blob: 7968da96bebbb5d1cd087cbfa2ece65c09cc8b4a [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 Viro8d020762014-04-05 04:27:08 -040035#include <linux/aio.h>
Al Viro06ae43f2013-03-20 13:19:30 -040036#include "internal.h"
Jens Axboe5274f052006-03-30 15:15:30 +020037
Jens Axboe83f91352006-04-02 23:05:09 +020038/*
39 * Attempt to steal a page from a pipe buffer. This should perhaps go into
40 * a vm helper function, it's already simplified quite a bit by the
41 * addition of remove_mapping(). If success is returned, the caller may
42 * attempt to reuse this page for another destination.
43 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020044static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020045 struct pipe_buffer *buf)
46{
47 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020048 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020049
Jens Axboe9e0267c2006-04-19 15:57:31 +020050 lock_page(page);
51
Jens Axboe9e94cd42006-06-20 15:01:12 +020052 mapping = page_mapping(page);
53 if (mapping) {
54 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020055
Jens Axboe9e94cd42006-06-20 15:01:12 +020056 /*
57 * At least for ext2 with nobh option, we need to wait on
58 * writeback completing on this page, since we'll remove it
59 * from the pagecache. Otherwise truncate wont wait on the
60 * page, allowing the disk blocks to be reused by someone else
61 * before we actually wrote our data to them. fs corruption
62 * ensues.
63 */
64 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020065
David Howells266cf652009-04-03 16:42:36 +010066 if (page_has_private(page) &&
67 !try_to_release_page(page, GFP_KERNEL))
Jens Axboeca39d652008-05-20 21:27:41 +020068 goto out_unlock;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020069
Jens Axboe9e94cd42006-06-20 15:01:12 +020070 /*
71 * If we succeeded in removing the mapping, set LRU flag
72 * and return good.
73 */
74 if (remove_mapping(mapping, page)) {
75 buf->flags |= PIPE_BUF_FLAG_LRU;
76 return 0;
77 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020078 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020079
Jens Axboe9e94cd42006-06-20 15:01:12 +020080 /*
81 * Raced with truncate or failed to remove page from current
82 * address space, unlock and return failure.
83 */
Jens Axboeca39d652008-05-20 21:27:41 +020084out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020085 unlock_page(page);
86 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020087}
88
Jens Axboe76ad4d12006-05-03 10:41:33 +020089static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020090 struct pipe_buffer *buf)
91{
92 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020093 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020094}
95
Jens Axboe08457182007-06-12 20:51:32 +020096/*
97 * Check whether the contents of buf is OK to access. Since the content
98 * is a page cache page, IO may be in flight.
99 */
Jens Axboecac36bb02007-06-14 13:10:48 +0200100static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
101 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +0200102{
103 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200104 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200105
106 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200107 lock_page(page);
108
109 /*
110 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200111 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200112 */
113 if (!page->mapping) {
114 err = -ENODATA;
115 goto error;
116 }
117
118 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200119 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200120 */
121 if (!PageUptodate(page)) {
122 err = -EIO;
123 goto error;
124 }
125
126 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200127 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200128 */
Jens Axboe5274f052006-03-30 15:15:30 +0200129 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200130 }
131
Jens Axboef84d7512006-05-01 19:59:03 +0200132 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200133error:
134 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200135 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200136}
137
Hugh Dickins708e3502011-07-25 17:12:32 -0700138const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200139 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +0200140 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200141 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200142 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200143 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200144};
145
Jens Axboe912d35f2006-04-26 10:59:21 +0200146static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
147 struct pipe_buffer *buf)
148{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200149 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
150 return 1;
151
Jens Axboe14328732006-05-03 10:35:26 +0200152 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200153 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200154}
155
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800156static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200157 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +0200158 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200159 .release = page_cache_pipe_buf_release,
160 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200161 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200162};
163
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200164static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
165{
166 smp_mb();
167 if (waitqueue_active(&pipe->wait))
168 wake_up_interruptible(&pipe->wait);
169 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
170}
171
Jens Axboe932cc6d2007-06-21 13:10:21 +0200172/**
173 * splice_to_pipe - fill passed data into a pipe
174 * @pipe: pipe to fill
175 * @spd: data to fill
176 *
177 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200178 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200179 * the struct pipe_buf_operations associated with these pages. This
180 * function will link that data to the pipe.
181 *
Jens Axboe83f91352006-04-02 23:05:09 +0200182 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200183ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
184 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200185{
Jens Axboe00de00b2007-06-15 13:14:22 +0200186 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200187 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200188
189 ret = 0;
190 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200191 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200192
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200193 pipe_lock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200194
Jens Axboe5274f052006-03-30 15:15:30 +0200195 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200196 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200197 send_sig(SIGPIPE, current, 0);
198 if (!ret)
199 ret = -EPIPE;
200 break;
201 }
202
Jens Axboe35f3d142010-05-20 10:43:18 +0200203 if (pipe->nrbufs < pipe->buffers) {
204 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200205 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200206
Jens Axboe912d35f2006-04-26 10:59:21 +0200207 buf->page = spd->pages[page_nr];
208 buf->offset = spd->partial[page_nr].offset;
209 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200210 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200211 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200212 if (spd->flags & SPLICE_F_GIFT)
213 buf->flags |= PIPE_BUF_FLAG_GIFT;
214
Jens Axboe6f767b02006-04-11 13:53:56 +0200215 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200216 page_nr++;
217 ret += buf->len;
218
Al Viro6447a3c2013-03-21 11:01:38 -0400219 if (pipe->files)
Jens Axboe6f767b02006-04-11 13:53:56 +0200220 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200221
Jens Axboe912d35f2006-04-26 10:59:21 +0200222 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200223 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200224 if (pipe->nrbufs < pipe->buffers)
Jens Axboe5274f052006-03-30 15:15:30 +0200225 continue;
226
227 break;
228 }
229
Jens Axboe912d35f2006-04-26 10:59:21 +0200230 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700231 if (!ret)
232 ret = -EAGAIN;
233 break;
234 }
235
Jens Axboe5274f052006-03-30 15:15:30 +0200236 if (signal_pending(current)) {
237 if (!ret)
238 ret = -ERESTARTSYS;
239 break;
240 }
241
242 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200243 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200244 if (waitqueue_active(&pipe->wait))
245 wake_up_interruptible_sync(&pipe->wait);
246 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200247 do_wakeup = 0;
248 }
249
Ingo Molnar3a326a22006-04-10 15:18:35 +0200250 pipe->waiting_writers++;
251 pipe_wait(pipe);
252 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200253 }
254
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200255 pipe_unlock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200256
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200257 if (do_wakeup)
258 wakeup_pipe_readers(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200259
Jens Axboe00de00b2007-06-15 13:14:22 +0200260 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800261 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200262
263 return ret;
264}
265
Hugh Dickins708e3502011-07-25 17:12:32 -0700266void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800267{
268 page_cache_release(spd->pages[i]);
269}
270
Jens Axboe35f3d142010-05-20 10:43:18 +0200271/*
272 * Check if we need to grow the arrays holding pages and partial page
273 * descriptions.
274 */
Eric Dumazet047fe362012-06-12 15:24:40 +0200275int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200276{
Eric Dumazet047fe362012-06-12 15:24:40 +0200277 unsigned int buffers = ACCESS_ONCE(pipe->buffers);
278
279 spd->nr_pages_max = buffers;
280 if (buffers <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200281 return 0;
282
Eric Dumazet047fe362012-06-12 15:24:40 +0200283 spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL);
284 spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200285
286 if (spd->pages && spd->partial)
287 return 0;
288
289 kfree(spd->pages);
290 kfree(spd->partial);
291 return -ENOMEM;
292}
293
Eric Dumazet047fe362012-06-12 15:24:40 +0200294void splice_shrink_spd(struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200295{
Eric Dumazet047fe362012-06-12 15:24:40 +0200296 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200297 return;
298
299 kfree(spd->pages);
300 kfree(spd->partial);
301}
302
Ingo Molnar3a326a22006-04-10 15:18:35 +0200303static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200304__generic_file_splice_read(struct file *in, loff_t *ppos,
305 struct pipe_inode_info *pipe, size_t len,
306 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200307{
308 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700309 unsigned int loff, nr_pages, req_pages;
Jens Axboe35f3d142010-05-20 10:43:18 +0200310 struct page *pages[PIPE_DEF_BUFFERS];
311 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200312 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200313 pgoff_t index, end_index;
314 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200315 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200316 struct splice_pipe_desc spd = {
317 .pages = pages,
318 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200319 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +0200320 .flags = flags,
321 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800322 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200323 };
Jens Axboe5274f052006-03-30 15:15:30 +0200324
Jens Axboe35f3d142010-05-20 10:43:18 +0200325 if (splice_grow_spd(pipe, &spd))
326 return -ENOMEM;
327
Jens Axboecbb7e572006-04-11 14:57:50 +0200328 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200329 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700330 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Eric Dumazet047fe362012-06-12 15:24:40 +0200331 nr_pages = min(req_pages, spd.nr_pages_max);
Jens Axboe5274f052006-03-30 15:15:30 +0200332
333 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200334 * Lookup the (hopefully) full range of pages we need.
335 */
Jens Axboe35f3d142010-05-20 10:43:18 +0200336 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700337 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200338
339 /*
340 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a48202007-07-19 01:48:05 -0700341 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200342 */
Fengguang Wu431a48202007-07-19 01:48:05 -0700343 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700344 page_cache_sync_readahead(mapping, &in->f_ra, in,
345 index, req_pages - spd.nr_pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700346
Jens Axboe932cc6d2007-06-21 13:10:21 +0200347 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200348 while (spd.nr_pages < nr_pages) {
349 /*
350 * Page could be there, find_get_pages_contig() breaks on
351 * the first hole.
352 */
353 page = find_get_page(mapping, index);
354 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200355 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200356 * page didn't exist, allocate one.
357 */
358 page = page_cache_alloc_cold(mapping);
359 if (!page)
360 break;
361
362 error = add_to_page_cache_lru(page, mapping, index,
Nick Piggin0ae0b5d2010-05-25 10:25:26 +0200363 GFP_KERNEL);
Jens Axboeeb207962006-04-27 11:05:22 +0200364 if (unlikely(error)) {
365 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200366 if (error == -EEXIST)
367 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200368 break;
369 }
370 /*
371 * add_to_page_cache() locks the page, unlock it
372 * to avoid convoluting the logic below even more.
373 */
374 unlock_page(page);
375 }
376
Jens Axboe35f3d142010-05-20 10:43:18 +0200377 spd.pages[spd.nr_pages++] = page;
Jens Axboeeb207962006-04-27 11:05:22 +0200378 index++;
379 }
380
381 /*
382 * Now loop over the map and see if we need to start IO on any
383 * pages, fill in the partial map, etc.
384 */
385 index = *ppos >> PAGE_CACHE_SHIFT;
386 nr_pages = spd.nr_pages;
387 spd.nr_pages = 0;
388 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200389 unsigned int this_len;
390
391 if (!len)
392 break;
393
394 /*
395 * this_len is the max we'll use from this page
396 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200397 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboe35f3d142010-05-20 10:43:18 +0200398 page = spd.pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200399
Fengguang Wua08a1662007-07-19 01:48:03 -0700400 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700401 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700402 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700403
Jens Axboe7480a902006-04-11 13:52:47 +0200404 /*
405 * If the page isn't uptodate, we may need to start io on it
406 */
407 if (!PageUptodate(page)) {
Miklos Szeredi69650312010-08-03 12:48:50 +0200408 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200409
410 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200411 * Page was truncated, or invalidated by the
412 * filesystem. Redo the find/create, but this time the
413 * page is kept locked, so there's no chance of another
414 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200415 */
416 if (!page->mapping) {
417 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200418 page = find_or_create_page(mapping, index,
419 mapping_gfp_mask(mapping));
420
421 if (!page) {
422 error = -ENOMEM;
423 break;
424 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200425 page_cache_release(spd.pages[page_nr]);
426 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200427 }
428 /*
429 * page was already under io and is now done, great
430 */
431 if (PageUptodate(page)) {
432 unlock_page(page);
433 goto fill_it;
434 }
435
Jens Axboe7480a902006-04-11 13:52:47 +0200436 /*
437 * need to read in the page
438 */
439 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200440 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200441 /*
442 * We really should re-lookup the page here,
443 * but it complicates things a lot. Instead
444 * lets just do what we already stored, and
445 * we'll get it the next time we are called.
446 */
Jens Axboe7480a902006-04-11 13:52:47 +0200447 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200448 error = 0;
449
Jens Axboe7480a902006-04-11 13:52:47 +0200450 break;
451 }
Jens Axboe620a3242007-06-07 09:39:42 +0200452 }
453fill_it:
454 /*
455 * i_size must be checked after PageUptodate.
456 */
457 isize = i_size_read(mapping->host);
458 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
459 if (unlikely(!isize || index > end_index))
460 break;
461
462 /*
463 * if this is the last page, see if we need to shrink
464 * the length and stop
465 */
466 if (end_index == index) {
467 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200468
469 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200470 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200471 */
Jens Axboe620a3242007-06-07 09:39:42 +0200472 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
473 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200474 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200475
476 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200477 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200478 */
Jens Axboe620a3242007-06-07 09:39:42 +0200479 this_len = min(this_len, plen - loff);
480 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200481 }
Jens Axboe620a3242007-06-07 09:39:42 +0200482
Jens Axboe35f3d142010-05-20 10:43:18 +0200483 spd.partial[page_nr].offset = loff;
484 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200485 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200486 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200487 spd.nr_pages++;
488 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200489 }
490
Jens Axboeeb207962006-04-27 11:05:22 +0200491 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200492 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200493 * we got, 'nr_pages' is how many pages are in the map.
494 */
495 while (page_nr < nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200496 page_cache_release(spd.pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700497 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200498
Jens Axboe912d35f2006-04-26 10:59:21 +0200499 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200500 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200501
Eric Dumazet047fe362012-06-12 15:24:40 +0200502 splice_shrink_spd(&spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200503 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200504}
505
Jens Axboe83f91352006-04-02 23:05:09 +0200506/**
507 * generic_file_splice_read - splice data from file to a pipe
508 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200509 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200510 * @pipe: pipe to splice to
511 * @len: number of bytes to splice
512 * @flags: splice modifier flags
513 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200514 * Description:
515 * Will read pages from given file and fill them into a pipe. Can be
516 * used as long as the address_space operations for the source implements
517 * a readpage() hook.
518 *
Jens Axboe83f91352006-04-02 23:05:09 +0200519 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200520ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
521 struct pipe_inode_info *pipe, size_t len,
522 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200523{
Jens Axboed366d3982007-06-01 14:54:11 +0200524 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200525 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200526
527 isize = i_size_read(in->f_mapping->host);
528 if (unlikely(*ppos >= isize))
529 return 0;
530
531 left = isize - *ppos;
532 if (unlikely(left < len))
533 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200534
Jens Axboe8191ecd2008-04-10 08:24:25 +0200535 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200536 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200537 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200538 file_accessed(in);
539 }
Jens Axboe5274f052006-03-30 15:15:30 +0200540
541 return ret;
542}
Jens Axboe059a8f32006-04-02 23:06:05 +0200543EXPORT_SYMBOL(generic_file_splice_read);
544
Miklos Szeredi68181732009-05-07 15:37:36 +0200545static const struct pipe_buf_operations default_pipe_buf_ops = {
546 .can_merge = 0,
Miklos Szeredi68181732009-05-07 15:37:36 +0200547 .confirm = generic_pipe_buf_confirm,
548 .release = generic_pipe_buf_release,
549 .steal = generic_pipe_buf_steal,
550 .get = generic_pipe_buf_get,
551};
552
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100553static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe,
554 struct pipe_buffer *buf)
555{
556 return 1;
557}
558
559/* Pipe buffer operations for a socket and similar. */
560const struct pipe_buf_operations nosteal_pipe_buf_ops = {
561 .can_merge = 0,
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100562 .confirm = generic_pipe_buf_confirm,
563 .release = generic_pipe_buf_release,
564 .steal = generic_pipe_buf_nosteal,
565 .get = generic_pipe_buf_get,
566};
567EXPORT_SYMBOL(nosteal_pipe_buf_ops);
568
Miklos Szeredi68181732009-05-07 15:37:36 +0200569static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
570 unsigned long vlen, loff_t offset)
571{
572 mm_segment_t old_fs;
573 loff_t pos = offset;
574 ssize_t res;
575
576 old_fs = get_fs();
577 set_fs(get_ds());
578 /* The cast to a user pointer is valid due to the set_fs() */
579 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
580 set_fs(old_fs);
581
582 return res;
583}
584
Al Viro7bb307e2013-02-23 14:51:48 -0500585ssize_t kernel_write(struct file *file, const char *buf, size_t count,
Miklos Szeredib2858d72009-05-19 11:37:46 +0200586 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200587{
588 mm_segment_t old_fs;
589 ssize_t res;
590
591 old_fs = get_fs();
592 set_fs(get_ds());
593 /* The cast to a user pointer is valid due to the set_fs() */
Al Viro7bb307e2013-02-23 14:51:48 -0500594 res = vfs_write(file, (__force const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200595 set_fs(old_fs);
596
597 return res;
598}
Al Viro7bb307e2013-02-23 14:51:48 -0500599EXPORT_SYMBOL(kernel_write);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200600
Miklos Szeredi68181732009-05-07 15:37:36 +0200601ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
602 struct pipe_inode_info *pipe, size_t len,
603 unsigned int flags)
604{
605 unsigned int nr_pages;
606 unsigned int nr_freed;
607 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200608 struct page *pages[PIPE_DEF_BUFFERS];
609 struct partial_page partial[PIPE_DEF_BUFFERS];
610 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200611 ssize_t res;
612 size_t this_len;
613 int error;
614 int i;
615 struct splice_pipe_desc spd = {
616 .pages = pages,
617 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200618 .nr_pages_max = PIPE_DEF_BUFFERS,
Miklos Szeredi68181732009-05-07 15:37:36 +0200619 .flags = flags,
620 .ops = &default_pipe_buf_ops,
621 .spd_release = spd_release_page,
622 };
623
Jens Axboe35f3d142010-05-20 10:43:18 +0200624 if (splice_grow_spd(pipe, &spd))
625 return -ENOMEM;
626
627 res = -ENOMEM;
628 vec = __vec;
Eric Dumazet047fe362012-06-12 15:24:40 +0200629 if (spd.nr_pages_max > PIPE_DEF_BUFFERS) {
630 vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200631 if (!vec)
632 goto shrink_ret;
633 }
634
Miklos Szeredi68181732009-05-07 15:37:36 +0200635 offset = *ppos & ~PAGE_CACHE_MASK;
636 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
637
Eric Dumazet047fe362012-06-12 15:24:40 +0200638 for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200639 struct page *page;
640
Jens Axboe4f231222009-05-13 08:35:35 +0200641 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200642 error = -ENOMEM;
643 if (!page)
644 goto err;
645
646 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200647 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200648 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200649 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200650 spd.nr_pages++;
651 len -= this_len;
652 offset = 0;
653 }
654
655 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200656 if (res < 0) {
657 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200658 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200659 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200660
661 error = 0;
662 if (!res)
663 goto err;
664
665 nr_freed = 0;
666 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200667 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200668 spd.partial[i].offset = 0;
669 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200670 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200671 __free_page(spd.pages[i]);
672 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200673 nr_freed++;
674 }
675 res -= this_len;
676 }
677 spd.nr_pages -= nr_freed;
678
679 res = splice_to_pipe(pipe, &spd);
680 if (res > 0)
681 *ppos += res;
682
Jens Axboe35f3d142010-05-20 10:43:18 +0200683shrink_ret:
684 if (vec != __vec)
685 kfree(vec);
Eric Dumazet047fe362012-06-12 15:24:40 +0200686 splice_shrink_spd(&spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200687 return res;
688
689err:
Jens Axboe4f231222009-05-13 08:35:35 +0200690 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200691 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200692
Jens Axboe35f3d142010-05-20 10:43:18 +0200693 res = error;
694 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200695}
696EXPORT_SYMBOL(default_file_splice_read);
697
Jens Axboe5274f052006-03-30 15:15:30 +0200698/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200699 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200700 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200701 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200702static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200703 struct pipe_buffer *buf, struct splice_desc *sd)
704{
Jens Axboe6a14b902007-06-14 13:08:55 +0200705 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200706 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100707 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200708
Al Viro72c2d532013-09-22 16:27:52 -0400709 if (!likely(file->f_op->sendpage))
Michał Mirosława8adbe32010-12-17 08:56:44 +0100710 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200711
Eric Dumazet35f9c092012-04-05 03:05:35 +0000712 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000713
714 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000715 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000716
Michał Mirosława8adbe32010-12-17 08:56:44 +0100717 return file->f_op->sendpage(file, buf->page, buf->offset,
718 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200719}
720
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200721static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
722{
723 smp_mb();
724 if (waitqueue_active(&pipe->wait))
725 wake_up_interruptible(&pipe->wait);
726 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
727}
728
729/**
730 * splice_from_pipe_feed - feed available data from a pipe to a file
731 * @pipe: pipe to splice from
732 * @sd: information to @actor
733 * @actor: handler that splices the data
734 *
735 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200736 * This function loops over the pipe and calls @actor to do the
737 * actual moving of a single struct pipe_buffer to the desired
738 * destination. It returns when there's no more buffers left in
739 * the pipe or if the requested number of bytes (@sd->total_len)
740 * have been copied. It returns a positive number (one) if the
741 * pipe needs to be filled with more data, zero if the required
742 * number of bytes have been copied and -errno on error.
743 *
744 * This, together with splice_from_pipe_{begin,end,next}, may be
745 * used to implement the functionality of __splice_from_pipe() when
746 * locking is required around copying the pipe buffers to the
747 * destination.
748 */
Al Viro96f9bc82014-04-05 04:35:49 -0400749static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200750 splice_actor *actor)
751{
752 int ret;
753
754 while (pipe->nrbufs) {
755 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
756 const struct pipe_buf_operations *ops = buf->ops;
757
758 sd->len = buf->len;
759 if (sd->len > sd->total_len)
760 sd->len = sd->total_len;
761
Michał Mirosława8adbe32010-12-17 08:56:44 +0100762 ret = buf->ops->confirm(pipe, buf);
763 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200764 if (ret == -ENODATA)
765 ret = 0;
766 return ret;
767 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100768
769 ret = actor(pipe, buf, sd);
770 if (ret <= 0)
771 return ret;
772
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200773 buf->offset += ret;
774 buf->len -= ret;
775
776 sd->num_spliced += ret;
777 sd->len -= ret;
778 sd->pos += ret;
779 sd->total_len -= ret;
780
781 if (!buf->len) {
782 buf->ops = NULL;
783 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200784 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200785 pipe->nrbufs--;
Al Viro6447a3c2013-03-21 11:01:38 -0400786 if (pipe->files)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200787 sd->need_wakeup = true;
788 }
789
790 if (!sd->total_len)
791 return 0;
792 }
793
794 return 1;
795}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200796
797/**
798 * splice_from_pipe_next - wait for some data to splice from
799 * @pipe: pipe to splice from
800 * @sd: information about the splice operation
801 *
802 * Description:
803 * This function will wait for some data and return a positive
804 * value (one) if pipe buffers are available. It will return zero
805 * or -errno if no more data needs to be spliced.
806 */
Al Viro96f9bc82014-04-05 04:35:49 -0400807static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200808{
809 while (!pipe->nrbufs) {
810 if (!pipe->writers)
811 return 0;
812
813 if (!pipe->waiting_writers && sd->num_spliced)
814 return 0;
815
816 if (sd->flags & SPLICE_F_NONBLOCK)
817 return -EAGAIN;
818
819 if (signal_pending(current))
820 return -ERESTARTSYS;
821
822 if (sd->need_wakeup) {
823 wakeup_pipe_writers(pipe);
824 sd->need_wakeup = false;
825 }
826
827 pipe_wait(pipe);
828 }
829
830 return 1;
831}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200832
833/**
834 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700835 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200836 *
837 * Description:
838 * This function should be called before a loop containing
839 * splice_from_pipe_next() and splice_from_pipe_feed() to
840 * initialize the necessary fields of @sd.
841 */
Al Viro96f9bc82014-04-05 04:35:49 -0400842static void splice_from_pipe_begin(struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200843{
844 sd->num_spliced = 0;
845 sd->need_wakeup = false;
846}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200847
848/**
849 * splice_from_pipe_end - finish splicing from pipe
850 * @pipe: pipe to splice from
851 * @sd: information about the splice operation
852 *
853 * Description:
854 * This function will wake up pipe writers if necessary. It should
855 * be called after a loop containing splice_from_pipe_next() and
856 * splice_from_pipe_feed().
857 */
Al Viro96f9bc82014-04-05 04:35:49 -0400858static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200859{
860 if (sd->need_wakeup)
861 wakeup_pipe_writers(pipe);
862}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200863
Jens Axboe932cc6d2007-06-21 13:10:21 +0200864/**
865 * __splice_from_pipe - splice data from a pipe to given actor
866 * @pipe: pipe to splice from
867 * @sd: information to @actor
868 * @actor: handler that splices the data
869 *
870 * Description:
871 * This function does little more than loop over the pipe and call
872 * @actor to do the actual moving of a single struct pipe_buffer to
873 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
874 * pipe_to_user.
875 *
Jens Axboe83f91352006-04-02 23:05:09 +0200876 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200877ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
878 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200879{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200880 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200881
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200882 splice_from_pipe_begin(sd);
883 do {
884 ret = splice_from_pipe_next(pipe, sd);
885 if (ret > 0)
886 ret = splice_from_pipe_feed(pipe, sd, actor);
887 } while (ret > 0);
888 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200889
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200890 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200891}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100892EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200893
Jens Axboe932cc6d2007-06-21 13:10:21 +0200894/**
895 * splice_from_pipe - splice data from a pipe to a file
896 * @pipe: pipe to splice from
897 * @out: file to splice to
898 * @ppos: position in @out
899 * @len: how many bytes to splice
900 * @flags: splice modifier flags
901 * @actor: handler that splices the data
902 *
903 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200904 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200905 * otherwise it's identical to __splice_from_pipe().
906 *
907 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200908ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
909 loff_t *ppos, size_t len, unsigned int flags,
910 splice_actor *actor)
911{
912 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200913 struct splice_desc sd = {
914 .total_len = len,
915 .flags = flags,
916 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200917 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200918 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200919
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200920 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200921 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200922 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200923
924 return ret;
925}
926
927/**
Al Viro8d020762014-04-05 04:27:08 -0400928 * iter_file_splice_write - splice data from a pipe to a file
929 * @pipe: pipe info
930 * @out: file to write to
931 * @ppos: position in @out
932 * @len: number of bytes to splice
933 * @flags: splice modifier flags
934 *
935 * Description:
936 * Will either move or copy pages (determined by @flags options) from
937 * the given pipe inode to the given file.
938 * This one is ->write_iter-based.
939 *
940 */
941ssize_t
942iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
943 loff_t *ppos, size_t len, unsigned int flags)
944{
945 struct splice_desc sd = {
946 .total_len = len,
947 .flags = flags,
948 .pos = *ppos,
949 .u.file = out,
950 };
951 int nbufs = pipe->buffers;
952 struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
953 GFP_KERNEL);
954 ssize_t ret;
955
956 if (unlikely(!array))
957 return -ENOMEM;
958
959 pipe_lock(pipe);
960
961 splice_from_pipe_begin(&sd);
962 while (sd.total_len) {
963 struct iov_iter from;
Al Viro8d020762014-04-05 04:27:08 -0400964 size_t left;
965 int n, idx;
966
967 ret = splice_from_pipe_next(pipe, &sd);
968 if (ret <= 0)
969 break;
970
971 if (unlikely(nbufs < pipe->buffers)) {
972 kfree(array);
973 nbufs = pipe->buffers;
974 array = kcalloc(nbufs, sizeof(struct bio_vec),
975 GFP_KERNEL);
976 if (!array) {
977 ret = -ENOMEM;
978 break;
979 }
980 }
981
982 /* build the vector */
983 left = sd.total_len;
984 for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) {
985 struct pipe_buffer *buf = pipe->bufs + idx;
986 size_t this_len = buf->len;
987
988 if (this_len > left)
989 this_len = left;
990
991 if (idx == pipe->buffers - 1)
992 idx = -1;
993
994 ret = buf->ops->confirm(pipe, buf);
995 if (unlikely(ret)) {
996 if (ret == -ENODATA)
997 ret = 0;
998 goto done;
999 }
1000
1001 array[n].bv_page = buf->page;
1002 array[n].bv_len = this_len;
1003 array[n].bv_offset = buf->offset;
1004 left -= this_len;
1005 }
1006
Al Viro05afcb72015-01-23 01:08:07 -05001007 iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n,
1008 sd.total_len - left);
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001009 ret = vfs_iter_write(out, &from, &sd.pos);
Al Viro8d020762014-04-05 04:27:08 -04001010 if (ret <= 0)
1011 break;
1012
1013 sd.num_spliced += ret;
1014 sd.total_len -= ret;
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001015 *ppos = sd.pos;
Al Viro8d020762014-04-05 04:27:08 -04001016
1017 /* dismiss the fully eaten buffers, adjust the partial one */
1018 while (ret) {
1019 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
1020 if (ret >= buf->len) {
1021 const struct pipe_buf_operations *ops = buf->ops;
1022 ret -= buf->len;
1023 buf->len = 0;
1024 buf->ops = NULL;
1025 ops->release(pipe, buf);
1026 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1027 pipe->nrbufs--;
1028 if (pipe->files)
1029 sd.need_wakeup = true;
1030 } else {
1031 buf->offset += ret;
1032 buf->len -= ret;
1033 ret = 0;
1034 }
1035 }
1036 }
1037done:
1038 kfree(array);
1039 splice_from_pipe_end(pipe, &sd);
1040
1041 pipe_unlock(pipe);
1042
1043 if (sd.num_spliced)
1044 ret = sd.num_spliced;
1045
1046 return ret;
1047}
1048
1049EXPORT_SYMBOL(iter_file_splice_write);
1050
Miklos Szeredib2858d72009-05-19 11:37:46 +02001051static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1052 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001053{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001054 int ret;
1055 void *data;
Al Viro06ae43f2013-03-20 13:19:30 -04001056 loff_t tmp = sd->pos;
Miklos Szeredib2858d72009-05-19 11:37:46 +02001057
Al Virofbb32752014-02-02 21:09:54 -05001058 data = kmap(buf->page);
Al Viro06ae43f2013-03-20 13:19:30 -04001059 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
Al Virofbb32752014-02-02 21:09:54 -05001060 kunmap(buf->page);
Miklos Szeredib2858d72009-05-19 11:37:46 +02001061
1062 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001063}
1064
1065static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1066 struct file *out, loff_t *ppos,
1067 size_t len, unsigned int flags)
1068{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001069 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001070
Miklos Szeredib2858d72009-05-19 11:37:46 +02001071 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1072 if (ret > 0)
1073 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001074
Miklos Szeredib2858d72009-05-19 11:37:46 +02001075 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001076}
1077
Jens Axboe83f91352006-04-02 23:05:09 +02001078/**
1079 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001080 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001081 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001082 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001083 * @len: number of bytes to splice
1084 * @flags: splice modifier flags
1085 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001086 * Description:
1087 * Will send @len bytes from the pipe to a network socket. No data copying
1088 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001089 *
1090 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001091ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001092 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001093{
Jens Axboe00522fb2006-04-26 14:39:29 +02001094 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001095}
1096
Jens Axboe059a8f32006-04-02 23:06:05 +02001097EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001098
Jens Axboe83f91352006-04-02 23:05:09 +02001099/*
1100 * Attempt to initiate a splice from pipe to file.
1101 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001102static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001103 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001104{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001105 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1106 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001107
Al Viro72c2d532013-09-22 16:27:52 -04001108 if (out->f_op->splice_write)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001109 splice_write = out->f_op->splice_write;
1110 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001111 splice_write = default_file_splice_write;
1112
Al Viro500368f2013-05-23 20:07:11 -04001113 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001114}
1115
Jens Axboe83f91352006-04-02 23:05:09 +02001116/*
1117 * Attempt to initiate a splice from a file to a pipe.
1118 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001119static long do_splice_to(struct file *in, loff_t *ppos,
1120 struct pipe_inode_info *pipe, size_t len,
1121 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001122{
Miklos Szeredi68181732009-05-07 15:37:36 +02001123 ssize_t (*splice_read)(struct file *, loff_t *,
1124 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001125 int ret;
1126
Jens Axboe49570e92006-04-11 13:56:09 +02001127 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001128 return -EBADF;
1129
Jens Axboecbb7e572006-04-11 14:57:50 +02001130 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001131 if (unlikely(ret < 0))
1132 return ret;
1133
Al Viro72c2d532013-09-22 16:27:52 -04001134 if (in->f_op->splice_read)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001135 splice_read = in->f_op->splice_read;
1136 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001137 splice_read = default_file_splice_read;
1138
1139 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001140}
1141
Jens Axboe932cc6d2007-06-21 13:10:21 +02001142/**
1143 * splice_direct_to_actor - splices data directly between two non-pipes
1144 * @in: file to splice from
1145 * @sd: actor information on where to splice to
1146 * @actor: handles the data splicing
1147 *
1148 * Description:
1149 * This is a special case helper to splice directly between two
1150 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001151 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001152 * that process.
1153 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001154 */
1155ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1156 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001157{
1158 struct pipe_inode_info *pipe;
1159 long ret, bytes;
1160 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001161 size_t len;
1162 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001163
1164 /*
1165 * We require the input being a regular file, as we don't want to
1166 * randomly drop data for eg socket -> socket splicing. Use the
1167 * piped splicing for that!
1168 */
Al Viro496ad9a2013-01-23 17:07:38 -05001169 i_mode = file_inode(in)->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001170 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1171 return -EINVAL;
1172
1173 /*
1174 * neither in nor out is a pipe, setup an internal pipe attached to
1175 * 'out' and transfer the wanted data from 'in' to 'out' through that
1176 */
1177 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001178 if (unlikely(!pipe)) {
Al Viro7bee1302013-03-21 11:04:15 -04001179 pipe = alloc_pipe_info();
Jens Axboeb92ce552006-04-11 13:52:07 +02001180 if (!pipe)
1181 return -ENOMEM;
1182
1183 /*
1184 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001185 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001186 * PIPE_READERS appropriately.
1187 */
1188 pipe->readers = 1;
1189
1190 current->splice_pipe = pipe;
1191 }
1192
1193 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001194 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001195 */
1196 ret = 0;
1197 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001198 len = sd->total_len;
1199 flags = sd->flags;
1200
1201 /*
1202 * Don't block on output, we have to drain the direct pipe.
1203 */
1204 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001205
1206 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001207 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001208 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001209
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001210 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001211 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001212 goto out_release;
1213
1214 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001215 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001216
1217 /*
1218 * NOTE: nonblocking mode only applies to the input. We
1219 * must not do the output in nonblocking mode as then we
1220 * could get stuck data in the internal pipe:
1221 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001222 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001223 if (unlikely(ret <= 0)) {
1224 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001225 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001226 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001227
1228 bytes += ret;
1229 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001230 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001231
Tom Zanussia82c53a2008-05-09 13:28:36 +02001232 if (ret < read_len) {
1233 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001234 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001235 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001236 }
1237
Jens Axboe9e971982008-01-29 21:05:57 +01001238done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001239 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001240 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001241 return bytes;
1242
1243out_release:
1244 /*
1245 * If we did an incomplete transfer we must release
1246 * the pipe buffers in question:
1247 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001248 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001249 struct pipe_buffer *buf = pipe->bufs + i;
1250
1251 if (buf->ops) {
1252 buf->ops->release(pipe, buf);
1253 buf->ops = NULL;
1254 }
1255 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001256
Jens Axboe9e971982008-01-29 21:05:57 +01001257 if (!bytes)
1258 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001259
Jens Axboe9e971982008-01-29 21:05:57 +01001260 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001261}
1262EXPORT_SYMBOL(splice_direct_to_actor);
1263
1264static int direct_splice_actor(struct pipe_inode_info *pipe,
1265 struct splice_desc *sd)
1266{
Jens Axboe6a14b902007-06-14 13:08:55 +02001267 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001268
Al Viro7995bd22013-06-20 18:58:36 +04001269 return do_splice_from(pipe, file, sd->opos, sd->total_len,
Changli Gao2cb4b052010-06-29 13:09:18 +02001270 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001271}
1272
Jens Axboe932cc6d2007-06-21 13:10:21 +02001273/**
1274 * do_splice_direct - splices data directly between two files
1275 * @in: file to splice from
1276 * @ppos: input file offset
1277 * @out: file to splice to
Randy Dunlapacdb37c2013-06-22 19:44:08 -07001278 * @opos: output file offset
Jens Axboe932cc6d2007-06-21 13:10:21 +02001279 * @len: number of bytes to splice
1280 * @flags: splice modifier flags
1281 *
1282 * Description:
1283 * For use by do_sendfile(). splice can easily emulate sendfile, but
1284 * doing it in the application would incur an extra system call
1285 * (splice in + splice out, as compared to just sendfile()). So this helper
1286 * can splice directly through a process-private pipe.
1287 *
1288 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001289long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
Al Viro7995bd22013-06-20 18:58:36 +04001290 loff_t *opos, size_t len, unsigned int flags)
Jens Axboec66ab6f2007-06-12 21:17:17 +02001291{
1292 struct splice_desc sd = {
1293 .len = len,
1294 .total_len = len,
1295 .flags = flags,
1296 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001297 .u.file = out,
Al Viro7995bd22013-06-20 18:58:36 +04001298 .opos = opos,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001299 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001300 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001301
Al Viro18c67cb2013-06-19 15:41:54 +04001302 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1303 return -EBADF;
1304
1305 if (unlikely(out->f_flags & O_APPEND))
1306 return -EINVAL;
1307
1308 ret = rw_verify_area(WRITE, out, opos, len);
1309 if (unlikely(ret < 0))
1310 return ret;
1311
Jens Axboec66ab6f2007-06-12 21:17:17 +02001312 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001313 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001314 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001315
Jens Axboec66ab6f2007-06-12 21:17:17 +02001316 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001317}
Miklos Szeredi1c118592014-10-24 00:14:35 +02001318EXPORT_SYMBOL(do_splice_direct);
Jens Axboeb92ce552006-04-11 13:52:07 +02001319
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001320static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1321 struct pipe_inode_info *opipe,
1322 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001323
1324/*
Jens Axboe83f91352006-04-02 23:05:09 +02001325 * Determine where to splice to/from.
1326 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001327static long do_splice(struct file *in, loff_t __user *off_in,
1328 struct file *out, loff_t __user *off_out,
1329 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001330{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001331 struct pipe_inode_info *ipipe;
1332 struct pipe_inode_info *opipe;
Al Viro7995bd22013-06-20 18:58:36 +04001333 loff_t offset;
Jens Axboea4514eb2006-04-19 15:57:05 +02001334 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001335
Linus Torvalds71993e62010-11-28 13:56:09 -08001336 ipipe = get_pipe_info(in);
1337 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001338
1339 if (ipipe && opipe) {
1340 if (off_in || off_out)
1341 return -ESPIPE;
1342
1343 if (!(in->f_mode & FMODE_READ))
1344 return -EBADF;
1345
1346 if (!(out->f_mode & FMODE_WRITE))
1347 return -EBADF;
1348
1349 /* Splicing to self would be fun, but... */
1350 if (ipipe == opipe)
1351 return -EINVAL;
1352
1353 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1354 }
1355
1356 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001357 if (off_in)
1358 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001359 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001360 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001361 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001362 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001363 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001364 } else {
1365 offset = out->f_pos;
1366 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001367
Al Viro18c67cb2013-06-19 15:41:54 +04001368 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1369 return -EBADF;
1370
1371 if (unlikely(out->f_flags & O_APPEND))
1372 return -EINVAL;
1373
1374 ret = rw_verify_area(WRITE, out, &offset, len);
1375 if (unlikely(ret < 0))
1376 return ret;
1377
Al Viro500368f2013-05-23 20:07:11 -04001378 file_start_write(out);
Al Viro7995bd22013-06-20 18:58:36 +04001379 ret = do_splice_from(ipipe, out, &offset, len, flags);
Al Viro500368f2013-05-23 20:07:11 -04001380 file_end_write(out);
Jens Axboea4514eb2006-04-19 15:57:05 +02001381
Al Viro7995bd22013-06-20 18:58:36 +04001382 if (!off_out)
1383 out->f_pos = offset;
1384 else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001385 ret = -EFAULT;
1386
1387 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001388 }
Jens Axboe5274f052006-03-30 15:15:30 +02001389
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001390 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001391 if (off_out)
1392 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001393 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001394 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001395 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001396 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001397 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001398 } else {
1399 offset = in->f_pos;
1400 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001401
Al Viro7995bd22013-06-20 18:58:36 +04001402 ret = do_splice_to(in, &offset, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001403
Al Viro7995bd22013-06-20 18:58:36 +04001404 if (!off_in)
1405 in->f_pos = offset;
1406 else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001407 ret = -EFAULT;
1408
1409 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001410 }
Jens Axboe5274f052006-03-30 15:15:30 +02001411
1412 return -EINVAL;
1413}
1414
Jens Axboe912d35f2006-04-26 10:59:21 +02001415/*
1416 * Map an iov into an array of pages and offset/length tupples. With the
1417 * partial_page structure, we can map several non-contiguous ranges into
1418 * our ones pages[] map instead of splitting that operation into pieces.
1419 * Could easily be exported as a generic helper for other users, in which
1420 * case one would probably want to add a 'max_nr_pages' parameter as well.
1421 */
1422static int get_iovec_page_array(const struct iovec __user *iov,
1423 unsigned int nr_vecs, struct page **pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001424 struct partial_page *partial, bool aligned,
Jens Axboe35f3d142010-05-20 10:43:18 +02001425 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001426{
1427 int buffers = 0, error = 0;
1428
Jens Axboe912d35f2006-04-26 10:59:21 +02001429 while (nr_vecs) {
1430 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001431 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001432 void __user *base;
1433 size_t len;
1434 int i;
1435
Linus Torvalds75723952007-10-01 13:17:28 -07001436 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001437 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001438 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001439
1440 base = entry.iov_base;
1441 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001442
1443 /*
1444 * Sanity check this iovec. 0 read succeeds.
1445 */
Linus Torvalds75723952007-10-01 13:17:28 -07001446 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001447 if (unlikely(!len))
1448 break;
1449 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001450 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001451 break;
1452
1453 /*
1454 * Get this base offset and number of pages, then map
1455 * in the user pages.
1456 */
1457 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001458
1459 /*
1460 * If asked for alignment, the offset must be zero and the
1461 * length a multiple of the PAGE_SIZE.
1462 */
1463 error = -EINVAL;
1464 if (aligned && (off || len & ~PAGE_MASK))
1465 break;
1466
Jens Axboe912d35f2006-04-26 10:59:21 +02001467 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001468 if (npages > pipe_buffers - buffers)
1469 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001470
Nick Pigginbc40d732008-07-25 19:45:26 -07001471 error = get_user_pages_fast((unsigned long)base, npages,
1472 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001473
1474 if (unlikely(error <= 0))
1475 break;
1476
1477 /*
1478 * Fill this contiguous range into the partial page map.
1479 */
1480 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001481 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001482
1483 partial[buffers].offset = off;
1484 partial[buffers].len = plen;
1485
1486 off = 0;
1487 len -= plen;
1488 buffers++;
1489 }
1490
1491 /*
1492 * We didn't complete this iov, stop here since it probably
1493 * means we have to move some of this into a pipe to
1494 * be able to continue.
1495 */
1496 if (len)
1497 break;
1498
1499 /*
1500 * Don't continue if we mapped fewer pages than we asked for,
1501 * or if we mapped the max number of pages that we have
1502 * room for.
1503 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001504 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001505 break;
1506
1507 nr_vecs--;
1508 iov++;
1509 }
1510
Jens Axboe912d35f2006-04-26 10:59:21 +02001511 if (buffers)
1512 return buffers;
1513
1514 return error;
1515}
1516
Jens Axboe6a14b902007-06-14 13:08:55 +02001517static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1518 struct splice_desc *sd)
1519{
Al Viro6130f532014-02-03 18:19:51 -05001520 int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1521 return n == sd->len ? n : -EFAULT;
Jens Axboe6a14b902007-06-14 13:08:55 +02001522}
1523
1524/*
1525 * For lack of a better implementation, implement vmsplice() to userspace
1526 * as a simple copy of the pipes pages to the user iov.
1527 */
Al Viro6130f532014-02-03 18:19:51 -05001528static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov,
Jens Axboe6a14b902007-06-14 13:08:55 +02001529 unsigned long nr_segs, unsigned int flags)
1530{
1531 struct pipe_inode_info *pipe;
1532 struct splice_desc sd;
Jens Axboe6a14b902007-06-14 13:08:55 +02001533 long ret;
Al Viro6130f532014-02-03 18:19:51 -05001534 struct iovec iovstack[UIO_FASTIOV];
1535 struct iovec *iov = iovstack;
1536 struct iov_iter iter;
Miklos Szeredib6dd6f42014-05-27 16:41:16 +02001537 ssize_t count;
Jens Axboe6a14b902007-06-14 13:08:55 +02001538
Linus Torvalds71993e62010-11-28 13:56:09 -08001539 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001540 if (!pipe)
1541 return -EBADF;
1542
Al Viro6130f532014-02-03 18:19:51 -05001543 ret = rw_copy_check_uvector(READ, uiov, nr_segs,
1544 ARRAY_SIZE(iovstack), iovstack, &iov);
1545 if (ret <= 0)
Miklos Szeredib6dd6f42014-05-27 16:41:16 +02001546 goto out;
Al Viro6130f532014-02-03 18:19:51 -05001547
Miklos Szeredib6dd6f42014-05-27 16:41:16 +02001548 count = ret;
Al Viro71d8e532014-03-05 19:28:09 -05001549 iov_iter_init(&iter, READ, iov, nr_segs, count);
Al Viro6130f532014-02-03 18:19:51 -05001550
1551 sd.len = 0;
1552 sd.total_len = count;
1553 sd.flags = flags;
1554 sd.u.data = &iter;
1555 sd.pos = 0;
1556
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001557 pipe_lock(pipe);
Al Viro6130f532014-02-03 18:19:51 -05001558 ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001559 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001560
Miklos Szeredib6dd6f42014-05-27 16:41:16 +02001561out:
Al Viro6130f532014-02-03 18:19:51 -05001562 if (iov != iovstack)
1563 kfree(iov);
Jens Axboe6a14b902007-06-14 13:08:55 +02001564
1565 return ret;
1566}
1567
Jens Axboe912d35f2006-04-26 10:59:21 +02001568/*
1569 * vmsplice splices a user address range into a pipe. It can be thought of
1570 * as splice-from-memory, where the regular splice is splice-from-file (or
1571 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001572 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001573static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1574 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001575{
Jens Axboeddac0d32006-11-04 12:49:32 +01001576 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001577 struct page *pages[PIPE_DEF_BUFFERS];
1578 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001579 struct splice_pipe_desc spd = {
1580 .pages = pages,
1581 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001582 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +02001583 .flags = flags,
1584 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001585 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001586 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001587 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001588
Linus Torvalds71993e62010-11-28 13:56:09 -08001589 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001590 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001591 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001592
Jens Axboe35f3d142010-05-20 10:43:18 +02001593 if (splice_grow_spd(pipe, &spd))
1594 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001595
Jens Axboe35f3d142010-05-20 10:43:18 +02001596 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001597 spd.partial, false,
Eric Dumazet047fe362012-06-12 15:24:40 +02001598 spd.nr_pages_max);
Jens Axboe35f3d142010-05-20 10:43:18 +02001599 if (spd.nr_pages <= 0)
1600 ret = spd.nr_pages;
1601 else
1602 ret = splice_to_pipe(pipe, &spd);
1603
Eric Dumazet047fe362012-06-12 15:24:40 +02001604 splice_shrink_spd(&spd);
Jens Axboe35f3d142010-05-20 10:43:18 +02001605 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001606}
1607
Jens Axboe6a14b902007-06-14 13:08:55 +02001608/*
1609 * Note that vmsplice only really supports true splicing _from_ user memory
1610 * to a pipe, not the other way around. Splicing from user memory is a simple
1611 * operation that can be supported without any funky alignment restrictions
1612 * or nasty vm tricks. We simply map in the user memory and fill them into
1613 * a pipe. The reverse isn't quite as easy, though. There are two possible
1614 * solutions for that:
1615 *
1616 * - memcpy() the data internally, at which point we might as well just
1617 * do a regular read() on the buffer anyway.
1618 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1619 * has restriction limitations on both ends of the pipe).
1620 *
1621 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1622 *
1623 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001624SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1625 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001626{
Al Viro2903ff02012-08-28 12:52:22 -04001627 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001628 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001629
Jens Axboe6a14b902007-06-14 13:08:55 +02001630 if (unlikely(nr_segs > UIO_MAXIOV))
1631 return -EINVAL;
1632 else if (unlikely(!nr_segs))
1633 return 0;
1634
Jens Axboe912d35f2006-04-26 10:59:21 +02001635 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001636 f = fdget(fd);
1637 if (f.file) {
1638 if (f.file->f_mode & FMODE_WRITE)
1639 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1640 else if (f.file->f_mode & FMODE_READ)
1641 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001642
Al Viro2903ff02012-08-28 12:52:22 -04001643 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001644 }
1645
1646 return error;
1647}
1648
Al Viro76b021d2013-03-02 10:19:56 -05001649#ifdef CONFIG_COMPAT
1650COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1651 unsigned int, nr_segs, unsigned int, flags)
1652{
1653 unsigned i;
1654 struct iovec __user *iov;
1655 if (nr_segs > UIO_MAXIOV)
1656 return -EINVAL;
1657 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1658 for (i = 0; i < nr_segs; i++) {
1659 struct compat_iovec v;
1660 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1661 get_user(v.iov_len, &iov32[i].iov_len) ||
1662 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1663 put_user(v.iov_len, &iov[i].iov_len))
1664 return -EFAULT;
1665 }
1666 return sys_vmsplice(fd, iov, nr_segs, flags);
1667}
1668#endif
1669
Heiko Carstens836f92a2009-01-14 14:14:33 +01001670SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1671 int, fd_out, loff_t __user *, off_out,
1672 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001673{
Al Viro2903ff02012-08-28 12:52:22 -04001674 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001675 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001676
1677 if (unlikely(!len))
1678 return 0;
1679
1680 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001681 in = fdget(fd_in);
1682 if (in.file) {
1683 if (in.file->f_mode & FMODE_READ) {
1684 out = fdget(fd_out);
1685 if (out.file) {
1686 if (out.file->f_mode & FMODE_WRITE)
1687 error = do_splice(in.file, off_in,
1688 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001689 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001690 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001691 }
1692 }
Al Viro2903ff02012-08-28 12:52:22 -04001693 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001694 }
Jens Axboe5274f052006-03-30 15:15:30 +02001695 return error;
1696}
Jens Axboe70524492006-04-11 15:51:17 +02001697
1698/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001699 * Make sure there's data to read. Wait for input if we can, otherwise
1700 * return an appropriate error.
1701 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001702static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001703{
1704 int ret;
1705
1706 /*
1707 * Check ->nrbufs without the inode lock first. This function
1708 * is speculative anyways, so missing one is ok.
1709 */
1710 if (pipe->nrbufs)
1711 return 0;
1712
1713 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001714 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001715
1716 while (!pipe->nrbufs) {
1717 if (signal_pending(current)) {
1718 ret = -ERESTARTSYS;
1719 break;
1720 }
1721 if (!pipe->writers)
1722 break;
1723 if (!pipe->waiting_writers) {
1724 if (flags & SPLICE_F_NONBLOCK) {
1725 ret = -EAGAIN;
1726 break;
1727 }
1728 }
1729 pipe_wait(pipe);
1730 }
1731
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001732 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001733 return ret;
1734}
1735
1736/*
1737 * Make sure there's writeable room. Wait for room if we can, otherwise
1738 * return an appropriate error.
1739 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001740static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001741{
1742 int ret;
1743
1744 /*
1745 * Check ->nrbufs without the inode lock first. This function
1746 * is speculative anyways, so missing one is ok.
1747 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001748 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001749 return 0;
1750
1751 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001752 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001753
Jens Axboe35f3d142010-05-20 10:43:18 +02001754 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001755 if (!pipe->readers) {
1756 send_sig(SIGPIPE, current, 0);
1757 ret = -EPIPE;
1758 break;
1759 }
1760 if (flags & SPLICE_F_NONBLOCK) {
1761 ret = -EAGAIN;
1762 break;
1763 }
1764 if (signal_pending(current)) {
1765 ret = -ERESTARTSYS;
1766 break;
1767 }
1768 pipe->waiting_writers++;
1769 pipe_wait(pipe);
1770 pipe->waiting_writers--;
1771 }
1772
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001773 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001774 return ret;
1775}
1776
1777/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001778 * Splice contents of ipipe to opipe.
1779 */
1780static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1781 struct pipe_inode_info *opipe,
1782 size_t len, unsigned int flags)
1783{
1784 struct pipe_buffer *ibuf, *obuf;
1785 int ret = 0, nbuf;
1786 bool input_wakeup = false;
1787
1788
1789retry:
1790 ret = ipipe_prep(ipipe, flags);
1791 if (ret)
1792 return ret;
1793
1794 ret = opipe_prep(opipe, flags);
1795 if (ret)
1796 return ret;
1797
1798 /*
1799 * Potential ABBA deadlock, work around it by ordering lock
1800 * grabbing by pipe info address. Otherwise two different processes
1801 * could deadlock (one doing tee from A -> B, the other from B -> A).
1802 */
1803 pipe_double_lock(ipipe, opipe);
1804
1805 do {
1806 if (!opipe->readers) {
1807 send_sig(SIGPIPE, current, 0);
1808 if (!ret)
1809 ret = -EPIPE;
1810 break;
1811 }
1812
1813 if (!ipipe->nrbufs && !ipipe->writers)
1814 break;
1815
1816 /*
1817 * Cannot make any progress, because either the input
1818 * pipe is empty or the output pipe is full.
1819 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001820 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001821 /* Already processed some buffers, break */
1822 if (ret)
1823 break;
1824
1825 if (flags & SPLICE_F_NONBLOCK) {
1826 ret = -EAGAIN;
1827 break;
1828 }
1829
1830 /*
1831 * We raced with another reader/writer and haven't
1832 * managed to process any buffers. A zero return
1833 * value means EOF, so retry instead.
1834 */
1835 pipe_unlock(ipipe);
1836 pipe_unlock(opipe);
1837 goto retry;
1838 }
1839
1840 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001841 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001842 obuf = opipe->bufs + nbuf;
1843
1844 if (len >= ibuf->len) {
1845 /*
1846 * Simply move the whole buffer from ipipe to opipe
1847 */
1848 *obuf = *ibuf;
1849 ibuf->ops = NULL;
1850 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001851 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001852 ipipe->nrbufs--;
1853 input_wakeup = true;
1854 } else {
1855 /*
1856 * Get a reference to this pipe buffer,
1857 * so we can copy the contents over.
1858 */
1859 ibuf->ops->get(ipipe, ibuf);
1860 *obuf = *ibuf;
1861
1862 /*
1863 * Don't inherit the gift flag, we need to
1864 * prevent multiple steals of this page.
1865 */
1866 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1867
1868 obuf->len = len;
1869 opipe->nrbufs++;
1870 ibuf->offset += obuf->len;
1871 ibuf->len -= obuf->len;
1872 }
1873 ret += obuf->len;
1874 len -= obuf->len;
1875 } while (len);
1876
1877 pipe_unlock(ipipe);
1878 pipe_unlock(opipe);
1879
1880 /*
1881 * If we put data in the output pipe, wakeup any potential readers.
1882 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001883 if (ret > 0)
1884 wakeup_pipe_readers(opipe);
1885
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001886 if (input_wakeup)
1887 wakeup_pipe_writers(ipipe);
1888
1889 return ret;
1890}
1891
1892/*
Jens Axboe70524492006-04-11 15:51:17 +02001893 * Link contents of ipipe to opipe.
1894 */
1895static int link_pipe(struct pipe_inode_info *ipipe,
1896 struct pipe_inode_info *opipe,
1897 size_t len, unsigned int flags)
1898{
1899 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001900 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001901
1902 /*
1903 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001904 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001905 * could deadlock (one doing tee from A -> B, the other from B -> A).
1906 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001907 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001908
Jens Axboeaadd06e2006-07-10 11:00:01 +02001909 do {
Jens Axboe70524492006-04-11 15:51:17 +02001910 if (!opipe->readers) {
1911 send_sig(SIGPIPE, current, 0);
1912 if (!ret)
1913 ret = -EPIPE;
1914 break;
1915 }
Jens Axboe70524492006-04-11 15:51:17 +02001916
1917 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001918 * If we have iterated all input buffers or ran out of
1919 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001920 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001921 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001922 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001923
Jens Axboe35f3d142010-05-20 10:43:18 +02001924 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1925 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001926
Jens Axboe2a27250e2006-04-19 15:56:40 +02001927 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001928 * Get a reference to this pipe buffer,
1929 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001930 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001931 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001932
Jens Axboeaadd06e2006-07-10 11:00:01 +02001933 obuf = opipe->bufs + nbuf;
1934 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001935
Jens Axboeaadd06e2006-07-10 11:00:01 +02001936 /*
1937 * Don't inherit the gift flag, we need to
1938 * prevent multiple steals of this page.
1939 */
1940 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1941
1942 if (obuf->len > len)
1943 obuf->len = len;
1944
1945 opipe->nrbufs++;
1946 ret += obuf->len;
1947 len -= obuf->len;
1948 i++;
1949 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001950
Jens Axboe02cf01a2008-02-20 10:34:51 +01001951 /*
1952 * return EAGAIN if we have the potential of some data in the
1953 * future, otherwise just return 0
1954 */
1955 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1956 ret = -EAGAIN;
1957
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001958 pipe_unlock(ipipe);
1959 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001960
Jens Axboeaadd06e2006-07-10 11:00:01 +02001961 /*
1962 * If we put data in the output pipe, wakeup any potential readers.
1963 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001964 if (ret > 0)
1965 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001966
1967 return ret;
1968}
1969
1970/*
1971 * This is a tee(1) implementation that works on pipes. It doesn't copy
1972 * any data, it simply references the 'in' pages on the 'out' pipe.
1973 * The 'flags' used are the SPLICE_F_* variants, currently the only
1974 * applicable one is SPLICE_F_NONBLOCK.
1975 */
1976static long do_tee(struct file *in, struct file *out, size_t len,
1977 unsigned int flags)
1978{
Linus Torvalds71993e62010-11-28 13:56:09 -08001979 struct pipe_inode_info *ipipe = get_pipe_info(in);
1980 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001981 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001982
1983 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001984 * Duplicate the contents of ipipe to opipe without actually
1985 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001986 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001987 if (ipipe && opipe && ipipe != opipe) {
1988 /*
1989 * Keep going, unless we encounter an error. The ipipe/opipe
1990 * ordering doesn't really matter.
1991 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001992 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001993 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001994 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01001995 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001996 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001997 }
1998 }
Jens Axboe70524492006-04-11 15:51:17 +02001999
Jens Axboeaadd06e2006-07-10 11:00:01 +02002000 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002001}
2002
Heiko Carstens836f92a2009-01-14 14:14:33 +01002003SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002004{
Al Viro2903ff02012-08-28 12:52:22 -04002005 struct fd in;
2006 int error;
Jens Axboe70524492006-04-11 15:51:17 +02002007
2008 if (unlikely(!len))
2009 return 0;
2010
2011 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04002012 in = fdget(fdin);
2013 if (in.file) {
2014 if (in.file->f_mode & FMODE_READ) {
2015 struct fd out = fdget(fdout);
2016 if (out.file) {
2017 if (out.file->f_mode & FMODE_WRITE)
2018 error = do_tee(in.file, out.file,
2019 len, flags);
2020 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02002021 }
2022 }
Al Viro2903ff02012-08-28 12:52:22 -04002023 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02002024 }
2025
2026 return error;
2027}