Lines Matching full:pipe
3 * linux/fs/pipe.c
35 * New pipe buffers will be restricted to this size while the user is exceeding
36 * their pipe buffer quota. The general pipe use case needs at least two
38 * than two, then a write to a non-empty pipe may block even if the pipe is not
41 * pipe before reading tokens: https://lore.kernel.org/lkml/1628086770.5rn8p04n6j.none@localhost/.
43 * Users can reduce their pipe buffers with F_SETPIPE_SZ below this at their
44 * own risk, namely: pipe writes to non-full pipes may block until the pipe is
50 * The max size that a non-root user is allowed to grow the pipe. Can
51 * be set by root in /proc/sys/fs/pipe-max-size
78 static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass) in pipe_lock_nested() argument
80 if (pipe->files) in pipe_lock_nested()
81 mutex_lock_nested(&pipe->mutex, subclass); in pipe_lock_nested()
84 void pipe_lock(struct pipe_inode_info *pipe) in pipe_lock() argument
87 * pipe_lock() nests non-pipe inode locks (for writing to a file) in pipe_lock()
89 pipe_lock_nested(pipe, I_MUTEX_PARENT); in pipe_lock()
93 void pipe_unlock(struct pipe_inode_info *pipe) in pipe_unlock() argument
95 if (pipe->files) in pipe_unlock()
96 mutex_unlock(&pipe->mutex); in pipe_unlock()
100 static inline void __pipe_lock(struct pipe_inode_info *pipe) in __pipe_lock() argument
102 mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT); in __pipe_lock()
105 static inline void __pipe_unlock(struct pipe_inode_info *pipe) in __pipe_unlock() argument
107 mutex_unlock(&pipe->mutex); in __pipe_unlock()
124 static void anon_pipe_buf_release(struct pipe_inode_info *pipe, in anon_pipe_buf_release() argument
134 if (page_count(page) == 1 && !pipe->tmp_page) in anon_pipe_buf_release()
135 pipe->tmp_page = page; in anon_pipe_buf_release()
140 static bool anon_pipe_buf_try_steal(struct pipe_inode_info *pipe, in anon_pipe_buf_try_steal() argument
154 * @pipe: the pipe that the buffer belongs to
164 bool generic_pipe_buf_try_steal(struct pipe_inode_info *pipe, in generic_pipe_buf_try_steal() argument
184 * @pipe: the pipe that the buffer belongs to
190 * pipe into another.
192 bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) in generic_pipe_buf_get() argument
200 * @pipe: the pipe that the buffer belongs to
206 void generic_pipe_buf_release(struct pipe_inode_info *pipe, in generic_pipe_buf_release() argument
219 /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */
220 static inline bool pipe_readable(const struct pipe_inode_info *pipe) in pipe_readable() argument
222 unsigned int head = READ_ONCE(pipe->head); in pipe_readable()
223 unsigned int tail = READ_ONCE(pipe->tail); in pipe_readable()
224 unsigned int writers = READ_ONCE(pipe->writers); in pipe_readable()
234 struct pipe_inode_info *pipe = filp->private_data; in pipe_read() local
243 __pipe_lock(pipe); in pipe_read()
246 * We only wake up writers if the pipe was full when we started in pipe_read()
253 was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); in pipe_read()
256 unsigned int head = smp_load_acquire(&pipe->head); in pipe_read()
257 unsigned int tail = pipe->tail; in pipe_read()
258 unsigned int mask = pipe->ring_size - 1; in pipe_read()
261 if (pipe->note_loss) { in pipe_read()
280 pipe->note_loss = false; in pipe_read()
285 struct pipe_buffer *buf = &pipe->bufs[tail & mask]; in pipe_read()
299 error = pipe_buf_confirm(pipe, buf); in pipe_read()
323 pipe_buf_release(pipe, buf); in pipe_read()
324 spin_lock_irq(&pipe->rd_wait.lock); in pipe_read()
327 pipe->note_loss = true; in pipe_read()
330 pipe->tail = tail; in pipe_read()
331 spin_unlock_irq(&pipe->rd_wait.lock); in pipe_read()
340 if (!pipe->writers) in pipe_read()
348 __pipe_unlock(pipe); in pipe_read()
354 * pipe buffer, and might have made space in the buffers in pipe_read()
357 * You can't make zero-sized pipe buffers by doing an empty in pipe_read()
364 * _very_ unlikely case that the pipe was full, but we got in pipe_read()
368 wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); in pipe_read()
369 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); in pipe_read()
377 if (wait_event_interruptible_exclusive(pipe->rd_wait, pipe_readable(pipe)) < 0) in pipe_read()
380 __pipe_lock(pipe); in pipe_read()
381 was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); in pipe_read()
384 if (pipe_empty(pipe->head, pipe->tail)) in pipe_read()
386 __pipe_unlock(pipe); in pipe_read()
389 wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); in pipe_read()
391 wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); in pipe_read()
392 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); in pipe_read()
403 /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */
404 static inline bool pipe_writable(const struct pipe_inode_info *pipe) in pipe_writable() argument
406 unsigned int head = READ_ONCE(pipe->head); in pipe_writable()
407 unsigned int tail = READ_ONCE(pipe->tail); in pipe_writable()
408 unsigned int max_usage = READ_ONCE(pipe->max_usage); in pipe_writable()
411 !READ_ONCE(pipe->readers); in pipe_writable()
418 struct pipe_inode_info *pipe = filp->private_data; in pipe_write() local
430 __pipe_lock(pipe); in pipe_write()
432 if (!pipe->readers) { in pipe_write()
438 if (pipe_has_watch_queue(pipe)) { in pipe_write()
451 head = pipe->head; in pipe_write()
452 was_empty = pipe_empty(head, pipe->tail); in pipe_write()
455 unsigned int mask = pipe->ring_size - 1; in pipe_write()
456 struct pipe_buffer *buf = &pipe->bufs[(head - 1) & mask]; in pipe_write()
461 ret = pipe_buf_confirm(pipe, buf); in pipe_write()
478 if (!pipe->readers) { in pipe_write()
485 head = pipe->head; in pipe_write()
486 if (!pipe_full(head, pipe->tail, pipe->max_usage)) { in pipe_write()
487 unsigned int mask = pipe->ring_size - 1; in pipe_write()
488 struct pipe_buffer *buf = &pipe->bufs[head & mask]; in pipe_write()
489 struct page *page = pipe->tmp_page; in pipe_write()
498 pipe->tmp_page = page; in pipe_write()
506 spin_lock_irq(&pipe->rd_wait.lock); in pipe_write()
508 head = pipe->head; in pipe_write()
509 if (pipe_full(head, pipe->tail, pipe->max_usage)) { in pipe_write()
510 spin_unlock_irq(&pipe->rd_wait.lock); in pipe_write()
514 pipe->head = head + 1; in pipe_write()
515 spin_unlock_irq(&pipe->rd_wait.lock); in pipe_write()
518 buf = &pipe->bufs[head & mask]; in pipe_write()
527 pipe->tmp_page = NULL; in pipe_write()
543 if (!pipe_full(head, pipe->tail, pipe->max_usage)) in pipe_write()
559 * We're going to release the pipe lock and wait for more in pipe_write()
561 * after waiting we need to re-check whether the pipe in pipe_write()
564 __pipe_unlock(pipe); in pipe_write()
566 wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); in pipe_write()
567 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); in pipe_write()
568 wait_event_interruptible_exclusive(pipe->wr_wait, pipe_writable(pipe)); in pipe_write()
569 __pipe_lock(pipe); in pipe_write()
570 was_empty = pipe_empty(pipe->head, pipe->tail); in pipe_write()
574 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) in pipe_write()
576 __pipe_unlock(pipe); in pipe_write()
587 * Epoll nonsensically wants a wakeup whether the pipe in pipe_write()
590 if (was_empty || pipe->poll_usage) in pipe_write()
591 wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); in pipe_write()
592 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); in pipe_write()
594 wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); in pipe_write()
606 struct pipe_inode_info *pipe = filp->private_data; in pipe_ioctl() local
611 __pipe_lock(pipe); in pipe_ioctl()
613 head = pipe->head; in pipe_ioctl()
614 tail = pipe->tail; in pipe_ioctl()
615 mask = pipe->ring_size - 1; in pipe_ioctl()
618 count += pipe->bufs[tail & mask].len; in pipe_ioctl()
621 __pipe_unlock(pipe); in pipe_ioctl()
628 __pipe_lock(pipe); in pipe_ioctl()
629 ret = watch_queue_set_size(pipe, arg); in pipe_ioctl()
630 __pipe_unlock(pipe); in pipe_ioctl()
636 pipe, (struct watch_notification_filter __user *)arg); in pipe_ioctl()
649 struct pipe_inode_info *pipe = filp->private_data; in pipe_poll() local
653 WRITE_ONCE(pipe->poll_usage, true); in pipe_poll()
656 * Reading pipe state only -- no need for acquiring the semaphore. in pipe_poll()
662 poll_wait(filp, &pipe->rd_wait, wait); in pipe_poll()
664 poll_wait(filp, &pipe->wr_wait, wait); in pipe_poll()
671 head = READ_ONCE(pipe->head); in pipe_poll()
672 tail = READ_ONCE(pipe->tail); in pipe_poll()
678 if (!pipe->writers && filp->f_version != pipe->w_counter) in pipe_poll()
683 if (!pipe_full(head, tail, pipe->max_usage)) in pipe_poll()
689 if (!pipe->readers) in pipe_poll()
696 static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe) in put_pipe_info() argument
701 if (!--pipe->files) { in put_pipe_info()
708 free_pipe_info(pipe); in put_pipe_info()
714 struct pipe_inode_info *pipe = file->private_data; in pipe_release() local
716 __pipe_lock(pipe); in pipe_release()
718 pipe->readers--; in pipe_release()
720 pipe->writers--; in pipe_release()
723 if (!pipe->readers != !pipe->writers) { in pipe_release()
724 wake_up_interruptible_all(&pipe->rd_wait); in pipe_release()
725 wake_up_interruptible_all(&pipe->wr_wait); in pipe_release()
726 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); in pipe_release()
727 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); in pipe_release()
729 __pipe_unlock(pipe); in pipe_release()
731 put_pipe_info(inode, pipe); in pipe_release()
738 struct pipe_inode_info *pipe = filp->private_data; in pipe_fasync() local
741 __pipe_lock(pipe); in pipe_fasync()
743 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); in pipe_fasync()
745 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); in pipe_fasync()
748 fasync_helper(-1, filp, 0, &pipe->fasync_readers); in pipe_fasync()
750 __pipe_unlock(pipe); in pipe_fasync()
781 struct pipe_inode_info *pipe; in alloc_pipe_info() local
787 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT); in alloc_pipe_info()
788 if (pipe == NULL) in alloc_pipe_info()
804 pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer), in alloc_pipe_info()
807 if (pipe->bufs) { in alloc_pipe_info()
808 init_waitqueue_head(&pipe->rd_wait); in alloc_pipe_info()
809 init_waitqueue_head(&pipe->wr_wait); in alloc_pipe_info()
810 pipe->r_counter = pipe->w_counter = 1; in alloc_pipe_info()
811 pipe->max_usage = pipe_bufs; in alloc_pipe_info()
812 pipe->ring_size = pipe_bufs; in alloc_pipe_info()
813 pipe->nr_accounted = pipe_bufs; in alloc_pipe_info()
814 pipe->user = user; in alloc_pipe_info()
815 mutex_init(&pipe->mutex); in alloc_pipe_info()
816 return pipe; in alloc_pipe_info()
821 kfree(pipe); in alloc_pipe_info()
827 void free_pipe_info(struct pipe_inode_info *pipe) in free_pipe_info() argument
832 if (pipe->watch_queue) in free_pipe_info()
833 watch_queue_clear(pipe->watch_queue); in free_pipe_info()
836 (void) account_pipe_buffers(pipe->user, pipe->nr_accounted, 0); in free_pipe_info()
837 free_uid(pipe->user); in free_pipe_info()
838 for (i = 0; i < pipe->ring_size; i++) { in free_pipe_info()
839 struct pipe_buffer *buf = pipe->bufs + i; in free_pipe_info()
841 pipe_buf_release(pipe, buf); in free_pipe_info()
844 if (pipe->watch_queue) in free_pipe_info()
845 put_watch_queue(pipe->watch_queue); in free_pipe_info()
847 if (pipe->tmp_page) in free_pipe_info()
848 __free_page(pipe->tmp_page); in free_pipe_info()
849 kfree(pipe->bufs); in free_pipe_info()
850 kfree(pipe); in free_pipe_info()
860 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]", in pipefs_dname()
871 struct pipe_inode_info *pipe; in get_pipe_inode() local
878 pipe = alloc_pipe_info(); in get_pipe_inode()
879 if (!pipe) in get_pipe_inode()
882 inode->i_pipe = pipe; in get_pipe_inode()
883 pipe->files = 2; in get_pipe_inode()
884 pipe->readers = pipe->writers = 1; in get_pipe_inode()
999 * a pipe. It's not the way Unix traditionally does this, though.
1028 SYSCALL_DEFINE1(pipe, int __user *, fildes) in SYSCALL_DEFINE1() argument
1034 * This is the stupid "wait for pipe to be readable or writable"
1042 void pipe_wait_readable(struct pipe_inode_info *pipe) in pipe_wait_readable() argument
1044 pipe_unlock(pipe); in pipe_wait_readable()
1045 wait_event_interruptible(pipe->rd_wait, pipe_readable(pipe)); in pipe_wait_readable()
1046 pipe_lock(pipe); in pipe_wait_readable()
1049 void pipe_wait_writable(struct pipe_inode_info *pipe) in pipe_wait_writable() argument
1051 pipe_unlock(pipe); in pipe_wait_writable()
1052 wait_event_interruptible(pipe->wr_wait, pipe_writable(pipe)); in pipe_wait_writable()
1053 pipe_lock(pipe); in pipe_wait_writable()
1058 * holding the pipe lock, so "*cnt" is stable and we know a wakeup cannot
1063 * because of the pipe lock, we can check the condition before being on
1066 * We use the 'rd_wait' waitqueue for pipe partner waiting.
1068 static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt) in wait_for_partner() argument
1074 prepare_to_wait(&pipe->rd_wait, &rdwait, TASK_INTERRUPTIBLE); in wait_for_partner()
1075 pipe_unlock(pipe); in wait_for_partner()
1077 finish_wait(&pipe->rd_wait, &rdwait); in wait_for_partner()
1078 pipe_lock(pipe); in wait_for_partner()
1085 static void wake_up_partner(struct pipe_inode_info *pipe) in wake_up_partner() argument
1087 wake_up_interruptible_all(&pipe->rd_wait); in wake_up_partner()
1092 struct pipe_inode_info *pipe; in fifo_open() local
1100 pipe = inode->i_pipe; in fifo_open()
1101 pipe->files++; in fifo_open()
1105 pipe = alloc_pipe_info(); in fifo_open()
1106 if (!pipe) in fifo_open()
1108 pipe->files = 1; in fifo_open()
1113 free_pipe_info(pipe); in fifo_open()
1114 pipe = inode->i_pipe; in fifo_open()
1116 inode->i_pipe = pipe; in fifo_open()
1120 filp->private_data = pipe; in fifo_open()
1121 /* OK, we have a pipe and it's pinned down */ in fifo_open()
1123 __pipe_lock(pipe); in fifo_open()
1135 pipe->r_counter++; in fifo_open()
1136 if (pipe->readers++ == 0) in fifo_open()
1137 wake_up_partner(pipe); in fifo_open()
1139 if (!is_pipe && !pipe->writers) { in fifo_open()
1143 filp->f_version = pipe->w_counter; in fifo_open()
1145 if (wait_for_partner(pipe, &pipe->w_counter)) in fifo_open()
1158 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers) in fifo_open()
1161 pipe->w_counter++; in fifo_open()
1162 if (!pipe->writers++) in fifo_open()
1163 wake_up_partner(pipe); in fifo_open()
1165 if (!is_pipe && !pipe->readers) { in fifo_open()
1166 if (wait_for_partner(pipe, &pipe->r_counter)) in fifo_open()
1179 pipe->readers++; in fifo_open()
1180 pipe->writers++; in fifo_open()
1181 pipe->r_counter++; in fifo_open()
1182 pipe->w_counter++; in fifo_open()
1183 if (pipe->readers == 1 || pipe->writers == 1) in fifo_open()
1184 wake_up_partner(pipe); in fifo_open()
1193 __pipe_unlock(pipe); in fifo_open()
1197 if (!--pipe->readers) in fifo_open()
1198 wake_up_interruptible(&pipe->wr_wait); in fifo_open()
1203 if (!--pipe->writers) in fifo_open()
1204 wake_up_interruptible_all(&pipe->rd_wait); in fifo_open()
1209 __pipe_unlock(pipe); in fifo_open()
1211 put_pipe_info(inode, pipe); in fifo_open()
1228 * Currently we rely on the pipe array holding a power-of-2 number
1236 /* Minimum pipe size, as required by POSIX */ in round_pipe_size()
1244 * Resize the pipe ring to a number of slots.
1246 * Note the pipe can be reduced in capacity, but only if the current
1250 int pipe_resize_ring(struct pipe_inode_info *pipe, unsigned int nr_slots) in pipe_resize_ring() argument
1260 spin_lock_irq(&pipe->rd_wait.lock); in pipe_resize_ring()
1261 mask = pipe->ring_size - 1; in pipe_resize_ring()
1262 head = pipe->head; in pipe_resize_ring()
1263 tail = pipe->tail; in pipe_resize_ring()
1267 spin_unlock_irq(&pipe->rd_wait.lock); in pipe_resize_ring()
1273 * The pipe array wraps around, so just start the new one at zero in pipe_resize_ring()
1280 memcpy(bufs, pipe->bufs + t, in pipe_resize_ring()
1283 unsigned int tsize = pipe->ring_size - t; in pipe_resize_ring()
1285 memcpy(bufs + tsize, pipe->bufs, in pipe_resize_ring()
1287 memcpy(bufs, pipe->bufs + t, in pipe_resize_ring()
1295 kfree(pipe->bufs); in pipe_resize_ring()
1296 pipe->bufs = bufs; in pipe_resize_ring()
1297 pipe->ring_size = nr_slots; in pipe_resize_ring()
1298 if (pipe->max_usage > nr_slots) in pipe_resize_ring()
1299 pipe->max_usage = nr_slots; in pipe_resize_ring()
1300 pipe->tail = tail; in pipe_resize_ring()
1301 pipe->head = head; in pipe_resize_ring()
1303 if (!pipe_has_watch_queue(pipe)) { in pipe_resize_ring()
1304 pipe->max_usage = nr_slots; in pipe_resize_ring()
1305 pipe->nr_accounted = nr_slots; in pipe_resize_ring()
1308 spin_unlock_irq(&pipe->rd_wait.lock); in pipe_resize_ring()
1311 wake_up_interruptible(&pipe->wr_wait); in pipe_resize_ring()
1316 * Allocate a new array of pipe buffers and copy the info over. Returns the
1317 * pipe size if successful, or return -ERROR on error.
1319 static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg) in pipe_set_size() argument
1325 if (pipe_has_watch_queue(pipe)) in pipe_set_size()
1335 * If trying to increase the pipe capacity, check that an in pipe_set_size()
1338 * Decreasing the pipe capacity is always permitted, even in pipe_set_size()
1341 if (nr_slots > pipe->max_usage && in pipe_set_size()
1345 user_bufs = account_pipe_buffers(pipe->user, pipe->nr_accounted, nr_slots); in pipe_set_size()
1347 if (nr_slots > pipe->max_usage && in pipe_set_size()
1355 ret = pipe_resize_ring(pipe, nr_slots); in pipe_set_size()
1359 return pipe->max_usage * PAGE_SIZE; in pipe_set_size()
1362 (void) account_pipe_buffers(pipe->user, nr_slots, pipe->nr_accounted); in pipe_set_size()
1369 * pipe.
1373 struct pipe_inode_info *pipe = file->private_data; in get_pipe_info() local
1375 if (file->f_op != &pipefifo_fops || !pipe) in get_pipe_info()
1377 if (for_splice && pipe_has_watch_queue(pipe)) in get_pipe_info()
1379 return pipe; in get_pipe_info()
1384 struct pipe_inode_info *pipe; in pipe_fcntl() local
1387 pipe = get_pipe_info(file, false); in pipe_fcntl()
1388 if (!pipe) in pipe_fcntl()
1391 __pipe_lock(pipe); in pipe_fcntl()
1395 ret = pipe_set_size(pipe, arg); in pipe_fcntl()
1398 ret = pipe->max_usage * PAGE_SIZE; in pipe_fcntl()
1405 __pipe_unlock(pipe); in pipe_fcntl()
1418 * d_name - pipe: will go nicely and kill the special-casing in procfs.