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@suse.de>
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/pipe_fs_i.h>
24 #include <linux/mm_inline.h>
25 #include <linux/swap.h>
26 #include <linux/writeback.h>
27 #include <linux/buffer_head.h>
28 #include <linux/module.h>
29 #include <linux/syscalls.h>
30 #include <linux/uio.h>
38 * Passed to splice_to_pipe
40 struct splice_pipe_desc
{
41 struct page
**pages
; /* page map */
42 struct partial_page
*partial
; /* pages[] may not be contig */
43 int nr_pages
; /* number of pages in map */
44 unsigned int flags
; /* splice flags */
45 struct pipe_buf_operations
*ops
;/* ops associated with output pipe */
49 * Attempt to steal a page from a pipe buffer. This should perhaps go into
50 * a vm helper function, it's already simplified quite a bit by the
51 * addition of remove_mapping(). If success is returned, the caller may
52 * attempt to reuse this page for another destination.
54 static int page_cache_pipe_buf_steal(struct pipe_inode_info
*info
,
55 struct pipe_buffer
*buf
)
57 struct page
*page
= buf
->page
;
58 struct address_space
*mapping
= page_mapping(page
);
62 WARN_ON(!PageUptodate(page
));
65 * At least for ext2 with nobh option, we need to wait on writeback
66 * completing on this page, since we'll remove it from the pagecache.
67 * Otherwise truncate wont wait on the page, allowing the disk
68 * blocks to be reused by someone else before we actually wrote our
69 * data to them. fs corruption ensues.
71 wait_on_page_writeback(page
);
73 if (PagePrivate(page
))
74 try_to_release_page(page
, mapping_gfp_mask(mapping
));
76 if (!remove_mapping(mapping
, page
)) {
84 static void page_cache_pipe_buf_release(struct pipe_inode_info
*info
,
85 struct pipe_buffer
*buf
)
87 page_cache_release(buf
->page
);
90 static int page_cache_pipe_buf_pin(struct pipe_inode_info
*info
,
91 struct pipe_buffer
*buf
)
93 struct page
*page
= buf
->page
;
96 if (!PageUptodate(page
)) {
100 * Page got truncated/unhashed. This will cause a 0-byte
101 * splice, if this is the first page.
103 if (!page
->mapping
) {
109 * Uh oh, read-error from disk.
111 if (!PageUptodate(page
)) {
117 * Page is ok afterall, we are done.
128 static struct pipe_buf_operations page_cache_pipe_buf_ops
= {
130 .map
= generic_pipe_buf_map
,
131 .unmap
= generic_pipe_buf_unmap
,
132 .pin
= page_cache_pipe_buf_pin
,
133 .release
= page_cache_pipe_buf_release
,
134 .steal
= page_cache_pipe_buf_steal
,
135 .get
= generic_pipe_buf_get
,
138 static int user_page_pipe_buf_steal(struct pipe_inode_info
*pipe
,
139 struct pipe_buffer
*buf
)
141 if (!(buf
->flags
& PIPE_BUF_FLAG_GIFT
))
144 return generic_pipe_buf_steal(pipe
, buf
);
147 static struct pipe_buf_operations user_page_pipe_buf_ops
= {
149 .map
= generic_pipe_buf_map
,
150 .unmap
= generic_pipe_buf_unmap
,
151 .pin
= generic_pipe_buf_pin
,
152 .release
= page_cache_pipe_buf_release
,
153 .steal
= user_page_pipe_buf_steal
,
154 .get
= generic_pipe_buf_get
,
158 * Pipe output worker. This sets up our pipe format with the page cache
159 * pipe buffer operations. Otherwise very similar to the regular pipe_writev().
161 static ssize_t
splice_to_pipe(struct pipe_inode_info
*pipe
,
162 struct splice_pipe_desc
*spd
)
164 int ret
, do_wakeup
, page_nr
;
171 mutex_lock(&pipe
->inode
->i_mutex
);
174 if (!pipe
->readers
) {
175 send_sig(SIGPIPE
, current
, 0);
181 if (pipe
->nrbufs
< PIPE_BUFFERS
) {
182 int newbuf
= (pipe
->curbuf
+ pipe
->nrbufs
) & (PIPE_BUFFERS
- 1);
183 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
185 buf
->page
= spd
->pages
[page_nr
];
186 buf
->offset
= spd
->partial
[page_nr
].offset
;
187 buf
->len
= spd
->partial
[page_nr
].len
;
189 if (spd
->flags
& SPLICE_F_GIFT
)
190 buf
->flags
|= PIPE_BUF_FLAG_GIFT
;
199 if (!--spd
->nr_pages
)
201 if (pipe
->nrbufs
< PIPE_BUFFERS
)
207 if (spd
->flags
& SPLICE_F_NONBLOCK
) {
213 if (signal_pending(current
)) {
221 if (waitqueue_active(&pipe
->wait
))
222 wake_up_interruptible_sync(&pipe
->wait
);
223 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
227 pipe
->waiting_writers
++;
229 pipe
->waiting_writers
--;
233 mutex_unlock(&pipe
->inode
->i_mutex
);
237 if (waitqueue_active(&pipe
->wait
))
238 wake_up_interruptible(&pipe
->wait
);
239 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
242 while (page_nr
< spd
->nr_pages
)
243 page_cache_release(spd
->pages
[page_nr
++]);
249 __generic_file_splice_read(struct file
*in
, loff_t
*ppos
,
250 struct pipe_inode_info
*pipe
, size_t len
,
253 struct address_space
*mapping
= in
->f_mapping
;
254 unsigned int loff
, nr_pages
;
255 struct page
*pages
[PIPE_BUFFERS
];
256 struct partial_page partial
[PIPE_BUFFERS
];
258 pgoff_t index
, end_index
;
262 struct splice_pipe_desc spd
= {
266 .ops
= &page_cache_pipe_buf_ops
,
269 index
= *ppos
>> PAGE_CACHE_SHIFT
;
270 loff
= *ppos
& ~PAGE_CACHE_MASK
;
271 nr_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
273 if (nr_pages
> PIPE_BUFFERS
)
274 nr_pages
= PIPE_BUFFERS
;
277 * Initiate read-ahead on this page range. however, don't call into
278 * read-ahead if this is a non-zero offset (we are likely doing small
279 * chunk splice and the page is already there) for a single page.
281 if (!loff
|| nr_pages
> 1)
282 page_cache_readahead(mapping
, &in
->f_ra
, in
, index
, nr_pages
);
285 * Now fill in the holes:
291 * Lookup the (hopefully) full range of pages we need.
293 spd
.nr_pages
= find_get_pages_contig(mapping
, index
, nr_pages
, pages
);
296 * If find_get_pages_contig() returned fewer pages than we needed,
299 index
+= spd
.nr_pages
;
300 while (spd
.nr_pages
< nr_pages
) {
302 * Page could be there, find_get_pages_contig() breaks on
305 page
= find_get_page(mapping
, index
);
308 * Make sure the read-ahead engine is notified
309 * about this failure.
311 handle_ra_miss(mapping
, &in
->f_ra
, index
);
314 * page didn't exist, allocate one.
316 page
= page_cache_alloc_cold(mapping
);
320 error
= add_to_page_cache_lru(page
, mapping
, index
,
321 mapping_gfp_mask(mapping
));
322 if (unlikely(error
)) {
323 page_cache_release(page
);
327 * add_to_page_cache() locks the page, unlock it
328 * to avoid convoluting the logic below even more.
333 pages
[spd
.nr_pages
++] = page
;
338 * Now loop over the map and see if we need to start IO on any
339 * pages, fill in the partial map, etc.
341 index
= *ppos
>> PAGE_CACHE_SHIFT
;
342 nr_pages
= spd
.nr_pages
;
344 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
345 unsigned int this_len
;
351 * this_len is the max we'll use from this page
353 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
354 page
= pages
[page_nr
];
357 * If the page isn't uptodate, we may need to start io on it
359 if (!PageUptodate(page
)) {
361 * If in nonblock mode then dont block on waiting
362 * for an in-flight io page
364 if (flags
& SPLICE_F_NONBLOCK
)
370 * page was truncated, stop here. if this isn't the
371 * first page, we'll just complete what we already
374 if (!page
->mapping
) {
379 * page was already under io and is now done, great
381 if (PageUptodate(page
)) {
387 * need to read in the page
389 error
= mapping
->a_ops
->readpage(in
, page
);
390 if (unlikely(error
)) {
392 * We really should re-lookup the page here,
393 * but it complicates things a lot. Instead
394 * lets just do what we already stored, and
395 * we'll get it the next time we are called.
397 if (error
== AOP_TRUNCATED_PAGE
)
404 * i_size must be checked after ->readpage().
406 isize
= i_size_read(mapping
->host
);
407 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
408 if (unlikely(!isize
|| index
> end_index
))
412 * if this is the last page, see if we need to shrink
413 * the length and stop
415 if (end_index
== index
) {
416 loff
= PAGE_CACHE_SIZE
- (isize
& ~PAGE_CACHE_MASK
);
417 if (total_len
+ loff
> isize
)
420 * force quit after adding this page
423 this_len
= min(this_len
, loff
);
428 partial
[page_nr
].offset
= loff
;
429 partial
[page_nr
].len
= this_len
;
431 total_len
+= this_len
;
438 * Release any pages at the end, if we quit early. 'i' is how far
439 * we got, 'nr_pages' is how many pages are in the map.
441 while (page_nr
< nr_pages
)
442 page_cache_release(pages
[page_nr
++]);
445 return splice_to_pipe(pipe
, &spd
);
451 * generic_file_splice_read - splice data from file to a pipe
452 * @in: file to splice from
453 * @pipe: pipe to splice to
454 * @len: number of bytes to splice
455 * @flags: splice modifier flags
457 * Will read pages from given file and fill them into a pipe.
459 ssize_t
generic_file_splice_read(struct file
*in
, loff_t
*ppos
,
460 struct pipe_inode_info
*pipe
, size_t len
,
470 ret
= __generic_file_splice_read(in
, ppos
, pipe
, len
, flags
);
477 if (flags
& SPLICE_F_NONBLOCK
) {
494 EXPORT_SYMBOL(generic_file_splice_read
);
497 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
498 * using sendpage(). Return the number of bytes sent.
500 static int pipe_to_sendpage(struct pipe_inode_info
*info
,
501 struct pipe_buffer
*buf
, struct splice_desc
*sd
)
503 struct file
*file
= sd
->file
;
504 loff_t pos
= sd
->pos
;
507 ret
= buf
->ops
->pin(info
, buf
);
509 more
= (sd
->flags
& SPLICE_F_MORE
) || sd
->len
< sd
->total_len
;
511 ret
= file
->f_op
->sendpage(file
, buf
->page
, buf
->offset
,
512 sd
->len
, &pos
, more
);
519 * This is a little more tricky than the file -> pipe splicing. There are
520 * basically three cases:
522 * - Destination page already exists in the address space and there
523 * are users of it. For that case we have no other option that
524 * copying the data. Tough luck.
525 * - Destination page already exists in the address space, but there
526 * are no users of it. Make sure it's uptodate, then drop it. Fall
527 * through to last case.
528 * - Destination page does not exist, we can add the pipe page to
529 * the page cache and avoid the copy.
531 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
532 * sd->flags), we attempt to migrate pages from the pipe to the output
533 * file address space page cache. This is possible if no one else has
534 * the pipe page referenced outside of the pipe and page cache. If
535 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
536 * a new page in the output file page cache and fill/dirty that.
538 static int pipe_to_file(struct pipe_inode_info
*info
, struct pipe_buffer
*buf
,
539 struct splice_desc
*sd
)
541 struct file
*file
= sd
->file
;
542 struct address_space
*mapping
= file
->f_mapping
;
543 gfp_t gfp_mask
= mapping_gfp_mask(mapping
);
544 unsigned int offset
, this_len
;
550 * make sure the data in this buffer is uptodate
552 ret
= buf
->ops
->pin(info
, buf
);
556 index
= sd
->pos
>> PAGE_CACHE_SHIFT
;
557 offset
= sd
->pos
& ~PAGE_CACHE_MASK
;
560 if (this_len
+ offset
> PAGE_CACHE_SIZE
)
561 this_len
= PAGE_CACHE_SIZE
- offset
;
564 * Reuse buf page, if SPLICE_F_MOVE is set and we are doing a full
567 if ((sd
->flags
& SPLICE_F_MOVE
) && this_len
== PAGE_CACHE_SIZE
) {
569 * If steal succeeds, buf->page is now pruned from the vm
570 * side (page cache) and we can reuse it. The page will also
571 * be locked on successful return.
573 if (buf
->ops
->steal(info
, buf
))
577 page_cache_get(page
);
580 * page must be on the LRU for adding to the pagecache.
581 * Check this without grabbing the zone lock, if it isn't
582 * the do grab the zone lock, recheck, and add if necessary.
584 if (!PageLRU(page
)) {
585 struct zone
*zone
= page_zone(page
);
587 spin_lock_irq(&zone
->lru_lock
);
588 if (!PageLRU(page
)) {
590 add_page_to_inactive_list(zone
, page
);
592 spin_unlock_irq(&zone
->lru_lock
);
595 if (add_to_page_cache(page
, mapping
, index
, gfp_mask
)) {
596 page_cache_release(page
);
602 page
= find_lock_page(mapping
, index
);
605 page
= page_cache_alloc_cold(mapping
);
610 * This will also lock the page
612 ret
= add_to_page_cache_lru(page
, mapping
, index
,
619 * We get here with the page locked. If the page is also
620 * uptodate, we don't need to do more. If it isn't, we
621 * may need to bring it in if we are not going to overwrite
624 if (!PageUptodate(page
)) {
625 if (this_len
< PAGE_CACHE_SIZE
) {
626 ret
= mapping
->a_ops
->readpage(file
, page
);
632 if (!PageUptodate(page
)) {
634 * Page got invalidated, repeat.
636 if (!page
->mapping
) {
638 page_cache_release(page
);
645 SetPageUptodate(page
);
649 ret
= mapping
->a_ops
->prepare_write(file
, page
, offset
, offset
+this_len
);
650 if (ret
== AOP_TRUNCATED_PAGE
) {
651 page_cache_release(page
);
656 if (buf
->page
!= page
) {
658 * Careful, ->map() uses KM_USER0!
660 char *src
= buf
->ops
->map(info
, buf
, 1);
661 char *dst
= kmap_atomic(page
, KM_USER1
);
663 memcpy(dst
+ offset
, src
+ buf
->offset
, this_len
);
664 flush_dcache_page(page
);
665 kunmap_atomic(dst
, KM_USER1
);
666 buf
->ops
->unmap(info
, buf
, src
);
669 ret
= mapping
->a_ops
->commit_write(file
, page
, offset
, offset
+this_len
);
672 * Return the number of bytes written and mark page as
673 * accessed, we are now done!
676 mark_page_accessed(page
);
677 balance_dirty_pages_ratelimited(mapping
);
678 } else if (ret
== AOP_TRUNCATED_PAGE
) {
679 page_cache_release(page
);
683 page_cache_release(page
);
690 * Pipe input worker. Most of this logic works like a regular pipe, the
691 * key here is the 'actor' worker passed in that actually moves the data
692 * to the wanted destination. See pipe_to_file/pipe_to_sendpage above.
694 ssize_t
splice_from_pipe(struct pipe_inode_info
*pipe
, struct file
*out
,
695 loff_t
*ppos
, size_t len
, unsigned int flags
,
698 int ret
, do_wakeup
, err
;
699 struct splice_desc sd
;
710 mutex_lock(&pipe
->inode
->i_mutex
);
714 struct pipe_buffer
*buf
= pipe
->bufs
+ pipe
->curbuf
;
715 struct pipe_buf_operations
*ops
= buf
->ops
;
718 if (sd
.len
> sd
.total_len
)
719 sd
.len
= sd
.total_len
;
721 err
= actor(pipe
, buf
, &sd
);
723 if (!ret
&& err
!= -ENODATA
)
741 ops
->release(pipe
, buf
);
742 pipe
->curbuf
= (pipe
->curbuf
+ 1) & (PIPE_BUFFERS
- 1);
756 if (!pipe
->waiting_writers
) {
761 if (flags
& SPLICE_F_NONBLOCK
) {
767 if (signal_pending(current
)) {
775 if (waitqueue_active(&pipe
->wait
))
776 wake_up_interruptible_sync(&pipe
->wait
);
777 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
785 mutex_unlock(&pipe
->inode
->i_mutex
);
789 if (waitqueue_active(&pipe
->wait
))
790 wake_up_interruptible(&pipe
->wait
);
791 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
798 * generic_file_splice_write - splice data from a pipe to a file
800 * @out: file to write to
801 * @len: number of bytes to splice
802 * @flags: splice modifier flags
804 * Will either move or copy pages (determined by @flags options) from
805 * the given pipe inode to the given file.
809 generic_file_splice_write(struct pipe_inode_info
*pipe
, struct file
*out
,
810 loff_t
*ppos
, size_t len
, unsigned int flags
)
812 struct address_space
*mapping
= out
->f_mapping
;
815 ret
= splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_file
);
817 struct inode
*inode
= mapping
->host
;
822 * If file or inode is SYNC and we actually wrote some data,
825 if (unlikely((out
->f_flags
& O_SYNC
) || IS_SYNC(inode
))) {
828 mutex_lock(&inode
->i_mutex
);
829 err
= generic_osync_inode(inode
, mapping
,
830 OSYNC_METADATA
|OSYNC_DATA
);
831 mutex_unlock(&inode
->i_mutex
);
841 EXPORT_SYMBOL(generic_file_splice_write
);
844 * generic_splice_sendpage - splice data from a pipe to a socket
846 * @out: socket to write to
847 * @len: number of bytes to splice
848 * @flags: splice modifier flags
850 * Will send @len bytes from the pipe to a network socket. No data copying
854 ssize_t
generic_splice_sendpage(struct pipe_inode_info
*pipe
, struct file
*out
,
855 loff_t
*ppos
, size_t len
, unsigned int flags
)
857 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_sendpage
);
860 EXPORT_SYMBOL(generic_splice_sendpage
);
863 * Attempt to initiate a splice from pipe to file.
865 static long do_splice_from(struct pipe_inode_info
*pipe
, struct file
*out
,
866 loff_t
*ppos
, size_t len
, unsigned int flags
)
870 if (unlikely(!out
->f_op
|| !out
->f_op
->splice_write
))
873 if (unlikely(!(out
->f_mode
& FMODE_WRITE
)))
876 ret
= rw_verify_area(WRITE
, out
, ppos
, len
);
877 if (unlikely(ret
< 0))
880 return out
->f_op
->splice_write(pipe
, out
, ppos
, len
, flags
);
884 * Attempt to initiate a splice from a file to a pipe.
886 static long do_splice_to(struct file
*in
, loff_t
*ppos
,
887 struct pipe_inode_info
*pipe
, size_t len
,
893 if (unlikely(!in
->f_op
|| !in
->f_op
->splice_read
))
896 if (unlikely(!(in
->f_mode
& FMODE_READ
)))
899 ret
= rw_verify_area(READ
, in
, ppos
, len
);
900 if (unlikely(ret
< 0))
903 isize
= i_size_read(in
->f_mapping
->host
);
904 if (unlikely(*ppos
>= isize
))
907 left
= isize
- *ppos
;
908 if (unlikely(left
< len
))
911 return in
->f_op
->splice_read(in
, ppos
, pipe
, len
, flags
);
914 long do_splice_direct(struct file
*in
, loff_t
*ppos
, struct file
*out
,
915 size_t len
, unsigned int flags
)
917 struct pipe_inode_info
*pipe
;
924 * We require the input being a regular file, as we don't want to
925 * randomly drop data for eg socket -> socket splicing. Use the
926 * piped splicing for that!
928 i_mode
= in
->f_dentry
->d_inode
->i_mode
;
929 if (unlikely(!S_ISREG(i_mode
) && !S_ISBLK(i_mode
)))
933 * neither in nor out is a pipe, setup an internal pipe attached to
934 * 'out' and transfer the wanted data from 'in' to 'out' through that
936 pipe
= current
->splice_pipe
;
937 if (unlikely(!pipe
)) {
938 pipe
= alloc_pipe_info(NULL
);
943 * We don't have an immediate reader, but we'll read the stuff
944 * out of the pipe right after the splice_to_pipe(). So set
945 * PIPE_READERS appropriately.
949 current
->splice_pipe
= pipe
;
960 size_t read_len
, max_read_len
;
963 * Do at most PIPE_BUFFERS pages worth of transfer:
965 max_read_len
= min(len
, (size_t)(PIPE_BUFFERS
*PAGE_SIZE
));
967 ret
= do_splice_to(in
, ppos
, pipe
, max_read_len
, flags
);
968 if (unlikely(ret
< 0))
974 * NOTE: nonblocking mode only applies to the input. We
975 * must not do the output in nonblocking mode as then we
976 * could get stuck data in the internal pipe:
978 ret
= do_splice_from(pipe
, out
, &out_off
, read_len
,
979 flags
& ~SPLICE_F_NONBLOCK
);
980 if (unlikely(ret
< 0))
987 * In nonblocking mode, if we got back a short read then
988 * that was due to either an IO error or due to the
989 * pagecache entry not being there. In the IO error case
990 * the _next_ splice attempt will produce a clean IO error
991 * return value (not a short read), so in both cases it's
992 * correct to break out of the loop here:
994 if ((flags
& SPLICE_F_NONBLOCK
) && (read_len
< max_read_len
))
998 pipe
->nrbufs
= pipe
->curbuf
= 0;
1004 * If we did an incomplete transfer we must release
1005 * the pipe buffers in question:
1007 for (i
= 0; i
< PIPE_BUFFERS
; i
++) {
1008 struct pipe_buffer
*buf
= pipe
->bufs
+ i
;
1011 buf
->ops
->release(pipe
, buf
);
1015 pipe
->nrbufs
= pipe
->curbuf
= 0;
1018 * If we transferred some data, return the number of bytes:
1026 EXPORT_SYMBOL(do_splice_direct
);
1029 * Determine where to splice to/from.
1031 static long do_splice(struct file
*in
, loff_t __user
*off_in
,
1032 struct file
*out
, loff_t __user
*off_out
,
1033 size_t len
, unsigned int flags
)
1035 struct pipe_inode_info
*pipe
;
1036 loff_t offset
, *off
;
1039 pipe
= in
->f_dentry
->d_inode
->i_pipe
;
1044 if (out
->f_op
->llseek
== no_llseek
)
1046 if (copy_from_user(&offset
, off_out
, sizeof(loff_t
)))
1052 ret
= do_splice_from(pipe
, out
, off
, len
, flags
);
1054 if (off_out
&& copy_to_user(off_out
, off
, sizeof(loff_t
)))
1060 pipe
= out
->f_dentry
->d_inode
->i_pipe
;
1065 if (in
->f_op
->llseek
== no_llseek
)
1067 if (copy_from_user(&offset
, off_in
, sizeof(loff_t
)))
1073 ret
= do_splice_to(in
, off
, pipe
, len
, flags
);
1075 if (off_in
&& copy_to_user(off_in
, off
, sizeof(loff_t
)))
1085 * Map an iov into an array of pages and offset/length tupples. With the
1086 * partial_page structure, we can map several non-contiguous ranges into
1087 * our ones pages[] map instead of splitting that operation into pieces.
1088 * Could easily be exported as a generic helper for other users, in which
1089 * case one would probably want to add a 'max_nr_pages' parameter as well.
1091 static int get_iovec_page_array(const struct iovec __user
*iov
,
1092 unsigned int nr_vecs
, struct page
**pages
,
1093 struct partial_page
*partial
, int aligned
)
1095 int buffers
= 0, error
= 0;
1098 * It's ok to take the mmap_sem for reading, even
1099 * across a "get_user()".
1101 down_read(¤t
->mm
->mmap_sem
);
1104 unsigned long off
, npages
;
1110 * Get user address base and length for this iovec.
1112 error
= get_user(base
, &iov
->iov_base
);
1113 if (unlikely(error
))
1115 error
= get_user(len
, &iov
->iov_len
);
1116 if (unlikely(error
))
1120 * Sanity check this iovec. 0 read succeeds.
1125 if (unlikely(!base
))
1129 * Get this base offset and number of pages, then map
1130 * in the user pages.
1132 off
= (unsigned long) base
& ~PAGE_MASK
;
1135 * If asked for alignment, the offset must be zero and the
1136 * length a multiple of the PAGE_SIZE.
1139 if (aligned
&& (off
|| len
& ~PAGE_MASK
))
1142 npages
= (off
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1143 if (npages
> PIPE_BUFFERS
- buffers
)
1144 npages
= PIPE_BUFFERS
- buffers
;
1146 error
= get_user_pages(current
, current
->mm
,
1147 (unsigned long) base
, npages
, 0, 0,
1148 &pages
[buffers
], NULL
);
1150 if (unlikely(error
<= 0))
1154 * Fill this contiguous range into the partial page map.
1156 for (i
= 0; i
< error
; i
++) {
1157 const int plen
= min_t(size_t, len
, PAGE_SIZE
- off
);
1159 partial
[buffers
].offset
= off
;
1160 partial
[buffers
].len
= plen
;
1168 * We didn't complete this iov, stop here since it probably
1169 * means we have to move some of this into a pipe to
1170 * be able to continue.
1176 * Don't continue if we mapped fewer pages than we asked for,
1177 * or if we mapped the max number of pages that we have
1180 if (error
< npages
|| buffers
== PIPE_BUFFERS
)
1187 up_read(¤t
->mm
->mmap_sem
);
1196 * vmsplice splices a user address range into a pipe. It can be thought of
1197 * as splice-from-memory, where the regular splice is splice-from-file (or
1198 * to file). In both cases the output is a pipe, naturally.
1200 * Note that vmsplice only supports splicing _from_ user memory to a pipe,
1201 * not the other way around. Splicing from user memory is a simple operation
1202 * that can be supported without any funky alignment restrictions or nasty
1203 * vm tricks. We simply map in the user memory and fill them into a pipe.
1204 * The reverse isn't quite as easy, though. There are two possible solutions
1207 * - memcpy() the data internally, at which point we might as well just
1208 * do a regular read() on the buffer anyway.
1209 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1210 * has restriction limitations on both ends of the pipe).
1212 * Alas, it isn't here.
1215 static long do_vmsplice(struct file
*file
, const struct iovec __user
*iov
,
1216 unsigned long nr_segs
, unsigned int flags
)
1218 struct pipe_inode_info
*pipe
= file
->f_dentry
->d_inode
->i_pipe
;
1219 struct page
*pages
[PIPE_BUFFERS
];
1220 struct partial_page partial
[PIPE_BUFFERS
];
1221 struct splice_pipe_desc spd
= {
1225 .ops
= &user_page_pipe_buf_ops
,
1228 if (unlikely(!pipe
))
1230 if (unlikely(nr_segs
> UIO_MAXIOV
))
1232 else if (unlikely(!nr_segs
))
1235 spd
.nr_pages
= get_iovec_page_array(iov
, nr_segs
, pages
, partial
,
1236 flags
& SPLICE_F_GIFT
);
1237 if (spd
.nr_pages
<= 0)
1238 return spd
.nr_pages
;
1240 return splice_to_pipe(pipe
, &spd
);
1243 asmlinkage
long sys_vmsplice(int fd
, const struct iovec __user
*iov
,
1244 unsigned long nr_segs
, unsigned int flags
)
1251 file
= fget_light(fd
, &fput
);
1253 if (file
->f_mode
& FMODE_WRITE
)
1254 error
= do_vmsplice(file
, iov
, nr_segs
, flags
);
1256 fput_light(file
, fput
);
1262 asmlinkage
long sys_splice(int fd_in
, loff_t __user
*off_in
,
1263 int fd_out
, loff_t __user
*off_out
,
1264 size_t len
, unsigned int flags
)
1267 struct file
*in
, *out
;
1268 int fput_in
, fput_out
;
1274 in
= fget_light(fd_in
, &fput_in
);
1276 if (in
->f_mode
& FMODE_READ
) {
1277 out
= fget_light(fd_out
, &fput_out
);
1279 if (out
->f_mode
& FMODE_WRITE
)
1280 error
= do_splice(in
, off_in
,
1283 fput_light(out
, fput_out
);
1287 fput_light(in
, fput_in
);
1294 * Link contents of ipipe to opipe.
1296 static int link_pipe(struct pipe_inode_info
*ipipe
,
1297 struct pipe_inode_info
*opipe
,
1298 size_t len
, unsigned int flags
)
1300 struct pipe_buffer
*ibuf
, *obuf
;
1301 int ret
, do_wakeup
, i
, ipipe_first
;
1303 ret
= do_wakeup
= ipipe_first
= 0;
1306 * Potential ABBA deadlock, work around it by ordering lock
1307 * grabbing by inode address. Otherwise two different processes
1308 * could deadlock (one doing tee from A -> B, the other from B -> A).
1310 if (ipipe
->inode
< opipe
->inode
) {
1312 mutex_lock(&ipipe
->inode
->i_mutex
);
1313 mutex_lock(&opipe
->inode
->i_mutex
);
1315 mutex_lock(&opipe
->inode
->i_mutex
);
1316 mutex_lock(&ipipe
->inode
->i_mutex
);
1320 if (!opipe
->readers
) {
1321 send_sig(SIGPIPE
, current
, 0);
1326 if (ipipe
->nrbufs
- i
) {
1327 ibuf
= ipipe
->bufs
+ ((ipipe
->curbuf
+ i
) & (PIPE_BUFFERS
- 1));
1330 * If we have room, fill this buffer
1332 if (opipe
->nrbufs
< PIPE_BUFFERS
) {
1333 int nbuf
= (opipe
->curbuf
+ opipe
->nrbufs
) & (PIPE_BUFFERS
- 1);
1336 * Get a reference to this pipe buffer,
1337 * so we can copy the contents over.
1339 ibuf
->ops
->get(ipipe
, ibuf
);
1341 obuf
= opipe
->bufs
+ nbuf
;
1345 * Don't inherit the gift flag, we need to
1346 * prevent multiple steals of this page.
1348 obuf
->flags
&= ~PIPE_BUF_FLAG_GIFT
;
1350 if (obuf
->len
> len
)
1360 if (opipe
->nrbufs
< PIPE_BUFFERS
)
1365 * We have input available, but no output room.
1366 * If we already copied data, return that. If we
1367 * need to drop the opipe lock, it must be ordered
1368 * last to avoid deadlocks.
1370 if ((flags
& SPLICE_F_NONBLOCK
) || !ipipe_first
) {
1375 if (signal_pending(current
)) {
1382 if (waitqueue_active(&opipe
->wait
))
1383 wake_up_interruptible(&opipe
->wait
);
1384 kill_fasync(&opipe
->fasync_readers
, SIGIO
, POLL_IN
);
1388 opipe
->waiting_writers
++;
1390 opipe
->waiting_writers
--;
1395 * No input buffers, do the usual checks for available
1396 * writers and blocking and wait if necessary
1398 if (!ipipe
->writers
)
1400 if (!ipipe
->waiting_writers
) {
1405 * pipe_wait() drops the ipipe mutex. To avoid deadlocks
1406 * with another process, we can only safely do that if
1407 * the ipipe lock is ordered last.
1409 if ((flags
& SPLICE_F_NONBLOCK
) || ipipe_first
) {
1414 if (signal_pending(current
)) {
1420 if (waitqueue_active(&ipipe
->wait
))
1421 wake_up_interruptible_sync(&ipipe
->wait
);
1422 kill_fasync(&ipipe
->fasync_writers
, SIGIO
, POLL_OUT
);
1427 mutex_unlock(&ipipe
->inode
->i_mutex
);
1428 mutex_unlock(&opipe
->inode
->i_mutex
);
1432 if (waitqueue_active(&opipe
->wait
))
1433 wake_up_interruptible(&opipe
->wait
);
1434 kill_fasync(&opipe
->fasync_readers
, SIGIO
, POLL_IN
);
1441 * This is a tee(1) implementation that works on pipes. It doesn't copy
1442 * any data, it simply references the 'in' pages on the 'out' pipe.
1443 * The 'flags' used are the SPLICE_F_* variants, currently the only
1444 * applicable one is SPLICE_F_NONBLOCK.
1446 static long do_tee(struct file
*in
, struct file
*out
, size_t len
,
1449 struct pipe_inode_info
*ipipe
= in
->f_dentry
->d_inode
->i_pipe
;
1450 struct pipe_inode_info
*opipe
= out
->f_dentry
->d_inode
->i_pipe
;
1453 * Link ipipe to the two output pipes, consuming as we go along.
1456 return link_pipe(ipipe
, opipe
, len
, flags
);
1461 asmlinkage
long sys_tee(int fdin
, int fdout
, size_t len
, unsigned int flags
)
1470 in
= fget_light(fdin
, &fput_in
);
1472 if (in
->f_mode
& FMODE_READ
) {
1474 struct file
*out
= fget_light(fdout
, &fput_out
);
1477 if (out
->f_mode
& FMODE_WRITE
)
1478 error
= do_tee(in
, out
, len
, flags
);
1479 fput_light(out
, fput_out
);
1482 fput_light(in
, fput_in
);