blob: 6909d89d0da56ffd929a2114b0228a03d1f35589 [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>
Jens Axboe5274f052006-03-30 15:15:30 +020034
Jens Axboe83f91352006-04-02 23:05:09 +020035/*
36 * Attempt to steal a page from a pipe buffer. This should perhaps go into
37 * a vm helper function, it's already simplified quite a bit by the
38 * addition of remove_mapping(). If success is returned, the caller may
39 * attempt to reuse this page for another destination.
40 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020041static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020042 struct pipe_buffer *buf)
43{
44 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020045 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020046
Jens Axboe9e0267c2006-04-19 15:57:31 +020047 lock_page(page);
48
Jens Axboe9e94cd42006-06-20 15:01:12 +020049 mapping = page_mapping(page);
50 if (mapping) {
51 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020052
Jens Axboe9e94cd42006-06-20 15:01:12 +020053 /*
54 * At least for ext2 with nobh option, we need to wait on
55 * writeback completing on this page, since we'll remove it
56 * from the pagecache. Otherwise truncate wont wait on the
57 * page, allowing the disk blocks to be reused by someone else
58 * before we actually wrote our data to them. fs corruption
59 * ensues.
60 */
61 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020062
David Howells266cf652009-04-03 16:42:36 +010063 if (page_has_private(page) &&
64 !try_to_release_page(page, GFP_KERNEL))
Jens Axboeca39d652008-05-20 21:27:41 +020065 goto out_unlock;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020066
Jens Axboe9e94cd42006-06-20 15:01:12 +020067 /*
68 * If we succeeded in removing the mapping, set LRU flag
69 * and return good.
70 */
71 if (remove_mapping(mapping, page)) {
72 buf->flags |= PIPE_BUF_FLAG_LRU;
73 return 0;
74 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020075 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020076
Jens Axboe9e94cd42006-06-20 15:01:12 +020077 /*
78 * Raced with truncate or failed to remove page from current
79 * address space, unlock and return failure.
80 */
Jens Axboeca39d652008-05-20 21:27:41 +020081out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020082 unlock_page(page);
83 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020084}
85
Jens Axboe76ad4d12006-05-03 10:41:33 +020086static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020087 struct pipe_buffer *buf)
88{
89 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020090 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020091}
92
Jens Axboe08457182007-06-12 20:51:32 +020093/*
94 * Check whether the contents of buf is OK to access. Since the content
95 * is a page cache page, IO may be in flight.
96 */
Jens Axboecac36bb02007-06-14 13:10:48 +020097static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
98 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +020099{
100 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200101 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200102
103 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200104 lock_page(page);
105
106 /*
107 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200108 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200109 */
110 if (!page->mapping) {
111 err = -ENODATA;
112 goto error;
113 }
114
115 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200116 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200117 */
118 if (!PageUptodate(page)) {
119 err = -EIO;
120 goto error;
121 }
122
123 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200124 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200125 */
Jens Axboe5274f052006-03-30 15:15:30 +0200126 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200127 }
128
Jens Axboef84d7512006-05-01 19:59:03 +0200129 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200130error:
131 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200132 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200133}
134
Hugh Dickins708e3502011-07-25 17:12:32 -0700135const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200136 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200137 .map = generic_pipe_buf_map,
138 .unmap = generic_pipe_buf_unmap,
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 Axboef84d7512006-05-01 19:59:03 +0200157 .map = generic_pipe_buf_map,
158 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200159 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200160 .release = page_cache_pipe_buf_release,
161 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200162 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200163};
164
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200165static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
166{
167 smp_mb();
168 if (waitqueue_active(&pipe->wait))
169 wake_up_interruptible(&pipe->wait);
170 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
171}
172
Jens Axboe932cc6d2007-06-21 13:10:21 +0200173/**
174 * splice_to_pipe - fill passed data into a pipe
175 * @pipe: pipe to fill
176 * @spd: data to fill
177 *
178 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200179 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200180 * the struct pipe_buf_operations associated with these pages. This
181 * function will link that data to the pipe.
182 *
Jens Axboe83f91352006-04-02 23:05:09 +0200183 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200184ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
185 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200186{
Jens Axboe00de00b2007-06-15 13:14:22 +0200187 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200188 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200189
190 ret = 0;
191 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200192 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200193
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200194 pipe_lock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200195
Jens Axboe5274f052006-03-30 15:15:30 +0200196 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200197 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200198 send_sig(SIGPIPE, current, 0);
199 if (!ret)
200 ret = -EPIPE;
201 break;
202 }
203
Jens Axboe35f3d142010-05-20 10:43:18 +0200204 if (pipe->nrbufs < pipe->buffers) {
205 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200206 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200207
Jens Axboe912d35f2006-04-26 10:59:21 +0200208 buf->page = spd->pages[page_nr];
209 buf->offset = spd->partial[page_nr].offset;
210 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200211 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200212 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200213 if (spd->flags & SPLICE_F_GIFT)
214 buf->flags |= PIPE_BUF_FLAG_GIFT;
215
Jens Axboe6f767b02006-04-11 13:53:56 +0200216 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200217 page_nr++;
218 ret += buf->len;
219
Jens Axboe6f767b02006-04-11 13:53:56 +0200220 if (pipe->inode)
221 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200222
Jens Axboe912d35f2006-04-26 10:59:21 +0200223 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200224 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200225 if (pipe->nrbufs < pipe->buffers)
Jens Axboe5274f052006-03-30 15:15:30 +0200226 continue;
227
228 break;
229 }
230
Jens Axboe912d35f2006-04-26 10:59:21 +0200231 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700232 if (!ret)
233 ret = -EAGAIN;
234 break;
235 }
236
Jens Axboe5274f052006-03-30 15:15:30 +0200237 if (signal_pending(current)) {
238 if (!ret)
239 ret = -ERESTARTSYS;
240 break;
241 }
242
243 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200244 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200245 if (waitqueue_active(&pipe->wait))
246 wake_up_interruptible_sync(&pipe->wait);
247 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200248 do_wakeup = 0;
249 }
250
Ingo Molnar3a326a22006-04-10 15:18:35 +0200251 pipe->waiting_writers++;
252 pipe_wait(pipe);
253 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200254 }
255
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200256 pipe_unlock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200257
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200258 if (do_wakeup)
259 wakeup_pipe_readers(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200260
Jens Axboe00de00b2007-06-15 13:14:22 +0200261 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800262 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200263
264 return ret;
265}
266
Hugh Dickins708e3502011-07-25 17:12:32 -0700267void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800268{
269 page_cache_release(spd->pages[i]);
270}
271
Jens Axboe35f3d142010-05-20 10:43:18 +0200272/*
273 * Check if we need to grow the arrays holding pages and partial page
274 * descriptions.
275 */
Eric Dumazet047fe362012-06-12 15:24:40 +0200276int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200277{
Eric Dumazet047fe362012-06-12 15:24:40 +0200278 unsigned int buffers = ACCESS_ONCE(pipe->buffers);
279
280 spd->nr_pages_max = buffers;
281 if (buffers <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200282 return 0;
283
Eric Dumazet047fe362012-06-12 15:24:40 +0200284 spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL);
285 spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200286
287 if (spd->pages && spd->partial)
288 return 0;
289
290 kfree(spd->pages);
291 kfree(spd->partial);
292 return -ENOMEM;
293}
294
Eric Dumazet047fe362012-06-12 15:24:40 +0200295void splice_shrink_spd(struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200296{
Eric Dumazet047fe362012-06-12 15:24:40 +0200297 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200298 return;
299
300 kfree(spd->pages);
301 kfree(spd->partial);
302}
303
Ingo Molnar3a326a22006-04-10 15:18:35 +0200304static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200305__generic_file_splice_read(struct file *in, loff_t *ppos,
306 struct pipe_inode_info *pipe, size_t len,
307 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200308{
309 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700310 unsigned int loff, nr_pages, req_pages;
Jens Axboe35f3d142010-05-20 10:43:18 +0200311 struct page *pages[PIPE_DEF_BUFFERS];
312 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200313 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200314 pgoff_t index, end_index;
315 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200316 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200317 struct splice_pipe_desc spd = {
318 .pages = pages,
319 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200320 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +0200321 .flags = flags,
322 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800323 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200324 };
Jens Axboe5274f052006-03-30 15:15:30 +0200325
Jens Axboe35f3d142010-05-20 10:43:18 +0200326 if (splice_grow_spd(pipe, &spd))
327 return -ENOMEM;
328
Jens Axboecbb7e572006-04-11 14:57:50 +0200329 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200330 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700331 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Eric Dumazet047fe362012-06-12 15:24:40 +0200332 nr_pages = min(req_pages, spd.nr_pages_max);
Jens Axboe5274f052006-03-30 15:15:30 +0200333
334 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200335 * Lookup the (hopefully) full range of pages we need.
336 */
Jens Axboe35f3d142010-05-20 10:43:18 +0200337 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700338 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200339
340 /*
341 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a4822007-07-19 01:48:05 -0700342 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200343 */
Fengguang Wu431a4822007-07-19 01:48:05 -0700344 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700345 page_cache_sync_readahead(mapping, &in->f_ra, in,
346 index, req_pages - spd.nr_pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700347
Jens Axboe932cc6d2007-06-21 13:10:21 +0200348 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200349 while (spd.nr_pages < nr_pages) {
350 /*
351 * Page could be there, find_get_pages_contig() breaks on
352 * the first hole.
353 */
354 page = find_get_page(mapping, index);
355 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200356 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200357 * page didn't exist, allocate one.
358 */
359 page = page_cache_alloc_cold(mapping);
360 if (!page)
361 break;
362
363 error = add_to_page_cache_lru(page, mapping, index,
Nick Piggin0ae0b5d2010-05-25 10:25:26 +0200364 GFP_KERNEL);
Jens Axboeeb207962006-04-27 11:05:22 +0200365 if (unlikely(error)) {
366 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200367 if (error == -EEXIST)
368 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200369 break;
370 }
371 /*
372 * add_to_page_cache() locks the page, unlock it
373 * to avoid convoluting the logic below even more.
374 */
375 unlock_page(page);
376 }
377
Jens Axboe35f3d142010-05-20 10:43:18 +0200378 spd.pages[spd.nr_pages++] = page;
Jens Axboeeb207962006-04-27 11:05:22 +0200379 index++;
380 }
381
382 /*
383 * Now loop over the map and see if we need to start IO on any
384 * pages, fill in the partial map, etc.
385 */
386 index = *ppos >> PAGE_CACHE_SHIFT;
387 nr_pages = spd.nr_pages;
388 spd.nr_pages = 0;
389 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200390 unsigned int this_len;
391
392 if (!len)
393 break;
394
395 /*
396 * this_len is the max we'll use from this page
397 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200398 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboe35f3d142010-05-20 10:43:18 +0200399 page = spd.pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200400
Fengguang Wua08a1662007-07-19 01:48:03 -0700401 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700402 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700403 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700404
Jens Axboe7480a902006-04-11 13:52:47 +0200405 /*
406 * If the page isn't uptodate, we may need to start io on it
407 */
408 if (!PageUptodate(page)) {
Miklos Szeredi69650312010-08-03 12:48:50 +0200409 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200410
411 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200412 * Page was truncated, or invalidated by the
413 * filesystem. Redo the find/create, but this time the
414 * page is kept locked, so there's no chance of another
415 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200416 */
417 if (!page->mapping) {
418 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200419 page = find_or_create_page(mapping, index,
420 mapping_gfp_mask(mapping));
421
422 if (!page) {
423 error = -ENOMEM;
424 break;
425 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200426 page_cache_release(spd.pages[page_nr]);
427 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200428 }
429 /*
430 * page was already under io and is now done, great
431 */
432 if (PageUptodate(page)) {
433 unlock_page(page);
434 goto fill_it;
435 }
436
Jens Axboe7480a902006-04-11 13:52:47 +0200437 /*
438 * need to read in the page
439 */
440 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200441 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200442 /*
443 * We really should re-lookup the page here,
444 * but it complicates things a lot. Instead
445 * lets just do what we already stored, and
446 * we'll get it the next time we are called.
447 */
Jens Axboe7480a902006-04-11 13:52:47 +0200448 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200449 error = 0;
450
Jens Axboe7480a902006-04-11 13:52:47 +0200451 break;
452 }
Jens Axboe620a3242007-06-07 09:39:42 +0200453 }
454fill_it:
455 /*
456 * i_size must be checked after PageUptodate.
457 */
458 isize = i_size_read(mapping->host);
459 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
460 if (unlikely(!isize || index > end_index))
461 break;
462
463 /*
464 * if this is the last page, see if we need to shrink
465 * the length and stop
466 */
467 if (end_index == index) {
468 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200469
470 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200471 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200472 */
Jens Axboe620a3242007-06-07 09:39:42 +0200473 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
474 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200475 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200476
477 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200478 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200479 */
Jens Axboe620a3242007-06-07 09:39:42 +0200480 this_len = min(this_len, plen - loff);
481 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200482 }
Jens Axboe620a3242007-06-07 09:39:42 +0200483
Jens Axboe35f3d142010-05-20 10:43:18 +0200484 spd.partial[page_nr].offset = loff;
485 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200486 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200487 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200488 spd.nr_pages++;
489 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200490 }
491
Jens Axboeeb207962006-04-27 11:05:22 +0200492 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200493 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200494 * we got, 'nr_pages' is how many pages are in the map.
495 */
496 while (page_nr < nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200497 page_cache_release(spd.pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700498 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200499
Jens Axboe912d35f2006-04-26 10:59:21 +0200500 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200501 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200502
Eric Dumazet047fe362012-06-12 15:24:40 +0200503 splice_shrink_spd(&spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200504 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200505}
506
Jens Axboe83f91352006-04-02 23:05:09 +0200507/**
508 * generic_file_splice_read - splice data from file to a pipe
509 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200510 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200511 * @pipe: pipe to splice to
512 * @len: number of bytes to splice
513 * @flags: splice modifier flags
514 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200515 * Description:
516 * Will read pages from given file and fill them into a pipe. Can be
517 * used as long as the address_space operations for the source implements
518 * a readpage() hook.
519 *
Jens Axboe83f91352006-04-02 23:05:09 +0200520 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200521ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
522 struct pipe_inode_info *pipe, size_t len,
523 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200524{
Jens Axboed366d3982007-06-01 14:54:11 +0200525 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200526 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200527
528 isize = i_size_read(in->f_mapping->host);
529 if (unlikely(*ppos >= isize))
530 return 0;
531
532 left = isize - *ppos;
533 if (unlikely(left < len))
534 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200535
Jens Axboe8191ecd2008-04-10 08:24:25 +0200536 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200537 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200538 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200539 file_accessed(in);
540 }
Jens Axboe5274f052006-03-30 15:15:30 +0200541
542 return ret;
543}
Jens Axboe059a8f32006-04-02 23:06:05 +0200544EXPORT_SYMBOL(generic_file_splice_read);
545
Miklos Szeredi68181732009-05-07 15:37:36 +0200546static const struct pipe_buf_operations default_pipe_buf_ops = {
547 .can_merge = 0,
548 .map = generic_pipe_buf_map,
549 .unmap = generic_pipe_buf_unmap,
550 .confirm = generic_pipe_buf_confirm,
551 .release = generic_pipe_buf_release,
552 .steal = generic_pipe_buf_steal,
553 .get = generic_pipe_buf_get,
554};
555
556static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
557 unsigned long vlen, loff_t offset)
558{
559 mm_segment_t old_fs;
560 loff_t pos = offset;
561 ssize_t res;
562
563 old_fs = get_fs();
564 set_fs(get_ds());
565 /* The cast to a user pointer is valid due to the set_fs() */
566 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
567 set_fs(old_fs);
568
569 return res;
570}
571
Miklos Szeredib2858d72009-05-19 11:37:46 +0200572static ssize_t kernel_write(struct file *file, const char *buf, size_t count,
573 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200574{
575 mm_segment_t old_fs;
576 ssize_t res;
577
578 old_fs = get_fs();
579 set_fs(get_ds());
580 /* The cast to a user pointer is valid due to the set_fs() */
Miklos Szeredib2858d72009-05-19 11:37:46 +0200581 res = vfs_write(file, (const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200582 set_fs(old_fs);
583
584 return res;
585}
586
Miklos Szeredi68181732009-05-07 15:37:36 +0200587ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
588 struct pipe_inode_info *pipe, size_t len,
589 unsigned int flags)
590{
591 unsigned int nr_pages;
592 unsigned int nr_freed;
593 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200594 struct page *pages[PIPE_DEF_BUFFERS];
595 struct partial_page partial[PIPE_DEF_BUFFERS];
596 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200597 ssize_t res;
598 size_t this_len;
599 int error;
600 int i;
601 struct splice_pipe_desc spd = {
602 .pages = pages,
603 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200604 .nr_pages_max = PIPE_DEF_BUFFERS,
Miklos Szeredi68181732009-05-07 15:37:36 +0200605 .flags = flags,
606 .ops = &default_pipe_buf_ops,
607 .spd_release = spd_release_page,
608 };
609
Jens Axboe35f3d142010-05-20 10:43:18 +0200610 if (splice_grow_spd(pipe, &spd))
611 return -ENOMEM;
612
613 res = -ENOMEM;
614 vec = __vec;
Eric Dumazet047fe362012-06-12 15:24:40 +0200615 if (spd.nr_pages_max > PIPE_DEF_BUFFERS) {
616 vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200617 if (!vec)
618 goto shrink_ret;
619 }
620
Miklos Szeredi68181732009-05-07 15:37:36 +0200621 offset = *ppos & ~PAGE_CACHE_MASK;
622 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
623
Eric Dumazet047fe362012-06-12 15:24:40 +0200624 for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200625 struct page *page;
626
Jens Axboe4f231222009-05-13 08:35:35 +0200627 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200628 error = -ENOMEM;
629 if (!page)
630 goto err;
631
632 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200633 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200634 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200635 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200636 spd.nr_pages++;
637 len -= this_len;
638 offset = 0;
639 }
640
641 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200642 if (res < 0) {
643 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200644 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200645 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200646
647 error = 0;
648 if (!res)
649 goto err;
650
651 nr_freed = 0;
652 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200653 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200654 spd.partial[i].offset = 0;
655 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200656 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200657 __free_page(spd.pages[i]);
658 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200659 nr_freed++;
660 }
661 res -= this_len;
662 }
663 spd.nr_pages -= nr_freed;
664
665 res = splice_to_pipe(pipe, &spd);
666 if (res > 0)
667 *ppos += res;
668
Jens Axboe35f3d142010-05-20 10:43:18 +0200669shrink_ret:
670 if (vec != __vec)
671 kfree(vec);
Eric Dumazet047fe362012-06-12 15:24:40 +0200672 splice_shrink_spd(&spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200673 return res;
674
675err:
Jens Axboe4f231222009-05-13 08:35:35 +0200676 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200677 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200678
Jens Axboe35f3d142010-05-20 10:43:18 +0200679 res = error;
680 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200681}
682EXPORT_SYMBOL(default_file_splice_read);
683
Jens Axboe5274f052006-03-30 15:15:30 +0200684/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200685 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200686 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200687 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200688static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200689 struct pipe_buffer *buf, struct splice_desc *sd)
690{
Jens Axboe6a14b902007-06-14 13:08:55 +0200691 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200692 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100693 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200694
Michał Mirosława8adbe32010-12-17 08:56:44 +0100695 if (!likely(file->f_op && file->f_op->sendpage))
696 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200697
Eric Dumazet35f9c092012-04-05 03:05:35 +0000698 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000699
700 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000701 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000702
Michał Mirosława8adbe32010-12-17 08:56:44 +0100703 return file->f_op->sendpage(file, buf->page, buf->offset,
704 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200705}
706
707/*
708 * This is a little more tricky than the file -> pipe splicing. There are
709 * basically three cases:
710 *
711 * - Destination page already exists in the address space and there
712 * are users of it. For that case we have no other option that
713 * copying the data. Tough luck.
714 * - Destination page already exists in the address space, but there
715 * are no users of it. Make sure it's uptodate, then drop it. Fall
716 * through to last case.
717 * - Destination page does not exist, we can add the pipe page to
718 * the page cache and avoid the copy.
719 *
Jens Axboe83f91352006-04-02 23:05:09 +0200720 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
721 * sd->flags), we attempt to migrate pages from the pipe to the output
722 * file address space page cache. This is possible if no one else has
723 * the pipe page referenced outside of the pipe and page cache. If
724 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
725 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200726 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200727int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
728 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200729{
Jens Axboe6a14b902007-06-14 13:08:55 +0200730 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200731 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200732 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200733 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700734 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200735 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200736
Jens Axboe5274f052006-03-30 15:15:30 +0200737 offset = sd->pos & ~PAGE_CACHE_MASK;
738
Jens Axboe016b6612006-04-25 15:42:00 +0200739 this_len = sd->len;
740 if (this_len + offset > PAGE_CACHE_SIZE)
741 this_len = PAGE_CACHE_SIZE - offset;
742
Nick Pigginafddba42007-10-16 01:25:01 -0700743 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
744 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
745 if (unlikely(ret))
746 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200747
Jens Axboe0568b402006-05-01 19:50:48 +0200748 if (buf->page != page) {
Jens Axboe76ad4d12006-05-03 10:41:33 +0200749 char *src = buf->ops->map(pipe, buf, 1);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800750 char *dst = kmap_atomic(page);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200751
Jens Axboe016b6612006-04-25 15:42:00 +0200752 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200753 flush_dcache_page(page);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800754 kunmap_atomic(dst);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200755 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200756 }
Nick Pigginafddba42007-10-16 01:25:01 -0700757 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
758 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200759out:
Jens Axboe5274f052006-03-30 15:15:30 +0200760 return ret;
761}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200762EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200763
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200764static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
765{
766 smp_mb();
767 if (waitqueue_active(&pipe->wait))
768 wake_up_interruptible(&pipe->wait);
769 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
770}
771
772/**
773 * splice_from_pipe_feed - feed available data from a pipe to a file
774 * @pipe: pipe to splice from
775 * @sd: information to @actor
776 * @actor: handler that splices the data
777 *
778 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200779 * This function loops over the pipe and calls @actor to do the
780 * actual moving of a single struct pipe_buffer to the desired
781 * destination. It returns when there's no more buffers left in
782 * the pipe or if the requested number of bytes (@sd->total_len)
783 * have been copied. It returns a positive number (one) if the
784 * pipe needs to be filled with more data, zero if the required
785 * number of bytes have been copied and -errno on error.
786 *
787 * This, together with splice_from_pipe_{begin,end,next}, may be
788 * used to implement the functionality of __splice_from_pipe() when
789 * locking is required around copying the pipe buffers to the
790 * destination.
791 */
792int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
793 splice_actor *actor)
794{
795 int ret;
796
797 while (pipe->nrbufs) {
798 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
799 const struct pipe_buf_operations *ops = buf->ops;
800
801 sd->len = buf->len;
802 if (sd->len > sd->total_len)
803 sd->len = sd->total_len;
804
Michał Mirosława8adbe32010-12-17 08:56:44 +0100805 ret = buf->ops->confirm(pipe, buf);
806 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200807 if (ret == -ENODATA)
808 ret = 0;
809 return ret;
810 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100811
812 ret = actor(pipe, buf, sd);
813 if (ret <= 0)
814 return ret;
815
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200816 buf->offset += ret;
817 buf->len -= ret;
818
819 sd->num_spliced += ret;
820 sd->len -= ret;
821 sd->pos += ret;
822 sd->total_len -= ret;
823
824 if (!buf->len) {
825 buf->ops = NULL;
826 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200827 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200828 pipe->nrbufs--;
829 if (pipe->inode)
830 sd->need_wakeup = true;
831 }
832
833 if (!sd->total_len)
834 return 0;
835 }
836
837 return 1;
838}
839EXPORT_SYMBOL(splice_from_pipe_feed);
840
841/**
842 * splice_from_pipe_next - wait for some data to splice from
843 * @pipe: pipe to splice from
844 * @sd: information about the splice operation
845 *
846 * Description:
847 * This function will wait for some data and return a positive
848 * value (one) if pipe buffers are available. It will return zero
849 * or -errno if no more data needs to be spliced.
850 */
851int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
852{
853 while (!pipe->nrbufs) {
854 if (!pipe->writers)
855 return 0;
856
857 if (!pipe->waiting_writers && sd->num_spliced)
858 return 0;
859
860 if (sd->flags & SPLICE_F_NONBLOCK)
861 return -EAGAIN;
862
863 if (signal_pending(current))
864 return -ERESTARTSYS;
865
866 if (sd->need_wakeup) {
867 wakeup_pipe_writers(pipe);
868 sd->need_wakeup = false;
869 }
870
871 pipe_wait(pipe);
872 }
873
874 return 1;
875}
876EXPORT_SYMBOL(splice_from_pipe_next);
877
878/**
879 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700880 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200881 *
882 * Description:
883 * This function should be called before a loop containing
884 * splice_from_pipe_next() and splice_from_pipe_feed() to
885 * initialize the necessary fields of @sd.
886 */
887void splice_from_pipe_begin(struct splice_desc *sd)
888{
889 sd->num_spliced = 0;
890 sd->need_wakeup = false;
891}
892EXPORT_SYMBOL(splice_from_pipe_begin);
893
894/**
895 * splice_from_pipe_end - finish splicing from pipe
896 * @pipe: pipe to splice from
897 * @sd: information about the splice operation
898 *
899 * Description:
900 * This function will wake up pipe writers if necessary. It should
901 * be called after a loop containing splice_from_pipe_next() and
902 * splice_from_pipe_feed().
903 */
904void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
905{
906 if (sd->need_wakeup)
907 wakeup_pipe_writers(pipe);
908}
909EXPORT_SYMBOL(splice_from_pipe_end);
910
Jens Axboe932cc6d2007-06-21 13:10:21 +0200911/**
912 * __splice_from_pipe - splice data from a pipe to given actor
913 * @pipe: pipe to splice from
914 * @sd: information to @actor
915 * @actor: handler that splices the data
916 *
917 * Description:
918 * This function does little more than loop over the pipe and call
919 * @actor to do the actual moving of a single struct pipe_buffer to
920 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
921 * pipe_to_user.
922 *
Jens Axboe83f91352006-04-02 23:05:09 +0200923 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200924ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
925 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200926{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200927 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200928
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200929 splice_from_pipe_begin(sd);
930 do {
931 ret = splice_from_pipe_next(pipe, sd);
932 if (ret > 0)
933 ret = splice_from_pipe_feed(pipe, sd, actor);
934 } while (ret > 0);
935 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200936
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200937 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200938}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100939EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200940
Jens Axboe932cc6d2007-06-21 13:10:21 +0200941/**
942 * splice_from_pipe - splice data from a pipe to a file
943 * @pipe: pipe to splice from
944 * @out: file to splice to
945 * @ppos: position in @out
946 * @len: how many bytes to splice
947 * @flags: splice modifier flags
948 * @actor: handler that splices the data
949 *
950 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200951 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200952 * otherwise it's identical to __splice_from_pipe().
953 *
954 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200955ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
956 loff_t *ppos, size_t len, unsigned int flags,
957 splice_actor *actor)
958{
959 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200960 struct splice_desc sd = {
961 .total_len = len,
962 .flags = flags,
963 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200964 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200965 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200966
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200967 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200968 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200969 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200970
971 return ret;
972}
973
974/**
Jens Axboe83f91352006-04-02 23:05:09 +0200975 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200976 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200977 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200978 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200979 * @len: number of bytes to splice
980 * @flags: splice modifier flags
981 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200982 * Description:
983 * Will either move or copy pages (determined by @flags options) from
984 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200985 *
986 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200987ssize_t
988generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200989 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200990{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200991 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200992 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200993 struct splice_desc sd = {
994 .total_len = len,
995 .flags = flags,
996 .pos = *ppos,
997 .u.file = out,
998 };
Ingo Molnar3a326a22006-04-10 15:18:35 +0200999 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +02001000
Jan Kara14da9202012-06-12 16:20:37 +02001001 sb_start_write(inode->i_sb);
1002
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001003 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001004
1005 splice_from_pipe_begin(&sd);
1006 do {
1007 ret = splice_from_pipe_next(pipe, &sd);
1008 if (ret <= 0)
1009 break;
1010
1011 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1012 ret = file_remove_suid(out);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001013 if (!ret) {
Josef Bacikc3b2da32012-03-26 09:59:21 -04001014 ret = file_update_time(out);
1015 if (!ret)
1016 ret = splice_from_pipe_feed(pipe, &sd,
1017 pipe_to_file);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001018 }
Miklos Szeredieb443e52009-04-14 19:48:38 +02001019 mutex_unlock(&inode->i_mutex);
1020 } while (ret > 0);
1021 splice_from_pipe_end(pipe, &sd);
1022
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001023 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001024
1025 if (sd.num_spliced)
1026 ret = sd.num_spliced;
1027
Jens Axboea4514eb2006-04-19 15:57:05 +02001028 if (ret > 0) {
Jan Kara148f9482009-08-17 19:52:36 +02001029 int err;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001030
Jan Kara148f9482009-08-17 19:52:36 +02001031 err = generic_write_sync(out, *ppos, ret);
1032 if (err)
1033 ret = err;
1034 else
1035 *ppos += ret;
Namjae Jeond0e1d662012-12-11 16:00:21 -08001036 balance_dirty_pages_ratelimited(mapping);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001037 }
Jan Kara14da9202012-06-12 16:20:37 +02001038 sb_end_write(inode->i_sb);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001039
1040 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001041}
1042
Jens Axboe059a8f32006-04-02 23:06:05 +02001043EXPORT_SYMBOL(generic_file_splice_write);
1044
Miklos Szeredib2858d72009-05-19 11:37:46 +02001045static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1046 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001047{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001048 int ret;
1049 void *data;
1050
Miklos Szeredib2858d72009-05-19 11:37:46 +02001051 data = buf->ops->map(pipe, buf, 0);
1052 ret = kernel_write(sd->u.file, data + buf->offset, sd->len, sd->pos);
1053 buf->ops->unmap(pipe, buf, data);
1054
1055 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001056}
1057
1058static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1059 struct file *out, loff_t *ppos,
1060 size_t len, unsigned int flags)
1061{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001062 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001063
Miklos Szeredib2858d72009-05-19 11:37:46 +02001064 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1065 if (ret > 0)
1066 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001067
Miklos Szeredib2858d72009-05-19 11:37:46 +02001068 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001069}
1070
Jens Axboe83f91352006-04-02 23:05:09 +02001071/**
1072 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001073 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001074 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001075 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001076 * @len: number of bytes to splice
1077 * @flags: splice modifier flags
1078 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001079 * Description:
1080 * Will send @len bytes from the pipe to a network socket. No data copying
1081 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001082 *
1083 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001084ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001085 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001086{
Jens Axboe00522fb2006-04-26 14:39:29 +02001087 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001088}
1089
Jens Axboe059a8f32006-04-02 23:06:05 +02001090EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001091
Jens Axboe83f91352006-04-02 23:05:09 +02001092/*
1093 * Attempt to initiate a splice from pipe to file.
1094 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001095static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001096 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001097{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001098 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1099 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001100 int ret;
1101
Jens Axboe49570e92006-04-11 13:56:09 +02001102 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +02001103 return -EBADF;
1104
Linus Torvaldsefc968d2008-10-09 14:04:54 -07001105 if (unlikely(out->f_flags & O_APPEND))
1106 return -EINVAL;
1107
Jens Axboecbb7e572006-04-11 14:57:50 +02001108 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001109 if (unlikely(ret < 0))
1110 return ret;
1111
Changli Gaocc56f7d2009-11-04 09:09:52 +01001112 if (out->f_op && out->f_op->splice_write)
1113 splice_write = out->f_op->splice_write;
1114 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001115 splice_write = default_file_splice_write;
1116
1117 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001118}
1119
Jens Axboe83f91352006-04-02 23:05:09 +02001120/*
1121 * Attempt to initiate a splice from a file to a pipe.
1122 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001123static long do_splice_to(struct file *in, loff_t *ppos,
1124 struct pipe_inode_info *pipe, size_t len,
1125 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001126{
Miklos Szeredi68181732009-05-07 15:37:36 +02001127 ssize_t (*splice_read)(struct file *, loff_t *,
1128 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001129 int ret;
1130
Jens Axboe49570e92006-04-11 13:56:09 +02001131 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001132 return -EBADF;
1133
Jens Axboecbb7e572006-04-11 14:57:50 +02001134 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001135 if (unlikely(ret < 0))
1136 return ret;
1137
Changli Gaocc56f7d2009-11-04 09:09:52 +01001138 if (in->f_op && in->f_op->splice_read)
1139 splice_read = in->f_op->splice_read;
1140 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001141 splice_read = default_file_splice_read;
1142
1143 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001144}
1145
Jens Axboe932cc6d2007-06-21 13:10:21 +02001146/**
1147 * splice_direct_to_actor - splices data directly between two non-pipes
1148 * @in: file to splice from
1149 * @sd: actor information on where to splice to
1150 * @actor: handles the data splicing
1151 *
1152 * Description:
1153 * This is a special case helper to splice directly between two
1154 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001155 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001156 * that process.
1157 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001158 */
1159ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1160 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001161{
1162 struct pipe_inode_info *pipe;
1163 long ret, bytes;
1164 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001165 size_t len;
1166 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001167
1168 /*
1169 * We require the input being a regular file, as we don't want to
1170 * randomly drop data for eg socket -> socket splicing. Use the
1171 * piped splicing for that!
1172 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001173 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001174 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1175 return -EINVAL;
1176
1177 /*
1178 * neither in nor out is a pipe, setup an internal pipe attached to
1179 * 'out' and transfer the wanted data from 'in' to 'out' through that
1180 */
1181 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001182 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001183 pipe = alloc_pipe_info(NULL);
1184 if (!pipe)
1185 return -ENOMEM;
1186
1187 /*
1188 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001189 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001190 * PIPE_READERS appropriately.
1191 */
1192 pipe->readers = 1;
1193
1194 current->splice_pipe = pipe;
1195 }
1196
1197 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001198 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001199 */
1200 ret = 0;
1201 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001202 len = sd->total_len;
1203 flags = sd->flags;
1204
1205 /*
1206 * Don't block on output, we have to drain the direct pipe.
1207 */
1208 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001209
1210 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001211 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001212 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001213
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001214 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001215 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001216 goto out_release;
1217
1218 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001219 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001220
1221 /*
1222 * NOTE: nonblocking mode only applies to the input. We
1223 * must not do the output in nonblocking mode as then we
1224 * could get stuck data in the internal pipe:
1225 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001226 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001227 if (unlikely(ret <= 0)) {
1228 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001229 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001230 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001231
1232 bytes += ret;
1233 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001234 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001235
Tom Zanussia82c53a2008-05-09 13:28:36 +02001236 if (ret < read_len) {
1237 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001238 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001239 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001240 }
1241
Jens Axboe9e971982008-01-29 21:05:57 +01001242done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001243 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001244 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001245 return bytes;
1246
1247out_release:
1248 /*
1249 * If we did an incomplete transfer we must release
1250 * the pipe buffers in question:
1251 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001252 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001253 struct pipe_buffer *buf = pipe->bufs + i;
1254
1255 if (buf->ops) {
1256 buf->ops->release(pipe, buf);
1257 buf->ops = NULL;
1258 }
1259 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001260
Jens Axboe9e971982008-01-29 21:05:57 +01001261 if (!bytes)
1262 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001263
Jens Axboe9e971982008-01-29 21:05:57 +01001264 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001265}
1266EXPORT_SYMBOL(splice_direct_to_actor);
1267
1268static int direct_splice_actor(struct pipe_inode_info *pipe,
1269 struct splice_desc *sd)
1270{
Jens Axboe6a14b902007-06-14 13:08:55 +02001271 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001272
Changli Gao2cb4b052010-06-29 13:09:18 +02001273 return do_splice_from(pipe, file, &file->f_pos, sd->total_len,
1274 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001275}
1276
Jens Axboe932cc6d2007-06-21 13:10:21 +02001277/**
1278 * do_splice_direct - splices data directly between two files
1279 * @in: file to splice from
1280 * @ppos: input file offset
1281 * @out: file to splice to
1282 * @len: number of bytes to splice
1283 * @flags: splice modifier flags
1284 *
1285 * Description:
1286 * For use by do_sendfile(). splice can easily emulate sendfile, but
1287 * doing it in the application would incur an extra system call
1288 * (splice in + splice out, as compared to just sendfile()). So this helper
1289 * can splice directly through a process-private pipe.
1290 *
1291 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001292long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1293 size_t len, unsigned int flags)
1294{
1295 struct splice_desc sd = {
1296 .len = len,
1297 .total_len = len,
1298 .flags = flags,
1299 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001300 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001301 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001302 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001303
1304 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001305 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001306 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001307
Jens Axboec66ab6f2007-06-12 21:17:17 +02001308 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001309}
1310
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001311static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1312 struct pipe_inode_info *opipe,
1313 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001314
1315/*
Jens Axboe83f91352006-04-02 23:05:09 +02001316 * Determine where to splice to/from.
1317 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001318static long do_splice(struct file *in, loff_t __user *off_in,
1319 struct file *out, loff_t __user *off_out,
1320 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001321{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001322 struct pipe_inode_info *ipipe;
1323 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001324 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001325 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001326
Linus Torvalds71993e62010-11-28 13:56:09 -08001327 ipipe = get_pipe_info(in);
1328 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001329
1330 if (ipipe && opipe) {
1331 if (off_in || off_out)
1332 return -ESPIPE;
1333
1334 if (!(in->f_mode & FMODE_READ))
1335 return -EBADF;
1336
1337 if (!(out->f_mode & FMODE_WRITE))
1338 return -EBADF;
1339
1340 /* Splicing to self would be fun, but... */
1341 if (ipipe == opipe)
1342 return -EINVAL;
1343
1344 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1345 }
1346
1347 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001348 if (off_in)
1349 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001350 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001351 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001352 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001353 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001354 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001355 off = &offset;
1356 } else
1357 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001358
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001359 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001360
1361 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1362 ret = -EFAULT;
1363
1364 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001365 }
Jens Axboe5274f052006-03-30 15:15:30 +02001366
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001367 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001368 if (off_out)
1369 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001370 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001371 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001372 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001373 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001374 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001375 off = &offset;
1376 } else
1377 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001378
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001379 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001380
1381 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1382 ret = -EFAULT;
1383
1384 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001385 }
Jens Axboe5274f052006-03-30 15:15:30 +02001386
1387 return -EINVAL;
1388}
1389
Jens Axboe912d35f2006-04-26 10:59:21 +02001390/*
1391 * Map an iov into an array of pages and offset/length tupples. With the
1392 * partial_page structure, we can map several non-contiguous ranges into
1393 * our ones pages[] map instead of splitting that operation into pieces.
1394 * Could easily be exported as a generic helper for other users, in which
1395 * case one would probably want to add a 'max_nr_pages' parameter as well.
1396 */
1397static int get_iovec_page_array(const struct iovec __user *iov,
1398 unsigned int nr_vecs, struct page **pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001399 struct partial_page *partial, bool aligned,
Jens Axboe35f3d142010-05-20 10:43:18 +02001400 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001401{
1402 int buffers = 0, error = 0;
1403
Jens Axboe912d35f2006-04-26 10:59:21 +02001404 while (nr_vecs) {
1405 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001406 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001407 void __user *base;
1408 size_t len;
1409 int i;
1410
Linus Torvalds75723952007-10-01 13:17:28 -07001411 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001412 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001413 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001414
1415 base = entry.iov_base;
1416 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001417
1418 /*
1419 * Sanity check this iovec. 0 read succeeds.
1420 */
Linus Torvalds75723952007-10-01 13:17:28 -07001421 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001422 if (unlikely(!len))
1423 break;
1424 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001425 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001426 break;
1427
1428 /*
1429 * Get this base offset and number of pages, then map
1430 * in the user pages.
1431 */
1432 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001433
1434 /*
1435 * If asked for alignment, the offset must be zero and the
1436 * length a multiple of the PAGE_SIZE.
1437 */
1438 error = -EINVAL;
1439 if (aligned && (off || len & ~PAGE_MASK))
1440 break;
1441
Jens Axboe912d35f2006-04-26 10:59:21 +02001442 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001443 if (npages > pipe_buffers - buffers)
1444 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001445
Nick Pigginbc40d732008-07-25 19:45:26 -07001446 error = get_user_pages_fast((unsigned long)base, npages,
1447 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001448
1449 if (unlikely(error <= 0))
1450 break;
1451
1452 /*
1453 * Fill this contiguous range into the partial page map.
1454 */
1455 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001456 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001457
1458 partial[buffers].offset = off;
1459 partial[buffers].len = plen;
1460
1461 off = 0;
1462 len -= plen;
1463 buffers++;
1464 }
1465
1466 /*
1467 * We didn't complete this iov, stop here since it probably
1468 * means we have to move some of this into a pipe to
1469 * be able to continue.
1470 */
1471 if (len)
1472 break;
1473
1474 /*
1475 * Don't continue if we mapped fewer pages than we asked for,
1476 * or if we mapped the max number of pages that we have
1477 * room for.
1478 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001479 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001480 break;
1481
1482 nr_vecs--;
1483 iov++;
1484 }
1485
Jens Axboe912d35f2006-04-26 10:59:21 +02001486 if (buffers)
1487 return buffers;
1488
1489 return error;
1490}
1491
Jens Axboe6a14b902007-06-14 13:08:55 +02001492static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1493 struct splice_desc *sd)
1494{
1495 char *src;
1496 int ret;
1497
Jens Axboe6a14b902007-06-14 13:08:55 +02001498 /*
1499 * See if we can use the atomic maps, by prefaulting in the
1500 * pages and doing an atomic copy
1501 */
1502 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1503 src = buf->ops->map(pipe, buf, 1);
1504 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1505 sd->len);
1506 buf->ops->unmap(pipe, buf, src);
1507 if (!ret) {
1508 ret = sd->len;
1509 goto out;
1510 }
1511 }
1512
1513 /*
1514 * No dice, use slow non-atomic map and copy
1515 */
1516 src = buf->ops->map(pipe, buf, 0);
1517
1518 ret = sd->len;
1519 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1520 ret = -EFAULT;
1521
Jens Axboe6866bef2007-10-16 10:01:29 +02001522 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001523out:
1524 if (ret > 0)
1525 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001526 return ret;
1527}
1528
1529/*
1530 * For lack of a better implementation, implement vmsplice() to userspace
1531 * as a simple copy of the pipes pages to the user iov.
1532 */
1533static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1534 unsigned long nr_segs, unsigned int flags)
1535{
1536 struct pipe_inode_info *pipe;
1537 struct splice_desc sd;
1538 ssize_t size;
1539 int error;
1540 long ret;
1541
Linus Torvalds71993e62010-11-28 13:56:09 -08001542 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001543 if (!pipe)
1544 return -EBADF;
1545
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001546 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001547
1548 error = ret = 0;
1549 while (nr_segs) {
1550 void __user *base;
1551 size_t len;
1552
1553 /*
1554 * Get user address base and length for this iovec.
1555 */
1556 error = get_user(base, &iov->iov_base);
1557 if (unlikely(error))
1558 break;
1559 error = get_user(len, &iov->iov_len);
1560 if (unlikely(error))
1561 break;
1562
1563 /*
1564 * Sanity check this iovec. 0 read succeeds.
1565 */
1566 if (unlikely(!len))
1567 break;
1568 if (unlikely(!base)) {
1569 error = -EFAULT;
1570 break;
1571 }
1572
Jens Axboe88119302008-02-08 08:49:14 -08001573 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1574 error = -EFAULT;
1575 break;
1576 }
1577
Jens Axboe6a14b902007-06-14 13:08:55 +02001578 sd.len = 0;
1579 sd.total_len = len;
1580 sd.flags = flags;
1581 sd.u.userptr = base;
1582 sd.pos = 0;
1583
1584 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1585 if (size < 0) {
1586 if (!ret)
1587 ret = size;
1588
1589 break;
1590 }
1591
1592 ret += size;
1593
1594 if (size < len)
1595 break;
1596
1597 nr_segs--;
1598 iov++;
1599 }
1600
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001601 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001602
1603 if (!ret)
1604 ret = error;
1605
1606 return ret;
1607}
1608
Jens Axboe912d35f2006-04-26 10:59:21 +02001609/*
1610 * vmsplice splices a user address range into a pipe. It can be thought of
1611 * as splice-from-memory, where the regular splice is splice-from-file (or
1612 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001613 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001614static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1615 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001616{
Jens Axboeddac0d32006-11-04 12:49:32 +01001617 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001618 struct page *pages[PIPE_DEF_BUFFERS];
1619 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001620 struct splice_pipe_desc spd = {
1621 .pages = pages,
1622 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001623 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +02001624 .flags = flags,
1625 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001626 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001627 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001628 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001629
Linus Torvalds71993e62010-11-28 13:56:09 -08001630 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001631 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001632 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001633
Jens Axboe35f3d142010-05-20 10:43:18 +02001634 if (splice_grow_spd(pipe, &spd))
1635 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001636
Jens Axboe35f3d142010-05-20 10:43:18 +02001637 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001638 spd.partial, false,
Eric Dumazet047fe362012-06-12 15:24:40 +02001639 spd.nr_pages_max);
Jens Axboe35f3d142010-05-20 10:43:18 +02001640 if (spd.nr_pages <= 0)
1641 ret = spd.nr_pages;
1642 else
1643 ret = splice_to_pipe(pipe, &spd);
1644
Eric Dumazet047fe362012-06-12 15:24:40 +02001645 splice_shrink_spd(&spd);
Jens Axboe35f3d142010-05-20 10:43:18 +02001646 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001647}
1648
Jens Axboe6a14b902007-06-14 13:08:55 +02001649/*
1650 * Note that vmsplice only really supports true splicing _from_ user memory
1651 * to a pipe, not the other way around. Splicing from user memory is a simple
1652 * operation that can be supported without any funky alignment restrictions
1653 * or nasty vm tricks. We simply map in the user memory and fill them into
1654 * a pipe. The reverse isn't quite as easy, though. There are two possible
1655 * solutions for that:
1656 *
1657 * - memcpy() the data internally, at which point we might as well just
1658 * do a regular read() on the buffer anyway.
1659 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1660 * has restriction limitations on both ends of the pipe).
1661 *
1662 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1663 *
1664 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001665SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1666 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001667{
Al Viro2903ff02012-08-28 12:52:22 -04001668 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001669 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001670
Jens Axboe6a14b902007-06-14 13:08:55 +02001671 if (unlikely(nr_segs > UIO_MAXIOV))
1672 return -EINVAL;
1673 else if (unlikely(!nr_segs))
1674 return 0;
1675
Jens Axboe912d35f2006-04-26 10:59:21 +02001676 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001677 f = fdget(fd);
1678 if (f.file) {
1679 if (f.file->f_mode & FMODE_WRITE)
1680 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1681 else if (f.file->f_mode & FMODE_READ)
1682 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001683
Al Viro2903ff02012-08-28 12:52:22 -04001684 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001685 }
1686
1687 return error;
1688}
1689
Heiko Carstens836f92a2009-01-14 14:14:33 +01001690SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1691 int, fd_out, loff_t __user *, off_out,
1692 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001693{
Al Viro2903ff02012-08-28 12:52:22 -04001694 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001695 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001696
1697 if (unlikely(!len))
1698 return 0;
1699
1700 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001701 in = fdget(fd_in);
1702 if (in.file) {
1703 if (in.file->f_mode & FMODE_READ) {
1704 out = fdget(fd_out);
1705 if (out.file) {
1706 if (out.file->f_mode & FMODE_WRITE)
1707 error = do_splice(in.file, off_in,
1708 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001709 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001710 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001711 }
1712 }
Al Viro2903ff02012-08-28 12:52:22 -04001713 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001714 }
Jens Axboe5274f052006-03-30 15:15:30 +02001715 return error;
1716}
Jens Axboe70524492006-04-11 15:51:17 +02001717
1718/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001719 * Make sure there's data to read. Wait for input if we can, otherwise
1720 * return an appropriate error.
1721 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001722static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001723{
1724 int ret;
1725
1726 /*
1727 * Check ->nrbufs without the inode lock first. This function
1728 * is speculative anyways, so missing one is ok.
1729 */
1730 if (pipe->nrbufs)
1731 return 0;
1732
1733 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001734 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001735
1736 while (!pipe->nrbufs) {
1737 if (signal_pending(current)) {
1738 ret = -ERESTARTSYS;
1739 break;
1740 }
1741 if (!pipe->writers)
1742 break;
1743 if (!pipe->waiting_writers) {
1744 if (flags & SPLICE_F_NONBLOCK) {
1745 ret = -EAGAIN;
1746 break;
1747 }
1748 }
1749 pipe_wait(pipe);
1750 }
1751
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001752 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001753 return ret;
1754}
1755
1756/*
1757 * Make sure there's writeable room. Wait for room if we can, otherwise
1758 * return an appropriate error.
1759 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001760static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001761{
1762 int ret;
1763
1764 /*
1765 * Check ->nrbufs without the inode lock first. This function
1766 * is speculative anyways, so missing one is ok.
1767 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001768 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001769 return 0;
1770
1771 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001772 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001773
Jens Axboe35f3d142010-05-20 10:43:18 +02001774 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001775 if (!pipe->readers) {
1776 send_sig(SIGPIPE, current, 0);
1777 ret = -EPIPE;
1778 break;
1779 }
1780 if (flags & SPLICE_F_NONBLOCK) {
1781 ret = -EAGAIN;
1782 break;
1783 }
1784 if (signal_pending(current)) {
1785 ret = -ERESTARTSYS;
1786 break;
1787 }
1788 pipe->waiting_writers++;
1789 pipe_wait(pipe);
1790 pipe->waiting_writers--;
1791 }
1792
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001793 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001794 return ret;
1795}
1796
1797/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001798 * Splice contents of ipipe to opipe.
1799 */
1800static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1801 struct pipe_inode_info *opipe,
1802 size_t len, unsigned int flags)
1803{
1804 struct pipe_buffer *ibuf, *obuf;
1805 int ret = 0, nbuf;
1806 bool input_wakeup = false;
1807
1808
1809retry:
1810 ret = ipipe_prep(ipipe, flags);
1811 if (ret)
1812 return ret;
1813
1814 ret = opipe_prep(opipe, flags);
1815 if (ret)
1816 return ret;
1817
1818 /*
1819 * Potential ABBA deadlock, work around it by ordering lock
1820 * grabbing by pipe info address. Otherwise two different processes
1821 * could deadlock (one doing tee from A -> B, the other from B -> A).
1822 */
1823 pipe_double_lock(ipipe, opipe);
1824
1825 do {
1826 if (!opipe->readers) {
1827 send_sig(SIGPIPE, current, 0);
1828 if (!ret)
1829 ret = -EPIPE;
1830 break;
1831 }
1832
1833 if (!ipipe->nrbufs && !ipipe->writers)
1834 break;
1835
1836 /*
1837 * Cannot make any progress, because either the input
1838 * pipe is empty or the output pipe is full.
1839 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001840 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001841 /* Already processed some buffers, break */
1842 if (ret)
1843 break;
1844
1845 if (flags & SPLICE_F_NONBLOCK) {
1846 ret = -EAGAIN;
1847 break;
1848 }
1849
1850 /*
1851 * We raced with another reader/writer and haven't
1852 * managed to process any buffers. A zero return
1853 * value means EOF, so retry instead.
1854 */
1855 pipe_unlock(ipipe);
1856 pipe_unlock(opipe);
1857 goto retry;
1858 }
1859
1860 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001861 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001862 obuf = opipe->bufs + nbuf;
1863
1864 if (len >= ibuf->len) {
1865 /*
1866 * Simply move the whole buffer from ipipe to opipe
1867 */
1868 *obuf = *ibuf;
1869 ibuf->ops = NULL;
1870 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001871 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001872 ipipe->nrbufs--;
1873 input_wakeup = true;
1874 } else {
1875 /*
1876 * Get a reference to this pipe buffer,
1877 * so we can copy the contents over.
1878 */
1879 ibuf->ops->get(ipipe, ibuf);
1880 *obuf = *ibuf;
1881
1882 /*
1883 * Don't inherit the gift flag, we need to
1884 * prevent multiple steals of this page.
1885 */
1886 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1887
1888 obuf->len = len;
1889 opipe->nrbufs++;
1890 ibuf->offset += obuf->len;
1891 ibuf->len -= obuf->len;
1892 }
1893 ret += obuf->len;
1894 len -= obuf->len;
1895 } while (len);
1896
1897 pipe_unlock(ipipe);
1898 pipe_unlock(opipe);
1899
1900 /*
1901 * If we put data in the output pipe, wakeup any potential readers.
1902 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001903 if (ret > 0)
1904 wakeup_pipe_readers(opipe);
1905
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001906 if (input_wakeup)
1907 wakeup_pipe_writers(ipipe);
1908
1909 return ret;
1910}
1911
1912/*
Jens Axboe70524492006-04-11 15:51:17 +02001913 * Link contents of ipipe to opipe.
1914 */
1915static int link_pipe(struct pipe_inode_info *ipipe,
1916 struct pipe_inode_info *opipe,
1917 size_t len, unsigned int flags)
1918{
1919 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001920 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001921
1922 /*
1923 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001924 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001925 * could deadlock (one doing tee from A -> B, the other from B -> A).
1926 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001927 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001928
Jens Axboeaadd06e2006-07-10 11:00:01 +02001929 do {
Jens Axboe70524492006-04-11 15:51:17 +02001930 if (!opipe->readers) {
1931 send_sig(SIGPIPE, current, 0);
1932 if (!ret)
1933 ret = -EPIPE;
1934 break;
1935 }
Jens Axboe70524492006-04-11 15:51:17 +02001936
1937 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001938 * If we have iterated all input buffers or ran out of
1939 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001940 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001941 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001942 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001943
Jens Axboe35f3d142010-05-20 10:43:18 +02001944 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1945 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001946
Jens Axboe2a27250e2006-04-19 15:56:40 +02001947 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001948 * Get a reference to this pipe buffer,
1949 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001950 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001951 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001952
Jens Axboeaadd06e2006-07-10 11:00:01 +02001953 obuf = opipe->bufs + nbuf;
1954 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001955
Jens Axboeaadd06e2006-07-10 11:00:01 +02001956 /*
1957 * Don't inherit the gift flag, we need to
1958 * prevent multiple steals of this page.
1959 */
1960 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1961
1962 if (obuf->len > len)
1963 obuf->len = len;
1964
1965 opipe->nrbufs++;
1966 ret += obuf->len;
1967 len -= obuf->len;
1968 i++;
1969 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001970
Jens Axboe02cf01a2008-02-20 10:34:51 +01001971 /*
1972 * return EAGAIN if we have the potential of some data in the
1973 * future, otherwise just return 0
1974 */
1975 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1976 ret = -EAGAIN;
1977
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001978 pipe_unlock(ipipe);
1979 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001980
Jens Axboeaadd06e2006-07-10 11:00:01 +02001981 /*
1982 * If we put data in the output pipe, wakeup any potential readers.
1983 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001984 if (ret > 0)
1985 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001986
1987 return ret;
1988}
1989
1990/*
1991 * This is a tee(1) implementation that works on pipes. It doesn't copy
1992 * any data, it simply references the 'in' pages on the 'out' pipe.
1993 * The 'flags' used are the SPLICE_F_* variants, currently the only
1994 * applicable one is SPLICE_F_NONBLOCK.
1995 */
1996static long do_tee(struct file *in, struct file *out, size_t len,
1997 unsigned int flags)
1998{
Linus Torvalds71993e62010-11-28 13:56:09 -08001999 struct pipe_inode_info *ipipe = get_pipe_info(in);
2000 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002001 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02002002
2003 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02002004 * Duplicate the contents of ipipe to opipe without actually
2005 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002006 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002007 if (ipipe && opipe && ipipe != opipe) {
2008 /*
2009 * Keep going, unless we encounter an error. The ipipe/opipe
2010 * ordering doesn't really matter.
2011 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002012 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002013 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002014 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002015 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002016 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002017 }
2018 }
Jens Axboe70524492006-04-11 15:51:17 +02002019
Jens Axboeaadd06e2006-07-10 11:00:01 +02002020 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002021}
2022
Heiko Carstens836f92a2009-01-14 14:14:33 +01002023SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002024{
Al Viro2903ff02012-08-28 12:52:22 -04002025 struct fd in;
2026 int error;
Jens Axboe70524492006-04-11 15:51:17 +02002027
2028 if (unlikely(!len))
2029 return 0;
2030
2031 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04002032 in = fdget(fdin);
2033 if (in.file) {
2034 if (in.file->f_mode & FMODE_READ) {
2035 struct fd out = fdget(fdout);
2036 if (out.file) {
2037 if (out.file->f_mode & FMODE_WRITE)
2038 error = do_tee(in.file, out.file,
2039 len, flags);
2040 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02002041 }
2042 }
Al Viro2903ff02012-08-28 12:52:22 -04002043 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02002044 }
2045
2046 return error;
2047}