2 * Public API and common code for kernel->userspace relay file support.
4 * See Documentation/filesystems/relay.txt 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 * close() vm_op implementation for relay file mapping.
33 static void relay_file_mmap_close(struct vm_area_struct *vma)
35 struct rchan_buf *buf = vma->vm_private_data;
36 buf->chan->cb->buf_unmapped(buf, vma->vm_file);
40 * fault() vm_op implementation for relay file mapping.
42 static int relay_buf_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
45 struct rchan_buf *buf = vma->vm_private_data;
46 pgoff_t pgoff = vmf->pgoff;
51 page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT));
53 return VM_FAULT_SIGBUS;
61 * vm_ops for relay file mappings.
63 static const struct vm_operations_struct relay_file_mmap_ops = {
64 .fault = relay_buf_fault,
65 .close = relay_file_mmap_close,
69 * allocate an array of pointers of struct page
71 static struct page **relay_alloc_page_array(unsigned int n_pages)
73 const size_t pa_size = n_pages * sizeof(struct page *);
74 if (pa_size > PAGE_SIZE)
75 return vzalloc(pa_size);
76 return kzalloc(pa_size, GFP_KERNEL);
80 * free an array of pointers of struct page
82 static void relay_free_page_array(struct page **array)
88 * relay_mmap_buf: - mmap channel buffer to process address space
89 * @buf: relay channel buffer
90 * @vma: vm_area_struct describing memory to be mapped
92 * Returns 0 if ok, negative on error
94 * Caller should already have grabbed mmap_sem.
96 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
98 unsigned long length = vma->vm_end - vma->vm_start;
99 struct file *filp = vma->vm_file;
104 if (length != (unsigned long)buf->chan->alloc_size)
107 vma->vm_ops = &relay_file_mmap_ops;
108 vma->vm_flags |= VM_DONTEXPAND;
109 vma->vm_private_data = buf;
110 buf->chan->cb->buf_mapped(buf, filp);
116 * relay_alloc_buf - allocate a channel buffer
117 * @buf: the buffer struct
118 * @size: total size of the buffer
120 * Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
121 * passed in size will get page aligned, if it isn't already.
123 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
126 unsigned int i, j, n_pages;
128 *size = PAGE_ALIGN(*size);
129 n_pages = *size >> PAGE_SHIFT;
131 buf->page_array = relay_alloc_page_array(n_pages);
132 if (!buf->page_array)
135 for (i = 0; i < n_pages; i++) {
136 buf->page_array[i] = alloc_page(GFP_KERNEL);
137 if (unlikely(!buf->page_array[i]))
139 set_page_private(buf->page_array[i], (unsigned long)buf);
141 mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
145 memset(mem, 0, *size);
146 buf->page_count = n_pages;
150 for (j = 0; j < i; j++)
151 __free_page(buf->page_array[j]);
152 relay_free_page_array(buf->page_array);
157 * relay_create_buf - allocate and initialize a channel buffer
158 * @chan: the relay channel
160 * Returns channel buffer if successful, %NULL otherwise.
162 static struct rchan_buf *relay_create_buf(struct rchan *chan)
164 struct rchan_buf *buf;
166 if (chan->n_subbufs > UINT_MAX / sizeof(size_t *))
169 buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
172 buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
176 buf->start = relay_alloc_buf(buf, &chan->alloc_size);
181 kref_get(&buf->chan->kref);
191 * relay_destroy_channel - free the channel struct
192 * @kref: target kernel reference that contains the relay channel
194 * Should only be called from kref_put().
196 static void relay_destroy_channel(struct kref *kref)
198 struct rchan *chan = container_of(kref, struct rchan, kref);
203 * relay_destroy_buf - destroy an rchan_buf struct and associated buffer
204 * @buf: the buffer struct
206 static void relay_destroy_buf(struct rchan_buf *buf)
208 struct rchan *chan = buf->chan;
211 if (likely(buf->start)) {
213 for (i = 0; i < buf->page_count; i++)
214 __free_page(buf->page_array[i]);
215 relay_free_page_array(buf->page_array);
217 chan->buf[buf->cpu] = NULL;
220 kref_put(&chan->kref, relay_destroy_channel);
224 * relay_remove_buf - remove a channel buffer
225 * @kref: target kernel reference that contains the relay buffer
227 * Removes the file from the filesystem, which also frees the
228 * rchan_buf_struct and the channel buffer. Should only be called from
231 static void relay_remove_buf(struct kref *kref)
233 struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
234 relay_destroy_buf(buf);
238 * relay_buf_empty - boolean, is the channel buffer empty?
239 * @buf: channel buffer
241 * Returns 1 if the buffer is empty, 0 otherwise.
243 static int relay_buf_empty(struct rchan_buf *buf)
245 return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
249 * relay_buf_full - boolean, is the channel buffer full?
250 * @buf: channel buffer
252 * Returns 1 if the buffer is full, 0 otherwise.
254 int relay_buf_full(struct rchan_buf *buf)
256 size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
257 return (ready >= buf->chan->n_subbufs) ? 1 : 0;
259 EXPORT_SYMBOL_GPL(relay_buf_full);
262 * High-level relay kernel API and associated functions.
266 * rchan_callback implementations defining default channel behavior. Used
267 * in place of corresponding NULL values in client callback struct.
271 * subbuf_start() default callback. Does nothing.
273 static int subbuf_start_default_callback (struct rchan_buf *buf,
278 if (relay_buf_full(buf))
285 * buf_mapped() default callback. Does nothing.
287 static void buf_mapped_default_callback(struct rchan_buf *buf,
293 * buf_unmapped() default callback. Does nothing.
295 static void buf_unmapped_default_callback(struct rchan_buf *buf,
301 * create_buf_file_create() default callback. Does nothing.
303 static struct dentry *create_buf_file_default_callback(const char *filename,
304 struct dentry *parent,
306 struct rchan_buf *buf,
313 * remove_buf_file() default callback. Does nothing.
315 static int remove_buf_file_default_callback(struct dentry *dentry)
320 /* relay channel default callbacks */
321 static struct rchan_callbacks default_channel_callbacks = {
322 .subbuf_start = subbuf_start_default_callback,
323 .buf_mapped = buf_mapped_default_callback,
324 .buf_unmapped = buf_unmapped_default_callback,
325 .create_buf_file = create_buf_file_default_callback,
326 .remove_buf_file = remove_buf_file_default_callback,
330 * wakeup_readers - wake up readers waiting on a channel
331 * @data: contains the channel buffer
333 * This is the timer function used to defer reader waking.
335 static void wakeup_readers(unsigned long data)
337 struct rchan_buf *buf = (struct rchan_buf *)data;
338 wake_up_interruptible(&buf->read_wait);
340 * Stupid polling for now:
342 mod_timer(&buf->timer, jiffies + 1);
346 * __relay_reset - reset a channel buffer
347 * @buf: the channel buffer
348 * @init: 1 if this is a first-time initialization
350 * See relay_reset() for description of effect.
352 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
357 init_waitqueue_head(&buf->read_wait);
358 kref_init(&buf->kref);
359 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
360 mod_timer(&buf->timer, jiffies + 1);
362 del_timer_sync(&buf->timer);
364 buf->subbufs_produced = 0;
365 buf->subbufs_consumed = 0;
366 buf->bytes_consumed = 0;
368 buf->data = buf->start;
371 for (i = 0; i < buf->chan->n_subbufs; i++)
374 buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
378 * relay_reset - reset the channel
381 * This has the effect of erasing all data from all channel buffers
382 * and restarting the channel in its initial state. The buffers
383 * are not freed, so any mappings are still in effect.
385 * NOTE. Care should be taken that the channel isn't actually
386 * being used by anything when this call is made.
388 void relay_reset(struct rchan *chan)
395 if (chan->is_global && chan->buf[0]) {
396 __relay_reset(chan->buf[0], 0);
400 mutex_lock(&relay_channels_mutex);
401 for_each_possible_cpu(i)
403 __relay_reset(chan->buf[i], 0);
404 mutex_unlock(&relay_channels_mutex);
406 EXPORT_SYMBOL_GPL(relay_reset);
408 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
409 struct dentry *dentry)
411 buf->dentry = dentry;
412 d_inode(buf->dentry)->i_size = buf->early_bytes;
415 static struct dentry *relay_create_buf_file(struct rchan *chan,
416 struct rchan_buf *buf,
419 struct dentry *dentry;
422 tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
425 snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
427 /* Create file in fs */
428 dentry = chan->cb->create_buf_file(tmpname, chan->parent,
438 * relay_open_buf - create a new relay channel buffer
440 * used by relay_open() and CPU hotplug.
442 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
444 struct rchan_buf *buf = NULL;
445 struct dentry *dentry;
450 buf = relay_create_buf(chan);
454 if (chan->has_base_filename) {
455 dentry = relay_create_buf_file(chan, buf, cpu);
458 relay_set_buf_dentry(buf, dentry);
462 __relay_reset(buf, 1);
464 if(chan->is_global) {
472 relay_destroy_buf(buf);
477 * relay_close_buf - close a channel buffer
478 * @buf: channel buffer
480 * Marks the buffer finalized and restores the default callbacks.
481 * The channel buffer and channel buffer data structure are then freed
482 * automatically when the last reference is given up.
484 static void relay_close_buf(struct rchan_buf *buf)
487 del_timer_sync(&buf->timer);
488 buf->chan->cb->remove_buf_file(buf->dentry);
489 kref_put(&buf->kref, relay_remove_buf);
492 static void setup_callbacks(struct rchan *chan,
493 struct rchan_callbacks *cb)
496 chan->cb = &default_channel_callbacks;
500 if (!cb->subbuf_start)
501 cb->subbuf_start = subbuf_start_default_callback;
503 cb->buf_mapped = buf_mapped_default_callback;
504 if (!cb->buf_unmapped)
505 cb->buf_unmapped = buf_unmapped_default_callback;
506 if (!cb->create_buf_file)
507 cb->create_buf_file = create_buf_file_default_callback;
508 if (!cb->remove_buf_file)
509 cb->remove_buf_file = remove_buf_file_default_callback;
514 * relay_hotcpu_callback - CPU hotplug callback
515 * @nb: notifier block
516 * @action: hotplug action to take
519 * Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
521 static int relay_hotcpu_callback(struct notifier_block *nb,
522 unsigned long action,
525 unsigned int hotcpu = (unsigned long)hcpu;
530 case CPU_UP_PREPARE_FROZEN:
531 mutex_lock(&relay_channels_mutex);
532 list_for_each_entry(chan, &relay_channels, list) {
533 if (chan->buf[hotcpu])
535 chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
536 if(!chan->buf[hotcpu]) {
538 "relay_hotcpu_callback: cpu %d buffer "
539 "creation failed\n", hotcpu);
540 mutex_unlock(&relay_channels_mutex);
541 return notifier_from_errno(-ENOMEM);
544 mutex_unlock(&relay_channels_mutex);
547 case CPU_DEAD_FROZEN:
548 /* No need to flush the cpu : will be flushed upon
549 * final relay_flush() call. */
556 * relay_open - create a new relay channel
557 * @base_filename: base name of files to create, %NULL for buffering only
558 * @parent: dentry of parent directory, %NULL for root directory or buffer
559 * @subbuf_size: size of sub-buffers
560 * @n_subbufs: number of sub-buffers
561 * @cb: client callback functions
562 * @private_data: user-defined data
564 * Returns channel pointer if successful, %NULL otherwise.
566 * Creates a channel buffer for each cpu using the sizes and
567 * attributes specified. The created channel buffer files
568 * will be named base_filename0...base_filenameN-1. File
569 * permissions will be %S_IRUSR.
571 struct rchan *relay_open(const char *base_filename,
572 struct dentry *parent,
575 struct rchan_callbacks *cb,
581 if (!(subbuf_size && n_subbufs))
583 if (subbuf_size > UINT_MAX / n_subbufs)
586 chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
590 chan->version = RELAYFS_CHANNEL_VERSION;
591 chan->n_subbufs = n_subbufs;
592 chan->subbuf_size = subbuf_size;
593 chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
594 chan->parent = parent;
595 chan->private_data = private_data;
597 chan->has_base_filename = 1;
598 strlcpy(chan->base_filename, base_filename, NAME_MAX);
600 setup_callbacks(chan, cb);
601 kref_init(&chan->kref);
603 mutex_lock(&relay_channels_mutex);
604 for_each_online_cpu(i) {
605 chan->buf[i] = relay_open_buf(chan, i);
609 list_add(&chan->list, &relay_channels);
610 mutex_unlock(&relay_channels_mutex);
615 for_each_possible_cpu(i) {
617 relay_close_buf(chan->buf[i]);
620 kref_put(&chan->kref, relay_destroy_channel);
621 mutex_unlock(&relay_channels_mutex);
624 EXPORT_SYMBOL_GPL(relay_open);
626 struct rchan_percpu_buf_dispatcher {
627 struct rchan_buf *buf;
628 struct dentry *dentry;
631 /* Called in atomic context. */
632 static void __relay_set_buf_dentry(void *info)
634 struct rchan_percpu_buf_dispatcher *p = info;
636 relay_set_buf_dentry(p->buf, p->dentry);
640 * relay_late_setup_files - triggers file creation
641 * @chan: channel to operate on
642 * @base_filename: base name of files to create
643 * @parent: dentry of parent directory, %NULL for root directory
645 * Returns 0 if successful, non-zero otherwise.
647 * Use to setup files for a previously buffer-only channel.
648 * Useful to do early tracing in kernel, before VFS is up, for example.
650 int relay_late_setup_files(struct rchan *chan,
651 const char *base_filename,
652 struct dentry *parent)
655 unsigned int i, curr_cpu;
657 struct dentry *dentry;
658 struct rchan_percpu_buf_dispatcher disp;
660 if (!chan || !base_filename)
663 strlcpy(chan->base_filename, base_filename, NAME_MAX);
665 mutex_lock(&relay_channels_mutex);
666 /* Is chan already set up? */
667 if (unlikely(chan->has_base_filename)) {
668 mutex_unlock(&relay_channels_mutex);
671 chan->has_base_filename = 1;
672 chan->parent = parent;
673 curr_cpu = get_cpu();
675 * The CPU hotplug notifier ran before us and created buffers with
676 * no files associated. So it's safe to call relay_setup_buf_file()
677 * on all currently online CPUs.
679 for_each_online_cpu(i) {
680 if (unlikely(!chan->buf[i])) {
681 WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
686 dentry = relay_create_buf_file(chan, chan->buf[i], i);
687 if (unlikely(!dentry)) {
693 local_irq_save(flags);
694 relay_set_buf_dentry(chan->buf[i], dentry);
695 local_irq_restore(flags);
697 disp.buf = chan->buf[i];
698 disp.dentry = dentry;
700 /* relay_channels_mutex must be held, so wait. */
701 err = smp_call_function_single(i,
702 __relay_set_buf_dentry,
709 mutex_unlock(&relay_channels_mutex);
715 * relay_switch_subbuf - switch to a new sub-buffer
716 * @buf: channel buffer
717 * @length: size of current event
719 * Returns either the length passed in or 0 if full.
721 * Performs sub-buffer-switch tasks such as invoking callbacks,
722 * updating padding counts, waking up readers, etc.
724 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
727 size_t old_subbuf, new_subbuf;
729 if (unlikely(length > buf->chan->subbuf_size))
732 if (buf->offset != buf->chan->subbuf_size + 1) {
733 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
734 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
735 buf->padding[old_subbuf] = buf->prev_padding;
736 buf->subbufs_produced++;
738 d_inode(buf->dentry)->i_size +=
739 buf->chan->subbuf_size -
740 buf->padding[old_subbuf];
742 buf->early_bytes += buf->chan->subbuf_size -
743 buf->padding[old_subbuf];
747 new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
748 new = buf->start + new_subbuf * buf->chan->subbuf_size;
750 if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
751 buf->offset = buf->chan->subbuf_size + 1;
755 buf->padding[new_subbuf] = 0;
757 if (unlikely(length + buf->offset > buf->chan->subbuf_size))
763 buf->chan->last_toobig = length;
766 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
769 * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
771 * @cpu: the cpu associated with the channel buffer to update
772 * @subbufs_consumed: number of sub-buffers to add to current buf's count
774 * Adds to the channel buffer's consumed sub-buffer count.
775 * subbufs_consumed should be the number of sub-buffers newly consumed,
776 * not the total consumed.
778 * NOTE. Kernel clients don't need to call this function if the channel
779 * mode is 'overwrite'.
781 void relay_subbufs_consumed(struct rchan *chan,
783 size_t subbufs_consumed)
785 struct rchan_buf *buf;
790 if (cpu >= NR_CPUS || !chan->buf[cpu] ||
791 subbufs_consumed > chan->n_subbufs)
794 buf = chan->buf[cpu];
795 if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
796 buf->subbufs_consumed = buf->subbufs_produced;
798 buf->subbufs_consumed += subbufs_consumed;
800 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
803 * relay_close - close the channel
806 * Closes all channel buffers and frees the channel.
808 void relay_close(struct rchan *chan)
815 mutex_lock(&relay_channels_mutex);
816 if (chan->is_global && chan->buf[0])
817 relay_close_buf(chan->buf[0]);
819 for_each_possible_cpu(i)
821 relay_close_buf(chan->buf[i]);
823 if (chan->last_toobig)
824 printk(KERN_WARNING "relay: one or more items not logged "
825 "[item size (%Zd) > sub-buffer size (%Zd)]\n",
826 chan->last_toobig, chan->subbuf_size);
828 list_del(&chan->list);
829 kref_put(&chan->kref, relay_destroy_channel);
830 mutex_unlock(&relay_channels_mutex);
832 EXPORT_SYMBOL_GPL(relay_close);
835 * relay_flush - close the channel
838 * Flushes all channel buffers, i.e. forces buffer switch.
840 void relay_flush(struct rchan *chan)
847 if (chan->is_global && chan->buf[0]) {
848 relay_switch_subbuf(chan->buf[0], 0);
852 mutex_lock(&relay_channels_mutex);
853 for_each_possible_cpu(i)
855 relay_switch_subbuf(chan->buf[i], 0);
856 mutex_unlock(&relay_channels_mutex);
858 EXPORT_SYMBOL_GPL(relay_flush);
861 * relay_file_open - open file op for relay files
865 * Increments the channel buffer refcount.
867 static int relay_file_open(struct inode *inode, struct file *filp)
869 struct rchan_buf *buf = inode->i_private;
870 kref_get(&buf->kref);
871 filp->private_data = buf;
873 return nonseekable_open(inode, filp);
877 * relay_file_mmap - mmap file op for relay files
879 * @vma: the vma describing what to map
881 * Calls upon relay_mmap_buf() to map the file into user space.
883 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
885 struct rchan_buf *buf = filp->private_data;
886 return relay_mmap_buf(buf, vma);
890 * relay_file_poll - poll file op for relay files
896 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
898 unsigned int mask = 0;
899 struct rchan_buf *buf = filp->private_data;
904 if (filp->f_mode & FMODE_READ) {
905 poll_wait(filp, &buf->read_wait, wait);
906 if (!relay_buf_empty(buf))
907 mask |= POLLIN | POLLRDNORM;
914 * relay_file_release - release file op for relay files
918 * Decrements the channel refcount, as the filesystem is
919 * no longer using it.
921 static int relay_file_release(struct inode *inode, struct file *filp)
923 struct rchan_buf *buf = filp->private_data;
924 kref_put(&buf->kref, relay_remove_buf);
930 * relay_file_read_consume - update the consumed count for the buffer
932 static void relay_file_read_consume(struct rchan_buf *buf,
934 size_t bytes_consumed)
936 size_t subbuf_size = buf->chan->subbuf_size;
937 size_t n_subbufs = buf->chan->n_subbufs;
940 if (buf->subbufs_produced == buf->subbufs_consumed &&
941 buf->offset == buf->bytes_consumed)
944 if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
945 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
946 buf->bytes_consumed = 0;
949 buf->bytes_consumed += bytes_consumed;
951 read_subbuf = buf->subbufs_consumed % n_subbufs;
953 read_subbuf = read_pos / buf->chan->subbuf_size;
954 if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
955 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
956 (buf->offset == subbuf_size))
958 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
959 buf->bytes_consumed = 0;
964 * relay_file_read_avail - boolean, are there unconsumed bytes available?
966 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
968 size_t subbuf_size = buf->chan->subbuf_size;
969 size_t n_subbufs = buf->chan->n_subbufs;
970 size_t produced = buf->subbufs_produced;
971 size_t consumed = buf->subbufs_consumed;
973 relay_file_read_consume(buf, read_pos, 0);
975 consumed = buf->subbufs_consumed;
977 if (unlikely(buf->offset > subbuf_size)) {
978 if (produced == consumed)
983 if (unlikely(produced - consumed >= n_subbufs)) {
984 consumed = produced - n_subbufs + 1;
985 buf->subbufs_consumed = consumed;
986 buf->bytes_consumed = 0;
989 produced = (produced % n_subbufs) * subbuf_size + buf->offset;
990 consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
992 if (consumed > produced)
993 produced += n_subbufs * subbuf_size;
995 if (consumed == produced) {
996 if (buf->offset == subbuf_size &&
997 buf->subbufs_produced > buf->subbufs_consumed)
1006 * relay_file_read_subbuf_avail - return bytes available in sub-buffer
1007 * @read_pos: file read position
1008 * @buf: relay channel buffer
1010 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1011 struct rchan_buf *buf)
1013 size_t padding, avail = 0;
1014 size_t read_subbuf, read_offset, write_subbuf, write_offset;
1015 size_t subbuf_size = buf->chan->subbuf_size;
1017 write_subbuf = (buf->data - buf->start) / subbuf_size;
1018 write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1019 read_subbuf = read_pos / subbuf_size;
1020 read_offset = read_pos % subbuf_size;
1021 padding = buf->padding[read_subbuf];
1023 if (read_subbuf == write_subbuf) {
1024 if (read_offset + padding < write_offset)
1025 avail = write_offset - (read_offset + padding);
1027 avail = (subbuf_size - padding) - read_offset;
1033 * relay_file_read_start_pos - find the first available byte to read
1034 * @read_pos: file read position
1035 * @buf: relay channel buffer
1037 * If the @read_pos is in the middle of padding, return the
1038 * position of the first actually available byte, otherwise
1039 * return the original value.
1041 static size_t relay_file_read_start_pos(size_t read_pos,
1042 struct rchan_buf *buf)
1044 size_t read_subbuf, padding, padding_start, padding_end;
1045 size_t subbuf_size = buf->chan->subbuf_size;
1046 size_t n_subbufs = buf->chan->n_subbufs;
1047 size_t consumed = buf->subbufs_consumed % n_subbufs;
1050 read_pos = consumed * subbuf_size + buf->bytes_consumed;
1051 read_subbuf = read_pos / subbuf_size;
1052 padding = buf->padding[read_subbuf];
1053 padding_start = (read_subbuf + 1) * subbuf_size - padding;
1054 padding_end = (read_subbuf + 1) * subbuf_size;
1055 if (read_pos >= padding_start && read_pos < padding_end) {
1056 read_subbuf = (read_subbuf + 1) % n_subbufs;
1057 read_pos = read_subbuf * subbuf_size;
1064 * relay_file_read_end_pos - return the new read position
1065 * @read_pos: file read position
1066 * @buf: relay channel buffer
1067 * @count: number of bytes to be read
1069 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1073 size_t read_subbuf, padding, end_pos;
1074 size_t subbuf_size = buf->chan->subbuf_size;
1075 size_t n_subbufs = buf->chan->n_subbufs;
1077 read_subbuf = read_pos / subbuf_size;
1078 padding = buf->padding[read_subbuf];
1079 if (read_pos % subbuf_size + count + padding == subbuf_size)
1080 end_pos = (read_subbuf + 1) * subbuf_size;
1082 end_pos = read_pos + count;
1083 if (end_pos >= subbuf_size * n_subbufs)
1090 * subbuf_read_actor - read up to one subbuf's worth of data
1092 static int subbuf_read_actor(size_t read_start,
1093 struct rchan_buf *buf,
1095 read_descriptor_t *desc)
1100 from = buf->start + read_start;
1102 if (copy_to_user(desc->arg.buf, from, avail)) {
1103 desc->error = -EFAULT;
1106 desc->arg.data += ret;
1107 desc->written += ret;
1113 typedef int (*subbuf_actor_t) (size_t read_start,
1114 struct rchan_buf *buf,
1116 read_descriptor_t *desc);
1119 * relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1121 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1122 subbuf_actor_t subbuf_actor,
1123 read_descriptor_t *desc)
1125 struct rchan_buf *buf = filp->private_data;
1126 size_t read_start, avail;
1132 mutex_lock(&file_inode(filp)->i_mutex);
1134 if (!relay_file_read_avail(buf, *ppos))
1137 read_start = relay_file_read_start_pos(*ppos, buf);
1138 avail = relay_file_read_subbuf_avail(read_start, buf);
1142 avail = min(desc->count, avail);
1143 ret = subbuf_actor(read_start, buf, avail, desc);
1144 if (desc->error < 0)
1148 relay_file_read_consume(buf, read_start, ret);
1149 *ppos = relay_file_read_end_pos(buf, read_start, ret);
1151 } while (desc->count && ret);
1152 mutex_unlock(&file_inode(filp)->i_mutex);
1154 return desc->written;
1157 static ssize_t relay_file_read(struct file *filp,
1158 char __user *buffer,
1162 read_descriptor_t desc;
1165 desc.arg.buf = buffer;
1167 return relay_file_read_subbufs(filp, ppos, subbuf_read_actor, &desc);
1170 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1172 rbuf->bytes_consumed += bytes_consumed;
1174 if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1175 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1176 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1180 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1181 struct pipe_buffer *buf)
1183 struct rchan_buf *rbuf;
1185 rbuf = (struct rchan_buf *)page_private(buf->page);
1186 relay_consume_bytes(rbuf, buf->private);
1189 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1191 .confirm = generic_pipe_buf_confirm,
1192 .release = relay_pipe_buf_release,
1193 .steal = generic_pipe_buf_steal,
1194 .get = generic_pipe_buf_get,
1197 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1202 * subbuf_splice_actor - splice up to one subbuf's worth of data
1204 static ssize_t subbuf_splice_actor(struct file *in,
1206 struct pipe_inode_info *pipe,
1211 unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1212 struct rchan_buf *rbuf = in->private_data;
1213 unsigned int subbuf_size = rbuf->chan->subbuf_size;
1214 uint64_t pos = (uint64_t) *ppos;
1215 uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1216 size_t read_start = (size_t) do_div(pos, alloc_size);
1217 size_t read_subbuf = read_start / subbuf_size;
1218 size_t padding = rbuf->padding[read_subbuf];
1219 size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1220 struct page *pages[PIPE_DEF_BUFFERS];
1221 struct partial_page partial[PIPE_DEF_BUFFERS];
1222 struct splice_pipe_desc spd = {
1225 .nr_pages_max = PIPE_DEF_BUFFERS,
1228 .ops = &relay_pipe_buf_ops,
1229 .spd_release = relay_page_release,
1233 if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1235 if (splice_grow_spd(pipe, &spd))
1239 * Adjust read len, if longer than what is available
1241 if (len > (subbuf_size - read_start % subbuf_size))
1242 len = subbuf_size - read_start % subbuf_size;
1244 subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1245 pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1246 poff = read_start & ~PAGE_MASK;
1247 nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1249 for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1250 unsigned int this_len, this_end, private;
1251 unsigned int cur_pos = read_start + total_len;
1256 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1259 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1260 spd.partial[spd.nr_pages].offset = poff;
1262 this_end = cur_pos + this_len;
1263 if (this_end >= nonpad_end) {
1264 this_len = nonpad_end - cur_pos;
1265 private = this_len + padding;
1267 spd.partial[spd.nr_pages].len = this_len;
1268 spd.partial[spd.nr_pages].private = private;
1271 total_len += this_len;
1273 pidx = (pidx + 1) % subbuf_pages;
1275 if (this_end >= nonpad_end) {
1285 ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1286 if (ret < 0 || ret < total_len)
1289 if (read_start + ret == nonpad_end)
1293 splice_shrink_spd(&spd);
1297 static ssize_t relay_file_splice_read(struct file *in,
1299 struct pipe_inode_info *pipe,
1310 while (len && !spliced) {
1311 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1315 if (flags & SPLICE_F_NONBLOCK)
1325 spliced += nonpad_ret;
1335 const struct file_operations relay_file_operations = {
1336 .open = relay_file_open,
1337 .poll = relay_file_poll,
1338 .mmap = relay_file_mmap,
1339 .read = relay_file_read,
1340 .llseek = no_llseek,
1341 .release = relay_file_release,
1342 .splice_read = relay_file_splice_read,
1344 EXPORT_SYMBOL_GPL(relay_file_operations);
1346 static __init int relay_init(void)
1349 hotcpu_notifier(relay_hotcpu_callback, 0);
1353 early_initcall(relay_init);