| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * "splice": joining two ropes together by interweaving their strands. |
| 4 | * |
| 5 | * This is the "extended pipe" functionality, where a pipe is used as |
| 6 | * an arbitrary in-memory buffer. Think of a pipe as a small kernel |
| 7 | * buffer that you can use to transfer data from one end to the other. |
| 8 | * |
| 9 | * The traditional unix read/write is extended with a "splice()" operation |
| 10 | * that transfers data buffers to or from a pipe buffer. |
| 11 | * |
| 12 | * Named by Larry McVoy, original implementation from Linus, extended by |
| 13 | * Jens to support splicing to files, network, direct splicing, etc and |
| 14 | * fixing lots of bugs. |
| 15 | * |
| 16 | * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk> |
| 17 | * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org> |
| 18 | * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu> |
| 19 | * |
| 20 | */ |
| 21 | #include <linux/bvec.h> |
| 22 | #include <linux/fs.h> |
| 23 | #include <linux/file.h> |
| 24 | #include <linux/pagemap.h> |
| 25 | #include <linux/splice.h> |
| 26 | #include <linux/memcontrol.h> |
| 27 | #include <linux/mm_inline.h> |
| 28 | #include <linux/swap.h> |
| 29 | #include <linux/writeback.h> |
| 30 | #include <linux/export.h> |
| 31 | #include <linux/syscalls.h> |
| 32 | #include <linux/uio.h> |
| 33 | #include <linux/fsnotify.h> |
| 34 | #include <linux/security.h> |
| 35 | #include <linux/gfp.h> |
| 36 | #include <linux/net.h> |
| 37 | #include <linux/socket.h> |
| 38 | #include <linux/sched/signal.h> |
| 39 | |
| 40 | #include "internal.h" |
| 41 | |
| 42 | /* |
| 43 | * Splice doesn't support FMODE_NOWAIT. Since pipes may set this flag to |
| 44 | * indicate they support non-blocking reads or writes, we must clear it |
| 45 | * here if set to avoid blocking other users of this pipe if splice is |
| 46 | * being done on it. |
| 47 | */ |
| 48 | static noinline void pipe_clear_nowait(struct file *file) |
| 49 | { |
| 50 | fmode_t fmode = READ_ONCE(file->f_mode); |
| 51 | |
| 52 | do { |
| 53 | if (!(fmode & FMODE_NOWAIT)) |
| 54 | break; |
| 55 | } while (!try_cmpxchg(&file->f_mode, &fmode, fmode & ~FMODE_NOWAIT)); |
| 56 | } |
| 57 | |
| 58 | /* |
| 59 | * Attempt to steal a page from a pipe buffer. This should perhaps go into |
| 60 | * a vm helper function, it's already simplified quite a bit by the |
| 61 | * addition of remove_mapping(). If success is returned, the caller may |
| 62 | * attempt to reuse this page for another destination. |
| 63 | */ |
| 64 | static bool page_cache_pipe_buf_try_steal(struct pipe_inode_info *pipe, |
| 65 | struct pipe_buffer *buf) |
| 66 | { |
| 67 | struct folio *folio = page_folio(buf->page); |
| 68 | struct address_space *mapping; |
| 69 | |
| 70 | folio_lock(folio); |
| 71 | |
| 72 | mapping = folio_mapping(folio); |
| 73 | if (mapping) { |
| 74 | WARN_ON(!folio_test_uptodate(folio)); |
| 75 | |
| 76 | /* |
| 77 | * At least for ext2 with nobh option, we need to wait on |
| 78 | * writeback completing on this folio, since we'll remove it |
| 79 | * from the pagecache. Otherwise truncate wont wait on the |
| 80 | * folio, allowing the disk blocks to be reused by someone else |
| 81 | * before we actually wrote our data to them. fs corruption |
| 82 | * ensues. |
| 83 | */ |
| 84 | folio_wait_writeback(folio); |
| 85 | |
| 86 | if (!filemap_release_folio(folio, GFP_KERNEL)) |
| 87 | goto out_unlock; |
| 88 | |
| 89 | /* |
| 90 | * If we succeeded in removing the mapping, set LRU flag |
| 91 | * and return good. |
| 92 | */ |
| 93 | if (remove_mapping(mapping, folio)) { |
| 94 | buf->flags |= PIPE_BUF_FLAG_LRU; |
| 95 | return true; |
| 96 | } |
| 97 | } |
| 98 | |
| 99 | /* |
| 100 | * Raced with truncate or failed to remove folio from current |
| 101 | * address space, unlock and return failure. |
| 102 | */ |
| 103 | out_unlock: |
| 104 | folio_unlock(folio); |
| 105 | return false; |
| 106 | } |
| 107 | |
| 108 | static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, |
| 109 | struct pipe_buffer *buf) |
| 110 | { |
| 111 | put_page(buf->page); |
| 112 | buf->flags &= ~PIPE_BUF_FLAG_LRU; |
| 113 | } |
| 114 | |
| 115 | /* |
| 116 | * Check whether the contents of buf is OK to access. Since the content |
| 117 | * is a page cache page, IO may be in flight. |
| 118 | */ |
| 119 | static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, |
| 120 | struct pipe_buffer *buf) |
| 121 | { |
| 122 | struct folio *folio = page_folio(buf->page); |
| 123 | int err; |
| 124 | |
| 125 | if (!folio_test_uptodate(folio)) { |
| 126 | folio_lock(folio); |
| 127 | |
| 128 | /* |
| 129 | * Folio got truncated/unhashed. This will cause a 0-byte |
| 130 | * splice, if this is the first page. |
| 131 | */ |
| 132 | if (!folio->mapping) { |
| 133 | err = -ENODATA; |
| 134 | goto error; |
| 135 | } |
| 136 | |
| 137 | /* |
| 138 | * Uh oh, read-error from disk. |
| 139 | */ |
| 140 | if (!folio_test_uptodate(folio)) { |
| 141 | err = -EIO; |
| 142 | goto error; |
| 143 | } |
| 144 | |
| 145 | /* Folio is ok after all, we are done */ |
| 146 | folio_unlock(folio); |
| 147 | } |
| 148 | |
| 149 | return 0; |
| 150 | error: |
| 151 | folio_unlock(folio); |
| 152 | return err; |
| 153 | } |
| 154 | |
| 155 | const struct pipe_buf_operations page_cache_pipe_buf_ops = { |
| 156 | .confirm = page_cache_pipe_buf_confirm, |
| 157 | .release = page_cache_pipe_buf_release, |
| 158 | .try_steal = page_cache_pipe_buf_try_steal, |
| 159 | .get = generic_pipe_buf_get, |
| 160 | }; |
| 161 | |
| 162 | static bool user_page_pipe_buf_try_steal(struct pipe_inode_info *pipe, |
| 163 | struct pipe_buffer *buf) |
| 164 | { |
| 165 | if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) |
| 166 | return false; |
| 167 | |
| 168 | buf->flags |= PIPE_BUF_FLAG_LRU; |
| 169 | return generic_pipe_buf_try_steal(pipe, buf); |
| 170 | } |
| 171 | |
| 172 | static const struct pipe_buf_operations user_page_pipe_buf_ops = { |
| 173 | .release = page_cache_pipe_buf_release, |
| 174 | .try_steal = user_page_pipe_buf_try_steal, |
| 175 | .get = generic_pipe_buf_get, |
| 176 | }; |
| 177 | |
| 178 | static void wakeup_pipe_readers(struct pipe_inode_info *pipe) |
| 179 | { |
| 180 | smp_mb(); |
| 181 | if (waitqueue_active(&pipe->rd_wait)) |
| 182 | wake_up_interruptible(&pipe->rd_wait); |
| 183 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
| 184 | } |
| 185 | |
| 186 | /** |
| 187 | * splice_to_pipe - fill passed data into a pipe |
| 188 | * @pipe: pipe to fill |
| 189 | * @spd: data to fill |
| 190 | * |
| 191 | * Description: |
| 192 | * @spd contains a map of pages and len/offset tuples, along with |
| 193 | * the struct pipe_buf_operations associated with these pages. This |
| 194 | * function will link that data to the pipe. |
| 195 | * |
| 196 | */ |
| 197 | ssize_t splice_to_pipe(struct pipe_inode_info *pipe, |
| 198 | struct splice_pipe_desc *spd) |
| 199 | { |
| 200 | unsigned int spd_pages = spd->nr_pages; |
| 201 | unsigned int tail = pipe->tail; |
| 202 | unsigned int head = pipe->head; |
| 203 | ssize_t ret = 0; |
| 204 | int page_nr = 0; |
| 205 | |
| 206 | if (!spd_pages) |
| 207 | return 0; |
| 208 | |
| 209 | if (unlikely(!pipe->readers)) { |
| 210 | send_sig(SIGPIPE, current, 0); |
| 211 | ret = -EPIPE; |
| 212 | goto out; |
| 213 | } |
| 214 | |
| 215 | while (!pipe_full(head, tail, pipe->max_usage)) { |
| 216 | struct pipe_buffer *buf = pipe_buf(pipe, head); |
| 217 | |
| 218 | buf->page = spd->pages[page_nr]; |
| 219 | buf->offset = spd->partial[page_nr].offset; |
| 220 | buf->len = spd->partial[page_nr].len; |
| 221 | buf->private = spd->partial[page_nr].private; |
| 222 | buf->ops = spd->ops; |
| 223 | buf->flags = 0; |
| 224 | |
| 225 | head++; |
| 226 | pipe->head = head; |
| 227 | page_nr++; |
| 228 | ret += buf->len; |
| 229 | |
| 230 | if (!--spd->nr_pages) |
| 231 | break; |
| 232 | } |
| 233 | |
| 234 | if (!ret) |
| 235 | ret = -EAGAIN; |
| 236 | |
| 237 | out: |
| 238 | while (page_nr < spd_pages) |
| 239 | spd->spd_release(spd, page_nr++); |
| 240 | |
| 241 | return ret; |
| 242 | } |
| 243 | EXPORT_SYMBOL_GPL(splice_to_pipe); |
| 244 | |
| 245 | ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf) |
| 246 | { |
| 247 | unsigned int head = pipe->head; |
| 248 | unsigned int tail = pipe->tail; |
| 249 | int ret; |
| 250 | |
| 251 | if (unlikely(!pipe->readers)) { |
| 252 | send_sig(SIGPIPE, current, 0); |
| 253 | ret = -EPIPE; |
| 254 | } else if (pipe_full(head, tail, pipe->max_usage)) { |
| 255 | ret = -EAGAIN; |
| 256 | } else { |
| 257 | *pipe_buf(pipe, head) = *buf; |
| 258 | pipe->head = head + 1; |
| 259 | return buf->len; |
| 260 | } |
| 261 | pipe_buf_release(pipe, buf); |
| 262 | return ret; |
| 263 | } |
| 264 | EXPORT_SYMBOL(add_to_pipe); |
| 265 | |
| 266 | /* |
| 267 | * Check if we need to grow the arrays holding pages and partial page |
| 268 | * descriptions. |
| 269 | */ |
| 270 | int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd) |
| 271 | { |
| 272 | unsigned int max_usage = READ_ONCE(pipe->max_usage); |
| 273 | |
| 274 | spd->nr_pages_max = max_usage; |
| 275 | if (max_usage <= PIPE_DEF_BUFFERS) |
| 276 | return 0; |
| 277 | |
| 278 | spd->pages = kmalloc_array(max_usage, sizeof(struct page *), GFP_KERNEL); |
| 279 | spd->partial = kmalloc_array(max_usage, sizeof(struct partial_page), |
| 280 | GFP_KERNEL); |
| 281 | |
| 282 | if (spd->pages && spd->partial) |
| 283 | return 0; |
| 284 | |
| 285 | kfree(spd->pages); |
| 286 | kfree(spd->partial); |
| 287 | return -ENOMEM; |
| 288 | } |
| 289 | |
| 290 | void splice_shrink_spd(struct splice_pipe_desc *spd) |
| 291 | { |
| 292 | if (spd->nr_pages_max <= PIPE_DEF_BUFFERS) |
| 293 | return; |
| 294 | |
| 295 | kfree(spd->pages); |
| 296 | kfree(spd->partial); |
| 297 | } |
| 298 | |
| 299 | /** |
| 300 | * copy_splice_read - Copy data from a file and splice the copy into a pipe |
| 301 | * @in: The file to read from |
| 302 | * @ppos: Pointer to the file position to read from |
| 303 | * @pipe: The pipe to splice into |
| 304 | * @len: The amount to splice |
| 305 | * @flags: The SPLICE_F_* flags |
| 306 | * |
| 307 | * This function allocates a bunch of pages sufficient to hold the requested |
| 308 | * amount of data (but limited by the remaining pipe capacity), passes it to |
| 309 | * the file's ->read_iter() to read into and then splices the used pages into |
| 310 | * the pipe. |
| 311 | * |
| 312 | * Return: On success, the number of bytes read will be returned and *@ppos |
| 313 | * will be updated if appropriate; 0 will be returned if there is no more data |
| 314 | * to be read; -EAGAIN will be returned if the pipe had no space, and some |
| 315 | * other negative error code will be returned on error. A short read may occur |
| 316 | * if the pipe has insufficient space, we reach the end of the data or we hit a |
| 317 | * hole. |
| 318 | */ |
| 319 | ssize_t copy_splice_read(struct file *in, loff_t *ppos, |
| 320 | struct pipe_inode_info *pipe, |
| 321 | size_t len, unsigned int flags) |
| 322 | { |
| 323 | struct iov_iter to; |
| 324 | struct bio_vec *bv; |
| 325 | struct kiocb kiocb; |
| 326 | struct page **pages; |
| 327 | ssize_t ret; |
| 328 | size_t used, npages, chunk, remain, keep = 0; |
| 329 | int i; |
| 330 | |
| 331 | /* Work out how much data we can actually add into the pipe */ |
| 332 | used = pipe_buf_usage(pipe); |
| 333 | npages = max_t(ssize_t, pipe->max_usage - used, 0); |
| 334 | len = min_t(size_t, len, npages * PAGE_SIZE); |
| 335 | npages = DIV_ROUND_UP(len, PAGE_SIZE); |
| 336 | |
| 337 | bv = kzalloc(array_size(npages, sizeof(bv[0])) + |
| 338 | array_size(npages, sizeof(struct page *)), GFP_KERNEL); |
| 339 | if (!bv) |
| 340 | return -ENOMEM; |
| 341 | |
| 342 | pages = (struct page **)(bv + npages); |
| 343 | npages = alloc_pages_bulk(GFP_USER, npages, pages); |
| 344 | if (!npages) { |
| 345 | kfree(bv); |
| 346 | return -ENOMEM; |
| 347 | } |
| 348 | |
| 349 | remain = len = min_t(size_t, len, npages * PAGE_SIZE); |
| 350 | |
| 351 | for (i = 0; i < npages; i++) { |
| 352 | chunk = min_t(size_t, PAGE_SIZE, remain); |
| 353 | bv[i].bv_page = pages[i]; |
| 354 | bv[i].bv_offset = 0; |
| 355 | bv[i].bv_len = chunk; |
| 356 | remain -= chunk; |
| 357 | } |
| 358 | |
| 359 | /* Do the I/O */ |
| 360 | iov_iter_bvec(&to, ITER_DEST, bv, npages, len); |
| 361 | init_sync_kiocb(&kiocb, in); |
| 362 | kiocb.ki_pos = *ppos; |
| 363 | ret = in->f_op->read_iter(&kiocb, &to); |
| 364 | |
| 365 | if (ret > 0) { |
| 366 | keep = DIV_ROUND_UP(ret, PAGE_SIZE); |
| 367 | *ppos = kiocb.ki_pos; |
| 368 | } |
| 369 | |
| 370 | /* |
| 371 | * Callers of ->splice_read() expect -EAGAIN on "can't put anything in |
| 372 | * there", rather than -EFAULT. |
| 373 | */ |
| 374 | if (ret == -EFAULT) |
| 375 | ret = -EAGAIN; |
| 376 | |
| 377 | /* Free any pages that didn't get touched at all. */ |
| 378 | if (keep < npages) |
| 379 | release_pages(pages + keep, npages - keep); |
| 380 | |
| 381 | /* Push the remaining pages into the pipe. */ |
| 382 | remain = ret; |
| 383 | for (i = 0; i < keep; i++) { |
| 384 | struct pipe_buffer *buf = pipe_head_buf(pipe); |
| 385 | |
| 386 | chunk = min_t(size_t, remain, PAGE_SIZE); |
| 387 | *buf = (struct pipe_buffer) { |
| 388 | .ops = &default_pipe_buf_ops, |
| 389 | .page = bv[i].bv_page, |
| 390 | .offset = 0, |
| 391 | .len = chunk, |
| 392 | }; |
| 393 | pipe->head++; |
| 394 | remain -= chunk; |
| 395 | } |
| 396 | |
| 397 | kfree(bv); |
| 398 | return ret; |
| 399 | } |
| 400 | EXPORT_SYMBOL(copy_splice_read); |
| 401 | |
| 402 | const struct pipe_buf_operations default_pipe_buf_ops = { |
| 403 | .release = generic_pipe_buf_release, |
| 404 | .try_steal = generic_pipe_buf_try_steal, |
| 405 | .get = generic_pipe_buf_get, |
| 406 | }; |
| 407 | |
| 408 | /* Pipe buffer operations for a socket and similar. */ |
| 409 | const struct pipe_buf_operations nosteal_pipe_buf_ops = { |
| 410 | .release = generic_pipe_buf_release, |
| 411 | .get = generic_pipe_buf_get, |
| 412 | }; |
| 413 | EXPORT_SYMBOL(nosteal_pipe_buf_ops); |
| 414 | |
| 415 | static void wakeup_pipe_writers(struct pipe_inode_info *pipe) |
| 416 | { |
| 417 | smp_mb(); |
| 418 | if (waitqueue_active(&pipe->wr_wait)) |
| 419 | wake_up_interruptible(&pipe->wr_wait); |
| 420 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
| 421 | } |
| 422 | |
| 423 | /** |
| 424 | * splice_from_pipe_feed - feed available data from a pipe to a file |
| 425 | * @pipe: pipe to splice from |
| 426 | * @sd: information to @actor |
| 427 | * @actor: handler that splices the data |
| 428 | * |
| 429 | * Description: |
| 430 | * This function loops over the pipe and calls @actor to do the |
| 431 | * actual moving of a single struct pipe_buffer to the desired |
| 432 | * destination. It returns when there's no more buffers left in |
| 433 | * the pipe or if the requested number of bytes (@sd->total_len) |
| 434 | * have been copied. It returns a positive number (one) if the |
| 435 | * pipe needs to be filled with more data, zero if the required |
| 436 | * number of bytes have been copied and -errno on error. |
| 437 | * |
| 438 | * This, together with splice_from_pipe_{begin,end,next}, may be |
| 439 | * used to implement the functionality of __splice_from_pipe() when |
| 440 | * locking is required around copying the pipe buffers to the |
| 441 | * destination. |
| 442 | */ |
| 443 | static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, |
| 444 | splice_actor *actor) |
| 445 | { |
| 446 | unsigned int head = pipe->head; |
| 447 | unsigned int tail = pipe->tail; |
| 448 | int ret; |
| 449 | |
| 450 | while (!pipe_empty(head, tail)) { |
| 451 | struct pipe_buffer *buf = pipe_buf(pipe, tail); |
| 452 | |
| 453 | sd->len = buf->len; |
| 454 | if (sd->len > sd->total_len) |
| 455 | sd->len = sd->total_len; |
| 456 | |
| 457 | ret = pipe_buf_confirm(pipe, buf); |
| 458 | if (unlikely(ret)) { |
| 459 | if (ret == -ENODATA) |
| 460 | ret = 0; |
| 461 | return ret; |
| 462 | } |
| 463 | |
| 464 | ret = actor(pipe, buf, sd); |
| 465 | if (ret <= 0) |
| 466 | return ret; |
| 467 | |
| 468 | buf->offset += ret; |
| 469 | buf->len -= ret; |
| 470 | |
| 471 | sd->num_spliced += ret; |
| 472 | sd->len -= ret; |
| 473 | sd->pos += ret; |
| 474 | sd->total_len -= ret; |
| 475 | |
| 476 | if (!buf->len) { |
| 477 | pipe_buf_release(pipe, buf); |
| 478 | tail++; |
| 479 | pipe->tail = tail; |
| 480 | if (pipe->files) |
| 481 | sd->need_wakeup = true; |
| 482 | } |
| 483 | |
| 484 | if (!sd->total_len) |
| 485 | return 0; |
| 486 | } |
| 487 | |
| 488 | return 1; |
| 489 | } |
| 490 | |
| 491 | /* We know we have a pipe buffer, but maybe it's empty? */ |
| 492 | static inline bool eat_empty_buffer(struct pipe_inode_info *pipe) |
| 493 | { |
| 494 | unsigned int tail = pipe->tail; |
| 495 | struct pipe_buffer *buf = pipe_buf(pipe, tail); |
| 496 | |
| 497 | if (unlikely(!buf->len)) { |
| 498 | pipe_buf_release(pipe, buf); |
| 499 | pipe->tail = tail+1; |
| 500 | return true; |
| 501 | } |
| 502 | |
| 503 | return false; |
| 504 | } |
| 505 | |
| 506 | /** |
| 507 | * splice_from_pipe_next - wait for some data to splice from |
| 508 | * @pipe: pipe to splice from |
| 509 | * @sd: information about the splice operation |
| 510 | * |
| 511 | * Description: |
| 512 | * This function will wait for some data and return a positive |
| 513 | * value (one) if pipe buffers are available. It will return zero |
| 514 | * or -errno if no more data needs to be spliced. |
| 515 | */ |
| 516 | static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) |
| 517 | { |
| 518 | /* |
| 519 | * Check for signal early to make process killable when there are |
| 520 | * always buffers available |
| 521 | */ |
| 522 | if (signal_pending(current)) |
| 523 | return -ERESTARTSYS; |
| 524 | |
| 525 | repeat: |
| 526 | while (pipe_is_empty(pipe)) { |
| 527 | if (!pipe->writers) |
| 528 | return 0; |
| 529 | |
| 530 | if (sd->num_spliced) |
| 531 | return 0; |
| 532 | |
| 533 | if (sd->flags & SPLICE_F_NONBLOCK) |
| 534 | return -EAGAIN; |
| 535 | |
| 536 | if (signal_pending(current)) |
| 537 | return -ERESTARTSYS; |
| 538 | |
| 539 | if (sd->need_wakeup) { |
| 540 | wakeup_pipe_writers(pipe); |
| 541 | sd->need_wakeup = false; |
| 542 | } |
| 543 | |
| 544 | pipe_wait_readable(pipe); |
| 545 | } |
| 546 | |
| 547 | if (eat_empty_buffer(pipe)) |
| 548 | goto repeat; |
| 549 | |
| 550 | return 1; |
| 551 | } |
| 552 | |
| 553 | /** |
| 554 | * splice_from_pipe_begin - start splicing from pipe |
| 555 | * @sd: information about the splice operation |
| 556 | * |
| 557 | * Description: |
| 558 | * This function should be called before a loop containing |
| 559 | * splice_from_pipe_next() and splice_from_pipe_feed() to |
| 560 | * initialize the necessary fields of @sd. |
| 561 | */ |
| 562 | static void splice_from_pipe_begin(struct splice_desc *sd) |
| 563 | { |
| 564 | sd->num_spliced = 0; |
| 565 | sd->need_wakeup = false; |
| 566 | } |
| 567 | |
| 568 | /** |
| 569 | * splice_from_pipe_end - finish splicing from pipe |
| 570 | * @pipe: pipe to splice from |
| 571 | * @sd: information about the splice operation |
| 572 | * |
| 573 | * Description: |
| 574 | * This function will wake up pipe writers if necessary. It should |
| 575 | * be called after a loop containing splice_from_pipe_next() and |
| 576 | * splice_from_pipe_feed(). |
| 577 | */ |
| 578 | static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) |
| 579 | { |
| 580 | if (sd->need_wakeup) |
| 581 | wakeup_pipe_writers(pipe); |
| 582 | } |
| 583 | |
| 584 | /** |
| 585 | * __splice_from_pipe - splice data from a pipe to given actor |
| 586 | * @pipe: pipe to splice from |
| 587 | * @sd: information to @actor |
| 588 | * @actor: handler that splices the data |
| 589 | * |
| 590 | * Description: |
| 591 | * This function does little more than loop over the pipe and call |
| 592 | * @actor to do the actual moving of a single struct pipe_buffer to |
| 593 | * the desired destination. See pipe_to_file, pipe_to_sendmsg, or |
| 594 | * pipe_to_user. |
| 595 | * |
| 596 | */ |
| 597 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, |
| 598 | splice_actor *actor) |
| 599 | { |
| 600 | int ret; |
| 601 | |
| 602 | splice_from_pipe_begin(sd); |
| 603 | do { |
| 604 | cond_resched(); |
| 605 | ret = splice_from_pipe_next(pipe, sd); |
| 606 | if (ret > 0) |
| 607 | ret = splice_from_pipe_feed(pipe, sd, actor); |
| 608 | } while (ret > 0); |
| 609 | splice_from_pipe_end(pipe, sd); |
| 610 | |
| 611 | return sd->num_spliced ? sd->num_spliced : ret; |
| 612 | } |
| 613 | EXPORT_SYMBOL(__splice_from_pipe); |
| 614 | |
| 615 | /** |
| 616 | * splice_from_pipe - splice data from a pipe to a file |
| 617 | * @pipe: pipe to splice from |
| 618 | * @out: file to splice to |
| 619 | * @ppos: position in @out |
| 620 | * @len: how many bytes to splice |
| 621 | * @flags: splice modifier flags |
| 622 | * @actor: handler that splices the data |
| 623 | * |
| 624 | * Description: |
| 625 | * See __splice_from_pipe. This function locks the pipe inode, |
| 626 | * otherwise it's identical to __splice_from_pipe(). |
| 627 | * |
| 628 | */ |
| 629 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
| 630 | loff_t *ppos, size_t len, unsigned int flags, |
| 631 | splice_actor *actor) |
| 632 | { |
| 633 | ssize_t ret; |
| 634 | struct splice_desc sd = { |
| 635 | .total_len = len, |
| 636 | .flags = flags, |
| 637 | .pos = *ppos, |
| 638 | .u.file = out, |
| 639 | }; |
| 640 | |
| 641 | pipe_lock(pipe); |
| 642 | ret = __splice_from_pipe(pipe, &sd, actor); |
| 643 | pipe_unlock(pipe); |
| 644 | |
| 645 | return ret; |
| 646 | } |
| 647 | |
| 648 | /** |
| 649 | * iter_file_splice_write - splice data from a pipe to a file |
| 650 | * @pipe: pipe info |
| 651 | * @out: file to write to |
| 652 | * @ppos: position in @out |
| 653 | * @len: number of bytes to splice |
| 654 | * @flags: splice modifier flags |
| 655 | * |
| 656 | * Description: |
| 657 | * Will either move or copy pages (determined by @flags options) from |
| 658 | * the given pipe inode to the given file. |
| 659 | * This one is ->write_iter-based. |
| 660 | * |
| 661 | */ |
| 662 | ssize_t |
| 663 | iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out, |
| 664 | loff_t *ppos, size_t len, unsigned int flags) |
| 665 | { |
| 666 | struct splice_desc sd = { |
| 667 | .total_len = len, |
| 668 | .flags = flags, |
| 669 | .pos = *ppos, |
| 670 | .u.file = out, |
| 671 | }; |
| 672 | int nbufs = pipe->max_usage; |
| 673 | struct bio_vec *array; |
| 674 | ssize_t ret; |
| 675 | |
| 676 | if (!out->f_op->write_iter) |
| 677 | return -EINVAL; |
| 678 | |
| 679 | array = kcalloc(nbufs, sizeof(struct bio_vec), GFP_KERNEL); |
| 680 | if (unlikely(!array)) |
| 681 | return -ENOMEM; |
| 682 | |
| 683 | pipe_lock(pipe); |
| 684 | |
| 685 | splice_from_pipe_begin(&sd); |
| 686 | while (sd.total_len) { |
| 687 | struct kiocb kiocb; |
| 688 | struct iov_iter from; |
| 689 | unsigned int head, tail; |
| 690 | size_t left; |
| 691 | int n; |
| 692 | |
| 693 | ret = splice_from_pipe_next(pipe, &sd); |
| 694 | if (ret <= 0) |
| 695 | break; |
| 696 | |
| 697 | if (unlikely(nbufs < pipe->max_usage)) { |
| 698 | kfree(array); |
| 699 | nbufs = pipe->max_usage; |
| 700 | array = kcalloc(nbufs, sizeof(struct bio_vec), |
| 701 | GFP_KERNEL); |
| 702 | if (!array) { |
| 703 | ret = -ENOMEM; |
| 704 | break; |
| 705 | } |
| 706 | } |
| 707 | |
| 708 | head = pipe->head; |
| 709 | tail = pipe->tail; |
| 710 | |
| 711 | /* build the vector */ |
| 712 | left = sd.total_len; |
| 713 | for (n = 0; !pipe_empty(head, tail) && left && n < nbufs; tail++) { |
| 714 | struct pipe_buffer *buf = pipe_buf(pipe, tail); |
| 715 | size_t this_len = buf->len; |
| 716 | |
| 717 | /* zero-length bvecs are not supported, skip them */ |
| 718 | if (!this_len) |
| 719 | continue; |
| 720 | this_len = min(this_len, left); |
| 721 | |
| 722 | ret = pipe_buf_confirm(pipe, buf); |
| 723 | if (unlikely(ret)) { |
| 724 | if (ret == -ENODATA) |
| 725 | ret = 0; |
| 726 | goto done; |
| 727 | } |
| 728 | |
| 729 | bvec_set_page(&array[n], buf->page, this_len, |
| 730 | buf->offset); |
| 731 | left -= this_len; |
| 732 | n++; |
| 733 | } |
| 734 | |
| 735 | iov_iter_bvec(&from, ITER_SOURCE, array, n, sd.total_len - left); |
| 736 | init_sync_kiocb(&kiocb, out); |
| 737 | kiocb.ki_pos = sd.pos; |
| 738 | ret = out->f_op->write_iter(&kiocb, &from); |
| 739 | sd.pos = kiocb.ki_pos; |
| 740 | if (ret <= 0) |
| 741 | break; |
| 742 | |
| 743 | sd.num_spliced += ret; |
| 744 | sd.total_len -= ret; |
| 745 | *ppos = sd.pos; |
| 746 | |
| 747 | /* dismiss the fully eaten buffers, adjust the partial one */ |
| 748 | tail = pipe->tail; |
| 749 | while (ret) { |
| 750 | struct pipe_buffer *buf = pipe_buf(pipe, tail); |
| 751 | if (ret >= buf->len) { |
| 752 | ret -= buf->len; |
| 753 | buf->len = 0; |
| 754 | pipe_buf_release(pipe, buf); |
| 755 | tail++; |
| 756 | pipe->tail = tail; |
| 757 | if (pipe->files) |
| 758 | sd.need_wakeup = true; |
| 759 | } else { |
| 760 | buf->offset += ret; |
| 761 | buf->len -= ret; |
| 762 | ret = 0; |
| 763 | } |
| 764 | } |
| 765 | } |
| 766 | done: |
| 767 | kfree(array); |
| 768 | splice_from_pipe_end(pipe, &sd); |
| 769 | |
| 770 | pipe_unlock(pipe); |
| 771 | |
| 772 | if (sd.num_spliced) |
| 773 | ret = sd.num_spliced; |
| 774 | |
| 775 | return ret; |
| 776 | } |
| 777 | |
| 778 | EXPORT_SYMBOL(iter_file_splice_write); |
| 779 | |
| 780 | #ifdef CONFIG_NET |
| 781 | /** |
| 782 | * splice_to_socket - splice data from a pipe to a socket |
| 783 | * @pipe: pipe to splice from |
| 784 | * @out: socket to write to |
| 785 | * @ppos: position in @out |
| 786 | * @len: number of bytes to splice |
| 787 | * @flags: splice modifier flags |
| 788 | * |
| 789 | * Description: |
| 790 | * Will send @len bytes from the pipe to a network socket. No data copying |
| 791 | * is involved. |
| 792 | * |
| 793 | */ |
| 794 | ssize_t splice_to_socket(struct pipe_inode_info *pipe, struct file *out, |
| 795 | loff_t *ppos, size_t len, unsigned int flags) |
| 796 | { |
| 797 | struct socket *sock = sock_from_file(out); |
| 798 | struct bio_vec bvec[16]; |
| 799 | struct msghdr msg = {}; |
| 800 | ssize_t ret = 0; |
| 801 | size_t spliced = 0; |
| 802 | bool need_wakeup = false; |
| 803 | |
| 804 | pipe_lock(pipe); |
| 805 | |
| 806 | while (len > 0) { |
| 807 | unsigned int head, tail, bc = 0; |
| 808 | size_t remain = len; |
| 809 | |
| 810 | /* |
| 811 | * Check for signal early to make process killable when there |
| 812 | * are always buffers available |
| 813 | */ |
| 814 | ret = -ERESTARTSYS; |
| 815 | if (signal_pending(current)) |
| 816 | break; |
| 817 | |
| 818 | while (pipe_is_empty(pipe)) { |
| 819 | ret = 0; |
| 820 | if (!pipe->writers) |
| 821 | goto out; |
| 822 | |
| 823 | if (spliced) |
| 824 | goto out; |
| 825 | |
| 826 | ret = -EAGAIN; |
| 827 | if (flags & SPLICE_F_NONBLOCK) |
| 828 | goto out; |
| 829 | |
| 830 | ret = -ERESTARTSYS; |
| 831 | if (signal_pending(current)) |
| 832 | goto out; |
| 833 | |
| 834 | if (need_wakeup) { |
| 835 | wakeup_pipe_writers(pipe); |
| 836 | need_wakeup = false; |
| 837 | } |
| 838 | |
| 839 | pipe_wait_readable(pipe); |
| 840 | } |
| 841 | |
| 842 | head = pipe->head; |
| 843 | tail = pipe->tail; |
| 844 | |
| 845 | while (!pipe_empty(head, tail)) { |
| 846 | struct pipe_buffer *buf = pipe_buf(pipe, tail); |
| 847 | size_t seg; |
| 848 | |
| 849 | if (!buf->len) { |
| 850 | tail++; |
| 851 | continue; |
| 852 | } |
| 853 | |
| 854 | seg = min_t(size_t, remain, buf->len); |
| 855 | |
| 856 | ret = pipe_buf_confirm(pipe, buf); |
| 857 | if (unlikely(ret)) { |
| 858 | if (ret == -ENODATA) |
| 859 | ret = 0; |
| 860 | break; |
| 861 | } |
| 862 | |
| 863 | bvec_set_page(&bvec[bc++], buf->page, seg, buf->offset); |
| 864 | remain -= seg; |
| 865 | if (remain == 0 || bc >= ARRAY_SIZE(bvec)) |
| 866 | break; |
| 867 | tail++; |
| 868 | } |
| 869 | |
| 870 | if (!bc) |
| 871 | break; |
| 872 | |
| 873 | msg.msg_flags = MSG_SPLICE_PAGES; |
| 874 | if (flags & SPLICE_F_MORE) |
| 875 | msg.msg_flags |= MSG_MORE; |
| 876 | if (remain && pipe_occupancy(pipe->head, tail) > 0) |
| 877 | msg.msg_flags |= MSG_MORE; |
| 878 | if (out->f_flags & O_NONBLOCK) |
| 879 | msg.msg_flags |= MSG_DONTWAIT; |
| 880 | |
| 881 | iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, bvec, bc, |
| 882 | len - remain); |
| 883 | ret = sock_sendmsg(sock, &msg); |
| 884 | if (ret <= 0) |
| 885 | break; |
| 886 | |
| 887 | spliced += ret; |
| 888 | len -= ret; |
| 889 | tail = pipe->tail; |
| 890 | while (ret > 0) { |
| 891 | struct pipe_buffer *buf = pipe_buf(pipe, tail); |
| 892 | size_t seg = min_t(size_t, ret, buf->len); |
| 893 | |
| 894 | buf->offset += seg; |
| 895 | buf->len -= seg; |
| 896 | ret -= seg; |
| 897 | |
| 898 | if (!buf->len) { |
| 899 | pipe_buf_release(pipe, buf); |
| 900 | tail++; |
| 901 | } |
| 902 | } |
| 903 | |
| 904 | if (tail != pipe->tail) { |
| 905 | pipe->tail = tail; |
| 906 | if (pipe->files) |
| 907 | need_wakeup = true; |
| 908 | } |
| 909 | } |
| 910 | |
| 911 | out: |
| 912 | pipe_unlock(pipe); |
| 913 | if (need_wakeup) |
| 914 | wakeup_pipe_writers(pipe); |
| 915 | return spliced ?: ret; |
| 916 | } |
| 917 | #endif |
| 918 | |
| 919 | static int warn_unsupported(struct file *file, const char *op) |
| 920 | { |
| 921 | pr_debug_ratelimited( |
| 922 | "splice %s not supported for file %pD4 (pid: %d comm: %.20s)\n", |
| 923 | op, file, current->pid, current->comm); |
| 924 | return -EINVAL; |
| 925 | } |
| 926 | |
| 927 | /* |
| 928 | * Attempt to initiate a splice from pipe to file. |
| 929 | */ |
| 930 | static ssize_t do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
| 931 | loff_t *ppos, size_t len, unsigned int flags) |
| 932 | { |
| 933 | if (unlikely(!out->f_op->splice_write)) |
| 934 | return warn_unsupported(out, "write"); |
| 935 | return out->f_op->splice_write(pipe, out, ppos, len, flags); |
| 936 | } |
| 937 | |
| 938 | /* |
| 939 | * Indicate to the caller that there was a premature EOF when reading from the |
| 940 | * source and the caller didn't indicate they would be sending more data after |
| 941 | * this. |
| 942 | */ |
| 943 | static void do_splice_eof(struct splice_desc *sd) |
| 944 | { |
| 945 | if (sd->splice_eof) |
| 946 | sd->splice_eof(sd); |
| 947 | } |
| 948 | |
| 949 | /* |
| 950 | * Callers already called rw_verify_area() on the entire range. |
| 951 | * No need to call it for sub ranges. |
| 952 | */ |
| 953 | static ssize_t do_splice_read(struct file *in, loff_t *ppos, |
| 954 | struct pipe_inode_info *pipe, size_t len, |
| 955 | unsigned int flags) |
| 956 | { |
| 957 | unsigned int p_space; |
| 958 | |
| 959 | if (unlikely(!(in->f_mode & FMODE_READ))) |
| 960 | return -EBADF; |
| 961 | if (!len) |
| 962 | return 0; |
| 963 | |
| 964 | /* Don't try to read more the pipe has space for. */ |
| 965 | p_space = pipe->max_usage - pipe_buf_usage(pipe); |
| 966 | len = min_t(size_t, len, p_space << PAGE_SHIFT); |
| 967 | |
| 968 | if (unlikely(len > MAX_RW_COUNT)) |
| 969 | len = MAX_RW_COUNT; |
| 970 | |
| 971 | if (unlikely(!in->f_op->splice_read)) |
| 972 | return warn_unsupported(in, "read"); |
| 973 | /* |
| 974 | * O_DIRECT and DAX don't deal with the pagecache, so we allocate a |
| 975 | * buffer, copy into it and splice that into the pipe. |
| 976 | */ |
| 977 | if ((in->f_flags & O_DIRECT) || IS_DAX(in->f_mapping->host)) |
| 978 | return copy_splice_read(in, ppos, pipe, len, flags); |
| 979 | return in->f_op->splice_read(in, ppos, pipe, len, flags); |
| 980 | } |
| 981 | |
| 982 | /** |
| 983 | * vfs_splice_read - Read data from a file and splice it into a pipe |
| 984 | * @in: File to splice from |
| 985 | * @ppos: Input file offset |
| 986 | * @pipe: Pipe to splice to |
| 987 | * @len: Number of bytes to splice |
| 988 | * @flags: Splice modifier flags (SPLICE_F_*) |
| 989 | * |
| 990 | * Splice the requested amount of data from the input file to the pipe. This |
| 991 | * is synchronous as the caller must hold the pipe lock across the entire |
| 992 | * operation. |
| 993 | * |
| 994 | * If successful, it returns the amount of data spliced, 0 if it hit the EOF or |
| 995 | * a hole and a negative error code otherwise. |
| 996 | */ |
| 997 | ssize_t vfs_splice_read(struct file *in, loff_t *ppos, |
| 998 | struct pipe_inode_info *pipe, size_t len, |
| 999 | unsigned int flags) |
| 1000 | { |
| 1001 | ssize_t ret; |
| 1002 | |
| 1003 | ret = rw_verify_area(READ, in, ppos, len); |
| 1004 | if (unlikely(ret < 0)) |
| 1005 | return ret; |
| 1006 | |
| 1007 | return do_splice_read(in, ppos, pipe, len, flags); |
| 1008 | } |
| 1009 | EXPORT_SYMBOL_GPL(vfs_splice_read); |
| 1010 | |
| 1011 | /** |
| 1012 | * splice_direct_to_actor - splices data directly between two non-pipes |
| 1013 | * @in: file to splice from |
| 1014 | * @sd: actor information on where to splice to |
| 1015 | * @actor: handles the data splicing |
| 1016 | * |
| 1017 | * Description: |
| 1018 | * This is a special case helper to splice directly between two |
| 1019 | * points, without requiring an explicit pipe. Internally an allocated |
| 1020 | * pipe is cached in the process, and reused during the lifetime of |
| 1021 | * that process. |
| 1022 | * |
| 1023 | */ |
| 1024 | ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, |
| 1025 | splice_direct_actor *actor) |
| 1026 | { |
| 1027 | struct pipe_inode_info *pipe; |
| 1028 | ssize_t ret, bytes; |
| 1029 | size_t len; |
| 1030 | int i, flags, more; |
| 1031 | |
| 1032 | /* |
| 1033 | * We require the input to be seekable, as we don't want to randomly |
| 1034 | * drop data for eg socket -> socket splicing. Use the piped splicing |
| 1035 | * for that! |
| 1036 | */ |
| 1037 | if (unlikely(!(in->f_mode & FMODE_LSEEK))) |
| 1038 | return -EINVAL; |
| 1039 | |
| 1040 | /* |
| 1041 | * neither in nor out is a pipe, setup an internal pipe attached to |
| 1042 | * 'out' and transfer the wanted data from 'in' to 'out' through that |
| 1043 | */ |
| 1044 | pipe = current->splice_pipe; |
| 1045 | if (unlikely(!pipe)) { |
| 1046 | pipe = alloc_pipe_info(); |
| 1047 | if (!pipe) |
| 1048 | return -ENOMEM; |
| 1049 | |
| 1050 | /* |
| 1051 | * We don't have an immediate reader, but we'll read the stuff |
| 1052 | * out of the pipe right after the splice_to_pipe(). So set |
| 1053 | * PIPE_READERS appropriately. |
| 1054 | */ |
| 1055 | pipe->readers = 1; |
| 1056 | |
| 1057 | current->splice_pipe = pipe; |
| 1058 | } |
| 1059 | |
| 1060 | /* |
| 1061 | * Do the splice. |
| 1062 | */ |
| 1063 | bytes = 0; |
| 1064 | len = sd->total_len; |
| 1065 | |
| 1066 | /* Don't block on output, we have to drain the direct pipe. */ |
| 1067 | flags = sd->flags; |
| 1068 | sd->flags &= ~SPLICE_F_NONBLOCK; |
| 1069 | |
| 1070 | /* |
| 1071 | * We signal MORE until we've read sufficient data to fulfill the |
| 1072 | * request and we keep signalling it if the caller set it. |
| 1073 | */ |
| 1074 | more = sd->flags & SPLICE_F_MORE; |
| 1075 | sd->flags |= SPLICE_F_MORE; |
| 1076 | |
| 1077 | WARN_ON_ONCE(!pipe_is_empty(pipe)); |
| 1078 | |
| 1079 | while (len) { |
| 1080 | size_t read_len; |
| 1081 | loff_t pos = sd->pos, prev_pos = pos; |
| 1082 | |
| 1083 | ret = do_splice_read(in, &pos, pipe, len, flags); |
| 1084 | if (unlikely(ret <= 0)) |
| 1085 | goto read_failure; |
| 1086 | |
| 1087 | read_len = ret; |
| 1088 | sd->total_len = read_len; |
| 1089 | |
| 1090 | /* |
| 1091 | * If we now have sufficient data to fulfill the request then |
| 1092 | * we clear SPLICE_F_MORE if it was not set initially. |
| 1093 | */ |
| 1094 | if (read_len >= len && !more) |
| 1095 | sd->flags &= ~SPLICE_F_MORE; |
| 1096 | |
| 1097 | /* |
| 1098 | * NOTE: nonblocking mode only applies to the input. We |
| 1099 | * must not do the output in nonblocking mode as then we |
| 1100 | * could get stuck data in the internal pipe: |
| 1101 | */ |
| 1102 | ret = actor(pipe, sd); |
| 1103 | if (unlikely(ret <= 0)) { |
| 1104 | sd->pos = prev_pos; |
| 1105 | goto out_release; |
| 1106 | } |
| 1107 | |
| 1108 | bytes += ret; |
| 1109 | len -= ret; |
| 1110 | sd->pos = pos; |
| 1111 | |
| 1112 | if (ret < read_len) { |
| 1113 | sd->pos = prev_pos + ret; |
| 1114 | goto out_release; |
| 1115 | } |
| 1116 | } |
| 1117 | |
| 1118 | done: |
| 1119 | pipe->tail = pipe->head = 0; |
| 1120 | file_accessed(in); |
| 1121 | return bytes; |
| 1122 | |
| 1123 | read_failure: |
| 1124 | /* |
| 1125 | * If the user did *not* set SPLICE_F_MORE *and* we didn't hit that |
| 1126 | * "use all of len" case that cleared SPLICE_F_MORE, *and* we did a |
| 1127 | * "->splice_in()" that returned EOF (ie zero) *and* we have sent at |
| 1128 | * least 1 byte *then* we will also do the ->splice_eof() call. |
| 1129 | */ |
| 1130 | if (ret == 0 && !more && len > 0 && bytes) |
| 1131 | do_splice_eof(sd); |
| 1132 | out_release: |
| 1133 | /* |
| 1134 | * If we did an incomplete transfer we must release |
| 1135 | * the pipe buffers in question: |
| 1136 | */ |
| 1137 | for (i = 0; i < pipe->ring_size; i++) { |
| 1138 | struct pipe_buffer *buf = &pipe->bufs[i]; |
| 1139 | |
| 1140 | if (buf->ops) |
| 1141 | pipe_buf_release(pipe, buf); |
| 1142 | } |
| 1143 | |
| 1144 | if (!bytes) |
| 1145 | bytes = ret; |
| 1146 | |
| 1147 | goto done; |
| 1148 | } |
| 1149 | EXPORT_SYMBOL(splice_direct_to_actor); |
| 1150 | |
| 1151 | static int direct_splice_actor(struct pipe_inode_info *pipe, |
| 1152 | struct splice_desc *sd) |
| 1153 | { |
| 1154 | struct file *file = sd->u.file; |
| 1155 | long ret; |
| 1156 | |
| 1157 | file_start_write(file); |
| 1158 | ret = do_splice_from(pipe, file, sd->opos, sd->total_len, sd->flags); |
| 1159 | file_end_write(file); |
| 1160 | return ret; |
| 1161 | } |
| 1162 | |
| 1163 | static int splice_file_range_actor(struct pipe_inode_info *pipe, |
| 1164 | struct splice_desc *sd) |
| 1165 | { |
| 1166 | struct file *file = sd->u.file; |
| 1167 | |
| 1168 | return do_splice_from(pipe, file, sd->opos, sd->total_len, sd->flags); |
| 1169 | } |
| 1170 | |
| 1171 | static void direct_file_splice_eof(struct splice_desc *sd) |
| 1172 | { |
| 1173 | struct file *file = sd->u.file; |
| 1174 | |
| 1175 | if (file->f_op->splice_eof) |
| 1176 | file->f_op->splice_eof(file); |
| 1177 | } |
| 1178 | |
| 1179 | static ssize_t do_splice_direct_actor(struct file *in, loff_t *ppos, |
| 1180 | struct file *out, loff_t *opos, |
| 1181 | size_t len, unsigned int flags, |
| 1182 | splice_direct_actor *actor) |
| 1183 | { |
| 1184 | struct splice_desc sd = { |
| 1185 | .len = len, |
| 1186 | .total_len = len, |
| 1187 | .flags = flags, |
| 1188 | .pos = *ppos, |
| 1189 | .u.file = out, |
| 1190 | .splice_eof = direct_file_splice_eof, |
| 1191 | .opos = opos, |
| 1192 | }; |
| 1193 | ssize_t ret; |
| 1194 | |
| 1195 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
| 1196 | return -EBADF; |
| 1197 | |
| 1198 | if (unlikely(out->f_flags & O_APPEND)) |
| 1199 | return -EINVAL; |
| 1200 | |
| 1201 | ret = splice_direct_to_actor(in, &sd, actor); |
| 1202 | if (ret > 0) |
| 1203 | *ppos = sd.pos; |
| 1204 | |
| 1205 | return ret; |
| 1206 | } |
| 1207 | /** |
| 1208 | * do_splice_direct - splices data directly between two files |
| 1209 | * @in: file to splice from |
| 1210 | * @ppos: input file offset |
| 1211 | * @out: file to splice to |
| 1212 | * @opos: output file offset |
| 1213 | * @len: number of bytes to splice |
| 1214 | * @flags: splice modifier flags |
| 1215 | * |
| 1216 | * Description: |
| 1217 | * For use by do_sendfile(). splice can easily emulate sendfile, but |
| 1218 | * doing it in the application would incur an extra system call |
| 1219 | * (splice in + splice out, as compared to just sendfile()). So this helper |
| 1220 | * can splice directly through a process-private pipe. |
| 1221 | * |
| 1222 | * Callers already called rw_verify_area() on the entire range. |
| 1223 | */ |
| 1224 | ssize_t do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
| 1225 | loff_t *opos, size_t len, unsigned int flags) |
| 1226 | { |
| 1227 | return do_splice_direct_actor(in, ppos, out, opos, len, flags, |
| 1228 | direct_splice_actor); |
| 1229 | } |
| 1230 | EXPORT_SYMBOL(do_splice_direct); |
| 1231 | |
| 1232 | /** |
| 1233 | * splice_file_range - splices data between two files for copy_file_range() |
| 1234 | * @in: file to splice from |
| 1235 | * @ppos: input file offset |
| 1236 | * @out: file to splice to |
| 1237 | * @opos: output file offset |
| 1238 | * @len: number of bytes to splice |
| 1239 | * |
| 1240 | * Description: |
| 1241 | * For use by ->copy_file_range() methods. |
| 1242 | * Like do_splice_direct(), but vfs_copy_file_range() already holds |
| 1243 | * start_file_write() on @out file. |
| 1244 | * |
| 1245 | * Callers already called rw_verify_area() on the entire range. |
| 1246 | */ |
| 1247 | ssize_t splice_file_range(struct file *in, loff_t *ppos, struct file *out, |
| 1248 | loff_t *opos, size_t len) |
| 1249 | { |
| 1250 | lockdep_assert(file_write_started(out)); |
| 1251 | |
| 1252 | return do_splice_direct_actor(in, ppos, out, opos, |
| 1253 | min_t(size_t, len, MAX_RW_COUNT), |
| 1254 | 0, splice_file_range_actor); |
| 1255 | } |
| 1256 | EXPORT_SYMBOL(splice_file_range); |
| 1257 | |
| 1258 | static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags) |
| 1259 | { |
| 1260 | for (;;) { |
| 1261 | if (unlikely(!pipe->readers)) { |
| 1262 | send_sig(SIGPIPE, current, 0); |
| 1263 | return -EPIPE; |
| 1264 | } |
| 1265 | if (!pipe_is_full(pipe)) |
| 1266 | return 0; |
| 1267 | if (flags & SPLICE_F_NONBLOCK) |
| 1268 | return -EAGAIN; |
| 1269 | if (signal_pending(current)) |
| 1270 | return -ERESTARTSYS; |
| 1271 | pipe_wait_writable(pipe); |
| 1272 | } |
| 1273 | } |
| 1274 | |
| 1275 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, |
| 1276 | struct pipe_inode_info *opipe, |
| 1277 | size_t len, unsigned int flags); |
| 1278 | |
| 1279 | ssize_t splice_file_to_pipe(struct file *in, |
| 1280 | struct pipe_inode_info *opipe, |
| 1281 | loff_t *offset, |
| 1282 | size_t len, unsigned int flags) |
| 1283 | { |
| 1284 | ssize_t ret; |
| 1285 | |
| 1286 | pipe_lock(opipe); |
| 1287 | ret = wait_for_space(opipe, flags); |
| 1288 | if (!ret) |
| 1289 | ret = do_splice_read(in, offset, opipe, len, flags); |
| 1290 | pipe_unlock(opipe); |
| 1291 | if (ret > 0) |
| 1292 | wakeup_pipe_readers(opipe); |
| 1293 | return ret; |
| 1294 | } |
| 1295 | |
| 1296 | /* |
| 1297 | * Determine where to splice to/from. |
| 1298 | */ |
| 1299 | ssize_t do_splice(struct file *in, loff_t *off_in, struct file *out, |
| 1300 | loff_t *off_out, size_t len, unsigned int flags) |
| 1301 | { |
| 1302 | struct pipe_inode_info *ipipe; |
| 1303 | struct pipe_inode_info *opipe; |
| 1304 | loff_t offset; |
| 1305 | ssize_t ret; |
| 1306 | |
| 1307 | if (unlikely(!(in->f_mode & FMODE_READ) || |
| 1308 | !(out->f_mode & FMODE_WRITE))) |
| 1309 | return -EBADF; |
| 1310 | |
| 1311 | ipipe = get_pipe_info(in, true); |
| 1312 | opipe = get_pipe_info(out, true); |
| 1313 | |
| 1314 | if (ipipe && opipe) { |
| 1315 | if (off_in || off_out) |
| 1316 | return -ESPIPE; |
| 1317 | |
| 1318 | /* Splicing to self would be fun, but... */ |
| 1319 | if (ipipe == opipe) |
| 1320 | return -EINVAL; |
| 1321 | |
| 1322 | if ((in->f_flags | out->f_flags) & O_NONBLOCK) |
| 1323 | flags |= SPLICE_F_NONBLOCK; |
| 1324 | |
| 1325 | ret = splice_pipe_to_pipe(ipipe, opipe, len, flags); |
| 1326 | } else if (ipipe) { |
| 1327 | if (off_in) |
| 1328 | return -ESPIPE; |
| 1329 | if (off_out) { |
| 1330 | if (!(out->f_mode & FMODE_PWRITE)) |
| 1331 | return -EINVAL; |
| 1332 | offset = *off_out; |
| 1333 | } else { |
| 1334 | offset = out->f_pos; |
| 1335 | } |
| 1336 | |
| 1337 | if (unlikely(out->f_flags & O_APPEND)) |
| 1338 | return -EINVAL; |
| 1339 | |
| 1340 | ret = rw_verify_area(WRITE, out, &offset, len); |
| 1341 | if (unlikely(ret < 0)) |
| 1342 | return ret; |
| 1343 | |
| 1344 | if (in->f_flags & O_NONBLOCK) |
| 1345 | flags |= SPLICE_F_NONBLOCK; |
| 1346 | |
| 1347 | file_start_write(out); |
| 1348 | ret = do_splice_from(ipipe, out, &offset, len, flags); |
| 1349 | file_end_write(out); |
| 1350 | |
| 1351 | if (!off_out) |
| 1352 | out->f_pos = offset; |
| 1353 | else |
| 1354 | *off_out = offset; |
| 1355 | } else if (opipe) { |
| 1356 | if (off_out) |
| 1357 | return -ESPIPE; |
| 1358 | if (off_in) { |
| 1359 | if (!(in->f_mode & FMODE_PREAD)) |
| 1360 | return -EINVAL; |
| 1361 | offset = *off_in; |
| 1362 | } else { |
| 1363 | offset = in->f_pos; |
| 1364 | } |
| 1365 | |
| 1366 | ret = rw_verify_area(READ, in, &offset, len); |
| 1367 | if (unlikely(ret < 0)) |
| 1368 | return ret; |
| 1369 | |
| 1370 | if (out->f_flags & O_NONBLOCK) |
| 1371 | flags |= SPLICE_F_NONBLOCK; |
| 1372 | |
| 1373 | ret = splice_file_to_pipe(in, opipe, &offset, len, flags); |
| 1374 | |
| 1375 | if (!off_in) |
| 1376 | in->f_pos = offset; |
| 1377 | else |
| 1378 | *off_in = offset; |
| 1379 | } else { |
| 1380 | ret = -EINVAL; |
| 1381 | } |
| 1382 | |
| 1383 | if (ret > 0) { |
| 1384 | /* |
| 1385 | * Generate modify out before access in: |
| 1386 | * do_splice_from() may've already sent modify out, |
| 1387 | * and this ensures the events get merged. |
| 1388 | */ |
| 1389 | fsnotify_modify(out); |
| 1390 | fsnotify_access(in); |
| 1391 | } |
| 1392 | |
| 1393 | return ret; |
| 1394 | } |
| 1395 | |
| 1396 | static ssize_t __do_splice(struct file *in, loff_t __user *off_in, |
| 1397 | struct file *out, loff_t __user *off_out, |
| 1398 | size_t len, unsigned int flags) |
| 1399 | { |
| 1400 | struct pipe_inode_info *ipipe; |
| 1401 | struct pipe_inode_info *opipe; |
| 1402 | loff_t offset, *__off_in = NULL, *__off_out = NULL; |
| 1403 | ssize_t ret; |
| 1404 | |
| 1405 | ipipe = get_pipe_info(in, true); |
| 1406 | opipe = get_pipe_info(out, true); |
| 1407 | |
| 1408 | if (ipipe) { |
| 1409 | if (off_in) |
| 1410 | return -ESPIPE; |
| 1411 | pipe_clear_nowait(in); |
| 1412 | } |
| 1413 | if (opipe) { |
| 1414 | if (off_out) |
| 1415 | return -ESPIPE; |
| 1416 | pipe_clear_nowait(out); |
| 1417 | } |
| 1418 | |
| 1419 | if (off_out) { |
| 1420 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
| 1421 | return -EFAULT; |
| 1422 | __off_out = &offset; |
| 1423 | } |
| 1424 | if (off_in) { |
| 1425 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
| 1426 | return -EFAULT; |
| 1427 | __off_in = &offset; |
| 1428 | } |
| 1429 | |
| 1430 | ret = do_splice(in, __off_in, out, __off_out, len, flags); |
| 1431 | if (ret < 0) |
| 1432 | return ret; |
| 1433 | |
| 1434 | if (__off_out && copy_to_user(off_out, __off_out, sizeof(loff_t))) |
| 1435 | return -EFAULT; |
| 1436 | if (__off_in && copy_to_user(off_in, __off_in, sizeof(loff_t))) |
| 1437 | return -EFAULT; |
| 1438 | |
| 1439 | return ret; |
| 1440 | } |
| 1441 | |
| 1442 | static ssize_t iter_to_pipe(struct iov_iter *from, |
| 1443 | struct pipe_inode_info *pipe, |
| 1444 | unsigned int flags) |
| 1445 | { |
| 1446 | struct pipe_buffer buf = { |
| 1447 | .ops = &user_page_pipe_buf_ops, |
| 1448 | .flags = flags |
| 1449 | }; |
| 1450 | size_t total = 0; |
| 1451 | ssize_t ret = 0; |
| 1452 | |
| 1453 | while (iov_iter_count(from)) { |
| 1454 | struct page *pages[16]; |
| 1455 | ssize_t left; |
| 1456 | size_t start; |
| 1457 | int i, n; |
| 1458 | |
| 1459 | left = iov_iter_get_pages2(from, pages, ~0UL, 16, &start); |
| 1460 | if (left <= 0) { |
| 1461 | ret = left; |
| 1462 | break; |
| 1463 | } |
| 1464 | |
| 1465 | n = DIV_ROUND_UP(left + start, PAGE_SIZE); |
| 1466 | for (i = 0; i < n; i++) { |
| 1467 | int size = min_t(int, left, PAGE_SIZE - start); |
| 1468 | |
| 1469 | buf.page = pages[i]; |
| 1470 | buf.offset = start; |
| 1471 | buf.len = size; |
| 1472 | ret = add_to_pipe(pipe, &buf); |
| 1473 | if (unlikely(ret < 0)) { |
| 1474 | iov_iter_revert(from, left); |
| 1475 | // this one got dropped by add_to_pipe() |
| 1476 | while (++i < n) |
| 1477 | put_page(pages[i]); |
| 1478 | goto out; |
| 1479 | } |
| 1480 | total += ret; |
| 1481 | left -= size; |
| 1482 | start = 0; |
| 1483 | } |
| 1484 | } |
| 1485 | out: |
| 1486 | return total ? total : ret; |
| 1487 | } |
| 1488 | |
| 1489 | static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
| 1490 | struct splice_desc *sd) |
| 1491 | { |
| 1492 | int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data); |
| 1493 | return n == sd->len ? n : -EFAULT; |
| 1494 | } |
| 1495 | |
| 1496 | /* |
| 1497 | * For lack of a better implementation, implement vmsplice() to userspace |
| 1498 | * as a simple copy of the pipes pages to the user iov. |
| 1499 | */ |
| 1500 | static ssize_t vmsplice_to_user(struct file *file, struct iov_iter *iter, |
| 1501 | unsigned int flags) |
| 1502 | { |
| 1503 | struct pipe_inode_info *pipe = get_pipe_info(file, true); |
| 1504 | struct splice_desc sd = { |
| 1505 | .total_len = iov_iter_count(iter), |
| 1506 | .flags = flags, |
| 1507 | .u.data = iter |
| 1508 | }; |
| 1509 | ssize_t ret = 0; |
| 1510 | |
| 1511 | if (!pipe) |
| 1512 | return -EBADF; |
| 1513 | |
| 1514 | pipe_clear_nowait(file); |
| 1515 | |
| 1516 | if (sd.total_len) { |
| 1517 | pipe_lock(pipe); |
| 1518 | ret = __splice_from_pipe(pipe, &sd, pipe_to_user); |
| 1519 | pipe_unlock(pipe); |
| 1520 | } |
| 1521 | |
| 1522 | if (ret > 0) |
| 1523 | fsnotify_access(file); |
| 1524 | |
| 1525 | return ret; |
| 1526 | } |
| 1527 | |
| 1528 | /* |
| 1529 | * vmsplice splices a user address range into a pipe. It can be thought of |
| 1530 | * as splice-from-memory, where the regular splice is splice-from-file (or |
| 1531 | * to file). In both cases the output is a pipe, naturally. |
| 1532 | */ |
| 1533 | static ssize_t vmsplice_to_pipe(struct file *file, struct iov_iter *iter, |
| 1534 | unsigned int flags) |
| 1535 | { |
| 1536 | struct pipe_inode_info *pipe; |
| 1537 | ssize_t ret = 0; |
| 1538 | unsigned buf_flag = 0; |
| 1539 | |
| 1540 | if (flags & SPLICE_F_GIFT) |
| 1541 | buf_flag = PIPE_BUF_FLAG_GIFT; |
| 1542 | |
| 1543 | pipe = get_pipe_info(file, true); |
| 1544 | if (!pipe) |
| 1545 | return -EBADF; |
| 1546 | |
| 1547 | pipe_clear_nowait(file); |
| 1548 | |
| 1549 | pipe_lock(pipe); |
| 1550 | ret = wait_for_space(pipe, flags); |
| 1551 | if (!ret) |
| 1552 | ret = iter_to_pipe(iter, pipe, buf_flag); |
| 1553 | pipe_unlock(pipe); |
| 1554 | if (ret > 0) { |
| 1555 | wakeup_pipe_readers(pipe); |
| 1556 | fsnotify_modify(file); |
| 1557 | } |
| 1558 | return ret; |
| 1559 | } |
| 1560 | |
| 1561 | /* |
| 1562 | * Note that vmsplice only really supports true splicing _from_ user memory |
| 1563 | * to a pipe, not the other way around. Splicing from user memory is a simple |
| 1564 | * operation that can be supported without any funky alignment restrictions |
| 1565 | * or nasty vm tricks. We simply map in the user memory and fill them into |
| 1566 | * a pipe. The reverse isn't quite as easy, though. There are two possible |
| 1567 | * solutions for that: |
| 1568 | * |
| 1569 | * - memcpy() the data internally, at which point we might as well just |
| 1570 | * do a regular read() on the buffer anyway. |
| 1571 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it |
| 1572 | * has restriction limitations on both ends of the pipe). |
| 1573 | * |
| 1574 | * Currently we punt and implement it as a normal copy, see pipe_to_user(). |
| 1575 | * |
| 1576 | */ |
| 1577 | SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, uiov, |
| 1578 | unsigned long, nr_segs, unsigned int, flags) |
| 1579 | { |
| 1580 | struct iovec iovstack[UIO_FASTIOV]; |
| 1581 | struct iovec *iov = iovstack; |
| 1582 | struct iov_iter iter; |
| 1583 | ssize_t error; |
| 1584 | int type; |
| 1585 | |
| 1586 | if (unlikely(flags & ~SPLICE_F_ALL)) |
| 1587 | return -EINVAL; |
| 1588 | |
| 1589 | CLASS(fd, f)(fd); |
| 1590 | if (fd_empty(f)) |
| 1591 | return -EBADF; |
| 1592 | if (fd_file(f)->f_mode & FMODE_WRITE) |
| 1593 | type = ITER_SOURCE; |
| 1594 | else if (fd_file(f)->f_mode & FMODE_READ) |
| 1595 | type = ITER_DEST; |
| 1596 | else |
| 1597 | return -EBADF; |
| 1598 | |
| 1599 | error = import_iovec(type, uiov, nr_segs, |
| 1600 | ARRAY_SIZE(iovstack), &iov, &iter); |
| 1601 | if (error < 0) |
| 1602 | return error; |
| 1603 | |
| 1604 | if (!iov_iter_count(&iter)) |
| 1605 | error = 0; |
| 1606 | else if (type == ITER_SOURCE) |
| 1607 | error = vmsplice_to_pipe(fd_file(f), &iter, flags); |
| 1608 | else |
| 1609 | error = vmsplice_to_user(fd_file(f), &iter, flags); |
| 1610 | |
| 1611 | kfree(iov); |
| 1612 | return error; |
| 1613 | } |
| 1614 | |
| 1615 | SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, |
| 1616 | int, fd_out, loff_t __user *, off_out, |
| 1617 | size_t, len, unsigned int, flags) |
| 1618 | { |
| 1619 | if (unlikely(!len)) |
| 1620 | return 0; |
| 1621 | |
| 1622 | if (unlikely(flags & ~SPLICE_F_ALL)) |
| 1623 | return -EINVAL; |
| 1624 | |
| 1625 | CLASS(fd, in)(fd_in); |
| 1626 | if (fd_empty(in)) |
| 1627 | return -EBADF; |
| 1628 | |
| 1629 | CLASS(fd, out)(fd_out); |
| 1630 | if (fd_empty(out)) |
| 1631 | return -EBADF; |
| 1632 | |
| 1633 | return __do_splice(fd_file(in), off_in, fd_file(out), off_out, |
| 1634 | len, flags); |
| 1635 | } |
| 1636 | |
| 1637 | /* |
| 1638 | * Make sure there's data to read. Wait for input if we can, otherwise |
| 1639 | * return an appropriate error. |
| 1640 | */ |
| 1641 | static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
| 1642 | { |
| 1643 | int ret; |
| 1644 | |
| 1645 | /* |
| 1646 | * Check the pipe occupancy without the inode lock first. This function |
| 1647 | * is speculative anyways, so missing one is ok. |
| 1648 | */ |
| 1649 | if (!pipe_is_empty(pipe)) |
| 1650 | return 0; |
| 1651 | |
| 1652 | ret = 0; |
| 1653 | pipe_lock(pipe); |
| 1654 | |
| 1655 | while (pipe_is_empty(pipe)) { |
| 1656 | if (signal_pending(current)) { |
| 1657 | ret = -ERESTARTSYS; |
| 1658 | break; |
| 1659 | } |
| 1660 | if (!pipe->writers) |
| 1661 | break; |
| 1662 | if (flags & SPLICE_F_NONBLOCK) { |
| 1663 | ret = -EAGAIN; |
| 1664 | break; |
| 1665 | } |
| 1666 | pipe_wait_readable(pipe); |
| 1667 | } |
| 1668 | |
| 1669 | pipe_unlock(pipe); |
| 1670 | return ret; |
| 1671 | } |
| 1672 | |
| 1673 | /* |
| 1674 | * Make sure there's writeable room. Wait for room if we can, otherwise |
| 1675 | * return an appropriate error. |
| 1676 | */ |
| 1677 | static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
| 1678 | { |
| 1679 | int ret; |
| 1680 | |
| 1681 | /* |
| 1682 | * Check pipe occupancy without the inode lock first. This function |
| 1683 | * is speculative anyways, so missing one is ok. |
| 1684 | */ |
| 1685 | if (!pipe_is_full(pipe)) |
| 1686 | return 0; |
| 1687 | |
| 1688 | ret = 0; |
| 1689 | pipe_lock(pipe); |
| 1690 | |
| 1691 | while (pipe_is_full(pipe)) { |
| 1692 | if (!pipe->readers) { |
| 1693 | send_sig(SIGPIPE, current, 0); |
| 1694 | ret = -EPIPE; |
| 1695 | break; |
| 1696 | } |
| 1697 | if (flags & SPLICE_F_NONBLOCK) { |
| 1698 | ret = -EAGAIN; |
| 1699 | break; |
| 1700 | } |
| 1701 | if (signal_pending(current)) { |
| 1702 | ret = -ERESTARTSYS; |
| 1703 | break; |
| 1704 | } |
| 1705 | pipe_wait_writable(pipe); |
| 1706 | } |
| 1707 | |
| 1708 | pipe_unlock(pipe); |
| 1709 | return ret; |
| 1710 | } |
| 1711 | |
| 1712 | /* |
| 1713 | * Splice contents of ipipe to opipe. |
| 1714 | */ |
| 1715 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, |
| 1716 | struct pipe_inode_info *opipe, |
| 1717 | size_t len, unsigned int flags) |
| 1718 | { |
| 1719 | struct pipe_buffer *ibuf, *obuf; |
| 1720 | unsigned int i_head, o_head; |
| 1721 | unsigned int i_tail, o_tail; |
| 1722 | int ret = 0; |
| 1723 | bool input_wakeup = false; |
| 1724 | |
| 1725 | |
| 1726 | retry: |
| 1727 | ret = ipipe_prep(ipipe, flags); |
| 1728 | if (ret) |
| 1729 | return ret; |
| 1730 | |
| 1731 | ret = opipe_prep(opipe, flags); |
| 1732 | if (ret) |
| 1733 | return ret; |
| 1734 | |
| 1735 | /* |
| 1736 | * Potential ABBA deadlock, work around it by ordering lock |
| 1737 | * grabbing by pipe info address. Otherwise two different processes |
| 1738 | * could deadlock (one doing tee from A -> B, the other from B -> A). |
| 1739 | */ |
| 1740 | pipe_double_lock(ipipe, opipe); |
| 1741 | |
| 1742 | i_tail = ipipe->tail; |
| 1743 | o_head = opipe->head; |
| 1744 | |
| 1745 | do { |
| 1746 | size_t o_len; |
| 1747 | |
| 1748 | if (!opipe->readers) { |
| 1749 | send_sig(SIGPIPE, current, 0); |
| 1750 | if (!ret) |
| 1751 | ret = -EPIPE; |
| 1752 | break; |
| 1753 | } |
| 1754 | |
| 1755 | i_head = ipipe->head; |
| 1756 | o_tail = opipe->tail; |
| 1757 | |
| 1758 | if (pipe_empty(i_head, i_tail) && !ipipe->writers) |
| 1759 | break; |
| 1760 | |
| 1761 | /* |
| 1762 | * Cannot make any progress, because either the input |
| 1763 | * pipe is empty or the output pipe is full. |
| 1764 | */ |
| 1765 | if (pipe_empty(i_head, i_tail) || |
| 1766 | pipe_full(o_head, o_tail, opipe->max_usage)) { |
| 1767 | /* Already processed some buffers, break */ |
| 1768 | if (ret) |
| 1769 | break; |
| 1770 | |
| 1771 | if (flags & SPLICE_F_NONBLOCK) { |
| 1772 | ret = -EAGAIN; |
| 1773 | break; |
| 1774 | } |
| 1775 | |
| 1776 | /* |
| 1777 | * We raced with another reader/writer and haven't |
| 1778 | * managed to process any buffers. A zero return |
| 1779 | * value means EOF, so retry instead. |
| 1780 | */ |
| 1781 | pipe_unlock(ipipe); |
| 1782 | pipe_unlock(opipe); |
| 1783 | goto retry; |
| 1784 | } |
| 1785 | |
| 1786 | ibuf = pipe_buf(ipipe, i_tail); |
| 1787 | obuf = pipe_buf(opipe, o_head); |
| 1788 | |
| 1789 | if (len >= ibuf->len) { |
| 1790 | /* |
| 1791 | * Simply move the whole buffer from ipipe to opipe |
| 1792 | */ |
| 1793 | *obuf = *ibuf; |
| 1794 | ibuf->ops = NULL; |
| 1795 | i_tail++; |
| 1796 | ipipe->tail = i_tail; |
| 1797 | input_wakeup = true; |
| 1798 | o_len = obuf->len; |
| 1799 | o_head++; |
| 1800 | opipe->head = o_head; |
| 1801 | } else { |
| 1802 | /* |
| 1803 | * Get a reference to this pipe buffer, |
| 1804 | * so we can copy the contents over. |
| 1805 | */ |
| 1806 | if (!pipe_buf_get(ipipe, ibuf)) { |
| 1807 | if (ret == 0) |
| 1808 | ret = -EFAULT; |
| 1809 | break; |
| 1810 | } |
| 1811 | *obuf = *ibuf; |
| 1812 | |
| 1813 | /* |
| 1814 | * Don't inherit the gift and merge flags, we need to |
| 1815 | * prevent multiple steals of this page. |
| 1816 | */ |
| 1817 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
| 1818 | obuf->flags &= ~PIPE_BUF_FLAG_CAN_MERGE; |
| 1819 | |
| 1820 | obuf->len = len; |
| 1821 | ibuf->offset += len; |
| 1822 | ibuf->len -= len; |
| 1823 | o_len = len; |
| 1824 | o_head++; |
| 1825 | opipe->head = o_head; |
| 1826 | } |
| 1827 | ret += o_len; |
| 1828 | len -= o_len; |
| 1829 | } while (len); |
| 1830 | |
| 1831 | pipe_unlock(ipipe); |
| 1832 | pipe_unlock(opipe); |
| 1833 | |
| 1834 | /* |
| 1835 | * If we put data in the output pipe, wakeup any potential readers. |
| 1836 | */ |
| 1837 | if (ret > 0) |
| 1838 | wakeup_pipe_readers(opipe); |
| 1839 | |
| 1840 | if (input_wakeup) |
| 1841 | wakeup_pipe_writers(ipipe); |
| 1842 | |
| 1843 | return ret; |
| 1844 | } |
| 1845 | |
| 1846 | /* |
| 1847 | * Link contents of ipipe to opipe. |
| 1848 | */ |
| 1849 | static ssize_t link_pipe(struct pipe_inode_info *ipipe, |
| 1850 | struct pipe_inode_info *opipe, |
| 1851 | size_t len, unsigned int flags) |
| 1852 | { |
| 1853 | struct pipe_buffer *ibuf, *obuf; |
| 1854 | unsigned int i_head, o_head; |
| 1855 | unsigned int i_tail, o_tail; |
| 1856 | ssize_t ret = 0; |
| 1857 | |
| 1858 | /* |
| 1859 | * Potential ABBA deadlock, work around it by ordering lock |
| 1860 | * grabbing by pipe info address. Otherwise two different processes |
| 1861 | * could deadlock (one doing tee from A -> B, the other from B -> A). |
| 1862 | */ |
| 1863 | pipe_double_lock(ipipe, opipe); |
| 1864 | |
| 1865 | i_tail = ipipe->tail; |
| 1866 | o_head = opipe->head; |
| 1867 | |
| 1868 | do { |
| 1869 | if (!opipe->readers) { |
| 1870 | send_sig(SIGPIPE, current, 0); |
| 1871 | if (!ret) |
| 1872 | ret = -EPIPE; |
| 1873 | break; |
| 1874 | } |
| 1875 | |
| 1876 | i_head = ipipe->head; |
| 1877 | o_tail = opipe->tail; |
| 1878 | |
| 1879 | /* |
| 1880 | * If we have iterated all input buffers or run out of |
| 1881 | * output room, break. |
| 1882 | */ |
| 1883 | if (pipe_empty(i_head, i_tail) || |
| 1884 | pipe_full(o_head, o_tail, opipe->max_usage)) |
| 1885 | break; |
| 1886 | |
| 1887 | ibuf = pipe_buf(ipipe, i_tail); |
| 1888 | obuf = pipe_buf(opipe, o_head); |
| 1889 | |
| 1890 | /* |
| 1891 | * Get a reference to this pipe buffer, |
| 1892 | * so we can copy the contents over. |
| 1893 | */ |
| 1894 | if (!pipe_buf_get(ipipe, ibuf)) { |
| 1895 | if (ret == 0) |
| 1896 | ret = -EFAULT; |
| 1897 | break; |
| 1898 | } |
| 1899 | |
| 1900 | *obuf = *ibuf; |
| 1901 | |
| 1902 | /* |
| 1903 | * Don't inherit the gift and merge flag, we need to prevent |
| 1904 | * multiple steals of this page. |
| 1905 | */ |
| 1906 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
| 1907 | obuf->flags &= ~PIPE_BUF_FLAG_CAN_MERGE; |
| 1908 | |
| 1909 | if (obuf->len > len) |
| 1910 | obuf->len = len; |
| 1911 | ret += obuf->len; |
| 1912 | len -= obuf->len; |
| 1913 | |
| 1914 | o_head++; |
| 1915 | opipe->head = o_head; |
| 1916 | i_tail++; |
| 1917 | } while (len); |
| 1918 | |
| 1919 | pipe_unlock(ipipe); |
| 1920 | pipe_unlock(opipe); |
| 1921 | |
| 1922 | /* |
| 1923 | * If we put data in the output pipe, wakeup any potential readers. |
| 1924 | */ |
| 1925 | if (ret > 0) |
| 1926 | wakeup_pipe_readers(opipe); |
| 1927 | |
| 1928 | return ret; |
| 1929 | } |
| 1930 | |
| 1931 | /* |
| 1932 | * This is a tee(1) implementation that works on pipes. It doesn't copy |
| 1933 | * any data, it simply references the 'in' pages on the 'out' pipe. |
| 1934 | * The 'flags' used are the SPLICE_F_* variants, currently the only |
| 1935 | * applicable one is SPLICE_F_NONBLOCK. |
| 1936 | */ |
| 1937 | ssize_t do_tee(struct file *in, struct file *out, size_t len, |
| 1938 | unsigned int flags) |
| 1939 | { |
| 1940 | struct pipe_inode_info *ipipe = get_pipe_info(in, true); |
| 1941 | struct pipe_inode_info *opipe = get_pipe_info(out, true); |
| 1942 | ssize_t ret = -EINVAL; |
| 1943 | |
| 1944 | if (unlikely(!(in->f_mode & FMODE_READ) || |
| 1945 | !(out->f_mode & FMODE_WRITE))) |
| 1946 | return -EBADF; |
| 1947 | |
| 1948 | /* |
| 1949 | * Duplicate the contents of ipipe to opipe without actually |
| 1950 | * copying the data. |
| 1951 | */ |
| 1952 | if (ipipe && opipe && ipipe != opipe) { |
| 1953 | if ((in->f_flags | out->f_flags) & O_NONBLOCK) |
| 1954 | flags |= SPLICE_F_NONBLOCK; |
| 1955 | |
| 1956 | /* |
| 1957 | * Keep going, unless we encounter an error. The ipipe/opipe |
| 1958 | * ordering doesn't really matter. |
| 1959 | */ |
| 1960 | ret = ipipe_prep(ipipe, flags); |
| 1961 | if (!ret) { |
| 1962 | ret = opipe_prep(opipe, flags); |
| 1963 | if (!ret) |
| 1964 | ret = link_pipe(ipipe, opipe, len, flags); |
| 1965 | } |
| 1966 | } |
| 1967 | |
| 1968 | if (ret > 0) { |
| 1969 | fsnotify_access(in); |
| 1970 | fsnotify_modify(out); |
| 1971 | } |
| 1972 | |
| 1973 | return ret; |
| 1974 | } |
| 1975 | |
| 1976 | SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) |
| 1977 | { |
| 1978 | if (unlikely(flags & ~SPLICE_F_ALL)) |
| 1979 | return -EINVAL; |
| 1980 | |
| 1981 | if (unlikely(!len)) |
| 1982 | return 0; |
| 1983 | |
| 1984 | CLASS(fd, in)(fdin); |
| 1985 | if (fd_empty(in)) |
| 1986 | return -EBADF; |
| 1987 | |
| 1988 | CLASS(fd, out)(fdout); |
| 1989 | if (fd_empty(out)) |
| 1990 | return -EBADF; |
| 1991 | |
| 1992 | return do_tee(fd_file(in), fd_file(out), len, flags); |
| 1993 | } |