projects/linux·blob v2.6.21 · refs

1097 lines · 26.2 KiB

1/*
2 * Public API and common code for kernel->userspace relay file support.
3 *
4 * See Documentation/filesystems/relayfs.txt for an overview of relayfs.
5 *
6 * Copyright (C) 2002-2005 - Tom Zanussi ([email protected]), IBM Corp
7 * Copyright (C) 1999-2005 - Karim Yaghmour ([email protected])
8 *
9 * Moved to kernel/relay.c by Paul Mundt, 2006.
10 * November 2006 - CPU hotplug support by Mathieu Desnoyers
11 * ([email protected])
12 *
13 * This file is released under the GPL.
14 */
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>
24
25/* list of open channels, for cpu hotplug */
26static DEFINE_MUTEX(relay_channels_mutex);
27static LIST_HEAD(relay_channels);
28
29/*
30 * close() vm_op implementation for relay file mapping.
31 */
32static void relay_file_mmap_close(struct vm_area_struct *vma)
33{
34 struct rchan_buf *buf = vma->vm_private_data;
35 buf->chan->cb->buf_unmapped(buf, vma->vm_file);
36}
37
38/*
39 * nopage() vm_op implementation for relay file mapping.
40 */
41static struct page *relay_buf_nopage(struct vm_area_struct *vma,
42 unsigned long address,
43 int *type)
44{
45 struct page *page;
46 struct rchan_buf *buf = vma->vm_private_data;
47 unsigned long offset = address - vma->vm_start;
48
49 if (address > vma->vm_end)
50 return NOPAGE_SIGBUS; /* Disallow mremap */
51 if (!buf)
52 return NOPAGE_OOM;
53
54 page = vmalloc_to_page(buf->start + offset);
55 if (!page)
56 return NOPAGE_OOM;
57 get_page(page);
58
59 if (type)
60 *type = VM_FAULT_MINOR;
61
62 return page;
63}
64
65/*
66 * vm_ops for relay file mappings.
67 */
68static struct vm_operations_struct relay_file_mmap_ops = {
69 .nopage = relay_buf_nopage,
70 .close = relay_file_mmap_close,
71};
72
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
77 *
78 * Returns 0 if ok, negative on error
79 *
80 * Caller should already have grabbed mmap_sem.
81 */
82int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
83{
84 unsigned long length = vma->vm_end - vma->vm_start;
85 struct file *filp = vma->vm_file;
86
87 if (!buf)
88 return -EBADF;
89
90 if (length != (unsigned long)buf->chan->alloc_size)
91 return -EINVAL;
92
93 vma->vm_ops = &relay_file_mmap_ops;
94 vma->vm_private_data = buf;
95 buf->chan->cb->buf_mapped(buf, filp);
96
97 return 0;
98}
99
100/**
101 * relay_alloc_buf - allocate a channel buffer
102 * @buf: the buffer struct
103 * @size: total size of the buffer
104 *
105 * Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
106 * passed in size will get page aligned, if it isn't already.
107 */
108static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
109{
110 void *mem;
111 unsigned int i, j, n_pages;
112
113 *size = PAGE_ALIGN(*size);
114 n_pages = *size >> PAGE_SHIFT;
115
116 buf->page_array = kcalloc(n_pages, sizeof(struct page *), GFP_KERNEL);
117 if (!buf->page_array)
118 return NULL;
119
120 for (i = 0; i < n_pages; i++) {
121 buf->page_array[i] = alloc_page(GFP_KERNEL);
122 if (unlikely(!buf->page_array[i]))
123 goto depopulate;
124 }
125 mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
126 if (!mem)
127 goto depopulate;
128
129 memset(mem, 0, *size);
130 buf->page_count = n_pages;
131 return mem;
132
133depopulate:
134 for (j = 0; j < i; j++)
135 __free_page(buf->page_array[j]);
136 kfree(buf->page_array);
137 return NULL;
138}
139
140/**
141 * relay_create_buf - allocate and initialize a channel buffer
142 * @chan: the relay channel
143 *
144 * Returns channel buffer if successful, %NULL otherwise.
145 */
146struct rchan_buf *relay_create_buf(struct rchan *chan)
147{
148 struct rchan_buf *buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
149 if (!buf)
150 return NULL;
151
152 buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
153 if (!buf->padding)
154 goto free_buf;
155
156 buf->start = relay_alloc_buf(buf, &chan->alloc_size);
157 if (!buf->start)
158 goto free_buf;
159
160 buf->chan = chan;
161 kref_get(&buf->chan->kref);
162 return buf;
163
164free_buf:
165 kfree(buf->padding);
166 kfree(buf);
167 return NULL;
168}
169
170/**
171 * relay_destroy_channel - free the channel struct
172 * @kref: target kernel reference that contains the relay channel
173 *
174 * Should only be called from kref_put().
175 */
176void relay_destroy_channel(struct kref *kref)
177{
178 struct rchan *chan = container_of(kref, struct rchan, kref);
179 kfree(chan);
180}
181
182/**
183 * relay_destroy_buf - destroy an rchan_buf struct and associated buffer
184 * @buf: the buffer struct
185 */
186void relay_destroy_buf(struct rchan_buf *buf)
187{
188 struct rchan *chan = buf->chan;
189 unsigned int i;
190
191 if (likely(buf->start)) {
192 vunmap(buf->start);
193 for (i = 0; i < buf->page_count; i++)
194 __free_page(buf->page_array[i]);
195 kfree(buf->page_array);
196 }
197 chan->buf[buf->cpu] = NULL;
198 kfree(buf->padding);
199 kfree(buf);
200 kref_put(&chan->kref, relay_destroy_channel);
201}
202
203/**
204 * relay_remove_buf - remove a channel buffer
205 * @kref: target kernel reference that contains the relay buffer
206 *
207 * Removes the file from the fileystem, which also frees the
208 * rchan_buf_struct and the channel buffer. Should only be called from
209 * kref_put().
210 */
211void relay_remove_buf(struct kref *kref)
212{
213 struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
214 buf->chan->cb->remove_buf_file(buf->dentry);
215 relay_destroy_buf(buf);
216}
217
218/**
219 * relay_buf_empty - boolean, is the channel buffer empty?
220 * @buf: channel buffer
221 *
222 * Returns 1 if the buffer is empty, 0 otherwise.
223 */
224int relay_buf_empty(struct rchan_buf *buf)
225{
226 return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
227}
228EXPORT_SYMBOL_GPL(relay_buf_empty);
229
230/**
231 * relay_buf_full - boolean, is the channel buffer full?
232 * @buf: channel buffer
233 *
234 * Returns 1 if the buffer is full, 0 otherwise.
235 */
236int relay_buf_full(struct rchan_buf *buf)
237{
238 size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
239 return (ready >= buf->chan->n_subbufs) ? 1 : 0;
240}
241EXPORT_SYMBOL_GPL(relay_buf_full);
242
243/*
244 * High-level relay kernel API and associated functions.
245 */
246
247/*
248 * rchan_callback implementations defining default channel behavior. Used
249 * in place of corresponding NULL values in client callback struct.
250 */
251
252/*
253 * subbuf_start() default callback. Does nothing.
254 */
255static int subbuf_start_default_callback (struct rchan_buf *buf,
256 void *subbuf,
257 void *prev_subbuf,
258 size_t prev_padding)
259{
260 if (relay_buf_full(buf))
261 return 0;
262
263 return 1;
264}
265
266/*
267 * buf_mapped() default callback. Does nothing.
268 */
269static void buf_mapped_default_callback(struct rchan_buf *buf,
270 struct file *filp)
271{
272}
273
274/*
275 * buf_unmapped() default callback. Does nothing.
276 */
277static void buf_unmapped_default_callback(struct rchan_buf *buf,
278 struct file *filp)
279{
280}
281
282/*
283 * create_buf_file_create() default callback. Does nothing.
284 */
285static struct dentry *create_buf_file_default_callback(const char *filename,
286 struct dentry *parent,
287 int mode,
288 struct rchan_buf *buf,
289 int *is_global)
290{
291 return NULL;
292}
293
294/*
295 * remove_buf_file() default callback. Does nothing.
296 */
297static int remove_buf_file_default_callback(struct dentry *dentry)
298{
299 return -EINVAL;
300}
301
302/* relay channel default callbacks */
303static struct rchan_callbacks default_channel_callbacks = {
304 .subbuf_start = subbuf_start_default_callback,
305 .buf_mapped = buf_mapped_default_callback,
306 .buf_unmapped = buf_unmapped_default_callback,
307 .create_buf_file = create_buf_file_default_callback,
308 .remove_buf_file = remove_buf_file_default_callback,
309};
310
311/**
312 * wakeup_readers - wake up readers waiting on a channel
313 * @work: work struct that contains the the channel buffer
314 *
315 * This is the work function used to defer reader waking. The
316 * reason waking is deferred is that calling directly from write
317 * causes problems if you're writing from say the scheduler.
318 */
319static void wakeup_readers(struct work_struct *work)
320{
321 struct rchan_buf *buf =
322 container_of(work, struct rchan_buf, wake_readers.work);
323 wake_up_interruptible(&buf->read_wait);
324}
325
326/**
327 * __relay_reset - reset a channel buffer
328 * @buf: the channel buffer
329 * @init: 1 if this is a first-time initialization
330 *
331 * See relay_reset() for description of effect.
332 */
333static void __relay_reset(struct rchan_buf *buf, unsigned int init)
334{
335 size_t i;
336
337 if (init) {
338 init_waitqueue_head(&buf->read_wait);
339 kref_init(&buf->kref);
340 INIT_DELAYED_WORK(&buf->wake_readers, NULL);
341 } else {
342 cancel_delayed_work(&buf->wake_readers);
343 flush_scheduled_work();
344 }
345
346 buf->subbufs_produced = 0;
347 buf->subbufs_consumed = 0;
348 buf->bytes_consumed = 0;
349 buf->finalized = 0;
350 buf->data = buf->start;
351 buf->offset = 0;
352
353 for (i = 0; i < buf->chan->n_subbufs; i++)
354 buf->padding[i] = 0;
355
356 buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
357}
358
359/**
360 * relay_reset - reset the channel
361 * @chan: the channel
362 *
363 * This has the effect of erasing all data from all channel buffers
364 * and restarting the channel in its initial state. The buffers
365 * are not freed, so any mappings are still in effect.
366 *
367 * NOTE. Care should be taken that the channel isn't actually
368 * being used by anything when this call is made.
369 */
370void relay_reset(struct rchan *chan)
371{
372 unsigned int i;
373
374 if (!chan)
375 return;
376
377 if (chan->is_global && chan->buf[0]) {
378 __relay_reset(chan->buf[0], 0);
379 return;
380 }
381
382 mutex_lock(&relay_channels_mutex);
383 for_each_online_cpu(i)
384 if (chan->buf[i])
385 __relay_reset(chan->buf[i], 0);
386 mutex_unlock(&relay_channels_mutex);
387}
388EXPORT_SYMBOL_GPL(relay_reset);
389
390/*
391 * relay_open_buf - create a new relay channel buffer
392 *
393 * used by relay_open() and CPU hotplug.
394 */
395static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
396{
397 struct rchan_buf *buf = NULL;
398 struct dentry *dentry;
399 char *tmpname;
400
401 if (chan->is_global)
402 return chan->buf[0];
403
404 tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
405 if (!tmpname)
406 goto end;
407 snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
408
409 buf = relay_create_buf(chan);
410 if (!buf)
411 goto free_name;
412
413 buf->cpu = cpu;
414 __relay_reset(buf, 1);
415
416 /* Create file in fs */
417 dentry = chan->cb->create_buf_file(tmpname, chan->parent, S_IRUSR,
418 buf, &chan->is_global);
419 if (!dentry)
420 goto free_buf;
421
422 buf->dentry = dentry;
423
424 if(chan->is_global) {
425 chan->buf[0] = buf;
426 buf->cpu = 0;
427 }
428
429 goto free_name;
430
431free_buf:
432 relay_destroy_buf(buf);
433free_name:
434 kfree(tmpname);
435end:
436 return buf;
437}
438
439/**
440 * relay_close_buf - close a channel buffer
441 * @buf: channel buffer
442 *
443 * Marks the buffer finalized and restores the default callbacks.
444 * The channel buffer and channel buffer data structure are then freed
445 * automatically when the last reference is given up.
446 */
447static void relay_close_buf(struct rchan_buf *buf)
448{
449 buf->finalized = 1;
450 cancel_delayed_work(&buf->wake_readers);
451 flush_scheduled_work();
452 kref_put(&buf->kref, relay_remove_buf);
453}
454
455static void setup_callbacks(struct rchan *chan,
456 struct rchan_callbacks *cb)
457{
458 if (!cb) {
459 chan->cb = &default_channel_callbacks;
460 return;
461 }
462
463 if (!cb->subbuf_start)
464 cb->subbuf_start = subbuf_start_default_callback;
465 if (!cb->buf_mapped)
466 cb->buf_mapped = buf_mapped_default_callback;
467 if (!cb->buf_unmapped)
468 cb->buf_unmapped = buf_unmapped_default_callback;
469 if (!cb->create_buf_file)
470 cb->create_buf_file = create_buf_file_default_callback;
471 if (!cb->remove_buf_file)
472 cb->remove_buf_file = remove_buf_file_default_callback;
473 chan->cb = cb;
474}
475
476/**
477 * relay_hotcpu_callback - CPU hotplug callback
478 * @nb: notifier block
479 * @action: hotplug action to take
480 * @hcpu: CPU number
481 *
482 * Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
483 */
484static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb,
485 unsigned long action,
486 void *hcpu)
487{
488 unsigned int hotcpu = (unsigned long)hcpu;
489 struct rchan *chan;
490
491 switch(action) {
492 case CPU_UP_PREPARE:
493 mutex_lock(&relay_channels_mutex);
494 list_for_each_entry(chan, &relay_channels, list) {
495 if (chan->buf[hotcpu])
496 continue;
497 chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
498 if(!chan->buf[hotcpu]) {
499 printk(KERN_ERR
500 "relay_hotcpu_callback: cpu %d buffer "
501 "creation failed\n", hotcpu);
502 mutex_unlock(&relay_channels_mutex);
503 return NOTIFY_BAD;
504 }
505 }
506 mutex_unlock(&relay_channels_mutex);
507 break;
508 case CPU_DEAD:
509 /* No need to flush the cpu : will be flushed upon
510 * final relay_flush() call. */
511 break;
512 }
513 return NOTIFY_OK;
514}
515
516/**
517 * relay_open - create a new relay channel
518 * @base_filename: base name of files to create
519 * @parent: dentry of parent directory, %NULL for root directory
520 * @subbuf_size: size of sub-buffers
521 * @n_subbufs: number of sub-buffers
522 * @cb: client callback functions
523 * @private_data: user-defined data
524 *
525 * Returns channel pointer if successful, %NULL otherwise.
526 *
527 * Creates a channel buffer for each cpu using the sizes and
528 * attributes specified. The created channel buffer files
529 * will be named base_filename0...base_filenameN-1. File
530 * permissions will be %S_IRUSR.
531 */
532struct rchan *relay_open(const char *base_filename,
533 struct dentry *parent,
534 size_t subbuf_size,
535 size_t n_subbufs,
536 struct rchan_callbacks *cb,
537 void *private_data)
538{
539 unsigned int i;
540 struct rchan *chan;
541 if (!base_filename)
542 return NULL;
543
544 if (!(subbuf_size && n_subbufs))
545 return NULL;
546
547 chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
548 if (!chan)
549 return NULL;
550
551 chan->version = RELAYFS_CHANNEL_VERSION;
552 chan->n_subbufs = n_subbufs;
553 chan->subbuf_size = subbuf_size;
554 chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
555 chan->parent = parent;
556 chan->private_data = private_data;
557 strlcpy(chan->base_filename, base_filename, NAME_MAX);
558 setup_callbacks(chan, cb);
559 kref_init(&chan->kref);
560
561 mutex_lock(&relay_channels_mutex);
562 for_each_online_cpu(i) {
563 chan->buf[i] = relay_open_buf(chan, i);
564 if (!chan->buf[i])
565 goto free_bufs;
566 }
567 list_add(&chan->list, &relay_channels);
568 mutex_unlock(&relay_channels_mutex);
569
570 return chan;
571
572free_bufs:
573 for_each_online_cpu(i) {
574 if (!chan->buf[i])
575 break;
576 relay_close_buf(chan->buf[i]);
577 }
578
579 kref_put(&chan->kref, relay_destroy_channel);
580 mutex_unlock(&relay_channels_mutex);
581 return NULL;
582}
583EXPORT_SYMBOL_GPL(relay_open);
584
585/**
586 * relay_switch_subbuf - switch to a new sub-buffer
587 * @buf: channel buffer
588 * @length: size of current event
589 *
590 * Returns either the length passed in or 0 if full.
591 *
592 * Performs sub-buffer-switch tasks such as invoking callbacks,
593 * updating padding counts, waking up readers, etc.
594 */
595size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
596{
597 void *old, *new;
598 size_t old_subbuf, new_subbuf;
599
600 if (unlikely(length > buf->chan->subbuf_size))
601 goto toobig;
602
603 if (buf->offset != buf->chan->subbuf_size + 1) {
604 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
605 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
606 buf->padding[old_subbuf] = buf->prev_padding;
607 buf->subbufs_produced++;
608 buf->dentry->d_inode->i_size += buf->chan->subbuf_size -
609 buf->padding[old_subbuf];
610 smp_mb();
611 if (waitqueue_active(&buf->read_wait)) {
612 PREPARE_DELAYED_WORK(&buf->wake_readers,
613 wakeup_readers);
614 schedule_delayed_work(&buf->wake_readers, 1);
615 }
616 }
617
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;
625 }
626 buf->data = new;
627 buf->padding[new_subbuf] = 0;
628
629 if (unlikely(length + buf->offset > buf->chan->subbuf_size))
630 goto toobig;
631
632 return length;
633
634toobig:
635 buf->chan->last_toobig = length;
636 return 0;
637}
638EXPORT_SYMBOL_GPL(relay_switch_subbuf);
639
640/**
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
645 *
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.
649 *
650 * NOTE. Kernel clients don't need to call this function if the channel
651 * mode is 'overwrite'.
652 */
653void relay_subbufs_consumed(struct rchan *chan,
654 unsigned int cpu,
655 size_t subbufs_consumed)
656{
657 struct rchan_buf *buf;
658
659 if (!chan)
660 return;
661
662 if (cpu >= NR_CPUS || !chan->buf[cpu])
663 return;
664
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;
669}
670EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
671
672/**
673 * relay_close - close the channel
674 * @chan: the channel
675 *
676 * Closes all channel buffers and frees the channel.
677 */
678void relay_close(struct rchan *chan)
679{
680 unsigned int i;
681
682 if (!chan)
683 return;
684
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]);
692
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);
697
698 list_del(&chan->list);
699 kref_put(&chan->kref, relay_destroy_channel);
700 mutex_unlock(&relay_channels_mutex);
701}
702EXPORT_SYMBOL_GPL(relay_close);
703
704/**
705 * relay_flush - close the channel
706 * @chan: the channel
707 *
708 * Flushes all channel buffers, i.e. forces buffer switch.
709 */
710void relay_flush(struct rchan *chan)
711{
712 unsigned int i;
713
714 if (!chan)
715 return;
716
717 if (chan->is_global && chan->buf[0]) {
718 relay_switch_subbuf(chan->buf[0], 0);
719 return;
720 }
721
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);
727}
728EXPORT_SYMBOL_GPL(relay_flush);
729
730/**
731 * relay_file_open - open file op for relay files
732 * @inode: the inode
733 * @filp: the file
734 *
735 * Increments the channel buffer refcount.
736 */
737static int relay_file_open(struct inode *inode, struct file *filp)
738{
739 struct rchan_buf *buf = inode->i_private;
740 kref_get(&buf->kref);
741 filp->private_data = buf;
742
743 return 0;
744}
745
746/**
747 * relay_file_mmap - mmap file op for relay files
748 * @filp: the file
749 * @vma: the vma describing what to map
750 *
751 * Calls upon relay_mmap_buf() to map the file into user space.
752 */
753static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
754{
755 struct rchan_buf *buf = filp->private_data;
756 return relay_mmap_buf(buf, vma);
757}
758
759/**
760 * relay_file_poll - poll file op for relay files
761 * @filp: the file
762 * @wait: poll table
763 *
764 * Poll implemention.
765 */
766static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
767{
768 unsigned int mask = 0;
769 struct rchan_buf *buf = filp->private_data;
770
771 if (buf->finalized)
772 return POLLERR;
773
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;
778 }
779
780 return mask;
781}
782
783/**
784 * relay_file_release - release file op for relay files
785 * @inode: the inode
786 * @filp: the file
787 *
788 * Decrements the channel refcount, as the filesystem is
789 * no longer using it.
790 */
791static int relay_file_release(struct inode *inode, struct file *filp)
792{
793 struct rchan_buf *buf = filp->private_data;
794 kref_put(&buf->kref, relay_remove_buf);
795
796 return 0;
797}
798
799/*
800 * relay_file_read_consume - update the consumed count for the buffer
801 */
802static void relay_file_read_consume(struct rchan_buf *buf,
803 size_t read_pos,
804 size_t bytes_consumed)
805{
806 size_t subbuf_size = buf->chan->subbuf_size;
807 size_t n_subbufs = buf->chan->n_subbufs;
808 size_t read_subbuf;
809
810 if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
811 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
812 buf->bytes_consumed = 0;
813 }
814
815 buf->bytes_consumed += bytes_consumed;
816 read_subbuf = read_pos / buf->chan->subbuf_size;
817 if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
818 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
819 (buf->offset == subbuf_size))
820 return;
821 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
822 buf->bytes_consumed = 0;
823 }
824}
825
826/*
827 * relay_file_read_avail - boolean, are there unconsumed bytes available?
828 */
829static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
830{
831 size_t subbuf_size = buf->chan->subbuf_size;
832 size_t n_subbufs = buf->chan->n_subbufs;
833 size_t produced = buf->subbufs_produced;
834 size_t consumed = buf->subbufs_consumed;
835
836 relay_file_read_consume(buf, read_pos, 0);
837
838 if (unlikely(buf->offset > subbuf_size)) {
839 if (produced == consumed)
840 return 0;
841 return 1;
842 }
843
844 if (unlikely(produced - consumed >= n_subbufs)) {
845 consumed = (produced / n_subbufs) * n_subbufs;
846 buf->subbufs_consumed = consumed;
847 }
848
849 produced = (produced % n_subbufs) * subbuf_size + buf->offset;
850 consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
851
852 if (consumed > produced)
853 produced += n_subbufs * subbuf_size;
854
855 if (consumed == produced)
856 return 0;
857
858 return 1;
859}
860
861/**
862 * relay_file_read_subbuf_avail - return bytes available in sub-buffer
863 * @read_pos: file read position
864 * @buf: relay channel buffer
865 */
866static size_t relay_file_read_subbuf_avail(size_t read_pos,
867 struct rchan_buf *buf)
868{
869 size_t padding, avail = 0;
870 size_t read_subbuf, read_offset, write_subbuf, write_offset;
871 size_t subbuf_size = buf->chan->subbuf_size;
872
873 write_subbuf = (buf->data - buf->start) / subbuf_size;
874 write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
875 read_subbuf = read_pos / subbuf_size;
876 read_offset = read_pos % subbuf_size;
877 padding = buf->padding[read_subbuf];
878
879 if (read_subbuf == write_subbuf) {
880 if (read_offset + padding < write_offset)
881 avail = write_offset - (read_offset + padding);
882 } else
883 avail = (subbuf_size - padding) - read_offset;
884
885 return avail;
886}
887
888/**
889 * relay_file_read_start_pos - find the first available byte to read
890 * @read_pos: file read position
891 * @buf: relay channel buffer
892 *
893 * If the @read_pos is in the middle of padding, return the
894 * position of the first actually available byte, otherwise
895 * return the original value.
896 */
897static size_t relay_file_read_start_pos(size_t read_pos,
898 struct rchan_buf *buf)
899{
900 size_t read_subbuf, padding, padding_start, padding_end;
901 size_t subbuf_size = buf->chan->subbuf_size;
902 size_t n_subbufs = buf->chan->n_subbufs;
903
904 read_subbuf = read_pos / subbuf_size;
905 padding = buf->padding[read_subbuf];
906 padding_start = (read_subbuf + 1) * subbuf_size - padding;
907 padding_end = (read_subbuf + 1) * subbuf_size;
908 if (read_pos >= padding_start && read_pos < padding_end) {
909 read_subbuf = (read_subbuf + 1) % n_subbufs;
910 read_pos = read_subbuf * subbuf_size;
911 }
912
913 return read_pos;
914}
915
916/**
917 * relay_file_read_end_pos - return the new read position
918 * @read_pos: file read position
919 * @buf: relay channel buffer
920 * @count: number of bytes to be read
921 */
922static size_t relay_file_read_end_pos(struct rchan_buf *buf,
923 size_t read_pos,
924 size_t count)
925{
926 size_t read_subbuf, padding, end_pos;
927 size_t subbuf_size = buf->chan->subbuf_size;
928 size_t n_subbufs = buf->chan->n_subbufs;
929
930 read_subbuf = read_pos / subbuf_size;
931 padding = buf->padding[read_subbuf];
932 if (read_pos % subbuf_size + count + padding == subbuf_size)
933 end_pos = (read_subbuf + 1) * subbuf_size;
934 else
935 end_pos = read_pos + count;
936 if (end_pos >= subbuf_size * n_subbufs)
937 end_pos = 0;
938
939 return end_pos;
940}
941
942/*
943 * subbuf_read_actor - read up to one subbuf's worth of data
944 */
945static int subbuf_read_actor(size_t read_start,
946 struct rchan_buf *buf,
947 size_t avail,
948 read_descriptor_t *desc,
949 read_actor_t actor)
950{
951 void *from;
952 int ret = 0;
953
954 from = buf->start + read_start;
955 ret = avail;
956 if (copy_to_user(desc->arg.buf, from, avail)) {
957 desc->error = -EFAULT;
958 ret = 0;
959 }
960 desc->arg.data += ret;
961 desc->written += ret;
962 desc->count -= ret;
963
964 return ret;
965}
966
967/*
968 * subbuf_send_actor - send up to one subbuf's worth of data
969 */
970static int subbuf_send_actor(size_t read_start,
971 struct rchan_buf *buf,
972 size_t avail,
973 read_descriptor_t *desc,
974 read_actor_t actor)
975{
976 unsigned long pidx, poff;
977 unsigned int subbuf_pages;
978 int ret = 0;
979
980 subbuf_pages = buf->chan->alloc_size >> PAGE_SHIFT;
981 pidx = (read_start / PAGE_SIZE) % subbuf_pages;
982 poff = read_start & ~PAGE_MASK;
983 while (avail) {
984 struct page *p = buf->page_array[pidx];
985 unsigned int len;
986
987 len = PAGE_SIZE - poff;
988 if (len > avail)
989 len = avail;
990
991 len = actor(desc, p, poff, len);
992 if (desc->error)
993 break;
994
995 avail -= len;
996 ret += len;
997 poff = 0;
998 pidx = (pidx + 1) % subbuf_pages;
999 }
1000
1001 return ret;
1002}
1003
1004typedef int (*subbuf_actor_t) (size_t read_start,
1005 struct rchan_buf *buf,
1006 size_t avail,
1007 read_descriptor_t *desc,
1008 read_actor_t actor);
1009
1010/*
1011 * relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1012 */
1013static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1014 subbuf_actor_t subbuf_actor,
1015 read_actor_t actor,
1016 read_descriptor_t *desc)
1017{
1018 struct rchan_buf *buf = filp->private_data;
1019 size_t read_start, avail;
1020 int ret;
1021
1022 if (!desc->count)
1023 return 0;
1024
1025 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
1026 do {
1027 if (!relay_file_read_avail(buf, *ppos))
1028 break;
1029
1030 read_start = relay_file_read_start_pos(*ppos, buf);
1031 avail = relay_file_read_subbuf_avail(read_start, buf);
1032 if (!avail)
1033 break;
1034
1035 avail = min(desc->count, avail);
1036 ret = subbuf_actor(read_start, buf, avail, desc, actor);
1037 if (desc->error < 0)
1038 break;
1039
1040 if (ret) {
1041 relay_file_read_consume(buf, read_start, ret);
1042 *ppos = relay_file_read_end_pos(buf, read_start, ret);
1043 }
1044 } while (desc->count && ret);
1045 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
1046
1047 return desc->written;
1048}
1049
1050static ssize_t relay_file_read(struct file *filp,
1051 char __user *buffer,
1052 size_t count,
1053 loff_t *ppos)
1054{
1055 read_descriptor_t desc;
1056 desc.written = 0;
1057 desc.count = count;
1058 desc.arg.buf = buffer;
1059 desc.error = 0;
1060 return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
1061 NULL, &desc);
1062}
1063
1064static ssize_t relay_file_sendfile(struct file *filp,
1065 loff_t *ppos,
1066 size_t count,
1067 read_actor_t actor,
1068 void *target)
1069{
1070 read_descriptor_t desc;
1071 desc.written = 0;
1072 desc.count = count;
1073 desc.arg.data = target;
1074 desc.error = 0;
1075 return relay_file_read_subbufs(filp, ppos, subbuf_send_actor,
1076 actor, &desc);
1077}
1078
1079const struct file_operations relay_file_operations = {
1080 .open = relay_file_open,
1081 .poll = relay_file_poll,
1082 .mmap = relay_file_mmap,
1083 .read = relay_file_read,
1084 .llseek = no_llseek,
1085 .release = relay_file_release,
1086 .sendfile = relay_file_sendfile,
1087};
1088EXPORT_SYMBOL_GPL(relay_file_operations);
1089
1090static __init int relay_init(void)
1091{
1092
1093 hotcpu_notifier(relay_hotcpu_callback, 0);
1094 return 0;
1095}
1096
1097module_init(relay_init);