blob: 01983bea760cc5e6309a0eee7efb3ff609ff60ca [file] [log] [blame]
Jens Axboe5274f052006-03-30 15:15:30 +02001/*
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
Jens Axboec2058e02006-04-11 13:56:34 +020012 * Jens to support splicing to files, network, direct splicing, etc and
13 * fixing lots of bugs.
Jens Axboe5274f052006-03-30 15:15:30 +020014 *
Jens Axboe0fe23472006-09-04 15:41:16 +020015 * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk>
Jens Axboec2058e02006-04-11 13:56:34 +020016 * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
17 * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
Jens Axboe5274f052006-03-30 15:15:30 +020018 *
19 */
20#include <linux/fs.h>
21#include <linux/file.h>
22#include <linux/pagemap.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020023#include <linux/splice.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080024#include <linux/memcontrol.h>
Jens Axboe5274f052006-03-30 15:15:30 +020025#include <linux/mm_inline.h>
Jens Axboe5abc97a2006-03-30 15:16:46 +020026#include <linux/swap.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020027#include <linux/writeback.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050028#include <linux/export.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020029#include <linux/syscalls.h>
Jens Axboe912d35f2006-04-26 10:59:21 +020030#include <linux/uio.h>
James Morris29ce2052007-07-13 11:44:32 +020031#include <linux/security.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090032#include <linux/gfp.h>
Eric Dumazet35f9c092012-04-05 03:05:35 +000033#include <linux/socket.h>
Al Viro76b021d2013-03-02 10:19:56 -050034#include <linux/compat.h>
Al Viro06ae43f2013-03-20 13:19:30 -040035#include "internal.h"
Jens Axboe5274f052006-03-30 15:15:30 +020036
Jens Axboe83f91352006-04-02 23:05:09 +020037/*
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 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020043static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020044 struct pipe_buffer *buf)
45{
46 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020047 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020048
Jens Axboe9e0267c2006-04-19 15:57:31 +020049 lock_page(page);
50
Jens Axboe9e94cd42006-06-20 15:01:12 +020051 mapping = page_mapping(page);
52 if (mapping) {
53 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020054
Jens Axboe9e94cd42006-06-20 15:01:12 +020055 /*
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);
Jens Axboead8d6f02006-04-02 23:10:32 +020064
David Howells266cf652009-04-03 16:42:36 +010065 if (page_has_private(page) &&
66 !try_to_release_page(page, GFP_KERNEL))
Jens Axboeca39d652008-05-20 21:27:41 +020067 goto out_unlock;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020068
Jens Axboe9e94cd42006-06-20 15:01:12 +020069 /*
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 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020077 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020078
Jens Axboe9e94cd42006-06-20 15:01:12 +020079 /*
80 * Raced with truncate or failed to remove page from current
81 * address space, unlock and return failure.
82 */
Jens Axboeca39d652008-05-20 21:27:41 +020083out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020084 unlock_page(page);
85 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020086}
87
Jens Axboe76ad4d12006-05-03 10:41:33 +020088static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020089 struct pipe_buffer *buf)
90{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030091 put_page(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020092 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020093}
94
Jens Axboe08457182007-06-12 20:51:32 +020095/*
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 */
Jens Axboecac36bb02007-06-14 13:10:48 +020099static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
100 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +0200101{
102 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200103 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200104
105 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200106 lock_page(page);
107
108 /*
109 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200110 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200111 */
112 if (!page->mapping) {
113 err = -ENODATA;
114 goto error;
115 }
116
117 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200118 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200119 */
120 if (!PageUptodate(page)) {
121 err = -EIO;
122 goto error;
123 }
124
125 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200126 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200127 */
Jens Axboe5274f052006-03-30 15:15:30 +0200128 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200129 }
130
Jens Axboef84d7512006-05-01 19:59:03 +0200131 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200132error:
133 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200134 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200135}
136
Hugh Dickins708e3502011-07-25 17:12:32 -0700137const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200138 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +0200139 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200140 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200141 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200142 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200143};
144
Jens Axboe912d35f2006-04-26 10:59:21 +0200145static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
146 struct pipe_buffer *buf)
147{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200148 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
149 return 1;
150
Jens Axboe14328732006-05-03 10:35:26 +0200151 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200152 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200153}
154
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800155static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200156 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +0200157 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200158 .release = page_cache_pipe_buf_release,
159 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200160 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200161};
162
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200163static 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
Jens Axboe932cc6d2007-06-21 13:10:21 +0200171/**
172 * splice_to_pipe - fill passed data into a pipe
173 * @pipe: pipe to fill
174 * @spd: data to fill
175 *
176 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200177 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200178 * the struct pipe_buf_operations associated with these pages. This
179 * function will link that data to the pipe.
180 *
Jens Axboe83f91352006-04-02 23:05:09 +0200181 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200182ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
183 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200184{
Jens Axboe00de00b2007-06-15 13:14:22 +0200185 unsigned int spd_pages = spd->nr_pages;
Al Viro8924fef2016-09-17 20:44:45 -0400186 int ret = 0, page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200187
Rabin Vincentd6785d92016-03-10 21:19:06 +0100188 if (!spd_pages)
189 return 0;
190
Al Viro8924fef2016-09-17 20:44:45 -0400191 if (unlikely(!pipe->readers)) {
192 send_sig(SIGPIPE, current, 0);
193 ret = -EPIPE;
194 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200195 }
196
Al Viro8924fef2016-09-17 20:44:45 -0400197 while (pipe->nrbufs < pipe->buffers) {
198 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
199 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200200
Al Viro8924fef2016-09-17 20:44:45 -0400201 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;
Miklos Szeredi166a2462017-02-16 17:49:02 +0100206 buf->flags = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200207
Al Viro8924fef2016-09-17 20:44:45 -0400208 pipe->nrbufs++;
209 page_nr++;
210 ret += buf->len;
211
212 if (!--spd->nr_pages)
213 break;
214 }
215
216 if (!ret)
217 ret = -EAGAIN;
218
219out:
Jens Axboe00de00b2007-06-15 13:14:22 +0200220 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800221 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200222
223 return ret;
224}
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +0200225EXPORT_SYMBOL_GPL(splice_to_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200226
Al Viro79fddc42016-09-17 22:38:20 -0400227ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
228{
229 int ret;
230
231 if (unlikely(!pipe->readers)) {
232 send_sig(SIGPIPE, current, 0);
233 ret = -EPIPE;
234 } else if (pipe->nrbufs == pipe->buffers) {
235 ret = -EAGAIN;
236 } else {
237 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
238 pipe->bufs[newbuf] = *buf;
239 pipe->nrbufs++;
240 return buf->len;
241 }
Miklos Szeredia7796382016-09-27 10:45:12 +0200242 pipe_buf_release(pipe, buf);
Al Viro79fddc42016-09-17 22:38:20 -0400243 return ret;
244}
245EXPORT_SYMBOL(add_to_pipe);
246
Hugh Dickins708e3502011-07-25 17:12:32 -0700247void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800248{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300249 put_page(spd->pages[i]);
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800250}
251
Jens Axboe35f3d142010-05-20 10:43:18 +0200252/*
253 * Check if we need to grow the arrays holding pages and partial page
254 * descriptions.
255 */
Eric Dumazet047fe362012-06-12 15:24:40 +0200256int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200257{
Eric Dumazet047fe362012-06-12 15:24:40 +0200258 unsigned int buffers = ACCESS_ONCE(pipe->buffers);
259
260 spd->nr_pages_max = buffers;
261 if (buffers <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200262 return 0;
263
Eric Dumazet047fe362012-06-12 15:24:40 +0200264 spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL);
265 spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200266
267 if (spd->pages && spd->partial)
268 return 0;
269
270 kfree(spd->pages);
271 kfree(spd->partial);
272 return -ENOMEM;
273}
274
Eric Dumazet047fe362012-06-12 15:24:40 +0200275void splice_shrink_spd(struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200276{
Eric Dumazet047fe362012-06-12 15:24:40 +0200277 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200278 return;
279
280 kfree(spd->pages);
281 kfree(spd->partial);
282}
283
Jens Axboe83f91352006-04-02 23:05:09 +0200284/**
285 * generic_file_splice_read - splice data from file to a pipe
286 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200287 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200288 * @pipe: pipe to splice to
289 * @len: number of bytes to splice
290 * @flags: splice modifier flags
291 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200292 * Description:
293 * Will read pages from given file and fill them into a pipe. Can be
Al Viro82c156f2016-09-22 23:35:42 -0400294 * used as long as it has more or less sane ->read_iter().
Jens Axboe932cc6d2007-06-21 13:10:21 +0200295 *
Jens Axboe83f91352006-04-02 23:05:09 +0200296 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200297ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
298 struct pipe_inode_info *pipe, size_t len,
299 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200300{
Al Viro82c156f2016-09-22 23:35:42 -0400301 struct iov_iter to;
302 struct kiocb kiocb;
Al Viro82c156f2016-09-22 23:35:42 -0400303 int idx, ret;
Boaz Harroshbe64f882015-04-15 16:15:17 -0700304
Al Viro82c156f2016-09-22 23:35:42 -0400305 iov_iter_pipe(&to, ITER_PIPE | READ, pipe, len);
306 idx = to.idx;
307 init_sync_kiocb(&kiocb, in);
308 kiocb.ki_pos = *ppos;
309 ret = in->f_op->read_iter(&kiocb, &to);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200310 if (ret > 0) {
Al Viro82c156f2016-09-22 23:35:42 -0400311 *ppos = kiocb.ki_pos;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200312 file_accessed(in);
Al Viro82c156f2016-09-22 23:35:42 -0400313 } else if (ret < 0) {
Al Viroc3a69022016-10-10 13:26:27 -0400314 to.idx = idx;
315 to.iov_offset = 0;
316 iov_iter_advance(&to, 0); /* to free what was emitted */
Al Viro82c156f2016-09-22 23:35:42 -0400317 /*
318 * callers of ->splice_read() expect -EAGAIN on
319 * "can't put anything in there", rather than -EFAULT.
320 */
321 if (ret == -EFAULT)
322 ret = -EAGAIN;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200323 }
Jens Axboe5274f052006-03-30 15:15:30 +0200324
325 return ret;
326}
Jens Axboe059a8f32006-04-02 23:06:05 +0200327EXPORT_SYMBOL(generic_file_splice_read);
328
Al Viro241699c2016-09-22 16:33:12 -0400329const struct pipe_buf_operations default_pipe_buf_ops = {
Miklos Szeredi68181732009-05-07 15:37:36 +0200330 .can_merge = 0,
Miklos Szeredi68181732009-05-07 15:37:36 +0200331 .confirm = generic_pipe_buf_confirm,
332 .release = generic_pipe_buf_release,
333 .steal = generic_pipe_buf_steal,
334 .get = generic_pipe_buf_get,
335};
336
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100337static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe,
338 struct pipe_buffer *buf)
339{
340 return 1;
341}
342
343/* Pipe buffer operations for a socket and similar. */
344const struct pipe_buf_operations nosteal_pipe_buf_ops = {
345 .can_merge = 0,
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100346 .confirm = generic_pipe_buf_confirm,
347 .release = generic_pipe_buf_release,
348 .steal = generic_pipe_buf_nosteal,
349 .get = generic_pipe_buf_get,
350};
351EXPORT_SYMBOL(nosteal_pipe_buf_ops);
352
Al Viro523ac9a2016-09-23 15:34:57 -0400353static ssize_t kernel_readv(struct file *file, const struct kvec *vec,
Miklos Szeredi68181732009-05-07 15:37:36 +0200354 unsigned long vlen, loff_t offset)
355{
356 mm_segment_t old_fs;
357 loff_t pos = offset;
358 ssize_t res;
359
360 old_fs = get_fs();
361 set_fs(get_ds());
362 /* The cast to a user pointer is valid due to the set_fs() */
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100363 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos, 0);
Miklos Szeredi68181732009-05-07 15:37:36 +0200364 set_fs(old_fs);
365
366 return res;
367}
368
Al Viro7bb307e2013-02-23 14:51:48 -0500369ssize_t kernel_write(struct file *file, const char *buf, size_t count,
Miklos Szeredib2858d72009-05-19 11:37:46 +0200370 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200371{
372 mm_segment_t old_fs;
373 ssize_t res;
374
375 old_fs = get_fs();
376 set_fs(get_ds());
377 /* The cast to a user pointer is valid due to the set_fs() */
Al Viro7bb307e2013-02-23 14:51:48 -0500378 res = vfs_write(file, (__force const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200379 set_fs(old_fs);
380
381 return res;
382}
Al Viro7bb307e2013-02-23 14:51:48 -0500383EXPORT_SYMBOL(kernel_write);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200384
Al Viro82c156f2016-09-22 23:35:42 -0400385static ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
Miklos Szeredi68181732009-05-07 15:37:36 +0200386 struct pipe_inode_info *pipe, size_t len,
387 unsigned int flags)
388{
Al Viro523ac9a2016-09-23 15:34:57 -0400389 struct kvec *vec, __vec[PIPE_DEF_BUFFERS];
390 struct iov_iter to;
391 struct page **pages;
Miklos Szeredi68181732009-05-07 15:37:36 +0200392 unsigned int nr_pages;
Al Viro523ac9a2016-09-23 15:34:57 -0400393 size_t offset, dummy, copied = 0;
Miklos Szeredi68181732009-05-07 15:37:36 +0200394 ssize_t res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200395 int i;
Miklos Szeredi68181732009-05-07 15:37:36 +0200396
Al Viro523ac9a2016-09-23 15:34:57 -0400397 if (pipe->nrbufs == pipe->buffers)
398 return -EAGAIN;
399
400 /*
401 * Try to keep page boundaries matching to source pagecache ones -
402 * it probably won't be much help, but...
403 */
404 offset = *ppos & ~PAGE_MASK;
405
406 iov_iter_pipe(&to, ITER_PIPE | READ, pipe, len + offset);
407
408 res = iov_iter_get_pages_alloc(&to, &pages, len + offset, &dummy);
409 if (res <= 0)
Jens Axboe35f3d142010-05-20 10:43:18 +0200410 return -ENOMEM;
411
Al Viro8e54cad2016-11-26 20:05:42 -0500412 BUG_ON(dummy);
413 nr_pages = DIV_ROUND_UP(res, PAGE_SIZE);
Al Viro523ac9a2016-09-23 15:34:57 -0400414
Jens Axboe35f3d142010-05-20 10:43:18 +0200415 vec = __vec;
Al Viro523ac9a2016-09-23 15:34:57 -0400416 if (nr_pages > PIPE_DEF_BUFFERS) {
417 vec = kmalloc(nr_pages * sizeof(struct kvec), GFP_KERNEL);
418 if (unlikely(!vec)) {
419 res = -ENOMEM;
420 goto out;
421 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200422 }
423
Al Viro523ac9a2016-09-23 15:34:57 -0400424 pipe->bufs[to.idx].offset = offset;
425 pipe->bufs[to.idx].len -= offset;
Miklos Szeredi68181732009-05-07 15:37:36 +0200426
Al Viro523ac9a2016-09-23 15:34:57 -0400427 for (i = 0; i < nr_pages; i++) {
428 size_t this_len = min_t(size_t, len, PAGE_SIZE - offset);
429 vec[i].iov_base = page_address(pages[i]) + offset;
Miklos Szeredi68181732009-05-07 15:37:36 +0200430 vec[i].iov_len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200431 len -= this_len;
432 offset = 0;
433 }
434
Al Viro523ac9a2016-09-23 15:34:57 -0400435 res = kernel_readv(in, vec, nr_pages, *ppos);
436 if (res > 0) {
437 copied = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200438 *ppos += res;
Al Viro523ac9a2016-09-23 15:34:57 -0400439 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200440
Jens Axboe35f3d142010-05-20 10:43:18 +0200441 if (vec != __vec)
442 kfree(vec);
Al Viro523ac9a2016-09-23 15:34:57 -0400443out:
444 for (i = 0; i < nr_pages; i++)
445 put_page(pages[i]);
446 kvfree(pages);
447 iov_iter_advance(&to, copied); /* truncates and discards */
Miklos Szeredi68181732009-05-07 15:37:36 +0200448 return res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200449}
Miklos Szeredi68181732009-05-07 15:37:36 +0200450
Jens Axboe5274f052006-03-30 15:15:30 +0200451/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200452 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200453 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200454 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200455static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200456 struct pipe_buffer *buf, struct splice_desc *sd)
457{
Jens Axboe6a14b902007-06-14 13:08:55 +0200458 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200459 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100460 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200461
Al Viro72c2d532013-09-22 16:27:52 -0400462 if (!likely(file->f_op->sendpage))
Michał Mirosława8adbe32010-12-17 08:56:44 +0100463 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200464
Eric Dumazet35f9c092012-04-05 03:05:35 +0000465 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000466
467 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000468 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000469
Michał Mirosława8adbe32010-12-17 08:56:44 +0100470 return file->f_op->sendpage(file, buf->page, buf->offset,
471 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200472}
473
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200474static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
475{
476 smp_mb();
477 if (waitqueue_active(&pipe->wait))
478 wake_up_interruptible(&pipe->wait);
479 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
480}
481
482/**
483 * splice_from_pipe_feed - feed available data from a pipe to a file
484 * @pipe: pipe to splice from
485 * @sd: information to @actor
486 * @actor: handler that splices the data
487 *
488 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200489 * This function loops over the pipe and calls @actor to do the
490 * actual moving of a single struct pipe_buffer to the desired
491 * destination. It returns when there's no more buffers left in
492 * the pipe or if the requested number of bytes (@sd->total_len)
493 * have been copied. It returns a positive number (one) if the
494 * pipe needs to be filled with more data, zero if the required
495 * number of bytes have been copied and -errno on error.
496 *
497 * This, together with splice_from_pipe_{begin,end,next}, may be
498 * used to implement the functionality of __splice_from_pipe() when
499 * locking is required around copying the pipe buffers to the
500 * destination.
501 */
Al Viro96f9bc82014-04-05 04:35:49 -0400502static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200503 splice_actor *actor)
504{
505 int ret;
506
507 while (pipe->nrbufs) {
508 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200509
510 sd->len = buf->len;
511 if (sd->len > sd->total_len)
512 sd->len = sd->total_len;
513
Miklos Szeredifba597d2016-09-27 10:45:12 +0200514 ret = pipe_buf_confirm(pipe, buf);
Michał Mirosława8adbe32010-12-17 08:56:44 +0100515 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200516 if (ret == -ENODATA)
517 ret = 0;
518 return ret;
519 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100520
521 ret = actor(pipe, buf, sd);
522 if (ret <= 0)
523 return ret;
524
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200525 buf->offset += ret;
526 buf->len -= ret;
527
528 sd->num_spliced += ret;
529 sd->len -= ret;
530 sd->pos += ret;
531 sd->total_len -= ret;
532
533 if (!buf->len) {
Miklos Szeredia7796382016-09-27 10:45:12 +0200534 pipe_buf_release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200535 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200536 pipe->nrbufs--;
Al Viro6447a3c2013-03-21 11:01:38 -0400537 if (pipe->files)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200538 sd->need_wakeup = true;
539 }
540
541 if (!sd->total_len)
542 return 0;
543 }
544
545 return 1;
546}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200547
548/**
549 * splice_from_pipe_next - wait for some data to splice from
550 * @pipe: pipe to splice from
551 * @sd: information about the splice operation
552 *
553 * Description:
554 * This function will wait for some data and return a positive
555 * value (one) if pipe buffers are available. It will return zero
556 * or -errno if no more data needs to be spliced.
557 */
Al Viro96f9bc82014-04-05 04:35:49 -0400558static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200559{
Jan Karac725bfc2015-11-23 13:09:50 +0100560 /*
561 * Check for signal early to make process killable when there are
562 * always buffers available
563 */
564 if (signal_pending(current))
565 return -ERESTARTSYS;
566
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200567 while (!pipe->nrbufs) {
568 if (!pipe->writers)
569 return 0;
570
571 if (!pipe->waiting_writers && sd->num_spliced)
572 return 0;
573
574 if (sd->flags & SPLICE_F_NONBLOCK)
575 return -EAGAIN;
576
577 if (signal_pending(current))
578 return -ERESTARTSYS;
579
580 if (sd->need_wakeup) {
581 wakeup_pipe_writers(pipe);
582 sd->need_wakeup = false;
583 }
584
585 pipe_wait(pipe);
586 }
587
588 return 1;
589}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200590
591/**
592 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700593 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200594 *
595 * Description:
596 * This function should be called before a loop containing
597 * splice_from_pipe_next() and splice_from_pipe_feed() to
598 * initialize the necessary fields of @sd.
599 */
Al Viro96f9bc82014-04-05 04:35:49 -0400600static void splice_from_pipe_begin(struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200601{
602 sd->num_spliced = 0;
603 sd->need_wakeup = false;
604}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200605
606/**
607 * splice_from_pipe_end - finish splicing from pipe
608 * @pipe: pipe to splice from
609 * @sd: information about the splice operation
610 *
611 * Description:
612 * This function will wake up pipe writers if necessary. It should
613 * be called after a loop containing splice_from_pipe_next() and
614 * splice_from_pipe_feed().
615 */
Al Viro96f9bc82014-04-05 04:35:49 -0400616static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200617{
618 if (sd->need_wakeup)
619 wakeup_pipe_writers(pipe);
620}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200621
Jens Axboe932cc6d2007-06-21 13:10:21 +0200622/**
623 * __splice_from_pipe - splice data from a pipe to given actor
624 * @pipe: pipe to splice from
625 * @sd: information to @actor
626 * @actor: handler that splices the data
627 *
628 * Description:
629 * This function does little more than loop over the pipe and call
630 * @actor to do the actual moving of a single struct pipe_buffer to
631 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
632 * pipe_to_user.
633 *
Jens Axboe83f91352006-04-02 23:05:09 +0200634 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200635ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
636 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200637{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200638 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200639
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200640 splice_from_pipe_begin(sd);
641 do {
Jan Karac2489e02015-11-23 13:09:51 +0100642 cond_resched();
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200643 ret = splice_from_pipe_next(pipe, sd);
644 if (ret > 0)
645 ret = splice_from_pipe_feed(pipe, sd, actor);
646 } while (ret > 0);
647 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200648
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200649 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200650}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100651EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200652
Jens Axboe932cc6d2007-06-21 13:10:21 +0200653/**
654 * splice_from_pipe - splice data from a pipe to a file
655 * @pipe: pipe to splice from
656 * @out: file to splice to
657 * @ppos: position in @out
658 * @len: how many bytes to splice
659 * @flags: splice modifier flags
660 * @actor: handler that splices the data
661 *
662 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200663 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200664 * otherwise it's identical to __splice_from_pipe().
665 *
666 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200667ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
668 loff_t *ppos, size_t len, unsigned int flags,
669 splice_actor *actor)
670{
671 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200672 struct splice_desc sd = {
673 .total_len = len,
674 .flags = flags,
675 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200676 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200677 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200678
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200679 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200680 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200681 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200682
683 return ret;
684}
685
686/**
Al Viro8d020762014-04-05 04:27:08 -0400687 * iter_file_splice_write - splice data from a pipe to a file
688 * @pipe: pipe info
689 * @out: file to write to
690 * @ppos: position in @out
691 * @len: number of bytes to splice
692 * @flags: splice modifier flags
693 *
694 * Description:
695 * Will either move or copy pages (determined by @flags options) from
696 * the given pipe inode to the given file.
697 * This one is ->write_iter-based.
698 *
699 */
700ssize_t
701iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
702 loff_t *ppos, size_t len, unsigned int flags)
703{
704 struct splice_desc sd = {
705 .total_len = len,
706 .flags = flags,
707 .pos = *ppos,
708 .u.file = out,
709 };
710 int nbufs = pipe->buffers;
711 struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
712 GFP_KERNEL);
713 ssize_t ret;
714
715 if (unlikely(!array))
716 return -ENOMEM;
717
718 pipe_lock(pipe);
719
720 splice_from_pipe_begin(&sd);
721 while (sd.total_len) {
722 struct iov_iter from;
Al Viro8d020762014-04-05 04:27:08 -0400723 size_t left;
724 int n, idx;
725
726 ret = splice_from_pipe_next(pipe, &sd);
727 if (ret <= 0)
728 break;
729
730 if (unlikely(nbufs < pipe->buffers)) {
731 kfree(array);
732 nbufs = pipe->buffers;
733 array = kcalloc(nbufs, sizeof(struct bio_vec),
734 GFP_KERNEL);
735 if (!array) {
736 ret = -ENOMEM;
737 break;
738 }
739 }
740
741 /* build the vector */
742 left = sd.total_len;
743 for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) {
744 struct pipe_buffer *buf = pipe->bufs + idx;
745 size_t this_len = buf->len;
746
747 if (this_len > left)
748 this_len = left;
749
750 if (idx == pipe->buffers - 1)
751 idx = -1;
752
Miklos Szeredifba597d2016-09-27 10:45:12 +0200753 ret = pipe_buf_confirm(pipe, buf);
Al Viro8d020762014-04-05 04:27:08 -0400754 if (unlikely(ret)) {
755 if (ret == -ENODATA)
756 ret = 0;
757 goto done;
758 }
759
760 array[n].bv_page = buf->page;
761 array[n].bv_len = this_len;
762 array[n].bv_offset = buf->offset;
763 left -= this_len;
764 }
765
Al Viro05afcb72015-01-23 01:08:07 -0500766 iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n,
767 sd.total_len - left);
Christoph Hellwigdbe4e192015-01-25 21:11:59 +0100768 ret = vfs_iter_write(out, &from, &sd.pos);
Al Viro8d020762014-04-05 04:27:08 -0400769 if (ret <= 0)
770 break;
771
772 sd.num_spliced += ret;
773 sd.total_len -= ret;
Christoph Hellwigdbe4e192015-01-25 21:11:59 +0100774 *ppos = sd.pos;
Al Viro8d020762014-04-05 04:27:08 -0400775
776 /* dismiss the fully eaten buffers, adjust the partial one */
777 while (ret) {
778 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
779 if (ret >= buf->len) {
Al Viro8d020762014-04-05 04:27:08 -0400780 ret -= buf->len;
781 buf->len = 0;
Miklos Szeredia7796382016-09-27 10:45:12 +0200782 pipe_buf_release(pipe, buf);
Al Viro8d020762014-04-05 04:27:08 -0400783 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
784 pipe->nrbufs--;
785 if (pipe->files)
786 sd.need_wakeup = true;
787 } else {
788 buf->offset += ret;
789 buf->len -= ret;
790 ret = 0;
791 }
792 }
793 }
794done:
795 kfree(array);
796 splice_from_pipe_end(pipe, &sd);
797
798 pipe_unlock(pipe);
799
800 if (sd.num_spliced)
801 ret = sd.num_spliced;
802
803 return ret;
804}
805
806EXPORT_SYMBOL(iter_file_splice_write);
807
Miklos Szeredib2858d72009-05-19 11:37:46 +0200808static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
809 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200810{
Miklos Szeredib2858d72009-05-19 11:37:46 +0200811 int ret;
812 void *data;
Al Viro06ae43f2013-03-20 13:19:30 -0400813 loff_t tmp = sd->pos;
Miklos Szeredib2858d72009-05-19 11:37:46 +0200814
Al Virofbb32752014-02-02 21:09:54 -0500815 data = kmap(buf->page);
Al Viro06ae43f2013-03-20 13:19:30 -0400816 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
Al Virofbb32752014-02-02 21:09:54 -0500817 kunmap(buf->page);
Miklos Szeredib2858d72009-05-19 11:37:46 +0200818
819 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200820}
821
822static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
823 struct file *out, loff_t *ppos,
824 size_t len, unsigned int flags)
825{
Miklos Szeredib2858d72009-05-19 11:37:46 +0200826 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200827
Miklos Szeredib2858d72009-05-19 11:37:46 +0200828 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
829 if (ret > 0)
830 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200831
Miklos Szeredib2858d72009-05-19 11:37:46 +0200832 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200833}
834
Jens Axboe83f91352006-04-02 23:05:09 +0200835/**
836 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +0200837 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +0200838 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200839 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200840 * @len: number of bytes to splice
841 * @flags: splice modifier flags
842 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200843 * Description:
844 * Will send @len bytes from the pipe to a network socket. No data copying
845 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +0200846 *
847 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200848ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200849 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200850{
Jens Axboe00522fb2006-04-26 14:39:29 +0200851 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +0200852}
853
Jens Axboe059a8f32006-04-02 23:06:05 +0200854EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -0500855
Jens Axboe83f91352006-04-02 23:05:09 +0200856/*
857 * Attempt to initiate a splice from pipe to file.
858 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200859static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200860 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200861{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200862 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
863 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +0200864
Al Viro72c2d532013-09-22 16:27:52 -0400865 if (out->f_op->splice_write)
Changli Gaocc56f7d2009-11-04 09:09:52 +0100866 splice_write = out->f_op->splice_write;
867 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200868 splice_write = default_file_splice_write;
869
Al Viro500368f2013-05-23 20:07:11 -0400870 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200871}
872
Jens Axboe83f91352006-04-02 23:05:09 +0200873/*
874 * Attempt to initiate a splice from a file to a pipe.
875 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200876static long do_splice_to(struct file *in, loff_t *ppos,
877 struct pipe_inode_info *pipe, size_t len,
878 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200879{
Miklos Szeredi68181732009-05-07 15:37:36 +0200880 ssize_t (*splice_read)(struct file *, loff_t *,
881 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +0200882 int ret;
883
Jens Axboe49570e92006-04-11 13:56:09 +0200884 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +0200885 return -EBADF;
886
Jens Axboecbb7e572006-04-11 14:57:50 +0200887 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200888 if (unlikely(ret < 0))
889 return ret;
890
Al Viro03cc0782016-04-02 14:56:58 -0400891 if (unlikely(len > MAX_RW_COUNT))
892 len = MAX_RW_COUNT;
893
Al Viro72c2d532013-09-22 16:27:52 -0400894 if (in->f_op->splice_read)
Changli Gaocc56f7d2009-11-04 09:09:52 +0100895 splice_read = in->f_op->splice_read;
896 else
Miklos Szeredi68181732009-05-07 15:37:36 +0200897 splice_read = default_file_splice_read;
898
899 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200900}
901
Jens Axboe932cc6d2007-06-21 13:10:21 +0200902/**
903 * splice_direct_to_actor - splices data directly between two non-pipes
904 * @in: file to splice from
905 * @sd: actor information on where to splice to
906 * @actor: handles the data splicing
907 *
908 * Description:
909 * This is a special case helper to splice directly between two
910 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +0200911 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +0200912 * that process.
913 *
Jens Axboec66ab6f2007-06-12 21:17:17 +0200914 */
915ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
916 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +0200917{
918 struct pipe_inode_info *pipe;
919 long ret, bytes;
920 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200921 size_t len;
Christophe Leroy0ff28d92015-05-06 17:26:47 +0200922 int i, flags, more;
Jens Axboeb92ce552006-04-11 13:52:07 +0200923
924 /*
925 * We require the input being a regular file, as we don't want to
926 * randomly drop data for eg socket -> socket splicing. Use the
927 * piped splicing for that!
928 */
Al Viro496ad9a2013-01-23 17:07:38 -0500929 i_mode = file_inode(in)->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +0200930 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
931 return -EINVAL;
932
933 /*
934 * neither in nor out is a pipe, setup an internal pipe attached to
935 * 'out' and transfer the wanted data from 'in' to 'out' through that
936 */
937 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +0200938 if (unlikely(!pipe)) {
Al Viro7bee1302013-03-21 11:04:15 -0400939 pipe = alloc_pipe_info();
Jens Axboeb92ce552006-04-11 13:52:07 +0200940 if (!pipe)
941 return -ENOMEM;
942
943 /*
944 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +0200945 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +0200946 * PIPE_READERS appropriately.
947 */
948 pipe->readers = 1;
949
950 current->splice_pipe = pipe;
951 }
952
953 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200954 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +0200955 */
956 ret = 0;
957 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200958 len = sd->total_len;
959 flags = sd->flags;
960
961 /*
962 * Don't block on output, we have to drain the direct pipe.
963 */
964 sd->flags &= ~SPLICE_F_NONBLOCK;
Christophe Leroy0ff28d92015-05-06 17:26:47 +0200965 more = sd->flags & SPLICE_F_MORE;
Jens Axboeb92ce552006-04-11 13:52:07 +0200966
967 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +0200968 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +0200969 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +0200970
Jens Axboebcd4f3a2007-07-16 14:41:49 +0200971 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +0200972 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +0200973 goto out_release;
974
975 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200976 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +0200977
978 /*
Christophe Leroy0ff28d92015-05-06 17:26:47 +0200979 * If more data is pending, set SPLICE_F_MORE
980 * If this is the last data and SPLICE_F_MORE was not set
981 * initially, clears it.
982 */
983 if (read_len < len)
984 sd->flags |= SPLICE_F_MORE;
985 else if (!more)
986 sd->flags &= ~SPLICE_F_MORE;
987 /*
Jens Axboeb92ce552006-04-11 13:52:07 +0200988 * NOTE: nonblocking mode only applies to the input. We
989 * must not do the output in nonblocking mode as then we
990 * could get stuck data in the internal pipe:
991 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200992 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +0200993 if (unlikely(ret <= 0)) {
994 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +0200995 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +0200996 }
Jens Axboeb92ce552006-04-11 13:52:07 +0200997
998 bytes += ret;
999 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001000 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001001
Tom Zanussia82c53a2008-05-09 13:28:36 +02001002 if (ret < read_len) {
1003 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001004 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001005 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001006 }
1007
Jens Axboe9e971982008-01-29 21:05:57 +01001008done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001009 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001010 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001011 return bytes;
1012
1013out_release:
1014 /*
1015 * If we did an incomplete transfer we must release
1016 * the pipe buffers in question:
1017 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001018 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001019 struct pipe_buffer *buf = pipe->bufs + i;
1020
Miklos Szeredia7796382016-09-27 10:45:12 +02001021 if (buf->ops)
1022 pipe_buf_release(pipe, buf);
Jens Axboeb92ce552006-04-11 13:52:07 +02001023 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001024
Jens Axboe9e971982008-01-29 21:05:57 +01001025 if (!bytes)
1026 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001027
Jens Axboe9e971982008-01-29 21:05:57 +01001028 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001029}
1030EXPORT_SYMBOL(splice_direct_to_actor);
1031
1032static int direct_splice_actor(struct pipe_inode_info *pipe,
1033 struct splice_desc *sd)
1034{
Jens Axboe6a14b902007-06-14 13:08:55 +02001035 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001036
Al Viro7995bd22013-06-20 18:58:36 +04001037 return do_splice_from(pipe, file, sd->opos, sd->total_len,
Changli Gao2cb4b052010-06-29 13:09:18 +02001038 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001039}
1040
Jens Axboe932cc6d2007-06-21 13:10:21 +02001041/**
1042 * do_splice_direct - splices data directly between two files
1043 * @in: file to splice from
1044 * @ppos: input file offset
1045 * @out: file to splice to
Randy Dunlapacdb37c2013-06-22 19:44:08 -07001046 * @opos: output file offset
Jens Axboe932cc6d2007-06-21 13:10:21 +02001047 * @len: number of bytes to splice
1048 * @flags: splice modifier flags
1049 *
1050 * Description:
1051 * For use by do_sendfile(). splice can easily emulate sendfile, but
1052 * doing it in the application would incur an extra system call
1053 * (splice in + splice out, as compared to just sendfile()). So this helper
1054 * can splice directly through a process-private pipe.
1055 *
1056 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001057long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
Al Viro7995bd22013-06-20 18:58:36 +04001058 loff_t *opos, size_t len, unsigned int flags)
Jens Axboec66ab6f2007-06-12 21:17:17 +02001059{
1060 struct splice_desc sd = {
1061 .len = len,
1062 .total_len = len,
1063 .flags = flags,
1064 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001065 .u.file = out,
Al Viro7995bd22013-06-20 18:58:36 +04001066 .opos = opos,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001067 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001068 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001069
Al Viro18c67cb2013-06-19 15:41:54 +04001070 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1071 return -EBADF;
1072
1073 if (unlikely(out->f_flags & O_APPEND))
1074 return -EINVAL;
1075
1076 ret = rw_verify_area(WRITE, out, opos, len);
1077 if (unlikely(ret < 0))
1078 return ret;
1079
Jens Axboec66ab6f2007-06-12 21:17:17 +02001080 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001081 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001082 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001083
Jens Axboec66ab6f2007-06-12 21:17:17 +02001084 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001085}
Miklos Szeredi1c118592014-10-24 00:14:35 +02001086EXPORT_SYMBOL(do_splice_direct);
Jens Axboeb92ce552006-04-11 13:52:07 +02001087
Al Viro8924fef2016-09-17 20:44:45 -04001088static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags)
1089{
Linus Torvaldsb6cce9b2016-12-21 10:59:34 -08001090 for (;;) {
1091 if (unlikely(!pipe->readers)) {
1092 send_sig(SIGPIPE, current, 0);
1093 return -EPIPE;
1094 }
1095 if (pipe->nrbufs != pipe->buffers)
1096 return 0;
Al Viro8924fef2016-09-17 20:44:45 -04001097 if (flags & SPLICE_F_NONBLOCK)
1098 return -EAGAIN;
1099 if (signal_pending(current))
1100 return -ERESTARTSYS;
1101 pipe->waiting_writers++;
1102 pipe_wait(pipe);
1103 pipe->waiting_writers--;
1104 }
Al Viro8924fef2016-09-17 20:44:45 -04001105}
1106
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001107static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1108 struct pipe_inode_info *opipe,
1109 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001110
1111/*
Jens Axboe83f91352006-04-02 23:05:09 +02001112 * Determine where to splice to/from.
1113 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001114static long do_splice(struct file *in, loff_t __user *off_in,
1115 struct file *out, loff_t __user *off_out,
1116 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001117{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001118 struct pipe_inode_info *ipipe;
1119 struct pipe_inode_info *opipe;
Al Viro7995bd22013-06-20 18:58:36 +04001120 loff_t offset;
Jens Axboea4514eb2006-04-19 15:57:05 +02001121 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001122
Linus Torvalds71993e62010-11-28 13:56:09 -08001123 ipipe = get_pipe_info(in);
1124 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001125
1126 if (ipipe && opipe) {
1127 if (off_in || off_out)
1128 return -ESPIPE;
1129
1130 if (!(in->f_mode & FMODE_READ))
1131 return -EBADF;
1132
1133 if (!(out->f_mode & FMODE_WRITE))
1134 return -EBADF;
1135
1136 /* Splicing to self would be fun, but... */
1137 if (ipipe == opipe)
1138 return -EINVAL;
1139
1140 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1141 }
1142
1143 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001144 if (off_in)
1145 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001146 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001147 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001148 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001149 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001150 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001151 } else {
1152 offset = out->f_pos;
1153 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001154
Al Viro18c67cb2013-06-19 15:41:54 +04001155 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1156 return -EBADF;
1157
1158 if (unlikely(out->f_flags & O_APPEND))
1159 return -EINVAL;
1160
1161 ret = rw_verify_area(WRITE, out, &offset, len);
1162 if (unlikely(ret < 0))
1163 return ret;
1164
Al Viro500368f2013-05-23 20:07:11 -04001165 file_start_write(out);
Al Viro7995bd22013-06-20 18:58:36 +04001166 ret = do_splice_from(ipipe, out, &offset, len, flags);
Al Viro500368f2013-05-23 20:07:11 -04001167 file_end_write(out);
Jens Axboea4514eb2006-04-19 15:57:05 +02001168
Al Viro7995bd22013-06-20 18:58:36 +04001169 if (!off_out)
1170 out->f_pos = offset;
1171 else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001172 ret = -EFAULT;
1173
1174 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001175 }
Jens Axboe5274f052006-03-30 15:15:30 +02001176
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001177 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001178 if (off_out)
1179 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001180 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001181 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001182 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001183 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001184 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001185 } else {
1186 offset = in->f_pos;
1187 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001188
Al Viro8924fef2016-09-17 20:44:45 -04001189 pipe_lock(opipe);
1190 ret = wait_for_space(opipe, flags);
1191 if (!ret)
1192 ret = do_splice_to(in, &offset, opipe, len, flags);
1193 pipe_unlock(opipe);
1194 if (ret > 0)
1195 wakeup_pipe_readers(opipe);
Al Viro7995bd22013-06-20 18:58:36 +04001196 if (!off_in)
1197 in->f_pos = offset;
1198 else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001199 ret = -EFAULT;
1200
1201 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001202 }
Jens Axboe5274f052006-03-30 15:15:30 +02001203
1204 return -EINVAL;
1205}
1206
Al Viro79fddc42016-09-17 22:38:20 -04001207static int iter_to_pipe(struct iov_iter *from,
1208 struct pipe_inode_info *pipe,
1209 unsigned flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001210{
Al Viro79fddc42016-09-17 22:38:20 -04001211 struct pipe_buffer buf = {
1212 .ops = &user_page_pipe_buf_ops,
1213 .flags = flags
1214 };
1215 size_t total = 0;
1216 int ret = 0;
1217 bool failed = false;
1218
1219 while (iov_iter_count(from) && !failed) {
1220 struct page *pages[16];
Al Virodb85a9e2016-09-17 20:25:06 -04001221 ssize_t copied;
1222 size_t start;
Al Viro79fddc42016-09-17 22:38:20 -04001223 int n;
Jens Axboe912d35f2006-04-26 10:59:21 +02001224
Al Viro79fddc42016-09-17 22:38:20 -04001225 copied = iov_iter_get_pages(from, pages, ~0UL, 16, &start);
1226 if (copied <= 0) {
1227 ret = copied;
1228 break;
1229 }
Jens Axboe912d35f2006-04-26 10:59:21 +02001230
Al Viro79fddc42016-09-17 22:38:20 -04001231 for (n = 0; copied; n++, start = 0) {
Al Virodb85a9e2016-09-17 20:25:06 -04001232 int size = min_t(int, copied, PAGE_SIZE - start);
Al Viro79fddc42016-09-17 22:38:20 -04001233 if (!failed) {
1234 buf.page = pages[n];
1235 buf.offset = start;
1236 buf.len = size;
1237 ret = add_to_pipe(pipe, &buf);
1238 if (unlikely(ret < 0)) {
1239 failed = true;
1240 } else {
1241 iov_iter_advance(from, ret);
1242 total += ret;
1243 }
1244 } else {
1245 put_page(pages[n]);
1246 }
Al Virodb85a9e2016-09-17 20:25:06 -04001247 copied -= size;
Jens Axboe912d35f2006-04-26 10:59:21 +02001248 }
Jens Axboe912d35f2006-04-26 10:59:21 +02001249 }
Al Viro79fddc42016-09-17 22:38:20 -04001250 return total ? total : ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001251}
1252
Jens Axboe6a14b902007-06-14 13:08:55 +02001253static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1254 struct splice_desc *sd)
1255{
Al Viro6130f532014-02-03 18:19:51 -05001256 int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1257 return n == sd->len ? n : -EFAULT;
Jens Axboe6a14b902007-06-14 13:08:55 +02001258}
1259
1260/*
1261 * For lack of a better implementation, implement vmsplice() to userspace
1262 * as a simple copy of the pipes pages to the user iov.
1263 */
Al Viro6130f532014-02-03 18:19:51 -05001264static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov,
Jens Axboe6a14b902007-06-14 13:08:55 +02001265 unsigned long nr_segs, unsigned int flags)
1266{
1267 struct pipe_inode_info *pipe;
1268 struct splice_desc sd;
Jens Axboe6a14b902007-06-14 13:08:55 +02001269 long ret;
Al Viro6130f532014-02-03 18:19:51 -05001270 struct iovec iovstack[UIO_FASTIOV];
1271 struct iovec *iov = iovstack;
1272 struct iov_iter iter;
Jens Axboe6a14b902007-06-14 13:08:55 +02001273
Linus Torvalds71993e62010-11-28 13:56:09 -08001274 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001275 if (!pipe)
1276 return -EBADF;
1277
Al Viro345995f2015-03-21 19:17:55 -04001278 ret = import_iovec(READ, uiov, nr_segs,
1279 ARRAY_SIZE(iovstack), &iov, &iter);
1280 if (ret < 0)
1281 return ret;
Al Viro6130f532014-02-03 18:19:51 -05001282
Al Viro345995f2015-03-21 19:17:55 -04001283 sd.total_len = iov_iter_count(&iter);
Al Viro6130f532014-02-03 18:19:51 -05001284 sd.len = 0;
Al Viro6130f532014-02-03 18:19:51 -05001285 sd.flags = flags;
1286 sd.u.data = &iter;
1287 sd.pos = 0;
1288
Al Viro345995f2015-03-21 19:17:55 -04001289 if (sd.total_len) {
1290 pipe_lock(pipe);
1291 ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
1292 pipe_unlock(pipe);
1293 }
Jens Axboe6a14b902007-06-14 13:08:55 +02001294
Al Viro345995f2015-03-21 19:17:55 -04001295 kfree(iov);
Jens Axboe6a14b902007-06-14 13:08:55 +02001296 return ret;
1297}
1298
Jens Axboe912d35f2006-04-26 10:59:21 +02001299/*
1300 * vmsplice splices a user address range into a pipe. It can be thought of
1301 * as splice-from-memory, where the regular splice is splice-from-file (or
1302 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001303 */
Al Virodb85a9e2016-09-17 20:25:06 -04001304static long vmsplice_to_pipe(struct file *file, const struct iovec __user *uiov,
Jens Axboe6a14b902007-06-14 13:08:55 +02001305 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001306{
Jens Axboeddac0d32006-11-04 12:49:32 +01001307 struct pipe_inode_info *pipe;
Al Virodb85a9e2016-09-17 20:25:06 -04001308 struct iovec iovstack[UIO_FASTIOV];
1309 struct iovec *iov = iovstack;
1310 struct iov_iter from;
Jens Axboe35f3d142010-05-20 10:43:18 +02001311 long ret;
Al Viro79fddc42016-09-17 22:38:20 -04001312 unsigned buf_flag = 0;
1313
1314 if (flags & SPLICE_F_GIFT)
1315 buf_flag = PIPE_BUF_FLAG_GIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +02001316
Linus Torvalds71993e62010-11-28 13:56:09 -08001317 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001318 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001319 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001320
Al Virodb85a9e2016-09-17 20:25:06 -04001321 ret = import_iovec(WRITE, uiov, nr_segs,
1322 ARRAY_SIZE(iovstack), &iov, &from);
1323 if (ret < 0)
1324 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001325
Al Viro8924fef2016-09-17 20:44:45 -04001326 pipe_lock(pipe);
1327 ret = wait_for_space(pipe, flags);
Al Viro79fddc42016-09-17 22:38:20 -04001328 if (!ret)
1329 ret = iter_to_pipe(&from, pipe, buf_flag);
Al Viro8924fef2016-09-17 20:44:45 -04001330 pipe_unlock(pipe);
1331 if (ret > 0)
1332 wakeup_pipe_readers(pipe);
Al Virodb85a9e2016-09-17 20:25:06 -04001333 kfree(iov);
Jens Axboe35f3d142010-05-20 10:43:18 +02001334 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001335}
1336
Jens Axboe6a14b902007-06-14 13:08:55 +02001337/*
1338 * Note that vmsplice only really supports true splicing _from_ user memory
1339 * to a pipe, not the other way around. Splicing from user memory is a simple
1340 * operation that can be supported without any funky alignment restrictions
1341 * or nasty vm tricks. We simply map in the user memory and fill them into
1342 * a pipe. The reverse isn't quite as easy, though. There are two possible
1343 * solutions for that:
1344 *
1345 * - memcpy() the data internally, at which point we might as well just
1346 * do a regular read() on the buffer anyway.
1347 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1348 * has restriction limitations on both ends of the pipe).
1349 *
1350 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1351 *
1352 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001353SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1354 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001355{
Al Viro2903ff02012-08-28 12:52:22 -04001356 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001357 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001358
Jens Axboe6a14b902007-06-14 13:08:55 +02001359 if (unlikely(nr_segs > UIO_MAXIOV))
1360 return -EINVAL;
1361 else if (unlikely(!nr_segs))
1362 return 0;
1363
Jens Axboe912d35f2006-04-26 10:59:21 +02001364 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001365 f = fdget(fd);
1366 if (f.file) {
1367 if (f.file->f_mode & FMODE_WRITE)
1368 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1369 else if (f.file->f_mode & FMODE_READ)
1370 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001371
Al Viro2903ff02012-08-28 12:52:22 -04001372 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001373 }
1374
1375 return error;
1376}
1377
Al Viro76b021d2013-03-02 10:19:56 -05001378#ifdef CONFIG_COMPAT
1379COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1380 unsigned int, nr_segs, unsigned int, flags)
1381{
1382 unsigned i;
1383 struct iovec __user *iov;
1384 if (nr_segs > UIO_MAXIOV)
1385 return -EINVAL;
1386 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1387 for (i = 0; i < nr_segs; i++) {
1388 struct compat_iovec v;
1389 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1390 get_user(v.iov_len, &iov32[i].iov_len) ||
1391 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1392 put_user(v.iov_len, &iov[i].iov_len))
1393 return -EFAULT;
1394 }
1395 return sys_vmsplice(fd, iov, nr_segs, flags);
1396}
1397#endif
1398
Heiko Carstens836f92a2009-01-14 14:14:33 +01001399SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1400 int, fd_out, loff_t __user *, off_out,
1401 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001402{
Al Viro2903ff02012-08-28 12:52:22 -04001403 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001404 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001405
1406 if (unlikely(!len))
1407 return 0;
1408
1409 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001410 in = fdget(fd_in);
1411 if (in.file) {
1412 if (in.file->f_mode & FMODE_READ) {
1413 out = fdget(fd_out);
1414 if (out.file) {
1415 if (out.file->f_mode & FMODE_WRITE)
1416 error = do_splice(in.file, off_in,
1417 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001418 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001419 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001420 }
1421 }
Al Viro2903ff02012-08-28 12:52:22 -04001422 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001423 }
Jens Axboe5274f052006-03-30 15:15:30 +02001424 return error;
1425}
Jens Axboe70524492006-04-11 15:51:17 +02001426
1427/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001428 * Make sure there's data to read. Wait for input if we can, otherwise
1429 * return an appropriate error.
1430 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001431static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001432{
1433 int ret;
1434
1435 /*
1436 * Check ->nrbufs without the inode lock first. This function
1437 * is speculative anyways, so missing one is ok.
1438 */
1439 if (pipe->nrbufs)
1440 return 0;
1441
1442 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001443 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001444
1445 while (!pipe->nrbufs) {
1446 if (signal_pending(current)) {
1447 ret = -ERESTARTSYS;
1448 break;
1449 }
1450 if (!pipe->writers)
1451 break;
1452 if (!pipe->waiting_writers) {
1453 if (flags & SPLICE_F_NONBLOCK) {
1454 ret = -EAGAIN;
1455 break;
1456 }
1457 }
1458 pipe_wait(pipe);
1459 }
1460
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001461 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001462 return ret;
1463}
1464
1465/*
1466 * Make sure there's writeable room. Wait for room if we can, otherwise
1467 * return an appropriate error.
1468 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001469static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001470{
1471 int ret;
1472
1473 /*
1474 * Check ->nrbufs without the inode lock first. This function
1475 * is speculative anyways, so missing one is ok.
1476 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001477 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001478 return 0;
1479
1480 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001481 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001482
Jens Axboe35f3d142010-05-20 10:43:18 +02001483 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001484 if (!pipe->readers) {
1485 send_sig(SIGPIPE, current, 0);
1486 ret = -EPIPE;
1487 break;
1488 }
1489 if (flags & SPLICE_F_NONBLOCK) {
1490 ret = -EAGAIN;
1491 break;
1492 }
1493 if (signal_pending(current)) {
1494 ret = -ERESTARTSYS;
1495 break;
1496 }
1497 pipe->waiting_writers++;
1498 pipe_wait(pipe);
1499 pipe->waiting_writers--;
1500 }
1501
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001502 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001503 return ret;
1504}
1505
1506/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001507 * Splice contents of ipipe to opipe.
1508 */
1509static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1510 struct pipe_inode_info *opipe,
1511 size_t len, unsigned int flags)
1512{
1513 struct pipe_buffer *ibuf, *obuf;
1514 int ret = 0, nbuf;
1515 bool input_wakeup = false;
1516
1517
1518retry:
1519 ret = ipipe_prep(ipipe, flags);
1520 if (ret)
1521 return ret;
1522
1523 ret = opipe_prep(opipe, flags);
1524 if (ret)
1525 return ret;
1526
1527 /*
1528 * Potential ABBA deadlock, work around it by ordering lock
1529 * grabbing by pipe info address. Otherwise two different processes
1530 * could deadlock (one doing tee from A -> B, the other from B -> A).
1531 */
1532 pipe_double_lock(ipipe, opipe);
1533
1534 do {
1535 if (!opipe->readers) {
1536 send_sig(SIGPIPE, current, 0);
1537 if (!ret)
1538 ret = -EPIPE;
1539 break;
1540 }
1541
1542 if (!ipipe->nrbufs && !ipipe->writers)
1543 break;
1544
1545 /*
1546 * Cannot make any progress, because either the input
1547 * pipe is empty or the output pipe is full.
1548 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001549 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001550 /* Already processed some buffers, break */
1551 if (ret)
1552 break;
1553
1554 if (flags & SPLICE_F_NONBLOCK) {
1555 ret = -EAGAIN;
1556 break;
1557 }
1558
1559 /*
1560 * We raced with another reader/writer and haven't
1561 * managed to process any buffers. A zero return
1562 * value means EOF, so retry instead.
1563 */
1564 pipe_unlock(ipipe);
1565 pipe_unlock(opipe);
1566 goto retry;
1567 }
1568
1569 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001570 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001571 obuf = opipe->bufs + nbuf;
1572
1573 if (len >= ibuf->len) {
1574 /*
1575 * Simply move the whole buffer from ipipe to opipe
1576 */
1577 *obuf = *ibuf;
1578 ibuf->ops = NULL;
1579 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001580 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001581 ipipe->nrbufs--;
1582 input_wakeup = true;
1583 } else {
1584 /*
1585 * Get a reference to this pipe buffer,
1586 * so we can copy the contents over.
1587 */
Miklos Szeredi7bf2d1d2016-09-27 10:45:12 +02001588 pipe_buf_get(ipipe, ibuf);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001589 *obuf = *ibuf;
1590
1591 /*
1592 * Don't inherit the gift flag, we need to
1593 * prevent multiple steals of this page.
1594 */
1595 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1596
Jann Horn943ebf42019-01-23 15:19:17 +01001597 pipe_buf_mark_unmergeable(obuf);
1598
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001599 obuf->len = len;
1600 opipe->nrbufs++;
1601 ibuf->offset += obuf->len;
1602 ibuf->len -= obuf->len;
1603 }
1604 ret += obuf->len;
1605 len -= obuf->len;
1606 } while (len);
1607
1608 pipe_unlock(ipipe);
1609 pipe_unlock(opipe);
1610
1611 /*
1612 * If we put data in the output pipe, wakeup any potential readers.
1613 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001614 if (ret > 0)
1615 wakeup_pipe_readers(opipe);
1616
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001617 if (input_wakeup)
1618 wakeup_pipe_writers(ipipe);
1619
1620 return ret;
1621}
1622
1623/*
Jens Axboe70524492006-04-11 15:51:17 +02001624 * Link contents of ipipe to opipe.
1625 */
1626static int link_pipe(struct pipe_inode_info *ipipe,
1627 struct pipe_inode_info *opipe,
1628 size_t len, unsigned int flags)
1629{
1630 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001631 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001632
1633 /*
1634 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001635 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001636 * could deadlock (one doing tee from A -> B, the other from B -> A).
1637 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001638 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001639
Jens Axboeaadd06e2006-07-10 11:00:01 +02001640 do {
Jens Axboe70524492006-04-11 15:51:17 +02001641 if (!opipe->readers) {
1642 send_sig(SIGPIPE, current, 0);
1643 if (!ret)
1644 ret = -EPIPE;
1645 break;
1646 }
Jens Axboe70524492006-04-11 15:51:17 +02001647
1648 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001649 * If we have iterated all input buffers or ran out of
1650 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001651 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001652 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001653 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001654
Jens Axboe35f3d142010-05-20 10:43:18 +02001655 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1656 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001657
Jens Axboe2a27250e2006-04-19 15:56:40 +02001658 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001659 * Get a reference to this pipe buffer,
1660 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001661 */
Miklos Szeredi7bf2d1d2016-09-27 10:45:12 +02001662 pipe_buf_get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001663
Jens Axboeaadd06e2006-07-10 11:00:01 +02001664 obuf = opipe->bufs + nbuf;
1665 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001666
Jens Axboeaadd06e2006-07-10 11:00:01 +02001667 /*
1668 * Don't inherit the gift flag, we need to
1669 * prevent multiple steals of this page.
1670 */
1671 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1672
Jann Horn943ebf42019-01-23 15:19:17 +01001673 pipe_buf_mark_unmergeable(obuf);
1674
Jens Axboeaadd06e2006-07-10 11:00:01 +02001675 if (obuf->len > len)
1676 obuf->len = len;
1677
1678 opipe->nrbufs++;
1679 ret += obuf->len;
1680 len -= obuf->len;
1681 i++;
1682 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001683
Jens Axboe02cf01a2008-02-20 10:34:51 +01001684 /*
1685 * return EAGAIN if we have the potential of some data in the
1686 * future, otherwise just return 0
1687 */
1688 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1689 ret = -EAGAIN;
1690
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001691 pipe_unlock(ipipe);
1692 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001693
Jens Axboeaadd06e2006-07-10 11:00:01 +02001694 /*
1695 * If we put data in the output pipe, wakeup any potential readers.
1696 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001697 if (ret > 0)
1698 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001699
1700 return ret;
1701}
1702
1703/*
1704 * This is a tee(1) implementation that works on pipes. It doesn't copy
1705 * any data, it simply references the 'in' pages on the 'out' pipe.
1706 * The 'flags' used are the SPLICE_F_* variants, currently the only
1707 * applicable one is SPLICE_F_NONBLOCK.
1708 */
1709static long do_tee(struct file *in, struct file *out, size_t len,
1710 unsigned int flags)
1711{
Linus Torvalds71993e62010-11-28 13:56:09 -08001712 struct pipe_inode_info *ipipe = get_pipe_info(in);
1713 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001714 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001715
1716 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001717 * Duplicate the contents of ipipe to opipe without actually
1718 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001719 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001720 if (ipipe && opipe && ipipe != opipe) {
1721 /*
1722 * Keep going, unless we encounter an error. The ipipe/opipe
1723 * ordering doesn't really matter.
1724 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001725 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001726 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001727 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01001728 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001729 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001730 }
1731 }
Jens Axboe70524492006-04-11 15:51:17 +02001732
Jens Axboeaadd06e2006-07-10 11:00:01 +02001733 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02001734}
1735
Heiko Carstens836f92a2009-01-14 14:14:33 +01001736SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02001737{
Al Viro2903ff02012-08-28 12:52:22 -04001738 struct fd in;
1739 int error;
Jens Axboe70524492006-04-11 15:51:17 +02001740
1741 if (unlikely(!len))
1742 return 0;
1743
1744 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001745 in = fdget(fdin);
1746 if (in.file) {
1747 if (in.file->f_mode & FMODE_READ) {
1748 struct fd out = fdget(fdout);
1749 if (out.file) {
1750 if (out.file->f_mode & FMODE_WRITE)
1751 error = do_tee(in.file, out.file,
1752 len, flags);
1753 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02001754 }
1755 }
Al Viro2903ff02012-08-28 12:52:22 -04001756 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02001757 }
1758
1759 return error;
1760}