1 /*
2  * "splice": joining two ropes together by interweaving their strands.
3  *
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.
7  *
8  * The traditional unix read/write is extended with a "splice()" operation
9  * that transfers data buffers to or from a pipe buffer.
10  *
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.
14  *
15  * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk>
16  * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
17  * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
18  *
19  */
20 #include <linux/bvec.h>
21 #include <linux/fs.h>
22 #include <linux/file.h>
23 #include <linux/pagemap.h>
24 #include <linux/splice.h>
25 #include <linux/memcontrol.h>
26 #include <linux/mm_inline.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/export.h>
30 #include <linux/syscalls.h>
31 #include <linux/uio.h>
32 #include <linux/security.h>
33 #include <linux/gfp.h>
34 #include <linux/socket.h>
35 #include <linux/compat.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_steal(struct pipe_inode_info * pipe,struct pipe_buffer * buf)46 static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
47 				     struct pipe_buffer *buf)
48 {
49 	struct page *page = buf->page;
50 	struct address_space *mapping;
51 
52 	lock_page(page);
53 
54 	mapping = page_mapping(page);
55 	if (mapping) {
56 		WARN_ON(!PageUptodate(page));
57 
58 		/*
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
64 		 * ensues.
65 		 */
66 		wait_on_page_writeback(page);
67 
68 		if (page_has_private(page) &&
69 		    !try_to_release_page(page, 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, page)) {
77 			buf->flags |= PIPE_BUF_FLAG_LRU;
78 			return 0;
79 		}
80 	}
81 
82 	/*
83 	 * Raced with truncate or failed to remove page from current
84 	 * address space, unlock and return failure.
85 	 */
86 out_unlock:
87 	unlock_page(page);
88 	return 1;
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 	.can_merge = 0,
142 	.confirm = page_cache_pipe_buf_confirm,
143 	.release = page_cache_pipe_buf_release,
144 	.steal = page_cache_pipe_buf_steal,
145 	.get = generic_pipe_buf_get,
146 };
147 
user_page_pipe_buf_steal(struct pipe_inode_info * pipe,struct pipe_buffer * buf)148 static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
149 				    struct pipe_buffer *buf)
150 {
151 	if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
152 		return 1;
153 
154 	buf->flags |= PIPE_BUF_FLAG_LRU;
155 	return generic_pipe_buf_steal(pipe, buf);
156 }
157 
158 static const struct pipe_buf_operations user_page_pipe_buf_ops = {
159 	.can_merge = 0,
160 	.confirm = generic_pipe_buf_confirm,
161 	.release = page_cache_pipe_buf_release,
162 	.steal = user_page_pipe_buf_steal,
163 	.get = generic_pipe_buf_get,
164 };
165 
wakeup_pipe_readers(struct pipe_inode_info * pipe)166 static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
167 {
168 	smp_mb();
169 	if (waitqueue_active(&pipe->wait))
170 		wake_up_interruptible(&pipe->wait);
171 	kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
172 }
173 
174 /**
175  * splice_to_pipe - fill passed data into a pipe
176  * @pipe:	pipe to fill
177  * @spd:	data to fill
178  *
179  * Description:
180  *    @spd contains a map of pages and len/offset tuples, along with
181  *    the struct pipe_buf_operations associated with these pages. This
182  *    function will link that data to the pipe.
183  *
184  */
splice_to_pipe(struct pipe_inode_info * pipe,struct splice_pipe_desc * spd)185 ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
186 		       struct splice_pipe_desc *spd)
187 {
188 	unsigned int spd_pages = spd->nr_pages;
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->nrbufs < pipe->buffers) {
201 		int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
202 		struct pipe_buffer *buf = pipe->bufs + newbuf;
203 
204 		buf->page = spd->pages[page_nr];
205 		buf->offset = spd->partial[page_nr].offset;
206 		buf->len = spd->partial[page_nr].len;
207 		buf->private = spd->partial[page_nr].private;
208 		buf->ops = spd->ops;
209 		buf->flags = 0;
210 
211 		pipe->nrbufs++;
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 	int ret;
233 
234 	if (unlikely(!pipe->readers)) {
235 		send_sig(SIGPIPE, current, 0);
236 		ret = -EPIPE;
237 	} else if (pipe->nrbufs == pipe->buffers) {
238 		ret = -EAGAIN;
239 	} else {
240 		int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
241 		pipe->bufs[newbuf] = *buf;
242 		pipe->nrbufs++;
243 		return buf->len;
244 	}
245 	pipe_buf_release(pipe, buf);
246 	return ret;
247 }
248 EXPORT_SYMBOL(add_to_pipe);
249 
250 /*
251  * Check if we need to grow the arrays holding pages and partial page
252  * descriptions.
253  */
splice_grow_spd(const struct pipe_inode_info * pipe,struct splice_pipe_desc * spd)254 int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
255 {
256 	unsigned int buffers = READ_ONCE(pipe->buffers);
257 
258 	spd->nr_pages_max = buffers;
259 	if (buffers <= PIPE_DEF_BUFFERS)
260 		return 0;
261 
262 	spd->pages = kmalloc_array(buffers, sizeof(struct page *), GFP_KERNEL);
263 	spd->partial = kmalloc_array(buffers, sizeof(struct partial_page),
264 				     GFP_KERNEL);
265 
266 	if (spd->pages && spd->partial)
267 		return 0;
268 
269 	kfree(spd->pages);
270 	kfree(spd->partial);
271 	return -ENOMEM;
272 }
273 
splice_shrink_spd(struct splice_pipe_desc * spd)274 void splice_shrink_spd(struct splice_pipe_desc *spd)
275 {
276 	if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
277 		return;
278 
279 	kfree(spd->pages);
280 	kfree(spd->partial);
281 }
282 
283 /**
284  * generic_file_splice_read - splice data from file to a pipe
285  * @in:		file to splice from
286  * @ppos:	position in @in
287  * @pipe:	pipe to splice to
288  * @len:	number of bytes to splice
289  * @flags:	splice modifier flags
290  *
291  * Description:
292  *    Will read pages from given file and fill them into a pipe. Can be
293  *    used as long as it has more or less sane ->read_iter().
294  *
295  */
generic_file_splice_read(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)296 ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
297 				 struct pipe_inode_info *pipe, size_t len,
298 				 unsigned int flags)
299 {
300 	struct iov_iter to;
301 	struct kiocb kiocb;
302 	int idx, ret;
303 
304 	iov_iter_pipe(&to, ITER_PIPE | READ, pipe, len);
305 	idx = to.idx;
306 	init_sync_kiocb(&kiocb, in);
307 	kiocb.ki_pos = *ppos;
308 	ret = call_read_iter(in, &kiocb, &to);
309 	if (ret > 0) {
310 		*ppos = kiocb.ki_pos;
311 		file_accessed(in);
312 	} else if (ret < 0) {
313 		to.idx = idx;
314 		to.iov_offset = 0;
315 		iov_iter_advance(&to, 0); /* to free what was emitted */
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 	.can_merge = 0,
330 	.confirm = generic_pipe_buf_confirm,
331 	.release = generic_pipe_buf_release,
332 	.steal = generic_pipe_buf_steal,
333 	.get = generic_pipe_buf_get,
334 };
335 
generic_pipe_buf_nosteal(struct pipe_inode_info * pipe,struct pipe_buffer * buf)336 int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe,
337 			     struct pipe_buffer *buf)
338 {
339 	return 1;
340 }
341 
342 /* Pipe buffer operations for a socket and similar. */
343 const struct pipe_buf_operations nosteal_pipe_buf_ops = {
344 	.can_merge = 0,
345 	.confirm = generic_pipe_buf_confirm,
346 	.release = generic_pipe_buf_release,
347 	.steal = generic_pipe_buf_nosteal,
348 	.get = generic_pipe_buf_get,
349 };
350 EXPORT_SYMBOL(nosteal_pipe_buf_ops);
351 
kernel_readv(struct file * file,const struct kvec * vec,unsigned long vlen,loff_t offset)352 static ssize_t kernel_readv(struct file *file, const struct kvec *vec,
353 			    unsigned long vlen, loff_t offset)
354 {
355 	mm_segment_t old_fs;
356 	loff_t pos = offset;
357 	ssize_t res;
358 
359 	old_fs = get_fs();
360 	set_fs(get_ds());
361 	/* The cast to a user pointer is valid due to the set_fs() */
362 	res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos, 0);
363 	set_fs(old_fs);
364 
365 	return res;
366 }
367 
default_file_splice_read(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)368 static ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
369 				 struct pipe_inode_info *pipe, size_t len,
370 				 unsigned int flags)
371 {
372 	struct kvec *vec, __vec[PIPE_DEF_BUFFERS];
373 	struct iov_iter to;
374 	struct page **pages;
375 	unsigned int nr_pages;
376 	size_t offset, base, copied = 0;
377 	ssize_t res;
378 	int i;
379 
380 	if (pipe->nrbufs == pipe->buffers)
381 		return -EAGAIN;
382 
383 	/*
384 	 * Try to keep page boundaries matching to source pagecache ones -
385 	 * it probably won't be much help, but...
386 	 */
387 	offset = *ppos & ~PAGE_MASK;
388 
389 	iov_iter_pipe(&to, ITER_PIPE | READ, pipe, len + offset);
390 
391 	res = iov_iter_get_pages_alloc(&to, &pages, len + offset, &base);
392 	if (res <= 0)
393 		return -ENOMEM;
394 
395 	nr_pages = DIV_ROUND_UP(res + base, PAGE_SIZE);
396 
397 	vec = __vec;
398 	if (nr_pages > PIPE_DEF_BUFFERS) {
399 		vec = kmalloc_array(nr_pages, sizeof(struct kvec), GFP_KERNEL);
400 		if (unlikely(!vec)) {
401 			res = -ENOMEM;
402 			goto out;
403 		}
404 	}
405 
406 	pipe->bufs[to.idx].offset = offset;
407 	pipe->bufs[to.idx].len -= offset;
408 
409 	for (i = 0; i < nr_pages; i++) {
410 		size_t this_len = min_t(size_t, len, PAGE_SIZE - offset);
411 		vec[i].iov_base = page_address(pages[i]) + offset;
412 		vec[i].iov_len = this_len;
413 		len -= this_len;
414 		offset = 0;
415 	}
416 
417 	res = kernel_readv(in, vec, nr_pages, *ppos);
418 	if (res > 0) {
419 		copied = res;
420 		*ppos += res;
421 	}
422 
423 	if (vec != __vec)
424 		kfree(vec);
425 out:
426 	for (i = 0; i < nr_pages; i++)
427 		put_page(pages[i]);
428 	kvfree(pages);
429 	iov_iter_advance(&to, copied);	/* truncates and discards */
430 	return res;
431 }
432 
433 /*
434  * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
435  * using sendpage(). Return the number of bytes sent.
436  */
pipe_to_sendpage(struct pipe_inode_info * pipe,struct pipe_buffer * buf,struct splice_desc * sd)437 static int pipe_to_sendpage(struct pipe_inode_info *pipe,
438 			    struct pipe_buffer *buf, struct splice_desc *sd)
439 {
440 	struct file *file = sd->u.file;
441 	loff_t pos = sd->pos;
442 	int more;
443 
444 	if (!likely(file->f_op->sendpage))
445 		return -EINVAL;
446 
447 	more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
448 
449 	if (sd->len < sd->total_len && pipe->nrbufs > 1)
450 		more |= MSG_SENDPAGE_NOTLAST;
451 
452 	return file->f_op->sendpage(file, buf->page, buf->offset,
453 				    sd->len, &pos, more);
454 }
455 
wakeup_pipe_writers(struct pipe_inode_info * pipe)456 static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
457 {
458 	smp_mb();
459 	if (waitqueue_active(&pipe->wait))
460 		wake_up_interruptible(&pipe->wait);
461 	kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
462 }
463 
464 /**
465  * splice_from_pipe_feed - feed available data from a pipe to a file
466  * @pipe:	pipe to splice from
467  * @sd:		information to @actor
468  * @actor:	handler that splices the data
469  *
470  * Description:
471  *    This function loops over the pipe and calls @actor to do the
472  *    actual moving of a single struct pipe_buffer to the desired
473  *    destination.  It returns when there's no more buffers left in
474  *    the pipe or if the requested number of bytes (@sd->total_len)
475  *    have been copied.  It returns a positive number (one) if the
476  *    pipe needs to be filled with more data, zero if the required
477  *    number of bytes have been copied and -errno on error.
478  *
479  *    This, together with splice_from_pipe_{begin,end,next}, may be
480  *    used to implement the functionality of __splice_from_pipe() when
481  *    locking is required around copying the pipe buffers to the
482  *    destination.
483  */
splice_from_pipe_feed(struct pipe_inode_info * pipe,struct splice_desc * sd,splice_actor * actor)484 static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
485 			  splice_actor *actor)
486 {
487 	int ret;
488 
489 	while (pipe->nrbufs) {
490 		struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
491 
492 		sd->len = buf->len;
493 		if (sd->len > sd->total_len)
494 			sd->len = sd->total_len;
495 
496 		ret = pipe_buf_confirm(pipe, buf);
497 		if (unlikely(ret)) {
498 			if (ret == -ENODATA)
499 				ret = 0;
500 			return ret;
501 		}
502 
503 		ret = actor(pipe, buf, sd);
504 		if (ret <= 0)
505 			return ret;
506 
507 		buf->offset += ret;
508 		buf->len -= ret;
509 
510 		sd->num_spliced += ret;
511 		sd->len -= ret;
512 		sd->pos += ret;
513 		sd->total_len -= ret;
514 
515 		if (!buf->len) {
516 			pipe_buf_release(pipe, buf);
517 			pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
518 			pipe->nrbufs--;
519 			if (pipe->files)
520 				sd->need_wakeup = true;
521 		}
522 
523 		if (!sd->total_len)
524 			return 0;
525 	}
526 
527 	return 1;
528 }
529 
530 /**
531  * splice_from_pipe_next - wait for some data to splice from
532  * @pipe:	pipe to splice from
533  * @sd:		information about the splice operation
534  *
535  * Description:
536  *    This function will wait for some data and return a positive
537  *    value (one) if pipe buffers are available.  It will return zero
538  *    or -errno if no more data needs to be spliced.
539  */
splice_from_pipe_next(struct pipe_inode_info * pipe,struct splice_desc * sd)540 static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
541 {
542 	/*
543 	 * Check for signal early to make process killable when there are
544 	 * always buffers available
545 	 */
546 	if (signal_pending(current))
547 		return -ERESTARTSYS;
548 
549 	while (!pipe->nrbufs) {
550 		if (!pipe->writers)
551 			return 0;
552 
553 		if (!pipe->waiting_writers && sd->num_spliced)
554 			return 0;
555 
556 		if (sd->flags & SPLICE_F_NONBLOCK)
557 			return -EAGAIN;
558 
559 		if (signal_pending(current))
560 			return -ERESTARTSYS;
561 
562 		if (sd->need_wakeup) {
563 			wakeup_pipe_writers(pipe);
564 			sd->need_wakeup = false;
565 		}
566 
567 		pipe_wait(pipe);
568 	}
569 
570 	return 1;
571 }
572 
573 /**
574  * splice_from_pipe_begin - start splicing from pipe
575  * @sd:		information about the splice operation
576  *
577  * Description:
578  *    This function should be called before a loop containing
579  *    splice_from_pipe_next() and splice_from_pipe_feed() to
580  *    initialize the necessary fields of @sd.
581  */
splice_from_pipe_begin(struct splice_desc * sd)582 static void splice_from_pipe_begin(struct splice_desc *sd)
583 {
584 	sd->num_spliced = 0;
585 	sd->need_wakeup = false;
586 }
587 
588 /**
589  * splice_from_pipe_end - finish splicing from pipe
590  * @pipe:	pipe to splice from
591  * @sd:		information about the splice operation
592  *
593  * Description:
594  *    This function will wake up pipe writers if necessary.  It should
595  *    be called after a loop containing splice_from_pipe_next() and
596  *    splice_from_pipe_feed().
597  */
splice_from_pipe_end(struct pipe_inode_info * pipe,struct splice_desc * sd)598 static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
599 {
600 	if (sd->need_wakeup)
601 		wakeup_pipe_writers(pipe);
602 }
603 
604 /**
605  * __splice_from_pipe - splice data from a pipe to given actor
606  * @pipe:	pipe to splice from
607  * @sd:		information to @actor
608  * @actor:	handler that splices the data
609  *
610  * Description:
611  *    This function does little more than loop over the pipe and call
612  *    @actor to do the actual moving of a single struct pipe_buffer to
613  *    the desired destination. See pipe_to_file, pipe_to_sendpage, or
614  *    pipe_to_user.
615  *
616  */
__splice_from_pipe(struct pipe_inode_info * pipe,struct splice_desc * sd,splice_actor * actor)617 ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
618 			   splice_actor *actor)
619 {
620 	int ret;
621 
622 	splice_from_pipe_begin(sd);
623 	do {
624 		cond_resched();
625 		ret = splice_from_pipe_next(pipe, sd);
626 		if (ret > 0)
627 			ret = splice_from_pipe_feed(pipe, sd, actor);
628 	} while (ret > 0);
629 	splice_from_pipe_end(pipe, sd);
630 
631 	return sd->num_spliced ? sd->num_spliced : ret;
632 }
633 EXPORT_SYMBOL(__splice_from_pipe);
634 
635 /**
636  * splice_from_pipe - splice data from a pipe to a file
637  * @pipe:	pipe to splice from
638  * @out:	file to splice to
639  * @ppos:	position in @out
640  * @len:	how many bytes to splice
641  * @flags:	splice modifier flags
642  * @actor:	handler that splices the data
643  *
644  * Description:
645  *    See __splice_from_pipe. This function locks the pipe inode,
646  *    otherwise it's identical to __splice_from_pipe().
647  *
648  */
splice_from_pipe(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags,splice_actor * actor)649 ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
650 			 loff_t *ppos, size_t len, unsigned int flags,
651 			 splice_actor *actor)
652 {
653 	ssize_t ret;
654 	struct splice_desc sd = {
655 		.total_len = len,
656 		.flags = flags,
657 		.pos = *ppos,
658 		.u.file = out,
659 	};
660 
661 	pipe_lock(pipe);
662 	ret = __splice_from_pipe(pipe, &sd, actor);
663 	pipe_unlock(pipe);
664 
665 	return ret;
666 }
667 
668 /**
669  * iter_file_splice_write - splice data from a pipe to a file
670  * @pipe:	pipe info
671  * @out:	file to write to
672  * @ppos:	position in @out
673  * @len:	number of bytes to splice
674  * @flags:	splice modifier flags
675  *
676  * Description:
677  *    Will either move or copy pages (determined by @flags options) from
678  *    the given pipe inode to the given file.
679  *    This one is ->write_iter-based.
680  *
681  */
682 ssize_t
iter_file_splice_write(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)683 iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
684 			  loff_t *ppos, size_t len, unsigned int flags)
685 {
686 	struct splice_desc sd = {
687 		.total_len = len,
688 		.flags = flags,
689 		.pos = *ppos,
690 		.u.file = out,
691 	};
692 	int nbufs = pipe->buffers;
693 	struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
694 					GFP_KERNEL);
695 	ssize_t ret;
696 
697 	if (unlikely(!array))
698 		return -ENOMEM;
699 
700 	pipe_lock(pipe);
701 
702 	splice_from_pipe_begin(&sd);
703 	while (sd.total_len) {
704 		struct iov_iter from;
705 		size_t left;
706 		int n, idx;
707 
708 		ret = splice_from_pipe_next(pipe, &sd);
709 		if (ret <= 0)
710 			break;
711 
712 		if (unlikely(nbufs < pipe->buffers)) {
713 			kfree(array);
714 			nbufs = pipe->buffers;
715 			array = kcalloc(nbufs, sizeof(struct bio_vec),
716 					GFP_KERNEL);
717 			if (!array) {
718 				ret = -ENOMEM;
719 				break;
720 			}
721 		}
722 
723 		/* build the vector */
724 		left = sd.total_len;
725 		for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) {
726 			struct pipe_buffer *buf = pipe->bufs + idx;
727 			size_t this_len = buf->len;
728 
729 			if (this_len > left)
730 				this_len = left;
731 
732 			if (idx == pipe->buffers - 1)
733 				idx = -1;
734 
735 			ret = pipe_buf_confirm(pipe, buf);
736 			if (unlikely(ret)) {
737 				if (ret == -ENODATA)
738 					ret = 0;
739 				goto done;
740 			}
741 
742 			array[n].bv_page = buf->page;
743 			array[n].bv_len = this_len;
744 			array[n].bv_offset = buf->offset;
745 			left -= this_len;
746 		}
747 
748 		iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n,
749 			      sd.total_len - left);
750 		ret = vfs_iter_write(out, &from, &sd.pos, 0);
751 		if (ret <= 0)
752 			break;
753 
754 		sd.num_spliced += ret;
755 		sd.total_len -= ret;
756 		*ppos = sd.pos;
757 
758 		/* dismiss the fully eaten buffers, adjust the partial one */
759 		while (ret) {
760 			struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
761 			if (ret >= buf->len) {
762 				ret -= buf->len;
763 				buf->len = 0;
764 				pipe_buf_release(pipe, buf);
765 				pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
766 				pipe->nrbufs--;
767 				if (pipe->files)
768 					sd.need_wakeup = true;
769 			} else {
770 				buf->offset += ret;
771 				buf->len -= ret;
772 				ret = 0;
773 			}
774 		}
775 	}
776 done:
777 	kfree(array);
778 	splice_from_pipe_end(pipe, &sd);
779 
780 	pipe_unlock(pipe);
781 
782 	if (sd.num_spliced)
783 		ret = sd.num_spliced;
784 
785 	return ret;
786 }
787 
788 EXPORT_SYMBOL(iter_file_splice_write);
789 
write_pipe_buf(struct pipe_inode_info * pipe,struct pipe_buffer * buf,struct splice_desc * sd)790 static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
791 			  struct splice_desc *sd)
792 {
793 	int ret;
794 	void *data;
795 	loff_t tmp = sd->pos;
796 
797 	data = kmap(buf->page);
798 	ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
799 	kunmap(buf->page);
800 
801 	return ret;
802 }
803 
default_file_splice_write(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)804 static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
805 					 struct file *out, loff_t *ppos,
806 					 size_t len, unsigned int flags)
807 {
808 	ssize_t ret;
809 
810 	ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
811 	if (ret > 0)
812 		*ppos += ret;
813 
814 	return ret;
815 }
816 
817 /**
818  * generic_splice_sendpage - splice data from a pipe to a socket
819  * @pipe:	pipe to splice from
820  * @out:	socket to write to
821  * @ppos:	position in @out
822  * @len:	number of bytes to splice
823  * @flags:	splice modifier flags
824  *
825  * Description:
826  *    Will send @len bytes from the pipe to a network socket. No data copying
827  *    is involved.
828  *
829  */
generic_splice_sendpage(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)830 ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
831 				loff_t *ppos, size_t len, unsigned int flags)
832 {
833 	return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
834 }
835 
836 EXPORT_SYMBOL(generic_splice_sendpage);
837 
838 /*
839  * Attempt to initiate a splice from pipe to file.
840  */
do_splice_from(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)841 static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
842 			   loff_t *ppos, size_t len, unsigned int flags)
843 {
844 	ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
845 				loff_t *, size_t, unsigned int);
846 
847 	if (out->f_op->splice_write)
848 		splice_write = out->f_op->splice_write;
849 	else
850 		splice_write = default_file_splice_write;
851 
852 	return splice_write(pipe, out, ppos, len, flags);
853 }
854 
855 /*
856  * Attempt to initiate a splice from a file to a pipe.
857  */
do_splice_to(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)858 static long do_splice_to(struct file *in, loff_t *ppos,
859 			 struct pipe_inode_info *pipe, size_t len,
860 			 unsigned int flags)
861 {
862 	ssize_t (*splice_read)(struct file *, loff_t *,
863 			       struct pipe_inode_info *, size_t, unsigned int);
864 	int ret;
865 
866 	if (unlikely(!(in->f_mode & FMODE_READ)))
867 		return -EBADF;
868 
869 	ret = rw_verify_area(READ, in, ppos, len);
870 	if (unlikely(ret < 0))
871 		return ret;
872 
873 	if (unlikely(len > MAX_RW_COUNT))
874 		len = MAX_RW_COUNT;
875 
876 	if (in->f_op->splice_read)
877 		splice_read = in->f_op->splice_read;
878 	else
879 		splice_read = default_file_splice_read;
880 
881 	return splice_read(in, ppos, pipe, len, flags);
882 }
883 
884 /**
885  * splice_direct_to_actor - splices data directly between two non-pipes
886  * @in:		file to splice from
887  * @sd:		actor information on where to splice to
888  * @actor:	handles the data splicing
889  *
890  * Description:
891  *    This is a special case helper to splice directly between two
892  *    points, without requiring an explicit pipe. Internally an allocated
893  *    pipe is cached in the process, and reused during the lifetime of
894  *    that process.
895  *
896  */
splice_direct_to_actor(struct file * in,struct splice_desc * sd,splice_direct_actor * actor)897 ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
898 			       splice_direct_actor *actor)
899 {
900 	struct pipe_inode_info *pipe;
901 	long ret, bytes;
902 	umode_t i_mode;
903 	size_t len;
904 	int i, flags, more;
905 
906 	/*
907 	 * We require the input being a regular file, as we don't want to
908 	 * randomly drop data for eg socket -> socket splicing. Use the
909 	 * piped splicing for that!
910 	 */
911 	i_mode = file_inode(in)->i_mode;
912 	if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
913 		return -EINVAL;
914 
915 	/*
916 	 * neither in nor out is a pipe, setup an internal pipe attached to
917 	 * 'out' and transfer the wanted data from 'in' to 'out' through that
918 	 */
919 	pipe = current->splice_pipe;
920 	if (unlikely(!pipe)) {
921 		pipe = alloc_pipe_info();
922 		if (!pipe)
923 			return -ENOMEM;
924 
925 		/*
926 		 * We don't have an immediate reader, but we'll read the stuff
927 		 * out of the pipe right after the splice_to_pipe(). So set
928 		 * PIPE_READERS appropriately.
929 		 */
930 		pipe->readers = 1;
931 
932 		current->splice_pipe = pipe;
933 	}
934 
935 	/*
936 	 * Do the splice.
937 	 */
938 	ret = 0;
939 	bytes = 0;
940 	len = sd->total_len;
941 	flags = sd->flags;
942 
943 	/*
944 	 * Don't block on output, we have to drain the direct pipe.
945 	 */
946 	sd->flags &= ~SPLICE_F_NONBLOCK;
947 	more = sd->flags & SPLICE_F_MORE;
948 
949 	WARN_ON_ONCE(pipe->nrbufs != 0);
950 
951 	while (len) {
952 		unsigned int pipe_pages;
953 		size_t read_len;
954 		loff_t pos = sd->pos, prev_pos = pos;
955 
956 		/* Don't try to read more the pipe has space for. */
957 		pipe_pages = pipe->buffers - pipe->nrbufs;
958 		read_len = min(len, (size_t)pipe_pages << PAGE_SHIFT);
959 		ret = do_splice_to(in, &pos, pipe, read_len, flags);
960 		if (unlikely(ret <= 0))
961 			goto out_release;
962 
963 		read_len = ret;
964 		sd->total_len = read_len;
965 
966 		/*
967 		 * If more data is pending, set SPLICE_F_MORE
968 		 * If this is the last data and SPLICE_F_MORE was not set
969 		 * initially, clears it.
970 		 */
971 		if (read_len < len)
972 			sd->flags |= SPLICE_F_MORE;
973 		else if (!more)
974 			sd->flags &= ~SPLICE_F_MORE;
975 		/*
976 		 * NOTE: nonblocking mode only applies to the input. We
977 		 * must not do the output in nonblocking mode as then we
978 		 * could get stuck data in the internal pipe:
979 		 */
980 		ret = actor(pipe, sd);
981 		if (unlikely(ret <= 0)) {
982 			sd->pos = prev_pos;
983 			goto out_release;
984 		}
985 
986 		bytes += ret;
987 		len -= ret;
988 		sd->pos = pos;
989 
990 		if (ret < read_len) {
991 			sd->pos = prev_pos + ret;
992 			goto out_release;
993 		}
994 	}
995 
996 done:
997 	pipe->nrbufs = pipe->curbuf = 0;
998 	file_accessed(in);
999 	return bytes;
1000 
1001 out_release:
1002 	/*
1003 	 * If we did an incomplete transfer we must release
1004 	 * the pipe buffers in question:
1005 	 */
1006 	for (i = 0; i < pipe->buffers; i++) {
1007 		struct pipe_buffer *buf = pipe->bufs + i;
1008 
1009 		if (buf->ops)
1010 			pipe_buf_release(pipe, buf);
1011 	}
1012 
1013 	if (!bytes)
1014 		bytes = ret;
1015 
1016 	goto done;
1017 }
1018 EXPORT_SYMBOL(splice_direct_to_actor);
1019 
direct_splice_actor(struct pipe_inode_info * pipe,struct splice_desc * sd)1020 static int direct_splice_actor(struct pipe_inode_info *pipe,
1021 			       struct splice_desc *sd)
1022 {
1023 	struct file *file = sd->u.file;
1024 
1025 	return do_splice_from(pipe, file, sd->opos, sd->total_len,
1026 			      sd->flags);
1027 }
1028 
1029 /**
1030  * do_splice_direct - splices data directly between two files
1031  * @in:		file to splice from
1032  * @ppos:	input file offset
1033  * @out:	file to splice to
1034  * @opos:	output file offset
1035  * @len:	number of bytes to splice
1036  * @flags:	splice modifier flags
1037  *
1038  * Description:
1039  *    For use by do_sendfile(). splice can easily emulate sendfile, but
1040  *    doing it in the application would incur an extra system call
1041  *    (splice in + splice out, as compared to just sendfile()). So this helper
1042  *    can splice directly through a process-private pipe.
1043  *
1044  */
do_splice_direct(struct file * in,loff_t * ppos,struct file * out,loff_t * opos,size_t len,unsigned int flags)1045 long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1046 		      loff_t *opos, size_t len, unsigned int flags)
1047 {
1048 	struct splice_desc sd = {
1049 		.len		= len,
1050 		.total_len	= len,
1051 		.flags		= flags,
1052 		.pos		= *ppos,
1053 		.u.file		= out,
1054 		.opos		= opos,
1055 	};
1056 	long ret;
1057 
1058 	if (unlikely(!(out->f_mode & FMODE_WRITE)))
1059 		return -EBADF;
1060 
1061 	if (unlikely(out->f_flags & O_APPEND))
1062 		return -EINVAL;
1063 
1064 	ret = rw_verify_area(WRITE, out, opos, len);
1065 	if (unlikely(ret < 0))
1066 		return ret;
1067 
1068 	ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
1069 	if (ret > 0)
1070 		*ppos = sd.pos;
1071 
1072 	return ret;
1073 }
1074 EXPORT_SYMBOL(do_splice_direct);
1075 
wait_for_space(struct pipe_inode_info * pipe,unsigned flags)1076 static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags)
1077 {
1078 	for (;;) {
1079 		if (unlikely(!pipe->readers)) {
1080 			send_sig(SIGPIPE, current, 0);
1081 			return -EPIPE;
1082 		}
1083 		if (pipe->nrbufs != pipe->buffers)
1084 			return 0;
1085 		if (flags & SPLICE_F_NONBLOCK)
1086 			return -EAGAIN;
1087 		if (signal_pending(current))
1088 			return -ERESTARTSYS;
1089 		pipe->waiting_writers++;
1090 		pipe_wait(pipe);
1091 		pipe->waiting_writers--;
1092 	}
1093 }
1094 
1095 static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1096 			       struct pipe_inode_info *opipe,
1097 			       size_t len, unsigned int flags);
1098 
1099 /*
1100  * Determine where to splice to/from.
1101  */
do_splice(struct file * in,loff_t __user * off_in,struct file * out,loff_t __user * off_out,size_t len,unsigned int flags)1102 static long do_splice(struct file *in, loff_t __user *off_in,
1103 		      struct file *out, loff_t __user *off_out,
1104 		      size_t len, unsigned int flags)
1105 {
1106 	struct pipe_inode_info *ipipe;
1107 	struct pipe_inode_info *opipe;
1108 	loff_t offset;
1109 	long ret;
1110 
1111 	ipipe = get_pipe_info(in);
1112 	opipe = get_pipe_info(out);
1113 
1114 	if (ipipe && opipe) {
1115 		if (off_in || off_out)
1116 			return -ESPIPE;
1117 
1118 		if (!(in->f_mode & FMODE_READ))
1119 			return -EBADF;
1120 
1121 		if (!(out->f_mode & FMODE_WRITE))
1122 			return -EBADF;
1123 
1124 		/* Splicing to self would be fun, but... */
1125 		if (ipipe == opipe)
1126 			return -EINVAL;
1127 
1128 		return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1129 	}
1130 
1131 	if (ipipe) {
1132 		if (off_in)
1133 			return -ESPIPE;
1134 		if (off_out) {
1135 			if (!(out->f_mode & FMODE_PWRITE))
1136 				return -EINVAL;
1137 			if (copy_from_user(&offset, off_out, sizeof(loff_t)))
1138 				return -EFAULT;
1139 		} else {
1140 			offset = out->f_pos;
1141 		}
1142 
1143 		if (unlikely(!(out->f_mode & FMODE_WRITE)))
1144 			return -EBADF;
1145 
1146 		if (unlikely(out->f_flags & O_APPEND))
1147 			return -EINVAL;
1148 
1149 		ret = rw_verify_area(WRITE, out, &offset, len);
1150 		if (unlikely(ret < 0))
1151 			return ret;
1152 
1153 		file_start_write(out);
1154 		ret = do_splice_from(ipipe, out, &offset, len, flags);
1155 		file_end_write(out);
1156 
1157 		if (!off_out)
1158 			out->f_pos = offset;
1159 		else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
1160 			ret = -EFAULT;
1161 
1162 		return ret;
1163 	}
1164 
1165 	if (opipe) {
1166 		if (off_out)
1167 			return -ESPIPE;
1168 		if (off_in) {
1169 			if (!(in->f_mode & FMODE_PREAD))
1170 				return -EINVAL;
1171 			if (copy_from_user(&offset, off_in, sizeof(loff_t)))
1172 				return -EFAULT;
1173 		} else {
1174 			offset = in->f_pos;
1175 		}
1176 
1177 		pipe_lock(opipe);
1178 		ret = wait_for_space(opipe, flags);
1179 		if (!ret) {
1180 			unsigned int pipe_pages;
1181 
1182 			/* Don't try to read more the pipe has space for. */
1183 			pipe_pages = opipe->buffers - opipe->nrbufs;
1184 			len = min(len, (size_t)pipe_pages << PAGE_SHIFT);
1185 
1186 			ret = do_splice_to(in, &offset, opipe, len, flags);
1187 		}
1188 		pipe_unlock(opipe);
1189 		if (ret > 0)
1190 			wakeup_pipe_readers(opipe);
1191 		if (!off_in)
1192 			in->f_pos = offset;
1193 		else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
1194 			ret = -EFAULT;
1195 
1196 		return ret;
1197 	}
1198 
1199 	return -EINVAL;
1200 }
1201 
iter_to_pipe(struct iov_iter * from,struct pipe_inode_info * pipe,unsigned flags)1202 static int iter_to_pipe(struct iov_iter *from,
1203 			struct pipe_inode_info *pipe,
1204 			unsigned flags)
1205 {
1206 	struct pipe_buffer buf = {
1207 		.ops = &user_page_pipe_buf_ops,
1208 		.flags = flags
1209 	};
1210 	size_t total = 0;
1211 	int ret = 0;
1212 	bool failed = false;
1213 
1214 	while (iov_iter_count(from) && !failed) {
1215 		struct page *pages[16];
1216 		ssize_t copied;
1217 		size_t start;
1218 		int n;
1219 
1220 		copied = iov_iter_get_pages(from, pages, ~0UL, 16, &start);
1221 		if (copied <= 0) {
1222 			ret = copied;
1223 			break;
1224 		}
1225 
1226 		for (n = 0; copied; n++, start = 0) {
1227 			int size = min_t(int, copied, PAGE_SIZE - start);
1228 			if (!failed) {
1229 				buf.page = pages[n];
1230 				buf.offset = start;
1231 				buf.len = size;
1232 				ret = add_to_pipe(pipe, &buf);
1233 				if (unlikely(ret < 0)) {
1234 					failed = true;
1235 				} else {
1236 					iov_iter_advance(from, ret);
1237 					total += ret;
1238 				}
1239 			} else {
1240 				put_page(pages[n]);
1241 			}
1242 			copied -= size;
1243 		}
1244 	}
1245 	return total ? total : ret;
1246 }
1247 
pipe_to_user(struct pipe_inode_info * pipe,struct pipe_buffer * buf,struct splice_desc * sd)1248 static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1249 			struct splice_desc *sd)
1250 {
1251 	int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1252 	return n == sd->len ? n : -EFAULT;
1253 }
1254 
1255 /*
1256  * For lack of a better implementation, implement vmsplice() to userspace
1257  * as a simple copy of the pipes pages to the user iov.
1258  */
vmsplice_to_user(struct file * file,struct iov_iter * iter,unsigned int flags)1259 static long vmsplice_to_user(struct file *file, struct iov_iter *iter,
1260 			     unsigned int flags)
1261 {
1262 	struct pipe_inode_info *pipe = get_pipe_info(file);
1263 	struct splice_desc sd = {
1264 		.total_len = iov_iter_count(iter),
1265 		.flags = flags,
1266 		.u.data = iter
1267 	};
1268 	long ret = 0;
1269 
1270 	if (!pipe)
1271 		return -EBADF;
1272 
1273 	if (sd.total_len) {
1274 		pipe_lock(pipe);
1275 		ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
1276 		pipe_unlock(pipe);
1277 	}
1278 
1279 	return ret;
1280 }
1281 
1282 /*
1283  * vmsplice splices a user address range into a pipe. It can be thought of
1284  * as splice-from-memory, where the regular splice is splice-from-file (or
1285  * to file). In both cases the output is a pipe, naturally.
1286  */
vmsplice_to_pipe(struct file * file,struct iov_iter * iter,unsigned int flags)1287 static long vmsplice_to_pipe(struct file *file, struct iov_iter *iter,
1288 			     unsigned int flags)
1289 {
1290 	struct pipe_inode_info *pipe;
1291 	long ret = 0;
1292 	unsigned buf_flag = 0;
1293 
1294 	if (flags & SPLICE_F_GIFT)
1295 		buf_flag = PIPE_BUF_FLAG_GIFT;
1296 
1297 	pipe = get_pipe_info(file);
1298 	if (!pipe)
1299 		return -EBADF;
1300 
1301 	pipe_lock(pipe);
1302 	ret = wait_for_space(pipe, flags);
1303 	if (!ret)
1304 		ret = iter_to_pipe(iter, pipe, buf_flag);
1305 	pipe_unlock(pipe);
1306 	if (ret > 0)
1307 		wakeup_pipe_readers(pipe);
1308 	return ret;
1309 }
1310 
vmsplice_type(struct fd f,int * type)1311 static int vmsplice_type(struct fd f, int *type)
1312 {
1313 	if (!f.file)
1314 		return -EBADF;
1315 	if (f.file->f_mode & FMODE_WRITE) {
1316 		*type = WRITE;
1317 	} else if (f.file->f_mode & FMODE_READ) {
1318 		*type = READ;
1319 	} else {
1320 		fdput(f);
1321 		return -EBADF;
1322 	}
1323 	return 0;
1324 }
1325 
1326 /*
1327  * Note that vmsplice only really supports true splicing _from_ user memory
1328  * to a pipe, not the other way around. Splicing from user memory is a simple
1329  * operation that can be supported without any funky alignment restrictions
1330  * or nasty vm tricks. We simply map in the user memory and fill them into
1331  * a pipe. The reverse isn't quite as easy, though. There are two possible
1332  * solutions for that:
1333  *
1334  *	- memcpy() the data internally, at which point we might as well just
1335  *	  do a regular read() on the buffer anyway.
1336  *	- Lots of nasty vm tricks, that are neither fast nor flexible (it
1337  *	  has restriction limitations on both ends of the pipe).
1338  *
1339  * Currently we punt and implement it as a normal copy, see pipe_to_user().
1340  *
1341  */
do_vmsplice(struct file * f,struct iov_iter * iter,unsigned int flags)1342 static long do_vmsplice(struct file *f, struct iov_iter *iter, unsigned int flags)
1343 {
1344 	if (unlikely(flags & ~SPLICE_F_ALL))
1345 		return -EINVAL;
1346 
1347 	if (!iov_iter_count(iter))
1348 		return 0;
1349 
1350 	if (iov_iter_rw(iter) == WRITE)
1351 		return vmsplice_to_pipe(f, iter, flags);
1352 	else
1353 		return vmsplice_to_user(f, iter, flags);
1354 }
1355 
SYSCALL_DEFINE4(vmsplice,int,fd,const struct iovec __user *,uiov,unsigned long,nr_segs,unsigned int,flags)1356 SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, uiov,
1357 		unsigned long, nr_segs, unsigned int, flags)
1358 {
1359 	struct iovec iovstack[UIO_FASTIOV];
1360 	struct iovec *iov = iovstack;
1361 	struct iov_iter iter;
1362 	long error;
1363 	struct fd f;
1364 	int type;
1365 
1366 	f = fdget(fd);
1367 	error = vmsplice_type(f, &type);
1368 	if (error)
1369 		return error;
1370 
1371 	error = import_iovec(type, uiov, nr_segs,
1372 			     ARRAY_SIZE(iovstack), &iov, &iter);
1373 	if (!error) {
1374 		error = do_vmsplice(f.file, &iter, flags);
1375 		kfree(iov);
1376 	}
1377 	fdput(f);
1378 	return error;
1379 }
1380 
1381 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(vmsplice,int,fd,const struct compat_iovec __user *,iov32,unsigned int,nr_segs,unsigned int,flags)1382 COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1383 		    unsigned int, nr_segs, unsigned int, flags)
1384 {
1385 	struct iovec iovstack[UIO_FASTIOV];
1386 	struct iovec *iov = iovstack;
1387 	struct iov_iter iter;
1388 	long error;
1389 	struct fd f;
1390 	int type;
1391 
1392 	f = fdget(fd);
1393 	error = vmsplice_type(f, &type);
1394 	if (error)
1395 		return error;
1396 
1397 	error = compat_import_iovec(type, iov32, nr_segs,
1398 			     ARRAY_SIZE(iovstack), &iov, &iter);
1399 	if (!error) {
1400 		error = do_vmsplice(f.file, &iter, flags);
1401 		kfree(iov);
1402 	}
1403 	fdput(f);
1404 	return error;
1405 }
1406 #endif
1407 
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)1408 SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1409 		int, fd_out, loff_t __user *, off_out,
1410 		size_t, len, unsigned int, flags)
1411 {
1412 	struct fd in, out;
1413 	long error;
1414 
1415 	if (unlikely(!len))
1416 		return 0;
1417 
1418 	if (unlikely(flags & ~SPLICE_F_ALL))
1419 		return -EINVAL;
1420 
1421 	error = -EBADF;
1422 	in = fdget(fd_in);
1423 	if (in.file) {
1424 		if (in.file->f_mode & FMODE_READ) {
1425 			out = fdget(fd_out);
1426 			if (out.file) {
1427 				if (out.file->f_mode & FMODE_WRITE)
1428 					error = do_splice(in.file, off_in,
1429 							  out.file, off_out,
1430 							  len, flags);
1431 				fdput(out);
1432 			}
1433 		}
1434 		fdput(in);
1435 	}
1436 	return error;
1437 }
1438 
1439 /*
1440  * Make sure there's data to read. Wait for input if we can, otherwise
1441  * return an appropriate error.
1442  */
ipipe_prep(struct pipe_inode_info * pipe,unsigned int flags)1443 static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1444 {
1445 	int ret;
1446 
1447 	/*
1448 	 * Check ->nrbufs without the inode lock first. This function
1449 	 * is speculative anyways, so missing one is ok.
1450 	 */
1451 	if (pipe->nrbufs)
1452 		return 0;
1453 
1454 	ret = 0;
1455 	pipe_lock(pipe);
1456 
1457 	while (!pipe->nrbufs) {
1458 		if (signal_pending(current)) {
1459 			ret = -ERESTARTSYS;
1460 			break;
1461 		}
1462 		if (!pipe->writers)
1463 			break;
1464 		if (!pipe->waiting_writers) {
1465 			if (flags & SPLICE_F_NONBLOCK) {
1466 				ret = -EAGAIN;
1467 				break;
1468 			}
1469 		}
1470 		pipe_wait(pipe);
1471 	}
1472 
1473 	pipe_unlock(pipe);
1474 	return ret;
1475 }
1476 
1477 /*
1478  * Make sure there's writeable room. Wait for room if we can, otherwise
1479  * return an appropriate error.
1480  */
opipe_prep(struct pipe_inode_info * pipe,unsigned int flags)1481 static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1482 {
1483 	int ret;
1484 
1485 	/*
1486 	 * Check ->nrbufs without the inode lock first. This function
1487 	 * is speculative anyways, so missing one is ok.
1488 	 */
1489 	if (pipe->nrbufs < pipe->buffers)
1490 		return 0;
1491 
1492 	ret = 0;
1493 	pipe_lock(pipe);
1494 
1495 	while (pipe->nrbufs >= pipe->buffers) {
1496 		if (!pipe->readers) {
1497 			send_sig(SIGPIPE, current, 0);
1498 			ret = -EPIPE;
1499 			break;
1500 		}
1501 		if (flags & SPLICE_F_NONBLOCK) {
1502 			ret = -EAGAIN;
1503 			break;
1504 		}
1505 		if (signal_pending(current)) {
1506 			ret = -ERESTARTSYS;
1507 			break;
1508 		}
1509 		pipe->waiting_writers++;
1510 		pipe_wait(pipe);
1511 		pipe->waiting_writers--;
1512 	}
1513 
1514 	pipe_unlock(pipe);
1515 	return ret;
1516 }
1517 
1518 /*
1519  * Splice contents of ipipe to opipe.
1520  */
splice_pipe_to_pipe(struct pipe_inode_info * ipipe,struct pipe_inode_info * opipe,size_t len,unsigned int flags)1521 static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1522 			       struct pipe_inode_info *opipe,
1523 			       size_t len, unsigned int flags)
1524 {
1525 	struct pipe_buffer *ibuf, *obuf;
1526 	int ret = 0, nbuf;
1527 	bool input_wakeup = false;
1528 
1529 
1530 retry:
1531 	ret = ipipe_prep(ipipe, flags);
1532 	if (ret)
1533 		return ret;
1534 
1535 	ret = opipe_prep(opipe, flags);
1536 	if (ret)
1537 		return ret;
1538 
1539 	/*
1540 	 * Potential ABBA deadlock, work around it by ordering lock
1541 	 * grabbing by pipe info address. Otherwise two different processes
1542 	 * could deadlock (one doing tee from A -> B, the other from B -> A).
1543 	 */
1544 	pipe_double_lock(ipipe, opipe);
1545 
1546 	do {
1547 		if (!opipe->readers) {
1548 			send_sig(SIGPIPE, current, 0);
1549 			if (!ret)
1550 				ret = -EPIPE;
1551 			break;
1552 		}
1553 
1554 		if (!ipipe->nrbufs && !ipipe->writers)
1555 			break;
1556 
1557 		/*
1558 		 * Cannot make any progress, because either the input
1559 		 * pipe is empty or the output pipe is full.
1560 		 */
1561 		if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
1562 			/* Already processed some buffers, break */
1563 			if (ret)
1564 				break;
1565 
1566 			if (flags & SPLICE_F_NONBLOCK) {
1567 				ret = -EAGAIN;
1568 				break;
1569 			}
1570 
1571 			/*
1572 			 * We raced with another reader/writer and haven't
1573 			 * managed to process any buffers.  A zero return
1574 			 * value means EOF, so retry instead.
1575 			 */
1576 			pipe_unlock(ipipe);
1577 			pipe_unlock(opipe);
1578 			goto retry;
1579 		}
1580 
1581 		ibuf = ipipe->bufs + ipipe->curbuf;
1582 		nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
1583 		obuf = opipe->bufs + nbuf;
1584 
1585 		if (len >= ibuf->len) {
1586 			/*
1587 			 * Simply move the whole buffer from ipipe to opipe
1588 			 */
1589 			*obuf = *ibuf;
1590 			ibuf->ops = NULL;
1591 			opipe->nrbufs++;
1592 			ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
1593 			ipipe->nrbufs--;
1594 			input_wakeup = true;
1595 		} else {
1596 			/*
1597 			 * Get a reference to this pipe buffer,
1598 			 * so we can copy the contents over.
1599 			 */
1600 			if (!pipe_buf_get(ipipe, ibuf)) {
1601 				if (ret == 0)
1602 					ret = -EFAULT;
1603 				break;
1604 			}
1605 			*obuf = *ibuf;
1606 
1607 			/*
1608 			 * Don't inherit the gift flag, we need to
1609 			 * prevent multiple steals of this page.
1610 			 */
1611 			obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1612 
1613 			pipe_buf_mark_unmergeable(obuf);
1614 
1615 			obuf->len = len;
1616 			opipe->nrbufs++;
1617 			ibuf->offset += obuf->len;
1618 			ibuf->len -= obuf->len;
1619 		}
1620 		ret += obuf->len;
1621 		len -= obuf->len;
1622 	} while (len);
1623 
1624 	pipe_unlock(ipipe);
1625 	pipe_unlock(opipe);
1626 
1627 	/*
1628 	 * If we put data in the output pipe, wakeup any potential readers.
1629 	 */
1630 	if (ret > 0)
1631 		wakeup_pipe_readers(opipe);
1632 
1633 	if (input_wakeup)
1634 		wakeup_pipe_writers(ipipe);
1635 
1636 	return ret;
1637 }
1638 
1639 /*
1640  * Link contents of ipipe to opipe.
1641  */
link_pipe(struct pipe_inode_info * ipipe,struct pipe_inode_info * opipe,size_t len,unsigned int flags)1642 static int link_pipe(struct pipe_inode_info *ipipe,
1643 		     struct pipe_inode_info *opipe,
1644 		     size_t len, unsigned int flags)
1645 {
1646 	struct pipe_buffer *ibuf, *obuf;
1647 	int ret = 0, i = 0, nbuf;
1648 
1649 	/*
1650 	 * Potential ABBA deadlock, work around it by ordering lock
1651 	 * grabbing by pipe info address. Otherwise two different processes
1652 	 * could deadlock (one doing tee from A -> B, the other from B -> A).
1653 	 */
1654 	pipe_double_lock(ipipe, opipe);
1655 
1656 	do {
1657 		if (!opipe->readers) {
1658 			send_sig(SIGPIPE, current, 0);
1659 			if (!ret)
1660 				ret = -EPIPE;
1661 			break;
1662 		}
1663 
1664 		/*
1665 		 * If we have iterated all input buffers or ran out of
1666 		 * output room, break.
1667 		 */
1668 		if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
1669 			break;
1670 
1671 		ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1672 		nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
1673 
1674 		/*
1675 		 * Get a reference to this pipe buffer,
1676 		 * so we can copy the contents over.
1677 		 */
1678 		if (!pipe_buf_get(ipipe, ibuf)) {
1679 			if (ret == 0)
1680 				ret = -EFAULT;
1681 			break;
1682 		}
1683 
1684 		obuf = opipe->bufs + nbuf;
1685 		*obuf = *ibuf;
1686 
1687 		/*
1688 		 * Don't inherit the gift flag, we need to
1689 		 * prevent multiple steals of this page.
1690 		 */
1691 		obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1692 
1693 		pipe_buf_mark_unmergeable(obuf);
1694 
1695 		if (obuf->len > len)
1696 			obuf->len = len;
1697 
1698 		opipe->nrbufs++;
1699 		ret += obuf->len;
1700 		len -= obuf->len;
1701 		i++;
1702 	} while (len);
1703 
1704 	/*
1705 	 * return EAGAIN if we have the potential of some data in the
1706 	 * future, otherwise just return 0
1707 	 */
1708 	if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1709 		ret = -EAGAIN;
1710 
1711 	pipe_unlock(ipipe);
1712 	pipe_unlock(opipe);
1713 
1714 	/*
1715 	 * If we put data in the output pipe, wakeup any potential readers.
1716 	 */
1717 	if (ret > 0)
1718 		wakeup_pipe_readers(opipe);
1719 
1720 	return ret;
1721 }
1722 
1723 /*
1724  * This is a tee(1) implementation that works on pipes. It doesn't copy
1725  * any data, it simply references the 'in' pages on the 'out' pipe.
1726  * The 'flags' used are the SPLICE_F_* variants, currently the only
1727  * applicable one is SPLICE_F_NONBLOCK.
1728  */
do_tee(struct file * in,struct file * out,size_t len,unsigned int flags)1729 static long do_tee(struct file *in, struct file *out, size_t len,
1730 		   unsigned int flags)
1731 {
1732 	struct pipe_inode_info *ipipe = get_pipe_info(in);
1733 	struct pipe_inode_info *opipe = get_pipe_info(out);
1734 	int ret = -EINVAL;
1735 
1736 	/*
1737 	 * Duplicate the contents of ipipe to opipe without actually
1738 	 * copying the data.
1739 	 */
1740 	if (ipipe && opipe && ipipe != opipe) {
1741 		/*
1742 		 * Keep going, unless we encounter an error. The ipipe/opipe
1743 		 * ordering doesn't really matter.
1744 		 */
1745 		ret = ipipe_prep(ipipe, flags);
1746 		if (!ret) {
1747 			ret = opipe_prep(opipe, flags);
1748 			if (!ret)
1749 				ret = link_pipe(ipipe, opipe, len, flags);
1750 		}
1751 	}
1752 
1753 	return ret;
1754 }
1755 
SYSCALL_DEFINE4(tee,int,fdin,int,fdout,size_t,len,unsigned int,flags)1756 SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
1757 {
1758 	struct fd in;
1759 	int error;
1760 
1761 	if (unlikely(flags & ~SPLICE_F_ALL))
1762 		return -EINVAL;
1763 
1764 	if (unlikely(!len))
1765 		return 0;
1766 
1767 	error = -EBADF;
1768 	in = fdget(fdin);
1769 	if (in.file) {
1770 		if (in.file->f_mode & FMODE_READ) {
1771 			struct fd out = fdget(fdout);
1772 			if (out.file) {
1773 				if (out.file->f_mode & FMODE_WRITE)
1774 					error = do_tee(in.file, out.file,
1775 							len, flags);
1776 				fdput(out);
1777 			}
1778 		}
1779  		fdput(in);
1780  	}
1781 
1782 	return error;
1783 }
1784