1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 25274f052SJens Axboe /* 35274f052SJens Axboe * "splice": joining two ropes together by interweaving their strands. 45274f052SJens Axboe * 55274f052SJens Axboe * This is the "extended pipe" functionality, where a pipe is used as 65274f052SJens Axboe * an arbitrary in-memory buffer. Think of a pipe as a small kernel 75274f052SJens Axboe * buffer that you can use to transfer data from one end to the other. 85274f052SJens Axboe * 95274f052SJens Axboe * The traditional unix read/write is extended with a "splice()" operation 105274f052SJens Axboe * that transfers data buffers to or from a pipe buffer. 115274f052SJens Axboe * 125274f052SJens Axboe * Named by Larry McVoy, original implementation from Linus, extended by 13c2058e06SJens Axboe * Jens to support splicing to files, network, direct splicing, etc and 14c2058e06SJens Axboe * fixing lots of bugs. 155274f052SJens Axboe * 160fe23479SJens Axboe * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk> 17c2058e06SJens Axboe * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org> 18c2058e06SJens Axboe * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu> 195274f052SJens Axboe * 205274f052SJens Axboe */ 21be297968SChristoph Hellwig #include <linux/bvec.h> 225274f052SJens Axboe #include <linux/fs.h> 235274f052SJens Axboe #include <linux/file.h> 245274f052SJens Axboe #include <linux/pagemap.h> 25d6b29d7cSJens Axboe #include <linux/splice.h> 2608e552c6SKAMEZAWA Hiroyuki #include <linux/memcontrol.h> 275274f052SJens Axboe #include <linux/mm_inline.h> 285abc97aaSJens Axboe #include <linux/swap.h> 294f6f0bd2SJens Axboe #include <linux/writeback.h> 30630d9c47SPaul Gortmaker #include <linux/export.h> 314f6f0bd2SJens Axboe #include <linux/syscalls.h> 32912d35f8SJens Axboe #include <linux/uio.h> 33983652c6SChung-Chiang Cheng #include <linux/fsnotify.h> 3429ce2058SJames Morris #include <linux/security.h> 355a0e3ad6STejun Heo #include <linux/gfp.h> 3635f9c09fSEric Dumazet #include <linux/socket.h> 37174cd4b1SIngo Molnar #include <linux/sched/signal.h> 38174cd4b1SIngo Molnar 3906ae43f3SAl Viro #include "internal.h" 405274f052SJens Axboe 4183f9135bSJens Axboe /* 420f99fc51SJens Axboe * Splice doesn't support FMODE_NOWAIT. Since pipes may set this flag to 430f99fc51SJens Axboe * indicate they support non-blocking reads or writes, we must clear it 440f99fc51SJens Axboe * here if set to avoid blocking other users of this pipe if splice is 450f99fc51SJens Axboe * being done on it. 460f99fc51SJens Axboe */ 470f99fc51SJens Axboe static noinline void noinline pipe_clear_nowait(struct file *file) 480f99fc51SJens Axboe { 490f99fc51SJens Axboe fmode_t fmode = READ_ONCE(file->f_mode); 500f99fc51SJens Axboe 510f99fc51SJens Axboe do { 520f99fc51SJens Axboe if (!(fmode & FMODE_NOWAIT)) 530f99fc51SJens Axboe break; 540f99fc51SJens Axboe } while (!try_cmpxchg(&file->f_mode, &fmode, fmode & ~FMODE_NOWAIT)); 550f99fc51SJens Axboe } 560f99fc51SJens Axboe 570f99fc51SJens Axboe /* 5883f9135bSJens Axboe * Attempt to steal a page from a pipe buffer. This should perhaps go into 5983f9135bSJens Axboe * a vm helper function, it's already simplified quite a bit by the 6083f9135bSJens Axboe * addition of remove_mapping(). If success is returned, the caller may 6183f9135bSJens Axboe * attempt to reuse this page for another destination. 6283f9135bSJens Axboe */ 63c928f642SChristoph Hellwig static bool page_cache_pipe_buf_try_steal(struct pipe_inode_info *pipe, 645abc97aaSJens Axboe struct pipe_buffer *buf) 655abc97aaSJens Axboe { 665100da38SMatthew Wilcox (Oracle) struct folio *folio = page_folio(buf->page); 679e94cd4fSJens Axboe struct address_space *mapping; 685abc97aaSJens Axboe 69b9ccad2eSMatthew Wilcox (Oracle) folio_lock(folio); 709e0267c2SJens Axboe 71b9ccad2eSMatthew Wilcox (Oracle) mapping = folio_mapping(folio); 729e94cd4fSJens Axboe if (mapping) { 73b9ccad2eSMatthew Wilcox (Oracle) WARN_ON(!folio_test_uptodate(folio)); 745abc97aaSJens Axboe 75ad8d6f0aSJens Axboe /* 769e94cd4fSJens Axboe * At least for ext2 with nobh option, we need to wait on 77b9ccad2eSMatthew Wilcox (Oracle) * writeback completing on this folio, since we'll remove it 789e94cd4fSJens Axboe * from the pagecache. Otherwise truncate wont wait on the 79b9ccad2eSMatthew Wilcox (Oracle) * folio, allowing the disk blocks to be reused by someone else 809e94cd4fSJens Axboe * before we actually wrote our data to them. fs corruption 819e94cd4fSJens Axboe * ensues. 82ad8d6f0aSJens Axboe */ 83b9ccad2eSMatthew Wilcox (Oracle) folio_wait_writeback(folio); 84ad8d6f0aSJens Axboe 85b9ccad2eSMatthew Wilcox (Oracle) if (folio_has_private(folio) && 86b9ccad2eSMatthew Wilcox (Oracle) !filemap_release_folio(folio, GFP_KERNEL)) 87ca39d651SJens Axboe goto out_unlock; 884f6f0bd2SJens Axboe 899e94cd4fSJens Axboe /* 909e94cd4fSJens Axboe * If we succeeded in removing the mapping, set LRU flag 919e94cd4fSJens Axboe * and return good. 929e94cd4fSJens Axboe */ 935100da38SMatthew Wilcox (Oracle) if (remove_mapping(mapping, folio)) { 941432873aSJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 95c928f642SChristoph Hellwig return true; 965abc97aaSJens Axboe } 979e94cd4fSJens Axboe } 989e94cd4fSJens Axboe 999e94cd4fSJens Axboe /* 100b9ccad2eSMatthew Wilcox (Oracle) * Raced with truncate or failed to remove folio from current 1019e94cd4fSJens Axboe * address space, unlock and return failure. 1029e94cd4fSJens Axboe */ 103ca39d651SJens Axboe out_unlock: 104b9ccad2eSMatthew Wilcox (Oracle) folio_unlock(folio); 105c928f642SChristoph Hellwig return false; 1069e94cd4fSJens Axboe } 1075abc97aaSJens Axboe 10876ad4d11SJens Axboe static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, 1095274f052SJens Axboe struct pipe_buffer *buf) 1105274f052SJens Axboe { 11109cbfeafSKirill A. Shutemov put_page(buf->page); 1121432873aSJens Axboe buf->flags &= ~PIPE_BUF_FLAG_LRU; 1135274f052SJens Axboe } 1145274f052SJens Axboe 1150845718dSJens Axboe /* 1160845718dSJens Axboe * Check whether the contents of buf is OK to access. Since the content 1170845718dSJens Axboe * is a page cache page, IO may be in flight. 1180845718dSJens Axboe */ 119cac36bb0SJens Axboe static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, 1205274f052SJens Axboe struct pipe_buffer *buf) 1215274f052SJens Axboe { 1225274f052SJens Axboe struct page *page = buf->page; 12349d0b21bSJens Axboe int err; 1245274f052SJens Axboe 1255274f052SJens Axboe if (!PageUptodate(page)) { 12649d0b21bSJens Axboe lock_page(page); 1275274f052SJens Axboe 12849d0b21bSJens Axboe /* 12949d0b21bSJens Axboe * Page got truncated/unhashed. This will cause a 0-byte 13073d62d83SIngo Molnar * splice, if this is the first page. 13149d0b21bSJens Axboe */ 1325274f052SJens Axboe if (!page->mapping) { 13349d0b21bSJens Axboe err = -ENODATA; 13449d0b21bSJens Axboe goto error; 1355274f052SJens Axboe } 1365274f052SJens Axboe 13749d0b21bSJens Axboe /* 13873d62d83SIngo Molnar * Uh oh, read-error from disk. 13949d0b21bSJens Axboe */ 14049d0b21bSJens Axboe if (!PageUptodate(page)) { 14149d0b21bSJens Axboe err = -EIO; 14249d0b21bSJens Axboe goto error; 14349d0b21bSJens Axboe } 14449d0b21bSJens Axboe 14549d0b21bSJens Axboe /* 146f84d7519SJens Axboe * Page is ok afterall, we are done. 14749d0b21bSJens Axboe */ 14849d0b21bSJens Axboe unlock_page(page); 14949d0b21bSJens Axboe } 15049d0b21bSJens Axboe 151f84d7519SJens Axboe return 0; 15249d0b21bSJens Axboe error: 15349d0b21bSJens Axboe unlock_page(page); 154f84d7519SJens Axboe return err; 15570524490SJens Axboe } 15670524490SJens Axboe 157708e3508SHugh Dickins const struct pipe_buf_operations page_cache_pipe_buf_ops = { 158cac36bb0SJens Axboe .confirm = page_cache_pipe_buf_confirm, 1595274f052SJens Axboe .release = page_cache_pipe_buf_release, 160c928f642SChristoph Hellwig .try_steal = page_cache_pipe_buf_try_steal, 161f84d7519SJens Axboe .get = generic_pipe_buf_get, 1625274f052SJens Axboe }; 1635274f052SJens Axboe 164c928f642SChristoph Hellwig static bool user_page_pipe_buf_try_steal(struct pipe_inode_info *pipe, 165912d35f8SJens Axboe struct pipe_buffer *buf) 166912d35f8SJens Axboe { 1677afa6fd0SJens Axboe if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) 168c928f642SChristoph Hellwig return false; 1697afa6fd0SJens Axboe 1701432873aSJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 171c928f642SChristoph Hellwig return generic_pipe_buf_try_steal(pipe, buf); 172912d35f8SJens Axboe } 173912d35f8SJens Axboe 174d4c3cca9SEric Dumazet static const struct pipe_buf_operations user_page_pipe_buf_ops = { 175912d35f8SJens Axboe .release = page_cache_pipe_buf_release, 176c928f642SChristoph Hellwig .try_steal = user_page_pipe_buf_try_steal, 177f84d7519SJens Axboe .get = generic_pipe_buf_get, 178912d35f8SJens Axboe }; 179912d35f8SJens Axboe 180825cdcb1SNamhyung Kim static void wakeup_pipe_readers(struct pipe_inode_info *pipe) 181825cdcb1SNamhyung Kim { 182825cdcb1SNamhyung Kim smp_mb(); 1830ddad21dSLinus Torvalds if (waitqueue_active(&pipe->rd_wait)) 1840ddad21dSLinus Torvalds wake_up_interruptible(&pipe->rd_wait); 185825cdcb1SNamhyung Kim kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); 186825cdcb1SNamhyung Kim } 187825cdcb1SNamhyung Kim 188932cc6d4SJens Axboe /** 189932cc6d4SJens Axboe * splice_to_pipe - fill passed data into a pipe 190932cc6d4SJens Axboe * @pipe: pipe to fill 191932cc6d4SJens Axboe * @spd: data to fill 192932cc6d4SJens Axboe * 193932cc6d4SJens Axboe * Description: 19479685b8dSRandy Dunlap * @spd contains a map of pages and len/offset tuples, along with 195932cc6d4SJens Axboe * the struct pipe_buf_operations associated with these pages. This 196932cc6d4SJens Axboe * function will link that data to the pipe. 197932cc6d4SJens Axboe * 19883f9135bSJens Axboe */ 199d6b29d7cSJens Axboe ssize_t splice_to_pipe(struct pipe_inode_info *pipe, 200912d35f8SJens Axboe struct splice_pipe_desc *spd) 2015274f052SJens Axboe { 20200de00bdSJens Axboe unsigned int spd_pages = spd->nr_pages; 2038cefc107SDavid Howells unsigned int tail = pipe->tail; 2048cefc107SDavid Howells unsigned int head = pipe->head; 2058cefc107SDavid Howells unsigned int mask = pipe->ring_size - 1; 2068924feffSAl Viro int ret = 0, page_nr = 0; 2075274f052SJens Axboe 208d6785d91SRabin Vincent if (!spd_pages) 209d6785d91SRabin Vincent return 0; 210d6785d91SRabin Vincent 2118924feffSAl Viro if (unlikely(!pipe->readers)) { 2125274f052SJens Axboe send_sig(SIGPIPE, current, 0); 2135274f052SJens Axboe ret = -EPIPE; 2148924feffSAl Viro goto out; 2155274f052SJens Axboe } 2165274f052SJens Axboe 2176718b6f8SDavid Howells while (!pipe_full(head, tail, pipe->max_usage)) { 2188cefc107SDavid Howells struct pipe_buffer *buf = &pipe->bufs[head & mask]; 2195274f052SJens Axboe 220912d35f8SJens Axboe buf->page = spd->pages[page_nr]; 221912d35f8SJens Axboe buf->offset = spd->partial[page_nr].offset; 222912d35f8SJens Axboe buf->len = spd->partial[page_nr].len; 223497f9625SJens Axboe buf->private = spd->partial[page_nr].private; 224912d35f8SJens Axboe buf->ops = spd->ops; 2255a81e6a1SMiklos Szeredi buf->flags = 0; 2267afa6fd0SJens Axboe 2278cefc107SDavid Howells head++; 2288cefc107SDavid Howells pipe->head = head; 229912d35f8SJens Axboe page_nr++; 230912d35f8SJens Axboe ret += buf->len; 231912d35f8SJens Axboe 232912d35f8SJens Axboe if (!--spd->nr_pages) 2335274f052SJens Axboe break; 2345274f052SJens Axboe } 2355274f052SJens Axboe 23629e35094SLinus Torvalds if (!ret) 23729e35094SLinus Torvalds ret = -EAGAIN; 23829e35094SLinus Torvalds 2398924feffSAl Viro out: 24000de00bdSJens Axboe while (page_nr < spd_pages) 241bbdfc2f7SJens Axboe spd->spd_release(spd, page_nr++); 2425274f052SJens Axboe 2435274f052SJens Axboe return ret; 2445274f052SJens Axboe } 2452b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(splice_to_pipe); 2465274f052SJens Axboe 24779fddc4eSAl Viro ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf) 24879fddc4eSAl Viro { 2498cefc107SDavid Howells unsigned int head = pipe->head; 2508cefc107SDavid Howells unsigned int tail = pipe->tail; 2518cefc107SDavid Howells unsigned int mask = pipe->ring_size - 1; 25279fddc4eSAl Viro int ret; 25379fddc4eSAl Viro 25479fddc4eSAl Viro if (unlikely(!pipe->readers)) { 25579fddc4eSAl Viro send_sig(SIGPIPE, current, 0); 25679fddc4eSAl Viro ret = -EPIPE; 2576718b6f8SDavid Howells } else if (pipe_full(head, tail, pipe->max_usage)) { 25879fddc4eSAl Viro ret = -EAGAIN; 25979fddc4eSAl Viro } else { 2608cefc107SDavid Howells pipe->bufs[head & mask] = *buf; 2618cefc107SDavid Howells pipe->head = head + 1; 26279fddc4eSAl Viro return buf->len; 26379fddc4eSAl Viro } 264a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 26579fddc4eSAl Viro return ret; 26679fddc4eSAl Viro } 26779fddc4eSAl Viro EXPORT_SYMBOL(add_to_pipe); 26879fddc4eSAl Viro 26935f3d14dSJens Axboe /* 27035f3d14dSJens Axboe * Check if we need to grow the arrays holding pages and partial page 27135f3d14dSJens Axboe * descriptions. 27235f3d14dSJens Axboe */ 273047fe360SEric Dumazet int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd) 27435f3d14dSJens Axboe { 2756718b6f8SDavid Howells unsigned int max_usage = READ_ONCE(pipe->max_usage); 276047fe360SEric Dumazet 2778cefc107SDavid Howells spd->nr_pages_max = max_usage; 2788cefc107SDavid Howells if (max_usage <= PIPE_DEF_BUFFERS) 27935f3d14dSJens Axboe return 0; 28035f3d14dSJens Axboe 2818cefc107SDavid Howells spd->pages = kmalloc_array(max_usage, sizeof(struct page *), GFP_KERNEL); 2828cefc107SDavid Howells spd->partial = kmalloc_array(max_usage, sizeof(struct partial_page), 2836da2ec56SKees Cook GFP_KERNEL); 28435f3d14dSJens Axboe 28535f3d14dSJens Axboe if (spd->pages && spd->partial) 28635f3d14dSJens Axboe return 0; 28735f3d14dSJens Axboe 28835f3d14dSJens Axboe kfree(spd->pages); 28935f3d14dSJens Axboe kfree(spd->partial); 29035f3d14dSJens Axboe return -ENOMEM; 29135f3d14dSJens Axboe } 29235f3d14dSJens Axboe 293047fe360SEric Dumazet void splice_shrink_spd(struct splice_pipe_desc *spd) 29435f3d14dSJens Axboe { 295047fe360SEric Dumazet if (spd->nr_pages_max <= PIPE_DEF_BUFFERS) 29635f3d14dSJens Axboe return; 29735f3d14dSJens Axboe 29835f3d14dSJens Axboe kfree(spd->pages); 29935f3d14dSJens Axboe kfree(spd->partial); 30035f3d14dSJens Axboe } 30135f3d14dSJens Axboe 302*9eee8bd8SDavid Howells /** 303*9eee8bd8SDavid Howells * copy_splice_read - Copy data from a file and splice the copy into a pipe 304*9eee8bd8SDavid Howells * @in: The file to read from 305*9eee8bd8SDavid Howells * @ppos: Pointer to the file position to read from 306*9eee8bd8SDavid Howells * @pipe: The pipe to splice into 307*9eee8bd8SDavid Howells * @len: The amount to splice 308*9eee8bd8SDavid Howells * @flags: The SPLICE_F_* flags 309*9eee8bd8SDavid Howells * 310*9eee8bd8SDavid Howells * This function allocates a bunch of pages sufficient to hold the requested 311*9eee8bd8SDavid Howells * amount of data (but limited by the remaining pipe capacity), passes it to 312*9eee8bd8SDavid Howells * the file's ->read_iter() to read into and then splices the used pages into 313*9eee8bd8SDavid Howells * the pipe. 314*9eee8bd8SDavid Howells * 315*9eee8bd8SDavid Howells * Return: On success, the number of bytes read will be returned and *@ppos 316*9eee8bd8SDavid Howells * will be updated if appropriate; 0 will be returned if there is no more data 317*9eee8bd8SDavid Howells * to be read; -EAGAIN will be returned if the pipe had no space, and some 318*9eee8bd8SDavid Howells * other negative error code will be returned on error. A short read may occur 319*9eee8bd8SDavid Howells * if the pipe has insufficient space, we reach the end of the data or we hit a 320*9eee8bd8SDavid Howells * hole. 32133b3b041SDavid Howells */ 32269df79a4SDavid Howells ssize_t copy_splice_read(struct file *in, loff_t *ppos, 32333b3b041SDavid Howells struct pipe_inode_info *pipe, 32433b3b041SDavid Howells size_t len, unsigned int flags) 32533b3b041SDavid Howells { 32633b3b041SDavid Howells struct iov_iter to; 32733b3b041SDavid Howells struct bio_vec *bv; 32833b3b041SDavid Howells struct kiocb kiocb; 32933b3b041SDavid Howells struct page **pages; 33033b3b041SDavid Howells ssize_t ret; 331e69f37bcSDavid Howells size_t used, npages, chunk, remain, keep = 0; 33233b3b041SDavid Howells int i; 33333b3b041SDavid Howells 33433b3b041SDavid Howells /* Work out how much data we can actually add into the pipe */ 33533b3b041SDavid Howells used = pipe_occupancy(pipe->head, pipe->tail); 33633b3b041SDavid Howells npages = max_t(ssize_t, pipe->max_usage - used, 0); 33733b3b041SDavid Howells len = min_t(size_t, len, npages * PAGE_SIZE); 33833b3b041SDavid Howells npages = DIV_ROUND_UP(len, PAGE_SIZE); 33933b3b041SDavid Howells 34033b3b041SDavid Howells bv = kzalloc(array_size(npages, sizeof(bv[0])) + 34133b3b041SDavid Howells array_size(npages, sizeof(struct page *)), GFP_KERNEL); 34233b3b041SDavid Howells if (!bv) 34333b3b041SDavid Howells return -ENOMEM; 34433b3b041SDavid Howells 345e69f37bcSDavid Howells pages = (struct page **)(bv + npages); 34633b3b041SDavid Howells npages = alloc_pages_bulk_array(GFP_USER, npages, pages); 34733b3b041SDavid Howells if (!npages) { 34833b3b041SDavid Howells kfree(bv); 34933b3b041SDavid Howells return -ENOMEM; 35033b3b041SDavid Howells } 35133b3b041SDavid Howells 35233b3b041SDavid Howells remain = len = min_t(size_t, len, npages * PAGE_SIZE); 35333b3b041SDavid Howells 35433b3b041SDavid Howells for (i = 0; i < npages; i++) { 35533b3b041SDavid Howells chunk = min_t(size_t, PAGE_SIZE, remain); 35633b3b041SDavid Howells bv[i].bv_page = pages[i]; 35733b3b041SDavid Howells bv[i].bv_offset = 0; 35833b3b041SDavid Howells bv[i].bv_len = chunk; 35933b3b041SDavid Howells remain -= chunk; 36033b3b041SDavid Howells } 36133b3b041SDavid Howells 36233b3b041SDavid Howells /* Do the I/O */ 36333b3b041SDavid Howells iov_iter_bvec(&to, ITER_DEST, bv, npages, len); 36433b3b041SDavid Howells init_sync_kiocb(&kiocb, in); 36533b3b041SDavid Howells kiocb.ki_pos = *ppos; 36633b3b041SDavid Howells ret = call_read_iter(in, &kiocb, &to); 36733b3b041SDavid Howells 36833b3b041SDavid Howells if (ret > 0) { 369e69f37bcSDavid Howells keep = DIV_ROUND_UP(ret, PAGE_SIZE); 37033b3b041SDavid Howells *ppos = kiocb.ki_pos; 37133b3b041SDavid Howells file_accessed(in); 37233b3b041SDavid Howells } else if (ret < 0) { 37333b3b041SDavid Howells /* 37433b3b041SDavid Howells * callers of ->splice_read() expect -EAGAIN on 37533b3b041SDavid Howells * "can't put anything in there", rather than -EFAULT. 37633b3b041SDavid Howells */ 37733b3b041SDavid Howells if (ret == -EFAULT) 37833b3b041SDavid Howells ret = -EAGAIN; 37933b3b041SDavid Howells } 38033b3b041SDavid Howells 38133b3b041SDavid Howells /* Free any pages that didn't get touched at all. */ 382e69f37bcSDavid Howells if (keep < npages) 383e69f37bcSDavid Howells release_pages(pages + keep, npages - keep); 38433b3b041SDavid Howells 38533b3b041SDavid Howells /* Push the remaining pages into the pipe. */ 386e69f37bcSDavid Howells remain = ret; 387e69f37bcSDavid Howells for (i = 0; i < keep; i++) { 38833b3b041SDavid Howells struct pipe_buffer *buf = pipe_head_buf(pipe); 38933b3b041SDavid Howells 39033b3b041SDavid Howells chunk = min_t(size_t, remain, PAGE_SIZE); 39133b3b041SDavid Howells *buf = (struct pipe_buffer) { 39233b3b041SDavid Howells .ops = &default_pipe_buf_ops, 39333b3b041SDavid Howells .page = bv[i].bv_page, 39433b3b041SDavid Howells .offset = 0, 39533b3b041SDavid Howells .len = chunk, 39633b3b041SDavid Howells }; 39733b3b041SDavid Howells pipe->head++; 39833b3b041SDavid Howells remain -= chunk; 39933b3b041SDavid Howells } 40033b3b041SDavid Howells 40133b3b041SDavid Howells kfree(bv); 40233b3b041SDavid Howells return ret; 40333b3b041SDavid Howells } 40469df79a4SDavid Howells EXPORT_SYMBOL(copy_splice_read); 40533b3b041SDavid Howells 406241699cdSAl Viro const struct pipe_buf_operations default_pipe_buf_ops = { 4076818173bSMiklos Szeredi .release = generic_pipe_buf_release, 408c928f642SChristoph Hellwig .try_steal = generic_pipe_buf_try_steal, 4096818173bSMiklos Szeredi .get = generic_pipe_buf_get, 4106818173bSMiklos Szeredi }; 4116818173bSMiklos Szeredi 41228a625cbSMiklos Szeredi /* Pipe buffer operations for a socket and similar. */ 41328a625cbSMiklos Szeredi const struct pipe_buf_operations nosteal_pipe_buf_ops = { 41428a625cbSMiklos Szeredi .release = generic_pipe_buf_release, 41528a625cbSMiklos Szeredi .get = generic_pipe_buf_get, 41628a625cbSMiklos Szeredi }; 41728a625cbSMiklos Szeredi EXPORT_SYMBOL(nosteal_pipe_buf_ops); 41828a625cbSMiklos Szeredi 4195274f052SJens Axboe /* 4204f6f0bd2SJens Axboe * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' 421016b661eSJens Axboe * using sendpage(). Return the number of bytes sent. 4225274f052SJens Axboe */ 42376ad4d11SJens Axboe static int pipe_to_sendpage(struct pipe_inode_info *pipe, 4245274f052SJens Axboe struct pipe_buffer *buf, struct splice_desc *sd) 4255274f052SJens Axboe { 4266a14b90bSJens Axboe struct file *file = sd->u.file; 4275274f052SJens Axboe loff_t pos = sd->pos; 428a8adbe37SMichał Mirosław int more; 4295274f052SJens Axboe 43072c2d531SAl Viro if (!likely(file->f_op->sendpage)) 431a8adbe37SMichał Mirosław return -EINVAL; 432a8adbe37SMichał Mirosław 43335f9c09fSEric Dumazet more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; 434ae62ca7bSEric Dumazet 4358cefc107SDavid Howells if (sd->len < sd->total_len && 4368cefc107SDavid Howells pipe_occupancy(pipe->head, pipe->tail) > 1) 43735f9c09fSEric Dumazet more |= MSG_SENDPAGE_NOTLAST; 438ae62ca7bSEric Dumazet 439a8adbe37SMichał Mirosław return file->f_op->sendpage(file, buf->page, buf->offset, 440f84d7519SJens Axboe sd->len, &pos, more); 4415274f052SJens Axboe } 4425274f052SJens Axboe 443b3c2d2ddSMiklos Szeredi static void wakeup_pipe_writers(struct pipe_inode_info *pipe) 444b3c2d2ddSMiklos Szeredi { 445b3c2d2ddSMiklos Szeredi smp_mb(); 4460ddad21dSLinus Torvalds if (waitqueue_active(&pipe->wr_wait)) 4470ddad21dSLinus Torvalds wake_up_interruptible(&pipe->wr_wait); 448b3c2d2ddSMiklos Szeredi kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); 449b3c2d2ddSMiklos Szeredi } 450b3c2d2ddSMiklos Szeredi 451b3c2d2ddSMiklos Szeredi /** 452b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed - feed available data from a pipe to a file 453b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 454b3c2d2ddSMiklos Szeredi * @sd: information to @actor 455b3c2d2ddSMiklos Szeredi * @actor: handler that splices the data 456b3c2d2ddSMiklos Szeredi * 457b3c2d2ddSMiklos Szeredi * Description: 458b3c2d2ddSMiklos Szeredi * This function loops over the pipe and calls @actor to do the 459b3c2d2ddSMiklos Szeredi * actual moving of a single struct pipe_buffer to the desired 460b3c2d2ddSMiklos Szeredi * destination. It returns when there's no more buffers left in 461b3c2d2ddSMiklos Szeredi * the pipe or if the requested number of bytes (@sd->total_len) 462b3c2d2ddSMiklos Szeredi * have been copied. It returns a positive number (one) if the 463b3c2d2ddSMiklos Szeredi * pipe needs to be filled with more data, zero if the required 464b3c2d2ddSMiklos Szeredi * number of bytes have been copied and -errno on error. 465b3c2d2ddSMiklos Szeredi * 466b3c2d2ddSMiklos Szeredi * This, together with splice_from_pipe_{begin,end,next}, may be 467b3c2d2ddSMiklos Szeredi * used to implement the functionality of __splice_from_pipe() when 468b3c2d2ddSMiklos Szeredi * locking is required around copying the pipe buffers to the 469b3c2d2ddSMiklos Szeredi * destination. 470b3c2d2ddSMiklos Szeredi */ 47196f9bc8fSAl Viro static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, 472b3c2d2ddSMiklos Szeredi splice_actor *actor) 473b3c2d2ddSMiklos Szeredi { 4748cefc107SDavid Howells unsigned int head = pipe->head; 4758cefc107SDavid Howells unsigned int tail = pipe->tail; 4768cefc107SDavid Howells unsigned int mask = pipe->ring_size - 1; 477b3c2d2ddSMiklos Szeredi int ret; 478b3c2d2ddSMiklos Szeredi 479ec057595SLinus Torvalds while (!pipe_empty(head, tail)) { 4808cefc107SDavid Howells struct pipe_buffer *buf = &pipe->bufs[tail & mask]; 481b3c2d2ddSMiklos Szeredi 482b3c2d2ddSMiklos Szeredi sd->len = buf->len; 483b3c2d2ddSMiklos Szeredi if (sd->len > sd->total_len) 484b3c2d2ddSMiklos Szeredi sd->len = sd->total_len; 485b3c2d2ddSMiklos Szeredi 486fba597dbSMiklos Szeredi ret = pipe_buf_confirm(pipe, buf); 487a8adbe37SMichał Mirosław if (unlikely(ret)) { 488b3c2d2ddSMiklos Szeredi if (ret == -ENODATA) 489b3c2d2ddSMiklos Szeredi ret = 0; 490b3c2d2ddSMiklos Szeredi return ret; 491b3c2d2ddSMiklos Szeredi } 492a8adbe37SMichał Mirosław 493a8adbe37SMichał Mirosław ret = actor(pipe, buf, sd); 494a8adbe37SMichał Mirosław if (ret <= 0) 495a8adbe37SMichał Mirosław return ret; 496a8adbe37SMichał Mirosław 497b3c2d2ddSMiklos Szeredi buf->offset += ret; 498b3c2d2ddSMiklos Szeredi buf->len -= ret; 499b3c2d2ddSMiklos Szeredi 500b3c2d2ddSMiklos Szeredi sd->num_spliced += ret; 501b3c2d2ddSMiklos Szeredi sd->len -= ret; 502b3c2d2ddSMiklos Szeredi sd->pos += ret; 503b3c2d2ddSMiklos Szeredi sd->total_len -= ret; 504b3c2d2ddSMiklos Szeredi 505b3c2d2ddSMiklos Szeredi if (!buf->len) { 506a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 5078cefc107SDavid Howells tail++; 5088cefc107SDavid Howells pipe->tail = tail; 5096447a3cfSAl Viro if (pipe->files) 510b3c2d2ddSMiklos Szeredi sd->need_wakeup = true; 511b3c2d2ddSMiklos Szeredi } 512b3c2d2ddSMiklos Szeredi 513b3c2d2ddSMiklos Szeredi if (!sd->total_len) 514b3c2d2ddSMiklos Szeredi return 0; 515b3c2d2ddSMiklos Szeredi } 516b3c2d2ddSMiklos Szeredi 517b3c2d2ddSMiklos Szeredi return 1; 518b3c2d2ddSMiklos Szeredi } 519b3c2d2ddSMiklos Szeredi 520d1a819a2SLinus Torvalds /* We know we have a pipe buffer, but maybe it's empty? */ 521d1a819a2SLinus Torvalds static inline bool eat_empty_buffer(struct pipe_inode_info *pipe) 522d1a819a2SLinus Torvalds { 523d1a819a2SLinus Torvalds unsigned int tail = pipe->tail; 524d1a819a2SLinus Torvalds unsigned int mask = pipe->ring_size - 1; 525d1a819a2SLinus Torvalds struct pipe_buffer *buf = &pipe->bufs[tail & mask]; 526d1a819a2SLinus Torvalds 527d1a819a2SLinus Torvalds if (unlikely(!buf->len)) { 528d1a819a2SLinus Torvalds pipe_buf_release(pipe, buf); 529d1a819a2SLinus Torvalds pipe->tail = tail+1; 530d1a819a2SLinus Torvalds return true; 531d1a819a2SLinus Torvalds } 532d1a819a2SLinus Torvalds 533d1a819a2SLinus Torvalds return false; 534d1a819a2SLinus Torvalds } 535d1a819a2SLinus Torvalds 536b3c2d2ddSMiklos Szeredi /** 537b3c2d2ddSMiklos Szeredi * splice_from_pipe_next - wait for some data to splice from 538b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 539b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 540b3c2d2ddSMiklos Szeredi * 541b3c2d2ddSMiklos Szeredi * Description: 542b3c2d2ddSMiklos Szeredi * This function will wait for some data and return a positive 543b3c2d2ddSMiklos Szeredi * value (one) if pipe buffers are available. It will return zero 544b3c2d2ddSMiklos Szeredi * or -errno if no more data needs to be spliced. 545b3c2d2ddSMiklos Szeredi */ 54696f9bc8fSAl Viro static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) 547b3c2d2ddSMiklos Szeredi { 548c725bfceSJan Kara /* 549c725bfceSJan Kara * Check for signal early to make process killable when there are 550c725bfceSJan Kara * always buffers available 551c725bfceSJan Kara */ 552c725bfceSJan Kara if (signal_pending(current)) 553c725bfceSJan Kara return -ERESTARTSYS; 554c725bfceSJan Kara 555d1a819a2SLinus Torvalds repeat: 5568cefc107SDavid Howells while (pipe_empty(pipe->head, pipe->tail)) { 557b3c2d2ddSMiklos Szeredi if (!pipe->writers) 558b3c2d2ddSMiklos Szeredi return 0; 559b3c2d2ddSMiklos Szeredi 560a28c8b9dSLinus Torvalds if (sd->num_spliced) 561b3c2d2ddSMiklos Szeredi return 0; 562b3c2d2ddSMiklos Szeredi 563b3c2d2ddSMiklos Szeredi if (sd->flags & SPLICE_F_NONBLOCK) 564b3c2d2ddSMiklos Szeredi return -EAGAIN; 565b3c2d2ddSMiklos Szeredi 566b3c2d2ddSMiklos Szeredi if (signal_pending(current)) 567b3c2d2ddSMiklos Szeredi return -ERESTARTSYS; 568b3c2d2ddSMiklos Szeredi 569b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) { 570b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 571b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 572b3c2d2ddSMiklos Szeredi } 573b3c2d2ddSMiklos Szeredi 574472e5b05SLinus Torvalds pipe_wait_readable(pipe); 575b3c2d2ddSMiklos Szeredi } 576b3c2d2ddSMiklos Szeredi 577d1a819a2SLinus Torvalds if (eat_empty_buffer(pipe)) 578d1a819a2SLinus Torvalds goto repeat; 579d1a819a2SLinus Torvalds 580b3c2d2ddSMiklos Szeredi return 1; 581b3c2d2ddSMiklos Szeredi } 582b3c2d2ddSMiklos Szeredi 583b3c2d2ddSMiklos Szeredi /** 584b3c2d2ddSMiklos Szeredi * splice_from_pipe_begin - start splicing from pipe 585b80901bbSRandy Dunlap * @sd: information about the splice operation 586b3c2d2ddSMiklos Szeredi * 587b3c2d2ddSMiklos Szeredi * Description: 588b3c2d2ddSMiklos Szeredi * This function should be called before a loop containing 589b3c2d2ddSMiklos Szeredi * splice_from_pipe_next() and splice_from_pipe_feed() to 590b3c2d2ddSMiklos Szeredi * initialize the necessary fields of @sd. 591b3c2d2ddSMiklos Szeredi */ 59296f9bc8fSAl Viro static void splice_from_pipe_begin(struct splice_desc *sd) 593b3c2d2ddSMiklos Szeredi { 594b3c2d2ddSMiklos Szeredi sd->num_spliced = 0; 595b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 596b3c2d2ddSMiklos Szeredi } 597b3c2d2ddSMiklos Szeredi 598b3c2d2ddSMiklos Szeredi /** 599b3c2d2ddSMiklos Szeredi * splice_from_pipe_end - finish splicing from pipe 600b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 601b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 602b3c2d2ddSMiklos Szeredi * 603b3c2d2ddSMiklos Szeredi * Description: 604b3c2d2ddSMiklos Szeredi * This function will wake up pipe writers if necessary. It should 605b3c2d2ddSMiklos Szeredi * be called after a loop containing splice_from_pipe_next() and 606b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed(). 607b3c2d2ddSMiklos Szeredi */ 60896f9bc8fSAl Viro static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) 609b3c2d2ddSMiklos Szeredi { 610b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) 611b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 612b3c2d2ddSMiklos Szeredi } 613b3c2d2ddSMiklos Szeredi 614932cc6d4SJens Axboe /** 615932cc6d4SJens Axboe * __splice_from_pipe - splice data from a pipe to given actor 616932cc6d4SJens Axboe * @pipe: pipe to splice from 617932cc6d4SJens Axboe * @sd: information to @actor 618932cc6d4SJens Axboe * @actor: handler that splices the data 619932cc6d4SJens Axboe * 620932cc6d4SJens Axboe * Description: 621932cc6d4SJens Axboe * This function does little more than loop over the pipe and call 622932cc6d4SJens Axboe * @actor to do the actual moving of a single struct pipe_buffer to 623932cc6d4SJens Axboe * the desired destination. See pipe_to_file, pipe_to_sendpage, or 624932cc6d4SJens Axboe * pipe_to_user. 625932cc6d4SJens Axboe * 62683f9135bSJens Axboe */ 627c66ab6faSJens Axboe ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, 628c66ab6faSJens Axboe splice_actor *actor) 6295274f052SJens Axboe { 630b3c2d2ddSMiklos Szeredi int ret; 6315274f052SJens Axboe 632b3c2d2ddSMiklos Szeredi splice_from_pipe_begin(sd); 633b3c2d2ddSMiklos Szeredi do { 634c2489e07SJan Kara cond_resched(); 635b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_next(pipe, sd); 636b3c2d2ddSMiklos Szeredi if (ret > 0) 637b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_feed(pipe, sd, actor); 638b3c2d2ddSMiklos Szeredi } while (ret > 0); 639b3c2d2ddSMiklos Szeredi splice_from_pipe_end(pipe, sd); 6405274f052SJens Axboe 641b3c2d2ddSMiklos Szeredi return sd->num_spliced ? sd->num_spliced : ret; 6425274f052SJens Axboe } 64340bee44eSMark Fasheh EXPORT_SYMBOL(__splice_from_pipe); 6445274f052SJens Axboe 645932cc6d4SJens Axboe /** 646932cc6d4SJens Axboe * splice_from_pipe - splice data from a pipe to a file 647932cc6d4SJens Axboe * @pipe: pipe to splice from 648932cc6d4SJens Axboe * @out: file to splice to 649932cc6d4SJens Axboe * @ppos: position in @out 650932cc6d4SJens Axboe * @len: how many bytes to splice 651932cc6d4SJens Axboe * @flags: splice modifier flags 652932cc6d4SJens Axboe * @actor: handler that splices the data 653932cc6d4SJens Axboe * 654932cc6d4SJens Axboe * Description: 6552933970bSMiklos Szeredi * See __splice_from_pipe. This function locks the pipe inode, 656932cc6d4SJens Axboe * otherwise it's identical to __splice_from_pipe(). 657932cc6d4SJens Axboe * 658932cc6d4SJens Axboe */ 6596da61809SMark Fasheh ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, 6606da61809SMark Fasheh loff_t *ppos, size_t len, unsigned int flags, 6616da61809SMark Fasheh splice_actor *actor) 6626da61809SMark Fasheh { 6636da61809SMark Fasheh ssize_t ret; 664c66ab6faSJens Axboe struct splice_desc sd = { 665c66ab6faSJens Axboe .total_len = len, 666c66ab6faSJens Axboe .flags = flags, 667c66ab6faSJens Axboe .pos = *ppos, 6686a14b90bSJens Axboe .u.file = out, 669c66ab6faSJens Axboe }; 6706da61809SMark Fasheh 67161e0d47cSMiklos Szeredi pipe_lock(pipe); 672c66ab6faSJens Axboe ret = __splice_from_pipe(pipe, &sd, actor); 67361e0d47cSMiklos Szeredi pipe_unlock(pipe); 6746da61809SMark Fasheh 6756da61809SMark Fasheh return ret; 6766da61809SMark Fasheh } 6776da61809SMark Fasheh 6786da61809SMark Fasheh /** 6798d020765SAl Viro * iter_file_splice_write - splice data from a pipe to a file 6808d020765SAl Viro * @pipe: pipe info 6818d020765SAl Viro * @out: file to write to 6828d020765SAl Viro * @ppos: position in @out 6838d020765SAl Viro * @len: number of bytes to splice 6848d020765SAl Viro * @flags: splice modifier flags 6858d020765SAl Viro * 6868d020765SAl Viro * Description: 6878d020765SAl Viro * Will either move or copy pages (determined by @flags options) from 6888d020765SAl Viro * the given pipe inode to the given file. 6898d020765SAl Viro * This one is ->write_iter-based. 6908d020765SAl Viro * 6918d020765SAl Viro */ 6928d020765SAl Viro ssize_t 6938d020765SAl Viro iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out, 6948d020765SAl Viro loff_t *ppos, size_t len, unsigned int flags) 6958d020765SAl Viro { 6968d020765SAl Viro struct splice_desc sd = { 6978d020765SAl Viro .total_len = len, 6988d020765SAl Viro .flags = flags, 6998d020765SAl Viro .pos = *ppos, 7008d020765SAl Viro .u.file = out, 7018d020765SAl Viro }; 7026718b6f8SDavid Howells int nbufs = pipe->max_usage; 7038d020765SAl Viro struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec), 7048d020765SAl Viro GFP_KERNEL); 7058d020765SAl Viro ssize_t ret; 7068d020765SAl Viro 7078d020765SAl Viro if (unlikely(!array)) 7088d020765SAl Viro return -ENOMEM; 7098d020765SAl Viro 7108d020765SAl Viro pipe_lock(pipe); 7118d020765SAl Viro 7128d020765SAl Viro splice_from_pipe_begin(&sd); 7138d020765SAl Viro while (sd.total_len) { 7148d020765SAl Viro struct iov_iter from; 715ec057595SLinus Torvalds unsigned int head, tail, mask; 7168d020765SAl Viro size_t left; 7178cefc107SDavid Howells int n; 7188d020765SAl Viro 7198d020765SAl Viro ret = splice_from_pipe_next(pipe, &sd); 7208d020765SAl Viro if (ret <= 0) 7218d020765SAl Viro break; 7228d020765SAl Viro 7236718b6f8SDavid Howells if (unlikely(nbufs < pipe->max_usage)) { 7248d020765SAl Viro kfree(array); 7256718b6f8SDavid Howells nbufs = pipe->max_usage; 7268d020765SAl Viro array = kcalloc(nbufs, sizeof(struct bio_vec), 7278d020765SAl Viro GFP_KERNEL); 7288d020765SAl Viro if (!array) { 7298d020765SAl Viro ret = -ENOMEM; 7308d020765SAl Viro break; 7318d020765SAl Viro } 7328d020765SAl Viro } 7338d020765SAl Viro 734ec057595SLinus Torvalds head = pipe->head; 735ec057595SLinus Torvalds tail = pipe->tail; 736ec057595SLinus Torvalds mask = pipe->ring_size - 1; 737ec057595SLinus Torvalds 7388d020765SAl Viro /* build the vector */ 7398d020765SAl Viro left = sd.total_len; 7400f1d344fSPavel Begunkov for (n = 0; !pipe_empty(head, tail) && left && n < nbufs; tail++) { 7418cefc107SDavid Howells struct pipe_buffer *buf = &pipe->bufs[tail & mask]; 7428d020765SAl Viro size_t this_len = buf->len; 7438d020765SAl Viro 7440f1d344fSPavel Begunkov /* zero-length bvecs are not supported, skip them */ 7450f1d344fSPavel Begunkov if (!this_len) 7460f1d344fSPavel Begunkov continue; 7470f1d344fSPavel Begunkov this_len = min(this_len, left); 7488d020765SAl Viro 749fba597dbSMiklos Szeredi ret = pipe_buf_confirm(pipe, buf); 7508d020765SAl Viro if (unlikely(ret)) { 7518d020765SAl Viro if (ret == -ENODATA) 7528d020765SAl Viro ret = 0; 7538d020765SAl Viro goto done; 7548d020765SAl Viro } 7558d020765SAl Viro 756664e4078SChristoph Hellwig bvec_set_page(&array[n], buf->page, this_len, 757664e4078SChristoph Hellwig buf->offset); 7588d020765SAl Viro left -= this_len; 7590f1d344fSPavel Begunkov n++; 7608d020765SAl Viro } 7618d020765SAl Viro 762de4eda9dSAl Viro iov_iter_bvec(&from, ITER_SOURCE, array, n, sd.total_len - left); 763abbb6589SChristoph Hellwig ret = vfs_iter_write(out, &from, &sd.pos, 0); 7648d020765SAl Viro if (ret <= 0) 7658d020765SAl Viro break; 7668d020765SAl Viro 7678d020765SAl Viro sd.num_spliced += ret; 7688d020765SAl Viro sd.total_len -= ret; 769dbe4e192SChristoph Hellwig *ppos = sd.pos; 7708d020765SAl Viro 7718d020765SAl Viro /* dismiss the fully eaten buffers, adjust the partial one */ 7728cefc107SDavid Howells tail = pipe->tail; 7738d020765SAl Viro while (ret) { 7748cefc107SDavid Howells struct pipe_buffer *buf = &pipe->bufs[tail & mask]; 7758d020765SAl Viro if (ret >= buf->len) { 7768d020765SAl Viro ret -= buf->len; 7778d020765SAl Viro buf->len = 0; 778a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 7798cefc107SDavid Howells tail++; 7808cefc107SDavid Howells pipe->tail = tail; 7818d020765SAl Viro if (pipe->files) 7828d020765SAl Viro sd.need_wakeup = true; 7838d020765SAl Viro } else { 7848d020765SAl Viro buf->offset += ret; 7858d020765SAl Viro buf->len -= ret; 7868d020765SAl Viro ret = 0; 7878d020765SAl Viro } 7888d020765SAl Viro } 7898d020765SAl Viro } 7908d020765SAl Viro done: 7918d020765SAl Viro kfree(array); 7928d020765SAl Viro splice_from_pipe_end(pipe, &sd); 7938d020765SAl Viro 7948d020765SAl Viro pipe_unlock(pipe); 7958d020765SAl Viro 7968d020765SAl Viro if (sd.num_spliced) 7978d020765SAl Viro ret = sd.num_spliced; 7988d020765SAl Viro 7998d020765SAl Viro return ret; 8008d020765SAl Viro } 8018d020765SAl Viro 8028d020765SAl Viro EXPORT_SYMBOL(iter_file_splice_write); 8038d020765SAl Viro 80483f9135bSJens Axboe /** 80583f9135bSJens Axboe * generic_splice_sendpage - splice data from a pipe to a socket 806932cc6d4SJens Axboe * @pipe: pipe to splice from 80783f9135bSJens Axboe * @out: socket to write to 808932cc6d4SJens Axboe * @ppos: position in @out 80983f9135bSJens Axboe * @len: number of bytes to splice 81083f9135bSJens Axboe * @flags: splice modifier flags 81183f9135bSJens Axboe * 812932cc6d4SJens Axboe * Description: 81383f9135bSJens Axboe * Will send @len bytes from the pipe to a network socket. No data copying 81483f9135bSJens Axboe * is involved. 81583f9135bSJens Axboe * 81683f9135bSJens Axboe */ 8173a326a2cSIngo Molnar ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, 818cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 8195274f052SJens Axboe { 82000522fb4SJens Axboe return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); 8215274f052SJens Axboe } 8225274f052SJens Axboe 823059a8f37SJens Axboe EXPORT_SYMBOL(generic_splice_sendpage); 824a0f06780SJeff Garzik 82536e2c742SChristoph Hellwig static int warn_unsupported(struct file *file, const char *op) 82636e2c742SChristoph Hellwig { 82736e2c742SChristoph Hellwig pr_debug_ratelimited( 82836e2c742SChristoph Hellwig "splice %s not supported for file %pD4 (pid: %d comm: %.20s)\n", 82936e2c742SChristoph Hellwig op, file, current->pid, current->comm); 83036e2c742SChristoph Hellwig return -EINVAL; 83136e2c742SChristoph Hellwig } 83236e2c742SChristoph Hellwig 83383f9135bSJens Axboe /* 83483f9135bSJens Axboe * Attempt to initiate a splice from pipe to file. 83583f9135bSJens Axboe */ 8363a326a2cSIngo Molnar static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, 837cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 8385274f052SJens Axboe { 83936e2c742SChristoph Hellwig if (unlikely(!out->f_op->splice_write)) 84036e2c742SChristoph Hellwig return warn_unsupported(out, "write"); 84100c285d0SChristoph Hellwig return out->f_op->splice_write(pipe, out, ppos, len, flags); 8425274f052SJens Axboe } 8435274f052SJens Axboe 8446a3f30b8SDavid Howells /** 8456a3f30b8SDavid Howells * vfs_splice_read - Read data from a file and splice it into a pipe 8466a3f30b8SDavid Howells * @in: File to splice from 8476a3f30b8SDavid Howells * @ppos: Input file offset 8486a3f30b8SDavid Howells * @pipe: Pipe to splice to 8496a3f30b8SDavid Howells * @len: Number of bytes to splice 8506a3f30b8SDavid Howells * @flags: Splice modifier flags (SPLICE_F_*) 8516a3f30b8SDavid Howells * 8526a3f30b8SDavid Howells * Splice the requested amount of data from the input file to the pipe. This 8536a3f30b8SDavid Howells * is synchronous as the caller must hold the pipe lock across the entire 8546a3f30b8SDavid Howells * operation. 8556a3f30b8SDavid Howells * 8566a3f30b8SDavid Howells * If successful, it returns the amount of data spliced, 0 if it hit the EOF or 8576a3f30b8SDavid Howells * a hole and a negative error code otherwise. 85883f9135bSJens Axboe */ 8596a3f30b8SDavid Howells long vfs_splice_read(struct file *in, loff_t *ppos, 860cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 861cbb7e577SJens Axboe unsigned int flags) 8625274f052SJens Axboe { 863313d64a3SAl Viro unsigned int p_space; 8645274f052SJens Axboe int ret; 8655274f052SJens Axboe 86649570e9bSJens Axboe if (unlikely(!(in->f_mode & FMODE_READ))) 8675274f052SJens Axboe return -EBADF; 868123856f0SDavid Howells if (!len) 869123856f0SDavid Howells return 0; 8705274f052SJens Axboe 871313d64a3SAl Viro /* Don't try to read more the pipe has space for. */ 872313d64a3SAl Viro p_space = pipe->max_usage - pipe_occupancy(pipe->head, pipe->tail); 873313d64a3SAl Viro len = min_t(size_t, len, p_space << PAGE_SHIFT); 874313d64a3SAl Viro 875cbb7e577SJens Axboe ret = rw_verify_area(READ, in, ppos, len); 8765274f052SJens Axboe if (unlikely(ret < 0)) 8775274f052SJens Axboe return ret; 8785274f052SJens Axboe 87903cc0789SAl Viro if (unlikely(len > MAX_RW_COUNT)) 88003cc0789SAl Viro len = MAX_RW_COUNT; 88103cc0789SAl Viro 88236e2c742SChristoph Hellwig if (unlikely(!in->f_op->splice_read)) 88336e2c742SChristoph Hellwig return warn_unsupported(in, "read"); 884aa3dbde8SDavid Howells /* 885b85930a0SDavid Howells * O_DIRECT and DAX don't deal with the pagecache, so we allocate a 886b85930a0SDavid Howells * buffer, copy into it and splice that into the pipe. 887aa3dbde8SDavid Howells */ 888b85930a0SDavid Howells if ((in->f_flags & O_DIRECT) || IS_DAX(in->f_mapping->host)) 889aa3dbde8SDavid Howells return copy_splice_read(in, ppos, pipe, len, flags); 8902bc01060SChristoph Hellwig return in->f_op->splice_read(in, ppos, pipe, len, flags); 8915274f052SJens Axboe } 8926a3f30b8SDavid Howells EXPORT_SYMBOL_GPL(vfs_splice_read); 8935274f052SJens Axboe 894932cc6d4SJens Axboe /** 895932cc6d4SJens Axboe * splice_direct_to_actor - splices data directly between two non-pipes 896932cc6d4SJens Axboe * @in: file to splice from 897932cc6d4SJens Axboe * @sd: actor information on where to splice to 898932cc6d4SJens Axboe * @actor: handles the data splicing 899932cc6d4SJens Axboe * 900932cc6d4SJens Axboe * Description: 901932cc6d4SJens Axboe * This is a special case helper to splice directly between two 902932cc6d4SJens Axboe * points, without requiring an explicit pipe. Internally an allocated 903932cc6d4SJens Axboe * pipe is cached in the process, and reused during the lifetime of 904932cc6d4SJens Axboe * that process. 905932cc6d4SJens Axboe * 906c66ab6faSJens Axboe */ 907c66ab6faSJens Axboe ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, 908c66ab6faSJens Axboe splice_direct_actor *actor) 909b92ce558SJens Axboe { 910b92ce558SJens Axboe struct pipe_inode_info *pipe; 911b92ce558SJens Axboe long ret, bytes; 912c66ab6faSJens Axboe size_t len; 9130ff28d9fSChristophe Leroy int i, flags, more; 914b92ce558SJens Axboe 915b92ce558SJens Axboe /* 91697ef77c5SJason A. Donenfeld * We require the input to be seekable, as we don't want to randomly 91797ef77c5SJason A. Donenfeld * drop data for eg socket -> socket splicing. Use the piped splicing 91897ef77c5SJason A. Donenfeld * for that! 919b92ce558SJens Axboe */ 92097ef77c5SJason A. Donenfeld if (unlikely(!(in->f_mode & FMODE_LSEEK))) 921b92ce558SJens Axboe return -EINVAL; 922b92ce558SJens Axboe 923b92ce558SJens Axboe /* 924b92ce558SJens Axboe * neither in nor out is a pipe, setup an internal pipe attached to 925b92ce558SJens Axboe * 'out' and transfer the wanted data from 'in' to 'out' through that 926b92ce558SJens Axboe */ 927b92ce558SJens Axboe pipe = current->splice_pipe; 92849570e9bSJens Axboe if (unlikely(!pipe)) { 9297bee130eSAl Viro pipe = alloc_pipe_info(); 930b92ce558SJens Axboe if (!pipe) 931b92ce558SJens Axboe return -ENOMEM; 932b92ce558SJens Axboe 933b92ce558SJens Axboe /* 934b92ce558SJens Axboe * We don't have an immediate reader, but we'll read the stuff 93500522fb4SJens Axboe * out of the pipe right after the splice_to_pipe(). So set 936b92ce558SJens Axboe * PIPE_READERS appropriately. 937b92ce558SJens Axboe */ 938b92ce558SJens Axboe pipe->readers = 1; 939b92ce558SJens Axboe 940b92ce558SJens Axboe current->splice_pipe = pipe; 941b92ce558SJens Axboe } 942b92ce558SJens Axboe 943b92ce558SJens Axboe /* 94473d62d83SIngo Molnar * Do the splice. 945b92ce558SJens Axboe */ 946b92ce558SJens Axboe bytes = 0; 947c66ab6faSJens Axboe len = sd->total_len; 948c66ab6faSJens Axboe flags = sd->flags; 949c66ab6faSJens Axboe 950c66ab6faSJens Axboe /* 951c66ab6faSJens Axboe * Don't block on output, we have to drain the direct pipe. 952c66ab6faSJens Axboe */ 953c66ab6faSJens Axboe sd->flags &= ~SPLICE_F_NONBLOCK; 9540ff28d9fSChristophe Leroy more = sd->flags & SPLICE_F_MORE; 955b92ce558SJens Axboe 9568cefc107SDavid Howells WARN_ON_ONCE(!pipe_empty(pipe->head, pipe->tail)); 95717614445SDarrick J. Wong 958b92ce558SJens Axboe while (len) { 95951a92c0fSJens Axboe size_t read_len; 960a82c53a0STom Zanussi loff_t pos = sd->pos, prev_pos = pos; 961b92ce558SJens Axboe 9626a3f30b8SDavid Howells ret = vfs_splice_read(in, &pos, pipe, len, flags); 96351a92c0fSJens Axboe if (unlikely(ret <= 0)) 964b92ce558SJens Axboe goto out_release; 965b92ce558SJens Axboe 966b92ce558SJens Axboe read_len = ret; 967c66ab6faSJens Axboe sd->total_len = read_len; 968b92ce558SJens Axboe 969b92ce558SJens Axboe /* 9700ff28d9fSChristophe Leroy * If more data is pending, set SPLICE_F_MORE 9710ff28d9fSChristophe Leroy * If this is the last data and SPLICE_F_MORE was not set 9720ff28d9fSChristophe Leroy * initially, clears it. 9730ff28d9fSChristophe Leroy */ 9740ff28d9fSChristophe Leroy if (read_len < len) 9750ff28d9fSChristophe Leroy sd->flags |= SPLICE_F_MORE; 9760ff28d9fSChristophe Leroy else if (!more) 9770ff28d9fSChristophe Leroy sd->flags &= ~SPLICE_F_MORE; 9780ff28d9fSChristophe Leroy /* 979b92ce558SJens Axboe * NOTE: nonblocking mode only applies to the input. We 980b92ce558SJens Axboe * must not do the output in nonblocking mode as then we 981b92ce558SJens Axboe * could get stuck data in the internal pipe: 982b92ce558SJens Axboe */ 983c66ab6faSJens Axboe ret = actor(pipe, sd); 984a82c53a0STom Zanussi if (unlikely(ret <= 0)) { 985a82c53a0STom Zanussi sd->pos = prev_pos; 986b92ce558SJens Axboe goto out_release; 987a82c53a0STom Zanussi } 988b92ce558SJens Axboe 989b92ce558SJens Axboe bytes += ret; 990b92ce558SJens Axboe len -= ret; 991bcd4f3acSJens Axboe sd->pos = pos; 992b92ce558SJens Axboe 993a82c53a0STom Zanussi if (ret < read_len) { 994a82c53a0STom Zanussi sd->pos = prev_pos + ret; 99551a92c0fSJens Axboe goto out_release; 996b92ce558SJens Axboe } 997a82c53a0STom Zanussi } 998b92ce558SJens Axboe 9999e97198dSJens Axboe done: 10008cefc107SDavid Howells pipe->tail = pipe->head = 0; 10019e97198dSJens Axboe file_accessed(in); 1002b92ce558SJens Axboe return bytes; 1003b92ce558SJens Axboe 1004b92ce558SJens Axboe out_release: 1005b92ce558SJens Axboe /* 1006b92ce558SJens Axboe * If we did an incomplete transfer we must release 1007b92ce558SJens Axboe * the pipe buffers in question: 1008b92ce558SJens Axboe */ 10098cefc107SDavid Howells for (i = 0; i < pipe->ring_size; i++) { 10108cefc107SDavid Howells struct pipe_buffer *buf = &pipe->bufs[i]; 1011b92ce558SJens Axboe 1012a779638cSMiklos Szeredi if (buf->ops) 1013a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 1014b92ce558SJens Axboe } 1015b92ce558SJens Axboe 10169e97198dSJens Axboe if (!bytes) 10179e97198dSJens Axboe bytes = ret; 1018b92ce558SJens Axboe 10199e97198dSJens Axboe goto done; 1020c66ab6faSJens Axboe } 1021c66ab6faSJens Axboe EXPORT_SYMBOL(splice_direct_to_actor); 1022c66ab6faSJens Axboe 1023c66ab6faSJens Axboe static int direct_splice_actor(struct pipe_inode_info *pipe, 1024c66ab6faSJens Axboe struct splice_desc *sd) 1025c66ab6faSJens Axboe { 10266a14b90bSJens Axboe struct file *file = sd->u.file; 1027c66ab6faSJens Axboe 10287995bd28SAl Viro return do_splice_from(pipe, file, sd->opos, sd->total_len, 10292cb4b05eSChangli Gao sd->flags); 1030c66ab6faSJens Axboe } 1031c66ab6faSJens Axboe 1032932cc6d4SJens Axboe /** 1033932cc6d4SJens Axboe * do_splice_direct - splices data directly between two files 1034932cc6d4SJens Axboe * @in: file to splice from 1035932cc6d4SJens Axboe * @ppos: input file offset 1036932cc6d4SJens Axboe * @out: file to splice to 1037acdb37c3SRandy Dunlap * @opos: output file offset 1038932cc6d4SJens Axboe * @len: number of bytes to splice 1039932cc6d4SJens Axboe * @flags: splice modifier flags 1040932cc6d4SJens Axboe * 1041932cc6d4SJens Axboe * Description: 1042932cc6d4SJens Axboe * For use by do_sendfile(). splice can easily emulate sendfile, but 1043932cc6d4SJens Axboe * doing it in the application would incur an extra system call 1044932cc6d4SJens Axboe * (splice in + splice out, as compared to just sendfile()). So this helper 1045932cc6d4SJens Axboe * can splice directly through a process-private pipe. 1046932cc6d4SJens Axboe * 1047932cc6d4SJens Axboe */ 1048c66ab6faSJens Axboe long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, 10497995bd28SAl Viro loff_t *opos, size_t len, unsigned int flags) 1050c66ab6faSJens Axboe { 1051c66ab6faSJens Axboe struct splice_desc sd = { 1052c66ab6faSJens Axboe .len = len, 1053c66ab6faSJens Axboe .total_len = len, 1054c66ab6faSJens Axboe .flags = flags, 1055c66ab6faSJens Axboe .pos = *ppos, 10566a14b90bSJens Axboe .u.file = out, 10577995bd28SAl Viro .opos = opos, 1058c66ab6faSJens Axboe }; 105951a92c0fSJens Axboe long ret; 1060c66ab6faSJens Axboe 106118c67cb9SAl Viro if (unlikely(!(out->f_mode & FMODE_WRITE))) 106218c67cb9SAl Viro return -EBADF; 106318c67cb9SAl Viro 106418c67cb9SAl Viro if (unlikely(out->f_flags & O_APPEND)) 106518c67cb9SAl Viro return -EINVAL; 106618c67cb9SAl Viro 106718c67cb9SAl Viro ret = rw_verify_area(WRITE, out, opos, len); 106818c67cb9SAl Viro if (unlikely(ret < 0)) 106918c67cb9SAl Viro return ret; 107018c67cb9SAl Viro 1071c66ab6faSJens Axboe ret = splice_direct_to_actor(in, &sd, direct_splice_actor); 107251a92c0fSJens Axboe if (ret > 0) 1073a82c53a0STom Zanussi *ppos = sd.pos; 107451a92c0fSJens Axboe 1075c66ab6faSJens Axboe return ret; 1076b92ce558SJens Axboe } 10771c118596SMiklos Szeredi EXPORT_SYMBOL(do_splice_direct); 1078b92ce558SJens Axboe 10798924feffSAl Viro static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags) 10808924feffSAl Viro { 108152bce911SLinus Torvalds for (;;) { 108252bce911SLinus Torvalds if (unlikely(!pipe->readers)) { 108352bce911SLinus Torvalds send_sig(SIGPIPE, current, 0); 108452bce911SLinus Torvalds return -EPIPE; 108552bce911SLinus Torvalds } 10866718b6f8SDavid Howells if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) 108752bce911SLinus Torvalds return 0; 10888924feffSAl Viro if (flags & SPLICE_F_NONBLOCK) 10898924feffSAl Viro return -EAGAIN; 10908924feffSAl Viro if (signal_pending(current)) 10918924feffSAl Viro return -ERESTARTSYS; 1092472e5b05SLinus Torvalds pipe_wait_writable(pipe); 10938924feffSAl Viro } 10948924feffSAl Viro } 10958924feffSAl Viro 10967c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, 10977c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe, 10987c77f0b3SMiklos Szeredi size_t len, unsigned int flags); 1099ddac0d39SJens Axboe 1100b964bf53SAl Viro long splice_file_to_pipe(struct file *in, 1101faa97c48SAl Viro struct pipe_inode_info *opipe, 1102faa97c48SAl Viro loff_t *offset, 1103faa97c48SAl Viro size_t len, unsigned int flags) 1104faa97c48SAl Viro { 1105faa97c48SAl Viro long ret; 1106faa97c48SAl Viro 1107faa97c48SAl Viro pipe_lock(opipe); 1108faa97c48SAl Viro ret = wait_for_space(opipe, flags); 1109faa97c48SAl Viro if (!ret) 11106a3f30b8SDavid Howells ret = vfs_splice_read(in, offset, opipe, len, flags); 1111faa97c48SAl Viro pipe_unlock(opipe); 1112faa97c48SAl Viro if (ret > 0) 1113faa97c48SAl Viro wakeup_pipe_readers(opipe); 1114faa97c48SAl Viro return ret; 1115faa97c48SAl Viro } 1116faa97c48SAl Viro 1117ddac0d39SJens Axboe /* 111883f9135bSJens Axboe * Determine where to splice to/from. 111983f9135bSJens Axboe */ 1120ee6e00c8SJens Axboe long do_splice(struct file *in, loff_t *off_in, struct file *out, 1121ee6e00c8SJens Axboe loff_t *off_out, size_t len, unsigned int flags) 11225274f052SJens Axboe { 11237c77f0b3SMiklos Szeredi struct pipe_inode_info *ipipe; 11247c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe; 11257995bd28SAl Viro loff_t offset; 1126a4514ebdSJens Axboe long ret; 11275274f052SJens Axboe 112890da2e3fSPavel Begunkov if (unlikely(!(in->f_mode & FMODE_READ) || 112990da2e3fSPavel Begunkov !(out->f_mode & FMODE_WRITE))) 113090da2e3fSPavel Begunkov return -EBADF; 113190da2e3fSPavel Begunkov 1132c73be61cSDavid Howells ipipe = get_pipe_info(in, true); 1133c73be61cSDavid Howells opipe = get_pipe_info(out, true); 11347c77f0b3SMiklos Szeredi 11357c77f0b3SMiklos Szeredi if (ipipe && opipe) { 11367c77f0b3SMiklos Szeredi if (off_in || off_out) 11377c77f0b3SMiklos Szeredi return -ESPIPE; 11387c77f0b3SMiklos Szeredi 11397c77f0b3SMiklos Szeredi /* Splicing to self would be fun, but... */ 11407c77f0b3SMiklos Szeredi if (ipipe == opipe) 11417c77f0b3SMiklos Szeredi return -EINVAL; 11427c77f0b3SMiklos Szeredi 1143ee5e0011SSlavomir Kaslev if ((in->f_flags | out->f_flags) & O_NONBLOCK) 1144ee5e0011SSlavomir Kaslev flags |= SPLICE_F_NONBLOCK; 1145ee5e0011SSlavomir Kaslev 11467c77f0b3SMiklos Szeredi return splice_pipe_to_pipe(ipipe, opipe, len, flags); 11477c77f0b3SMiklos Szeredi } 11487c77f0b3SMiklos Szeredi 11497c77f0b3SMiklos Szeredi if (ipipe) { 1150529565dcSIngo Molnar if (off_in) 1151529565dcSIngo Molnar return -ESPIPE; 1152b92ce558SJens Axboe if (off_out) { 115319c9a49bSChangli Gao if (!(out->f_mode & FMODE_PWRITE)) 1154b92ce558SJens Axboe return -EINVAL; 1155ee6e00c8SJens Axboe offset = *off_out; 11567995bd28SAl Viro } else { 11577995bd28SAl Viro offset = out->f_pos; 11587995bd28SAl Viro } 1159529565dcSIngo Molnar 116018c67cb9SAl Viro if (unlikely(out->f_flags & O_APPEND)) 116118c67cb9SAl Viro return -EINVAL; 116218c67cb9SAl Viro 116318c67cb9SAl Viro ret = rw_verify_area(WRITE, out, &offset, len); 116418c67cb9SAl Viro if (unlikely(ret < 0)) 116518c67cb9SAl Viro return ret; 116618c67cb9SAl Viro 1167ee5e0011SSlavomir Kaslev if (in->f_flags & O_NONBLOCK) 1168ee5e0011SSlavomir Kaslev flags |= SPLICE_F_NONBLOCK; 1169ee5e0011SSlavomir Kaslev 1170500368f7SAl Viro file_start_write(out); 11717995bd28SAl Viro ret = do_splice_from(ipipe, out, &offset, len, flags); 1172500368f7SAl Viro file_end_write(out); 1173a4514ebdSJens Axboe 1174983652c6SChung-Chiang Cheng if (ret > 0) 1175983652c6SChung-Chiang Cheng fsnotify_modify(out); 1176983652c6SChung-Chiang Cheng 11777995bd28SAl Viro if (!off_out) 11787995bd28SAl Viro out->f_pos = offset; 1179ee6e00c8SJens Axboe else 1180ee6e00c8SJens Axboe *off_out = offset; 1181a4514ebdSJens Axboe 1182a4514ebdSJens Axboe return ret; 1183529565dcSIngo Molnar } 11845274f052SJens Axboe 11857c77f0b3SMiklos Szeredi if (opipe) { 1186529565dcSIngo Molnar if (off_out) 1187529565dcSIngo Molnar return -ESPIPE; 1188b92ce558SJens Axboe if (off_in) { 118919c9a49bSChangli Gao if (!(in->f_mode & FMODE_PREAD)) 1190b92ce558SJens Axboe return -EINVAL; 1191ee6e00c8SJens Axboe offset = *off_in; 11927995bd28SAl Viro } else { 11937995bd28SAl Viro offset = in->f_pos; 11947995bd28SAl Viro } 1195529565dcSIngo Molnar 1196ee5e0011SSlavomir Kaslev if (out->f_flags & O_NONBLOCK) 1197ee5e0011SSlavomir Kaslev flags |= SPLICE_F_NONBLOCK; 1198ee5e0011SSlavomir Kaslev 1199faa97c48SAl Viro ret = splice_file_to_pipe(in, opipe, &offset, len, flags); 1200983652c6SChung-Chiang Cheng 1201983652c6SChung-Chiang Cheng if (ret > 0) 1202983652c6SChung-Chiang Cheng fsnotify_access(in); 1203983652c6SChung-Chiang Cheng 12047995bd28SAl Viro if (!off_in) 12057995bd28SAl Viro in->f_pos = offset; 1206ee6e00c8SJens Axboe else 1207ee6e00c8SJens Axboe *off_in = offset; 1208a4514ebdSJens Axboe 1209a4514ebdSJens Axboe return ret; 1210529565dcSIngo Molnar } 12115274f052SJens Axboe 12125274f052SJens Axboe return -EINVAL; 12135274f052SJens Axboe } 12145274f052SJens Axboe 1215ee6e00c8SJens Axboe static long __do_splice(struct file *in, loff_t __user *off_in, 1216ee6e00c8SJens Axboe struct file *out, loff_t __user *off_out, 1217ee6e00c8SJens Axboe size_t len, unsigned int flags) 1218ee6e00c8SJens Axboe { 1219ee6e00c8SJens Axboe struct pipe_inode_info *ipipe; 1220ee6e00c8SJens Axboe struct pipe_inode_info *opipe; 1221ee6e00c8SJens Axboe loff_t offset, *__off_in = NULL, *__off_out = NULL; 1222ee6e00c8SJens Axboe long ret; 1223ee6e00c8SJens Axboe 1224ee6e00c8SJens Axboe ipipe = get_pipe_info(in, true); 1225ee6e00c8SJens Axboe opipe = get_pipe_info(out, true); 1226ee6e00c8SJens Axboe 12270f99fc51SJens Axboe if (ipipe) { 12280f99fc51SJens Axboe if (off_in) 1229ee6e00c8SJens Axboe return -ESPIPE; 12300f99fc51SJens Axboe pipe_clear_nowait(in); 12310f99fc51SJens Axboe } 12320f99fc51SJens Axboe if (opipe) { 12330f99fc51SJens Axboe if (off_out) 1234ee6e00c8SJens Axboe return -ESPIPE; 12350f99fc51SJens Axboe pipe_clear_nowait(out); 12360f99fc51SJens Axboe } 1237ee6e00c8SJens Axboe 1238ee6e00c8SJens Axboe if (off_out) { 1239ee6e00c8SJens Axboe if (copy_from_user(&offset, off_out, sizeof(loff_t))) 1240ee6e00c8SJens Axboe return -EFAULT; 1241ee6e00c8SJens Axboe __off_out = &offset; 1242ee6e00c8SJens Axboe } 1243ee6e00c8SJens Axboe if (off_in) { 1244ee6e00c8SJens Axboe if (copy_from_user(&offset, off_in, sizeof(loff_t))) 1245ee6e00c8SJens Axboe return -EFAULT; 1246ee6e00c8SJens Axboe __off_in = &offset; 1247ee6e00c8SJens Axboe } 1248ee6e00c8SJens Axboe 1249ee6e00c8SJens Axboe ret = do_splice(in, __off_in, out, __off_out, len, flags); 1250ee6e00c8SJens Axboe if (ret < 0) 1251ee6e00c8SJens Axboe return ret; 1252ee6e00c8SJens Axboe 1253ee6e00c8SJens Axboe if (__off_out && copy_to_user(off_out, __off_out, sizeof(loff_t))) 1254ee6e00c8SJens Axboe return -EFAULT; 1255ee6e00c8SJens Axboe if (__off_in && copy_to_user(off_in, __off_in, sizeof(loff_t))) 1256ee6e00c8SJens Axboe return -EFAULT; 1257ee6e00c8SJens Axboe 1258ee6e00c8SJens Axboe return ret; 1259ee6e00c8SJens Axboe } 1260ee6e00c8SJens Axboe 126179fddc4eSAl Viro static int iter_to_pipe(struct iov_iter *from, 126279fddc4eSAl Viro struct pipe_inode_info *pipe, 126379fddc4eSAl Viro unsigned flags) 1264912d35f8SJens Axboe { 126579fddc4eSAl Viro struct pipe_buffer buf = { 126679fddc4eSAl Viro .ops = &user_page_pipe_buf_ops, 126779fddc4eSAl Viro .flags = flags 126879fddc4eSAl Viro }; 126979fddc4eSAl Viro size_t total = 0; 127079fddc4eSAl Viro int ret = 0; 127179fddc4eSAl Viro 12727d690c15SAl Viro while (iov_iter_count(from)) { 127379fddc4eSAl Viro struct page *pages[16]; 12747d690c15SAl Viro ssize_t left; 1275db85a9ebSAl Viro size_t start; 12767d690c15SAl Viro int i, n; 1277912d35f8SJens Axboe 12787d690c15SAl Viro left = iov_iter_get_pages2(from, pages, ~0UL, 16, &start); 12797d690c15SAl Viro if (left <= 0) { 12807d690c15SAl Viro ret = left; 128179fddc4eSAl Viro break; 128279fddc4eSAl Viro } 1283912d35f8SJens Axboe 12847d690c15SAl Viro n = DIV_ROUND_UP(left + start, PAGE_SIZE); 12857d690c15SAl Viro for (i = 0; i < n; i++) { 12867d690c15SAl Viro int size = min_t(int, left, PAGE_SIZE - start); 12877d690c15SAl Viro 12887d690c15SAl Viro buf.page = pages[i]; 128979fddc4eSAl Viro buf.offset = start; 129079fddc4eSAl Viro buf.len = size; 129179fddc4eSAl Viro ret = add_to_pipe(pipe, &buf); 129279fddc4eSAl Viro if (unlikely(ret < 0)) { 12937d690c15SAl Viro iov_iter_revert(from, left); 12947d690c15SAl Viro // this one got dropped by add_to_pipe() 12957d690c15SAl Viro while (++i < n) 12967d690c15SAl Viro put_page(pages[i]); 12977d690c15SAl Viro goto out; 12987d690c15SAl Viro } 129979fddc4eSAl Viro total += ret; 13007d690c15SAl Viro left -= size; 13017d690c15SAl Viro start = 0; 1302912d35f8SJens Axboe } 1303912d35f8SJens Axboe } 13047d690c15SAl Viro out: 130579fddc4eSAl Viro return total ? total : ret; 1306912d35f8SJens Axboe } 1307912d35f8SJens Axboe 13086a14b90bSJens Axboe static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 13096a14b90bSJens Axboe struct splice_desc *sd) 13106a14b90bSJens Axboe { 13116130f531SAl Viro int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data); 13126130f531SAl Viro return n == sd->len ? n : -EFAULT; 13136a14b90bSJens Axboe } 13146a14b90bSJens Axboe 13156a14b90bSJens Axboe /* 13166a14b90bSJens Axboe * For lack of a better implementation, implement vmsplice() to userspace 13176a14b90bSJens Axboe * as a simple copy of the pipes pages to the user iov. 13186a14b90bSJens Axboe */ 131987a3002aSAl Viro static long vmsplice_to_user(struct file *file, struct iov_iter *iter, 132087a3002aSAl Viro unsigned int flags) 13216a14b90bSJens Axboe { 1322c73be61cSDavid Howells struct pipe_inode_info *pipe = get_pipe_info(file, true); 132387a3002aSAl Viro struct splice_desc sd = { 132487a3002aSAl Viro .total_len = iov_iter_count(iter), 132587a3002aSAl Viro .flags = flags, 132687a3002aSAl Viro .u.data = iter 132787a3002aSAl Viro }; 132887a3002aSAl Viro long ret = 0; 13296a14b90bSJens Axboe 13306a14b90bSJens Axboe if (!pipe) 13316a14b90bSJens Axboe return -EBADF; 13326a14b90bSJens Axboe 13330f99fc51SJens Axboe pipe_clear_nowait(file); 13340f99fc51SJens Axboe 1335345995faSAl Viro if (sd.total_len) { 13366130f531SAl Viro pipe_lock(pipe); 13376130f531SAl Viro ret = __splice_from_pipe(pipe, &sd, pipe_to_user); 133861e0d47cSMiklos Szeredi pipe_unlock(pipe); 1339345995faSAl Viro } 13406a14b90bSJens Axboe 13416a14b90bSJens Axboe return ret; 13426a14b90bSJens Axboe } 13436a14b90bSJens Axboe 1344912d35f8SJens Axboe /* 1345912d35f8SJens Axboe * vmsplice splices a user address range into a pipe. It can be thought of 1346912d35f8SJens Axboe * as splice-from-memory, where the regular splice is splice-from-file (or 1347912d35f8SJens Axboe * to file). In both cases the output is a pipe, naturally. 1348912d35f8SJens Axboe */ 134987a3002aSAl Viro static long vmsplice_to_pipe(struct file *file, struct iov_iter *iter, 135087a3002aSAl Viro unsigned int flags) 1351912d35f8SJens Axboe { 1352ddac0d39SJens Axboe struct pipe_inode_info *pipe; 135387a3002aSAl Viro long ret = 0; 135479fddc4eSAl Viro unsigned buf_flag = 0; 135579fddc4eSAl Viro 135679fddc4eSAl Viro if (flags & SPLICE_F_GIFT) 135779fddc4eSAl Viro buf_flag = PIPE_BUF_FLAG_GIFT; 1358912d35f8SJens Axboe 1359c73be61cSDavid Howells pipe = get_pipe_info(file, true); 1360ddac0d39SJens Axboe if (!pipe) 1361912d35f8SJens Axboe return -EBADF; 1362912d35f8SJens Axboe 13630f99fc51SJens Axboe pipe_clear_nowait(file); 13640f99fc51SJens Axboe 13658924feffSAl Viro pipe_lock(pipe); 13668924feffSAl Viro ret = wait_for_space(pipe, flags); 136779fddc4eSAl Viro if (!ret) 136887a3002aSAl Viro ret = iter_to_pipe(iter, pipe, buf_flag); 13698924feffSAl Viro pipe_unlock(pipe); 13708924feffSAl Viro if (ret > 0) 13718924feffSAl Viro wakeup_pipe_readers(pipe); 137235f3d14dSJens Axboe return ret; 1373912d35f8SJens Axboe } 1374912d35f8SJens Axboe 137587a3002aSAl Viro static int vmsplice_type(struct fd f, int *type) 137687a3002aSAl Viro { 137787a3002aSAl Viro if (!f.file) 137887a3002aSAl Viro return -EBADF; 137987a3002aSAl Viro if (f.file->f_mode & FMODE_WRITE) { 1380de4eda9dSAl Viro *type = ITER_SOURCE; 138187a3002aSAl Viro } else if (f.file->f_mode & FMODE_READ) { 1382de4eda9dSAl Viro *type = ITER_DEST; 138387a3002aSAl Viro } else { 138487a3002aSAl Viro fdput(f); 138587a3002aSAl Viro return -EBADF; 138687a3002aSAl Viro } 138787a3002aSAl Viro return 0; 138887a3002aSAl Viro } 138987a3002aSAl Viro 13906a14b90bSJens Axboe /* 13916a14b90bSJens Axboe * Note that vmsplice only really supports true splicing _from_ user memory 13926a14b90bSJens Axboe * to a pipe, not the other way around. Splicing from user memory is a simple 13936a14b90bSJens Axboe * operation that can be supported without any funky alignment restrictions 13946a14b90bSJens Axboe * or nasty vm tricks. We simply map in the user memory and fill them into 13956a14b90bSJens Axboe * a pipe. The reverse isn't quite as easy, though. There are two possible 13966a14b90bSJens Axboe * solutions for that: 13976a14b90bSJens Axboe * 13986a14b90bSJens Axboe * - memcpy() the data internally, at which point we might as well just 13996a14b90bSJens Axboe * do a regular read() on the buffer anyway. 14006a14b90bSJens Axboe * - Lots of nasty vm tricks, that are neither fast nor flexible (it 14016a14b90bSJens Axboe * has restriction limitations on both ends of the pipe). 14026a14b90bSJens Axboe * 14036a14b90bSJens Axboe * Currently we punt and implement it as a normal copy, see pipe_to_user(). 14046a14b90bSJens Axboe * 14056a14b90bSJens Axboe */ 140687a3002aSAl Viro SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, uiov, 140730cfe4efSDominik Brodowski unsigned long, nr_segs, unsigned int, flags) 140830cfe4efSDominik Brodowski { 140987a3002aSAl Viro struct iovec iovstack[UIO_FASTIOV]; 141087a3002aSAl Viro struct iovec *iov = iovstack; 141187a3002aSAl Viro struct iov_iter iter; 141287e5e6daSJens Axboe ssize_t error; 141387a3002aSAl Viro struct fd f; 141487a3002aSAl Viro int type; 141587a3002aSAl Viro 1416598b3cecSChristoph Hellwig if (unlikely(flags & ~SPLICE_F_ALL)) 1417598b3cecSChristoph Hellwig return -EINVAL; 1418598b3cecSChristoph Hellwig 141987a3002aSAl Viro f = fdget(fd); 142087a3002aSAl Viro error = vmsplice_type(f, &type); 142187a3002aSAl Viro if (error) 142287a3002aSAl Viro return error; 142387a3002aSAl Viro 142487a3002aSAl Viro error = import_iovec(type, uiov, nr_segs, 142587a3002aSAl Viro ARRAY_SIZE(iovstack), &iov, &iter); 1426598b3cecSChristoph Hellwig if (error < 0) 1427598b3cecSChristoph Hellwig goto out_fdput; 1428598b3cecSChristoph Hellwig 1429598b3cecSChristoph Hellwig if (!iov_iter_count(&iter)) 1430598b3cecSChristoph Hellwig error = 0; 1431de4eda9dSAl Viro else if (type == ITER_SOURCE) 1432598b3cecSChristoph Hellwig error = vmsplice_to_pipe(f.file, &iter, flags); 1433598b3cecSChristoph Hellwig else 1434598b3cecSChristoph Hellwig error = vmsplice_to_user(f.file, &iter, flags); 1435598b3cecSChristoph Hellwig 143687a3002aSAl Viro kfree(iov); 1437598b3cecSChristoph Hellwig out_fdput: 143887a3002aSAl Viro fdput(f); 143987a3002aSAl Viro return error; 144030cfe4efSDominik Brodowski } 144130cfe4efSDominik Brodowski 1442836f92adSHeiko Carstens SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, 1443836f92adSHeiko Carstens int, fd_out, loff_t __user *, off_out, 1444836f92adSHeiko Carstens size_t, len, unsigned int, flags) 14455274f052SJens Axboe { 14462903ff01SAl Viro struct fd in, out; 14475274f052SJens Axboe long error; 14485274f052SJens Axboe 14495274f052SJens Axboe if (unlikely(!len)) 14505274f052SJens Axboe return 0; 14515274f052SJens Axboe 14523d6ea290SAl Viro if (unlikely(flags & ~SPLICE_F_ALL)) 14533d6ea290SAl Viro return -EINVAL; 14543d6ea290SAl Viro 14555274f052SJens Axboe error = -EBADF; 14562903ff01SAl Viro in = fdget(fd_in); 14572903ff01SAl Viro if (in.file) { 14582903ff01SAl Viro out = fdget(fd_out); 14592903ff01SAl Viro if (out.file) { 1460ee6e00c8SJens Axboe error = __do_splice(in.file, off_in, out.file, off_out, 1461529565dcSIngo Molnar len, flags); 14622903ff01SAl Viro fdput(out); 14635274f052SJens Axboe } 14642903ff01SAl Viro fdput(in); 14655274f052SJens Axboe } 14665274f052SJens Axboe return error; 14675274f052SJens Axboe } 146870524490SJens Axboe 146970524490SJens Axboe /* 1470aadd06e5SJens Axboe * Make sure there's data to read. Wait for input if we can, otherwise 1471aadd06e5SJens Axboe * return an appropriate error. 1472aadd06e5SJens Axboe */ 14737c77f0b3SMiklos Szeredi static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1474aadd06e5SJens Axboe { 1475aadd06e5SJens Axboe int ret; 1476aadd06e5SJens Axboe 1477aadd06e5SJens Axboe /* 14788cefc107SDavid Howells * Check the pipe occupancy without the inode lock first. This function 1479aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1480aadd06e5SJens Axboe */ 14818cefc107SDavid Howells if (!pipe_empty(pipe->head, pipe->tail)) 1482aadd06e5SJens Axboe return 0; 1483aadd06e5SJens Axboe 1484aadd06e5SJens Axboe ret = 0; 148561e0d47cSMiklos Szeredi pipe_lock(pipe); 1486aadd06e5SJens Axboe 14878cefc107SDavid Howells while (pipe_empty(pipe->head, pipe->tail)) { 1488aadd06e5SJens Axboe if (signal_pending(current)) { 1489aadd06e5SJens Axboe ret = -ERESTARTSYS; 1490aadd06e5SJens Axboe break; 1491aadd06e5SJens Axboe } 1492aadd06e5SJens Axboe if (!pipe->writers) 1493aadd06e5SJens Axboe break; 1494aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1495aadd06e5SJens Axboe ret = -EAGAIN; 1496aadd06e5SJens Axboe break; 1497aadd06e5SJens Axboe } 1498472e5b05SLinus Torvalds pipe_wait_readable(pipe); 1499aadd06e5SJens Axboe } 1500aadd06e5SJens Axboe 150161e0d47cSMiklos Szeredi pipe_unlock(pipe); 1502aadd06e5SJens Axboe return ret; 1503aadd06e5SJens Axboe } 1504aadd06e5SJens Axboe 1505aadd06e5SJens Axboe /* 1506aadd06e5SJens Axboe * Make sure there's writeable room. Wait for room if we can, otherwise 1507aadd06e5SJens Axboe * return an appropriate error. 1508aadd06e5SJens Axboe */ 15097c77f0b3SMiklos Szeredi static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1510aadd06e5SJens Axboe { 1511aadd06e5SJens Axboe int ret; 1512aadd06e5SJens Axboe 1513aadd06e5SJens Axboe /* 15148cefc107SDavid Howells * Check pipe occupancy without the inode lock first. This function 1515aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1516aadd06e5SJens Axboe */ 1517566d1362STetsuo Handa if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) 1518aadd06e5SJens Axboe return 0; 1519aadd06e5SJens Axboe 1520aadd06e5SJens Axboe ret = 0; 152161e0d47cSMiklos Szeredi pipe_lock(pipe); 1522aadd06e5SJens Axboe 15236718b6f8SDavid Howells while (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) { 1524aadd06e5SJens Axboe if (!pipe->readers) { 1525aadd06e5SJens Axboe send_sig(SIGPIPE, current, 0); 1526aadd06e5SJens Axboe ret = -EPIPE; 1527aadd06e5SJens Axboe break; 1528aadd06e5SJens Axboe } 1529aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1530aadd06e5SJens Axboe ret = -EAGAIN; 1531aadd06e5SJens Axboe break; 1532aadd06e5SJens Axboe } 1533aadd06e5SJens Axboe if (signal_pending(current)) { 1534aadd06e5SJens Axboe ret = -ERESTARTSYS; 1535aadd06e5SJens Axboe break; 1536aadd06e5SJens Axboe } 1537472e5b05SLinus Torvalds pipe_wait_writable(pipe); 1538aadd06e5SJens Axboe } 1539aadd06e5SJens Axboe 154061e0d47cSMiklos Szeredi pipe_unlock(pipe); 1541aadd06e5SJens Axboe return ret; 1542aadd06e5SJens Axboe } 1543aadd06e5SJens Axboe 1544aadd06e5SJens Axboe /* 15457c77f0b3SMiklos Szeredi * Splice contents of ipipe to opipe. 15467c77f0b3SMiklos Szeredi */ 15477c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, 15487c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe, 15497c77f0b3SMiklos Szeredi size_t len, unsigned int flags) 15507c77f0b3SMiklos Szeredi { 15517c77f0b3SMiklos Szeredi struct pipe_buffer *ibuf, *obuf; 15528cefc107SDavid Howells unsigned int i_head, o_head; 15538cefc107SDavid Howells unsigned int i_tail, o_tail; 15548cefc107SDavid Howells unsigned int i_mask, o_mask; 15558cefc107SDavid Howells int ret = 0; 15567c77f0b3SMiklos Szeredi bool input_wakeup = false; 15577c77f0b3SMiklos Szeredi 15587c77f0b3SMiklos Szeredi 15597c77f0b3SMiklos Szeredi retry: 15607c77f0b3SMiklos Szeredi ret = ipipe_prep(ipipe, flags); 15617c77f0b3SMiklos Szeredi if (ret) 15627c77f0b3SMiklos Szeredi return ret; 15637c77f0b3SMiklos Szeredi 15647c77f0b3SMiklos Szeredi ret = opipe_prep(opipe, flags); 15657c77f0b3SMiklos Szeredi if (ret) 15667c77f0b3SMiklos Szeredi return ret; 15677c77f0b3SMiklos Szeredi 15687c77f0b3SMiklos Szeredi /* 15697c77f0b3SMiklos Szeredi * Potential ABBA deadlock, work around it by ordering lock 15707c77f0b3SMiklos Szeredi * grabbing by pipe info address. Otherwise two different processes 15717c77f0b3SMiklos Szeredi * could deadlock (one doing tee from A -> B, the other from B -> A). 15727c77f0b3SMiklos Szeredi */ 15737c77f0b3SMiklos Szeredi pipe_double_lock(ipipe, opipe); 15747c77f0b3SMiklos Szeredi 15758cefc107SDavid Howells i_tail = ipipe->tail; 15768cefc107SDavid Howells i_mask = ipipe->ring_size - 1; 15778cefc107SDavid Howells o_head = opipe->head; 15788cefc107SDavid Howells o_mask = opipe->ring_size - 1; 15798cefc107SDavid Howells 15807c77f0b3SMiklos Szeredi do { 15818cefc107SDavid Howells size_t o_len; 15828cefc107SDavid Howells 15837c77f0b3SMiklos Szeredi if (!opipe->readers) { 15847c77f0b3SMiklos Szeredi send_sig(SIGPIPE, current, 0); 15857c77f0b3SMiklos Szeredi if (!ret) 15867c77f0b3SMiklos Szeredi ret = -EPIPE; 15877c77f0b3SMiklos Szeredi break; 15887c77f0b3SMiklos Szeredi } 15897c77f0b3SMiklos Szeredi 15908cefc107SDavid Howells i_head = ipipe->head; 15918cefc107SDavid Howells o_tail = opipe->tail; 15928cefc107SDavid Howells 15938cefc107SDavid Howells if (pipe_empty(i_head, i_tail) && !ipipe->writers) 15947c77f0b3SMiklos Szeredi break; 15957c77f0b3SMiklos Szeredi 15967c77f0b3SMiklos Szeredi /* 15977c77f0b3SMiklos Szeredi * Cannot make any progress, because either the input 15987c77f0b3SMiklos Szeredi * pipe is empty or the output pipe is full. 15997c77f0b3SMiklos Szeredi */ 16008cefc107SDavid Howells if (pipe_empty(i_head, i_tail) || 16016718b6f8SDavid Howells pipe_full(o_head, o_tail, opipe->max_usage)) { 16027c77f0b3SMiklos Szeredi /* Already processed some buffers, break */ 16037c77f0b3SMiklos Szeredi if (ret) 16047c77f0b3SMiklos Szeredi break; 16057c77f0b3SMiklos Szeredi 16067c77f0b3SMiklos Szeredi if (flags & SPLICE_F_NONBLOCK) { 16077c77f0b3SMiklos Szeredi ret = -EAGAIN; 16087c77f0b3SMiklos Szeredi break; 16097c77f0b3SMiklos Szeredi } 16107c77f0b3SMiklos Szeredi 16117c77f0b3SMiklos Szeredi /* 16127c77f0b3SMiklos Szeredi * We raced with another reader/writer and haven't 16137c77f0b3SMiklos Szeredi * managed to process any buffers. A zero return 16147c77f0b3SMiklos Szeredi * value means EOF, so retry instead. 16157c77f0b3SMiklos Szeredi */ 16167c77f0b3SMiklos Szeredi pipe_unlock(ipipe); 16177c77f0b3SMiklos Szeredi pipe_unlock(opipe); 16187c77f0b3SMiklos Szeredi goto retry; 16197c77f0b3SMiklos Szeredi } 16207c77f0b3SMiklos Szeredi 16218cefc107SDavid Howells ibuf = &ipipe->bufs[i_tail & i_mask]; 16228cefc107SDavid Howells obuf = &opipe->bufs[o_head & o_mask]; 16237c77f0b3SMiklos Szeredi 16247c77f0b3SMiklos Szeredi if (len >= ibuf->len) { 16257c77f0b3SMiklos Szeredi /* 16267c77f0b3SMiklos Szeredi * Simply move the whole buffer from ipipe to opipe 16277c77f0b3SMiklos Szeredi */ 16287c77f0b3SMiklos Szeredi *obuf = *ibuf; 16297c77f0b3SMiklos Szeredi ibuf->ops = NULL; 16308cefc107SDavid Howells i_tail++; 16318cefc107SDavid Howells ipipe->tail = i_tail; 16327c77f0b3SMiklos Szeredi input_wakeup = true; 16338cefc107SDavid Howells o_len = obuf->len; 16348cefc107SDavid Howells o_head++; 16358cefc107SDavid Howells opipe->head = o_head; 16367c77f0b3SMiklos Szeredi } else { 16377c77f0b3SMiklos Szeredi /* 16387c77f0b3SMiklos Szeredi * Get a reference to this pipe buffer, 16397c77f0b3SMiklos Szeredi * so we can copy the contents over. 16407c77f0b3SMiklos Szeredi */ 164115fab63eSMatthew Wilcox if (!pipe_buf_get(ipipe, ibuf)) { 164215fab63eSMatthew Wilcox if (ret == 0) 164315fab63eSMatthew Wilcox ret = -EFAULT; 164415fab63eSMatthew Wilcox break; 164515fab63eSMatthew Wilcox } 16467c77f0b3SMiklos Szeredi *obuf = *ibuf; 16477c77f0b3SMiklos Szeredi 16487c77f0b3SMiklos Szeredi /* 1649f6dd9755SChristoph Hellwig * Don't inherit the gift and merge flags, we need to 16507c77f0b3SMiklos Szeredi * prevent multiple steals of this page. 16517c77f0b3SMiklos Szeredi */ 16527c77f0b3SMiklos Szeredi obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 1653f6dd9755SChristoph Hellwig obuf->flags &= ~PIPE_BUF_FLAG_CAN_MERGE; 1654a0ce2f0aSJann Horn 16557c77f0b3SMiklos Szeredi obuf->len = len; 16568cefc107SDavid Howells ibuf->offset += len; 16578cefc107SDavid Howells ibuf->len -= len; 16588cefc107SDavid Howells o_len = len; 16598cefc107SDavid Howells o_head++; 16608cefc107SDavid Howells opipe->head = o_head; 16617c77f0b3SMiklos Szeredi } 16628cefc107SDavid Howells ret += o_len; 16638cefc107SDavid Howells len -= o_len; 16647c77f0b3SMiklos Szeredi } while (len); 16657c77f0b3SMiklos Szeredi 16667c77f0b3SMiklos Szeredi pipe_unlock(ipipe); 16677c77f0b3SMiklos Szeredi pipe_unlock(opipe); 16687c77f0b3SMiklos Szeredi 16697c77f0b3SMiklos Szeredi /* 16707c77f0b3SMiklos Szeredi * If we put data in the output pipe, wakeup any potential readers. 16717c77f0b3SMiklos Szeredi */ 1672825cdcb1SNamhyung Kim if (ret > 0) 1673825cdcb1SNamhyung Kim wakeup_pipe_readers(opipe); 1674825cdcb1SNamhyung Kim 16757c77f0b3SMiklos Szeredi if (input_wakeup) 16767c77f0b3SMiklos Szeredi wakeup_pipe_writers(ipipe); 16777c77f0b3SMiklos Szeredi 16787c77f0b3SMiklos Szeredi return ret; 16797c77f0b3SMiklos Szeredi } 16807c77f0b3SMiklos Szeredi 16817c77f0b3SMiklos Szeredi /* 168270524490SJens Axboe * Link contents of ipipe to opipe. 168370524490SJens Axboe */ 168470524490SJens Axboe static int link_pipe(struct pipe_inode_info *ipipe, 168570524490SJens Axboe struct pipe_inode_info *opipe, 168670524490SJens Axboe size_t len, unsigned int flags) 168770524490SJens Axboe { 168870524490SJens Axboe struct pipe_buffer *ibuf, *obuf; 16898cefc107SDavid Howells unsigned int i_head, o_head; 16908cefc107SDavid Howells unsigned int i_tail, o_tail; 16918cefc107SDavid Howells unsigned int i_mask, o_mask; 16928cefc107SDavid Howells int ret = 0; 169370524490SJens Axboe 169470524490SJens Axboe /* 169570524490SJens Axboe * Potential ABBA deadlock, work around it by ordering lock 169661e0d47cSMiklos Szeredi * grabbing by pipe info address. Otherwise two different processes 169770524490SJens Axboe * could deadlock (one doing tee from A -> B, the other from B -> A). 169870524490SJens Axboe */ 169961e0d47cSMiklos Szeredi pipe_double_lock(ipipe, opipe); 170070524490SJens Axboe 17018cefc107SDavid Howells i_tail = ipipe->tail; 17028cefc107SDavid Howells i_mask = ipipe->ring_size - 1; 17038cefc107SDavid Howells o_head = opipe->head; 17048cefc107SDavid Howells o_mask = opipe->ring_size - 1; 17058cefc107SDavid Howells 1706aadd06e5SJens Axboe do { 170770524490SJens Axboe if (!opipe->readers) { 170870524490SJens Axboe send_sig(SIGPIPE, current, 0); 170970524490SJens Axboe if (!ret) 171070524490SJens Axboe ret = -EPIPE; 171170524490SJens Axboe break; 171270524490SJens Axboe } 171370524490SJens Axboe 17148cefc107SDavid Howells i_head = ipipe->head; 17158cefc107SDavid Howells o_tail = opipe->tail; 17168cefc107SDavid Howells 171770524490SJens Axboe /* 17188cefc107SDavid Howells * If we have iterated all input buffers or run out of 1719aadd06e5SJens Axboe * output room, break. 172070524490SJens Axboe */ 17218cefc107SDavid Howells if (pipe_empty(i_head, i_tail) || 17226718b6f8SDavid Howells pipe_full(o_head, o_tail, opipe->max_usage)) 1723aadd06e5SJens Axboe break; 1724aadd06e5SJens Axboe 17258cefc107SDavid Howells ibuf = &ipipe->bufs[i_tail & i_mask]; 17268cefc107SDavid Howells obuf = &opipe->bufs[o_head & o_mask]; 172770524490SJens Axboe 172870524490SJens Axboe /* 172970524490SJens Axboe * Get a reference to this pipe buffer, 173070524490SJens Axboe * so we can copy the contents over. 173170524490SJens Axboe */ 173215fab63eSMatthew Wilcox if (!pipe_buf_get(ipipe, ibuf)) { 173315fab63eSMatthew Wilcox if (ret == 0) 173415fab63eSMatthew Wilcox ret = -EFAULT; 173515fab63eSMatthew Wilcox break; 173615fab63eSMatthew Wilcox } 173770524490SJens Axboe 173870524490SJens Axboe *obuf = *ibuf; 173970524490SJens Axboe 17407afa6fd0SJens Axboe /* 1741f6dd9755SChristoph Hellwig * Don't inherit the gift and merge flag, we need to prevent 1742f6dd9755SChristoph Hellwig * multiple steals of this page. 17437afa6fd0SJens Axboe */ 17447afa6fd0SJens Axboe obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 1745f6dd9755SChristoph Hellwig obuf->flags &= ~PIPE_BUF_FLAG_CAN_MERGE; 1746a0ce2f0aSJann Horn 174770524490SJens Axboe if (obuf->len > len) 174870524490SJens Axboe obuf->len = len; 174970524490SJens Axboe ret += obuf->len; 175070524490SJens Axboe len -= obuf->len; 17518cefc107SDavid Howells 17528cefc107SDavid Howells o_head++; 17538cefc107SDavid Howells opipe->head = o_head; 17548cefc107SDavid Howells i_tail++; 1755aadd06e5SJens Axboe } while (len); 175670524490SJens Axboe 175761e0d47cSMiklos Szeredi pipe_unlock(ipipe); 175861e0d47cSMiklos Szeredi pipe_unlock(opipe); 175970524490SJens Axboe 1760aadd06e5SJens Axboe /* 1761aadd06e5SJens Axboe * If we put data in the output pipe, wakeup any potential readers. 1762aadd06e5SJens Axboe */ 1763825cdcb1SNamhyung Kim if (ret > 0) 1764825cdcb1SNamhyung Kim wakeup_pipe_readers(opipe); 176570524490SJens Axboe 176670524490SJens Axboe return ret; 176770524490SJens Axboe } 176870524490SJens Axboe 176970524490SJens Axboe /* 177070524490SJens Axboe * This is a tee(1) implementation that works on pipes. It doesn't copy 177170524490SJens Axboe * any data, it simply references the 'in' pages on the 'out' pipe. 177270524490SJens Axboe * The 'flags' used are the SPLICE_F_* variants, currently the only 177370524490SJens Axboe * applicable one is SPLICE_F_NONBLOCK. 177470524490SJens Axboe */ 17759dafdfc2SPavel Begunkov long do_tee(struct file *in, struct file *out, size_t len, unsigned int flags) 177670524490SJens Axboe { 1777c73be61cSDavid Howells struct pipe_inode_info *ipipe = get_pipe_info(in, true); 1778c73be61cSDavid Howells struct pipe_inode_info *opipe = get_pipe_info(out, true); 1779aadd06e5SJens Axboe int ret = -EINVAL; 178070524490SJens Axboe 178190da2e3fSPavel Begunkov if (unlikely(!(in->f_mode & FMODE_READ) || 178290da2e3fSPavel Begunkov !(out->f_mode & FMODE_WRITE))) 178390da2e3fSPavel Begunkov return -EBADF; 178490da2e3fSPavel Begunkov 178570524490SJens Axboe /* 1786aadd06e5SJens Axboe * Duplicate the contents of ipipe to opipe without actually 1787aadd06e5SJens Axboe * copying the data. 178870524490SJens Axboe */ 1789aadd06e5SJens Axboe if (ipipe && opipe && ipipe != opipe) { 1790ee5e0011SSlavomir Kaslev if ((in->f_flags | out->f_flags) & O_NONBLOCK) 1791ee5e0011SSlavomir Kaslev flags |= SPLICE_F_NONBLOCK; 1792ee5e0011SSlavomir Kaslev 1793aadd06e5SJens Axboe /* 1794aadd06e5SJens Axboe * Keep going, unless we encounter an error. The ipipe/opipe 1795aadd06e5SJens Axboe * ordering doesn't really matter. 1796aadd06e5SJens Axboe */ 17977c77f0b3SMiklos Szeredi ret = ipipe_prep(ipipe, flags); 1798aadd06e5SJens Axboe if (!ret) { 17997c77f0b3SMiklos Szeredi ret = opipe_prep(opipe, flags); 180002cf01aeSJens Axboe if (!ret) 1801aadd06e5SJens Axboe ret = link_pipe(ipipe, opipe, len, flags); 1802aadd06e5SJens Axboe } 1803aadd06e5SJens Axboe } 180470524490SJens Axboe 1805aadd06e5SJens Axboe return ret; 180670524490SJens Axboe } 180770524490SJens Axboe 1808836f92adSHeiko Carstens SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) 180970524490SJens Axboe { 181090da2e3fSPavel Begunkov struct fd in, out; 18112903ff01SAl Viro int error; 181270524490SJens Axboe 18133d6ea290SAl Viro if (unlikely(flags & ~SPLICE_F_ALL)) 18143d6ea290SAl Viro return -EINVAL; 18153d6ea290SAl Viro 181670524490SJens Axboe if (unlikely(!len)) 181770524490SJens Axboe return 0; 181870524490SJens Axboe 181970524490SJens Axboe error = -EBADF; 18202903ff01SAl Viro in = fdget(fdin); 18212903ff01SAl Viro if (in.file) { 182290da2e3fSPavel Begunkov out = fdget(fdout); 18232903ff01SAl Viro if (out.file) { 182490da2e3fSPavel Begunkov error = do_tee(in.file, out.file, len, flags); 18252903ff01SAl Viro fdput(out); 182670524490SJens Axboe } 18272903ff01SAl Viro fdput(in); 182870524490SJens Axboe } 182970524490SJens Axboe 183070524490SJens Axboe return error; 183170524490SJens Axboe } 1832