]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/splice.c
tracing: Make tracepoint_printk a static_key
[mirror_ubuntu-bionic-kernel.git] / fs / splice.c
CommitLineData
5274f052
JA
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
c2058e06
JA
12 * Jens to support splicing to files, network, direct splicing, etc and
13 * fixing lots of bugs.
5274f052 14 *
0fe23479 15 * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk>
c2058e06
JA
16 * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
17 * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
5274f052
JA
18 *
19 */
20#include <linux/fs.h>
21#include <linux/file.h>
22#include <linux/pagemap.h>
d6b29d7c 23#include <linux/splice.h>
08e552c6 24#include <linux/memcontrol.h>
5274f052 25#include <linux/mm_inline.h>
5abc97aa 26#include <linux/swap.h>
4f6f0bd2 27#include <linux/writeback.h>
630d9c47 28#include <linux/export.h>
4f6f0bd2 29#include <linux/syscalls.h>
912d35f8 30#include <linux/uio.h>
29ce2058 31#include <linux/security.h>
5a0e3ad6 32#include <linux/gfp.h>
35f9c09f 33#include <linux/socket.h>
76b021d0 34#include <linux/compat.h>
06ae43f3 35#include "internal.h"
5274f052 36
83f9135b
JA
37/*
38 * Attempt to steal a page from a pipe buffer. This should perhaps go into
39 * a vm helper function, it's already simplified quite a bit by the
40 * addition of remove_mapping(). If success is returned, the caller may
41 * attempt to reuse this page for another destination.
42 */
76ad4d11 43static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
5abc97aa
JA
44 struct pipe_buffer *buf)
45{
46 struct page *page = buf->page;
9e94cd4f 47 struct address_space *mapping;
5abc97aa 48
9e0267c2
JA
49 lock_page(page);
50
9e94cd4f
JA
51 mapping = page_mapping(page);
52 if (mapping) {
53 WARN_ON(!PageUptodate(page));
5abc97aa 54
9e94cd4f
JA
55 /*
56 * At least for ext2 with nobh option, we need to wait on
57 * writeback completing on this page, since we'll remove it
58 * from the pagecache. Otherwise truncate wont wait on the
59 * page, allowing the disk blocks to be reused by someone else
60 * before we actually wrote our data to them. fs corruption
61 * ensues.
62 */
63 wait_on_page_writeback(page);
ad8d6f0a 64
266cf658
DH
65 if (page_has_private(page) &&
66 !try_to_release_page(page, GFP_KERNEL))
ca39d651 67 goto out_unlock;
4f6f0bd2 68
9e94cd4f
JA
69 /*
70 * If we succeeded in removing the mapping, set LRU flag
71 * and return good.
72 */
73 if (remove_mapping(mapping, page)) {
74 buf->flags |= PIPE_BUF_FLAG_LRU;
75 return 0;
76 }
9e0267c2 77 }
5abc97aa 78
9e94cd4f
JA
79 /*
80 * Raced with truncate or failed to remove page from current
81 * address space, unlock and return failure.
82 */
ca39d651 83out_unlock:
9e94cd4f
JA
84 unlock_page(page);
85 return 1;
5abc97aa
JA
86}
87
76ad4d11 88static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
5274f052
JA
89 struct pipe_buffer *buf)
90{
09cbfeaf 91 put_page(buf->page);
1432873a 92 buf->flags &= ~PIPE_BUF_FLAG_LRU;
5274f052
JA
93}
94
0845718d
JA
95/*
96 * Check whether the contents of buf is OK to access. Since the content
97 * is a page cache page, IO may be in flight.
98 */
cac36bb0
JA
99static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
100 struct pipe_buffer *buf)
5274f052
JA
101{
102 struct page *page = buf->page;
49d0b21b 103 int err;
5274f052
JA
104
105 if (!PageUptodate(page)) {
49d0b21b
JA
106 lock_page(page);
107
108 /*
109 * Page got truncated/unhashed. This will cause a 0-byte
73d62d83 110 * splice, if this is the first page.
49d0b21b
JA
111 */
112 if (!page->mapping) {
113 err = -ENODATA;
114 goto error;
115 }
5274f052 116
49d0b21b 117 /*
73d62d83 118 * Uh oh, read-error from disk.
49d0b21b
JA
119 */
120 if (!PageUptodate(page)) {
121 err = -EIO;
122 goto error;
123 }
124
125 /*
f84d7519 126 * Page is ok afterall, we are done.
49d0b21b 127 */
5274f052 128 unlock_page(page);
5274f052
JA
129 }
130
f84d7519 131 return 0;
49d0b21b
JA
132error:
133 unlock_page(page);
f84d7519 134 return err;
70524490
JA
135}
136
708e3508 137const struct pipe_buf_operations page_cache_pipe_buf_ops = {
5274f052 138 .can_merge = 0,
cac36bb0 139 .confirm = page_cache_pipe_buf_confirm,
5274f052 140 .release = page_cache_pipe_buf_release,
5abc97aa 141 .steal = page_cache_pipe_buf_steal,
f84d7519 142 .get = generic_pipe_buf_get,
5274f052
JA
143};
144
912d35f8
JA
145static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
146 struct pipe_buffer *buf)
147{
7afa6fd0
JA
148 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
149 return 1;
150
1432873a 151 buf->flags |= PIPE_BUF_FLAG_LRU;
330ab716 152 return generic_pipe_buf_steal(pipe, buf);
912d35f8
JA
153}
154
d4c3cca9 155static const struct pipe_buf_operations user_page_pipe_buf_ops = {
912d35f8 156 .can_merge = 0,
cac36bb0 157 .confirm = generic_pipe_buf_confirm,
912d35f8
JA
158 .release = page_cache_pipe_buf_release,
159 .steal = user_page_pipe_buf_steal,
f84d7519 160 .get = generic_pipe_buf_get,
912d35f8
JA
161};
162
825cdcb1
NK
163static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
164{
165 smp_mb();
166 if (waitqueue_active(&pipe->wait))
167 wake_up_interruptible(&pipe->wait);
168 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
169}
170
932cc6d4
JA
171/**
172 * splice_to_pipe - fill passed data into a pipe
173 * @pipe: pipe to fill
174 * @spd: data to fill
175 *
176 * Description:
79685b8d 177 * @spd contains a map of pages and len/offset tuples, along with
932cc6d4
JA
178 * the struct pipe_buf_operations associated with these pages. This
179 * function will link that data to the pipe.
180 *
83f9135b 181 */
d6b29d7c
JA
182ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
183 struct splice_pipe_desc *spd)
5274f052 184{
00de00bd 185 unsigned int spd_pages = spd->nr_pages;
8924feff 186 int ret = 0, page_nr = 0;
5274f052 187
d6785d91
RV
188 if (!spd_pages)
189 return 0;
190
8924feff
AV
191 if (unlikely(!pipe->readers)) {
192 send_sig(SIGPIPE, current, 0);
193 ret = -EPIPE;
194 goto out;
195 }
5274f052 196
8924feff
AV
197 while (pipe->nrbufs < pipe->buffers) {
198 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
199 struct pipe_buffer *buf = pipe->bufs + newbuf;
5274f052 200
8924feff
AV
201 buf->page = spd->pages[page_nr];
202 buf->offset = spd->partial[page_nr].offset;
203 buf->len = spd->partial[page_nr].len;
204 buf->private = spd->partial[page_nr].private;
205 buf->ops = spd->ops;
5274f052 206
8924feff
AV
207 pipe->nrbufs++;
208 page_nr++;
209 ret += buf->len;
29e35094 210
8924feff 211 if (!--spd->nr_pages)
5274f052 212 break;
5274f052
JA
213 }
214
8924feff
AV
215 if (!ret)
216 ret = -EAGAIN;
5274f052 217
8924feff 218out:
00de00bd 219 while (page_nr < spd_pages)
bbdfc2f7 220 spd->spd_release(spd, page_nr++);
5274f052
JA
221
222 return ret;
223}
2b514574 224EXPORT_SYMBOL_GPL(splice_to_pipe);
5274f052 225
79fddc4e
AV
226ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
227{
228 int ret;
229
230 if (unlikely(!pipe->readers)) {
231 send_sig(SIGPIPE, current, 0);
232 ret = -EPIPE;
233 } else if (pipe->nrbufs == pipe->buffers) {
234 ret = -EAGAIN;
235 } else {
236 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
237 pipe->bufs[newbuf] = *buf;
238 pipe->nrbufs++;
239 return buf->len;
240 }
a779638c 241 pipe_buf_release(pipe, buf);
79fddc4e
AV
242 return ret;
243}
244EXPORT_SYMBOL(add_to_pipe);
245
708e3508 246void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
bbdfc2f7 247{
09cbfeaf 248 put_page(spd->pages[i]);
bbdfc2f7
JA
249}
250
35f3d14d
JA
251/*
252 * Check if we need to grow the arrays holding pages and partial page
253 * descriptions.
254 */
047fe360 255int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
35f3d14d 256{
047fe360
ED
257 unsigned int buffers = ACCESS_ONCE(pipe->buffers);
258
259 spd->nr_pages_max = buffers;
260 if (buffers <= PIPE_DEF_BUFFERS)
35f3d14d
JA
261 return 0;
262
047fe360
ED
263 spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL);
264 spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL);
35f3d14d
JA
265
266 if (spd->pages && spd->partial)
267 return 0;
268
269 kfree(spd->pages);
270 kfree(spd->partial);
271 return -ENOMEM;
272}
273
047fe360 274void splice_shrink_spd(struct splice_pipe_desc *spd)
35f3d14d 275{
047fe360 276 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
35f3d14d
JA
277 return;
278
279 kfree(spd->pages);
280 kfree(spd->partial);
281}
282
83f9135b
JA
283/**
284 * generic_file_splice_read - splice data from file to a pipe
285 * @in: file to splice from
932cc6d4 286 * @ppos: position in @in
83f9135b
JA
287 * @pipe: pipe to splice to
288 * @len: number of bytes to splice
289 * @flags: splice modifier flags
290 *
932cc6d4
JA
291 * Description:
292 * Will read pages from given file and fill them into a pipe. Can be
82c156f8 293 * used as long as it has more or less sane ->read_iter().
932cc6d4 294 *
83f9135b 295 */
cbb7e577
JA
296ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
297 struct pipe_inode_info *pipe, size_t len,
298 unsigned int flags)
5274f052 299{
82c156f8
AV
300 struct iov_iter to;
301 struct kiocb kiocb;
82c156f8 302 int idx, ret;
be64f884 303
82c156f8
AV
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 = in->f_op->read_iter(&kiocb, &to);
723590ed 309 if (ret > 0) {
82c156f8 310 *ppos = kiocb.ki_pos;
723590ed 311 file_accessed(in);
82c156f8 312 } else if (ret < 0) {
c3a69024
AV
313 to.idx = idx;
314 to.iov_offset = 0;
315 iov_iter_advance(&to, 0); /* to free what was emitted */
82c156f8
AV
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;
723590ed 322 }
5274f052
JA
323
324 return ret;
325}
059a8f37
JA
326EXPORT_SYMBOL(generic_file_splice_read);
327
241699cd 328const struct pipe_buf_operations default_pipe_buf_ops = {
6818173b 329 .can_merge = 0,
6818173b
MS
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
28a625cb
MS
336static 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. */
343const struct pipe_buf_operations nosteal_pipe_buf_ops = {
344 .can_merge = 0,
28a625cb
MS
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};
350EXPORT_SYMBOL(nosteal_pipe_buf_ops);
351
523ac9af 352static ssize_t kernel_readv(struct file *file, const struct kvec *vec,
6818173b
MS
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() */
793b80ef 362 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos, 0);
6818173b
MS
363 set_fs(old_fs);
364
365 return res;
366}
367
7bb307e8 368ssize_t kernel_write(struct file *file, const char *buf, size_t count,
b2858d7d 369 loff_t pos)
0b0a47f5
MS
370{
371 mm_segment_t old_fs;
372 ssize_t res;
373
374 old_fs = get_fs();
375 set_fs(get_ds());
376 /* The cast to a user pointer is valid due to the set_fs() */
7bb307e8 377 res = vfs_write(file, (__force const char __user *)buf, count, &pos);
0b0a47f5
MS
378 set_fs(old_fs);
379
380 return res;
381}
7bb307e8 382EXPORT_SYMBOL(kernel_write);
0b0a47f5 383
82c156f8 384static ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
6818173b
MS
385 struct pipe_inode_info *pipe, size_t len,
386 unsigned int flags)
387{
523ac9af
AV
388 struct kvec *vec, __vec[PIPE_DEF_BUFFERS];
389 struct iov_iter to;
390 struct page **pages;
6818173b 391 unsigned int nr_pages;
523ac9af 392 size_t offset, dummy, copied = 0;
6818173b 393 ssize_t res;
6818173b 394 int i;
35f3d14d 395
523ac9af
AV
396 if (pipe->nrbufs == pipe->buffers)
397 return -EAGAIN;
35f3d14d 398
523ac9af
AV
399 /*
400 * Try to keep page boundaries matching to source pagecache ones -
401 * it probably won't be much help, but...
402 */
09cbfeaf 403 offset = *ppos & ~PAGE_MASK;
6818173b 404
523ac9af 405 iov_iter_pipe(&to, ITER_PIPE | READ, pipe, len + offset);
6818173b 406
523ac9af
AV
407 res = iov_iter_get_pages_alloc(&to, &pages, len + offset, &dummy);
408 if (res <= 0)
409 return -ENOMEM;
6818173b 410
523ac9af 411 nr_pages = res / PAGE_SIZE;
6818173b 412
523ac9af
AV
413 vec = __vec;
414 if (nr_pages > PIPE_DEF_BUFFERS) {
415 vec = kmalloc(nr_pages * sizeof(struct kvec), GFP_KERNEL);
416 if (unlikely(!vec)) {
417 res = -ENOMEM;
418 goto out;
419 }
77f6bf57 420 }
6818173b 421
523ac9af
AV
422 pipe->bufs[to.idx].offset = offset;
423 pipe->bufs[to.idx].len -= offset;
424
425 for (i = 0; i < nr_pages; i++) {
426 size_t this_len = min_t(size_t, len, PAGE_SIZE - offset);
427 vec[i].iov_base = page_address(pages[i]) + offset;
428 vec[i].iov_len = this_len;
429 len -= this_len;
430 offset = 0;
6818173b 431 }
6818173b 432
523ac9af
AV
433 res = kernel_readv(in, vec, nr_pages, *ppos);
434 if (res > 0) {
435 copied = res;
6818173b 436 *ppos += res;
523ac9af 437 }
6818173b 438
35f3d14d
JA
439 if (vec != __vec)
440 kfree(vec);
523ac9af
AV
441out:
442 for (i = 0; i < nr_pages; i++)
443 put_page(pages[i]);
444 kvfree(pages);
445 iov_iter_advance(&to, copied); /* truncates and discards */
6818173b 446 return res;
6818173b 447}
6818173b 448
5274f052 449/*
4f6f0bd2 450 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
016b661e 451 * using sendpage(). Return the number of bytes sent.
5274f052 452 */
76ad4d11 453static int pipe_to_sendpage(struct pipe_inode_info *pipe,
5274f052
JA
454 struct pipe_buffer *buf, struct splice_desc *sd)
455{
6a14b90b 456 struct file *file = sd->u.file;
5274f052 457 loff_t pos = sd->pos;
a8adbe37 458 int more;
5274f052 459
72c2d531 460 if (!likely(file->f_op->sendpage))
a8adbe37
MM
461 return -EINVAL;
462
35f9c09f 463 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
ae62ca7b
ED
464
465 if (sd->len < sd->total_len && pipe->nrbufs > 1)
35f9c09f 466 more |= MSG_SENDPAGE_NOTLAST;
ae62ca7b 467
a8adbe37
MM
468 return file->f_op->sendpage(file, buf->page, buf->offset,
469 sd->len, &pos, more);
5274f052
JA
470}
471
b3c2d2dd
MS
472static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
473{
474 smp_mb();
475 if (waitqueue_active(&pipe->wait))
476 wake_up_interruptible(&pipe->wait);
477 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
478}
479
932cc6d4 480/**
b3c2d2dd 481 * splice_from_pipe_feed - feed available data from a pipe to a file
932cc6d4
JA
482 * @pipe: pipe to splice from
483 * @sd: information to @actor
484 * @actor: handler that splices the data
485 *
486 * Description:
b3c2d2dd
MS
487 * This function loops over the pipe and calls @actor to do the
488 * actual moving of a single struct pipe_buffer to the desired
489 * destination. It returns when there's no more buffers left in
490 * the pipe or if the requested number of bytes (@sd->total_len)
491 * have been copied. It returns a positive number (one) if the
492 * pipe needs to be filled with more data, zero if the required
493 * number of bytes have been copied and -errno on error.
932cc6d4 494 *
b3c2d2dd
MS
495 * This, together with splice_from_pipe_{begin,end,next}, may be
496 * used to implement the functionality of __splice_from_pipe() when
497 * locking is required around copying the pipe buffers to the
498 * destination.
83f9135b 499 */
96f9bc8f 500static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
b3c2d2dd 501 splice_actor *actor)
5274f052 502{
b3c2d2dd 503 int ret;
5274f052 504
b3c2d2dd
MS
505 while (pipe->nrbufs) {
506 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
5274f052 507
b3c2d2dd
MS
508 sd->len = buf->len;
509 if (sd->len > sd->total_len)
510 sd->len = sd->total_len;
5274f052 511
fba597db 512 ret = pipe_buf_confirm(pipe, buf);
a8adbe37 513 if (unlikely(ret)) {
b3c2d2dd
MS
514 if (ret == -ENODATA)
515 ret = 0;
516 return ret;
517 }
a8adbe37
MM
518
519 ret = actor(pipe, buf, sd);
520 if (ret <= 0)
521 return ret;
522
b3c2d2dd
MS
523 buf->offset += ret;
524 buf->len -= ret;
525
526 sd->num_spliced += ret;
527 sd->len -= ret;
528 sd->pos += ret;
529 sd->total_len -= ret;
530
531 if (!buf->len) {
a779638c 532 pipe_buf_release(pipe, buf);
35f3d14d 533 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
b3c2d2dd 534 pipe->nrbufs--;
6447a3cf 535 if (pipe->files)
b3c2d2dd
MS
536 sd->need_wakeup = true;
537 }
5274f052 538
b3c2d2dd
MS
539 if (!sd->total_len)
540 return 0;
541 }
5274f052 542
b3c2d2dd
MS
543 return 1;
544}
5274f052 545
b3c2d2dd
MS
546/**
547 * splice_from_pipe_next - wait for some data to splice from
548 * @pipe: pipe to splice from
549 * @sd: information about the splice operation
550 *
551 * Description:
552 * This function will wait for some data and return a positive
553 * value (one) if pipe buffers are available. It will return zero
554 * or -errno if no more data needs to be spliced.
555 */
96f9bc8f 556static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
b3c2d2dd 557{
c725bfce
JK
558 /*
559 * Check for signal early to make process killable when there are
560 * always buffers available
561 */
562 if (signal_pending(current))
563 return -ERESTARTSYS;
564
b3c2d2dd
MS
565 while (!pipe->nrbufs) {
566 if (!pipe->writers)
567 return 0;
016b661e 568
b3c2d2dd
MS
569 if (!pipe->waiting_writers && sd->num_spliced)
570 return 0;
73d62d83 571
b3c2d2dd
MS
572 if (sd->flags & SPLICE_F_NONBLOCK)
573 return -EAGAIN;
5274f052 574
b3c2d2dd
MS
575 if (signal_pending(current))
576 return -ERESTARTSYS;
5274f052 577
b3c2d2dd
MS
578 if (sd->need_wakeup) {
579 wakeup_pipe_writers(pipe);
580 sd->need_wakeup = false;
5274f052
JA
581 }
582
b3c2d2dd
MS
583 pipe_wait(pipe);
584 }
29e35094 585
b3c2d2dd
MS
586 return 1;
587}
5274f052 588
b3c2d2dd
MS
589/**
590 * splice_from_pipe_begin - start splicing from pipe
b80901bb 591 * @sd: information about the splice operation
b3c2d2dd
MS
592 *
593 * Description:
594 * This function should be called before a loop containing
595 * splice_from_pipe_next() and splice_from_pipe_feed() to
596 * initialize the necessary fields of @sd.
597 */
96f9bc8f 598static void splice_from_pipe_begin(struct splice_desc *sd)
b3c2d2dd
MS
599{
600 sd->num_spliced = 0;
601 sd->need_wakeup = false;
602}
5274f052 603
b3c2d2dd
MS
604/**
605 * splice_from_pipe_end - finish splicing from pipe
606 * @pipe: pipe to splice from
607 * @sd: information about the splice operation
608 *
609 * Description:
610 * This function will wake up pipe writers if necessary. It should
611 * be called after a loop containing splice_from_pipe_next() and
612 * splice_from_pipe_feed().
613 */
96f9bc8f 614static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
b3c2d2dd
MS
615{
616 if (sd->need_wakeup)
617 wakeup_pipe_writers(pipe);
618}
5274f052 619
b3c2d2dd
MS
620/**
621 * __splice_from_pipe - splice data from a pipe to given actor
622 * @pipe: pipe to splice from
623 * @sd: information to @actor
624 * @actor: handler that splices the data
625 *
626 * Description:
627 * This function does little more than loop over the pipe and call
628 * @actor to do the actual moving of a single struct pipe_buffer to
629 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
630 * pipe_to_user.
631 *
632 */
633ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
634 splice_actor *actor)
635{
636 int ret;
5274f052 637
b3c2d2dd
MS
638 splice_from_pipe_begin(sd);
639 do {
c2489e07 640 cond_resched();
b3c2d2dd
MS
641 ret = splice_from_pipe_next(pipe, sd);
642 if (ret > 0)
643 ret = splice_from_pipe_feed(pipe, sd, actor);
644 } while (ret > 0);
645 splice_from_pipe_end(pipe, sd);
646
647 return sd->num_spliced ? sd->num_spliced : ret;
5274f052 648}
40bee44e 649EXPORT_SYMBOL(__splice_from_pipe);
5274f052 650
932cc6d4
JA
651/**
652 * splice_from_pipe - splice data from a pipe to a file
653 * @pipe: pipe to splice from
654 * @out: file to splice to
655 * @ppos: position in @out
656 * @len: how many bytes to splice
657 * @flags: splice modifier flags
658 * @actor: handler that splices the data
659 *
660 * Description:
2933970b 661 * See __splice_from_pipe. This function locks the pipe inode,
932cc6d4
JA
662 * otherwise it's identical to __splice_from_pipe().
663 *
664 */
6da61809
MF
665ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
666 loff_t *ppos, size_t len, unsigned int flags,
667 splice_actor *actor)
668{
669 ssize_t ret;
c66ab6fa
JA
670 struct splice_desc sd = {
671 .total_len = len,
672 .flags = flags,
673 .pos = *ppos,
6a14b90b 674 .u.file = out,
c66ab6fa 675 };
6da61809 676
61e0d47c 677 pipe_lock(pipe);
c66ab6fa 678 ret = __splice_from_pipe(pipe, &sd, actor);
61e0d47c 679 pipe_unlock(pipe);
6da61809
MF
680
681 return ret;
682}
683
8d020765
AV
684/**
685 * iter_file_splice_write - splice data from a pipe to a file
686 * @pipe: pipe info
687 * @out: file to write to
688 * @ppos: position in @out
689 * @len: number of bytes to splice
690 * @flags: splice modifier flags
691 *
692 * Description:
693 * Will either move or copy pages (determined by @flags options) from
694 * the given pipe inode to the given file.
695 * This one is ->write_iter-based.
696 *
697 */
698ssize_t
699iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
700 loff_t *ppos, size_t len, unsigned int flags)
701{
702 struct splice_desc sd = {
703 .total_len = len,
704 .flags = flags,
705 .pos = *ppos,
706 .u.file = out,
707 };
708 int nbufs = pipe->buffers;
709 struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
710 GFP_KERNEL);
711 ssize_t ret;
712
713 if (unlikely(!array))
714 return -ENOMEM;
715
716 pipe_lock(pipe);
717
718 splice_from_pipe_begin(&sd);
719 while (sd.total_len) {
720 struct iov_iter from;
8d020765
AV
721 size_t left;
722 int n, idx;
723
724 ret = splice_from_pipe_next(pipe, &sd);
725 if (ret <= 0)
726 break;
727
728 if (unlikely(nbufs < pipe->buffers)) {
729 kfree(array);
730 nbufs = pipe->buffers;
731 array = kcalloc(nbufs, sizeof(struct bio_vec),
732 GFP_KERNEL);
733 if (!array) {
734 ret = -ENOMEM;
735 break;
736 }
737 }
738
739 /* build the vector */
740 left = sd.total_len;
741 for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) {
742 struct pipe_buffer *buf = pipe->bufs + idx;
743 size_t this_len = buf->len;
744
745 if (this_len > left)
746 this_len = left;
747
748 if (idx == pipe->buffers - 1)
749 idx = -1;
750
fba597db 751 ret = pipe_buf_confirm(pipe, buf);
8d020765
AV
752 if (unlikely(ret)) {
753 if (ret == -ENODATA)
754 ret = 0;
755 goto done;
756 }
757
758 array[n].bv_page = buf->page;
759 array[n].bv_len = this_len;
760 array[n].bv_offset = buf->offset;
761 left -= this_len;
762 }
763
05afcb77
AV
764 iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n,
765 sd.total_len - left);
dbe4e192 766 ret = vfs_iter_write(out, &from, &sd.pos);
8d020765
AV
767 if (ret <= 0)
768 break;
769
770 sd.num_spliced += ret;
771 sd.total_len -= ret;
dbe4e192 772 *ppos = sd.pos;
8d020765
AV
773
774 /* dismiss the fully eaten buffers, adjust the partial one */
775 while (ret) {
776 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
777 if (ret >= buf->len) {
8d020765
AV
778 ret -= buf->len;
779 buf->len = 0;
a779638c 780 pipe_buf_release(pipe, buf);
8d020765
AV
781 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
782 pipe->nrbufs--;
783 if (pipe->files)
784 sd.need_wakeup = true;
785 } else {
786 buf->offset += ret;
787 buf->len -= ret;
788 ret = 0;
789 }
790 }
791 }
792done:
793 kfree(array);
794 splice_from_pipe_end(pipe, &sd);
795
796 pipe_unlock(pipe);
797
798 if (sd.num_spliced)
799 ret = sd.num_spliced;
800
801 return ret;
802}
803
804EXPORT_SYMBOL(iter_file_splice_write);
805
b2858d7d
MS
806static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
807 struct splice_desc *sd)
0b0a47f5 808{
b2858d7d
MS
809 int ret;
810 void *data;
06ae43f3 811 loff_t tmp = sd->pos;
b2858d7d 812
fbb32750 813 data = kmap(buf->page);
06ae43f3 814 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
fbb32750 815 kunmap(buf->page);
b2858d7d
MS
816
817 return ret;
0b0a47f5
MS
818}
819
820static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
821 struct file *out, loff_t *ppos,
822 size_t len, unsigned int flags)
823{
b2858d7d 824 ssize_t ret;
0b0a47f5 825
b2858d7d
MS
826 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
827 if (ret > 0)
828 *ppos += ret;
0b0a47f5 829
b2858d7d 830 return ret;
0b0a47f5
MS
831}
832
83f9135b
JA
833/**
834 * generic_splice_sendpage - splice data from a pipe to a socket
932cc6d4 835 * @pipe: pipe to splice from
83f9135b 836 * @out: socket to write to
932cc6d4 837 * @ppos: position in @out
83f9135b
JA
838 * @len: number of bytes to splice
839 * @flags: splice modifier flags
840 *
932cc6d4
JA
841 * Description:
842 * Will send @len bytes from the pipe to a network socket. No data copying
843 * is involved.
83f9135b
JA
844 *
845 */
3a326a2c 846ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
cbb7e577 847 loff_t *ppos, size_t len, unsigned int flags)
5274f052 848{
00522fb4 849 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
5274f052
JA
850}
851
059a8f37 852EXPORT_SYMBOL(generic_splice_sendpage);
a0f06780 853
83f9135b
JA
854/*
855 * Attempt to initiate a splice from pipe to file.
856 */
3a326a2c 857static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
cbb7e577 858 loff_t *ppos, size_t len, unsigned int flags)
5274f052 859{
0b0a47f5
MS
860 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
861 loff_t *, size_t, unsigned int);
5274f052 862
72c2d531 863 if (out->f_op->splice_write)
cc56f7de
CG
864 splice_write = out->f_op->splice_write;
865 else
0b0a47f5
MS
866 splice_write = default_file_splice_write;
867
500368f7 868 return splice_write(pipe, out, ppos, len, flags);
5274f052
JA
869}
870
83f9135b
JA
871/*
872 * Attempt to initiate a splice from a file to a pipe.
873 */
cbb7e577
JA
874static long do_splice_to(struct file *in, loff_t *ppos,
875 struct pipe_inode_info *pipe, size_t len,
876 unsigned int flags)
5274f052 877{
6818173b
MS
878 ssize_t (*splice_read)(struct file *, loff_t *,
879 struct pipe_inode_info *, size_t, unsigned int);
5274f052
JA
880 int ret;
881
49570e9b 882 if (unlikely(!(in->f_mode & FMODE_READ)))
5274f052
JA
883 return -EBADF;
884
cbb7e577 885 ret = rw_verify_area(READ, in, ppos, len);
5274f052
JA
886 if (unlikely(ret < 0))
887 return ret;
888
03cc0789
AV
889 if (unlikely(len > MAX_RW_COUNT))
890 len = MAX_RW_COUNT;
891
72c2d531 892 if (in->f_op->splice_read)
cc56f7de
CG
893 splice_read = in->f_op->splice_read;
894 else
6818173b
MS
895 splice_read = default_file_splice_read;
896
897 return splice_read(in, ppos, pipe, len, flags);
5274f052
JA
898}
899
932cc6d4
JA
900/**
901 * splice_direct_to_actor - splices data directly between two non-pipes
902 * @in: file to splice from
903 * @sd: actor information on where to splice to
904 * @actor: handles the data splicing
905 *
906 * Description:
907 * This is a special case helper to splice directly between two
908 * points, without requiring an explicit pipe. Internally an allocated
79685b8d 909 * pipe is cached in the process, and reused during the lifetime of
932cc6d4
JA
910 * that process.
911 *
c66ab6fa
JA
912 */
913ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
914 splice_direct_actor *actor)
b92ce558
JA
915{
916 struct pipe_inode_info *pipe;
917 long ret, bytes;
918 umode_t i_mode;
c66ab6fa 919 size_t len;
0ff28d9f 920 int i, flags, more;
b92ce558
JA
921
922 /*
923 * We require the input being a regular file, as we don't want to
924 * randomly drop data for eg socket -> socket splicing. Use the
925 * piped splicing for that!
926 */
496ad9aa 927 i_mode = file_inode(in)->i_mode;
b92ce558
JA
928 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
929 return -EINVAL;
930
931 /*
932 * neither in nor out is a pipe, setup an internal pipe attached to
933 * 'out' and transfer the wanted data from 'in' to 'out' through that
934 */
935 pipe = current->splice_pipe;
49570e9b 936 if (unlikely(!pipe)) {
7bee130e 937 pipe = alloc_pipe_info();
b92ce558
JA
938 if (!pipe)
939 return -ENOMEM;
940
941 /*
942 * We don't have an immediate reader, but we'll read the stuff
00522fb4 943 * out of the pipe right after the splice_to_pipe(). So set
b92ce558
JA
944 * PIPE_READERS appropriately.
945 */
946 pipe->readers = 1;
947
948 current->splice_pipe = pipe;
949 }
950
951 /*
73d62d83 952 * Do the splice.
b92ce558
JA
953 */
954 ret = 0;
955 bytes = 0;
c66ab6fa
JA
956 len = sd->total_len;
957 flags = sd->flags;
958
959 /*
960 * Don't block on output, we have to drain the direct pipe.
961 */
962 sd->flags &= ~SPLICE_F_NONBLOCK;
0ff28d9f 963 more = sd->flags & SPLICE_F_MORE;
b92ce558
JA
964
965 while (len) {
51a92c0f 966 size_t read_len;
a82c53a0 967 loff_t pos = sd->pos, prev_pos = pos;
b92ce558 968
bcd4f3ac 969 ret = do_splice_to(in, &pos, pipe, len, flags);
51a92c0f 970 if (unlikely(ret <= 0))
b92ce558
JA
971 goto out_release;
972
973 read_len = ret;
c66ab6fa 974 sd->total_len = read_len;
b92ce558 975
0ff28d9f
CL
976 /*
977 * If more data is pending, set SPLICE_F_MORE
978 * If this is the last data and SPLICE_F_MORE was not set
979 * initially, clears it.
980 */
981 if (read_len < len)
982 sd->flags |= SPLICE_F_MORE;
983 else if (!more)
984 sd->flags &= ~SPLICE_F_MORE;
b92ce558
JA
985 /*
986 * NOTE: nonblocking mode only applies to the input. We
987 * must not do the output in nonblocking mode as then we
988 * could get stuck data in the internal pipe:
989 */
c66ab6fa 990 ret = actor(pipe, sd);
a82c53a0
TZ
991 if (unlikely(ret <= 0)) {
992 sd->pos = prev_pos;
b92ce558 993 goto out_release;
a82c53a0 994 }
b92ce558
JA
995
996 bytes += ret;
997 len -= ret;
bcd4f3ac 998 sd->pos = pos;
b92ce558 999
a82c53a0
TZ
1000 if (ret < read_len) {
1001 sd->pos = prev_pos + ret;
51a92c0f 1002 goto out_release;
a82c53a0 1003 }
b92ce558
JA
1004 }
1005
9e97198d 1006done:
b92ce558 1007 pipe->nrbufs = pipe->curbuf = 0;
80848708 1008 file_accessed(in);
b92ce558
JA
1009 return bytes;
1010
1011out_release:
1012 /*
1013 * If we did an incomplete transfer we must release
1014 * the pipe buffers in question:
1015 */
35f3d14d 1016 for (i = 0; i < pipe->buffers; i++) {
b92ce558
JA
1017 struct pipe_buffer *buf = pipe->bufs + i;
1018
a779638c
MS
1019 if (buf->ops)
1020 pipe_buf_release(pipe, buf);
b92ce558 1021 }
b92ce558 1022
9e97198d
JA
1023 if (!bytes)
1024 bytes = ret;
c66ab6fa 1025
9e97198d 1026 goto done;
c66ab6fa
JA
1027}
1028EXPORT_SYMBOL(splice_direct_to_actor);
1029
1030static int direct_splice_actor(struct pipe_inode_info *pipe,
1031 struct splice_desc *sd)
1032{
6a14b90b 1033 struct file *file = sd->u.file;
c66ab6fa 1034
7995bd28 1035 return do_splice_from(pipe, file, sd->opos, sd->total_len,
2cb4b05e 1036 sd->flags);
c66ab6fa
JA
1037}
1038
932cc6d4
JA
1039/**
1040 * do_splice_direct - splices data directly between two files
1041 * @in: file to splice from
1042 * @ppos: input file offset
1043 * @out: file to splice to
acdb37c3 1044 * @opos: output file offset
932cc6d4
JA
1045 * @len: number of bytes to splice
1046 * @flags: splice modifier flags
1047 *
1048 * Description:
1049 * For use by do_sendfile(). splice can easily emulate sendfile, but
1050 * doing it in the application would incur an extra system call
1051 * (splice in + splice out, as compared to just sendfile()). So this helper
1052 * can splice directly through a process-private pipe.
1053 *
1054 */
c66ab6fa 1055long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
7995bd28 1056 loff_t *opos, size_t len, unsigned int flags)
c66ab6fa
JA
1057{
1058 struct splice_desc sd = {
1059 .len = len,
1060 .total_len = len,
1061 .flags = flags,
1062 .pos = *ppos,
6a14b90b 1063 .u.file = out,
7995bd28 1064 .opos = opos,
c66ab6fa 1065 };
51a92c0f 1066 long ret;
c66ab6fa 1067
18c67cb9
AV
1068 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1069 return -EBADF;
1070
1071 if (unlikely(out->f_flags & O_APPEND))
1072 return -EINVAL;
1073
1074 ret = rw_verify_area(WRITE, out, opos, len);
1075 if (unlikely(ret < 0))
1076 return ret;
1077
c66ab6fa 1078 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
51a92c0f 1079 if (ret > 0)
a82c53a0 1080 *ppos = sd.pos;
51a92c0f 1081
c66ab6fa 1082 return ret;
b92ce558 1083}
1c118596 1084EXPORT_SYMBOL(do_splice_direct);
b92ce558 1085
8924feff
AV
1086static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags)
1087{
1088 while (pipe->nrbufs == pipe->buffers) {
1089 if (flags & SPLICE_F_NONBLOCK)
1090 return -EAGAIN;
1091 if (signal_pending(current))
1092 return -ERESTARTSYS;
1093 pipe->waiting_writers++;
1094 pipe_wait(pipe);
1095 pipe->waiting_writers--;
1096 }
1097 return 0;
1098}
1099
7c77f0b3
MS
1100static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1101 struct pipe_inode_info *opipe,
1102 size_t len, unsigned int flags);
ddac0d39 1103
83f9135b
JA
1104/*
1105 * Determine where to splice to/from.
1106 */
529565dc
IM
1107static long do_splice(struct file *in, loff_t __user *off_in,
1108 struct file *out, loff_t __user *off_out,
1109 size_t len, unsigned int flags)
5274f052 1110{
7c77f0b3
MS
1111 struct pipe_inode_info *ipipe;
1112 struct pipe_inode_info *opipe;
7995bd28 1113 loff_t offset;
a4514ebd 1114 long ret;
5274f052 1115
71993e62
LT
1116 ipipe = get_pipe_info(in);
1117 opipe = get_pipe_info(out);
7c77f0b3
MS
1118
1119 if (ipipe && opipe) {
1120 if (off_in || off_out)
1121 return -ESPIPE;
1122
1123 if (!(in->f_mode & FMODE_READ))
1124 return -EBADF;
1125
1126 if (!(out->f_mode & FMODE_WRITE))
1127 return -EBADF;
1128
1129 /* Splicing to self would be fun, but... */
1130 if (ipipe == opipe)
1131 return -EINVAL;
1132
1133 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1134 }
1135
1136 if (ipipe) {
529565dc
IM
1137 if (off_in)
1138 return -ESPIPE;
b92ce558 1139 if (off_out) {
19c9a49b 1140 if (!(out->f_mode & FMODE_PWRITE))
b92ce558 1141 return -EINVAL;
cbb7e577 1142 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
b92ce558 1143 return -EFAULT;
7995bd28
AV
1144 } else {
1145 offset = out->f_pos;
1146 }
529565dc 1147
18c67cb9
AV
1148 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1149 return -EBADF;
1150
1151 if (unlikely(out->f_flags & O_APPEND))
1152 return -EINVAL;
1153
1154 ret = rw_verify_area(WRITE, out, &offset, len);
1155 if (unlikely(ret < 0))
1156 return ret;
1157
500368f7 1158 file_start_write(out);
7995bd28 1159 ret = do_splice_from(ipipe, out, &offset, len, flags);
500368f7 1160 file_end_write(out);
a4514ebd 1161
7995bd28
AV
1162 if (!off_out)
1163 out->f_pos = offset;
1164 else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
a4514ebd
JA
1165 ret = -EFAULT;
1166
1167 return ret;
529565dc 1168 }
5274f052 1169
7c77f0b3 1170 if (opipe) {
529565dc
IM
1171 if (off_out)
1172 return -ESPIPE;
b92ce558 1173 if (off_in) {
19c9a49b 1174 if (!(in->f_mode & FMODE_PREAD))
b92ce558 1175 return -EINVAL;
cbb7e577 1176 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
b92ce558 1177 return -EFAULT;
7995bd28
AV
1178 } else {
1179 offset = in->f_pos;
1180 }
529565dc 1181
8924feff
AV
1182 pipe_lock(opipe);
1183 ret = wait_for_space(opipe, flags);
1184 if (!ret)
1185 ret = do_splice_to(in, &offset, opipe, len, flags);
1186 pipe_unlock(opipe);
1187 if (ret > 0)
1188 wakeup_pipe_readers(opipe);
7995bd28
AV
1189 if (!off_in)
1190 in->f_pos = offset;
1191 else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
a4514ebd
JA
1192 ret = -EFAULT;
1193
1194 return ret;
529565dc 1195 }
5274f052
JA
1196
1197 return -EINVAL;
1198}
1199
79fddc4e
AV
1200static int iter_to_pipe(struct iov_iter *from,
1201 struct pipe_inode_info *pipe,
1202 unsigned flags)
912d35f8 1203{
79fddc4e
AV
1204 struct pipe_buffer buf = {
1205 .ops = &user_page_pipe_buf_ops,
1206 .flags = flags
1207 };
1208 size_t total = 0;
1209 int ret = 0;
1210 bool failed = false;
1211
1212 while (iov_iter_count(from) && !failed) {
1213 struct page *pages[16];
db85a9eb
AV
1214 ssize_t copied;
1215 size_t start;
79fddc4e 1216 int n;
db85a9eb 1217
79fddc4e
AV
1218 copied = iov_iter_get_pages(from, pages, ~0UL, 16, &start);
1219 if (copied <= 0) {
1220 ret = copied;
1221 break;
1222 }
db85a9eb 1223
79fddc4e 1224 for (n = 0; copied; n++, start = 0) {
db85a9eb 1225 int size = min_t(int, copied, PAGE_SIZE - start);
79fddc4e
AV
1226 if (!failed) {
1227 buf.page = pages[n];
1228 buf.offset = start;
1229 buf.len = size;
1230 ret = add_to_pipe(pipe, &buf);
1231 if (unlikely(ret < 0)) {
1232 failed = true;
1233 } else {
1234 iov_iter_advance(from, ret);
1235 total += ret;
1236 }
1237 } else {
1238 put_page(pages[n]);
1239 }
db85a9eb 1240 copied -= size;
912d35f8 1241 }
912d35f8 1242 }
79fddc4e 1243 return total ? total : ret;
912d35f8
JA
1244}
1245
6a14b90b
JA
1246static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1247 struct splice_desc *sd)
1248{
6130f531
AV
1249 int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1250 return n == sd->len ? n : -EFAULT;
6a14b90b
JA
1251}
1252
1253/*
1254 * For lack of a better implementation, implement vmsplice() to userspace
1255 * as a simple copy of the pipes pages to the user iov.
1256 */
6130f531 1257static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov,
6a14b90b
JA
1258 unsigned long nr_segs, unsigned int flags)
1259{
1260 struct pipe_inode_info *pipe;
1261 struct splice_desc sd;
6a14b90b 1262 long ret;
6130f531
AV
1263 struct iovec iovstack[UIO_FASTIOV];
1264 struct iovec *iov = iovstack;
1265 struct iov_iter iter;
6a14b90b 1266
71993e62 1267 pipe = get_pipe_info(file);
6a14b90b
JA
1268 if (!pipe)
1269 return -EBADF;
1270
345995fa
AV
1271 ret = import_iovec(READ, uiov, nr_segs,
1272 ARRAY_SIZE(iovstack), &iov, &iter);
1273 if (ret < 0)
1274 return ret;
6a14b90b 1275
345995fa 1276 sd.total_len = iov_iter_count(&iter);
6130f531 1277 sd.len = 0;
6130f531
AV
1278 sd.flags = flags;
1279 sd.u.data = &iter;
1280 sd.pos = 0;
6a14b90b 1281
345995fa
AV
1282 if (sd.total_len) {
1283 pipe_lock(pipe);
1284 ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
1285 pipe_unlock(pipe);
1286 }
6a14b90b 1287
345995fa 1288 kfree(iov);
6a14b90b
JA
1289 return ret;
1290}
1291
912d35f8
JA
1292/*
1293 * vmsplice splices a user address range into a pipe. It can be thought of
1294 * as splice-from-memory, where the regular splice is splice-from-file (or
1295 * to file). In both cases the output is a pipe, naturally.
912d35f8 1296 */
db85a9eb 1297static long vmsplice_to_pipe(struct file *file, const struct iovec __user *uiov,
6a14b90b 1298 unsigned long nr_segs, unsigned int flags)
912d35f8 1299{
ddac0d39 1300 struct pipe_inode_info *pipe;
db85a9eb
AV
1301 struct iovec iovstack[UIO_FASTIOV];
1302 struct iovec *iov = iovstack;
1303 struct iov_iter from;
35f3d14d 1304 long ret;
79fddc4e
AV
1305 unsigned buf_flag = 0;
1306
1307 if (flags & SPLICE_F_GIFT)
1308 buf_flag = PIPE_BUF_FLAG_GIFT;
912d35f8 1309
71993e62 1310 pipe = get_pipe_info(file);
ddac0d39 1311 if (!pipe)
912d35f8 1312 return -EBADF;
912d35f8 1313
db85a9eb
AV
1314 ret = import_iovec(WRITE, uiov, nr_segs,
1315 ARRAY_SIZE(iovstack), &iov, &from);
1316 if (ret < 0)
1317 return ret;
1318
8924feff
AV
1319 pipe_lock(pipe);
1320 ret = wait_for_space(pipe, flags);
79fddc4e
AV
1321 if (!ret)
1322 ret = iter_to_pipe(&from, pipe, buf_flag);
8924feff
AV
1323 pipe_unlock(pipe);
1324 if (ret > 0)
1325 wakeup_pipe_readers(pipe);
db85a9eb 1326 kfree(iov);
35f3d14d 1327 return ret;
912d35f8
JA
1328}
1329
6a14b90b
JA
1330/*
1331 * Note that vmsplice only really supports true splicing _from_ user memory
1332 * to a pipe, not the other way around. Splicing from user memory is a simple
1333 * operation that can be supported without any funky alignment restrictions
1334 * or nasty vm tricks. We simply map in the user memory and fill them into
1335 * a pipe. The reverse isn't quite as easy, though. There are two possible
1336 * solutions for that:
1337 *
1338 * - memcpy() the data internally, at which point we might as well just
1339 * do a regular read() on the buffer anyway.
1340 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1341 * has restriction limitations on both ends of the pipe).
1342 *
1343 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1344 *
1345 */
836f92ad
HC
1346SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1347 unsigned long, nr_segs, unsigned int, flags)
912d35f8 1348{
2903ff01 1349 struct fd f;
912d35f8 1350 long error;
912d35f8 1351
6a14b90b
JA
1352 if (unlikely(nr_segs > UIO_MAXIOV))
1353 return -EINVAL;
1354 else if (unlikely(!nr_segs))
1355 return 0;
1356
912d35f8 1357 error = -EBADF;
2903ff01
AV
1358 f = fdget(fd);
1359 if (f.file) {
1360 if (f.file->f_mode & FMODE_WRITE)
1361 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1362 else if (f.file->f_mode & FMODE_READ)
1363 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
1364
1365 fdput(f);
912d35f8
JA
1366 }
1367
1368 return error;
1369}
1370
76b021d0
AV
1371#ifdef CONFIG_COMPAT
1372COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1373 unsigned int, nr_segs, unsigned int, flags)
1374{
1375 unsigned i;
1376 struct iovec __user *iov;
1377 if (nr_segs > UIO_MAXIOV)
1378 return -EINVAL;
1379 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1380 for (i = 0; i < nr_segs; i++) {
1381 struct compat_iovec v;
1382 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1383 get_user(v.iov_len, &iov32[i].iov_len) ||
1384 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1385 put_user(v.iov_len, &iov[i].iov_len))
1386 return -EFAULT;
1387 }
1388 return sys_vmsplice(fd, iov, nr_segs, flags);
1389}
1390#endif
1391
836f92ad
HC
1392SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1393 int, fd_out, loff_t __user *, off_out,
1394 size_t, len, unsigned int, flags)
5274f052 1395{
2903ff01 1396 struct fd in, out;
5274f052 1397 long error;
5274f052
JA
1398
1399 if (unlikely(!len))
1400 return 0;
1401
1402 error = -EBADF;
2903ff01
AV
1403 in = fdget(fd_in);
1404 if (in.file) {
1405 if (in.file->f_mode & FMODE_READ) {
1406 out = fdget(fd_out);
1407 if (out.file) {
1408 if (out.file->f_mode & FMODE_WRITE)
1409 error = do_splice(in.file, off_in,
1410 out.file, off_out,
529565dc 1411 len, flags);
2903ff01 1412 fdput(out);
5274f052
JA
1413 }
1414 }
2903ff01 1415 fdput(in);
5274f052 1416 }
5274f052
JA
1417 return error;
1418}
70524490 1419
aadd06e5
JA
1420/*
1421 * Make sure there's data to read. Wait for input if we can, otherwise
1422 * return an appropriate error.
1423 */
7c77f0b3 1424static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
aadd06e5
JA
1425{
1426 int ret;
1427
1428 /*
1429 * Check ->nrbufs without the inode lock first. This function
1430 * is speculative anyways, so missing one is ok.
1431 */
1432 if (pipe->nrbufs)
1433 return 0;
1434
1435 ret = 0;
61e0d47c 1436 pipe_lock(pipe);
aadd06e5
JA
1437
1438 while (!pipe->nrbufs) {
1439 if (signal_pending(current)) {
1440 ret = -ERESTARTSYS;
1441 break;
1442 }
1443 if (!pipe->writers)
1444 break;
1445 if (!pipe->waiting_writers) {
1446 if (flags & SPLICE_F_NONBLOCK) {
1447 ret = -EAGAIN;
1448 break;
1449 }
1450 }
1451 pipe_wait(pipe);
1452 }
1453
61e0d47c 1454 pipe_unlock(pipe);
aadd06e5
JA
1455 return ret;
1456}
1457
1458/*
1459 * Make sure there's writeable room. Wait for room if we can, otherwise
1460 * return an appropriate error.
1461 */
7c77f0b3 1462static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
aadd06e5
JA
1463{
1464 int ret;
1465
1466 /*
1467 * Check ->nrbufs without the inode lock first. This function
1468 * is speculative anyways, so missing one is ok.
1469 */
35f3d14d 1470 if (pipe->nrbufs < pipe->buffers)
aadd06e5
JA
1471 return 0;
1472
1473 ret = 0;
61e0d47c 1474 pipe_lock(pipe);
aadd06e5 1475
35f3d14d 1476 while (pipe->nrbufs >= pipe->buffers) {
aadd06e5
JA
1477 if (!pipe->readers) {
1478 send_sig(SIGPIPE, current, 0);
1479 ret = -EPIPE;
1480 break;
1481 }
1482 if (flags & SPLICE_F_NONBLOCK) {
1483 ret = -EAGAIN;
1484 break;
1485 }
1486 if (signal_pending(current)) {
1487 ret = -ERESTARTSYS;
1488 break;
1489 }
1490 pipe->waiting_writers++;
1491 pipe_wait(pipe);
1492 pipe->waiting_writers--;
1493 }
1494
61e0d47c 1495 pipe_unlock(pipe);
aadd06e5
JA
1496 return ret;
1497}
1498
7c77f0b3
MS
1499/*
1500 * Splice contents of ipipe to opipe.
1501 */
1502static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1503 struct pipe_inode_info *opipe,
1504 size_t len, unsigned int flags)
1505{
1506 struct pipe_buffer *ibuf, *obuf;
1507 int ret = 0, nbuf;
1508 bool input_wakeup = false;
1509
1510
1511retry:
1512 ret = ipipe_prep(ipipe, flags);
1513 if (ret)
1514 return ret;
1515
1516 ret = opipe_prep(opipe, flags);
1517 if (ret)
1518 return ret;
1519
1520 /*
1521 * Potential ABBA deadlock, work around it by ordering lock
1522 * grabbing by pipe info address. Otherwise two different processes
1523 * could deadlock (one doing tee from A -> B, the other from B -> A).
1524 */
1525 pipe_double_lock(ipipe, opipe);
1526
1527 do {
1528 if (!opipe->readers) {
1529 send_sig(SIGPIPE, current, 0);
1530 if (!ret)
1531 ret = -EPIPE;
1532 break;
1533 }
1534
1535 if (!ipipe->nrbufs && !ipipe->writers)
1536 break;
1537
1538 /*
1539 * Cannot make any progress, because either the input
1540 * pipe is empty or the output pipe is full.
1541 */
35f3d14d 1542 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
7c77f0b3
MS
1543 /* Already processed some buffers, break */
1544 if (ret)
1545 break;
1546
1547 if (flags & SPLICE_F_NONBLOCK) {
1548 ret = -EAGAIN;
1549 break;
1550 }
1551
1552 /*
1553 * We raced with another reader/writer and haven't
1554 * managed to process any buffers. A zero return
1555 * value means EOF, so retry instead.
1556 */
1557 pipe_unlock(ipipe);
1558 pipe_unlock(opipe);
1559 goto retry;
1560 }
1561
1562 ibuf = ipipe->bufs + ipipe->curbuf;
35f3d14d 1563 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
7c77f0b3
MS
1564 obuf = opipe->bufs + nbuf;
1565
1566 if (len >= ibuf->len) {
1567 /*
1568 * Simply move the whole buffer from ipipe to opipe
1569 */
1570 *obuf = *ibuf;
1571 ibuf->ops = NULL;
1572 opipe->nrbufs++;
35f3d14d 1573 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
7c77f0b3
MS
1574 ipipe->nrbufs--;
1575 input_wakeup = true;
1576 } else {
1577 /*
1578 * Get a reference to this pipe buffer,
1579 * so we can copy the contents over.
1580 */
7bf2d1df 1581 pipe_buf_get(ipipe, ibuf);
7c77f0b3
MS
1582 *obuf = *ibuf;
1583
1584 /*
1585 * Don't inherit the gift flag, we need to
1586 * prevent multiple steals of this page.
1587 */
1588 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1589
1590 obuf->len = len;
1591 opipe->nrbufs++;
1592 ibuf->offset += obuf->len;
1593 ibuf->len -= obuf->len;
1594 }
1595 ret += obuf->len;
1596 len -= obuf->len;
1597 } while (len);
1598
1599 pipe_unlock(ipipe);
1600 pipe_unlock(opipe);
1601
1602 /*
1603 * If we put data in the output pipe, wakeup any potential readers.
1604 */
825cdcb1
NK
1605 if (ret > 0)
1606 wakeup_pipe_readers(opipe);
1607
7c77f0b3
MS
1608 if (input_wakeup)
1609 wakeup_pipe_writers(ipipe);
1610
1611 return ret;
1612}
1613
70524490
JA
1614/*
1615 * Link contents of ipipe to opipe.
1616 */
1617static int link_pipe(struct pipe_inode_info *ipipe,
1618 struct pipe_inode_info *opipe,
1619 size_t len, unsigned int flags)
1620{
1621 struct pipe_buffer *ibuf, *obuf;
aadd06e5 1622 int ret = 0, i = 0, nbuf;
70524490
JA
1623
1624 /*
1625 * Potential ABBA deadlock, work around it by ordering lock
61e0d47c 1626 * grabbing by pipe info address. Otherwise two different processes
70524490
JA
1627 * could deadlock (one doing tee from A -> B, the other from B -> A).
1628 */
61e0d47c 1629 pipe_double_lock(ipipe, opipe);
70524490 1630
aadd06e5 1631 do {
70524490
JA
1632 if (!opipe->readers) {
1633 send_sig(SIGPIPE, current, 0);
1634 if (!ret)
1635 ret = -EPIPE;
1636 break;
1637 }
70524490 1638
aadd06e5
JA
1639 /*
1640 * If we have iterated all input buffers or ran out of
1641 * output room, break.
1642 */
35f3d14d 1643 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
aadd06e5 1644 break;
70524490 1645
35f3d14d
JA
1646 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1647 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
70524490
JA
1648
1649 /*
aadd06e5
JA
1650 * Get a reference to this pipe buffer,
1651 * so we can copy the contents over.
70524490 1652 */
7bf2d1df 1653 pipe_buf_get(ipipe, ibuf);
aadd06e5
JA
1654
1655 obuf = opipe->bufs + nbuf;
1656 *obuf = *ibuf;
1657
2a27250e 1658 /*
aadd06e5
JA
1659 * Don't inherit the gift flag, we need to
1660 * prevent multiple steals of this page.
2a27250e 1661 */
aadd06e5 1662 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
70524490 1663
aadd06e5
JA
1664 if (obuf->len > len)
1665 obuf->len = len;
70524490 1666
aadd06e5
JA
1667 opipe->nrbufs++;
1668 ret += obuf->len;
1669 len -= obuf->len;
1670 i++;
1671 } while (len);
70524490 1672
02cf01ae
JA
1673 /*
1674 * return EAGAIN if we have the potential of some data in the
1675 * future, otherwise just return 0
1676 */
1677 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1678 ret = -EAGAIN;
1679
61e0d47c
MS
1680 pipe_unlock(ipipe);
1681 pipe_unlock(opipe);
70524490 1682
aadd06e5
JA
1683 /*
1684 * If we put data in the output pipe, wakeup any potential readers.
1685 */
825cdcb1
NK
1686 if (ret > 0)
1687 wakeup_pipe_readers(opipe);
70524490
JA
1688
1689 return ret;
1690}
1691
1692/*
1693 * This is a tee(1) implementation that works on pipes. It doesn't copy
1694 * any data, it simply references the 'in' pages on the 'out' pipe.
1695 * The 'flags' used are the SPLICE_F_* variants, currently the only
1696 * applicable one is SPLICE_F_NONBLOCK.
1697 */
1698static long do_tee(struct file *in, struct file *out, size_t len,
1699 unsigned int flags)
1700{
71993e62
LT
1701 struct pipe_inode_info *ipipe = get_pipe_info(in);
1702 struct pipe_inode_info *opipe = get_pipe_info(out);
aadd06e5 1703 int ret = -EINVAL;
70524490
JA
1704
1705 /*
aadd06e5
JA
1706 * Duplicate the contents of ipipe to opipe without actually
1707 * copying the data.
70524490 1708 */
aadd06e5
JA
1709 if (ipipe && opipe && ipipe != opipe) {
1710 /*
1711 * Keep going, unless we encounter an error. The ipipe/opipe
1712 * ordering doesn't really matter.
1713 */
7c77f0b3 1714 ret = ipipe_prep(ipipe, flags);
aadd06e5 1715 if (!ret) {
7c77f0b3 1716 ret = opipe_prep(opipe, flags);
02cf01ae 1717 if (!ret)
aadd06e5 1718 ret = link_pipe(ipipe, opipe, len, flags);
aadd06e5
JA
1719 }
1720 }
70524490 1721
aadd06e5 1722 return ret;
70524490
JA
1723}
1724
836f92ad 1725SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
70524490 1726{
2903ff01
AV
1727 struct fd in;
1728 int error;
70524490
JA
1729
1730 if (unlikely(!len))
1731 return 0;
1732
1733 error = -EBADF;
2903ff01
AV
1734 in = fdget(fdin);
1735 if (in.file) {
1736 if (in.file->f_mode & FMODE_READ) {
1737 struct fd out = fdget(fdout);
1738 if (out.file) {
1739 if (out.file->f_mode & FMODE_WRITE)
1740 error = do_tee(in.file, out.file,
1741 len, flags);
1742 fdput(out);
70524490
JA
1743 }
1744 }
2903ff01 1745 fdput(in);
70524490
JA
1746 }
1747
1748 return error;
1749}