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> 28630d9c47SPaul Gortmaker #include <linux/export.h> 294f6f0bd2SJens Axboe #include <linux/syscalls.h> 30912d35f8SJens Axboe #include <linux/uio.h> 3129ce2058SJames Morris #include <linux/security.h> 325a0e3ad6STejun Heo #include <linux/gfp.h> 3335f9c09fSEric Dumazet #include <linux/socket.h> 3476b021d0SAl Viro #include <linux/compat.h> 3506ae43f3SAl Viro #include "internal.h" 365274f052SJens Axboe 3783f9135bSJens Axboe /* 3883f9135bSJens Axboe * Attempt to steal a page from a pipe buffer. This should perhaps go into 3983f9135bSJens Axboe * a vm helper function, it's already simplified quite a bit by the 4083f9135bSJens Axboe * addition of remove_mapping(). If success is returned, the caller may 4183f9135bSJens Axboe * attempt to reuse this page for another destination. 4283f9135bSJens Axboe */ 4376ad4d11SJens Axboe static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe, 445abc97aaSJens Axboe struct pipe_buffer *buf) 455abc97aaSJens Axboe { 465abc97aaSJens Axboe struct page *page = buf->page; 479e94cd4fSJens Axboe struct address_space *mapping; 485abc97aaSJens Axboe 499e0267c2SJens Axboe lock_page(page); 509e0267c2SJens Axboe 519e94cd4fSJens Axboe mapping = page_mapping(page); 529e94cd4fSJens Axboe if (mapping) { 535abc97aaSJens Axboe WARN_ON(!PageUptodate(page)); 545abc97aaSJens Axboe 55ad8d6f0aSJens Axboe /* 569e94cd4fSJens Axboe * At least for ext2 with nobh option, we need to wait on 579e94cd4fSJens Axboe * writeback completing on this page, since we'll remove it 589e94cd4fSJens Axboe * from the pagecache. Otherwise truncate wont wait on the 599e94cd4fSJens Axboe * page, allowing the disk blocks to be reused by someone else 609e94cd4fSJens Axboe * before we actually wrote our data to them. fs corruption 619e94cd4fSJens Axboe * ensues. 62ad8d6f0aSJens Axboe */ 63ad8d6f0aSJens Axboe wait_on_page_writeback(page); 64ad8d6f0aSJens Axboe 65266cf658SDavid Howells if (page_has_private(page) && 66266cf658SDavid Howells !try_to_release_page(page, GFP_KERNEL)) 67ca39d651SJens Axboe goto out_unlock; 684f6f0bd2SJens Axboe 699e94cd4fSJens Axboe /* 709e94cd4fSJens Axboe * If we succeeded in removing the mapping, set LRU flag 719e94cd4fSJens Axboe * and return good. 729e94cd4fSJens Axboe */ 739e94cd4fSJens Axboe if (remove_mapping(mapping, page)) { 741432873aSJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 755abc97aaSJens Axboe return 0; 765abc97aaSJens Axboe } 779e94cd4fSJens Axboe } 789e94cd4fSJens Axboe 799e94cd4fSJens Axboe /* 809e94cd4fSJens Axboe * Raced with truncate or failed to remove page from current 819e94cd4fSJens Axboe * address space, unlock and return failure. 829e94cd4fSJens Axboe */ 83ca39d651SJens Axboe out_unlock: 849e94cd4fSJens Axboe unlock_page(page); 859e94cd4fSJens Axboe return 1; 869e94cd4fSJens Axboe } 875abc97aaSJens Axboe 8876ad4d11SJens Axboe static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, 895274f052SJens Axboe struct pipe_buffer *buf) 905274f052SJens Axboe { 9109cbfeafSKirill A. Shutemov put_page(buf->page); 921432873aSJens Axboe buf->flags &= ~PIPE_BUF_FLAG_LRU; 935274f052SJens Axboe } 945274f052SJens Axboe 950845718dSJens Axboe /* 960845718dSJens Axboe * Check whether the contents of buf is OK to access. Since the content 970845718dSJens Axboe * is a page cache page, IO may be in flight. 980845718dSJens Axboe */ 99cac36bb0SJens Axboe static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, 1005274f052SJens Axboe struct pipe_buffer *buf) 1015274f052SJens Axboe { 1025274f052SJens Axboe struct page *page = buf->page; 10349d0b21bSJens Axboe int err; 1045274f052SJens Axboe 1055274f052SJens Axboe if (!PageUptodate(page)) { 10649d0b21bSJens Axboe lock_page(page); 1075274f052SJens Axboe 10849d0b21bSJens Axboe /* 10949d0b21bSJens Axboe * Page got truncated/unhashed. This will cause a 0-byte 11073d62d83SIngo Molnar * splice, if this is the first page. 11149d0b21bSJens Axboe */ 1125274f052SJens Axboe if (!page->mapping) { 11349d0b21bSJens Axboe err = -ENODATA; 11449d0b21bSJens Axboe goto error; 1155274f052SJens Axboe } 1165274f052SJens Axboe 11749d0b21bSJens Axboe /* 11873d62d83SIngo Molnar * Uh oh, read-error from disk. 11949d0b21bSJens Axboe */ 12049d0b21bSJens Axboe if (!PageUptodate(page)) { 12149d0b21bSJens Axboe err = -EIO; 12249d0b21bSJens Axboe goto error; 12349d0b21bSJens Axboe } 12449d0b21bSJens Axboe 12549d0b21bSJens Axboe /* 126f84d7519SJens Axboe * Page is ok afterall, we are done. 12749d0b21bSJens Axboe */ 12849d0b21bSJens Axboe unlock_page(page); 12949d0b21bSJens Axboe } 13049d0b21bSJens Axboe 131f84d7519SJens Axboe return 0; 13249d0b21bSJens Axboe error: 13349d0b21bSJens Axboe unlock_page(page); 134f84d7519SJens Axboe return err; 13570524490SJens Axboe } 13670524490SJens Axboe 137708e3508SHugh Dickins const struct pipe_buf_operations page_cache_pipe_buf_ops = { 1385274f052SJens Axboe .can_merge = 0, 139cac36bb0SJens Axboe .confirm = page_cache_pipe_buf_confirm, 1405274f052SJens Axboe .release = page_cache_pipe_buf_release, 1415abc97aaSJens Axboe .steal = page_cache_pipe_buf_steal, 142f84d7519SJens Axboe .get = generic_pipe_buf_get, 1435274f052SJens Axboe }; 1445274f052SJens Axboe 145912d35f8SJens Axboe static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, 146912d35f8SJens Axboe struct pipe_buffer *buf) 147912d35f8SJens Axboe { 1487afa6fd0SJens Axboe if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) 149912d35f8SJens Axboe return 1; 1507afa6fd0SJens Axboe 1511432873aSJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 152330ab716SJens Axboe return generic_pipe_buf_steal(pipe, buf); 153912d35f8SJens Axboe } 154912d35f8SJens Axboe 155d4c3cca9SEric Dumazet static const struct pipe_buf_operations user_page_pipe_buf_ops = { 156912d35f8SJens Axboe .can_merge = 0, 157cac36bb0SJens Axboe .confirm = generic_pipe_buf_confirm, 158912d35f8SJens Axboe .release = page_cache_pipe_buf_release, 159912d35f8SJens Axboe .steal = user_page_pipe_buf_steal, 160f84d7519SJens Axboe .get = generic_pipe_buf_get, 161912d35f8SJens Axboe }; 162912d35f8SJens Axboe 163825cdcb1SNamhyung Kim static void wakeup_pipe_readers(struct pipe_inode_info *pipe) 164825cdcb1SNamhyung Kim { 165825cdcb1SNamhyung Kim smp_mb(); 166825cdcb1SNamhyung Kim if (waitqueue_active(&pipe->wait)) 167825cdcb1SNamhyung Kim wake_up_interruptible(&pipe->wait); 168825cdcb1SNamhyung Kim kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); 169825cdcb1SNamhyung Kim } 170825cdcb1SNamhyung Kim 171932cc6d4SJens Axboe /** 172932cc6d4SJens Axboe * splice_to_pipe - fill passed data into a pipe 173932cc6d4SJens Axboe * @pipe: pipe to fill 174932cc6d4SJens Axboe * @spd: data to fill 175932cc6d4SJens Axboe * 176932cc6d4SJens Axboe * Description: 17779685b8dSRandy Dunlap * @spd contains a map of pages and len/offset tuples, along with 178932cc6d4SJens Axboe * the struct pipe_buf_operations associated with these pages. This 179932cc6d4SJens Axboe * function will link that data to the pipe. 180932cc6d4SJens Axboe * 18183f9135bSJens Axboe */ 182d6b29d7cSJens Axboe ssize_t splice_to_pipe(struct pipe_inode_info *pipe, 183912d35f8SJens Axboe struct splice_pipe_desc *spd) 1845274f052SJens Axboe { 18500de00bdSJens Axboe unsigned int spd_pages = spd->nr_pages; 1868924feffSAl Viro int ret = 0, page_nr = 0; 1875274f052SJens Axboe 188d6785d91SRabin Vincent if (!spd_pages) 189d6785d91SRabin Vincent return 0; 190d6785d91SRabin Vincent 1918924feffSAl Viro if (unlikely(!pipe->readers)) { 1925274f052SJens Axboe send_sig(SIGPIPE, current, 0); 1935274f052SJens Axboe ret = -EPIPE; 1948924feffSAl Viro goto out; 1955274f052SJens Axboe } 1965274f052SJens Axboe 1978924feffSAl Viro while (pipe->nrbufs < pipe->buffers) { 19835f3d14dSJens Axboe int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); 1993a326a2cSIngo Molnar struct pipe_buffer *buf = pipe->bufs + newbuf; 2005274f052SJens Axboe 201912d35f8SJens Axboe buf->page = spd->pages[page_nr]; 202912d35f8SJens Axboe buf->offset = spd->partial[page_nr].offset; 203912d35f8SJens Axboe buf->len = spd->partial[page_nr].len; 204497f9625SJens Axboe buf->private = spd->partial[page_nr].private; 205912d35f8SJens Axboe buf->ops = spd->ops; 2067afa6fd0SJens Axboe 2076f767b04SJens Axboe pipe->nrbufs++; 208912d35f8SJens Axboe page_nr++; 209912d35f8SJens Axboe ret += buf->len; 210912d35f8SJens Axboe 211912d35f8SJens Axboe if (!--spd->nr_pages) 2125274f052SJens Axboe break; 2135274f052SJens Axboe } 2145274f052SJens Axboe 21529e35094SLinus Torvalds if (!ret) 21629e35094SLinus Torvalds ret = -EAGAIN; 21729e35094SLinus Torvalds 2188924feffSAl Viro out: 21900de00bdSJens Axboe while (page_nr < spd_pages) 220bbdfc2f7SJens Axboe spd->spd_release(spd, page_nr++); 2215274f052SJens Axboe 2225274f052SJens Axboe return ret; 2235274f052SJens Axboe } 2242b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(splice_to_pipe); 2255274f052SJens Axboe 22679fddc4eSAl Viro ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf) 22779fddc4eSAl Viro { 22879fddc4eSAl Viro int ret; 22979fddc4eSAl Viro 23079fddc4eSAl Viro if (unlikely(!pipe->readers)) { 23179fddc4eSAl Viro send_sig(SIGPIPE, current, 0); 23279fddc4eSAl Viro ret = -EPIPE; 23379fddc4eSAl Viro } else if (pipe->nrbufs == pipe->buffers) { 23479fddc4eSAl Viro ret = -EAGAIN; 23579fddc4eSAl Viro } else { 23679fddc4eSAl Viro int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); 23779fddc4eSAl Viro pipe->bufs[newbuf] = *buf; 23879fddc4eSAl Viro pipe->nrbufs++; 23979fddc4eSAl Viro return buf->len; 24079fddc4eSAl Viro } 24179fddc4eSAl Viro buf->ops->release(pipe, buf); 24279fddc4eSAl Viro buf->ops = NULL; 24379fddc4eSAl Viro return ret; 24479fddc4eSAl Viro } 24579fddc4eSAl Viro EXPORT_SYMBOL(add_to_pipe); 24679fddc4eSAl Viro 247708e3508SHugh Dickins void spd_release_page(struct splice_pipe_desc *spd, unsigned int i) 248bbdfc2f7SJens Axboe { 24909cbfeafSKirill A. Shutemov put_page(spd->pages[i]); 250bbdfc2f7SJens Axboe } 251bbdfc2f7SJens Axboe 25235f3d14dSJens Axboe /* 25335f3d14dSJens Axboe * Check if we need to grow the arrays holding pages and partial page 25435f3d14dSJens Axboe * descriptions. 25535f3d14dSJens Axboe */ 256047fe360SEric Dumazet int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd) 25735f3d14dSJens Axboe { 258047fe360SEric Dumazet unsigned int buffers = ACCESS_ONCE(pipe->buffers); 259047fe360SEric Dumazet 260047fe360SEric Dumazet spd->nr_pages_max = buffers; 261047fe360SEric Dumazet if (buffers <= PIPE_DEF_BUFFERS) 26235f3d14dSJens Axboe return 0; 26335f3d14dSJens Axboe 264047fe360SEric Dumazet spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL); 265047fe360SEric Dumazet spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL); 26635f3d14dSJens Axboe 26735f3d14dSJens Axboe if (spd->pages && spd->partial) 26835f3d14dSJens Axboe return 0; 26935f3d14dSJens Axboe 27035f3d14dSJens Axboe kfree(spd->pages); 27135f3d14dSJens Axboe kfree(spd->partial); 27235f3d14dSJens Axboe return -ENOMEM; 27335f3d14dSJens Axboe } 27435f3d14dSJens Axboe 275047fe360SEric Dumazet void splice_shrink_spd(struct splice_pipe_desc *spd) 27635f3d14dSJens Axboe { 277047fe360SEric Dumazet if (spd->nr_pages_max <= PIPE_DEF_BUFFERS) 27835f3d14dSJens Axboe return; 27935f3d14dSJens Axboe 28035f3d14dSJens Axboe kfree(spd->pages); 28135f3d14dSJens Axboe kfree(spd->partial); 28235f3d14dSJens Axboe } 28335f3d14dSJens Axboe 2843a326a2cSIngo Molnar static int 285cbb7e577SJens Axboe __generic_file_splice_read(struct file *in, loff_t *ppos, 286cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 287cbb7e577SJens Axboe unsigned int flags) 2885274f052SJens Axboe { 2895274f052SJens Axboe struct address_space *mapping = in->f_mapping; 290d8983910SFengguang Wu unsigned int loff, nr_pages, req_pages; 29135f3d14dSJens Axboe struct page *pages[PIPE_DEF_BUFFERS]; 29235f3d14dSJens Axboe struct partial_page partial[PIPE_DEF_BUFFERS]; 2935274f052SJens Axboe struct page *page; 29491ad66efSJens Axboe pgoff_t index, end_index; 29591ad66efSJens Axboe loff_t isize; 296eb20796bSJens Axboe int error, page_nr; 297912d35f8SJens Axboe struct splice_pipe_desc spd = { 298912d35f8SJens Axboe .pages = pages, 299912d35f8SJens Axboe .partial = partial, 300047fe360SEric Dumazet .nr_pages_max = PIPE_DEF_BUFFERS, 301912d35f8SJens Axboe .flags = flags, 302912d35f8SJens Axboe .ops = &page_cache_pipe_buf_ops, 303bbdfc2f7SJens Axboe .spd_release = spd_release_page, 304912d35f8SJens Axboe }; 3055274f052SJens Axboe 30635f3d14dSJens Axboe if (splice_grow_spd(pipe, &spd)) 30735f3d14dSJens Axboe return -ENOMEM; 30835f3d14dSJens Axboe 30909cbfeafSKirill A. Shutemov index = *ppos >> PAGE_SHIFT; 31009cbfeafSKirill A. Shutemov loff = *ppos & ~PAGE_MASK; 31109cbfeafSKirill A. Shutemov req_pages = (len + loff + PAGE_SIZE - 1) >> PAGE_SHIFT; 312047fe360SEric Dumazet nr_pages = min(req_pages, spd.nr_pages_max); 3135274f052SJens Axboe 3145274f052SJens Axboe /* 315eb20796bSJens Axboe * Lookup the (hopefully) full range of pages we need. 31682aa5d61SJens Axboe */ 31735f3d14dSJens Axboe spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages); 318431a4820SFengguang Wu index += spd.nr_pages; 319eb20796bSJens Axboe 3205274f052SJens Axboe /* 321eb20796bSJens Axboe * If find_get_pages_contig() returned fewer pages than we needed, 322431a4820SFengguang Wu * readahead/allocate the rest and fill in the holes. 323eb20796bSJens Axboe */ 324431a4820SFengguang Wu if (spd.nr_pages < nr_pages) 325cf914a7dSRusty Russell page_cache_sync_readahead(mapping, &in->f_ra, in, 326cf914a7dSRusty Russell index, req_pages - spd.nr_pages); 327431a4820SFengguang Wu 328932cc6d4SJens Axboe error = 0; 329eb20796bSJens Axboe while (spd.nr_pages < nr_pages) { 330eb20796bSJens Axboe /* 331eb20796bSJens Axboe * Page could be there, find_get_pages_contig() breaks on 332eb20796bSJens Axboe * the first hole. 3335274f052SJens Axboe */ 3347480a904SJens Axboe page = find_get_page(mapping, index); 3357480a904SJens Axboe if (!page) { 336e27dedd8SJens Axboe /* 337eb20796bSJens Axboe * page didn't exist, allocate one. 3387480a904SJens Axboe */ 3397480a904SJens Axboe page = page_cache_alloc_cold(mapping); 3405274f052SJens Axboe if (!page) 3415274f052SJens Axboe break; 3425274f052SJens Axboe 3437480a904SJens Axboe error = add_to_page_cache_lru(page, mapping, index, 344c62d2555SMichal Hocko mapping_gfp_constraint(mapping, GFP_KERNEL)); 3455274f052SJens Axboe if (unlikely(error)) { 34609cbfeafSKirill A. Shutemov put_page(page); 347a0548871SJens Axboe if (error == -EEXIST) 348a0548871SJens Axboe continue; 3495274f052SJens Axboe break; 3505274f052SJens Axboe } 351eb20796bSJens Axboe /* 352eb20796bSJens Axboe * add_to_page_cache() locks the page, unlock it 353eb20796bSJens Axboe * to avoid convoluting the logic below even more. 354eb20796bSJens Axboe */ 355eb20796bSJens Axboe unlock_page(page); 3565274f052SJens Axboe } 3577480a904SJens Axboe 35835f3d14dSJens Axboe spd.pages[spd.nr_pages++] = page; 359eb20796bSJens Axboe index++; 360eb20796bSJens Axboe } 361eb20796bSJens Axboe 362eb20796bSJens Axboe /* 363eb20796bSJens Axboe * Now loop over the map and see if we need to start IO on any 364eb20796bSJens Axboe * pages, fill in the partial map, etc. 365eb20796bSJens Axboe */ 36609cbfeafSKirill A. Shutemov index = *ppos >> PAGE_SHIFT; 367eb20796bSJens Axboe nr_pages = spd.nr_pages; 368eb20796bSJens Axboe spd.nr_pages = 0; 369eb20796bSJens Axboe for (page_nr = 0; page_nr < nr_pages; page_nr++) { 370eb20796bSJens Axboe unsigned int this_len; 371eb20796bSJens Axboe 372eb20796bSJens Axboe if (!len) 373eb20796bSJens Axboe break; 374eb20796bSJens Axboe 375eb20796bSJens Axboe /* 376eb20796bSJens Axboe * this_len is the max we'll use from this page 377eb20796bSJens Axboe */ 37809cbfeafSKirill A. Shutemov this_len = min_t(unsigned long, len, PAGE_SIZE - loff); 37935f3d14dSJens Axboe page = spd.pages[page_nr]; 380eb20796bSJens Axboe 381a08a166fSFengguang Wu if (PageReadahead(page)) 382cf914a7dSRusty Russell page_cache_async_readahead(mapping, &in->f_ra, in, 383d8983910SFengguang Wu page, index, req_pages - page_nr); 384a08a166fSFengguang Wu 3857480a904SJens Axboe /* 3867480a904SJens Axboe * If the page isn't uptodate, we may need to start io on it 3877480a904SJens Axboe */ 3887480a904SJens Axboe if (!PageUptodate(page)) { 3897480a904SJens Axboe lock_page(page); 3907480a904SJens Axboe 3917480a904SJens Axboe /* 39232502b84SMiklos Szeredi * Page was truncated, or invalidated by the 39332502b84SMiklos Szeredi * filesystem. Redo the find/create, but this time the 39432502b84SMiklos Szeredi * page is kept locked, so there's no chance of another 39532502b84SMiklos Szeredi * race with truncate/invalidate. 3967480a904SJens Axboe */ 3977480a904SJens Axboe if (!page->mapping) { 3987480a904SJens Axboe unlock_page(page); 39990330e68SAbhi Das retry_lookup: 40032502b84SMiklos Szeredi page = find_or_create_page(mapping, index, 40132502b84SMiklos Szeredi mapping_gfp_mask(mapping)); 40232502b84SMiklos Szeredi 40332502b84SMiklos Szeredi if (!page) { 40432502b84SMiklos Szeredi error = -ENOMEM; 4057480a904SJens Axboe break; 4067480a904SJens Axboe } 40709cbfeafSKirill A. Shutemov put_page(spd.pages[page_nr]); 40835f3d14dSJens Axboe spd.pages[page_nr] = page; 40932502b84SMiklos Szeredi } 4107480a904SJens Axboe /* 4117480a904SJens Axboe * page was already under io and is now done, great 4127480a904SJens Axboe */ 4137480a904SJens Axboe if (PageUptodate(page)) { 4147480a904SJens Axboe unlock_page(page); 4157480a904SJens Axboe goto fill_it; 4167480a904SJens Axboe } 4177480a904SJens Axboe 4187480a904SJens Axboe /* 4197480a904SJens Axboe * need to read in the page 4207480a904SJens Axboe */ 4217480a904SJens Axboe error = mapping->a_ops->readpage(in, page); 4227480a904SJens Axboe if (unlikely(error)) { 423eb20796bSJens Axboe /* 42490330e68SAbhi Das * Re-lookup the page 425eb20796bSJens Axboe */ 4267480a904SJens Axboe if (error == AOP_TRUNCATED_PAGE) 42790330e68SAbhi Das goto retry_lookup; 428eb20796bSJens Axboe 4297480a904SJens Axboe break; 4307480a904SJens Axboe } 431620a324bSJens Axboe } 432620a324bSJens Axboe fill_it: 43391ad66efSJens Axboe /* 434620a324bSJens Axboe * i_size must be checked after PageUptodate. 43591ad66efSJens Axboe */ 43691ad66efSJens Axboe isize = i_size_read(mapping->host); 43709cbfeafSKirill A. Shutemov end_index = (isize - 1) >> PAGE_SHIFT; 438eb20796bSJens Axboe if (unlikely(!isize || index > end_index)) 43991ad66efSJens Axboe break; 44091ad66efSJens Axboe 44191ad66efSJens Axboe /* 44291ad66efSJens Axboe * if this is the last page, see if we need to shrink 44391ad66efSJens Axboe * the length and stop 44491ad66efSJens Axboe */ 44591ad66efSJens Axboe if (end_index == index) { 446475ecadeSHugh Dickins unsigned int plen; 447475ecadeSHugh Dickins 448475ecadeSHugh Dickins /* 449475ecadeSHugh Dickins * max good bytes in this page 450475ecadeSHugh Dickins */ 45109cbfeafSKirill A. Shutemov plen = ((isize - 1) & ~PAGE_MASK) + 1; 452475ecadeSHugh Dickins if (plen <= loff) 45391ad66efSJens Axboe break; 454475ecadeSHugh Dickins 45591ad66efSJens Axboe /* 45691ad66efSJens Axboe * force quit after adding this page 45791ad66efSJens Axboe */ 458475ecadeSHugh Dickins this_len = min(this_len, plen - loff); 459eb20796bSJens Axboe len = this_len; 46091ad66efSJens Axboe } 461620a324bSJens Axboe 46235f3d14dSJens Axboe spd.partial[page_nr].offset = loff; 46335f3d14dSJens Axboe spd.partial[page_nr].len = this_len; 46482aa5d61SJens Axboe len -= this_len; 46591ad66efSJens Axboe loff = 0; 466eb20796bSJens Axboe spd.nr_pages++; 467eb20796bSJens Axboe index++; 4685274f052SJens Axboe } 4695274f052SJens Axboe 470eb20796bSJens Axboe /* 471475ecadeSHugh Dickins * Release any pages at the end, if we quit early. 'page_nr' is how far 472eb20796bSJens Axboe * we got, 'nr_pages' is how many pages are in the map. 473eb20796bSJens Axboe */ 474eb20796bSJens Axboe while (page_nr < nr_pages) 47509cbfeafSKirill A. Shutemov put_page(spd.pages[page_nr++]); 47609cbfeafSKirill A. Shutemov in->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT; 477eb20796bSJens Axboe 478912d35f8SJens Axboe if (spd.nr_pages) 47935f3d14dSJens Axboe error = splice_to_pipe(pipe, &spd); 48016c523ddSJens Axboe 481047fe360SEric Dumazet splice_shrink_spd(&spd); 4827480a904SJens Axboe return error; 4835274f052SJens Axboe } 4845274f052SJens Axboe 48583f9135bSJens Axboe /** 48683f9135bSJens Axboe * generic_file_splice_read - splice data from file to a pipe 48783f9135bSJens Axboe * @in: file to splice from 488932cc6d4SJens Axboe * @ppos: position in @in 48983f9135bSJens Axboe * @pipe: pipe to splice to 49083f9135bSJens Axboe * @len: number of bytes to splice 49183f9135bSJens Axboe * @flags: splice modifier flags 49283f9135bSJens Axboe * 493932cc6d4SJens Axboe * Description: 494932cc6d4SJens Axboe * Will read pages from given file and fill them into a pipe. Can be 495932cc6d4SJens Axboe * used as long as the address_space operations for the source implements 496932cc6d4SJens Axboe * a readpage() hook. 497932cc6d4SJens Axboe * 49883f9135bSJens Axboe */ 499cbb7e577SJens Axboe ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, 500cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 501cbb7e577SJens Axboe unsigned int flags) 5025274f052SJens Axboe { 503d366d398SJens Axboe loff_t isize, left; 5048191ecd1SJens Axboe int ret; 505d366d398SJens Axboe 506be64f884SBoaz Harrosh if (IS_DAX(in->f_mapping->host)) 507be64f884SBoaz Harrosh return default_file_splice_read(in, ppos, pipe, len, flags); 508be64f884SBoaz Harrosh 509d366d398SJens Axboe isize = i_size_read(in->f_mapping->host); 510d366d398SJens Axboe if (unlikely(*ppos >= isize)) 511d366d398SJens Axboe return 0; 512d366d398SJens Axboe 513d366d398SJens Axboe left = isize - *ppos; 514d366d398SJens Axboe if (unlikely(left < len)) 515d366d398SJens Axboe len = left; 5165274f052SJens Axboe 517cbb7e577SJens Axboe ret = __generic_file_splice_read(in, ppos, pipe, len, flags); 518723590edSMiklos Szeredi if (ret > 0) { 519cbb7e577SJens Axboe *ppos += ret; 520723590edSMiklos Szeredi file_accessed(in); 521723590edSMiklos Szeredi } 5225274f052SJens Axboe 5235274f052SJens Axboe return ret; 5245274f052SJens Axboe } 525059a8f37SJens Axboe EXPORT_SYMBOL(generic_file_splice_read); 526059a8f37SJens Axboe 527241699cdSAl Viro const struct pipe_buf_operations default_pipe_buf_ops = { 5286818173bSMiklos Szeredi .can_merge = 0, 5296818173bSMiklos Szeredi .confirm = generic_pipe_buf_confirm, 5306818173bSMiklos Szeredi .release = generic_pipe_buf_release, 5316818173bSMiklos Szeredi .steal = generic_pipe_buf_steal, 5326818173bSMiklos Szeredi .get = generic_pipe_buf_get, 5336818173bSMiklos Szeredi }; 5346818173bSMiklos Szeredi 53528a625cbSMiklos Szeredi static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe, 53628a625cbSMiklos Szeredi struct pipe_buffer *buf) 53728a625cbSMiklos Szeredi { 53828a625cbSMiklos Szeredi return 1; 53928a625cbSMiklos Szeredi } 54028a625cbSMiklos Szeredi 54128a625cbSMiklos Szeredi /* Pipe buffer operations for a socket and similar. */ 54228a625cbSMiklos Szeredi const struct pipe_buf_operations nosteal_pipe_buf_ops = { 54328a625cbSMiklos Szeredi .can_merge = 0, 54428a625cbSMiklos Szeredi .confirm = generic_pipe_buf_confirm, 54528a625cbSMiklos Szeredi .release = generic_pipe_buf_release, 54628a625cbSMiklos Szeredi .steal = generic_pipe_buf_nosteal, 54728a625cbSMiklos Szeredi .get = generic_pipe_buf_get, 54828a625cbSMiklos Szeredi }; 54928a625cbSMiklos Szeredi EXPORT_SYMBOL(nosteal_pipe_buf_ops); 55028a625cbSMiklos Szeredi 5516818173bSMiklos Szeredi static ssize_t kernel_readv(struct file *file, const struct iovec *vec, 5526818173bSMiklos Szeredi unsigned long vlen, loff_t offset) 5536818173bSMiklos Szeredi { 5546818173bSMiklos Szeredi mm_segment_t old_fs; 5556818173bSMiklos Szeredi loff_t pos = offset; 5566818173bSMiklos Szeredi ssize_t res; 5576818173bSMiklos Szeredi 5586818173bSMiklos Szeredi old_fs = get_fs(); 5596818173bSMiklos Szeredi set_fs(get_ds()); 5606818173bSMiklos Szeredi /* The cast to a user pointer is valid due to the set_fs() */ 561793b80efSChristoph Hellwig res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos, 0); 5626818173bSMiklos Szeredi set_fs(old_fs); 5636818173bSMiklos Szeredi 5646818173bSMiklos Szeredi return res; 5656818173bSMiklos Szeredi } 5666818173bSMiklos Szeredi 5677bb307e8SAl Viro ssize_t kernel_write(struct file *file, const char *buf, size_t count, 568b2858d7dSMiklos Szeredi loff_t pos) 5690b0a47f5SMiklos Szeredi { 5700b0a47f5SMiklos Szeredi mm_segment_t old_fs; 5710b0a47f5SMiklos Szeredi ssize_t res; 5720b0a47f5SMiklos Szeredi 5730b0a47f5SMiklos Szeredi old_fs = get_fs(); 5740b0a47f5SMiklos Szeredi set_fs(get_ds()); 5750b0a47f5SMiklos Szeredi /* The cast to a user pointer is valid due to the set_fs() */ 5767bb307e8SAl Viro res = vfs_write(file, (__force const char __user *)buf, count, &pos); 5770b0a47f5SMiklos Szeredi set_fs(old_fs); 5780b0a47f5SMiklos Szeredi 5790b0a47f5SMiklos Szeredi return res; 5800b0a47f5SMiklos Szeredi } 5817bb307e8SAl Viro EXPORT_SYMBOL(kernel_write); 5820b0a47f5SMiklos Szeredi 5836818173bSMiklos Szeredi ssize_t default_file_splice_read(struct file *in, loff_t *ppos, 5846818173bSMiklos Szeredi struct pipe_inode_info *pipe, size_t len, 5856818173bSMiklos Szeredi unsigned int flags) 5866818173bSMiklos Szeredi { 5876818173bSMiklos Szeredi unsigned int nr_pages; 5886818173bSMiklos Szeredi unsigned int nr_freed; 5896818173bSMiklos Szeredi size_t offset; 59035f3d14dSJens Axboe struct page *pages[PIPE_DEF_BUFFERS]; 59135f3d14dSJens Axboe struct partial_page partial[PIPE_DEF_BUFFERS]; 59235f3d14dSJens Axboe struct iovec *vec, __vec[PIPE_DEF_BUFFERS]; 5936818173bSMiklos Szeredi ssize_t res; 5946818173bSMiklos Szeredi size_t this_len; 5956818173bSMiklos Szeredi int error; 5966818173bSMiklos Szeredi int i; 5976818173bSMiklos Szeredi struct splice_pipe_desc spd = { 5986818173bSMiklos Szeredi .pages = pages, 5996818173bSMiklos Szeredi .partial = partial, 600047fe360SEric Dumazet .nr_pages_max = PIPE_DEF_BUFFERS, 6016818173bSMiklos Szeredi .flags = flags, 6026818173bSMiklos Szeredi .ops = &default_pipe_buf_ops, 6036818173bSMiklos Szeredi .spd_release = spd_release_page, 6046818173bSMiklos Szeredi }; 6056818173bSMiklos Szeredi 60635f3d14dSJens Axboe if (splice_grow_spd(pipe, &spd)) 60735f3d14dSJens Axboe return -ENOMEM; 60835f3d14dSJens Axboe 60935f3d14dSJens Axboe res = -ENOMEM; 61035f3d14dSJens Axboe vec = __vec; 611047fe360SEric Dumazet if (spd.nr_pages_max > PIPE_DEF_BUFFERS) { 612047fe360SEric Dumazet vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL); 61335f3d14dSJens Axboe if (!vec) 61435f3d14dSJens Axboe goto shrink_ret; 61535f3d14dSJens Axboe } 61635f3d14dSJens Axboe 61709cbfeafSKirill A. Shutemov offset = *ppos & ~PAGE_MASK; 61809cbfeafSKirill A. Shutemov nr_pages = (len + offset + PAGE_SIZE - 1) >> PAGE_SHIFT; 6196818173bSMiklos Szeredi 620047fe360SEric Dumazet for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) { 6216818173bSMiklos Szeredi struct page *page; 6226818173bSMiklos Szeredi 6234f231228SJens Axboe page = alloc_page(GFP_USER); 6246818173bSMiklos Szeredi error = -ENOMEM; 6256818173bSMiklos Szeredi if (!page) 6266818173bSMiklos Szeredi goto err; 6276818173bSMiklos Szeredi 62809cbfeafSKirill A. Shutemov this_len = min_t(size_t, len, PAGE_SIZE - offset); 6294f231228SJens Axboe vec[i].iov_base = (void __user *) page_address(page); 6306818173bSMiklos Szeredi vec[i].iov_len = this_len; 63135f3d14dSJens Axboe spd.pages[i] = page; 6326818173bSMiklos Szeredi spd.nr_pages++; 6336818173bSMiklos Szeredi len -= this_len; 6346818173bSMiklos Szeredi offset = 0; 6356818173bSMiklos Szeredi } 6366818173bSMiklos Szeredi 6376818173bSMiklos Szeredi res = kernel_readv(in, vec, spd.nr_pages, *ppos); 63877f6bf57SAndrew Morton if (res < 0) { 63977f6bf57SAndrew Morton error = res; 6406818173bSMiklos Szeredi goto err; 64177f6bf57SAndrew Morton } 6426818173bSMiklos Szeredi 6436818173bSMiklos Szeredi error = 0; 6446818173bSMiklos Szeredi if (!res) 6456818173bSMiklos Szeredi goto err; 6466818173bSMiklos Szeredi 6476818173bSMiklos Szeredi nr_freed = 0; 6486818173bSMiklos Szeredi for (i = 0; i < spd.nr_pages; i++) { 6496818173bSMiklos Szeredi this_len = min_t(size_t, vec[i].iov_len, res); 65035f3d14dSJens Axboe spd.partial[i].offset = 0; 65135f3d14dSJens Axboe spd.partial[i].len = this_len; 6526818173bSMiklos Szeredi if (!this_len) { 65335f3d14dSJens Axboe __free_page(spd.pages[i]); 65435f3d14dSJens Axboe spd.pages[i] = NULL; 6556818173bSMiklos Szeredi nr_freed++; 6566818173bSMiklos Szeredi } 6576818173bSMiklos Szeredi res -= this_len; 6586818173bSMiklos Szeredi } 6596818173bSMiklos Szeredi spd.nr_pages -= nr_freed; 6606818173bSMiklos Szeredi 6616818173bSMiklos Szeredi res = splice_to_pipe(pipe, &spd); 6626818173bSMiklos Szeredi if (res > 0) 6636818173bSMiklos Szeredi *ppos += res; 6646818173bSMiklos Szeredi 66535f3d14dSJens Axboe shrink_ret: 66635f3d14dSJens Axboe if (vec != __vec) 66735f3d14dSJens Axboe kfree(vec); 668047fe360SEric Dumazet splice_shrink_spd(&spd); 6696818173bSMiklos Szeredi return res; 6706818173bSMiklos Szeredi 6716818173bSMiklos Szeredi err: 6724f231228SJens Axboe for (i = 0; i < spd.nr_pages; i++) 67335f3d14dSJens Axboe __free_page(spd.pages[i]); 6744f231228SJens Axboe 67535f3d14dSJens Axboe res = error; 67635f3d14dSJens Axboe goto shrink_ret; 6776818173bSMiklos Szeredi } 6786818173bSMiklos Szeredi EXPORT_SYMBOL(default_file_splice_read); 6796818173bSMiklos Szeredi 6805274f052SJens Axboe /* 6814f6f0bd2SJens Axboe * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' 682016b661eSJens Axboe * using sendpage(). Return the number of bytes sent. 6835274f052SJens Axboe */ 68476ad4d11SJens Axboe static int pipe_to_sendpage(struct pipe_inode_info *pipe, 6855274f052SJens Axboe struct pipe_buffer *buf, struct splice_desc *sd) 6865274f052SJens Axboe { 6876a14b90bSJens Axboe struct file *file = sd->u.file; 6885274f052SJens Axboe loff_t pos = sd->pos; 689a8adbe37SMichał Mirosław int more; 6905274f052SJens Axboe 69172c2d531SAl Viro if (!likely(file->f_op->sendpage)) 692a8adbe37SMichał Mirosław return -EINVAL; 693a8adbe37SMichał Mirosław 69435f9c09fSEric Dumazet more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; 695ae62ca7bSEric Dumazet 696ae62ca7bSEric Dumazet if (sd->len < sd->total_len && pipe->nrbufs > 1) 69735f9c09fSEric Dumazet more |= MSG_SENDPAGE_NOTLAST; 698ae62ca7bSEric Dumazet 699a8adbe37SMichał Mirosław return file->f_op->sendpage(file, buf->page, buf->offset, 700f84d7519SJens Axboe sd->len, &pos, more); 7015274f052SJens Axboe } 7025274f052SJens Axboe 703b3c2d2ddSMiklos Szeredi static void wakeup_pipe_writers(struct pipe_inode_info *pipe) 704b3c2d2ddSMiklos Szeredi { 705b3c2d2ddSMiklos Szeredi smp_mb(); 706b3c2d2ddSMiklos Szeredi if (waitqueue_active(&pipe->wait)) 707b3c2d2ddSMiklos Szeredi wake_up_interruptible(&pipe->wait); 708b3c2d2ddSMiklos Szeredi kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); 709b3c2d2ddSMiklos Szeredi } 710b3c2d2ddSMiklos Szeredi 711b3c2d2ddSMiklos Szeredi /** 712b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed - feed available data from a pipe to a file 713b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 714b3c2d2ddSMiklos Szeredi * @sd: information to @actor 715b3c2d2ddSMiklos Szeredi * @actor: handler that splices the data 716b3c2d2ddSMiklos Szeredi * 717b3c2d2ddSMiklos Szeredi * Description: 718b3c2d2ddSMiklos Szeredi * This function loops over the pipe and calls @actor to do the 719b3c2d2ddSMiklos Szeredi * actual moving of a single struct pipe_buffer to the desired 720b3c2d2ddSMiklos Szeredi * destination. It returns when there's no more buffers left in 721b3c2d2ddSMiklos Szeredi * the pipe or if the requested number of bytes (@sd->total_len) 722b3c2d2ddSMiklos Szeredi * have been copied. It returns a positive number (one) if the 723b3c2d2ddSMiklos Szeredi * pipe needs to be filled with more data, zero if the required 724b3c2d2ddSMiklos Szeredi * number of bytes have been copied and -errno on error. 725b3c2d2ddSMiklos Szeredi * 726b3c2d2ddSMiklos Szeredi * This, together with splice_from_pipe_{begin,end,next}, may be 727b3c2d2ddSMiklos Szeredi * used to implement the functionality of __splice_from_pipe() when 728b3c2d2ddSMiklos Szeredi * locking is required around copying the pipe buffers to the 729b3c2d2ddSMiklos Szeredi * destination. 730b3c2d2ddSMiklos Szeredi */ 73196f9bc8fSAl Viro static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, 732b3c2d2ddSMiklos Szeredi splice_actor *actor) 733b3c2d2ddSMiklos Szeredi { 734b3c2d2ddSMiklos Szeredi int ret; 735b3c2d2ddSMiklos Szeredi 736b3c2d2ddSMiklos Szeredi while (pipe->nrbufs) { 737b3c2d2ddSMiklos Szeredi struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; 738b3c2d2ddSMiklos Szeredi const struct pipe_buf_operations *ops = buf->ops; 739b3c2d2ddSMiklos Szeredi 740b3c2d2ddSMiklos Szeredi sd->len = buf->len; 741b3c2d2ddSMiklos Szeredi if (sd->len > sd->total_len) 742b3c2d2ddSMiklos Szeredi sd->len = sd->total_len; 743b3c2d2ddSMiklos Szeredi 744a8adbe37SMichał Mirosław ret = buf->ops->confirm(pipe, buf); 745a8adbe37SMichał Mirosław if (unlikely(ret)) { 746b3c2d2ddSMiklos Szeredi if (ret == -ENODATA) 747b3c2d2ddSMiklos Szeredi ret = 0; 748b3c2d2ddSMiklos Szeredi return ret; 749b3c2d2ddSMiklos Szeredi } 750a8adbe37SMichał Mirosław 751a8adbe37SMichał Mirosław ret = actor(pipe, buf, sd); 752a8adbe37SMichał Mirosław if (ret <= 0) 753a8adbe37SMichał Mirosław return ret; 754a8adbe37SMichał Mirosław 755b3c2d2ddSMiklos Szeredi buf->offset += ret; 756b3c2d2ddSMiklos Szeredi buf->len -= ret; 757b3c2d2ddSMiklos Szeredi 758b3c2d2ddSMiklos Szeredi sd->num_spliced += ret; 759b3c2d2ddSMiklos Szeredi sd->len -= ret; 760b3c2d2ddSMiklos Szeredi sd->pos += ret; 761b3c2d2ddSMiklos Szeredi sd->total_len -= ret; 762b3c2d2ddSMiklos Szeredi 763b3c2d2ddSMiklos Szeredi if (!buf->len) { 764b3c2d2ddSMiklos Szeredi buf->ops = NULL; 765b3c2d2ddSMiklos Szeredi ops->release(pipe, buf); 76635f3d14dSJens Axboe pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); 767b3c2d2ddSMiklos Szeredi pipe->nrbufs--; 7686447a3cfSAl Viro if (pipe->files) 769b3c2d2ddSMiklos Szeredi sd->need_wakeup = true; 770b3c2d2ddSMiklos Szeredi } 771b3c2d2ddSMiklos Szeredi 772b3c2d2ddSMiklos Szeredi if (!sd->total_len) 773b3c2d2ddSMiklos Szeredi return 0; 774b3c2d2ddSMiklos Szeredi } 775b3c2d2ddSMiklos Szeredi 776b3c2d2ddSMiklos Szeredi return 1; 777b3c2d2ddSMiklos Szeredi } 778b3c2d2ddSMiklos Szeredi 779b3c2d2ddSMiklos Szeredi /** 780b3c2d2ddSMiklos Szeredi * splice_from_pipe_next - wait for some data to splice from 781b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 782b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 783b3c2d2ddSMiklos Szeredi * 784b3c2d2ddSMiklos Szeredi * Description: 785b3c2d2ddSMiklos Szeredi * This function will wait for some data and return a positive 786b3c2d2ddSMiklos Szeredi * value (one) if pipe buffers are available. It will return zero 787b3c2d2ddSMiklos Szeredi * or -errno if no more data needs to be spliced. 788b3c2d2ddSMiklos Szeredi */ 78996f9bc8fSAl Viro static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) 790b3c2d2ddSMiklos Szeredi { 791c725bfceSJan Kara /* 792c725bfceSJan Kara * Check for signal early to make process killable when there are 793c725bfceSJan Kara * always buffers available 794c725bfceSJan Kara */ 795c725bfceSJan Kara if (signal_pending(current)) 796c725bfceSJan Kara return -ERESTARTSYS; 797c725bfceSJan Kara 798b3c2d2ddSMiklos Szeredi while (!pipe->nrbufs) { 799b3c2d2ddSMiklos Szeredi if (!pipe->writers) 800b3c2d2ddSMiklos Szeredi return 0; 801b3c2d2ddSMiklos Szeredi 802b3c2d2ddSMiklos Szeredi if (!pipe->waiting_writers && sd->num_spliced) 803b3c2d2ddSMiklos Szeredi return 0; 804b3c2d2ddSMiklos Szeredi 805b3c2d2ddSMiklos Szeredi if (sd->flags & SPLICE_F_NONBLOCK) 806b3c2d2ddSMiklos Szeredi return -EAGAIN; 807b3c2d2ddSMiklos Szeredi 808b3c2d2ddSMiklos Szeredi if (signal_pending(current)) 809b3c2d2ddSMiklos Szeredi return -ERESTARTSYS; 810b3c2d2ddSMiklos Szeredi 811b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) { 812b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 813b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 814b3c2d2ddSMiklos Szeredi } 815b3c2d2ddSMiklos Szeredi 816b3c2d2ddSMiklos Szeredi pipe_wait(pipe); 817b3c2d2ddSMiklos Szeredi } 818b3c2d2ddSMiklos Szeredi 819b3c2d2ddSMiklos Szeredi return 1; 820b3c2d2ddSMiklos Szeredi } 821b3c2d2ddSMiklos Szeredi 822b3c2d2ddSMiklos Szeredi /** 823b3c2d2ddSMiklos Szeredi * splice_from_pipe_begin - start splicing from pipe 824b80901bbSRandy Dunlap * @sd: information about the splice operation 825b3c2d2ddSMiklos Szeredi * 826b3c2d2ddSMiklos Szeredi * Description: 827b3c2d2ddSMiklos Szeredi * This function should be called before a loop containing 828b3c2d2ddSMiklos Szeredi * splice_from_pipe_next() and splice_from_pipe_feed() to 829b3c2d2ddSMiklos Szeredi * initialize the necessary fields of @sd. 830b3c2d2ddSMiklos Szeredi */ 83196f9bc8fSAl Viro static void splice_from_pipe_begin(struct splice_desc *sd) 832b3c2d2ddSMiklos Szeredi { 833b3c2d2ddSMiklos Szeredi sd->num_spliced = 0; 834b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 835b3c2d2ddSMiklos Szeredi } 836b3c2d2ddSMiklos Szeredi 837b3c2d2ddSMiklos Szeredi /** 838b3c2d2ddSMiklos Szeredi * splice_from_pipe_end - finish splicing from pipe 839b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 840b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 841b3c2d2ddSMiklos Szeredi * 842b3c2d2ddSMiklos Szeredi * Description: 843b3c2d2ddSMiklos Szeredi * This function will wake up pipe writers if necessary. It should 844b3c2d2ddSMiklos Szeredi * be called after a loop containing splice_from_pipe_next() and 845b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed(). 846b3c2d2ddSMiklos Szeredi */ 84796f9bc8fSAl Viro static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) 848b3c2d2ddSMiklos Szeredi { 849b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) 850b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 851b3c2d2ddSMiklos Szeredi } 852b3c2d2ddSMiklos Szeredi 853932cc6d4SJens Axboe /** 854932cc6d4SJens Axboe * __splice_from_pipe - splice data from a pipe to given actor 855932cc6d4SJens Axboe * @pipe: pipe to splice from 856932cc6d4SJens Axboe * @sd: information to @actor 857932cc6d4SJens Axboe * @actor: handler that splices the data 858932cc6d4SJens Axboe * 859932cc6d4SJens Axboe * Description: 860932cc6d4SJens Axboe * This function does little more than loop over the pipe and call 861932cc6d4SJens Axboe * @actor to do the actual moving of a single struct pipe_buffer to 862932cc6d4SJens Axboe * the desired destination. See pipe_to_file, pipe_to_sendpage, or 863932cc6d4SJens Axboe * pipe_to_user. 864932cc6d4SJens Axboe * 86583f9135bSJens Axboe */ 866c66ab6faSJens Axboe ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, 867c66ab6faSJens Axboe splice_actor *actor) 8685274f052SJens Axboe { 869b3c2d2ddSMiklos Szeredi int ret; 8705274f052SJens Axboe 871b3c2d2ddSMiklos Szeredi splice_from_pipe_begin(sd); 872b3c2d2ddSMiklos Szeredi do { 873c2489e07SJan Kara cond_resched(); 874b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_next(pipe, sd); 875b3c2d2ddSMiklos Szeredi if (ret > 0) 876b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_feed(pipe, sd, actor); 877b3c2d2ddSMiklos Szeredi } while (ret > 0); 878b3c2d2ddSMiklos Szeredi splice_from_pipe_end(pipe, sd); 8795274f052SJens Axboe 880b3c2d2ddSMiklos Szeredi return sd->num_spliced ? sd->num_spliced : ret; 8815274f052SJens Axboe } 88240bee44eSMark Fasheh EXPORT_SYMBOL(__splice_from_pipe); 8835274f052SJens Axboe 884932cc6d4SJens Axboe /** 885932cc6d4SJens Axboe * splice_from_pipe - splice data from a pipe to a file 886932cc6d4SJens Axboe * @pipe: pipe to splice from 887932cc6d4SJens Axboe * @out: file to splice to 888932cc6d4SJens Axboe * @ppos: position in @out 889932cc6d4SJens Axboe * @len: how many bytes to splice 890932cc6d4SJens Axboe * @flags: splice modifier flags 891932cc6d4SJens Axboe * @actor: handler that splices the data 892932cc6d4SJens Axboe * 893932cc6d4SJens Axboe * Description: 8942933970bSMiklos Szeredi * See __splice_from_pipe. This function locks the pipe inode, 895932cc6d4SJens Axboe * otherwise it's identical to __splice_from_pipe(). 896932cc6d4SJens Axboe * 897932cc6d4SJens Axboe */ 8986da61809SMark Fasheh ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, 8996da61809SMark Fasheh loff_t *ppos, size_t len, unsigned int flags, 9006da61809SMark Fasheh splice_actor *actor) 9016da61809SMark Fasheh { 9026da61809SMark Fasheh ssize_t ret; 903c66ab6faSJens Axboe struct splice_desc sd = { 904c66ab6faSJens Axboe .total_len = len, 905c66ab6faSJens Axboe .flags = flags, 906c66ab6faSJens Axboe .pos = *ppos, 9076a14b90bSJens Axboe .u.file = out, 908c66ab6faSJens Axboe }; 9096da61809SMark Fasheh 91061e0d47cSMiklos Szeredi pipe_lock(pipe); 911c66ab6faSJens Axboe ret = __splice_from_pipe(pipe, &sd, actor); 91261e0d47cSMiklos Szeredi pipe_unlock(pipe); 9136da61809SMark Fasheh 9146da61809SMark Fasheh return ret; 9156da61809SMark Fasheh } 9166da61809SMark Fasheh 9176da61809SMark Fasheh /** 9188d020765SAl Viro * iter_file_splice_write - splice data from a pipe to a file 9198d020765SAl Viro * @pipe: pipe info 9208d020765SAl Viro * @out: file to write to 9218d020765SAl Viro * @ppos: position in @out 9228d020765SAl Viro * @len: number of bytes to splice 9238d020765SAl Viro * @flags: splice modifier flags 9248d020765SAl Viro * 9258d020765SAl Viro * Description: 9268d020765SAl Viro * Will either move or copy pages (determined by @flags options) from 9278d020765SAl Viro * the given pipe inode to the given file. 9288d020765SAl Viro * This one is ->write_iter-based. 9298d020765SAl Viro * 9308d020765SAl Viro */ 9318d020765SAl Viro ssize_t 9328d020765SAl Viro iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out, 9338d020765SAl Viro loff_t *ppos, size_t len, unsigned int flags) 9348d020765SAl Viro { 9358d020765SAl Viro struct splice_desc sd = { 9368d020765SAl Viro .total_len = len, 9378d020765SAl Viro .flags = flags, 9388d020765SAl Viro .pos = *ppos, 9398d020765SAl Viro .u.file = out, 9408d020765SAl Viro }; 9418d020765SAl Viro int nbufs = pipe->buffers; 9428d020765SAl Viro struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec), 9438d020765SAl Viro GFP_KERNEL); 9448d020765SAl Viro ssize_t ret; 9458d020765SAl Viro 9468d020765SAl Viro if (unlikely(!array)) 9478d020765SAl Viro return -ENOMEM; 9488d020765SAl Viro 9498d020765SAl Viro pipe_lock(pipe); 9508d020765SAl Viro 9518d020765SAl Viro splice_from_pipe_begin(&sd); 9528d020765SAl Viro while (sd.total_len) { 9538d020765SAl Viro struct iov_iter from; 9548d020765SAl Viro size_t left; 9558d020765SAl Viro int n, idx; 9568d020765SAl Viro 9578d020765SAl Viro ret = splice_from_pipe_next(pipe, &sd); 9588d020765SAl Viro if (ret <= 0) 9598d020765SAl Viro break; 9608d020765SAl Viro 9618d020765SAl Viro if (unlikely(nbufs < pipe->buffers)) { 9628d020765SAl Viro kfree(array); 9638d020765SAl Viro nbufs = pipe->buffers; 9648d020765SAl Viro array = kcalloc(nbufs, sizeof(struct bio_vec), 9658d020765SAl Viro GFP_KERNEL); 9668d020765SAl Viro if (!array) { 9678d020765SAl Viro ret = -ENOMEM; 9688d020765SAl Viro break; 9698d020765SAl Viro } 9708d020765SAl Viro } 9718d020765SAl Viro 9728d020765SAl Viro /* build the vector */ 9738d020765SAl Viro left = sd.total_len; 9748d020765SAl Viro for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) { 9758d020765SAl Viro struct pipe_buffer *buf = pipe->bufs + idx; 9768d020765SAl Viro size_t this_len = buf->len; 9778d020765SAl Viro 9788d020765SAl Viro if (this_len > left) 9798d020765SAl Viro this_len = left; 9808d020765SAl Viro 9818d020765SAl Viro if (idx == pipe->buffers - 1) 9828d020765SAl Viro idx = -1; 9838d020765SAl Viro 9848d020765SAl Viro ret = buf->ops->confirm(pipe, buf); 9858d020765SAl Viro if (unlikely(ret)) { 9868d020765SAl Viro if (ret == -ENODATA) 9878d020765SAl Viro ret = 0; 9888d020765SAl Viro goto done; 9898d020765SAl Viro } 9908d020765SAl Viro 9918d020765SAl Viro array[n].bv_page = buf->page; 9928d020765SAl Viro array[n].bv_len = this_len; 9938d020765SAl Viro array[n].bv_offset = buf->offset; 9948d020765SAl Viro left -= this_len; 9958d020765SAl Viro } 9968d020765SAl Viro 99705afcb77SAl Viro iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n, 99805afcb77SAl Viro sd.total_len - left); 999dbe4e192SChristoph Hellwig ret = vfs_iter_write(out, &from, &sd.pos); 10008d020765SAl Viro if (ret <= 0) 10018d020765SAl Viro break; 10028d020765SAl Viro 10038d020765SAl Viro sd.num_spliced += ret; 10048d020765SAl Viro sd.total_len -= ret; 1005dbe4e192SChristoph Hellwig *ppos = sd.pos; 10068d020765SAl Viro 10078d020765SAl Viro /* dismiss the fully eaten buffers, adjust the partial one */ 10088d020765SAl Viro while (ret) { 10098d020765SAl Viro struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; 10108d020765SAl Viro if (ret >= buf->len) { 10118d020765SAl Viro const struct pipe_buf_operations *ops = buf->ops; 10128d020765SAl Viro ret -= buf->len; 10138d020765SAl Viro buf->len = 0; 10148d020765SAl Viro buf->ops = NULL; 10158d020765SAl Viro ops->release(pipe, buf); 10168d020765SAl Viro pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); 10178d020765SAl Viro pipe->nrbufs--; 10188d020765SAl Viro if (pipe->files) 10198d020765SAl Viro sd.need_wakeup = true; 10208d020765SAl Viro } else { 10218d020765SAl Viro buf->offset += ret; 10228d020765SAl Viro buf->len -= ret; 10238d020765SAl Viro ret = 0; 10248d020765SAl Viro } 10258d020765SAl Viro } 10268d020765SAl Viro } 10278d020765SAl Viro done: 10288d020765SAl Viro kfree(array); 10298d020765SAl Viro splice_from_pipe_end(pipe, &sd); 10308d020765SAl Viro 10318d020765SAl Viro pipe_unlock(pipe); 10328d020765SAl Viro 10338d020765SAl Viro if (sd.num_spliced) 10348d020765SAl Viro ret = sd.num_spliced; 10358d020765SAl Viro 10368d020765SAl Viro return ret; 10378d020765SAl Viro } 10388d020765SAl Viro 10398d020765SAl Viro EXPORT_SYMBOL(iter_file_splice_write); 10408d020765SAl Viro 1041b2858d7dSMiklos Szeredi static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 1042b2858d7dSMiklos Szeredi struct splice_desc *sd) 10430b0a47f5SMiklos Szeredi { 1044b2858d7dSMiklos Szeredi int ret; 1045b2858d7dSMiklos Szeredi void *data; 104606ae43f3SAl Viro loff_t tmp = sd->pos; 1047b2858d7dSMiklos Szeredi 1048fbb32750SAl Viro data = kmap(buf->page); 104906ae43f3SAl Viro ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp); 1050fbb32750SAl Viro kunmap(buf->page); 1051b2858d7dSMiklos Szeredi 1052b2858d7dSMiklos Szeredi return ret; 10530b0a47f5SMiklos Szeredi } 10540b0a47f5SMiklos Szeredi 10550b0a47f5SMiklos Szeredi static ssize_t default_file_splice_write(struct pipe_inode_info *pipe, 10560b0a47f5SMiklos Szeredi struct file *out, loff_t *ppos, 10570b0a47f5SMiklos Szeredi size_t len, unsigned int flags) 10580b0a47f5SMiklos Szeredi { 1059b2858d7dSMiklos Szeredi ssize_t ret; 10600b0a47f5SMiklos Szeredi 1061b2858d7dSMiklos Szeredi ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf); 1062b2858d7dSMiklos Szeredi if (ret > 0) 1063b2858d7dSMiklos Szeredi *ppos += ret; 10640b0a47f5SMiklos Szeredi 1065b2858d7dSMiklos Szeredi return ret; 10660b0a47f5SMiklos Szeredi } 10670b0a47f5SMiklos Szeredi 106883f9135bSJens Axboe /** 106983f9135bSJens Axboe * generic_splice_sendpage - splice data from a pipe to a socket 1070932cc6d4SJens Axboe * @pipe: pipe to splice from 107183f9135bSJens Axboe * @out: socket to write to 1072932cc6d4SJens Axboe * @ppos: position in @out 107383f9135bSJens Axboe * @len: number of bytes to splice 107483f9135bSJens Axboe * @flags: splice modifier flags 107583f9135bSJens Axboe * 1076932cc6d4SJens Axboe * Description: 107783f9135bSJens Axboe * Will send @len bytes from the pipe to a network socket. No data copying 107883f9135bSJens Axboe * is involved. 107983f9135bSJens Axboe * 108083f9135bSJens Axboe */ 10813a326a2cSIngo Molnar ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, 1082cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 10835274f052SJens Axboe { 108400522fb4SJens Axboe return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); 10855274f052SJens Axboe } 10865274f052SJens Axboe 1087059a8f37SJens Axboe EXPORT_SYMBOL(generic_splice_sendpage); 1088a0f06780SJeff Garzik 108983f9135bSJens Axboe /* 109083f9135bSJens Axboe * Attempt to initiate a splice from pipe to file. 109183f9135bSJens Axboe */ 10923a326a2cSIngo Molnar static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, 1093cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 10945274f052SJens Axboe { 10950b0a47f5SMiklos Szeredi ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, 10960b0a47f5SMiklos Szeredi loff_t *, size_t, unsigned int); 10975274f052SJens Axboe 109872c2d531SAl Viro if (out->f_op->splice_write) 10990b0a47f5SMiklos Szeredi splice_write = out->f_op->splice_write; 1100cc56f7deSChangli Gao else 11010b0a47f5SMiklos Szeredi splice_write = default_file_splice_write; 11020b0a47f5SMiklos Szeredi 1103500368f7SAl Viro return splice_write(pipe, out, ppos, len, flags); 11045274f052SJens Axboe } 11055274f052SJens Axboe 110683f9135bSJens Axboe /* 110783f9135bSJens Axboe * Attempt to initiate a splice from a file to a pipe. 110883f9135bSJens Axboe */ 1109cbb7e577SJens Axboe static long do_splice_to(struct file *in, loff_t *ppos, 1110cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 1111cbb7e577SJens Axboe unsigned int flags) 11125274f052SJens Axboe { 11136818173bSMiklos Szeredi ssize_t (*splice_read)(struct file *, loff_t *, 11146818173bSMiklos Szeredi struct pipe_inode_info *, size_t, unsigned int); 11155274f052SJens Axboe int ret; 11165274f052SJens Axboe 111749570e9bSJens Axboe if (unlikely(!(in->f_mode & FMODE_READ))) 11185274f052SJens Axboe return -EBADF; 11195274f052SJens Axboe 1120cbb7e577SJens Axboe ret = rw_verify_area(READ, in, ppos, len); 11215274f052SJens Axboe if (unlikely(ret < 0)) 11225274f052SJens Axboe return ret; 11235274f052SJens Axboe 112403cc0789SAl Viro if (unlikely(len > MAX_RW_COUNT)) 112503cc0789SAl Viro len = MAX_RW_COUNT; 112603cc0789SAl Viro 112772c2d531SAl Viro if (in->f_op->splice_read) 11286818173bSMiklos Szeredi splice_read = in->f_op->splice_read; 1129cc56f7deSChangli Gao else 11306818173bSMiklos Szeredi splice_read = default_file_splice_read; 11316818173bSMiklos Szeredi 11326818173bSMiklos Szeredi return splice_read(in, ppos, pipe, len, flags); 11335274f052SJens Axboe } 11345274f052SJens Axboe 1135932cc6d4SJens Axboe /** 1136932cc6d4SJens Axboe * splice_direct_to_actor - splices data directly between two non-pipes 1137932cc6d4SJens Axboe * @in: file to splice from 1138932cc6d4SJens Axboe * @sd: actor information on where to splice to 1139932cc6d4SJens Axboe * @actor: handles the data splicing 1140932cc6d4SJens Axboe * 1141932cc6d4SJens Axboe * Description: 1142932cc6d4SJens Axboe * This is a special case helper to splice directly between two 1143932cc6d4SJens Axboe * points, without requiring an explicit pipe. Internally an allocated 1144932cc6d4SJens Axboe * pipe is cached in the process, and reused during the lifetime of 1145932cc6d4SJens Axboe * that process. 1146932cc6d4SJens Axboe * 1147c66ab6faSJens Axboe */ 1148c66ab6faSJens Axboe ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, 1149c66ab6faSJens Axboe splice_direct_actor *actor) 1150b92ce558SJens Axboe { 1151b92ce558SJens Axboe struct pipe_inode_info *pipe; 1152b92ce558SJens Axboe long ret, bytes; 1153b92ce558SJens Axboe umode_t i_mode; 1154c66ab6faSJens Axboe size_t len; 11550ff28d9fSChristophe Leroy int i, flags, more; 1156b92ce558SJens Axboe 1157b92ce558SJens Axboe /* 1158b92ce558SJens Axboe * We require the input being a regular file, as we don't want to 1159b92ce558SJens Axboe * randomly drop data for eg socket -> socket splicing. Use the 1160b92ce558SJens Axboe * piped splicing for that! 1161b92ce558SJens Axboe */ 1162496ad9aaSAl Viro i_mode = file_inode(in)->i_mode; 1163b92ce558SJens Axboe if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) 1164b92ce558SJens Axboe return -EINVAL; 1165b92ce558SJens Axboe 1166b92ce558SJens Axboe /* 1167b92ce558SJens Axboe * neither in nor out is a pipe, setup an internal pipe attached to 1168b92ce558SJens Axboe * 'out' and transfer the wanted data from 'in' to 'out' through that 1169b92ce558SJens Axboe */ 1170b92ce558SJens Axboe pipe = current->splice_pipe; 117149570e9bSJens Axboe if (unlikely(!pipe)) { 11727bee130eSAl Viro pipe = alloc_pipe_info(); 1173b92ce558SJens Axboe if (!pipe) 1174b92ce558SJens Axboe return -ENOMEM; 1175b92ce558SJens Axboe 1176b92ce558SJens Axboe /* 1177b92ce558SJens Axboe * We don't have an immediate reader, but we'll read the stuff 117800522fb4SJens Axboe * out of the pipe right after the splice_to_pipe(). So set 1179b92ce558SJens Axboe * PIPE_READERS appropriately. 1180b92ce558SJens Axboe */ 1181b92ce558SJens Axboe pipe->readers = 1; 1182b92ce558SJens Axboe 1183b92ce558SJens Axboe current->splice_pipe = pipe; 1184b92ce558SJens Axboe } 1185b92ce558SJens Axboe 1186b92ce558SJens Axboe /* 118773d62d83SIngo Molnar * Do the splice. 1188b92ce558SJens Axboe */ 1189b92ce558SJens Axboe ret = 0; 1190b92ce558SJens Axboe bytes = 0; 1191c66ab6faSJens Axboe len = sd->total_len; 1192c66ab6faSJens Axboe flags = sd->flags; 1193c66ab6faSJens Axboe 1194c66ab6faSJens Axboe /* 1195c66ab6faSJens Axboe * Don't block on output, we have to drain the direct pipe. 1196c66ab6faSJens Axboe */ 1197c66ab6faSJens Axboe sd->flags &= ~SPLICE_F_NONBLOCK; 11980ff28d9fSChristophe Leroy more = sd->flags & SPLICE_F_MORE; 1199b92ce558SJens Axboe 1200b92ce558SJens Axboe while (len) { 120151a92c0fSJens Axboe size_t read_len; 1202a82c53a0STom Zanussi loff_t pos = sd->pos, prev_pos = pos; 1203b92ce558SJens Axboe 1204bcd4f3acSJens Axboe ret = do_splice_to(in, &pos, pipe, len, flags); 120551a92c0fSJens Axboe if (unlikely(ret <= 0)) 1206b92ce558SJens Axboe goto out_release; 1207b92ce558SJens Axboe 1208b92ce558SJens Axboe read_len = ret; 1209c66ab6faSJens Axboe sd->total_len = read_len; 1210b92ce558SJens Axboe 1211b92ce558SJens Axboe /* 12120ff28d9fSChristophe Leroy * If more data is pending, set SPLICE_F_MORE 12130ff28d9fSChristophe Leroy * If this is the last data and SPLICE_F_MORE was not set 12140ff28d9fSChristophe Leroy * initially, clears it. 12150ff28d9fSChristophe Leroy */ 12160ff28d9fSChristophe Leroy if (read_len < len) 12170ff28d9fSChristophe Leroy sd->flags |= SPLICE_F_MORE; 12180ff28d9fSChristophe Leroy else if (!more) 12190ff28d9fSChristophe Leroy sd->flags &= ~SPLICE_F_MORE; 12200ff28d9fSChristophe Leroy /* 1221b92ce558SJens Axboe * NOTE: nonblocking mode only applies to the input. We 1222b92ce558SJens Axboe * must not do the output in nonblocking mode as then we 1223b92ce558SJens Axboe * could get stuck data in the internal pipe: 1224b92ce558SJens Axboe */ 1225c66ab6faSJens Axboe ret = actor(pipe, sd); 1226a82c53a0STom Zanussi if (unlikely(ret <= 0)) { 1227a82c53a0STom Zanussi sd->pos = prev_pos; 1228b92ce558SJens Axboe goto out_release; 1229a82c53a0STom Zanussi } 1230b92ce558SJens Axboe 1231b92ce558SJens Axboe bytes += ret; 1232b92ce558SJens Axboe len -= ret; 1233bcd4f3acSJens Axboe sd->pos = pos; 1234b92ce558SJens Axboe 1235a82c53a0STom Zanussi if (ret < read_len) { 1236a82c53a0STom Zanussi sd->pos = prev_pos + ret; 123751a92c0fSJens Axboe goto out_release; 1238b92ce558SJens Axboe } 1239a82c53a0STom Zanussi } 1240b92ce558SJens Axboe 12419e97198dSJens Axboe done: 1242b92ce558SJens Axboe pipe->nrbufs = pipe->curbuf = 0; 12439e97198dSJens Axboe file_accessed(in); 1244b92ce558SJens Axboe return bytes; 1245b92ce558SJens Axboe 1246b92ce558SJens Axboe out_release: 1247b92ce558SJens Axboe /* 1248b92ce558SJens Axboe * If we did an incomplete transfer we must release 1249b92ce558SJens Axboe * the pipe buffers in question: 1250b92ce558SJens Axboe */ 125135f3d14dSJens Axboe for (i = 0; i < pipe->buffers; i++) { 1252b92ce558SJens Axboe struct pipe_buffer *buf = pipe->bufs + i; 1253b92ce558SJens Axboe 1254b92ce558SJens Axboe if (buf->ops) { 1255b92ce558SJens Axboe buf->ops->release(pipe, buf); 1256b92ce558SJens Axboe buf->ops = NULL; 1257b92ce558SJens Axboe } 1258b92ce558SJens Axboe } 1259b92ce558SJens Axboe 12609e97198dSJens Axboe if (!bytes) 12619e97198dSJens Axboe bytes = ret; 1262b92ce558SJens Axboe 12639e97198dSJens Axboe goto done; 1264c66ab6faSJens Axboe } 1265c66ab6faSJens Axboe EXPORT_SYMBOL(splice_direct_to_actor); 1266c66ab6faSJens Axboe 1267c66ab6faSJens Axboe static int direct_splice_actor(struct pipe_inode_info *pipe, 1268c66ab6faSJens Axboe struct splice_desc *sd) 1269c66ab6faSJens Axboe { 12706a14b90bSJens Axboe struct file *file = sd->u.file; 1271c66ab6faSJens Axboe 12727995bd28SAl Viro return do_splice_from(pipe, file, sd->opos, sd->total_len, 12732cb4b05eSChangli Gao sd->flags); 1274c66ab6faSJens Axboe } 1275c66ab6faSJens Axboe 1276932cc6d4SJens Axboe /** 1277932cc6d4SJens Axboe * do_splice_direct - splices data directly between two files 1278932cc6d4SJens Axboe * @in: file to splice from 1279932cc6d4SJens Axboe * @ppos: input file offset 1280932cc6d4SJens Axboe * @out: file to splice to 1281acdb37c3SRandy Dunlap * @opos: output file offset 1282932cc6d4SJens Axboe * @len: number of bytes to splice 1283932cc6d4SJens Axboe * @flags: splice modifier flags 1284932cc6d4SJens Axboe * 1285932cc6d4SJens Axboe * Description: 1286932cc6d4SJens Axboe * For use by do_sendfile(). splice can easily emulate sendfile, but 1287932cc6d4SJens Axboe * doing it in the application would incur an extra system call 1288932cc6d4SJens Axboe * (splice in + splice out, as compared to just sendfile()). So this helper 1289932cc6d4SJens Axboe * can splice directly through a process-private pipe. 1290932cc6d4SJens Axboe * 1291932cc6d4SJens Axboe */ 1292c66ab6faSJens Axboe long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, 12937995bd28SAl Viro loff_t *opos, size_t len, unsigned int flags) 1294c66ab6faSJens Axboe { 1295c66ab6faSJens Axboe struct splice_desc sd = { 1296c66ab6faSJens Axboe .len = len, 1297c66ab6faSJens Axboe .total_len = len, 1298c66ab6faSJens Axboe .flags = flags, 1299c66ab6faSJens Axboe .pos = *ppos, 13006a14b90bSJens Axboe .u.file = out, 13017995bd28SAl Viro .opos = opos, 1302c66ab6faSJens Axboe }; 130351a92c0fSJens Axboe long ret; 1304c66ab6faSJens Axboe 130518c67cb9SAl Viro if (unlikely(!(out->f_mode & FMODE_WRITE))) 130618c67cb9SAl Viro return -EBADF; 130718c67cb9SAl Viro 130818c67cb9SAl Viro if (unlikely(out->f_flags & O_APPEND)) 130918c67cb9SAl Viro return -EINVAL; 131018c67cb9SAl Viro 131118c67cb9SAl Viro ret = rw_verify_area(WRITE, out, opos, len); 131218c67cb9SAl Viro if (unlikely(ret < 0)) 131318c67cb9SAl Viro return ret; 131418c67cb9SAl Viro 1315c66ab6faSJens Axboe ret = splice_direct_to_actor(in, &sd, direct_splice_actor); 131651a92c0fSJens Axboe if (ret > 0) 1317a82c53a0STom Zanussi *ppos = sd.pos; 131851a92c0fSJens Axboe 1319c66ab6faSJens Axboe return ret; 1320b92ce558SJens Axboe } 13211c118596SMiklos Szeredi EXPORT_SYMBOL(do_splice_direct); 1322b92ce558SJens Axboe 13238924feffSAl Viro static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags) 13248924feffSAl Viro { 13258924feffSAl Viro while (pipe->nrbufs == pipe->buffers) { 13268924feffSAl Viro if (flags & SPLICE_F_NONBLOCK) 13278924feffSAl Viro return -EAGAIN; 13288924feffSAl Viro if (signal_pending(current)) 13298924feffSAl Viro return -ERESTARTSYS; 13308924feffSAl Viro pipe->waiting_writers++; 13318924feffSAl Viro pipe_wait(pipe); 13328924feffSAl Viro pipe->waiting_writers--; 13338924feffSAl Viro } 13348924feffSAl Viro return 0; 13358924feffSAl Viro } 13368924feffSAl Viro 13377c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, 13387c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe, 13397c77f0b3SMiklos Szeredi size_t len, unsigned int flags); 1340ddac0d39SJens Axboe 1341ddac0d39SJens Axboe /* 134283f9135bSJens Axboe * Determine where to splice to/from. 134383f9135bSJens Axboe */ 1344529565dcSIngo Molnar static long do_splice(struct file *in, loff_t __user *off_in, 1345529565dcSIngo Molnar struct file *out, loff_t __user *off_out, 1346529565dcSIngo Molnar size_t len, unsigned int flags) 13475274f052SJens Axboe { 13487c77f0b3SMiklos Szeredi struct pipe_inode_info *ipipe; 13497c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe; 13507995bd28SAl Viro loff_t offset; 1351a4514ebdSJens Axboe long ret; 13525274f052SJens Axboe 135371993e62SLinus Torvalds ipipe = get_pipe_info(in); 135471993e62SLinus Torvalds opipe = get_pipe_info(out); 13557c77f0b3SMiklos Szeredi 13567c77f0b3SMiklos Szeredi if (ipipe && opipe) { 13577c77f0b3SMiklos Szeredi if (off_in || off_out) 13587c77f0b3SMiklos Szeredi return -ESPIPE; 13597c77f0b3SMiklos Szeredi 13607c77f0b3SMiklos Szeredi if (!(in->f_mode & FMODE_READ)) 13617c77f0b3SMiklos Szeredi return -EBADF; 13627c77f0b3SMiklos Szeredi 13637c77f0b3SMiklos Szeredi if (!(out->f_mode & FMODE_WRITE)) 13647c77f0b3SMiklos Szeredi return -EBADF; 13657c77f0b3SMiklos Szeredi 13667c77f0b3SMiklos Szeredi /* Splicing to self would be fun, but... */ 13677c77f0b3SMiklos Szeredi if (ipipe == opipe) 13687c77f0b3SMiklos Szeredi return -EINVAL; 13697c77f0b3SMiklos Szeredi 13707c77f0b3SMiklos Szeredi return splice_pipe_to_pipe(ipipe, opipe, len, flags); 13717c77f0b3SMiklos Szeredi } 13727c77f0b3SMiklos Szeredi 13737c77f0b3SMiklos Szeredi if (ipipe) { 1374529565dcSIngo Molnar if (off_in) 1375529565dcSIngo Molnar return -ESPIPE; 1376b92ce558SJens Axboe if (off_out) { 137719c9a49bSChangli Gao if (!(out->f_mode & FMODE_PWRITE)) 1378b92ce558SJens Axboe return -EINVAL; 1379cbb7e577SJens Axboe if (copy_from_user(&offset, off_out, sizeof(loff_t))) 1380b92ce558SJens Axboe return -EFAULT; 13817995bd28SAl Viro } else { 13827995bd28SAl Viro offset = out->f_pos; 13837995bd28SAl Viro } 1384529565dcSIngo Molnar 138518c67cb9SAl Viro if (unlikely(!(out->f_mode & FMODE_WRITE))) 138618c67cb9SAl Viro return -EBADF; 138718c67cb9SAl Viro 138818c67cb9SAl Viro if (unlikely(out->f_flags & O_APPEND)) 138918c67cb9SAl Viro return -EINVAL; 139018c67cb9SAl Viro 139118c67cb9SAl Viro ret = rw_verify_area(WRITE, out, &offset, len); 139218c67cb9SAl Viro if (unlikely(ret < 0)) 139318c67cb9SAl Viro return ret; 139418c67cb9SAl Viro 1395500368f7SAl Viro file_start_write(out); 13967995bd28SAl Viro ret = do_splice_from(ipipe, out, &offset, len, flags); 1397500368f7SAl Viro file_end_write(out); 1398a4514ebdSJens Axboe 13997995bd28SAl Viro if (!off_out) 14007995bd28SAl Viro out->f_pos = offset; 14017995bd28SAl Viro else if (copy_to_user(off_out, &offset, sizeof(loff_t))) 1402a4514ebdSJens Axboe ret = -EFAULT; 1403a4514ebdSJens Axboe 1404a4514ebdSJens Axboe return ret; 1405529565dcSIngo Molnar } 14065274f052SJens Axboe 14077c77f0b3SMiklos Szeredi if (opipe) { 1408529565dcSIngo Molnar if (off_out) 1409529565dcSIngo Molnar return -ESPIPE; 1410b92ce558SJens Axboe if (off_in) { 141119c9a49bSChangli Gao if (!(in->f_mode & FMODE_PREAD)) 1412b92ce558SJens Axboe return -EINVAL; 1413cbb7e577SJens Axboe if (copy_from_user(&offset, off_in, sizeof(loff_t))) 1414b92ce558SJens Axboe return -EFAULT; 14157995bd28SAl Viro } else { 14167995bd28SAl Viro offset = in->f_pos; 14177995bd28SAl Viro } 1418529565dcSIngo Molnar 14198924feffSAl Viro pipe_lock(opipe); 14208924feffSAl Viro ret = wait_for_space(opipe, flags); 14218924feffSAl Viro if (!ret) 14227995bd28SAl Viro ret = do_splice_to(in, &offset, opipe, len, flags); 14238924feffSAl Viro pipe_unlock(opipe); 14248924feffSAl Viro if (ret > 0) 14258924feffSAl Viro wakeup_pipe_readers(opipe); 14267995bd28SAl Viro if (!off_in) 14277995bd28SAl Viro in->f_pos = offset; 14287995bd28SAl Viro else if (copy_to_user(off_in, &offset, sizeof(loff_t))) 1429a4514ebdSJens Axboe ret = -EFAULT; 1430a4514ebdSJens Axboe 1431a4514ebdSJens Axboe return ret; 1432529565dcSIngo Molnar } 14335274f052SJens Axboe 14345274f052SJens Axboe return -EINVAL; 14355274f052SJens Axboe } 14365274f052SJens Axboe 143779fddc4eSAl Viro static int iter_to_pipe(struct iov_iter *from, 143879fddc4eSAl Viro struct pipe_inode_info *pipe, 143979fddc4eSAl Viro unsigned flags) 1440912d35f8SJens Axboe { 144179fddc4eSAl Viro struct pipe_buffer buf = { 144279fddc4eSAl Viro .ops = &user_page_pipe_buf_ops, 144379fddc4eSAl Viro .flags = flags 144479fddc4eSAl Viro }; 144579fddc4eSAl Viro size_t total = 0; 144679fddc4eSAl Viro int ret = 0; 144779fddc4eSAl Viro bool failed = false; 144879fddc4eSAl Viro 144979fddc4eSAl Viro while (iov_iter_count(from) && !failed) { 145079fddc4eSAl Viro struct page *pages[16]; 1451db85a9ebSAl Viro ssize_t copied; 1452db85a9ebSAl Viro size_t start; 145379fddc4eSAl Viro int n; 1454912d35f8SJens Axboe 145579fddc4eSAl Viro copied = iov_iter_get_pages(from, pages, ~0UL, 16, &start); 145679fddc4eSAl Viro if (copied <= 0) { 145779fddc4eSAl Viro ret = copied; 145879fddc4eSAl Viro break; 145979fddc4eSAl Viro } 1460912d35f8SJens Axboe 146179fddc4eSAl Viro for (n = 0; copied; n++, start = 0) { 1462db85a9ebSAl Viro int size = min_t(int, copied, PAGE_SIZE - start); 146379fddc4eSAl Viro if (!failed) { 146479fddc4eSAl Viro buf.page = pages[n]; 146579fddc4eSAl Viro buf.offset = start; 146679fddc4eSAl Viro buf.len = size; 146779fddc4eSAl Viro ret = add_to_pipe(pipe, &buf); 146879fddc4eSAl Viro if (unlikely(ret < 0)) { 146979fddc4eSAl Viro failed = true; 147079fddc4eSAl Viro } else { 147179fddc4eSAl Viro iov_iter_advance(from, ret); 147279fddc4eSAl Viro total += ret; 147379fddc4eSAl Viro } 147479fddc4eSAl Viro } else { 147579fddc4eSAl Viro put_page(pages[n]); 147679fddc4eSAl Viro } 1477db85a9ebSAl Viro copied -= size; 1478912d35f8SJens Axboe } 1479912d35f8SJens Axboe } 148079fddc4eSAl Viro return total ? total : ret; 1481912d35f8SJens Axboe } 1482912d35f8SJens Axboe 14836a14b90bSJens Axboe static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 14846a14b90bSJens Axboe struct splice_desc *sd) 14856a14b90bSJens Axboe { 14866130f531SAl Viro int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data); 14876130f531SAl Viro return n == sd->len ? n : -EFAULT; 14886a14b90bSJens Axboe } 14896a14b90bSJens Axboe 14906a14b90bSJens Axboe /* 14916a14b90bSJens Axboe * For lack of a better implementation, implement vmsplice() to userspace 14926a14b90bSJens Axboe * as a simple copy of the pipes pages to the user iov. 14936a14b90bSJens Axboe */ 14946130f531SAl Viro static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov, 14956a14b90bSJens Axboe unsigned long nr_segs, unsigned int flags) 14966a14b90bSJens Axboe { 14976a14b90bSJens Axboe struct pipe_inode_info *pipe; 14986a14b90bSJens Axboe struct splice_desc sd; 14996a14b90bSJens Axboe long ret; 15006130f531SAl Viro struct iovec iovstack[UIO_FASTIOV]; 15016130f531SAl Viro struct iovec *iov = iovstack; 15026130f531SAl Viro struct iov_iter iter; 15036a14b90bSJens Axboe 150471993e62SLinus Torvalds pipe = get_pipe_info(file); 15056a14b90bSJens Axboe if (!pipe) 15066a14b90bSJens Axboe return -EBADF; 15076a14b90bSJens Axboe 1508345995faSAl Viro ret = import_iovec(READ, uiov, nr_segs, 1509345995faSAl Viro ARRAY_SIZE(iovstack), &iov, &iter); 1510345995faSAl Viro if (ret < 0) 1511345995faSAl Viro return ret; 15126a14b90bSJens Axboe 1513345995faSAl Viro sd.total_len = iov_iter_count(&iter); 15146a14b90bSJens Axboe sd.len = 0; 15156a14b90bSJens Axboe sd.flags = flags; 15166130f531SAl Viro sd.u.data = &iter; 15176a14b90bSJens Axboe sd.pos = 0; 15186a14b90bSJens Axboe 1519345995faSAl Viro if (sd.total_len) { 15206130f531SAl Viro pipe_lock(pipe); 15216130f531SAl Viro ret = __splice_from_pipe(pipe, &sd, pipe_to_user); 152261e0d47cSMiklos Szeredi pipe_unlock(pipe); 1523345995faSAl Viro } 15246a14b90bSJens Axboe 15256130f531SAl Viro kfree(iov); 15266a14b90bSJens Axboe return ret; 15276a14b90bSJens Axboe } 15286a14b90bSJens Axboe 1529912d35f8SJens Axboe /* 1530912d35f8SJens Axboe * vmsplice splices a user address range into a pipe. It can be thought of 1531912d35f8SJens Axboe * as splice-from-memory, where the regular splice is splice-from-file (or 1532912d35f8SJens Axboe * to file). In both cases the output is a pipe, naturally. 1533912d35f8SJens Axboe */ 1534db85a9ebSAl Viro static long vmsplice_to_pipe(struct file *file, const struct iovec __user *uiov, 1535912d35f8SJens Axboe unsigned long nr_segs, unsigned int flags) 1536912d35f8SJens Axboe { 1537ddac0d39SJens Axboe struct pipe_inode_info *pipe; 1538db85a9ebSAl Viro struct iovec iovstack[UIO_FASTIOV]; 1539db85a9ebSAl Viro struct iovec *iov = iovstack; 1540db85a9ebSAl Viro struct iov_iter from; 154135f3d14dSJens Axboe long ret; 154279fddc4eSAl Viro unsigned buf_flag = 0; 154379fddc4eSAl Viro 154479fddc4eSAl Viro if (flags & SPLICE_F_GIFT) 154579fddc4eSAl Viro buf_flag = PIPE_BUF_FLAG_GIFT; 1546912d35f8SJens Axboe 154771993e62SLinus Torvalds pipe = get_pipe_info(file); 1548ddac0d39SJens Axboe if (!pipe) 1549912d35f8SJens Axboe return -EBADF; 1550912d35f8SJens Axboe 1551db85a9ebSAl Viro ret = import_iovec(WRITE, uiov, nr_segs, 1552db85a9ebSAl Viro ARRAY_SIZE(iovstack), &iov, &from); 1553db85a9ebSAl Viro if (ret < 0) 1554db85a9ebSAl Viro return ret; 1555912d35f8SJens Axboe 15568924feffSAl Viro pipe_lock(pipe); 15578924feffSAl Viro ret = wait_for_space(pipe, flags); 155879fddc4eSAl Viro if (!ret) 155979fddc4eSAl Viro ret = iter_to_pipe(&from, pipe, buf_flag); 15608924feffSAl Viro pipe_unlock(pipe); 15618924feffSAl Viro if (ret > 0) 15628924feffSAl Viro wakeup_pipe_readers(pipe); 1563db85a9ebSAl Viro kfree(iov); 156435f3d14dSJens Axboe return ret; 1565912d35f8SJens Axboe } 1566912d35f8SJens Axboe 15676a14b90bSJens Axboe /* 15686a14b90bSJens Axboe * Note that vmsplice only really supports true splicing _from_ user memory 15696a14b90bSJens Axboe * to a pipe, not the other way around. Splicing from user memory is a simple 15706a14b90bSJens Axboe * operation that can be supported without any funky alignment restrictions 15716a14b90bSJens Axboe * or nasty vm tricks. We simply map in the user memory and fill them into 15726a14b90bSJens Axboe * a pipe. The reverse isn't quite as easy, though. There are two possible 15736a14b90bSJens Axboe * solutions for that: 15746a14b90bSJens Axboe * 15756a14b90bSJens Axboe * - memcpy() the data internally, at which point we might as well just 15766a14b90bSJens Axboe * do a regular read() on the buffer anyway. 15776a14b90bSJens Axboe * - Lots of nasty vm tricks, that are neither fast nor flexible (it 15786a14b90bSJens Axboe * has restriction limitations on both ends of the pipe). 15796a14b90bSJens Axboe * 15806a14b90bSJens Axboe * Currently we punt and implement it as a normal copy, see pipe_to_user(). 15816a14b90bSJens Axboe * 15826a14b90bSJens Axboe */ 1583836f92adSHeiko Carstens SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov, 1584836f92adSHeiko Carstens unsigned long, nr_segs, unsigned int, flags) 1585912d35f8SJens Axboe { 15862903ff01SAl Viro struct fd f; 1587912d35f8SJens Axboe long error; 1588912d35f8SJens Axboe 15896a14b90bSJens Axboe if (unlikely(nr_segs > UIO_MAXIOV)) 15906a14b90bSJens Axboe return -EINVAL; 15916a14b90bSJens Axboe else if (unlikely(!nr_segs)) 15926a14b90bSJens Axboe return 0; 15936a14b90bSJens Axboe 1594912d35f8SJens Axboe error = -EBADF; 15952903ff01SAl Viro f = fdget(fd); 15962903ff01SAl Viro if (f.file) { 15972903ff01SAl Viro if (f.file->f_mode & FMODE_WRITE) 15982903ff01SAl Viro error = vmsplice_to_pipe(f.file, iov, nr_segs, flags); 15992903ff01SAl Viro else if (f.file->f_mode & FMODE_READ) 16002903ff01SAl Viro error = vmsplice_to_user(f.file, iov, nr_segs, flags); 1601912d35f8SJens Axboe 16022903ff01SAl Viro fdput(f); 1603912d35f8SJens Axboe } 1604912d35f8SJens Axboe 1605912d35f8SJens Axboe return error; 1606912d35f8SJens Axboe } 1607912d35f8SJens Axboe 160876b021d0SAl Viro #ifdef CONFIG_COMPAT 160976b021d0SAl Viro COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32, 161076b021d0SAl Viro unsigned int, nr_segs, unsigned int, flags) 161176b021d0SAl Viro { 161276b021d0SAl Viro unsigned i; 161376b021d0SAl Viro struct iovec __user *iov; 161476b021d0SAl Viro if (nr_segs > UIO_MAXIOV) 161576b021d0SAl Viro return -EINVAL; 161676b021d0SAl Viro iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec)); 161776b021d0SAl Viro for (i = 0; i < nr_segs; i++) { 161876b021d0SAl Viro struct compat_iovec v; 161976b021d0SAl Viro if (get_user(v.iov_base, &iov32[i].iov_base) || 162076b021d0SAl Viro get_user(v.iov_len, &iov32[i].iov_len) || 162176b021d0SAl Viro put_user(compat_ptr(v.iov_base), &iov[i].iov_base) || 162276b021d0SAl Viro put_user(v.iov_len, &iov[i].iov_len)) 162376b021d0SAl Viro return -EFAULT; 162476b021d0SAl Viro } 162576b021d0SAl Viro return sys_vmsplice(fd, iov, nr_segs, flags); 162676b021d0SAl Viro } 162776b021d0SAl Viro #endif 162876b021d0SAl Viro 1629836f92adSHeiko Carstens SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, 1630836f92adSHeiko Carstens int, fd_out, loff_t __user *, off_out, 1631836f92adSHeiko Carstens size_t, len, unsigned int, flags) 16325274f052SJens Axboe { 16332903ff01SAl Viro struct fd in, out; 16345274f052SJens Axboe long error; 16355274f052SJens Axboe 16365274f052SJens Axboe if (unlikely(!len)) 16375274f052SJens Axboe return 0; 16385274f052SJens Axboe 16395274f052SJens Axboe error = -EBADF; 16402903ff01SAl Viro in = fdget(fd_in); 16412903ff01SAl Viro if (in.file) { 16422903ff01SAl Viro if (in.file->f_mode & FMODE_READ) { 16432903ff01SAl Viro out = fdget(fd_out); 16442903ff01SAl Viro if (out.file) { 16452903ff01SAl Viro if (out.file->f_mode & FMODE_WRITE) 16462903ff01SAl Viro error = do_splice(in.file, off_in, 16472903ff01SAl Viro out.file, off_out, 1648529565dcSIngo Molnar len, flags); 16492903ff01SAl Viro fdput(out); 16505274f052SJens Axboe } 16515274f052SJens Axboe } 16522903ff01SAl Viro fdput(in); 16535274f052SJens Axboe } 16545274f052SJens Axboe return error; 16555274f052SJens Axboe } 165670524490SJens Axboe 165770524490SJens Axboe /* 1658aadd06e5SJens Axboe * Make sure there's data to read. Wait for input if we can, otherwise 1659aadd06e5SJens Axboe * return an appropriate error. 1660aadd06e5SJens Axboe */ 16617c77f0b3SMiklos Szeredi static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1662aadd06e5SJens Axboe { 1663aadd06e5SJens Axboe int ret; 1664aadd06e5SJens Axboe 1665aadd06e5SJens Axboe /* 1666aadd06e5SJens Axboe * Check ->nrbufs without the inode lock first. This function 1667aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1668aadd06e5SJens Axboe */ 1669aadd06e5SJens Axboe if (pipe->nrbufs) 1670aadd06e5SJens Axboe return 0; 1671aadd06e5SJens Axboe 1672aadd06e5SJens Axboe ret = 0; 167361e0d47cSMiklos Szeredi pipe_lock(pipe); 1674aadd06e5SJens Axboe 1675aadd06e5SJens Axboe while (!pipe->nrbufs) { 1676aadd06e5SJens Axboe if (signal_pending(current)) { 1677aadd06e5SJens Axboe ret = -ERESTARTSYS; 1678aadd06e5SJens Axboe break; 1679aadd06e5SJens Axboe } 1680aadd06e5SJens Axboe if (!pipe->writers) 1681aadd06e5SJens Axboe break; 1682aadd06e5SJens Axboe if (!pipe->waiting_writers) { 1683aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1684aadd06e5SJens Axboe ret = -EAGAIN; 1685aadd06e5SJens Axboe break; 1686aadd06e5SJens Axboe } 1687aadd06e5SJens Axboe } 1688aadd06e5SJens Axboe pipe_wait(pipe); 1689aadd06e5SJens Axboe } 1690aadd06e5SJens Axboe 169161e0d47cSMiklos Szeredi pipe_unlock(pipe); 1692aadd06e5SJens Axboe return ret; 1693aadd06e5SJens Axboe } 1694aadd06e5SJens Axboe 1695aadd06e5SJens Axboe /* 1696aadd06e5SJens Axboe * Make sure there's writeable room. Wait for room if we can, otherwise 1697aadd06e5SJens Axboe * return an appropriate error. 1698aadd06e5SJens Axboe */ 16997c77f0b3SMiklos Szeredi static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1700aadd06e5SJens Axboe { 1701aadd06e5SJens Axboe int ret; 1702aadd06e5SJens Axboe 1703aadd06e5SJens Axboe /* 1704aadd06e5SJens Axboe * Check ->nrbufs without the inode lock first. This function 1705aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1706aadd06e5SJens Axboe */ 170735f3d14dSJens Axboe if (pipe->nrbufs < pipe->buffers) 1708aadd06e5SJens Axboe return 0; 1709aadd06e5SJens Axboe 1710aadd06e5SJens Axboe ret = 0; 171161e0d47cSMiklos Szeredi pipe_lock(pipe); 1712aadd06e5SJens Axboe 171335f3d14dSJens Axboe while (pipe->nrbufs >= pipe->buffers) { 1714aadd06e5SJens Axboe if (!pipe->readers) { 1715aadd06e5SJens Axboe send_sig(SIGPIPE, current, 0); 1716aadd06e5SJens Axboe ret = -EPIPE; 1717aadd06e5SJens Axboe break; 1718aadd06e5SJens Axboe } 1719aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1720aadd06e5SJens Axboe ret = -EAGAIN; 1721aadd06e5SJens Axboe break; 1722aadd06e5SJens Axboe } 1723aadd06e5SJens Axboe if (signal_pending(current)) { 1724aadd06e5SJens Axboe ret = -ERESTARTSYS; 1725aadd06e5SJens Axboe break; 1726aadd06e5SJens Axboe } 1727aadd06e5SJens Axboe pipe->waiting_writers++; 1728aadd06e5SJens Axboe pipe_wait(pipe); 1729aadd06e5SJens Axboe pipe->waiting_writers--; 1730aadd06e5SJens Axboe } 1731aadd06e5SJens Axboe 173261e0d47cSMiklos Szeredi pipe_unlock(pipe); 1733aadd06e5SJens Axboe return ret; 1734aadd06e5SJens Axboe } 1735aadd06e5SJens Axboe 1736aadd06e5SJens Axboe /* 17377c77f0b3SMiklos Szeredi * Splice contents of ipipe to opipe. 17387c77f0b3SMiklos Szeredi */ 17397c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, 17407c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe, 17417c77f0b3SMiklos Szeredi size_t len, unsigned int flags) 17427c77f0b3SMiklos Szeredi { 17437c77f0b3SMiklos Szeredi struct pipe_buffer *ibuf, *obuf; 17447c77f0b3SMiklos Szeredi int ret = 0, nbuf; 17457c77f0b3SMiklos Szeredi bool input_wakeup = false; 17467c77f0b3SMiklos Szeredi 17477c77f0b3SMiklos Szeredi 17487c77f0b3SMiklos Szeredi retry: 17497c77f0b3SMiklos Szeredi ret = ipipe_prep(ipipe, flags); 17507c77f0b3SMiklos Szeredi if (ret) 17517c77f0b3SMiklos Szeredi return ret; 17527c77f0b3SMiklos Szeredi 17537c77f0b3SMiklos Szeredi ret = opipe_prep(opipe, flags); 17547c77f0b3SMiklos Szeredi if (ret) 17557c77f0b3SMiklos Szeredi return ret; 17567c77f0b3SMiklos Szeredi 17577c77f0b3SMiklos Szeredi /* 17587c77f0b3SMiklos Szeredi * Potential ABBA deadlock, work around it by ordering lock 17597c77f0b3SMiklos Szeredi * grabbing by pipe info address. Otherwise two different processes 17607c77f0b3SMiklos Szeredi * could deadlock (one doing tee from A -> B, the other from B -> A). 17617c77f0b3SMiklos Szeredi */ 17627c77f0b3SMiklos Szeredi pipe_double_lock(ipipe, opipe); 17637c77f0b3SMiklos Szeredi 17647c77f0b3SMiklos Szeredi do { 17657c77f0b3SMiklos Szeredi if (!opipe->readers) { 17667c77f0b3SMiklos Szeredi send_sig(SIGPIPE, current, 0); 17677c77f0b3SMiklos Szeredi if (!ret) 17687c77f0b3SMiklos Szeredi ret = -EPIPE; 17697c77f0b3SMiklos Szeredi break; 17707c77f0b3SMiklos Szeredi } 17717c77f0b3SMiklos Szeredi 17727c77f0b3SMiklos Szeredi if (!ipipe->nrbufs && !ipipe->writers) 17737c77f0b3SMiklos Szeredi break; 17747c77f0b3SMiklos Szeredi 17757c77f0b3SMiklos Szeredi /* 17767c77f0b3SMiklos Szeredi * Cannot make any progress, because either the input 17777c77f0b3SMiklos Szeredi * pipe is empty or the output pipe is full. 17787c77f0b3SMiklos Szeredi */ 177935f3d14dSJens Axboe if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) { 17807c77f0b3SMiklos Szeredi /* Already processed some buffers, break */ 17817c77f0b3SMiklos Szeredi if (ret) 17827c77f0b3SMiklos Szeredi break; 17837c77f0b3SMiklos Szeredi 17847c77f0b3SMiklos Szeredi if (flags & SPLICE_F_NONBLOCK) { 17857c77f0b3SMiklos Szeredi ret = -EAGAIN; 17867c77f0b3SMiklos Szeredi break; 17877c77f0b3SMiklos Szeredi } 17887c77f0b3SMiklos Szeredi 17897c77f0b3SMiklos Szeredi /* 17907c77f0b3SMiklos Szeredi * We raced with another reader/writer and haven't 17917c77f0b3SMiklos Szeredi * managed to process any buffers. A zero return 17927c77f0b3SMiklos Szeredi * value means EOF, so retry instead. 17937c77f0b3SMiklos Szeredi */ 17947c77f0b3SMiklos Szeredi pipe_unlock(ipipe); 17957c77f0b3SMiklos Szeredi pipe_unlock(opipe); 17967c77f0b3SMiklos Szeredi goto retry; 17977c77f0b3SMiklos Szeredi } 17987c77f0b3SMiklos Szeredi 17997c77f0b3SMiklos Szeredi ibuf = ipipe->bufs + ipipe->curbuf; 180035f3d14dSJens Axboe nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); 18017c77f0b3SMiklos Szeredi obuf = opipe->bufs + nbuf; 18027c77f0b3SMiklos Szeredi 18037c77f0b3SMiklos Szeredi if (len >= ibuf->len) { 18047c77f0b3SMiklos Szeredi /* 18057c77f0b3SMiklos Szeredi * Simply move the whole buffer from ipipe to opipe 18067c77f0b3SMiklos Szeredi */ 18077c77f0b3SMiklos Szeredi *obuf = *ibuf; 18087c77f0b3SMiklos Szeredi ibuf->ops = NULL; 18097c77f0b3SMiklos Szeredi opipe->nrbufs++; 181035f3d14dSJens Axboe ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1); 18117c77f0b3SMiklos Szeredi ipipe->nrbufs--; 18127c77f0b3SMiklos Szeredi input_wakeup = true; 18137c77f0b3SMiklos Szeredi } else { 18147c77f0b3SMiklos Szeredi /* 18157c77f0b3SMiklos Szeredi * Get a reference to this pipe buffer, 18167c77f0b3SMiklos Szeredi * so we can copy the contents over. 18177c77f0b3SMiklos Szeredi */ 18187c77f0b3SMiklos Szeredi ibuf->ops->get(ipipe, ibuf); 18197c77f0b3SMiklos Szeredi *obuf = *ibuf; 18207c77f0b3SMiklos Szeredi 18217c77f0b3SMiklos Szeredi /* 18227c77f0b3SMiklos Szeredi * Don't inherit the gift flag, we need to 18237c77f0b3SMiklos Szeredi * prevent multiple steals of this page. 18247c77f0b3SMiklos Szeredi */ 18257c77f0b3SMiklos Szeredi obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 18267c77f0b3SMiklos Szeredi 18277c77f0b3SMiklos Szeredi obuf->len = len; 18287c77f0b3SMiklos Szeredi opipe->nrbufs++; 18297c77f0b3SMiklos Szeredi ibuf->offset += obuf->len; 18307c77f0b3SMiklos Szeredi ibuf->len -= obuf->len; 18317c77f0b3SMiklos Szeredi } 18327c77f0b3SMiklos Szeredi ret += obuf->len; 18337c77f0b3SMiklos Szeredi len -= obuf->len; 18347c77f0b3SMiklos Szeredi } while (len); 18357c77f0b3SMiklos Szeredi 18367c77f0b3SMiklos Szeredi pipe_unlock(ipipe); 18377c77f0b3SMiklos Szeredi pipe_unlock(opipe); 18387c77f0b3SMiklos Szeredi 18397c77f0b3SMiklos Szeredi /* 18407c77f0b3SMiklos Szeredi * If we put data in the output pipe, wakeup any potential readers. 18417c77f0b3SMiklos Szeredi */ 1842825cdcb1SNamhyung Kim if (ret > 0) 1843825cdcb1SNamhyung Kim wakeup_pipe_readers(opipe); 1844825cdcb1SNamhyung Kim 18457c77f0b3SMiklos Szeredi if (input_wakeup) 18467c77f0b3SMiklos Szeredi wakeup_pipe_writers(ipipe); 18477c77f0b3SMiklos Szeredi 18487c77f0b3SMiklos Szeredi return ret; 18497c77f0b3SMiklos Szeredi } 18507c77f0b3SMiklos Szeredi 18517c77f0b3SMiklos Szeredi /* 185270524490SJens Axboe * Link contents of ipipe to opipe. 185370524490SJens Axboe */ 185470524490SJens Axboe static int link_pipe(struct pipe_inode_info *ipipe, 185570524490SJens Axboe struct pipe_inode_info *opipe, 185670524490SJens Axboe size_t len, unsigned int flags) 185770524490SJens Axboe { 185870524490SJens Axboe struct pipe_buffer *ibuf, *obuf; 1859aadd06e5SJens Axboe int ret = 0, i = 0, nbuf; 186070524490SJens Axboe 186170524490SJens Axboe /* 186270524490SJens Axboe * Potential ABBA deadlock, work around it by ordering lock 186361e0d47cSMiklos Szeredi * grabbing by pipe info address. Otherwise two different processes 186470524490SJens Axboe * could deadlock (one doing tee from A -> B, the other from B -> A). 186570524490SJens Axboe */ 186661e0d47cSMiklos Szeredi pipe_double_lock(ipipe, opipe); 186770524490SJens Axboe 1868aadd06e5SJens Axboe do { 186970524490SJens Axboe if (!opipe->readers) { 187070524490SJens Axboe send_sig(SIGPIPE, current, 0); 187170524490SJens Axboe if (!ret) 187270524490SJens Axboe ret = -EPIPE; 187370524490SJens Axboe break; 187470524490SJens Axboe } 187570524490SJens Axboe 187670524490SJens Axboe /* 1877aadd06e5SJens Axboe * If we have iterated all input buffers or ran out of 1878aadd06e5SJens Axboe * output room, break. 187970524490SJens Axboe */ 188035f3d14dSJens Axboe if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) 1881aadd06e5SJens Axboe break; 1882aadd06e5SJens Axboe 188335f3d14dSJens Axboe ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1)); 188435f3d14dSJens Axboe nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); 188570524490SJens Axboe 188670524490SJens Axboe /* 188770524490SJens Axboe * Get a reference to this pipe buffer, 188870524490SJens Axboe * so we can copy the contents over. 188970524490SJens Axboe */ 189070524490SJens Axboe ibuf->ops->get(ipipe, ibuf); 189170524490SJens Axboe 189270524490SJens Axboe obuf = opipe->bufs + nbuf; 189370524490SJens Axboe *obuf = *ibuf; 189470524490SJens Axboe 18957afa6fd0SJens Axboe /* 18967afa6fd0SJens Axboe * Don't inherit the gift flag, we need to 18977afa6fd0SJens Axboe * prevent multiple steals of this page. 18987afa6fd0SJens Axboe */ 18997afa6fd0SJens Axboe obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 19007afa6fd0SJens Axboe 190170524490SJens Axboe if (obuf->len > len) 190270524490SJens Axboe obuf->len = len; 190370524490SJens Axboe 190470524490SJens Axboe opipe->nrbufs++; 190570524490SJens Axboe ret += obuf->len; 190670524490SJens Axboe len -= obuf->len; 1907aadd06e5SJens Axboe i++; 1908aadd06e5SJens Axboe } while (len); 190970524490SJens Axboe 191002cf01aeSJens Axboe /* 191102cf01aeSJens Axboe * return EAGAIN if we have the potential of some data in the 191202cf01aeSJens Axboe * future, otherwise just return 0 191302cf01aeSJens Axboe */ 191402cf01aeSJens Axboe if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) 191502cf01aeSJens Axboe ret = -EAGAIN; 191602cf01aeSJens Axboe 191761e0d47cSMiklos Szeredi pipe_unlock(ipipe); 191861e0d47cSMiklos Szeredi pipe_unlock(opipe); 191970524490SJens Axboe 1920aadd06e5SJens Axboe /* 1921aadd06e5SJens Axboe * If we put data in the output pipe, wakeup any potential readers. 1922aadd06e5SJens Axboe */ 1923825cdcb1SNamhyung Kim if (ret > 0) 1924825cdcb1SNamhyung Kim wakeup_pipe_readers(opipe); 192570524490SJens Axboe 192670524490SJens Axboe return ret; 192770524490SJens Axboe } 192870524490SJens Axboe 192970524490SJens Axboe /* 193070524490SJens Axboe * This is a tee(1) implementation that works on pipes. It doesn't copy 193170524490SJens Axboe * any data, it simply references the 'in' pages on the 'out' pipe. 193270524490SJens Axboe * The 'flags' used are the SPLICE_F_* variants, currently the only 193370524490SJens Axboe * applicable one is SPLICE_F_NONBLOCK. 193470524490SJens Axboe */ 193570524490SJens Axboe static long do_tee(struct file *in, struct file *out, size_t len, 193670524490SJens Axboe unsigned int flags) 193770524490SJens Axboe { 193871993e62SLinus Torvalds struct pipe_inode_info *ipipe = get_pipe_info(in); 193971993e62SLinus Torvalds struct pipe_inode_info *opipe = get_pipe_info(out); 1940aadd06e5SJens Axboe int ret = -EINVAL; 194170524490SJens Axboe 194270524490SJens Axboe /* 1943aadd06e5SJens Axboe * Duplicate the contents of ipipe to opipe without actually 1944aadd06e5SJens Axboe * copying the data. 194570524490SJens Axboe */ 1946aadd06e5SJens Axboe if (ipipe && opipe && ipipe != opipe) { 1947aadd06e5SJens Axboe /* 1948aadd06e5SJens Axboe * Keep going, unless we encounter an error. The ipipe/opipe 1949aadd06e5SJens Axboe * ordering doesn't really matter. 1950aadd06e5SJens Axboe */ 19517c77f0b3SMiklos Szeredi ret = ipipe_prep(ipipe, flags); 1952aadd06e5SJens Axboe if (!ret) { 19537c77f0b3SMiklos Szeredi ret = opipe_prep(opipe, flags); 195402cf01aeSJens Axboe if (!ret) 1955aadd06e5SJens Axboe ret = link_pipe(ipipe, opipe, len, flags); 1956aadd06e5SJens Axboe } 1957aadd06e5SJens Axboe } 195870524490SJens Axboe 1959aadd06e5SJens Axboe return ret; 196070524490SJens Axboe } 196170524490SJens Axboe 1962836f92adSHeiko Carstens SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) 196370524490SJens Axboe { 19642903ff01SAl Viro struct fd in; 19652903ff01SAl Viro int error; 196670524490SJens Axboe 196770524490SJens Axboe if (unlikely(!len)) 196870524490SJens Axboe return 0; 196970524490SJens Axboe 197070524490SJens Axboe error = -EBADF; 19712903ff01SAl Viro in = fdget(fdin); 19722903ff01SAl Viro if (in.file) { 19732903ff01SAl Viro if (in.file->f_mode & FMODE_READ) { 19742903ff01SAl Viro struct fd out = fdget(fdout); 19752903ff01SAl Viro if (out.file) { 19762903ff01SAl Viro if (out.file->f_mode & FMODE_WRITE) 19772903ff01SAl Viro error = do_tee(in.file, out.file, 19782903ff01SAl Viro len, flags); 19792903ff01SAl Viro fdput(out); 198070524490SJens Axboe } 198170524490SJens Axboe } 19822903ff01SAl Viro fdput(in); 198370524490SJens Axboe } 198470524490SJens Axboe 198570524490SJens Axboe return error; 198670524490SJens Axboe } 1987