1 // SPDX-License-Identifier: GPL-2.0-only
3 * "splice": joining two ropes together by interweaving their strands.
5 * This is the "extended pipe" functionality, where a pipe is used as
6 * an arbitrary in-memory buffer. Think of a pipe as a small kernel
7 * buffer that you can use to transfer data from one end to the other.
9 * The traditional unix read/write is extended with a "splice()" operation
10 * that transfers data buffers to or from a pipe buffer.
12 * Named by Larry McVoy, original implementation from Linus, extended by
13 * Jens to support splicing to files, network, direct splicing, etc and
14 * fixing lots of bugs.
16 * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk>
17 * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
18 * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
21 #include <linux/bvec.h>
23 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/splice.h>
26 #include <linux/memcontrol.h>
27 #include <linux/mm_inline.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/export.h>
31 #include <linux/syscalls.h>
32 #include <linux/uio.h>
33 #include <linux/security.h>
34 #include <linux/gfp.h>
35 #include <linux/socket.h>
36 #include <linux/sched/signal.h>
41 * Attempt to steal a page from a pipe buffer. This should perhaps go into
42 * a vm helper function, it's already simplified quite a bit by the
43 * addition of remove_mapping(). If success is returned, the caller may
44 * attempt to reuse this page for another destination.
46 static bool page_cache_pipe_buf_try_steal(struct pipe_inode_info *pipe,
47 struct pipe_buffer *buf)
49 struct page *page = buf->page;
50 struct address_space *mapping;
54 mapping = page_mapping(page);
56 WARN_ON(!PageUptodate(page));
59 * At least for ext2 with nobh option, we need to wait on
60 * writeback completing on this page, since we'll remove it
61 * from the pagecache. Otherwise truncate wont wait on the
62 * page, allowing the disk blocks to be reused by someone else
63 * before we actually wrote our data to them. fs corruption
66 wait_on_page_writeback(page);
68 if (page_has_private(page) &&
69 !try_to_release_page(page, GFP_KERNEL))
73 * If we succeeded in removing the mapping, set LRU flag
76 if (remove_mapping(mapping, page)) {
77 buf->flags |= PIPE_BUF_FLAG_LRU;
83 * Raced with truncate or failed to remove page from current
84 * address space, unlock and return failure.
91 static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
92 struct pipe_buffer *buf)
95 buf->flags &= ~PIPE_BUF_FLAG_LRU;
99 * Check whether the contents of buf is OK to access. Since the content
100 * is a page cache page, IO may be in flight.
102 static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
103 struct pipe_buffer *buf)
105 struct page *page = buf->page;
108 if (!PageUptodate(page)) {
112 * Page got truncated/unhashed. This will cause a 0-byte
113 * splice, if this is the first page.
115 if (!page->mapping) {
121 * Uh oh, read-error from disk.
123 if (!PageUptodate(page)) {
129 * Page is ok afterall, we are done.
140 const struct pipe_buf_operations page_cache_pipe_buf_ops = {
141 .confirm = page_cache_pipe_buf_confirm,
142 .release = page_cache_pipe_buf_release,
143 .try_steal = page_cache_pipe_buf_try_steal,
144 .get = generic_pipe_buf_get,
147 static bool user_page_pipe_buf_try_steal(struct pipe_inode_info *pipe,
148 struct pipe_buffer *buf)
150 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
153 buf->flags |= PIPE_BUF_FLAG_LRU;
154 return generic_pipe_buf_try_steal(pipe, buf);
157 static const struct pipe_buf_operations user_page_pipe_buf_ops = {
158 .release = page_cache_pipe_buf_release,
159 .try_steal = user_page_pipe_buf_try_steal,
160 .get = generic_pipe_buf_get,
163 static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
166 if (waitqueue_active(&pipe->rd_wait))
167 wake_up_interruptible(&pipe->rd_wait);
168 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
172 * splice_to_pipe - fill passed data into a pipe
173 * @pipe: pipe to fill
177 * @spd contains a map of pages and len/offset tuples, along with
178 * the struct pipe_buf_operations associated with these pages. This
179 * function will link that data to the pipe.
182 ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
183 struct splice_pipe_desc *spd)
185 unsigned int spd_pages = spd->nr_pages;
186 unsigned int tail = pipe->tail;
187 unsigned int head = pipe->head;
188 unsigned int mask = pipe->ring_size - 1;
189 int ret = 0, page_nr = 0;
194 if (unlikely(!pipe->readers)) {
195 send_sig(SIGPIPE, current, 0);
200 while (!pipe_full(head, tail, pipe->max_usage)) {
201 struct pipe_buffer *buf = &pipe->bufs[head & mask];
203 buf->page = spd->pages[page_nr];
204 buf->offset = spd->partial[page_nr].offset;
205 buf->len = spd->partial[page_nr].len;
206 buf->private = spd->partial[page_nr].private;
215 if (!--spd->nr_pages)
223 while (page_nr < spd_pages)
224 spd->spd_release(spd, page_nr++);
228 EXPORT_SYMBOL_GPL(splice_to_pipe);
230 ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
232 unsigned int head = pipe->head;
233 unsigned int tail = pipe->tail;
234 unsigned int mask = pipe->ring_size - 1;
237 if (unlikely(!pipe->readers)) {
238 send_sig(SIGPIPE, current, 0);
240 } else if (pipe_full(head, tail, pipe->max_usage)) {
243 pipe->bufs[head & mask] = *buf;
244 pipe->head = head + 1;
247 pipe_buf_release(pipe, buf);
250 EXPORT_SYMBOL(add_to_pipe);
253 * Check if we need to grow the arrays holding pages and partial page
256 int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
258 unsigned int max_usage = READ_ONCE(pipe->max_usage);
260 spd->nr_pages_max = max_usage;
261 if (max_usage <= PIPE_DEF_BUFFERS)
264 spd->pages = kmalloc_array(max_usage, sizeof(struct page *), GFP_KERNEL);
265 spd->partial = kmalloc_array(max_usage, sizeof(struct partial_page),
268 if (spd->pages && spd->partial)
276 void splice_shrink_spd(struct splice_pipe_desc *spd)
278 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
286 * generic_file_splice_read - splice data from file to a pipe
287 * @in: file to splice from
288 * @ppos: position in @in
289 * @pipe: pipe to splice to
290 * @len: number of bytes to splice
291 * @flags: splice modifier flags
294 * Will read pages from given file and fill them into a pipe. Can be
295 * used as long as it has more or less sane ->read_iter().
298 ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
299 struct pipe_inode_info *pipe, size_t len,
307 iov_iter_pipe(&to, READ, pipe, len);
309 init_sync_kiocb(&kiocb, in);
310 kiocb.ki_pos = *ppos;
311 ret = call_read_iter(in, &kiocb, &to);
313 *ppos = kiocb.ki_pos;
315 } else if (ret < 0) {
318 iov_iter_advance(&to, 0); /* to free what was emitted */
320 * callers of ->splice_read() expect -EAGAIN on
321 * "can't put anything in there", rather than -EFAULT.
329 EXPORT_SYMBOL(generic_file_splice_read);
331 const struct pipe_buf_operations default_pipe_buf_ops = {
332 .release = generic_pipe_buf_release,
333 .try_steal = generic_pipe_buf_try_steal,
334 .get = generic_pipe_buf_get,
337 /* Pipe buffer operations for a socket and similar. */
338 const struct pipe_buf_operations nosteal_pipe_buf_ops = {
339 .release = generic_pipe_buf_release,
340 .get = generic_pipe_buf_get,
342 EXPORT_SYMBOL(nosteal_pipe_buf_ops);
345 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
346 * using sendpage(). Return the number of bytes sent.
348 static int pipe_to_sendpage(struct pipe_inode_info *pipe,
349 struct pipe_buffer *buf, struct splice_desc *sd)
351 struct file *file = sd->u.file;
352 loff_t pos = sd->pos;
355 if (!likely(file->f_op->sendpage))
358 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
360 if (sd->len < sd->total_len &&
361 pipe_occupancy(pipe->head, pipe->tail) > 1)
362 more |= MSG_SENDPAGE_NOTLAST;
364 return file->f_op->sendpage(file, buf->page, buf->offset,
365 sd->len, &pos, more);
368 static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
371 if (waitqueue_active(&pipe->wr_wait))
372 wake_up_interruptible(&pipe->wr_wait);
373 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
377 * splice_from_pipe_feed - feed available data from a pipe to a file
378 * @pipe: pipe to splice from
379 * @sd: information to @actor
380 * @actor: handler that splices the data
383 * This function loops over the pipe and calls @actor to do the
384 * actual moving of a single struct pipe_buffer to the desired
385 * destination. It returns when there's no more buffers left in
386 * the pipe or if the requested number of bytes (@sd->total_len)
387 * have been copied. It returns a positive number (one) if the
388 * pipe needs to be filled with more data, zero if the required
389 * number of bytes have been copied and -errno on error.
391 * This, together with splice_from_pipe_{begin,end,next}, may be
392 * used to implement the functionality of __splice_from_pipe() when
393 * locking is required around copying the pipe buffers to the
396 static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
399 unsigned int head = pipe->head;
400 unsigned int tail = pipe->tail;
401 unsigned int mask = pipe->ring_size - 1;
404 while (!pipe_empty(head, tail)) {
405 struct pipe_buffer *buf = &pipe->bufs[tail & mask];
408 if (sd->len > sd->total_len)
409 sd->len = sd->total_len;
411 ret = pipe_buf_confirm(pipe, buf);
418 ret = actor(pipe, buf, sd);
425 sd->num_spliced += ret;
428 sd->total_len -= ret;
431 pipe_buf_release(pipe, buf);
435 sd->need_wakeup = true;
445 /* We know we have a pipe buffer, but maybe it's empty? */
446 static inline bool eat_empty_buffer(struct pipe_inode_info *pipe)
448 unsigned int tail = pipe->tail;
449 unsigned int mask = pipe->ring_size - 1;
450 struct pipe_buffer *buf = &pipe->bufs[tail & mask];
452 if (unlikely(!buf->len)) {
453 pipe_buf_release(pipe, buf);
462 * splice_from_pipe_next - wait for some data to splice from
463 * @pipe: pipe to splice from
464 * @sd: information about the splice operation
467 * This function will wait for some data and return a positive
468 * value (one) if pipe buffers are available. It will return zero
469 * or -errno if no more data needs to be spliced.
471 static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
474 * Check for signal early to make process killable when there are
475 * always buffers available
477 if (signal_pending(current))
481 while (pipe_empty(pipe->head, pipe->tail)) {
488 if (sd->flags & SPLICE_F_NONBLOCK)
491 if (signal_pending(current))
494 if (sd->need_wakeup) {
495 wakeup_pipe_writers(pipe);
496 sd->need_wakeup = false;
499 pipe_wait_readable(pipe);
502 if (eat_empty_buffer(pipe))
509 * splice_from_pipe_begin - start splicing from pipe
510 * @sd: information about the splice operation
513 * This function should be called before a loop containing
514 * splice_from_pipe_next() and splice_from_pipe_feed() to
515 * initialize the necessary fields of @sd.
517 static void splice_from_pipe_begin(struct splice_desc *sd)
520 sd->need_wakeup = false;
524 * splice_from_pipe_end - finish splicing from pipe
525 * @pipe: pipe to splice from
526 * @sd: information about the splice operation
529 * This function will wake up pipe writers if necessary. It should
530 * be called after a loop containing splice_from_pipe_next() and
531 * splice_from_pipe_feed().
533 static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
536 wakeup_pipe_writers(pipe);
540 * __splice_from_pipe - splice data from a pipe to given actor
541 * @pipe: pipe to splice from
542 * @sd: information to @actor
543 * @actor: handler that splices the data
546 * This function does little more than loop over the pipe and call
547 * @actor to do the actual moving of a single struct pipe_buffer to
548 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
552 ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
557 splice_from_pipe_begin(sd);
560 ret = splice_from_pipe_next(pipe, sd);
562 ret = splice_from_pipe_feed(pipe, sd, actor);
564 splice_from_pipe_end(pipe, sd);
566 return sd->num_spliced ? sd->num_spliced : ret;
568 EXPORT_SYMBOL(__splice_from_pipe);
571 * splice_from_pipe - splice data from a pipe to a file
572 * @pipe: pipe to splice from
573 * @out: file to splice to
574 * @ppos: position in @out
575 * @len: how many bytes to splice
576 * @flags: splice modifier flags
577 * @actor: handler that splices the data
580 * See __splice_from_pipe. This function locks the pipe inode,
581 * otherwise it's identical to __splice_from_pipe().
584 ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
585 loff_t *ppos, size_t len, unsigned int flags,
589 struct splice_desc sd = {
597 ret = __splice_from_pipe(pipe, &sd, actor);
604 * iter_file_splice_write - splice data from a pipe to a file
606 * @out: file to write to
607 * @ppos: position in @out
608 * @len: number of bytes to splice
609 * @flags: splice modifier flags
612 * Will either move or copy pages (determined by @flags options) from
613 * the given pipe inode to the given file.
614 * This one is ->write_iter-based.
618 iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
619 loff_t *ppos, size_t len, unsigned int flags)
621 struct splice_desc sd = {
627 int nbufs = pipe->max_usage;
628 struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
632 if (unlikely(!array))
637 splice_from_pipe_begin(&sd);
638 while (sd.total_len) {
639 struct iov_iter from;
640 unsigned int head, tail, mask;
644 ret = splice_from_pipe_next(pipe, &sd);
648 if (unlikely(nbufs < pipe->max_usage)) {
650 nbufs = pipe->max_usage;
651 array = kcalloc(nbufs, sizeof(struct bio_vec),
661 mask = pipe->ring_size - 1;
663 /* build the vector */
665 for (n = 0; !pipe_empty(head, tail) && left && n < nbufs; tail++, n++) {
666 struct pipe_buffer *buf = &pipe->bufs[tail & mask];
667 size_t this_len = buf->len;
672 ret = pipe_buf_confirm(pipe, buf);
679 array[n].bv_page = buf->page;
680 array[n].bv_len = this_len;
681 array[n].bv_offset = buf->offset;
685 iov_iter_bvec(&from, WRITE, array, n, sd.total_len - left);
686 ret = vfs_iter_write(out, &from, &sd.pos, 0);
690 sd.num_spliced += ret;
694 /* dismiss the fully eaten buffers, adjust the partial one */
697 struct pipe_buffer *buf = &pipe->bufs[tail & mask];
698 if (ret >= buf->len) {
701 pipe_buf_release(pipe, buf);
705 sd.need_wakeup = true;
715 splice_from_pipe_end(pipe, &sd);
720 ret = sd.num_spliced;
725 EXPORT_SYMBOL(iter_file_splice_write);
728 * generic_splice_sendpage - splice data from a pipe to a socket
729 * @pipe: pipe to splice from
730 * @out: socket to write to
731 * @ppos: position in @out
732 * @len: number of bytes to splice
733 * @flags: splice modifier flags
736 * Will send @len bytes from the pipe to a network socket. No data copying
740 ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
741 loff_t *ppos, size_t len, unsigned int flags)
743 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
746 EXPORT_SYMBOL(generic_splice_sendpage);
748 static int warn_unsupported(struct file *file, const char *op)
750 pr_debug_ratelimited(
751 "splice %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
752 op, file, current->pid, current->comm);
757 * Attempt to initiate a splice from pipe to file.
759 static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
760 loff_t *ppos, size_t len, unsigned int flags)
762 if (unlikely(!out->f_op->splice_write))
763 return warn_unsupported(out, "write");
764 return out->f_op->splice_write(pipe, out, ppos, len, flags);
768 * Attempt to initiate a splice from a file to a pipe.
770 static long do_splice_to(struct file *in, loff_t *ppos,
771 struct pipe_inode_info *pipe, size_t len,
774 unsigned int p_space;
777 if (unlikely(!(in->f_mode & FMODE_READ)))
780 /* Don't try to read more the pipe has space for. */
781 p_space = pipe->max_usage - pipe_occupancy(pipe->head, pipe->tail);
782 len = min_t(size_t, len, p_space << PAGE_SHIFT);
784 ret = rw_verify_area(READ, in, ppos, len);
785 if (unlikely(ret < 0))
788 if (unlikely(len > MAX_RW_COUNT))
791 if (unlikely(!in->f_op->splice_read))
792 return warn_unsupported(in, "read");
793 return in->f_op->splice_read(in, ppos, pipe, len, flags);
797 * splice_direct_to_actor - splices data directly between two non-pipes
798 * @in: file to splice from
799 * @sd: actor information on where to splice to
800 * @actor: handles the data splicing
803 * This is a special case helper to splice directly between two
804 * points, without requiring an explicit pipe. Internally an allocated
805 * pipe is cached in the process, and reused during the lifetime of
809 ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
810 splice_direct_actor *actor)
812 struct pipe_inode_info *pipe;
819 * We require the input being a regular file, as we don't want to
820 * randomly drop data for eg socket -> socket splicing. Use the
821 * piped splicing for that!
823 i_mode = file_inode(in)->i_mode;
824 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
828 * neither in nor out is a pipe, setup an internal pipe attached to
829 * 'out' and transfer the wanted data from 'in' to 'out' through that
831 pipe = current->splice_pipe;
832 if (unlikely(!pipe)) {
833 pipe = alloc_pipe_info();
838 * We don't have an immediate reader, but we'll read the stuff
839 * out of the pipe right after the splice_to_pipe(). So set
840 * PIPE_READERS appropriately.
844 current->splice_pipe = pipe;
856 * Don't block on output, we have to drain the direct pipe.
858 sd->flags &= ~SPLICE_F_NONBLOCK;
859 more = sd->flags & SPLICE_F_MORE;
861 WARN_ON_ONCE(!pipe_empty(pipe->head, pipe->tail));
865 loff_t pos = sd->pos, prev_pos = pos;
867 ret = do_splice_to(in, &pos, pipe, len, flags);
868 if (unlikely(ret <= 0))
872 sd->total_len = read_len;
875 * If more data is pending, set SPLICE_F_MORE
876 * If this is the last data and SPLICE_F_MORE was not set
877 * initially, clears it.
880 sd->flags |= SPLICE_F_MORE;
882 sd->flags &= ~SPLICE_F_MORE;
884 * NOTE: nonblocking mode only applies to the input. We
885 * must not do the output in nonblocking mode as then we
886 * could get stuck data in the internal pipe:
888 ret = actor(pipe, sd);
889 if (unlikely(ret <= 0)) {
898 if (ret < read_len) {
899 sd->pos = prev_pos + ret;
905 pipe->tail = pipe->head = 0;
911 * If we did an incomplete transfer we must release
912 * the pipe buffers in question:
914 for (i = 0; i < pipe->ring_size; i++) {
915 struct pipe_buffer *buf = &pipe->bufs[i];
918 pipe_buf_release(pipe, buf);
926 EXPORT_SYMBOL(splice_direct_to_actor);
928 static int direct_splice_actor(struct pipe_inode_info *pipe,
929 struct splice_desc *sd)
931 struct file *file = sd->u.file;
933 return do_splice_from(pipe, file, sd->opos, sd->total_len,
938 * do_splice_direct - splices data directly between two files
939 * @in: file to splice from
940 * @ppos: input file offset
941 * @out: file to splice to
942 * @opos: output file offset
943 * @len: number of bytes to splice
944 * @flags: splice modifier flags
947 * For use by do_sendfile(). splice can easily emulate sendfile, but
948 * doing it in the application would incur an extra system call
949 * (splice in + splice out, as compared to just sendfile()). So this helper
950 * can splice directly through a process-private pipe.
953 long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
954 loff_t *opos, size_t len, unsigned int flags)
956 struct splice_desc sd = {
966 if (unlikely(!(out->f_mode & FMODE_WRITE)))
969 if (unlikely(out->f_flags & O_APPEND))
972 ret = rw_verify_area(WRITE, out, opos, len);
973 if (unlikely(ret < 0))
976 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
982 EXPORT_SYMBOL(do_splice_direct);
984 static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags)
987 if (unlikely(!pipe->readers)) {
988 send_sig(SIGPIPE, current, 0);
991 if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage))
993 if (flags & SPLICE_F_NONBLOCK)
995 if (signal_pending(current))
997 pipe_wait_writable(pipe);
1001 static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1002 struct pipe_inode_info *opipe,
1003 size_t len, unsigned int flags);
1005 long splice_file_to_pipe(struct file *in,
1006 struct pipe_inode_info *opipe,
1008 size_t len, unsigned int flags)
1013 ret = wait_for_space(opipe, flags);
1015 ret = do_splice_to(in, offset, opipe, len, flags);
1018 wakeup_pipe_readers(opipe);
1023 * Determine where to splice to/from.
1025 long do_splice(struct file *in, loff_t *off_in, struct file *out,
1026 loff_t *off_out, size_t len, unsigned int flags)
1028 struct pipe_inode_info *ipipe;
1029 struct pipe_inode_info *opipe;
1033 if (unlikely(!(in->f_mode & FMODE_READ) ||
1034 !(out->f_mode & FMODE_WRITE)))
1037 ipipe = get_pipe_info(in, true);
1038 opipe = get_pipe_info(out, true);
1040 if (ipipe && opipe) {
1041 if (off_in || off_out)
1044 /* Splicing to self would be fun, but... */
1048 if ((in->f_flags | out->f_flags) & O_NONBLOCK)
1049 flags |= SPLICE_F_NONBLOCK;
1051 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1058 if (!(out->f_mode & FMODE_PWRITE))
1062 offset = out->f_pos;
1065 if (unlikely(out->f_flags & O_APPEND))
1068 ret = rw_verify_area(WRITE, out, &offset, len);
1069 if (unlikely(ret < 0))
1072 if (in->f_flags & O_NONBLOCK)
1073 flags |= SPLICE_F_NONBLOCK;
1075 file_start_write(out);
1076 ret = do_splice_from(ipipe, out, &offset, len, flags);
1077 file_end_write(out);
1080 out->f_pos = offset;
1091 if (!(in->f_mode & FMODE_PREAD))
1098 if (out->f_flags & O_NONBLOCK)
1099 flags |= SPLICE_F_NONBLOCK;
1101 ret = splice_file_to_pipe(in, opipe, &offset, len, flags);
1113 static long __do_splice(struct file *in, loff_t __user *off_in,
1114 struct file *out, loff_t __user *off_out,
1115 size_t len, unsigned int flags)
1117 struct pipe_inode_info *ipipe;
1118 struct pipe_inode_info *opipe;
1119 loff_t offset, *__off_in = NULL, *__off_out = NULL;
1122 ipipe = get_pipe_info(in, true);
1123 opipe = get_pipe_info(out, true);
1125 if (ipipe && off_in)
1127 if (opipe && off_out)
1131 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
1133 __off_out = &offset;
1136 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
1141 ret = do_splice(in, __off_in, out, __off_out, len, flags);
1145 if (__off_out && copy_to_user(off_out, __off_out, sizeof(loff_t)))
1147 if (__off_in && copy_to_user(off_in, __off_in, sizeof(loff_t)))
1153 static int iter_to_pipe(struct iov_iter *from,
1154 struct pipe_inode_info *pipe,
1157 struct pipe_buffer buf = {
1158 .ops = &user_page_pipe_buf_ops,
1163 bool failed = false;
1165 while (iov_iter_count(from) && !failed) {
1166 struct page *pages[16];
1171 copied = iov_iter_get_pages(from, pages, ~0UL, 16, &start);
1177 for (n = 0; copied; n++, start = 0) {
1178 int size = min_t(int, copied, PAGE_SIZE - start);
1180 buf.page = pages[n];
1183 ret = add_to_pipe(pipe, &buf);
1184 if (unlikely(ret < 0)) {
1187 iov_iter_advance(from, ret);
1196 return total ? total : ret;
1199 static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1200 struct splice_desc *sd)
1202 int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1203 return n == sd->len ? n : -EFAULT;
1207 * For lack of a better implementation, implement vmsplice() to userspace
1208 * as a simple copy of the pipes pages to the user iov.
1210 static long vmsplice_to_user(struct file *file, struct iov_iter *iter,
1213 struct pipe_inode_info *pipe = get_pipe_info(file, true);
1214 struct splice_desc sd = {
1215 .total_len = iov_iter_count(iter),
1226 ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
1234 * vmsplice splices a user address range into a pipe. It can be thought of
1235 * as splice-from-memory, where the regular splice is splice-from-file (or
1236 * to file). In both cases the output is a pipe, naturally.
1238 static long vmsplice_to_pipe(struct file *file, struct iov_iter *iter,
1241 struct pipe_inode_info *pipe;
1243 unsigned buf_flag = 0;
1245 if (flags & SPLICE_F_GIFT)
1246 buf_flag = PIPE_BUF_FLAG_GIFT;
1248 pipe = get_pipe_info(file, true);
1253 ret = wait_for_space(pipe, flags);
1255 ret = iter_to_pipe(iter, pipe, buf_flag);
1258 wakeup_pipe_readers(pipe);
1262 static int vmsplice_type(struct fd f, int *type)
1266 if (f.file->f_mode & FMODE_WRITE) {
1268 } else if (f.file->f_mode & FMODE_READ) {
1278 * Note that vmsplice only really supports true splicing _from_ user memory
1279 * to a pipe, not the other way around. Splicing from user memory is a simple
1280 * operation that can be supported without any funky alignment restrictions
1281 * or nasty vm tricks. We simply map in the user memory and fill them into
1282 * a pipe. The reverse isn't quite as easy, though. There are two possible
1283 * solutions for that:
1285 * - memcpy() the data internally, at which point we might as well just
1286 * do a regular read() on the buffer anyway.
1287 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1288 * has restriction limitations on both ends of the pipe).
1290 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1293 SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, uiov,
1294 unsigned long, nr_segs, unsigned int, flags)
1296 struct iovec iovstack[UIO_FASTIOV];
1297 struct iovec *iov = iovstack;
1298 struct iov_iter iter;
1303 if (unlikely(flags & ~SPLICE_F_ALL))
1307 error = vmsplice_type(f, &type);
1311 error = import_iovec(type, uiov, nr_segs,
1312 ARRAY_SIZE(iovstack), &iov, &iter);
1316 if (!iov_iter_count(&iter))
1318 else if (iov_iter_rw(&iter) == WRITE)
1319 error = vmsplice_to_pipe(f.file, &iter, flags);
1321 error = vmsplice_to_user(f.file, &iter, flags);
1329 SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1330 int, fd_out, loff_t __user *, off_out,
1331 size_t, len, unsigned int, flags)
1339 if (unlikely(flags & ~SPLICE_F_ALL))
1345 out = fdget(fd_out);
1347 error = __do_splice(in.file, off_in, out.file, off_out,
1357 * Make sure there's data to read. Wait for input if we can, otherwise
1358 * return an appropriate error.
1360 static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1365 * Check the pipe occupancy without the inode lock first. This function
1366 * is speculative anyways, so missing one is ok.
1368 if (!pipe_empty(pipe->head, pipe->tail))
1374 while (pipe_empty(pipe->head, pipe->tail)) {
1375 if (signal_pending(current)) {
1381 if (flags & SPLICE_F_NONBLOCK) {
1385 pipe_wait_readable(pipe);
1393 * Make sure there's writeable room. Wait for room if we can, otherwise
1394 * return an appropriate error.
1396 static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1401 * Check pipe occupancy without the inode lock first. This function
1402 * is speculative anyways, so missing one is ok.
1404 if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage))
1410 while (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
1411 if (!pipe->readers) {
1412 send_sig(SIGPIPE, current, 0);
1416 if (flags & SPLICE_F_NONBLOCK) {
1420 if (signal_pending(current)) {
1424 pipe_wait_writable(pipe);
1432 * Splice contents of ipipe to opipe.
1434 static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1435 struct pipe_inode_info *opipe,
1436 size_t len, unsigned int flags)
1438 struct pipe_buffer *ibuf, *obuf;
1439 unsigned int i_head, o_head;
1440 unsigned int i_tail, o_tail;
1441 unsigned int i_mask, o_mask;
1443 bool input_wakeup = false;
1447 ret = ipipe_prep(ipipe, flags);
1451 ret = opipe_prep(opipe, flags);
1456 * Potential ABBA deadlock, work around it by ordering lock
1457 * grabbing by pipe info address. Otherwise two different processes
1458 * could deadlock (one doing tee from A -> B, the other from B -> A).
1460 pipe_double_lock(ipipe, opipe);
1462 i_tail = ipipe->tail;
1463 i_mask = ipipe->ring_size - 1;
1464 o_head = opipe->head;
1465 o_mask = opipe->ring_size - 1;
1470 if (!opipe->readers) {
1471 send_sig(SIGPIPE, current, 0);
1477 i_head = ipipe->head;
1478 o_tail = opipe->tail;
1480 if (pipe_empty(i_head, i_tail) && !ipipe->writers)
1484 * Cannot make any progress, because either the input
1485 * pipe is empty or the output pipe is full.
1487 if (pipe_empty(i_head, i_tail) ||
1488 pipe_full(o_head, o_tail, opipe->max_usage)) {
1489 /* Already processed some buffers, break */
1493 if (flags & SPLICE_F_NONBLOCK) {
1499 * We raced with another reader/writer and haven't
1500 * managed to process any buffers. A zero return
1501 * value means EOF, so retry instead.
1508 ibuf = &ipipe->bufs[i_tail & i_mask];
1509 obuf = &opipe->bufs[o_head & o_mask];
1511 if (len >= ibuf->len) {
1513 * Simply move the whole buffer from ipipe to opipe
1518 ipipe->tail = i_tail;
1519 input_wakeup = true;
1522 opipe->head = o_head;
1525 * Get a reference to this pipe buffer,
1526 * so we can copy the contents over.
1528 if (!pipe_buf_get(ipipe, ibuf)) {
1536 * Don't inherit the gift and merge flags, we need to
1537 * prevent multiple steals of this page.
1539 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1540 obuf->flags &= ~PIPE_BUF_FLAG_CAN_MERGE;
1543 ibuf->offset += len;
1547 opipe->head = o_head;
1557 * If we put data in the output pipe, wakeup any potential readers.
1560 wakeup_pipe_readers(opipe);
1563 wakeup_pipe_writers(ipipe);
1569 * Link contents of ipipe to opipe.
1571 static int link_pipe(struct pipe_inode_info *ipipe,
1572 struct pipe_inode_info *opipe,
1573 size_t len, unsigned int flags)
1575 struct pipe_buffer *ibuf, *obuf;
1576 unsigned int i_head, o_head;
1577 unsigned int i_tail, o_tail;
1578 unsigned int i_mask, o_mask;
1582 * Potential ABBA deadlock, work around it by ordering lock
1583 * grabbing by pipe info address. Otherwise two different processes
1584 * could deadlock (one doing tee from A -> B, the other from B -> A).
1586 pipe_double_lock(ipipe, opipe);
1588 i_tail = ipipe->tail;
1589 i_mask = ipipe->ring_size - 1;
1590 o_head = opipe->head;
1591 o_mask = opipe->ring_size - 1;
1594 if (!opipe->readers) {
1595 send_sig(SIGPIPE, current, 0);
1601 i_head = ipipe->head;
1602 o_tail = opipe->tail;
1605 * If we have iterated all input buffers or run out of
1606 * output room, break.
1608 if (pipe_empty(i_head, i_tail) ||
1609 pipe_full(o_head, o_tail, opipe->max_usage))
1612 ibuf = &ipipe->bufs[i_tail & i_mask];
1613 obuf = &opipe->bufs[o_head & o_mask];
1616 * Get a reference to this pipe buffer,
1617 * so we can copy the contents over.
1619 if (!pipe_buf_get(ipipe, ibuf)) {
1628 * Don't inherit the gift and merge flag, we need to prevent
1629 * multiple steals of this page.
1631 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1632 obuf->flags &= ~PIPE_BUF_FLAG_CAN_MERGE;
1634 if (obuf->len > len)
1640 opipe->head = o_head;
1648 * If we put data in the output pipe, wakeup any potential readers.
1651 wakeup_pipe_readers(opipe);
1657 * This is a tee(1) implementation that works on pipes. It doesn't copy
1658 * any data, it simply references the 'in' pages on the 'out' pipe.
1659 * The 'flags' used are the SPLICE_F_* variants, currently the only
1660 * applicable one is SPLICE_F_NONBLOCK.
1662 long do_tee(struct file *in, struct file *out, size_t len, unsigned int flags)
1664 struct pipe_inode_info *ipipe = get_pipe_info(in, true);
1665 struct pipe_inode_info *opipe = get_pipe_info(out, true);
1668 if (unlikely(!(in->f_mode & FMODE_READ) ||
1669 !(out->f_mode & FMODE_WRITE)))
1673 * Duplicate the contents of ipipe to opipe without actually
1676 if (ipipe && opipe && ipipe != opipe) {
1677 if ((in->f_flags | out->f_flags) & O_NONBLOCK)
1678 flags |= SPLICE_F_NONBLOCK;
1681 * Keep going, unless we encounter an error. The ipipe/opipe
1682 * ordering doesn't really matter.
1684 ret = ipipe_prep(ipipe, flags);
1686 ret = opipe_prep(opipe, flags);
1688 ret = link_pipe(ipipe, opipe, len, flags);
1695 SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
1700 if (unlikely(flags & ~SPLICE_F_ALL))
1711 error = do_tee(in.file, out.file, len, flags);