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
, do_wakeup
, page_nr
;
198 if (!pipe
->readers
) {
199 send_sig(SIGPIPE
, current
, 0);
205 if (pipe
->nrbufs
< pipe
->buffers
) {
206 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
207 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
209 buf
->page
= spd
->pages
[page_nr
];
210 buf
->offset
= spd
->partial
[page_nr
].offset
;
211 buf
->len
= spd
->partial
[page_nr
].len
;
212 buf
->private = spd
->partial
[page_nr
].private;
214 if (spd
->flags
& SPLICE_F_GIFT
)
215 buf
->flags
|= PIPE_BUF_FLAG_GIFT
;
224 if (!--spd
->nr_pages
)
226 if (pipe
->nrbufs
< pipe
->buffers
)
232 if (spd
->flags
& SPLICE_F_NONBLOCK
) {
238 if (signal_pending(current
)) {
245 wakeup_pipe_readers(pipe
);
249 pipe
->waiting_writers
++;
251 pipe
->waiting_writers
--;
257 wakeup_pipe_readers(pipe
);
259 while (page_nr
< spd_pages
)
260 spd
->spd_release(spd
, page_nr
++);
264 EXPORT_SYMBOL_GPL(splice_to_pipe
);
266 void spd_release_page(struct splice_pipe_desc
*spd
, unsigned int i
)
268 put_page(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_SHIFT
;
329 loff
= *ppos
& ~PAGE_MASK
;
330 req_pages
= (len
+ loff
+ PAGE_SIZE
- 1) >> PAGE_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
,
363 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
364 if (unlikely(error
)) {
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_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_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
) {
419 page
= find_or_create_page(mapping
, index
,
420 mapping_gfp_mask(mapping
));
426 put_page(spd
.pages
[page_nr
]);
427 spd
.pages
[page_nr
] = page
;
430 * page was already under io and is now done, great
432 if (PageUptodate(page
)) {
438 * need to read in the page
440 error
= mapping
->a_ops
->readpage(in
, page
);
441 if (unlikely(error
)) {
445 if (error
== AOP_TRUNCATED_PAGE
)
453 * i_size must be checked after PageUptodate.
455 isize
= i_size_read(mapping
->host
);
456 end_index
= (isize
- 1) >> PAGE_SHIFT
;
457 if (unlikely(!isize
|| index
> end_index
))
461 * if this is the last page, see if we need to shrink
462 * the length and stop
464 if (end_index
== index
) {
468 * max good bytes in this page
470 plen
= ((isize
- 1) & ~PAGE_MASK
) + 1;
475 * force quit after adding this page
477 this_len
= min(this_len
, plen
- loff
);
481 spd
.partial
[page_nr
].offset
= loff
;
482 spd
.partial
[page_nr
].len
= this_len
;
490 * Release any pages at the end, if we quit early. 'page_nr' is how far
491 * we got, 'nr_pages' is how many pages are in the map.
493 while (page_nr
< nr_pages
)
494 put_page(spd
.pages
[page_nr
++]);
495 in
->f_ra
.prev_pos
= (loff_t
)index
<< PAGE_SHIFT
;
498 error
= splice_to_pipe(pipe
, &spd
);
500 splice_shrink_spd(&spd
);
505 * generic_file_splice_read - splice data from file to a pipe
506 * @in: file to splice from
507 * @ppos: position in @in
508 * @pipe: pipe to splice to
509 * @len: number of bytes to splice
510 * @flags: splice modifier flags
513 * Will read pages from given file and fill them into a pipe. Can be
514 * used as long as the address_space operations for the source implements
518 ssize_t
generic_file_splice_read(struct file
*in
, loff_t
*ppos
,
519 struct pipe_inode_info
*pipe
, size_t len
,
525 if (IS_DAX(in
->f_mapping
->host
))
526 return default_file_splice_read(in
, ppos
, pipe
, len
, flags
);
528 isize
= i_size_read(in
->f_mapping
->host
);
529 if (unlikely(*ppos
>= isize
))
532 left
= isize
- *ppos
;
533 if (unlikely(left
< len
))
536 ret
= __generic_file_splice_read(in
, ppos
, pipe
, len
, flags
);
544 EXPORT_SYMBOL(generic_file_splice_read
);
546 static const struct pipe_buf_operations default_pipe_buf_ops
= {
548 .confirm
= generic_pipe_buf_confirm
,
549 .release
= generic_pipe_buf_release
,
550 .steal
= generic_pipe_buf_steal
,
551 .get
= generic_pipe_buf_get
,
554 static int generic_pipe_buf_nosteal(struct pipe_inode_info
*pipe
,
555 struct pipe_buffer
*buf
)
560 /* Pipe buffer operations for a socket and similar. */
561 const struct pipe_buf_operations nosteal_pipe_buf_ops
= {
563 .confirm
= generic_pipe_buf_confirm
,
564 .release
= generic_pipe_buf_release
,
565 .steal
= generic_pipe_buf_nosteal
,
566 .get
= generic_pipe_buf_get
,
568 EXPORT_SYMBOL(nosteal_pipe_buf_ops
);
570 static ssize_t
kernel_readv(struct file
*file
, const struct iovec
*vec
,
571 unsigned long vlen
, loff_t offset
)
579 /* The cast to a user pointer is valid due to the set_fs() */
580 res
= vfs_readv(file
, (const struct iovec __user
*)vec
, vlen
, &pos
, 0);
586 ssize_t
kernel_write(struct file
*file
, const char *buf
, size_t count
,
594 /* The cast to a user pointer is valid due to the set_fs() */
595 res
= vfs_write(file
, (__force
const char __user
*)buf
, count
, &pos
);
600 EXPORT_SYMBOL(kernel_write
);
602 ssize_t
default_file_splice_read(struct file
*in
, loff_t
*ppos
,
603 struct pipe_inode_info
*pipe
, size_t len
,
606 unsigned int nr_pages
;
607 unsigned int nr_freed
;
609 struct page
*pages
[PIPE_DEF_BUFFERS
];
610 struct partial_page partial
[PIPE_DEF_BUFFERS
];
611 struct iovec
*vec
, __vec
[PIPE_DEF_BUFFERS
];
616 struct splice_pipe_desc spd
= {
619 .nr_pages_max
= PIPE_DEF_BUFFERS
,
621 .ops
= &default_pipe_buf_ops
,
622 .spd_release
= spd_release_page
,
625 if (splice_grow_spd(pipe
, &spd
))
630 if (spd
.nr_pages_max
> PIPE_DEF_BUFFERS
) {
631 vec
= kmalloc(spd
.nr_pages_max
* sizeof(struct iovec
), GFP_KERNEL
);
636 offset
= *ppos
& ~PAGE_MASK
;
637 nr_pages
= (len
+ offset
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
639 for (i
= 0; i
< nr_pages
&& i
< spd
.nr_pages_max
&& len
; i
++) {
642 page
= alloc_page(GFP_USER
);
647 this_len
= min_t(size_t, len
, PAGE_SIZE
- offset
);
648 vec
[i
].iov_base
= (void __user
*) page_address(page
);
649 vec
[i
].iov_len
= this_len
;
656 res
= kernel_readv(in
, vec
, spd
.nr_pages
, *ppos
);
667 for (i
= 0; i
< spd
.nr_pages
; i
++) {
668 this_len
= min_t(size_t, vec
[i
].iov_len
, res
);
669 spd
.partial
[i
].offset
= 0;
670 spd
.partial
[i
].len
= this_len
;
672 __free_page(spd
.pages
[i
]);
678 spd
.nr_pages
-= nr_freed
;
680 res
= splice_to_pipe(pipe
, &spd
);
687 splice_shrink_spd(&spd
);
691 for (i
= 0; i
< spd
.nr_pages
; i
++)
692 __free_page(spd
.pages
[i
]);
697 EXPORT_SYMBOL(default_file_splice_read
);
700 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
701 * using sendpage(). Return the number of bytes sent.
703 static int pipe_to_sendpage(struct pipe_inode_info
*pipe
,
704 struct pipe_buffer
*buf
, struct splice_desc
*sd
)
706 struct file
*file
= sd
->u
.file
;
707 loff_t pos
= sd
->pos
;
710 if (!likely(file
->f_op
->sendpage
))
713 more
= (sd
->flags
& SPLICE_F_MORE
) ? MSG_MORE
: 0;
715 if (sd
->len
< sd
->total_len
&& pipe
->nrbufs
> 1)
716 more
|= MSG_SENDPAGE_NOTLAST
;
718 return file
->f_op
->sendpage(file
, buf
->page
, buf
->offset
,
719 sd
->len
, &pos
, more
);
722 static void wakeup_pipe_writers(struct pipe_inode_info
*pipe
)
725 if (waitqueue_active(&pipe
->wait
))
726 wake_up_interruptible(&pipe
->wait
);
727 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
731 * splice_from_pipe_feed - feed available data from a pipe to a file
732 * @pipe: pipe to splice from
733 * @sd: information to @actor
734 * @actor: handler that splices the data
737 * This function loops over the pipe and calls @actor to do the
738 * actual moving of a single struct pipe_buffer to the desired
739 * destination. It returns when there's no more buffers left in
740 * the pipe or if the requested number of bytes (@sd->total_len)
741 * have been copied. It returns a positive number (one) if the
742 * pipe needs to be filled with more data, zero if the required
743 * number of bytes have been copied and -errno on error.
745 * This, together with splice_from_pipe_{begin,end,next}, may be
746 * used to implement the functionality of __splice_from_pipe() when
747 * locking is required around copying the pipe buffers to the
750 static int splice_from_pipe_feed(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
755 while (pipe
->nrbufs
) {
756 struct pipe_buffer
*buf
= pipe
->bufs
+ pipe
->curbuf
;
757 const struct pipe_buf_operations
*ops
= buf
->ops
;
760 if (sd
->len
> sd
->total_len
)
761 sd
->len
= sd
->total_len
;
763 ret
= buf
->ops
->confirm(pipe
, buf
);
770 ret
= actor(pipe
, buf
, sd
);
777 sd
->num_spliced
+= ret
;
780 sd
->total_len
-= ret
;
784 ops
->release(pipe
, buf
);
785 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
788 sd
->need_wakeup
= true;
799 * splice_from_pipe_next - wait for some data to splice from
800 * @pipe: pipe to splice from
801 * @sd: information about the splice operation
804 * This function will wait for some data and return a positive
805 * value (one) if pipe buffers are available. It will return zero
806 * or -errno if no more data needs to be spliced.
808 static int splice_from_pipe_next(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
811 * Check for signal early to make process killable when there are
812 * always buffers available
814 if (signal_pending(current
))
817 while (!pipe
->nrbufs
) {
821 if (!pipe
->waiting_writers
&& sd
->num_spliced
)
824 if (sd
->flags
& SPLICE_F_NONBLOCK
)
827 if (signal_pending(current
))
830 if (sd
->need_wakeup
) {
831 wakeup_pipe_writers(pipe
);
832 sd
->need_wakeup
= false;
842 * splice_from_pipe_begin - start splicing from pipe
843 * @sd: information about the splice operation
846 * This function should be called before a loop containing
847 * splice_from_pipe_next() and splice_from_pipe_feed() to
848 * initialize the necessary fields of @sd.
850 static void splice_from_pipe_begin(struct splice_desc
*sd
)
853 sd
->need_wakeup
= false;
857 * splice_from_pipe_end - finish splicing from pipe
858 * @pipe: pipe to splice from
859 * @sd: information about the splice operation
862 * This function will wake up pipe writers if necessary. It should
863 * be called after a loop containing splice_from_pipe_next() and
864 * splice_from_pipe_feed().
866 static void splice_from_pipe_end(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
869 wakeup_pipe_writers(pipe
);
873 * __splice_from_pipe - splice data from a pipe to given actor
874 * @pipe: pipe to splice from
875 * @sd: information to @actor
876 * @actor: handler that splices the data
879 * This function does little more than loop over the pipe and call
880 * @actor to do the actual moving of a single struct pipe_buffer to
881 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
885 ssize_t
__splice_from_pipe(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
,
890 splice_from_pipe_begin(sd
);
893 ret
= splice_from_pipe_next(pipe
, sd
);
895 ret
= splice_from_pipe_feed(pipe
, sd
, actor
);
897 splice_from_pipe_end(pipe
, sd
);
899 return sd
->num_spliced
? sd
->num_spliced
: ret
;
901 EXPORT_SYMBOL(__splice_from_pipe
);
904 * splice_from_pipe - splice data from a pipe to a file
905 * @pipe: pipe to splice from
906 * @out: file to splice to
907 * @ppos: position in @out
908 * @len: how many bytes to splice
909 * @flags: splice modifier flags
910 * @actor: handler that splices the data
913 * See __splice_from_pipe. This function locks the pipe inode,
914 * otherwise it's identical to __splice_from_pipe().
917 ssize_t
splice_from_pipe(struct pipe_inode_info
*pipe
, struct file
*out
,
918 loff_t
*ppos
, size_t len
, unsigned int flags
,
922 struct splice_desc sd
= {
930 ret
= __splice_from_pipe(pipe
, &sd
, actor
);
937 * iter_file_splice_write - splice data from a pipe to a file
939 * @out: file to write to
940 * @ppos: position in @out
941 * @len: number of bytes to splice
942 * @flags: splice modifier flags
945 * Will either move or copy pages (determined by @flags options) from
946 * the given pipe inode to the given file.
947 * This one is ->write_iter-based.
951 iter_file_splice_write(struct pipe_inode_info
*pipe
, struct file
*out
,
952 loff_t
*ppos
, size_t len
, unsigned int flags
)
954 struct splice_desc sd
= {
960 int nbufs
= pipe
->buffers
;
961 struct bio_vec
*array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
965 if (unlikely(!array
))
970 splice_from_pipe_begin(&sd
);
971 while (sd
.total_len
) {
972 struct iov_iter from
;
976 ret
= splice_from_pipe_next(pipe
, &sd
);
980 if (unlikely(nbufs
< pipe
->buffers
)) {
982 nbufs
= pipe
->buffers
;
983 array
= kcalloc(nbufs
, sizeof(struct bio_vec
),
991 /* build the vector */
993 for (n
= 0, idx
= pipe
->curbuf
; left
&& n
< pipe
->nrbufs
; n
++, idx
++) {
994 struct pipe_buffer
*buf
= pipe
->bufs
+ idx
;
995 size_t this_len
= buf
->len
;
1000 if (idx
== pipe
->buffers
- 1)
1003 ret
= buf
->ops
->confirm(pipe
, buf
);
1004 if (unlikely(ret
)) {
1005 if (ret
== -ENODATA
)
1010 array
[n
].bv_page
= buf
->page
;
1011 array
[n
].bv_len
= this_len
;
1012 array
[n
].bv_offset
= buf
->offset
;
1016 iov_iter_bvec(&from
, ITER_BVEC
| WRITE
, array
, n
,
1017 sd
.total_len
- left
);
1018 ret
= vfs_iter_write(out
, &from
, &sd
.pos
);
1022 sd
.num_spliced
+= ret
;
1023 sd
.total_len
-= ret
;
1026 /* dismiss the fully eaten buffers, adjust the partial one */
1028 struct pipe_buffer
*buf
= pipe
->bufs
+ pipe
->curbuf
;
1029 if (ret
>= buf
->len
) {
1030 const struct pipe_buf_operations
*ops
= buf
->ops
;
1034 ops
->release(pipe
, buf
);
1035 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (pipe
->buffers
- 1);
1038 sd
.need_wakeup
= true;
1048 splice_from_pipe_end(pipe
, &sd
);
1053 ret
= sd
.num_spliced
;
1058 EXPORT_SYMBOL(iter_file_splice_write
);
1060 static int write_pipe_buf(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
1061 struct splice_desc
*sd
)
1065 loff_t tmp
= sd
->pos
;
1067 data
= kmap(buf
->page
);
1068 ret
= __kernel_write(sd
->u
.file
, data
+ buf
->offset
, sd
->len
, &tmp
);
1074 static ssize_t
default_file_splice_write(struct pipe_inode_info
*pipe
,
1075 struct file
*out
, loff_t
*ppos
,
1076 size_t len
, unsigned int flags
)
1080 ret
= splice_from_pipe(pipe
, out
, ppos
, len
, flags
, write_pipe_buf
);
1088 * generic_splice_sendpage - splice data from a pipe to a socket
1089 * @pipe: pipe to splice from
1090 * @out: socket to write to
1091 * @ppos: position in @out
1092 * @len: number of bytes to splice
1093 * @flags: splice modifier flags
1096 * Will send @len bytes from the pipe to a network socket. No data copying
1100 ssize_t
generic_splice_sendpage(struct pipe_inode_info
*pipe
, struct file
*out
,
1101 loff_t
*ppos
, size_t len
, unsigned int flags
)
1103 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_sendpage
);
1106 EXPORT_SYMBOL(generic_splice_sendpage
);
1109 * Attempt to initiate a splice from pipe to file.
1111 static long do_splice_from(struct pipe_inode_info
*pipe
, struct file
*out
,
1112 loff_t
*ppos
, size_t len
, unsigned int flags
)
1114 ssize_t (*splice_write
)(struct pipe_inode_info
*, struct file
*,
1115 loff_t
*, size_t, unsigned int);
1117 if (out
->f_op
->splice_write
)
1118 splice_write
= out
->f_op
->splice_write
;
1120 splice_write
= default_file_splice_write
;
1122 return splice_write(pipe
, out
, ppos
, len
, flags
);
1126 * Attempt to initiate a splice from a file to a pipe.
1128 static long do_splice_to(struct file
*in
, loff_t
*ppos
,
1129 struct pipe_inode_info
*pipe
, size_t len
,
1132 ssize_t (*splice_read
)(struct file
*, loff_t
*,
1133 struct pipe_inode_info
*, size_t, unsigned int);
1136 if (unlikely(!(in
->f_mode
& FMODE_READ
)))
1139 ret
= rw_verify_area(READ
, in
, ppos
, len
);
1140 if (unlikely(ret
< 0))
1143 if (unlikely(len
> MAX_RW_COUNT
))
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
;
1217 more
= sd
->flags
& SPLICE_F_MORE
;
1221 loff_t pos
= sd
->pos
, prev_pos
= pos
;
1223 ret
= do_splice_to(in
, &pos
, pipe
, len
, flags
);
1224 if (unlikely(ret
<= 0))
1228 sd
->total_len
= read_len
;
1231 * If more data is pending, set SPLICE_F_MORE
1232 * If this is the last data and SPLICE_F_MORE was not set
1233 * initially, clears it.
1236 sd
->flags
|= SPLICE_F_MORE
;
1238 sd
->flags
&= ~SPLICE_F_MORE
;
1240 * NOTE: nonblocking mode only applies to the input. We
1241 * must not do the output in nonblocking mode as then we
1242 * could get stuck data in the internal pipe:
1244 ret
= actor(pipe
, sd
);
1245 if (unlikely(ret
<= 0)) {
1254 if (ret
< read_len
) {
1255 sd
->pos
= prev_pos
+ ret
;
1261 pipe
->nrbufs
= pipe
->curbuf
= 0;
1267 * If we did an incomplete transfer we must release
1268 * the pipe buffers in question:
1270 for (i
= 0; i
< pipe
->buffers
; i
++) {
1271 struct pipe_buffer
*buf
= pipe
->bufs
+ i
;
1274 buf
->ops
->release(pipe
, buf
);
1284 EXPORT_SYMBOL(splice_direct_to_actor
);
1286 static int direct_splice_actor(struct pipe_inode_info
*pipe
,
1287 struct splice_desc
*sd
)
1289 struct file
*file
= sd
->u
.file
;
1291 return do_splice_from(pipe
, file
, sd
->opos
, sd
->total_len
,
1296 * do_splice_direct - splices data directly between two files
1297 * @in: file to splice from
1298 * @ppos: input file offset
1299 * @out: file to splice to
1300 * @opos: output file offset
1301 * @len: number of bytes to splice
1302 * @flags: splice modifier flags
1305 * For use by do_sendfile(). splice can easily emulate sendfile, but
1306 * doing it in the application would incur an extra system call
1307 * (splice in + splice out, as compared to just sendfile()). So this helper
1308 * can splice directly through a process-private pipe.
1311 long do_splice_direct(struct file
*in
, loff_t
*ppos
, struct file
*out
,
1312 loff_t
*opos
, size_t len
, unsigned int flags
)
1314 struct splice_desc sd
= {
1324 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
1327 if (unlikely(out
->f_flags
& O_APPEND
))
1330 ret
= rw_verify_area(WRITE
, out
, opos
, len
);
1331 if (unlikely(ret
< 0))
1334 ret
= splice_direct_to_actor(in
, &sd
, direct_splice_actor
);
1340 EXPORT_SYMBOL(do_splice_direct
);
1342 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1343 struct pipe_inode_info
*opipe
,
1344 size_t len
, unsigned int flags
);
1347 * Determine where to splice to/from.
1349 static long do_splice(struct file
*in
, loff_t __user
*off_in
,
1350 struct file
*out
, loff_t __user
*off_out
,
1351 size_t len
, unsigned int flags
)
1353 struct pipe_inode_info
*ipipe
;
1354 struct pipe_inode_info
*opipe
;
1358 ipipe
= get_pipe_info(in
);
1359 opipe
= get_pipe_info(out
);
1361 if (ipipe
&& opipe
) {
1362 if (off_in
|| off_out
)
1365 if (!(in
->f_mode
& FMODE_READ
))
1368 if (!(out
->f_mode
& FMODE_WRITE
))
1371 /* Splicing to self would be fun, but... */
1375 return splice_pipe_to_pipe(ipipe
, opipe
, len
, flags
);
1382 if (!(out
->f_mode
& FMODE_PWRITE
))
1384 if (copy_from_user(&offset
, off_out
, sizeof(loff_t
)))
1387 offset
= out
->f_pos
;
1390 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
1393 if (unlikely(out
->f_flags
& O_APPEND
))
1396 ret
= rw_verify_area(WRITE
, out
, &offset
, len
);
1397 if (unlikely(ret
< 0))
1400 file_start_write(out
);
1401 ret
= do_splice_from(ipipe
, out
, &offset
, len
, flags
);
1402 file_end_write(out
);
1405 out
->f_pos
= offset
;
1406 else if (copy_to_user(off_out
, &offset
, sizeof(loff_t
)))
1416 if (!(in
->f_mode
& FMODE_PREAD
))
1418 if (copy_from_user(&offset
, off_in
, sizeof(loff_t
)))
1424 ret
= do_splice_to(in
, &offset
, opipe
, len
, flags
);
1428 else if (copy_to_user(off_in
, &offset
, sizeof(loff_t
)))
1438 * Map an iov into an array of pages and offset/length tupples. With the
1439 * partial_page structure, we can map several non-contiguous ranges into
1440 * our ones pages[] map instead of splitting that operation into pieces.
1441 * Could easily be exported as a generic helper for other users, in which
1442 * case one would probably want to add a 'max_nr_pages' parameter as well.
1444 static int get_iovec_page_array(const struct iovec __user
*iov
,
1445 unsigned int nr_vecs
, struct page
**pages
,
1446 struct partial_page
*partial
, bool aligned
,
1447 unsigned int pipe_buffers
)
1449 int buffers
= 0, error
= 0;
1452 unsigned long off
, npages
;
1459 if (copy_from_user(&entry
, iov
, sizeof(entry
)))
1462 base
= entry
.iov_base
;
1463 len
= entry
.iov_len
;
1466 * Sanity check this iovec. 0 read succeeds.
1472 if (!access_ok(VERIFY_READ
, base
, len
))
1476 * Get this base offset and number of pages, then map
1477 * in the user pages.
1479 off
= (unsigned long) base
& ~PAGE_MASK
;
1482 * If asked for alignment, the offset must be zero and the
1483 * length a multiple of the PAGE_SIZE.
1486 if (aligned
&& (off
|| len
& ~PAGE_MASK
))
1489 npages
= (off
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1490 if (npages
> pipe_buffers
- buffers
)
1491 npages
= pipe_buffers
- buffers
;
1493 error
= get_user_pages_fast((unsigned long)base
, npages
,
1494 0, &pages
[buffers
]);
1496 if (unlikely(error
<= 0))
1500 * Fill this contiguous range into the partial page map.
1502 for (i
= 0; i
< error
; i
++) {
1503 const int plen
= min_t(size_t, len
, PAGE_SIZE
- off
);
1505 partial
[buffers
].offset
= off
;
1506 partial
[buffers
].len
= plen
;
1514 * We didn't complete this iov, stop here since it probably
1515 * means we have to move some of this into a pipe to
1516 * be able to continue.
1522 * Don't continue if we mapped fewer pages than we asked for,
1523 * or if we mapped the max number of pages that we have
1526 if (error
< npages
|| buffers
== pipe_buffers
)
1539 static int pipe_to_user(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
1540 struct splice_desc
*sd
)
1542 int n
= copy_page_to_iter(buf
->page
, buf
->offset
, sd
->len
, sd
->u
.data
);
1543 return n
== sd
->len
? n
: -EFAULT
;
1547 * For lack of a better implementation, implement vmsplice() to userspace
1548 * as a simple copy of the pipes pages to the user iov.
1550 static long vmsplice_to_user(struct file
*file
, const struct iovec __user
*uiov
,
1551 unsigned long nr_segs
, unsigned int flags
)
1553 struct pipe_inode_info
*pipe
;
1554 struct splice_desc sd
;
1556 struct iovec iovstack
[UIO_FASTIOV
];
1557 struct iovec
*iov
= iovstack
;
1558 struct iov_iter iter
;
1560 pipe
= get_pipe_info(file
);
1564 ret
= import_iovec(READ
, uiov
, nr_segs
,
1565 ARRAY_SIZE(iovstack
), &iov
, &iter
);
1569 sd
.total_len
= iov_iter_count(&iter
);
1577 ret
= __splice_from_pipe(pipe
, &sd
, pipe_to_user
);
1586 * vmsplice splices a user address range into a pipe. It can be thought of
1587 * as splice-from-memory, where the regular splice is splice-from-file (or
1588 * to file). In both cases the output is a pipe, naturally.
1590 static long vmsplice_to_pipe(struct file
*file
, const struct iovec __user
*iov
,
1591 unsigned long nr_segs
, unsigned int flags
)
1593 struct pipe_inode_info
*pipe
;
1594 struct page
*pages
[PIPE_DEF_BUFFERS
];
1595 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1596 struct splice_pipe_desc spd
= {
1599 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1601 .ops
= &user_page_pipe_buf_ops
,
1602 .spd_release
= spd_release_page
,
1606 pipe
= get_pipe_info(file
);
1610 if (splice_grow_spd(pipe
, &spd
))
1613 spd
.nr_pages
= get_iovec_page_array(iov
, nr_segs
, spd
.pages
,
1616 if (spd
.nr_pages
<= 0)
1619 ret
= splice_to_pipe(pipe
, &spd
);
1621 splice_shrink_spd(&spd
);
1626 * Note that vmsplice only really supports true splicing _from_ user memory
1627 * to a pipe, not the other way around. Splicing from user memory is a simple
1628 * operation that can be supported without any funky alignment restrictions
1629 * or nasty vm tricks. We simply map in the user memory and fill them into
1630 * a pipe. The reverse isn't quite as easy, though. There are two possible
1631 * solutions for that:
1633 * - memcpy() the data internally, at which point we might as well just
1634 * do a regular read() on the buffer anyway.
1635 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1636 * has restriction limitations on both ends of the pipe).
1638 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1641 SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct iovec __user
*, iov
,
1642 unsigned long, nr_segs
, unsigned int, flags
)
1647 if (unlikely(nr_segs
> UIO_MAXIOV
))
1649 else if (unlikely(!nr_segs
))
1655 if (f
.file
->f_mode
& FMODE_WRITE
)
1656 error
= vmsplice_to_pipe(f
.file
, iov
, nr_segs
, flags
);
1657 else if (f
.file
->f_mode
& FMODE_READ
)
1658 error
= vmsplice_to_user(f
.file
, iov
, nr_segs
, flags
);
1666 #ifdef CONFIG_COMPAT
1667 COMPAT_SYSCALL_DEFINE4(vmsplice
, int, fd
, const struct compat_iovec __user
*, iov32
,
1668 unsigned int, nr_segs
, unsigned int, flags
)
1671 struct iovec __user
*iov
;
1672 if (nr_segs
> UIO_MAXIOV
)
1674 iov
= compat_alloc_user_space(nr_segs
* sizeof(struct iovec
));
1675 for (i
= 0; i
< nr_segs
; i
++) {
1676 struct compat_iovec v
;
1677 if (get_user(v
.iov_base
, &iov32
[i
].iov_base
) ||
1678 get_user(v
.iov_len
, &iov32
[i
].iov_len
) ||
1679 put_user(compat_ptr(v
.iov_base
), &iov
[i
].iov_base
) ||
1680 put_user(v
.iov_len
, &iov
[i
].iov_len
))
1683 return sys_vmsplice(fd
, iov
, nr_segs
, flags
);
1687 SYSCALL_DEFINE6(splice
, int, fd_in
, loff_t __user
*, off_in
,
1688 int, fd_out
, loff_t __user
*, off_out
,
1689 size_t, len
, unsigned int, flags
)
1700 if (in
.file
->f_mode
& FMODE_READ
) {
1701 out
= fdget(fd_out
);
1703 if (out
.file
->f_mode
& FMODE_WRITE
)
1704 error
= do_splice(in
.file
, off_in
,
1716 * Make sure there's data to read. Wait for input if we can, otherwise
1717 * return an appropriate error.
1719 static int ipipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1724 * Check ->nrbufs without the inode lock first. This function
1725 * is speculative anyways, so missing one is ok.
1733 while (!pipe
->nrbufs
) {
1734 if (signal_pending(current
)) {
1740 if (!pipe
->waiting_writers
) {
1741 if (flags
& SPLICE_F_NONBLOCK
) {
1754 * Make sure there's writeable room. Wait for room if we can, otherwise
1755 * return an appropriate error.
1757 static int opipe_prep(struct pipe_inode_info
*pipe
, unsigned int flags
)
1762 * Check ->nrbufs without the inode lock first. This function
1763 * is speculative anyways, so missing one is ok.
1765 if (pipe
->nrbufs
< pipe
->buffers
)
1771 while (pipe
->nrbufs
>= pipe
->buffers
) {
1772 if (!pipe
->readers
) {
1773 send_sig(SIGPIPE
, current
, 0);
1777 if (flags
& SPLICE_F_NONBLOCK
) {
1781 if (signal_pending(current
)) {
1785 pipe
->waiting_writers
++;
1787 pipe
->waiting_writers
--;
1795 * Splice contents of ipipe to opipe.
1797 static int splice_pipe_to_pipe(struct pipe_inode_info
*ipipe
,
1798 struct pipe_inode_info
*opipe
,
1799 size_t len
, unsigned int flags
)
1801 struct pipe_buffer
*ibuf
, *obuf
;
1803 bool input_wakeup
= false;
1807 ret
= ipipe_prep(ipipe
, flags
);
1811 ret
= opipe_prep(opipe
, flags
);
1816 * Potential ABBA deadlock, work around it by ordering lock
1817 * grabbing by pipe info address. Otherwise two different processes
1818 * could deadlock (one doing tee from A -> B, the other from B -> A).
1820 pipe_double_lock(ipipe
, opipe
);
1823 if (!opipe
->readers
) {
1824 send_sig(SIGPIPE
, current
, 0);
1830 if (!ipipe
->nrbufs
&& !ipipe
->writers
)
1834 * Cannot make any progress, because either the input
1835 * pipe is empty or the output pipe is full.
1837 if (!ipipe
->nrbufs
|| opipe
->nrbufs
>= opipe
->buffers
) {
1838 /* Already processed some buffers, break */
1842 if (flags
& SPLICE_F_NONBLOCK
) {
1848 * We raced with another reader/writer and haven't
1849 * managed to process any buffers. A zero return
1850 * value means EOF, so retry instead.
1857 ibuf
= ipipe
->bufs
+ ipipe
->curbuf
;
1858 nbuf
= (opipe
->curbuf
+ opipe
->nrbufs
) & (opipe
->buffers
- 1);
1859 obuf
= opipe
->bufs
+ nbuf
;
1861 if (len
>= ibuf
->len
) {
1863 * Simply move the whole buffer from ipipe to opipe
1868 ipipe
->curbuf
= (ipipe
->curbuf
+ 1) & (ipipe
->buffers
- 1);
1870 input_wakeup
= true;
1873 * Get a reference to this pipe buffer,
1874 * so we can copy the contents over.
1876 ibuf
->ops
->get(ipipe
, ibuf
);
1880 * Don't inherit the gift flag, we need to
1881 * prevent multiple steals of this page.
1883 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1887 ibuf
->offset
+= obuf
->len
;
1888 ibuf
->len
-= obuf
->len
;
1898 * If we put data in the output pipe, wakeup any potential readers.
1901 wakeup_pipe_readers(opipe
);
1904 wakeup_pipe_writers(ipipe
);
1910 * Link contents of ipipe to opipe.
1912 static int link_pipe(struct pipe_inode_info
*ipipe
,
1913 struct pipe_inode_info
*opipe
,
1914 size_t len
, unsigned int flags
)
1916 struct pipe_buffer
*ibuf
, *obuf
;
1917 int ret
= 0, i
= 0, nbuf
;
1920 * Potential ABBA deadlock, work around it by ordering lock
1921 * grabbing by pipe info address. Otherwise two different processes
1922 * could deadlock (one doing tee from A -> B, the other from B -> A).
1924 pipe_double_lock(ipipe
, opipe
);
1927 if (!opipe
->readers
) {
1928 send_sig(SIGPIPE
, current
, 0);
1935 * If we have iterated all input buffers or ran out of
1936 * output room, break.
1938 if (i
>= ipipe
->nrbufs
|| opipe
->nrbufs
>= opipe
->buffers
)
1941 ibuf
= ipipe
->bufs
+ ((ipipe
->curbuf
+ i
) & (ipipe
->buffers
-1));
1942 nbuf
= (opipe
->curbuf
+ opipe
->nrbufs
) & (opipe
->buffers
- 1);
1945 * Get a reference to this pipe buffer,
1946 * so we can copy the contents over.
1948 ibuf
->ops
->get(ipipe
, ibuf
);
1950 obuf
= opipe
->bufs
+ nbuf
;
1954 * Don't inherit the gift flag, we need to
1955 * prevent multiple steals of this page.
1957 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1959 if (obuf
->len
> len
)
1969 * return EAGAIN if we have the potential of some data in the
1970 * future, otherwise just return 0
1972 if (!ret
&& ipipe
->waiting_writers
&& (flags
& SPLICE_F_NONBLOCK
))
1979 * If we put data in the output pipe, wakeup any potential readers.
1982 wakeup_pipe_readers(opipe
);
1988 * This is a tee(1) implementation that works on pipes. It doesn't copy
1989 * any data, it simply references the 'in' pages on the 'out' pipe.
1990 * The 'flags' used are the SPLICE_F_* variants, currently the only
1991 * applicable one is SPLICE_F_NONBLOCK.
1993 static long do_tee(struct file
*in
, struct file
*out
, size_t len
,
1996 struct pipe_inode_info
*ipipe
= get_pipe_info(in
);
1997 struct pipe_inode_info
*opipe
= get_pipe_info(out
);
2001 * Duplicate the contents of ipipe to opipe without actually
2004 if (ipipe
&& opipe
&& ipipe
!= opipe
) {
2006 * Keep going, unless we encounter an error. The ipipe/opipe
2007 * ordering doesn't really matter.
2009 ret
= ipipe_prep(ipipe
, flags
);
2011 ret
= opipe_prep(opipe
, flags
);
2013 ret
= link_pipe(ipipe
, opipe
, len
, flags
);
2020 SYSCALL_DEFINE4(tee
, int, fdin
, int, fdout
, size_t, len
, unsigned int, flags
)
2031 if (in
.file
->f_mode
& FMODE_READ
) {
2032 struct fd out
= fdget(fdout
);
2034 if (out
.file
->f_mode
& FMODE_WRITE
)
2035 error
= do_tee(in
.file
, out
.file
,