2 * Public API and common code for kernel->userspace relay file support.
4 * See Documentation/filesystems/relayfs.txt for an overview of relayfs.
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.
11 * This file is released under the GPL.
13 #include <linux/errno.h>
14 #include <linux/stddef.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/string.h>
18 #include <linux/relay.h>
19 #include <linux/vmalloc.h>
23 * close() vm_op implementation for relay file mapping.
25 static void relay_file_mmap_close(struct vm_area_struct *vma)
27 struct rchan_buf *buf = vma->vm_private_data;
28 buf->chan->cb->buf_unmapped(buf, vma->vm_file);
32 * nopage() vm_op implementation for relay file mapping.
34 static struct page *relay_buf_nopage(struct vm_area_struct *vma,
35 unsigned long address,
39 struct rchan_buf *buf = vma->vm_private_data;
40 unsigned long offset = address - vma->vm_start;
42 if (address > vma->vm_end)
43 return NOPAGE_SIGBUS; /* Disallow mremap */
47 page = vmalloc_to_page(buf->start + offset);
53 *type = VM_FAULT_MINOR;
59 * vm_ops for relay file mappings.
61 static struct vm_operations_struct relay_file_mmap_ops = {
62 .nopage = relay_buf_nopage,
63 .close = relay_file_mmap_close,
67 * relay_mmap_buf: - mmap channel buffer to process address space
68 * @buf: relay channel buffer
69 * @vma: vm_area_struct describing memory to be mapped
71 * Returns 0 if ok, negative on error
73 * Caller should already have grabbed mmap_sem.
75 int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
77 unsigned long length = vma->vm_end - vma->vm_start;
78 struct file *filp = vma->vm_file;
83 if (length != (unsigned long)buf->chan->alloc_size)
86 vma->vm_ops = &relay_file_mmap_ops;
87 vma->vm_private_data = buf;
88 buf->chan->cb->buf_mapped(buf, filp);
94 * relay_alloc_buf - allocate a channel buffer
95 * @buf: the buffer struct
96 * @size: total size of the buffer
98 * Returns a pointer to the resulting buffer, NULL if unsuccessful
100 static void *relay_alloc_buf(struct rchan_buf *buf, unsigned long size)
103 unsigned int i, j, n_pages;
105 size = PAGE_ALIGN(size);
106 n_pages = size >> PAGE_SHIFT;
108 buf->page_array = kcalloc(n_pages, sizeof(struct page *), GFP_KERNEL);
109 if (!buf->page_array)
112 for (i = 0; i < n_pages; i++) {
113 buf->page_array[i] = alloc_page(GFP_KERNEL);
114 if (unlikely(!buf->page_array[i]))
117 mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
121 memset(mem, 0, size);
122 buf->page_count = n_pages;
126 for (j = 0; j < i; j++)
127 __free_page(buf->page_array[j]);
128 kfree(buf->page_array);
133 * relay_create_buf - allocate and initialize a channel buffer
134 * @alloc_size: size of the buffer to allocate
135 * @n_subbufs: number of sub-buffers in the channel
137 * Returns channel buffer if successful, NULL otherwise
139 struct rchan_buf *relay_create_buf(struct rchan *chan)
141 struct rchan_buf *buf = kcalloc(1, sizeof(struct rchan_buf), GFP_KERNEL);
145 buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
149 buf->start = relay_alloc_buf(buf, chan->alloc_size);
154 kref_get(&buf->chan->kref);
164 * relay_destroy_channel - free the channel struct
166 * Should only be called from kref_put().
168 void relay_destroy_channel(struct kref *kref)
170 struct rchan *chan = container_of(kref, struct rchan, kref);
175 * relay_destroy_buf - destroy an rchan_buf struct and associated buffer
176 * @buf: the buffer struct
178 void relay_destroy_buf(struct rchan_buf *buf)
180 struct rchan *chan = buf->chan;
183 if (likely(buf->start)) {
185 for (i = 0; i < buf->page_count; i++)
186 __free_page(buf->page_array[i]);
187 kfree(buf->page_array);
191 kref_put(&chan->kref, relay_destroy_channel);
195 * relay_remove_buf - remove a channel buffer
197 * Removes the file from the fileystem, which also frees the
198 * rchan_buf_struct and the channel buffer. Should only be called from
201 void relay_remove_buf(struct kref *kref)
203 struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
204 buf->chan->cb->remove_buf_file(buf->dentry);
205 relay_destroy_buf(buf);
209 * relay_buf_empty - boolean, is the channel buffer empty?
210 * @buf: channel buffer
212 * Returns 1 if the buffer is empty, 0 otherwise.
214 int relay_buf_empty(struct rchan_buf *buf)
216 return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
218 EXPORT_SYMBOL_GPL(relay_buf_empty);
221 * relay_buf_full - boolean, is the channel buffer full?
222 * @buf: channel buffer
224 * Returns 1 if the buffer is full, 0 otherwise.
226 int relay_buf_full(struct rchan_buf *buf)
228 size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
229 return (ready >= buf->chan->n_subbufs) ? 1 : 0;
231 EXPORT_SYMBOL_GPL(relay_buf_full);
234 * High-level relay kernel API and associated functions.
238 * rchan_callback implementations defining default channel behavior. Used
239 * in place of corresponding NULL values in client callback struct.
243 * subbuf_start() default callback. Does nothing.
245 static int subbuf_start_default_callback (struct rchan_buf *buf,
250 if (relay_buf_full(buf))
257 * buf_mapped() default callback. Does nothing.
259 static void buf_mapped_default_callback(struct rchan_buf *buf,
265 * buf_unmapped() default callback. Does nothing.
267 static void buf_unmapped_default_callback(struct rchan_buf *buf,
273 * create_buf_file_create() default callback. Does nothing.
275 static struct dentry *create_buf_file_default_callback(const char *filename,
276 struct dentry *parent,
278 struct rchan_buf *buf,
285 * remove_buf_file() default callback. Does nothing.
287 static int remove_buf_file_default_callback(struct dentry *dentry)
292 /* relay channel default callbacks */
293 static struct rchan_callbacks default_channel_callbacks = {
294 .subbuf_start = subbuf_start_default_callback,
295 .buf_mapped = buf_mapped_default_callback,
296 .buf_unmapped = buf_unmapped_default_callback,
297 .create_buf_file = create_buf_file_default_callback,
298 .remove_buf_file = remove_buf_file_default_callback,
302 * wakeup_readers - wake up readers waiting on a channel
303 * @private: the channel buffer
305 * This is the work function used to defer reader waking. The
306 * reason waking is deferred is that calling directly from write
307 * causes problems if you're writing from say the scheduler.
309 static void wakeup_readers(void *private)
311 struct rchan_buf *buf = private;
312 wake_up_interruptible(&buf->read_wait);
316 * __relay_reset - reset a channel buffer
317 * @buf: the channel buffer
318 * @init: 1 if this is a first-time initialization
320 * See relay_reset for description of effect.
322 static inline void __relay_reset(struct rchan_buf *buf, unsigned int init)
327 init_waitqueue_head(&buf->read_wait);
328 kref_init(&buf->kref);
329 INIT_WORK(&buf->wake_readers, NULL, NULL);
331 cancel_delayed_work(&buf->wake_readers);
332 flush_scheduled_work();
335 buf->subbufs_produced = 0;
336 buf->subbufs_consumed = 0;
337 buf->bytes_consumed = 0;
339 buf->data = buf->start;
342 for (i = 0; i < buf->chan->n_subbufs; i++)
345 buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
349 * relay_reset - reset the channel
352 * This has the effect of erasing all data from all channel buffers
353 * and restarting the channel in its initial state. The buffers
354 * are not freed, so any mappings are still in effect.
356 * NOTE: Care should be taken that the channel isn't actually
357 * being used by anything when this call is made.
359 void relay_reset(struct rchan *chan)
362 struct rchan_buf *prev = NULL;
367 for (i = 0; i < NR_CPUS; i++) {
368 if (!chan->buf[i] || chan->buf[i] == prev)
370 __relay_reset(chan->buf[i], 0);
374 EXPORT_SYMBOL_GPL(relay_reset);
377 * relay_open_buf - create a new relay channel buffer
379 * Internal - used by relay_open().
381 static struct rchan_buf *relay_open_buf(struct rchan *chan,
382 const char *filename,
383 struct dentry *parent,
386 struct rchan_buf *buf;
387 struct dentry *dentry;
392 buf = relay_create_buf(chan);
396 /* Create file in fs */
397 dentry = chan->cb->create_buf_file(filename, parent, S_IRUSR,
400 relay_destroy_buf(buf);
404 buf->dentry = dentry;
405 __relay_reset(buf, 1);
411 * relay_close_buf - close a channel buffer
412 * @buf: channel buffer
414 * Marks the buffer finalized and restores the default callbacks.
415 * The channel buffer and channel buffer data structure are then freed
416 * automatically when the last reference is given up.
418 static inline void relay_close_buf(struct rchan_buf *buf)
421 cancel_delayed_work(&buf->wake_readers);
422 flush_scheduled_work();
423 kref_put(&buf->kref, relay_remove_buf);
426 static inline void setup_callbacks(struct rchan *chan,
427 struct rchan_callbacks *cb)
430 chan->cb = &default_channel_callbacks;
434 if (!cb->subbuf_start)
435 cb->subbuf_start = subbuf_start_default_callback;
437 cb->buf_mapped = buf_mapped_default_callback;
438 if (!cb->buf_unmapped)
439 cb->buf_unmapped = buf_unmapped_default_callback;
440 if (!cb->create_buf_file)
441 cb->create_buf_file = create_buf_file_default_callback;
442 if (!cb->remove_buf_file)
443 cb->remove_buf_file = remove_buf_file_default_callback;
448 * relay_open - create a new relay channel
449 * @base_filename: base name of files to create
450 * @parent: dentry of parent directory, NULL for root directory
451 * @subbuf_size: size of sub-buffers
452 * @n_subbufs: number of sub-buffers
453 * @cb: client callback functions
455 * Returns channel pointer if successful, NULL otherwise.
457 * Creates a channel buffer for each cpu using the sizes and
458 * attributes specified. The created channel buffer files
459 * will be named base_filename0...base_filenameN-1. File
460 * permissions will be S_IRUSR.
462 struct rchan *relay_open(const char *base_filename,
463 struct dentry *parent,
466 struct rchan_callbacks *cb)
476 if (!(subbuf_size && n_subbufs))
479 chan = kcalloc(1, sizeof(struct rchan), GFP_KERNEL);
483 chan->version = RELAYFS_CHANNEL_VERSION;
484 chan->n_subbufs = n_subbufs;
485 chan->subbuf_size = subbuf_size;
486 chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
487 setup_callbacks(chan, cb);
488 kref_init(&chan->kref);
490 tmpname = kmalloc(NAME_MAX + 1, GFP_KERNEL);
494 for_each_online_cpu(i) {
495 sprintf(tmpname, "%s%d", base_filename, i);
496 chan->buf[i] = relay_open_buf(chan, tmpname, parent,
501 chan->buf[i]->cpu = i;
508 for (i = 0; i < NR_CPUS; i++) {
511 relay_close_buf(chan->buf[i]);
518 kref_put(&chan->kref, relay_destroy_channel);
521 EXPORT_SYMBOL_GPL(relay_open);
524 * relay_switch_subbuf - switch to a new sub-buffer
525 * @buf: channel buffer
526 * @length: size of current event
528 * Returns either the length passed in or 0 if full.
530 * Performs sub-buffer-switch tasks such as invoking callbacks,
531 * updating padding counts, waking up readers, etc.
533 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
536 size_t old_subbuf, new_subbuf;
538 if (unlikely(length > buf->chan->subbuf_size))
541 if (buf->offset != buf->chan->subbuf_size + 1) {
542 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
543 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
544 buf->padding[old_subbuf] = buf->prev_padding;
545 buf->subbufs_produced++;
546 if (waitqueue_active(&buf->read_wait)) {
547 PREPARE_WORK(&buf->wake_readers, wakeup_readers, buf);
548 schedule_delayed_work(&buf->wake_readers, 1);
553 new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
554 new = buf->start + new_subbuf * buf->chan->subbuf_size;
556 if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
557 buf->offset = buf->chan->subbuf_size + 1;
561 buf->padding[new_subbuf] = 0;
563 if (unlikely(length + buf->offset > buf->chan->subbuf_size))
569 buf->chan->last_toobig = length;
572 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
575 * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
577 * @cpu: the cpu associated with the channel buffer to update
578 * @subbufs_consumed: number of sub-buffers to add to current buf's count
580 * Adds to the channel buffer's consumed sub-buffer count.
581 * subbufs_consumed should be the number of sub-buffers newly consumed,
582 * not the total consumed.
584 * NOTE: kernel clients don't need to call this function if the channel
585 * mode is 'overwrite'.
587 void relay_subbufs_consumed(struct rchan *chan,
589 size_t subbufs_consumed)
591 struct rchan_buf *buf;
596 if (cpu >= NR_CPUS || !chan->buf[cpu])
599 buf = chan->buf[cpu];
600 buf->subbufs_consumed += subbufs_consumed;
601 if (buf->subbufs_consumed > buf->subbufs_produced)
602 buf->subbufs_consumed = buf->subbufs_produced;
604 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
607 * relay_close - close the channel
610 * Closes all channel buffers and frees the channel.
612 void relay_close(struct rchan *chan)
615 struct rchan_buf *prev = NULL;
620 for (i = 0; i < NR_CPUS; i++) {
621 if (!chan->buf[i] || chan->buf[i] == prev)
623 relay_close_buf(chan->buf[i]);
627 if (chan->last_toobig)
628 printk(KERN_WARNING "relay: one or more items not logged "
629 "[item size (%Zd) > sub-buffer size (%Zd)]\n",
630 chan->last_toobig, chan->subbuf_size);
632 kref_put(&chan->kref, relay_destroy_channel);
634 EXPORT_SYMBOL_GPL(relay_close);
637 * relay_flush - close the channel
640 * Flushes all channel buffers i.e. forces buffer switch.
642 void relay_flush(struct rchan *chan)
645 struct rchan_buf *prev = NULL;
650 for (i = 0; i < NR_CPUS; i++) {
651 if (!chan->buf[i] || chan->buf[i] == prev)
653 relay_switch_subbuf(chan->buf[i], 0);
657 EXPORT_SYMBOL_GPL(relay_flush);
660 * relay_file_open - open file op for relay files
664 * Increments the channel buffer refcount.
666 static int relay_file_open(struct inode *inode, struct file *filp)
668 struct rchan_buf *buf = inode->u.generic_ip;
669 kref_get(&buf->kref);
670 filp->private_data = buf;
676 * relay_file_mmap - mmap file op for relay files
678 * @vma: the vma describing what to map
680 * Calls upon relay_mmap_buf to map the file into user space.
682 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
684 struct rchan_buf *buf = filp->private_data;
685 return relay_mmap_buf(buf, vma);
689 * relay_file_poll - poll file op for relay files
695 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
697 unsigned int mask = 0;
698 struct rchan_buf *buf = filp->private_data;
703 if (filp->f_mode & FMODE_READ) {
704 poll_wait(filp, &buf->read_wait, wait);
705 if (!relay_buf_empty(buf))
706 mask |= POLLIN | POLLRDNORM;
713 * relay_file_release - release file op for relay files
717 * Decrements the channel refcount, as the filesystem is
718 * no longer using it.
720 static int relay_file_release(struct inode *inode, struct file *filp)
722 struct rchan_buf *buf = filp->private_data;
723 kref_put(&buf->kref, relay_remove_buf);
729 * relay_file_read_consume - update the consumed count for the buffer
731 static void relay_file_read_consume(struct rchan_buf *buf,
733 size_t bytes_consumed)
735 size_t subbuf_size = buf->chan->subbuf_size;
736 size_t n_subbufs = buf->chan->n_subbufs;
739 if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
740 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
741 buf->bytes_consumed = 0;
744 buf->bytes_consumed += bytes_consumed;
745 read_subbuf = read_pos / buf->chan->subbuf_size;
746 if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
747 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
748 (buf->offset == subbuf_size))
750 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
751 buf->bytes_consumed = 0;
756 * relay_file_read_avail - boolean, are there unconsumed bytes available?
758 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
760 size_t bytes_produced, bytes_consumed, write_offset;
761 size_t subbuf_size = buf->chan->subbuf_size;
762 size_t n_subbufs = buf->chan->n_subbufs;
763 size_t produced = buf->subbufs_produced % n_subbufs;
764 size_t consumed = buf->subbufs_consumed % n_subbufs;
766 write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
768 if (consumed > produced) {
769 if ((produced > n_subbufs) &&
770 (produced + n_subbufs - consumed <= n_subbufs))
771 produced += n_subbufs;
772 } else if (consumed == produced) {
773 if (buf->offset > subbuf_size) {
774 produced += n_subbufs;
775 if (buf->subbufs_produced == buf->subbufs_consumed)
776 consumed += n_subbufs;
780 if (buf->offset > subbuf_size)
781 bytes_produced = (produced - 1) * subbuf_size + write_offset;
783 bytes_produced = produced * subbuf_size + write_offset;
784 bytes_consumed = consumed * subbuf_size + buf->bytes_consumed;
786 if (bytes_produced == bytes_consumed)
789 relay_file_read_consume(buf, read_pos, 0);
795 * relay_file_read_subbuf_avail - return bytes available in sub-buffer
797 static size_t relay_file_read_subbuf_avail(size_t read_pos,
798 struct rchan_buf *buf)
800 size_t padding, avail = 0;
801 size_t read_subbuf, read_offset, write_subbuf, write_offset;
802 size_t subbuf_size = buf->chan->subbuf_size;
804 write_subbuf = (buf->data - buf->start) / subbuf_size;
805 write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
806 read_subbuf = read_pos / subbuf_size;
807 read_offset = read_pos % subbuf_size;
808 padding = buf->padding[read_subbuf];
810 if (read_subbuf == write_subbuf) {
811 if (read_offset + padding < write_offset)
812 avail = write_offset - (read_offset + padding);
814 avail = (subbuf_size - padding) - read_offset;
820 * relay_file_read_start_pos - find the first available byte to read
822 * If the read_pos is in the middle of padding, return the
823 * position of the first actually available byte, otherwise
824 * return the original value.
826 static size_t relay_file_read_start_pos(size_t read_pos,
827 struct rchan_buf *buf)
829 size_t read_subbuf, padding, padding_start, padding_end;
830 size_t subbuf_size = buf->chan->subbuf_size;
831 size_t n_subbufs = buf->chan->n_subbufs;
833 read_subbuf = read_pos / subbuf_size;
834 padding = buf->padding[read_subbuf];
835 padding_start = (read_subbuf + 1) * subbuf_size - padding;
836 padding_end = (read_subbuf + 1) * subbuf_size;
837 if (read_pos >= padding_start && read_pos < padding_end) {
838 read_subbuf = (read_subbuf + 1) % n_subbufs;
839 read_pos = read_subbuf * subbuf_size;
846 * relay_file_read_end_pos - return the new read position
848 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
852 size_t read_subbuf, padding, end_pos;
853 size_t subbuf_size = buf->chan->subbuf_size;
854 size_t n_subbufs = buf->chan->n_subbufs;
856 read_subbuf = read_pos / subbuf_size;
857 padding = buf->padding[read_subbuf];
858 if (read_pos % subbuf_size + count + padding == subbuf_size)
859 end_pos = (read_subbuf + 1) * subbuf_size;
861 end_pos = read_pos + count;
862 if (end_pos >= subbuf_size * n_subbufs)
869 * relay_file_read - read file op for relay files
871 * @buffer: the userspace buffer
872 * @count: number of bytes to read
873 * @ppos: position to read from
875 * Reads count bytes or the number of bytes available in the
876 * current sub-buffer being read, whichever is smaller.
878 static ssize_t relay_file_read(struct file *filp,
883 struct rchan_buf *buf = filp->private_data;
884 struct inode *inode = filp->f_dentry->d_inode;
885 size_t read_start, avail;
889 mutex_lock(&inode->i_mutex);
890 if(!relay_file_read_avail(buf, *ppos))
893 read_start = relay_file_read_start_pos(*ppos, buf);
894 avail = relay_file_read_subbuf_avail(read_start, buf);
898 from = buf->start + read_start;
899 ret = count = min(count, avail);
900 if (copy_to_user(buffer, from, count)) {
904 relay_file_read_consume(buf, read_start, count);
905 *ppos = relay_file_read_end_pos(buf, read_start, count);
907 mutex_unlock(&inode->i_mutex);
911 struct file_operations relay_file_operations = {
912 .open = relay_file_open,
913 .poll = relay_file_poll,
914 .mmap = relay_file_mmap,
915 .read = relay_file_read,
917 .release = relay_file_release,
919 EXPORT_SYMBOL_GPL(relay_file_operations);