]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - fs/pipe.c
Merge tag 'sound-4.19-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
[mirror_ubuntu-eoan-kernel.git] / fs / pipe.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
1da177e4
LT
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>
35f3d14d 15#include <linux/log2.h>
1da177e4 16#include <linux/mount.h>
b502bd11 17#include <linux/magic.h>
1da177e4
LT
18#include <linux/pipe_fs_i.h>
19#include <linux/uio.h>
20#include <linux/highmem.h>
5274f052 21#include <linux/pagemap.h>
db349509 22#include <linux/audit.h>
ba719bae 23#include <linux/syscalls.h>
b492e95b 24#include <linux/fcntl.h>
d86133bd 25#include <linux/memcontrol.h>
1da177e4 26
7c0f6ba6 27#include <linux/uaccess.h>
1da177e4
LT
28#include <asm/ioctls.h>
29
599a0ac1
AV
30#include "internal.h"
31
b492e95b
JA
32/*
33 * The max size that a non-root user is allowed to grow the pipe. Can
ff9da691 34 * be set by root in /proc/sys/fs/pipe-max-size
b492e95b 35 */
ff9da691
JA
36unsigned int pipe_max_size = 1048576;
37
759c0114
WT
38/* Maximum allocatable pages per user. Hard limit is unset by default, soft
39 * matches default values.
40 */
41unsigned long pipe_user_pages_hard;
42unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;
43
1da177e4
LT
44/*
45 * We use a start+len construction, which provides full use of the
46 * allocated memory.
47 * -- Florian Coosmann (FGC)
48 *
49 * Reads with count = 0 should always return 0.
50 * -- Julian Bradfield 1999-06-07.
51 *
52 * FIFOs and Pipes now generate SIGIO for both readers and writers.
53 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
54 *
55 * pipe_read & write cleanup
56 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
57 */
58
61e0d47c
MS
59static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
60{
6447a3cf 61 if (pipe->files)
72b0d9aa 62 mutex_lock_nested(&pipe->mutex, subclass);
61e0d47c
MS
63}
64
65void pipe_lock(struct pipe_inode_info *pipe)
66{
67 /*
68 * pipe_lock() nests non-pipe inode locks (for writing to a file)
69 */
70 pipe_lock_nested(pipe, I_MUTEX_PARENT);
71}
72EXPORT_SYMBOL(pipe_lock);
73
74void pipe_unlock(struct pipe_inode_info *pipe)
75{
6447a3cf 76 if (pipe->files)
72b0d9aa 77 mutex_unlock(&pipe->mutex);
61e0d47c
MS
78}
79EXPORT_SYMBOL(pipe_unlock);
80
ebec73f4
AV
81static inline void __pipe_lock(struct pipe_inode_info *pipe)
82{
83 mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
84}
85
86static inline void __pipe_unlock(struct pipe_inode_info *pipe)
87{
88 mutex_unlock(&pipe->mutex);
89}
90
61e0d47c
MS
91void pipe_double_lock(struct pipe_inode_info *pipe1,
92 struct pipe_inode_info *pipe2)
93{
94 BUG_ON(pipe1 == pipe2);
95
96 if (pipe1 < pipe2) {
97 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
98 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
99 } else {
023d43c7
PZ
100 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
101 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
61e0d47c
MS
102 }
103}
104
1da177e4 105/* Drop the inode semaphore and wait for a pipe event, atomically */
3a326a2c 106void pipe_wait(struct pipe_inode_info *pipe)
1da177e4
LT
107{
108 DEFINE_WAIT(wait);
109
d79fc0fc
IM
110 /*
111 * Pipes are system-local resources, so sleeping on them
112 * is considered a noninteractive wait:
113 */
af927232 114 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
61e0d47c 115 pipe_unlock(pipe);
1da177e4 116 schedule();
3a326a2c 117 finish_wait(&pipe->wait, &wait);
61e0d47c 118 pipe_lock(pipe);
1da177e4
LT
119}
120
341b446b
IM
121static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
122 struct pipe_buffer *buf)
1da177e4
LT
123{
124 struct page *page = buf->page;
125
5274f052
JA
126 /*
127 * If nobody else uses this page, and we don't already have a
128 * temporary page, let's keep track of it as a one-deep
341b446b 129 * allocation cache. (Otherwise just release our reference to it)
5274f052 130 */
341b446b 131 if (page_count(page) == 1 && !pipe->tmp_page)
923f4f23 132 pipe->tmp_page = page;
341b446b 133 else
09cbfeaf 134 put_page(page);
1da177e4
LT
135}
136
d86133bd
VD
137static int anon_pipe_buf_steal(struct pipe_inode_info *pipe,
138 struct pipe_buffer *buf)
139{
140 struct page *page = buf->page;
141
142 if (page_count(page) == 1) {
c4159a75 143 if (memcg_kmem_enabled())
d86133bd 144 memcg_kmem_uncharge(page, 0);
d86133bd
VD
145 __SetPageLocked(page);
146 return 0;
147 }
148 return 1;
149}
150
0845718d 151/**
b51d63c6 152 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
0845718d
JA
153 * @pipe: the pipe that the buffer belongs to
154 * @buf: the buffer to attempt to steal
155 *
156 * Description:
b51d63c6 157 * This function attempts to steal the &struct page attached to
0845718d
JA
158 * @buf. If successful, this function returns 0 and returns with
159 * the page locked. The caller may then reuse the page for whatever
b51d63c6 160 * he wishes; the typical use is insertion into a different file
0845718d
JA
161 * page cache.
162 */
330ab716
JA
163int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
164 struct pipe_buffer *buf)
5abc97aa 165{
46e678c9
JA
166 struct page *page = buf->page;
167
0845718d
JA
168 /*
169 * A reference of one is golden, that means that the owner of this
170 * page is the only one holding a reference to it. lock the page
171 * and return OK.
172 */
46e678c9 173 if (page_count(page) == 1) {
46e678c9
JA
174 lock_page(page);
175 return 0;
176 }
177
178 return 1;
5abc97aa 179}
51921cb7 180EXPORT_SYMBOL(generic_pipe_buf_steal);
5abc97aa 181
0845718d 182/**
b51d63c6 183 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
0845718d
JA
184 * @pipe: the pipe that the buffer belongs to
185 * @buf: the buffer to get a reference to
186 *
187 * Description:
188 * This function grabs an extra reference to @buf. It's used in
189 * in the tee() system call, when we duplicate the buffers in one
190 * pipe into another.
191 */
192void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
70524490 193{
09cbfeaf 194 get_page(buf->page);
70524490 195}
51921cb7 196EXPORT_SYMBOL(generic_pipe_buf_get);
70524490 197
0845718d
JA
198/**
199 * generic_pipe_buf_confirm - verify contents of the pipe buffer
79685b8d 200 * @info: the pipe that the buffer belongs to
0845718d
JA
201 * @buf: the buffer to confirm
202 *
203 * Description:
204 * This function does nothing, because the generic pipe code uses
205 * pages that are always good when inserted into the pipe.
206 */
cac36bb0
JA
207int generic_pipe_buf_confirm(struct pipe_inode_info *info,
208 struct pipe_buffer *buf)
f84d7519
JA
209{
210 return 0;
211}
51921cb7 212EXPORT_SYMBOL(generic_pipe_buf_confirm);
f84d7519 213
6818173b
MS
214/**
215 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
216 * @pipe: the pipe that the buffer belongs to
217 * @buf: the buffer to put a reference to
218 *
219 * Description:
220 * This function releases a reference to @buf.
221 */
222void generic_pipe_buf_release(struct pipe_inode_info *pipe,
223 struct pipe_buffer *buf)
224{
09cbfeaf 225 put_page(buf->page);
6818173b 226}
51921cb7 227EXPORT_SYMBOL(generic_pipe_buf_release);
6818173b 228
d4c3cca9 229static const struct pipe_buf_operations anon_pipe_buf_ops = {
1da177e4 230 .can_merge = 1,
cac36bb0 231 .confirm = generic_pipe_buf_confirm,
1da177e4 232 .release = anon_pipe_buf_release,
d86133bd 233 .steal = anon_pipe_buf_steal,
f84d7519 234 .get = generic_pipe_buf_get,
1da177e4
LT
235};
236
9883035a
LT
237static const struct pipe_buf_operations packet_pipe_buf_ops = {
238 .can_merge = 0,
9883035a
LT
239 .confirm = generic_pipe_buf_confirm,
240 .release = anon_pipe_buf_release,
d86133bd 241 .steal = anon_pipe_buf_steal,
9883035a
LT
242 .get = generic_pipe_buf_get,
243};
244
1da177e4 245static ssize_t
fb9096a3 246pipe_read(struct kiocb *iocb, struct iov_iter *to)
1da177e4 247{
fb9096a3 248 size_t total_len = iov_iter_count(to);
ee0b3e67 249 struct file *filp = iocb->ki_filp;
de32ec4c 250 struct pipe_inode_info *pipe = filp->private_data;
1da177e4
LT
251 int do_wakeup;
252 ssize_t ret;
1da177e4 253
1da177e4
LT
254 /* Null read succeeds. */
255 if (unlikely(total_len == 0))
256 return 0;
257
258 do_wakeup = 0;
259 ret = 0;
ebec73f4 260 __pipe_lock(pipe);
1da177e4 261 for (;;) {
923f4f23 262 int bufs = pipe->nrbufs;
1da177e4 263 if (bufs) {
923f4f23
IM
264 int curbuf = pipe->curbuf;
265 struct pipe_buffer *buf = pipe->bufs + curbuf;
1da177e4 266 size_t chars = buf->len;
637b58c2
AV
267 size_t written;
268 int error;
1da177e4
LT
269
270 if (chars > total_len)
271 chars = total_len;
272
fba597db 273 error = pipe_buf_confirm(pipe, buf);
f84d7519 274 if (error) {
5274f052 275 if (!ret)
e5953cbd 276 ret = error;
5274f052
JA
277 break;
278 }
f84d7519 279
fb9096a3 280 written = copy_page_to_iter(buf->page, buf->offset, chars, to);
637b58c2 281 if (unlikely(written < chars)) {
341b446b 282 if (!ret)
637b58c2 283 ret = -EFAULT;
1da177e4
LT
284 break;
285 }
286 ret += chars;
287 buf->offset += chars;
288 buf->len -= chars;
9883035a
LT
289
290 /* Was it a packet buffer? Clean up and exit */
291 if (buf->flags & PIPE_BUF_FLAG_PACKET) {
292 total_len = chars;
293 buf->len = 0;
294 }
295
1da177e4 296 if (!buf->len) {
a779638c 297 pipe_buf_release(pipe, buf);
35f3d14d 298 curbuf = (curbuf + 1) & (pipe->buffers - 1);
923f4f23
IM
299 pipe->curbuf = curbuf;
300 pipe->nrbufs = --bufs;
1da177e4
LT
301 do_wakeup = 1;
302 }
303 total_len -= chars;
304 if (!total_len)
305 break; /* common path: read succeeded */
306 }
307 if (bufs) /* More to do? */
308 continue;
923f4f23 309 if (!pipe->writers)
1da177e4 310 break;
923f4f23 311 if (!pipe->waiting_writers) {
1da177e4
LT
312 /* syscall merging: Usually we must not sleep
313 * if O_NONBLOCK is set, or if we got some data.
314 * But if a writer sleeps in kernel space, then
315 * we can wait for that data without violating POSIX.
316 */
317 if (ret)
318 break;
319 if (filp->f_flags & O_NONBLOCK) {
320 ret = -EAGAIN;
321 break;
322 }
323 }
324 if (signal_pending(current)) {
341b446b
IM
325 if (!ret)
326 ret = -ERESTARTSYS;
1da177e4
LT
327 break;
328 }
329 if (do_wakeup) {
a9a08845 330 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
923f4f23 331 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
1da177e4 332 }
923f4f23 333 pipe_wait(pipe);
1da177e4 334 }
ebec73f4 335 __pipe_unlock(pipe);
341b446b
IM
336
337 /* Signal writers asynchronously that there is more room. */
1da177e4 338 if (do_wakeup) {
a9a08845 339 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
923f4f23 340 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
1da177e4
LT
341 }
342 if (ret > 0)
343 file_accessed(filp);
344 return ret;
345}
346
9883035a
LT
347static inline int is_packetized(struct file *file)
348{
349 return (file->f_flags & O_DIRECT) != 0;
350}
351
1da177e4 352static ssize_t
f0d1bec9 353pipe_write(struct kiocb *iocb, struct iov_iter *from)
1da177e4 354{
ee0b3e67 355 struct file *filp = iocb->ki_filp;
de32ec4c 356 struct pipe_inode_info *pipe = filp->private_data;
f0d1bec9
AV
357 ssize_t ret = 0;
358 int do_wakeup = 0;
359 size_t total_len = iov_iter_count(from);
1da177e4
LT
360 ssize_t chars;
361
1da177e4
LT
362 /* Null write succeeds. */
363 if (unlikely(total_len == 0))
364 return 0;
365
ebec73f4 366 __pipe_lock(pipe);
1da177e4 367
923f4f23 368 if (!pipe->readers) {
1da177e4
LT
369 send_sig(SIGPIPE, current, 0);
370 ret = -EPIPE;
371 goto out;
372 }
373
374 /* We try to merge small writes */
375 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
923f4f23 376 if (pipe->nrbufs && chars != 0) {
341b446b 377 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
35f3d14d 378 (pipe->buffers - 1);
923f4f23 379 struct pipe_buffer *buf = pipe->bufs + lastbuf;
1da177e4 380 int offset = buf->offset + buf->len;
341b446b 381
fba597db
MS
382 if (buf->ops->can_merge && offset + chars <= PAGE_SIZE) {
383 ret = pipe_buf_confirm(pipe, buf);
6ae08069 384 if (ret)
5274f052 385 goto out;
f84d7519 386
f0d1bec9
AV
387 ret = copy_page_from_iter(buf->page, offset, chars, from);
388 if (unlikely(ret < chars)) {
6ae08069 389 ret = -EFAULT;
1da177e4 390 goto out;
f6762b7a 391 }
f0d1bec9 392 do_wakeup = 1;
6ae08069 393 buf->len += ret;
f0d1bec9 394 if (!iov_iter_count(from))
1da177e4
LT
395 goto out;
396 }
397 }
398
399 for (;;) {
400 int bufs;
341b446b 401
923f4f23 402 if (!pipe->readers) {
1da177e4 403 send_sig(SIGPIPE, current, 0);
341b446b
IM
404 if (!ret)
405 ret = -EPIPE;
1da177e4
LT
406 break;
407 }
923f4f23 408 bufs = pipe->nrbufs;
35f3d14d
JA
409 if (bufs < pipe->buffers) {
410 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
923f4f23
IM
411 struct pipe_buffer *buf = pipe->bufs + newbuf;
412 struct page *page = pipe->tmp_page;
f0d1bec9 413 int copied;
1da177e4
LT
414
415 if (!page) {
d86133bd 416 page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
1da177e4
LT
417 if (unlikely(!page)) {
418 ret = ret ? : -ENOMEM;
419 break;
420 }
923f4f23 421 pipe->tmp_page = page;
1da177e4 422 }
341b446b 423 /* Always wake up, even if the copy fails. Otherwise
1da177e4
LT
424 * we lock up (O_NONBLOCK-)readers that sleep due to
425 * syscall merging.
426 * FIXME! Is this really true?
427 */
428 do_wakeup = 1;
f0d1bec9
AV
429 copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
430 if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
341b446b 431 if (!ret)
f0d1bec9 432 ret = -EFAULT;
1da177e4
LT
433 break;
434 }
f0d1bec9 435 ret += copied;
1da177e4
LT
436
437 /* Insert it into the buffer array */
438 buf->page = page;
439 buf->ops = &anon_pipe_buf_ops;
440 buf->offset = 0;
f0d1bec9 441 buf->len = copied;
9883035a
LT
442 buf->flags = 0;
443 if (is_packetized(filp)) {
444 buf->ops = &packet_pipe_buf_ops;
445 buf->flags = PIPE_BUF_FLAG_PACKET;
446 }
923f4f23
IM
447 pipe->nrbufs = ++bufs;
448 pipe->tmp_page = NULL;
1da177e4 449
f0d1bec9 450 if (!iov_iter_count(from))
1da177e4
LT
451 break;
452 }
35f3d14d 453 if (bufs < pipe->buffers)
1da177e4
LT
454 continue;
455 if (filp->f_flags & O_NONBLOCK) {
341b446b
IM
456 if (!ret)
457 ret = -EAGAIN;
1da177e4
LT
458 break;
459 }
460 if (signal_pending(current)) {
341b446b
IM
461 if (!ret)
462 ret = -ERESTARTSYS;
1da177e4
LT
463 break;
464 }
465 if (do_wakeup) {
a9a08845 466 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
923f4f23 467 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1da177e4
LT
468 do_wakeup = 0;
469 }
923f4f23
IM
470 pipe->waiting_writers++;
471 pipe_wait(pipe);
472 pipe->waiting_writers--;
1da177e4
LT
473 }
474out:
ebec73f4 475 __pipe_unlock(pipe);
1da177e4 476 if (do_wakeup) {
a9a08845 477 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
923f4f23 478 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1da177e4 479 }
7e775f46 480 if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
c3b2da31
JB
481 int err = file_update_time(filp);
482 if (err)
483 ret = err;
7e775f46 484 sb_end_write(file_inode(filp)->i_sb);
c3b2da31 485 }
1da177e4
LT
486 return ret;
487}
488
d59d0b1b 489static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1da177e4 490{
de32ec4c 491 struct pipe_inode_info *pipe = filp->private_data;
1da177e4
LT
492 int count, buf, nrbufs;
493
494 switch (cmd) {
495 case FIONREAD:
ebec73f4 496 __pipe_lock(pipe);
1da177e4 497 count = 0;
923f4f23
IM
498 buf = pipe->curbuf;
499 nrbufs = pipe->nrbufs;
1da177e4 500 while (--nrbufs >= 0) {
923f4f23 501 count += pipe->bufs[buf].len;
35f3d14d 502 buf = (buf+1) & (pipe->buffers - 1);
1da177e4 503 }
ebec73f4 504 __pipe_unlock(pipe);
923f4f23 505
1da177e4
LT
506 return put_user(count, (int __user *)arg);
507 default:
46ce341b 508 return -ENOIOCTLCMD;
1da177e4
LT
509 }
510}
511
dd67081b 512/* No kernel lock held - fine */
a11e1d43
LT
513static __poll_t
514pipe_poll(struct file *filp, poll_table *wait)
dd67081b 515{
a11e1d43 516 __poll_t mask;
dd67081b 517 struct pipe_inode_info *pipe = filp->private_data;
a11e1d43
LT
518 int nrbufs;
519
520 poll_wait(filp, &pipe->wait, wait);
1da177e4
LT
521
522 /* Reading only -- no need for acquiring the semaphore. */
a11e1d43
LT
523 nrbufs = pipe->nrbufs;
524 mask = 0;
1da177e4 525 if (filp->f_mode & FMODE_READ) {
a9a08845 526 mask = (nrbufs > 0) ? EPOLLIN | EPOLLRDNORM : 0;
923f4f23 527 if (!pipe->writers && filp->f_version != pipe->w_counter)
a9a08845 528 mask |= EPOLLHUP;
1da177e4
LT
529 }
530
531 if (filp->f_mode & FMODE_WRITE) {
a9a08845 532 mask |= (nrbufs < pipe->buffers) ? EPOLLOUT | EPOLLWRNORM : 0;
5e5d7a22 533 /*
a9a08845 534 * Most Unices do not set EPOLLERR for FIFOs but on Linux they
5e5d7a22
PE
535 * behave exactly like pipes for poll().
536 */
923f4f23 537 if (!pipe->readers)
a9a08845 538 mask |= EPOLLERR;
1da177e4
LT
539 }
540
541 return mask;
542}
543
b0d8d229
LT
544static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
545{
546 int kill = 0;
547
548 spin_lock(&inode->i_lock);
549 if (!--pipe->files) {
550 inode->i_pipe = NULL;
551 kill = 1;
552 }
553 spin_unlock(&inode->i_lock);
554
555 if (kill)
556 free_pipe_info(pipe);
557}
558
1da177e4 559static int
599a0ac1 560pipe_release(struct inode *inode, struct file *file)
1da177e4 561{
b0d8d229 562 struct pipe_inode_info *pipe = file->private_data;
923f4f23 563
ebec73f4 564 __pipe_lock(pipe);
599a0ac1
AV
565 if (file->f_mode & FMODE_READ)
566 pipe->readers--;
567 if (file->f_mode & FMODE_WRITE)
568 pipe->writers--;
341b446b 569
ba5bb147 570 if (pipe->readers || pipe->writers) {
a9a08845 571 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM | EPOLLERR | EPOLLHUP);
923f4f23
IM
572 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
573 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
1da177e4 574 }
ebec73f4 575 __pipe_unlock(pipe);
ba5bb147 576
b0d8d229 577 put_pipe_info(inode, pipe);
1da177e4
LT
578 return 0;
579}
580
581static int
599a0ac1 582pipe_fasync(int fd, struct file *filp, int on)
1da177e4 583{
de32ec4c 584 struct pipe_inode_info *pipe = filp->private_data;
599a0ac1 585 int retval = 0;
1da177e4 586
ebec73f4 587 __pipe_lock(pipe);
599a0ac1
AV
588 if (filp->f_mode & FMODE_READ)
589 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
590 if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
341b446b 591 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
599a0ac1
AV
592 if (retval < 0 && (filp->f_mode & FMODE_READ))
593 /* this can happen only if on == T */
e5bc49ba
ON
594 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
595 }
ebec73f4 596 __pipe_unlock(pipe);
60aa4924 597 return retval;
1da177e4
LT
598}
599
9c87bcf0 600static unsigned long account_pipe_buffers(struct user_struct *user,
759c0114
WT
601 unsigned long old, unsigned long new)
602{
9c87bcf0 603 return atomic_long_add_return(new - old, &user->pipe_bufs);
759c0114
WT
604}
605
9c87bcf0 606static bool too_many_pipe_buffers_soft(unsigned long user_bufs)
759c0114 607{
f7340761
EB
608 unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft);
609
610 return soft_limit && user_bufs > soft_limit;
759c0114
WT
611}
612
9c87bcf0 613static bool too_many_pipe_buffers_hard(unsigned long user_bufs)
759c0114 614{
f7340761
EB
615 unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard);
616
617 return hard_limit && user_bufs > hard_limit;
759c0114
WT
618}
619
85c2dd54
EB
620static bool is_unprivileged_user(void)
621{
622 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
623}
624
7bee130e 625struct pipe_inode_info *alloc_pipe_info(void)
3a326a2c 626{
923f4f23 627 struct pipe_inode_info *pipe;
09b4d199
MK
628 unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
629 struct user_struct *user = get_current_user();
9c87bcf0 630 unsigned long user_bufs;
f7340761 631 unsigned int max_size = READ_ONCE(pipe_max_size);
3a326a2c 632
d86133bd 633 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT);
09b4d199
MK
634 if (pipe == NULL)
635 goto out_free_uid;
636
f7340761
EB
637 if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE))
638 pipe_bufs = max_size >> PAGE_SHIFT;
086e774a 639
9c87bcf0 640 user_bufs = account_pipe_buffers(user, 0, pipe_bufs);
a005ca0e 641
85c2dd54 642 if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) {
9c87bcf0 643 user_bufs = account_pipe_buffers(user, pipe_bufs, 1);
a005ca0e 644 pipe_bufs = 1;
09b4d199 645 }
759c0114 646
85c2dd54 647 if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user())
a005ca0e
MK
648 goto out_revert_acct;
649
650 pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer),
651 GFP_KERNEL_ACCOUNT);
652
09b4d199
MK
653 if (pipe->bufs) {
654 init_waitqueue_head(&pipe->wait);
655 pipe->r_counter = pipe->w_counter = 1;
656 pipe->buffers = pipe_bufs;
657 pipe->user = user;
09b4d199
MK
658 mutex_init(&pipe->mutex);
659 return pipe;
3a326a2c
IM
660 }
661
a005ca0e 662out_revert_acct:
9c87bcf0 663 (void) account_pipe_buffers(user, pipe_bufs, 0);
09b4d199
MK
664 kfree(pipe);
665out_free_uid:
666 free_uid(user);
35f3d14d 667 return NULL;
3a326a2c
IM
668}
669
4b8a8f1e 670void free_pipe_info(struct pipe_inode_info *pipe)
1da177e4
LT
671{
672 int i;
1da177e4 673
9c87bcf0 674 (void) account_pipe_buffers(pipe->user, pipe->buffers, 0);
759c0114 675 free_uid(pipe->user);
35f3d14d 676 for (i = 0; i < pipe->buffers; i++) {
923f4f23 677 struct pipe_buffer *buf = pipe->bufs + i;
1da177e4 678 if (buf->ops)
a779638c 679 pipe_buf_release(pipe, buf);
1da177e4 680 }
923f4f23
IM
681 if (pipe->tmp_page)
682 __free_page(pipe->tmp_page);
35f3d14d 683 kfree(pipe->bufs);
923f4f23 684 kfree(pipe);
1da177e4
LT
685}
686
fa3536cc 687static struct vfsmount *pipe_mnt __read_mostly;
341b446b 688
c23fbb6b
ED
689/*
690 * pipefs_dname() is called from d_path().
691 */
692static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
693{
694 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
75c3cfa8 695 d_inode(dentry)->i_ino);
c23fbb6b
ED
696}
697
3ba13d17 698static const struct dentry_operations pipefs_dentry_operations = {
c23fbb6b 699 .d_dname = pipefs_dname,
1da177e4
LT
700};
701
702static struct inode * get_pipe_inode(void)
703{
a209dfc7 704 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
923f4f23 705 struct pipe_inode_info *pipe;
1da177e4
LT
706
707 if (!inode)
708 goto fail_inode;
709
85fe4025
CH
710 inode->i_ino = get_next_ino();
711
7bee130e 712 pipe = alloc_pipe_info();
923f4f23 713 if (!pipe)
1da177e4 714 goto fail_iput;
3a326a2c 715
ba5bb147
AV
716 inode->i_pipe = pipe;
717 pipe->files = 2;
923f4f23 718 pipe->readers = pipe->writers = 1;
599a0ac1 719 inode->i_fop = &pipefifo_fops;
1da177e4
LT
720
721 /*
722 * Mark the inode dirty from the very beginning,
723 * that way it will never be moved to the dirty
724 * list because "mark_inode_dirty()" will think
725 * that it already _is_ on the dirty list.
726 */
727 inode->i_state = I_DIRTY;
728 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
da9592ed
DH
729 inode->i_uid = current_fsuid();
730 inode->i_gid = current_fsgid();
078cd827 731 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
923f4f23 732
1da177e4
LT
733 return inode;
734
735fail_iput:
736 iput(inode);
341b446b 737
1da177e4
LT
738fail_inode:
739 return NULL;
740}
741
e4fad8e5 742int create_pipe_files(struct file **res, int flags)
1da177e4 743{
e4fad8e5 744 struct inode *inode = get_pipe_inode();
d6cbd281 745 struct file *f;
1da177e4 746
1da177e4 747 if (!inode)
e4fad8e5 748 return -ENFILE;
1da177e4 749
152b6372
AV
750 f = alloc_file_pseudo(inode, pipe_mnt, "",
751 O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)),
752 &pipefifo_fops);
e9bb1f9b 753 if (IS_ERR(f)) {
152b6372
AV
754 free_pipe_info(inode->i_pipe);
755 iput(inode);
756 return PTR_ERR(f);
e9bb1f9b 757 }
341b446b 758
de32ec4c 759 f->private_data = inode->i_pipe;
d6cbd281 760
183266f2
AV
761 res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK),
762 &pipefifo_fops);
e9bb1f9b 763 if (IS_ERR(res[0])) {
b10a4a9f
AV
764 put_pipe_info(inode, inode->i_pipe);
765 fput(f);
766 return PTR_ERR(res[0]);
e9bb1f9b 767 }
de32ec4c 768 res[0]->private_data = inode->i_pipe;
e4fad8e5
AV
769 res[1] = f;
770 return 0;
d6cbd281
AK
771}
772
5b249b1b 773static int __do_pipe_flags(int *fd, struct file **files, int flags)
d6cbd281 774{
d6cbd281
AK
775 int error;
776 int fdw, fdr;
777
9883035a 778 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
ed8cae8b
UD
779 return -EINVAL;
780
e4fad8e5
AV
781 error = create_pipe_files(files, flags);
782 if (error)
783 return error;
d6cbd281 784
ed8cae8b 785 error = get_unused_fd_flags(flags);
d6cbd281
AK
786 if (error < 0)
787 goto err_read_pipe;
788 fdr = error;
789
ed8cae8b 790 error = get_unused_fd_flags(flags);
d6cbd281
AK
791 if (error < 0)
792 goto err_fdr;
793 fdw = error;
794
157cf649 795 audit_fd_pair(fdr, fdw);
d6cbd281
AK
796 fd[0] = fdr;
797 fd[1] = fdw;
d6cbd281
AK
798 return 0;
799
800 err_fdr:
801 put_unused_fd(fdr);
802 err_read_pipe:
e4fad8e5
AV
803 fput(files[0]);
804 fput(files[1]);
d6cbd281 805 return error;
1da177e4
LT
806}
807
5b249b1b
AV
808int do_pipe_flags(int *fd, int flags)
809{
810 struct file *files[2];
811 int error = __do_pipe_flags(fd, files, flags);
812 if (!error) {
813 fd_install(fd[0], files[0]);
814 fd_install(fd[1], files[1]);
815 }
816 return error;
817}
818
d35c7b0e
UD
819/*
820 * sys_pipe() is the normal C calling standard for creating
821 * a pipe. It's not the way Unix traditionally does this, though.
822 */
0a216dd1 823static int do_pipe2(int __user *fildes, int flags)
d35c7b0e 824{
5b249b1b 825 struct file *files[2];
d35c7b0e
UD
826 int fd[2];
827 int error;
828
5b249b1b 829 error = __do_pipe_flags(fd, files, flags);
d35c7b0e 830 if (!error) {
5b249b1b
AV
831 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
832 fput(files[0]);
833 fput(files[1]);
834 put_unused_fd(fd[0]);
835 put_unused_fd(fd[1]);
d35c7b0e 836 error = -EFAULT;
5b249b1b
AV
837 } else {
838 fd_install(fd[0], files[0]);
839 fd_install(fd[1], files[1]);
ba719bae 840 }
d35c7b0e
UD
841 }
842 return error;
843}
844
0a216dd1
DB
845SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
846{
847 return do_pipe2(fildes, flags);
848}
849
2b664219 850SYSCALL_DEFINE1(pipe, int __user *, fildes)
ed8cae8b 851{
0a216dd1 852 return do_pipe2(fildes, 0);
ed8cae8b
UD
853}
854
fc7478a2 855static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
f776c738
AV
856{
857 int cur = *cnt;
858
859 while (cur == *cnt) {
fc7478a2 860 pipe_wait(pipe);
f776c738
AV
861 if (signal_pending(current))
862 break;
863 }
864 return cur == *cnt ? -ERESTARTSYS : 0;
865}
866
fc7478a2 867static void wake_up_partner(struct pipe_inode_info *pipe)
f776c738 868{
fc7478a2 869 wake_up_interruptible(&pipe->wait);
f776c738
AV
870}
871
872static int fifo_open(struct inode *inode, struct file *filp)
873{
874 struct pipe_inode_info *pipe;
599a0ac1 875 bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
f776c738
AV
876 int ret;
877
ba5bb147
AV
878 filp->f_version = 0;
879
880 spin_lock(&inode->i_lock);
881 if (inode->i_pipe) {
882 pipe = inode->i_pipe;
883 pipe->files++;
884 spin_unlock(&inode->i_lock);
885 } else {
886 spin_unlock(&inode->i_lock);
7bee130e 887 pipe = alloc_pipe_info();
f776c738 888 if (!pipe)
ba5bb147
AV
889 return -ENOMEM;
890 pipe->files = 1;
891 spin_lock(&inode->i_lock);
892 if (unlikely(inode->i_pipe)) {
893 inode->i_pipe->files++;
894 spin_unlock(&inode->i_lock);
4b8a8f1e 895 free_pipe_info(pipe);
ba5bb147
AV
896 pipe = inode->i_pipe;
897 } else {
898 inode->i_pipe = pipe;
899 spin_unlock(&inode->i_lock);
900 }
f776c738 901 }
de32ec4c 902 filp->private_data = pipe;
ba5bb147
AV
903 /* OK, we have a pipe and it's pinned down */
904
ebec73f4 905 __pipe_lock(pipe);
f776c738
AV
906
907 /* We can only do regular read/write on fifos */
908 filp->f_mode &= (FMODE_READ | FMODE_WRITE);
909
910 switch (filp->f_mode) {
911 case FMODE_READ:
912 /*
913 * O_RDONLY
914 * POSIX.1 says that O_NONBLOCK means return with the FIFO
915 * opened, even when there is no process writing the FIFO.
916 */
f776c738
AV
917 pipe->r_counter++;
918 if (pipe->readers++ == 0)
fc7478a2 919 wake_up_partner(pipe);
f776c738 920
599a0ac1 921 if (!is_pipe && !pipe->writers) {
f776c738 922 if ((filp->f_flags & O_NONBLOCK)) {
a9a08845 923 /* suppress EPOLLHUP until we have
f776c738
AV
924 * seen a writer */
925 filp->f_version = pipe->w_counter;
926 } else {
fc7478a2 927 if (wait_for_partner(pipe, &pipe->w_counter))
f776c738
AV
928 goto err_rd;
929 }
930 }
931 break;
932
933 case FMODE_WRITE:
934 /*
935 * O_WRONLY
936 * POSIX.1 says that O_NONBLOCK means return -1 with
937 * errno=ENXIO when there is no process reading the FIFO.
938 */
939 ret = -ENXIO;
599a0ac1 940 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
f776c738
AV
941 goto err;
942
f776c738
AV
943 pipe->w_counter++;
944 if (!pipe->writers++)
fc7478a2 945 wake_up_partner(pipe);
f776c738 946
599a0ac1 947 if (!is_pipe && !pipe->readers) {
fc7478a2 948 if (wait_for_partner(pipe, &pipe->r_counter))
f776c738
AV
949 goto err_wr;
950 }
951 break;
952
953 case FMODE_READ | FMODE_WRITE:
954 /*
955 * O_RDWR
956 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
957 * This implementation will NEVER block on a O_RDWR open, since
958 * the process can at least talk to itself.
959 */
f776c738
AV
960
961 pipe->readers++;
962 pipe->writers++;
963 pipe->r_counter++;
964 pipe->w_counter++;
965 if (pipe->readers == 1 || pipe->writers == 1)
fc7478a2 966 wake_up_partner(pipe);
f776c738
AV
967 break;
968
969 default:
970 ret = -EINVAL;
971 goto err;
972 }
973
974 /* Ok! */
ebec73f4 975 __pipe_unlock(pipe);
f776c738
AV
976 return 0;
977
978err_rd:
979 if (!--pipe->readers)
980 wake_up_interruptible(&pipe->wait);
981 ret = -ERESTARTSYS;
982 goto err;
983
984err_wr:
985 if (!--pipe->writers)
986 wake_up_interruptible(&pipe->wait);
987 ret = -ERESTARTSYS;
988 goto err;
989
990err:
ebec73f4 991 __pipe_unlock(pipe);
b0d8d229
LT
992
993 put_pipe_info(inode, pipe);
f776c738
AV
994 return ret;
995}
996
599a0ac1
AV
997const struct file_operations pipefifo_fops = {
998 .open = fifo_open,
999 .llseek = no_llseek,
fb9096a3 1000 .read_iter = pipe_read,
f0d1bec9 1001 .write_iter = pipe_write,
a11e1d43 1002 .poll = pipe_poll,
599a0ac1
AV
1003 .unlocked_ioctl = pipe_ioctl,
1004 .release = pipe_release,
1005 .fasync = pipe_fasync,
f776c738
AV
1006};
1007
f491bd71
MK
1008/*
1009 * Currently we rely on the pipe array holding a power-of-2 number
d3f14c48 1010 * of pages. Returns 0 on error.
f491bd71 1011 */
96e99be4 1012unsigned int round_pipe_size(unsigned long size)
f491bd71 1013{
c4fed5a9 1014 if (size > (1U << 31))
96e99be4
EB
1015 return 0;
1016
4c2e4bef
EB
1017 /* Minimum pipe size, as required by POSIX */
1018 if (size < PAGE_SIZE)
c4fed5a9 1019 return PAGE_SIZE;
d3f14c48 1020
c4fed5a9 1021 return roundup_pow_of_two(size);
f491bd71
MK
1022}
1023
35f3d14d
JA
1024/*
1025 * Allocate a new array of pipe buffers and copy the info over. Returns the
1026 * pipe size if successful, or return -ERROR on error.
1027 */
d37d4166 1028static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
35f3d14d
JA
1029{
1030 struct pipe_buffer *bufs;
d37d4166 1031 unsigned int size, nr_pages;
9c87bcf0 1032 unsigned long user_bufs;
b0b91d18 1033 long ret = 0;
d37d4166
MK
1034
1035 size = round_pipe_size(arg);
1036 nr_pages = size >> PAGE_SHIFT;
1037
1038 if (!nr_pages)
1039 return -EINVAL;
1040
b0b91d18
MK
1041 /*
1042 * If trying to increase the pipe capacity, check that an
1043 * unprivileged user is not trying to exceed various limits
1044 * (soft limit check here, hard limit check just below).
1045 * Decreasing the pipe capacity is always permitted, even
1046 * if the user is currently over a limit.
1047 */
1048 if (nr_pages > pipe->buffers &&
1049 size > pipe_max_size && !capable(CAP_SYS_RESOURCE))
d37d4166
MK
1050 return -EPERM;
1051
9c87bcf0 1052 user_bufs = account_pipe_buffers(pipe->user, pipe->buffers, nr_pages);
b0b91d18
MK
1053
1054 if (nr_pages > pipe->buffers &&
9c87bcf0
MK
1055 (too_many_pipe_buffers_hard(user_bufs) ||
1056 too_many_pipe_buffers_soft(user_bufs)) &&
85c2dd54 1057 is_unprivileged_user()) {
b0b91d18
MK
1058 ret = -EPERM;
1059 goto out_revert_acct;
1060 }
35f3d14d 1061
35f3d14d
JA
1062 /*
1063 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1064 * expect a lot of shrink+grow operations, just free and allocate
1065 * again like we would do for growing. If the pipe currently
1066 * contains more buffers than arg, then return busy.
1067 */
b0b91d18
MK
1068 if (nr_pages < pipe->nrbufs) {
1069 ret = -EBUSY;
1070 goto out_revert_acct;
1071 }
35f3d14d 1072
d86133bd
VD
1073 bufs = kcalloc(nr_pages, sizeof(*bufs),
1074 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
b0b91d18
MK
1075 if (unlikely(!bufs)) {
1076 ret = -ENOMEM;
1077 goto out_revert_acct;
1078 }
35f3d14d
JA
1079
1080 /*
1081 * The pipe array wraps around, so just start the new one at zero
1082 * and adjust the indexes.
1083 */
1084 if (pipe->nrbufs) {
1d862f41
MS
1085 unsigned int tail;
1086 unsigned int head;
35f3d14d 1087
1d862f41
MS
1088 tail = pipe->curbuf + pipe->nrbufs;
1089 if (tail < pipe->buffers)
1090 tail = 0;
1091 else
1092 tail &= (pipe->buffers - 1);
1093
1094 head = pipe->nrbufs - tail;
35f3d14d
JA
1095 if (head)
1096 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1097 if (tail)
1d862f41 1098 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
35f3d14d
JA
1099 }
1100
1101 pipe->curbuf = 0;
1102 kfree(pipe->bufs);
1103 pipe->bufs = bufs;
b9598db3
JA
1104 pipe->buffers = nr_pages;
1105 return nr_pages * PAGE_SIZE;
b0b91d18
MK
1106
1107out_revert_acct:
9c87bcf0 1108 (void) account_pipe_buffers(pipe->user, nr_pages, pipe->buffers);
b0b91d18 1109 return ret;
35f3d14d
JA
1110}
1111
72083646
LT
1112/*
1113 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1114 * location, so checking ->i_pipe is not enough to verify that this is a
1115 * pipe.
1116 */
1117struct pipe_inode_info *get_pipe_info(struct file *file)
1118{
de32ec4c 1119 return file->f_op == &pipefifo_fops ? file->private_data : NULL;
72083646
LT
1120}
1121
35f3d14d
JA
1122long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1123{
1124 struct pipe_inode_info *pipe;
1125 long ret;
1126
c66fb347 1127 pipe = get_pipe_info(file);
35f3d14d
JA
1128 if (!pipe)
1129 return -EBADF;
1130
ebec73f4 1131 __pipe_lock(pipe);
35f3d14d
JA
1132
1133 switch (cmd) {
d37d4166
MK
1134 case F_SETPIPE_SZ:
1135 ret = pipe_set_size(pipe, arg);
35f3d14d
JA
1136 break;
1137 case F_GETPIPE_SZ:
b9598db3 1138 ret = pipe->buffers * PAGE_SIZE;
35f3d14d
JA
1139 break;
1140 default:
1141 ret = -EINVAL;
1142 break;
1143 }
1144
ebec73f4 1145 __pipe_unlock(pipe);
35f3d14d
JA
1146 return ret;
1147}
1148
ff0c7d15
NP
1149static const struct super_operations pipefs_ops = {
1150 .destroy_inode = free_inode_nonrcu,
d70ef97b 1151 .statfs = simple_statfs,
ff0c7d15
NP
1152};
1153
1da177e4
LT
1154/*
1155 * pipefs should _never_ be mounted by userland - too much of security hassle,
1156 * no real gain from having the whole whorehouse mounted. So we don't need
1157 * any operations on the root directory. However, we need a non-trivial
1158 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1159 */
51139ada
AV
1160static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1161 int flags, const char *dev_name, void *data)
1da177e4 1162{
c74a1cbb
AV
1163 return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1164 &pipefs_dentry_operations, PIPEFS_MAGIC);
1da177e4
LT
1165}
1166
1167static struct file_system_type pipe_fs_type = {
1168 .name = "pipefs",
51139ada 1169 .mount = pipefs_mount,
1da177e4
LT
1170 .kill_sb = kill_anon_super,
1171};
1172
1173static int __init init_pipe_fs(void)
1174{
1175 int err = register_filesystem(&pipe_fs_type);
341b446b 1176
1da177e4
LT
1177 if (!err) {
1178 pipe_mnt = kern_mount(&pipe_fs_type);
1179 if (IS_ERR(pipe_mnt)) {
1180 err = PTR_ERR(pipe_mnt);
1181 unregister_filesystem(&pipe_fs_type);
1182 }
1183 }
1184 return err;
1185}
1186
1da177e4 1187fs_initcall(init_pipe_fs);