v2.6.22.24-op1
[linux-2.6.22.y-op.git] / kernel / relay.c
blob24db7e81cde4e6b235af935bf7cfdb8d801e7a21
1 /*
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.
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/module.h>
19 #include <linux/string.h>
20 #include <linux/relay.h>
21 #include <linux/vmalloc.h>
22 #include <linux/mm.h>
23 #include <linux/cpu.h>
25 /* list of open channels, for cpu hotplug */
26 static DEFINE_MUTEX(relay_channels_mutex);
27 static LIST_HEAD(relay_channels);
30 * close() vm_op implementation for relay file mapping.
32 static void relay_file_mmap_close(struct vm_area_struct *vma)
34 struct rchan_buf *buf = vma->vm_private_data;
35 buf->chan->cb->buf_unmapped(buf, vma->vm_file);
39 * nopage() vm_op implementation for relay file mapping.
41 static struct page *relay_buf_nopage(struct vm_area_struct *vma,
42 unsigned long address,
43 int *type)
45 struct page *page;
46 struct rchan_buf *buf = vma->vm_private_data;
47 unsigned long offset = address - vma->vm_start;
49 if (address > vma->vm_end)
50 return NOPAGE_SIGBUS; /* Disallow mremap */
51 if (!buf)
52 return NOPAGE_OOM;
54 page = vmalloc_to_page(buf->start + offset);
55 if (!page)
56 return NOPAGE_OOM;
57 get_page(page);
59 if (type)
60 *type = VM_FAULT_MINOR;
62 return page;
66 * vm_ops for relay file mappings.
68 static struct vm_operations_struct relay_file_mmap_ops = {
69 .nopage = relay_buf_nopage,
70 .close = relay_file_mmap_close,
73 /**
74 * relay_mmap_buf: - mmap channel buffer to process address space
75 * @buf: relay channel buffer
76 * @vma: vm_area_struct describing memory to be mapped
78 * Returns 0 if ok, negative on error
80 * Caller should already have grabbed mmap_sem.
82 int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
84 unsigned long length = vma->vm_end - vma->vm_start;
85 struct file *filp = vma->vm_file;
87 if (!buf)
88 return -EBADF;
90 if (length != (unsigned long)buf->chan->alloc_size)
91 return -EINVAL;
93 vma->vm_ops = &relay_file_mmap_ops;
94 vma->vm_flags |= VM_DONTEXPAND;
95 vma->vm_private_data = buf;
96 buf->chan->cb->buf_mapped(buf, filp);
98 return 0;
102 * relay_alloc_buf - allocate a channel buffer
103 * @buf: the buffer struct
104 * @size: total size of the buffer
106 * Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
107 * passed in size will get page aligned, if it isn't already.
109 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
111 void *mem;
112 unsigned int i, j, n_pages;
114 *size = PAGE_ALIGN(*size);
115 n_pages = *size >> PAGE_SHIFT;
117 buf->page_array = kcalloc(n_pages, sizeof(struct page *), GFP_KERNEL);
118 if (!buf->page_array)
119 return NULL;
121 for (i = 0; i < n_pages; i++) {
122 buf->page_array[i] = alloc_page(GFP_KERNEL);
123 if (unlikely(!buf->page_array[i]))
124 goto depopulate;
126 mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
127 if (!mem)
128 goto depopulate;
130 memset(mem, 0, *size);
131 buf->page_count = n_pages;
132 return mem;
134 depopulate:
135 for (j = 0; j < i; j++)
136 __free_page(buf->page_array[j]);
137 kfree(buf->page_array);
138 return NULL;
142 * relay_create_buf - allocate and initialize a channel buffer
143 * @chan: the relay channel
145 * Returns channel buffer if successful, %NULL otherwise.
147 struct rchan_buf *relay_create_buf(struct rchan *chan)
149 struct rchan_buf *buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
150 if (!buf)
151 return NULL;
153 buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
154 if (!buf->padding)
155 goto free_buf;
157 buf->start = relay_alloc_buf(buf, &chan->alloc_size);
158 if (!buf->start)
159 goto free_buf;
161 buf->chan = chan;
162 kref_get(&buf->chan->kref);
163 return buf;
165 free_buf:
166 kfree(buf->padding);
167 kfree(buf);
168 return NULL;
172 * relay_destroy_channel - free the channel struct
173 * @kref: target kernel reference that contains the relay channel
175 * Should only be called from kref_put().
177 void relay_destroy_channel(struct kref *kref)
179 struct rchan *chan = container_of(kref, struct rchan, kref);
180 kfree(chan);
184 * relay_destroy_buf - destroy an rchan_buf struct and associated buffer
185 * @buf: the buffer struct
187 void relay_destroy_buf(struct rchan_buf *buf)
189 struct rchan *chan = buf->chan;
190 unsigned int i;
192 if (likely(buf->start)) {
193 vunmap(buf->start);
194 for (i = 0; i < buf->page_count; i++)
195 __free_page(buf->page_array[i]);
196 kfree(buf->page_array);
198 chan->buf[buf->cpu] = NULL;
199 kfree(buf->padding);
200 kfree(buf);
201 kref_put(&chan->kref, relay_destroy_channel);
205 * relay_remove_buf - remove a channel buffer
206 * @kref: target kernel reference that contains the relay buffer
208 * Removes the file from the fileystem, which also frees the
209 * rchan_buf_struct and the channel buffer. Should only be called from
210 * kref_put().
212 void relay_remove_buf(struct kref *kref)
214 struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
215 buf->chan->cb->remove_buf_file(buf->dentry);
216 relay_destroy_buf(buf);
220 * relay_buf_empty - boolean, is the channel buffer empty?
221 * @buf: channel buffer
223 * Returns 1 if the buffer is empty, 0 otherwise.
225 int relay_buf_empty(struct rchan_buf *buf)
227 return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
229 EXPORT_SYMBOL_GPL(relay_buf_empty);
232 * relay_buf_full - boolean, is the channel buffer full?
233 * @buf: channel buffer
235 * Returns 1 if the buffer is full, 0 otherwise.
237 int relay_buf_full(struct rchan_buf *buf)
239 size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
240 return (ready >= buf->chan->n_subbufs) ? 1 : 0;
242 EXPORT_SYMBOL_GPL(relay_buf_full);
245 * High-level relay kernel API and associated functions.
249 * rchan_callback implementations defining default channel behavior. Used
250 * in place of corresponding NULL values in client callback struct.
254 * subbuf_start() default callback. Does nothing.
256 static int subbuf_start_default_callback (struct rchan_buf *buf,
257 void *subbuf,
258 void *prev_subbuf,
259 size_t prev_padding)
261 if (relay_buf_full(buf))
262 return 0;
264 return 1;
268 * buf_mapped() default callback. Does nothing.
270 static void buf_mapped_default_callback(struct rchan_buf *buf,
271 struct file *filp)
276 * buf_unmapped() default callback. Does nothing.
278 static void buf_unmapped_default_callback(struct rchan_buf *buf,
279 struct file *filp)
284 * create_buf_file_create() default callback. Does nothing.
286 static struct dentry *create_buf_file_default_callback(const char *filename,
287 struct dentry *parent,
288 int mode,
289 struct rchan_buf *buf,
290 int *is_global)
292 return NULL;
296 * remove_buf_file() default callback. Does nothing.
298 static int remove_buf_file_default_callback(struct dentry *dentry)
300 return -EINVAL;
303 /* relay channel default callbacks */
304 static struct rchan_callbacks default_channel_callbacks = {
305 .subbuf_start = subbuf_start_default_callback,
306 .buf_mapped = buf_mapped_default_callback,
307 .buf_unmapped = buf_unmapped_default_callback,
308 .create_buf_file = create_buf_file_default_callback,
309 .remove_buf_file = remove_buf_file_default_callback,
313 * wakeup_readers - wake up readers waiting on a channel
314 * @data: contains the channel buffer
316 * This is the timer function used to defer reader waking.
318 static void wakeup_readers(unsigned long data)
320 struct rchan_buf *buf = (struct rchan_buf *)data;
321 wake_up_interruptible(&buf->read_wait);
325 * __relay_reset - reset a channel buffer
326 * @buf: the channel buffer
327 * @init: 1 if this is a first-time initialization
329 * See relay_reset() for description of effect.
331 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
333 size_t i;
335 if (init) {
336 init_waitqueue_head(&buf->read_wait);
337 kref_init(&buf->kref);
338 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
339 } else
340 del_timer_sync(&buf->timer);
342 buf->subbufs_produced = 0;
343 buf->subbufs_consumed = 0;
344 buf->bytes_consumed = 0;
345 buf->finalized = 0;
346 buf->data = buf->start;
347 buf->offset = 0;
349 for (i = 0; i < buf->chan->n_subbufs; i++)
350 buf->padding[i] = 0;
352 buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
356 * relay_reset - reset the channel
357 * @chan: the channel
359 * This has the effect of erasing all data from all channel buffers
360 * and restarting the channel in its initial state. The buffers
361 * are not freed, so any mappings are still in effect.
363 * NOTE. Care should be taken that the channel isn't actually
364 * being used by anything when this call is made.
366 void relay_reset(struct rchan *chan)
368 unsigned int i;
370 if (!chan)
371 return;
373 if (chan->is_global && chan->buf[0]) {
374 __relay_reset(chan->buf[0], 0);
375 return;
378 mutex_lock(&relay_channels_mutex);
379 for_each_online_cpu(i)
380 if (chan->buf[i])
381 __relay_reset(chan->buf[i], 0);
382 mutex_unlock(&relay_channels_mutex);
384 EXPORT_SYMBOL_GPL(relay_reset);
387 * relay_open_buf - create a new relay channel buffer
389 * used by relay_open() and CPU hotplug.
391 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
393 struct rchan_buf *buf = NULL;
394 struct dentry *dentry;
395 char *tmpname;
397 if (chan->is_global)
398 return chan->buf[0];
400 tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
401 if (!tmpname)
402 goto end;
403 snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
405 buf = relay_create_buf(chan);
406 if (!buf)
407 goto free_name;
409 buf->cpu = cpu;
410 __relay_reset(buf, 1);
412 /* Create file in fs */
413 dentry = chan->cb->create_buf_file(tmpname, chan->parent, S_IRUSR,
414 buf, &chan->is_global);
415 if (!dentry)
416 goto free_buf;
418 buf->dentry = dentry;
420 if(chan->is_global) {
421 chan->buf[0] = buf;
422 buf->cpu = 0;
425 goto free_name;
427 free_buf:
428 relay_destroy_buf(buf);
429 free_name:
430 kfree(tmpname);
431 end:
432 return buf;
436 * relay_close_buf - close a channel buffer
437 * @buf: channel buffer
439 * Marks the buffer finalized and restores the default callbacks.
440 * The channel buffer and channel buffer data structure are then freed
441 * automatically when the last reference is given up.
443 static void relay_close_buf(struct rchan_buf *buf)
445 buf->finalized = 1;
446 del_timer_sync(&buf->timer);
447 kref_put(&buf->kref, relay_remove_buf);
450 static void setup_callbacks(struct rchan *chan,
451 struct rchan_callbacks *cb)
453 if (!cb) {
454 chan->cb = &default_channel_callbacks;
455 return;
458 if (!cb->subbuf_start)
459 cb->subbuf_start = subbuf_start_default_callback;
460 if (!cb->buf_mapped)
461 cb->buf_mapped = buf_mapped_default_callback;
462 if (!cb->buf_unmapped)
463 cb->buf_unmapped = buf_unmapped_default_callback;
464 if (!cb->create_buf_file)
465 cb->create_buf_file = create_buf_file_default_callback;
466 if (!cb->remove_buf_file)
467 cb->remove_buf_file = remove_buf_file_default_callback;
468 chan->cb = cb;
472 * relay_hotcpu_callback - CPU hotplug callback
473 * @nb: notifier block
474 * @action: hotplug action to take
475 * @hcpu: CPU number
477 * Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
479 static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb,
480 unsigned long action,
481 void *hcpu)
483 unsigned int hotcpu = (unsigned long)hcpu;
484 struct rchan *chan;
486 switch(action) {
487 case CPU_UP_PREPARE:
488 case CPU_UP_PREPARE_FROZEN:
489 mutex_lock(&relay_channels_mutex);
490 list_for_each_entry(chan, &relay_channels, list) {
491 if (chan->buf[hotcpu])
492 continue;
493 chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
494 if(!chan->buf[hotcpu]) {
495 printk(KERN_ERR
496 "relay_hotcpu_callback: cpu %d buffer "
497 "creation failed\n", hotcpu);
498 mutex_unlock(&relay_channels_mutex);
499 return NOTIFY_BAD;
502 mutex_unlock(&relay_channels_mutex);
503 break;
504 case CPU_DEAD:
505 case CPU_DEAD_FROZEN:
506 /* No need to flush the cpu : will be flushed upon
507 * final relay_flush() call. */
508 break;
510 return NOTIFY_OK;
514 * relay_open - create a new relay channel
515 * @base_filename: base name of files to create
516 * @parent: dentry of parent directory, %NULL for root directory
517 * @subbuf_size: size of sub-buffers
518 * @n_subbufs: number of sub-buffers
519 * @cb: client callback functions
520 * @private_data: user-defined data
522 * Returns channel pointer if successful, %NULL otherwise.
524 * Creates a channel buffer for each cpu using the sizes and
525 * attributes specified. The created channel buffer files
526 * will be named base_filename0...base_filenameN-1. File
527 * permissions will be %S_IRUSR.
529 struct rchan *relay_open(const char *base_filename,
530 struct dentry *parent,
531 size_t subbuf_size,
532 size_t n_subbufs,
533 struct rchan_callbacks *cb,
534 void *private_data)
536 unsigned int i;
537 struct rchan *chan;
538 if (!base_filename)
539 return NULL;
541 if (!(subbuf_size && n_subbufs))
542 return NULL;
544 chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
545 if (!chan)
546 return NULL;
548 chan->version = RELAYFS_CHANNEL_VERSION;
549 chan->n_subbufs = n_subbufs;
550 chan->subbuf_size = subbuf_size;
551 chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
552 chan->parent = parent;
553 chan->private_data = private_data;
554 strlcpy(chan->base_filename, base_filename, NAME_MAX);
555 setup_callbacks(chan, cb);
556 kref_init(&chan->kref);
558 mutex_lock(&relay_channels_mutex);
559 for_each_online_cpu(i) {
560 chan->buf[i] = relay_open_buf(chan, i);
561 if (!chan->buf[i])
562 goto free_bufs;
564 list_add(&chan->list, &relay_channels);
565 mutex_unlock(&relay_channels_mutex);
567 return chan;
569 free_bufs:
570 for_each_online_cpu(i) {
571 if (!chan->buf[i])
572 break;
573 relay_close_buf(chan->buf[i]);
576 kref_put(&chan->kref, relay_destroy_channel);
577 mutex_unlock(&relay_channels_mutex);
578 return NULL;
580 EXPORT_SYMBOL_GPL(relay_open);
583 * relay_switch_subbuf - switch to a new sub-buffer
584 * @buf: channel buffer
585 * @length: size of current event
587 * Returns either the length passed in or 0 if full.
589 * Performs sub-buffer-switch tasks such as invoking callbacks,
590 * updating padding counts, waking up readers, etc.
592 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
594 void *old, *new;
595 size_t old_subbuf, new_subbuf;
597 if (unlikely(length > buf->chan->subbuf_size))
598 goto toobig;
600 if (buf->offset != buf->chan->subbuf_size + 1) {
601 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
602 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
603 buf->padding[old_subbuf] = buf->prev_padding;
604 buf->subbufs_produced++;
605 buf->dentry->d_inode->i_size += buf->chan->subbuf_size -
606 buf->padding[old_subbuf];
607 smp_mb();
608 if (waitqueue_active(&buf->read_wait))
610 * Calling wake_up_interruptible() from here
611 * will deadlock if we happen to be logging
612 * from the scheduler (trying to re-grab
613 * rq->lock), so defer it.
615 __mod_timer(&buf->timer, jiffies + 1);
618 old = buf->data;
619 new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
620 new = buf->start + new_subbuf * buf->chan->subbuf_size;
621 buf->offset = 0;
622 if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
623 buf->offset = buf->chan->subbuf_size + 1;
624 return 0;
626 buf->data = new;
627 buf->padding[new_subbuf] = 0;
629 if (unlikely(length + buf->offset > buf->chan->subbuf_size))
630 goto toobig;
632 return length;
634 toobig:
635 buf->chan->last_toobig = length;
636 return 0;
638 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
641 * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
642 * @chan: the channel
643 * @cpu: the cpu associated with the channel buffer to update
644 * @subbufs_consumed: number of sub-buffers to add to current buf's count
646 * Adds to the channel buffer's consumed sub-buffer count.
647 * subbufs_consumed should be the number of sub-buffers newly consumed,
648 * not the total consumed.
650 * NOTE. Kernel clients don't need to call this function if the channel
651 * mode is 'overwrite'.
653 void relay_subbufs_consumed(struct rchan *chan,
654 unsigned int cpu,
655 size_t subbufs_consumed)
657 struct rchan_buf *buf;
659 if (!chan)
660 return;
662 if (cpu >= NR_CPUS || !chan->buf[cpu])
663 return;
665 buf = chan->buf[cpu];
666 buf->subbufs_consumed += subbufs_consumed;
667 if (buf->subbufs_consumed > buf->subbufs_produced)
668 buf->subbufs_consumed = buf->subbufs_produced;
670 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
673 * relay_close - close the channel
674 * @chan: the channel
676 * Closes all channel buffers and frees the channel.
678 void relay_close(struct rchan *chan)
680 unsigned int i;
682 if (!chan)
683 return;
685 mutex_lock(&relay_channels_mutex);
686 if (chan->is_global && chan->buf[0])
687 relay_close_buf(chan->buf[0]);
688 else
689 for_each_possible_cpu(i)
690 if (chan->buf[i])
691 relay_close_buf(chan->buf[i]);
693 if (chan->last_toobig)
694 printk(KERN_WARNING "relay: one or more items not logged "
695 "[item size (%Zd) > sub-buffer size (%Zd)]\n",
696 chan->last_toobig, chan->subbuf_size);
698 list_del(&chan->list);
699 kref_put(&chan->kref, relay_destroy_channel);
700 mutex_unlock(&relay_channels_mutex);
702 EXPORT_SYMBOL_GPL(relay_close);
705 * relay_flush - close the channel
706 * @chan: the channel
708 * Flushes all channel buffers, i.e. forces buffer switch.
710 void relay_flush(struct rchan *chan)
712 unsigned int i;
714 if (!chan)
715 return;
717 if (chan->is_global && chan->buf[0]) {
718 relay_switch_subbuf(chan->buf[0], 0);
719 return;
722 mutex_lock(&relay_channels_mutex);
723 for_each_possible_cpu(i)
724 if (chan->buf[i])
725 relay_switch_subbuf(chan->buf[i], 0);
726 mutex_unlock(&relay_channels_mutex);
728 EXPORT_SYMBOL_GPL(relay_flush);
731 * relay_file_open - open file op for relay files
732 * @inode: the inode
733 * @filp: the file
735 * Increments the channel buffer refcount.
737 static int relay_file_open(struct inode *inode, struct file *filp)
739 struct rchan_buf *buf = inode->i_private;
740 kref_get(&buf->kref);
741 filp->private_data = buf;
743 return 0;
747 * relay_file_mmap - mmap file op for relay files
748 * @filp: the file
749 * @vma: the vma describing what to map
751 * Calls upon relay_mmap_buf() to map the file into user space.
753 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
755 struct rchan_buf *buf = filp->private_data;
756 return relay_mmap_buf(buf, vma);
760 * relay_file_poll - poll file op for relay files
761 * @filp: the file
762 * @wait: poll table
764 * Poll implemention.
766 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
768 unsigned int mask = 0;
769 struct rchan_buf *buf = filp->private_data;
771 if (buf->finalized)
772 return POLLERR;
774 if (filp->f_mode & FMODE_READ) {
775 poll_wait(filp, &buf->read_wait, wait);
776 if (!relay_buf_empty(buf))
777 mask |= POLLIN | POLLRDNORM;
780 return mask;
784 * relay_file_release - release file op for relay files
785 * @inode: the inode
786 * @filp: the file
788 * Decrements the channel refcount, as the filesystem is
789 * no longer using it.
791 static int relay_file_release(struct inode *inode, struct file *filp)
793 struct rchan_buf *buf = filp->private_data;
794 kref_put(&buf->kref, relay_remove_buf);
796 return 0;
800 * relay_file_read_consume - update the consumed count for the buffer
802 static void relay_file_read_consume(struct rchan_buf *buf,
803 size_t read_pos,
804 size_t bytes_consumed)
806 size_t subbuf_size = buf->chan->subbuf_size;
807 size_t n_subbufs = buf->chan->n_subbufs;
808 size_t read_subbuf;
810 if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
811 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
812 buf->bytes_consumed = 0;
815 buf->bytes_consumed += bytes_consumed;
816 if (!read_pos)
817 read_subbuf = buf->subbufs_consumed % n_subbufs;
818 else
819 read_subbuf = read_pos / buf->chan->subbuf_size;
820 if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
821 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
822 (buf->offset == subbuf_size))
823 return;
824 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
825 buf->bytes_consumed = 0;
830 * relay_file_read_avail - boolean, are there unconsumed bytes available?
832 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
834 size_t subbuf_size = buf->chan->subbuf_size;
835 size_t n_subbufs = buf->chan->n_subbufs;
836 size_t produced = buf->subbufs_produced;
837 size_t consumed = buf->subbufs_consumed;
839 relay_file_read_consume(buf, read_pos, 0);
841 if (unlikely(buf->offset > subbuf_size)) {
842 if (produced == consumed)
843 return 0;
844 return 1;
847 if (unlikely(produced - consumed >= n_subbufs)) {
848 consumed = produced - n_subbufs + 1;
849 buf->subbufs_consumed = consumed;
850 buf->bytes_consumed = 0;
853 produced = (produced % n_subbufs) * subbuf_size + buf->offset;
854 consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
856 if (consumed > produced)
857 produced += n_subbufs * subbuf_size;
859 if (consumed == produced)
860 return 0;
862 return 1;
866 * relay_file_read_subbuf_avail - return bytes available in sub-buffer
867 * @read_pos: file read position
868 * @buf: relay channel buffer
870 static size_t relay_file_read_subbuf_avail(size_t read_pos,
871 struct rchan_buf *buf)
873 size_t padding, avail = 0;
874 size_t read_subbuf, read_offset, write_subbuf, write_offset;
875 size_t subbuf_size = buf->chan->subbuf_size;
877 write_subbuf = (buf->data - buf->start) / subbuf_size;
878 write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
879 read_subbuf = read_pos / subbuf_size;
880 read_offset = read_pos % subbuf_size;
881 padding = buf->padding[read_subbuf];
883 if (read_subbuf == write_subbuf) {
884 if (read_offset + padding < write_offset)
885 avail = write_offset - (read_offset + padding);
886 } else
887 avail = (subbuf_size - padding) - read_offset;
889 return avail;
893 * relay_file_read_start_pos - find the first available byte to read
894 * @read_pos: file read position
895 * @buf: relay channel buffer
897 * If the @read_pos is in the middle of padding, return the
898 * position of the first actually available byte, otherwise
899 * return the original value.
901 static size_t relay_file_read_start_pos(size_t read_pos,
902 struct rchan_buf *buf)
904 size_t read_subbuf, padding, padding_start, padding_end;
905 size_t subbuf_size = buf->chan->subbuf_size;
906 size_t n_subbufs = buf->chan->n_subbufs;
907 size_t consumed = buf->subbufs_consumed % n_subbufs;
909 if (!read_pos)
910 read_pos = consumed * subbuf_size + buf->bytes_consumed;
911 read_subbuf = read_pos / subbuf_size;
912 padding = buf->padding[read_subbuf];
913 padding_start = (read_subbuf + 1) * subbuf_size - padding;
914 padding_end = (read_subbuf + 1) * subbuf_size;
915 if (read_pos >= padding_start && read_pos < padding_end) {
916 read_subbuf = (read_subbuf + 1) % n_subbufs;
917 read_pos = read_subbuf * subbuf_size;
920 return read_pos;
924 * relay_file_read_end_pos - return the new read position
925 * @read_pos: file read position
926 * @buf: relay channel buffer
927 * @count: number of bytes to be read
929 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
930 size_t read_pos,
931 size_t count)
933 size_t read_subbuf, padding, end_pos;
934 size_t subbuf_size = buf->chan->subbuf_size;
935 size_t n_subbufs = buf->chan->n_subbufs;
937 read_subbuf = read_pos / subbuf_size;
938 padding = buf->padding[read_subbuf];
939 if (read_pos % subbuf_size + count + padding == subbuf_size)
940 end_pos = (read_subbuf + 1) * subbuf_size;
941 else
942 end_pos = read_pos + count;
943 if (end_pos >= subbuf_size * n_subbufs)
944 end_pos = 0;
946 return end_pos;
950 * subbuf_read_actor - read up to one subbuf's worth of data
952 static int subbuf_read_actor(size_t read_start,
953 struct rchan_buf *buf,
954 size_t avail,
955 read_descriptor_t *desc,
956 read_actor_t actor)
958 void *from;
959 int ret = 0;
961 from = buf->start + read_start;
962 ret = avail;
963 if (copy_to_user(desc->arg.buf, from, avail)) {
964 desc->error = -EFAULT;
965 ret = 0;
967 desc->arg.data += ret;
968 desc->written += ret;
969 desc->count -= ret;
971 return ret;
975 * subbuf_send_actor - send up to one subbuf's worth of data
977 static int subbuf_send_actor(size_t read_start,
978 struct rchan_buf *buf,
979 size_t avail,
980 read_descriptor_t *desc,
981 read_actor_t actor)
983 unsigned long pidx, poff;
984 unsigned int subbuf_pages;
985 int ret = 0;
987 subbuf_pages = buf->chan->alloc_size >> PAGE_SHIFT;
988 pidx = (read_start / PAGE_SIZE) % subbuf_pages;
989 poff = read_start & ~PAGE_MASK;
990 while (avail) {
991 struct page *p = buf->page_array[pidx];
992 unsigned int len;
994 len = PAGE_SIZE - poff;
995 if (len > avail)
996 len = avail;
998 len = actor(desc, p, poff, len);
999 if (desc->error)
1000 break;
1002 avail -= len;
1003 ret += len;
1004 poff = 0;
1005 pidx = (pidx + 1) % subbuf_pages;
1008 return ret;
1011 typedef int (*subbuf_actor_t) (size_t read_start,
1012 struct rchan_buf *buf,
1013 size_t avail,
1014 read_descriptor_t *desc,
1015 read_actor_t actor);
1018 * relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1020 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1021 subbuf_actor_t subbuf_actor,
1022 read_actor_t actor,
1023 read_descriptor_t *desc)
1025 struct rchan_buf *buf = filp->private_data;
1026 size_t read_start, avail;
1027 int ret;
1029 if (!desc->count)
1030 return 0;
1032 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
1033 do {
1034 if (!relay_file_read_avail(buf, *ppos))
1035 break;
1037 read_start = relay_file_read_start_pos(*ppos, buf);
1038 avail = relay_file_read_subbuf_avail(read_start, buf);
1039 if (!avail)
1040 break;
1042 avail = min(desc->count, avail);
1043 ret = subbuf_actor(read_start, buf, avail, desc, actor);
1044 if (desc->error < 0)
1045 break;
1047 if (ret) {
1048 relay_file_read_consume(buf, read_start, ret);
1049 *ppos = relay_file_read_end_pos(buf, read_start, ret);
1051 } while (desc->count && ret);
1052 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
1054 return desc->written;
1057 static ssize_t relay_file_read(struct file *filp,
1058 char __user *buffer,
1059 size_t count,
1060 loff_t *ppos)
1062 read_descriptor_t desc;
1063 desc.written = 0;
1064 desc.count = count;
1065 desc.arg.buf = buffer;
1066 desc.error = 0;
1067 return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
1068 NULL, &desc);
1071 static ssize_t relay_file_sendfile(struct file *filp,
1072 loff_t *ppos,
1073 size_t count,
1074 read_actor_t actor,
1075 void *target)
1077 read_descriptor_t desc;
1078 desc.written = 0;
1079 desc.count = count;
1080 desc.arg.data = target;
1081 desc.error = 0;
1082 return relay_file_read_subbufs(filp, ppos, subbuf_send_actor,
1083 actor, &desc);
1086 const struct file_operations relay_file_operations = {
1087 .open = relay_file_open,
1088 .poll = relay_file_poll,
1089 .mmap = relay_file_mmap,
1090 .read = relay_file_read,
1091 .llseek = no_llseek,
1092 .release = relay_file_release,
1093 .sendfile = relay_file_sendfile,
1095 EXPORT_SYMBOL_GPL(relay_file_operations);
1097 static __init int relay_init(void)
1100 hotcpu_notifier(relay_hotcpu_callback, 0);
1101 return 0;
1104 module_init(relay_init);