Lines Matching full:pipe
3 * linux/fs/pipe.c
36 * New pipe buffers will be restricted to this size while the user is exceeding
37 * their pipe buffer quota. The general pipe use case needs at least two
39 * than two, then a write to a non-empty pipe may block even if the pipe is not
42 * pipe before reading tokens: https://lore.kernel.org/lkml/1628086770.5rn8p04n6j.none@localhost/.
44 * Users can reduce their pipe buffers with F_SETPIPE_SZ below this at their
45 * own risk, namely: pipe writes to non-full pipes may block until the pipe is
51 * The max size that a non-root user is allowed to grow the pipe. Can
52 * be set by root in /proc/sys/fs/pipe-max-size
79 static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass) in pipe_lock_nested() argument
81 if (pipe->files) in pipe_lock_nested()
82 mutex_lock_nested(&pipe->mutex, subclass); in pipe_lock_nested()
85 void pipe_lock(struct pipe_inode_info *pipe) in pipe_lock() argument
88 * pipe_lock() nests non-pipe inode locks (for writing to a file) in pipe_lock()
90 pipe_lock_nested(pipe, I_MUTEX_PARENT); in pipe_lock()
94 void pipe_unlock(struct pipe_inode_info *pipe) in pipe_unlock() argument
96 if (pipe->files) in pipe_unlock()
97 mutex_unlock(&pipe->mutex); in pipe_unlock()
101 static inline void __pipe_lock(struct pipe_inode_info *pipe) in __pipe_lock() argument
103 mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT); in __pipe_lock()
106 static inline void __pipe_unlock(struct pipe_inode_info *pipe) in __pipe_unlock() argument
108 mutex_unlock(&pipe->mutex); in __pipe_unlock()
125 static void anon_pipe_buf_release(struct pipe_inode_info *pipe, in anon_pipe_buf_release() argument
135 if (page_count(page) == 1 && !pipe->tmp_page) in anon_pipe_buf_release()
136 pipe->tmp_page = page; in anon_pipe_buf_release()
141 static bool anon_pipe_buf_try_steal(struct pipe_inode_info *pipe, in anon_pipe_buf_try_steal() argument
155 * @pipe: the pipe that the buffer belongs to
165 bool generic_pipe_buf_try_steal(struct pipe_inode_info *pipe, in generic_pipe_buf_try_steal() argument
185 * @pipe: the pipe that the buffer belongs to
191 * pipe into another.
193 bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) in generic_pipe_buf_get() argument
201 * @pipe: the pipe that the buffer belongs to
207 void generic_pipe_buf_release(struct pipe_inode_info *pipe, in generic_pipe_buf_release() argument
220 /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */
221 static inline bool pipe_readable(const struct pipe_inode_info *pipe) in pipe_readable() argument
223 unsigned int head = READ_ONCE(pipe->head); in pipe_readable()
224 unsigned int tail = READ_ONCE(pipe->tail); in pipe_readable()
225 unsigned int writers = READ_ONCE(pipe->writers); in pipe_readable()
235 struct pipe_inode_info *pipe = filp->private_data; in pipe_read() local
244 __pipe_lock(pipe); in pipe_read()
247 * We only wake up writers if the pipe was full when we started in pipe_read()
254 was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); in pipe_read()
257 unsigned int head = smp_load_acquire(&pipe->head); in pipe_read()
258 unsigned int tail = pipe->tail; in pipe_read()
259 unsigned int mask = pipe->ring_size - 1; in pipe_read()
262 if (pipe->note_loss) { in pipe_read()
281 pipe->note_loss = false; in pipe_read()
286 struct pipe_buffer *buf = &pipe->bufs[tail & mask]; in pipe_read()
300 error = pipe_buf_confirm(pipe, buf); in pipe_read()
324 pipe_buf_release(pipe, buf); in pipe_read()
325 spin_lock_irq(&pipe->rd_wait.lock); in pipe_read()
328 pipe->note_loss = true; in pipe_read()
331 pipe->tail = tail; in pipe_read()
332 spin_unlock_irq(&pipe->rd_wait.lock); in pipe_read()
341 if (!pipe->writers) in pipe_read()
350 __pipe_unlock(pipe); in pipe_read()
356 * pipe buffer, and might have made space in the buffers in pipe_read()
359 * You can't make zero-sized pipe buffers by doing an empty in pipe_read()
366 * _very_ unlikely case that the pipe was full, but we got in pipe_read()
370 wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); in pipe_read()
371 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); in pipe_read()
379 if (wait_event_interruptible_exclusive(pipe->rd_wait, pipe_readable(pipe)) < 0) in pipe_read()
382 __pipe_lock(pipe); in pipe_read()
383 was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); in pipe_read()
386 if (pipe_empty(pipe->head, pipe->tail)) in pipe_read()
388 __pipe_unlock(pipe); in pipe_read()
391 wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); in pipe_read()
393 wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); in pipe_read()
394 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); in pipe_read()
405 /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */
406 static inline bool pipe_writable(const struct pipe_inode_info *pipe) in pipe_writable() argument
408 unsigned int head = READ_ONCE(pipe->head); in pipe_writable()
409 unsigned int tail = READ_ONCE(pipe->tail); in pipe_writable()
410 unsigned int max_usage = READ_ONCE(pipe->max_usage); in pipe_writable()
413 !READ_ONCE(pipe->readers); in pipe_writable()
420 struct pipe_inode_info *pipe = filp->private_data; in pipe_write() local
430 * the pipe. in pipe_write()
432 * pipe locked. in pipe_write()
433 * If we had to support locking a normal pipe and a notification pipe at in pipe_write()
437 if (pipe_has_watch_queue(pipe)) in pipe_write()
444 __pipe_lock(pipe); in pipe_write()
446 if (!pipe->readers) { in pipe_write()
460 head = pipe->head; in pipe_write()
461 was_empty = pipe_empty(head, pipe->tail); in pipe_write()
464 unsigned int mask = pipe->ring_size - 1; in pipe_write()
465 struct pipe_buffer *buf = &pipe->bufs[(head - 1) & mask]; in pipe_write()
470 ret = pipe_buf_confirm(pipe, buf); in pipe_write()
487 if (!pipe->readers) { in pipe_write()
494 head = pipe->head; in pipe_write()
495 if (!pipe_full(head, pipe->tail, pipe->max_usage)) { in pipe_write()
496 unsigned int mask = pipe->ring_size - 1; in pipe_write()
498 struct page *page = pipe->tmp_page; in pipe_write()
507 pipe->tmp_page = page; in pipe_write()
515 spin_lock_irq(&pipe->rd_wait.lock); in pipe_write()
517 head = pipe->head; in pipe_write()
518 if (pipe_full(head, pipe->tail, pipe->max_usage)) { in pipe_write()
519 spin_unlock_irq(&pipe->rd_wait.lock); in pipe_write()
523 pipe->head = head + 1; in pipe_write()
524 spin_unlock_irq(&pipe->rd_wait.lock); in pipe_write()
527 buf = &pipe->bufs[head & mask]; in pipe_write()
536 pipe->tmp_page = NULL; in pipe_write()
551 if (!pipe_full(head, pipe->tail, pipe->max_usage)) in pipe_write()
568 * We're going to release the pipe lock and wait for more in pipe_write()
570 * after waiting we need to re-check whether the pipe in pipe_write()
573 __pipe_unlock(pipe); in pipe_write()
575 wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); in pipe_write()
576 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); in pipe_write()
577 wait_event_interruptible_exclusive(pipe->wr_wait, pipe_writable(pipe)); in pipe_write()
578 __pipe_lock(pipe); in pipe_write()
579 was_empty = pipe_empty(pipe->head, pipe->tail); in pipe_write()
583 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) in pipe_write()
585 __pipe_unlock(pipe); in pipe_write()
596 * Epoll nonsensically wants a wakeup whether the pipe in pipe_write()
599 if (was_empty || pipe->poll_usage) in pipe_write()
600 wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); in pipe_write()
601 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); in pipe_write()
603 wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); in pipe_write()
615 struct pipe_inode_info *pipe = filp->private_data; in pipe_ioctl() local
620 __pipe_lock(pipe); in pipe_ioctl()
622 head = pipe->head; in pipe_ioctl()
623 tail = pipe->tail; in pipe_ioctl()
624 mask = pipe->ring_size - 1; in pipe_ioctl()
627 count += pipe->bufs[tail & mask].len; in pipe_ioctl()
630 __pipe_unlock(pipe); in pipe_ioctl()
637 __pipe_lock(pipe); in pipe_ioctl()
638 ret = watch_queue_set_size(pipe, arg); in pipe_ioctl()
639 __pipe_unlock(pipe); in pipe_ioctl()
645 pipe, (struct watch_notification_filter __user *)arg); in pipe_ioctl()
658 struct pipe_inode_info *pipe = filp->private_data; in pipe_poll() local
662 WRITE_ONCE(pipe->poll_usage, true); in pipe_poll()
665 * Reading pipe state only -- no need for acquiring the semaphore. in pipe_poll()
671 poll_wait(filp, &pipe->rd_wait, wait); in pipe_poll()
673 poll_wait(filp, &pipe->wr_wait, wait); in pipe_poll()
680 head = READ_ONCE(pipe->head); in pipe_poll()
681 tail = READ_ONCE(pipe->tail); in pipe_poll()
687 if (!pipe->writers && filp->f_version != pipe->w_counter) in pipe_poll()
692 if (!pipe_full(head, tail, pipe->max_usage)) in pipe_poll()
698 if (!pipe->readers) in pipe_poll()
705 static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe) in put_pipe_info() argument
710 if (!--pipe->files) { in put_pipe_info()
717 free_pipe_info(pipe); in put_pipe_info()
723 struct pipe_inode_info *pipe = file->private_data; in pipe_release() local
725 __pipe_lock(pipe); in pipe_release()
727 pipe->readers--; in pipe_release()
729 pipe->writers--; in pipe_release()
732 if (!pipe->readers != !pipe->writers) { in pipe_release()
733 wake_up_interruptible_all(&pipe->rd_wait); in pipe_release()
734 wake_up_interruptible_all(&pipe->wr_wait); in pipe_release()
735 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); in pipe_release()
736 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); in pipe_release()
738 __pipe_unlock(pipe); in pipe_release()
740 put_pipe_info(inode, pipe); in pipe_release()
747 struct pipe_inode_info *pipe = filp->private_data; in pipe_fasync() local
750 __pipe_lock(pipe); in pipe_fasync()
752 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); in pipe_fasync()
754 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); in pipe_fasync()
757 fasync_helper(-1, filp, 0, &pipe->fasync_readers); in pipe_fasync()
759 __pipe_unlock(pipe); in pipe_fasync()
790 struct pipe_inode_info *pipe; in alloc_pipe_info() local
796 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT); in alloc_pipe_info()
797 if (pipe == NULL) in alloc_pipe_info()
813 pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer), in alloc_pipe_info()
816 if (pipe->bufs) { in alloc_pipe_info()
817 init_waitqueue_head(&pipe->rd_wait); in alloc_pipe_info()
818 init_waitqueue_head(&pipe->wr_wait); in alloc_pipe_info()
819 pipe->r_counter = pipe->w_counter = 1; in alloc_pipe_info()
820 pipe->max_usage = pipe_bufs; in alloc_pipe_info()
821 pipe->ring_size = pipe_bufs; in alloc_pipe_info()
822 pipe->nr_accounted = pipe_bufs; in alloc_pipe_info()
823 pipe->user = user; in alloc_pipe_info()
824 mutex_init(&pipe->mutex); in alloc_pipe_info()
825 return pipe; in alloc_pipe_info()
830 kfree(pipe); in alloc_pipe_info()
836 void free_pipe_info(struct pipe_inode_info *pipe) in free_pipe_info() argument
841 if (pipe->watch_queue) in free_pipe_info()
842 watch_queue_clear(pipe->watch_queue); in free_pipe_info()
845 (void) account_pipe_buffers(pipe->user, pipe->nr_accounted, 0); in free_pipe_info()
846 free_uid(pipe->user); in free_pipe_info()
847 for (i = 0; i < pipe->ring_size; i++) { in free_pipe_info()
848 struct pipe_buffer *buf = pipe->bufs + i; in free_pipe_info()
850 pipe_buf_release(pipe, buf); in free_pipe_info()
853 if (pipe->watch_queue) in free_pipe_info()
854 put_watch_queue(pipe->watch_queue); in free_pipe_info()
856 if (pipe->tmp_page) in free_pipe_info()
857 __free_page(pipe->tmp_page); in free_pipe_info()
858 kfree(pipe->bufs); in free_pipe_info()
859 kfree(pipe); in free_pipe_info()
869 return dynamic_dname(buffer, buflen, "pipe:[%lu]", in pipefs_dname()
880 struct pipe_inode_info *pipe; in get_pipe_inode() local
887 pipe = alloc_pipe_info(); in get_pipe_inode()
888 if (!pipe) in get_pipe_inode()
891 inode->i_pipe = pipe; in get_pipe_inode()
892 pipe->files = 2; in get_pipe_inode()
893 pipe->readers = pipe->writers = 1; in get_pipe_inode()
985 /* pipe groks IOCB_NOWAIT */ in __do_pipe_flags()
1011 * a pipe. It's not the way Unix traditionally does this, though.
1040 SYSCALL_DEFINE1(pipe, int __user *, fildes) in SYSCALL_DEFINE1() argument
1046 * This is the stupid "wait for pipe to be readable or writable"
1054 void pipe_wait_readable(struct pipe_inode_info *pipe) in pipe_wait_readable() argument
1056 pipe_unlock(pipe); in pipe_wait_readable()
1057 wait_event_interruptible(pipe->rd_wait, pipe_readable(pipe)); in pipe_wait_readable()
1058 pipe_lock(pipe); in pipe_wait_readable()
1061 void pipe_wait_writable(struct pipe_inode_info *pipe) in pipe_wait_writable() argument
1063 pipe_unlock(pipe); in pipe_wait_writable()
1064 wait_event_interruptible(pipe->wr_wait, pipe_writable(pipe)); in pipe_wait_writable()
1065 pipe_lock(pipe); in pipe_wait_writable()
1070 * holding the pipe lock, so "*cnt" is stable and we know a wakeup cannot
1075 * because of the pipe lock, we can check the condition before being on
1078 * We use the 'rd_wait' waitqueue for pipe partner waiting.
1080 static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt) in wait_for_partner() argument
1086 prepare_to_wait(&pipe->rd_wait, &rdwait, TASK_INTERRUPTIBLE); in wait_for_partner()
1087 pipe_unlock(pipe); in wait_for_partner()
1089 finish_wait(&pipe->rd_wait, &rdwait); in wait_for_partner()
1090 pipe_lock(pipe); in wait_for_partner()
1097 static void wake_up_partner(struct pipe_inode_info *pipe) in wake_up_partner() argument
1099 wake_up_interruptible_all(&pipe->rd_wait); in wake_up_partner()
1104 struct pipe_inode_info *pipe; in fifo_open() local
1112 pipe = inode->i_pipe; in fifo_open()
1113 pipe->files++; in fifo_open()
1117 pipe = alloc_pipe_info(); in fifo_open()
1118 if (!pipe) in fifo_open()
1120 pipe->files = 1; in fifo_open()
1125 free_pipe_info(pipe); in fifo_open()
1126 pipe = inode->i_pipe; in fifo_open()
1128 inode->i_pipe = pipe; in fifo_open()
1132 filp->private_data = pipe; in fifo_open()
1133 /* OK, we have a pipe and it's pinned down */ in fifo_open()
1135 __pipe_lock(pipe); in fifo_open()
1147 pipe->r_counter++; in fifo_open()
1148 if (pipe->readers++ == 0) in fifo_open()
1149 wake_up_partner(pipe); in fifo_open()
1151 if (!is_pipe && !pipe->writers) { in fifo_open()
1155 filp->f_version = pipe->w_counter; in fifo_open()
1157 if (wait_for_partner(pipe, &pipe->w_counter)) in fifo_open()
1170 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers) in fifo_open()
1173 pipe->w_counter++; in fifo_open()
1174 if (!pipe->writers++) in fifo_open()
1175 wake_up_partner(pipe); in fifo_open()
1177 if (!is_pipe && !pipe->readers) { in fifo_open()
1178 if (wait_for_partner(pipe, &pipe->r_counter)) in fifo_open()
1191 pipe->readers++; in fifo_open()
1192 pipe->writers++; in fifo_open()
1193 pipe->r_counter++; in fifo_open()
1194 pipe->w_counter++; in fifo_open()
1195 if (pipe->readers == 1 || pipe->writers == 1) in fifo_open()
1196 wake_up_partner(pipe); in fifo_open()
1205 __pipe_unlock(pipe); in fifo_open()
1209 if (!--pipe->readers) in fifo_open()
1210 wake_up_interruptible(&pipe->wr_wait); in fifo_open()
1215 if (!--pipe->writers) in fifo_open()
1216 wake_up_interruptible_all(&pipe->rd_wait); in fifo_open()
1221 __pipe_unlock(pipe); in fifo_open()
1223 put_pipe_info(inode, pipe); in fifo_open()
1240 * Currently we rely on the pipe array holding a power-of-2 number
1248 /* Minimum pipe size, as required by POSIX */ in round_pipe_size()
1256 * Resize the pipe ring to a number of slots.
1258 * Note the pipe can be reduced in capacity, but only if the current
1262 int pipe_resize_ring(struct pipe_inode_info *pipe, unsigned int nr_slots) in pipe_resize_ring() argument
1272 spin_lock_irq(&pipe->rd_wait.lock); in pipe_resize_ring()
1273 mask = pipe->ring_size - 1; in pipe_resize_ring()
1274 head = pipe->head; in pipe_resize_ring()
1275 tail = pipe->tail; in pipe_resize_ring()
1279 spin_unlock_irq(&pipe->rd_wait.lock); in pipe_resize_ring()
1285 * The pipe array wraps around, so just start the new one at zero in pipe_resize_ring()
1292 memcpy(bufs, pipe->bufs + t, in pipe_resize_ring()
1295 unsigned int tsize = pipe->ring_size - t; in pipe_resize_ring()
1297 memcpy(bufs + tsize, pipe->bufs, in pipe_resize_ring()
1299 memcpy(bufs, pipe->bufs + t, in pipe_resize_ring()
1307 kfree(pipe->bufs); in pipe_resize_ring()
1308 pipe->bufs = bufs; in pipe_resize_ring()
1309 pipe->ring_size = nr_slots; in pipe_resize_ring()
1310 if (pipe->max_usage > nr_slots) in pipe_resize_ring()
1311 pipe->max_usage = nr_slots; in pipe_resize_ring()
1312 pipe->tail = tail; in pipe_resize_ring()
1313 pipe->head = head; in pipe_resize_ring()
1315 if (!pipe_has_watch_queue(pipe)) { in pipe_resize_ring()
1316 pipe->max_usage = nr_slots; in pipe_resize_ring()
1317 pipe->nr_accounted = nr_slots; in pipe_resize_ring()
1320 spin_unlock_irq(&pipe->rd_wait.lock); in pipe_resize_ring()
1323 wake_up_interruptible(&pipe->wr_wait); in pipe_resize_ring()
1328 * Allocate a new array of pipe buffers and copy the info over. Returns the
1329 * pipe size if successful, or return -ERROR on error.
1331 static long pipe_set_size(struct pipe_inode_info *pipe, unsigned int arg) in pipe_set_size() argument
1337 if (pipe_has_watch_queue(pipe)) in pipe_set_size()
1347 * If trying to increase the pipe capacity, check that an in pipe_set_size()
1350 * Decreasing the pipe capacity is always permitted, even in pipe_set_size()
1353 if (nr_slots > pipe->max_usage && in pipe_set_size()
1357 user_bufs = account_pipe_buffers(pipe->user, pipe->nr_accounted, nr_slots); in pipe_set_size()
1359 if (nr_slots > pipe->max_usage && in pipe_set_size()
1367 ret = pipe_resize_ring(pipe, nr_slots); in pipe_set_size()
1371 return pipe->max_usage * PAGE_SIZE; in pipe_set_size()
1374 (void) account_pipe_buffers(pipe->user, nr_slots, pipe->nr_accounted); in pipe_set_size()
1380 * not enough to verify that this is a pipe.
1384 struct pipe_inode_info *pipe = file->private_data; in get_pipe_info() local
1386 if (file->f_op != &pipefifo_fops || !pipe) in get_pipe_info()
1388 if (for_splice && pipe_has_watch_queue(pipe)) in get_pipe_info()
1390 return pipe; in get_pipe_info()
1395 struct pipe_inode_info *pipe; in pipe_fcntl() local
1398 pipe = get_pipe_info(file, false); in pipe_fcntl()
1399 if (!pipe) in pipe_fcntl()
1402 __pipe_lock(pipe); in pipe_fcntl()
1406 ret = pipe_set_size(pipe, arg); in pipe_fcntl()
1409 ret = pipe->max_usage * PAGE_SIZE; in pipe_fcntl()
1416 __pipe_unlock(pipe); in pipe_fcntl()
1429 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1478 .procname = "pipe-max-size",
1485 .procname = "pipe-user-pages-hard",
1492 .procname = "pipe-user-pages-soft",