Loading...
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/magic.h>
18#include <linux/pipe_fs_i.h>
19#include <linux/uio.h>
20#include <linux/highmem.h>
21#include <linux/pagemap.h>
22#include <linux/audit.h>
23#include <linux/syscalls.h>
24#include <linux/fcntl.h>
25#include <linux/memcontrol.h>
26
27#include <linux/uaccess.h>
28#include <asm/ioctls.h>
29
30#include "internal.h"
31
32/*
33 * The max size that a non-root user is allowed to grow the pipe. Can
34 * be set by root in /proc/sys/fs/pipe-max-size
35 */
36unsigned int pipe_max_size = 1048576;
37
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
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
59static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
60{
61 if (pipe->files)
62 mutex_lock_nested(&pipe->mutex, subclass);
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{
76 if (pipe->files)
77 mutex_unlock(&pipe->mutex);
78}
79EXPORT_SYMBOL(pipe_unlock);
80
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
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 {
100 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
101 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
102 }
103}
104
105/* Drop the inode semaphore and wait for a pipe event, atomically */
106void pipe_wait(struct pipe_inode_info *pipe)
107{
108 DEFINE_WAIT(wait);
109
110 /*
111 * Pipes are system-local resources, so sleeping on them
112 * is considered a noninteractive wait:
113 */
114 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
115 pipe_unlock(pipe);
116 schedule();
117 finish_wait(&pipe->wait, &wait);
118 pipe_lock(pipe);
119}
120
121static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
122 struct pipe_buffer *buf)
123{
124 struct page *page = buf->page;
125
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
129 * allocation cache. (Otherwise just release our reference to it)
130 */
131 if (page_count(page) == 1 && !pipe->tmp_page)
132 pipe->tmp_page = page;
133 else
134 put_page(page);
135}
136
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) {
143 if (memcg_kmem_enabled())
144 memcg_kmem_uncharge(page, 0);
145 __SetPageLocked(page);
146 return 0;
147 }
148 return 1;
149}
150
151/**
152 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
153 * @pipe: the pipe that the buffer belongs to
154 * @buf: the buffer to attempt to steal
155 *
156 * Description:
157 * This function attempts to steal the &struct page attached to
158 * @buf. If successful, this function returns 0 and returns with
159 * the page locked. The caller may then reuse the page for whatever
160 * he wishes; the typical use is insertion into a different file
161 * page cache.
162 */
163int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
164 struct pipe_buffer *buf)
165{
166 struct page *page = buf->page;
167
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 */
173 if (page_count(page) == 1) {
174 lock_page(page);
175 return 0;
176 }
177
178 return 1;
179}
180EXPORT_SYMBOL(generic_pipe_buf_steal);
181
182/**
183 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
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)
193{
194 get_page(buf->page);
195}
196EXPORT_SYMBOL(generic_pipe_buf_get);
197
198/**
199 * generic_pipe_buf_confirm - verify contents of the pipe buffer
200 * @info: the pipe that the buffer belongs to
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 */
207int generic_pipe_buf_confirm(struct pipe_inode_info *info,
208 struct pipe_buffer *buf)
209{
210 return 0;
211}
212EXPORT_SYMBOL(generic_pipe_buf_confirm);
213
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{
225 put_page(buf->page);
226}
227EXPORT_SYMBOL(generic_pipe_buf_release);
228
229static const struct pipe_buf_operations anon_pipe_buf_ops = {
230 .can_merge = 1,
231 .confirm = generic_pipe_buf_confirm,
232 .release = anon_pipe_buf_release,
233 .steal = anon_pipe_buf_steal,
234 .get = generic_pipe_buf_get,
235};
236
237static const struct pipe_buf_operations packet_pipe_buf_ops = {
238 .can_merge = 0,
239 .confirm = generic_pipe_buf_confirm,
240 .release = anon_pipe_buf_release,
241 .steal = anon_pipe_buf_steal,
242 .get = generic_pipe_buf_get,
243};
244
245static ssize_t
246pipe_read(struct kiocb *iocb, struct iov_iter *to)
247{
248 size_t total_len = iov_iter_count(to);
249 struct file *filp = iocb->ki_filp;
250 struct pipe_inode_info *pipe = filp->private_data;
251 int do_wakeup;
252 ssize_t ret;
253
254 /* Null read succeeds. */
255 if (unlikely(total_len == 0))
256 return 0;
257
258 do_wakeup = 0;
259 ret = 0;
260 __pipe_lock(pipe);
261 for (;;) {
262 int bufs = pipe->nrbufs;
263 if (bufs) {
264 int curbuf = pipe->curbuf;
265 struct pipe_buffer *buf = pipe->bufs + curbuf;
266 size_t chars = buf->len;
267 size_t written;
268 int error;
269
270 if (chars > total_len)
271 chars = total_len;
272
273 error = pipe_buf_confirm(pipe, buf);
274 if (error) {
275 if (!ret)
276 ret = error;
277 break;
278 }
279
280 written = copy_page_to_iter(buf->page, buf->offset, chars, to);
281 if (unlikely(written < chars)) {
282 if (!ret)
283 ret = -EFAULT;
284 break;
285 }
286 ret += chars;
287 buf->offset += chars;
288 buf->len -= chars;
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
296 if (!buf->len) {
297 pipe_buf_release(pipe, buf);
298 curbuf = (curbuf + 1) & (pipe->buffers - 1);
299 pipe->curbuf = curbuf;
300 pipe->nrbufs = --bufs;
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;
309 if (!pipe->writers)
310 break;
311 if (!pipe->waiting_writers) {
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)) {
325 if (!ret)
326 ret = -ERESTARTSYS;
327 break;
328 }
329 if (do_wakeup) {
330 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
331 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
332 }
333 pipe_wait(pipe);
334 }
335 __pipe_unlock(pipe);
336
337 /* Signal writers asynchronously that there is more room. */
338 if (do_wakeup) {
339 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
340 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
341 }
342 if (ret > 0)
343 file_accessed(filp);
344 return ret;
345}
346
347static inline int is_packetized(struct file *file)
348{
349 return (file->f_flags & O_DIRECT) != 0;
350}
351
352static ssize_t
353pipe_write(struct kiocb *iocb, struct iov_iter *from)
354{
355 struct file *filp = iocb->ki_filp;
356 struct pipe_inode_info *pipe = filp->private_data;
357 ssize_t ret = 0;
358 int do_wakeup = 0;
359 size_t total_len = iov_iter_count(from);
360 ssize_t chars;
361
362 /* Null write succeeds. */
363 if (unlikely(total_len == 0))
364 return 0;
365
366 __pipe_lock(pipe);
367
368 if (!pipe->readers) {
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 */
376 if (pipe->nrbufs && chars != 0) {
377 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
378 (pipe->buffers - 1);
379 struct pipe_buffer *buf = pipe->bufs + lastbuf;
380 int offset = buf->offset + buf->len;
381
382 if (buf->ops->can_merge && offset + chars <= PAGE_SIZE) {
383 ret = pipe_buf_confirm(pipe, buf);
384 if (ret)
385 goto out;
386
387 ret = copy_page_from_iter(buf->page, offset, chars, from);
388 if (unlikely(ret < chars)) {
389 ret = -EFAULT;
390 goto out;
391 }
392 do_wakeup = 1;
393 buf->len += ret;
394 if (!iov_iter_count(from))
395 goto out;
396 }
397 }
398
399 for (;;) {
400 int bufs;
401
402 if (!pipe->readers) {
403 send_sig(SIGPIPE, current, 0);
404 if (!ret)
405 ret = -EPIPE;
406 break;
407 }
408 bufs = pipe->nrbufs;
409 if (bufs < pipe->buffers) {
410 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
411 struct pipe_buffer *buf = pipe->bufs + newbuf;
412 struct page *page = pipe->tmp_page;
413 int copied;
414
415 if (!page) {
416 page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
417 if (unlikely(!page)) {
418 ret = ret ? : -ENOMEM;
419 break;
420 }
421 pipe->tmp_page = page;
422 }
423 /* Always wake up, even if the copy fails. Otherwise
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;
429 copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
430 if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
431 if (!ret)
432 ret = -EFAULT;
433 break;
434 }
435 ret += copied;
436
437 /* Insert it into the buffer array */
438 buf->page = page;
439 buf->ops = &anon_pipe_buf_ops;
440 buf->offset = 0;
441 buf->len = copied;
442 buf->flags = 0;
443 if (is_packetized(filp)) {
444 buf->ops = &packet_pipe_buf_ops;
445 buf->flags = PIPE_BUF_FLAG_PACKET;
446 }
447 pipe->nrbufs = ++bufs;
448 pipe->tmp_page = NULL;
449
450 if (!iov_iter_count(from))
451 break;
452 }
453 if (bufs < pipe->buffers)
454 continue;
455 if (filp->f_flags & O_NONBLOCK) {
456 if (!ret)
457 ret = -EAGAIN;
458 break;
459 }
460 if (signal_pending(current)) {
461 if (!ret)
462 ret = -ERESTARTSYS;
463 break;
464 }
465 if (do_wakeup) {
466 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
467 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
468 do_wakeup = 0;
469 }
470 pipe->waiting_writers++;
471 pipe_wait(pipe);
472 pipe->waiting_writers--;
473 }
474out:
475 __pipe_unlock(pipe);
476 if (do_wakeup) {
477 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
478 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
479 }
480 if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
481 int err = file_update_time(filp);
482 if (err)
483 ret = err;
484 sb_end_write(file_inode(filp)->i_sb);
485 }
486 return ret;
487}
488
489static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
490{
491 struct pipe_inode_info *pipe = filp->private_data;
492 int count, buf, nrbufs;
493
494 switch (cmd) {
495 case FIONREAD:
496 __pipe_lock(pipe);
497 count = 0;
498 buf = pipe->curbuf;
499 nrbufs = pipe->nrbufs;
500 while (--nrbufs >= 0) {
501 count += pipe->bufs[buf].len;
502 buf = (buf+1) & (pipe->buffers - 1);
503 }
504 __pipe_unlock(pipe);
505
506 return put_user(count, (int __user *)arg);
507 default:
508 return -ENOIOCTLCMD;
509 }
510}
511
512/* No kernel lock held - fine */
513static __poll_t
514pipe_poll(struct file *filp, poll_table *wait)
515{
516 __poll_t mask;
517 struct pipe_inode_info *pipe = filp->private_data;
518 int nrbufs;
519
520 poll_wait(filp, &pipe->wait, wait);
521
522 /* Reading only -- no need for acquiring the semaphore. */
523 nrbufs = pipe->nrbufs;
524 mask = 0;
525 if (filp->f_mode & FMODE_READ) {
526 mask = (nrbufs > 0) ? EPOLLIN | EPOLLRDNORM : 0;
527 if (!pipe->writers && filp->f_version != pipe->w_counter)
528 mask |= EPOLLHUP;
529 }
530
531 if (filp->f_mode & FMODE_WRITE) {
532 mask |= (nrbufs < pipe->buffers) ? EPOLLOUT | EPOLLWRNORM : 0;
533 /*
534 * Most Unices do not set EPOLLERR for FIFOs but on Linux they
535 * behave exactly like pipes for poll().
536 */
537 if (!pipe->readers)
538 mask |= EPOLLERR;
539 }
540
541 return mask;
542}
543
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
559static int
560pipe_release(struct inode *inode, struct file *file)
561{
562 struct pipe_inode_info *pipe = file->private_data;
563
564 __pipe_lock(pipe);
565 if (file->f_mode & FMODE_READ)
566 pipe->readers--;
567 if (file->f_mode & FMODE_WRITE)
568 pipe->writers--;
569
570 if (pipe->readers || pipe->writers) {
571 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM | EPOLLERR | EPOLLHUP);
572 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
573 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
574 }
575 __pipe_unlock(pipe);
576
577 put_pipe_info(inode, pipe);
578 return 0;
579}
580
581static int
582pipe_fasync(int fd, struct file *filp, int on)
583{
584 struct pipe_inode_info *pipe = filp->private_data;
585 int retval = 0;
586
587 __pipe_lock(pipe);
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) {
591 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
592 if (retval < 0 && (filp->f_mode & FMODE_READ))
593 /* this can happen only if on == T */
594 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
595 }
596 __pipe_unlock(pipe);
597 return retval;
598}
599
600static unsigned long account_pipe_buffers(struct user_struct *user,
601 unsigned long old, unsigned long new)
602{
603 return atomic_long_add_return(new - old, &user->pipe_bufs);
604}
605
606static bool too_many_pipe_buffers_soft(unsigned long user_bufs)
607{
608 unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft);
609
610 return soft_limit && user_bufs > soft_limit;
611}
612
613static bool too_many_pipe_buffers_hard(unsigned long user_bufs)
614{
615 unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard);
616
617 return hard_limit && user_bufs > hard_limit;
618}
619
620static bool is_unprivileged_user(void)
621{
622 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
623}
624
625struct pipe_inode_info *alloc_pipe_info(void)
626{
627 struct pipe_inode_info *pipe;
628 unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
629 struct user_struct *user = get_current_user();
630 unsigned long user_bufs;
631 unsigned int max_size = READ_ONCE(pipe_max_size);
632
633 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT);
634 if (pipe == NULL)
635 goto out_free_uid;
636
637 if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE))
638 pipe_bufs = max_size >> PAGE_SHIFT;
639
640 user_bufs = account_pipe_buffers(user, 0, pipe_bufs);
641
642 if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) {
643 user_bufs = account_pipe_buffers(user, pipe_bufs, 1);
644 pipe_bufs = 1;
645 }
646
647 if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user())
648 goto out_revert_acct;
649
650 pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer),
651 GFP_KERNEL_ACCOUNT);
652
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;
658 mutex_init(&pipe->mutex);
659 return pipe;
660 }
661
662out_revert_acct:
663 (void) account_pipe_buffers(user, pipe_bufs, 0);
664 kfree(pipe);
665out_free_uid:
666 free_uid(user);
667 return NULL;
668}
669
670void free_pipe_info(struct pipe_inode_info *pipe)
671{
672 int i;
673
674 (void) account_pipe_buffers(pipe->user, pipe->buffers, 0);
675 free_uid(pipe->user);
676 for (i = 0; i < pipe->buffers; i++) {
677 struct pipe_buffer *buf = pipe->bufs + i;
678 if (buf->ops)
679 pipe_buf_release(pipe, buf);
680 }
681 if (pipe->tmp_page)
682 __free_page(pipe->tmp_page);
683 kfree(pipe->bufs);
684 kfree(pipe);
685}
686
687static struct vfsmount *pipe_mnt __read_mostly;
688
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]",
695 d_inode(dentry)->i_ino);
696}
697
698static const struct dentry_operations pipefs_dentry_operations = {
699 .d_dname = pipefs_dname,
700};
701
702static struct inode * get_pipe_inode(void)
703{
704 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
705 struct pipe_inode_info *pipe;
706
707 if (!inode)
708 goto fail_inode;
709
710 inode->i_ino = get_next_ino();
711
712 pipe = alloc_pipe_info();
713 if (!pipe)
714 goto fail_iput;
715
716 inode->i_pipe = pipe;
717 pipe->files = 2;
718 pipe->readers = pipe->writers = 1;
719 inode->i_fop = &pipefifo_fops;
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;
729 inode->i_uid = current_fsuid();
730 inode->i_gid = current_fsgid();
731 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
732
733 return inode;
734
735fail_iput:
736 iput(inode);
737
738fail_inode:
739 return NULL;
740}
741
742int create_pipe_files(struct file **res, int flags)
743{
744 int err;
745 struct inode *inode = get_pipe_inode();
746 struct file *f;
747 struct path path;
748
749 if (!inode)
750 return -ENFILE;
751
752 err = -ENOMEM;
753 path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &empty_name);
754 if (!path.dentry)
755 goto err_inode;
756 path.mnt = mntget(pipe_mnt);
757
758 d_instantiate(path.dentry, inode);
759
760 f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops);
761 if (IS_ERR(f)) {
762 err = PTR_ERR(f);
763 goto err_dentry;
764 }
765
766 f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
767 f->private_data = inode->i_pipe;
768
769 res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops);
770 if (IS_ERR(res[0])) {
771 err = PTR_ERR(res[0]);
772 goto err_file;
773 }
774
775 path_get(&path);
776 res[0]->private_data = inode->i_pipe;
777 res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK);
778 res[1] = f;
779 return 0;
780
781err_file:
782 put_filp(f);
783err_dentry:
784 free_pipe_info(inode->i_pipe);
785 path_put(&path);
786 return err;
787
788err_inode:
789 free_pipe_info(inode->i_pipe);
790 iput(inode);
791 return err;
792}
793
794static int __do_pipe_flags(int *fd, struct file **files, int flags)
795{
796 int error;
797 int fdw, fdr;
798
799 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
800 return -EINVAL;
801
802 error = create_pipe_files(files, flags);
803 if (error)
804 return error;
805
806 error = get_unused_fd_flags(flags);
807 if (error < 0)
808 goto err_read_pipe;
809 fdr = error;
810
811 error = get_unused_fd_flags(flags);
812 if (error < 0)
813 goto err_fdr;
814 fdw = error;
815
816 audit_fd_pair(fdr, fdw);
817 fd[0] = fdr;
818 fd[1] = fdw;
819 return 0;
820
821 err_fdr:
822 put_unused_fd(fdr);
823 err_read_pipe:
824 fput(files[0]);
825 fput(files[1]);
826 return error;
827}
828
829int do_pipe_flags(int *fd, int flags)
830{
831 struct file *files[2];
832 int error = __do_pipe_flags(fd, files, flags);
833 if (!error) {
834 fd_install(fd[0], files[0]);
835 fd_install(fd[1], files[1]);
836 }
837 return error;
838}
839
840/*
841 * sys_pipe() is the normal C calling standard for creating
842 * a pipe. It's not the way Unix traditionally does this, though.
843 */
844static int do_pipe2(int __user *fildes, int flags)
845{
846 struct file *files[2];
847 int fd[2];
848 int error;
849
850 error = __do_pipe_flags(fd, files, flags);
851 if (!error) {
852 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
853 fput(files[0]);
854 fput(files[1]);
855 put_unused_fd(fd[0]);
856 put_unused_fd(fd[1]);
857 error = -EFAULT;
858 } else {
859 fd_install(fd[0], files[0]);
860 fd_install(fd[1], files[1]);
861 }
862 }
863 return error;
864}
865
866SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
867{
868 return do_pipe2(fildes, flags);
869}
870
871SYSCALL_DEFINE1(pipe, int __user *, fildes)
872{
873 return do_pipe2(fildes, 0);
874}
875
876static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
877{
878 int cur = *cnt;
879
880 while (cur == *cnt) {
881 pipe_wait(pipe);
882 if (signal_pending(current))
883 break;
884 }
885 return cur == *cnt ? -ERESTARTSYS : 0;
886}
887
888static void wake_up_partner(struct pipe_inode_info *pipe)
889{
890 wake_up_interruptible(&pipe->wait);
891}
892
893static int fifo_open(struct inode *inode, struct file *filp)
894{
895 struct pipe_inode_info *pipe;
896 bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
897 int ret;
898
899 filp->f_version = 0;
900
901 spin_lock(&inode->i_lock);
902 if (inode->i_pipe) {
903 pipe = inode->i_pipe;
904 pipe->files++;
905 spin_unlock(&inode->i_lock);
906 } else {
907 spin_unlock(&inode->i_lock);
908 pipe = alloc_pipe_info();
909 if (!pipe)
910 return -ENOMEM;
911 pipe->files = 1;
912 spin_lock(&inode->i_lock);
913 if (unlikely(inode->i_pipe)) {
914 inode->i_pipe->files++;
915 spin_unlock(&inode->i_lock);
916 free_pipe_info(pipe);
917 pipe = inode->i_pipe;
918 } else {
919 inode->i_pipe = pipe;
920 spin_unlock(&inode->i_lock);
921 }
922 }
923 filp->private_data = pipe;
924 /* OK, we have a pipe and it's pinned down */
925
926 __pipe_lock(pipe);
927
928 /* We can only do regular read/write on fifos */
929 filp->f_mode &= (FMODE_READ | FMODE_WRITE);
930
931 switch (filp->f_mode) {
932 case FMODE_READ:
933 /*
934 * O_RDONLY
935 * POSIX.1 says that O_NONBLOCK means return with the FIFO
936 * opened, even when there is no process writing the FIFO.
937 */
938 pipe->r_counter++;
939 if (pipe->readers++ == 0)
940 wake_up_partner(pipe);
941
942 if (!is_pipe && !pipe->writers) {
943 if ((filp->f_flags & O_NONBLOCK)) {
944 /* suppress EPOLLHUP until we have
945 * seen a writer */
946 filp->f_version = pipe->w_counter;
947 } else {
948 if (wait_for_partner(pipe, &pipe->w_counter))
949 goto err_rd;
950 }
951 }
952 break;
953
954 case FMODE_WRITE:
955 /*
956 * O_WRONLY
957 * POSIX.1 says that O_NONBLOCK means return -1 with
958 * errno=ENXIO when there is no process reading the FIFO.
959 */
960 ret = -ENXIO;
961 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
962 goto err;
963
964 pipe->w_counter++;
965 if (!pipe->writers++)
966 wake_up_partner(pipe);
967
968 if (!is_pipe && !pipe->readers) {
969 if (wait_for_partner(pipe, &pipe->r_counter))
970 goto err_wr;
971 }
972 break;
973
974 case FMODE_READ | FMODE_WRITE:
975 /*
976 * O_RDWR
977 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
978 * This implementation will NEVER block on a O_RDWR open, since
979 * the process can at least talk to itself.
980 */
981
982 pipe->readers++;
983 pipe->writers++;
984 pipe->r_counter++;
985 pipe->w_counter++;
986 if (pipe->readers == 1 || pipe->writers == 1)
987 wake_up_partner(pipe);
988 break;
989
990 default:
991 ret = -EINVAL;
992 goto err;
993 }
994
995 /* Ok! */
996 __pipe_unlock(pipe);
997 return 0;
998
999err_rd:
1000 if (!--pipe->readers)
1001 wake_up_interruptible(&pipe->wait);
1002 ret = -ERESTARTSYS;
1003 goto err;
1004
1005err_wr:
1006 if (!--pipe->writers)
1007 wake_up_interruptible(&pipe->wait);
1008 ret = -ERESTARTSYS;
1009 goto err;
1010
1011err:
1012 __pipe_unlock(pipe);
1013
1014 put_pipe_info(inode, pipe);
1015 return ret;
1016}
1017
1018const struct file_operations pipefifo_fops = {
1019 .open = fifo_open,
1020 .llseek = no_llseek,
1021 .read_iter = pipe_read,
1022 .write_iter = pipe_write,
1023 .poll = pipe_poll,
1024 .unlocked_ioctl = pipe_ioctl,
1025 .release = pipe_release,
1026 .fasync = pipe_fasync,
1027};
1028
1029/*
1030 * Currently we rely on the pipe array holding a power-of-2 number
1031 * of pages. Returns 0 on error.
1032 */
1033unsigned int round_pipe_size(unsigned long size)
1034{
1035 if (size > (1U << 31))
1036 return 0;
1037
1038 /* Minimum pipe size, as required by POSIX */
1039 if (size < PAGE_SIZE)
1040 return PAGE_SIZE;
1041
1042 return roundup_pow_of_two(size);
1043}
1044
1045/*
1046 * Allocate a new array of pipe buffers and copy the info over. Returns the
1047 * pipe size if successful, or return -ERROR on error.
1048 */
1049static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
1050{
1051 struct pipe_buffer *bufs;
1052 unsigned int size, nr_pages;
1053 unsigned long user_bufs;
1054 long ret = 0;
1055
1056 size = round_pipe_size(arg);
1057 nr_pages = size >> PAGE_SHIFT;
1058
1059 if (!nr_pages)
1060 return -EINVAL;
1061
1062 /*
1063 * If trying to increase the pipe capacity, check that an
1064 * unprivileged user is not trying to exceed various limits
1065 * (soft limit check here, hard limit check just below).
1066 * Decreasing the pipe capacity is always permitted, even
1067 * if the user is currently over a limit.
1068 */
1069 if (nr_pages > pipe->buffers &&
1070 size > pipe_max_size && !capable(CAP_SYS_RESOURCE))
1071 return -EPERM;
1072
1073 user_bufs = account_pipe_buffers(pipe->user, pipe->buffers, nr_pages);
1074
1075 if (nr_pages > pipe->buffers &&
1076 (too_many_pipe_buffers_hard(user_bufs) ||
1077 too_many_pipe_buffers_soft(user_bufs)) &&
1078 is_unprivileged_user()) {
1079 ret = -EPERM;
1080 goto out_revert_acct;
1081 }
1082
1083 /*
1084 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1085 * expect a lot of shrink+grow operations, just free and allocate
1086 * again like we would do for growing. If the pipe currently
1087 * contains more buffers than arg, then return busy.
1088 */
1089 if (nr_pages < pipe->nrbufs) {
1090 ret = -EBUSY;
1091 goto out_revert_acct;
1092 }
1093
1094 bufs = kcalloc(nr_pages, sizeof(*bufs),
1095 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
1096 if (unlikely(!bufs)) {
1097 ret = -ENOMEM;
1098 goto out_revert_acct;
1099 }
1100
1101 /*
1102 * The pipe array wraps around, so just start the new one at zero
1103 * and adjust the indexes.
1104 */
1105 if (pipe->nrbufs) {
1106 unsigned int tail;
1107 unsigned int head;
1108
1109 tail = pipe->curbuf + pipe->nrbufs;
1110 if (tail < pipe->buffers)
1111 tail = 0;
1112 else
1113 tail &= (pipe->buffers - 1);
1114
1115 head = pipe->nrbufs - tail;
1116 if (head)
1117 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1118 if (tail)
1119 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
1120 }
1121
1122 pipe->curbuf = 0;
1123 kfree(pipe->bufs);
1124 pipe->bufs = bufs;
1125 pipe->buffers = nr_pages;
1126 return nr_pages * PAGE_SIZE;
1127
1128out_revert_acct:
1129 (void) account_pipe_buffers(pipe->user, nr_pages, pipe->buffers);
1130 return ret;
1131}
1132
1133/*
1134 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1135 * location, so checking ->i_pipe is not enough to verify that this is a
1136 * pipe.
1137 */
1138struct pipe_inode_info *get_pipe_info(struct file *file)
1139{
1140 return file->f_op == &pipefifo_fops ? file->private_data : NULL;
1141}
1142
1143long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1144{
1145 struct pipe_inode_info *pipe;
1146 long ret;
1147
1148 pipe = get_pipe_info(file);
1149 if (!pipe)
1150 return -EBADF;
1151
1152 __pipe_lock(pipe);
1153
1154 switch (cmd) {
1155 case F_SETPIPE_SZ:
1156 ret = pipe_set_size(pipe, arg);
1157 break;
1158 case F_GETPIPE_SZ:
1159 ret = pipe->buffers * PAGE_SIZE;
1160 break;
1161 default:
1162 ret = -EINVAL;
1163 break;
1164 }
1165
1166 __pipe_unlock(pipe);
1167 return ret;
1168}
1169
1170static const struct super_operations pipefs_ops = {
1171 .destroy_inode = free_inode_nonrcu,
1172 .statfs = simple_statfs,
1173};
1174
1175/*
1176 * pipefs should _never_ be mounted by userland - too much of security hassle,
1177 * no real gain from having the whole whorehouse mounted. So we don't need
1178 * any operations on the root directory. However, we need a non-trivial
1179 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1180 */
1181static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1182 int flags, const char *dev_name, void *data)
1183{
1184 return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1185 &pipefs_dentry_operations, PIPEFS_MAGIC);
1186}
1187
1188static struct file_system_type pipe_fs_type = {
1189 .name = "pipefs",
1190 .mount = pipefs_mount,
1191 .kill_sb = kill_anon_super,
1192};
1193
1194static int __init init_pipe_fs(void)
1195{
1196 int err = register_filesystem(&pipe_fs_type);
1197
1198 if (!err) {
1199 pipe_mnt = kern_mount(&pipe_fs_type);
1200 if (IS_ERR(pipe_mnt)) {
1201 err = PTR_ERR(pipe_mnt);
1202 unregister_filesystem(&pipe_fs_type);
1203 }
1204 }
1205 return err;
1206}
1207
1208fs_initcall(init_pipe_fs);
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
28#include <linux/uaccess.h>
29#include <asm/ioctls.h>
30
31#include "internal.h"
32
33/*
34 * The max size that a non-root user is allowed to grow the pipe. Can
35 * be set by root in /proc/sys/fs/pipe-max-size
36 */
37unsigned int pipe_max_size = 1048576;
38
39/* Maximum allocatable pages per user. Hard limit is unset by default, soft
40 * matches default values.
41 */
42unsigned long pipe_user_pages_hard;
43unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;
44
45/*
46 * We use a start+len construction, which provides full use of the
47 * allocated memory.
48 * -- Florian Coosmann (FGC)
49 *
50 * Reads with count = 0 should always return 0.
51 * -- Julian Bradfield 1999-06-07.
52 *
53 * FIFOs and Pipes now generate SIGIO for both readers and writers.
54 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
55 *
56 * pipe_read & write cleanup
57 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
58 */
59
60static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
61{
62 if (pipe->files)
63 mutex_lock_nested(&pipe->mutex, subclass);
64}
65
66void pipe_lock(struct pipe_inode_info *pipe)
67{
68 /*
69 * pipe_lock() nests non-pipe inode locks (for writing to a file)
70 */
71 pipe_lock_nested(pipe, I_MUTEX_PARENT);
72}
73EXPORT_SYMBOL(pipe_lock);
74
75void pipe_unlock(struct pipe_inode_info *pipe)
76{
77 if (pipe->files)
78 mutex_unlock(&pipe->mutex);
79}
80EXPORT_SYMBOL(pipe_unlock);
81
82static inline void __pipe_lock(struct pipe_inode_info *pipe)
83{
84 mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
85}
86
87static inline void __pipe_unlock(struct pipe_inode_info *pipe)
88{
89 mutex_unlock(&pipe->mutex);
90}
91
92void pipe_double_lock(struct pipe_inode_info *pipe1,
93 struct pipe_inode_info *pipe2)
94{
95 BUG_ON(pipe1 == pipe2);
96
97 if (pipe1 < pipe2) {
98 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
99 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
100 } else {
101 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
102 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
103 }
104}
105
106/* Drop the inode semaphore and wait for a pipe event, atomically */
107void pipe_wait(struct pipe_inode_info *pipe)
108{
109 DEFINE_WAIT(wait);
110
111 /*
112 * Pipes are system-local resources, so sleeping on them
113 * is considered a noninteractive wait:
114 */
115 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
116 pipe_unlock(pipe);
117 schedule();
118 finish_wait(&pipe->wait, &wait);
119 pipe_lock(pipe);
120}
121
122static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
123 struct pipe_buffer *buf)
124{
125 struct page *page = buf->page;
126
127 /*
128 * If nobody else uses this page, and we don't already have a
129 * temporary page, let's keep track of it as a one-deep
130 * allocation cache. (Otherwise just release our reference to it)
131 */
132 if (page_count(page) == 1 && !pipe->tmp_page)
133 pipe->tmp_page = page;
134 else
135 put_page(page);
136}
137
138static int anon_pipe_buf_steal(struct pipe_inode_info *pipe,
139 struct pipe_buffer *buf)
140{
141 struct page *page = buf->page;
142
143 if (page_count(page) == 1) {
144 memcg_kmem_uncharge(page, 0);
145 __SetPageLocked(page);
146 return 0;
147 }
148 return 1;
149}
150
151/**
152 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
153 * @pipe: the pipe that the buffer belongs to
154 * @buf: the buffer to attempt to steal
155 *
156 * Description:
157 * This function attempts to steal the &struct page attached to
158 * @buf. If successful, this function returns 0 and returns with
159 * the page locked. The caller may then reuse the page for whatever
160 * he wishes; the typical use is insertion into a different file
161 * page cache.
162 */
163int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
164 struct pipe_buffer *buf)
165{
166 struct page *page = buf->page;
167
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 */
173 if (page_count(page) == 1) {
174 lock_page(page);
175 return 0;
176 }
177
178 return 1;
179}
180EXPORT_SYMBOL(generic_pipe_buf_steal);
181
182/**
183 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
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 */
192bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
193{
194 return try_get_page(buf->page);
195}
196EXPORT_SYMBOL(generic_pipe_buf_get);
197
198/**
199 * generic_pipe_buf_confirm - verify contents of the pipe buffer
200 * @info: the pipe that the buffer belongs to
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 */
207int generic_pipe_buf_confirm(struct pipe_inode_info *info,
208 struct pipe_buffer *buf)
209{
210 return 0;
211}
212EXPORT_SYMBOL(generic_pipe_buf_confirm);
213
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{
225 put_page(buf->page);
226}
227EXPORT_SYMBOL(generic_pipe_buf_release);
228
229/* New data written to a pipe may be appended to a buffer with this type. */
230static const struct pipe_buf_operations anon_pipe_buf_ops = {
231 .confirm = generic_pipe_buf_confirm,
232 .release = anon_pipe_buf_release,
233 .steal = anon_pipe_buf_steal,
234 .get = generic_pipe_buf_get,
235};
236
237static const struct pipe_buf_operations anon_pipe_buf_nomerge_ops = {
238 .confirm = generic_pipe_buf_confirm,
239 .release = anon_pipe_buf_release,
240 .steal = anon_pipe_buf_steal,
241 .get = generic_pipe_buf_get,
242};
243
244static const struct pipe_buf_operations packet_pipe_buf_ops = {
245 .confirm = generic_pipe_buf_confirm,
246 .release = anon_pipe_buf_release,
247 .steal = anon_pipe_buf_steal,
248 .get = generic_pipe_buf_get,
249};
250
251/**
252 * pipe_buf_mark_unmergeable - mark a &struct pipe_buffer as unmergeable
253 * @buf: the buffer to mark
254 *
255 * Description:
256 * This function ensures that no future writes will be merged into the
257 * given &struct pipe_buffer. This is necessary when multiple pipe buffers
258 * share the same backing page.
259 */
260void pipe_buf_mark_unmergeable(struct pipe_buffer *buf)
261{
262 if (buf->ops == &anon_pipe_buf_ops)
263 buf->ops = &anon_pipe_buf_nomerge_ops;
264}
265
266static bool pipe_buf_can_merge(struct pipe_buffer *buf)
267{
268 return buf->ops == &anon_pipe_buf_ops;
269}
270
271static ssize_t
272pipe_read(struct kiocb *iocb, struct iov_iter *to)
273{
274 size_t total_len = iov_iter_count(to);
275 struct file *filp = iocb->ki_filp;
276 struct pipe_inode_info *pipe = filp->private_data;
277 int do_wakeup;
278 ssize_t ret;
279
280 /* Null read succeeds. */
281 if (unlikely(total_len == 0))
282 return 0;
283
284 do_wakeup = 0;
285 ret = 0;
286 __pipe_lock(pipe);
287 for (;;) {
288 int bufs = pipe->nrbufs;
289 if (bufs) {
290 int curbuf = pipe->curbuf;
291 struct pipe_buffer *buf = pipe->bufs + curbuf;
292 size_t chars = buf->len;
293 size_t written;
294 int error;
295
296 if (chars > total_len)
297 chars = total_len;
298
299 error = pipe_buf_confirm(pipe, buf);
300 if (error) {
301 if (!ret)
302 ret = error;
303 break;
304 }
305
306 written = copy_page_to_iter(buf->page, buf->offset, chars, to);
307 if (unlikely(written < chars)) {
308 if (!ret)
309 ret = -EFAULT;
310 break;
311 }
312 ret += chars;
313 buf->offset += chars;
314 buf->len -= chars;
315
316 /* Was it a packet buffer? Clean up and exit */
317 if (buf->flags & PIPE_BUF_FLAG_PACKET) {
318 total_len = chars;
319 buf->len = 0;
320 }
321
322 if (!buf->len) {
323 pipe_buf_release(pipe, buf);
324 curbuf = (curbuf + 1) & (pipe->buffers - 1);
325 pipe->curbuf = curbuf;
326 pipe->nrbufs = --bufs;
327 do_wakeup = 1;
328 }
329 total_len -= chars;
330 if (!total_len)
331 break; /* common path: read succeeded */
332 }
333 if (bufs) /* More to do? */
334 continue;
335 if (!pipe->writers)
336 break;
337 if (!pipe->waiting_writers) {
338 /* syscall merging: Usually we must not sleep
339 * if O_NONBLOCK is set, or if we got some data.
340 * But if a writer sleeps in kernel space, then
341 * we can wait for that data without violating POSIX.
342 */
343 if (ret)
344 break;
345 if (filp->f_flags & O_NONBLOCK) {
346 ret = -EAGAIN;
347 break;
348 }
349 }
350 if (signal_pending(current)) {
351 if (!ret)
352 ret = -ERESTARTSYS;
353 break;
354 }
355 if (do_wakeup) {
356 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
357 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
358 }
359 pipe_wait(pipe);
360 }
361 __pipe_unlock(pipe);
362
363 /* Signal writers asynchronously that there is more room. */
364 if (do_wakeup) {
365 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
366 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
367 }
368 if (ret > 0)
369 file_accessed(filp);
370 return ret;
371}
372
373static inline int is_packetized(struct file *file)
374{
375 return (file->f_flags & O_DIRECT) != 0;
376}
377
378static ssize_t
379pipe_write(struct kiocb *iocb, struct iov_iter *from)
380{
381 struct file *filp = iocb->ki_filp;
382 struct pipe_inode_info *pipe = filp->private_data;
383 ssize_t ret = 0;
384 int do_wakeup = 0;
385 size_t total_len = iov_iter_count(from);
386 ssize_t chars;
387
388 /* Null write succeeds. */
389 if (unlikely(total_len == 0))
390 return 0;
391
392 __pipe_lock(pipe);
393
394 if (!pipe->readers) {
395 send_sig(SIGPIPE, current, 0);
396 ret = -EPIPE;
397 goto out;
398 }
399
400 /* We try to merge small writes */
401 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
402 if (pipe->nrbufs && chars != 0) {
403 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
404 (pipe->buffers - 1);
405 struct pipe_buffer *buf = pipe->bufs + lastbuf;
406 int offset = buf->offset + buf->len;
407
408 if (pipe_buf_can_merge(buf) && offset + chars <= PAGE_SIZE) {
409 ret = pipe_buf_confirm(pipe, buf);
410 if (ret)
411 goto out;
412
413 ret = copy_page_from_iter(buf->page, offset, chars, from);
414 if (unlikely(ret < chars)) {
415 ret = -EFAULT;
416 goto out;
417 }
418 do_wakeup = 1;
419 buf->len += ret;
420 if (!iov_iter_count(from))
421 goto out;
422 }
423 }
424
425 for (;;) {
426 int bufs;
427
428 if (!pipe->readers) {
429 send_sig(SIGPIPE, current, 0);
430 if (!ret)
431 ret = -EPIPE;
432 break;
433 }
434 bufs = pipe->nrbufs;
435 if (bufs < pipe->buffers) {
436 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
437 struct pipe_buffer *buf = pipe->bufs + newbuf;
438 struct page *page = pipe->tmp_page;
439 int copied;
440
441 if (!page) {
442 page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
443 if (unlikely(!page)) {
444 ret = ret ? : -ENOMEM;
445 break;
446 }
447 pipe->tmp_page = page;
448 }
449 /* Always wake up, even if the copy fails. Otherwise
450 * we lock up (O_NONBLOCK-)readers that sleep due to
451 * syscall merging.
452 * FIXME! Is this really true?
453 */
454 do_wakeup = 1;
455 copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
456 if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
457 if (!ret)
458 ret = -EFAULT;
459 break;
460 }
461 ret += copied;
462
463 /* Insert it into the buffer array */
464 buf->page = page;
465 buf->ops = &anon_pipe_buf_ops;
466 buf->offset = 0;
467 buf->len = copied;
468 buf->flags = 0;
469 if (is_packetized(filp)) {
470 buf->ops = &packet_pipe_buf_ops;
471 buf->flags = PIPE_BUF_FLAG_PACKET;
472 }
473 pipe->nrbufs = ++bufs;
474 pipe->tmp_page = NULL;
475
476 if (!iov_iter_count(from))
477 break;
478 }
479 if (bufs < pipe->buffers)
480 continue;
481 if (filp->f_flags & O_NONBLOCK) {
482 if (!ret)
483 ret = -EAGAIN;
484 break;
485 }
486 if (signal_pending(current)) {
487 if (!ret)
488 ret = -ERESTARTSYS;
489 break;
490 }
491 if (do_wakeup) {
492 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
493 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
494 do_wakeup = 0;
495 }
496 pipe->waiting_writers++;
497 pipe_wait(pipe);
498 pipe->waiting_writers--;
499 }
500out:
501 __pipe_unlock(pipe);
502 if (do_wakeup) {
503 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
504 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
505 }
506 if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
507 int err = file_update_time(filp);
508 if (err)
509 ret = err;
510 sb_end_write(file_inode(filp)->i_sb);
511 }
512 return ret;
513}
514
515static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
516{
517 struct pipe_inode_info *pipe = filp->private_data;
518 int count, buf, nrbufs;
519
520 switch (cmd) {
521 case FIONREAD:
522 __pipe_lock(pipe);
523 count = 0;
524 buf = pipe->curbuf;
525 nrbufs = pipe->nrbufs;
526 while (--nrbufs >= 0) {
527 count += pipe->bufs[buf].len;
528 buf = (buf+1) & (pipe->buffers - 1);
529 }
530 __pipe_unlock(pipe);
531
532 return put_user(count, (int __user *)arg);
533 default:
534 return -ENOIOCTLCMD;
535 }
536}
537
538/* No kernel lock held - fine */
539static __poll_t
540pipe_poll(struct file *filp, poll_table *wait)
541{
542 __poll_t mask;
543 struct pipe_inode_info *pipe = filp->private_data;
544 int nrbufs;
545
546 poll_wait(filp, &pipe->wait, wait);
547
548 /* Reading only -- no need for acquiring the semaphore. */
549 nrbufs = pipe->nrbufs;
550 mask = 0;
551 if (filp->f_mode & FMODE_READ) {
552 mask = (nrbufs > 0) ? EPOLLIN | EPOLLRDNORM : 0;
553 if (!pipe->writers && filp->f_version != pipe->w_counter)
554 mask |= EPOLLHUP;
555 }
556
557 if (filp->f_mode & FMODE_WRITE) {
558 mask |= (nrbufs < pipe->buffers) ? EPOLLOUT | EPOLLWRNORM : 0;
559 /*
560 * Most Unices do not set EPOLLERR for FIFOs but on Linux they
561 * behave exactly like pipes for poll().
562 */
563 if (!pipe->readers)
564 mask |= EPOLLERR;
565 }
566
567 return mask;
568}
569
570static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
571{
572 int kill = 0;
573
574 spin_lock(&inode->i_lock);
575 if (!--pipe->files) {
576 inode->i_pipe = NULL;
577 kill = 1;
578 }
579 spin_unlock(&inode->i_lock);
580
581 if (kill)
582 free_pipe_info(pipe);
583}
584
585static int
586pipe_release(struct inode *inode, struct file *file)
587{
588 struct pipe_inode_info *pipe = file->private_data;
589
590 __pipe_lock(pipe);
591 if (file->f_mode & FMODE_READ)
592 pipe->readers--;
593 if (file->f_mode & FMODE_WRITE)
594 pipe->writers--;
595
596 if (pipe->readers || pipe->writers) {
597 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM | EPOLLERR | EPOLLHUP);
598 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
599 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
600 }
601 __pipe_unlock(pipe);
602
603 put_pipe_info(inode, pipe);
604 return 0;
605}
606
607static int
608pipe_fasync(int fd, struct file *filp, int on)
609{
610 struct pipe_inode_info *pipe = filp->private_data;
611 int retval = 0;
612
613 __pipe_lock(pipe);
614 if (filp->f_mode & FMODE_READ)
615 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
616 if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
617 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
618 if (retval < 0 && (filp->f_mode & FMODE_READ))
619 /* this can happen only if on == T */
620 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
621 }
622 __pipe_unlock(pipe);
623 return retval;
624}
625
626static unsigned long account_pipe_buffers(struct user_struct *user,
627 unsigned long old, unsigned long new)
628{
629 return atomic_long_add_return(new - old, &user->pipe_bufs);
630}
631
632static bool too_many_pipe_buffers_soft(unsigned long user_bufs)
633{
634 unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft);
635
636 return soft_limit && user_bufs > soft_limit;
637}
638
639static bool too_many_pipe_buffers_hard(unsigned long user_bufs)
640{
641 unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard);
642
643 return hard_limit && user_bufs > hard_limit;
644}
645
646static bool is_unprivileged_user(void)
647{
648 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
649}
650
651struct pipe_inode_info *alloc_pipe_info(void)
652{
653 struct pipe_inode_info *pipe;
654 unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
655 struct user_struct *user = get_current_user();
656 unsigned long user_bufs;
657 unsigned int max_size = READ_ONCE(pipe_max_size);
658
659 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT);
660 if (pipe == NULL)
661 goto out_free_uid;
662
663 if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE))
664 pipe_bufs = max_size >> PAGE_SHIFT;
665
666 user_bufs = account_pipe_buffers(user, 0, pipe_bufs);
667
668 if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) {
669 user_bufs = account_pipe_buffers(user, pipe_bufs, 1);
670 pipe_bufs = 1;
671 }
672
673 if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user())
674 goto out_revert_acct;
675
676 pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer),
677 GFP_KERNEL_ACCOUNT);
678
679 if (pipe->bufs) {
680 init_waitqueue_head(&pipe->wait);
681 pipe->r_counter = pipe->w_counter = 1;
682 pipe->buffers = pipe_bufs;
683 pipe->user = user;
684 mutex_init(&pipe->mutex);
685 return pipe;
686 }
687
688out_revert_acct:
689 (void) account_pipe_buffers(user, pipe_bufs, 0);
690 kfree(pipe);
691out_free_uid:
692 free_uid(user);
693 return NULL;
694}
695
696void free_pipe_info(struct pipe_inode_info *pipe)
697{
698 int i;
699
700 (void) account_pipe_buffers(pipe->user, pipe->buffers, 0);
701 free_uid(pipe->user);
702 for (i = 0; i < pipe->buffers; i++) {
703 struct pipe_buffer *buf = pipe->bufs + i;
704 if (buf->ops)
705 pipe_buf_release(pipe, buf);
706 }
707 if (pipe->tmp_page)
708 __free_page(pipe->tmp_page);
709 kfree(pipe->bufs);
710 kfree(pipe);
711}
712
713static struct vfsmount *pipe_mnt __read_mostly;
714
715/*
716 * pipefs_dname() is called from d_path().
717 */
718static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
719{
720 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
721 d_inode(dentry)->i_ino);
722}
723
724static const struct dentry_operations pipefs_dentry_operations = {
725 .d_dname = pipefs_dname,
726};
727
728static struct inode * get_pipe_inode(void)
729{
730 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
731 struct pipe_inode_info *pipe;
732
733 if (!inode)
734 goto fail_inode;
735
736 inode->i_ino = get_next_ino();
737
738 pipe = alloc_pipe_info();
739 if (!pipe)
740 goto fail_iput;
741
742 inode->i_pipe = pipe;
743 pipe->files = 2;
744 pipe->readers = pipe->writers = 1;
745 inode->i_fop = &pipefifo_fops;
746
747 /*
748 * Mark the inode dirty from the very beginning,
749 * that way it will never be moved to the dirty
750 * list because "mark_inode_dirty()" will think
751 * that it already _is_ on the dirty list.
752 */
753 inode->i_state = I_DIRTY;
754 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
755 inode->i_uid = current_fsuid();
756 inode->i_gid = current_fsgid();
757 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
758
759 return inode;
760
761fail_iput:
762 iput(inode);
763
764fail_inode:
765 return NULL;
766}
767
768int create_pipe_files(struct file **res, int flags)
769{
770 struct inode *inode = get_pipe_inode();
771 struct file *f;
772
773 if (!inode)
774 return -ENFILE;
775
776 f = alloc_file_pseudo(inode, pipe_mnt, "",
777 O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)),
778 &pipefifo_fops);
779 if (IS_ERR(f)) {
780 free_pipe_info(inode->i_pipe);
781 iput(inode);
782 return PTR_ERR(f);
783 }
784
785 f->private_data = inode->i_pipe;
786
787 res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK),
788 &pipefifo_fops);
789 if (IS_ERR(res[0])) {
790 put_pipe_info(inode, inode->i_pipe);
791 fput(f);
792 return PTR_ERR(res[0]);
793 }
794 res[0]->private_data = inode->i_pipe;
795 res[1] = f;
796 return 0;
797}
798
799static int __do_pipe_flags(int *fd, struct file **files, int flags)
800{
801 int error;
802 int fdw, fdr;
803
804 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
805 return -EINVAL;
806
807 error = create_pipe_files(files, flags);
808 if (error)
809 return error;
810
811 error = get_unused_fd_flags(flags);
812 if (error < 0)
813 goto err_read_pipe;
814 fdr = error;
815
816 error = get_unused_fd_flags(flags);
817 if (error < 0)
818 goto err_fdr;
819 fdw = error;
820
821 audit_fd_pair(fdr, fdw);
822 fd[0] = fdr;
823 fd[1] = fdw;
824 return 0;
825
826 err_fdr:
827 put_unused_fd(fdr);
828 err_read_pipe:
829 fput(files[0]);
830 fput(files[1]);
831 return error;
832}
833
834int do_pipe_flags(int *fd, int flags)
835{
836 struct file *files[2];
837 int error = __do_pipe_flags(fd, files, flags);
838 if (!error) {
839 fd_install(fd[0], files[0]);
840 fd_install(fd[1], files[1]);
841 }
842 return error;
843}
844
845/*
846 * sys_pipe() is the normal C calling standard for creating
847 * a pipe. It's not the way Unix traditionally does this, though.
848 */
849static int do_pipe2(int __user *fildes, int flags)
850{
851 struct file *files[2];
852 int fd[2];
853 int error;
854
855 error = __do_pipe_flags(fd, files, flags);
856 if (!error) {
857 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
858 fput(files[0]);
859 fput(files[1]);
860 put_unused_fd(fd[0]);
861 put_unused_fd(fd[1]);
862 error = -EFAULT;
863 } else {
864 fd_install(fd[0], files[0]);
865 fd_install(fd[1], files[1]);
866 }
867 }
868 return error;
869}
870
871SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
872{
873 return do_pipe2(fildes, flags);
874}
875
876SYSCALL_DEFINE1(pipe, int __user *, fildes)
877{
878 return do_pipe2(fildes, 0);
879}
880
881static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
882{
883 int cur = *cnt;
884
885 while (cur == *cnt) {
886 pipe_wait(pipe);
887 if (signal_pending(current))
888 break;
889 }
890 return cur == *cnt ? -ERESTARTSYS : 0;
891}
892
893static void wake_up_partner(struct pipe_inode_info *pipe)
894{
895 wake_up_interruptible(&pipe->wait);
896}
897
898static int fifo_open(struct inode *inode, struct file *filp)
899{
900 struct pipe_inode_info *pipe;
901 bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
902 int ret;
903
904 filp->f_version = 0;
905
906 spin_lock(&inode->i_lock);
907 if (inode->i_pipe) {
908 pipe = inode->i_pipe;
909 pipe->files++;
910 spin_unlock(&inode->i_lock);
911 } else {
912 spin_unlock(&inode->i_lock);
913 pipe = alloc_pipe_info();
914 if (!pipe)
915 return -ENOMEM;
916 pipe->files = 1;
917 spin_lock(&inode->i_lock);
918 if (unlikely(inode->i_pipe)) {
919 inode->i_pipe->files++;
920 spin_unlock(&inode->i_lock);
921 free_pipe_info(pipe);
922 pipe = inode->i_pipe;
923 } else {
924 inode->i_pipe = pipe;
925 spin_unlock(&inode->i_lock);
926 }
927 }
928 filp->private_data = pipe;
929 /* OK, we have a pipe and it's pinned down */
930
931 __pipe_lock(pipe);
932
933 /* We can only do regular read/write on fifos */
934 filp->f_mode &= (FMODE_READ | FMODE_WRITE);
935
936 switch (filp->f_mode) {
937 case FMODE_READ:
938 /*
939 * O_RDONLY
940 * POSIX.1 says that O_NONBLOCK means return with the FIFO
941 * opened, even when there is no process writing the FIFO.
942 */
943 pipe->r_counter++;
944 if (pipe->readers++ == 0)
945 wake_up_partner(pipe);
946
947 if (!is_pipe && !pipe->writers) {
948 if ((filp->f_flags & O_NONBLOCK)) {
949 /* suppress EPOLLHUP until we have
950 * seen a writer */
951 filp->f_version = pipe->w_counter;
952 } else {
953 if (wait_for_partner(pipe, &pipe->w_counter))
954 goto err_rd;
955 }
956 }
957 break;
958
959 case FMODE_WRITE:
960 /*
961 * O_WRONLY
962 * POSIX.1 says that O_NONBLOCK means return -1 with
963 * errno=ENXIO when there is no process reading the FIFO.
964 */
965 ret = -ENXIO;
966 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
967 goto err;
968
969 pipe->w_counter++;
970 if (!pipe->writers++)
971 wake_up_partner(pipe);
972
973 if (!is_pipe && !pipe->readers) {
974 if (wait_for_partner(pipe, &pipe->r_counter))
975 goto err_wr;
976 }
977 break;
978
979 case FMODE_READ | FMODE_WRITE:
980 /*
981 * O_RDWR
982 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
983 * This implementation will NEVER block on a O_RDWR open, since
984 * the process can at least talk to itself.
985 */
986
987 pipe->readers++;
988 pipe->writers++;
989 pipe->r_counter++;
990 pipe->w_counter++;
991 if (pipe->readers == 1 || pipe->writers == 1)
992 wake_up_partner(pipe);
993 break;
994
995 default:
996 ret = -EINVAL;
997 goto err;
998 }
999
1000 /* Ok! */
1001 __pipe_unlock(pipe);
1002 return 0;
1003
1004err_rd:
1005 if (!--pipe->readers)
1006 wake_up_interruptible(&pipe->wait);
1007 ret = -ERESTARTSYS;
1008 goto err;
1009
1010err_wr:
1011 if (!--pipe->writers)
1012 wake_up_interruptible(&pipe->wait);
1013 ret = -ERESTARTSYS;
1014 goto err;
1015
1016err:
1017 __pipe_unlock(pipe);
1018
1019 put_pipe_info(inode, pipe);
1020 return ret;
1021}
1022
1023const struct file_operations pipefifo_fops = {
1024 .open = fifo_open,
1025 .llseek = no_llseek,
1026 .read_iter = pipe_read,
1027 .write_iter = pipe_write,
1028 .poll = pipe_poll,
1029 .unlocked_ioctl = pipe_ioctl,
1030 .release = pipe_release,
1031 .fasync = pipe_fasync,
1032};
1033
1034/*
1035 * Currently we rely on the pipe array holding a power-of-2 number
1036 * of pages. Returns 0 on error.
1037 */
1038unsigned int round_pipe_size(unsigned long size)
1039{
1040 if (size > (1U << 31))
1041 return 0;
1042
1043 /* Minimum pipe size, as required by POSIX */
1044 if (size < PAGE_SIZE)
1045 return PAGE_SIZE;
1046
1047 return roundup_pow_of_two(size);
1048}
1049
1050/*
1051 * Allocate a new array of pipe buffers and copy the info over. Returns the
1052 * pipe size if successful, or return -ERROR on error.
1053 */
1054static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
1055{
1056 struct pipe_buffer *bufs;
1057 unsigned int size, nr_pages;
1058 unsigned long user_bufs;
1059 long ret = 0;
1060
1061 size = round_pipe_size(arg);
1062 nr_pages = size >> PAGE_SHIFT;
1063
1064 if (!nr_pages)
1065 return -EINVAL;
1066
1067 /*
1068 * If trying to increase the pipe capacity, check that an
1069 * unprivileged user is not trying to exceed various limits
1070 * (soft limit check here, hard limit check just below).
1071 * Decreasing the pipe capacity is always permitted, even
1072 * if the user is currently over a limit.
1073 */
1074 if (nr_pages > pipe->buffers &&
1075 size > pipe_max_size && !capable(CAP_SYS_RESOURCE))
1076 return -EPERM;
1077
1078 user_bufs = account_pipe_buffers(pipe->user, pipe->buffers, nr_pages);
1079
1080 if (nr_pages > pipe->buffers &&
1081 (too_many_pipe_buffers_hard(user_bufs) ||
1082 too_many_pipe_buffers_soft(user_bufs)) &&
1083 is_unprivileged_user()) {
1084 ret = -EPERM;
1085 goto out_revert_acct;
1086 }
1087
1088 /*
1089 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1090 * expect a lot of shrink+grow operations, just free and allocate
1091 * again like we would do for growing. If the pipe currently
1092 * contains more buffers than arg, then return busy.
1093 */
1094 if (nr_pages < pipe->nrbufs) {
1095 ret = -EBUSY;
1096 goto out_revert_acct;
1097 }
1098
1099 bufs = kcalloc(nr_pages, sizeof(*bufs),
1100 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
1101 if (unlikely(!bufs)) {
1102 ret = -ENOMEM;
1103 goto out_revert_acct;
1104 }
1105
1106 /*
1107 * The pipe array wraps around, so just start the new one at zero
1108 * and adjust the indexes.
1109 */
1110 if (pipe->nrbufs) {
1111 unsigned int tail;
1112 unsigned int head;
1113
1114 tail = pipe->curbuf + pipe->nrbufs;
1115 if (tail < pipe->buffers)
1116 tail = 0;
1117 else
1118 tail &= (pipe->buffers - 1);
1119
1120 head = pipe->nrbufs - tail;
1121 if (head)
1122 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1123 if (tail)
1124 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
1125 }
1126
1127 pipe->curbuf = 0;
1128 kfree(pipe->bufs);
1129 pipe->bufs = bufs;
1130 pipe->buffers = nr_pages;
1131 return nr_pages * PAGE_SIZE;
1132
1133out_revert_acct:
1134 (void) account_pipe_buffers(pipe->user, nr_pages, pipe->buffers);
1135 return ret;
1136}
1137
1138/*
1139 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1140 * location, so checking ->i_pipe is not enough to verify that this is a
1141 * pipe.
1142 */
1143struct pipe_inode_info *get_pipe_info(struct file *file)
1144{
1145 return file->f_op == &pipefifo_fops ? file->private_data : NULL;
1146}
1147
1148long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1149{
1150 struct pipe_inode_info *pipe;
1151 long ret;
1152
1153 pipe = get_pipe_info(file);
1154 if (!pipe)
1155 return -EBADF;
1156
1157 __pipe_lock(pipe);
1158
1159 switch (cmd) {
1160 case F_SETPIPE_SZ:
1161 ret = pipe_set_size(pipe, arg);
1162 break;
1163 case F_GETPIPE_SZ:
1164 ret = pipe->buffers * PAGE_SIZE;
1165 break;
1166 default:
1167 ret = -EINVAL;
1168 break;
1169 }
1170
1171 __pipe_unlock(pipe);
1172 return ret;
1173}
1174
1175static const struct super_operations pipefs_ops = {
1176 .destroy_inode = free_inode_nonrcu,
1177 .statfs = simple_statfs,
1178};
1179
1180/*
1181 * pipefs should _never_ be mounted by userland - too much of security hassle,
1182 * no real gain from having the whole whorehouse mounted. So we don't need
1183 * any operations on the root directory. However, we need a non-trivial
1184 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1185 */
1186
1187static int pipefs_init_fs_context(struct fs_context *fc)
1188{
1189 struct pseudo_fs_context *ctx = init_pseudo(fc, PIPEFS_MAGIC);
1190 if (!ctx)
1191 return -ENOMEM;
1192 ctx->ops = &pipefs_ops;
1193 ctx->dops = &pipefs_dentry_operations;
1194 return 0;
1195}
1196
1197static struct file_system_type pipe_fs_type = {
1198 .name = "pipefs",
1199 .init_fs_context = pipefs_init_fs_context,
1200 .kill_sb = kill_anon_super,
1201};
1202
1203static int __init init_pipe_fs(void)
1204{
1205 int err = register_filesystem(&pipe_fs_type);
1206
1207 if (!err) {
1208 pipe_mnt = kern_mount(&pipe_fs_type);
1209 if (IS_ERR(pipe_mnt)) {
1210 err = PTR_ERR(pipe_mnt);
1211 unregister_filesystem(&pipe_fs_type);
1212 }
1213 }
1214 return err;
1215}
1216
1217fs_initcall(init_pipe_fs);