Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[platform/adaptation/renesas_rcar/renesas_kernel.git] / fs / pipe.c
1 /*
2  *  linux/fs/pipe.c
3  *
4  *  Copyright (C) 1991, 1992, 1999  Linus Torvalds
5  */
6
7 #include <linux/mm.h>
8 #include <linux/file.h>
9 #include <linux/poll.h>
10 #include <linux/slab.h>
11 #include <linux/module.h>
12 #include <linux/init.h>
13 #include <linux/fs.h>
14 #include <linux/log2.h>
15 #include <linux/mount.h>
16 #include <linux/pipe_fs_i.h>
17 #include <linux/uio.h>
18 #include <linux/highmem.h>
19 #include <linux/pagemap.h>
20 #include <linux/audit.h>
21 #include <linux/syscalls.h>
22 #include <linux/fcntl.h>
23
24 #include <asm/uaccess.h>
25 #include <asm/ioctls.h>
26
27 /*
28  * The max size that a non-root user is allowed to grow the pipe. Can
29  * be set by root in /proc/sys/fs/pipe-max-size
30  */
31 unsigned int pipe_max_size = 1048576;
32
33 /*
34  * Minimum pipe size, as required by POSIX
35  */
36 unsigned int pipe_min_size = PAGE_SIZE;
37
38 /*
39  * We use a start+len construction, which provides full use of the 
40  * allocated memory.
41  * -- Florian Coosmann (FGC)
42  * 
43  * Reads with count = 0 should always return 0.
44  * -- Julian Bradfield 1999-06-07.
45  *
46  * FIFOs and Pipes now generate SIGIO for both readers and writers.
47  * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
48  *
49  * pipe_read & write cleanup
50  * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
51  */
52
53 static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
54 {
55         if (pipe->inode)
56                 mutex_lock_nested(&pipe->inode->i_mutex, subclass);
57 }
58
59 void pipe_lock(struct pipe_inode_info *pipe)
60 {
61         /*
62          * pipe_lock() nests non-pipe inode locks (for writing to a file)
63          */
64         pipe_lock_nested(pipe, I_MUTEX_PARENT);
65 }
66 EXPORT_SYMBOL(pipe_lock);
67
68 void pipe_unlock(struct pipe_inode_info *pipe)
69 {
70         if (pipe->inode)
71                 mutex_unlock(&pipe->inode->i_mutex);
72 }
73 EXPORT_SYMBOL(pipe_unlock);
74
75 void pipe_double_lock(struct pipe_inode_info *pipe1,
76                       struct pipe_inode_info *pipe2)
77 {
78         BUG_ON(pipe1 == pipe2);
79
80         if (pipe1 < pipe2) {
81                 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
82                 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
83         } else {
84                 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
85                 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
86         }
87 }
88
89 /* Drop the inode semaphore and wait for a pipe event, atomically */
90 void pipe_wait(struct pipe_inode_info *pipe)
91 {
92         DEFINE_WAIT(wait);
93
94         /*
95          * Pipes are system-local resources, so sleeping on them
96          * is considered a noninteractive wait:
97          */
98         prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
99         pipe_unlock(pipe);
100         schedule();
101         finish_wait(&pipe->wait, &wait);
102         pipe_lock(pipe);
103 }
104
105 static int
106 pipe_iov_copy_from_user(void *to, struct iovec *iov, unsigned long len,
107                         int atomic)
108 {
109         unsigned long copy;
110
111         while (len > 0) {
112                 while (!iov->iov_len)
113                         iov++;
114                 copy = min_t(unsigned long, len, iov->iov_len);
115
116                 if (atomic) {
117                         if (__copy_from_user_inatomic(to, iov->iov_base, copy))
118                                 return -EFAULT;
119                 } else {
120                         if (copy_from_user(to, iov->iov_base, copy))
121                                 return -EFAULT;
122                 }
123                 to += copy;
124                 len -= copy;
125                 iov->iov_base += copy;
126                 iov->iov_len -= copy;
127         }
128         return 0;
129 }
130
131 static int
132 pipe_iov_copy_to_user(struct iovec *iov, const void *from, unsigned long len,
133                       int atomic)
134 {
135         unsigned long copy;
136
137         while (len > 0) {
138                 while (!iov->iov_len)
139                         iov++;
140                 copy = min_t(unsigned long, len, iov->iov_len);
141
142                 if (atomic) {
143                         if (__copy_to_user_inatomic(iov->iov_base, from, copy))
144                                 return -EFAULT;
145                 } else {
146                         if (copy_to_user(iov->iov_base, from, copy))
147                                 return -EFAULT;
148                 }
149                 from += copy;
150                 len -= copy;
151                 iov->iov_base += copy;
152                 iov->iov_len -= copy;
153         }
154         return 0;
155 }
156
157 /*
158  * Attempt to pre-fault in the user memory, so we can use atomic copies.
159  * Returns the number of bytes not faulted in.
160  */
161 static int iov_fault_in_pages_write(struct iovec *iov, unsigned long len)
162 {
163         while (!iov->iov_len)
164                 iov++;
165
166         while (len > 0) {
167                 unsigned long this_len;
168
169                 this_len = min_t(unsigned long, len, iov->iov_len);
170                 if (fault_in_pages_writeable(iov->iov_base, this_len))
171                         break;
172
173                 len -= this_len;
174                 iov++;
175         }
176
177         return len;
178 }
179
180 /*
181  * Pre-fault in the user memory, so we can use atomic copies.
182  */
183 static void iov_fault_in_pages_read(struct iovec *iov, unsigned long len)
184 {
185         while (!iov->iov_len)
186                 iov++;
187
188         while (len > 0) {
189                 unsigned long this_len;
190
191                 this_len = min_t(unsigned long, len, iov->iov_len);
192                 fault_in_pages_readable(iov->iov_base, this_len);
193                 len -= this_len;
194                 iov++;
195         }
196 }
197
198 static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
199                                   struct pipe_buffer *buf)
200 {
201         struct page *page = buf->page;
202
203         /*
204          * If nobody else uses this page, and we don't already have a
205          * temporary page, let's keep track of it as a one-deep
206          * allocation cache. (Otherwise just release our reference to it)
207          */
208         if (page_count(page) == 1 && !pipe->tmp_page)
209                 pipe->tmp_page = page;
210         else
211                 page_cache_release(page);
212 }
213
214 /**
215  * generic_pipe_buf_map - virtually map a pipe buffer
216  * @pipe:       the pipe that the buffer belongs to
217  * @buf:        the buffer that should be mapped
218  * @atomic:     whether to use an atomic map
219  *
220  * Description:
221  *      This function returns a kernel virtual address mapping for the
222  *      pipe_buffer passed in @buf. If @atomic is set, an atomic map is provided
223  *      and the caller has to be careful not to fault before calling
224  *      the unmap function.
225  *
226  *      Note that this function occupies KM_USER0 if @atomic != 0.
227  */
228 void *generic_pipe_buf_map(struct pipe_inode_info *pipe,
229                            struct pipe_buffer *buf, int atomic)
230 {
231         if (atomic) {
232                 buf->flags |= PIPE_BUF_FLAG_ATOMIC;
233                 return kmap_atomic(buf->page, KM_USER0);
234         }
235
236         return kmap(buf->page);
237 }
238 EXPORT_SYMBOL(generic_pipe_buf_map);
239
240 /**
241  * generic_pipe_buf_unmap - unmap a previously mapped pipe buffer
242  * @pipe:       the pipe that the buffer belongs to
243  * @buf:        the buffer that should be unmapped
244  * @map_data:   the data that the mapping function returned
245  *
246  * Description:
247  *      This function undoes the mapping that ->map() provided.
248  */
249 void generic_pipe_buf_unmap(struct pipe_inode_info *pipe,
250                             struct pipe_buffer *buf, void *map_data)
251 {
252         if (buf->flags & PIPE_BUF_FLAG_ATOMIC) {
253                 buf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
254                 kunmap_atomic(map_data, KM_USER0);
255         } else
256                 kunmap(buf->page);
257 }
258 EXPORT_SYMBOL(generic_pipe_buf_unmap);
259
260 /**
261  * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
262  * @pipe:       the pipe that the buffer belongs to
263  * @buf:        the buffer to attempt to steal
264  *
265  * Description:
266  *      This function attempts to steal the &struct page attached to
267  *      @buf. If successful, this function returns 0 and returns with
268  *      the page locked. The caller may then reuse the page for whatever
269  *      he wishes; the typical use is insertion into a different file
270  *      page cache.
271  */
272 int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
273                            struct pipe_buffer *buf)
274 {
275         struct page *page = buf->page;
276
277         /*
278          * A reference of one is golden, that means that the owner of this
279          * page is the only one holding a reference to it. lock the page
280          * and return OK.
281          */
282         if (page_count(page) == 1) {
283                 lock_page(page);
284                 return 0;
285         }
286
287         return 1;
288 }
289 EXPORT_SYMBOL(generic_pipe_buf_steal);
290
291 /**
292  * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
293  * @pipe:       the pipe that the buffer belongs to
294  * @buf:        the buffer to get a reference to
295  *
296  * Description:
297  *      This function grabs an extra reference to @buf. It's used in
298  *      in the tee() system call, when we duplicate the buffers in one
299  *      pipe into another.
300  */
301 void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
302 {
303         page_cache_get(buf->page);
304 }
305 EXPORT_SYMBOL(generic_pipe_buf_get);
306
307 /**
308  * generic_pipe_buf_confirm - verify contents of the pipe buffer
309  * @info:       the pipe that the buffer belongs to
310  * @buf:        the buffer to confirm
311  *
312  * Description:
313  *      This function does nothing, because the generic pipe code uses
314  *      pages that are always good when inserted into the pipe.
315  */
316 int generic_pipe_buf_confirm(struct pipe_inode_info *info,
317                              struct pipe_buffer *buf)
318 {
319         return 0;
320 }
321 EXPORT_SYMBOL(generic_pipe_buf_confirm);
322
323 /**
324  * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
325  * @pipe:       the pipe that the buffer belongs to
326  * @buf:        the buffer to put a reference to
327  *
328  * Description:
329  *      This function releases a reference to @buf.
330  */
331 void generic_pipe_buf_release(struct pipe_inode_info *pipe,
332                               struct pipe_buffer *buf)
333 {
334         page_cache_release(buf->page);
335 }
336 EXPORT_SYMBOL(generic_pipe_buf_release);
337
338 static const struct pipe_buf_operations anon_pipe_buf_ops = {
339         .can_merge = 1,
340         .map = generic_pipe_buf_map,
341         .unmap = generic_pipe_buf_unmap,
342         .confirm = generic_pipe_buf_confirm,
343         .release = anon_pipe_buf_release,
344         .steal = generic_pipe_buf_steal,
345         .get = generic_pipe_buf_get,
346 };
347
348 static ssize_t
349 pipe_read(struct kiocb *iocb, const struct iovec *_iov,
350            unsigned long nr_segs, loff_t pos)
351 {
352         struct file *filp = iocb->ki_filp;
353         struct inode *inode = filp->f_path.dentry->d_inode;
354         struct pipe_inode_info *pipe;
355         int do_wakeup;
356         ssize_t ret;
357         struct iovec *iov = (struct iovec *)_iov;
358         size_t total_len;
359
360         total_len = iov_length(iov, nr_segs);
361         /* Null read succeeds. */
362         if (unlikely(total_len == 0))
363                 return 0;
364
365         do_wakeup = 0;
366         ret = 0;
367         mutex_lock(&inode->i_mutex);
368         pipe = inode->i_pipe;
369         for (;;) {
370                 int bufs = pipe->nrbufs;
371                 if (bufs) {
372                         int curbuf = pipe->curbuf;
373                         struct pipe_buffer *buf = pipe->bufs + curbuf;
374                         const struct pipe_buf_operations *ops = buf->ops;
375                         void *addr;
376                         size_t chars = buf->len;
377                         int error, atomic;
378
379                         if (chars > total_len)
380                                 chars = total_len;
381
382                         error = ops->confirm(pipe, buf);
383                         if (error) {
384                                 if (!ret)
385                                         ret = error;
386                                 break;
387                         }
388
389                         atomic = !iov_fault_in_pages_write(iov, chars);
390 redo:
391                         addr = ops->map(pipe, buf, atomic);
392                         error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars, atomic);
393                         ops->unmap(pipe, buf, addr);
394                         if (unlikely(error)) {
395                                 /*
396                                  * Just retry with the slow path if we failed.
397                                  */
398                                 if (atomic) {
399                                         atomic = 0;
400                                         goto redo;
401                                 }
402                                 if (!ret)
403                                         ret = error;
404                                 break;
405                         }
406                         ret += chars;
407                         buf->offset += chars;
408                         buf->len -= chars;
409                         if (!buf->len) {
410                                 buf->ops = NULL;
411                                 ops->release(pipe, buf);
412                                 curbuf = (curbuf + 1) & (pipe->buffers - 1);
413                                 pipe->curbuf = curbuf;
414                                 pipe->nrbufs = --bufs;
415                                 do_wakeup = 1;
416                         }
417                         total_len -= chars;
418                         if (!total_len)
419                                 break;  /* common path: read succeeded */
420                 }
421                 if (bufs)       /* More to do? */
422                         continue;
423                 if (!pipe->writers)
424                         break;
425                 if (!pipe->waiting_writers) {
426                         /* syscall merging: Usually we must not sleep
427                          * if O_NONBLOCK is set, or if we got some data.
428                          * But if a writer sleeps in kernel space, then
429                          * we can wait for that data without violating POSIX.
430                          */
431                         if (ret)
432                                 break;
433                         if (filp->f_flags & O_NONBLOCK) {
434                                 ret = -EAGAIN;
435                                 break;
436                         }
437                 }
438                 if (signal_pending(current)) {
439                         if (!ret)
440                                 ret = -ERESTARTSYS;
441                         break;
442                 }
443                 if (do_wakeup) {
444                         wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
445                         kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
446                 }
447                 pipe_wait(pipe);
448         }
449         mutex_unlock(&inode->i_mutex);
450
451         /* Signal writers asynchronously that there is more room. */
452         if (do_wakeup) {
453                 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
454                 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
455         }
456         if (ret > 0)
457                 file_accessed(filp);
458         return ret;
459 }
460
461 static ssize_t
462 pipe_write(struct kiocb *iocb, const struct iovec *_iov,
463             unsigned long nr_segs, loff_t ppos)
464 {
465         struct file *filp = iocb->ki_filp;
466         struct inode *inode = filp->f_path.dentry->d_inode;
467         struct pipe_inode_info *pipe;
468         ssize_t ret;
469         int do_wakeup;
470         struct iovec *iov = (struct iovec *)_iov;
471         size_t total_len;
472         ssize_t chars;
473
474         total_len = iov_length(iov, nr_segs);
475         /* Null write succeeds. */
476         if (unlikely(total_len == 0))
477                 return 0;
478
479         do_wakeup = 0;
480         ret = 0;
481         mutex_lock(&inode->i_mutex);
482         pipe = inode->i_pipe;
483
484         if (!pipe->readers) {
485                 send_sig(SIGPIPE, current, 0);
486                 ret = -EPIPE;
487                 goto out;
488         }
489
490         /* We try to merge small writes */
491         chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
492         if (pipe->nrbufs && chars != 0) {
493                 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
494                                                         (pipe->buffers - 1);
495                 struct pipe_buffer *buf = pipe->bufs + lastbuf;
496                 const struct pipe_buf_operations *ops = buf->ops;
497                 int offset = buf->offset + buf->len;
498
499                 if (ops->can_merge && offset + chars <= PAGE_SIZE) {
500                         int error, atomic = 1;
501                         void *addr;
502
503                         error = ops->confirm(pipe, buf);
504                         if (error)
505                                 goto out;
506
507                         iov_fault_in_pages_read(iov, chars);
508 redo1:
509                         addr = ops->map(pipe, buf, atomic);
510                         error = pipe_iov_copy_from_user(offset + addr, iov,
511                                                         chars, atomic);
512                         ops->unmap(pipe, buf, addr);
513                         ret = error;
514                         do_wakeup = 1;
515                         if (error) {
516                                 if (atomic) {
517                                         atomic = 0;
518                                         goto redo1;
519                                 }
520                                 goto out;
521                         }
522                         buf->len += chars;
523                         total_len -= chars;
524                         ret = chars;
525                         if (!total_len)
526                                 goto out;
527                 }
528         }
529
530         for (;;) {
531                 int bufs;
532
533                 if (!pipe->readers) {
534                         send_sig(SIGPIPE, current, 0);
535                         if (!ret)
536                                 ret = -EPIPE;
537                         break;
538                 }
539                 bufs = pipe->nrbufs;
540                 if (bufs < pipe->buffers) {
541                         int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
542                         struct pipe_buffer *buf = pipe->bufs + newbuf;
543                         struct page *page = pipe->tmp_page;
544                         char *src;
545                         int error, atomic = 1;
546
547                         if (!page) {
548                                 page = alloc_page(GFP_HIGHUSER);
549                                 if (unlikely(!page)) {
550                                         ret = ret ? : -ENOMEM;
551                                         break;
552                                 }
553                                 pipe->tmp_page = page;
554                         }
555                         /* Always wake up, even if the copy fails. Otherwise
556                          * we lock up (O_NONBLOCK-)readers that sleep due to
557                          * syscall merging.
558                          * FIXME! Is this really true?
559                          */
560                         do_wakeup = 1;
561                         chars = PAGE_SIZE;
562                         if (chars > total_len)
563                                 chars = total_len;
564
565                         iov_fault_in_pages_read(iov, chars);
566 redo2:
567                         if (atomic)
568                                 src = kmap_atomic(page, KM_USER0);
569                         else
570                                 src = kmap(page);
571
572                         error = pipe_iov_copy_from_user(src, iov, chars,
573                                                         atomic);
574                         if (atomic)
575                                 kunmap_atomic(src, KM_USER0);
576                         else
577                                 kunmap(page);
578
579                         if (unlikely(error)) {
580                                 if (atomic) {
581                                         atomic = 0;
582                                         goto redo2;
583                                 }
584                                 if (!ret)
585                                         ret = error;
586                                 break;
587                         }
588                         ret += chars;
589
590                         /* Insert it into the buffer array */
591                         buf->page = page;
592                         buf->ops = &anon_pipe_buf_ops;
593                         buf->offset = 0;
594                         buf->len = chars;
595                         pipe->nrbufs = ++bufs;
596                         pipe->tmp_page = NULL;
597
598                         total_len -= chars;
599                         if (!total_len)
600                                 break;
601                 }
602                 if (bufs < pipe->buffers)
603                         continue;
604                 if (filp->f_flags & O_NONBLOCK) {
605                         if (!ret)
606                                 ret = -EAGAIN;
607                         break;
608                 }
609                 if (signal_pending(current)) {
610                         if (!ret)
611                                 ret = -ERESTARTSYS;
612                         break;
613                 }
614                 if (do_wakeup) {
615                         wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
616                         kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
617                         do_wakeup = 0;
618                 }
619                 pipe->waiting_writers++;
620                 pipe_wait(pipe);
621                 pipe->waiting_writers--;
622         }
623 out:
624         mutex_unlock(&inode->i_mutex);
625         if (do_wakeup) {
626                 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
627                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
628         }
629         if (ret > 0)
630                 file_update_time(filp);
631         return ret;
632 }
633
634 static ssize_t
635 bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
636 {
637         return -EBADF;
638 }
639
640 static ssize_t
641 bad_pipe_w(struct file *filp, const char __user *buf, size_t count,
642            loff_t *ppos)
643 {
644         return -EBADF;
645 }
646
647 static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
648 {
649         struct inode *inode = filp->f_path.dentry->d_inode;
650         struct pipe_inode_info *pipe;
651         int count, buf, nrbufs;
652
653         switch (cmd) {
654                 case FIONREAD:
655                         mutex_lock(&inode->i_mutex);
656                         pipe = inode->i_pipe;
657                         count = 0;
658                         buf = pipe->curbuf;
659                         nrbufs = pipe->nrbufs;
660                         while (--nrbufs >= 0) {
661                                 count += pipe->bufs[buf].len;
662                                 buf = (buf+1) & (pipe->buffers - 1);
663                         }
664                         mutex_unlock(&inode->i_mutex);
665
666                         return put_user(count, (int __user *)arg);
667                 default:
668                         return -EINVAL;
669         }
670 }
671
672 /* No kernel lock held - fine */
673 static unsigned int
674 pipe_poll(struct file *filp, poll_table *wait)
675 {
676         unsigned int mask;
677         struct inode *inode = filp->f_path.dentry->d_inode;
678         struct pipe_inode_info *pipe = inode->i_pipe;
679         int nrbufs;
680
681         poll_wait(filp, &pipe->wait, wait);
682
683         /* Reading only -- no need for acquiring the semaphore.  */
684         nrbufs = pipe->nrbufs;
685         mask = 0;
686         if (filp->f_mode & FMODE_READ) {
687                 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
688                 if (!pipe->writers && filp->f_version != pipe->w_counter)
689                         mask |= POLLHUP;
690         }
691
692         if (filp->f_mode & FMODE_WRITE) {
693                 mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
694                 /*
695                  * Most Unices do not set POLLERR for FIFOs but on Linux they
696                  * behave exactly like pipes for poll().
697                  */
698                 if (!pipe->readers)
699                         mask |= POLLERR;
700         }
701
702         return mask;
703 }
704
705 static int
706 pipe_release(struct inode *inode, int decr, int decw)
707 {
708         struct pipe_inode_info *pipe;
709
710         mutex_lock(&inode->i_mutex);
711         pipe = inode->i_pipe;
712         pipe->readers -= decr;
713         pipe->writers -= decw;
714
715         if (!pipe->readers && !pipe->writers) {
716                 free_pipe_info(inode);
717         } else {
718                 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
719                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
720                 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
721         }
722         mutex_unlock(&inode->i_mutex);
723
724         return 0;
725 }
726
727 static int
728 pipe_read_fasync(int fd, struct file *filp, int on)
729 {
730         struct inode *inode = filp->f_path.dentry->d_inode;
731         int retval;
732
733         mutex_lock(&inode->i_mutex);
734         retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_readers);
735         mutex_unlock(&inode->i_mutex);
736
737         return retval;
738 }
739
740
741 static int
742 pipe_write_fasync(int fd, struct file *filp, int on)
743 {
744         struct inode *inode = filp->f_path.dentry->d_inode;
745         int retval;
746
747         mutex_lock(&inode->i_mutex);
748         retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_writers);
749         mutex_unlock(&inode->i_mutex);
750
751         return retval;
752 }
753
754
755 static int
756 pipe_rdwr_fasync(int fd, struct file *filp, int on)
757 {
758         struct inode *inode = filp->f_path.dentry->d_inode;
759         struct pipe_inode_info *pipe = inode->i_pipe;
760         int retval;
761
762         mutex_lock(&inode->i_mutex);
763         retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
764         if (retval >= 0) {
765                 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
766                 if (retval < 0) /* this can happen only if on == T */
767                         fasync_helper(-1, filp, 0, &pipe->fasync_readers);
768         }
769         mutex_unlock(&inode->i_mutex);
770         return retval;
771 }
772
773
774 static int
775 pipe_read_release(struct inode *inode, struct file *filp)
776 {
777         return pipe_release(inode, 1, 0);
778 }
779
780 static int
781 pipe_write_release(struct inode *inode, struct file *filp)
782 {
783         return pipe_release(inode, 0, 1);
784 }
785
786 static int
787 pipe_rdwr_release(struct inode *inode, struct file *filp)
788 {
789         int decr, decw;
790
791         decr = (filp->f_mode & FMODE_READ) != 0;
792         decw = (filp->f_mode & FMODE_WRITE) != 0;
793         return pipe_release(inode, decr, decw);
794 }
795
796 static int
797 pipe_read_open(struct inode *inode, struct file *filp)
798 {
799         int ret = -ENOENT;
800
801         mutex_lock(&inode->i_mutex);
802
803         if (inode->i_pipe) {
804                 ret = 0;
805                 inode->i_pipe->readers++;
806         }
807
808         mutex_unlock(&inode->i_mutex);
809
810         return ret;
811 }
812
813 static int
814 pipe_write_open(struct inode *inode, struct file *filp)
815 {
816         int ret = -ENOENT;
817
818         mutex_lock(&inode->i_mutex);
819
820         if (inode->i_pipe) {
821                 ret = 0;
822                 inode->i_pipe->writers++;
823         }
824
825         mutex_unlock(&inode->i_mutex);
826
827         return ret;
828 }
829
830 static int
831 pipe_rdwr_open(struct inode *inode, struct file *filp)
832 {
833         int ret = -ENOENT;
834
835         mutex_lock(&inode->i_mutex);
836
837         if (inode->i_pipe) {
838                 ret = 0;
839                 if (filp->f_mode & FMODE_READ)
840                         inode->i_pipe->readers++;
841                 if (filp->f_mode & FMODE_WRITE)
842                         inode->i_pipe->writers++;
843         }
844
845         mutex_unlock(&inode->i_mutex);
846
847         return ret;
848 }
849
850 /*
851  * The file_operations structs are not static because they
852  * are also used in linux/fs/fifo.c to do operations on FIFOs.
853  *
854  * Pipes reuse fifos' file_operations structs.
855  */
856 const struct file_operations read_pipefifo_fops = {
857         .llseek         = no_llseek,
858         .read           = do_sync_read,
859         .aio_read       = pipe_read,
860         .write          = bad_pipe_w,
861         .poll           = pipe_poll,
862         .unlocked_ioctl = pipe_ioctl,
863         .open           = pipe_read_open,
864         .release        = pipe_read_release,
865         .fasync         = pipe_read_fasync,
866 };
867
868 const struct file_operations write_pipefifo_fops = {
869         .llseek         = no_llseek,
870         .read           = bad_pipe_r,
871         .write          = do_sync_write,
872         .aio_write      = pipe_write,
873         .poll           = pipe_poll,
874         .unlocked_ioctl = pipe_ioctl,
875         .open           = pipe_write_open,
876         .release        = pipe_write_release,
877         .fasync         = pipe_write_fasync,
878 };
879
880 const struct file_operations rdwr_pipefifo_fops = {
881         .llseek         = no_llseek,
882         .read           = do_sync_read,
883         .aio_read       = pipe_read,
884         .write          = do_sync_write,
885         .aio_write      = pipe_write,
886         .poll           = pipe_poll,
887         .unlocked_ioctl = pipe_ioctl,
888         .open           = pipe_rdwr_open,
889         .release        = pipe_rdwr_release,
890         .fasync         = pipe_rdwr_fasync,
891 };
892
893 struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
894 {
895         struct pipe_inode_info *pipe;
896
897         pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
898         if (pipe) {
899                 pipe->bufs = kzalloc(sizeof(struct pipe_buffer) * PIPE_DEF_BUFFERS, GFP_KERNEL);
900                 if (pipe->bufs) {
901                         init_waitqueue_head(&pipe->wait);
902                         pipe->r_counter = pipe->w_counter = 1;
903                         pipe->inode = inode;
904                         pipe->buffers = PIPE_DEF_BUFFERS;
905                         return pipe;
906                 }
907                 kfree(pipe);
908         }
909
910         return NULL;
911 }
912
913 void __free_pipe_info(struct pipe_inode_info *pipe)
914 {
915         int i;
916
917         for (i = 0; i < pipe->buffers; i++) {
918                 struct pipe_buffer *buf = pipe->bufs + i;
919                 if (buf->ops)
920                         buf->ops->release(pipe, buf);
921         }
922         if (pipe->tmp_page)
923                 __free_page(pipe->tmp_page);
924         kfree(pipe->bufs);
925         kfree(pipe);
926 }
927
928 void free_pipe_info(struct inode *inode)
929 {
930         __free_pipe_info(inode->i_pipe);
931         inode->i_pipe = NULL;
932 }
933
934 static struct vfsmount *pipe_mnt __read_mostly;
935
936 /*
937  * pipefs_dname() is called from d_path().
938  */
939 static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
940 {
941         return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
942                                 dentry->d_inode->i_ino);
943 }
944
945 static const struct dentry_operations pipefs_dentry_operations = {
946         .d_dname        = pipefs_dname,
947 };
948
949 static struct inode * get_pipe_inode(void)
950 {
951         struct inode *inode = new_inode(pipe_mnt->mnt_sb);
952         struct pipe_inode_info *pipe;
953
954         if (!inode)
955                 goto fail_inode;
956
957         inode->i_ino = get_next_ino();
958
959         pipe = alloc_pipe_info(inode);
960         if (!pipe)
961                 goto fail_iput;
962         inode->i_pipe = pipe;
963
964         pipe->readers = pipe->writers = 1;
965         inode->i_fop = &rdwr_pipefifo_fops;
966
967         /*
968          * Mark the inode dirty from the very beginning,
969          * that way it will never be moved to the dirty
970          * list because "mark_inode_dirty()" will think
971          * that it already _is_ on the dirty list.
972          */
973         inode->i_state = I_DIRTY;
974         inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
975         inode->i_uid = current_fsuid();
976         inode->i_gid = current_fsgid();
977         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
978
979         return inode;
980
981 fail_iput:
982         iput(inode);
983
984 fail_inode:
985         return NULL;
986 }
987
988 struct file *create_write_pipe(int flags)
989 {
990         int err;
991         struct inode *inode;
992         struct file *f;
993         struct path path;
994         struct qstr name = { .name = "" };
995
996         err = -ENFILE;
997         inode = get_pipe_inode();
998         if (!inode)
999                 goto err;
1000
1001         err = -ENOMEM;
1002         path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name);
1003         if (!path.dentry)
1004                 goto err_inode;
1005         path.mnt = mntget(pipe_mnt);
1006
1007         d_instantiate(path.dentry, inode);
1008
1009         err = -ENFILE;
1010         f = alloc_file(&path, FMODE_WRITE, &write_pipefifo_fops);
1011         if (!f)
1012                 goto err_dentry;
1013         f->f_mapping = inode->i_mapping;
1014
1015         f->f_flags = O_WRONLY | (flags & O_NONBLOCK);
1016         f->f_version = 0;
1017
1018         return f;
1019
1020  err_dentry:
1021         free_pipe_info(inode);
1022         path_put(&path);
1023         return ERR_PTR(err);
1024
1025  err_inode:
1026         free_pipe_info(inode);
1027         iput(inode);
1028  err:
1029         return ERR_PTR(err);
1030 }
1031
1032 void free_write_pipe(struct file *f)
1033 {
1034         free_pipe_info(f->f_dentry->d_inode);
1035         path_put(&f->f_path);
1036         put_filp(f);
1037 }
1038
1039 struct file *create_read_pipe(struct file *wrf, int flags)
1040 {
1041         /* Grab pipe from the writer */
1042         struct file *f = alloc_file(&wrf->f_path, FMODE_READ,
1043                                     &read_pipefifo_fops);
1044         if (!f)
1045                 return ERR_PTR(-ENFILE);
1046
1047         path_get(&wrf->f_path);
1048         f->f_flags = O_RDONLY | (flags & O_NONBLOCK);
1049
1050         return f;
1051 }
1052
1053 int do_pipe_flags(int *fd, int flags)
1054 {
1055         struct file *fw, *fr;
1056         int error;
1057         int fdw, fdr;
1058
1059         if (flags & ~(O_CLOEXEC | O_NONBLOCK))
1060                 return -EINVAL;
1061
1062         fw = create_write_pipe(flags);
1063         if (IS_ERR(fw))
1064                 return PTR_ERR(fw);
1065         fr = create_read_pipe(fw, flags);
1066         error = PTR_ERR(fr);
1067         if (IS_ERR(fr))
1068                 goto err_write_pipe;
1069
1070         error = get_unused_fd_flags(flags);
1071         if (error < 0)
1072                 goto err_read_pipe;
1073         fdr = error;
1074
1075         error = get_unused_fd_flags(flags);
1076         if (error < 0)
1077                 goto err_fdr;
1078         fdw = error;
1079
1080         audit_fd_pair(fdr, fdw);
1081         fd_install(fdr, fr);
1082         fd_install(fdw, fw);
1083         fd[0] = fdr;
1084         fd[1] = fdw;
1085
1086         return 0;
1087
1088  err_fdr:
1089         put_unused_fd(fdr);
1090  err_read_pipe:
1091         path_put(&fr->f_path);
1092         put_filp(fr);
1093  err_write_pipe:
1094         free_write_pipe(fw);
1095         return error;
1096 }
1097
1098 /*
1099  * sys_pipe() is the normal C calling standard for creating
1100  * a pipe. It's not the way Unix traditionally does this, though.
1101  */
1102 SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
1103 {
1104         int fd[2];
1105         int error;
1106
1107         error = do_pipe_flags(fd, flags);
1108         if (!error) {
1109                 if (copy_to_user(fildes, fd, sizeof(fd))) {
1110                         sys_close(fd[0]);
1111                         sys_close(fd[1]);
1112                         error = -EFAULT;
1113                 }
1114         }
1115         return error;
1116 }
1117
1118 SYSCALL_DEFINE1(pipe, int __user *, fildes)
1119 {
1120         return sys_pipe2(fildes, 0);
1121 }
1122
1123 /*
1124  * Allocate a new array of pipe buffers and copy the info over. Returns the
1125  * pipe size if successful, or return -ERROR on error.
1126  */
1127 static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long nr_pages)
1128 {
1129         struct pipe_buffer *bufs;
1130
1131         /*
1132          * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1133          * expect a lot of shrink+grow operations, just free and allocate
1134          * again like we would do for growing. If the pipe currently
1135          * contains more buffers than arg, then return busy.
1136          */
1137         if (nr_pages < pipe->nrbufs)
1138                 return -EBUSY;
1139
1140         bufs = kcalloc(nr_pages, sizeof(struct pipe_buffer), GFP_KERNEL);
1141         if (unlikely(!bufs))
1142                 return -ENOMEM;
1143
1144         /*
1145          * The pipe array wraps around, so just start the new one at zero
1146          * and adjust the indexes.
1147          */
1148         if (pipe->nrbufs) {
1149                 unsigned int tail;
1150                 unsigned int head;
1151
1152                 tail = pipe->curbuf + pipe->nrbufs;
1153                 if (tail < pipe->buffers)
1154                         tail = 0;
1155                 else
1156                         tail &= (pipe->buffers - 1);
1157
1158                 head = pipe->nrbufs - tail;
1159                 if (head)
1160                         memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1161                 if (tail)
1162                         memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
1163         }
1164
1165         pipe->curbuf = 0;
1166         kfree(pipe->bufs);
1167         pipe->bufs = bufs;
1168         pipe->buffers = nr_pages;
1169         return nr_pages * PAGE_SIZE;
1170 }
1171
1172 /*
1173  * Currently we rely on the pipe array holding a power-of-2 number
1174  * of pages.
1175  */
1176 static inline unsigned int round_pipe_size(unsigned int size)
1177 {
1178         unsigned long nr_pages;
1179
1180         nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1181         return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
1182 }
1183
1184 /*
1185  * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1186  * will return an error.
1187  */
1188 int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
1189                  size_t *lenp, loff_t *ppos)
1190 {
1191         int ret;
1192
1193         ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
1194         if (ret < 0 || !write)
1195                 return ret;
1196
1197         pipe_max_size = round_pipe_size(pipe_max_size);
1198         return ret;
1199 }
1200
1201 /*
1202  * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1203  * location, so checking ->i_pipe is not enough to verify that this is a
1204  * pipe.
1205  */
1206 struct pipe_inode_info *get_pipe_info(struct file *file)
1207 {
1208         struct inode *i = file->f_path.dentry->d_inode;
1209
1210         return S_ISFIFO(i->i_mode) ? i->i_pipe : NULL;
1211 }
1212
1213 long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1214 {
1215         struct pipe_inode_info *pipe;
1216         long ret;
1217
1218         pipe = get_pipe_info(file);
1219         if (!pipe)
1220                 return -EBADF;
1221
1222         mutex_lock(&pipe->inode->i_mutex);
1223
1224         switch (cmd) {
1225         case F_SETPIPE_SZ: {
1226                 unsigned int size, nr_pages;
1227
1228                 size = round_pipe_size(arg);
1229                 nr_pages = size >> PAGE_SHIFT;
1230
1231                 ret = -EINVAL;
1232                 if (!nr_pages)
1233                         goto out;
1234
1235                 if (!capable(CAP_SYS_RESOURCE) && size > pipe_max_size) {
1236                         ret = -EPERM;
1237                         goto out;
1238                 }
1239                 ret = pipe_set_size(pipe, nr_pages);
1240                 break;
1241                 }
1242         case F_GETPIPE_SZ:
1243                 ret = pipe->buffers * PAGE_SIZE;
1244                 break;
1245         default:
1246                 ret = -EINVAL;
1247                 break;
1248         }
1249
1250 out:
1251         mutex_unlock(&pipe->inode->i_mutex);
1252         return ret;
1253 }
1254
1255 static const struct super_operations pipefs_ops = {
1256         .destroy_inode = free_inode_nonrcu,
1257 };
1258
1259 /*
1260  * pipefs should _never_ be mounted by userland - too much of security hassle,
1261  * no real gain from having the whole whorehouse mounted. So we don't need
1262  * any operations on the root directory. However, we need a non-trivial
1263  * d_name - pipe: will go nicely and kill the special-casing in procfs.
1264  */
1265 static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1266                          int flags, const char *dev_name, void *data)
1267 {
1268         return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1269                         &pipefs_dentry_operations, PIPEFS_MAGIC);
1270 }
1271
1272 static struct file_system_type pipe_fs_type = {
1273         .name           = "pipefs",
1274         .mount          = pipefs_mount,
1275         .kill_sb        = kill_anon_super,
1276 };
1277
1278 static int __init init_pipe_fs(void)
1279 {
1280         int err = register_filesystem(&pipe_fs_type);
1281
1282         if (!err) {
1283                 pipe_mnt = kern_mount(&pipe_fs_type);
1284                 if (IS_ERR(pipe_mnt)) {
1285                         err = PTR_ERR(pipe_mnt);
1286                         unregister_filesystem(&pipe_fs_type);
1287                 }
1288         }
1289         return err;
1290 }
1291
1292 static void __exit exit_pipe_fs(void)
1293 {
1294         unregister_filesystem(&pipe_fs_type);
1295         mntput(pipe_mnt);
1296 }
1297
1298 fs_initcall(init_pipe_fs);
1299 module_exit(exit_pipe_fs);