
)(struct file
*, loff_t
*,
1136 struct pipe_inode_info
*, size_t, unsigned int);
1139 if (unlikely(!(in
->f_mode
& FMODE_READ
)))
1142 ret
= rw_verify_area(READ
, in
, ppos
, len
);
1143 if (unlikely(ret
< 0))
1146 if (in
->f_op
->splice_read
)
1147 splice_read
= in
->f_op
->splice_read
;
1149 splice_read
= default_file_splice_read
;
1151 return splice_read(in
, ppos
, pipe
, len
, flags
);
1155 * splice_direct_to_actor - splices data directly between two non-pipes
1156 * @in: file to splice from
1157 * @sd: actor information on where to splice to
1158 * @actor: handles the data splicing
1161 * This is a special case helper to splice directly between two
1162 * points, without requiring an explicit pipe. Internally an allocated
1163 * pipe is cached in the process, and reused during the lifetime of
1167 ssize_t
splice_direct_to_actor(struct file
*in
, struct splice_desc
*sd
,
1168 splice_direct_actor
*actor
)
1170 struct pipe_inode_info
*pipe
;
1177 * We require the input being a regular file, as we don't want to
1178 * randomly drop data for eg socket -> socket splicing. Use the
1179 * piped splicing for that!
1181 i_mode
= file_inode(in
)->i_mode
;
1182 if (unlikely(!S_ISREG(i_mode
) && !S_ISBLK(i_mode
)))
1186 * neither in nor out is a pipe, setup an internal pipe attached to
1187 * 'out' and transfer the wanted data from 'in' to 'out' through that
1189 pipe
= current
->splice_pipe
;
1190 if (unlikely(!pipe
)) {
1191 pipe
= alloc_pipe_info();
1196 * We don't have an immediate reader, but we'll read the stuff
1197 * out of the pipe right after the splice_to_pipe(). So set
1198 * PIPE_READERS appropriately.
1202 current
->splice_pipe
= pipe
;
1210 len
= sd
->total_len
;
1214 * Don't block on output, we have to drain the direct pipe.
1216 sd
->flags
&= ~SPLICE_F_NONBLOCK
;
1220 loff_t pos
= sd
->pos
, prev_pos
= pos
;
1222 ret
= do_splice_to(in
, &pos
, pipe
, len
, flags
);
1223 if (unlikely(ret
<= 0))
1227 sd
->total_len
= read_len
;
1230 * NOTE: nonblocking mode only applies to the input. We
1231 * must not do the output in nonblocking mode as then we
1232 * could get stuck data in the internal pipe:
1234 ret
= actor(pipe
, sd
);
1235 if (unlikely(ret
<= 0)) {
1244 if (ret
< read_len
) {
1245 sd
->pos
= prev_pos
+ ret
;
1251 pipe
->nrbufs
= pipe
->curbuf
= 0;
1257 * If we did an incomplete transfer we must release
1258 * the pipe buffers in question:
1260 for (i
= 0; i
< pipe
->buffers
; i
++) {
1261 struct pipe_buffer
*buf
= pipe
->bufs
+ i
;
1264 buf
->ops
->release(pipe
, buf
);
1274 EXPORT_SYMBOL(splice_direct_to_actor
);
1276 static int direct_splice_actor(struct pipe_inode_info
*pipe
,
1277 struct splice_desc
*sd
)
1279 struct file
*file
= sd
->u
.file
;
1281 return do_splice_from(pipe
, file
, sd
->opos
, sd
->total_len
,
1286 * do_splice_direct - splices data directly between two files
1287 * @in: file to splice from
1288 * @ppos: input file offset
1289 * @out: file to splice to
1290 * @opos: output file offset
1291 * @len: number of bytes to splice
1292 * @flags: splice modifier flags
1295 * For use by do_sendfile(). splice can easily emulate sendfile, but
1296 * doing it in the application would incur an extra system call
1297 * (splice in + splice out, as compared to just sendfile()). So this helper
1298 * can splice directly through a process-private pipe.
1301 long do_splice_direct(struct file
*in
, loff_t
*ppos
, struct file
*out
,
1302 loff_t
*opos
, size_t len
, unsigned int flags
)
1304 struct splice_desc sd
= {
1314 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
1317 if (unlikely(out
->f_flags
& O_APPEND
))
1320 ret
= rw_verify_area(WRITE
, out
, opos
, len
);
1321 if (unlikely(ret
< 0))
1324 ret
= splice_direct_to_actor(in
, &sd
, direct_splice_actor
);
1330 EXPORT_SYMBOL(do_splice_direct
);
1332 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1333 struct pipe_inode_info
*opipe
,
1334 size_t len
, unsigned int flags
);
1337 * Determine where to splice to/from.
1339 static long do_splice(struct file
*in
, loff_t __user
*off_in
,
1340 struct file
*out
, loff_t __user
*off_out
,
1341 size_t len
, unsigned int flags
)
1343 struct pipe_inode_info
*ipipe
;
1344 struct pipe_inode_info
*opipe
;
1348 ipipe
= get_pipe_info(in
);
1349 opipe
= get_pipe_info(out
);
1351 if (ipipe
&& opipe
) {
1352 if (off_in
|| off_out
)
1355 if (!(in
->f_mode
& FMODE_READ
))
1358 if (!(out
->f_mode
& FMODE_WRITE
))
1361 /* Splicing to self would be fun, but... */
1365 return splice_pipe_to_pipe(ipipe
, opipe
, len
, flags
);
1372 if (!(out
->f_mode
& FMODE_PWRITE
))
1374 if (copy_from_user(&offset
, off_out
, sizeof(loff_t
)))
1377 offset
= out
->f_pos
;
1380 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
1383 if (unlikely(out
->f_flags
& O_APPEND
))
1386 ret
= rw_verify_area(WRITE
, out
, &offset
, len
);
1387 if (unlikely(ret
< 0))
1390 file_start_write(out
);
1391 ret
= do_splice_from(ipipe
, out
, &offset
, len
, flags
);
1392 file_end_write(out
);
1395 out
->f_pos
= offset
;
1396 else if (copy_to_user(off_out
, &offset
, sizeof(loff_t
)))
1406 if (!(in
->f_mode
& FMODE_PREAD
))
1408 if (copy_from_user(&offset
, off_in
, sizeof(loff_t
)))
1414 ret
= do_splice_to(in
, &offset
, opipe
, len
, flags
);
1418 else if (copy_to_user(off_in
, &offset
, sizeof(loff_t
)))
1428 * Map an iov into an array of pages and offset/length tupples. With the
1429 * partial_page structure, we can map several non-contiguous ranges into
1430 * our ones pages[] map instead of splitting that operation into pieces.
1431 * Could easily be exported as a generic helper for other users, in which
1432 * case one would probably want to add a 'max_nr_pages' parameter as well.
1434 static int get_iovec_page_array(const struct iovec __user
*iov
,
1435 unsigned int nr_vecs
, struct page
**pages
,
1436 struct partial_page
*partial
, bool aligned
,
1437 unsigned int pipe_buffers
)
1439 int buffers
= 0, error
= 0;
1442 unsigned long off
, npages
;
1449 if (copy_from_user(&entry
, iov
, sizeof(entry
)))
1452 base
= entry
.iov_base
;
1453 len
= entry
.iov_len
;
1456 * Sanity check this iovec. 0 read succeeds.
1462 if (!access_ok(VERIFY_READ
, base
, len
))
1466 * Get this base offset and number of pages, then map
1467 * in the user pages.
1469 off
= (unsigned long) base
& ~PAGE_MASK
;
1472 * If asked for alignment, the offset must be zero and the
1473 * length a multiple of the PAGE_SIZE.
1476 if (aligned
&& (off
|| len
& ~PAGE_MASK
))
1479 npages
= (off
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1480 if (npages
> pipe_buffers
- buffers
)
1481 npages
= pipe_buffers
- buffers
;
1483 error
= get_user_pages_fast((unsigned long)base
, npages
,
1484 0, &pages
[buffers
]);
1486 if (unlikely(error
<= 0))
1490 * Fill this contiguous range into the partial page map.
1492 for (i
= 0; i
< error
; i
++) {
1493 const int plen
= min_t(size_t, len
, PAGE_SIZE
- off
);
1495 partial
[buffers
].offset
= off
;
1496 partial
[buffers
].len
= plen
;
1504 * We didn't complete this iov, stop here since it probably
1505 * means we have to move some of this into a pipe to
1506 * be able to continue.
1512 * Don't continue if we mapped fewer pages than we asked for,
1513 * or if we mapped the max number of pages that we have
1516 if (error
< npages
|| buffers
== pipe_buffers
)
1529 static int pipe_to_user(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
1530 struct splice_desc
*sd
)
1532 int n
= copy_page_to_iter(buf
->page
, buf
->offset
, sd
->len
, sd
->u
.data
);
1533 return n
== sd
->len
? n
: -EFAULT
;
1537 * For lack of a better implementation, implement vmsplice() to userspace
1538 * as a simple copy of the pipes pages to the user iov.
1540 static long vmsplice_to_user(struct file
*file
, const struct iovec __user
*uiov
,
1541 unsigned long nr_segs
, unsigned int flags
)
1543 struct pipe_inode_info
*pipe
;
1544 struct splice_desc sd
;
1546 struct iovec iovstack
[UIO_FASTIOV
];
1547 struct iovec
*iov
= iovstack
;
1548 struct iov_iter iter
;
1551 pipe
= get_pipe_info(file
);
1555 ret
= rw_copy_check_uvector(READ
, uiov
, nr_segs
,
1556 ARRAY_SIZE(iovstack
), iovstack
, &iov
);
1561 iov_iter_init(&iter
, READ
, iov
, nr_segs
, count
);
1564 sd
.total_len
= count
;
1570 ret
= __splice_from_pipe(pipe
, &sd
, pipe_to_user
);
1574 if (iov
!= iovstack
)
1581 * vmsplice splices a user address range into a pipe. It can be thought of
1582 * as splice-from-memory, where the regular splice is splice-from-file (or
1583 * to file). In both cases the output is a pipe, naturally.
1585 static long vmsplice_to_pipe(struct file
*file
, const struct iovec __user
*iov
,
1586 unsigned long nr_segs
, unsigned int flags
)
1588 struct pipe_inode_info
*pipe
;
1589 struct page
*pages
[PIPE_DEF_BUFFERS
];
1590 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1591 struct splice_pipe_desc spd
= {
1594 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1596 .ops
= &user_page_pipe_buf_ops
,
1597 .spd_release
= spd_release_page
,
1601 pipe
= get_pipe_info(file
);
1605 if (splice_grow_spd(pipe
, &spd
))
1608 spd
.nr_pages
= get_iovec_page_array(iov
, nr_segs
, spd
.pages
,
1611 if (spd
.nr_pages
<= 0)
1614 ret
= splice_to_pipe(pipe
, &spd
);
1616 splice_shrink_spd(&spd
);
1621 * Note that vmsplice only really supports true splicing _from_ user memory
1622 * to a pipe, not the other way around. Splicing from user memory is a simple
1623 * operation that can be supported without any funky alignment restrictions
1624 * or nasty vm tricks. We simply map in the user memory and fill them into
1625 * a pipe. The reverse isn't quite as easy, though. There are two possible
1626 * solutions for that:
1628 * - memcpy() the data internally, at which point we might as well just
1629 * do a regular read() on the buffer anyway.
1630 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1631 * has restriction limitations on both ends of the pipe).
1633 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1636 SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct iovec __user
*, iov
,
1637 unsigned long, nr_segs
, unsigned int, flags
)
1642 if (unlikely(nr_segs
> UIO_MAXIOV
))
1644 else if (unlikely(!nr_segs
))
1650 if (f
.file
->f_mode
& FMODE_WRITE
)
1651 error
= vmsplice_to_pipe(f
.file
, iov
, nr_segs
, flags
);
1652 else if (f
.file
->f_mode
& FMODE_READ
)
1653 error
= vmsplice_to_user(f
.file
, iov
, nr_segs
, flags
);
1661 #ifdef CONFIG_COMPAT
1662 COMPAT_SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct compat_iovec __user
*, iov32
,
1663 unsigned int, nr_segs
, unsigned int, flags
)
1666 struct iovec __user
*iov
;
1667 if (nr_segs
> UIO_MAXIOV
)
1669 iov
= compat_alloc_user_space(nr_segs
* sizeof(struct iovec
));
1670 for (i
= 0; i
< nr_segs
; i
++) {
1671 struct compat_iovec v
;
1672 if (get_user(v
.iov_base
, &iov32
[i
].iov_base
) ||
1673 get_user(v
.iov_len
, &iov32
[i
].iov_len
) ||
1674 put_user(compat_ptr(v
.iov_base
), &iov
[i
].iov_base
) ||
1675 put_user(v
.iov_len
, &iov
[i
].iov_len
))
1678 return sys_vmsplice(fd
, iov
, nr_segs
, flags
);
1682 SYSCALL_DEFINE6(splice
, int, fd_in
, loff_t __user
*, off_in
,
1683 int, fd_out
, loff_t __user
*, off_out
,
1684 size_t, len
, unsigned int, flags
)
1695 if (in
.file
->f_mode
& FMODE_READ
) {
1696 out
= fdget(fd_out
);
1698 if (out
.file
->f_mode
& FMODE_WRITE
)
1699 error
= do_splice(in
.file
, off_in
,
1711 * Make sure there's data to read. Wait for input if we can, otherwise
1712 * return an appropriate error.
1714 static int ipipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1719 * Check ->nrbufs without the inode lock first. This function
1720 * is speculative anyways, so missing one is ok.
1728 while (!pipe
->nrbufs
) {
1729 if (signal_pending(current
)) {
1735 if (!pipe
->waiting_writers
) {
1736 if (flags
& SPLICE_F_NONBLOCK
) {
1749 * Make sure there's writeable room. Wait for room if we can, otherwise
1750 * return an appropriate error.
1752 static int opipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1757 * Check ->nrbufs without the inode lock first. This function
1758 * is speculative anyways, so missing one is ok.
1760 if (pipe
->nrbufs
< pipe
->buffers
)
1766 while (pipe
->nrbufs
>= pipe
->buffers
) {
1767 if (!pipe
->readers
) {
1768 send_sig(SIGPIPE
, current
, 0);
1772 if (flags
& SPLICE_F_NONBLOCK
) {
1776 if (signal_pending(current
)) {
1780 pipe
->waiting_writers
++;
1782 pipe
->waiting_writers
--;
1790 * Splice contents of ipipe to opipe.
1792 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1793 struct pipe_inode_info
*opipe
,
1794 size_t len
, unsigned int flags
)
1796 struct pipe_buffer
*ibuf
, *obuf
;
1798 bool input_wakeup
= false;
1802 ret
= ipipe_prep(ipipe
, flags
);
1806 ret
= opipe_prep(opipe
, flags
);
1811 * Potential ABBA deadlock, work around it by ordering lock
1812 * grabbing by pipe info address. Otherwise two different processes
1813 * could deadlock (one doing tee from A -> B, the other from B -> A).
1815 pipe_double_lock(ipipe
, opipe
);
1818 if (!opipe
->readers
) {
1819 send_sig(SIGPIPE
, current
, 0);
1825 if (!ipipe
->nrbufs
&& !ipipe
->writers
)
1829 * Cannot make any progress, because either the input
1830 * pipe is empty or the output pipe is full.
1832 if (!ipipe
->nrbufs
|| opipe
->nrbufs
>= opipe
->buffers
) {
1833 /* Already processed some buffers, break */
1837 if (flags
& SPLICE_F_NONBLOCK
) {
1843 * We raced with another reader/writer and haven't
1844 * managed to process any buffers. A zero return
1845 * value means EOF, so retry instead.
1852 ibuf
= ipipe
->bufs
+ ipipe
->curbuf
;
1853 nbuf
= (opipe
->curbuf
+ opipe
->nrbufs
) & (opipe
->buffers
- 1);
1854 obuf
= opipe
->bufs
+ nbuf
;
1856 if (len
>= ibuf
->len
) {
1858 * Simply move the whole buffer from ipipe to opipe
1863 ipipe
->curbuf
= (ipipe
->curbuf
+ 1) & (ipipe
->buffers
- 1);
1865 input_wakeup
= true;
1868 * Get a reference to this pipe buffer,
1869 * so we can copy the contents over.
1871 ibuf
->ops
->get(ipipe
, ibuf
);
1875 * Don't inherit the gift flag, we need to
1876 * prevent multiple steals of this page.
1878 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1882 ibuf
->offset
+= obuf
->len
;
1883 ibuf
->len
-= obuf
->len
;
1893 * If we put data in the output pipe, wakeup any potential readers.
1896 wakeup_pipe_readers(opipe
);
1899 wakeup_pipe_writers(ipipe
);
1905 * Link contents of ipipe to opipe.
1907 static int link_pipe(struct pipe_inode_info
*ipipe
,
1908 struct pipe_inode_info
*opipe
,
1909 size_t len
, unsigned int flags
)
1911 struct pipe_buffer
*ibuf
, *obuf
;
1912 int ret
= 0, i
= 0, nbuf
;
1915 * Potential ABBA deadlock, work around it by ordering lock
1916 * grabbing by pipe info address. Otherwise two different processes
1917 * could deadlock (one doing tee from A -> B, the other from B -> A).
1919 pipe_double_lock(ipipe
, opipe
);
1922 if (!opipe
->readers
) {
1923 send_sig(SIGPIPE
, current
, 0);
1930 * If we have iterated all input buffers or ran out of
1931 * output room, break.
1933 if (i
>= ipipe
->nrbufs
|| opipe
->nrbufs
>= opipe
->buffers
)
1936 ibuf
= ipipe
->bufs
+ ((ipipe
->curbuf
+ i
) & (ipipe
->buffers
-1));
1937 nbuf
= (opipe
->curbuf
+ opipe
->nrbufs
) & (opipe
->buffers
- 1);
1940 * Get a reference to this pipe buffer,
1941 * so we can copy the contents over.
1943 ibuf
->ops
->get(ipipe
, ibuf
);
1945 obuf
= opipe
->bufs
+ nbuf
;
1949 * Don't inherit the gift flag, we need to
1950 * prevent multiple steals of this page.
1952 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1954 if (obuf
->len
> len
)
1964 * return EAGAIN if we have the potential of some data in the
1965 * future, otherwise just return 0
1967 if (!ret
&& ipipe
->waiting_writers
&& (flags
& SPLICE_F_NONBLOCK
))
1974 * If we put data in the output pipe, wakeup any potential readers.
1977 wakeup_pipe_readers(opipe
);
1983 * This is a tee(1) implementation that works on pipes. It doesn't copy
1984 * any data, it simply references the 'in' pages on the 'out' pipe.
1985 * The 'flags' used are the SPLICE_F_* variants, currently the only
1986 * applicable one is SPLICE_F_NONBLOCK.
1988 static long do_tee(struct file
*in
, struct file
*out
, size_t len
,
1991 struct pipe_inode_info
*ipipe
= get_pipe_info(in
);
1992 struct pipe_inode_info
*opipe
= get_pipe_info(out
);
1996 * Duplicate the contents of ipipe to opipe without actually
1999 if (ipipe
&& opipe
&& ipipe
!= opipe
) {
2001 * Keep going, unless we encounter an error. The ipipe/opipe
2002 * ordering doesn't really matter.
2004 ret
= ipipe_prep(ipipe
, flags
);
2006 ret
= opipe_prep(opipe
, flags
);
2008 ret
= link_pipe(ipipe
, opipe
, len
, flags
);
2015 SYSCALL_DEFINE4(tee
, int, fdin
, int, fdout
, size_t, len
, unsigned int, flags
)
2026 if (in
.file
->f_mode
& FMODE_READ
) {
2027 struct fd out
= fdget(fdout
);
2029 if (out
.file
->f_mode
& FMODE_WRITE
)
2030 error
= do_tee(in
.file
, out
.file
,