1 | // SPDX-License-Identifier: GPL-2.0 |
---|---|
2 | /* |
3 | * linux/fs/pipe.c |
4 | * |
5 | * Copyright (C) 1991, 1992, 1999 Linus Torvalds |
6 | */ |
7 | |
8 | #include <linux/mm.h> |
9 | #include <linux/file.h> |
10 | #include <linux/poll.h> |
11 | #include <linux/slab.h> |
12 | #include <linux/module.h> |
13 | #include <linux/init.h> |
14 | #include <linux/fs.h> |
15 | #include <linux/log2.h> |
16 | #include <linux/mount.h> |
17 | #include <linux/pseudo_fs.h> |
18 | #include <linux/magic.h> |
19 | #include <linux/pipe_fs_i.h> |
20 | #include <linux/uio.h> |
21 | #include <linux/highmem.h> |
22 | #include <linux/pagemap.h> |
23 | #include <linux/audit.h> |
24 | #include <linux/syscalls.h> |
25 | #include <linux/fcntl.h> |
26 | #include <linux/memcontrol.h> |
27 | #include <linux/watch_queue.h> |
28 | #include <linux/sysctl.h> |
29 | #include <linux/sort.h> |
30 | |
31 | #include <linux/uaccess.h> |
32 | #include <asm/ioctls.h> |
33 | |
34 | #include "internal.h" |
35 | |
36 | /* |
37 | * New pipe buffers will be restricted to this size while the user is exceeding |
38 | * their pipe buffer quota. The general pipe use case needs at least two |
39 | * buffers: one for data yet to be read, and one for new data. If this is less |
40 | * than two, then a write to a non-empty pipe may block even if the pipe is not |
41 | * full. This can occur with GNU make jobserver or similar uses of pipes as |
42 | * semaphores: multiple processes may be waiting to write tokens back to the |
43 | * pipe before reading tokens: https://lore.kernel.org/lkml/1628086770.5rn8p04n6j.none@localhost/. |
44 | * |
45 | * Users can reduce their pipe buffers with F_SETPIPE_SZ below this at their |
46 | * own risk, namely: pipe writes to non-full pipes may block until the pipe is |
47 | * emptied. |
48 | */ |
49 | #define PIPE_MIN_DEF_BUFFERS 2 |
50 | |
51 | /* |
52 | * The max size that a non-root user is allowed to grow the pipe. Can |
53 | * be set by root in /proc/sys/fs/pipe-max-size |
54 | */ |
55 | static unsigned int pipe_max_size = 1048576; |
56 | |
57 | /* Maximum allocatable pages per user. Hard limit is unset by default, soft |
58 | * matches default values. |
59 | */ |
60 | static unsigned long pipe_user_pages_hard; |
61 | static unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR; |
62 | |
63 | /* |
64 | * We use head and tail indices that aren't masked off, except at the point of |
65 | * dereference, but rather they're allowed to wrap naturally. This means there |
66 | * isn't a dead spot in the buffer, but the ring has to be a power of two and |
67 | * <= 2^31. |
68 | * -- David Howells 2019-09-23. |
69 | * |
70 | * Reads with count = 0 should always return 0. |
71 | * -- Julian Bradfield 1999-06-07. |
72 | * |
73 | * FIFOs and Pipes now generate SIGIO for both readers and writers. |
74 | * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16 |
75 | * |
76 | * pipe_read & write cleanup |
77 | * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09 |
78 | */ |
79 | |
80 | #ifdef CONFIG_PROVE_LOCKING |
81 | static int pipe_lock_cmp_fn(const struct lockdep_map *a, |
82 | const struct lockdep_map *b) |
83 | { |
84 | return cmp_int((unsigned long) a, (unsigned long) b); |
85 | } |
86 | #endif |
87 | |
88 | void pipe_lock(struct pipe_inode_info *pipe) |
89 | { |
90 | if (pipe->files) |
91 | mutex_lock(&pipe->mutex); |
92 | } |
93 | EXPORT_SYMBOL(pipe_lock); |
94 | |
95 | void pipe_unlock(struct pipe_inode_info *pipe) |
96 | { |
97 | if (pipe->files) |
98 | mutex_unlock(lock: &pipe->mutex); |
99 | } |
100 | EXPORT_SYMBOL(pipe_unlock); |
101 | |
102 | void pipe_double_lock(struct pipe_inode_info *pipe1, |
103 | struct pipe_inode_info *pipe2) |
104 | { |
105 | BUG_ON(pipe1 == pipe2); |
106 | |
107 | if (pipe1 > pipe2) |
108 | swap(pipe1, pipe2); |
109 | |
110 | pipe_lock(pipe1); |
111 | pipe_lock(pipe2); |
112 | } |
113 | |
114 | static struct page *anon_pipe_get_page(struct pipe_inode_info *pipe) |
115 | { |
116 | for (int i = 0; i < ARRAY_SIZE(pipe->tmp_page); i++) { |
117 | if (pipe->tmp_page[i]) { |
118 | struct page *page = pipe->tmp_page[i]; |
119 | pipe->tmp_page[i] = NULL; |
120 | return page; |
121 | } |
122 | } |
123 | |
124 | return alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT); |
125 | } |
126 | |
127 | static void anon_pipe_put_page(struct pipe_inode_info *pipe, |
128 | struct page *page) |
129 | { |
130 | if (page_count(page) == 1) { |
131 | for (int i = 0; i < ARRAY_SIZE(pipe->tmp_page); i++) { |
132 | if (!pipe->tmp_page[i]) { |
133 | pipe->tmp_page[i] = page; |
134 | return; |
135 | } |
136 | } |
137 | } |
138 | |
139 | put_page(page); |
140 | } |
141 | |
142 | static void anon_pipe_buf_release(struct pipe_inode_info *pipe, |
143 | struct pipe_buffer *buf) |
144 | { |
145 | struct page *page = buf->page; |
146 | |
147 | anon_pipe_put_page(pipe, page); |
148 | } |
149 | |
150 | static bool anon_pipe_buf_try_steal(struct pipe_inode_info *pipe, |
151 | struct pipe_buffer *buf) |
152 | { |
153 | struct page *page = buf->page; |
154 | |
155 | if (page_count(page) != 1) |
156 | return false; |
157 | memcg_kmem_uncharge_page(page, order: 0); |
158 | __SetPageLocked(page); |
159 | return true; |
160 | } |
161 | |
162 | /** |
163 | * generic_pipe_buf_try_steal - attempt to take ownership of a &pipe_buffer |
164 | * @pipe: the pipe that the buffer belongs to |
165 | * @buf: the buffer to attempt to steal |
166 | * |
167 | * Description: |
168 | * This function attempts to steal the &struct page attached to |
169 | * @buf. If successful, this function returns 0 and returns with |
170 | * the page locked. The caller may then reuse the page for whatever |
171 | * he wishes; the typical use is insertion into a different file |
172 | * page cache. |
173 | */ |
174 | bool generic_pipe_buf_try_steal(struct pipe_inode_info *pipe, |
175 | struct pipe_buffer *buf) |
176 | { |
177 | struct page *page = buf->page; |
178 | |
179 | /* |
180 | * A reference of one is golden, that means that the owner of this |
181 | * page is the only one holding a reference to it. lock the page |
182 | * and return OK. |
183 | */ |
184 | if (page_count(page) == 1) { |
185 | lock_page(page); |
186 | return true; |
187 | } |
188 | return false; |
189 | } |
190 | EXPORT_SYMBOL(generic_pipe_buf_try_steal); |
191 | |
192 | /** |
193 | * generic_pipe_buf_get - get a reference to a &struct pipe_buffer |
194 | * @pipe: the pipe that the buffer belongs to |
195 | * @buf: the buffer to get a reference to |
196 | * |
197 | * Description: |
198 | * This function grabs an extra reference to @buf. It's used in |
199 | * the tee() system call, when we duplicate the buffers in one |
200 | * pipe into another. |
201 | */ |
202 | bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) |
203 | { |
204 | return try_get_page(page: buf->page); |
205 | } |
206 | EXPORT_SYMBOL(generic_pipe_buf_get); |
207 | |
208 | /** |
209 | * generic_pipe_buf_release - put a reference to a &struct pipe_buffer |
210 | * @pipe: the pipe that the buffer belongs to |
211 | * @buf: the buffer to put a reference to |
212 | * |
213 | * Description: |
214 | * This function releases a reference to @buf. |
215 | */ |
216 | void generic_pipe_buf_release(struct pipe_inode_info *pipe, |
217 | struct pipe_buffer *buf) |
218 | { |
219 | put_page(page: buf->page); |
220 | } |
221 | EXPORT_SYMBOL(generic_pipe_buf_release); |
222 | |
223 | static const struct pipe_buf_operations anon_pipe_buf_ops = { |
224 | .release = anon_pipe_buf_release, |
225 | .try_steal = anon_pipe_buf_try_steal, |
226 | .get = generic_pipe_buf_get, |
227 | }; |
228 | |
229 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
230 | static inline bool pipe_readable(const struct pipe_inode_info *pipe) |
231 | { |
232 | union pipe_index idx = { .head_tail = READ_ONCE(pipe->head_tail) }; |
233 | unsigned int writers = READ_ONCE(pipe->writers); |
234 | |
235 | return !pipe_empty(head: idx.head, tail: idx.tail) || !writers; |
236 | } |
237 | |
238 | static inline unsigned int pipe_update_tail(struct pipe_inode_info *pipe, |
239 | struct pipe_buffer *buf, |
240 | unsigned int tail) |
241 | { |
242 | pipe_buf_release(pipe, buf); |
243 | |
244 | /* |
245 | * If the pipe has a watch_queue, we need additional protection |
246 | * by the spinlock because notifications get posted with only |
247 | * this spinlock, no mutex |
248 | */ |
249 | if (pipe_has_watch_queue(pipe)) { |
250 | spin_lock_irq(lock: &pipe->rd_wait.lock); |
251 | #ifdef CONFIG_WATCH_QUEUE |
252 | if (buf->flags & PIPE_BUF_FLAG_LOSS) |
253 | pipe->note_loss = true; |
254 | #endif |
255 | pipe->tail = ++tail; |
256 | spin_unlock_irq(lock: &pipe->rd_wait.lock); |
257 | return tail; |
258 | } |
259 | |
260 | /* |
261 | * Without a watch_queue, we can simply increment the tail |
262 | * without the spinlock - the mutex is enough. |
263 | */ |
264 | pipe->tail = ++tail; |
265 | return tail; |
266 | } |
267 | |
268 | static ssize_t |
269 | anon_pipe_read(struct kiocb *iocb, struct iov_iter *to) |
270 | { |
271 | size_t total_len = iov_iter_count(i: to); |
272 | struct file *filp = iocb->ki_filp; |
273 | struct pipe_inode_info *pipe = filp->private_data; |
274 | bool wake_writer = false, wake_next_reader = false; |
275 | ssize_t ret; |
276 | |
277 | /* Null read succeeds. */ |
278 | if (unlikely(total_len == 0)) |
279 | return 0; |
280 | |
281 | ret = 0; |
282 | mutex_lock(&pipe->mutex); |
283 | |
284 | /* |
285 | * We only wake up writers if the pipe was full when we started reading |
286 | * and it is no longer full after reading to avoid unnecessary wakeups. |
287 | * |
288 | * But when we do wake up writers, we do so using a sync wakeup |
289 | * (WF_SYNC), because we want them to get going and generate more |
290 | * data for us. |
291 | */ |
292 | for (;;) { |
293 | /* Read ->head with a barrier vs post_one_notification() */ |
294 | unsigned int head = smp_load_acquire(&pipe->head); |
295 | unsigned int tail = pipe->tail; |
296 | |
297 | #ifdef CONFIG_WATCH_QUEUE |
298 | if (pipe->note_loss) { |
299 | struct watch_notification n; |
300 | |
301 | if (total_len < 8) { |
302 | if (ret == 0) |
303 | ret = -ENOBUFS; |
304 | break; |
305 | } |
306 | |
307 | n.type = WATCH_TYPE_META; |
308 | n.subtype = WATCH_META_LOSS_NOTIFICATION; |
309 | n.info = watch_sizeof(n); |
310 | if (copy_to_iter(addr: &n, bytes: sizeof(n), i: to) != sizeof(n)) { |
311 | if (ret == 0) |
312 | ret = -EFAULT; |
313 | break; |
314 | } |
315 | ret += sizeof(n); |
316 | total_len -= sizeof(n); |
317 | pipe->note_loss = false; |
318 | } |
319 | #endif |
320 | |
321 | if (!pipe_empty(head, tail)) { |
322 | struct pipe_buffer *buf = pipe_buf(pipe, slot: tail); |
323 | size_t chars = buf->len; |
324 | size_t written; |
325 | int error; |
326 | |
327 | if (chars > total_len) { |
328 | if (buf->flags & PIPE_BUF_FLAG_WHOLE) { |
329 | if (ret == 0) |
330 | ret = -ENOBUFS; |
331 | break; |
332 | } |
333 | chars = total_len; |
334 | } |
335 | |
336 | error = pipe_buf_confirm(pipe, buf); |
337 | if (error) { |
338 | if (!ret) |
339 | ret = error; |
340 | break; |
341 | } |
342 | |
343 | written = copy_page_to_iter(page: buf->page, offset: buf->offset, bytes: chars, i: to); |
344 | if (unlikely(written < chars)) { |
345 | if (!ret) |
346 | ret = -EFAULT; |
347 | break; |
348 | } |
349 | ret += chars; |
350 | buf->offset += chars; |
351 | buf->len -= chars; |
352 | |
353 | /* Was it a packet buffer? Clean up and exit */ |
354 | if (buf->flags & PIPE_BUF_FLAG_PACKET) { |
355 | total_len = chars; |
356 | buf->len = 0; |
357 | } |
358 | |
359 | if (!buf->len) { |
360 | wake_writer |= pipe_full(head, tail, limit: pipe->max_usage); |
361 | tail = pipe_update_tail(pipe, buf, tail); |
362 | } |
363 | total_len -= chars; |
364 | if (!total_len) |
365 | break; /* common path: read succeeded */ |
366 | if (!pipe_empty(head, tail)) /* More to do? */ |
367 | continue; |
368 | } |
369 | |
370 | if (!pipe->writers) |
371 | break; |
372 | if (ret) |
373 | break; |
374 | if ((filp->f_flags & O_NONBLOCK) || |
375 | (iocb->ki_flags & IOCB_NOWAIT)) { |
376 | ret = -EAGAIN; |
377 | break; |
378 | } |
379 | mutex_unlock(lock: &pipe->mutex); |
380 | /* |
381 | * We only get here if we didn't actually read anything. |
382 | * |
383 | * But because we didn't read anything, at this point we can |
384 | * just return directly with -ERESTARTSYS if we're interrupted, |
385 | * since we've done any required wakeups and there's no need |
386 | * to mark anything accessed. And we've dropped the lock. |
387 | */ |
388 | if (wait_event_interruptible_exclusive(pipe->rd_wait, pipe_readable(pipe)) < 0) |
389 | return -ERESTARTSYS; |
390 | |
391 | wake_next_reader = true; |
392 | mutex_lock(&pipe->mutex); |
393 | } |
394 | if (pipe_is_empty(pipe)) |
395 | wake_next_reader = false; |
396 | mutex_unlock(lock: &pipe->mutex); |
397 | |
398 | if (wake_writer) |
399 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); |
400 | if (wake_next_reader) |
401 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); |
402 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
403 | return ret; |
404 | } |
405 | |
406 | static ssize_t |
407 | fifo_pipe_read(struct kiocb *iocb, struct iov_iter *to) |
408 | { |
409 | int ret = anon_pipe_read(iocb, to); |
410 | if (ret > 0) |
411 | file_accessed(file: iocb->ki_filp); |
412 | return ret; |
413 | } |
414 | |
415 | static inline int is_packetized(struct file *file) |
416 | { |
417 | return (file->f_flags & O_DIRECT) != 0; |
418 | } |
419 | |
420 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
421 | static inline bool pipe_writable(const struct pipe_inode_info *pipe) |
422 | { |
423 | union pipe_index idx = { .head_tail = READ_ONCE(pipe->head_tail) }; |
424 | unsigned int max_usage = READ_ONCE(pipe->max_usage); |
425 | |
426 | return !pipe_full(head: idx.head, tail: idx.tail, limit: max_usage) || |
427 | !READ_ONCE(pipe->readers); |
428 | } |
429 | |
430 | static ssize_t |
431 | anon_pipe_write(struct kiocb *iocb, struct iov_iter *from) |
432 | { |
433 | struct file *filp = iocb->ki_filp; |
434 | struct pipe_inode_info *pipe = filp->private_data; |
435 | unsigned int head; |
436 | ssize_t ret = 0; |
437 | size_t total_len = iov_iter_count(i: from); |
438 | ssize_t chars; |
439 | bool was_empty = false; |
440 | bool wake_next_writer = false; |
441 | |
442 | /* |
443 | * Reject writing to watch queue pipes before the point where we lock |
444 | * the pipe. |
445 | * Otherwise, lockdep would be unhappy if the caller already has another |
446 | * pipe locked. |
447 | * If we had to support locking a normal pipe and a notification pipe at |
448 | * the same time, we could set up lockdep annotations for that, but |
449 | * since we don't actually need that, it's simpler to just bail here. |
450 | */ |
451 | if (pipe_has_watch_queue(pipe)) |
452 | return -EXDEV; |
453 | |
454 | /* Null write succeeds. */ |
455 | if (unlikely(total_len == 0)) |
456 | return 0; |
457 | |
458 | mutex_lock(&pipe->mutex); |
459 | |
460 | if (!pipe->readers) { |
461 | send_sig(SIGPIPE, current, 0); |
462 | ret = -EPIPE; |
463 | goto out; |
464 | } |
465 | |
466 | /* |
467 | * If it wasn't empty we try to merge new data into |
468 | * the last buffer. |
469 | * |
470 | * That naturally merges small writes, but it also |
471 | * page-aligns the rest of the writes for large writes |
472 | * spanning multiple pages. |
473 | */ |
474 | head = pipe->head; |
475 | was_empty = pipe_empty(head, tail: pipe->tail); |
476 | chars = total_len & (PAGE_SIZE-1); |
477 | if (chars && !was_empty) { |
478 | struct pipe_buffer *buf = pipe_buf(pipe, slot: head - 1); |
479 | int offset = buf->offset + buf->len; |
480 | |
481 | if ((buf->flags & PIPE_BUF_FLAG_CAN_MERGE) && |
482 | offset + chars <= PAGE_SIZE) { |
483 | ret = pipe_buf_confirm(pipe, buf); |
484 | if (ret) |
485 | goto out; |
486 | |
487 | ret = copy_page_from_iter(page: buf->page, offset, bytes: chars, i: from); |
488 | if (unlikely(ret < chars)) { |
489 | ret = -EFAULT; |
490 | goto out; |
491 | } |
492 | |
493 | buf->len += ret; |
494 | if (!iov_iter_count(i: from)) |
495 | goto out; |
496 | } |
497 | } |
498 | |
499 | for (;;) { |
500 | if (!pipe->readers) { |
501 | send_sig(SIGPIPE, current, 0); |
502 | if (!ret) |
503 | ret = -EPIPE; |
504 | break; |
505 | } |
506 | |
507 | head = pipe->head; |
508 | if (!pipe_full(head, tail: pipe->tail, limit: pipe->max_usage)) { |
509 | struct pipe_buffer *buf; |
510 | struct page *page; |
511 | int copied; |
512 | |
513 | page = anon_pipe_get_page(pipe); |
514 | if (unlikely(!page)) { |
515 | if (!ret) |
516 | ret = -ENOMEM; |
517 | break; |
518 | } |
519 | |
520 | copied = copy_page_from_iter(page, offset: 0, PAGE_SIZE, i: from); |
521 | if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) { |
522 | anon_pipe_put_page(pipe, page); |
523 | if (!ret) |
524 | ret = -EFAULT; |
525 | break; |
526 | } |
527 | |
528 | pipe->head = head + 1; |
529 | /* Insert it into the buffer array */ |
530 | buf = pipe_buf(pipe, slot: head); |
531 | buf->page = page; |
532 | buf->ops = &anon_pipe_buf_ops; |
533 | buf->offset = 0; |
534 | if (is_packetized(file: filp)) |
535 | buf->flags = PIPE_BUF_FLAG_PACKET; |
536 | else |
537 | buf->flags = PIPE_BUF_FLAG_CAN_MERGE; |
538 | |
539 | buf->len = copied; |
540 | ret += copied; |
541 | |
542 | if (!iov_iter_count(i: from)) |
543 | break; |
544 | |
545 | continue; |
546 | } |
547 | |
548 | /* Wait for buffer space to become available. */ |
549 | if ((filp->f_flags & O_NONBLOCK) || |
550 | (iocb->ki_flags & IOCB_NOWAIT)) { |
551 | if (!ret) |
552 | ret = -EAGAIN; |
553 | break; |
554 | } |
555 | if (signal_pending(current)) { |
556 | if (!ret) |
557 | ret = -ERESTARTSYS; |
558 | break; |
559 | } |
560 | |
561 | /* |
562 | * We're going to release the pipe lock and wait for more |
563 | * space. We wake up any readers if necessary, and then |
564 | * after waiting we need to re-check whether the pipe |
565 | * become empty while we dropped the lock. |
566 | */ |
567 | mutex_unlock(lock: &pipe->mutex); |
568 | if (was_empty) |
569 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); |
570 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
571 | wait_event_interruptible_exclusive(pipe->wr_wait, pipe_writable(pipe)); |
572 | mutex_lock(&pipe->mutex); |
573 | was_empty = pipe_is_empty(pipe); |
574 | wake_next_writer = true; |
575 | } |
576 | out: |
577 | if (pipe_is_full(pipe)) |
578 | wake_next_writer = false; |
579 | mutex_unlock(lock: &pipe->mutex); |
580 | |
581 | /* |
582 | * If we do do a wakeup event, we do a 'sync' wakeup, because we |
583 | * want the reader to start processing things asap, rather than |
584 | * leave the data pending. |
585 | * |
586 | * This is particularly important for small writes, because of |
587 | * how (for example) the GNU make jobserver uses small writes to |
588 | * wake up pending jobs |
589 | * |
590 | * Epoll nonsensically wants a wakeup whether the pipe |
591 | * was already empty or not. |
592 | */ |
593 | if (was_empty || pipe->poll_usage) |
594 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); |
595 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
596 | if (wake_next_writer) |
597 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); |
598 | return ret; |
599 | } |
600 | |
601 | static ssize_t |
602 | fifo_pipe_write(struct kiocb *iocb, struct iov_iter *from) |
603 | { |
604 | int ret = anon_pipe_write(iocb, from); |
605 | if (ret > 0) { |
606 | struct file *filp = iocb->ki_filp; |
607 | if (sb_start_write_trylock(sb: file_inode(f: filp)->i_sb)) { |
608 | int err = file_update_time(file: filp); |
609 | if (err) |
610 | ret = err; |
611 | sb_end_write(sb: file_inode(f: filp)->i_sb); |
612 | } |
613 | } |
614 | return ret; |
615 | } |
616 | |
617 | static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
618 | { |
619 | struct pipe_inode_info *pipe = filp->private_data; |
620 | unsigned int count, head, tail; |
621 | |
622 | switch (cmd) { |
623 | case FIONREAD: |
624 | mutex_lock(&pipe->mutex); |
625 | count = 0; |
626 | head = pipe->head; |
627 | tail = pipe->tail; |
628 | |
629 | while (!pipe_empty(head, tail)) { |
630 | count += pipe_buf(pipe, slot: tail)->len; |
631 | tail++; |
632 | } |
633 | mutex_unlock(lock: &pipe->mutex); |
634 | |
635 | return put_user(count, (int __user *)arg); |
636 | |
637 | #ifdef CONFIG_WATCH_QUEUE |
638 | case IOC_WATCH_QUEUE_SET_SIZE: { |
639 | int ret; |
640 | mutex_lock(&pipe->mutex); |
641 | ret = watch_queue_set_size(pipe, arg); |
642 | mutex_unlock(lock: &pipe->mutex); |
643 | return ret; |
644 | } |
645 | |
646 | case IOC_WATCH_QUEUE_SET_FILTER: |
647 | return watch_queue_set_filter( |
648 | pipe, (struct watch_notification_filter __user *)arg); |
649 | #endif |
650 | |
651 | default: |
652 | return -ENOIOCTLCMD; |
653 | } |
654 | } |
655 | |
656 | /* No kernel lock held - fine */ |
657 | static __poll_t |
658 | pipe_poll(struct file *filp, poll_table *wait) |
659 | { |
660 | __poll_t mask; |
661 | struct pipe_inode_info *pipe = filp->private_data; |
662 | union pipe_index idx; |
663 | |
664 | /* Epoll has some historical nasty semantics, this enables them */ |
665 | WRITE_ONCE(pipe->poll_usage, true); |
666 | |
667 | /* |
668 | * Reading pipe state only -- no need for acquiring the semaphore. |
669 | * |
670 | * But because this is racy, the code has to add the |
671 | * entry to the poll table _first_ .. |
672 | */ |
673 | if (filp->f_mode & FMODE_READ) |
674 | poll_wait(filp, wait_address: &pipe->rd_wait, p: wait); |
675 | if (filp->f_mode & FMODE_WRITE) |
676 | poll_wait(filp, wait_address: &pipe->wr_wait, p: wait); |
677 | |
678 | /* |
679 | * .. and only then can you do the racy tests. That way, |
680 | * if something changes and you got it wrong, the poll |
681 | * table entry will wake you up and fix it. |
682 | */ |
683 | idx.head_tail = READ_ONCE(pipe->head_tail); |
684 | |
685 | mask = 0; |
686 | if (filp->f_mode & FMODE_READ) { |
687 | if (!pipe_empty(head: idx.head, tail: idx.tail)) |
688 | mask |= EPOLLIN | EPOLLRDNORM; |
689 | if (!pipe->writers && filp->f_pipe != pipe->w_counter) |
690 | mask |= EPOLLHUP; |
691 | } |
692 | |
693 | if (filp->f_mode & FMODE_WRITE) { |
694 | if (!pipe_full(head: idx.head, tail: idx.tail, limit: pipe->max_usage)) |
695 | mask |= EPOLLOUT | EPOLLWRNORM; |
696 | /* |
697 | * Most Unices do not set EPOLLERR for FIFOs but on Linux they |
698 | * behave exactly like pipes for poll(). |
699 | */ |
700 | if (!pipe->readers) |
701 | mask |= EPOLLERR; |
702 | } |
703 | |
704 | return mask; |
705 | } |
706 | |
707 | static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe) |
708 | { |
709 | int kill = 0; |
710 | |
711 | spin_lock(lock: &inode->i_lock); |
712 | if (!--pipe->files) { |
713 | inode->i_pipe = NULL; |
714 | kill = 1; |
715 | } |
716 | spin_unlock(lock: &inode->i_lock); |
717 | |
718 | if (kill) |
719 | free_pipe_info(pipe); |
720 | } |
721 | |
722 | static int |
723 | pipe_release(struct inode *inode, struct file *file) |
724 | { |
725 | struct pipe_inode_info *pipe = file->private_data; |
726 | |
727 | mutex_lock(&pipe->mutex); |
728 | if (file->f_mode & FMODE_READ) |
729 | pipe->readers--; |
730 | if (file->f_mode & FMODE_WRITE) |
731 | pipe->writers--; |
732 | |
733 | /* Was that the last reader or writer, but not the other side? */ |
734 | if (!pipe->readers != !pipe->writers) { |
735 | wake_up_interruptible_all(&pipe->rd_wait); |
736 | wake_up_interruptible_all(&pipe->wr_wait); |
737 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
738 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
739 | } |
740 | mutex_unlock(lock: &pipe->mutex); |
741 | |
742 | put_pipe_info(inode, pipe); |
743 | return 0; |
744 | } |
745 | |
746 | static int |
747 | pipe_fasync(int fd, struct file *filp, int on) |
748 | { |
749 | struct pipe_inode_info *pipe = filp->private_data; |
750 | int retval = 0; |
751 | |
752 | mutex_lock(&pipe->mutex); |
753 | if (filp->f_mode & FMODE_READ) |
754 | retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); |
755 | if ((filp->f_mode & FMODE_WRITE) && retval >= 0) { |
756 | retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); |
757 | if (retval < 0 && (filp->f_mode & FMODE_READ)) |
758 | /* this can happen only if on == T */ |
759 | fasync_helper(-1, filp, 0, &pipe->fasync_readers); |
760 | } |
761 | mutex_unlock(lock: &pipe->mutex); |
762 | return retval; |
763 | } |
764 | |
765 | unsigned long account_pipe_buffers(struct user_struct *user, |
766 | unsigned long old, unsigned long new) |
767 | { |
768 | return atomic_long_add_return(i: new - old, v: &user->pipe_bufs); |
769 | } |
770 | |
771 | bool too_many_pipe_buffers_soft(unsigned long user_bufs) |
772 | { |
773 | unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft); |
774 | |
775 | return soft_limit && user_bufs > soft_limit; |
776 | } |
777 | |
778 | bool too_many_pipe_buffers_hard(unsigned long user_bufs) |
779 | { |
780 | unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard); |
781 | |
782 | return hard_limit && user_bufs > hard_limit; |
783 | } |
784 | |
785 | bool pipe_is_unprivileged_user(void) |
786 | { |
787 | return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN); |
788 | } |
789 | |
790 | struct pipe_inode_info *alloc_pipe_info(void) |
791 | { |
792 | struct pipe_inode_info *pipe; |
793 | unsigned long pipe_bufs = PIPE_DEF_BUFFERS; |
794 | struct user_struct *user = get_current_user(); |
795 | unsigned long user_bufs; |
796 | unsigned int max_size = READ_ONCE(pipe_max_size); |
797 | |
798 | pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT); |
799 | if (pipe == NULL) |
800 | goto out_free_uid; |
801 | |
802 | if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE)) |
803 | pipe_bufs = max_size >> PAGE_SHIFT; |
804 | |
805 | user_bufs = account_pipe_buffers(user, old: 0, new: pipe_bufs); |
806 | |
807 | if (too_many_pipe_buffers_soft(user_bufs) && pipe_is_unprivileged_user()) { |
808 | user_bufs = account_pipe_buffers(user, old: pipe_bufs, PIPE_MIN_DEF_BUFFERS); |
809 | pipe_bufs = PIPE_MIN_DEF_BUFFERS; |
810 | } |
811 | |
812 | if (too_many_pipe_buffers_hard(user_bufs) && pipe_is_unprivileged_user()) |
813 | goto out_revert_acct; |
814 | |
815 | pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer), |
816 | GFP_KERNEL_ACCOUNT); |
817 | |
818 | if (pipe->bufs) { |
819 | init_waitqueue_head(&pipe->rd_wait); |
820 | init_waitqueue_head(&pipe->wr_wait); |
821 | pipe->r_counter = pipe->w_counter = 1; |
822 | pipe->max_usage = pipe_bufs; |
823 | pipe->ring_size = pipe_bufs; |
824 | pipe->nr_accounted = pipe_bufs; |
825 | pipe->user = user; |
826 | mutex_init(&pipe->mutex); |
827 | lock_set_cmp_fn(&pipe->mutex, pipe_lock_cmp_fn, NULL); |
828 | return pipe; |
829 | } |
830 | |
831 | out_revert_acct: |
832 | (void) account_pipe_buffers(user, old: pipe_bufs, new: 0); |
833 | kfree(objp: pipe); |
834 | out_free_uid: |
835 | free_uid(user); |
836 | return NULL; |
837 | } |
838 | |
839 | void free_pipe_info(struct pipe_inode_info *pipe) |
840 | { |
841 | unsigned int i; |
842 | |
843 | #ifdef CONFIG_WATCH_QUEUE |
844 | if (pipe->watch_queue) |
845 | watch_queue_clear(pipe->watch_queue); |
846 | #endif |
847 | |
848 | (void) account_pipe_buffers(user: pipe->user, old: pipe->nr_accounted, new: 0); |
849 | free_uid(pipe->user); |
850 | for (i = 0; i < pipe->ring_size; i++) { |
851 | struct pipe_buffer *buf = pipe->bufs + i; |
852 | if (buf->ops) |
853 | pipe_buf_release(pipe, buf); |
854 | } |
855 | #ifdef CONFIG_WATCH_QUEUE |
856 | if (pipe->watch_queue) |
857 | put_watch_queue(pipe->watch_queue); |
858 | #endif |
859 | for (i = 0; i < ARRAY_SIZE(pipe->tmp_page); i++) { |
860 | if (pipe->tmp_page[i]) |
861 | __free_page(pipe->tmp_page[i]); |
862 | } |
863 | kfree(objp: pipe->bufs); |
864 | kfree(objp: pipe); |
865 | } |
866 | |
867 | static struct vfsmount *pipe_mnt __ro_after_init; |
868 | |
869 | /* |
870 | * pipefs_dname() is called from d_path(). |
871 | */ |
872 | static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen) |
873 | { |
874 | return dynamic_dname(buffer, buflen, "pipe:[%lu]", |
875 | d_inode(dentry)->i_ino); |
876 | } |
877 | |
878 | static const struct dentry_operations pipefs_dentry_operations = { |
879 | .d_dname = pipefs_dname, |
880 | }; |
881 | |
882 | static const struct file_operations pipeanon_fops; |
883 | |
884 | static struct inode * get_pipe_inode(void) |
885 | { |
886 | struct inode *inode = new_inode_pseudo(sb: pipe_mnt->mnt_sb); |
887 | struct pipe_inode_info *pipe; |
888 | |
889 | if (!inode) |
890 | goto fail_inode; |
891 | |
892 | inode->i_ino = get_next_ino(); |
893 | |
894 | pipe = alloc_pipe_info(); |
895 | if (!pipe) |
896 | goto fail_iput; |
897 | |
898 | inode->i_pipe = pipe; |
899 | pipe->files = 2; |
900 | pipe->readers = pipe->writers = 1; |
901 | inode->i_fop = &pipeanon_fops; |
902 | |
903 | /* |
904 | * Mark the inode dirty from the very beginning, |
905 | * that way it will never be moved to the dirty |
906 | * list because "mark_inode_dirty()" will think |
907 | * that it already _is_ on the dirty list. |
908 | */ |
909 | inode->i_state = I_DIRTY; |
910 | inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR; |
911 | inode->i_uid = current_fsuid(); |
912 | inode->i_gid = current_fsgid(); |
913 | simple_inode_init_ts(inode); |
914 | |
915 | return inode; |
916 | |
917 | fail_iput: |
918 | iput(inode); |
919 | |
920 | fail_inode: |
921 | return NULL; |
922 | } |
923 | |
924 | int create_pipe_files(struct file **res, int flags) |
925 | { |
926 | struct inode *inode = get_pipe_inode(); |
927 | struct file *f; |
928 | int error; |
929 | |
930 | if (!inode) |
931 | return -ENFILE; |
932 | |
933 | if (flags & O_NOTIFICATION_PIPE) { |
934 | error = watch_queue_init(inode->i_pipe); |
935 | if (error) { |
936 | free_pipe_info(pipe: inode->i_pipe); |
937 | iput(inode); |
938 | return error; |
939 | } |
940 | } |
941 | |
942 | f = alloc_file_pseudo(inode, pipe_mnt, "", |
943 | O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)), |
944 | &pipeanon_fops); |
945 | if (IS_ERR(ptr: f)) { |
946 | free_pipe_info(pipe: inode->i_pipe); |
947 | iput(inode); |
948 | return PTR_ERR(ptr: f); |
949 | } |
950 | |
951 | f->private_data = inode->i_pipe; |
952 | f->f_pipe = 0; |
953 | |
954 | res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK), |
955 | &pipeanon_fops); |
956 | if (IS_ERR(ptr: res[0])) { |
957 | put_pipe_info(inode, pipe: inode->i_pipe); |
958 | fput(f); |
959 | return PTR_ERR(ptr: res[0]); |
960 | } |
961 | res[0]->private_data = inode->i_pipe; |
962 | res[0]->f_pipe = 0; |
963 | res[1] = f; |
964 | stream_open(inode, filp: res[0]); |
965 | stream_open(inode, filp: res[1]); |
966 | /* |
967 | * Disable permission and pre-content events, but enable legacy |
968 | * inotify events for legacy users. |
969 | */ |
970 | file_set_fsnotify_mode(file: res[0], FMODE_NONOTIFY_PERM); |
971 | file_set_fsnotify_mode(file: res[1], FMODE_NONOTIFY_PERM); |
972 | return 0; |
973 | } |
974 | |
975 | static int __do_pipe_flags(int *fd, struct file **files, int flags) |
976 | { |
977 | int error; |
978 | int fdw, fdr; |
979 | |
980 | if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT | O_NOTIFICATION_PIPE)) |
981 | return -EINVAL; |
982 | |
983 | error = create_pipe_files(res: files, flags); |
984 | if (error) |
985 | return error; |
986 | |
987 | error = get_unused_fd_flags(flags); |
988 | if (error < 0) |
989 | goto err_read_pipe; |
990 | fdr = error; |
991 | |
992 | error = get_unused_fd_flags(flags); |
993 | if (error < 0) |
994 | goto err_fdr; |
995 | fdw = error; |
996 | |
997 | audit_fd_pair(fd1: fdr, fd2: fdw); |
998 | fd[0] = fdr; |
999 | fd[1] = fdw; |
1000 | /* pipe groks IOCB_NOWAIT */ |
1001 | files[0]->f_mode |= FMODE_NOWAIT; |
1002 | files[1]->f_mode |= FMODE_NOWAIT; |
1003 | return 0; |
1004 | |
1005 | err_fdr: |
1006 | put_unused_fd(fd: fdr); |
1007 | err_read_pipe: |
1008 | fput(files[0]); |
1009 | fput(files[1]); |
1010 | return error; |
1011 | } |
1012 | |
1013 | int do_pipe_flags(int *fd, int flags) |
1014 | { |
1015 | struct file *files[2]; |
1016 | int error = __do_pipe_flags(fd, files, flags); |
1017 | if (!error) { |
1018 | fd_install(fd: fd[0], file: files[0]); |
1019 | fd_install(fd: fd[1], file: files[1]); |
1020 | } |
1021 | return error; |
1022 | } |
1023 | |
1024 | /* |
1025 | * sys_pipe() is the normal C calling standard for creating |
1026 | * a pipe. It's not the way Unix traditionally does this, though. |
1027 | */ |
1028 | static int do_pipe2(int __user *fildes, int flags) |
1029 | { |
1030 | struct file *files[2]; |
1031 | int fd[2]; |
1032 | int error; |
1033 | |
1034 | error = __do_pipe_flags(fd, files, flags); |
1035 | if (!error) { |
1036 | if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) { |
1037 | fput(files[0]); |
1038 | fput(files[1]); |
1039 | put_unused_fd(fd: fd[0]); |
1040 | put_unused_fd(fd: fd[1]); |
1041 | error = -EFAULT; |
1042 | } else { |
1043 | fd_install(fd: fd[0], file: files[0]); |
1044 | fd_install(fd: fd[1], file: files[1]); |
1045 | } |
1046 | } |
1047 | return error; |
1048 | } |
1049 | |
1050 | SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags) |
1051 | { |
1052 | return do_pipe2(fildes, flags); |
1053 | } |
1054 | |
1055 | SYSCALL_DEFINE1(pipe, int __user *, fildes) |
1056 | { |
1057 | return do_pipe2(fildes, flags: 0); |
1058 | } |
1059 | |
1060 | /* |
1061 | * This is the stupid "wait for pipe to be readable or writable" |
1062 | * model. |
1063 | * |
1064 | * See pipe_read/write() for the proper kind of exclusive wait, |
1065 | * but that requires that we wake up any other readers/writers |
1066 | * if we then do not end up reading everything (ie the whole |
1067 | * "wake_next_reader/writer" logic in pipe_read/write()). |
1068 | */ |
1069 | void pipe_wait_readable(struct pipe_inode_info *pipe) |
1070 | { |
1071 | pipe_unlock(pipe); |
1072 | wait_event_interruptible(pipe->rd_wait, pipe_readable(pipe)); |
1073 | pipe_lock(pipe); |
1074 | } |
1075 | |
1076 | void pipe_wait_writable(struct pipe_inode_info *pipe) |
1077 | { |
1078 | pipe_unlock(pipe); |
1079 | wait_event_interruptible(pipe->wr_wait, pipe_writable(pipe)); |
1080 | pipe_lock(pipe); |
1081 | } |
1082 | |
1083 | /* |
1084 | * This depends on both the wait (here) and the wakeup (wake_up_partner) |
1085 | * holding the pipe lock, so "*cnt" is stable and we know a wakeup cannot |
1086 | * race with the count check and waitqueue prep. |
1087 | * |
1088 | * Normally in order to avoid races, you'd do the prepare_to_wait() first, |
1089 | * then check the condition you're waiting for, and only then sleep. But |
1090 | * because of the pipe lock, we can check the condition before being on |
1091 | * the wait queue. |
1092 | * |
1093 | * We use the 'rd_wait' waitqueue for pipe partner waiting. |
1094 | */ |
1095 | static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt) |
1096 | { |
1097 | DEFINE_WAIT(rdwait); |
1098 | int cur = *cnt; |
1099 | |
1100 | while (cur == *cnt) { |
1101 | prepare_to_wait(wq_head: &pipe->rd_wait, wq_entry: &rdwait, TASK_INTERRUPTIBLE); |
1102 | pipe_unlock(pipe); |
1103 | schedule(); |
1104 | finish_wait(wq_head: &pipe->rd_wait, wq_entry: &rdwait); |
1105 | pipe_lock(pipe); |
1106 | if (signal_pending(current)) |
1107 | break; |
1108 | } |
1109 | return cur == *cnt ? -ERESTARTSYS : 0; |
1110 | } |
1111 | |
1112 | static void wake_up_partner(struct pipe_inode_info *pipe) |
1113 | { |
1114 | wake_up_interruptible_all(&pipe->rd_wait); |
1115 | } |
1116 | |
1117 | static int fifo_open(struct inode *inode, struct file *filp) |
1118 | { |
1119 | bool is_pipe = inode->i_fop == &pipeanon_fops; |
1120 | struct pipe_inode_info *pipe; |
1121 | int ret; |
1122 | |
1123 | filp->f_pipe = 0; |
1124 | |
1125 | spin_lock(lock: &inode->i_lock); |
1126 | if (inode->i_pipe) { |
1127 | pipe = inode->i_pipe; |
1128 | pipe->files++; |
1129 | spin_unlock(lock: &inode->i_lock); |
1130 | } else { |
1131 | spin_unlock(lock: &inode->i_lock); |
1132 | pipe = alloc_pipe_info(); |
1133 | if (!pipe) |
1134 | return -ENOMEM; |
1135 | pipe->files = 1; |
1136 | spin_lock(lock: &inode->i_lock); |
1137 | if (unlikely(inode->i_pipe)) { |
1138 | inode->i_pipe->files++; |
1139 | spin_unlock(lock: &inode->i_lock); |
1140 | free_pipe_info(pipe); |
1141 | pipe = inode->i_pipe; |
1142 | } else { |
1143 | inode->i_pipe = pipe; |
1144 | spin_unlock(lock: &inode->i_lock); |
1145 | } |
1146 | } |
1147 | filp->private_data = pipe; |
1148 | /* OK, we have a pipe and it's pinned down */ |
1149 | |
1150 | mutex_lock(&pipe->mutex); |
1151 | |
1152 | /* We can only do regular read/write on fifos */ |
1153 | stream_open(inode, filp); |
1154 | |
1155 | switch (filp->f_mode & (FMODE_READ | FMODE_WRITE)) { |
1156 | case FMODE_READ: |
1157 | /* |
1158 | * O_RDONLY |
1159 | * POSIX.1 says that O_NONBLOCK means return with the FIFO |
1160 | * opened, even when there is no process writing the FIFO. |
1161 | */ |
1162 | pipe->r_counter++; |
1163 | if (pipe->readers++ == 0) |
1164 | wake_up_partner(pipe); |
1165 | |
1166 | if (!is_pipe && !pipe->writers) { |
1167 | if ((filp->f_flags & O_NONBLOCK)) { |
1168 | /* suppress EPOLLHUP until we have |
1169 | * seen a writer */ |
1170 | filp->f_pipe = pipe->w_counter; |
1171 | } else { |
1172 | if (wait_for_partner(pipe, cnt: &pipe->w_counter)) |
1173 | goto err_rd; |
1174 | } |
1175 | } |
1176 | break; |
1177 | |
1178 | case FMODE_WRITE: |
1179 | /* |
1180 | * O_WRONLY |
1181 | * POSIX.1 says that O_NONBLOCK means return -1 with |
1182 | * errno=ENXIO when there is no process reading the FIFO. |
1183 | */ |
1184 | ret = -ENXIO; |
1185 | if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers) |
1186 | goto err; |
1187 | |
1188 | pipe->w_counter++; |
1189 | if (!pipe->writers++) |
1190 | wake_up_partner(pipe); |
1191 | |
1192 | if (!is_pipe && !pipe->readers) { |
1193 | if (wait_for_partner(pipe, cnt: &pipe->r_counter)) |
1194 | goto err_wr; |
1195 | } |
1196 | break; |
1197 | |
1198 | case FMODE_READ | FMODE_WRITE: |
1199 | /* |
1200 | * O_RDWR |
1201 | * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set. |
1202 | * This implementation will NEVER block on a O_RDWR open, since |
1203 | * the process can at least talk to itself. |
1204 | */ |
1205 | |
1206 | pipe->readers++; |
1207 | pipe->writers++; |
1208 | pipe->r_counter++; |
1209 | pipe->w_counter++; |
1210 | if (pipe->readers == 1 || pipe->writers == 1) |
1211 | wake_up_partner(pipe); |
1212 | break; |
1213 | |
1214 | default: |
1215 | ret = -EINVAL; |
1216 | goto err; |
1217 | } |
1218 | |
1219 | /* Ok! */ |
1220 | mutex_unlock(lock: &pipe->mutex); |
1221 | return 0; |
1222 | |
1223 | err_rd: |
1224 | if (!--pipe->readers) |
1225 | wake_up_interruptible(&pipe->wr_wait); |
1226 | ret = -ERESTARTSYS; |
1227 | goto err; |
1228 | |
1229 | err_wr: |
1230 | if (!--pipe->writers) |
1231 | wake_up_interruptible_all(&pipe->rd_wait); |
1232 | ret = -ERESTARTSYS; |
1233 | goto err; |
1234 | |
1235 | err: |
1236 | mutex_unlock(lock: &pipe->mutex); |
1237 | |
1238 | put_pipe_info(inode, pipe); |
1239 | return ret; |
1240 | } |
1241 | |
1242 | const struct file_operations pipefifo_fops = { |
1243 | .open = fifo_open, |
1244 | .read_iter = fifo_pipe_read, |
1245 | .write_iter = fifo_pipe_write, |
1246 | .poll = pipe_poll, |
1247 | .unlocked_ioctl = pipe_ioctl, |
1248 | .release = pipe_release, |
1249 | .fasync = pipe_fasync, |
1250 | .splice_write = iter_file_splice_write, |
1251 | }; |
1252 | |
1253 | static const struct file_operations pipeanon_fops = { |
1254 | .open = fifo_open, |
1255 | .read_iter = anon_pipe_read, |
1256 | .write_iter = anon_pipe_write, |
1257 | .poll = pipe_poll, |
1258 | .unlocked_ioctl = pipe_ioctl, |
1259 | .release = pipe_release, |
1260 | .fasync = pipe_fasync, |
1261 | .splice_write = iter_file_splice_write, |
1262 | }; |
1263 | |
1264 | /* |
1265 | * Currently we rely on the pipe array holding a power-of-2 number |
1266 | * of pages. Returns 0 on error. |
1267 | */ |
1268 | unsigned int round_pipe_size(unsigned int size) |
1269 | { |
1270 | if (size > (1U << 31)) |
1271 | return 0; |
1272 | |
1273 | /* Minimum pipe size, as required by POSIX */ |
1274 | if (size < PAGE_SIZE) |
1275 | return PAGE_SIZE; |
1276 | |
1277 | return roundup_pow_of_two(size); |
1278 | } |
1279 | |
1280 | /* |
1281 | * Resize the pipe ring to a number of slots. |
1282 | * |
1283 | * Note the pipe can be reduced in capacity, but only if the current |
1284 | * occupancy doesn't exceed nr_slots; if it does, EBUSY will be |
1285 | * returned instead. |
1286 | */ |
1287 | int pipe_resize_ring(struct pipe_inode_info *pipe, unsigned int nr_slots) |
1288 | { |
1289 | struct pipe_buffer *bufs; |
1290 | unsigned int head, tail, mask, n; |
1291 | |
1292 | /* nr_slots larger than limits of pipe->{head,tail} */ |
1293 | if (unlikely(nr_slots > (pipe_index_t)-1u)) |
1294 | return -EINVAL; |
1295 | |
1296 | bufs = kcalloc(nr_slots, sizeof(*bufs), |
1297 | GFP_KERNEL_ACCOUNT | __GFP_NOWARN); |
1298 | if (unlikely(!bufs)) |
1299 | return -ENOMEM; |
1300 | |
1301 | spin_lock_irq(lock: &pipe->rd_wait.lock); |
1302 | mask = pipe->ring_size - 1; |
1303 | head = pipe->head; |
1304 | tail = pipe->tail; |
1305 | |
1306 | n = pipe_occupancy(head, tail); |
1307 | if (nr_slots < n) { |
1308 | spin_unlock_irq(lock: &pipe->rd_wait.lock); |
1309 | kfree(objp: bufs); |
1310 | return -EBUSY; |
1311 | } |
1312 | |
1313 | /* |
1314 | * The pipe array wraps around, so just start the new one at zero |
1315 | * and adjust the indices. |
1316 | */ |
1317 | if (n > 0) { |
1318 | unsigned int h = head & mask; |
1319 | unsigned int t = tail & mask; |
1320 | if (h > t) { |
1321 | memcpy(bufs, pipe->bufs + t, |
1322 | n * sizeof(struct pipe_buffer)); |
1323 | } else { |
1324 | unsigned int tsize = pipe->ring_size - t; |
1325 | if (h > 0) |
1326 | memcpy(bufs + tsize, pipe->bufs, |
1327 | h * sizeof(struct pipe_buffer)); |
1328 | memcpy(bufs, pipe->bufs + t, |
1329 | tsize * sizeof(struct pipe_buffer)); |
1330 | } |
1331 | } |
1332 | |
1333 | head = n; |
1334 | tail = 0; |
1335 | |
1336 | kfree(objp: pipe->bufs); |
1337 | pipe->bufs = bufs; |
1338 | pipe->ring_size = nr_slots; |
1339 | if (pipe->max_usage > nr_slots) |
1340 | pipe->max_usage = nr_slots; |
1341 | pipe->tail = tail; |
1342 | pipe->head = head; |
1343 | |
1344 | if (!pipe_has_watch_queue(pipe)) { |
1345 | pipe->max_usage = nr_slots; |
1346 | pipe->nr_accounted = nr_slots; |
1347 | } |
1348 | |
1349 | spin_unlock_irq(lock: &pipe->rd_wait.lock); |
1350 | |
1351 | /* This might have made more room for writers */ |
1352 | wake_up_interruptible(&pipe->wr_wait); |
1353 | return 0; |
1354 | } |
1355 | |
1356 | /* |
1357 | * Allocate a new array of pipe buffers and copy the info over. Returns the |
1358 | * pipe size if successful, or return -ERROR on error. |
1359 | */ |
1360 | static long pipe_set_size(struct pipe_inode_info *pipe, unsigned int arg) |
1361 | { |
1362 | unsigned long user_bufs; |
1363 | unsigned int nr_slots, size; |
1364 | long ret = 0; |
1365 | |
1366 | if (pipe_has_watch_queue(pipe)) |
1367 | return -EBUSY; |
1368 | |
1369 | size = round_pipe_size(size: arg); |
1370 | nr_slots = size >> PAGE_SHIFT; |
1371 | |
1372 | if (!nr_slots) |
1373 | return -EINVAL; |
1374 | |
1375 | /* |
1376 | * If trying to increase the pipe capacity, check that an |
1377 | * unprivileged user is not trying to exceed various limits |
1378 | * (soft limit check here, hard limit check just below). |
1379 | * Decreasing the pipe capacity is always permitted, even |
1380 | * if the user is currently over a limit. |
1381 | */ |
1382 | if (nr_slots > pipe->max_usage && |
1383 | size > pipe_max_size && !capable(CAP_SYS_RESOURCE)) |
1384 | return -EPERM; |
1385 | |
1386 | user_bufs = account_pipe_buffers(user: pipe->user, old: pipe->nr_accounted, new: nr_slots); |
1387 | |
1388 | if (nr_slots > pipe->max_usage && |
1389 | (too_many_pipe_buffers_hard(user_bufs) || |
1390 | too_many_pipe_buffers_soft(user_bufs)) && |
1391 | pipe_is_unprivileged_user()) { |
1392 | ret = -EPERM; |
1393 | goto out_revert_acct; |
1394 | } |
1395 | |
1396 | ret = pipe_resize_ring(pipe, nr_slots); |
1397 | if (ret < 0) |
1398 | goto out_revert_acct; |
1399 | |
1400 | return pipe->max_usage * PAGE_SIZE; |
1401 | |
1402 | out_revert_acct: |
1403 | (void) account_pipe_buffers(user: pipe->user, old: nr_slots, new: pipe->nr_accounted); |
1404 | return ret; |
1405 | } |
1406 | |
1407 | /* |
1408 | * Note that i_pipe and i_cdev share the same location, so checking ->i_pipe is |
1409 | * not enough to verify that this is a pipe. |
1410 | */ |
1411 | struct pipe_inode_info *get_pipe_info(struct file *file, bool for_splice) |
1412 | { |
1413 | struct pipe_inode_info *pipe = file->private_data; |
1414 | |
1415 | if (!pipe) |
1416 | return NULL; |
1417 | if (file->f_op != &pipefifo_fops && file->f_op != &pipeanon_fops) |
1418 | return NULL; |
1419 | if (for_splice && pipe_has_watch_queue(pipe)) |
1420 | return NULL; |
1421 | return pipe; |
1422 | } |
1423 | |
1424 | long pipe_fcntl(struct file *file, unsigned int cmd, unsigned int arg) |
1425 | { |
1426 | struct pipe_inode_info *pipe; |
1427 | long ret; |
1428 | |
1429 | pipe = get_pipe_info(file, for_splice: false); |
1430 | if (!pipe) |
1431 | return -EBADF; |
1432 | |
1433 | mutex_lock(&pipe->mutex); |
1434 | |
1435 | switch (cmd) { |
1436 | case F_SETPIPE_SZ: |
1437 | ret = pipe_set_size(pipe, arg); |
1438 | break; |
1439 | case F_GETPIPE_SZ: |
1440 | ret = pipe->max_usage * PAGE_SIZE; |
1441 | break; |
1442 | default: |
1443 | ret = -EINVAL; |
1444 | break; |
1445 | } |
1446 | |
1447 | mutex_unlock(lock: &pipe->mutex); |
1448 | return ret; |
1449 | } |
1450 | |
1451 | static const struct super_operations pipefs_ops = { |
1452 | .destroy_inode = free_inode_nonrcu, |
1453 | .statfs = simple_statfs, |
1454 | }; |
1455 | |
1456 | /* |
1457 | * pipefs should _never_ be mounted by userland - too much of security hassle, |
1458 | * no real gain from having the whole file system mounted. So we don't need |
1459 | * any operations on the root directory. However, we need a non-trivial |
1460 | * d_name - pipe: will go nicely and kill the special-casing in procfs. |
1461 | */ |
1462 | |
1463 | static int pipefs_init_fs_context(struct fs_context *fc) |
1464 | { |
1465 | struct pseudo_fs_context *ctx = init_pseudo(fc, PIPEFS_MAGIC); |
1466 | if (!ctx) |
1467 | return -ENOMEM; |
1468 | ctx->ops = &pipefs_ops; |
1469 | ctx->dops = &pipefs_dentry_operations; |
1470 | return 0; |
1471 | } |
1472 | |
1473 | static struct file_system_type pipe_fs_type = { |
1474 | .name = "pipefs", |
1475 | .init_fs_context = pipefs_init_fs_context, |
1476 | .kill_sb = kill_anon_super, |
1477 | }; |
1478 | |
1479 | #ifdef CONFIG_SYSCTL |
1480 | static int do_proc_dopipe_max_size_conv(unsigned long *lvalp, |
1481 | unsigned int *valp, |
1482 | int write, void *data) |
1483 | { |
1484 | if (write) { |
1485 | unsigned int val; |
1486 | |
1487 | val = round_pipe_size(size: *lvalp); |
1488 | if (val == 0) |
1489 | return -EINVAL; |
1490 | |
1491 | *valp = val; |
1492 | } else { |
1493 | unsigned int val = *valp; |
1494 | *lvalp = (unsigned long) val; |
1495 | } |
1496 | |
1497 | return 0; |
1498 | } |
1499 | |
1500 | static int proc_dopipe_max_size(const struct ctl_table *table, int write, |
1501 | void *buffer, size_t *lenp, loff_t *ppos) |
1502 | { |
1503 | return do_proc_douintvec(table, write, buffer, lenp, ppos, |
1504 | conv: do_proc_dopipe_max_size_conv, NULL); |
1505 | } |
1506 | |
1507 | static const struct ctl_table fs_pipe_sysctls[] = { |
1508 | { |
1509 | .procname = "pipe-max-size", |
1510 | .data = &pipe_max_size, |
1511 | .maxlen = sizeof(pipe_max_size), |
1512 | .mode = 0644, |
1513 | .proc_handler = proc_dopipe_max_size, |
1514 | }, |
1515 | { |
1516 | .procname = "pipe-user-pages-hard", |
1517 | .data = &pipe_user_pages_hard, |
1518 | .maxlen = sizeof(pipe_user_pages_hard), |
1519 | .mode = 0644, |
1520 | .proc_handler = proc_doulongvec_minmax, |
1521 | }, |
1522 | { |
1523 | .procname = "pipe-user-pages-soft", |
1524 | .data = &pipe_user_pages_soft, |
1525 | .maxlen = sizeof(pipe_user_pages_soft), |
1526 | .mode = 0644, |
1527 | .proc_handler = proc_doulongvec_minmax, |
1528 | }, |
1529 | }; |
1530 | #endif |
1531 | |
1532 | static int __init init_pipe_fs(void) |
1533 | { |
1534 | int err = register_filesystem(&pipe_fs_type); |
1535 | |
1536 | if (!err) { |
1537 | pipe_mnt = kern_mount(&pipe_fs_type); |
1538 | if (IS_ERR(ptr: pipe_mnt)) { |
1539 | err = PTR_ERR(ptr: pipe_mnt); |
1540 | unregister_filesystem(&pipe_fs_type); |
1541 | } |
1542 | } |
1543 | #ifdef CONFIG_SYSCTL |
1544 | register_sysctl_init("fs", fs_pipe_sysctls); |
1545 | #endif |
1546 | return err; |
1547 | } |
1548 | |
1549 | fs_initcall(init_pipe_fs); |
1550 |
Definitions
- pipe_max_size
- pipe_user_pages_hard
- pipe_user_pages_soft
- pipe_lock_cmp_fn
- pipe_lock
- pipe_unlock
- pipe_double_lock
- anon_pipe_get_page
- anon_pipe_put_page
- anon_pipe_buf_release
- anon_pipe_buf_try_steal
- generic_pipe_buf_try_steal
- generic_pipe_buf_get
- generic_pipe_buf_release
- anon_pipe_buf_ops
- pipe_readable
- pipe_update_tail
- anon_pipe_read
- fifo_pipe_read
- is_packetized
- pipe_writable
- anon_pipe_write
- fifo_pipe_write
- pipe_ioctl
- pipe_poll
- put_pipe_info
- pipe_release
- pipe_fasync
- account_pipe_buffers
- too_many_pipe_buffers_soft
- too_many_pipe_buffers_hard
- pipe_is_unprivileged_user
- alloc_pipe_info
- free_pipe_info
- pipe_mnt
- pipefs_dname
- pipefs_dentry_operations
- pipeanon_fops
- get_pipe_inode
- create_pipe_files
- __do_pipe_flags
- do_pipe_flags
- do_pipe2
- pipe_wait_readable
- pipe_wait_writable
- wait_for_partner
- wake_up_partner
- fifo_open
- pipefifo_fops
- pipeanon_fops
- round_pipe_size
- pipe_resize_ring
- pipe_set_size
- get_pipe_info
- pipe_fcntl
- pipefs_ops
- pipefs_init_fs_context
- pipe_fs_type
- do_proc_dopipe_max_size_conv
- proc_dopipe_max_size
- fs_pipe_sysctls
Improve your Profiling and Debugging skills
Find out more