1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * "splice": joining two ropes together by interweaving their strands.
4  *
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.
8  *
9  * The traditional unix read/write is extended with a "splice()" operation
10  * that transfers data buffers to or from a pipe buffer.
11  *
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.
15  *
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>
19  *
20  */
21 #include <linux/bvec.h>
22 #include <linux/fs.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>
37 
38 #include "internal.h"
39 
40 /*
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.
45  */
page_cache_pipe_buf_try_steal(struct pipe_inode_info * pipe,struct pipe_buffer * buf)46 static bool page_cache_pipe_buf_try_steal(struct pipe_inode_info *pipe,
47 		struct pipe_buffer *buf)
48 {
49 	struct folio *folio = page_folio(buf->page);
50 	struct address_space *mapping;
51 
52 	folio_lock(folio);
53 
54 	mapping = folio_mapping(folio);
55 	if (mapping) {
56 		WARN_ON(!folio_test_uptodate(folio));
57 
58 		/*
59 		 * At least for ext2 with nobh option, we need to wait on
60 		 * writeback completing on this folio, since we'll remove it
61 		 * from the pagecache.  Otherwise truncate wont wait on the
62 		 * folio, allowing the disk blocks to be reused by someone else
63 		 * before we actually wrote our data to them. fs corruption
64 		 * ensues.
65 		 */
66 		folio_wait_writeback(folio);
67 
68 		if (folio_has_private(folio) &&
69 		    !filemap_release_folio(folio, GFP_KERNEL))
70 			goto out_unlock;
71 
72 		/*
73 		 * If we succeeded in removing the mapping, set LRU flag
74 		 * and return good.
75 		 */
76 		if (remove_mapping(mapping, folio)) {
77 			buf->flags |= PIPE_BUF_FLAG_LRU;
78 			return true;
79 		}
80 	}
81 
82 	/*
83 	 * Raced with truncate or failed to remove folio from current
84 	 * address space, unlock and return failure.
85 	 */
86 out_unlock:
87 	folio_unlock(folio);
88 	return false;
89 }
90 
page_cache_pipe_buf_release(struct pipe_inode_info * pipe,struct pipe_buffer * buf)91 static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
92 					struct pipe_buffer *buf)
93 {
94 	put_page(buf->page);
95 	buf->flags &= ~PIPE_BUF_FLAG_LRU;
96 }
97 
98 /*
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.
101  */
page_cache_pipe_buf_confirm(struct pipe_inode_info * pipe,struct pipe_buffer * buf)102 static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
103 				       struct pipe_buffer *buf)
104 {
105 	struct page *page = buf->page;
106 	int err;
107 
108 	if (!PageUptodate(page)) {
109 		lock_page(page);
110 
111 		/*
112 		 * Page got truncated/unhashed. This will cause a 0-byte
113 		 * splice, if this is the first page.
114 		 */
115 		if (!page->mapping) {
116 			err = -ENODATA;
117 			goto error;
118 		}
119 
120 		/*
121 		 * Uh oh, read-error from disk.
122 		 */
123 		if (!PageUptodate(page)) {
124 			err = -EIO;
125 			goto error;
126 		}
127 
128 		/*
129 		 * Page is ok afterall, we are done.
130 		 */
131 		unlock_page(page);
132 	}
133 
134 	return 0;
135 error:
136 	unlock_page(page);
137 	return err;
138 }
139 
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,
145 };
146 
user_page_pipe_buf_try_steal(struct pipe_inode_info * pipe,struct pipe_buffer * buf)147 static bool user_page_pipe_buf_try_steal(struct pipe_inode_info *pipe,
148 		struct pipe_buffer *buf)
149 {
150 	if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
151 		return false;
152 
153 	buf->flags |= PIPE_BUF_FLAG_LRU;
154 	return generic_pipe_buf_try_steal(pipe, buf);
155 }
156 
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,
161 };
162 
wakeup_pipe_readers(struct pipe_inode_info * pipe)163 static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
164 {
165 	smp_mb();
166 	if (waitqueue_active(&pipe->rd_wait))
167 		wake_up_interruptible(&pipe->rd_wait);
168 	kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
169 }
170 
171 /**
172  * splice_to_pipe - fill passed data into a pipe
173  * @pipe:	pipe to fill
174  * @spd:	data to fill
175  *
176  * Description:
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.
180  *
181  */
splice_to_pipe(struct pipe_inode_info * pipe,struct splice_pipe_desc * spd)182 ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
183 		       struct splice_pipe_desc *spd)
184 {
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;
190 
191 	if (!spd_pages)
192 		return 0;
193 
194 	if (unlikely(!pipe->readers)) {
195 		send_sig(SIGPIPE, current, 0);
196 		ret = -EPIPE;
197 		goto out;
198 	}
199 
200 	while (!pipe_full(head, tail, pipe->max_usage)) {
201 		struct pipe_buffer *buf = &pipe->bufs[head & mask];
202 
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;
207 		buf->ops = spd->ops;
208 		buf->flags = 0;
209 
210 		head++;
211 		pipe->head = head;
212 		page_nr++;
213 		ret += buf->len;
214 
215 		if (!--spd->nr_pages)
216 			break;
217 	}
218 
219 	if (!ret)
220 		ret = -EAGAIN;
221 
222 out:
223 	while (page_nr < spd_pages)
224 		spd->spd_release(spd, page_nr++);
225 
226 	return ret;
227 }
228 EXPORT_SYMBOL_GPL(splice_to_pipe);
229 
add_to_pipe(struct pipe_inode_info * pipe,struct pipe_buffer * buf)230 ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
231 {
232 	unsigned int head = pipe->head;
233 	unsigned int tail = pipe->tail;
234 	unsigned int mask = pipe->ring_size - 1;
235 	int ret;
236 
237 	if (unlikely(!pipe->readers)) {
238 		send_sig(SIGPIPE, current, 0);
239 		ret = -EPIPE;
240 	} else if (pipe_full(head, tail, pipe->max_usage)) {
241 		ret = -EAGAIN;
242 	} else {
243 		pipe->bufs[head & mask] = *buf;
244 		pipe->head = head + 1;
245 		return buf->len;
246 	}
247 	pipe_buf_release(pipe, buf);
248 	return ret;
249 }
250 EXPORT_SYMBOL(add_to_pipe);
251 
252 /*
253  * Check if we need to grow the arrays holding pages and partial page
254  * descriptions.
255  */
splice_grow_spd(const struct pipe_inode_info * pipe,struct splice_pipe_desc * spd)256 int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
257 {
258 	unsigned int max_usage = READ_ONCE(pipe->max_usage);
259 
260 	spd->nr_pages_max = max_usage;
261 	if (max_usage <= PIPE_DEF_BUFFERS)
262 		return 0;
263 
264 	spd->pages = kmalloc_array(max_usage, sizeof(struct page *), GFP_KERNEL);
265 	spd->partial = kmalloc_array(max_usage, sizeof(struct partial_page),
266 				     GFP_KERNEL);
267 
268 	if (spd->pages && spd->partial)
269 		return 0;
270 
271 	kfree(spd->pages);
272 	kfree(spd->partial);
273 	return -ENOMEM;
274 }
275 
splice_shrink_spd(struct splice_pipe_desc * spd)276 void splice_shrink_spd(struct splice_pipe_desc *spd)
277 {
278 	if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
279 		return;
280 
281 	kfree(spd->pages);
282 	kfree(spd->partial);
283 }
284 
285 /**
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
292  *
293  * Description:
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().
296  *
297  */
generic_file_splice_read(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)298 ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
299 				 struct pipe_inode_info *pipe, size_t len,
300 				 unsigned int flags)
301 {
302 	struct iov_iter to;
303 	struct kiocb kiocb;
304 	int ret;
305 
306 	iov_iter_pipe(&to, READ, pipe, len);
307 	init_sync_kiocb(&kiocb, in);
308 	kiocb.ki_pos = *ppos;
309 	ret = call_read_iter(in, &kiocb, &to);
310 	if (ret > 0) {
311 		*ppos = kiocb.ki_pos;
312 		file_accessed(in);
313 	} else if (ret < 0) {
314 		/* free what was emitted */
315 		pipe_discard_from(pipe, to.start_head);
316 		/*
317 		 * callers of ->splice_read() expect -EAGAIN on
318 		 * "can't put anything in there", rather than -EFAULT.
319 		 */
320 		if (ret == -EFAULT)
321 			ret = -EAGAIN;
322 	}
323 
324 	return ret;
325 }
326 EXPORT_SYMBOL(generic_file_splice_read);
327 
328 const struct pipe_buf_operations default_pipe_buf_ops = {
329 	.release	= generic_pipe_buf_release,
330 	.try_steal	= generic_pipe_buf_try_steal,
331 	.get		= generic_pipe_buf_get,
332 };
333 
334 /* Pipe buffer operations for a socket and similar. */
335 const struct pipe_buf_operations nosteal_pipe_buf_ops = {
336 	.release	= generic_pipe_buf_release,
337 	.get		= generic_pipe_buf_get,
338 };
339 EXPORT_SYMBOL(nosteal_pipe_buf_ops);
340 
341 /*
342  * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
343  * using sendpage(). Return the number of bytes sent.
344  */
pipe_to_sendpage(struct pipe_inode_info * pipe,struct pipe_buffer * buf,struct splice_desc * sd)345 static int pipe_to_sendpage(struct pipe_inode_info *pipe,
346 			    struct pipe_buffer *buf, struct splice_desc *sd)
347 {
348 	struct file *file = sd->u.file;
349 	loff_t pos = sd->pos;
350 	int more;
351 
352 	if (!likely(file->f_op->sendpage))
353 		return -EINVAL;
354 
355 	more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
356 
357 	if (sd->len < sd->total_len &&
358 	    pipe_occupancy(pipe->head, pipe->tail) > 1)
359 		more |= MSG_SENDPAGE_NOTLAST;
360 
361 	return file->f_op->sendpage(file, buf->page, buf->offset,
362 				    sd->len, &pos, more);
363 }
364 
wakeup_pipe_writers(struct pipe_inode_info * pipe)365 static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
366 {
367 	smp_mb();
368 	if (waitqueue_active(&pipe->wr_wait))
369 		wake_up_interruptible(&pipe->wr_wait);
370 	kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
371 }
372 
373 /**
374  * splice_from_pipe_feed - feed available data from a pipe to a file
375  * @pipe:	pipe to splice from
376  * @sd:		information to @actor
377  * @actor:	handler that splices the data
378  *
379  * Description:
380  *    This function loops over the pipe and calls @actor to do the
381  *    actual moving of a single struct pipe_buffer to the desired
382  *    destination.  It returns when there's no more buffers left in
383  *    the pipe or if the requested number of bytes (@sd->total_len)
384  *    have been copied.  It returns a positive number (one) if the
385  *    pipe needs to be filled with more data, zero if the required
386  *    number of bytes have been copied and -errno on error.
387  *
388  *    This, together with splice_from_pipe_{begin,end,next}, may be
389  *    used to implement the functionality of __splice_from_pipe() when
390  *    locking is required around copying the pipe buffers to the
391  *    destination.
392  */
splice_from_pipe_feed(struct pipe_inode_info * pipe,struct splice_desc * sd,splice_actor * actor)393 static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
394 			  splice_actor *actor)
395 {
396 	unsigned int head = pipe->head;
397 	unsigned int tail = pipe->tail;
398 	unsigned int mask = pipe->ring_size - 1;
399 	int ret;
400 
401 	while (!pipe_empty(head, tail)) {
402 		struct pipe_buffer *buf = &pipe->bufs[tail & mask];
403 
404 		sd->len = buf->len;
405 		if (sd->len > sd->total_len)
406 			sd->len = sd->total_len;
407 
408 		ret = pipe_buf_confirm(pipe, buf);
409 		if (unlikely(ret)) {
410 			if (ret == -ENODATA)
411 				ret = 0;
412 			return ret;
413 		}
414 
415 		ret = actor(pipe, buf, sd);
416 		if (ret <= 0)
417 			return ret;
418 
419 		buf->offset += ret;
420 		buf->len -= ret;
421 
422 		sd->num_spliced += ret;
423 		sd->len -= ret;
424 		sd->pos += ret;
425 		sd->total_len -= ret;
426 
427 		if (!buf->len) {
428 			pipe_buf_release(pipe, buf);
429 			tail++;
430 			pipe->tail = tail;
431 			if (pipe->files)
432 				sd->need_wakeup = true;
433 		}
434 
435 		if (!sd->total_len)
436 			return 0;
437 	}
438 
439 	return 1;
440 }
441 
442 /* We know we have a pipe buffer, but maybe it's empty? */
eat_empty_buffer(struct pipe_inode_info * pipe)443 static inline bool eat_empty_buffer(struct pipe_inode_info *pipe)
444 {
445 	unsigned int tail = pipe->tail;
446 	unsigned int mask = pipe->ring_size - 1;
447 	struct pipe_buffer *buf = &pipe->bufs[tail & mask];
448 
449 	if (unlikely(!buf->len)) {
450 		pipe_buf_release(pipe, buf);
451 		pipe->tail = tail+1;
452 		return true;
453 	}
454 
455 	return false;
456 }
457 
458 /**
459  * splice_from_pipe_next - wait for some data to splice from
460  * @pipe:	pipe to splice from
461  * @sd:		information about the splice operation
462  *
463  * Description:
464  *    This function will wait for some data and return a positive
465  *    value (one) if pipe buffers are available.  It will return zero
466  *    or -errno if no more data needs to be spliced.
467  */
splice_from_pipe_next(struct pipe_inode_info * pipe,struct splice_desc * sd)468 static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
469 {
470 	/*
471 	 * Check for signal early to make process killable when there are
472 	 * always buffers available
473 	 */
474 	if (signal_pending(current))
475 		return -ERESTARTSYS;
476 
477 repeat:
478 	while (pipe_empty(pipe->head, pipe->tail)) {
479 		if (!pipe->writers)
480 			return 0;
481 
482 		if (sd->num_spliced)
483 			return 0;
484 
485 		if (sd->flags & SPLICE_F_NONBLOCK)
486 			return -EAGAIN;
487 
488 		if (signal_pending(current))
489 			return -ERESTARTSYS;
490 
491 		if (sd->need_wakeup) {
492 			wakeup_pipe_writers(pipe);
493 			sd->need_wakeup = false;
494 		}
495 
496 		pipe_wait_readable(pipe);
497 	}
498 
499 	if (eat_empty_buffer(pipe))
500 		goto repeat;
501 
502 	return 1;
503 }
504 
505 /**
506  * splice_from_pipe_begin - start splicing from pipe
507  * @sd:		information about the splice operation
508  *
509  * Description:
510  *    This function should be called before a loop containing
511  *    splice_from_pipe_next() and splice_from_pipe_feed() to
512  *    initialize the necessary fields of @sd.
513  */
splice_from_pipe_begin(struct splice_desc * sd)514 static void splice_from_pipe_begin(struct splice_desc *sd)
515 {
516 	sd->num_spliced = 0;
517 	sd->need_wakeup = false;
518 }
519 
520 /**
521  * splice_from_pipe_end - finish splicing from pipe
522  * @pipe:	pipe to splice from
523  * @sd:		information about the splice operation
524  *
525  * Description:
526  *    This function will wake up pipe writers if necessary.  It should
527  *    be called after a loop containing splice_from_pipe_next() and
528  *    splice_from_pipe_feed().
529  */
splice_from_pipe_end(struct pipe_inode_info * pipe,struct splice_desc * sd)530 static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
531 {
532 	if (sd->need_wakeup)
533 		wakeup_pipe_writers(pipe);
534 }
535 
536 /**
537  * __splice_from_pipe - splice data from a pipe to given actor
538  * @pipe:	pipe to splice from
539  * @sd:		information to @actor
540  * @actor:	handler that splices the data
541  *
542  * Description:
543  *    This function does little more than loop over the pipe and call
544  *    @actor to do the actual moving of a single struct pipe_buffer to
545  *    the desired destination. See pipe_to_file, pipe_to_sendpage, or
546  *    pipe_to_user.
547  *
548  */
__splice_from_pipe(struct pipe_inode_info * pipe,struct splice_desc * sd,splice_actor * actor)549 ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
550 			   splice_actor *actor)
551 {
552 	int ret;
553 
554 	splice_from_pipe_begin(sd);
555 	do {
556 		cond_resched();
557 		ret = splice_from_pipe_next(pipe, sd);
558 		if (ret > 0)
559 			ret = splice_from_pipe_feed(pipe, sd, actor);
560 	} while (ret > 0);
561 	splice_from_pipe_end(pipe, sd);
562 
563 	return sd->num_spliced ? sd->num_spliced : ret;
564 }
565 EXPORT_SYMBOL(__splice_from_pipe);
566 
567 /**
568  * splice_from_pipe - splice data from a pipe to a file
569  * @pipe:	pipe to splice from
570  * @out:	file to splice to
571  * @ppos:	position in @out
572  * @len:	how many bytes to splice
573  * @flags:	splice modifier flags
574  * @actor:	handler that splices the data
575  *
576  * Description:
577  *    See __splice_from_pipe. This function locks the pipe inode,
578  *    otherwise it's identical to __splice_from_pipe().
579  *
580  */
splice_from_pipe(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags,splice_actor * actor)581 ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
582 			 loff_t *ppos, size_t len, unsigned int flags,
583 			 splice_actor *actor)
584 {
585 	ssize_t ret;
586 	struct splice_desc sd = {
587 		.total_len = len,
588 		.flags = flags,
589 		.pos = *ppos,
590 		.u.file = out,
591 	};
592 
593 	pipe_lock(pipe);
594 	ret = __splice_from_pipe(pipe, &sd, actor);
595 	pipe_unlock(pipe);
596 
597 	return ret;
598 }
599 
600 /**
601  * iter_file_splice_write - splice data from a pipe to a file
602  * @pipe:	pipe info
603  * @out:	file to write to
604  * @ppos:	position in @out
605  * @len:	number of bytes to splice
606  * @flags:	splice modifier flags
607  *
608  * Description:
609  *    Will either move or copy pages (determined by @flags options) from
610  *    the given pipe inode to the given file.
611  *    This one is ->write_iter-based.
612  *
613  */
614 ssize_t
iter_file_splice_write(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)615 iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
616 			  loff_t *ppos, size_t len, unsigned int flags)
617 {
618 	struct splice_desc sd = {
619 		.total_len = len,
620 		.flags = flags,
621 		.pos = *ppos,
622 		.u.file = out,
623 	};
624 	int nbufs = pipe->max_usage;
625 	struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
626 					GFP_KERNEL);
627 	ssize_t ret;
628 
629 	if (unlikely(!array))
630 		return -ENOMEM;
631 
632 	pipe_lock(pipe);
633 
634 	splice_from_pipe_begin(&sd);
635 	while (sd.total_len) {
636 		struct iov_iter from;
637 		unsigned int head, tail, mask;
638 		size_t left;
639 		int n;
640 
641 		ret = splice_from_pipe_next(pipe, &sd);
642 		if (ret <= 0)
643 			break;
644 
645 		if (unlikely(nbufs < pipe->max_usage)) {
646 			kfree(array);
647 			nbufs = pipe->max_usage;
648 			array = kcalloc(nbufs, sizeof(struct bio_vec),
649 					GFP_KERNEL);
650 			if (!array) {
651 				ret = -ENOMEM;
652 				break;
653 			}
654 		}
655 
656 		head = pipe->head;
657 		tail = pipe->tail;
658 		mask = pipe->ring_size - 1;
659 
660 		/* build the vector */
661 		left = sd.total_len;
662 		for (n = 0; !pipe_empty(head, tail) && left && n < nbufs; tail++) {
663 			struct pipe_buffer *buf = &pipe->bufs[tail & mask];
664 			size_t this_len = buf->len;
665 
666 			/* zero-length bvecs are not supported, skip them */
667 			if (!this_len)
668 				continue;
669 			this_len = min(this_len, left);
670 
671 			ret = pipe_buf_confirm(pipe, buf);
672 			if (unlikely(ret)) {
673 				if (ret == -ENODATA)
674 					ret = 0;
675 				goto done;
676 			}
677 
678 			array[n].bv_page = buf->page;
679 			array[n].bv_len = this_len;
680 			array[n].bv_offset = buf->offset;
681 			left -= this_len;
682 			n++;
683 		}
684 
685 		iov_iter_bvec(&from, WRITE, array, n, sd.total_len - left);
686 		ret = vfs_iter_write(out, &from, &sd.pos, 0);
687 		if (ret <= 0)
688 			break;
689 
690 		sd.num_spliced += ret;
691 		sd.total_len -= ret;
692 		*ppos = sd.pos;
693 
694 		/* dismiss the fully eaten buffers, adjust the partial one */
695 		tail = pipe->tail;
696 		while (ret) {
697 			struct pipe_buffer *buf = &pipe->bufs[tail & mask];
698 			if (ret >= buf->len) {
699 				ret -= buf->len;
700 				buf->len = 0;
701 				pipe_buf_release(pipe, buf);
702 				tail++;
703 				pipe->tail = tail;
704 				if (pipe->files)
705 					sd.need_wakeup = true;
706 			} else {
707 				buf->offset += ret;
708 				buf->len -= ret;
709 				ret = 0;
710 			}
711 		}
712 	}
713 done:
714 	kfree(array);
715 	splice_from_pipe_end(pipe, &sd);
716 
717 	pipe_unlock(pipe);
718 
719 	if (sd.num_spliced)
720 		ret = sd.num_spliced;
721 
722 	return ret;
723 }
724 
725 EXPORT_SYMBOL(iter_file_splice_write);
726 
727 /**
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
734  *
735  * Description:
736  *    Will send @len bytes from the pipe to a network socket. No data copying
737  *    is involved.
738  *
739  */
generic_splice_sendpage(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)740 ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
741 				loff_t *ppos, size_t len, unsigned int flags)
742 {
743 	return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
744 }
745 
746 EXPORT_SYMBOL(generic_splice_sendpage);
747 
warn_unsupported(struct file * file,const char * op)748 static int warn_unsupported(struct file *file, const char *op)
749 {
750 	pr_debug_ratelimited(
751 		"splice %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
752 		op, file, current->pid, current->comm);
753 	return -EINVAL;
754 }
755 
756 /*
757  * Attempt to initiate a splice from pipe to file.
758  */
do_splice_from(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)759 static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
760 			   loff_t *ppos, size_t len, unsigned int flags)
761 {
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);
765 }
766 
767 /*
768  * Attempt to initiate a splice from a file to a pipe.
769  */
do_splice_to(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)770 static long do_splice_to(struct file *in, loff_t *ppos,
771 			 struct pipe_inode_info *pipe, size_t len,
772 			 unsigned int flags)
773 {
774 	unsigned int p_space;
775 	int ret;
776 
777 	if (unlikely(!(in->f_mode & FMODE_READ)))
778 		return -EBADF;
779 
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);
783 
784 	ret = rw_verify_area(READ, in, ppos, len);
785 	if (unlikely(ret < 0))
786 		return ret;
787 
788 	if (unlikely(len > MAX_RW_COUNT))
789 		len = MAX_RW_COUNT;
790 
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);
794 }
795 
796 /**
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
801  *
802  * Description:
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
806  *    that process.
807  *
808  */
splice_direct_to_actor(struct file * in,struct splice_desc * sd,splice_direct_actor * actor)809 ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
810 			       splice_direct_actor *actor)
811 {
812 	struct pipe_inode_info *pipe;
813 	long ret, bytes;
814 	size_t len;
815 	int i, flags, more;
816 
817 	/*
818 	 * We require the input to be seekable, as we don't want to randomly
819 	 * drop data for eg socket -> socket splicing. Use the piped splicing
820 	 * for that!
821 	 */
822 	if (unlikely(!(in->f_mode & FMODE_LSEEK)))
823 		return -EINVAL;
824 
825 	/*
826 	 * neither in nor out is a pipe, setup an internal pipe attached to
827 	 * 'out' and transfer the wanted data from 'in' to 'out' through that
828 	 */
829 	pipe = current->splice_pipe;
830 	if (unlikely(!pipe)) {
831 		pipe = alloc_pipe_info();
832 		if (!pipe)
833 			return -ENOMEM;
834 
835 		/*
836 		 * We don't have an immediate reader, but we'll read the stuff
837 		 * out of the pipe right after the splice_to_pipe(). So set
838 		 * PIPE_READERS appropriately.
839 		 */
840 		pipe->readers = 1;
841 
842 		current->splice_pipe = pipe;
843 	}
844 
845 	/*
846 	 * Do the splice.
847 	 */
848 	ret = 0;
849 	bytes = 0;
850 	len = sd->total_len;
851 	flags = sd->flags;
852 
853 	/*
854 	 * Don't block on output, we have to drain the direct pipe.
855 	 */
856 	sd->flags &= ~SPLICE_F_NONBLOCK;
857 	more = sd->flags & SPLICE_F_MORE;
858 
859 	WARN_ON_ONCE(!pipe_empty(pipe->head, pipe->tail));
860 
861 	while (len) {
862 		size_t read_len;
863 		loff_t pos = sd->pos, prev_pos = pos;
864 
865 		ret = do_splice_to(in, &pos, pipe, len, flags);
866 		if (unlikely(ret <= 0))
867 			goto out_release;
868 
869 		read_len = ret;
870 		sd->total_len = read_len;
871 
872 		/*
873 		 * If more data is pending, set SPLICE_F_MORE
874 		 * If this is the last data and SPLICE_F_MORE was not set
875 		 * initially, clears it.
876 		 */
877 		if (read_len < len)
878 			sd->flags |= SPLICE_F_MORE;
879 		else if (!more)
880 			sd->flags &= ~SPLICE_F_MORE;
881 		/*
882 		 * NOTE: nonblocking mode only applies to the input. We
883 		 * must not do the output in nonblocking mode as then we
884 		 * could get stuck data in the internal pipe:
885 		 */
886 		ret = actor(pipe, sd);
887 		if (unlikely(ret <= 0)) {
888 			sd->pos = prev_pos;
889 			goto out_release;
890 		}
891 
892 		bytes += ret;
893 		len -= ret;
894 		sd->pos = pos;
895 
896 		if (ret < read_len) {
897 			sd->pos = prev_pos + ret;
898 			goto out_release;
899 		}
900 	}
901 
902 done:
903 	pipe->tail = pipe->head = 0;
904 	file_accessed(in);
905 	return bytes;
906 
907 out_release:
908 	/*
909 	 * If we did an incomplete transfer we must release
910 	 * the pipe buffers in question:
911 	 */
912 	for (i = 0; i < pipe->ring_size; i++) {
913 		struct pipe_buffer *buf = &pipe->bufs[i];
914 
915 		if (buf->ops)
916 			pipe_buf_release(pipe, buf);
917 	}
918 
919 	if (!bytes)
920 		bytes = ret;
921 
922 	goto done;
923 }
924 EXPORT_SYMBOL(splice_direct_to_actor);
925 
direct_splice_actor(struct pipe_inode_info * pipe,struct splice_desc * sd)926 static int direct_splice_actor(struct pipe_inode_info *pipe,
927 			       struct splice_desc *sd)
928 {
929 	struct file *file = sd->u.file;
930 
931 	return do_splice_from(pipe, file, sd->opos, sd->total_len,
932 			      sd->flags);
933 }
934 
935 /**
936  * do_splice_direct - splices data directly between two files
937  * @in:		file to splice from
938  * @ppos:	input file offset
939  * @out:	file to splice to
940  * @opos:	output file offset
941  * @len:	number of bytes to splice
942  * @flags:	splice modifier flags
943  *
944  * Description:
945  *    For use by do_sendfile(). splice can easily emulate sendfile, but
946  *    doing it in the application would incur an extra system call
947  *    (splice in + splice out, as compared to just sendfile()). So this helper
948  *    can splice directly through a process-private pipe.
949  *
950  */
do_splice_direct(struct file * in,loff_t * ppos,struct file * out,loff_t * opos,size_t len,unsigned int flags)951 long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
952 		      loff_t *opos, size_t len, unsigned int flags)
953 {
954 	struct splice_desc sd = {
955 		.len		= len,
956 		.total_len	= len,
957 		.flags		= flags,
958 		.pos		= *ppos,
959 		.u.file		= out,
960 		.opos		= opos,
961 	};
962 	long ret;
963 
964 	if (unlikely(!(out->f_mode & FMODE_WRITE)))
965 		return -EBADF;
966 
967 	if (unlikely(out->f_flags & O_APPEND))
968 		return -EINVAL;
969 
970 	ret = rw_verify_area(WRITE, out, opos, len);
971 	if (unlikely(ret < 0))
972 		return ret;
973 
974 	ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
975 	if (ret > 0)
976 		*ppos = sd.pos;
977 
978 	return ret;
979 }
980 EXPORT_SYMBOL(do_splice_direct);
981 
wait_for_space(struct pipe_inode_info * pipe,unsigned flags)982 static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags)
983 {
984 	for (;;) {
985 		if (unlikely(!pipe->readers)) {
986 			send_sig(SIGPIPE, current, 0);
987 			return -EPIPE;
988 		}
989 		if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage))
990 			return 0;
991 		if (flags & SPLICE_F_NONBLOCK)
992 			return -EAGAIN;
993 		if (signal_pending(current))
994 			return -ERESTARTSYS;
995 		pipe_wait_writable(pipe);
996 	}
997 }
998 
999 static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1000 			       struct pipe_inode_info *opipe,
1001 			       size_t len, unsigned int flags);
1002 
splice_file_to_pipe(struct file * in,struct pipe_inode_info * opipe,loff_t * offset,size_t len,unsigned int flags)1003 long splice_file_to_pipe(struct file *in,
1004 			 struct pipe_inode_info *opipe,
1005 			 loff_t *offset,
1006 			 size_t len, unsigned int flags)
1007 {
1008 	long ret;
1009 
1010 	pipe_lock(opipe);
1011 	ret = wait_for_space(opipe, flags);
1012 	if (!ret)
1013 		ret = do_splice_to(in, offset, opipe, len, flags);
1014 	pipe_unlock(opipe);
1015 	if (ret > 0)
1016 		wakeup_pipe_readers(opipe);
1017 	return ret;
1018 }
1019 
1020 /*
1021  * Determine where to splice to/from.
1022  */
do_splice(struct file * in,loff_t * off_in,struct file * out,loff_t * off_out,size_t len,unsigned int flags)1023 long do_splice(struct file *in, loff_t *off_in, struct file *out,
1024 	       loff_t *off_out, size_t len, unsigned int flags)
1025 {
1026 	struct pipe_inode_info *ipipe;
1027 	struct pipe_inode_info *opipe;
1028 	loff_t offset;
1029 	long ret;
1030 
1031 	if (unlikely(!(in->f_mode & FMODE_READ) ||
1032 		     !(out->f_mode & FMODE_WRITE)))
1033 		return -EBADF;
1034 
1035 	ipipe = get_pipe_info(in, true);
1036 	opipe = get_pipe_info(out, true);
1037 
1038 	if (ipipe && opipe) {
1039 		if (off_in || off_out)
1040 			return -ESPIPE;
1041 
1042 		/* Splicing to self would be fun, but... */
1043 		if (ipipe == opipe)
1044 			return -EINVAL;
1045 
1046 		if ((in->f_flags | out->f_flags) & O_NONBLOCK)
1047 			flags |= SPLICE_F_NONBLOCK;
1048 
1049 		return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1050 	}
1051 
1052 	if (ipipe) {
1053 		if (off_in)
1054 			return -ESPIPE;
1055 		if (off_out) {
1056 			if (!(out->f_mode & FMODE_PWRITE))
1057 				return -EINVAL;
1058 			offset = *off_out;
1059 		} else {
1060 			offset = out->f_pos;
1061 		}
1062 
1063 		if (unlikely(out->f_flags & O_APPEND))
1064 			return -EINVAL;
1065 
1066 		ret = rw_verify_area(WRITE, out, &offset, len);
1067 		if (unlikely(ret < 0))
1068 			return ret;
1069 
1070 		if (in->f_flags & O_NONBLOCK)
1071 			flags |= SPLICE_F_NONBLOCK;
1072 
1073 		file_start_write(out);
1074 		ret = do_splice_from(ipipe, out, &offset, len, flags);
1075 		file_end_write(out);
1076 
1077 		if (!off_out)
1078 			out->f_pos = offset;
1079 		else
1080 			*off_out = offset;
1081 
1082 		return ret;
1083 	}
1084 
1085 	if (opipe) {
1086 		if (off_out)
1087 			return -ESPIPE;
1088 		if (off_in) {
1089 			if (!(in->f_mode & FMODE_PREAD))
1090 				return -EINVAL;
1091 			offset = *off_in;
1092 		} else {
1093 			offset = in->f_pos;
1094 		}
1095 
1096 		if (out->f_flags & O_NONBLOCK)
1097 			flags |= SPLICE_F_NONBLOCK;
1098 
1099 		ret = splice_file_to_pipe(in, opipe, &offset, len, flags);
1100 		if (!off_in)
1101 			in->f_pos = offset;
1102 		else
1103 			*off_in = offset;
1104 
1105 		return ret;
1106 	}
1107 
1108 	return -EINVAL;
1109 }
1110 
__do_splice(struct file * in,loff_t __user * off_in,struct file * out,loff_t __user * off_out,size_t len,unsigned int flags)1111 static long __do_splice(struct file *in, loff_t __user *off_in,
1112 			struct file *out, loff_t __user *off_out,
1113 			size_t len, unsigned int flags)
1114 {
1115 	struct pipe_inode_info *ipipe;
1116 	struct pipe_inode_info *opipe;
1117 	loff_t offset, *__off_in = NULL, *__off_out = NULL;
1118 	long ret;
1119 
1120 	ipipe = get_pipe_info(in, true);
1121 	opipe = get_pipe_info(out, true);
1122 
1123 	if (ipipe && off_in)
1124 		return -ESPIPE;
1125 	if (opipe && off_out)
1126 		return -ESPIPE;
1127 
1128 	if (off_out) {
1129 		if (copy_from_user(&offset, off_out, sizeof(loff_t)))
1130 			return -EFAULT;
1131 		__off_out = &offset;
1132 	}
1133 	if (off_in) {
1134 		if (copy_from_user(&offset, off_in, sizeof(loff_t)))
1135 			return -EFAULT;
1136 		__off_in = &offset;
1137 	}
1138 
1139 	ret = do_splice(in, __off_in, out, __off_out, len, flags);
1140 	if (ret < 0)
1141 		return ret;
1142 
1143 	if (__off_out && copy_to_user(off_out, __off_out, sizeof(loff_t)))
1144 		return -EFAULT;
1145 	if (__off_in && copy_to_user(off_in, __off_in, sizeof(loff_t)))
1146 		return -EFAULT;
1147 
1148 	return ret;
1149 }
1150 
iter_to_pipe(struct iov_iter * from,struct pipe_inode_info * pipe,unsigned flags)1151 static int iter_to_pipe(struct iov_iter *from,
1152 			struct pipe_inode_info *pipe,
1153 			unsigned flags)
1154 {
1155 	struct pipe_buffer buf = {
1156 		.ops = &user_page_pipe_buf_ops,
1157 		.flags = flags
1158 	};
1159 	size_t total = 0;
1160 	int ret = 0;
1161 
1162 	while (iov_iter_count(from)) {
1163 		struct page *pages[16];
1164 		ssize_t left;
1165 		size_t start;
1166 		int i, n;
1167 
1168 		left = iov_iter_get_pages2(from, pages, ~0UL, 16, &start);
1169 		if (left <= 0) {
1170 			ret = left;
1171 			break;
1172 		}
1173 
1174 		n = DIV_ROUND_UP(left + start, PAGE_SIZE);
1175 		for (i = 0; i < n; i++) {
1176 			int size = min_t(int, left, PAGE_SIZE - start);
1177 
1178 			buf.page = pages[i];
1179 			buf.offset = start;
1180 			buf.len = size;
1181 			ret = add_to_pipe(pipe, &buf);
1182 			if (unlikely(ret < 0)) {
1183 				iov_iter_revert(from, left);
1184 				// this one got dropped by add_to_pipe()
1185 				while (++i < n)
1186 					put_page(pages[i]);
1187 				goto out;
1188 			}
1189 			total += ret;
1190 			left -= size;
1191 			start = 0;
1192 		}
1193 	}
1194 out:
1195 	return total ? total : ret;
1196 }
1197 
pipe_to_user(struct pipe_inode_info * pipe,struct pipe_buffer * buf,struct splice_desc * sd)1198 static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1199 			struct splice_desc *sd)
1200 {
1201 	int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1202 	return n == sd->len ? n : -EFAULT;
1203 }
1204 
1205 /*
1206  * For lack of a better implementation, implement vmsplice() to userspace
1207  * as a simple copy of the pipes pages to the user iov.
1208  */
vmsplice_to_user(struct file * file,struct iov_iter * iter,unsigned int flags)1209 static long vmsplice_to_user(struct file *file, struct iov_iter *iter,
1210 			     unsigned int flags)
1211 {
1212 	struct pipe_inode_info *pipe = get_pipe_info(file, true);
1213 	struct splice_desc sd = {
1214 		.total_len = iov_iter_count(iter),
1215 		.flags = flags,
1216 		.u.data = iter
1217 	};
1218 	long ret = 0;
1219 
1220 	if (!pipe)
1221 		return -EBADF;
1222 
1223 	if (sd.total_len) {
1224 		pipe_lock(pipe);
1225 		ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
1226 		pipe_unlock(pipe);
1227 	}
1228 
1229 	return ret;
1230 }
1231 
1232 /*
1233  * vmsplice splices a user address range into a pipe. It can be thought of
1234  * as splice-from-memory, where the regular splice is splice-from-file (or
1235  * to file). In both cases the output is a pipe, naturally.
1236  */
vmsplice_to_pipe(struct file * file,struct iov_iter * iter,unsigned int flags)1237 static long vmsplice_to_pipe(struct file *file, struct iov_iter *iter,
1238 			     unsigned int flags)
1239 {
1240 	struct pipe_inode_info *pipe;
1241 	long ret = 0;
1242 	unsigned buf_flag = 0;
1243 
1244 	if (flags & SPLICE_F_GIFT)
1245 		buf_flag = PIPE_BUF_FLAG_GIFT;
1246 
1247 	pipe = get_pipe_info(file, true);
1248 	if (!pipe)
1249 		return -EBADF;
1250 
1251 	pipe_lock(pipe);
1252 	ret = wait_for_space(pipe, flags);
1253 	if (!ret)
1254 		ret = iter_to_pipe(iter, pipe, buf_flag);
1255 	pipe_unlock(pipe);
1256 	if (ret > 0)
1257 		wakeup_pipe_readers(pipe);
1258 	return ret;
1259 }
1260 
vmsplice_type(struct fd f,int * type)1261 static int vmsplice_type(struct fd f, int *type)
1262 {
1263 	if (!f.file)
1264 		return -EBADF;
1265 	if (f.file->f_mode & FMODE_WRITE) {
1266 		*type = WRITE;
1267 	} else if (f.file->f_mode & FMODE_READ) {
1268 		*type = READ;
1269 	} else {
1270 		fdput(f);
1271 		return -EBADF;
1272 	}
1273 	return 0;
1274 }
1275 
1276 /*
1277  * Note that vmsplice only really supports true splicing _from_ user memory
1278  * to a pipe, not the other way around. Splicing from user memory is a simple
1279  * operation that can be supported without any funky alignment restrictions
1280  * or nasty vm tricks. We simply map in the user memory and fill them into
1281  * a pipe. The reverse isn't quite as easy, though. There are two possible
1282  * solutions for that:
1283  *
1284  *	- memcpy() the data internally, at which point we might as well just
1285  *	  do a regular read() on the buffer anyway.
1286  *	- Lots of nasty vm tricks, that are neither fast nor flexible (it
1287  *	  has restriction limitations on both ends of the pipe).
1288  *
1289  * Currently we punt and implement it as a normal copy, see pipe_to_user().
1290  *
1291  */
SYSCALL_DEFINE4(vmsplice,int,fd,const struct iovec __user *,uiov,unsigned long,nr_segs,unsigned int,flags)1292 SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, uiov,
1293 		unsigned long, nr_segs, unsigned int, flags)
1294 {
1295 	struct iovec iovstack[UIO_FASTIOV];
1296 	struct iovec *iov = iovstack;
1297 	struct iov_iter iter;
1298 	ssize_t error;
1299 	struct fd f;
1300 	int type;
1301 
1302 	if (unlikely(flags & ~SPLICE_F_ALL))
1303 		return -EINVAL;
1304 
1305 	f = fdget(fd);
1306 	error = vmsplice_type(f, &type);
1307 	if (error)
1308 		return error;
1309 
1310 	error = import_iovec(type, uiov, nr_segs,
1311 			     ARRAY_SIZE(iovstack), &iov, &iter);
1312 	if (error < 0)
1313 		goto out_fdput;
1314 
1315 	if (!iov_iter_count(&iter))
1316 		error = 0;
1317 	else if (iov_iter_rw(&iter) == WRITE)
1318 		error = vmsplice_to_pipe(f.file, &iter, flags);
1319 	else
1320 		error = vmsplice_to_user(f.file, &iter, flags);
1321 
1322 	kfree(iov);
1323 out_fdput:
1324 	fdput(f);
1325 	return error;
1326 }
1327 
SYSCALL_DEFINE6(splice,int,fd_in,loff_t __user *,off_in,int,fd_out,loff_t __user *,off_out,size_t,len,unsigned int,flags)1328 SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1329 		int, fd_out, loff_t __user *, off_out,
1330 		size_t, len, unsigned int, flags)
1331 {
1332 	struct fd in, out;
1333 	long error;
1334 
1335 	if (unlikely(!len))
1336 		return 0;
1337 
1338 	if (unlikely(flags & ~SPLICE_F_ALL))
1339 		return -EINVAL;
1340 
1341 	error = -EBADF;
1342 	in = fdget(fd_in);
1343 	if (in.file) {
1344 		out = fdget(fd_out);
1345 		if (out.file) {
1346 			error = __do_splice(in.file, off_in, out.file, off_out,
1347 						len, flags);
1348 			fdput(out);
1349 		}
1350 		fdput(in);
1351 	}
1352 	return error;
1353 }
1354 
1355 /*
1356  * Make sure there's data to read. Wait for input if we can, otherwise
1357  * return an appropriate error.
1358  */
ipipe_prep(struct pipe_inode_info * pipe,unsigned int flags)1359 static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1360 {
1361 	int ret;
1362 
1363 	/*
1364 	 * Check the pipe occupancy without the inode lock first. This function
1365 	 * is speculative anyways, so missing one is ok.
1366 	 */
1367 	if (!pipe_empty(pipe->head, pipe->tail))
1368 		return 0;
1369 
1370 	ret = 0;
1371 	pipe_lock(pipe);
1372 
1373 	while (pipe_empty(pipe->head, pipe->tail)) {
1374 		if (signal_pending(current)) {
1375 			ret = -ERESTARTSYS;
1376 			break;
1377 		}
1378 		if (!pipe->writers)
1379 			break;
1380 		if (flags & SPLICE_F_NONBLOCK) {
1381 			ret = -EAGAIN;
1382 			break;
1383 		}
1384 		pipe_wait_readable(pipe);
1385 	}
1386 
1387 	pipe_unlock(pipe);
1388 	return ret;
1389 }
1390 
1391 /*
1392  * Make sure there's writeable room. Wait for room if we can, otherwise
1393  * return an appropriate error.
1394  */
opipe_prep(struct pipe_inode_info * pipe,unsigned int flags)1395 static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1396 {
1397 	int ret;
1398 
1399 	/*
1400 	 * Check pipe occupancy without the inode lock first. This function
1401 	 * is speculative anyways, so missing one is ok.
1402 	 */
1403 	if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage))
1404 		return 0;
1405 
1406 	ret = 0;
1407 	pipe_lock(pipe);
1408 
1409 	while (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
1410 		if (!pipe->readers) {
1411 			send_sig(SIGPIPE, current, 0);
1412 			ret = -EPIPE;
1413 			break;
1414 		}
1415 		if (flags & SPLICE_F_NONBLOCK) {
1416 			ret = -EAGAIN;
1417 			break;
1418 		}
1419 		if (signal_pending(current)) {
1420 			ret = -ERESTARTSYS;
1421 			break;
1422 		}
1423 		pipe_wait_writable(pipe);
1424 	}
1425 
1426 	pipe_unlock(pipe);
1427 	return ret;
1428 }
1429 
1430 /*
1431  * Splice contents of ipipe to opipe.
1432  */
splice_pipe_to_pipe(struct pipe_inode_info * ipipe,struct pipe_inode_info * opipe,size_t len,unsigned int flags)1433 static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1434 			       struct pipe_inode_info *opipe,
1435 			       size_t len, unsigned int flags)
1436 {
1437 	struct pipe_buffer *ibuf, *obuf;
1438 	unsigned int i_head, o_head;
1439 	unsigned int i_tail, o_tail;
1440 	unsigned int i_mask, o_mask;
1441 	int ret = 0;
1442 	bool input_wakeup = false;
1443 
1444 
1445 retry:
1446 	ret = ipipe_prep(ipipe, flags);
1447 	if (ret)
1448 		return ret;
1449 
1450 	ret = opipe_prep(opipe, flags);
1451 	if (ret)
1452 		return ret;
1453 
1454 	/*
1455 	 * Potential ABBA deadlock, work around it by ordering lock
1456 	 * grabbing by pipe info address. Otherwise two different processes
1457 	 * could deadlock (one doing tee from A -> B, the other from B -> A).
1458 	 */
1459 	pipe_double_lock(ipipe, opipe);
1460 
1461 	i_tail = ipipe->tail;
1462 	i_mask = ipipe->ring_size - 1;
1463 	o_head = opipe->head;
1464 	o_mask = opipe->ring_size - 1;
1465 
1466 	do {
1467 		size_t o_len;
1468 
1469 		if (!opipe->readers) {
1470 			send_sig(SIGPIPE, current, 0);
1471 			if (!ret)
1472 				ret = -EPIPE;
1473 			break;
1474 		}
1475 
1476 		i_head = ipipe->head;
1477 		o_tail = opipe->tail;
1478 
1479 		if (pipe_empty(i_head, i_tail) && !ipipe->writers)
1480 			break;
1481 
1482 		/*
1483 		 * Cannot make any progress, because either the input
1484 		 * pipe is empty or the output pipe is full.
1485 		 */
1486 		if (pipe_empty(i_head, i_tail) ||
1487 		    pipe_full(o_head, o_tail, opipe->max_usage)) {
1488 			/* Already processed some buffers, break */
1489 			if (ret)
1490 				break;
1491 
1492 			if (flags & SPLICE_F_NONBLOCK) {
1493 				ret = -EAGAIN;
1494 				break;
1495 			}
1496 
1497 			/*
1498 			 * We raced with another reader/writer and haven't
1499 			 * managed to process any buffers.  A zero return
1500 			 * value means EOF, so retry instead.
1501 			 */
1502 			pipe_unlock(ipipe);
1503 			pipe_unlock(opipe);
1504 			goto retry;
1505 		}
1506 
1507 		ibuf = &ipipe->bufs[i_tail & i_mask];
1508 		obuf = &opipe->bufs[o_head & o_mask];
1509 
1510 		if (len >= ibuf->len) {
1511 			/*
1512 			 * Simply move the whole buffer from ipipe to opipe
1513 			 */
1514 			*obuf = *ibuf;
1515 			ibuf->ops = NULL;
1516 			i_tail++;
1517 			ipipe->tail = i_tail;
1518 			input_wakeup = true;
1519 			o_len = obuf->len;
1520 			o_head++;
1521 			opipe->head = o_head;
1522 		} else {
1523 			/*
1524 			 * Get a reference to this pipe buffer,
1525 			 * so we can copy the contents over.
1526 			 */
1527 			if (!pipe_buf_get(ipipe, ibuf)) {
1528 				if (ret == 0)
1529 					ret = -EFAULT;
1530 				break;
1531 			}
1532 			*obuf = *ibuf;
1533 
1534 			/*
1535 			 * Don't inherit the gift and merge flags, we need to
1536 			 * prevent multiple steals of this page.
1537 			 */
1538 			obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1539 			obuf->flags &= ~PIPE_BUF_FLAG_CAN_MERGE;
1540 
1541 			obuf->len = len;
1542 			ibuf->offset += len;
1543 			ibuf->len -= len;
1544 			o_len = len;
1545 			o_head++;
1546 			opipe->head = o_head;
1547 		}
1548 		ret += o_len;
1549 		len -= o_len;
1550 	} while (len);
1551 
1552 	pipe_unlock(ipipe);
1553 	pipe_unlock(opipe);
1554 
1555 	/*
1556 	 * If we put data in the output pipe, wakeup any potential readers.
1557 	 */
1558 	if (ret > 0)
1559 		wakeup_pipe_readers(opipe);
1560 
1561 	if (input_wakeup)
1562 		wakeup_pipe_writers(ipipe);
1563 
1564 	return ret;
1565 }
1566 
1567 /*
1568  * Link contents of ipipe to opipe.
1569  */
link_pipe(struct pipe_inode_info * ipipe,struct pipe_inode_info * opipe,size_t len,unsigned int flags)1570 static int link_pipe(struct pipe_inode_info *ipipe,
1571 		     struct pipe_inode_info *opipe,
1572 		     size_t len, unsigned int flags)
1573 {
1574 	struct pipe_buffer *ibuf, *obuf;
1575 	unsigned int i_head, o_head;
1576 	unsigned int i_tail, o_tail;
1577 	unsigned int i_mask, o_mask;
1578 	int ret = 0;
1579 
1580 	/*
1581 	 * Potential ABBA deadlock, work around it by ordering lock
1582 	 * grabbing by pipe info address. Otherwise two different processes
1583 	 * could deadlock (one doing tee from A -> B, the other from B -> A).
1584 	 */
1585 	pipe_double_lock(ipipe, opipe);
1586 
1587 	i_tail = ipipe->tail;
1588 	i_mask = ipipe->ring_size - 1;
1589 	o_head = opipe->head;
1590 	o_mask = opipe->ring_size - 1;
1591 
1592 	do {
1593 		if (!opipe->readers) {
1594 			send_sig(SIGPIPE, current, 0);
1595 			if (!ret)
1596 				ret = -EPIPE;
1597 			break;
1598 		}
1599 
1600 		i_head = ipipe->head;
1601 		o_tail = opipe->tail;
1602 
1603 		/*
1604 		 * If we have iterated all input buffers or run out of
1605 		 * output room, break.
1606 		 */
1607 		if (pipe_empty(i_head, i_tail) ||
1608 		    pipe_full(o_head, o_tail, opipe->max_usage))
1609 			break;
1610 
1611 		ibuf = &ipipe->bufs[i_tail & i_mask];
1612 		obuf = &opipe->bufs[o_head & o_mask];
1613 
1614 		/*
1615 		 * Get a reference to this pipe buffer,
1616 		 * so we can copy the contents over.
1617 		 */
1618 		if (!pipe_buf_get(ipipe, ibuf)) {
1619 			if (ret == 0)
1620 				ret = -EFAULT;
1621 			break;
1622 		}
1623 
1624 		*obuf = *ibuf;
1625 
1626 		/*
1627 		 * Don't inherit the gift and merge flag, we need to prevent
1628 		 * multiple steals of this page.
1629 		 */
1630 		obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1631 		obuf->flags &= ~PIPE_BUF_FLAG_CAN_MERGE;
1632 
1633 		if (obuf->len > len)
1634 			obuf->len = len;
1635 		ret += obuf->len;
1636 		len -= obuf->len;
1637 
1638 		o_head++;
1639 		opipe->head = o_head;
1640 		i_tail++;
1641 	} while (len);
1642 
1643 	pipe_unlock(ipipe);
1644 	pipe_unlock(opipe);
1645 
1646 	/*
1647 	 * If we put data in the output pipe, wakeup any potential readers.
1648 	 */
1649 	if (ret > 0)
1650 		wakeup_pipe_readers(opipe);
1651 
1652 	return ret;
1653 }
1654 
1655 /*
1656  * This is a tee(1) implementation that works on pipes. It doesn't copy
1657  * any data, it simply references the 'in' pages on the 'out' pipe.
1658  * The 'flags' used are the SPLICE_F_* variants, currently the only
1659  * applicable one is SPLICE_F_NONBLOCK.
1660  */
do_tee(struct file * in,struct file * out,size_t len,unsigned int flags)1661 long do_tee(struct file *in, struct file *out, size_t len, unsigned int flags)
1662 {
1663 	struct pipe_inode_info *ipipe = get_pipe_info(in, true);
1664 	struct pipe_inode_info *opipe = get_pipe_info(out, true);
1665 	int ret = -EINVAL;
1666 
1667 	if (unlikely(!(in->f_mode & FMODE_READ) ||
1668 		     !(out->f_mode & FMODE_WRITE)))
1669 		return -EBADF;
1670 
1671 	/*
1672 	 * Duplicate the contents of ipipe to opipe without actually
1673 	 * copying the data.
1674 	 */
1675 	if (ipipe && opipe && ipipe != opipe) {
1676 		if ((in->f_flags | out->f_flags) & O_NONBLOCK)
1677 			flags |= SPLICE_F_NONBLOCK;
1678 
1679 		/*
1680 		 * Keep going, unless we encounter an error. The ipipe/opipe
1681 		 * ordering doesn't really matter.
1682 		 */
1683 		ret = ipipe_prep(ipipe, flags);
1684 		if (!ret) {
1685 			ret = opipe_prep(opipe, flags);
1686 			if (!ret)
1687 				ret = link_pipe(ipipe, opipe, len, flags);
1688 		}
1689 	}
1690 
1691 	return ret;
1692 }
1693 
SYSCALL_DEFINE4(tee,int,fdin,int,fdout,size_t,len,unsigned int,flags)1694 SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
1695 {
1696 	struct fd in, out;
1697 	int error;
1698 
1699 	if (unlikely(flags & ~SPLICE_F_ALL))
1700 		return -EINVAL;
1701 
1702 	if (unlikely(!len))
1703 		return 0;
1704 
1705 	error = -EBADF;
1706 	in = fdget(fdin);
1707 	if (in.file) {
1708 		out = fdget(fdout);
1709 		if (out.file) {
1710 			error = do_tee(in.file, out.file, len, flags);
1711 			fdput(out);
1712 		}
1713  		fdput(in);
1714  	}
1715 
1716 	return error;
1717 }
1718