blob: e95a36228863affa3d6c19a37ac4766f8e147412 [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>
Jens Axboe5274f052006-03-30 15:15:30 +020024#include <linux/mm_inline.h>
Jens Axboe5abc97a2006-03-30 15:16:46 +020025#include <linux/swap.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020026#include <linux/writeback.h>
27#include <linux/buffer_head.h>
Jeff Garzika0f06782006-03-30 23:06:13 -050028#include <linux/module.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>
Jens Axboe5274f052006-03-30 15:15:30 +020032
Jens Axboe83f91352006-04-02 23:05:09 +020033/*
34 * Attempt to steal a page from a pipe buffer. This should perhaps go into
35 * a vm helper function, it's already simplified quite a bit by the
36 * addition of remove_mapping(). If success is returned, the caller may
37 * attempt to reuse this page for another destination.
38 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020039static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020040 struct pipe_buffer *buf)
41{
42 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020043 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020044
Jens Axboe9e0267c2006-04-19 15:57:31 +020045 lock_page(page);
46
Jens Axboe9e94cd42006-06-20 15:01:12 +020047 mapping = page_mapping(page);
48 if (mapping) {
49 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020050
Jens Axboe9e94cd42006-06-20 15:01:12 +020051 /*
52 * At least for ext2 with nobh option, we need to wait on
53 * writeback completing on this page, since we'll remove it
54 * from the pagecache. Otherwise truncate wont wait on the
55 * page, allowing the disk blocks to be reused by someone else
56 * before we actually wrote our data to them. fs corruption
57 * ensues.
58 */
59 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020060
Jens Axboe9e94cd42006-06-20 15:01:12 +020061 if (PagePrivate(page))
Nick Piggin2ae88142006-10-28 10:38:23 -070062 try_to_release_page(page, GFP_KERNEL);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020063
Jens Axboe9e94cd42006-06-20 15:01:12 +020064 /*
65 * If we succeeded in removing the mapping, set LRU flag
66 * and return good.
67 */
68 if (remove_mapping(mapping, page)) {
69 buf->flags |= PIPE_BUF_FLAG_LRU;
70 return 0;
71 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020072 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020073
Jens Axboe9e94cd42006-06-20 15:01:12 +020074 /*
75 * Raced with truncate or failed to remove page from current
76 * address space, unlock and return failure.
77 */
78 unlock_page(page);
79 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020080}
81
Jens Axboe76ad4d12006-05-03 10:41:33 +020082static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020083 struct pipe_buffer *buf)
84{
85 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020086 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020087}
88
Jens Axboe08457182007-06-12 20:51:32 +020089/*
90 * Check whether the contents of buf is OK to access. Since the content
91 * is a page cache page, IO may be in flight.
92 */
Jens Axboecac36bb02007-06-14 13:10:48 +020093static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
94 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +020095{
96 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +020097 int err;
Jens Axboe5274f052006-03-30 15:15:30 +020098
99 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200100 lock_page(page);
101
102 /*
103 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200104 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200105 */
106 if (!page->mapping) {
107 err = -ENODATA;
108 goto error;
109 }
110
111 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200112 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200113 */
114 if (!PageUptodate(page)) {
115 err = -EIO;
116 goto error;
117 }
118
119 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200120 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200121 */
Jens Axboe5274f052006-03-30 15:15:30 +0200122 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200123 }
124
Jens Axboef84d7512006-05-01 19:59:03 +0200125 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200126error:
127 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200128 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200129}
130
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800131static const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200132 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200133 .map = generic_pipe_buf_map,
134 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200135 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200136 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200137 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200138 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200139};
140
Jens Axboe912d35f2006-04-26 10:59:21 +0200141static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
142 struct pipe_buffer *buf)
143{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200144 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
145 return 1;
146
Jens Axboe14328732006-05-03 10:35:26 +0200147 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200148 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200149}
150
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800151static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200152 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200153 .map = generic_pipe_buf_map,
154 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200155 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200156 .release = page_cache_pipe_buf_release,
157 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200158 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200159};
160
Jens Axboe932cc6d2007-06-21 13:10:21 +0200161/**
162 * splice_to_pipe - fill passed data into a pipe
163 * @pipe: pipe to fill
164 * @spd: data to fill
165 *
166 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200167 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200168 * the struct pipe_buf_operations associated with these pages. This
169 * function will link that data to the pipe.
170 *
Jens Axboe83f91352006-04-02 23:05:09 +0200171 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200172ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
173 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200174{
Jens Axboe00de00b2007-06-15 13:14:22 +0200175 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200176 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200177
178 ret = 0;
179 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200180 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200181
Ingo Molnar3a326a22006-04-10 15:18:35 +0200182 if (pipe->inode)
183 mutex_lock(&pipe->inode->i_mutex);
Jens Axboe5274f052006-03-30 15:15:30 +0200184
Jens Axboe5274f052006-03-30 15:15:30 +0200185 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200186 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200187 send_sig(SIGPIPE, current, 0);
188 if (!ret)
189 ret = -EPIPE;
190 break;
191 }
192
Jens Axboe6f767b02006-04-11 13:53:56 +0200193 if (pipe->nrbufs < PIPE_BUFFERS) {
194 int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200195 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200196
Jens Axboe912d35f2006-04-26 10:59:21 +0200197 buf->page = spd->pages[page_nr];
198 buf->offset = spd->partial[page_nr].offset;
199 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200200 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200201 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200202 if (spd->flags & SPLICE_F_GIFT)
203 buf->flags |= PIPE_BUF_FLAG_GIFT;
204
Jens Axboe6f767b02006-04-11 13:53:56 +0200205 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200206 page_nr++;
207 ret += buf->len;
208
Jens Axboe6f767b02006-04-11 13:53:56 +0200209 if (pipe->inode)
210 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200211
Jens Axboe912d35f2006-04-26 10:59:21 +0200212 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200213 break;
Jens Axboe6f767b02006-04-11 13:53:56 +0200214 if (pipe->nrbufs < PIPE_BUFFERS)
Jens Axboe5274f052006-03-30 15:15:30 +0200215 continue;
216
217 break;
218 }
219
Jens Axboe912d35f2006-04-26 10:59:21 +0200220 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700221 if (!ret)
222 ret = -EAGAIN;
223 break;
224 }
225
Jens Axboe5274f052006-03-30 15:15:30 +0200226 if (signal_pending(current)) {
227 if (!ret)
228 ret = -ERESTARTSYS;
229 break;
230 }
231
232 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200233 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200234 if (waitqueue_active(&pipe->wait))
235 wake_up_interruptible_sync(&pipe->wait);
236 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200237 do_wakeup = 0;
238 }
239
Ingo Molnar3a326a22006-04-10 15:18:35 +0200240 pipe->waiting_writers++;
241 pipe_wait(pipe);
242 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200243 }
244
Jens Axboe02676e52007-06-15 13:16:13 +0200245 if (pipe->inode) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200246 mutex_unlock(&pipe->inode->i_mutex);
Jens Axboe5274f052006-03-30 15:15:30 +0200247
Jens Axboe02676e52007-06-15 13:16:13 +0200248 if (do_wakeup) {
249 smp_mb();
250 if (waitqueue_active(&pipe->wait))
251 wake_up_interruptible(&pipe->wait);
252 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
253 }
Jens Axboe5274f052006-03-30 15:15:30 +0200254 }
255
Jens Axboe00de00b2007-06-15 13:14:22 +0200256 while (page_nr < spd_pages)
Jens Axboe912d35f2006-04-26 10:59:21 +0200257 page_cache_release(spd->pages[page_nr++]);
Jens Axboe5274f052006-03-30 15:15:30 +0200258
259 return ret;
260}
261
Ingo Molnar3a326a22006-04-10 15:18:35 +0200262static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200263__generic_file_splice_read(struct file *in, loff_t *ppos,
264 struct pipe_inode_info *pipe, size_t len,
265 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200266{
267 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700268 unsigned int loff, nr_pages, req_pages;
Jens Axboe16c523d2006-04-10 09:03:58 +0200269 struct page *pages[PIPE_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +0200270 struct partial_page partial[PIPE_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200271 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200272 pgoff_t index, end_index;
273 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200274 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200275 struct splice_pipe_desc spd = {
276 .pages = pages,
277 .partial = partial,
278 .flags = flags,
279 .ops = &page_cache_pipe_buf_ops,
280 };
Jens Axboe5274f052006-03-30 15:15:30 +0200281
Jens Axboecbb7e572006-04-11 14:57:50 +0200282 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200283 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700284 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
285 nr_pages = min(req_pages, (unsigned)PIPE_BUFFERS);
Jens Axboe5274f052006-03-30 15:15:30 +0200286
287 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200288 * Lookup the (hopefully) full range of pages we need.
289 */
290 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700291 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200292
293 /*
294 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a4822007-07-19 01:48:05 -0700295 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200296 */
Fengguang Wu431a4822007-07-19 01:48:05 -0700297 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700298 page_cache_sync_readahead(mapping, &in->f_ra, in,
299 index, req_pages - spd.nr_pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700300
Jens Axboe932cc6d2007-06-21 13:10:21 +0200301 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200302 while (spd.nr_pages < nr_pages) {
303 /*
304 * Page could be there, find_get_pages_contig() breaks on
305 * the first hole.
306 */
307 page = find_get_page(mapping, index);
308 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200309 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200310 * page didn't exist, allocate one.
311 */
312 page = page_cache_alloc_cold(mapping);
313 if (!page)
314 break;
315
316 error = add_to_page_cache_lru(page, mapping, index,
Nick Piggin2ae88142006-10-28 10:38:23 -0700317 GFP_KERNEL);
Jens Axboeeb207962006-04-27 11:05:22 +0200318 if (unlikely(error)) {
319 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200320 if (error == -EEXIST)
321 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200322 break;
323 }
324 /*
325 * add_to_page_cache() locks the page, unlock it
326 * to avoid convoluting the logic below even more.
327 */
328 unlock_page(page);
329 }
330
331 pages[spd.nr_pages++] = page;
332 index++;
333 }
334
335 /*
336 * Now loop over the map and see if we need to start IO on any
337 * pages, fill in the partial map, etc.
338 */
339 index = *ppos >> PAGE_CACHE_SHIFT;
340 nr_pages = spd.nr_pages;
341 spd.nr_pages = 0;
342 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200343 unsigned int this_len;
344
345 if (!len)
346 break;
347
348 /*
349 * this_len is the max we'll use from this page
350 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200351 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboeeb207962006-04-27 11:05:22 +0200352 page = pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200353
Fengguang Wua08a1662007-07-19 01:48:03 -0700354 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700355 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700356 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700357
Jens Axboe7480a902006-04-11 13:52:47 +0200358 /*
359 * If the page isn't uptodate, we may need to start io on it
360 */
361 if (!PageUptodate(page)) {
Jens Axboec4f895c2006-04-19 15:56:12 +0200362 /*
363 * If in nonblock mode then dont block on waiting
364 * for an in-flight io page
365 */
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200366 if (flags & SPLICE_F_NONBLOCK) {
367 if (TestSetPageLocked(page))
368 break;
369 } else
370 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200371
372 /*
373 * page was truncated, stop here. if this isn't the
374 * first page, we'll just complete what we already
375 * added
376 */
377 if (!page->mapping) {
378 unlock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200379 break;
380 }
381 /*
382 * page was already under io and is now done, great
383 */
384 if (PageUptodate(page)) {
385 unlock_page(page);
386 goto fill_it;
387 }
388
Jens Axboe7480a902006-04-11 13:52:47 +0200389 /*
390 * need to read in the page
391 */
392 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200393 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200394 /*
395 * We really should re-lookup the page here,
396 * but it complicates things a lot. Instead
397 * lets just do what we already stored, and
398 * we'll get it the next time we are called.
399 */
Jens Axboe7480a902006-04-11 13:52:47 +0200400 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200401 error = 0;
402
Jens Axboe7480a902006-04-11 13:52:47 +0200403 break;
404 }
Jens Axboe620a3242007-06-07 09:39:42 +0200405 }
406fill_it:
407 /*
408 * i_size must be checked after PageUptodate.
409 */
410 isize = i_size_read(mapping->host);
411 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
412 if (unlikely(!isize || index > end_index))
413 break;
414
415 /*
416 * if this is the last page, see if we need to shrink
417 * the length and stop
418 */
419 if (end_index == index) {
420 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200421
422 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200423 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200424 */
Jens Axboe620a3242007-06-07 09:39:42 +0200425 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
426 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200427 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200428
429 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200430 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200431 */
Jens Axboe620a3242007-06-07 09:39:42 +0200432 this_len = min(this_len, plen - loff);
433 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200434 }
Jens Axboe620a3242007-06-07 09:39:42 +0200435
Jens Axboeeb207962006-04-27 11:05:22 +0200436 partial[page_nr].offset = loff;
437 partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200438 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200439 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200440 spd.nr_pages++;
441 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200442 }
443
Jens Axboeeb207962006-04-27 11:05:22 +0200444 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200445 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200446 * we got, 'nr_pages' is how many pages are in the map.
447 */
448 while (page_nr < nr_pages)
449 page_cache_release(pages[page_nr++]);
Fengguang Wua08a1662007-07-19 01:48:03 -0700450 in->f_ra.prev_index = index;
Jens Axboeeb207962006-04-27 11:05:22 +0200451
Jens Axboe912d35f2006-04-26 10:59:21 +0200452 if (spd.nr_pages)
Jens Axboe00522fb2006-04-26 14:39:29 +0200453 return splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200454
Jens Axboe7480a902006-04-11 13:52:47 +0200455 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200456}
457
Jens Axboe83f91352006-04-02 23:05:09 +0200458/**
459 * generic_file_splice_read - splice data from file to a pipe
460 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200461 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200462 * @pipe: pipe to splice to
463 * @len: number of bytes to splice
464 * @flags: splice modifier flags
465 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200466 * Description:
467 * Will read pages from given file and fill them into a pipe. Can be
468 * used as long as the address_space operations for the source implements
469 * a readpage() hook.
470 *
Jens Axboe83f91352006-04-02 23:05:09 +0200471 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200472ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
473 struct pipe_inode_info *pipe, size_t len,
474 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200475{
476 ssize_t spliced;
477 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200478 loff_t isize, left;
479
480 isize = i_size_read(in->f_mapping->host);
481 if (unlikely(*ppos >= isize))
482 return 0;
483
484 left = isize - *ppos;
485 if (unlikely(left < len))
486 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200487
488 ret = 0;
489 spliced = 0;
Jens Axboe51a92c02007-07-13 14:11:43 +0200490 while (len && !spliced) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200491 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200492
Jens Axboec4f895c2006-04-19 15:56:12 +0200493 if (ret < 0)
Jens Axboe5274f052006-03-30 15:15:30 +0200494 break;
Jens Axboec4f895c2006-04-19 15:56:12 +0200495 else if (!ret) {
496 if (spliced)
497 break;
498 if (flags & SPLICE_F_NONBLOCK) {
499 ret = -EAGAIN;
500 break;
501 }
502 }
Jens Axboe5274f052006-03-30 15:15:30 +0200503
Jens Axboecbb7e572006-04-11 14:57:50 +0200504 *ppos += ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200505 len -= ret;
506 spliced += ret;
507 }
508
509 if (spliced)
510 return spliced;
511
512 return ret;
513}
514
Jens Axboe059a8f32006-04-02 23:06:05 +0200515EXPORT_SYMBOL(generic_file_splice_read);
516
Jens Axboe5274f052006-03-30 15:15:30 +0200517/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200518 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200519 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200520 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200521static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200522 struct pipe_buffer *buf, struct splice_desc *sd)
523{
Jens Axboe6a14b902007-06-14 13:08:55 +0200524 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200525 loff_t pos = sd->pos;
Jens Axboef84d7512006-05-01 19:59:03 +0200526 int ret, more;
Jens Axboe5274f052006-03-30 15:15:30 +0200527
Jens Axboecac36bb02007-06-14 13:10:48 +0200528 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200529 if (!ret) {
530 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200531
Jens Axboef84d7512006-05-01 19:59:03 +0200532 ret = file->f_op->sendpage(file, buf->page, buf->offset,
533 sd->len, &pos, more);
534 }
Jens Axboe5274f052006-03-30 15:15:30 +0200535
Jens Axboe016b6612006-04-25 15:42:00 +0200536 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200537}
538
539/*
540 * This is a little more tricky than the file -> pipe splicing. There are
541 * basically three cases:
542 *
543 * - Destination page already exists in the address space and there
544 * are users of it. For that case we have no other option that
545 * copying the data. Tough luck.
546 * - Destination page already exists in the address space, but there
547 * are no users of it. Make sure it's uptodate, then drop it. Fall
548 * through to last case.
549 * - Destination page does not exist, we can add the pipe page to
550 * the page cache and avoid the copy.
551 *
Jens Axboe83f91352006-04-02 23:05:09 +0200552 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
553 * sd->flags), we attempt to migrate pages from the pipe to the output
554 * file address space page cache. This is possible if no one else has
555 * the pipe page referenced outside of the pipe and page cache. If
556 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
557 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200558 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200559static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
Jens Axboe5274f052006-03-30 15:15:30 +0200560 struct splice_desc *sd)
561{
Jens Axboe6a14b902007-06-14 13:08:55 +0200562 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200563 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200564 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200565 struct page *page;
Jens Axboe5274f052006-03-30 15:15:30 +0200566 pgoff_t index;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200567 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200568
569 /*
Jens Axboe49d0b212006-04-10 09:04:41 +0200570 * make sure the data in this buffer is uptodate
Jens Axboe5274f052006-03-30 15:15:30 +0200571 */
Jens Axboecac36bb02007-06-14 13:10:48 +0200572 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200573 if (unlikely(ret))
574 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200575
576 index = sd->pos >> PAGE_CACHE_SHIFT;
577 offset = sd->pos & ~PAGE_CACHE_MASK;
578
Jens Axboe016b6612006-04-25 15:42:00 +0200579 this_len = sd->len;
580 if (this_len + offset > PAGE_CACHE_SIZE)
581 this_len = PAGE_CACHE_SIZE - offset;
582
Jens Axboe5abc97a2006-03-30 15:16:46 +0200583find_page:
Nick Piggin485ddb42007-03-27 08:55:08 +0200584 page = find_lock_page(mapping, index);
585 if (!page) {
586 ret = -ENOMEM;
587 page = page_cache_alloc_cold(mapping);
588 if (unlikely(!page))
589 goto out_ret;
Jens Axboe9e0267c2006-04-19 15:57:31 +0200590
Nick Piggin485ddb42007-03-27 08:55:08 +0200591 /*
592 * This will also lock the page
593 */
594 ret = add_to_page_cache_lru(page, mapping, index,
595 GFP_KERNEL);
596 if (unlikely(ret))
Jens Axboe6a860c92007-07-20 15:18:12 +0200597 goto out_release;
Nick Piggin485ddb42007-03-27 08:55:08 +0200598 }
599
Jens Axboe016b6612006-04-25 15:42:00 +0200600 ret = mapping->a_ops->prepare_write(file, page, offset, offset+this_len);
Jens Axboebfc4ee32006-05-03 10:35:10 +0200601 if (unlikely(ret)) {
602 loff_t isize = i_size_read(mapping->host);
603
604 if (ret != AOP_TRUNCATED_PAGE)
605 unlock_page(page);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200606 page_cache_release(page);
Jens Axboebfc4ee32006-05-03 10:35:10 +0200607 if (ret == AOP_TRUNCATED_PAGE)
608 goto find_page;
609
610 /*
611 * prepare_write() may have instantiated a few blocks
612 * outside i_size. Trim these off again.
613 */
614 if (sd->pos + this_len > isize)
615 vmtruncate(mapping->host, isize);
616
Jens Axboee6e80f22006-10-11 10:03:09 +0200617 goto out_ret;
Jens Axboebfc4ee32006-05-03 10:35:10 +0200618 }
Jens Axboe5274f052006-03-30 15:15:30 +0200619
Jens Axboe0568b402006-05-01 19:50:48 +0200620 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200621 /*
622 * Careful, ->map() uses KM_USER0!
623 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200624 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200625 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200626
Jens Axboe016b6612006-04-25 15:42:00 +0200627 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200628 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200629 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200630 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200631 }
Jens Axboe5274f052006-03-30 15:15:30 +0200632
Jens Axboe016b6612006-04-25 15:42:00 +0200633 ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len);
Dmitriy Monakhovd9993c32007-03-29 14:24:09 +0200634 if (ret) {
635 if (ret == AOP_TRUNCATED_PAGE) {
636 page_cache_release(page);
637 goto find_page;
638 }
639 if (ret < 0)
640 goto out;
Jens Axboe0568b402006-05-01 19:50:48 +0200641 /*
Dmitriy Monakhovd9993c32007-03-29 14:24:09 +0200642 * Partial write has happened, so 'ret' already initialized by
643 * number of bytes written, Where is nothing we have to do here.
Jens Axboe0568b402006-05-01 19:50:48 +0200644 */
Dmitriy Monakhovd9993c32007-03-29 14:24:09 +0200645 } else
Jens Axboe0568b402006-05-01 19:50:48 +0200646 ret = this_len;
Dmitriy Monakhovd9993c32007-03-29 14:24:09 +0200647 /*
648 * Return the number of bytes written and mark page as
649 * accessed, we are now done!
650 */
651 mark_page_accessed(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200652out:
Jens Axboe9e0267c2006-04-19 15:57:31 +0200653 unlock_page(page);
Jens Axboe6a860c92007-07-20 15:18:12 +0200654out_release:
655 page_cache_release(page);
Jens Axboee6e80f22006-10-11 10:03:09 +0200656out_ret:
Jens Axboe5274f052006-03-30 15:15:30 +0200657 return ret;
658}
659
Jens Axboe932cc6d2007-06-21 13:10:21 +0200660/**
661 * __splice_from_pipe - splice data from a pipe to given actor
662 * @pipe: pipe to splice from
663 * @sd: information to @actor
664 * @actor: handler that splices the data
665 *
666 * Description:
667 * This function does little more than loop over the pipe and call
668 * @actor to do the actual moving of a single struct pipe_buffer to
669 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
670 * pipe_to_user.
671 *
Jens Axboe83f91352006-04-02 23:05:09 +0200672 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200673ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
674 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200675{
Jens Axboe5274f052006-03-30 15:15:30 +0200676 int ret, do_wakeup, err;
Jens Axboe5274f052006-03-30 15:15:30 +0200677
678 ret = 0;
679 do_wakeup = 0;
680
Jens Axboe5274f052006-03-30 15:15:30 +0200681 for (;;) {
Jens Axboe6f767b02006-04-11 13:53:56 +0200682 if (pipe->nrbufs) {
683 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800684 const struct pipe_buf_operations *ops = buf->ops;
Jens Axboe5274f052006-03-30 15:15:30 +0200685
Jens Axboec66ab6f2007-06-12 21:17:17 +0200686 sd->len = buf->len;
687 if (sd->len > sd->total_len)
688 sd->len = sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200689
Jens Axboec66ab6f2007-06-12 21:17:17 +0200690 err = actor(pipe, buf, sd);
Jens Axboe016b6612006-04-25 15:42:00 +0200691 if (err <= 0) {
Jens Axboe5274f052006-03-30 15:15:30 +0200692 if (!ret && err != -ENODATA)
693 ret = err;
694
695 break;
696 }
697
Jens Axboe016b6612006-04-25 15:42:00 +0200698 ret += err;
699 buf->offset += err;
700 buf->len -= err;
701
Jens Axboec66ab6f2007-06-12 21:17:17 +0200702 sd->len -= err;
703 sd->pos += err;
704 sd->total_len -= err;
705 if (sd->len)
Jens Axboe016b6612006-04-25 15:42:00 +0200706 continue;
Ingo Molnar73d62d82006-04-11 13:57:21 +0200707
Jens Axboe5274f052006-03-30 15:15:30 +0200708 if (!buf->len) {
709 buf->ops = NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200710 ops->release(pipe, buf);
Jens Axboe6f767b02006-04-11 13:53:56 +0200711 pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1);
712 pipe->nrbufs--;
713 if (pipe->inode)
714 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200715 }
716
Jens Axboec66ab6f2007-06-12 21:17:17 +0200717 if (!sd->total_len)
Jens Axboe5274f052006-03-30 15:15:30 +0200718 break;
719 }
720
Jens Axboe6f767b02006-04-11 13:53:56 +0200721 if (pipe->nrbufs)
Jens Axboe5274f052006-03-30 15:15:30 +0200722 continue;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200723 if (!pipe->writers)
Jens Axboe5274f052006-03-30 15:15:30 +0200724 break;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200725 if (!pipe->waiting_writers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200726 if (ret)
727 break;
728 }
729
Jens Axboec66ab6f2007-06-12 21:17:17 +0200730 if (sd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700731 if (!ret)
732 ret = -EAGAIN;
733 break;
734 }
735
Jens Axboe5274f052006-03-30 15:15:30 +0200736 if (signal_pending(current)) {
737 if (!ret)
738 ret = -ERESTARTSYS;
739 break;
740 }
741
742 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200743 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200744 if (waitqueue_active(&pipe->wait))
745 wake_up_interruptible_sync(&pipe->wait);
746 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200747 do_wakeup = 0;
748 }
749
Ingo Molnar3a326a22006-04-10 15:18:35 +0200750 pipe_wait(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200751 }
752
Jens Axboe5274f052006-03-30 15:15:30 +0200753 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200754 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200755 if (waitqueue_active(&pipe->wait))
756 wake_up_interruptible(&pipe->wait);
757 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200758 }
759
Jens Axboe5274f052006-03-30 15:15:30 +0200760 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200761}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100762EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200763
Jens Axboe932cc6d2007-06-21 13:10:21 +0200764/**
765 * splice_from_pipe - splice data from a pipe to a file
766 * @pipe: pipe to splice from
767 * @out: file to splice to
768 * @ppos: position in @out
769 * @len: how many bytes to splice
770 * @flags: splice modifier flags
771 * @actor: handler that splices the data
772 *
773 * Description:
774 * See __splice_from_pipe. This function locks the input and output inodes,
775 * otherwise it's identical to __splice_from_pipe().
776 *
777 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200778ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
779 loff_t *ppos, size_t len, unsigned int flags,
780 splice_actor *actor)
781{
782 ssize_t ret;
783 struct inode *inode = out->f_mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200784 struct splice_desc sd = {
785 .total_len = len,
786 .flags = flags,
787 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200788 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200789 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200790
791 /*
792 * The actor worker might be calling ->prepare_write and
793 * ->commit_write. Most of the time, these expect i_mutex to
794 * be held. Since this may result in an ABBA deadlock with
795 * pipe->inode, we have to order lock acquiry here.
796 */
797 inode_double_lock(inode, pipe->inode);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200798 ret = __splice_from_pipe(pipe, &sd, actor);
Mark Fasheh6da61802006-10-17 18:43:07 +0200799 inode_double_unlock(inode, pipe->inode);
800
801 return ret;
802}
803
804/**
805 * generic_file_splice_write_nolock - generic_file_splice_write without mutexes
806 * @pipe: pipe info
807 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200808 * @ppos: position in @out
Mark Fasheh6da61802006-10-17 18:43:07 +0200809 * @len: number of bytes to splice
810 * @flags: splice modifier flags
811 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200812 * Description:
813 * Will either move or copy pages (determined by @flags options) from
814 * the given pipe inode to the given file. The caller is responsible
815 * for acquiring i_mutex on both inodes.
Mark Fasheh6da61802006-10-17 18:43:07 +0200816 *
817 */
818ssize_t
819generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out,
820 loff_t *ppos, size_t len, unsigned int flags)
821{
822 struct address_space *mapping = out->f_mapping;
823 struct inode *inode = mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200824 struct splice_desc sd = {
825 .total_len = len,
826 .flags = flags,
827 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200828 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200829 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200830 ssize_t ret;
831 int err;
832
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800833 err = remove_suid(out->f_path.dentry);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200834 if (unlikely(err))
835 return err;
836
Jens Axboec66ab6f2007-06-12 21:17:17 +0200837 ret = __splice_from_pipe(pipe, &sd, pipe_to_file);
Mark Fasheh6da61802006-10-17 18:43:07 +0200838 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200839 unsigned long nr_pages;
840
Mark Fasheh6da61802006-10-17 18:43:07 +0200841 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200842 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Mark Fasheh6da61802006-10-17 18:43:07 +0200843
844 /*
845 * If file or inode is SYNC and we actually wrote some data,
846 * sync it.
847 */
848 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
849 err = generic_osync_inode(inode, mapping,
850 OSYNC_METADATA|OSYNC_DATA);
851
852 if (err)
853 ret = err;
854 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200855 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Mark Fasheh6da61802006-10-17 18:43:07 +0200856 }
857
858 return ret;
859}
860
861EXPORT_SYMBOL(generic_file_splice_write_nolock);
862
Jens Axboe83f91352006-04-02 23:05:09 +0200863/**
864 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200865 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200866 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200867 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200868 * @len: number of bytes to splice
869 * @flags: splice modifier flags
870 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200871 * Description:
872 * Will either move or copy pages (determined by @flags options) from
873 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200874 *
875 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200876ssize_t
877generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200878 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200879{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200880 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200881 struct inode *inode = mapping->host;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200882 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200883 int err;
884
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800885 err = should_remove_suid(out->f_path.dentry);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200886 if (unlikely(err)) {
887 mutex_lock(&inode->i_mutex);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800888 err = __remove_suid(out->f_path.dentry, err);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200889 mutex_unlock(&inode->i_mutex);
890 if (err)
891 return err;
892 }
Ingo Molnar3a326a22006-04-10 15:18:35 +0200893
Jens Axboe00522fb2006-04-26 14:39:29 +0200894 ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file);
Jens Axboea4514eb2006-04-19 15:57:05 +0200895 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200896 unsigned long nr_pages;
897
Jens Axboea4514eb2006-04-19 15:57:05 +0200898 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200899 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200900
Jens Axboea4514eb2006-04-19 15:57:05 +0200901 /*
902 * If file or inode is SYNC and we actually wrote some data,
903 * sync it.
904 */
905 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
Jens Axboea4514eb2006-04-19 15:57:05 +0200906 mutex_lock(&inode->i_mutex);
907 err = generic_osync_inode(inode, mapping,
908 OSYNC_METADATA|OSYNC_DATA);
909 mutex_unlock(&inode->i_mutex);
910
911 if (err)
912 ret = err;
913 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200914 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200915 }
916
917 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200918}
919
Jens Axboe059a8f32006-04-02 23:06:05 +0200920EXPORT_SYMBOL(generic_file_splice_write);
921
Jens Axboe83f91352006-04-02 23:05:09 +0200922/**
923 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +0200924 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +0200925 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200926 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200927 * @len: number of bytes to splice
928 * @flags: splice modifier flags
929 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200930 * Description:
931 * Will send @len bytes from the pipe to a network socket. No data copying
932 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +0200933 *
934 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200935ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200936 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200937{
Jens Axboe00522fb2006-04-26 14:39:29 +0200938 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +0200939}
940
Jens Axboe059a8f32006-04-02 23:06:05 +0200941EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -0500942
Jens Axboe83f91352006-04-02 23:05:09 +0200943/*
944 * Attempt to initiate a splice from pipe to file.
945 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200946static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200947 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200948{
Jens Axboe5274f052006-03-30 15:15:30 +0200949 int ret;
950
Jens Axboe49570e92006-04-11 13:56:09 +0200951 if (unlikely(!out->f_op || !out->f_op->splice_write))
Jens Axboe5274f052006-03-30 15:15:30 +0200952 return -EINVAL;
953
Jens Axboe49570e92006-04-11 13:56:09 +0200954 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +0200955 return -EBADF;
956
Jens Axboecbb7e572006-04-11 14:57:50 +0200957 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200958 if (unlikely(ret < 0))
959 return ret;
960
James Morris29ce2052007-07-13 11:44:32 +0200961 ret = security_file_permission(out, MAY_WRITE);
962 if (unlikely(ret < 0))
963 return ret;
964
Jens Axboecbb7e572006-04-11 14:57:50 +0200965 return out->f_op->splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200966}
967
Jens Axboe83f91352006-04-02 23:05:09 +0200968/*
969 * Attempt to initiate a splice from a file to a pipe.
970 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200971static long do_splice_to(struct file *in, loff_t *ppos,
972 struct pipe_inode_info *pipe, size_t len,
973 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200974{
Jens Axboe5274f052006-03-30 15:15:30 +0200975 int ret;
976
Jens Axboe49570e92006-04-11 13:56:09 +0200977 if (unlikely(!in->f_op || !in->f_op->splice_read))
Jens Axboe5274f052006-03-30 15:15:30 +0200978 return -EINVAL;
979
Jens Axboe49570e92006-04-11 13:56:09 +0200980 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +0200981 return -EBADF;
982
Jens Axboecbb7e572006-04-11 14:57:50 +0200983 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200984 if (unlikely(ret < 0))
985 return ret;
986
James Morris29ce2052007-07-13 11:44:32 +0200987 ret = security_file_permission(in, MAY_READ);
988 if (unlikely(ret < 0))
989 return ret;
990
Jens Axboecbb7e572006-04-11 14:57:50 +0200991 return in->f_op->splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200992}
993
Jens Axboe932cc6d2007-06-21 13:10:21 +0200994/**
995 * splice_direct_to_actor - splices data directly between two non-pipes
996 * @in: file to splice from
997 * @sd: actor information on where to splice to
998 * @actor: handles the data splicing
999 *
1000 * Description:
1001 * This is a special case helper to splice directly between two
1002 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001003 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001004 * that process.
1005 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001006 */
1007ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1008 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001009{
1010 struct pipe_inode_info *pipe;
1011 long ret, bytes;
1012 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001013 size_t len;
1014 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001015
1016 /*
1017 * We require the input being a regular file, as we don't want to
1018 * randomly drop data for eg socket -> socket splicing. Use the
1019 * piped splicing for that!
1020 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001021 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001022 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1023 return -EINVAL;
1024
1025 /*
1026 * neither in nor out is a pipe, setup an internal pipe attached to
1027 * 'out' and transfer the wanted data from 'in' to 'out' through that
1028 */
1029 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001030 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001031 pipe = alloc_pipe_info(NULL);
1032 if (!pipe)
1033 return -ENOMEM;
1034
1035 /*
1036 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001037 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001038 * PIPE_READERS appropriately.
1039 */
1040 pipe->readers = 1;
1041
1042 current->splice_pipe = pipe;
1043 }
1044
1045 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001046 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001047 */
1048 ret = 0;
1049 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001050 len = sd->total_len;
1051 flags = sd->flags;
1052
1053 /*
1054 * Don't block on output, we have to drain the direct pipe.
1055 */
1056 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001057
1058 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001059 size_t read_len;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001060 loff_t pos = sd->pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001061
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001062 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001063 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001064 goto out_release;
1065
1066 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001067 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001068
1069 /*
1070 * NOTE: nonblocking mode only applies to the input. We
1071 * must not do the output in nonblocking mode as then we
1072 * could get stuck data in the internal pipe:
1073 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001074 ret = actor(pipe, sd);
Jens Axboe51a92c02007-07-13 14:11:43 +02001075 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001076 goto out_release;
1077
1078 bytes += ret;
1079 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001080 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001081
Jens Axboe51a92c02007-07-13 14:11:43 +02001082 if (ret < read_len)
1083 goto out_release;
Jens Axboeb92ce552006-04-11 13:52:07 +02001084 }
1085
1086 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboeb92ce552006-04-11 13:52:07 +02001087 return bytes;
1088
1089out_release:
1090 /*
1091 * If we did an incomplete transfer we must release
1092 * the pipe buffers in question:
1093 */
1094 for (i = 0; i < PIPE_BUFFERS; i++) {
1095 struct pipe_buffer *buf = pipe->bufs + i;
1096
1097 if (buf->ops) {
1098 buf->ops->release(pipe, buf);
1099 buf->ops = NULL;
1100 }
1101 }
1102 pipe->nrbufs = pipe->curbuf = 0;
1103
1104 /*
1105 * If we transferred some data, return the number of bytes:
1106 */
1107 if (bytes > 0)
1108 return bytes;
1109
1110 return ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001111
1112}
1113EXPORT_SYMBOL(splice_direct_to_actor);
1114
1115static int direct_splice_actor(struct pipe_inode_info *pipe,
1116 struct splice_desc *sd)
1117{
Jens Axboe6a14b902007-06-14 13:08:55 +02001118 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001119
1120 return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags);
1121}
1122
Jens Axboe932cc6d2007-06-21 13:10:21 +02001123/**
1124 * do_splice_direct - splices data directly between two files
1125 * @in: file to splice from
1126 * @ppos: input file offset
1127 * @out: file to splice to
1128 * @len: number of bytes to splice
1129 * @flags: splice modifier flags
1130 *
1131 * Description:
1132 * For use by do_sendfile(). splice can easily emulate sendfile, but
1133 * doing it in the application would incur an extra system call
1134 * (splice in + splice out, as compared to just sendfile()). So this helper
1135 * can splice directly through a process-private pipe.
1136 *
1137 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001138long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1139 size_t len, unsigned int flags)
1140{
1141 struct splice_desc sd = {
1142 .len = len,
1143 .total_len = len,
1144 .flags = flags,
1145 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001146 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001147 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001148 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001149
1150 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001151 if (ret > 0)
1152 *ppos += ret;
1153
Jens Axboec66ab6f2007-06-12 21:17:17 +02001154 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001155}
1156
Jens Axboe83f91352006-04-02 23:05:09 +02001157/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001158 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1159 * location, so checking ->i_pipe is not enough to verify that this is a
1160 * pipe.
1161 */
1162static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1163{
1164 if (S_ISFIFO(inode->i_mode))
1165 return inode->i_pipe;
1166
1167 return NULL;
1168}
1169
1170/*
Jens Axboe83f91352006-04-02 23:05:09 +02001171 * Determine where to splice to/from.
1172 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001173static long do_splice(struct file *in, loff_t __user *off_in,
1174 struct file *out, loff_t __user *off_out,
1175 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001176{
Ingo Molnar3a326a22006-04-10 15:18:35 +02001177 struct pipe_inode_info *pipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001178 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001179 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001180
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001181 pipe = pipe_info(in->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001182 if (pipe) {
1183 if (off_in)
1184 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001185 if (off_out) {
1186 if (out->f_op->llseek == no_llseek)
1187 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001188 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001189 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001190 off = &offset;
1191 } else
1192 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001193
Jens Axboea4514eb2006-04-19 15:57:05 +02001194 ret = do_splice_from(pipe, out, off, len, flags);
1195
1196 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1197 ret = -EFAULT;
1198
1199 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001200 }
Jens Axboe5274f052006-03-30 15:15:30 +02001201
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001202 pipe = pipe_info(out->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001203 if (pipe) {
1204 if (off_out)
1205 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001206 if (off_in) {
1207 if (in->f_op->llseek == no_llseek)
1208 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001209 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001210 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001211 off = &offset;
1212 } else
1213 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001214
Jens Axboea4514eb2006-04-19 15:57:05 +02001215 ret = do_splice_to(in, off, pipe, len, flags);
1216
1217 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1218 ret = -EFAULT;
1219
1220 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001221 }
Jens Axboe5274f052006-03-30 15:15:30 +02001222
1223 return -EINVAL;
1224}
1225
Jens Axboe912d35f2006-04-26 10:59:21 +02001226/*
Linus Torvalds75723952007-10-01 13:17:28 -07001227 * Do a copy-from-user while holding the mmap_semaphore for reading, in a
1228 * manner safe from deadlocking with simultaneous mmap() (grabbing mmap_sem
1229 * for writing) and page faulting on the user memory pointed to by src.
1230 * This assumes that we will very rarely hit the partial != 0 path, or this
1231 * will not be a win.
1232 */
1233static int copy_from_user_mmap_sem(void *dst, const void __user *src, size_t n)
1234{
1235 int partial;
1236
1237 pagefault_disable();
1238 partial = __copy_from_user_inatomic(dst, src, n);
1239 pagefault_enable();
1240
1241 /*
1242 * Didn't copy everything, drop the mmap_sem and do a faulting copy
1243 */
1244 if (unlikely(partial)) {
1245 up_read(&current->mm->mmap_sem);
1246 partial = copy_from_user(dst, src, n);
1247 down_read(&current->mm->mmap_sem);
1248 }
1249
1250 return partial;
1251}
1252
1253/*
Jens Axboe912d35f2006-04-26 10:59:21 +02001254 * Map an iov into an array of pages and offset/length tupples. With the
1255 * partial_page structure, we can map several non-contiguous ranges into
1256 * our ones pages[] map instead of splitting that operation into pieces.
1257 * Could easily be exported as a generic helper for other users, in which
1258 * case one would probably want to add a 'max_nr_pages' parameter as well.
1259 */
1260static int get_iovec_page_array(const struct iovec __user *iov,
1261 unsigned int nr_vecs, struct page **pages,
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001262 struct partial_page *partial, int aligned)
Jens Axboe912d35f2006-04-26 10:59:21 +02001263{
1264 int buffers = 0, error = 0;
1265
Jens Axboe912d35f2006-04-26 10:59:21 +02001266 down_read(&current->mm->mmap_sem);
1267
1268 while (nr_vecs) {
1269 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001270 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001271 void __user *base;
1272 size_t len;
1273 int i;
1274
Linus Torvalds75723952007-10-01 13:17:28 -07001275 error = -EFAULT;
1276 if (copy_from_user_mmap_sem(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001277 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001278
1279 base = entry.iov_base;
1280 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001281
1282 /*
1283 * Sanity check this iovec. 0 read succeeds.
1284 */
Linus Torvalds75723952007-10-01 13:17:28 -07001285 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001286 if (unlikely(!len))
1287 break;
1288 error = -EFAULT;
1289 if (unlikely(!base))
1290 break;
1291
1292 /*
1293 * Get this base offset and number of pages, then map
1294 * in the user pages.
1295 */
1296 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001297
1298 /*
1299 * If asked for alignment, the offset must be zero and the
1300 * length a multiple of the PAGE_SIZE.
1301 */
1302 error = -EINVAL;
1303 if (aligned && (off || len & ~PAGE_MASK))
1304 break;
1305
Jens Axboe912d35f2006-04-26 10:59:21 +02001306 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1307 if (npages > PIPE_BUFFERS - buffers)
1308 npages = PIPE_BUFFERS - buffers;
1309
1310 error = get_user_pages(current, current->mm,
1311 (unsigned long) base, npages, 0, 0,
1312 &pages[buffers], NULL);
1313
1314 if (unlikely(error <= 0))
1315 break;
1316
1317 /*
1318 * Fill this contiguous range into the partial page map.
1319 */
1320 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001321 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001322
1323 partial[buffers].offset = off;
1324 partial[buffers].len = plen;
1325
1326 off = 0;
1327 len -= plen;
1328 buffers++;
1329 }
1330
1331 /*
1332 * We didn't complete this iov, stop here since it probably
1333 * means we have to move some of this into a pipe to
1334 * be able to continue.
1335 */
1336 if (len)
1337 break;
1338
1339 /*
1340 * Don't continue if we mapped fewer pages than we asked for,
1341 * or if we mapped the max number of pages that we have
1342 * room for.
1343 */
1344 if (error < npages || buffers == PIPE_BUFFERS)
1345 break;
1346
1347 nr_vecs--;
1348 iov++;
1349 }
1350
1351 up_read(&current->mm->mmap_sem);
1352
1353 if (buffers)
1354 return buffers;
1355
1356 return error;
1357}
1358
Jens Axboe6a14b902007-06-14 13:08:55 +02001359static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1360 struct splice_desc *sd)
1361{
1362 char *src;
1363 int ret;
1364
Jens Axboecac36bb02007-06-14 13:10:48 +02001365 ret = buf->ops->confirm(pipe, buf);
Jens Axboe6a14b902007-06-14 13:08:55 +02001366 if (unlikely(ret))
1367 return ret;
1368
1369 /*
1370 * See if we can use the atomic maps, by prefaulting in the
1371 * pages and doing an atomic copy
1372 */
1373 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1374 src = buf->ops->map(pipe, buf, 1);
1375 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1376 sd->len);
1377 buf->ops->unmap(pipe, buf, src);
1378 if (!ret) {
1379 ret = sd->len;
1380 goto out;
1381 }
1382 }
1383
1384 /*
1385 * No dice, use slow non-atomic map and copy
1386 */
1387 src = buf->ops->map(pipe, buf, 0);
1388
1389 ret = sd->len;
1390 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1391 ret = -EFAULT;
1392
1393out:
1394 if (ret > 0)
1395 sd->u.userptr += ret;
1396 buf->ops->unmap(pipe, buf, src);
1397 return ret;
1398}
1399
1400/*
1401 * For lack of a better implementation, implement vmsplice() to userspace
1402 * as a simple copy of the pipes pages to the user iov.
1403 */
1404static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1405 unsigned long nr_segs, unsigned int flags)
1406{
1407 struct pipe_inode_info *pipe;
1408 struct splice_desc sd;
1409 ssize_t size;
1410 int error;
1411 long ret;
1412
1413 pipe = pipe_info(file->f_path.dentry->d_inode);
1414 if (!pipe)
1415 return -EBADF;
1416
1417 if (pipe->inode)
1418 mutex_lock(&pipe->inode->i_mutex);
1419
1420 error = ret = 0;
1421 while (nr_segs) {
1422 void __user *base;
1423 size_t len;
1424
1425 /*
1426 * Get user address base and length for this iovec.
1427 */
1428 error = get_user(base, &iov->iov_base);
1429 if (unlikely(error))
1430 break;
1431 error = get_user(len, &iov->iov_len);
1432 if (unlikely(error))
1433 break;
1434
1435 /*
1436 * Sanity check this iovec. 0 read succeeds.
1437 */
1438 if (unlikely(!len))
1439 break;
1440 if (unlikely(!base)) {
1441 error = -EFAULT;
1442 break;
1443 }
1444
1445 sd.len = 0;
1446 sd.total_len = len;
1447 sd.flags = flags;
1448 sd.u.userptr = base;
1449 sd.pos = 0;
1450
1451 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1452 if (size < 0) {
1453 if (!ret)
1454 ret = size;
1455
1456 break;
1457 }
1458
1459 ret += size;
1460
1461 if (size < len)
1462 break;
1463
1464 nr_segs--;
1465 iov++;
1466 }
1467
1468 if (pipe->inode)
1469 mutex_unlock(&pipe->inode->i_mutex);
1470
1471 if (!ret)
1472 ret = error;
1473
1474 return ret;
1475}
1476
Jens Axboe912d35f2006-04-26 10:59:21 +02001477/*
1478 * vmsplice splices a user address range into a pipe. It can be thought of
1479 * as splice-from-memory, where the regular splice is splice-from-file (or
1480 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001481 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001482static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1483 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001484{
Jens Axboeddac0d32006-11-04 12:49:32 +01001485 struct pipe_inode_info *pipe;
Jens Axboe912d35f2006-04-26 10:59:21 +02001486 struct page *pages[PIPE_BUFFERS];
1487 struct partial_page partial[PIPE_BUFFERS];
1488 struct splice_pipe_desc spd = {
1489 .pages = pages,
1490 .partial = partial,
1491 .flags = flags,
1492 .ops = &user_page_pipe_buf_ops,
1493 };
1494
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001495 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001496 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001497 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001498
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001499 spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial,
1500 flags & SPLICE_F_GIFT);
Jens Axboe912d35f2006-04-26 10:59:21 +02001501 if (spd.nr_pages <= 0)
1502 return spd.nr_pages;
1503
Jens Axboe00522fb2006-04-26 14:39:29 +02001504 return splice_to_pipe(pipe, &spd);
Jens Axboe912d35f2006-04-26 10:59:21 +02001505}
1506
Jens Axboe6a14b902007-06-14 13:08:55 +02001507/*
1508 * Note that vmsplice only really supports true splicing _from_ user memory
1509 * to a pipe, not the other way around. Splicing from user memory is a simple
1510 * operation that can be supported without any funky alignment restrictions
1511 * or nasty vm tricks. We simply map in the user memory and fill them into
1512 * a pipe. The reverse isn't quite as easy, though. There are two possible
1513 * solutions for that:
1514 *
1515 * - memcpy() the data internally, at which point we might as well just
1516 * do a regular read() on the buffer anyway.
1517 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1518 * has restriction limitations on both ends of the pipe).
1519 *
1520 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1521 *
1522 */
Jens Axboe912d35f2006-04-26 10:59:21 +02001523asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov,
1524 unsigned long nr_segs, unsigned int flags)
1525{
1526 struct file *file;
1527 long error;
1528 int fput;
1529
Jens Axboe6a14b902007-06-14 13:08:55 +02001530 if (unlikely(nr_segs > UIO_MAXIOV))
1531 return -EINVAL;
1532 else if (unlikely(!nr_segs))
1533 return 0;
1534
Jens Axboe912d35f2006-04-26 10:59:21 +02001535 error = -EBADF;
1536 file = fget_light(fd, &fput);
1537 if (file) {
1538 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001539 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1540 else if (file->f_mode & FMODE_READ)
1541 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001542
1543 fput_light(file, fput);
1544 }
1545
1546 return error;
1547}
1548
Ingo Molnar529565d2006-04-10 15:18:58 +02001549asmlinkage long sys_splice(int fd_in, loff_t __user *off_in,
1550 int fd_out, loff_t __user *off_out,
1551 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001552{
1553 long error;
1554 struct file *in, *out;
1555 int fput_in, fput_out;
1556
1557 if (unlikely(!len))
1558 return 0;
1559
1560 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001561 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001562 if (in) {
1563 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001564 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001565 if (out) {
1566 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001567 error = do_splice(in, off_in,
1568 out, off_out,
1569 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001570 fput_light(out, fput_out);
1571 }
1572 }
1573
1574 fput_light(in, fput_in);
1575 }
1576
1577 return error;
1578}
Jens Axboe70524492006-04-11 15:51:17 +02001579
1580/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001581 * Make sure there's data to read. Wait for input if we can, otherwise
1582 * return an appropriate error.
1583 */
1584static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1585{
1586 int ret;
1587
1588 /*
1589 * Check ->nrbufs without the inode lock first. This function
1590 * is speculative anyways, so missing one is ok.
1591 */
1592 if (pipe->nrbufs)
1593 return 0;
1594
1595 ret = 0;
1596 mutex_lock(&pipe->inode->i_mutex);
1597
1598 while (!pipe->nrbufs) {
1599 if (signal_pending(current)) {
1600 ret = -ERESTARTSYS;
1601 break;
1602 }
1603 if (!pipe->writers)
1604 break;
1605 if (!pipe->waiting_writers) {
1606 if (flags & SPLICE_F_NONBLOCK) {
1607 ret = -EAGAIN;
1608 break;
1609 }
1610 }
1611 pipe_wait(pipe);
1612 }
1613
1614 mutex_unlock(&pipe->inode->i_mutex);
1615 return ret;
1616}
1617
1618/*
1619 * Make sure there's writeable room. Wait for room if we can, otherwise
1620 * return an appropriate error.
1621 */
1622static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1623{
1624 int ret;
1625
1626 /*
1627 * Check ->nrbufs without the inode lock first. This function
1628 * is speculative anyways, so missing one is ok.
1629 */
1630 if (pipe->nrbufs < PIPE_BUFFERS)
1631 return 0;
1632
1633 ret = 0;
1634 mutex_lock(&pipe->inode->i_mutex);
1635
1636 while (pipe->nrbufs >= PIPE_BUFFERS) {
1637 if (!pipe->readers) {
1638 send_sig(SIGPIPE, current, 0);
1639 ret = -EPIPE;
1640 break;
1641 }
1642 if (flags & SPLICE_F_NONBLOCK) {
1643 ret = -EAGAIN;
1644 break;
1645 }
1646 if (signal_pending(current)) {
1647 ret = -ERESTARTSYS;
1648 break;
1649 }
1650 pipe->waiting_writers++;
1651 pipe_wait(pipe);
1652 pipe->waiting_writers--;
1653 }
1654
1655 mutex_unlock(&pipe->inode->i_mutex);
1656 return ret;
1657}
1658
1659/*
Jens Axboe70524492006-04-11 15:51:17 +02001660 * Link contents of ipipe to opipe.
1661 */
1662static int link_pipe(struct pipe_inode_info *ipipe,
1663 struct pipe_inode_info *opipe,
1664 size_t len, unsigned int flags)
1665{
1666 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001667 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001668
1669 /*
1670 * Potential ABBA deadlock, work around it by ordering lock
1671 * grabbing by inode address. Otherwise two different processes
1672 * could deadlock (one doing tee from A -> B, the other from B -> A).
1673 */
Mark Fasheh62752ee2006-10-17 10:31:38 +02001674 inode_double_lock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001675
Jens Axboeaadd06e2006-07-10 11:00:01 +02001676 do {
Jens Axboe70524492006-04-11 15:51:17 +02001677 if (!opipe->readers) {
1678 send_sig(SIGPIPE, current, 0);
1679 if (!ret)
1680 ret = -EPIPE;
1681 break;
1682 }
Jens Axboe70524492006-04-11 15:51:17 +02001683
1684 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001685 * If we have iterated all input buffers or ran out of
1686 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001687 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001688 if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS)
Jens Axboe70524492006-04-11 15:51:17 +02001689 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001690
1691 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
1692 nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
1693
Jens Axboe2a27250e2006-04-19 15:56:40 +02001694 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001695 * Get a reference to this pipe buffer,
1696 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001697 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001698 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001699
Jens Axboeaadd06e2006-07-10 11:00:01 +02001700 obuf = opipe->bufs + nbuf;
1701 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001702
Jens Axboeaadd06e2006-07-10 11:00:01 +02001703 /*
1704 * Don't inherit the gift flag, we need to
1705 * prevent multiple steals of this page.
1706 */
1707 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1708
1709 if (obuf->len > len)
1710 obuf->len = len;
1711
1712 opipe->nrbufs++;
1713 ret += obuf->len;
1714 len -= obuf->len;
1715 i++;
1716 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001717
Mark Fasheh62752ee2006-10-17 10:31:38 +02001718 inode_double_unlock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001719
Jens Axboeaadd06e2006-07-10 11:00:01 +02001720 /*
1721 * If we put data in the output pipe, wakeup any potential readers.
1722 */
1723 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02001724 smp_mb();
1725 if (waitqueue_active(&opipe->wait))
1726 wake_up_interruptible(&opipe->wait);
1727 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1728 }
1729
1730 return ret;
1731}
1732
1733/*
1734 * This is a tee(1) implementation that works on pipes. It doesn't copy
1735 * any data, it simply references the 'in' pages on the 'out' pipe.
1736 * The 'flags' used are the SPLICE_F_* variants, currently the only
1737 * applicable one is SPLICE_F_NONBLOCK.
1738 */
1739static long do_tee(struct file *in, struct file *out, size_t len,
1740 unsigned int flags)
1741{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001742 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
1743 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001744 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001745
1746 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001747 * Duplicate the contents of ipipe to opipe without actually
1748 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001749 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001750 if (ipipe && opipe && ipipe != opipe) {
1751 /*
1752 * Keep going, unless we encounter an error. The ipipe/opipe
1753 * ordering doesn't really matter.
1754 */
1755 ret = link_ipipe_prep(ipipe, flags);
1756 if (!ret) {
1757 ret = link_opipe_prep(opipe, flags);
1758 if (!ret) {
1759 ret = link_pipe(ipipe, opipe, len, flags);
1760 if (!ret && (flags & SPLICE_F_NONBLOCK))
1761 ret = -EAGAIN;
1762 }
1763 }
1764 }
Jens Axboe70524492006-04-11 15:51:17 +02001765
Jens Axboeaadd06e2006-07-10 11:00:01 +02001766 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02001767}
1768
1769asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags)
1770{
1771 struct file *in;
1772 int error, fput_in;
1773
1774 if (unlikely(!len))
1775 return 0;
1776
1777 error = -EBADF;
1778 in = fget_light(fdin, &fput_in);
1779 if (in) {
1780 if (in->f_mode & FMODE_READ) {
1781 int fput_out;
1782 struct file *out = fget_light(fdout, &fput_out);
1783
1784 if (out) {
1785 if (out->f_mode & FMODE_WRITE)
1786 error = do_tee(in, out, len, flags);
1787 fput_light(out, fput_out);
1788 }
1789 }
1790 fput_light(in, fput_in);
1791 }
1792
1793 return error;
1794}