2 * Public API and common code for kernel->userspace relay file support.
4 * See Documentation/filesystems/relay.rst for an overview.
6 * Copyright (C) 2002-2005 - Tom Zanussi (zanussi@us.ibm.com), IBM Corp
7 * Copyright (C) 1999-2005 - Karim Yaghmour (karim@opersys.com)
9 * Moved to kernel/relay.c by Paul Mundt, 2006.
10 * November 2006 - CPU hotplug support by Mathieu Desnoyers
11 * (mathieu.desnoyers@polymtl.ca)
13 * This file is released under the GPL.
15 #include <linux/errno.h>
16 #include <linux/stddef.h>
17 #include <linux/slab.h>
18 #include <linux/export.h>
19 #include <linux/string.h>
20 #include <linux/relay.h>
21 #include <linux/vmalloc.h>
23 #include <linux/cpu.h>
24 #include <linux/splice.h>
26 /* list of open channels, for cpu hotplug */
27 static DEFINE_MUTEX(relay_channels_mutex);
28 static LIST_HEAD(relay_channels);
31 * fault() vm_op implementation for relay file mapping.
33 static vm_fault_t relay_buf_fault(struct vm_fault *vmf)
36 struct rchan_buf *buf = vmf->vma->vm_private_data;
37 pgoff_t pgoff = vmf->pgoff;
42 page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT));
44 return VM_FAULT_SIGBUS;
52 * vm_ops for relay file mappings.
54 static const struct vm_operations_struct relay_file_mmap_ops = {
55 .fault = relay_buf_fault,
59 * allocate an array of pointers of struct page
61 static struct page **relay_alloc_page_array(unsigned int n_pages)
63 const size_t pa_size = n_pages * sizeof(struct page *);
64 if (pa_size > PAGE_SIZE)
65 return vzalloc(pa_size);
66 return kzalloc(pa_size, GFP_KERNEL);
70 * free an array of pointers of struct page
72 static void relay_free_page_array(struct page **array)
78 * relay_mmap_buf: - mmap channel buffer to process address space
79 * @buf: relay channel buffer
80 * @vma: vm_area_struct describing memory to be mapped
82 * Returns 0 if ok, negative on error
84 * Caller should already have grabbed mmap_lock.
86 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
88 unsigned long length = vma->vm_end - vma->vm_start;
93 if (length != (unsigned long)buf->chan->alloc_size)
96 vma->vm_ops = &relay_file_mmap_ops;
97 vma->vm_flags |= VM_DONTEXPAND;
98 vma->vm_private_data = buf;
104 * relay_alloc_buf - allocate a channel buffer
105 * @buf: the buffer struct
106 * @size: total size of the buffer
108 * Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
109 * passed in size will get page aligned, if it isn't already.
111 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
114 unsigned int i, j, n_pages;
116 *size = PAGE_ALIGN(*size);
117 n_pages = *size >> PAGE_SHIFT;
119 buf->page_array = relay_alloc_page_array(n_pages);
120 if (!buf->page_array)
123 for (i = 0; i < n_pages; i++) {
124 buf->page_array[i] = alloc_page(GFP_KERNEL);
125 if (unlikely(!buf->page_array[i]))
127 set_page_private(buf->page_array[i], (unsigned long)buf);
129 mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
133 memset(mem, 0, *size);
134 buf->page_count = n_pages;
138 for (j = 0; j < i; j++)
139 __free_page(buf->page_array[j]);
140 relay_free_page_array(buf->page_array);
145 * relay_create_buf - allocate and initialize a channel buffer
146 * @chan: the relay channel
148 * Returns channel buffer if successful, %NULL otherwise.
150 static struct rchan_buf *relay_create_buf(struct rchan *chan)
152 struct rchan_buf *buf;
154 if (chan->n_subbufs > KMALLOC_MAX_SIZE / sizeof(size_t *))
157 buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
160 buf->padding = kmalloc_array(chan->n_subbufs, sizeof(size_t *),
165 buf->start = relay_alloc_buf(buf, &chan->alloc_size);
170 kref_get(&buf->chan->kref);
180 * relay_destroy_channel - free the channel struct
181 * @kref: target kernel reference that contains the relay channel
183 * Should only be called from kref_put().
185 static void relay_destroy_channel(struct kref *kref)
187 struct rchan *chan = container_of(kref, struct rchan, kref);
188 free_percpu(chan->buf);
193 * relay_destroy_buf - destroy an rchan_buf struct and associated buffer
194 * @buf: the buffer struct
196 static void relay_destroy_buf(struct rchan_buf *buf)
198 struct rchan *chan = buf->chan;
201 if (likely(buf->start)) {
203 for (i = 0; i < buf->page_count; i++)
204 __free_page(buf->page_array[i]);
205 relay_free_page_array(buf->page_array);
207 *per_cpu_ptr(chan->buf, buf->cpu) = NULL;
210 kref_put(&chan->kref, relay_destroy_channel);
214 * relay_remove_buf - remove a channel buffer
215 * @kref: target kernel reference that contains the relay buffer
217 * Removes the file from the filesystem, which also frees the
218 * rchan_buf_struct and the channel buffer. Should only be called from
221 static void relay_remove_buf(struct kref *kref)
223 struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
224 relay_destroy_buf(buf);
228 * relay_buf_empty - boolean, is the channel buffer empty?
229 * @buf: channel buffer
231 * Returns 1 if the buffer is empty, 0 otherwise.
233 static int relay_buf_empty(struct rchan_buf *buf)
235 return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
239 * relay_buf_full - boolean, is the channel buffer full?
240 * @buf: channel buffer
242 * Returns 1 if the buffer is full, 0 otherwise.
244 int relay_buf_full(struct rchan_buf *buf)
246 size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
247 return (ready >= buf->chan->n_subbufs) ? 1 : 0;
249 EXPORT_SYMBOL_GPL(relay_buf_full);
252 * High-level relay kernel API and associated functions.
256 * rchan_callback implementations defining default channel behavior. Used
257 * in place of corresponding NULL values in client callback struct.
261 * subbuf_start() default callback. Does nothing.
263 static int subbuf_start_default_callback (struct rchan_buf *buf,
268 if (relay_buf_full(buf))
275 * create_buf_file_create() default callback. Does nothing.
277 static struct dentry *create_buf_file_default_callback(const char *filename,
278 struct dentry *parent,
280 struct rchan_buf *buf,
287 * remove_buf_file() default callback. Does nothing.
289 static int remove_buf_file_default_callback(struct dentry *dentry)
295 * wakeup_readers - wake up readers waiting on a channel
296 * @work: contains the channel buffer
298 * This is the function used to defer reader waking
300 static void wakeup_readers(struct irq_work *work)
302 struct rchan_buf *buf;
304 buf = container_of(work, struct rchan_buf, wakeup_work);
305 wake_up_interruptible(&buf->read_wait);
309 * __relay_reset - reset a channel buffer
310 * @buf: the channel buffer
311 * @init: 1 if this is a first-time initialization
313 * See relay_reset() for description of effect.
315 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
320 init_waitqueue_head(&buf->read_wait);
321 kref_init(&buf->kref);
322 init_irq_work(&buf->wakeup_work, wakeup_readers);
324 irq_work_sync(&buf->wakeup_work);
327 buf->subbufs_produced = 0;
328 buf->subbufs_consumed = 0;
329 buf->bytes_consumed = 0;
331 buf->data = buf->start;
334 for (i = 0; i < buf->chan->n_subbufs; i++)
337 buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
341 * relay_reset - reset the channel
344 * This has the effect of erasing all data from all channel buffers
345 * and restarting the channel in its initial state. The buffers
346 * are not freed, so any mappings are still in effect.
348 * NOTE. Care should be taken that the channel isn't actually
349 * being used by anything when this call is made.
351 void relay_reset(struct rchan *chan)
353 struct rchan_buf *buf;
359 if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
360 __relay_reset(buf, 0);
364 mutex_lock(&relay_channels_mutex);
365 for_each_possible_cpu(i)
366 if ((buf = *per_cpu_ptr(chan->buf, i)))
367 __relay_reset(buf, 0);
368 mutex_unlock(&relay_channels_mutex);
370 EXPORT_SYMBOL_GPL(relay_reset);
372 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
373 struct dentry *dentry)
375 buf->dentry = dentry;
376 d_inode(buf->dentry)->i_size = buf->early_bytes;
379 static struct dentry *relay_create_buf_file(struct rchan *chan,
380 struct rchan_buf *buf,
383 struct dentry *dentry;
386 tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
389 snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
391 /* Create file in fs */
392 dentry = chan->cb->create_buf_file(tmpname, chan->parent,
404 * relay_open_buf - create a new relay channel buffer
406 * used by relay_open() and CPU hotplug.
408 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
410 struct rchan_buf *buf = NULL;
411 struct dentry *dentry;
414 return *per_cpu_ptr(chan->buf, 0);
416 buf = relay_create_buf(chan);
420 if (chan->has_base_filename) {
421 dentry = relay_create_buf_file(chan, buf, cpu);
424 relay_set_buf_dentry(buf, dentry);
426 /* Only retrieve global info, nothing more, nothing less */
427 dentry = chan->cb->create_buf_file(NULL, NULL,
430 if (IS_ERR_OR_NULL(dentry))
435 __relay_reset(buf, 1);
437 if(chan->is_global) {
438 *per_cpu_ptr(chan->buf, 0) = buf;
445 relay_destroy_buf(buf);
450 * relay_close_buf - close a channel buffer
451 * @buf: channel buffer
453 * Marks the buffer finalized and restores the default callbacks.
454 * The channel buffer and channel buffer data structure are then freed
455 * automatically when the last reference is given up.
457 static void relay_close_buf(struct rchan_buf *buf)
460 irq_work_sync(&buf->wakeup_work);
461 buf->chan->cb->remove_buf_file(buf->dentry);
462 kref_put(&buf->kref, relay_remove_buf);
465 static void setup_callbacks(struct rchan *chan,
466 struct rchan_callbacks *cb)
468 if (!cb->subbuf_start)
469 cb->subbuf_start = subbuf_start_default_callback;
470 if (!cb->create_buf_file)
471 cb->create_buf_file = create_buf_file_default_callback;
472 if (!cb->remove_buf_file)
473 cb->remove_buf_file = remove_buf_file_default_callback;
477 int relay_prepare_cpu(unsigned int cpu)
480 struct rchan_buf *buf;
482 mutex_lock(&relay_channels_mutex);
483 list_for_each_entry(chan, &relay_channels, list) {
484 if ((buf = *per_cpu_ptr(chan->buf, cpu)))
486 buf = relay_open_buf(chan, cpu);
488 pr_err("relay: cpu %d buffer creation failed\n", cpu);
489 mutex_unlock(&relay_channels_mutex);
492 *per_cpu_ptr(chan->buf, cpu) = buf;
494 mutex_unlock(&relay_channels_mutex);
499 * relay_open - create a new relay channel
500 * @base_filename: base name of files to create, %NULL for buffering only
501 * @parent: dentry of parent directory, %NULL for root directory or buffer
502 * @subbuf_size: size of sub-buffers
503 * @n_subbufs: number of sub-buffers
504 * @cb: client callback functions
505 * @private_data: user-defined data
507 * Returns channel pointer if successful, %NULL otherwise.
509 * Creates a channel buffer for each cpu using the sizes and
510 * attributes specified. The created channel buffer files
511 * will be named base_filename0...base_filenameN-1. File
512 * permissions will be %S_IRUSR.
514 * If opening a buffer (@parent = NULL) that you later wish to register
515 * in a filesystem, call relay_late_setup_files() once the @parent dentry
518 struct rchan *relay_open(const char *base_filename,
519 struct dentry *parent,
522 struct rchan_callbacks *cb,
527 struct rchan_buf *buf;
529 if (!(subbuf_size && n_subbufs))
531 if (subbuf_size > UINT_MAX / n_subbufs)
536 chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
540 chan->buf = alloc_percpu(struct rchan_buf *);
546 chan->version = RELAYFS_CHANNEL_VERSION;
547 chan->n_subbufs = n_subbufs;
548 chan->subbuf_size = subbuf_size;
549 chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
550 chan->parent = parent;
551 chan->private_data = private_data;
553 chan->has_base_filename = 1;
554 strlcpy(chan->base_filename, base_filename, NAME_MAX);
556 setup_callbacks(chan, cb);
557 kref_init(&chan->kref);
559 mutex_lock(&relay_channels_mutex);
560 for_each_online_cpu(i) {
561 buf = relay_open_buf(chan, i);
564 *per_cpu_ptr(chan->buf, i) = buf;
566 list_add(&chan->list, &relay_channels);
567 mutex_unlock(&relay_channels_mutex);
572 for_each_possible_cpu(i) {
573 if ((buf = *per_cpu_ptr(chan->buf, i)))
574 relay_close_buf(buf);
577 kref_put(&chan->kref, relay_destroy_channel);
578 mutex_unlock(&relay_channels_mutex);
581 EXPORT_SYMBOL_GPL(relay_open);
583 struct rchan_percpu_buf_dispatcher {
584 struct rchan_buf *buf;
585 struct dentry *dentry;
588 /* Called in atomic context. */
589 static void __relay_set_buf_dentry(void *info)
591 struct rchan_percpu_buf_dispatcher *p = info;
593 relay_set_buf_dentry(p->buf, p->dentry);
597 * relay_late_setup_files - triggers file creation
598 * @chan: channel to operate on
599 * @base_filename: base name of files to create
600 * @parent: dentry of parent directory, %NULL for root directory
602 * Returns 0 if successful, non-zero otherwise.
604 * Use to setup files for a previously buffer-only channel created
605 * by relay_open() with a NULL parent dentry.
607 * For example, this is useful for perfomring early tracing in kernel,
608 * before VFS is up and then exposing the early results once the dentry
611 int relay_late_setup_files(struct rchan *chan,
612 const char *base_filename,
613 struct dentry *parent)
616 unsigned int i, curr_cpu;
618 struct dentry *dentry;
619 struct rchan_buf *buf;
620 struct rchan_percpu_buf_dispatcher disp;
622 if (!chan || !base_filename)
625 strlcpy(chan->base_filename, base_filename, NAME_MAX);
627 mutex_lock(&relay_channels_mutex);
628 /* Is chan already set up? */
629 if (unlikely(chan->has_base_filename)) {
630 mutex_unlock(&relay_channels_mutex);
633 chan->has_base_filename = 1;
634 chan->parent = parent;
636 if (chan->is_global) {
638 buf = *per_cpu_ptr(chan->buf, 0);
639 if (!WARN_ON_ONCE(!buf)) {
640 dentry = relay_create_buf_file(chan, buf, 0);
641 if (dentry && !WARN_ON_ONCE(!chan->is_global)) {
642 relay_set_buf_dentry(buf, dentry);
646 mutex_unlock(&relay_channels_mutex);
650 curr_cpu = get_cpu();
652 * The CPU hotplug notifier ran before us and created buffers with
653 * no files associated. So it's safe to call relay_setup_buf_file()
654 * on all currently online CPUs.
656 for_each_online_cpu(i) {
657 buf = *per_cpu_ptr(chan->buf, i);
658 if (unlikely(!buf)) {
659 WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
664 dentry = relay_create_buf_file(chan, buf, i);
665 if (unlikely(!dentry)) {
671 local_irq_save(flags);
672 relay_set_buf_dentry(buf, dentry);
673 local_irq_restore(flags);
676 disp.dentry = dentry;
678 /* relay_channels_mutex must be held, so wait. */
679 err = smp_call_function_single(i,
680 __relay_set_buf_dentry,
687 mutex_unlock(&relay_channels_mutex);
691 EXPORT_SYMBOL_GPL(relay_late_setup_files);
694 * relay_switch_subbuf - switch to a new sub-buffer
695 * @buf: channel buffer
696 * @length: size of current event
698 * Returns either the length passed in or 0 if full.
700 * Performs sub-buffer-switch tasks such as invoking callbacks,
701 * updating padding counts, waking up readers, etc.
703 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
706 size_t old_subbuf, new_subbuf;
708 if (unlikely(length > buf->chan->subbuf_size))
711 if (buf->offset != buf->chan->subbuf_size + 1) {
712 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
713 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
714 buf->padding[old_subbuf] = buf->prev_padding;
715 buf->subbufs_produced++;
717 d_inode(buf->dentry)->i_size +=
718 buf->chan->subbuf_size -
719 buf->padding[old_subbuf];
721 buf->early_bytes += buf->chan->subbuf_size -
722 buf->padding[old_subbuf];
724 if (waitqueue_active(&buf->read_wait)) {
726 * Calling wake_up_interruptible() from here
727 * will deadlock if we happen to be logging
728 * from the scheduler (trying to re-grab
729 * rq->lock), so defer it.
731 irq_work_queue(&buf->wakeup_work);
736 new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
737 new = buf->start + new_subbuf * buf->chan->subbuf_size;
739 if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
740 buf->offset = buf->chan->subbuf_size + 1;
744 buf->padding[new_subbuf] = 0;
746 if (unlikely(length + buf->offset > buf->chan->subbuf_size))
752 buf->chan->last_toobig = length;
755 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
758 * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
760 * @cpu: the cpu associated with the channel buffer to update
761 * @subbufs_consumed: number of sub-buffers to add to current buf's count
763 * Adds to the channel buffer's consumed sub-buffer count.
764 * subbufs_consumed should be the number of sub-buffers newly consumed,
765 * not the total consumed.
767 * NOTE. Kernel clients don't need to call this function if the channel
768 * mode is 'overwrite'.
770 void relay_subbufs_consumed(struct rchan *chan,
772 size_t subbufs_consumed)
774 struct rchan_buf *buf;
776 if (!chan || cpu >= NR_CPUS)
779 buf = *per_cpu_ptr(chan->buf, cpu);
780 if (!buf || subbufs_consumed > chan->n_subbufs)
783 if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
784 buf->subbufs_consumed = buf->subbufs_produced;
786 buf->subbufs_consumed += subbufs_consumed;
788 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
791 * relay_close - close the channel
794 * Closes all channel buffers and frees the channel.
796 void relay_close(struct rchan *chan)
798 struct rchan_buf *buf;
804 mutex_lock(&relay_channels_mutex);
805 if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0)))
806 relay_close_buf(buf);
808 for_each_possible_cpu(i)
809 if ((buf = *per_cpu_ptr(chan->buf, i)))
810 relay_close_buf(buf);
812 if (chan->last_toobig)
813 printk(KERN_WARNING "relay: one or more items not logged "
814 "[item size (%zd) > sub-buffer size (%zd)]\n",
815 chan->last_toobig, chan->subbuf_size);
817 list_del(&chan->list);
818 kref_put(&chan->kref, relay_destroy_channel);
819 mutex_unlock(&relay_channels_mutex);
821 EXPORT_SYMBOL_GPL(relay_close);
824 * relay_flush - close the channel
827 * Flushes all channel buffers, i.e. forces buffer switch.
829 void relay_flush(struct rchan *chan)
831 struct rchan_buf *buf;
837 if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
838 relay_switch_subbuf(buf, 0);
842 mutex_lock(&relay_channels_mutex);
843 for_each_possible_cpu(i)
844 if ((buf = *per_cpu_ptr(chan->buf, i)))
845 relay_switch_subbuf(buf, 0);
846 mutex_unlock(&relay_channels_mutex);
848 EXPORT_SYMBOL_GPL(relay_flush);
851 * relay_file_open - open file op for relay files
855 * Increments the channel buffer refcount.
857 static int relay_file_open(struct inode *inode, struct file *filp)
859 struct rchan_buf *buf = inode->i_private;
860 kref_get(&buf->kref);
861 filp->private_data = buf;
863 return nonseekable_open(inode, filp);
867 * relay_file_mmap - mmap file op for relay files
869 * @vma: the vma describing what to map
871 * Calls upon relay_mmap_buf() to map the file into user space.
873 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
875 struct rchan_buf *buf = filp->private_data;
876 return relay_mmap_buf(buf, vma);
880 * relay_file_poll - poll file op for relay files
886 static __poll_t relay_file_poll(struct file *filp, poll_table *wait)
889 struct rchan_buf *buf = filp->private_data;
894 if (filp->f_mode & FMODE_READ) {
895 poll_wait(filp, &buf->read_wait, wait);
896 if (!relay_buf_empty(buf))
897 mask |= EPOLLIN | EPOLLRDNORM;
904 * relay_file_release - release file op for relay files
908 * Decrements the channel refcount, as the filesystem is
909 * no longer using it.
911 static int relay_file_release(struct inode *inode, struct file *filp)
913 struct rchan_buf *buf = filp->private_data;
914 kref_put(&buf->kref, relay_remove_buf);
920 * relay_file_read_consume - update the consumed count for the buffer
922 static void relay_file_read_consume(struct rchan_buf *buf,
924 size_t bytes_consumed)
926 size_t subbuf_size = buf->chan->subbuf_size;
927 size_t n_subbufs = buf->chan->n_subbufs;
930 if (buf->subbufs_produced == buf->subbufs_consumed &&
931 buf->offset == buf->bytes_consumed)
934 if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
935 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
936 buf->bytes_consumed = 0;
939 buf->bytes_consumed += bytes_consumed;
941 read_subbuf = buf->subbufs_consumed % n_subbufs;
943 read_subbuf = read_pos / buf->chan->subbuf_size;
944 if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
945 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
946 (buf->offset == subbuf_size))
948 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
949 buf->bytes_consumed = 0;
954 * relay_file_read_avail - boolean, are there unconsumed bytes available?
956 static int relay_file_read_avail(struct rchan_buf *buf)
958 size_t subbuf_size = buf->chan->subbuf_size;
959 size_t n_subbufs = buf->chan->n_subbufs;
960 size_t produced = buf->subbufs_produced;
963 relay_file_read_consume(buf, 0, 0);
965 consumed = buf->subbufs_consumed;
967 if (unlikely(buf->offset > subbuf_size)) {
968 if (produced == consumed)
973 if (unlikely(produced - consumed >= n_subbufs)) {
974 consumed = produced - n_subbufs + 1;
975 buf->subbufs_consumed = consumed;
976 buf->bytes_consumed = 0;
979 produced = (produced % n_subbufs) * subbuf_size + buf->offset;
980 consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
982 if (consumed > produced)
983 produced += n_subbufs * subbuf_size;
985 if (consumed == produced) {
986 if (buf->offset == subbuf_size &&
987 buf->subbufs_produced > buf->subbufs_consumed)
996 * relay_file_read_subbuf_avail - return bytes available in sub-buffer
997 * @read_pos: file read position
998 * @buf: relay channel buffer
1000 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1001 struct rchan_buf *buf)
1003 size_t padding, avail = 0;
1004 size_t read_subbuf, read_offset, write_subbuf, write_offset;
1005 size_t subbuf_size = buf->chan->subbuf_size;
1007 write_subbuf = (buf->data - buf->start) / subbuf_size;
1008 write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1009 read_subbuf = read_pos / subbuf_size;
1010 read_offset = read_pos % subbuf_size;
1011 padding = buf->padding[read_subbuf];
1013 if (read_subbuf == write_subbuf) {
1014 if (read_offset + padding < write_offset)
1015 avail = write_offset - (read_offset + padding);
1017 avail = (subbuf_size - padding) - read_offset;
1023 * relay_file_read_start_pos - find the first available byte to read
1024 * @buf: relay channel buffer
1026 * If the read_pos is in the middle of padding, return the
1027 * position of the first actually available byte, otherwise
1028 * return the original value.
1030 static size_t relay_file_read_start_pos(struct rchan_buf *buf)
1032 size_t read_subbuf, padding, padding_start, padding_end;
1033 size_t subbuf_size = buf->chan->subbuf_size;
1034 size_t n_subbufs = buf->chan->n_subbufs;
1035 size_t consumed = buf->subbufs_consumed % n_subbufs;
1036 size_t read_pos = consumed * subbuf_size + buf->bytes_consumed;
1038 read_subbuf = read_pos / subbuf_size;
1039 padding = buf->padding[read_subbuf];
1040 padding_start = (read_subbuf + 1) * subbuf_size - padding;
1041 padding_end = (read_subbuf + 1) * subbuf_size;
1042 if (read_pos >= padding_start && read_pos < padding_end) {
1043 read_subbuf = (read_subbuf + 1) % n_subbufs;
1044 read_pos = read_subbuf * subbuf_size;
1051 * relay_file_read_end_pos - return the new read position
1052 * @read_pos: file read position
1053 * @buf: relay channel buffer
1054 * @count: number of bytes to be read
1056 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1060 size_t read_subbuf, padding, end_pos;
1061 size_t subbuf_size = buf->chan->subbuf_size;
1062 size_t n_subbufs = buf->chan->n_subbufs;
1064 read_subbuf = read_pos / subbuf_size;
1065 padding = buf->padding[read_subbuf];
1066 if (read_pos % subbuf_size + count + padding == subbuf_size)
1067 end_pos = (read_subbuf + 1) * subbuf_size;
1069 end_pos = read_pos + count;
1070 if (end_pos >= subbuf_size * n_subbufs)
1076 static ssize_t relay_file_read(struct file *filp,
1077 char __user *buffer,
1081 struct rchan_buf *buf = filp->private_data;
1082 size_t read_start, avail;
1089 inode_lock(file_inode(filp));
1093 if (!relay_file_read_avail(buf))
1096 read_start = relay_file_read_start_pos(buf);
1097 avail = relay_file_read_subbuf_avail(read_start, buf);
1101 avail = min(count, avail);
1102 from = buf->start + read_start;
1104 if (copy_to_user(buffer, from, avail))
1111 relay_file_read_consume(buf, read_start, ret);
1112 *ppos = relay_file_read_end_pos(buf, read_start, ret);
1114 inode_unlock(file_inode(filp));
1119 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1121 rbuf->bytes_consumed += bytes_consumed;
1123 if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1124 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1125 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1129 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1130 struct pipe_buffer *buf)
1132 struct rchan_buf *rbuf;
1134 rbuf = (struct rchan_buf *)page_private(buf->page);
1135 relay_consume_bytes(rbuf, buf->private);
1138 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1139 .release = relay_pipe_buf_release,
1140 .try_steal = generic_pipe_buf_try_steal,
1141 .get = generic_pipe_buf_get,
1144 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1149 * subbuf_splice_actor - splice up to one subbuf's worth of data
1151 static ssize_t subbuf_splice_actor(struct file *in,
1153 struct pipe_inode_info *pipe,
1158 unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1159 struct rchan_buf *rbuf = in->private_data;
1160 unsigned int subbuf_size = rbuf->chan->subbuf_size;
1161 uint64_t pos = (uint64_t) *ppos;
1162 uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1163 size_t read_start = (size_t) do_div(pos, alloc_size);
1164 size_t read_subbuf = read_start / subbuf_size;
1165 size_t padding = rbuf->padding[read_subbuf];
1166 size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1167 struct page *pages[PIPE_DEF_BUFFERS];
1168 struct partial_page partial[PIPE_DEF_BUFFERS];
1169 struct splice_pipe_desc spd = {
1172 .nr_pages_max = PIPE_DEF_BUFFERS,
1174 .ops = &relay_pipe_buf_ops,
1175 .spd_release = relay_page_release,
1179 if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1181 if (splice_grow_spd(pipe, &spd))
1185 * Adjust read len, if longer than what is available
1187 if (len > (subbuf_size - read_start % subbuf_size))
1188 len = subbuf_size - read_start % subbuf_size;
1190 subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1191 pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1192 poff = read_start & ~PAGE_MASK;
1193 nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1195 for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1196 unsigned int this_len, this_end, private;
1197 unsigned int cur_pos = read_start + total_len;
1202 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1205 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1206 spd.partial[spd.nr_pages].offset = poff;
1208 this_end = cur_pos + this_len;
1209 if (this_end >= nonpad_end) {
1210 this_len = nonpad_end - cur_pos;
1211 private = this_len + padding;
1213 spd.partial[spd.nr_pages].len = this_len;
1214 spd.partial[spd.nr_pages].private = private;
1217 total_len += this_len;
1219 pidx = (pidx + 1) % subbuf_pages;
1221 if (this_end >= nonpad_end) {
1231 ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1232 if (ret < 0 || ret < total_len)
1235 if (read_start + ret == nonpad_end)
1239 splice_shrink_spd(&spd);
1243 static ssize_t relay_file_splice_read(struct file *in,
1245 struct pipe_inode_info *pipe,
1256 while (len && !spliced) {
1257 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1261 if (flags & SPLICE_F_NONBLOCK)
1271 spliced += nonpad_ret;
1281 const struct file_operations relay_file_operations = {
1282 .open = relay_file_open,
1283 .poll = relay_file_poll,
1284 .mmap = relay_file_mmap,
1285 .read = relay_file_read,
1286 .llseek = no_llseek,
1287 .release = relay_file_release,
1288 .splice_read = relay_file_splice_read,
1290 EXPORT_SYMBOL_GPL(relay_file_operations);