15274f052SJens Axboe /* 25274f052SJens Axboe * "splice": joining two ropes together by interweaving their strands. 35274f052SJens Axboe * 45274f052SJens Axboe * This is the "extended pipe" functionality, where a pipe is used as 55274f052SJens Axboe * an arbitrary in-memory buffer. Think of a pipe as a small kernel 65274f052SJens Axboe * buffer that you can use to transfer data from one end to the other. 75274f052SJens Axboe * 85274f052SJens Axboe * The traditional unix read/write is extended with a "splice()" operation 95274f052SJens Axboe * that transfers data buffers to or from a pipe buffer. 105274f052SJens Axboe * 115274f052SJens Axboe * Named by Larry McVoy, original implementation from Linus, extended by 12c2058e06SJens Axboe * Jens to support splicing to files, network, direct splicing, etc and 13c2058e06SJens Axboe * fixing lots of bugs. 145274f052SJens Axboe * 150fe23479SJens Axboe * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk> 16c2058e06SJens Axboe * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org> 17c2058e06SJens Axboe * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu> 185274f052SJens Axboe * 195274f052SJens Axboe */ 205274f052SJens Axboe #include <linux/fs.h> 215274f052SJens Axboe #include <linux/file.h> 225274f052SJens Axboe #include <linux/pagemap.h> 23d6b29d7cSJens Axboe #include <linux/splice.h> 2408e552c6SKAMEZAWA Hiroyuki #include <linux/memcontrol.h> 255274f052SJens Axboe #include <linux/mm_inline.h> 265abc97aaSJens Axboe #include <linux/swap.h> 274f6f0bd2SJens Axboe #include <linux/writeback.h> 284f6f0bd2SJens Axboe #include <linux/buffer_head.h> 29a0f06780SJeff Garzik #include <linux/module.h> 304f6f0bd2SJens Axboe #include <linux/syscalls.h> 31912d35f8SJens Axboe #include <linux/uio.h> 3229ce2058SJames Morris #include <linux/security.h> 335274f052SJens Axboe 3483f9135bSJens Axboe /* 3583f9135bSJens Axboe * Attempt to steal a page from a pipe buffer. This should perhaps go into 3683f9135bSJens Axboe * a vm helper function, it's already simplified quite a bit by the 3783f9135bSJens Axboe * addition of remove_mapping(). If success is returned, the caller may 3883f9135bSJens Axboe * attempt to reuse this page for another destination. 3983f9135bSJens Axboe */ 4076ad4d11SJens Axboe static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe, 415abc97aaSJens Axboe struct pipe_buffer *buf) 425abc97aaSJens Axboe { 435abc97aaSJens Axboe struct page *page = buf->page; 449e94cd4fSJens Axboe struct address_space *mapping; 455abc97aaSJens Axboe 469e0267c2SJens Axboe lock_page(page); 479e0267c2SJens Axboe 489e94cd4fSJens Axboe mapping = page_mapping(page); 499e94cd4fSJens Axboe if (mapping) { 505abc97aaSJens Axboe WARN_ON(!PageUptodate(page)); 515abc97aaSJens Axboe 52ad8d6f0aSJens Axboe /* 539e94cd4fSJens Axboe * At least for ext2 with nobh option, we need to wait on 549e94cd4fSJens Axboe * writeback completing on this page, since we'll remove it 559e94cd4fSJens Axboe * from the pagecache. Otherwise truncate wont wait on the 569e94cd4fSJens Axboe * page, allowing the disk blocks to be reused by someone else 579e94cd4fSJens Axboe * before we actually wrote our data to them. fs corruption 589e94cd4fSJens Axboe * ensues. 59ad8d6f0aSJens Axboe */ 60ad8d6f0aSJens Axboe wait_on_page_writeback(page); 61ad8d6f0aSJens Axboe 62266cf658SDavid Howells if (page_has_private(page) && 63266cf658SDavid Howells !try_to_release_page(page, GFP_KERNEL)) 64ca39d651SJens Axboe goto out_unlock; 654f6f0bd2SJens Axboe 669e94cd4fSJens Axboe /* 679e94cd4fSJens Axboe * If we succeeded in removing the mapping, set LRU flag 689e94cd4fSJens Axboe * and return good. 699e94cd4fSJens Axboe */ 709e94cd4fSJens Axboe if (remove_mapping(mapping, page)) { 711432873aSJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 725abc97aaSJens Axboe return 0; 735abc97aaSJens Axboe } 749e94cd4fSJens Axboe } 759e94cd4fSJens Axboe 769e94cd4fSJens Axboe /* 779e94cd4fSJens Axboe * Raced with truncate or failed to remove page from current 789e94cd4fSJens Axboe * address space, unlock and return failure. 799e94cd4fSJens Axboe */ 80ca39d651SJens Axboe out_unlock: 819e94cd4fSJens Axboe unlock_page(page); 829e94cd4fSJens Axboe return 1; 839e94cd4fSJens Axboe } 845abc97aaSJens Axboe 8576ad4d11SJens Axboe static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, 865274f052SJens Axboe struct pipe_buffer *buf) 875274f052SJens Axboe { 885274f052SJens Axboe page_cache_release(buf->page); 891432873aSJens Axboe buf->flags &= ~PIPE_BUF_FLAG_LRU; 905274f052SJens Axboe } 915274f052SJens Axboe 920845718dSJens Axboe /* 930845718dSJens Axboe * Check whether the contents of buf is OK to access. Since the content 940845718dSJens Axboe * is a page cache page, IO may be in flight. 950845718dSJens Axboe */ 96cac36bb0SJens Axboe static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, 975274f052SJens Axboe struct pipe_buffer *buf) 985274f052SJens Axboe { 995274f052SJens Axboe struct page *page = buf->page; 10049d0b21bSJens Axboe int err; 1015274f052SJens Axboe 1025274f052SJens Axboe if (!PageUptodate(page)) { 10349d0b21bSJens Axboe lock_page(page); 1045274f052SJens Axboe 10549d0b21bSJens Axboe /* 10649d0b21bSJens Axboe * Page got truncated/unhashed. This will cause a 0-byte 10773d62d83SIngo Molnar * splice, if this is the first page. 10849d0b21bSJens Axboe */ 1095274f052SJens Axboe if (!page->mapping) { 11049d0b21bSJens Axboe err = -ENODATA; 11149d0b21bSJens Axboe goto error; 1125274f052SJens Axboe } 1135274f052SJens Axboe 11449d0b21bSJens Axboe /* 11573d62d83SIngo Molnar * Uh oh, read-error from disk. 11649d0b21bSJens Axboe */ 11749d0b21bSJens Axboe if (!PageUptodate(page)) { 11849d0b21bSJens Axboe err = -EIO; 11949d0b21bSJens Axboe goto error; 12049d0b21bSJens Axboe } 12149d0b21bSJens Axboe 12249d0b21bSJens Axboe /* 123f84d7519SJens Axboe * Page is ok afterall, we are done. 12449d0b21bSJens Axboe */ 12549d0b21bSJens Axboe unlock_page(page); 12649d0b21bSJens Axboe } 12749d0b21bSJens Axboe 128f84d7519SJens Axboe return 0; 12949d0b21bSJens Axboe error: 13049d0b21bSJens Axboe unlock_page(page); 131f84d7519SJens Axboe return err; 13270524490SJens Axboe } 13370524490SJens Axboe 134d4c3cca9SEric Dumazet static const struct pipe_buf_operations page_cache_pipe_buf_ops = { 1355274f052SJens Axboe .can_merge = 0, 136f84d7519SJens Axboe .map = generic_pipe_buf_map, 137f84d7519SJens Axboe .unmap = generic_pipe_buf_unmap, 138cac36bb0SJens Axboe .confirm = page_cache_pipe_buf_confirm, 1395274f052SJens Axboe .release = page_cache_pipe_buf_release, 1405abc97aaSJens Axboe .steal = page_cache_pipe_buf_steal, 141f84d7519SJens Axboe .get = generic_pipe_buf_get, 1425274f052SJens Axboe }; 1435274f052SJens Axboe 144912d35f8SJens Axboe static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, 145912d35f8SJens Axboe struct pipe_buffer *buf) 146912d35f8SJens Axboe { 1477afa6fd0SJens Axboe if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) 148912d35f8SJens Axboe return 1; 1497afa6fd0SJens Axboe 1501432873aSJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 151330ab716SJens Axboe return generic_pipe_buf_steal(pipe, buf); 152912d35f8SJens Axboe } 153912d35f8SJens Axboe 154d4c3cca9SEric Dumazet static const struct pipe_buf_operations user_page_pipe_buf_ops = { 155912d35f8SJens Axboe .can_merge = 0, 156f84d7519SJens Axboe .map = generic_pipe_buf_map, 157f84d7519SJens Axboe .unmap = generic_pipe_buf_unmap, 158cac36bb0SJens Axboe .confirm = generic_pipe_buf_confirm, 159912d35f8SJens Axboe .release = page_cache_pipe_buf_release, 160912d35f8SJens Axboe .steal = user_page_pipe_buf_steal, 161f84d7519SJens Axboe .get = generic_pipe_buf_get, 162912d35f8SJens Axboe }; 163912d35f8SJens Axboe 164932cc6d4SJens Axboe /** 165932cc6d4SJens Axboe * splice_to_pipe - fill passed data into a pipe 166932cc6d4SJens Axboe * @pipe: pipe to fill 167932cc6d4SJens Axboe * @spd: data to fill 168932cc6d4SJens Axboe * 169932cc6d4SJens Axboe * Description: 17079685b8dSRandy Dunlap * @spd contains a map of pages and len/offset tuples, along with 171932cc6d4SJens Axboe * the struct pipe_buf_operations associated with these pages. This 172932cc6d4SJens Axboe * function will link that data to the pipe. 173932cc6d4SJens Axboe * 17483f9135bSJens Axboe */ 175d6b29d7cSJens Axboe ssize_t splice_to_pipe(struct pipe_inode_info *pipe, 176912d35f8SJens Axboe struct splice_pipe_desc *spd) 1775274f052SJens Axboe { 17800de00bdSJens Axboe unsigned int spd_pages = spd->nr_pages; 179912d35f8SJens Axboe int ret, do_wakeup, page_nr; 1805274f052SJens Axboe 1815274f052SJens Axboe ret = 0; 1825274f052SJens Axboe do_wakeup = 0; 183912d35f8SJens Axboe page_nr = 0; 1845274f052SJens Axboe 1853a326a2cSIngo Molnar if (pipe->inode) 1863a326a2cSIngo Molnar mutex_lock(&pipe->inode->i_mutex); 1875274f052SJens Axboe 1885274f052SJens Axboe for (;;) { 1893a326a2cSIngo Molnar if (!pipe->readers) { 1905274f052SJens Axboe send_sig(SIGPIPE, current, 0); 1915274f052SJens Axboe if (!ret) 1925274f052SJens Axboe ret = -EPIPE; 1935274f052SJens Axboe break; 1945274f052SJens Axboe } 1955274f052SJens Axboe 1966f767b04SJens Axboe if (pipe->nrbufs < PIPE_BUFFERS) { 1976f767b04SJens Axboe int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1); 1983a326a2cSIngo Molnar struct pipe_buffer *buf = pipe->bufs + newbuf; 1995274f052SJens Axboe 200912d35f8SJens Axboe buf->page = spd->pages[page_nr]; 201912d35f8SJens Axboe buf->offset = spd->partial[page_nr].offset; 202912d35f8SJens Axboe buf->len = spd->partial[page_nr].len; 203497f9625SJens Axboe buf->private = spd->partial[page_nr].private; 204912d35f8SJens Axboe buf->ops = spd->ops; 2057afa6fd0SJens Axboe if (spd->flags & SPLICE_F_GIFT) 2067afa6fd0SJens Axboe buf->flags |= PIPE_BUF_FLAG_GIFT; 2077afa6fd0SJens Axboe 2086f767b04SJens Axboe pipe->nrbufs++; 209912d35f8SJens Axboe page_nr++; 210912d35f8SJens Axboe ret += buf->len; 211912d35f8SJens Axboe 2126f767b04SJens Axboe if (pipe->inode) 2135274f052SJens Axboe do_wakeup = 1; 2145274f052SJens Axboe 215912d35f8SJens Axboe if (!--spd->nr_pages) 2165274f052SJens Axboe break; 2176f767b04SJens Axboe if (pipe->nrbufs < PIPE_BUFFERS) 2185274f052SJens Axboe continue; 2195274f052SJens Axboe 2205274f052SJens Axboe break; 2215274f052SJens Axboe } 2225274f052SJens Axboe 223912d35f8SJens Axboe if (spd->flags & SPLICE_F_NONBLOCK) { 22429e35094SLinus Torvalds if (!ret) 22529e35094SLinus Torvalds ret = -EAGAIN; 22629e35094SLinus Torvalds break; 22729e35094SLinus Torvalds } 22829e35094SLinus Torvalds 2295274f052SJens Axboe if (signal_pending(current)) { 2305274f052SJens Axboe if (!ret) 2315274f052SJens Axboe ret = -ERESTARTSYS; 2325274f052SJens Axboe break; 2335274f052SJens Axboe } 2345274f052SJens Axboe 2355274f052SJens Axboe if (do_wakeup) { 236c0bd1f65SJens Axboe smp_mb(); 2373a326a2cSIngo Molnar if (waitqueue_active(&pipe->wait)) 2383a326a2cSIngo Molnar wake_up_interruptible_sync(&pipe->wait); 2393a326a2cSIngo Molnar kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); 2405274f052SJens Axboe do_wakeup = 0; 2415274f052SJens Axboe } 2425274f052SJens Axboe 2433a326a2cSIngo Molnar pipe->waiting_writers++; 2443a326a2cSIngo Molnar pipe_wait(pipe); 2453a326a2cSIngo Molnar pipe->waiting_writers--; 2465274f052SJens Axboe } 2475274f052SJens Axboe 24802676e5aSJens Axboe if (pipe->inode) { 2493a326a2cSIngo Molnar mutex_unlock(&pipe->inode->i_mutex); 2505274f052SJens Axboe 2515274f052SJens Axboe if (do_wakeup) { 252c0bd1f65SJens Axboe smp_mb(); 2533a326a2cSIngo Molnar if (waitqueue_active(&pipe->wait)) 2543a326a2cSIngo Molnar wake_up_interruptible(&pipe->wait); 2553a326a2cSIngo Molnar kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); 2565274f052SJens Axboe } 25702676e5aSJens Axboe } 2585274f052SJens Axboe 25900de00bdSJens Axboe while (page_nr < spd_pages) 260bbdfc2f7SJens Axboe spd->spd_release(spd, page_nr++); 2615274f052SJens Axboe 2625274f052SJens Axboe return ret; 2635274f052SJens Axboe } 2645274f052SJens Axboe 265bbdfc2f7SJens Axboe static void spd_release_page(struct splice_pipe_desc *spd, unsigned int i) 266bbdfc2f7SJens Axboe { 267bbdfc2f7SJens Axboe page_cache_release(spd->pages[i]); 268bbdfc2f7SJens Axboe } 269bbdfc2f7SJens Axboe 2703a326a2cSIngo Molnar static int 271cbb7e577SJens Axboe __generic_file_splice_read(struct file *in, loff_t *ppos, 272cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 273cbb7e577SJens Axboe unsigned int flags) 2745274f052SJens Axboe { 2755274f052SJens Axboe struct address_space *mapping = in->f_mapping; 276d8983910SFengguang Wu unsigned int loff, nr_pages, req_pages; 27716c523ddSJens Axboe struct page *pages[PIPE_BUFFERS]; 278912d35f8SJens Axboe struct partial_page partial[PIPE_BUFFERS]; 2795274f052SJens Axboe struct page *page; 28091ad66efSJens Axboe pgoff_t index, end_index; 28191ad66efSJens Axboe loff_t isize; 282eb20796bSJens Axboe int error, page_nr; 283912d35f8SJens Axboe struct splice_pipe_desc spd = { 284912d35f8SJens Axboe .pages = pages, 285912d35f8SJens Axboe .partial = partial, 286912d35f8SJens Axboe .flags = flags, 287912d35f8SJens Axboe .ops = &page_cache_pipe_buf_ops, 288bbdfc2f7SJens Axboe .spd_release = spd_release_page, 289912d35f8SJens Axboe }; 2905274f052SJens Axboe 291cbb7e577SJens Axboe index = *ppos >> PAGE_CACHE_SHIFT; 292912d35f8SJens Axboe loff = *ppos & ~PAGE_CACHE_MASK; 293d8983910SFengguang Wu req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 294d8983910SFengguang Wu nr_pages = min(req_pages, (unsigned)PIPE_BUFFERS); 2955274f052SJens Axboe 2965274f052SJens Axboe /* 297eb20796bSJens Axboe * Lookup the (hopefully) full range of pages we need. 29882aa5d61SJens Axboe */ 299eb20796bSJens Axboe spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages); 300431a4820SFengguang Wu index += spd.nr_pages; 301eb20796bSJens Axboe 3025274f052SJens Axboe /* 303eb20796bSJens Axboe * If find_get_pages_contig() returned fewer pages than we needed, 304431a4820SFengguang Wu * readahead/allocate the rest and fill in the holes. 305eb20796bSJens Axboe */ 306431a4820SFengguang Wu if (spd.nr_pages < nr_pages) 307cf914a7dSRusty Russell page_cache_sync_readahead(mapping, &in->f_ra, in, 308cf914a7dSRusty Russell index, req_pages - spd.nr_pages); 309431a4820SFengguang Wu 310932cc6d4SJens Axboe error = 0; 311eb20796bSJens Axboe while (spd.nr_pages < nr_pages) { 312eb20796bSJens Axboe /* 313eb20796bSJens Axboe * Page could be there, find_get_pages_contig() breaks on 314eb20796bSJens Axboe * the first hole. 3155274f052SJens Axboe */ 3167480a904SJens Axboe page = find_get_page(mapping, index); 3177480a904SJens Axboe if (!page) { 318e27dedd8SJens Axboe /* 319eb20796bSJens Axboe * page didn't exist, allocate one. 3207480a904SJens Axboe */ 3217480a904SJens Axboe page = page_cache_alloc_cold(mapping); 3225274f052SJens Axboe if (!page) 3235274f052SJens Axboe break; 3245274f052SJens Axboe 3257480a904SJens Axboe error = add_to_page_cache_lru(page, mapping, index, 3264cd13504SHugh Dickins mapping_gfp_mask(mapping)); 3275274f052SJens Axboe if (unlikely(error)) { 3285274f052SJens Axboe page_cache_release(page); 329a0548871SJens Axboe if (error == -EEXIST) 330a0548871SJens Axboe continue; 3315274f052SJens Axboe break; 3325274f052SJens Axboe } 333eb20796bSJens Axboe /* 334eb20796bSJens Axboe * add_to_page_cache() locks the page, unlock it 335eb20796bSJens Axboe * to avoid convoluting the logic below even more. 336eb20796bSJens Axboe */ 337eb20796bSJens Axboe unlock_page(page); 3385274f052SJens Axboe } 3397480a904SJens Axboe 340eb20796bSJens Axboe pages[spd.nr_pages++] = page; 341eb20796bSJens Axboe index++; 342eb20796bSJens Axboe } 343eb20796bSJens Axboe 344eb20796bSJens Axboe /* 345eb20796bSJens Axboe * Now loop over the map and see if we need to start IO on any 346eb20796bSJens Axboe * pages, fill in the partial map, etc. 347eb20796bSJens Axboe */ 348eb20796bSJens Axboe index = *ppos >> PAGE_CACHE_SHIFT; 349eb20796bSJens Axboe nr_pages = spd.nr_pages; 350eb20796bSJens Axboe spd.nr_pages = 0; 351eb20796bSJens Axboe for (page_nr = 0; page_nr < nr_pages; page_nr++) { 352eb20796bSJens Axboe unsigned int this_len; 353eb20796bSJens Axboe 354eb20796bSJens Axboe if (!len) 355eb20796bSJens Axboe break; 356eb20796bSJens Axboe 357eb20796bSJens Axboe /* 358eb20796bSJens Axboe * this_len is the max we'll use from this page 359eb20796bSJens Axboe */ 360eb20796bSJens Axboe this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); 361eb20796bSJens Axboe page = pages[page_nr]; 362eb20796bSJens Axboe 363a08a166fSFengguang Wu if (PageReadahead(page)) 364cf914a7dSRusty Russell page_cache_async_readahead(mapping, &in->f_ra, in, 365d8983910SFengguang Wu page, index, req_pages - page_nr); 366a08a166fSFengguang Wu 3677480a904SJens Axboe /* 3687480a904SJens Axboe * If the page isn't uptodate, we may need to start io on it 3697480a904SJens Axboe */ 3707480a904SJens Axboe if (!PageUptodate(page)) { 371c4f895cbSJens Axboe /* 372c4f895cbSJens Axboe * If in nonblock mode then dont block on waiting 373c4f895cbSJens Axboe * for an in-flight io page 374c4f895cbSJens Axboe */ 3759ae9d68cSFengguang Wu if (flags & SPLICE_F_NONBLOCK) { 376529ae9aaSNick Piggin if (!trylock_page(page)) { 3778191ecd1SJens Axboe error = -EAGAIN; 378c4f895cbSJens Axboe break; 3798191ecd1SJens Axboe } 3809ae9d68cSFengguang Wu } else 3817480a904SJens Axboe lock_page(page); 3827480a904SJens Axboe 3837480a904SJens Axboe /* 38432502b84SMiklos Szeredi * Page was truncated, or invalidated by the 38532502b84SMiklos Szeredi * filesystem. Redo the find/create, but this time the 38632502b84SMiklos Szeredi * page is kept locked, so there's no chance of another 38732502b84SMiklos Szeredi * race with truncate/invalidate. 3887480a904SJens Axboe */ 3897480a904SJens Axboe if (!page->mapping) { 3907480a904SJens Axboe unlock_page(page); 39132502b84SMiklos Szeredi page = find_or_create_page(mapping, index, 39232502b84SMiklos Szeredi mapping_gfp_mask(mapping)); 39332502b84SMiklos Szeredi 39432502b84SMiklos Szeredi if (!page) { 39532502b84SMiklos Szeredi error = -ENOMEM; 3967480a904SJens Axboe break; 3977480a904SJens Axboe } 39832502b84SMiklos Szeredi page_cache_release(pages[page_nr]); 39932502b84SMiklos Szeredi pages[page_nr] = page; 40032502b84SMiklos Szeredi } 4017480a904SJens Axboe /* 4027480a904SJens Axboe * page was already under io and is now done, great 4037480a904SJens Axboe */ 4047480a904SJens Axboe if (PageUptodate(page)) { 4057480a904SJens Axboe unlock_page(page); 4067480a904SJens Axboe goto fill_it; 4077480a904SJens Axboe } 4087480a904SJens Axboe 4097480a904SJens Axboe /* 4107480a904SJens Axboe * need to read in the page 4117480a904SJens Axboe */ 4127480a904SJens Axboe error = mapping->a_ops->readpage(in, page); 4137480a904SJens Axboe if (unlikely(error)) { 414eb20796bSJens Axboe /* 415eb20796bSJens Axboe * We really should re-lookup the page here, 416eb20796bSJens Axboe * but it complicates things a lot. Instead 417eb20796bSJens Axboe * lets just do what we already stored, and 418eb20796bSJens Axboe * we'll get it the next time we are called. 419eb20796bSJens Axboe */ 4207480a904SJens Axboe if (error == AOP_TRUNCATED_PAGE) 421eb20796bSJens Axboe error = 0; 422eb20796bSJens Axboe 4237480a904SJens Axboe break; 4247480a904SJens Axboe } 425620a324bSJens Axboe } 426620a324bSJens Axboe fill_it: 42791ad66efSJens Axboe /* 428620a324bSJens Axboe * i_size must be checked after PageUptodate. 42991ad66efSJens Axboe */ 43091ad66efSJens Axboe isize = i_size_read(mapping->host); 43191ad66efSJens Axboe end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 432eb20796bSJens Axboe if (unlikely(!isize || index > end_index)) 43391ad66efSJens Axboe break; 43491ad66efSJens Axboe 43591ad66efSJens Axboe /* 43691ad66efSJens Axboe * if this is the last page, see if we need to shrink 43791ad66efSJens Axboe * the length and stop 43891ad66efSJens Axboe */ 43991ad66efSJens Axboe if (end_index == index) { 440475ecadeSHugh Dickins unsigned int plen; 441475ecadeSHugh Dickins 442475ecadeSHugh Dickins /* 443475ecadeSHugh Dickins * max good bytes in this page 444475ecadeSHugh Dickins */ 445475ecadeSHugh Dickins plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; 446475ecadeSHugh Dickins if (plen <= loff) 44791ad66efSJens Axboe break; 448475ecadeSHugh Dickins 44991ad66efSJens Axboe /* 45091ad66efSJens Axboe * force quit after adding this page 45191ad66efSJens Axboe */ 452475ecadeSHugh Dickins this_len = min(this_len, plen - loff); 453eb20796bSJens Axboe len = this_len; 45491ad66efSJens Axboe } 455620a324bSJens Axboe 456eb20796bSJens Axboe partial[page_nr].offset = loff; 457eb20796bSJens Axboe partial[page_nr].len = this_len; 45882aa5d61SJens Axboe len -= this_len; 45991ad66efSJens Axboe loff = 0; 460eb20796bSJens Axboe spd.nr_pages++; 461eb20796bSJens Axboe index++; 4625274f052SJens Axboe } 4635274f052SJens Axboe 464eb20796bSJens Axboe /* 465475ecadeSHugh Dickins * Release any pages at the end, if we quit early. 'page_nr' is how far 466eb20796bSJens Axboe * we got, 'nr_pages' is how many pages are in the map. 467eb20796bSJens Axboe */ 468eb20796bSJens Axboe while (page_nr < nr_pages) 469eb20796bSJens Axboe page_cache_release(pages[page_nr++]); 470f4e6b498SFengguang Wu in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT; 471eb20796bSJens Axboe 472912d35f8SJens Axboe if (spd.nr_pages) 47300522fb4SJens Axboe return splice_to_pipe(pipe, &spd); 47416c523ddSJens Axboe 4757480a904SJens Axboe return error; 4765274f052SJens Axboe } 4775274f052SJens Axboe 47883f9135bSJens Axboe /** 47983f9135bSJens Axboe * generic_file_splice_read - splice data from file to a pipe 48083f9135bSJens Axboe * @in: file to splice from 481932cc6d4SJens Axboe * @ppos: position in @in 48283f9135bSJens Axboe * @pipe: pipe to splice to 48383f9135bSJens Axboe * @len: number of bytes to splice 48483f9135bSJens Axboe * @flags: splice modifier flags 48583f9135bSJens Axboe * 486932cc6d4SJens Axboe * Description: 487932cc6d4SJens Axboe * Will read pages from given file and fill them into a pipe. Can be 488932cc6d4SJens Axboe * used as long as the address_space operations for the source implements 489932cc6d4SJens Axboe * a readpage() hook. 490932cc6d4SJens Axboe * 49183f9135bSJens Axboe */ 492cbb7e577SJens Axboe ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, 493cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 494cbb7e577SJens Axboe unsigned int flags) 4955274f052SJens Axboe { 496d366d398SJens Axboe loff_t isize, left; 4978191ecd1SJens Axboe int ret; 498d366d398SJens Axboe 499d366d398SJens Axboe isize = i_size_read(in->f_mapping->host); 500d366d398SJens Axboe if (unlikely(*ppos >= isize)) 501d366d398SJens Axboe return 0; 502d366d398SJens Axboe 503d366d398SJens Axboe left = isize - *ppos; 504d366d398SJens Axboe if (unlikely(left < len)) 505d366d398SJens Axboe len = left; 5065274f052SJens Axboe 507cbb7e577SJens Axboe ret = __generic_file_splice_read(in, ppos, pipe, len, flags); 5088191ecd1SJens Axboe if (ret > 0) 509cbb7e577SJens Axboe *ppos += ret; 5105274f052SJens Axboe 5115274f052SJens Axboe return ret; 5125274f052SJens Axboe } 5135274f052SJens Axboe 514059a8f37SJens Axboe EXPORT_SYMBOL(generic_file_splice_read); 515059a8f37SJens Axboe 5165274f052SJens Axboe /* 5174f6f0bd2SJens Axboe * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' 518016b661eSJens Axboe * using sendpage(). Return the number of bytes sent. 5195274f052SJens Axboe */ 52076ad4d11SJens Axboe static int pipe_to_sendpage(struct pipe_inode_info *pipe, 5215274f052SJens Axboe struct pipe_buffer *buf, struct splice_desc *sd) 5225274f052SJens Axboe { 5236a14b90bSJens Axboe struct file *file = sd->u.file; 5245274f052SJens Axboe loff_t pos = sd->pos; 525f84d7519SJens Axboe int ret, more; 5265274f052SJens Axboe 527cac36bb0SJens Axboe ret = buf->ops->confirm(pipe, buf); 528f84d7519SJens Axboe if (!ret) { 529b2b39fa4SJens Axboe more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; 5305274f052SJens Axboe 531f84d7519SJens Axboe ret = file->f_op->sendpage(file, buf->page, buf->offset, 532f84d7519SJens Axboe sd->len, &pos, more); 533f84d7519SJens Axboe } 5345274f052SJens Axboe 535016b661eSJens Axboe return ret; 5365274f052SJens Axboe } 5375274f052SJens Axboe 5385274f052SJens Axboe /* 5395274f052SJens Axboe * This is a little more tricky than the file -> pipe splicing. There are 5405274f052SJens Axboe * basically three cases: 5415274f052SJens Axboe * 5425274f052SJens Axboe * - Destination page already exists in the address space and there 5435274f052SJens Axboe * are users of it. For that case we have no other option that 5445274f052SJens Axboe * copying the data. Tough luck. 5455274f052SJens Axboe * - Destination page already exists in the address space, but there 5465274f052SJens Axboe * are no users of it. Make sure it's uptodate, then drop it. Fall 5475274f052SJens Axboe * through to last case. 5485274f052SJens Axboe * - Destination page does not exist, we can add the pipe page to 5495274f052SJens Axboe * the page cache and avoid the copy. 5505274f052SJens Axboe * 55183f9135bSJens Axboe * If asked to move pages to the output file (SPLICE_F_MOVE is set in 55283f9135bSJens Axboe * sd->flags), we attempt to migrate pages from the pipe to the output 55383f9135bSJens Axboe * file address space page cache. This is possible if no one else has 55483f9135bSJens Axboe * the pipe page referenced outside of the pipe and page cache. If 55583f9135bSJens Axboe * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create 55683f9135bSJens Axboe * a new page in the output file page cache and fill/dirty that. 5575274f052SJens Axboe */ 55876ad4d11SJens Axboe static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 5595274f052SJens Axboe struct splice_desc *sd) 5605274f052SJens Axboe { 5616a14b90bSJens Axboe struct file *file = sd->u.file; 5625274f052SJens Axboe struct address_space *mapping = file->f_mapping; 563016b661eSJens Axboe unsigned int offset, this_len; 5645274f052SJens Axboe struct page *page; 565afddba49SNick Piggin void *fsdata; 5663e7ee3e7SJens Axboe int ret; 5675274f052SJens Axboe 5685274f052SJens Axboe /* 56949d0b21bSJens Axboe * make sure the data in this buffer is uptodate 5705274f052SJens Axboe */ 571cac36bb0SJens Axboe ret = buf->ops->confirm(pipe, buf); 572f84d7519SJens Axboe if (unlikely(ret)) 573f84d7519SJens Axboe return ret; 5745274f052SJens Axboe 5755274f052SJens Axboe offset = sd->pos & ~PAGE_CACHE_MASK; 5765274f052SJens Axboe 577016b661eSJens Axboe this_len = sd->len; 578016b661eSJens Axboe if (this_len + offset > PAGE_CACHE_SIZE) 579016b661eSJens Axboe this_len = PAGE_CACHE_SIZE - offset; 580016b661eSJens Axboe 581afddba49SNick Piggin ret = pagecache_write_begin(file, mapping, sd->pos, this_len, 582afddba49SNick Piggin AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); 5839e0267c2SJens Axboe if (unlikely(ret)) 584afddba49SNick Piggin goto out; 5855274f052SJens Axboe 5860568b409SJens Axboe if (buf->page != page) { 587f84d7519SJens Axboe /* 588f84d7519SJens Axboe * Careful, ->map() uses KM_USER0! 589f84d7519SJens Axboe */ 59076ad4d11SJens Axboe char *src = buf->ops->map(pipe, buf, 1); 591f84d7519SJens Axboe char *dst = kmap_atomic(page, KM_USER1); 5925abc97aaSJens Axboe 593016b661eSJens Axboe memcpy(dst + offset, src + buf->offset, this_len); 5945274f052SJens Axboe flush_dcache_page(page); 595f84d7519SJens Axboe kunmap_atomic(dst, KM_USER1); 59676ad4d11SJens Axboe buf->ops->unmap(pipe, buf, src); 5975abc97aaSJens Axboe } 598afddba49SNick Piggin ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len, 599afddba49SNick Piggin page, fsdata); 6000568b409SJens Axboe out: 6015274f052SJens Axboe return ret; 6025274f052SJens Axboe } 6035274f052SJens Axboe 604b3c2d2ddSMiklos Szeredi static void wakeup_pipe_writers(struct pipe_inode_info *pipe) 605b3c2d2ddSMiklos Szeredi { 606b3c2d2ddSMiklos Szeredi smp_mb(); 607b3c2d2ddSMiklos Szeredi if (waitqueue_active(&pipe->wait)) 608b3c2d2ddSMiklos Szeredi wake_up_interruptible(&pipe->wait); 609b3c2d2ddSMiklos Szeredi kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); 610b3c2d2ddSMiklos Szeredi } 611b3c2d2ddSMiklos Szeredi 612b3c2d2ddSMiklos Szeredi /** 613b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed - feed available data from a pipe to a file 614b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 615b3c2d2ddSMiklos Szeredi * @sd: information to @actor 616b3c2d2ddSMiklos Szeredi * @actor: handler that splices the data 617b3c2d2ddSMiklos Szeredi * 618b3c2d2ddSMiklos Szeredi * Description: 619b3c2d2ddSMiklos Szeredi 620b3c2d2ddSMiklos Szeredi * This function loops over the pipe and calls @actor to do the 621b3c2d2ddSMiklos Szeredi * actual moving of a single struct pipe_buffer to the desired 622b3c2d2ddSMiklos Szeredi * destination. It returns when there's no more buffers left in 623b3c2d2ddSMiklos Szeredi * the pipe or if the requested number of bytes (@sd->total_len) 624b3c2d2ddSMiklos Szeredi * have been copied. It returns a positive number (one) if the 625b3c2d2ddSMiklos Szeredi * pipe needs to be filled with more data, zero if the required 626b3c2d2ddSMiklos Szeredi * number of bytes have been copied and -errno on error. 627b3c2d2ddSMiklos Szeredi * 628b3c2d2ddSMiklos Szeredi * This, together with splice_from_pipe_{begin,end,next}, may be 629b3c2d2ddSMiklos Szeredi * used to implement the functionality of __splice_from_pipe() when 630b3c2d2ddSMiklos Szeredi * locking is required around copying the pipe buffers to the 631b3c2d2ddSMiklos Szeredi * destination. 632b3c2d2ddSMiklos Szeredi */ 633b3c2d2ddSMiklos Szeredi int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, 634b3c2d2ddSMiklos Szeredi splice_actor *actor) 635b3c2d2ddSMiklos Szeredi { 636b3c2d2ddSMiklos Szeredi int ret; 637b3c2d2ddSMiklos Szeredi 638b3c2d2ddSMiklos Szeredi while (pipe->nrbufs) { 639b3c2d2ddSMiklos Szeredi struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; 640b3c2d2ddSMiklos Szeredi const struct pipe_buf_operations *ops = buf->ops; 641b3c2d2ddSMiklos Szeredi 642b3c2d2ddSMiklos Szeredi sd->len = buf->len; 643b3c2d2ddSMiklos Szeredi if (sd->len > sd->total_len) 644b3c2d2ddSMiklos Szeredi sd->len = sd->total_len; 645b3c2d2ddSMiklos Szeredi 646b3c2d2ddSMiklos Szeredi ret = actor(pipe, buf, sd); 647b3c2d2ddSMiklos Szeredi if (ret <= 0) { 648b3c2d2ddSMiklos Szeredi if (ret == -ENODATA) 649b3c2d2ddSMiklos Szeredi ret = 0; 650b3c2d2ddSMiklos Szeredi return ret; 651b3c2d2ddSMiklos Szeredi } 652b3c2d2ddSMiklos Szeredi buf->offset += ret; 653b3c2d2ddSMiklos Szeredi buf->len -= ret; 654b3c2d2ddSMiklos Szeredi 655b3c2d2ddSMiklos Szeredi sd->num_spliced += ret; 656b3c2d2ddSMiklos Szeredi sd->len -= ret; 657b3c2d2ddSMiklos Szeredi sd->pos += ret; 658b3c2d2ddSMiklos Szeredi sd->total_len -= ret; 659b3c2d2ddSMiklos Szeredi 660b3c2d2ddSMiklos Szeredi if (!buf->len) { 661b3c2d2ddSMiklos Szeredi buf->ops = NULL; 662b3c2d2ddSMiklos Szeredi ops->release(pipe, buf); 663b3c2d2ddSMiklos Szeredi pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); 664b3c2d2ddSMiklos Szeredi pipe->nrbufs--; 665b3c2d2ddSMiklos Szeredi if (pipe->inode) 666b3c2d2ddSMiklos Szeredi sd->need_wakeup = true; 667b3c2d2ddSMiklos Szeredi } 668b3c2d2ddSMiklos Szeredi 669b3c2d2ddSMiklos Szeredi if (!sd->total_len) 670b3c2d2ddSMiklos Szeredi return 0; 671b3c2d2ddSMiklos Szeredi } 672b3c2d2ddSMiklos Szeredi 673b3c2d2ddSMiklos Szeredi return 1; 674b3c2d2ddSMiklos Szeredi } 675b3c2d2ddSMiklos Szeredi EXPORT_SYMBOL(splice_from_pipe_feed); 676b3c2d2ddSMiklos Szeredi 677b3c2d2ddSMiklos Szeredi /** 678b3c2d2ddSMiklos Szeredi * splice_from_pipe_next - wait for some data to splice from 679b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 680b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 681b3c2d2ddSMiklos Szeredi * 682b3c2d2ddSMiklos Szeredi * Description: 683b3c2d2ddSMiklos Szeredi * This function will wait for some data and return a positive 684b3c2d2ddSMiklos Szeredi * value (one) if pipe buffers are available. It will return zero 685b3c2d2ddSMiklos Szeredi * or -errno if no more data needs to be spliced. 686b3c2d2ddSMiklos Szeredi */ 687b3c2d2ddSMiklos Szeredi int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) 688b3c2d2ddSMiklos Szeredi { 689b3c2d2ddSMiklos Szeredi while (!pipe->nrbufs) { 690b3c2d2ddSMiklos Szeredi if (!pipe->writers) 691b3c2d2ddSMiklos Szeredi return 0; 692b3c2d2ddSMiklos Szeredi 693b3c2d2ddSMiklos Szeredi if (!pipe->waiting_writers && sd->num_spliced) 694b3c2d2ddSMiklos Szeredi return 0; 695b3c2d2ddSMiklos Szeredi 696b3c2d2ddSMiklos Szeredi if (sd->flags & SPLICE_F_NONBLOCK) 697b3c2d2ddSMiklos Szeredi return -EAGAIN; 698b3c2d2ddSMiklos Szeredi 699b3c2d2ddSMiklos Szeredi if (signal_pending(current)) 700b3c2d2ddSMiklos Szeredi return -ERESTARTSYS; 701b3c2d2ddSMiklos Szeredi 702b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) { 703b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 704b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 705b3c2d2ddSMiklos Szeredi } 706b3c2d2ddSMiklos Szeredi 707b3c2d2ddSMiklos Szeredi pipe_wait(pipe); 708b3c2d2ddSMiklos Szeredi } 709b3c2d2ddSMiklos Szeredi 710b3c2d2ddSMiklos Szeredi return 1; 711b3c2d2ddSMiklos Szeredi } 712b3c2d2ddSMiklos Szeredi EXPORT_SYMBOL(splice_from_pipe_next); 713b3c2d2ddSMiklos Szeredi 714b3c2d2ddSMiklos Szeredi /** 715b3c2d2ddSMiklos Szeredi * splice_from_pipe_begin - start splicing from pipe 716b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 717b3c2d2ddSMiklos Szeredi * 718b3c2d2ddSMiklos Szeredi * Description: 719b3c2d2ddSMiklos Szeredi * This function should be called before a loop containing 720b3c2d2ddSMiklos Szeredi * splice_from_pipe_next() and splice_from_pipe_feed() to 721b3c2d2ddSMiklos Szeredi * initialize the necessary fields of @sd. 722b3c2d2ddSMiklos Szeredi */ 723b3c2d2ddSMiklos Szeredi void splice_from_pipe_begin(struct splice_desc *sd) 724b3c2d2ddSMiklos Szeredi { 725b3c2d2ddSMiklos Szeredi sd->num_spliced = 0; 726b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 727b3c2d2ddSMiklos Szeredi } 728b3c2d2ddSMiklos Szeredi EXPORT_SYMBOL(splice_from_pipe_begin); 729b3c2d2ddSMiklos Szeredi 730b3c2d2ddSMiklos Szeredi /** 731b3c2d2ddSMiklos Szeredi * splice_from_pipe_end - finish splicing from pipe 732b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 733b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 734b3c2d2ddSMiklos Szeredi * 735b3c2d2ddSMiklos Szeredi * Description: 736b3c2d2ddSMiklos Szeredi * This function will wake up pipe writers if necessary. It should 737b3c2d2ddSMiklos Szeredi * be called after a loop containing splice_from_pipe_next() and 738b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed(). 739b3c2d2ddSMiklos Szeredi */ 740b3c2d2ddSMiklos Szeredi void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) 741b3c2d2ddSMiklos Szeredi { 742b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) 743b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 744b3c2d2ddSMiklos Szeredi } 745b3c2d2ddSMiklos Szeredi EXPORT_SYMBOL(splice_from_pipe_end); 746b3c2d2ddSMiklos Szeredi 747932cc6d4SJens Axboe /** 748932cc6d4SJens Axboe * __splice_from_pipe - splice data from a pipe to given actor 749932cc6d4SJens Axboe * @pipe: pipe to splice from 750932cc6d4SJens Axboe * @sd: information to @actor 751932cc6d4SJens Axboe * @actor: handler that splices the data 752932cc6d4SJens Axboe * 753932cc6d4SJens Axboe * Description: 754932cc6d4SJens Axboe * This function does little more than loop over the pipe and call 755932cc6d4SJens Axboe * @actor to do the actual moving of a single struct pipe_buffer to 756932cc6d4SJens Axboe * the desired destination. See pipe_to_file, pipe_to_sendpage, or 757932cc6d4SJens Axboe * pipe_to_user. 758932cc6d4SJens Axboe * 75983f9135bSJens Axboe */ 760c66ab6faSJens Axboe ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, 761c66ab6faSJens Axboe splice_actor *actor) 7625274f052SJens Axboe { 763b3c2d2ddSMiklos Szeredi int ret; 7645274f052SJens Axboe 765b3c2d2ddSMiklos Szeredi splice_from_pipe_begin(sd); 766b3c2d2ddSMiklos Szeredi do { 767b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_next(pipe, sd); 768b3c2d2ddSMiklos Szeredi if (ret > 0) 769b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_feed(pipe, sd, actor); 770b3c2d2ddSMiklos Szeredi } while (ret > 0); 771b3c2d2ddSMiklos Szeredi splice_from_pipe_end(pipe, sd); 7725274f052SJens Axboe 773b3c2d2ddSMiklos Szeredi return sd->num_spliced ? sd->num_spliced : ret; 7745274f052SJens Axboe } 77540bee44eSMark Fasheh EXPORT_SYMBOL(__splice_from_pipe); 7765274f052SJens Axboe 777932cc6d4SJens Axboe /** 778932cc6d4SJens Axboe * splice_from_pipe - splice data from a pipe to a file 779932cc6d4SJens Axboe * @pipe: pipe to splice from 780932cc6d4SJens Axboe * @out: file to splice to 781932cc6d4SJens Axboe * @ppos: position in @out 782932cc6d4SJens Axboe * @len: how many bytes to splice 783932cc6d4SJens Axboe * @flags: splice modifier flags 784932cc6d4SJens Axboe * @actor: handler that splices the data 785932cc6d4SJens Axboe * 786932cc6d4SJens Axboe * Description: 7872933970bSMiklos Szeredi * See __splice_from_pipe. This function locks the pipe inode, 788932cc6d4SJens Axboe * otherwise it's identical to __splice_from_pipe(). 789932cc6d4SJens Axboe * 790932cc6d4SJens Axboe */ 7916da61809SMark Fasheh ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, 7926da61809SMark Fasheh loff_t *ppos, size_t len, unsigned int flags, 7936da61809SMark Fasheh splice_actor *actor) 7946da61809SMark Fasheh { 7956da61809SMark Fasheh ssize_t ret; 796c66ab6faSJens Axboe struct splice_desc sd = { 797c66ab6faSJens Axboe .total_len = len, 798c66ab6faSJens Axboe .flags = flags, 799c66ab6faSJens Axboe .pos = *ppos, 8006a14b90bSJens Axboe .u.file = out, 801c66ab6faSJens Axboe }; 8026da61809SMark Fasheh 8037bfac9ecSMiklos Szeredi if (pipe->inode) 8042933970bSMiklos Szeredi mutex_lock(&pipe->inode->i_mutex); 805c66ab6faSJens Axboe ret = __splice_from_pipe(pipe, &sd, actor); 8067bfac9ecSMiklos Szeredi if (pipe->inode) 8077bfac9ecSMiklos Szeredi mutex_unlock(&pipe->inode->i_mutex); 8086da61809SMark Fasheh 8096da61809SMark Fasheh return ret; 8106da61809SMark Fasheh } 8116da61809SMark Fasheh 8126da61809SMark Fasheh /** 8136da61809SMark Fasheh * generic_file_splice_write_nolock - generic_file_splice_write without mutexes 8146da61809SMark Fasheh * @pipe: pipe info 8156da61809SMark Fasheh * @out: file to write to 816932cc6d4SJens Axboe * @ppos: position in @out 8176da61809SMark Fasheh * @len: number of bytes to splice 8186da61809SMark Fasheh * @flags: splice modifier flags 8196da61809SMark Fasheh * 820932cc6d4SJens Axboe * Description: 8216da61809SMark Fasheh * Will either move or copy pages (determined by @flags options) from 8226da61809SMark Fasheh * the given pipe inode to the given file. The caller is responsible 8236da61809SMark Fasheh * for acquiring i_mutex on both inodes. 8246da61809SMark Fasheh * 8256da61809SMark Fasheh */ 8266da61809SMark Fasheh ssize_t 8276da61809SMark Fasheh generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out, 8286da61809SMark Fasheh loff_t *ppos, size_t len, unsigned int flags) 8296da61809SMark Fasheh { 8306da61809SMark Fasheh struct address_space *mapping = out->f_mapping; 8316da61809SMark Fasheh struct inode *inode = mapping->host; 832c66ab6faSJens Axboe struct splice_desc sd = { 833c66ab6faSJens Axboe .total_len = len, 834c66ab6faSJens Axboe .flags = flags, 835c66ab6faSJens Axboe .pos = *ppos, 8366a14b90bSJens Axboe .u.file = out, 837c66ab6faSJens Axboe }; 8386da61809SMark Fasheh ssize_t ret; 8396da61809SMark Fasheh int err; 8406da61809SMark Fasheh 8412f1936b8SMiklos Szeredi err = file_remove_suid(out); 8428c34e2d6SJens Axboe if (unlikely(err)) 8438c34e2d6SJens Axboe return err; 8448c34e2d6SJens Axboe 845c66ab6faSJens Axboe ret = __splice_from_pipe(pipe, &sd, pipe_to_file); 8466da61809SMark Fasheh if (ret > 0) { 84717ee4f49SJens Axboe unsigned long nr_pages; 84817ee4f49SJens Axboe 8496da61809SMark Fasheh *ppos += ret; 85017ee4f49SJens Axboe nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 8516da61809SMark Fasheh 8526da61809SMark Fasheh /* 8536da61809SMark Fasheh * If file or inode is SYNC and we actually wrote some data, 8546da61809SMark Fasheh * sync it. 8556da61809SMark Fasheh */ 8566da61809SMark Fasheh if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { 8576da61809SMark Fasheh err = generic_osync_inode(inode, mapping, 8586da61809SMark Fasheh OSYNC_METADATA|OSYNC_DATA); 8596da61809SMark Fasheh 8606da61809SMark Fasheh if (err) 8616da61809SMark Fasheh ret = err; 8626da61809SMark Fasheh } 86317ee4f49SJens Axboe balance_dirty_pages_ratelimited_nr(mapping, nr_pages); 8646da61809SMark Fasheh } 8656da61809SMark Fasheh 8666da61809SMark Fasheh return ret; 8676da61809SMark Fasheh } 8686da61809SMark Fasheh 8696da61809SMark Fasheh EXPORT_SYMBOL(generic_file_splice_write_nolock); 8706da61809SMark Fasheh 87183f9135bSJens Axboe /** 87283f9135bSJens Axboe * generic_file_splice_write - splice data from a pipe to a file 8733a326a2cSIngo Molnar * @pipe: pipe info 87483f9135bSJens Axboe * @out: file to write to 875932cc6d4SJens Axboe * @ppos: position in @out 87683f9135bSJens Axboe * @len: number of bytes to splice 87783f9135bSJens Axboe * @flags: splice modifier flags 87883f9135bSJens Axboe * 879932cc6d4SJens Axboe * Description: 88083f9135bSJens Axboe * Will either move or copy pages (determined by @flags options) from 88183f9135bSJens Axboe * the given pipe inode to the given file. 88283f9135bSJens Axboe * 88383f9135bSJens Axboe */ 8843a326a2cSIngo Molnar ssize_t 8853a326a2cSIngo Molnar generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, 886cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 8875274f052SJens Axboe { 8884f6f0bd2SJens Axboe struct address_space *mapping = out->f_mapping; 8898c34e2d6SJens Axboe struct inode *inode = mapping->host; 8907f3d4ee1SMiklos Szeredi struct splice_desc sd = { 8917f3d4ee1SMiklos Szeredi .total_len = len, 8927f3d4ee1SMiklos Szeredi .flags = flags, 8937f3d4ee1SMiklos Szeredi .pos = *ppos, 8947f3d4ee1SMiklos Szeredi .u.file = out, 8957f3d4ee1SMiklos Szeredi }; 8963a326a2cSIngo Molnar ssize_t ret; 8978c34e2d6SJens Axboe 8987bfac9ecSMiklos Szeredi if (pipe->inode) 899eb443e5aSMiklos Szeredi mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT); 900eb443e5aSMiklos Szeredi 901eb443e5aSMiklos Szeredi splice_from_pipe_begin(&sd); 902eb443e5aSMiklos Szeredi do { 903eb443e5aSMiklos Szeredi ret = splice_from_pipe_next(pipe, &sd); 904eb443e5aSMiklos Szeredi if (ret <= 0) 905eb443e5aSMiklos Szeredi break; 906eb443e5aSMiklos Szeredi 907eb443e5aSMiklos Szeredi mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); 908eb443e5aSMiklos Szeredi ret = file_remove_suid(out); 909eb443e5aSMiklos Szeredi if (!ret) 910eb443e5aSMiklos Szeredi ret = splice_from_pipe_feed(pipe, &sd, pipe_to_file); 911eb443e5aSMiklos Szeredi mutex_unlock(&inode->i_mutex); 912eb443e5aSMiklos Szeredi } while (ret > 0); 913eb443e5aSMiklos Szeredi splice_from_pipe_end(pipe, &sd); 914eb443e5aSMiklos Szeredi 9157bfac9ecSMiklos Szeredi if (pipe->inode) 9167bfac9ecSMiklos Szeredi mutex_unlock(&pipe->inode->i_mutex); 917eb443e5aSMiklos Szeredi 918eb443e5aSMiklos Szeredi if (sd.num_spliced) 919eb443e5aSMiklos Szeredi ret = sd.num_spliced; 920eb443e5aSMiklos Szeredi 921a4514ebdSJens Axboe if (ret > 0) { 92217ee4f49SJens Axboe unsigned long nr_pages; 92317ee4f49SJens Axboe 924a4514ebdSJens Axboe *ppos += ret; 92517ee4f49SJens Axboe nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 9264f6f0bd2SJens Axboe 9274f6f0bd2SJens Axboe /* 928a4514ebdSJens Axboe * If file or inode is SYNC and we actually wrote some data, 929a4514ebdSJens Axboe * sync it. 9304f6f0bd2SJens Axboe */ 931a4514ebdSJens Axboe if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { 9327f3d4ee1SMiklos Szeredi int err; 9337f3d4ee1SMiklos Szeredi 9344f6f0bd2SJens Axboe mutex_lock(&inode->i_mutex); 935a4514ebdSJens Axboe err = generic_osync_inode(inode, mapping, 9364f6f0bd2SJens Axboe OSYNC_METADATA|OSYNC_DATA); 9374f6f0bd2SJens Axboe mutex_unlock(&inode->i_mutex); 9384f6f0bd2SJens Axboe 9394f6f0bd2SJens Axboe if (err) 9404f6f0bd2SJens Axboe ret = err; 9414f6f0bd2SJens Axboe } 94217ee4f49SJens Axboe balance_dirty_pages_ratelimited_nr(mapping, nr_pages); 943a4514ebdSJens Axboe } 9444f6f0bd2SJens Axboe 9454f6f0bd2SJens Axboe return ret; 9465274f052SJens Axboe } 9475274f052SJens Axboe 948059a8f37SJens Axboe EXPORT_SYMBOL(generic_file_splice_write); 949059a8f37SJens Axboe 95083f9135bSJens Axboe /** 95183f9135bSJens Axboe * generic_splice_sendpage - splice data from a pipe to a socket 952932cc6d4SJens Axboe * @pipe: pipe to splice from 95383f9135bSJens Axboe * @out: socket to write to 954932cc6d4SJens Axboe * @ppos: position in @out 95583f9135bSJens Axboe * @len: number of bytes to splice 95683f9135bSJens Axboe * @flags: splice modifier flags 95783f9135bSJens Axboe * 958932cc6d4SJens Axboe * Description: 95983f9135bSJens Axboe * Will send @len bytes from the pipe to a network socket. No data copying 96083f9135bSJens Axboe * is involved. 96183f9135bSJens Axboe * 96283f9135bSJens Axboe */ 9633a326a2cSIngo Molnar ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, 964cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 9655274f052SJens Axboe { 96600522fb4SJens Axboe return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); 9675274f052SJens Axboe } 9685274f052SJens Axboe 969059a8f37SJens Axboe EXPORT_SYMBOL(generic_splice_sendpage); 970a0f06780SJeff Garzik 97183f9135bSJens Axboe /* 97283f9135bSJens Axboe * Attempt to initiate a splice from pipe to file. 97383f9135bSJens Axboe */ 9743a326a2cSIngo Molnar static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, 975cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 9765274f052SJens Axboe { 9775274f052SJens Axboe int ret; 9785274f052SJens Axboe 97949570e9bSJens Axboe if (unlikely(!out->f_op || !out->f_op->splice_write)) 9805274f052SJens Axboe return -EINVAL; 9815274f052SJens Axboe 98249570e9bSJens Axboe if (unlikely(!(out->f_mode & FMODE_WRITE))) 9835274f052SJens Axboe return -EBADF; 9845274f052SJens Axboe 985efc968d4SLinus Torvalds if (unlikely(out->f_flags & O_APPEND)) 986efc968d4SLinus Torvalds return -EINVAL; 987efc968d4SLinus Torvalds 988cbb7e577SJens Axboe ret = rw_verify_area(WRITE, out, ppos, len); 9895274f052SJens Axboe if (unlikely(ret < 0)) 9905274f052SJens Axboe return ret; 9915274f052SJens Axboe 992cbb7e577SJens Axboe return out->f_op->splice_write(pipe, out, ppos, len, flags); 9935274f052SJens Axboe } 9945274f052SJens Axboe 99583f9135bSJens Axboe /* 99683f9135bSJens Axboe * Attempt to initiate a splice from a file to a pipe. 99783f9135bSJens Axboe */ 998cbb7e577SJens Axboe static long do_splice_to(struct file *in, loff_t *ppos, 999cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 1000cbb7e577SJens Axboe unsigned int flags) 10015274f052SJens Axboe { 10025274f052SJens Axboe int ret; 10035274f052SJens Axboe 100449570e9bSJens Axboe if (unlikely(!in->f_op || !in->f_op->splice_read)) 10055274f052SJens Axboe return -EINVAL; 10065274f052SJens Axboe 100749570e9bSJens Axboe if (unlikely(!(in->f_mode & FMODE_READ))) 10085274f052SJens Axboe return -EBADF; 10095274f052SJens Axboe 1010cbb7e577SJens Axboe ret = rw_verify_area(READ, in, ppos, len); 10115274f052SJens Axboe if (unlikely(ret < 0)) 10125274f052SJens Axboe return ret; 10135274f052SJens Axboe 1014cbb7e577SJens Axboe return in->f_op->splice_read(in, ppos, pipe, len, flags); 10155274f052SJens Axboe } 10165274f052SJens Axboe 1017932cc6d4SJens Axboe /** 1018932cc6d4SJens Axboe * splice_direct_to_actor - splices data directly between two non-pipes 1019932cc6d4SJens Axboe * @in: file to splice from 1020932cc6d4SJens Axboe * @sd: actor information on where to splice to 1021932cc6d4SJens Axboe * @actor: handles the data splicing 1022932cc6d4SJens Axboe * 1023932cc6d4SJens Axboe * Description: 1024932cc6d4SJens Axboe * This is a special case helper to splice directly between two 1025932cc6d4SJens Axboe * points, without requiring an explicit pipe. Internally an allocated 1026932cc6d4SJens Axboe * pipe is cached in the process, and reused during the lifetime of 1027932cc6d4SJens Axboe * that process. 1028932cc6d4SJens Axboe * 1029c66ab6faSJens Axboe */ 1030c66ab6faSJens Axboe ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, 1031c66ab6faSJens Axboe splice_direct_actor *actor) 1032b92ce558SJens Axboe { 1033b92ce558SJens Axboe struct pipe_inode_info *pipe; 1034b92ce558SJens Axboe long ret, bytes; 1035b92ce558SJens Axboe umode_t i_mode; 1036c66ab6faSJens Axboe size_t len; 1037c66ab6faSJens Axboe int i, flags; 1038b92ce558SJens Axboe 1039b92ce558SJens Axboe /* 1040b92ce558SJens Axboe * We require the input being a regular file, as we don't want to 1041b92ce558SJens Axboe * randomly drop data for eg socket -> socket splicing. Use the 1042b92ce558SJens Axboe * piped splicing for that! 1043b92ce558SJens Axboe */ 10440f7fc9e4SJosef "Jeff" Sipek i_mode = in->f_path.dentry->d_inode->i_mode; 1045b92ce558SJens Axboe if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) 1046b92ce558SJens Axboe return -EINVAL; 1047b92ce558SJens Axboe 1048b92ce558SJens Axboe /* 1049b92ce558SJens Axboe * neither in nor out is a pipe, setup an internal pipe attached to 1050b92ce558SJens Axboe * 'out' and transfer the wanted data from 'in' to 'out' through that 1051b92ce558SJens Axboe */ 1052b92ce558SJens Axboe pipe = current->splice_pipe; 105349570e9bSJens Axboe if (unlikely(!pipe)) { 1054b92ce558SJens Axboe pipe = alloc_pipe_info(NULL); 1055b92ce558SJens Axboe if (!pipe) 1056b92ce558SJens Axboe return -ENOMEM; 1057b92ce558SJens Axboe 1058b92ce558SJens Axboe /* 1059b92ce558SJens Axboe * We don't have an immediate reader, but we'll read the stuff 106000522fb4SJens Axboe * out of the pipe right after the splice_to_pipe(). So set 1061b92ce558SJens Axboe * PIPE_READERS appropriately. 1062b92ce558SJens Axboe */ 1063b92ce558SJens Axboe pipe->readers = 1; 1064b92ce558SJens Axboe 1065b92ce558SJens Axboe current->splice_pipe = pipe; 1066b92ce558SJens Axboe } 1067b92ce558SJens Axboe 1068b92ce558SJens Axboe /* 106973d62d83SIngo Molnar * Do the splice. 1070b92ce558SJens Axboe */ 1071b92ce558SJens Axboe ret = 0; 1072b92ce558SJens Axboe bytes = 0; 1073c66ab6faSJens Axboe len = sd->total_len; 1074c66ab6faSJens Axboe flags = sd->flags; 1075c66ab6faSJens Axboe 1076c66ab6faSJens Axboe /* 1077c66ab6faSJens Axboe * Don't block on output, we have to drain the direct pipe. 1078c66ab6faSJens Axboe */ 1079c66ab6faSJens Axboe sd->flags &= ~SPLICE_F_NONBLOCK; 1080b92ce558SJens Axboe 1081b92ce558SJens Axboe while (len) { 108251a92c0fSJens Axboe size_t read_len; 1083a82c53a0STom Zanussi loff_t pos = sd->pos, prev_pos = pos; 1084b92ce558SJens Axboe 1085bcd4f3acSJens Axboe ret = do_splice_to(in, &pos, pipe, len, flags); 108651a92c0fSJens Axboe if (unlikely(ret <= 0)) 1087b92ce558SJens Axboe goto out_release; 1088b92ce558SJens Axboe 1089b92ce558SJens Axboe read_len = ret; 1090c66ab6faSJens Axboe sd->total_len = read_len; 1091b92ce558SJens Axboe 1092b92ce558SJens Axboe /* 1093b92ce558SJens Axboe * NOTE: nonblocking mode only applies to the input. We 1094b92ce558SJens Axboe * must not do the output in nonblocking mode as then we 1095b92ce558SJens Axboe * could get stuck data in the internal pipe: 1096b92ce558SJens Axboe */ 1097c66ab6faSJens Axboe ret = actor(pipe, sd); 1098a82c53a0STom Zanussi if (unlikely(ret <= 0)) { 1099a82c53a0STom Zanussi sd->pos = prev_pos; 1100b92ce558SJens Axboe goto out_release; 1101a82c53a0STom Zanussi } 1102b92ce558SJens Axboe 1103b92ce558SJens Axboe bytes += ret; 1104b92ce558SJens Axboe len -= ret; 1105bcd4f3acSJens Axboe sd->pos = pos; 1106b92ce558SJens Axboe 1107a82c53a0STom Zanussi if (ret < read_len) { 1108a82c53a0STom Zanussi sd->pos = prev_pos + ret; 110951a92c0fSJens Axboe goto out_release; 1110b92ce558SJens Axboe } 1111a82c53a0STom Zanussi } 1112b92ce558SJens Axboe 11139e97198dSJens Axboe done: 1114b92ce558SJens Axboe pipe->nrbufs = pipe->curbuf = 0; 11159e97198dSJens Axboe file_accessed(in); 1116b92ce558SJens Axboe return bytes; 1117b92ce558SJens Axboe 1118b92ce558SJens Axboe out_release: 1119b92ce558SJens Axboe /* 1120b92ce558SJens Axboe * If we did an incomplete transfer we must release 1121b92ce558SJens Axboe * the pipe buffers in question: 1122b92ce558SJens Axboe */ 1123b92ce558SJens Axboe for (i = 0; i < PIPE_BUFFERS; i++) { 1124b92ce558SJens Axboe struct pipe_buffer *buf = pipe->bufs + i; 1125b92ce558SJens Axboe 1126b92ce558SJens Axboe if (buf->ops) { 1127b92ce558SJens Axboe buf->ops->release(pipe, buf); 1128b92ce558SJens Axboe buf->ops = NULL; 1129b92ce558SJens Axboe } 1130b92ce558SJens Axboe } 1131b92ce558SJens Axboe 11329e97198dSJens Axboe if (!bytes) 11339e97198dSJens Axboe bytes = ret; 1134b92ce558SJens Axboe 11359e97198dSJens Axboe goto done; 1136c66ab6faSJens Axboe } 1137c66ab6faSJens Axboe EXPORT_SYMBOL(splice_direct_to_actor); 1138c66ab6faSJens Axboe 1139c66ab6faSJens Axboe static int direct_splice_actor(struct pipe_inode_info *pipe, 1140c66ab6faSJens Axboe struct splice_desc *sd) 1141c66ab6faSJens Axboe { 11426a14b90bSJens Axboe struct file *file = sd->u.file; 1143c66ab6faSJens Axboe 1144c66ab6faSJens Axboe return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags); 1145c66ab6faSJens Axboe } 1146c66ab6faSJens Axboe 1147932cc6d4SJens Axboe /** 1148932cc6d4SJens Axboe * do_splice_direct - splices data directly between two files 1149932cc6d4SJens Axboe * @in: file to splice from 1150932cc6d4SJens Axboe * @ppos: input file offset 1151932cc6d4SJens Axboe * @out: file to splice to 1152932cc6d4SJens Axboe * @len: number of bytes to splice 1153932cc6d4SJens Axboe * @flags: splice modifier flags 1154932cc6d4SJens Axboe * 1155932cc6d4SJens Axboe * Description: 1156932cc6d4SJens Axboe * For use by do_sendfile(). splice can easily emulate sendfile, but 1157932cc6d4SJens Axboe * doing it in the application would incur an extra system call 1158932cc6d4SJens Axboe * (splice in + splice out, as compared to just sendfile()). So this helper 1159932cc6d4SJens Axboe * can splice directly through a process-private pipe. 1160932cc6d4SJens Axboe * 1161932cc6d4SJens Axboe */ 1162c66ab6faSJens Axboe long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, 1163c66ab6faSJens Axboe size_t len, unsigned int flags) 1164c66ab6faSJens Axboe { 1165c66ab6faSJens Axboe struct splice_desc sd = { 1166c66ab6faSJens Axboe .len = len, 1167c66ab6faSJens Axboe .total_len = len, 1168c66ab6faSJens Axboe .flags = flags, 1169c66ab6faSJens Axboe .pos = *ppos, 11706a14b90bSJens Axboe .u.file = out, 1171c66ab6faSJens Axboe }; 117251a92c0fSJens Axboe long ret; 1173c66ab6faSJens Axboe 1174c66ab6faSJens Axboe ret = splice_direct_to_actor(in, &sd, direct_splice_actor); 117551a92c0fSJens Axboe if (ret > 0) 1176a82c53a0STom Zanussi *ppos = sd.pos; 117751a92c0fSJens Axboe 1178c66ab6faSJens Axboe return ret; 1179b92ce558SJens Axboe } 1180b92ce558SJens Axboe 118183f9135bSJens Axboe /* 1182ddac0d39SJens Axboe * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same 1183ddac0d39SJens Axboe * location, so checking ->i_pipe is not enough to verify that this is a 1184ddac0d39SJens Axboe * pipe. 1185ddac0d39SJens Axboe */ 1186ddac0d39SJens Axboe static inline struct pipe_inode_info *pipe_info(struct inode *inode) 1187ddac0d39SJens Axboe { 1188ddac0d39SJens Axboe if (S_ISFIFO(inode->i_mode)) 1189ddac0d39SJens Axboe return inode->i_pipe; 1190ddac0d39SJens Axboe 1191ddac0d39SJens Axboe return NULL; 1192ddac0d39SJens Axboe } 1193ddac0d39SJens Axboe 1194ddac0d39SJens Axboe /* 119583f9135bSJens Axboe * Determine where to splice to/from. 119683f9135bSJens Axboe */ 1197529565dcSIngo Molnar static long do_splice(struct file *in, loff_t __user *off_in, 1198529565dcSIngo Molnar struct file *out, loff_t __user *off_out, 1199529565dcSIngo Molnar size_t len, unsigned int flags) 12005274f052SJens Axboe { 12013a326a2cSIngo Molnar struct pipe_inode_info *pipe; 1202cbb7e577SJens Axboe loff_t offset, *off; 1203a4514ebdSJens Axboe long ret; 12045274f052SJens Axboe 12050f7fc9e4SJosef "Jeff" Sipek pipe = pipe_info(in->f_path.dentry->d_inode); 1206529565dcSIngo Molnar if (pipe) { 1207529565dcSIngo Molnar if (off_in) 1208529565dcSIngo Molnar return -ESPIPE; 1209b92ce558SJens Axboe if (off_out) { 1210b92ce558SJens Axboe if (out->f_op->llseek == no_llseek) 1211b92ce558SJens Axboe return -EINVAL; 1212cbb7e577SJens Axboe if (copy_from_user(&offset, off_out, sizeof(loff_t))) 1213b92ce558SJens Axboe return -EFAULT; 1214cbb7e577SJens Axboe off = &offset; 1215cbb7e577SJens Axboe } else 1216cbb7e577SJens Axboe off = &out->f_pos; 1217529565dcSIngo Molnar 1218a4514ebdSJens Axboe ret = do_splice_from(pipe, out, off, len, flags); 1219a4514ebdSJens Axboe 1220a4514ebdSJens Axboe if (off_out && copy_to_user(off_out, off, sizeof(loff_t))) 1221a4514ebdSJens Axboe ret = -EFAULT; 1222a4514ebdSJens Axboe 1223a4514ebdSJens Axboe return ret; 1224529565dcSIngo Molnar } 12255274f052SJens Axboe 12260f7fc9e4SJosef "Jeff" Sipek pipe = pipe_info(out->f_path.dentry->d_inode); 1227529565dcSIngo Molnar if (pipe) { 1228529565dcSIngo Molnar if (off_out) 1229529565dcSIngo Molnar return -ESPIPE; 1230b92ce558SJens Axboe if (off_in) { 1231b92ce558SJens Axboe if (in->f_op->llseek == no_llseek) 1232b92ce558SJens Axboe return -EINVAL; 1233cbb7e577SJens Axboe if (copy_from_user(&offset, off_in, sizeof(loff_t))) 1234b92ce558SJens Axboe return -EFAULT; 1235cbb7e577SJens Axboe off = &offset; 1236cbb7e577SJens Axboe } else 1237cbb7e577SJens Axboe off = &in->f_pos; 1238529565dcSIngo Molnar 1239a4514ebdSJens Axboe ret = do_splice_to(in, off, pipe, len, flags); 1240a4514ebdSJens Axboe 1241a4514ebdSJens Axboe if (off_in && copy_to_user(off_in, off, sizeof(loff_t))) 1242a4514ebdSJens Axboe ret = -EFAULT; 1243a4514ebdSJens Axboe 1244a4514ebdSJens Axboe return ret; 1245529565dcSIngo Molnar } 12465274f052SJens Axboe 12475274f052SJens Axboe return -EINVAL; 12485274f052SJens Axboe } 12495274f052SJens Axboe 1250912d35f8SJens Axboe /* 1251912d35f8SJens Axboe * Map an iov into an array of pages and offset/length tupples. With the 1252912d35f8SJens Axboe * partial_page structure, we can map several non-contiguous ranges into 1253912d35f8SJens Axboe * our ones pages[] map instead of splitting that operation into pieces. 1254912d35f8SJens Axboe * Could easily be exported as a generic helper for other users, in which 1255912d35f8SJens Axboe * case one would probably want to add a 'max_nr_pages' parameter as well. 1256912d35f8SJens Axboe */ 1257912d35f8SJens Axboe static int get_iovec_page_array(const struct iovec __user *iov, 1258912d35f8SJens Axboe unsigned int nr_vecs, struct page **pages, 12597afa6fd0SJens Axboe struct partial_page *partial, int aligned) 1260912d35f8SJens Axboe { 1261912d35f8SJens Axboe int buffers = 0, error = 0; 1262912d35f8SJens Axboe 1263912d35f8SJens Axboe while (nr_vecs) { 1264912d35f8SJens Axboe unsigned long off, npages; 126575723957SLinus Torvalds struct iovec entry; 1266912d35f8SJens Axboe void __user *base; 1267912d35f8SJens Axboe size_t len; 1268912d35f8SJens Axboe int i; 1269912d35f8SJens Axboe 127075723957SLinus Torvalds error = -EFAULT; 1271bc40d73cSNick Piggin if (copy_from_user(&entry, iov, sizeof(entry))) 1272912d35f8SJens Axboe break; 127375723957SLinus Torvalds 127475723957SLinus Torvalds base = entry.iov_base; 127575723957SLinus Torvalds len = entry.iov_len; 1276912d35f8SJens Axboe 1277912d35f8SJens Axboe /* 1278912d35f8SJens Axboe * Sanity check this iovec. 0 read succeeds. 1279912d35f8SJens Axboe */ 128075723957SLinus Torvalds error = 0; 1281912d35f8SJens Axboe if (unlikely(!len)) 1282912d35f8SJens Axboe break; 1283912d35f8SJens Axboe error = -EFAULT; 1284712a30e6SBastian Blank if (!access_ok(VERIFY_READ, base, len)) 1285912d35f8SJens Axboe break; 1286912d35f8SJens Axboe 1287912d35f8SJens Axboe /* 1288912d35f8SJens Axboe * Get this base offset and number of pages, then map 1289912d35f8SJens Axboe * in the user pages. 1290912d35f8SJens Axboe */ 1291912d35f8SJens Axboe off = (unsigned long) base & ~PAGE_MASK; 12927afa6fd0SJens Axboe 12937afa6fd0SJens Axboe /* 12947afa6fd0SJens Axboe * If asked for alignment, the offset must be zero and the 12957afa6fd0SJens Axboe * length a multiple of the PAGE_SIZE. 12967afa6fd0SJens Axboe */ 12977afa6fd0SJens Axboe error = -EINVAL; 12987afa6fd0SJens Axboe if (aligned && (off || len & ~PAGE_MASK)) 12997afa6fd0SJens Axboe break; 13007afa6fd0SJens Axboe 1301912d35f8SJens Axboe npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1302912d35f8SJens Axboe if (npages > PIPE_BUFFERS - buffers) 1303912d35f8SJens Axboe npages = PIPE_BUFFERS - buffers; 1304912d35f8SJens Axboe 1305bc40d73cSNick Piggin error = get_user_pages_fast((unsigned long)base, npages, 1306bc40d73cSNick Piggin 0, &pages[buffers]); 1307912d35f8SJens Axboe 1308912d35f8SJens Axboe if (unlikely(error <= 0)) 1309912d35f8SJens Axboe break; 1310912d35f8SJens Axboe 1311912d35f8SJens Axboe /* 1312912d35f8SJens Axboe * Fill this contiguous range into the partial page map. 1313912d35f8SJens Axboe */ 1314912d35f8SJens Axboe for (i = 0; i < error; i++) { 13157591489aSJens Axboe const int plen = min_t(size_t, len, PAGE_SIZE - off); 1316912d35f8SJens Axboe 1317912d35f8SJens Axboe partial[buffers].offset = off; 1318912d35f8SJens Axboe partial[buffers].len = plen; 1319912d35f8SJens Axboe 1320912d35f8SJens Axboe off = 0; 1321912d35f8SJens Axboe len -= plen; 1322912d35f8SJens Axboe buffers++; 1323912d35f8SJens Axboe } 1324912d35f8SJens Axboe 1325912d35f8SJens Axboe /* 1326912d35f8SJens Axboe * We didn't complete this iov, stop here since it probably 1327912d35f8SJens Axboe * means we have to move some of this into a pipe to 1328912d35f8SJens Axboe * be able to continue. 1329912d35f8SJens Axboe */ 1330912d35f8SJens Axboe if (len) 1331912d35f8SJens Axboe break; 1332912d35f8SJens Axboe 1333912d35f8SJens Axboe /* 1334912d35f8SJens Axboe * Don't continue if we mapped fewer pages than we asked for, 1335912d35f8SJens Axboe * or if we mapped the max number of pages that we have 1336912d35f8SJens Axboe * room for. 1337912d35f8SJens Axboe */ 1338912d35f8SJens Axboe if (error < npages || buffers == PIPE_BUFFERS) 1339912d35f8SJens Axboe break; 1340912d35f8SJens Axboe 1341912d35f8SJens Axboe nr_vecs--; 1342912d35f8SJens Axboe iov++; 1343912d35f8SJens Axboe } 1344912d35f8SJens Axboe 1345912d35f8SJens Axboe if (buffers) 1346912d35f8SJens Axboe return buffers; 1347912d35f8SJens Axboe 1348912d35f8SJens Axboe return error; 1349912d35f8SJens Axboe } 1350912d35f8SJens Axboe 13516a14b90bSJens Axboe static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 13526a14b90bSJens Axboe struct splice_desc *sd) 13536a14b90bSJens Axboe { 13546a14b90bSJens Axboe char *src; 13556a14b90bSJens Axboe int ret; 13566a14b90bSJens Axboe 1357cac36bb0SJens Axboe ret = buf->ops->confirm(pipe, buf); 13586a14b90bSJens Axboe if (unlikely(ret)) 13596a14b90bSJens Axboe return ret; 13606a14b90bSJens Axboe 13616a14b90bSJens Axboe /* 13626a14b90bSJens Axboe * See if we can use the atomic maps, by prefaulting in the 13636a14b90bSJens Axboe * pages and doing an atomic copy 13646a14b90bSJens Axboe */ 13656a14b90bSJens Axboe if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) { 13666a14b90bSJens Axboe src = buf->ops->map(pipe, buf, 1); 13676a14b90bSJens Axboe ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset, 13686a14b90bSJens Axboe sd->len); 13696a14b90bSJens Axboe buf->ops->unmap(pipe, buf, src); 13706a14b90bSJens Axboe if (!ret) { 13716a14b90bSJens Axboe ret = sd->len; 13726a14b90bSJens Axboe goto out; 13736a14b90bSJens Axboe } 13746a14b90bSJens Axboe } 13756a14b90bSJens Axboe 13766a14b90bSJens Axboe /* 13776a14b90bSJens Axboe * No dice, use slow non-atomic map and copy 13786a14b90bSJens Axboe */ 13796a14b90bSJens Axboe src = buf->ops->map(pipe, buf, 0); 13806a14b90bSJens Axboe 13816a14b90bSJens Axboe ret = sd->len; 13826a14b90bSJens Axboe if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len)) 13836a14b90bSJens Axboe ret = -EFAULT; 13846a14b90bSJens Axboe 13856866bef4SJens Axboe buf->ops->unmap(pipe, buf, src); 13866a14b90bSJens Axboe out: 13876a14b90bSJens Axboe if (ret > 0) 13886a14b90bSJens Axboe sd->u.userptr += ret; 13896a14b90bSJens Axboe return ret; 13906a14b90bSJens Axboe } 13916a14b90bSJens Axboe 13926a14b90bSJens Axboe /* 13936a14b90bSJens Axboe * For lack of a better implementation, implement vmsplice() to userspace 13946a14b90bSJens Axboe * as a simple copy of the pipes pages to the user iov. 13956a14b90bSJens Axboe */ 13966a14b90bSJens Axboe static long vmsplice_to_user(struct file *file, const struct iovec __user *iov, 13976a14b90bSJens Axboe unsigned long nr_segs, unsigned int flags) 13986a14b90bSJens Axboe { 13996a14b90bSJens Axboe struct pipe_inode_info *pipe; 14006a14b90bSJens Axboe struct splice_desc sd; 14016a14b90bSJens Axboe ssize_t size; 14026a14b90bSJens Axboe int error; 14036a14b90bSJens Axboe long ret; 14046a14b90bSJens Axboe 14056a14b90bSJens Axboe pipe = pipe_info(file->f_path.dentry->d_inode); 14066a14b90bSJens Axboe if (!pipe) 14076a14b90bSJens Axboe return -EBADF; 14086a14b90bSJens Axboe 14096a14b90bSJens Axboe if (pipe->inode) 14106a14b90bSJens Axboe mutex_lock(&pipe->inode->i_mutex); 14116a14b90bSJens Axboe 14126a14b90bSJens Axboe error = ret = 0; 14136a14b90bSJens Axboe while (nr_segs) { 14146a14b90bSJens Axboe void __user *base; 14156a14b90bSJens Axboe size_t len; 14166a14b90bSJens Axboe 14176a14b90bSJens Axboe /* 14186a14b90bSJens Axboe * Get user address base and length for this iovec. 14196a14b90bSJens Axboe */ 14206a14b90bSJens Axboe error = get_user(base, &iov->iov_base); 14216a14b90bSJens Axboe if (unlikely(error)) 14226a14b90bSJens Axboe break; 14236a14b90bSJens Axboe error = get_user(len, &iov->iov_len); 14246a14b90bSJens Axboe if (unlikely(error)) 14256a14b90bSJens Axboe break; 14266a14b90bSJens Axboe 14276a14b90bSJens Axboe /* 14286a14b90bSJens Axboe * Sanity check this iovec. 0 read succeeds. 14296a14b90bSJens Axboe */ 14306a14b90bSJens Axboe if (unlikely(!len)) 14316a14b90bSJens Axboe break; 14326a14b90bSJens Axboe if (unlikely(!base)) { 14336a14b90bSJens Axboe error = -EFAULT; 14346a14b90bSJens Axboe break; 14356a14b90bSJens Axboe } 14366a14b90bSJens Axboe 14378811930dSJens Axboe if (unlikely(!access_ok(VERIFY_WRITE, base, len))) { 14388811930dSJens Axboe error = -EFAULT; 14398811930dSJens Axboe break; 14408811930dSJens Axboe } 14418811930dSJens Axboe 14426a14b90bSJens Axboe sd.len = 0; 14436a14b90bSJens Axboe sd.total_len = len; 14446a14b90bSJens Axboe sd.flags = flags; 14456a14b90bSJens Axboe sd.u.userptr = base; 14466a14b90bSJens Axboe sd.pos = 0; 14476a14b90bSJens Axboe 14486a14b90bSJens Axboe size = __splice_from_pipe(pipe, &sd, pipe_to_user); 14496a14b90bSJens Axboe if (size < 0) { 14506a14b90bSJens Axboe if (!ret) 14516a14b90bSJens Axboe ret = size; 14526a14b90bSJens Axboe 14536a14b90bSJens Axboe break; 14546a14b90bSJens Axboe } 14556a14b90bSJens Axboe 14566a14b90bSJens Axboe ret += size; 14576a14b90bSJens Axboe 14586a14b90bSJens Axboe if (size < len) 14596a14b90bSJens Axboe break; 14606a14b90bSJens Axboe 14616a14b90bSJens Axboe nr_segs--; 14626a14b90bSJens Axboe iov++; 14636a14b90bSJens Axboe } 14646a14b90bSJens Axboe 14656a14b90bSJens Axboe if (pipe->inode) 14666a14b90bSJens Axboe mutex_unlock(&pipe->inode->i_mutex); 14676a14b90bSJens Axboe 14686a14b90bSJens Axboe if (!ret) 14696a14b90bSJens Axboe ret = error; 14706a14b90bSJens Axboe 14716a14b90bSJens Axboe return ret; 14726a14b90bSJens Axboe } 14736a14b90bSJens Axboe 1474912d35f8SJens Axboe /* 1475912d35f8SJens Axboe * vmsplice splices a user address range into a pipe. It can be thought of 1476912d35f8SJens Axboe * as splice-from-memory, where the regular splice is splice-from-file (or 1477912d35f8SJens Axboe * to file). In both cases the output is a pipe, naturally. 1478912d35f8SJens Axboe */ 14796a14b90bSJens Axboe static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov, 1480912d35f8SJens Axboe unsigned long nr_segs, unsigned int flags) 1481912d35f8SJens Axboe { 1482ddac0d39SJens Axboe struct pipe_inode_info *pipe; 1483912d35f8SJens Axboe struct page *pages[PIPE_BUFFERS]; 1484912d35f8SJens Axboe struct partial_page partial[PIPE_BUFFERS]; 1485912d35f8SJens Axboe struct splice_pipe_desc spd = { 1486912d35f8SJens Axboe .pages = pages, 1487912d35f8SJens Axboe .partial = partial, 1488912d35f8SJens Axboe .flags = flags, 1489912d35f8SJens Axboe .ops = &user_page_pipe_buf_ops, 1490bbdfc2f7SJens Axboe .spd_release = spd_release_page, 1491912d35f8SJens Axboe }; 1492912d35f8SJens Axboe 14930f7fc9e4SJosef "Jeff" Sipek pipe = pipe_info(file->f_path.dentry->d_inode); 1494ddac0d39SJens Axboe if (!pipe) 1495912d35f8SJens Axboe return -EBADF; 1496912d35f8SJens Axboe 14977afa6fd0SJens Axboe spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial, 14987afa6fd0SJens Axboe flags & SPLICE_F_GIFT); 1499912d35f8SJens Axboe if (spd.nr_pages <= 0) 1500912d35f8SJens Axboe return spd.nr_pages; 1501912d35f8SJens Axboe 150200522fb4SJens Axboe return splice_to_pipe(pipe, &spd); 1503912d35f8SJens Axboe } 1504912d35f8SJens Axboe 15056a14b90bSJens Axboe /* 15066a14b90bSJens Axboe * Note that vmsplice only really supports true splicing _from_ user memory 15076a14b90bSJens Axboe * to a pipe, not the other way around. Splicing from user memory is a simple 15086a14b90bSJens Axboe * operation that can be supported without any funky alignment restrictions 15096a14b90bSJens Axboe * or nasty vm tricks. We simply map in the user memory and fill them into 15106a14b90bSJens Axboe * a pipe. The reverse isn't quite as easy, though. There are two possible 15116a14b90bSJens Axboe * solutions for that: 15126a14b90bSJens Axboe * 15136a14b90bSJens Axboe * - memcpy() the data internally, at which point we might as well just 15146a14b90bSJens Axboe * do a regular read() on the buffer anyway. 15156a14b90bSJens Axboe * - Lots of nasty vm tricks, that are neither fast nor flexible (it 15166a14b90bSJens Axboe * has restriction limitations on both ends of the pipe). 15176a14b90bSJens Axboe * 15186a14b90bSJens Axboe * Currently we punt and implement it as a normal copy, see pipe_to_user(). 15196a14b90bSJens Axboe * 15206a14b90bSJens Axboe */ 1521836f92adSHeiko Carstens SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov, 1522836f92adSHeiko Carstens unsigned long, nr_segs, unsigned int, flags) 1523912d35f8SJens Axboe { 1524912d35f8SJens Axboe struct file *file; 1525912d35f8SJens Axboe long error; 1526912d35f8SJens Axboe int fput; 1527912d35f8SJens Axboe 15286a14b90bSJens Axboe if (unlikely(nr_segs > UIO_MAXIOV)) 15296a14b90bSJens Axboe return -EINVAL; 15306a14b90bSJens Axboe else if (unlikely(!nr_segs)) 15316a14b90bSJens Axboe return 0; 15326a14b90bSJens Axboe 1533912d35f8SJens Axboe error = -EBADF; 1534912d35f8SJens Axboe file = fget_light(fd, &fput); 1535912d35f8SJens Axboe if (file) { 1536912d35f8SJens Axboe if (file->f_mode & FMODE_WRITE) 15376a14b90bSJens Axboe error = vmsplice_to_pipe(file, iov, nr_segs, flags); 15386a14b90bSJens Axboe else if (file->f_mode & FMODE_READ) 15396a14b90bSJens Axboe error = vmsplice_to_user(file, iov, nr_segs, flags); 1540912d35f8SJens Axboe 1541912d35f8SJens Axboe fput_light(file, fput); 1542912d35f8SJens Axboe } 1543912d35f8SJens Axboe 1544912d35f8SJens Axboe return error; 1545912d35f8SJens Axboe } 1546912d35f8SJens Axboe 1547836f92adSHeiko Carstens SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, 1548836f92adSHeiko Carstens int, fd_out, loff_t __user *, off_out, 1549836f92adSHeiko Carstens size_t, len, unsigned int, flags) 15505274f052SJens Axboe { 15515274f052SJens Axboe long error; 15525274f052SJens Axboe struct file *in, *out; 15535274f052SJens Axboe int fput_in, fput_out; 15545274f052SJens Axboe 15555274f052SJens Axboe if (unlikely(!len)) 15565274f052SJens Axboe return 0; 15575274f052SJens Axboe 15585274f052SJens Axboe error = -EBADF; 1559529565dcSIngo Molnar in = fget_light(fd_in, &fput_in); 15605274f052SJens Axboe if (in) { 15615274f052SJens Axboe if (in->f_mode & FMODE_READ) { 1562529565dcSIngo Molnar out = fget_light(fd_out, &fput_out); 15635274f052SJens Axboe if (out) { 15645274f052SJens Axboe if (out->f_mode & FMODE_WRITE) 1565529565dcSIngo Molnar error = do_splice(in, off_in, 1566529565dcSIngo Molnar out, off_out, 1567529565dcSIngo Molnar len, flags); 15685274f052SJens Axboe fput_light(out, fput_out); 15695274f052SJens Axboe } 15705274f052SJens Axboe } 15715274f052SJens Axboe 15725274f052SJens Axboe fput_light(in, fput_in); 15735274f052SJens Axboe } 15745274f052SJens Axboe 15755274f052SJens Axboe return error; 15765274f052SJens Axboe } 157770524490SJens Axboe 157870524490SJens Axboe /* 1579aadd06e5SJens Axboe * Make sure there's data to read. Wait for input if we can, otherwise 1580aadd06e5SJens Axboe * return an appropriate error. 1581aadd06e5SJens Axboe */ 1582aadd06e5SJens Axboe static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1583aadd06e5SJens Axboe { 1584aadd06e5SJens Axboe int ret; 1585aadd06e5SJens Axboe 1586aadd06e5SJens Axboe /* 1587aadd06e5SJens Axboe * Check ->nrbufs without the inode lock first. This function 1588aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1589aadd06e5SJens Axboe */ 1590aadd06e5SJens Axboe if (pipe->nrbufs) 1591aadd06e5SJens Axboe return 0; 1592aadd06e5SJens Axboe 1593aadd06e5SJens Axboe ret = 0; 1594aadd06e5SJens Axboe mutex_lock(&pipe->inode->i_mutex); 1595aadd06e5SJens Axboe 1596aadd06e5SJens Axboe while (!pipe->nrbufs) { 1597aadd06e5SJens Axboe if (signal_pending(current)) { 1598aadd06e5SJens Axboe ret = -ERESTARTSYS; 1599aadd06e5SJens Axboe break; 1600aadd06e5SJens Axboe } 1601aadd06e5SJens Axboe if (!pipe->writers) 1602aadd06e5SJens Axboe break; 1603aadd06e5SJens Axboe if (!pipe->waiting_writers) { 1604aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1605aadd06e5SJens Axboe ret = -EAGAIN; 1606aadd06e5SJens Axboe break; 1607aadd06e5SJens Axboe } 1608aadd06e5SJens Axboe } 1609aadd06e5SJens Axboe pipe_wait(pipe); 1610aadd06e5SJens Axboe } 1611aadd06e5SJens Axboe 1612aadd06e5SJens Axboe mutex_unlock(&pipe->inode->i_mutex); 1613aadd06e5SJens Axboe return ret; 1614aadd06e5SJens Axboe } 1615aadd06e5SJens Axboe 1616aadd06e5SJens Axboe /* 1617aadd06e5SJens Axboe * Make sure there's writeable room. Wait for room if we can, otherwise 1618aadd06e5SJens Axboe * return an appropriate error. 1619aadd06e5SJens Axboe */ 1620aadd06e5SJens Axboe static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1621aadd06e5SJens Axboe { 1622aadd06e5SJens Axboe int ret; 1623aadd06e5SJens Axboe 1624aadd06e5SJens Axboe /* 1625aadd06e5SJens Axboe * Check ->nrbufs without the inode lock first. This function 1626aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1627aadd06e5SJens Axboe */ 1628aadd06e5SJens Axboe if (pipe->nrbufs < PIPE_BUFFERS) 1629aadd06e5SJens Axboe return 0; 1630aadd06e5SJens Axboe 1631aadd06e5SJens Axboe ret = 0; 1632aadd06e5SJens Axboe mutex_lock(&pipe->inode->i_mutex); 1633aadd06e5SJens Axboe 1634aadd06e5SJens Axboe while (pipe->nrbufs >= PIPE_BUFFERS) { 1635aadd06e5SJens Axboe if (!pipe->readers) { 1636aadd06e5SJens Axboe send_sig(SIGPIPE, current, 0); 1637aadd06e5SJens Axboe ret = -EPIPE; 1638aadd06e5SJens Axboe break; 1639aadd06e5SJens Axboe } 1640aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1641aadd06e5SJens Axboe ret = -EAGAIN; 1642aadd06e5SJens Axboe break; 1643aadd06e5SJens Axboe } 1644aadd06e5SJens Axboe if (signal_pending(current)) { 1645aadd06e5SJens Axboe ret = -ERESTARTSYS; 1646aadd06e5SJens Axboe break; 1647aadd06e5SJens Axboe } 1648aadd06e5SJens Axboe pipe->waiting_writers++; 1649aadd06e5SJens Axboe pipe_wait(pipe); 1650aadd06e5SJens Axboe pipe->waiting_writers--; 1651aadd06e5SJens Axboe } 1652aadd06e5SJens Axboe 1653aadd06e5SJens Axboe mutex_unlock(&pipe->inode->i_mutex); 1654aadd06e5SJens Axboe return ret; 1655aadd06e5SJens Axboe } 1656aadd06e5SJens Axboe 1657aadd06e5SJens Axboe /* 165870524490SJens Axboe * Link contents of ipipe to opipe. 165970524490SJens Axboe */ 166070524490SJens Axboe static int link_pipe(struct pipe_inode_info *ipipe, 166170524490SJens Axboe struct pipe_inode_info *opipe, 166270524490SJens Axboe size_t len, unsigned int flags) 166370524490SJens Axboe { 166470524490SJens Axboe struct pipe_buffer *ibuf, *obuf; 1665aadd06e5SJens Axboe int ret = 0, i = 0, nbuf; 166670524490SJens Axboe 166770524490SJens Axboe /* 166870524490SJens Axboe * Potential ABBA deadlock, work around it by ordering lock 166970524490SJens Axboe * grabbing by inode address. Otherwise two different processes 167070524490SJens Axboe * could deadlock (one doing tee from A -> B, the other from B -> A). 167170524490SJens Axboe */ 167262752ee1SMark Fasheh inode_double_lock(ipipe->inode, opipe->inode); 167370524490SJens Axboe 1674aadd06e5SJens Axboe do { 167570524490SJens Axboe if (!opipe->readers) { 167670524490SJens Axboe send_sig(SIGPIPE, current, 0); 167770524490SJens Axboe if (!ret) 167870524490SJens Axboe ret = -EPIPE; 167970524490SJens Axboe break; 168070524490SJens Axboe } 168170524490SJens Axboe 168270524490SJens Axboe /* 1683aadd06e5SJens Axboe * If we have iterated all input buffers or ran out of 1684aadd06e5SJens Axboe * output room, break. 168570524490SJens Axboe */ 1686aadd06e5SJens Axboe if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) 1687aadd06e5SJens Axboe break; 1688aadd06e5SJens Axboe 1689aadd06e5SJens Axboe ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); 1690aadd06e5SJens Axboe nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); 169170524490SJens Axboe 169270524490SJens Axboe /* 169370524490SJens Axboe * Get a reference to this pipe buffer, 169470524490SJens Axboe * so we can copy the contents over. 169570524490SJens Axboe */ 169670524490SJens Axboe ibuf->ops->get(ipipe, ibuf); 169770524490SJens Axboe 169870524490SJens Axboe obuf = opipe->bufs + nbuf; 169970524490SJens Axboe *obuf = *ibuf; 170070524490SJens Axboe 17017afa6fd0SJens Axboe /* 17027afa6fd0SJens Axboe * Don't inherit the gift flag, we need to 17037afa6fd0SJens Axboe * prevent multiple steals of this page. 17047afa6fd0SJens Axboe */ 17057afa6fd0SJens Axboe obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 17067afa6fd0SJens Axboe 170770524490SJens Axboe if (obuf->len > len) 170870524490SJens Axboe obuf->len = len; 170970524490SJens Axboe 171070524490SJens Axboe opipe->nrbufs++; 171170524490SJens Axboe ret += obuf->len; 171270524490SJens Axboe len -= obuf->len; 1713aadd06e5SJens Axboe i++; 1714aadd06e5SJens Axboe } while (len); 171570524490SJens Axboe 171602cf01aeSJens Axboe /* 171702cf01aeSJens Axboe * return EAGAIN if we have the potential of some data in the 171802cf01aeSJens Axboe * future, otherwise just return 0 171902cf01aeSJens Axboe */ 172002cf01aeSJens Axboe if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) 172102cf01aeSJens Axboe ret = -EAGAIN; 172202cf01aeSJens Axboe 172362752ee1SMark Fasheh inode_double_unlock(ipipe->inode, opipe->inode); 172470524490SJens Axboe 1725aadd06e5SJens Axboe /* 1726aadd06e5SJens Axboe * If we put data in the output pipe, wakeup any potential readers. 1727aadd06e5SJens Axboe */ 1728aadd06e5SJens Axboe if (ret > 0) { 172970524490SJens Axboe smp_mb(); 173070524490SJens Axboe if (waitqueue_active(&opipe->wait)) 173170524490SJens Axboe wake_up_interruptible(&opipe->wait); 173270524490SJens Axboe kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); 173370524490SJens Axboe } 173470524490SJens Axboe 173570524490SJens Axboe return ret; 173670524490SJens Axboe } 173770524490SJens Axboe 173870524490SJens Axboe /* 173970524490SJens Axboe * This is a tee(1) implementation that works on pipes. It doesn't copy 174070524490SJens Axboe * any data, it simply references the 'in' pages on the 'out' pipe. 174170524490SJens Axboe * The 'flags' used are the SPLICE_F_* variants, currently the only 174270524490SJens Axboe * applicable one is SPLICE_F_NONBLOCK. 174370524490SJens Axboe */ 174470524490SJens Axboe static long do_tee(struct file *in, struct file *out, size_t len, 174570524490SJens Axboe unsigned int flags) 174670524490SJens Axboe { 17470f7fc9e4SJosef "Jeff" Sipek struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode); 17480f7fc9e4SJosef "Jeff" Sipek struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode); 1749aadd06e5SJens Axboe int ret = -EINVAL; 175070524490SJens Axboe 175170524490SJens Axboe /* 1752aadd06e5SJens Axboe * Duplicate the contents of ipipe to opipe without actually 1753aadd06e5SJens Axboe * copying the data. 175470524490SJens Axboe */ 1755aadd06e5SJens Axboe if (ipipe && opipe && ipipe != opipe) { 1756aadd06e5SJens Axboe /* 1757aadd06e5SJens Axboe * Keep going, unless we encounter an error. The ipipe/opipe 1758aadd06e5SJens Axboe * ordering doesn't really matter. 1759aadd06e5SJens Axboe */ 1760aadd06e5SJens Axboe ret = link_ipipe_prep(ipipe, flags); 1761aadd06e5SJens Axboe if (!ret) { 1762aadd06e5SJens Axboe ret = link_opipe_prep(opipe, flags); 176302cf01aeSJens Axboe if (!ret) 1764aadd06e5SJens Axboe ret = link_pipe(ipipe, opipe, len, flags); 1765aadd06e5SJens Axboe } 1766aadd06e5SJens Axboe } 176770524490SJens Axboe 1768aadd06e5SJens Axboe return ret; 176970524490SJens Axboe } 177070524490SJens Axboe 1771836f92adSHeiko Carstens SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) 177270524490SJens Axboe { 177370524490SJens Axboe struct file *in; 177470524490SJens Axboe int error, fput_in; 177570524490SJens Axboe 177670524490SJens Axboe if (unlikely(!len)) 177770524490SJens Axboe return 0; 177870524490SJens Axboe 177970524490SJens Axboe error = -EBADF; 178070524490SJens Axboe in = fget_light(fdin, &fput_in); 178170524490SJens Axboe if (in) { 178270524490SJens Axboe if (in->f_mode & FMODE_READ) { 178370524490SJens Axboe int fput_out; 178470524490SJens Axboe struct file *out = fget_light(fdout, &fput_out); 178570524490SJens Axboe 178670524490SJens Axboe if (out) { 178770524490SJens Axboe if (out->f_mode & FMODE_WRITE) 178870524490SJens Axboe error = do_tee(in, out, len, flags); 178970524490SJens Axboe fput_light(out, fput_out); 179070524490SJens Axboe } 179170524490SJens Axboe } 179270524490SJens Axboe fput_light(in, fput_in); 179370524490SJens Axboe } 179470524490SJens Axboe 179570524490SJens Axboe return error; 179670524490SJens Axboe } 1797