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)
84 if (is_vmalloc_addr(array))
91 * relay_mmap_buf: - mmap channel buffer to process address space
92 * @buf: relay channel buffer
93 * @vma: vm_area_struct describing memory to be mapped
95 * Returns 0 if ok, negative on error
97 * Caller should already have grabbed mmap_sem.
99 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
101 unsigned long length = vma->vm_end - vma->vm_start;
102 struct file *filp = vma->vm_file;
107 if (length != (unsigned long)buf->chan->alloc_size)
110 vma->vm_ops = &relay_file_mmap_ops;
111 vma->vm_flags |= VM_DONTEXPAND;
112 vma->vm_private_data = buf;
113 buf->chan->cb->buf_mapped(buf, filp);
119 * relay_alloc_buf - allocate a channel buffer
120 * @buf: the buffer struct
121 * @size: total size of the buffer
123 * Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
124 * passed in size will get page aligned, if it isn't already.
126 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
129 unsigned int i, j, n_pages;
131 *size = PAGE_ALIGN(*size);
132 n_pages = *size >> PAGE_SHIFT;
134 buf->page_array = relay_alloc_page_array(n_pages);
135 if (!buf->page_array)
138 for (i = 0; i < n_pages; i++) {
139 buf->page_array[i] = alloc_page(GFP_KERNEL);
140 if (unlikely(!buf->page_array[i]))
142 set_page_private(buf->page_array[i], (unsigned long)buf);
144 mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
148 memset(mem, 0, *size);
149 buf->page_count = n_pages;
153 for (j = 0; j < i; j++)
154 __free_page(buf->page_array[j]);
155 relay_free_page_array(buf->page_array);
160 * relay_create_buf - allocate and initialize a channel buffer
161 * @chan: the relay channel
163 * Returns channel buffer if successful, %NULL otherwise.
165 static struct rchan_buf *relay_create_buf(struct rchan *chan)
167 struct rchan_buf *buf;
169 if (chan->n_subbufs > UINT_MAX / sizeof(size_t *))
172 buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
175 buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
179 buf->start = relay_alloc_buf(buf, &chan->alloc_size);
184 kref_get(&buf->chan->kref);
194 * relay_destroy_channel - free the channel struct
195 * @kref: target kernel reference that contains the relay channel
197 * Should only be called from kref_put().
199 static void relay_destroy_channel(struct kref *kref)
201 struct rchan *chan = container_of(kref, struct rchan, kref);
206 * relay_destroy_buf - destroy an rchan_buf struct and associated buffer
207 * @buf: the buffer struct
209 static void relay_destroy_buf(struct rchan_buf *buf)
211 struct rchan *chan = buf->chan;
214 if (likely(buf->start)) {
216 for (i = 0; i < buf->page_count; i++)
217 __free_page(buf->page_array[i]);
218 relay_free_page_array(buf->page_array);
220 chan->buf[buf->cpu] = NULL;
223 kref_put(&chan->kref, relay_destroy_channel);
227 * relay_remove_buf - remove a channel buffer
228 * @kref: target kernel reference that contains the relay buffer
230 * Removes the file from the filesystem, which also frees the
231 * rchan_buf_struct and the channel buffer. Should only be called from
234 static void relay_remove_buf(struct kref *kref)
236 struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
237 relay_destroy_buf(buf);
241 * relay_buf_empty - boolean, is the channel buffer empty?
242 * @buf: channel buffer
244 * Returns 1 if the buffer is empty, 0 otherwise.
246 static int relay_buf_empty(struct rchan_buf *buf)
248 return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
252 * relay_buf_full - boolean, is the channel buffer full?
253 * @buf: channel buffer
255 * Returns 1 if the buffer is full, 0 otherwise.
257 int relay_buf_full(struct rchan_buf *buf)
259 size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
260 return (ready >= buf->chan->n_subbufs) ? 1 : 0;
262 EXPORT_SYMBOL_GPL(relay_buf_full);
265 * High-level relay kernel API and associated functions.
269 * rchan_callback implementations defining default channel behavior. Used
270 * in place of corresponding NULL values in client callback struct.
274 * subbuf_start() default callback. Does nothing.
276 static int subbuf_start_default_callback (struct rchan_buf *buf,
281 if (relay_buf_full(buf))
288 * buf_mapped() default callback. Does nothing.
290 static void buf_mapped_default_callback(struct rchan_buf *buf,
296 * buf_unmapped() default callback. Does nothing.
298 static void buf_unmapped_default_callback(struct rchan_buf *buf,
304 * create_buf_file_create() default callback. Does nothing.
306 static struct dentry *create_buf_file_default_callback(const char *filename,
307 struct dentry *parent,
309 struct rchan_buf *buf,
316 * remove_buf_file() default callback. Does nothing.
318 static int remove_buf_file_default_callback(struct dentry *dentry)
323 /* relay channel default callbacks */
324 static struct rchan_callbacks default_channel_callbacks = {
325 .subbuf_start = subbuf_start_default_callback,
326 .buf_mapped = buf_mapped_default_callback,
327 .buf_unmapped = buf_unmapped_default_callback,
328 .create_buf_file = create_buf_file_default_callback,
329 .remove_buf_file = remove_buf_file_default_callback,
333 * wakeup_readers - wake up readers waiting on a channel
334 * @data: contains the channel buffer
336 * This is the timer function used to defer reader waking.
338 static void wakeup_readers(unsigned long data)
340 struct rchan_buf *buf = (struct rchan_buf *)data;
341 wake_up_interruptible(&buf->read_wait);
343 * Stupid polling for now:
345 mod_timer(&buf->timer, jiffies + 1);
349 * __relay_reset - reset a channel buffer
350 * @buf: the channel buffer
351 * @init: 1 if this is a first-time initialization
353 * See relay_reset() for description of effect.
355 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
360 init_waitqueue_head(&buf->read_wait);
361 kref_init(&buf->kref);
362 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
363 mod_timer(&buf->timer, jiffies + 1);
365 del_timer_sync(&buf->timer);
367 buf->subbufs_produced = 0;
368 buf->subbufs_consumed = 0;
369 buf->bytes_consumed = 0;
371 buf->data = buf->start;
374 for (i = 0; i < buf->chan->n_subbufs; i++)
377 buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
381 * relay_reset - reset the channel
384 * This has the effect of erasing all data from all channel buffers
385 * and restarting the channel in its initial state. The buffers
386 * are not freed, so any mappings are still in effect.
388 * NOTE. Care should be taken that the channel isn't actually
389 * being used by anything when this call is made.
391 void relay_reset(struct rchan *chan)
398 if (chan->is_global && chan->buf[0]) {
399 __relay_reset(chan->buf[0], 0);
403 mutex_lock(&relay_channels_mutex);
404 for_each_possible_cpu(i)
406 __relay_reset(chan->buf[i], 0);
407 mutex_unlock(&relay_channels_mutex);
409 EXPORT_SYMBOL_GPL(relay_reset);
411 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
412 struct dentry *dentry)
414 buf->dentry = dentry;
415 d_inode(buf->dentry)->i_size = buf->early_bytes;
418 static struct dentry *relay_create_buf_file(struct rchan *chan,
419 struct rchan_buf *buf,
422 struct dentry *dentry;
425 tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
428 snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
430 /* Create file in fs */
431 dentry = chan->cb->create_buf_file(tmpname, chan->parent,
441 * relay_open_buf - create a new relay channel buffer
443 * used by relay_open() and CPU hotplug.
445 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
447 struct rchan_buf *buf = NULL;
448 struct dentry *dentry;
453 buf = relay_create_buf(chan);
457 if (chan->has_base_filename) {
458 dentry = relay_create_buf_file(chan, buf, cpu);
461 relay_set_buf_dentry(buf, dentry);
465 __relay_reset(buf, 1);
467 if(chan->is_global) {
475 relay_destroy_buf(buf);
480 * relay_close_buf - close a channel buffer
481 * @buf: channel buffer
483 * Marks the buffer finalized and restores the default callbacks.
484 * The channel buffer and channel buffer data structure are then freed
485 * automatically when the last reference is given up.
487 static void relay_close_buf(struct rchan_buf *buf)
490 del_timer_sync(&buf->timer);
491 buf->chan->cb->remove_buf_file(buf->dentry);
492 kref_put(&buf->kref, relay_remove_buf);
495 static void setup_callbacks(struct rchan *chan,
496 struct rchan_callbacks *cb)
499 chan->cb = &default_channel_callbacks;
503 if (!cb->subbuf_start)
504 cb->subbuf_start = subbuf_start_default_callback;
506 cb->buf_mapped = buf_mapped_default_callback;
507 if (!cb->buf_unmapped)
508 cb->buf_unmapped = buf_unmapped_default_callback;
509 if (!cb->create_buf_file)
510 cb->create_buf_file = create_buf_file_default_callback;
511 if (!cb->remove_buf_file)
512 cb->remove_buf_file = remove_buf_file_default_callback;
517 * relay_hotcpu_callback - CPU hotplug callback
518 * @nb: notifier block
519 * @action: hotplug action to take
522 * Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
524 static int relay_hotcpu_callback(struct notifier_block *nb,
525 unsigned long action,
528 unsigned int hotcpu = (unsigned long)hcpu;
533 case CPU_UP_PREPARE_FROZEN:
534 mutex_lock(&relay_channels_mutex);
535 list_for_each_entry(chan, &relay_channels, list) {
536 if (chan->buf[hotcpu])
538 chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
539 if(!chan->buf[hotcpu]) {
541 "relay_hotcpu_callback: cpu %d buffer "
542 "creation failed\n", hotcpu);
543 mutex_unlock(&relay_channels_mutex);
544 return notifier_from_errno(-ENOMEM);
547 mutex_unlock(&relay_channels_mutex);
550 case CPU_DEAD_FROZEN:
551 /* No need to flush the cpu : will be flushed upon
552 * final relay_flush() call. */
559 * relay_open - create a new relay channel
560 * @base_filename: base name of files to create, %NULL for buffering only
561 * @parent: dentry of parent directory, %NULL for root directory or buffer
562 * @subbuf_size: size of sub-buffers
563 * @n_subbufs: number of sub-buffers
564 * @cb: client callback functions
565 * @private_data: user-defined data
567 * Returns channel pointer if successful, %NULL otherwise.
569 * Creates a channel buffer for each cpu using the sizes and
570 * attributes specified. The created channel buffer files
571 * will be named base_filename0...base_filenameN-1. File
572 * permissions will be %S_IRUSR.
574 struct rchan *relay_open(const char *base_filename,
575 struct dentry *parent,
578 struct rchan_callbacks *cb,
584 if (!(subbuf_size && n_subbufs))
586 if (subbuf_size > UINT_MAX / n_subbufs)
589 chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
593 chan->version = RELAYFS_CHANNEL_VERSION;
594 chan->n_subbufs = n_subbufs;
595 chan->subbuf_size = subbuf_size;
596 chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
597 chan->parent = parent;
598 chan->private_data = private_data;
600 chan->has_base_filename = 1;
601 strlcpy(chan->base_filename, base_filename, NAME_MAX);
603 setup_callbacks(chan, cb);
604 kref_init(&chan->kref);
606 mutex_lock(&relay_channels_mutex);
607 for_each_online_cpu(i) {
608 chan->buf[i] = relay_open_buf(chan, i);
612 list_add(&chan->list, &relay_channels);
613 mutex_unlock(&relay_channels_mutex);
618 for_each_possible_cpu(i) {
620 relay_close_buf(chan->buf[i]);
623 kref_put(&chan->kref, relay_destroy_channel);
624 mutex_unlock(&relay_channels_mutex);
627 EXPORT_SYMBOL_GPL(relay_open);
629 struct rchan_percpu_buf_dispatcher {
630 struct rchan_buf *buf;
631 struct dentry *dentry;
634 /* Called in atomic context. */
635 static void __relay_set_buf_dentry(void *info)
637 struct rchan_percpu_buf_dispatcher *p = info;
639 relay_set_buf_dentry(p->buf, p->dentry);
643 * relay_late_setup_files - triggers file creation
644 * @chan: channel to operate on
645 * @base_filename: base name of files to create
646 * @parent: dentry of parent directory, %NULL for root directory
648 * Returns 0 if successful, non-zero otherwise.
650 * Use to setup files for a previously buffer-only channel.
651 * Useful to do early tracing in kernel, before VFS is up, for example.
653 int relay_late_setup_files(struct rchan *chan,
654 const char *base_filename,
655 struct dentry *parent)
658 unsigned int i, curr_cpu;
660 struct dentry *dentry;
661 struct rchan_percpu_buf_dispatcher disp;
663 if (!chan || !base_filename)
666 strlcpy(chan->base_filename, base_filename, NAME_MAX);
668 mutex_lock(&relay_channels_mutex);
669 /* Is chan already set up? */
670 if (unlikely(chan->has_base_filename)) {
671 mutex_unlock(&relay_channels_mutex);
674 chan->has_base_filename = 1;
675 chan->parent = parent;
676 curr_cpu = get_cpu();
678 * The CPU hotplug notifier ran before us and created buffers with
679 * no files associated. So it's safe to call relay_setup_buf_file()
680 * on all currently online CPUs.
682 for_each_online_cpu(i) {
683 if (unlikely(!chan->buf[i])) {
684 WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
689 dentry = relay_create_buf_file(chan, chan->buf[i], i);
690 if (unlikely(!dentry)) {
696 local_irq_save(flags);
697 relay_set_buf_dentry(chan->buf[i], dentry);
698 local_irq_restore(flags);
700 disp.buf = chan->buf[i];
701 disp.dentry = dentry;
703 /* relay_channels_mutex must be held, so wait. */
704 err = smp_call_function_single(i,
705 __relay_set_buf_dentry,
712 mutex_unlock(&relay_channels_mutex);
718 * relay_switch_subbuf - switch to a new sub-buffer
719 * @buf: channel buffer
720 * @length: size of current event
722 * Returns either the length passed in or 0 if full.
724 * Performs sub-buffer-switch tasks such as invoking callbacks,
725 * updating padding counts, waking up readers, etc.
727 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
730 size_t old_subbuf, new_subbuf;
732 if (unlikely(length > buf->chan->subbuf_size))
735 if (buf->offset != buf->chan->subbuf_size + 1) {
736 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
737 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
738 buf->padding[old_subbuf] = buf->prev_padding;
739 buf->subbufs_produced++;
741 d_inode(buf->dentry)->i_size +=
742 buf->chan->subbuf_size -
743 buf->padding[old_subbuf];
745 buf->early_bytes += buf->chan->subbuf_size -
746 buf->padding[old_subbuf];
750 new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
751 new = buf->start + new_subbuf * buf->chan->subbuf_size;
753 if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
754 buf->offset = buf->chan->subbuf_size + 1;
758 buf->padding[new_subbuf] = 0;
760 if (unlikely(length + buf->offset > buf->chan->subbuf_size))
766 buf->chan->last_toobig = length;
769 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
772 * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
774 * @cpu: the cpu associated with the channel buffer to update
775 * @subbufs_consumed: number of sub-buffers to add to current buf's count
777 * Adds to the channel buffer's consumed sub-buffer count.
778 * subbufs_consumed should be the number of sub-buffers newly consumed,
779 * not the total consumed.
781 * NOTE. Kernel clients don't need to call this function if the channel
782 * mode is 'overwrite'.
784 void relay_subbufs_consumed(struct rchan *chan,
786 size_t subbufs_consumed)
788 struct rchan_buf *buf;
793 if (cpu >= NR_CPUS || !chan->buf[cpu] ||
794 subbufs_consumed > chan->n_subbufs)
797 buf = chan->buf[cpu];
798 if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
799 buf->subbufs_consumed = buf->subbufs_produced;
801 buf->subbufs_consumed += subbufs_consumed;
803 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
806 * relay_close - close the channel
809 * Closes all channel buffers and frees the channel.
811 void relay_close(struct rchan *chan)
818 mutex_lock(&relay_channels_mutex);
819 if (chan->is_global && chan->buf[0])
820 relay_close_buf(chan->buf[0]);
822 for_each_possible_cpu(i)
824 relay_close_buf(chan->buf[i]);
826 if (chan->last_toobig)
827 printk(KERN_WARNING "relay: one or more items not logged "
828 "[item size (%Zd) > sub-buffer size (%Zd)]\n",
829 chan->last_toobig, chan->subbuf_size);
831 list_del(&chan->list);
832 kref_put(&chan->kref, relay_destroy_channel);
833 mutex_unlock(&relay_channels_mutex);
835 EXPORT_SYMBOL_GPL(relay_close);
838 * relay_flush - close the channel
841 * Flushes all channel buffers, i.e. forces buffer switch.
843 void relay_flush(struct rchan *chan)
850 if (chan->is_global && chan->buf[0]) {
851 relay_switch_subbuf(chan->buf[0], 0);
855 mutex_lock(&relay_channels_mutex);
856 for_each_possible_cpu(i)
858 relay_switch_subbuf(chan->buf[i], 0);
859 mutex_unlock(&relay_channels_mutex);
861 EXPORT_SYMBOL_GPL(relay_flush);
864 * relay_file_open - open file op for relay files
868 * Increments the channel buffer refcount.
870 static int relay_file_open(struct inode *inode, struct file *filp)
872 struct rchan_buf *buf = inode->i_private;
873 kref_get(&buf->kref);
874 filp->private_data = buf;
876 return nonseekable_open(inode, filp);
880 * relay_file_mmap - mmap file op for relay files
882 * @vma: the vma describing what to map
884 * Calls upon relay_mmap_buf() to map the file into user space.
886 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
888 struct rchan_buf *buf = filp->private_data;
889 return relay_mmap_buf(buf, vma);
893 * relay_file_poll - poll file op for relay files
899 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
901 unsigned int mask = 0;
902 struct rchan_buf *buf = filp->private_data;
907 if (filp->f_mode & FMODE_READ) {
908 poll_wait(filp, &buf->read_wait, wait);
909 if (!relay_buf_empty(buf))
910 mask |= POLLIN | POLLRDNORM;
917 * relay_file_release - release file op for relay files
921 * Decrements the channel refcount, as the filesystem is
922 * no longer using it.
924 static int relay_file_release(struct inode *inode, struct file *filp)
926 struct rchan_buf *buf = filp->private_data;
927 kref_put(&buf->kref, relay_remove_buf);
933 * relay_file_read_consume - update the consumed count for the buffer
935 static void relay_file_read_consume(struct rchan_buf *buf,
937 size_t bytes_consumed)
939 size_t subbuf_size = buf->chan->subbuf_size;
940 size_t n_subbufs = buf->chan->n_subbufs;
943 if (buf->subbufs_produced == buf->subbufs_consumed &&
944 buf->offset == buf->bytes_consumed)
947 if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
948 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
949 buf->bytes_consumed = 0;
952 buf->bytes_consumed += bytes_consumed;
954 read_subbuf = buf->subbufs_consumed % n_subbufs;
956 read_subbuf = read_pos / buf->chan->subbuf_size;
957 if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
958 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
959 (buf->offset == subbuf_size))
961 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
962 buf->bytes_consumed = 0;
967 * relay_file_read_avail - boolean, are there unconsumed bytes available?
969 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
971 size_t subbuf_size = buf->chan->subbuf_size;
972 size_t n_subbufs = buf->chan->n_subbufs;
973 size_t produced = buf->subbufs_produced;
974 size_t consumed = buf->subbufs_consumed;
976 relay_file_read_consume(buf, read_pos, 0);
978 consumed = buf->subbufs_consumed;
980 if (unlikely(buf->offset > subbuf_size)) {
981 if (produced == consumed)
986 if (unlikely(produced - consumed >= n_subbufs)) {
987 consumed = produced - n_subbufs + 1;
988 buf->subbufs_consumed = consumed;
989 buf->bytes_consumed = 0;
992 produced = (produced % n_subbufs) * subbuf_size + buf->offset;
993 consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
995 if (consumed > produced)
996 produced += n_subbufs * subbuf_size;
998 if (consumed == produced) {
999 if (buf->offset == subbuf_size &&
1000 buf->subbufs_produced > buf->subbufs_consumed)
1009 * relay_file_read_subbuf_avail - return bytes available in sub-buffer
1010 * @read_pos: file read position
1011 * @buf: relay channel buffer
1013 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1014 struct rchan_buf *buf)
1016 size_t padding, avail = 0;
1017 size_t read_subbuf, read_offset, write_subbuf, write_offset;
1018 size_t subbuf_size = buf->chan->subbuf_size;
1020 write_subbuf = (buf->data - buf->start) / subbuf_size;
1021 write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1022 read_subbuf = read_pos / subbuf_size;
1023 read_offset = read_pos % subbuf_size;
1024 padding = buf->padding[read_subbuf];
1026 if (read_subbuf == write_subbuf) {
1027 if (read_offset + padding < write_offset)
1028 avail = write_offset - (read_offset + padding);
1030 avail = (subbuf_size - padding) - read_offset;
1036 * relay_file_read_start_pos - find the first available byte to read
1037 * @read_pos: file read position
1038 * @buf: relay channel buffer
1040 * If the @read_pos is in the middle of padding, return the
1041 * position of the first actually available byte, otherwise
1042 * return the original value.
1044 static size_t relay_file_read_start_pos(size_t read_pos,
1045 struct rchan_buf *buf)
1047 size_t read_subbuf, padding, padding_start, padding_end;
1048 size_t subbuf_size = buf->chan->subbuf_size;
1049 size_t n_subbufs = buf->chan->n_subbufs;
1050 size_t consumed = buf->subbufs_consumed % n_subbufs;
1053 read_pos = consumed * subbuf_size + buf->bytes_consumed;
1054 read_subbuf = read_pos / subbuf_size;
1055 padding = buf->padding[read_subbuf];
1056 padding_start = (read_subbuf + 1) * subbuf_size - padding;
1057 padding_end = (read_subbuf + 1) * subbuf_size;
1058 if (read_pos >= padding_start && read_pos < padding_end) {
1059 read_subbuf = (read_subbuf + 1) % n_subbufs;
1060 read_pos = read_subbuf * subbuf_size;
1067 * relay_file_read_end_pos - return the new read position
1068 * @read_pos: file read position
1069 * @buf: relay channel buffer
1070 * @count: number of bytes to be read
1072 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1076 size_t read_subbuf, padding, end_pos;
1077 size_t subbuf_size = buf->chan->subbuf_size;
1078 size_t n_subbufs = buf->chan->n_subbufs;
1080 read_subbuf = read_pos / subbuf_size;
1081 padding = buf->padding[read_subbuf];
1082 if (read_pos % subbuf_size + count + padding == subbuf_size)
1083 end_pos = (read_subbuf + 1) * subbuf_size;
1085 end_pos = read_pos + count;
1086 if (end_pos >= subbuf_size * n_subbufs)
1093 * subbuf_read_actor - read up to one subbuf's worth of data
1095 static int subbuf_read_actor(size_t read_start,
1096 struct rchan_buf *buf,
1098 read_descriptor_t *desc)
1103 from = buf->start + read_start;
1105 if (copy_to_user(desc->arg.buf, from, avail)) {
1106 desc->error = -EFAULT;
1109 desc->arg.data += ret;
1110 desc->written += ret;
1116 typedef int (*subbuf_actor_t) (size_t read_start,
1117 struct rchan_buf *buf,
1119 read_descriptor_t *desc);
1122 * relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1124 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1125 subbuf_actor_t subbuf_actor,
1126 read_descriptor_t *desc)
1128 struct rchan_buf *buf = filp->private_data;
1129 size_t read_start, avail;
1135 mutex_lock(&file_inode(filp)->i_mutex);
1137 if (!relay_file_read_avail(buf, *ppos))
1140 read_start = relay_file_read_start_pos(*ppos, buf);
1141 avail = relay_file_read_subbuf_avail(read_start, buf);
1145 avail = min(desc->count, avail);
1146 ret = subbuf_actor(read_start, buf, avail, desc);
1147 if (desc->error < 0)
1151 relay_file_read_consume(buf, read_start, ret);
1152 *ppos = relay_file_read_end_pos(buf, read_start, ret);
1154 } while (desc->count && ret);
1155 mutex_unlock(&file_inode(filp)->i_mutex);
1157 return desc->written;
1160 static ssize_t relay_file_read(struct file *filp,
1161 char __user *buffer,
1165 read_descriptor_t desc;
1168 desc.arg.buf = buffer;
1170 return relay_file_read_subbufs(filp, ppos, subbuf_read_actor, &desc);
1173 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1175 rbuf->bytes_consumed += bytes_consumed;
1177 if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1178 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1179 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1183 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1184 struct pipe_buffer *buf)
1186 struct rchan_buf *rbuf;
1188 rbuf = (struct rchan_buf *)page_private(buf->page);
1189 relay_consume_bytes(rbuf, buf->private);
1192 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1194 .confirm = generic_pipe_buf_confirm,
1195 .release = relay_pipe_buf_release,
1196 .steal = generic_pipe_buf_steal,
1197 .get = generic_pipe_buf_get,
1200 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1205 * subbuf_splice_actor - splice up to one subbuf's worth of data
1207 static ssize_t subbuf_splice_actor(struct file *in,
1209 struct pipe_inode_info *pipe,
1214 unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1215 struct rchan_buf *rbuf = in->private_data;
1216 unsigned int subbuf_size = rbuf->chan->subbuf_size;
1217 uint64_t pos = (uint64_t) *ppos;
1218 uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1219 size_t read_start = (size_t) do_div(pos, alloc_size);
1220 size_t read_subbuf = read_start / subbuf_size;
1221 size_t padding = rbuf->padding[read_subbuf];
1222 size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1223 struct page *pages[PIPE_DEF_BUFFERS];
1224 struct partial_page partial[PIPE_DEF_BUFFERS];
1225 struct splice_pipe_desc spd = {
1228 .nr_pages_max = PIPE_DEF_BUFFERS,
1231 .ops = &relay_pipe_buf_ops,
1232 .spd_release = relay_page_release,
1236 if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1238 if (splice_grow_spd(pipe, &spd))
1242 * Adjust read len, if longer than what is available
1244 if (len > (subbuf_size - read_start % subbuf_size))
1245 len = subbuf_size - read_start % subbuf_size;
1247 subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1248 pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1249 poff = read_start & ~PAGE_MASK;
1250 nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1252 for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1253 unsigned int this_len, this_end, private;
1254 unsigned int cur_pos = read_start + total_len;
1259 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1262 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1263 spd.partial[spd.nr_pages].offset = poff;
1265 this_end = cur_pos + this_len;
1266 if (this_end >= nonpad_end) {
1267 this_len = nonpad_end - cur_pos;
1268 private = this_len + padding;
1270 spd.partial[spd.nr_pages].len = this_len;
1271 spd.partial[spd.nr_pages].private = private;
1274 total_len += this_len;
1276 pidx = (pidx + 1) % subbuf_pages;
1278 if (this_end >= nonpad_end) {
1288 ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1289 if (ret < 0 || ret < total_len)
1292 if (read_start + ret == nonpad_end)
1296 splice_shrink_spd(&spd);
1300 static ssize_t relay_file_splice_read(struct file *in,
1302 struct pipe_inode_info *pipe,
1313 while (len && !spliced) {
1314 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1318 if (flags & SPLICE_F_NONBLOCK)
1328 spliced += nonpad_ret;
1338 const struct file_operations relay_file_operations = {
1339 .open = relay_file_open,
1340 .poll = relay_file_poll,
1341 .mmap = relay_file_mmap,
1342 .read = relay_file_read,
1343 .llseek = no_llseek,
1344 .release = relay_file_release,
1345 .splice_read = relay_file_splice_read,
1347 EXPORT_SYMBOL_GPL(relay_file_operations);
1349 static __init int relay_init(void)
1352 hotcpu_notifier(relay_hotcpu_callback, 0);
1356 early_initcall(relay_init);